Patch series "1G transparent hugepage support for device dax", v2.
The following series implements support for 1G trasparent hugepage on
x86 for device dax. The bulk of the code was written by Mathew Wilcox a
while back supporting transparent 1G hugepage for fs DAX. I have
forward ported the relevant bits to 4.10-rc. The current submission has
only the necessary code to support device DAX.
Comments from Dan Williams: So the motivation and intended user of this
functionality mirrors the motivation and users of 1GB page support in
hugetlbfs. Given expected capacities of persistent memory devices an
in-memory database may want to reduce tlb pressure beyond what they can
already achieve with 2MB mappings of a device-dax file. We have
customer feedback to that effect as Willy mentioned in his previous
version of these patches [1].
[1]: https://lkml.org/lkml/2016/1/31/52
Comments from Nilesh @ Oracle:
There are applications which have a process model; and if you assume
10,000 processes attempting to mmap all the 6TB memory available on a
server; we are looking at the following:
processes : 10,000
memory : 6TB
pte @ 4k page size: 8 bytes / 4K of memory * #processes = 6TB / 4k * 8 * 10000 = 1.5GB * 80000 = 120,000GB
pmd @ 2M page size: 120,000 / 512 = ~240GB
pud @ 1G page size: 240GB / 512 = ~480MB
As you can see with 2M pages, this system will use up an exorbitant
amount of DRAM to hold the page tables; but the 1G pages finally brings
it down to a reasonable level. Memory sizes will keep increasing; so
this number will keep increasing.
An argument can be made to convert the applications from process model
to thread model, but in the real world that may not be always practical.
Hopefully this helps explain the use case where this is valuable.
This patch (of 3):
In preparation for adding the ability to handle PUD pages, convert
vm_operations_struct.pmd_fault to vm_operations_struct.huge_fault. The
vm_fault structure is extended to include a union of the different page
table pointers that may be needed, and three flag bits are reserved to
indicate which type of pointer is in the union.
[ross.zwisler@linux.intel.com: remove unused function ext4_dax_huge_fault()]
Link: http://lkml.kernel.org/r/1485813172-7284-1-git-send-email-ross.zwisler@linux.intel.com
[dave.jiang@intel.com: clear PMD or PUD size flags when in fall through path]
Link: http://lkml.kernel.org/r/148589842696.5820.16078080610311444794.stgit@djiang5-desk3.ch.intel.com
Link: http://lkml.kernel.org/r/148545058784.17912.6353162518188733642.stgit@djiang5-desk3.ch.intel.com
Signed-off-by: Matthew Wilcox <mawilcox@microsoft.com>
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Jan Kara <jack@suse.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Nilesh Choudhury <nilesh.choudhury@oracle.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Dave Jiang <dave.jiang@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
->fault(), ->page_mkwrite(), and ->pfn_mkwrite() calls do not need to
take a vma and vmf parameter when the vma already resides in vmf.
Remove the vma parameter to simplify things.
[arnd@arndb.de: fix ARM build]
Link: http://lkml.kernel.org/r/20170125223558.1451224-1-arnd@arndb.de
Link: http://lkml.kernel.org/r/148521301778.19116.10840599906674778980.stgit@djiang5-desk3.ch.intel.com
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Reviewed-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Cc: Theodore Ts'o <tytso@mit.edu>
Cc: Darrick J. Wong <darrick.wong@oracle.com>
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Jan Kara <jack@suse.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
pmd_fault() and related functions really only need the vmf parameter since
the additional parameters are all included in the vmf struct. Remove the
additional parameter and simplify pmd_fault() and friends.
Link: http://lkml.kernel.org/r/1484085142-2297-8-git-send-email-ross.zwisler@linux.intel.com
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Reviewed-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Dave Jiang <dave.jiang@intel.com>
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Instead of passing in multiple parameters in the pmd_fault() handler,
a vmf can be passed in just like a fault() handler. This will simplify
code and remove the need for the actual pmd fault handlers to allocate a
vmf. Related functions are also modified to do the same.
[dave.jiang@intel.com: fix issue with xfs_tests stall when DAX option is off]
Link: http://lkml.kernel.org/r/148469861071.195597.3619476895250028518.stgit@djiang5-desk3.ch.intel.com
Link: http://lkml.kernel.org/r/1484085142-2297-7-git-send-email-ross.zwisler@linux.intel.com
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Reviewed-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Matthew Wilcox <mawilcox@microsoft.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
* Dynamic label support: To date namespace label support has been
limited to disambiguating cases where PMEM (direct load/store) and BLK
(mmio aperture) accessed-capacity alias on the same DIMM. Since 4.9 added
support for multiple namespaces per PMEM-region there is value to
support namespace labels even in the non-aliasing case. The presence of
a valid namespace index block force-enables label support when the
kernel would otherwise rely on region boundaries, and permits the region
to be sub-divided.
* Handle media errors in namespace metadata: Complement the error
handling for media errors in namespace data areas with support for
clearing errors on writes, and downgrading potential machine-check
exceptions to simple i/o errors on read.
* Device-DAX region attributes: Add 'align', 'id', and 'size' as
attributes for device-dax regions. In particular this enables userspace
tooling to generically size memory mapping and i/o operations. Prevent
userspace from growing assumptions / dependencies about the parent
device topology for a dax region. A libnvdimm namespace may not always
be the parent device of a dax region.
* Various cleanups and small fixes.
-----BEGIN PGP SIGNATURE-----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=YWfP
-----END PGP SIGNATURE-----
Merge tag 'libnvdimm-for-4.10' of git://git.kernel.org/pub/scm/linux/kernel/git/nvdimm/nvdimm
Pull libnvdimm updates from Dan Williams:
"The libnvdimm pull request is relatively small this time around due to
some development topics being deferred to 4.11.
As for this pull request the bulk of it has been in -next for several
releases leading to one late fix being added (commit 868f036fee
("libnvdimm: fix mishandled nvdimm_clear_poison() return value")). It
has received a build success notification from the 0day-kbuild robot
and passes the latest libnvdimm unit tests.
Summary:
- Dynamic label support: To date namespace label support has been
limited to disambiguating cases where PMEM (direct load/store) and
BLK (mmio aperture) accessed-capacity alias on the same DIMM. Since
4.9 added support for multiple namespaces per PMEM-region there is
value to support namespace labels even in the non-aliasing case.
The presence of a valid namespace index block force-enables label
support when the kernel would otherwise rely on region boundaries,
and permits the region to be sub-divided.
- Handle media errors in namespace metadata: Complement the error
handling for media errors in namespace data areas with support for
clearing errors on writes, and downgrading potential machine-check
exceptions to simple i/o errors on read.
- Device-DAX region attributes: Add 'align', 'id', and 'size' as
attributes for device-dax regions. In particular this enables
userspace tooling to generically size memory mapping and i/o
operations. Prevent userspace from growing assumptions /
dependencies about the parent device topology for a dax region. A
libnvdimm namespace may not always be the parent device of a dax
region.
- Various cleanups and small fixes"
* tag 'libnvdimm-for-4.10' of git://git.kernel.org/pub/scm/linux/kernel/git/nvdimm/nvdimm:
dax: add region 'id', 'size', and 'align' attributes
libnvdimm: fix mishandled nvdimm_clear_poison() return value
libnvdimm: replace mutex_is_locked() warnings with lockdep_assert_held
libnvdimm, pfn: fix align attribute
libnvdimm, e820: use module_platform_driver
libnvdimm, namespace: use octal for permissions
libnvdimm, namespace: avoid multiple sector calculations
libnvdimm: remove else after return in nsio_rw_bytes()
libnvdimm, namespace: fix the type of name variable
libnvdimm: use consistent naming for request_mem_region()
nvdimm: use the right length of "pmem"
libnvdimm: check and clear poison before writing to pmem
tools/testing/nvdimm: dynamic label support
libnvdimm: allow a platform to force enable label support
libnvdimm: use generic iostat interfaces
While this information is available by looking at the nvdimm parent
device that may not always be the case when/if we add support for other
memory regions. Tooling should not depend on walking a given ancestor
topology that is not guaranteed by the device's class. For example, a
device-dax instance will always have a dax_region parent, but it may not
always have a libnvdimm "dax" device as a grandparent.
Reported-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Every single user of vmf->virtual_address typed that entry to unsigned
long before doing anything with it so the type of virtual_address does
not really provide us any additional safety. Just use masked
vmf->address which already has the appropriate type.
Link: http://lkml.kernel.org/r/1479460644-25076-3-git-send-email-jack@suse.cz
Signed-off-by: Jan Kara <jack@suse.cz>
Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Hugh notes in response to commit 4cb19355ea "device-dax: fail all
private mapping attempts":
"I think that is more restrictive than you intended: haven't tried, but I
believe it rejects a PROT_READ, MAP_SHARED, O_RDONLY fd mmap, leaving no
way to mmap /dev/dax without write permission to it."
Indeed it does restrict read-only mappings, switch to checking
VM_MAYSHARE, not VM_SHARED.
Cc: <stable@vger.kernel.org>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Pawel Lebioda <pawel.lebioda@intel.com>
Fixes: 4cb19355ea ("device-dax: fail all private mapping attempts")
Reported-by: Hugh Dickins <hughd@google.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Here is an example /proc/iomem listing for a system with 2 namespaces,
one in "sector" mode and one in "memory" mode:
1fc000000-2fbffffff : Persistent Memory (legacy)
1fc000000-2fbffffff : namespace1.0
340000000-34fffffff : Persistent Memory
340000000-34fffffff : btt0.1
Here is the corresponding ndctl listing:
# ndctl list
[
{
"dev":"namespace1.0",
"mode":"memory",
"size":4294967296,
"blockdev":"pmem1"
},
{
"dev":"namespace0.0",
"mode":"sector",
"size":267091968,
"uuid":"f7594f86-badb-4592-875f-ded577da2eaf",
"sector_size":4096,
"blockdev":"pmem0s"
}
]
Notice that the ndctl listing is purely in terms of namespace devices,
while the iomem listing leaks the internal "btt0.1" implementation
detail. Given that ndctl requires the namespace device name to change
the mode, for example:
# ndctl create-namespace --reconfig=namespace0.0 --mode=raw --force
...use the namespace name in the iomem listing to keep the claiming
device name consistent across different mode settings.
Cc: Vishal Verma <vishal.l.verma@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
The device-dax implementation originally tried to be tricky and allow
private read-only mappings, but in the process allowed writable
MAP_PRIVATE + MAP_NORESERVE mappings. For simplicity and predictability
just fail all private mapping attempts since device-dax memory is
statically allocated and will never support overcommit.
Cc: <stable@vger.kernel.org>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Fixes: dee4107924 ("/dev/dax, core: file operations and dax-mmap")
Reported-by: Pawel Lebioda <pawel.lebioda@intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
If the dax_pmem driver is passed a resource that is already busy the
driver probe attempt should fail with a message like the following:
dax_pmem dax0.1: could not reserve region [mem 0x100000000-0x11fffffff]
However, if we do not catch the error we crash for the obvious reason of
accessing memory that is not mapped.
BUG: unable to handle kernel paging request at ffffc90020001000
IP: [<ffffffff81496712>] __memcpy+0x12/0x20
[..]
Call Trace:
[<ffffffff815c4960>] ? nsio_rw_bytes+0x60/0x180
[<ffffffff815c6045>] nd_pfn_validate+0x75/0x320
[<ffffffff815c63a9>] nvdimm_setup_pfn+0xb9/0x5d0
[<ffffffff815c48ef>] ? devm_nsio_enable+0xff/0x110
[<ffffffff815cb699>] dax_pmem_probe+0x59/0x260
Cc: <stable@vger.kernel.org>
Fixes: ab68f26221 ("/dev/dax, pmem: direct access to persistent memory")
Reported-by: Dave Hansen <dave.hansen@linux.intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
We need to wait until the percpu_ref is released before exit. Otherwise,
we sometimes lose the race and trigger this new warning that was added
in v4.9 (commit a67823c1ed "percpu-refcount: init ->confirm_switch
member properly"):
WARNING: CPU: 0 PID: 3629 at lib/percpu-refcount.c:107 percpu_ref_exit+0x51/0x60
[..]
Call Trace:
[<ffffffff814bf093>] dump_stack+0x85/0xc2
[<ffffffff810b15db>] __warn+0xcb/0xf0
[<ffffffff810b170d>] warn_slowpath_null+0x1d/0x20
[<ffffffff814d70c1>] percpu_ref_exit+0x51/0x60
[<ffffffffa005706a>] dax_pmem_percpu_exit+0x1a/0x50 [dax_pmem]
[<ffffffff81615f1f>] devm_action_release+0xf/0x20
Cc: <stable@vger.kernel.org>
Fixes: ab68f26221 ("/dev/dax, pmem: direct access to persistent memory")
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
A bugfix just tried to address a randconfig build problem and introduced
a variant of the same problem: with CONFIG_LIBNVDIMM=y and
CONFIG_NVDIMM_DAX=m, the nvdimm module now fails to link:
drivers/nvdimm/built-in.o: In function `to_nd_device_type':
bus.c:(.text+0x1b5d): undefined reference to `is_nd_dax'
drivers/nvdimm/built-in.o: In function `nd_region_notify_driver_action.constprop.2':
region_devs.c:(.text+0x6b6c): undefined reference to `is_nd_dax'
region_devs.c:(.text+0x6b8c): undefined reference to `to_nd_dax'
drivers/nvdimm/built-in.o: In function `nd_region_probe':
region.c:(.text+0x70f3): undefined reference to `nd_dax_create'
drivers/nvdimm/built-in.o: In function `mode_show':
namespace_devs.c:(.text+0xa196): undefined reference to `is_nd_dax'
drivers/nvdimm/built-in.o: In function `nvdimm_namespace_common_probe':
(.text+0xa55f): undefined reference to `is_nd_dax'
drivers/nvdimm/built-in.o: In function `nvdimm_namespace_common_probe':
(.text+0xa56e): undefined reference to `to_nd_dax'
This reverts the earlier fix, making NVDIMM_DAX a 'bool' option again
as it should be (it gets linked into the libnvdimm module). To fix
the original problem, I'm adding a dependency on LIBNVDIMM to
DEV_DAX_PMEM, which ensures we can't have that one built-in if the
rest is a module.
Fixes: 4e65e9381c ("/dev/dax: fix Kconfig dependency build breakage")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Reviewed-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
The dev_t variable in devm_create_dax_dev() is used before it's
first set:
drivers/dax/dax.c: In function 'devm_create_dax_dev':
drivers/dax/dax.c:205:39: error: 'dev_t' may be used uninitialized in this function [-Werror=maybe-uninitialized]
inode = iget5_locked(dax_superblock, hash_32(devt + DAXFS_MAGIC, 31),
^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
drivers/dax/dax.c:688:8: note: 'dev_t' was declared here
This reorders the code to how it looks correct to me.
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Fixes: 3bc52c45ba ("dax: define a unified inode/address_space for device-dax mappings")
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
pgoff_to_phys() validates that both the starting address and the length
of the mapping against the resource list. We need to check for a
mapping size of PMD_SIZE not PAGE_SIZE in the pmd fault path.
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
The data offset for a dax region needs to account for a reservation in
the resource range. Otherwise, device-dax is allowing mappings directly
into the memmap or device-info-block area with crash signatures like the
following:
BUG: unable to handle kernel NULL pointer dereference at 0000000000000008
IP: get_zone_device_page+0x11/0x30
Call Trace:
follow_devmap_pmd+0x298/0x2c0
follow_page_mask+0x275/0x530
__get_user_pages+0xe3/0x750
__gfn_to_pfn_memslot+0x1b2/0x450 [kvm]
tdp_page_fault+0x130/0x280 [kvm]
kvm_mmu_page_fault+0x5f/0xf0 [kvm]
handle_ept_violation+0x94/0x180 [kvm_intel]
vmx_handle_exit+0x1d3/0x1440 [kvm_intel]
kvm_arch_vcpu_ioctl_run+0x81d/0x16a0 [kvm]
kvm_vcpu_ioctl+0x33c/0x620 [kvm]
do_vfs_ioctl+0xa2/0x5d0
SyS_ioctl+0x79/0x90
entry_SYSCALL_64_fastpath+0x1a/0xa4
Fixes: ab68f26221 ("/dev/dax, pmem: direct access to persistent memory")
Link: http://lkml.kernel.org/r/147205536732.1606.8994275381938837346.stgit@dwillia2-desk3.amr.corp.intel.com
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Reported-by: Abhilash Kumar Mulumudi <m.abhilash-kumar@hpe.com>
Reported-by: Toshi Kani <toshi.kani@hpe.com>
Tested-by: Toshi Kani <toshi.kani@hpe.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
All the extents of a dax-device must match the alignment of the region.
Otherwise, we are unable to guarantee fault semantics of a given page
size. The region must be self-consistent itself as well.
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
In support of enabling resize / truncate of device-dax instances, define
a pseudo-fs to provide a unified inode/address space for vm operations.
Cc: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
A goal of the device-DAX interface is to be able to support many
exclusive allocations (partitions) of performance / feature
differentiated memory. This count may exceed the default minors limit
of 256.
As a result of switching to an embedded cdev the inode-to-dax_dev
conversion is simplified, as well as reference counting which can switch
to the cdev kobject lifetime.
Cc: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
The kref in dax_dev can be made redundant if the final put_device() on
the device associated with the dax_dev frees the dax_dev. This can be
accomplished by embedding a struct device in struct dax_dev, open coding
device_create() and specifying a custom release method.
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Shorten the prefix of the file operations to distinguish them from
operations on the struct device associated with the dax_dev.
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
In order to convert devm_create_dax_dev() to use cdev, it will need
access to dax_fops. Move dax_fops and related function definitions
before devm_create_dax_dev().
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
drivers/dax/dax.c:75:6: warning: symbol 'dax_region_put' was not declared.
drivers/dax/dax.c:95:19: warning: symbol 'alloc_dax_region' was not declared.
drivers/dax/dax.c:173:5: warning: symbol 'devm_create_dax_dev' was not declared.
drivers/dax/pmem.c:27:17: warning: symbol 'to_dax_pmem' was not declared.
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
If devm_add_action() fails, we are explicitly calling the cleanup to free
the resources allocated. Use the helper devm_add_action_or_reset()
and return directly in case of error, since the cleanup function
has been already called by the helper if there was any error.
Reported-by: Sudip Mukherjee <sudipm.mukherjee@gmail.com>
Signed-off-by: Vikas C Sajjan <vikas.cha.sajjan@hpe.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
The "Device DAX" core enables dax mappings of performance / feature
differentiated memory. An open mapping or file handle keeps the backing
struct device live, but new mappings are only possible while the device
is enabled. Faults are handled under rcu_read_lock to synchronize
with the enabled state of the device.
Similar to the filesystem-dax case the backing memory may optionally
have struct page entries. However, unlike fs-dax there is no support
for private mappings, or mappings that are not backed by media (see
use of zero-page in fs-dax).
Mappings are always guaranteed to match the alignment of the dax_region.
If the dax_region is configured to have a 2MB alignment, all mappings
are guaranteed to be backed by a pmd entry. Contrast this determinism
with the fs-dax case where pmd mappings are opportunistic. If userspace
attempts to force a misaligned mapping, the driver will fail the mmap
attempt. See dax_dev_check_vma() for other scenarios that are rejected,
like MAP_PRIVATE mappings.
Cc: Hannes Reinecke <hare@suse.de>
Cc: Jeff Moyer <jmoyer@redhat.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Acked-by: "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Device DAX is the device-centric analogue of Filesystem DAX
(CONFIG_FS_DAX). It allows memory ranges to be allocated and mapped
without need of an intervening file system. Device DAX is strict,
precise and predictable. Specifically this interface:
1/ Guarantees fault granularity with respect to a given page size (pte,
pmd, or pud) set at configuration time.
2/ Enforces deterministic behavior by being strict about what fault
scenarios are supported.
For example, by forcing MADV_DONTFORK semantics and omitting MAP_PRIVATE
support device-dax guarantees that a mapping always behaves/performs the
same once established. It is the "what you see is what you get" access
mechanism to differentiated memory vs filesystem DAX which has
filesystem specific implementation semantics.
Persistent memory is the first target, but the mechanism is also
targeted for exclusive allocations of performance differentiated memory
ranges.
This commit is limited to the base device driver infrastructure to
associate a dax device with pmem range.
Cc: Jeff Moyer <jmoyer@redhat.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Dave Jiang
Linus Torvalds
Dan Williams
Jan Kara
Arnd Bergmann
Sajjan, Vikas C