We can't take traps with relocation off, so blacklist enter_rtas() and
rtas_return_loc(). However, instead of blacklisting all of enter_rtas(),
introduce a new symbol __enter_rtas from where on we can't take a trap
and blacklist that.
Signed-off-by: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Blacklist all functions involved while handling a trap. We:
- convert some of the symbols into private symbols, and
- blacklist most functions involved while handling a trap.
Reviewed-by: Masami Hiramatsu <mhiramat@kernel.org>
Signed-off-by: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
It is actually safe to probe system_call() in entry_64.S, but only till
we unset MSR_RI. To allow this, add a new symbol system_call_exit()
after the mtmsrd and blacklist that.
Suggested-by: Michael Ellerman <mpe@ellerman.id.au>
Reviewed-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
It is common to get a PMU interrupt right after the mtmsr instruction that
enables interrupts. Due to this, the stack trace profile gets needlessly split
across system_call_common() and system_call().
Previously, system_call() symbol was at the current place to hide a few
earlier symbols which have since been made private or removed entirely.
So, let's move system_call() slightly higher up, right after the mtmsr
instruction that enables interrupts. Convert existing references to
system_call to a local syscall symbol.
Suggested-by: Nicholas Piggin <npiggin@gmail.com>
Reviewed-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Convert some of the symbols into private symbols and blacklist
system_call_common() and system_call() from kprobes. We can't take a
trap at parts of these functions as either MSR_RI is unset or the
kernel stack pointer is not yet setup.
Reviewed-by: Masami Hiramatsu <mhiramat@kernel.org>
Reviewed-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com>
[mpe: Don't convert system_call_common to _GLOBAL()]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
The ISA v3.0B copy-paste facility only requires cpabort when switching
to a process that has foreign real addresses mapped (direct access to
accelerators), to clear a potential copy buffer filled by a previous
thread. There is no accelerator driver implemented yet, so cpabort can
be removed. It can be be re-added when a driver is implemented.
POWER9 DD1 requires the copy buffer to always be cleared on context
switch, but if accelerators are not in use, then an unpaired copy from
a dummy region is sufficient to clear data out of the copy buffer.
This increases context switch performance by about 5% on POWER9.
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
The sync (aka. hwsync, aka. heavyweight sync) in the context switch
code to prevent MMIO access being reordered from the point of view of
a single process if it gets migrated to a different CPU is not
required because there is an hwsync performed earlier in the context
switch path.
Comment this so it's clear enough if anything changes on the scheduler
or the powerpc sides. Remove the hwsync from _switch.
This improves context switch performance by 2-3% on POWER8.
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
There is no need to explicitly break the reservation in _switch,
because we are guaranteed that the context switch path will include a
larx/stcx.
Comment the guarantee and remove the reservation clear from _switch.
This is worth 1-2% in context switch performance.
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Commit 4387e9ff25 ("[POWERPC] Fix PMU + soft interrupt disable bug")
hard disabled interrupts over the low level context switch, because
the SLB management can't cope with a PMU interrupt accesing the stack
in that window.
Radix based kernel mapping does not use the SLB so it does not require
interrupts hard disabled here.
This is worth 1-2% in context switch performance on POWER9.
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
The syscall exit code that branches to restore_math is quite heavy on
Book3S, consisting of 2 mtmsr instructions. Threads that don't use both
FP and vector can get caught here if the kernel ever uses FP or vector.
Lazy-FP/vec context switching also trips this case.
So check for lazy FP and vector before switching RI for restore_math.
Move most of this case out of line.
For threads that do want to restore math registers, the MSR switches are
still suboptimal. Future direction may be to use a soft-RI bit to avoid
MSR switches in kernel (similar to soft-EE), but for now at least the
no-restore
POWER9 context switch rate increases by about 5% due to sched_yield(2)
return performance. I haven't constructed a test to measure the syscall
cost.
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Highlights include:
- Larger virtual address space on 64-bit server CPUs. By default we use a 128TB
virtual address space, but a process can request access to the full 512TB by
passing a hint to mmap().
- Support for the new Power9 "XIVE" interrupt controller.
- TLB flushing optimisations for the radix MMU on Power9.
- Support for CAPI cards on Power9, using the "Coherent Accelerator Interface
Architecture 2.0".
- The ability to configure the mmap randomisation limits at build and runtime.
- Several small fixes and cleanups to the kprobes code, as well as support for
KPROBES_ON_FTRACE.
- Major improvements to handling of system reset interrupts, correctly treating
them as NMIs, giving them a dedicated stack and using a new hypervisor call
to trigger them, all of which should aid debugging and robustness.
Many fixes and other minor enhancements.
Thanks to:
Alastair D'Silva, Alexey Kardashevskiy, Alistair Popple, Andrew Donnellan,
Aneesh Kumar K.V, Anshuman Khandual, Anton Blanchard, Balbir Singh, Ben
Hutchings, Benjamin Herrenschmidt, Bhupesh Sharma, Chris Packham, Christian
Zigotzky, Christophe Leroy, Christophe Lombard, Daniel Axtens, David Gibson,
Gautham R. Shenoy, Gavin Shan, Geert Uytterhoeven, Guilherme G. Piccoli,
Hamish Martin, Hari Bathini, Kees Cook, Laurent Dufour, Madhavan Srinivasan,
Mahesh J Salgaonkar, Mahesh Salgaonkar, Masami Hiramatsu, Matt Brown, Matthew
R. Ochs, Michael Neuling, Naveen N. Rao, Nicholas Piggin, Oliver O'Halloran,
Pan Xinhui, Paul Mackerras, Rashmica Gupta, Russell Currey, Sukadev
Bhattiprolu, Thadeu Lima de Souza Cascardo, Tobin C. Harding, Tyrel Datwyler,
Uma Krishnan, Vaibhav Jain, Vipin K Parashar, Yang Shi.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1
iQIcBAABAgAGBQJZDHUMAAoJEFHr6jzI4aWAT7oQALkE2Nj3gjcn1z0SkFhq/1iO
Py9Elmqm4E+L6NKYtBY5dS8xVAJ088ffzERyqJ1FY1LHkB8tn8bWRcMQmbjAFzTI
V4TAzDNI890BN/F4ptrYRwNFxRBHAvZ4NDunTzagwYnwmTzW9PYHmOi4pvWTo3Tw
KFUQ0joLSEgHzyfXxYB3fyj41u8N0FZvhfazdNSqia2Y5Vwwv/ION5jKplDM+09Y
EtVEXFvaKAS1sjbM/d/Jo5rblHfR0D9/lYV10+jjyIokjzslIpyTbnj3izeYoM5V
I4h99372zfsEjBGPPXyM3khL3zizGMSDYRmJHQSaKxjtecS9SPywPTZ8ufO/aSzV
Ngq6nlND+f1zep29VQ0cxd3Jh40skWOXzxJaFjfDT25xa6FbfsWP2NCtk8PGylZ7
EyqTuCWkMgIP02KlX3oHvEB2LRRPCDmRU2zECecRGNJrIQwYC2xjoiVi7Q8Qe8rY
gr7Ib5Jj/a+uiTcCIy37+5nXq2s14/JBOKqxuYZIxeuZFvKYuRUipbKWO05WDOAz
m/pSzeC3J8AAoYiqR0gcSOuJTOnJpGhs7zrQFqnEISbXIwLW+ICumzOmTAiBqOEY
Rt8uW2gYkPwKLrE05445RfVUoERaAjaE06eRMOWS6slnngHmmnRJbf3PcoALiJkT
ediqGEj0/N1HMB31V5tS
=vSF3
-----END PGP SIGNATURE-----
Merge tag 'powerpc-4.12-1' of git://git.kernel.org/pub/scm/linux/kernel/git/powerpc/linux
Pull powerpc updates from Michael Ellerman:
"Highlights include:
- Larger virtual address space on 64-bit server CPUs. By default we
use a 128TB virtual address space, but a process can request access
to the full 512TB by passing a hint to mmap().
- Support for the new Power9 "XIVE" interrupt controller.
- TLB flushing optimisations for the radix MMU on Power9.
- Support for CAPI cards on Power9, using the "Coherent Accelerator
Interface Architecture 2.0".
- The ability to configure the mmap randomisation limits at build and
runtime.
- Several small fixes and cleanups to the kprobes code, as well as
support for KPROBES_ON_FTRACE.
- Major improvements to handling of system reset interrupts,
correctly treating them as NMIs, giving them a dedicated stack and
using a new hypervisor call to trigger them, all of which should
aid debugging and robustness.
- Many fixes and other minor enhancements.
Thanks to: Alastair D'Silva, Alexey Kardashevskiy, Alistair Popple,
Andrew Donnellan, Aneesh Kumar K.V, Anshuman Khandual, Anton
Blanchard, Balbir Singh, Ben Hutchings, Benjamin Herrenschmidt,
Bhupesh Sharma, Chris Packham, Christian Zigotzky, Christophe Leroy,
Christophe Lombard, Daniel Axtens, David Gibson, Gautham R. Shenoy,
Gavin Shan, Geert Uytterhoeven, Guilherme G. Piccoli, Hamish Martin,
Hari Bathini, Kees Cook, Laurent Dufour, Madhavan Srinivasan, Mahesh J
Salgaonkar, Mahesh Salgaonkar, Masami Hiramatsu, Matt Brown, Matthew
R. Ochs, Michael Neuling, Naveen N. Rao, Nicholas Piggin, Oliver
O'Halloran, Pan Xinhui, Paul Mackerras, Rashmica Gupta, Russell
Currey, Sukadev Bhattiprolu, Thadeu Lima de Souza Cascardo, Tobin C.
Harding, Tyrel Datwyler, Uma Krishnan, Vaibhav Jain, Vipin K Parashar,
Yang Shi"
* tag 'powerpc-4.12-1' of git://git.kernel.org/pub/scm/linux/kernel/git/powerpc/linux: (214 commits)
powerpc/64s: Power9 has no LPCR[VRMASD] field so don't set it
powerpc/powernv: Fix TCE kill on NVLink2
powerpc/mm/radix: Drop support for CPUs without lockless tlbie
powerpc/book3s/mce: Move add_taint() later in virtual mode
powerpc/sysfs: Move #ifdef CONFIG_HOTPLUG_CPU out of the function body
powerpc/smp: Document irq enable/disable after migrating IRQs
powerpc/mpc52xx: Don't select user-visible RTAS_PROC
powerpc/powernv: Document cxl dependency on special case in pnv_eeh_reset()
powerpc/eeh: Clean up and document event handling functions
powerpc/eeh: Avoid use after free in eeh_handle_special_event()
cxl: Mask slice error interrupts after first occurrence
cxl: Route eeh events to all drivers in cxl_pci_error_detected()
cxl: Force context lock during EEH flow
powerpc/64: Allow CONFIG_RELOCATABLE if COMPILE_TEST
powerpc/xmon: Teach xmon oops about radix vectors
powerpc/mm/hash: Fix off-by-one in comment about kernel contexts ids
powerpc/pseries: Enable VFIO
powerpc/powernv: Fix iommu table size calculation hook for small tables
powerpc/powernv: Check kzalloc() return value in pnv_pci_table_alloc
powerpc: Add arch/powerpc/tools directory
...
entry_*.S now includes a lot more than just kernel entry/exit code. As a
first step at cleaning this up, let's split out the ftrace bits into
separate files. Also move all related tracing code into a new trace/
subdirectory.
No functional changes.
Suggested-by: Michael Ellerman <mpe@ellerman.id.au>
Signed-off-by: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Although most of these kprobes patches are powerpc specific, there's a couple
that touch generic code (with Acks). At the moment there's one conflict with
acme's tree, but it's not too bad. Still just in case some other conflicts show
up, we've put these in a topic branch so another tree could merge some or all of
it if necessary.
Pass the real LR to the ftrace handler. This is needed for KPROBES_ON_FTRACE for
the pre handlers.
Also, with KPROBES_ON_FTRACE, the link register may be updated by the pre
handlers or by a registed kretprobe. Honor updated LR by restoring it from
pt_regs, rather than from the stack save area.
Live patch and function graph continue to work fine with this change.
Signed-off-by: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Move the stack setup and teardown code into ftrace_graph_caller(). This way, we
don't incur the cost of setting it up unless function graph is enabled for this
function.
Also, remove the extraneous LR restore code after the function graph stub. LR
has previously been restored and neither livepatch_handler() nor
ftrace_graph_caller() return back here.
Signed-off-by: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com>
[mpe: Drop bad change to non-mprofile-kernel version of ftrace_graph_caller]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
If we set a kprobe on a 'stdu' instruction on powerpc64, we see a kernel
OOPS:
Bad kernel stack pointer cd93c840 at c000000000009868
Oops: Bad kernel stack pointer, sig: 6 [#1]
...
GPR00: c000001fcd93cb30 00000000cd93c840 c0000000015c5e00 00000000cd93c840
...
NIP [c000000000009868] resume_kernel+0x2c/0x58
LR [c000000000006208] program_check_common+0x108/0x180
On a 64-bit system when the user probes on a 'stdu' instruction, the kernel does
not emulate actual store in emulate_step() because it may corrupt the exception
frame. So the kernel does the actual store operation in exception return code
i.e. resume_kernel().
resume_kernel() loads the saved stack pointer from memory using lwz, which only
loads the low 32-bits of the address, causing the kernel crash.
Fix this by loading the 64-bit value instead.
Fixes: be96f63375 ("powerpc: Split out instruction analysis part of emulate_step()")
Cc: stable@vger.kernel.org # v3.18+
Signed-off-by: Ravi Bangoria <ravi.bangoria@linux.vnet.ibm.com>
Reviewed-by: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com>
Reviewed-by: Ananth N Mavinakayanahalli <ananth@linux.vnet.ibm.com>
[mpe: Change log massage, add stable tag]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Pull kbuild updates from Michal Marek:
- EXPORT_SYMBOL for asm source by Al Viro.
This does bring a regression, because genksyms no longer generates
checksums for these symbols (CONFIG_MODVERSIONS). Nick Piggin is
working on a patch to fix this.
Plus, we are talking about functions like strcpy(), which rarely
change prototypes.
- Fixes for PPC fallout of the above by Stephen Rothwell and Nick
Piggin
- fixdep speedup by Alexey Dobriyan.
- preparatory work by Nick Piggin to allow architectures to build with
-ffunction-sections, -fdata-sections and --gc-sections
- CONFIG_THIN_ARCHIVES support by Stephen Rothwell
- fix for filenames with colons in the initramfs source by me.
* 'kbuild' of git://git.kernel.org/pub/scm/linux/kernel/git/mmarek/kbuild: (22 commits)
initramfs: Escape colons in depfile
ppc: there is no clear_pages to export
powerpc/64: whitelist unresolved modversions CRCs
kbuild: -ffunction-sections fix for archs with conflicting sections
kbuild: add arch specific post-link Makefile
kbuild: allow archs to select link dead code/data elimination
kbuild: allow architectures to use thin archives instead of ld -r
kbuild: Regenerate genksyms lexer
kbuild: genksyms fix for typeof handling
fixdep: faster CONFIG_ search
ia64: move exports to definitions
sparc32: debride memcpy.S a bit
[sparc] unify 32bit and 64bit string.h
sparc: move exports to definitions
ppc: move exports to definitions
arm: move exports to definitions
s390: move exports to definitions
m68k: move exports to definitions
alpha: move exports to actual definitions
x86: move exports to actual definitions
...
mtmsrd with L=1 only affects MSR_EE and MSR_RI bits, and we always
know what state those bits are, so the kernel MSR does not need to be
loaded when modifying them.
mtmsrd is often in the critical execution path, so avoiding dependency
on even L1 load is noticable. On a POWER8 this saves about 3 cycles
from the syscall path, and possibly a few from other exception returns
(not measured).
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
tabort_syscall runs with RI=1, so a nested recoverable machine
check will load the paca into r13 and overwrite what we loaded
it with, because exceptions returning to privileged mode do not
restore r13.
Fixes: b4b56f9eca (powerpc/tm: Abort syscalls in active transactions)
Cc: stable@vger.kernel.org
Signed-off-by: Nick Piggin <npiggin@gmail.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
MMU feature bits are defined such that we use the lower half to
present MMU family features. Remove the strict split of half and
also move Radix to a mmu family feature. Radix introduce a new MMU
model and strictly speaking it is a new MMU family. This also free
up bits which can be used for individual features later.
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
This patch provides VIRT_CPU_ACCOUTING to PPC32 architecture.
PPC32 doesn't have the PACA structure, so we use the task_info
structure to store the accounting data.
In order to reuse on PPC32 the PPC64 functions, all u64 data has
been replaced by 'unsigned long' so that it is u32 on PPC32 and
u64 on PPC64
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Scott Wood <oss@buserror.net>
We're approaching 20 locations where we need to check for ELF ABI v2.
That's fine, except the logic is a bit awkward, because we have to check
that _CALL_ELF is defined and then what its value is.
So check it once in asm/types.h and define PPC64_ELF_ABI_v2 when ELF ABI
v2 is detected.
We also have a few places where what we're really trying to check is
that we are using the 64-bit v1 ABI, ie. function descriptors. So also
add a #define for that, which simplifies several checks.
Signed-off-by: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
We also use MMU_FTR_RADIX to branch out from code path specific to
hash.
No functionality change.
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
The copy paste facility introduced in POWER9 provides an optimised
mechanism for a userspace application to copy a cacheline. This is
provided by a pair of instructions, copy and paste, while a third,
cp_abort (copy paste abort), provides a clean up of the state in case of
a failure.
The copy instruction will read a 128 byte cacheline and store it in an
internal buffer. The subsequent paste instruction will store this
internal buffer to memory and set a CR field if the paste succeeds.
Since the state of the copy paste buffer is internal (and not
architecturally visible), in the unlikely event of a context switch, the
state cannot be stored and the paste should therefore fail.
The cp_abort instruction exists to fail and clean up any such
interrupted copy paste sequence and is to be called by the kernel as
part of the context switch. Doing so prevents data from a preceding copy
in one process leaking into the paste of another.
This code enables use of the cp_abort instruction if a supported
processor is detected.
NOTE: this is for userspace only, not in kernel, and does not deal
with KVM guests.
Patch created with much assistance from Michael Neuling
<mikey@neuling.org>
Signed-off-by: Chris Smart <chris@distroguy.com>
Reviewed-by: Cyril Bur <cyrilbur@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Add the kconfig logic & assembly support for handling live patched
functions. This depends on DYNAMIC_FTRACE_WITH_REGS, which in turn
depends on the new -mprofile-kernel ftrace ABI, which is only supported
currently on ppc64le.
Live patching is handled by a special ftrace handler. This means it runs
from ftrace_caller(). The live patch handler modifies the NIP so as to
redirect the return from ftrace_caller() to the new patched function.
However there is one particularly tricky case we need to handle.
If a function A calls another function B, and it is known at link time
that they share the same TOC, then A will not save or restore its TOC,
and will call the local entry point of B.
When we live patch B, we replace it with a new function C, which may
not have the same TOC as A. At live patch time it's too late to modify A
to do the TOC save/restore, so the live patching code must interpose
itself between A and C, and do the TOC save/restore that A omitted.
An additionaly complication is that the livepatch code can not create a
stack frame in order to save the TOC. That is because if C takes > 8
arguments, or is varargs, A will have written the arguments for C in
A's stack frame.
To solve this, we introduce a "livepatch stack" which grows upward from
the base of the regular stack, and is used to store the TOC & LR when
calling a live patched function.
When the patched function returns, we retrieve the real LR & TOC from
the livepatch stack, restore them, and pop the livepatch "stack frame".
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Reviewed-by: Torsten Duwe <duwe@suse.de>
Reviewed-by: Balbir Singh <bsingharora@gmail.com>
Commit 70fe3d9 "powerpc: Restore FPU/VEC/VSX if previously used" introduces a
call to restore_math() late in the syscall return path, after MSR_RI has been
cleared. The MSR_RI flag is used to indicate whether the kernel can take
another exception or not. A cleared MSR_RI flag indicates that the kernel
cannot.
Unfortunately when a machine is under SLB pressure an SLB miss can occur
in restore_math() which (with MSR_RI cleared) leads to an unrecoverable
exception.
Unrecoverable exception 4100 at c0000000000088d8
cpu 0x0: Vector: 4100 at [c0000003fa473b20]
pc: c0000000000088d8: .load_vr_state+0x70/0x110
lr: c00000000000f710: .restore_math+0x130/0x188
sp: c0000003fa473da0
msr: 9000000002003030
current = 0xc0000007f876f180
paca = 0xc00000000fff0000 softe: 0 irq_happened: 0x01
pid = 1944, comm = K08umountfs
[link register ] c00000000000f710 .restore_math+0x130/0x188
[c0000003fa473da0] c0000003fa473e30 (unreliable)
[c0000003fa473e30] c000000000007b6c system_call+0x84/0xfc
The clearing of MSR_RI is actually an optimisation to avoid multiple MSR
writes, what must be disabled are interrupts. See comment in entry_64.S:
/*
* For performance reasons we clear RI the same time that we
* clear EE. We only need to clear RI just before we restore r13
* below, but batching it with EE saves us one expensive mtmsrd call.
* We have to be careful to restore RI if we branch anywhere from
* here (eg syscall_exit_work).
*/
At the point of calling restore_math() r13 has not been restored, as such, the
quick fix of turning MSR_RI back on for the call to restore_math() will
eliminate the occurrence of an unrecoverable exception.
We'd like to do a better fix in future.
Fixes: 70fe3d980f ("powerpc: Restore FPU/VEC/VSX if previously used")
Signed-off-by: Cyril Bur <cyrilbur@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Merge the ftrace changes to support -mprofile-kernel on ppc64le. This is
a prerequisite for live patching, the support for which will be merged
via the livepatch tree based on this topic branch.
The gcc switch -mprofile-kernel defines a new ABI for calling _mcount()
very early in the function with minimal overhead.
Although mprofile-kernel has been available since GCC 3.4, there were
bugs which were only fixed recently. Currently it is known to work in
GCC 4.9, 5 and 6.
Additionally there are two possible code sequences generated by the
flag, the first uses mflr/std/bl and the second is optimised to omit the
std. Currently only gcc 6 has the optimised sequence. This patch
supports both sequences.
Initial work started by Vojtech Pavlik, used with permission.
Key changes:
- rework _mcount() to work for both the old and new ABIs.
- implement new versions of ftrace_caller() and ftrace_graph_caller()
which deal with the new ABI.
- updates to __ftrace_make_nop() to recognise the new mcount calling
sequence.
- updates to __ftrace_make_call() to recognise the nop'ed sequence.
- implement ftrace_modify_call().
- updates to the module loader to surpress the toc save in the module
stub when calling mcount with the new ABI.
Reviewed-by: Balbir Singh <bsingharora@gmail.com>
Signed-off-by: Torsten Duwe <duwe@suse.de>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Currently the FPU, VEC and VSX facilities are lazily loaded. This is not
a problem unless a process is using these facilities.
Modern versions of GCC are very good at automatically vectorising code,
new and modernised workloads make use of floating point and vector
facilities, even the kernel makes use of vectorised memcpy.
All this combined greatly increases the cost of a syscall since the
kernel uses the facilities sometimes even in syscall fast-path making it
increasingly common for a thread to take an *_unavailable exception soon
after a syscall, not to mention potentially taking all three.
The obvious overcompensation to this problem is to simply always load
all the facilities on every exit to userspace. Loading up all FPU, VEC
and VSX registers every time can be expensive and if a workload does
avoid using them, it should not be forced to incur this penalty.
An 8bit counter is used to detect if the registers have been used in the
past and the registers are always loaded until the value wraps to back
to zero.
Several versions of the assembly in entry_64.S were tested:
1. Always calling C.
2. Performing a common case check and then calling C.
3. A complex check in asm.
After some benchmarking it was determined that avoiding C in the common
case is a performance benefit (option 2). The full check in asm (option
3) greatly complicated that codepath for a negligible performance gain
and the trade-off was deemed not worth it.
Signed-off-by: Cyril Bur <cyrilbur@gmail.com>
[mpe: Move load_vec in the struct to fill an existing hole, reword change log]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
fixup
Move all our context switch SPR save and restore code into two
helpers. We do a few optimisations:
- Group all mfsprs and all mtsprs. In many cases an mtspr sets a
scoreboarding bit that an mfspr waits on, so the current practise of
mfspr A; mtspr A; mfpsr B; mtspr B is the worst scheduling we can
do.
- SPR writes are slow, so check that the value is changing before
writing it.
A context switch microbenchmark using yield():
http://ozlabs.org/~anton/junkcode/context_switch2.c
./context_switch2 --test=yield 0 0
shows an improvement of almost 10% on POWER8.
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Writing the MSR is slow, so we want to avoid it whenever possible.
A subsequent patch will add a debug option that strictly manages the
FP/VMX/VSX unavailable bits. For now just remove it, matching what
we do in other areas of the kernel (eg enable_kernel_altivec()).
A context switch microbenchmark using yield():
http://ozlabs.org/~anton/junkcode/context_switch2.c
./context_switch2 --test=yield --fp 0 0
shows an improvement of almost 3% on POWER8.
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
The API for calling do_syscall_trace_enter() is currently sensible
enough, it just returns the (modified) syscall number.
However once we enable seccomp filter it will get more complicated. When
seccomp filter runs, the seccomp kernel code (via SECCOMP_RET_ERRNO), or
a ptracer (via SECCOMP_RET_TRACE), may reject the syscall and *may* or may
*not* set a return value in r3.
That means the assembler that calls do_syscall_trace_enter() can not
blindly return ENOSYS, it needs to only return ENOSYS if a return value
has not already been set.
There is no way to implement that logic with the current API. So change
the do_syscall_trace_enter() API to make it deal with the return code
juggling, and the assembler can then just return whatever return code it
is given.
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Reviewed-by: Kees Cook <keescook@chromium.org>
Currently on powerpc we have our own #define for the highest (negative)
errno value, called _LAST_ERRNO. This is defined to be 516, for reasons
which are not clear.
The generic code, and x86, use MAX_ERRNO, which is defined to be 4095.
In particular seccomp uses MAX_ERRNO to restrict the value that a
seccomp filter can return.
Currently with the mismatch between _LAST_ERRNO and MAX_ERRNO, a seccomp
tracer wanting to return 600, expecting it to be seen as an error, would
instead find on powerpc that userspace sees a successful syscall with a
return value of 600.
To avoid this inconsistency, switch powerpc to use MAX_ERRNO.
We are somewhat confident that generic syscalls that can return a
non-error value above negative MAX_ERRNO have already been updated to
use force_successful_syscall_return().
I have also checked all the powerpc specific syscalls, and believe that
none of them expect to return a non-error value between -MAX_ERRNO and
-516. So this change should be safe ...
Acked-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Reviewed-by: Kees Cook <keescook@chromium.org>
This patch changes the syscall handler to doom (tabort) active
transactions when a syscall is made and return very early without
performing the syscall and keeping side effects to a minimum (no CPU
accounting or system call tracing is performed). Also included is a
new HWCAP2 bit, PPC_FEATURE2_HTM_NOSC, to indicate this
behaviour to userspace.
Currently, the system call instruction automatically suspends an
active transaction which causes side effects to persist when an active
transaction fails.
This does change the kernel's behaviour, but in a way that was
documented as unsupported. It doesn't reduce functionality as
syscalls will still be performed after tsuspend; it just requires that
the transaction be explicitly suspended. It also provides a
consistent interface and makes the behaviour of user code
substantially the same across powerpc and platforms that do not
support suspended transactions (e.g. x86 and s390).
Performance measurements using
http://ozlabs.org/~anton/junkcode/null_syscall.c indicate the cost of
a normal (non-aborted) system call increases by about 0.25%.
Signed-off-by: Sam Bobroff <sam.bobroff@au1.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
PACA_DSCR offset macro tracks dscr_default element in the paca
structure. Better change the name of this macro to match that of the
data element it tracks. Makes the code more readable.
Signed-off-by: Anshuman Khandual <khandual@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
This reverts commit feba40362b.
Although the principle of this change is good, the implementation has a
few issues.
Firstly we can sometimes fail to abort a syscall because r12 may have
been clobbered by C code if we went down the virtual CPU accounting
path, or if syscall tracing was enabled.
Secondly we have decided that it is safer to abort the syscall even
earlier in the syscall entry path, so that we avoid the syscall tracing
path when we are transactional.
So that we have time to thoroughly test those changes we have decided to
revert this for this merge window and will merge the fixed version in
the next window.
NB. Rather than reverting the selftest we just drop tm-syscall from
TEST_PROGS so that it's not run by default.
Fixes: feba40362b ("powerpc/tm: Abort syscalls in active transactions")
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
This patch changes the syscall handler to doom (tabort) active
transactions when a syscall is made and return immediately without
performing the syscall.
Currently, the system call instruction automatically suspends an
active transaction which causes side effects to persist when an active
transaction fails.
This does change the kernel's behaviour, but in a way that was
documented as unsupported. It doesn't reduce functionality because
syscalls will still be performed after tsuspend. It also provides a
consistent interface and makes the behaviour of user code
substantially the same across powerpc and platforms that do not
support suspended transactions (e.g. x86 and s390).
Performance measurements using
http://ozlabs.org/~anton/junkcode/null_syscall.c
indicate the cost of a system call increases by about 0.5%.
Signed-off-by: Sam Bobroff <sam.bobroff@au1.ibm.com>
Acked-By: Michael Neuling <mikey@neuling.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
We currently have a "special" syscall for switching endianness. This is
syscall number 0x1ebe, which is handled explicitly in the 64-bit syscall
exception entry.
That has a few problems, firstly the syscall number is outside of the
usual range, which confuses various tools. For example strace doesn't
recognise the syscall at all.
Secondly it's handled explicitly as a special case in the syscall
exception entry, which is complicated enough without it.
As a first step toward removing the special syscall, we need to add a
regular syscall that implements the same functionality.
The logic is simple, it simply toggles the MSR_LE bit in the userspace
MSR. This is the same as the special syscall, with the caveat that the
special syscall clobbers fewer registers.
This version clobbers r9-r12, XER, CTR, and CR0-1,5-7.
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
We have code to do syscall tracing which is disabled at compile time by
default. It's not been touched since the dawn of time (ie. v2.6.12).
There are now better ways to do syscall tracing, ie. using the
raw_syscall, or syscall tracepoints.
For the specific case of tracing syscalls at boot on a system that
doesn't get to userspace, you can boot with:
trace_event=syscalls tp_printk=on
Which will trace syscalls from boot, and echo all output to the console.
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Currently when we back trace something that is in a syscall we see
something like this:
[c000000000000000] [c000000000000000] SyS_read+0x6c/0x110
[c000000000000000] [c000000000000000] syscall_exit+0x0/0x98
Although it's entirely correct, seeing syscall_exit at the bottom can be
confusing - we were exiting from a syscall and then called SyS_read() ?
If we instead change syscall_exit to be a local label we get something
more intuitive:
[c0000001fa46fde0] [c00000000026719c] SyS_read+0x6c/0x110
[c0000001fa46fe30] [c000000000009264] system_call+0x38/0xd0
ie. we were handling a system call, and it was SyS_read().
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Once upon a time, at least 9 years ago (< 2.6.12), _TIF_SYSCALL_T_OR_A
meant "TRACE or AUDIT". But these days it means TRACE or AUDIT or
SECCOMP or TRACEPOINT or NOHZ.
All of those are implemented via syscall_dotrace() so rename the flag to
that to try and clarify things.
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Instead of passing in the stack address of the link register
to be modified, just pass in the old value and return the
new value and rely on ftrace_graph_caller to do the
modification.
This removes the exception handling around the stack update -
it isn't needed and we weren't consistent about it. Later on
we would do an unprotected modification:
if (!ftrace_graph_entry(&trace)) {
*parent = old;
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
mod_return_to_handler is the same as return_to_handler, except
it handles the change of the TOC (r2). Add this into
return_to_handler and remove mod_return_to_handler.
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Back in 7230c56441 ("powerpc: Rework lazy-interrupt handling") we
added a call out to restore_interrupts() (written in c) before calling
do_notify_resume:
bl restore_interrupts
addi r3,r1,STACK_FRAME_OVERHEAD
bl do_notify_resume
Unfortunately do_notify_resume takes two arguments, the second one
being the thread_info flags:
void do_notify_resume(struct pt_regs *regs, unsigned long thread_info_flags)
We do populate r4 (the second argument) earlier, but
restore_interrupts() is free to muck it up all it wants. My guess is
the gcc compiler gods shone down on us and its register allocator
never used r4. Sometimes, rarely, luck is on our side.
LLVM on the other hand did trample r4.
Signed-off-by: Anton Blanchard <anton@samba.org>
Cc: stable@vger.kernel.org
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Michael Ellerman