| Age | Commit message (Collapse) | Author |
|---|
|
commit 9a32a7e78bd0cd9a9b6332cbdc345ee5ffd0c5de upstream.
IBM Power9 processors can speculatively operate on data in the L1 cache before
it has been completely validated, via a way-prediction mechanism. It is not possible
for an attacker to determine the contents of impermissible memory using this method,
since these systems implement a combination of hardware and software security measures
to prevent scenarios where protected data could be leaked.
However these measures don't address the scenario where an attacker induces
the operating system to speculatively execute instructions using data that the
attacker controls. This can be used for example to speculatively bypass "kernel
user access prevention" techniques, as discovered by Anthony Steinhauser of
Google's Safeside Project. This is not an attack by itself, but there is a possibility
it could be used in conjunction with side-channels or other weaknesses in the
privileged code to construct an attack.
This issue can be mitigated by flushing the L1 cache between privilege boundaries
of concern. This patch flushes the L1 cache after user accesses.
This is part of the fix for CVE-2020-4788.
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Daniel Axtens <dja@axtens.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
|
|
commit 61e3acd8c693a14fc69b824cb5b08d02cb90a6e7 upstream.
The KUAP implementation adds calls in clear_user() to enable and
disable access to userspace memory. However, it doesn't add these to
__clear_user(), which is used in the ptrace regset code.
As there's only one direct user of __clear_user() (the regset code),
and the time taken to set the AMR for KUAP purposes is going to
dominate the cost of a quick access_ok(), there's not much point
having a separate path.
Rename __clear_user() to __arch_clear_user(), and make __clear_user()
just call clear_user().
Reported-by: syzbot+f25ecf4b2982d8c7a640@syzkaller-ppc64.appspotmail.com
Reported-by: Daniel Axtens <dja@axtens.net>
Suggested-by: Michael Ellerman <mpe@ellerman.id.au>
Fixes: de78a9c42a79 ("powerpc: Add a framework for Kernel Userspace Access Protection")
Signed-off-by: Andrew Donnellan <ajd@linux.ibm.com>
[mpe: Use __arch_clear_user() for the asm version like arm64 & nds32]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20191209132221.15328-1-ajd@linux.ibm.com
Signed-off-by: Daniel Axtens <dja@axtens.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
|
|
Backported from commit de78a9c42a79 ("powerpc: Add a framework
for Kernel Userspace Access Protection"). Here we don't try to
add the KUAP framework, we just want the helper functions
because we want to put uaccess flush helpers in them.
In terms of fixes, we don't need commit 1d8f739b07bd ("powerpc/kuap:
Fix set direction in allow/prevent_user_access()") as we don't have
real KUAP. Likewise as all our allows are noops and all our prevents
are just flushes, we don't need commit 9dc086f1e9ef ("powerpc/futex:
Fix incorrect user access blocking") The other 2 fixes we do need.
The original description is:
This patch implements a framework for Kernel Userspace Access
Protection.
Then subarches will have the possibility to provide their own
implementation by providing setup_kuap() and
allow/prevent_user_access().
Some platforms will need to know the area accessed and whether it is
accessed from read, write or both. Therefore source, destination and
size and handed over to the two functions.
mpe: Rename to allow/prevent rather than unlock/lock, and add
read/write wrappers. Drop the 32-bit code for now until we have an
implementation for it. Add kuap to pt_regs for 64-bit as well as
32-bit. Don't split strings, use pr_crit_ratelimited().
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Russell Currey <ruscur@russell.cc>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Signed-off-by: Daniel Axtens <dja@axtens.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
|
|
commit f79643787e0a0762d2409b7b8334e83f22d85695 upstream.
IBM Power9 processors can speculatively operate on data in the L1 cache before
it has been completely validated, via a way-prediction mechanism. It is not possible
for an attacker to determine the contents of impermissible memory using this method,
since these systems implement a combination of hardware and software security measures
to prevent scenarios where protected data could be leaked.
However these measures don't address the scenario where an attacker induces
the operating system to speculatively execute instructions using data that the
attacker controls. This can be used for example to speculatively bypass "kernel
user access prevention" techniques, as discovered by Anthony Steinhauser of
Google's Safeside Project. This is not an attack by itself, but there is a possibility
it could be used in conjunction with side-channels or other weaknesses in the
privileged code to construct an attack.
This issue can be mitigated by flushing the L1 cache between privilege boundaries
of concern. This patch flushes the L1 cache on kernel entry.
This is part of the fix for CVE-2020-4788.
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Daniel Axtens <dja@axtens.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
|
|
commit d9470757398a700d9450a43508000bcfd010c7a4 upstream.
Chandan reported that fstests' generic/026 test hit a crash:
BUG: Unable to handle kernel data access at 0xc00000062ac40000
Faulting instruction address: 0xc000000000092240
Oops: Kernel access of bad area, sig: 11 [#1]
LE SMP NR_CPUS=2048 DEBUG_PAGEALLOC NUMA pSeries
CPU: 0 PID: 27828 Comm: chacl Not tainted 5.0.0-rc2-next-20190115-00001-g6de6dba64dda #1
NIP: c000000000092240 LR: c00000000066a55c CTR: 0000000000000000
REGS: c00000062c0c3430 TRAP: 0300 Not tainted (5.0.0-rc2-next-20190115-00001-g6de6dba64dda)
MSR: 8000000002009033 <SF,VEC,EE,ME,IR,DR,RI,LE> CR: 44000842 XER: 20000000
CFAR: 00007fff7f3108ac DAR: c00000062ac40000 DSISR: 40000000 IRQMASK: 0
GPR00: 0000000000000000 c00000062c0c36c0 c0000000017f4c00 c00000000121a660
GPR04: c00000062ac3fff9 0000000000000004 0000000000000020 00000000275b19c4
GPR08: 000000000000000c 46494c4500000000 5347495f41434c5f c0000000026073a0
GPR12: 0000000000000000 c0000000027a0000 0000000000000000 0000000000000000
GPR16: 0000000000000000 0000000000000000 0000000000000000 0000000000000000
GPR20: c00000062ea70020 c00000062c0c38d0 0000000000000002 0000000000000002
GPR24: c00000062ac3ffe8 00000000275b19c4 0000000000000001 c00000062ac30000
GPR28: c00000062c0c38d0 c00000062ac30050 c00000062ac30058 0000000000000000
NIP memcmp+0x120/0x690
LR xfs_attr3_leaf_lookup_int+0x53c/0x5b0
Call Trace:
xfs_attr3_leaf_lookup_int+0x78/0x5b0 (unreliable)
xfs_da3_node_lookup_int+0x32c/0x5a0
xfs_attr_node_addname+0x170/0x6b0
xfs_attr_set+0x2ac/0x340
__xfs_set_acl+0xf0/0x230
xfs_set_acl+0xd0/0x160
set_posix_acl+0xc0/0x130
posix_acl_xattr_set+0x68/0x110
__vfs_setxattr+0xa4/0x110
__vfs_setxattr_noperm+0xac/0x240
vfs_setxattr+0x128/0x130
setxattr+0x248/0x600
path_setxattr+0x108/0x120
sys_setxattr+0x28/0x40
system_call+0x5c/0x70
Instruction dump:
7d201c28 7d402428 7c295040 38630008 38840008 408201f0 4200ffe8 2c050000
4182ff6c 20c50008 54c61838 7d201c28 <7d402428> 7d293436 7d4a3436 7c295040
The instruction dump decodes as:
subfic r6,r5,8
rlwinm r6,r6,3,0,28
ldbrx r9,0,r3
ldbrx r10,0,r4 <-
Which shows us doing an 8 byte load from c00000062ac3fff9, which
crosses the page boundary at c00000062ac40000 and faults.
It's not OK for memcmp to read past the end of the source or
destination buffers if that would cross a page boundary, because we
don't know that the next page is mapped.
As pointed out by Segher, we can read past the end of the source or
destination as long as we don't cross a 4K boundary, because that's
our minimum page size on all platforms.
The bug is in the code at the .Lcmp_rest_lt8bytes label. When we get
there we know that s1 is 8-byte aligned and we have at least 1 byte to
read, so a single 8-byte load won't read past the end of s1 and cross
a page boundary.
But we have to be more careful with s2. So check if it's within 8
bytes of a 4K boundary and if so go to the byte-by-byte loop.
Fixes: 2d9ee327adce ("powerpc/64: Align bytes before fall back to .Lshort in powerpc64 memcmp()")
Cc: stable@vger.kernel.org # v4.19+
Reported-by: Chandan Rajendra <chandan@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Reviewed-by: Segher Boessenkool <segher@kernel.crashing.org>
Tested-by: Chandan Rajendra <chandan@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
|
|
commit 76a5eaa38b15dda92cd6964248c39b5a6f3a4e9d upstream.
In order to protect against speculation attacks (Spectre
variant 2) on NXP PowerPC platforms, the branch predictor
should be flushed when the privillege level is changed.
This patch is adding the infrastructure to fixup at runtime
the code sections that are performing the branch predictor flush
depending on a boot arg parameter which is added later in a
separate patch.
Signed-off-by: Diana Craciun <diana.craciun@nxp.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
|
|
Commit 51c3c62b58b3 ("powerpc: Avoid code patching freed init
sections") accesses 'init_mem_is_free' flag too early, before the
kernel is relocated. This provokes early boot failure (before the
console is active).
As it is not necessary to do this verification that early, this
patch moves the test into patch_instruction() instead of
__patch_instruction().
This modification also has the advantage of avoiding unnecessary
remappings.
Fixes: 51c3c62b58b3 ("powerpc: Avoid code patching freed init sections")
Cc: stable@vger.kernel.org # 4.13+
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
On little endian platforms, csum_ipv6_magic() keeps len and proto in
CPU byte order. This generates a bad results leading to ICMPv6 packets
from other hosts being dropped by powerpc64le platforms.
In order to fix this, len and proto should be converted to network
byte order ie bigendian byte order. However checksumming 0x12345678
and 0x56341278 provide the exact same result so it is enough to
rotate the sum of len and proto by 1 byte.
PPC32 only support bigendian so the fix is needed for PPC64 only
Fixes: e9c4943a107b ("powerpc: Implement csum_ipv6_magic in assembly")
Reported-by: Jianlin Shi <jishi@redhat.com>
Reported-by: Xin Long <lucien.xin@gmail.com>
Cc: <stable@vger.kernel.org> # 4.18+
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Tested-by: Xin Long <lucien.xin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
This stops us from doing code patching in init sections after they've
been freed.
In this chain:
kvm_guest_init() ->
kvm_use_magic_page() ->
fault_in_pages_readable() ->
__get_user() ->
__get_user_nocheck() ->
barrier_nospec();
We have a code patching location at barrier_nospec() and
kvm_guest_init() is an init function. This whole chain gets inlined,
so when we free the init section (hence kvm_guest_init()), this code
goes away and hence should no longer be patched.
We seen this as userspace memory corruption when using a memory
checker while doing partition migration testing on powervm (this
starts the code patching post migration via
/sys/kernel/mobility/migration). In theory, it could also happen when
using /sys/kernel/debug/powerpc/barrier_nospec.
Cc: stable@vger.kernel.org # 4.13+
Signed-off-by: Michael Neuling <mikey@neuling.org>
Reviewed-by: Nicholas Piggin <npiggin@gmail.com>
Reviewed-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
The symbol memcpy_nocache_branch defined in order to allow patching
of memset function once cache is enabled leads to confusing reports
by perf tool.
Using the new patch_site functionality solves this issue.
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
In __copy_tofrom_user, if we encounter an exception on a store, we
stop copying and return the number of bytes not copied. However,
if the store is wider than one byte and is to an unaligned address,
it is possible that the store operand overlaps a page boundary
and the exception occurred on the latter part of the store operand,
meaning that it would be possible to copy a few more bytes. Since
copy_to_user is generally expected to copy as much as possible,
it would be better to copy those extra few bytes. This adds code
to do that. Since this edge case is not performance-critical,
the code has been written to be compact rather than as fast as
possible.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
The hand-coded assembler 64-bit copy routines include feature sections
that select one code path or another depending on which CPU we are
executing on. The self-tests for these copy routines end up testing
just one path. This adds a mechanism for selecting any desired code
path at compile time, and makes 2 or 3 versions of each test, each
using a different code path, so as to cover all the possible paths.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
[mpe: Add -mcpu=power4 to CFLAGS for older compilers]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
This aims to make the generation of exception table entries for the
loads and stores in __copy_tofrom_user_base clearer and easier to
verify. Instead of having a series of local labels on the loads and
stores, with a series of corresponding labels later for the exception
handlers, we now use macros to generate exception table entries at the
point of each load and store that could potentially trap. We do this
with the macros lex (load exception) and stex (store exception).
These macros are used right before the load or store to which they
apply.
Some complexity is introduced by the fact that we have some more work
to do after hitting an exception, because we need to calculate and
return the number of bytes not copied. The code uses r3 as the
current pointer into the destination buffer, that is, the address of
the first byte of the destination that has not been modified.
However, at various points in the copy loops, r3 can be 4, 8, 16 or 24
bytes behind that point.
To express this offset in an understandable way, we define a symbol
r3_offset which is updated at various points so that it equal to the
difference between the address of the first unmodified byte of the
destination and the value in r3. (In fact it only needs to be
accurate at the point of each lex or stex macro invocation.)
The rules for updating r3_offset are as follows:
* It starts out at 0
* An addi r3,r3,N instruction decreases r3_offset by N
* A store instruction (stb, sth, stw, std) to N(r3)
increases r3_offset by the width of the store (1, 2, 4, 8)
* A store with update instruction (stbu, sthu, stwu, stdu) to N(r3)
sets r3_offset to the width of the store.
There is some trickiness to the way that the lex and stex macros and
the associated exception handlers work. I would have liked to use
the current value of r3_offset in the name of the symbol used as
the exception handler, as in ".Lld_exc_$(r3_offset)" and then
have symbols .Lld_exc_0, .Lld_exc_8, .Lld_exc_16 etc. corresponding
to the offsets that needed to be added to r3. However, I couldn't
see a way to do that with gas.
Instead, the exception handler address is .Lld_exc - r3_offset or
.Lst_exc - r3_offset, that is, the distance ahead of .Lld_exc/.Lst_exc
that we start executing is equal to the amount that we need to add to
r3. This works because r3_offset is always a small multiple of 4,
and our instructions are 4 bytes long. This means that before
.Lld_exc and .Lst_exc, we have a sequence of instructions that
increments r3 by 4, 8, 16 or 24 depending on where we start. The
sequence increments r3 by 4 per instruction (on average).
We also replace the exception table for the 4k copy loop by a
macro per load or store. These loads and stores all use exactly
the same exception handler, which simply resets the argument registers
r3, r4 and r5 to there original values and re-does the whole copy
using the slower loop.
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
Add a macro and some helper C functions for patching single asm
instructions.
The gas macro means we can do something like:
1: nop
patch_site 1b, patch__foo
Which is less visually distracting than defining a GLOBAL symbol at 1,
and also doesn't pollute the symbol table which can confuse eg. perf.
These are obviously similar to our existing feature sections, but are
not automatically patched based on CPU/MMU features, rather they are
designed to be manually patched by C code at some arbitrary point.
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
Implement the barrier_nospec as a isync;sync instruction sequence.
The implementation uses the infrastructure built for BOOK3S 64.
Signed-off-by: Diana Craciun <diana.craciun@nxp.com>
[mpe: Split out of larger patch]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
Add a config symbol to encode which platforms support the
barrier_nospec speculation barrier. Currently this is just Book3S 64
but we will add Book3E in a future patch.
Signed-off-by: Diana Craciun <diana.craciun@nxp.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
The generic implementation of strlen() reads strings byte per byte.
This patch implements strlen() in assembly based on a read of entire
words, in the same spirit as what some other arches and glibc do.
On a 8xx the time spent in strlen is reduced by 3/4 for long strings.
strlen() selftest on an 8xx provides the following values:
Before the patch (ie with the generic strlen() in lib/string.c):
len 256 : time = 1.195055
len 016 : time = 0.083745
len 008 : time = 0.046828
len 004 : time = 0.028390
After the patch:
len 256 : time = 0.272185 ==> 78% improvment
len 016 : time = 0.040632 ==> 51% improvment
len 008 : time = 0.033060 ==> 29% improvment
len 004 : time = 0.029149 ==> 2% degradation
On a 832x:
Before the patch:
len 256 : time = 0.236125
len 016 : time = 0.018136
len 008 : time = 0.011000
len 004 : time = 0.007229
After the patch:
len 256 : time = 0.094950 ==> 60% improvment
len 016 : time = 0.013357 ==> 26% improvment
len 008 : time = 0.010586 ==> 4% improvment
len 004 : time = 0.008784
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
files not using feature fixup don't need asm/feature-fixups.h
files using feature fixup need asm/feature-fixups.h
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
Only include linux/stringify.h is files using __stringify()
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
This patch moves ASM_CONST() and stringify_in_c() into
dedicated asm-const.h, then cleans all related inclusions.
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
[mpe: asm-compat.h should include asm-const.h]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
This patch is based on the previous VMX patch on memcmp().
To optimize ppc64 memcmp() with VMX instruction, we need to think about
the VMX penalty brought with: If kernel uses VMX instruction, it needs
to save/restore current thread's VMX registers. There are 32 x 128 bits
VMX registers in PPC, which means 32 x 16 = 512 bytes for load and store.
The major concern regarding the memcmp() performance in kernel is KSM,
who will use memcmp() frequently to merge identical pages. So it will
make sense to take some measures/enhancement on KSM to see whether any
improvement can be done here. Cyril Bur indicates that the memcmp() for
KSM has a higher possibility to fail (unmatch) early in previous bytes
in following mail.
https://patchwork.ozlabs.org/patch/817322/#1773629
And I am taking a follow-up on this with this patch.
Per some testing, it shows KSM memcmp() will fail early at previous 32
bytes. More specifically:
- 76% cases will fail/unmatch before 16 bytes;
- 83% cases will fail/unmatch before 32 bytes;
- 84% cases will fail/unmatch before 64 bytes;
So 32 bytes looks a better choice than other bytes for pre-checking.
The early failure is also true for memcmp() for non-KSM case. With a
non-typical call load, it shows ~73% cases fail before first 32 bytes.
This patch adds a 32 bytes pre-checking firstly before jumping into VMX
operations, to avoid the unnecessary VMX penalty. It is not limited to
KSM case. And the testing shows ~20% improvement on memcmp() average
execution time with this patch.
And note the 32B pre-checking is only performed when the compare size
is long enough (>=4K currently) to allow VMX operation.
The detail data and analysis is at:
https://github.com/justdoitqd/publicFiles/blob/master/memcmp/README.md
Signed-off-by: Simon Guo <wei.guo.simon@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
This patch add VMX primitives to do memcmp() in case the compare size
is equal or greater than 4K bytes. KSM feature can benefit from this.
Test result with following test program(replace the "^>" with ""):
------
># cat tools/testing/selftests/powerpc/stringloops/memcmp.c
>#include <malloc.h>
>#include <stdlib.h>
>#include <string.h>
>#include <time.h>
>#include "utils.h"
>#define SIZE (1024 * 1024 * 900)
>#define ITERATIONS 40
int test_memcmp(const void *s1, const void *s2, size_t n);
static int testcase(void)
{
char *s1;
char *s2;
unsigned long i;
s1 = memalign(128, SIZE);
if (!s1) {
perror("memalign");
exit(1);
}
s2 = memalign(128, SIZE);
if (!s2) {
perror("memalign");
exit(1);
}
for (i = 0; i < SIZE; i++) {
s1[i] = i & 0xff;
s2[i] = i & 0xff;
}
for (i = 0; i < ITERATIONS; i++) {
int ret = test_memcmp(s1, s2, SIZE);
if (ret) {
printf("return %d at[%ld]! should have returned zero\n", ret, i);
abort();
}
}
return 0;
}
int main(void)
{
return test_harness(testcase, "memcmp");
}
------
Without this patch (but with the first patch "powerpc/64: Align bytes
before fall back to .Lshort in powerpc64 memcmp()." in the series):
4.726728762 seconds time elapsed ( +- 3.54%)
With VMX patch:
4.234335473 seconds time elapsed ( +- 2.63%)
There is ~+10% improvement.
Testing with unaligned and different offset version (make s1 and s2 shift
random offset within 16 bytes) can archieve higher improvement than 10%..
Signed-off-by: Simon Guo <wei.guo.simon@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
Currently memcmp() 64bytes version in powerpc will fall back to .Lshort
(compare per byte mode) if either src or dst address is not 8 bytes aligned.
It can be opmitized in 2 situations:
1) if both addresses are with the same offset with 8 bytes boundary:
memcmp() can compare the unaligned bytes within 8 bytes boundary firstly
and then compare the rest 8-bytes-aligned content with .Llong mode.
2) If src/dst addrs are not with the same offset of 8 bytes boundary:
memcmp() can align src addr with 8 bytes, increment dst addr accordingly,
then load src with aligned mode and load dst with unaligned mode.
This patch optmizes memcmp() behavior in the above 2 situations.
Tested with both little/big endian. Performance result below is based on
little endian.
Following is the test result with src/dst having the same offset case:
(a similar result was observed when src/dst having different offset):
(1) 256 bytes
Test with the existing tools/testing/selftests/powerpc/stringloops/memcmp:
- without patch
29.773018302 seconds time elapsed ( +- 0.09% )
- with patch
16.485568173 seconds time elapsed ( +- 0.02% )
-> There is ~+80% percent improvement
(2) 32 bytes
To observe performance impact on < 32 bytes, modify
tools/testing/selftests/powerpc/stringloops/memcmp.c with following:
-------
#include <string.h>
#include "utils.h"
-#define SIZE 256
+#define SIZE 32
#define ITERATIONS 10000
int test_memcmp(const void *s1, const void *s2, size_t n);
--------
- Without patch
0.244746482 seconds time elapsed ( +- 0.36%)
- with patch
0.215069477 seconds time elapsed ( +- 0.51%)
-> There is ～+13% improvement
(3) 0~8 bytes
To observe <8 bytes performance impact, modify
tools/testing/selftests/powerpc/stringloops/memcmp.c with following:
-------
#include <string.h>
#include "utils.h"
-#define SIZE 256
-#define ITERATIONS 10000
+#define SIZE 8
+#define ITERATIONS 1000000
int test_memcmp(const void *s1, const void *s2, size_t n);
-------
- Without patch
1.845642503 seconds time elapsed ( +- 0.12% )
- With patch
1.849767135 seconds time elapsed ( +- 0.26% )
-> They are nearly the same. (-0.2%)
Signed-off-by: Simon Guo <wei.guo.simon@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
The kmalloc() function has a 2-factor argument form, kmalloc_array(). This
patch replaces cases of:
kmalloc(a * b, gfp)
with:
kmalloc_array(a * b, gfp)
as well as handling cases of:
kmalloc(a * b * c, gfp)
with:
kmalloc(array3_size(a, b, c), gfp)
as it's slightly less ugly than:
kmalloc_array(array_size(a, b), c, gfp)
This does, however, attempt to ignore constant size factors like:
kmalloc(4 * 1024, gfp)
though any constants defined via macros get caught up in the conversion.
Any factors with a sizeof() of "unsigned char", "char", and "u8" were
dropped, since they're redundant.
The tools/ directory was manually excluded, since it has its own
implementation of kmalloc().
The Coccinelle script used for this was:
// Fix redundant parens around sizeof().
@@
type TYPE;
expression THING, E;
@@
(
kmalloc(
- (sizeof(TYPE)) * E
+ sizeof(TYPE) * E
, ...)
|
kmalloc(
- (sizeof(THING)) * E
+ sizeof(THING) * E
, ...)
)
// Drop single-byte sizes and redundant parens.
@@
expression COUNT;
typedef u8;
typedef __u8;
@@
(
kmalloc(
- sizeof(u8) * (COUNT)
+ COUNT
, ...)
|
kmalloc(
- sizeof(__u8) * (COUNT)
+ COUNT
, ...)
|
kmalloc(
- sizeof(char) * (COUNT)
+ COUNT
, ...)
|
kmalloc(
- sizeof(unsigned char) * (COUNT)
+ COUNT
, ...)
|
kmalloc(
- sizeof(u8) * COUNT
+ COUNT
, ...)
|
kmalloc(
- sizeof(__u8) * COUNT
+ COUNT
, ...)
|
kmalloc(
- sizeof(char) * COUNT
+ COUNT
, ...)
|
kmalloc(
- sizeof(unsigned char) * COUNT
+ COUNT
, ...)
)
// 2-factor product with sizeof(type/expression) and identifier or constant.
@@
type TYPE;
expression THING;
identifier COUNT_ID;
constant COUNT_CONST;
@@
(
- kmalloc
+ kmalloc_array
(
- sizeof(TYPE) * (COUNT_ID)
+ COUNT_ID, sizeof(TYPE)
, ...)
|
- kmalloc
+ kmalloc_array
(
- sizeof(TYPE) * COUNT_ID
+ COUNT_ID, sizeof(TYPE)
, ...)
|
- kmalloc
+ kmalloc_array
(
- sizeof(TYPE) * (COUNT_CONST)
+ COUNT_CONST, sizeof(TYPE)
, ...)
|
- kmalloc
+ kmalloc_array
(
- sizeof(TYPE) * COUNT_CONST
+ COUNT_CONST, sizeof(TYPE)
, ...)
|
- kmalloc
+ kmalloc_array
(
- sizeof(THING) * (COUNT_ID)
+ COUNT_ID, sizeof(THING)
, ...)
|
- kmalloc
+ kmalloc_array
(
- sizeof(THING) * COUNT_ID
+ COUNT_ID, sizeof(THING)
, ...)
|
- kmalloc
+ kmalloc_array
(
- sizeof(THING) * (COUNT_CONST)
+ COUNT_CONST, sizeof(THING)
, ...)
|
- kmalloc
+ kmalloc_array
(
- sizeof(THING) * COUNT_CONST
+ COUNT_CONST, sizeof(THING)
, ...)
)
// 2-factor product, only identifiers.
@@
identifier SIZE, COUNT;
@@
- kmalloc
+ kmalloc_array
(
- SIZE * COUNT
+ COUNT, SIZE
, ...)
// 3-factor product with 1 sizeof(type) or sizeof(expression), with
// redundant parens removed.
@@
expression THING;
identifier STRIDE, COUNT;
type TYPE;
@@
(
kmalloc(
- sizeof(TYPE) * (COUNT) * (STRIDE)
+ array3_size(COUNT, STRIDE, sizeof(TYPE))
, ...)
|
kmalloc(
- sizeof(TYPE) * (COUNT) * STRIDE
+ array3_size(COUNT, STRIDE, sizeof(TYPE))
, ...)
|
kmalloc(
- sizeof(TYPE) * COUNT * (STRIDE)
+ array3_size(COUNT, STRIDE, sizeof(TYPE))
, ...)
|
kmalloc(
- sizeof(TYPE) * COUNT * STRIDE
+ array3_size(COUNT, STRIDE, sizeof(TYPE))
, ...)
|
kmalloc(
- sizeof(THING) * (COUNT) * (STRIDE)
+ array3_size(COUNT, STRIDE, sizeof(THING))
, ...)
|
kmalloc(
- sizeof(THING) * (COUNT) * STRIDE
+ array3_size(COUNT, STRIDE, sizeof(THING))
, ...)
|
kmalloc(
- sizeof(THING) * COUNT * (STRIDE)
+ array3_size(COUNT, STRIDE, sizeof(THING))
, ...)
|
kmalloc(
- sizeof(THING) * COUNT * STRIDE
+ array3_size(COUNT, STRIDE, sizeof(THING))
, ...)
)
// 3-factor product with 2 sizeof(variable), with redundant parens removed.
@@
expression THING1, THING2;
identifier COUNT;
type TYPE1, TYPE2;
@@
(
kmalloc(
- sizeof(TYPE1) * sizeof(TYPE2) * COUNT
+ array3_size(COUNT, sizeof(TYPE1), sizeof(TYPE2))
, ...)
|
kmalloc(
- sizeof(TYPE1) * sizeof(THING2) * (COUNT)
+ array3_size(COUNT, sizeof(TYPE1), sizeof(TYPE2))
, ...)
|
kmalloc(
- sizeof(THING1) * sizeof(THING2) * COUNT
+ array3_size(COUNT, sizeof(THING1), sizeof(THING2))
, ...)
|
kmalloc(
- sizeof(THING1) * sizeof(THING2) * (COUNT)
+ array3_size(COUNT, sizeof(THING1), sizeof(THING2))
, ...)
|
kmalloc(
- sizeof(TYPE1) * sizeof(THING2) * COUNT
+ array3_size(COUNT, sizeof(TYPE1), sizeof(THING2))
, ...)
|
kmalloc(
- sizeof(TYPE1) * sizeof(THING2) * (COUNT)
+ array3_size(COUNT, sizeof(TYPE1), sizeof(THING2))
, ...)
)
// 3-factor product, only identifiers, with redundant parens removed.
@@
identifier STRIDE, SIZE, COUNT;
@@
(
kmalloc(
- (COUNT) * STRIDE * SIZE
+ array3_size(COUNT, STRIDE, SIZE)
, ...)
|
kmalloc(
- COUNT * (STRIDE) * SIZE
+ array3_size(COUNT, STRIDE, SIZE)
, ...)
|
kmalloc(
- COUNT * STRIDE * (SIZE)
+ array3_size(COUNT, STRIDE, SIZE)
, ...)
|
kmalloc(
- (COUNT) * (STRIDE) * SIZE
+ array3_size(COUNT, STRIDE, SIZE)
, ...)
|
kmalloc(
- COUNT * (STRIDE) * (SIZE)
+ array3_size(COUNT, STRIDE, SIZE)
, ...)
|
kmalloc(
- (COUNT) * STRIDE * (SIZE)
+ array3_size(COUNT, STRIDE, SIZE)
, ...)
|
kmalloc(
- (COUNT) * (STRIDE) * (SIZE)
+ array3_size(COUNT, STRIDE, SIZE)
, ...)
|
kmalloc(
- COUNT * STRIDE * SIZE
+ array3_size(COUNT, STRIDE, SIZE)
, ...)
)
// Any remaining multi-factor products, first at least 3-factor products,
// when they're not all constants...
@@
expression E1, E2, E3;
constant C1, C2, C3;
@@
(
kmalloc(C1 * C2 * C3, ...)
|
kmalloc(
- (E1) * E2 * E3
+ array3_size(E1, E2, E3)
, ...)
|
kmalloc(
- (E1) * (E2) * E3
+ array3_size(E1, E2, E3)
, ...)
|
kmalloc(
- (E1) * (E2) * (E3)
+ array3_size(E1, E2, E3)
, ...)
|
kmalloc(
- E1 * E2 * E3
+ array3_size(E1, E2, E3)
, ...)
)
// And then all remaining 2 factors products when they're not all constants,
// keeping sizeof() as the second factor argument.
@@
expression THING, E1, E2;
type TYPE;
constant C1, C2, C3;
@@
(
kmalloc(sizeof(THING) * C2, ...)
|
kmalloc(sizeof(TYPE) * C2, ...)
|
kmalloc(C1 * C2 * C3, ...)
|
kmalloc(C1 * C2, ...)
|
- kmalloc
+ kmalloc_array
(
- sizeof(TYPE) * (E2)
+ E2, sizeof(TYPE)
, ...)
|
- kmalloc
+ kmalloc_array
(
- sizeof(TYPE) * E2
+ E2, sizeof(TYPE)
, ...)
|
- kmalloc
+ kmalloc_array
(
- sizeof(THING) * (E2)
+ E2, sizeof(THING)
, ...)
|
- kmalloc
+ kmalloc_array
(
- sizeof(THING) * E2
+ E2, sizeof(THING)
, ...)
|
- kmalloc
+ kmalloc_array
(
- (E1) * E2
+ E1, E2
, ...)
|
- kmalloc
+ kmalloc_array
(
- (E1) * (E2)
+ E1, E2
, ...)
|
- kmalloc
+ kmalloc_array
(
- E1 * E2
+ E1, E2
, ...)
)
Signed-off-by: Kees Cook <keescook@chromium.org>
|
|
git://git.kernel.org/pub/scm/linux/kernel/git/powerpc/linux
Pull powerpc updates from Michael Ellerman:
"Notable changes:
- Support for split PMD page table lock on 64-bit Book3S (Power8/9).
- Add support for HAVE_RELIABLE_STACKTRACE, so we properly support
live patching again.
- Add support for patching barrier_nospec in copy_from_user() and
syscall entry.
- A couple of fixes for our data breakpoints on Book3S.
- A series from Nick optimising TLB/mm handling with the Radix MMU.
- Numerous small cleanups to squash sparse/gcc warnings from Mathieu
Malaterre.
- Several series optimising various parts of the 32-bit code from
Christophe Leroy.
- Removal of support for two old machines, "SBC834xE" and "C2K"
("GEFanuc,C2K"), which is why the diffstat has so many deletions.
And many other small improvements & fixes.
There's a few out-of-area changes. Some minor ftrace changes OK'ed by
Steve, and a fix to our powernv cpuidle driver. Then there's a series
touching mm, x86 and fs/proc/task_mmu.c, which cleans up some details
around pkey support. It was ack'ed/reviewed by Ingo & Dave and has
been in next for several weeks.
Thanks to: Akshay Adiga, Alastair D'Silva, Alexey Kardashevskiy, Al
Viro, Andrew Donnellan, Aneesh Kumar K.V, Anju T Sudhakar, Arnd
Bergmann, Balbir Singh, Cédric Le Goater, Christophe Leroy, Christophe
Lombard, Colin Ian King, Dave Hansen, Fabio Estevam, Finn Thain,
Frederic Barrat, Gautham R. Shenoy, Haren Myneni, Hari Bathini, Ingo
Molnar, Jonathan Neuschäfer, Josh Poimboeuf, Kamalesh Babulal,
Madhavan Srinivasan, Mahesh Salgaonkar, Mark Greer, Mathieu Malaterre,
Matthew Wilcox, Michael Neuling, Michal Suchanek, Naveen N. Rao,
Nicholas Piggin, Nicolai Stange, Olof Johansson, Paul Gortmaker, Paul
Mackerras, Peter Rosin, Pridhiviraj Paidipeddi, Ram Pai, Rashmica
Gupta, Ravi Bangoria, Russell Currey, Sam Bobroff, Samuel
Mendoza-Jonas, Segher Boessenkool, Shilpasri G Bhat, Simon Guo,
Souptick Joarder, Stewart Smith, Thiago Jung Bauermann, Torsten Duwe,
Vaibhav Jain, Wei Yongjun, Wolfram Sang, Yisheng Xie, YueHaibing"
* tag 'powerpc-4.18-1' of git://git.kernel.org/pub/scm/linux/kernel/git/powerpc/linux: (251 commits)
powerpc/64s/radix: Fix missing ptesync in flush_cache_vmap
cpuidle: powernv: Fix promotion from snooze if next state disabled
powerpc: fix build failure by disabling attribute-alias warning in pci_32
ocxl: Fix missing unlock on error in afu_ioctl_enable_p9_wait()
powerpc-opal: fix spelling mistake "Uniterrupted" -> "Uninterrupted"
powerpc: fix spelling mistake: "Usupported" -> "Unsupported"
powerpc/pkeys: Detach execute_only key on !PROT_EXEC
powerpc/powernv: copy/paste - Mask SO bit in CR
powerpc: Remove core support for Marvell mv64x60 hostbridges
powerpc/boot: Remove core support for Marvell mv64x60 hostbridges
powerpc/boot: Remove support for Marvell mv64x60 i2c controller
powerpc/boot: Remove support for Marvell MPSC serial controller
powerpc/embedded6xx: Remove C2K board support
powerpc/lib: optimise PPC32 memcmp
powerpc/lib: optimise 32 bits __clear_user()
powerpc/time: inline arch_vtime_task_switch()
powerpc/Makefile: set -mcpu=860 flag for the 8xx
powerpc: Implement csum_ipv6_magic in assembly
powerpc/32: Optimise __csum_partial()
powerpc/lib: Adjust .balign inside string functions for PPC32
...
|
|
At the time being, memcmp() compares two chunks of memory
byte per byte.
This patch optimises the comparison by comparing word by word.
On the same way as commit 15c2d45d17418 ("powerpc: Add 64bit
optimised memcmp"), this patch moves memcmp() into a dedicated
file named memcmp_32.S
A small benchmark performed on an 8xx comparing two chuncks
of 512 bytes performed 100000 times gives:
Before : 5852274 TB ticks
After: 1488638 TB ticks
This is almost 4 times faster
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
Rewrite clear_user() on the same principle as memset(0), making use
of dcbz to clear complete cache lines.
This code is a copy/paste of memset(), with some modifications
in order to retrieve remaining number of bytes to be cleared,
as it needs to be returned in case of error.
On the same way as done on PPC64 in commit 17968fbbd19f1
("powerpc: 64bit optimised __clear_user"), the patch moves
__clear_user() into a dedicated file string_32.S
On a MPC885, throughput is almost doubled:
Before:
~# dd if=/dev/zero of=/dev/null bs=1M count=1000
1048576000 bytes (1000.0MB) copied, 18.990779 seconds, 52.7MB/s
After:
~# dd if=/dev/zero of=/dev/null bs=1M count=1000
1048576000 bytes (1000.0MB) copied, 9.611468 seconds, 104.0MB/s
On a MPC8321, throughput is multiplied by 2.12:
Before:
root@vgoippro:~# dd if=/dev/zero of=/dev/null bs=1M count=1000
1048576000 bytes (1000.0MB) copied, 6.844352 seconds, 146.1MB/s
After:
root@vgoippro:~# dd if=/dev/zero of=/dev/null bs=1M count=1000
1048576000 bytes (1000.0MB) copied, 3.218854 seconds, 310.7MB/s
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
The generic csum_ipv6_magic() generates a pretty bad result
00000000 <csum_ipv6_magic>: (PPC32)
0: 81 23 00 00 lwz r9,0(r3)
4: 81 03 00 04 lwz r8,4(r3)
8: 7c e7 4a 14 add r7,r7,r9
c: 7d 29 38 10 subfc r9,r9,r7
10: 7d 4a 51 10 subfe r10,r10,r10
14: 7d 27 42 14 add r9,r7,r8
18: 7d 2a 48 50 subf r9,r10,r9
1c: 80 e3 00 08 lwz r7,8(r3)
20: 7d 08 48 10 subfc r8,r8,r9
24: 7d 4a 51 10 subfe r10,r10,r10
28: 7d 29 3a 14 add r9,r9,r7
2c: 81 03 00 0c lwz r8,12(r3)
30: 7d 2a 48 50 subf r9,r10,r9
34: 7c e7 48 10 subfc r7,r7,r9
38: 7d 4a 51 10 subfe r10,r10,r10
3c: 7d 29 42 14 add r9,r9,r8
40: 7d 2a 48 50 subf r9,r10,r9
44: 80 e4 00 00 lwz r7,0(r4)
48: 7d 08 48 10 subfc r8,r8,r9
4c: 7d 4a 51 10 subfe r10,r10,r10
50: 7d 29 3a 14 add r9,r9,r7
54: 7d 2a 48 50 subf r9,r10,r9
58: 81 04 00 04 lwz r8,4(r4)
5c: 7c e7 48 10 subfc r7,r7,r9
60: 7d 4a 51 10 subfe r10,r10,r10
64: 7d 29 42 14 add r9,r9,r8
68: 7d 2a 48 50 subf r9,r10,r9
6c: 80 e4 00 08 lwz r7,8(r4)
70: 7d 08 48 10 subfc r8,r8,r9
74: 7d 4a 51 10 subfe r10,r10,r10
78: 7d 29 3a 14 add r9,r9,r7
7c: 7d 2a 48 50 subf r9,r10,r9
80: 81 04 00 0c lwz r8,12(r4)
84: 7c e7 48 10 subfc r7,r7,r9
88: 7d 4a 51 10 subfe r10,r10,r10
8c: 7d 29 42 14 add r9,r9,r8
90: 7d 2a 48 50 subf r9,r10,r9
94: 7d 08 48 10 subfc r8,r8,r9
98: 7d 4a 51 10 subfe r10,r10,r10
9c: 7d 29 2a 14 add r9,r9,r5
a0: 7d 2a 48 50 subf r9,r10,r9
a4: 7c a5 48 10 subfc r5,r5,r9
a8: 7c 63 19 10 subfe r3,r3,r3
ac: 7d 29 32 14 add r9,r9,r6
b0: 7d 23 48 50 subf r9,r3,r9
b4: 7c c6 48 10 subfc r6,r6,r9
b8: 7c 63 19 10 subfe r3,r3,r3
bc: 7c 63 48 50 subf r3,r3,r9
c0: 54 6a 80 3e rotlwi r10,r3,16
c4: 7c 63 52 14 add r3,r3,r10
c8: 7c 63 18 f8 not r3,r3
cc: 54 63 84 3e rlwinm r3,r3,16,16,31
d0: 4e 80 00 20 blr
0000000000000000 <.csum_ipv6_magic>: (PPC64)
0: 81 23 00 00 lwz r9,0(r3)
4: 80 03 00 04 lwz r0,4(r3)
8: 81 63 00 08 lwz r11,8(r3)
c: 7c e7 4a 14 add r7,r7,r9
10: 7f 89 38 40 cmplw cr7,r9,r7
14: 7d 47 02 14 add r10,r7,r0
18: 7d 30 10 26 mfocrf r9,1
1c: 55 29 f7 fe rlwinm r9,r9,30,31,31
20: 7d 4a 4a 14 add r10,r10,r9
24: 7f 80 50 40 cmplw cr7,r0,r10
28: 7d 2a 5a 14 add r9,r10,r11
2c: 80 03 00 0c lwz r0,12(r3)
30: 81 44 00 00 lwz r10,0(r4)
34: 7d 10 10 26 mfocrf r8,1
38: 55 08 f7 fe rlwinm r8,r8,30,31,31
3c: 7d 29 42 14 add r9,r9,r8
40: 81 04 00 04 lwz r8,4(r4)
44: 7f 8b 48 40 cmplw cr7,r11,r9
48: 7d 29 02 14 add r9,r9,r0
4c: 7d 70 10 26 mfocrf r11,1
50: 55 6b f7 fe rlwinm r11,r11,30,31,31
54: 7d 29 5a 14 add r9,r9,r11
58: 7f 80 48 40 cmplw cr7,r0,r9
5c: 7d 29 52 14 add r9,r9,r10
60: 7c 10 10 26 mfocrf r0,1
64: 54 00 f7 fe rlwinm r0,r0,30,31,31
68: 7d 69 02 14 add r11,r9,r0
6c: 7f 8a 58 40 cmplw cr7,r10,r11
70: 7c 0b 42 14 add r0,r11,r8
74: 81 44 00 08 lwz r10,8(r4)
78: 7c f0 10 26 mfocrf r7,1
7c: 54 e7 f7 fe rlwinm r7,r7,30,31,31
80: 7c 00 3a 14 add r0,r0,r7
84: 7f 88 00 40 cmplw cr7,r8,r0
88: 7d 20 52 14 add r9,r0,r10
8c: 80 04 00 0c lwz r0,12(r4)
90: 7d 70 10 26 mfocrf r11,1
94: 55 6b f7 fe rlwinm r11,r11,30,31,31
98: 7d 29 5a 14 add r9,r9,r11
9c: 7f 8a 48 40 cmplw cr7,r10,r9
a0: 7d 29 02 14 add r9,r9,r0
a4: 7d 70 10 26 mfocrf r11,1
a8: 55 6b f7 fe rlwinm r11,r11,30,31,31
ac: 7d 29 5a 14 add r9,r9,r11
b0: 7f 80 48 40 cmplw cr7,r0,r9
b4: 7d 29 2a 14 add r9,r9,r5
b8: 7c 10 10 26 mfocrf r0,1
bc: 54 00 f7 fe rlwinm r0,r0,30,31,31
c0: 7d 29 02 14 add r9,r9,r0
c4: 7f 85 48 40 cmplw cr7,r5,r9
c8: 7c 09 32 14 add r0,r9,r6
cc: 7d 50 10 26 mfocrf r10,1
d0: 55 4a f7 fe rlwinm r10,r10,30,31,31
d4: 7c 00 52 14 add r0,r0,r10
d8: 7f 80 30 40 cmplw cr7,r0,r6
dc: 7d 30 10 26 mfocrf r9,1
e0: 55 29 ef fe rlwinm r9,r9,29,31,31
e4: 7c 09 02 14 add r0,r9,r0
e8: 54 03 80 3e rotlwi r3,r0,16
ec: 7c 03 02 14 add r0,r3,r0
f0: 7c 03 00 f8 not r3,r0
f4: 78 63 84 22 rldicl r3,r3,48,48
f8: 4e 80 00 20 blr
This patch implements it in assembly for both PPC32 and PPC64
Link: https://github.com/linuxppc/linux/issues/9
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Reviewed-by: Segher Boessenkool <segher@kernel.crashing.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
Improve __csum_partial by interleaving loads and adds.
On a 8xx, it brings neither improvement nor degradation.
On a 83xx, it brings a 25% improvement.
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Reviewed-by: Segher Boessenkool <segher@kernel.crashing.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
commit 87a156fb18fe1 ("Align hot loops of some string functions")
degraded the performance of string functions by adding useless
nops
A simple benchmark on an 8xx calling 100000x a memchr() that
matches the first byte runs in 41668 TB ticks before this patch
and in 35986 TB ticks after this patch. So this gives an
improvement of approx 10%
Another benchmark doing the same with a memchr() matching the 128th
byte runs in 1011365 TB ticks before this patch and 1005682 TB ticks
after this patch, so regardless on the number of loops, removing
those useless nops improves the test by 5683 TB ticks.
Fixes: 87a156fb18fe1 ("Align hot loops of some string functions")
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
emulate_step() tests are failing if VSX is not supported or disabled.
emulate_step_test: lxvd2x : FAIL
emulate_step_test: stxvd2x : FAIL
If !CPU_FTR_VSX, emulate_step() failure is expected and testcase should
PASS with a valid justification. After patch:
emulate_step_test: lxvd2x : PASS (!CPU_FTR_VSX)
emulate_step_test: stxvd2x : PASS (!CPU_FTR_VSX)
Signed-off-by: Ravi Bangoria <ravi.bangoria@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
emulate_step() is not checking runtime VSX feature flag before
emulating an instruction. This is causing kernel crash when kernel
is compiled with CONFIG_VSX=y but running on a machine where VSX
is not supported or disabled. Ex, while running emulate_step tests
on P6 machine:
Oops: Exception in kernel mode, sig: 4 [#1]
NIP [c000000000095c24] .load_vsrn+0x28/0x54
LR [c000000000094bdc] .emulate_loadstore+0x167c/0x17b0
Call Trace:
0x40fe240c7ae147ae (unreliable)
.emulate_loadstore+0x167c/0x17b0
.emulate_step+0x25c/0x5bc
.test_lxvd2x_stxvd2x+0x64/0x154
.test_emulate_step+0x38/0x4c
.do_one_initcall+0x5c/0x2c0
.kernel_init_freeable+0x314/0x4cc
.kernel_init+0x24/0x160
.ret_from_kernel_thread+0x58/0xb4
With fix:
emulate_step_test: lxvd2x : FAIL
emulate_step_test: stxvd2x : FAIL
Reported-by: Michael Ellerman <mpe@ellerman.id.au>
Signed-off-by: Ravi Bangoria <ravi.bangoria@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
Replace 'op->type & INSTR_TYPE_MASK' expression with GETTYPE(op->type)
macro.
Signed-off-by: Ravi Bangoria <ravi.bangoria@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
Note that unlike RFI which is patched only in kernel the nospec state
reflects settings at the time the module was loaded.
Iterating all modules and re-patching every time the settings change
is not implemented.
Based on lwsync patching.
Signed-off-by: Michal Suchanek <msuchanek@suse.de>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
Based on the RFI patching. This is required to be able to disable the
speculation barrier.
Only one barrier type is supported and it does nothing when the
firmware does not enable it. Also re-patching modules is not supported
So the only meaningful thing that can be done is patching out the
speculation barrier at boot when the user says it is not wanted.
Signed-off-by: Michal Suchanek <msuchanek@suse.de>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
Some functions prototypes were missing for the non-altivec code. Add the
missing prototypes in a new header file, fix warnings treated as errors
with W=1:
arch/powerpc/lib/xor_vmx_glue.c:18:6: error: no previous prototype for ‘xor_altivec_2’ [-Werror=missing-prototypes]
arch/powerpc/lib/xor_vmx_glue.c:29:6: error: no previous prototype for ‘xor_altivec_3’ [-Werror=missing-prototypes]
arch/powerpc/lib/xor_vmx_glue.c:40:6: error: no previous prototype for ‘xor_altivec_4’ [-Werror=missing-prototypes]
arch/powerpc/lib/xor_vmx_glue.c:52:6: error: no previous prototype for ‘xor_altivec_5’ [-Werror=missing-prototypes]
The prototypes were already present in <asm/xor.h> but this header file is
meant to be included after <include/linux/raid/xor.h>. Trying to re-use
<asm/xor.h> directly would lead to warnings such as:
arch/powerpc/include/asm/xor.h:39:15: error: variable ‘xor_block_altivec’ has initializer but incomplete type
Trying to re-use <asm/xor.h> after <include/linux/raid/xor.h> in
xor_vmx_glue.c would in turn trigger the following warnings:
include/asm-generic/xor.h:688:34: error: ‘xor_block_32regs’ defined but not used [-Werror=unused-variable]
Signed-off-by: Mathieu Malaterre <malat@debian.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
On some CPUs we can prevent a vulnerability related to store-to-load
forwarding by preventing store forwarding between privilege domains,
by inserting a barrier in kernel entry and exit paths.
This is known to be the case on at least Power7, Power8 and Power9
powerpc CPUs.
Barriers must be inserted generally before the first load after moving
to a higher privilege, and after the last store before moving to a
lower privilege, HV and PR privilege transitions must be protected.
Barriers are added as patch sections, with all kernel/hypervisor entry
points patched, and the exit points to lower privilge levels patched
similarly to the RFI flush patching.
Firmware advertisement is not implemented yet, so CPU flush types
are hard coded.
Thanks to Michal Suchánek for bug fixes and review.
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Mauricio Faria de Oliveira <mauricfo@linux.vnet.ibm.com>
Signed-off-by: Michael Neuling <mikey@neuling.org>
Signed-off-by: Michal Suchánek <msuchanek@suse.de>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
My powerpc-linux-gnu-gcc v4.4.5 compiler can't build a 32-bit kernel
any more:
arch/powerpc/lib/sstep.c: In function 'do_popcnt':
arch/powerpc/lib/sstep.c:1068: error: integer constant is too large for 'long' type
arch/powerpc/lib/sstep.c:1069: error: integer constant is too large for 'long' type
arch/powerpc/lib/sstep.c:1069: error: integer constant is too large for 'long' type
arch/powerpc/lib/sstep.c:1070: error: integer constant is too large for 'long' type
arch/powerpc/lib/sstep.c:1079: error: integer constant is too large for 'long' type
arch/powerpc/lib/sstep.c: In function 'do_prty':
arch/powerpc/lib/sstep.c:1117: error: integer constant is too large for 'long' type
This file gets compiled with -std=gnu89 which means a constant can be
given the type 'long' even if it won't fit. Fix the errors with a 'ULL'
suffix on the relevant constants.
Fixes: 2c979c489fee ("powerpc/lib/sstep: Add prty instruction emulation")
Fixes: dcbd19b48d31 ("powerpc/lib/sstep: Add popcnt instruction emulation")
Signed-off-by: Finn Thain <fthain@telegraphics.com.au>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
Add a test of the relative branch patching logic in the alternate
section feature fixup code. This tests that if we branch past the last
instruction of the alternate section, the branch is not patched.
That's because the assembler will have created a branch that already
points to the first instruction after the patched section, which is
correct and needs no further patching.
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
We want this to remain the last test (because it's disabled by
default), so give it a non-numbered name so we don't have to renumber
it when adding new tests before it.
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
The code patching code has always been a bit confused about whether
it's best to use void *, unsigned int *, char *, etc. to point to
instructions. In fact in the feature fixups tests we use both unsigned
int[] and u8[] in different places.
Unfortunately the tests that use unsigned int[] calculate the size of
the code blocks using subtraction of those unsigned int pointers, and
then pass the result to memcmp(). This means we're only comparing 1/4
of the bytes we need to, because we need to multiply by
sizeof(unsigned int) to get the number of *bytes*.
The result is that the tests do all the patching and then only compare
some of the resulting code, so patching bugs that only effect that
last 3/4 of the code could slip through undetected. It turns out that
hasn't been happening, although one test had a bad expected case (see
previous commit).
Fix it for now by multiplying the size by 4 in the affected functions.
Fixes: 362e7701fd18 ("powerpc: Add self-tests of the feature fixup code")
Epic-brown-paper-bag-by: Michael Ellerman <mpe@ellerman.id.au>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
The expected case for this test was wrong, the source of the alternate
code sequence is:
FTR_SECTION_ELSE
2: or 2,2,2
PPC_LCMPI r3,1
beq 3f
blt 2b
b 3f
b 1b
ALT_FTR_SECTION_END(0, 1)
3: or 1,1,1
or 2,2,2
4: or 3,3,3
So when it's patched the '3' label should still be on the 'or 1,1,1',
and the 4 label is irrelevant and can be removed.
Fixes: 362e7701fd18 ("powerpc: Add self-tests of the feature fixup code")
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
When we patch an alternate feature section, we have to adjust any
relative branches that branch out of the alternate section.
But currently we have a bug if we have a branch that points to past
the last instruction of the alternate section, eg:
FTR_SECTION_ELSE
1: b 2f
or 6,6,6
2:
ALT_FTR_SECTION_END(...)
nop
This will result in a relative branch at 1 with a target that equals
the end of the alternate section.
That branch does not need adjusting when it's moved to the non-else
location. Currently we do adjust it, resulting in a branch that goes
off into the link-time location of the else section, which is junk.
The fix is to not patch branches that have a target == end of the
alternate section.
Fixes: d20fe50a7b3c ("KVM: PPC: Book3S HV: Branch inside feature section")
Fixes: 9b1a735de64c ("powerpc: Add logic to patch alternative feature sections")
Cc: stable@vger.kernel.org # v2.6.27+
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
Rather than override the machine type in .S code (which can hide wrong
or ambiguous code generation for the target), set the type to power4
for all assembly.
This also means we need to be careful not to build power4-only code
when we're not building for Book3S, such as the "power7" versions of
copyuser/page/memcpy.
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
[mpe: Fix Book3E build, don't build the "power7" variants for non-Book3S]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
Currently the rfi-flush messages print 'Using <type> flush' for all
enabled_flush_types, but that is not necessarily true -- as now the
fallback flush is always enabled on pseries, but the fixup function
overwrites its nop/branch slot with other flush types, if available.
So, replace the 'Using <type> flush' messages with '<type> flush is
available'.
Also, print the patched flush types in the fixup function, so users
can know what is (not) being used (e.g., the slower, fallback flush,
or no flush type at all if flush is disabled via the debugfs switch).
Suggested-by: Michael Ellerman <mpe@ellerman.id.au>
Signed-off-by: Mauricio Faria de Oliveira <mauricfo@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
The inline keyword was not at the beginning of the function declaration.
Fix the following warning (treated as error in W=1):
arch/powerpc/lib/sstep.c:283:1: error: ‘inline’ is not at beginning of declaration
static int nokprobe_inline copy_mem_in(u8 *dest, unsigned long ea, int nb,
arch/powerpc/lib/sstep.c:388:1: error: ‘inline’ is not at beginning of declaration
static int nokprobe_inline copy_mem_out(u8 *dest, unsigned long ea, int nb,
Signed-off-by: Mathieu Malaterre <malat@debian.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
Merge our fixes branch from the 4.15 cycle.
Unusually the fixes branch saw some significant features merged,
notably the RFI flush patches, so we want the code in next to be
tested against that, to avoid any surprises when the two are merged.
There's also some other work on the panic handling that was reverted
in fixes and we now want to do properly in next, which would conflict.
And we also fix a few other minor merge conflicts.
|
|
feature fixups need to use patch_instruction() early in the boot,
even before the code is relocated to its final address, requiring
patch_instruction() to use PTRRELOC() in order to address data.
But feature fixups applies on code before it is set to read only,
even for modules. Therefore, feature fixups can use
raw_patch_instruction() instead.
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
patch_instruction() uses almost the same sequence as
__patch_instruction()
This patch refactor it so that patch_instruction() uses
__patch_instruction() instead of duplicating code.
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
Acked-by: Balbir Singh <bsingharora@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
|
On some CPUs we can prevent the Meltdown vulnerability by flushing the
L1-D cache on exit from kernel to user mode, and from hypervisor to
guest.
This is known to be the case on at least Power7, Power8 and Power9. At
this time we do not know the status of the vulnerability on other CPUs
such as the 970 (Apple G5), pasemi CPUs (AmigaOne X1000) or Freescale
CPUs. As more information comes to light we can enable this, or other
mechanisms on those CPUs.
The vulnerability occurs when the load of an architecturally
inaccessible memory region (eg. userspace load of kernel memory) is
speculatively executed to the point where its result can influence the
address of a subsequent speculatively executed load.
In order for that to happen, the first load must hit in the L1,
because before the load is sent to the L2 the permission check is
performed. Therefore if no kernel addresses hit in the L1 the
vulnerability can not occur. We can ensure that is the case by
flushing the L1 whenever we return to userspace. Similarly for
hypervisor vs guest.
In order to flush the L1-D cache on exit, we add a section of nops at
each (h)rfi location that returns to a lower privileged context, and
patch that with some sequence. Newer firmwares are able to advertise
to us that there is a special nop instruction that flushes the L1-D.
If we do not see that advertised, we fall back to doing a displacement
flush in software.
For guest kernels we support migration between some CPU versions, and
different CPUs may use different flush instructions. So that we are
prepared to migrate to a machine with a different flush instruction
activated, we may have to patch more than one flush instruction at
boot if the hypervisor tells us to.
In the end this patch is mostly the work of Nicholas Piggin and
Michael Ellerman. However a cast of thousands contributed to analysis
of the issue, earlier versions of the patch, back ports testing etc.
Many thanks to all of them.
Tested-by: Jon Masters <jcm@redhat.com>
Signed-off-by: Nicholas Piggin <npiggin@gmail.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
|
Nicholas Piggin
Andrew Donnellan
Christophe Leroy
Michael Ellerman
Diana Craciun
Michael Neuling
Paul Mackerras
Simon Guo
Kees Cook
Linus Torvalds
Ravi Bangoria
Michal Suchanek
Mathieu Malaterre
Finn Thain
Mauricio Faria de Oliveira