flush_dcache_page() is only a few lines, it is worth
inlining.
ia64, csky, mips, openrisc and riscv have a similar
flush_dcache_page() and inline it.
On pmac32_defconfig, we get a small size reduction.
On ppc64_defconfig, we get a very small size increase.
In both case that's in the noise (less than 0.1%).
text data bss dec hex filename
18991155 5934744 1497624 26423523 19330e3 vmlinux64.before
18994829 59367321497624 26429185 1934701 vmlinux64.after
9150963 2467502 184548 11803013 b41985 vmlinux32.before
9149689 2467302 184548 11801539 b413c3 vmlinux32.after
Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/21c417488b70b7629dae316539fb7bb8bdef4fdd.1617895813.git.christophe.leroy@csgroup.eu
Guest LPCR depends on hardware type, and future changes will add
restrictions based on errata and guest MMU mode. Move this logic
to a common function and use it for the cases where the guest
wants to update its LPCR (or the LPCR of a nested guest).
This also adds a warning in other places that set or update LPCR
if we try to set something that would have been disallowed by
the filter, as a sanity check.
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Reviewed-by: Fabiano Rosas <farosas@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210412014845.1517916-4-npiggin@gmail.com
For unknown reason, old commit d27dfd388715 ("Import pre2.0.8")
changed 'ptrdiff_t' from 'int' to 'long'.
GCC expects it as 'int' really, and this leads to the following
warning when building KFENCE:
CC mm/kfence/report.o
In file included from ./include/linux/printk.h:7,
from ./include/linux/kernel.h:16,
from mm/kfence/report.c:10:
mm/kfence/report.c: In function 'kfence_report_error':
./include/linux/kern_levels.h:5:18: warning: format '%td' expects argument of type 'ptrdiff_t', but argument 6 has type 'long int' [-Wformat=]
5 | #define KERN_SOH "\001" /* ASCII Start Of Header */
| ^~~~~~
./include/linux/kern_levels.h:11:18: note: in expansion of macro 'KERN_SOH'
11 | #define KERN_ERR KERN_SOH "3" /* error conditions */
| ^~~~~~~~
./include/linux/printk.h:343:9: note: in expansion of macro 'KERN_ERR'
343 | printk(KERN_ERR pr_fmt(fmt), ##__VA_ARGS__)
| ^~~~~~~~
mm/kfence/report.c:213:3: note: in expansion of macro 'pr_err'
213 | pr_err("Out-of-bounds %s at 0x%p (%luB %s of kfence-#%td):\n",
| ^~~~~~
<asm-generic/uapi/posix-types.h> defines it as 'int', and
defines 'size_t' and 'ssize_t' exactly as powerpc do, so
remove the powerpc specific definitions and fallback on
generic ones.
Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Acked-by: Segher Boessenkool <segher@kernel.crashing.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/e43d133bf52fa19e577f64f3a3a38cedc570377d.1617616601.git.christophe.leroy@csgroup.eu
In the past we had a fallback definition for _PAGE_KERNEL_ROX, but we
removed that in commit d82fd29c5a ("powerpc/mm: Distribute platform
specific PAGE and PMD flags and definitions") and added definitions
for each MMU family.
However we missed adding a definition for 64s, which was not really a
bug because it's currently not used.
But we'd like to use PAGE_KERNEL_ROX in a future patch so add a
definition now.
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210331003845.216246-1-mpe@ellerman.id.au
When adding a PTE a ptesync is needed to order the update of the PTE
with subsequent accesses otherwise a spurious fault may be raised.
radix__set_pte_at() does not do this for performance gains. For
non-kernel memory this is not an issue as any faults of this kind are
corrected by the page fault handler. For kernel memory these faults
are not handled. The current solution is that there is a ptesync in
flush_cache_vmap() which should be called when mapping from the
vmalloc region.
However, map_kernel_page() does not call flush_cache_vmap(). This is
troublesome in particular for code patching with Strict RWX on radix.
In do_patch_instruction() the page frame that contains the instruction
to be patched is mapped and then immediately patched. With no ordering
or synchronization between setting up the PTE and writing to the page
it is possible for faults.
As the code patching is done using __put_user_asm_goto() the resulting
fault is obscured - but using a normal store instead it can be seen:
BUG: Unable to handle kernel data access on write at 0xc008000008f24a3c
Faulting instruction address: 0xc00000000008bd74
Oops: Kernel access of bad area, sig: 11 [#1]
LE PAGE_SIZE=64K MMU=Radix SMP NR_CPUS=2048 NUMA PowerNV
Modules linked in: nop_module(PO+) [last unloaded: nop_module]
CPU: 4 PID: 757 Comm: sh Tainted: P O 5.10.0-rc5-01361-ge3c1b78c8440-dirty #43
NIP: c00000000008bd74 LR: c00000000008bd50 CTR: c000000000025810
REGS: c000000016f634a0 TRAP: 0300 Tainted: P O (5.10.0-rc5-01361-ge3c1b78c8440-dirty)
MSR: 9000000000009033 <SF,HV,EE,ME,IR,DR,RI,LE> CR: 44002884 XER: 00000000
CFAR: c00000000007c68c DAR: c008000008f24a3c DSISR: 42000000 IRQMASK: 1
This results in the kind of issue reported here:
https://lore.kernel.org/linuxppc-dev/15AC5B0E-A221-4B8C-9039-FA96B8EF7C88@lca.pw/
Chris Riedl suggested a reliable way to reproduce the issue:
$ mount -t debugfs none /sys/kernel/debug
$ (while true; do echo function > /sys/kernel/debug/tracing/current_tracer ; echo nop > /sys/kernel/debug/tracing/current_tracer ; done) &
Turning ftrace on and off does a large amount of code patching which
in usually less then 5min will crash giving a trace like:
ftrace-powerpc: (____ptrval____): replaced (4b473b11) != old (60000000)
------------[ ftrace bug ]------------
ftrace failed to modify
[<c000000000bf8e5c>] napi_busy_loop+0xc/0x390
actual: 11:3b:47:4b
Setting ftrace call site to call ftrace function
ftrace record flags: 80000001
(1)
expected tramp: c00000000006c96c
------------[ cut here ]------------
WARNING: CPU: 4 PID: 809 at kernel/trace/ftrace.c:2065 ftrace_bug+0x28c/0x2e8
Modules linked in: nop_module(PO-) [last unloaded: nop_module]
CPU: 4 PID: 809 Comm: sh Tainted: P O 5.10.0-rc5-01360-gf878ccaf250a #1
NIP: c00000000024f334 LR: c00000000024f330 CTR: c0000000001a5af0
REGS: c000000004c8b760 TRAP: 0700 Tainted: P O (5.10.0-rc5-01360-gf878ccaf250a)
MSR: 900000000282b033 <SF,HV,VEC,VSX,EE,FP,ME,IR,DR,RI,LE> CR: 28008848 XER: 20040000
CFAR: c0000000001a9c98 IRQMASK: 0
GPR00: c00000000024f330 c000000004c8b9f0 c000000002770600 0000000000000022
GPR04: 00000000ffff7fff c000000004c8b6d0 0000000000000027 c0000007fe9bcdd8
GPR08: 0000000000000023 ffffffffffffffd8 0000000000000027 c000000002613118
GPR12: 0000000000008000 c0000007fffdca00 0000000000000000 0000000000000000
GPR16: 0000000023ec37c5 0000000000000000 0000000000000000 0000000000000008
GPR20: c000000004c8bc90 c0000000027a2d20 c000000004c8bcd0 c000000002612fe8
GPR24: 0000000000000038 0000000000000030 0000000000000028 0000000000000020
GPR28: c000000000ff1b68 c000000000bf8e5c c00000000312f700 c000000000fbb9b0
NIP ftrace_bug+0x28c/0x2e8
LR ftrace_bug+0x288/0x2e8
Call Trace:
ftrace_bug+0x288/0x2e8 (unreliable)
ftrace_modify_all_code+0x168/0x210
arch_ftrace_update_code+0x18/0x30
ftrace_run_update_code+0x44/0xc0
ftrace_startup+0xf8/0x1c0
register_ftrace_function+0x4c/0xc0
function_trace_init+0x80/0xb0
tracing_set_tracer+0x2a4/0x4f0
tracing_set_trace_write+0xd4/0x130
vfs_write+0xf0/0x330
ksys_write+0x84/0x140
system_call_exception+0x14c/0x230
system_call_common+0xf0/0x27c
To fix this when updating kernel memory PTEs using ptesync.
Fixes: f1cb8f9beb ("powerpc/64s/radix: avoid ptesync after set_pte and ptep_set_access_flags")
Signed-off-by: Jordan Niethe <jniethe5@gmail.com>
Reviewed-by: Nicholas Piggin <npiggin@gmail.com>
[mpe: Tidy up change log slightly]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210208032957.1232102-1-jniethe5@gmail.com
Convert powerpc to relative jump labels.
Before the patch, pseries_defconfig vmlinux.o has:
9074 __jump_table 0003f2a0 0000000000000000 0000000000000000 01321fa8 2**0
With the patch, the same config gets:
9074 __jump_table 0002a0e0 0000000000000000 0000000000000000 01321fb4 2**0
Size is 258720 without the patch, 172256 with the patch.
That's a 33% size reduction.
Largely copied from commit c296146c05 ("arm64/kernel: jump_label:
Switch to relative references")
Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/828348da7868eda953ce023994404dfc49603b64.1616514473.git.christophe.leroy@csgroup.eu
Commit d02f6b7dab ("powerpc/uaccess: Evaluate macro arguments once,
before user access is allowed") changed the __chk_user_ptr()
argument from the passed ptr pointer to the locally
declared __gu_addr. But __gu_addr is locally defined as __user
so the check is pointless.
During kernel build __chk_user_ptr() voids and is only evaluated
during sparse checks so it should have been armless to leave the
original pointer check there.
Nevertheless, this check is indeed redundant with the assignment
above which casts the ptr pointer to the local __user __gu_addr.
In case of mismatch, sparse will detect it there, so the
__check_user_ptr() is not needed anywhere else than in access_ok().
Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/69f17d75046733b891ab2e668dbf464787cdf598.1615398265.git.christophe.leroy@csgroup.eu
Commit 6bfd93c32a ("powerpc: Fix incorrect might_sleep in
__get_user/__put_user on kernel addresses") added a check to not call
might_sleep() on kernel addresses. This was to enable the use of
__get_user() in the alignment exception handler for any address.
Then commit 95156f0051 ("lockdep, mm: fix might_fault() annotation")
added a check of the address space in might_fault(), based on
set_fs() logic. But this didn't solve the powerpc alignment exception
case as it didn't call set_fs(KERNEL_DS).
Nowadays, set_fs() is gone, previous patch fixed the alignment
exception handler and __get_user/__put_user are not supposed to be
used anymore to read kernel memory.
Therefore the is_kernel_addr() check has become useless and can be
removed.
Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Reviewed-by: Daniel Axtens <dja@axtens.net>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/e0a980a4dc7a2551183dd5cb30f46eafdbee390c.1615398265.git.christophe.leroy@csgroup.eu
Powerpc is the only architecture having _inatomic variants of
__get_user() and __put_user() accessors. They were introduced
by commit e68c825bb0 ("[POWERPC] Add inatomic versions of __get_user
and __put_user").
Those variants expand to the _nosleep macros instead of expanding
to the _nocheck macros. The only difference between the _nocheck
and the _nosleep macros is the call to might_fault().
Since commit 662bbcb274 ("mm, sched: Allow uaccess in atomic with
pagefault_disable()"), __get/put_user() can be used in atomic parts
of the code, therefore __get/put_user_inatomic() have become useless.
Remove __get_user_inatomic() and __put_user_inatomic().
Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Reviewed-by: Daniel Axtens <dja@axtens.net>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/1e5c895669e8d54a7810b62dc61eb111f33c2c37.1615398265.git.christophe.leroy@csgroup.eu
call_do_irq() and call_do_softirq() are simple enough to be
worth inlining.
Inlining them avoids an mflr/mtlr pair plus a save/reload on stack.
This is inspired from S390 arch. Several other arches do more or
less the same. The way sparc arch does seems odd thought.
Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210320122227.345427-1-mpe@ellerman.id.au
Sparse reports the following problems:
arch/powerpc/math-emu/math.c:228:21: warning: Using plain integer as NULL pointer
arch/powerpc/math-emu/math.c:228:31: warning: Using plain integer as NULL pointer
arch/powerpc/math-emu/math.c:228:41: warning: Using plain integer as NULL pointer
arch/powerpc/math-emu/math.c:228:51: warning: Using plain integer as NULL pointer
arch/powerpc/math-emu/math.c:237:13: warning: incorrect type in initializer (different address spaces)
arch/powerpc/math-emu/math.c:237:13: expected unsigned int [noderef] __user *_gu_addr
arch/powerpc/math-emu/math.c:237:13: got unsigned int [usertype] *
arch/powerpc/math-emu/math.c:226:1: warning: symbol 'do_mathemu' was not declared. Should it be static?
Add missing __user qualifier when casting pointer used in get_user()
Use NULL instead of 0 to initialise opX local variables.
Add a prototype for do_mathemu() (Added in processor.h like sparc)
Reported-by: kernel test robot <lkp@intel.com>
Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/e4d1aae7604d89c98a52dfd8ce8443462e595670.1615809591.git.christophe.leroy@csgroup.eu
Unlike the other MSR_TM_* macros, MSR_TM_ACTIVE does not reference or
use its parameter unless CONFIG_PPC_TRANSACTIONAL_MEM is defined. This
causes an 'unused variable' compile warning unless the variable is also
guarded with CONFIG_PPC_TRANSACTIONAL_MEM.
Reference but do nothing with the argument in the macro to avoid a
potential compile warning.
Signed-off-by: Christopher M. Riedl <cmr@codefail.de>
Reviewed-by: Daniel Axtens <dja@axtens.net>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210227011259.11992-5-cmr@codefail.de
49a7d46a06 (powerpc: Implement smp_cond_load_relaxed()) added
busy-waiting pausing with a preferred SMT priority pattern, lowering
the priority (reducing decode cycles) during the whole loop slowpath.
However, data shows that while this pattern works well with simple
spinlocks, queued spinlocks benefit more being kept in medium priority,
with a cpu_relax() instead, being a low+medium combo on powerpc.
Data is from three benchmarks on a Power9: 9008-22L 64 CPUs with
2 sockets and 8 threads per core.
1. locktorture.
This is data for the lowest and most artificial/pathological level,
with increasing thread counts pounding on the lock. Metrics are total
ops/minute. Despite some small hits in the 4-8 range, scenarios are
either neutral or favorable to this patch.
+=========+==========+==========+=======+
| # tasks | vanilla | dirty | %diff |
+=========+==========+==========+=======+
| 2 | 46718565 | 48751350 | 4.35 |
+---------+----------+----------+-------+
| 4 | 51740198 | 50369082 | -2.65 |
+---------+----------+----------+-------+
| 8 | 63756510 | 62568821 | -1.86 |
+---------+----------+----------+-------+
| 16 | 67824531 | 70966546 | 4.63 |
+---------+----------+----------+-------+
| 32 | 53843519 | 61155508 | 13.58 |
+---------+----------+----------+-------+
| 64 | 53005778 | 53104412 | 0.18 |
+---------+----------+----------+-------+
| 128 | 53331980 | 54606910 | 2.39 |
+=========+==========+==========+=======+
2. sockperf (tcp throughput)
Here a client will do one-way throughput tests to a localhost server, with
increasing message sizes, dealing with the sk_lock. This patch shows to put
the performance of the qspinlock back to par with that of the simple lock:
simple-spinlock vanilla dirty
Hmean 14 73.50 ( 0.00%) 54.44 * -25.93%* 73.45 * -0.07%*
Hmean 100 654.47 ( 0.00%) 385.61 * -41.08%* 771.43 * 17.87%*
Hmean 300 2719.39 ( 0.00%) 2181.67 * -19.77%* 2666.50 * -1.94%*
Hmean 500 4400.59 ( 0.00%) 3390.77 * -22.95%* 4322.14 * -1.78%*
Hmean 850 6726.21 ( 0.00%) 5264.03 * -21.74%* 6863.12 * 2.04%*
3. dbench (tmpfs)
Configured to run with up to ncpusx8 clients, it shows both latency and
throughput metrics. For the latency, with the exception of the 64 case,
there is really nothing to go by:
vanilla dirty
Amean latency-1 1.67 ( 0.00%) 1.67 * 0.09%*
Amean latency-2 2.15 ( 0.00%) 2.08 * 3.36%*
Amean latency-4 2.50 ( 0.00%) 2.56 * -2.27%*
Amean latency-8 2.49 ( 0.00%) 2.48 * 0.31%*
Amean latency-16 2.69 ( 0.00%) 2.72 * -1.37%*
Amean latency-32 2.96 ( 0.00%) 3.04 * -2.60%*
Amean latency-64 7.78 ( 0.00%) 8.17 * -5.07%*
Amean latency-512 186.91 ( 0.00%) 186.41 * 0.27%*
For the dbench4 Throughput (misleading but traditional) there's a small
but rather constant improvement:
vanilla dirty
Hmean 1 849.13 ( 0.00%) 851.51 * 0.28%*
Hmean 2 1664.03 ( 0.00%) 1663.94 * -0.01%*
Hmean 4 3073.70 ( 0.00%) 3104.29 * 1.00%*
Hmean 8 5624.02 ( 0.00%) 5694.16 * 1.25%*
Hmean 16 9169.49 ( 0.00%) 9324.43 * 1.69%*
Hmean 32 11969.37 ( 0.00%) 12127.09 * 1.32%*
Hmean 64 15021.12 ( 0.00%) 15243.14 * 1.48%*
Hmean 512 14891.27 ( 0.00%) 15162.11 * 1.82%*
Measuring the dbench4 Per-VFS Operation latency, shows some very minor
differences within the noise level, around the 0-1% ranges.
Fixes: 49a7d46a06 ("powerpc: Implement smp_cond_load_relaxed()")
Acked-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Davidlohr Bueso <dbueso@suse.de>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20210318204702.71417-1-dave@stgolabs.net
