WSL2-Linux-Kernel/drivers/mtd/ubi/kapi.c

778 строки
23 KiB
C
Исходник Обычный вид История

UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 12:22:22 +04:00
/*
* Copyright (c) International Business Machines Corp., 2006
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation; either version 2 of the License, or
* (at your option) any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See
* the GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program; if not, write to the Free Software
* Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
*
* Author: Artem Bityutskiy (Битюцкий Артём)
*/
/* This file mostly implements UBI kernel API functions */
#include <linux/module.h>
#include <linux/err.h>
include cleanup: Update gfp.h and slab.h includes to prepare for breaking implicit slab.h inclusion from percpu.h percpu.h is included by sched.h and module.h and thus ends up being included when building most .c files. percpu.h includes slab.h which in turn includes gfp.h making everything defined by the two files universally available and complicating inclusion dependencies. percpu.h -> slab.h dependency is about to be removed. Prepare for this change by updating users of gfp and slab facilities include those headers directly instead of assuming availability. As this conversion needs to touch large number of source files, the following script is used as the basis of conversion. http://userweb.kernel.org/~tj/misc/slabh-sweep.py The script does the followings. * Scan files for gfp and slab usages and update includes such that only the necessary includes are there. ie. if only gfp is used, gfp.h, if slab is used, slab.h. * When the script inserts a new include, it looks at the include blocks and try to put the new include such that its order conforms to its surrounding. It's put in the include block which contains core kernel includes, in the same order that the rest are ordered - alphabetical, Christmas tree, rev-Xmas-tree or at the end if there doesn't seem to be any matching order. * If the script can't find a place to put a new include (mostly because the file doesn't have fitting include block), it prints out an error message indicating which .h file needs to be added to the file. The conversion was done in the following steps. 1. The initial automatic conversion of all .c files updated slightly over 4000 files, deleting around 700 includes and adding ~480 gfp.h and ~3000 slab.h inclusions. The script emitted errors for ~400 files. 2. Each error was manually checked. Some didn't need the inclusion, some needed manual addition while adding it to implementation .h or embedding .c file was more appropriate for others. This step added inclusions to around 150 files. 3. The script was run again and the output was compared to the edits from #2 to make sure no file was left behind. 4. Several build tests were done and a couple of problems were fixed. e.g. lib/decompress_*.c used malloc/free() wrappers around slab APIs requiring slab.h to be added manually. 5. The script was run on all .h files but without automatically editing them as sprinkling gfp.h and slab.h inclusions around .h files could easily lead to inclusion dependency hell. Most gfp.h inclusion directives were ignored as stuff from gfp.h was usually wildly available and often used in preprocessor macros. Each slab.h inclusion directive was examined and added manually as necessary. 6. percpu.h was updated not to include slab.h. 7. Build test were done on the following configurations and failures were fixed. CONFIG_GCOV_KERNEL was turned off for all tests (as my distributed build env didn't work with gcov compiles) and a few more options had to be turned off depending on archs to make things build (like ipr on powerpc/64 which failed due to missing writeq). * x86 and x86_64 UP and SMP allmodconfig and a custom test config. * powerpc and powerpc64 SMP allmodconfig * sparc and sparc64 SMP allmodconfig * ia64 SMP allmodconfig * s390 SMP allmodconfig * alpha SMP allmodconfig * um on x86_64 SMP allmodconfig 8. percpu.h modifications were reverted so that it could be applied as a separate patch and serve as bisection point. Given the fact that I had only a couple of failures from tests on step 6, I'm fairly confident about the coverage of this conversion patch. If there is a breakage, it's likely to be something in one of the arch headers which should be easily discoverable easily on most builds of the specific arch. Signed-off-by: Tejun Heo <tj@kernel.org> Guess-its-ok-by: Christoph Lameter <cl@linux-foundation.org> Cc: Ingo Molnar <mingo@redhat.com> Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
2010-03-24 11:04:11 +03:00
#include <linux/slab.h>
#include <linux/namei.h>
#include <linux/fs.h>
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 12:22:22 +04:00
#include <asm/div64.h>
#include "ubi.h"
/**
* ubi_do_get_device_info - get information about UBI device.
* @ubi: UBI device description object
* @di: the information is stored here
*
* This function is the same as 'ubi_get_device_info()', but it assumes the UBI
* device is locked and cannot disappear.
*/
void ubi_do_get_device_info(struct ubi_device *ubi, struct ubi_device_info *di)
{
di->ubi_num = ubi->ubi_num;
di->leb_size = ubi->leb_size;
di->min_io_size = ubi->min_io_size;
di->ro_mode = ubi->ro_mode;
di->cdev = ubi->cdev.dev;
}
EXPORT_SYMBOL_GPL(ubi_do_get_device_info);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 12:22:22 +04:00
/**
* ubi_get_device_info - get information about UBI device.
* @ubi_num: UBI device number
* @di: the information is stored here
*
* This function returns %0 in case of success, %-EINVAL if the UBI device
* number is invalid, and %-ENODEV if there is no such UBI device.
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 12:22:22 +04:00
*/
int ubi_get_device_info(int ubi_num, struct ubi_device_info *di)
{
struct ubi_device *ubi;
if (ubi_num < 0 || ubi_num >= UBI_MAX_DEVICES)
return -EINVAL;
ubi = ubi_get_device(ubi_num);
if (!ubi)
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 12:22:22 +04:00
return -ENODEV;
ubi_do_get_device_info(ubi, di);
ubi_put_device(ubi);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 12:22:22 +04:00
return 0;
}
EXPORT_SYMBOL_GPL(ubi_get_device_info);
/**
* ubi_do_get_volume_info - get information about UBI volume.
* @ubi: UBI device description object
* @vol: volume description object
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 12:22:22 +04:00
* @vi: the information is stored here
*/
void ubi_do_get_volume_info(struct ubi_device *ubi, struct ubi_volume *vol,
struct ubi_volume_info *vi)
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 12:22:22 +04:00
{
vi->vol_id = vol->vol_id;
vi->ubi_num = ubi->ubi_num;
vi->size = vol->reserved_pebs;
vi->used_bytes = vol->used_bytes;
vi->vol_type = vol->vol_type;
vi->corrupted = vol->corrupted;
vi->upd_marker = vol->upd_marker;
vi->alignment = vol->alignment;
vi->usable_leb_size = vol->usable_leb_size;
vi->name_len = vol->name_len;
vi->name = vol->name;
vi->cdev = vol->cdev.dev;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 12:22:22 +04:00
}
/**
* ubi_get_volume_info - get information about UBI volume.
* @desc: volume descriptor
* @vi: the information is stored here
*/
void ubi_get_volume_info(struct ubi_volume_desc *desc,
struct ubi_volume_info *vi)
{
ubi_do_get_volume_info(desc->vol->ubi, desc->vol, vi);
}
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 12:22:22 +04:00
EXPORT_SYMBOL_GPL(ubi_get_volume_info);
/**
* ubi_open_volume - open UBI volume.
* @ubi_num: UBI device number
* @vol_id: volume ID
* @mode: open mode
*
* The @mode parameter specifies if the volume should be opened in read-only
* mode, read-write mode, or exclusive mode. The exclusive mode guarantees that
* nobody else will be able to open this volume. UBI allows to have many volume
* readers and one writer at a time.
*
* If a static volume is being opened for the first time since boot, it will be
* checked by this function, which means it will be fully read and the CRC
* checksum of each logical eraseblock will be checked.
*
* This function returns volume descriptor in case of success and a negative
* error code in case of failure.
*/
struct ubi_volume_desc *ubi_open_volume(int ubi_num, int vol_id, int mode)
{
int err;
struct ubi_volume_desc *desc;
struct ubi_device *ubi;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 12:22:22 +04:00
struct ubi_volume *vol;
dbg_gen("open device %d, volume %d, mode %d", ubi_num, vol_id, mode);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 12:22:22 +04:00
if (ubi_num < 0 || ubi_num >= UBI_MAX_DEVICES)
return ERR_PTR(-EINVAL);
if (mode != UBI_READONLY && mode != UBI_READWRITE &&
mode != UBI_EXCLUSIVE)
return ERR_PTR(-EINVAL);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 12:22:22 +04:00
/*
* First of all, we have to get the UBI device to prevent its removal.
*/
ubi = ubi_get_device(ubi_num);
if (!ubi)
return ERR_PTR(-ENODEV);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 12:22:22 +04:00
if (vol_id < 0 || vol_id >= ubi->vtbl_slots) {
err = -EINVAL;
goto out_put_ubi;
}
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 12:22:22 +04:00
desc = kmalloc(sizeof(struct ubi_volume_desc), GFP_KERNEL);
if (!desc) {
err = -ENOMEM;
goto out_put_ubi;
}
err = -ENODEV;
if (!try_module_get(THIS_MODULE))
goto out_free;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 12:22:22 +04:00
spin_lock(&ubi->volumes_lock);
vol = ubi->volumes[vol_id];
if (!vol)
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 12:22:22 +04:00
goto out_unlock;
err = -EBUSY;
switch (mode) {
case UBI_READONLY:
if (vol->exclusive)
goto out_unlock;
vol->readers += 1;
break;
case UBI_READWRITE:
if (vol->exclusive || vol->writers > 0)
goto out_unlock;
vol->writers += 1;
break;
case UBI_EXCLUSIVE:
if (vol->exclusive || vol->writers || vol->readers)
goto out_unlock;
vol->exclusive = 1;
break;
}
get_device(&vol->dev);
vol->ref_count += 1;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 12:22:22 +04:00
spin_unlock(&ubi->volumes_lock);
desc->vol = vol;
desc->mode = mode;
mutex_lock(&ubi->ckvol_mutex);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 12:22:22 +04:00
if (!vol->checked) {
/* This is the first open - check the volume */
err = ubi_check_volume(ubi, vol_id);
if (err < 0) {
mutex_unlock(&ubi->ckvol_mutex);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 12:22:22 +04:00
ubi_close_volume(desc);
return ERR_PTR(err);
}
if (err == 1) {
ubi_warn("volume %d on UBI device %d is corrupted",
vol_id, ubi->ubi_num);
vol->corrupted = 1;
}
vol->checked = 1;
}
mutex_unlock(&ubi->ckvol_mutex);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 12:22:22 +04:00
return desc;
out_unlock:
spin_unlock(&ubi->volumes_lock);
module_put(THIS_MODULE);
out_free:
kfree(desc);
out_put_ubi:
ubi_put_device(ubi);
dbg_err("cannot open device %d, volume %d, error %d",
ubi_num, vol_id, err);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 12:22:22 +04:00
return ERR_PTR(err);
}
EXPORT_SYMBOL_GPL(ubi_open_volume);
/**
* ubi_open_volume_nm - open UBI volume by name.
* @ubi_num: UBI device number
* @name: volume name
* @mode: open mode
*
* This function is similar to 'ubi_open_volume()', but opens a volume by name.
*/
struct ubi_volume_desc *ubi_open_volume_nm(int ubi_num, const char *name,
int mode)
{
int i, vol_id = -1, len;
struct ubi_device *ubi;
struct ubi_volume_desc *ret;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 12:22:22 +04:00
dbg_gen("open device %d, volume %s, mode %d", ubi_num, name, mode);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 12:22:22 +04:00
if (!name)
return ERR_PTR(-EINVAL);
len = strnlen(name, UBI_VOL_NAME_MAX + 1);
if (len > UBI_VOL_NAME_MAX)
return ERR_PTR(-EINVAL);
if (ubi_num < 0 || ubi_num >= UBI_MAX_DEVICES)
return ERR_PTR(-EINVAL);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 12:22:22 +04:00
ubi = ubi_get_device(ubi_num);
if (!ubi)
return ERR_PTR(-ENODEV);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 12:22:22 +04:00
spin_lock(&ubi->volumes_lock);
/* Walk all volumes of this UBI device */
for (i = 0; i < ubi->vtbl_slots; i++) {
struct ubi_volume *vol = ubi->volumes[i];
if (vol && len == vol->name_len && !strcmp(name, vol->name)) {
vol_id = i;
break;
}
}
spin_unlock(&ubi->volumes_lock);
if (vol_id >= 0)
ret = ubi_open_volume(ubi_num, vol_id, mode);
else
ret = ERR_PTR(-ENODEV);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 12:22:22 +04:00
/*
* We should put the UBI device even in case of success, because
* 'ubi_open_volume()' took a reference as well.
*/
ubi_put_device(ubi);
return ret;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 12:22:22 +04:00
}
EXPORT_SYMBOL_GPL(ubi_open_volume_nm);
/**
* ubi_open_volume_path - open UBI volume by its character device node path.
* @pathname: volume character device node path
* @mode: open mode
*
* This function is similar to 'ubi_open_volume()', but opens a volume the path
* to its character device node.
*/
struct ubi_volume_desc *ubi_open_volume_path(const char *pathname, int mode)
{
int error, ubi_num, vol_id, mod;
struct inode *inode;
struct path path;
dbg_gen("open volume %s, mode %d", pathname, mode);
if (!pathname || !*pathname)
return ERR_PTR(-EINVAL);
error = kern_path(pathname, LOOKUP_FOLLOW, &path);
if (error)
return ERR_PTR(error);
inode = path.dentry->d_inode;
mod = inode->i_mode;
ubi_num = ubi_major2num(imajor(inode));
vol_id = iminor(inode) - 1;
path_put(&path);
if (!S_ISCHR(mod))
return ERR_PTR(-EINVAL);
if (vol_id >= 0 && ubi_num >= 0)
return ubi_open_volume(ubi_num, vol_id, mode);
return ERR_PTR(-ENODEV);
}
EXPORT_SYMBOL_GPL(ubi_open_volume_path);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 12:22:22 +04:00
/**
* ubi_close_volume - close UBI volume.
* @desc: volume descriptor
*/
void ubi_close_volume(struct ubi_volume_desc *desc)
{
struct ubi_volume *vol = desc->vol;
struct ubi_device *ubi = vol->ubi;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 12:22:22 +04:00
dbg_gen("close device %d, volume %d, mode %d",
ubi->ubi_num, vol->vol_id, desc->mode);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 12:22:22 +04:00
spin_lock(&ubi->volumes_lock);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 12:22:22 +04:00
switch (desc->mode) {
case UBI_READONLY:
vol->readers -= 1;
break;
case UBI_READWRITE:
vol->writers -= 1;
break;
case UBI_EXCLUSIVE:
vol->exclusive = 0;
}
vol->ref_count -= 1;
spin_unlock(&ubi->volumes_lock);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 12:22:22 +04:00
kfree(desc);
put_device(&vol->dev);
ubi_put_device(ubi);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 12:22:22 +04:00
module_put(THIS_MODULE);
}
EXPORT_SYMBOL_GPL(ubi_close_volume);
/**
* ubi_leb_read - read data.
* @desc: volume descriptor
* @lnum: logical eraseblock number to read from
* @buf: buffer where to store the read data
* @offset: offset within the logical eraseblock to read from
* @len: how many bytes to read
* @check: whether UBI has to check the read data's CRC or not.
*
* This function reads data from offset @offset of logical eraseblock @lnum and
* stores the data at @buf. When reading from static volumes, @check specifies
* whether the data has to be checked or not. If yes, the whole logical
* eraseblock will be read and its CRC checksum will be checked (i.e., the CRC
* checksum is per-eraseblock). So checking may substantially slow down the
* read speed. The @check argument is ignored for dynamic volumes.
*
* In case of success, this function returns zero. In case of failure, this
* function returns a negative error code.
*
* %-EBADMSG error code is returned:
* o for both static and dynamic volumes if MTD driver has detected a data
* integrity problem (unrecoverable ECC checksum mismatch in case of NAND);
* o for static volumes in case of data CRC mismatch.
*
* If the volume is damaged because of an interrupted update this function just
* returns immediately with %-EBADF error code.
*/
int ubi_leb_read(struct ubi_volume_desc *desc, int lnum, char *buf, int offset,
int len, int check)
{
struct ubi_volume *vol = desc->vol;
struct ubi_device *ubi = vol->ubi;
int err, vol_id = vol->vol_id;
dbg_gen("read %d bytes from LEB %d:%d:%d", len, vol_id, lnum, offset);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 12:22:22 +04:00
if (vol_id < 0 || vol_id >= ubi->vtbl_slots || lnum < 0 ||
lnum >= vol->used_ebs || offset < 0 || len < 0 ||
offset + len > vol->usable_leb_size)
return -EINVAL;
if (vol->vol_type == UBI_STATIC_VOLUME) {
if (vol->used_ebs == 0)
/* Empty static UBI volume */
return 0;
if (lnum == vol->used_ebs - 1 &&
offset + len > vol->last_eb_bytes)
return -EINVAL;
}
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 12:22:22 +04:00
if (vol->upd_marker)
return -EBADF;
if (len == 0)
return 0;
err = ubi_eba_read_leb(ubi, vol, lnum, buf, offset, len, check);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 12:22:22 +04:00
if (err && err == -EBADMSG && vol->vol_type == UBI_STATIC_VOLUME) {
ubi_warn("mark volume %d as corrupted", vol_id);
vol->corrupted = 1;
}
return err;
}
EXPORT_SYMBOL_GPL(ubi_leb_read);
/**
* ubi_leb_write - write data.
* @desc: volume descriptor
* @lnum: logical eraseblock number to write to
* @buf: data to write
* @offset: offset within the logical eraseblock where to write
* @len: how many bytes to write
* @dtype: expected data type
*
* This function writes @len bytes of data from @buf to offset @offset of
* logical eraseblock @lnum. The @dtype argument describes expected lifetime of
* the data.
*
* This function takes care of physical eraseblock write failures. If write to
* the physical eraseblock write operation fails, the logical eraseblock is
* re-mapped to another physical eraseblock, the data is recovered, and the
* write finishes. UBI has a pool of reserved physical eraseblocks for this.
*
* If all the data were successfully written, zero is returned. If an error
* occurred and UBI has not been able to recover from it, this function returns
* a negative error code. Note, in case of an error, it is possible that
* something was still written to the flash media, but that may be some
* garbage.
*
* If the volume is damaged because of an interrupted update this function just
* returns immediately with %-EBADF code.
*/
int ubi_leb_write(struct ubi_volume_desc *desc, int lnum, const void *buf,
int offset, int len, int dtype)
{
struct ubi_volume *vol = desc->vol;
struct ubi_device *ubi = vol->ubi;
int vol_id = vol->vol_id;
dbg_gen("write %d bytes to LEB %d:%d:%d", len, vol_id, lnum, offset);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 12:22:22 +04:00
if (vol_id < 0 || vol_id >= ubi->vtbl_slots)
return -EINVAL;
if (desc->mode == UBI_READONLY || vol->vol_type == UBI_STATIC_VOLUME)
return -EROFS;
if (lnum < 0 || lnum >= vol->reserved_pebs || offset < 0 || len < 0 ||
offset + len > vol->usable_leb_size ||
offset & (ubi->min_io_size - 1) || len & (ubi->min_io_size - 1))
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 12:22:22 +04:00
return -EINVAL;
if (dtype != UBI_LONGTERM && dtype != UBI_SHORTTERM &&
dtype != UBI_UNKNOWN)
return -EINVAL;
if (vol->upd_marker)
return -EBADF;
if (len == 0)
return 0;
return ubi_eba_write_leb(ubi, vol, lnum, buf, offset, len, dtype);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 12:22:22 +04:00
}
EXPORT_SYMBOL_GPL(ubi_leb_write);
/*
* ubi_leb_change - change logical eraseblock atomically.
* @desc: volume descriptor
* @lnum: logical eraseblock number to change
* @buf: data to write
* @len: how many bytes to write
* @dtype: expected data type
*
* This function changes the contents of a logical eraseblock atomically. @buf
* has to contain new logical eraseblock data, and @len - the length of the
* data, which has to be aligned. The length may be shorter than the logical
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 12:22:22 +04:00
* eraseblock size, ant the logical eraseblock may be appended to more times
* later on. This function guarantees that in case of an unclean reboot the old
* contents is preserved. Returns zero in case of success and a negative error
* code in case of failure.
*/
int ubi_leb_change(struct ubi_volume_desc *desc, int lnum, const void *buf,
int len, int dtype)
{
struct ubi_volume *vol = desc->vol;
struct ubi_device *ubi = vol->ubi;
int vol_id = vol->vol_id;
dbg_gen("atomically write %d bytes to LEB %d:%d", len, vol_id, lnum);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 12:22:22 +04:00
if (vol_id < 0 || vol_id >= ubi->vtbl_slots)
return -EINVAL;
if (desc->mode == UBI_READONLY || vol->vol_type == UBI_STATIC_VOLUME)
return -EROFS;
if (lnum < 0 || lnum >= vol->reserved_pebs || len < 0 ||
len > vol->usable_leb_size || len & (ubi->min_io_size - 1))
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 12:22:22 +04:00
return -EINVAL;
if (dtype != UBI_LONGTERM && dtype != UBI_SHORTTERM &&
dtype != UBI_UNKNOWN)
return -EINVAL;
if (vol->upd_marker)
return -EBADF;
if (len == 0)
return 0;
return ubi_eba_atomic_leb_change(ubi, vol, lnum, buf, len, dtype);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 12:22:22 +04:00
}
EXPORT_SYMBOL_GPL(ubi_leb_change);
/**
* ubi_leb_erase - erase logical eraseblock.
* @desc: volume descriptor
* @lnum: logical eraseblock number
*
* This function un-maps logical eraseblock @lnum and synchronously erases the
* correspondent physical eraseblock. Returns zero in case of success and a
* negative error code in case of failure.
*
* If the volume is damaged because of an interrupted update this function just
* returns immediately with %-EBADF code.
*/
int ubi_leb_erase(struct ubi_volume_desc *desc, int lnum)
{
struct ubi_volume *vol = desc->vol;
struct ubi_device *ubi = vol->ubi;
int err;
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 12:22:22 +04:00
dbg_gen("erase LEB %d:%d", vol->vol_id, lnum);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 12:22:22 +04:00
if (desc->mode == UBI_READONLY || vol->vol_type == UBI_STATIC_VOLUME)
return -EROFS;
if (lnum < 0 || lnum >= vol->reserved_pebs)
return -EINVAL;
if (vol->upd_marker)
return -EBADF;
err = ubi_eba_unmap_leb(ubi, vol, lnum);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 12:22:22 +04:00
if (err)
return err;
return ubi_wl_flush(ubi);
}
EXPORT_SYMBOL_GPL(ubi_leb_erase);
/**
* ubi_leb_unmap - un-map logical eraseblock.
* @desc: volume descriptor
* @lnum: logical eraseblock number
*
* This function un-maps logical eraseblock @lnum and schedules the
* corresponding physical eraseblock for erasure, so that it will eventually be
* physically erased in background. This operation is much faster than the
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 12:22:22 +04:00
* erase operation.
*
* Unlike erase, the un-map operation does not guarantee that the logical
* eraseblock will contain all 0xFF bytes when UBI is initialized again. For
* example, if several logical eraseblocks are un-mapped, and an unclean reboot
* happens after this, the logical eraseblocks will not necessarily be
* un-mapped again when this MTD device is attached. They may actually be
* mapped to the same physical eraseblocks again. So, this function has to be
* used with care.
*
* In other words, when un-mapping a logical eraseblock, UBI does not store
* any information about this on the flash media, it just marks the logical
* eraseblock as "un-mapped" in RAM. If UBI is detached before the physical
* eraseblock is physically erased, it will be mapped again to the same logical
* eraseblock when the MTD device is attached again.
*
* The main and obvious use-case of this function is when the contents of a
* logical eraseblock has to be re-written. Then it is much more efficient to
* first un-map it, then write new data, rather than first erase it, then write
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 12:22:22 +04:00
* new data. Note, once new data has been written to the logical eraseblock,
* UBI guarantees that the old contents has gone forever. In other words, if an
* unclean reboot happens after the logical eraseblock has been un-mapped and
* then written to, it will contain the last written data.
*
* This function returns zero in case of success and a negative error code in
* case of failure. If the volume is damaged because of an interrupted update
* this function just returns immediately with %-EBADF code.
*/
int ubi_leb_unmap(struct ubi_volume_desc *desc, int lnum)
{
struct ubi_volume *vol = desc->vol;
struct ubi_device *ubi = vol->ubi;
dbg_gen("unmap LEB %d:%d", vol->vol_id, lnum);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 12:22:22 +04:00
if (desc->mode == UBI_READONLY || vol->vol_type == UBI_STATIC_VOLUME)
return -EROFS;
if (lnum < 0 || lnum >= vol->reserved_pebs)
return -EINVAL;
if (vol->upd_marker)
return -EBADF;
return ubi_eba_unmap_leb(ubi, vol, lnum);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 12:22:22 +04:00
}
EXPORT_SYMBOL_GPL(ubi_leb_unmap);
/**
* ubi_leb_map - map logical eraseblock to a physical eraseblock.
* @desc: volume descriptor
* @lnum: logical eraseblock number
* @dtype: expected data type
*
* This function maps an un-mapped logical eraseblock @lnum to a physical
* eraseblock. This means, that after a successful invocation of this
* function the logical eraseblock @lnum will be empty (contain only %0xFF
* bytes) and be mapped to a physical eraseblock, even if an unclean reboot
* happens.
*
* This function returns zero in case of success, %-EBADF if the volume is
* damaged because of an interrupted update, %-EBADMSG if the logical
* eraseblock is already mapped, and other negative error codes in case of
* other failures.
*/
int ubi_leb_map(struct ubi_volume_desc *desc, int lnum, int dtype)
{
struct ubi_volume *vol = desc->vol;
struct ubi_device *ubi = vol->ubi;
dbg_gen("unmap LEB %d:%d", vol->vol_id, lnum);
if (desc->mode == UBI_READONLY || vol->vol_type == UBI_STATIC_VOLUME)
return -EROFS;
if (lnum < 0 || lnum >= vol->reserved_pebs)
return -EINVAL;
if (dtype != UBI_LONGTERM && dtype != UBI_SHORTTERM &&
dtype != UBI_UNKNOWN)
return -EINVAL;
if (vol->upd_marker)
return -EBADF;
if (vol->eba_tbl[lnum] >= 0)
return -EBADMSG;
return ubi_eba_write_leb(ubi, vol, lnum, NULL, 0, 0, dtype);
}
EXPORT_SYMBOL_GPL(ubi_leb_map);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 12:22:22 +04:00
/**
* ubi_is_mapped - check if logical eraseblock is mapped.
* @desc: volume descriptor
* @lnum: logical eraseblock number
*
* This function checks if logical eraseblock @lnum is mapped to a physical
* eraseblock. If a logical eraseblock is un-mapped, this does not necessarily
* mean it will still be un-mapped after the UBI device is re-attached. The
* logical eraseblock may become mapped to the physical eraseblock it was last
* mapped to.
*
* This function returns %1 if the LEB is mapped, %0 if not, and a negative
* error code in case of failure. If the volume is damaged because of an
* interrupted update this function just returns immediately with %-EBADF error
* code.
*/
int ubi_is_mapped(struct ubi_volume_desc *desc, int lnum)
{
struct ubi_volume *vol = desc->vol;
dbg_gen("test LEB %d:%d", vol->vol_id, lnum);
UBI: Unsorted Block Images UBI (Latin: "where?") manages multiple logical volumes on a single flash device, specifically supporting NAND flash devices. UBI provides a flexible partitioning concept which still allows for wear-levelling across the whole flash device. In a sense, UBI may be compared to the Logical Volume Manager (LVM). Whereas LVM maps logical sector numbers to physical HDD sector numbers, UBI maps logical eraseblocks to physical eraseblocks. More information may be found at http://www.linux-mtd.infradead.org/doc/ubi.html Partitioning/Re-partitioning An UBI volume occupies a certain number of erase blocks. This is limited by a configured maximum volume size, which could also be viewed as the partition size. Each individual UBI volume's size can be changed independently of the other UBI volumes, provided that the sum of all volume sizes doesn't exceed a certain limit. UBI supports dynamic volumes and static volumes. Static volumes are read-only and their contents are protected by CRC check sums. Bad eraseblocks handling UBI transparently handles bad eraseblocks. When a physical eraseblock becomes bad, it is substituted by a good physical eraseblock, and the user does not even notice this. Scrubbing On a NAND flash bit flips can occur on any write operation, sometimes also on read. If bit flips persist on the device, at first they can still be corrected by ECC, but once they accumulate, correction will become impossible. Thus it is best to actively scrub the affected eraseblock, by first copying it to a free eraseblock and then erasing the original. The UBI layer performs this type of scrubbing under the covers, transparently to the UBI volume users. Erase Counts UBI maintains an erase count header per eraseblock. This frees higher-level layers (like file systems) from doing this and allows for centralized erase count management instead. The erase counts are used by the wear-levelling algorithm in the UBI layer. The algorithm itself is exchangeable. Booting from NAND For booting directly from NAND flash the hardware must at least be capable of fetching and executing a small portion of the NAND flash. Some NAND flash controllers have this kind of support. They usually limit the window to a few kilobytes in erase block 0. This "initial program loader" (IPL) must then contain sufficient logic to load and execute the next boot phase. Due to bad eraseblocks, which may be randomly scattered over the flash device, it is problematic to store the "secondary program loader" (SPL) statically. Also, due to bit-flips it may become corrupted over time. UBI allows to solve this problem gracefully by storing the SPL in a small static UBI volume. UBI volumes vs. static partitions UBI volumes are still very similar to static MTD partitions: * both consist of eraseblocks (logical eraseblocks in case of UBI volumes, and physical eraseblocks in case of static partitions; * both support three basic operations - read, write, erase. But UBI volumes have the following advantages over traditional static MTD partitions: * there are no eraseblock wear-leveling constraints in case of UBI volumes, so the user should not care about this; * there are no bit-flips and bad eraseblocks in case of UBI volumes. So, UBI volumes may be considered as flash devices with relaxed restrictions. Where can it be found? Documentation, kernel code and applications can be found in the MTD gits. What are the applications for? The applications help to create binary flash images for two purposes: pfi files (partial flash images) for in-system update of UBI volumes, and plain binary images, with or without OOB data in case of NAND, for a manufacturing step. Furthermore some tools are/and will be created that allow flash content analysis after a system has crashed.. Who did UBI? The original ideas, where UBI is based on, were developed by Andreas Arnez, Frank Haverkamp and Thomas Gleixner. Josh W. Boyer and some others were involved too. The implementation of the kernel layer was done by Artem B. Bityutskiy. The user-space applications and tools were written by Oliver Lohmann with contributions from Frank Haverkamp, Andreas Arnez, and Artem. Joern Engel contributed a patch which modifies JFFS2 so that it can be run on a UBI volume. Thomas Gleixner did modifications to the NAND layer. Alexander Schmidt made some testing work as well as core functionality improvements. Signed-off-by: Artem B. Bityutskiy <dedekind@linutronix.de> Signed-off-by: Frank Haverkamp <haver@vnet.ibm.com>
2006-06-27 12:22:22 +04:00
if (lnum < 0 || lnum >= vol->reserved_pebs)
return -EINVAL;
if (vol->upd_marker)
return -EBADF;
return vol->eba_tbl[lnum] >= 0;
}
EXPORT_SYMBOL_GPL(ubi_is_mapped);
/**
* ubi_sync - synchronize UBI device buffers.
* @ubi_num: UBI device to synchronize
*
* The underlying MTD device may cache data in hardware or in software. This
* function ensures the caches are flushed. Returns zero in case of success and
* a negative error code in case of failure.
*/
int ubi_sync(int ubi_num)
{
struct ubi_device *ubi;
ubi = ubi_get_device(ubi_num);
if (!ubi)
return -ENODEV;
if (ubi->mtd->sync)
ubi->mtd->sync(ubi->mtd);
ubi_put_device(ubi);
return 0;
}
EXPORT_SYMBOL_GPL(ubi_sync);
BLOCKING_NOTIFIER_HEAD(ubi_notifiers);
/**
* ubi_register_volume_notifier - register a volume notifier.
* @nb: the notifier description object
* @ignore_existing: if non-zero, do not send "added" notification for all
* already existing volumes
*
* This function registers a volume notifier, which means that
* 'nb->notifier_call()' will be invoked when an UBI volume is created,
* removed, re-sized, re-named, or updated. The first argument of the function
* is the notification type. The second argument is pointer to a
* &struct ubi_notification object which describes the notification event.
* Using UBI API from the volume notifier is prohibited.
*
* This function returns zero in case of success and a negative error code
* in case of failure.
*/
int ubi_register_volume_notifier(struct notifier_block *nb,
int ignore_existing)
{
int err;
err = blocking_notifier_chain_register(&ubi_notifiers, nb);
if (err != 0)
return err;
if (ignore_existing)
return 0;
/*
* We are going to walk all UBI devices and all volumes, and
* notify the user about existing volumes by the %UBI_VOLUME_ADDED
* event. We have to lock the @ubi_devices_mutex to make sure UBI
* devices do not disappear.
*/
mutex_lock(&ubi_devices_mutex);
ubi_enumerate_volumes(nb);
mutex_unlock(&ubi_devices_mutex);
return err;
}
EXPORT_SYMBOL_GPL(ubi_register_volume_notifier);
/**
* ubi_unregister_volume_notifier - unregister the volume notifier.
* @nb: the notifier description object
*
* This function unregisters volume notifier @nm and returns zero in case of
* success and a negative error code in case of failure.
*/
int ubi_unregister_volume_notifier(struct notifier_block *nb)
{
return blocking_notifier_chain_unregister(&ubi_notifiers, nb);
}
EXPORT_SYMBOL_GPL(ubi_unregister_volume_notifier);