s390 allows hotpatching the mask of a conditional jump instruction.
Make use of this feature in order to avoid the expensive stop_machine()
call.
The new trampolines are split in 3 stages:
- A first stage is a 6-byte relative conditional long branch located at
each function's entry point. Its offset always points to the second
stage for the corresponding function, and its mask is either all 0s
(ftrace off) or all 1s (ftrace on). The code for flipping the mask is
borrowed from ftrace_{enable,disable}_ftrace_graph_caller. After
flipping, ftrace_arch_code_modify_post_process() syncs with all the
other CPUs by sending SIGPs.
- Second stages for vmlinux are stored in a separate part of the .text
section reserved by the linker script, and in dynamically allocated
memory for modules. This prevents the icache pollution. The total
size of second stages is about 1.5% of that of the kernel image.
Putting second stages in the .bss section is possible and decreases
the size of the non-compressed vmlinux, but splits the kernel 1:1
mapping, which is a bad tradeoff.
Each second stage contains a call to the third stage, a pointer to
the part of the intercepted function right after the first stage, and
a pointer to an interceptor function (e.g. ftrace_caller).
Second stages are 8-byte aligned for the future direct calls
implementation.
- There are only two copies of the third stage: in the .text section
for vmlinux and in dynamically allocated memory for modules. It can be
an expoline, which is relatively large, so inlining it into each
second stage is prohibitively expensive.
As a result of this organization, phoronix-test-suite with ftrace off
does not show any performance degradation.
Suggested-by: Sven Schnelle <svens@linux.ibm.com>
Suggested-by: Vasily Gorbik <gor@linux.ibm.com>
Co-developed-by: Heiko Carstens <hca@linux.ibm.com>
Signed-off-by: Ilya Leoshkevich <iii@linux.ibm.com>
Link: https://lore.kernel.org/r/20210728212546.128248-3-iii@linux.ibm.com
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
Currently, the install target in arch/s390/Makefile descends into
arch/s390/boot/Makefile to invoke the shell script, but there is no
good reason to do so.
arch/s390/Makefile can run the shell script directly.
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
Link: https://lore.kernel.org/r/20210729142338.446002-1-masahiroy@kernel.org
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
After all the changes to delay.c there are many includes which are not
needed anymore. Get rid of them.
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
Commit 7f16d7e787 ("s390: show virtualization support in /proc/cpuinfo")
introduced special handling for sie capability, saying this should not be
exposed via hwcaps, without giving a reason.
However this leads to an inconsistent /proc/cpuinfo features line
where all features except the sie capability are also present in
hwcaps. I really don't see a reason to not add that to hwcaps - it
might be quite pointless, but at least this way it is possible to get
rid of some special handling.
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
Remove the not so obvious "(elf_hwcap & (1UL << 2)" which only checks
if stfle is available. This used to be required for old code before
test_facility() was introduced. test_facility() will do the right
thing, regardless if stfle is available or not.
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
Remove a leftover from the common 31/64 bit code. z/Architecture mode
is now always active, there is no need to check.
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
The first six hwcap bits are initialized in a rather odd way: an array
contains the stfl(e) bits which need to be set, so that the
corresponding bit position (= array index) within hwcaps are set.
Better open code it like it is done for all other bits, making it
obvious which bit is set when.
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
setup_hwcaps() is a quite large function. Make it smaller by moving
the elf platform setup code into an independent setup function.
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
Move setup_hwcaps() to processor.c for two reasons:
- make setup.c a bit smaller
- have allmost all of the hwcap code in one file
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
Add BUILD_BUG_ON() sanity checks to make sure the hwcap string array
contains a string for each hwcap.
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
Introduce HWCAP bit numbers, making it easier to tell at which bit
number we currently are. Also use these bits with the BIT macro to
define the real HWCAP masks.
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
Remove s390 part of all HWCAP defines, just to make them shorter and
easier to handle. The namespace is anyway per architecture.
This is similar to what arm64 has.
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
In order to support the use of enhanced PCI instructions in both kernel-
and userspace we need both hardware support and proper setup in the
kernel. The latter can be toggled off with the pci=nomio command line
option.
Thus availability of this feature in userspace depends on all of kernel
configuration (CONFIG_PCI), hardware support and the current kernel
command line and can thus not rely solely on a facility bit. Instead
let's introduce a new ELF hardware capability bit HWCAP_S390_PCI_MIO to
tell userspace whether these PCI instructions can be used.
Reviewed-by: Heiko Carstens <hca@linux.ibm.com>
Signed-off-by: Niklas Schnelle <schnelle@linux.ibm.com>
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
Kernel support for the newer PCI mio instructions can be toggled off
with the pci=nomio command line option which needs to integrate with
common code PCI option parsing. However this option then toggles static
branches which can't be toggled yet in an early_param() call.
Thus commit 9964f396f1 ("s390: fix setting of mio addressing control")
moved toggling the static branches to the PCI init routine.
With this setup however we can't check for mio support outside the PCI
code during early boot, i.e. before switching the static branches, which
we need to be able to export this as an ELF HWCAP.
Improve on this by turning mio availability into a machine flag that
gets initially set based on CONFIG_PCI and the facility bit and gets
toggled off if pci=nomio is found during PCI option parsing allowing
simple access to this machine flag after early init.
Reviewed-by: Heiko Carstens <hca@linux.ibm.com>
Signed-off-by: Niklas Schnelle <schnelle@linux.ibm.com>
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
Add hardware capability bits and feature tags to /proc/cpuinfo
for NNPA and Vector-Packed-Decimal-Enhancement Facility 2.
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
These macros haven't seen any use in a long time. Also note that the
queue_irqs_*() ones wouldn't even compile anymore.
Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
Reviewed-by: Benjamin Block <bblock@linux.ibm.com>
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
Now that all drivers use qdio_inspect_queue() and qdio's internal
queue tasklets are gone, the driver-specified queue handlers are
only called for async error reporting (eg. for an error condition in
the QEBSM code).
So take a moment to clean up the Output Queue handlers (they are
_always_ called with qdio_error != 0), and clarify which error types
can be reported through what interface. As Benjamin already suggested
a while ago, we should turn these into distinct enums at some point.
Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
Reviewed-by: Benjamin Block <bblock@linux.ibm.com>
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
Both qdio drivers have moved away from using qdio's internal tasklet
and timer mechanisms for Output Queues. Rip out all the leftovers.
Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
Reviewed-by: Benjamin Block <bblock@linux.ibm.com>
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
There is no useful information within [STARTUP_NORMAL_OFFSET, HEAD_END] now.
But the memory region [0, STARTUP_NORMAL_OFFSET] is used by:
* lowcore
* kdump for swapping memory
* stand-alone zipl dumpers for code, data, stack and heap
Signed-off-by: Alexander Egorenkov <egorenar@linux.ibm.com>
Acked-by: Heiko Carstens <hca@linux.ibm.com>
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
This change simplifies the task of making the decompressor relocatable.
The decompressor's image contains special DMA sections between _sdma and
_edma. This DMA segment is loaded at boot as part of the decompressor and
then simply handed over to the decompressed kernel. The decompressor itself
never uses it in any way. The primary reason for this is the need to keep
the aforementioned DMA segment below 2GB which is required by architecture,
and because the decompressor is always loaded at a fixed low physical
address, it is guaranteed that the DMA region will not cross the 2GB
memory limit. If the DMA region had been placed in the decompressed kernel,
then KASLR would make this guarantee impossible to fulfill or it would
be restricted to the first 2GB of memory address space.
This commit moves all DMA sections between _sdma and _edma from
the decompressor's image to the decompressed kernel's image. The complete
DMA region is placed in the init section of the decompressed kernel and
immediately relocated below 2GB at start-up before it is needed by other
parts of the decompressed kernel. The relocation of the DMA region happens
even if the decompressed kernel is already located below 2GB in order
to keep the first implementation simple. The relocation should not have
any noticeable impact on boot time because the DMA segment is only a couple
of pages.
After relocating the DMA sections, the kernel has to fix all references
which point into it. In order to automate this, place all variables
pointing into the DMA sections in a special .dma.refs section. All such
variables must be defined using the new __dma_ref macro. Only variables
containing addresses within the DMA sections must be placed in the new
.dma.refs section.
Furthermore, move the initialization of control registers from
the decompressor to the decompressed kernel because some control registers
reference tables that must be placed in the DMA data section to
guarantee that their addresses are below 2G. Because the decompressed
kernel relocates the DMA sections at startup, the content of control
registers CR2, CR5 and CR15 must be updated with new addresses after
the relocation. The decompressed kernel initializes all control registers
early at boot and then updates the content of CR2, CR5 and CR15
as soon as the DMA relocation has occurred. This practically reverts
the commit a80313ff91 ("s390/kernel: introduce .dma sections").
Signed-off-by: Alexander Egorenkov <egorenar@linux.ibm.com>
Acked-by: Heiko Carstens <hca@linux.ibm.com>
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
As a preparation for moving the .dma.data section from the decompressor to
the decompressed kernel, the .dma.data section must be made relocatable
by replacing absolute memory addressing with relative one. This is required
in order to be able to relocate the DMA section to a memory address <= 2G
as required by the hardware architecture. The DMA section must be
relocated in case the decompressed kernel was loaded to an address >= 2G
which can occur if KASAN is enabled. By making the whole DMA section
position-independent we avoid applying relocations to it whenever it is
moved to a different address, which becomes possible as soon as it becomes
a part of the decompressed kernel.
Signed-off-by: Alexander Egorenkov <egorenar@linux.ibm.com>
Acked-by: Heiko Carstens <hca@linux.ibm.com>
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
Both macros are used only in decompressor's head.S, unnecessary to put
them in a global header used in many places like setup.h is.
Signed-off-by: Alexander Egorenkov <egorenar@linux.ibm.com>
Acked-by: Heiko Carstens <hca@linux.ibm.com>
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
To reduce duplication, replace error-prone and hard-coded parameter area
offsets with auto-generated ones.
Signed-off-by: Alexander Egorenkov <egorenar@linux.ibm.com>
Acked-by: Heiko Carstens <hca@linux.ibm.com>
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
The macros
* IPL_DEVICE_OFFSET
* INITRD_START_OFFSET
* INITRD_SIZE_OFFSET
* OLDMEM_BASE_OFFSET
* OLDMEM_SIZE_OFFSET
* KERNEL_VERSION_OFFSET
* COMMAND_LINE_OFFSET
are no longer necessary and used only to define another set of macros
with the same names but w/o the suffix _OFFSET. Therefore, drop this
unnecessary indirection.
Drop the macro KERNEL_VERSION_OFFSET w/o renaming it to KERNEL_VERSION
because it is used nowhere.
Signed-off-by: Alexander Egorenkov <egorenar@linux.ibm.com>
Acked-by: Heiko Carstens <hca@linux.ibm.com>
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
These symbolic constants are used only by assembler code now:
* COMMAND_LINE
* IPL_DEVICE
C code of the decompressed kernel should use boot data passed
by the decompressor instead.
Signed-off-by: Alexander Egorenkov <egorenar@linux.ibm.com>
Acked-by: Heiko Carstens <hca@linux.ibm.com>
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
The new boot data struct shall replace global variables OLDMEM_BASE and
OLDMEM_SIZE. It is initialized in the decompressor and passed
to the decompressed kernel. In comparison to the old solution, this one
doesn't access data at fixed physical addresses which will become important
when the decompressor becomes relocatable.
Signed-off-by: Alexander Egorenkov <egorenar@linux.ibm.com>
Acked-by: Heiko Carstens <hca@linux.ibm.com>
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
The new boot data struct shall replace global variables INITRD_START and
INITRD_SIZE. It is initialized in the decompressor and passed
to the decompressed kernel. In comparison to the old solution, this one
doesn't access data at fixed physical addresses which will become important
when the decompressor becomes relocatable.
Signed-off-by: Alexander Egorenkov <egorenar@linux.ibm.com>
Acked-by: Heiko Carstens <hca@linux.ibm.com>
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
To make the decompressor relocatable, the early SCLP buffer with a fixed
address must be replaced with a relocatable C buffer of the according size
and alignment as required by SCLP.
Introduce a new function sclp_early_set_buffer() into the SCLP driver
which enables the decompressor to change the SCLP early buffer at any time.
This will be useful when the decompressor becomes fully relocatable and
might need to change the SCLP early buffer to one with an address < 2G
as required by SCLP because it was loaded at an address >= 2G.
Signed-off-by: Alexander Egorenkov <egorenar@linux.ibm.com>
Acked-by: Heiko Carstens <hca@linux.ibm.com>
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
Use STARTUP_NORMAL_OFFSET and STARTUP_KDUMP_OFFSET instead of magic numbers.
Signed-off-by: Alexander Egorenkov <egorenar@linux.ibm.com>
Acked-by: Heiko Carstens <hca@linux.ibm.com>
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
Use system call functions instead of open-coding svc inline
assemblies. This is mostly to get rid of even more register asm
constructs.
Besides that, it makes the code also a bit easier to understand.
The generated code is identical to what is was before.
Reviewed-by: Sven Schnelle <svens@linux.ibm.com>
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
Provide generic system call functions which should be used whenever a
system call needs to be done from user space. The only in-kernel code
is vdso, which will be converted with a follow on patch.
Reviewed-by: Sven Schnelle <svens@linux.ibm.com>
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
Using register asm statements has been proven to be very error prone,
especially when using code instrumentation where gcc may add function
calls, which clobbers register contents in an unexpected way.
Therefore get rid of register asm statements in cpacf code, and make
sure this bug class cannot happen.
Reviewed-by: Patrick Steuer <patrick.steuer@de.ibm.com>
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
De-duplicate checks for Protected Host Virtualization in decompressor and
kernel.
Set prot_virt_host=0 in the decompressor in *any* of the following cases
and hand it over to the decompressed kernel:
* No explicit prot_virt=1 is given on the kernel command-line
* Protected Guest Virtualization is enabled
* Hardware support not present
* kdump or stand-alone dump
The decompressed kernel needs to use only is_prot_virt_host() instead of
performing again all checks done by the decompressor.
Signed-off-by: Alexander Egorenkov <egorenar@linux.ibm.com>
Reviewed-by: Vasily Gorbik <gor@linux.ibm.com>
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
A dump kernel is neither required nor able to support Secure Execution.
Signed-off-by: Alexander Egorenkov <egorenar@linux.ibm.com>
Reviewed-by: Vasily Gorbik <gor@linux.ibm.com>
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
The functions adjust_to_uv_max() and uv_query_info() are used only
in the decompressor. Therefore, move the function declarations from
the global arch/s390/include/asm/uv.h to arch/s390/boot/uv.h.
Signed-off-by: Alexander Egorenkov <egorenar@linux.ibm.com>
Reviewed-by: Vasily Gorbik <gor@linux.ibm.com>
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
In case of a jump label print the real address of the piece of code
where a mismatch was detected. This is right before the system panics,
so there is nothing revealed.
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
Print the real pte, pmd, etc. values instead of some hashed
value. Otherwise debugging would be even more difficult.
This also matches what most other architectures are doing.
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
A buffer that can be used for communication with SCLP is required
to lie below 2GB memory address. Therefore, both sclp_info_sccb
and sclp_early_sccb must fulfill this requirement if passed directly
to the sclp_early_cmd() function. Instead, use only sclp_early_sccb
for communication with SCLP. This allows the buffer sclp_info_sccb
to be placed anywhere in the memory address space and, therefore,
simplifies the process of making the decompressor relocatable later on,
one thing less to relocate. And make sure that the length of the new unified
early SCLP buffer is no less than the length of the removed sclp_info_sccb
buffer which might be larger than the length of the sclp_early_sccb buffer.
Signed-off-by: Alexander Egorenkov <egorenar@linux.ibm.com>
Acked-by: Heiko Carstens <hca@linux.ibm.com>
Reviewed-by: Vasily Gorbik <gor@linux.ibm.com>
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>
Ilya Leoshkevich
Masahiro Yamada
Sven Schnelle
Heiko Carstens
Niklas Schnelle
Julian Wiedmann
Alexander Egorenkov