| Message ID | 1446019445-14421-1-git-send-email-huawei.libin@huawei.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Wed, 28 Oct 2015 16:04:05 +0800 Li Bin <huawei.libin@huawei.com> wrote: > By now, the recordmcount only records the function that in > following sections: > .text/.ref.text/.sched.text/.spinlock.text/.irqentry.text/ > .kprobes.text/.text.unlikely > > For the function that not in these sections, the call mcount > will be in place and not be replaced when kernel boot up. And > it will bring performance overhead, such as do_mem_abort (in > .exception.text section). This patch make the call mcount to > nop for this case in recordmcount. I can take this in my tree with an Ack from the ARM64 maintainers. -- Steve > > Signed-off-by: Li Bin <huawei.libin@huawei.com> > --- > scripts/recordmcount.c | 23 ++++++++++++++++++++++- > 1 file changed, 22 insertions(+), 1 deletion(-) > > diff --git a/scripts/recordmcount.c b/scripts/recordmcount.c > index 3d1984e..f697226 100644 > --- a/scripts/recordmcount.c > +++ b/scripts/recordmcount.c > @@ -42,6 +42,7 @@ > > #ifndef EM_AARCH64 > #define EM_AARCH64 183 > +#define R_AARCH64_NONE 0 > #define R_AARCH64_ABS64 257 > #endif > > @@ -160,6 +161,21 @@ static int make_nop_x86(void *map, size_t const offset) > return 0; > } > > +static unsigned char ideal_nop4_arm64[4] = {0x1f, 0x20, 0x03, 0xd5}; > +static int make_nop_arm64(void *map, size_t const offset) > +{ > + uint32_t *ptr; > + > + ptr = map + offset; > + if (*ptr != 0x94000000) > + return -1; > + > + /* Convert to nop */ > + ulseek(fd_map, offset, SEEK_SET); > + uwrite(fd_map, ideal_nop, 4); > + return 0; > +} > + > /* > * Get the whole file as a programming convenience in order to avoid > * malloc+lseek+read+free of many pieces. If successful, then mmap > @@ -353,7 +369,12 @@ do_file(char const *const fname) > altmcount = "__gnu_mcount_nc"; > break; > case EM_AARCH64: > - reltype = R_AARCH64_ABS64; gpfx = '_'; break; > + reltype = R_AARCH64_ABS64; > + make_nop = make_nop_arm64; > + rel_type_nop = R_AARCH64_NONE; > + ideal_nop = ideal_nop4_arm64; > + gpfx = '_'; > + break; > case EM_IA_64: reltype = R_IA64_IMM64; gpfx = '_'; break; > case EM_METAG: reltype = R_METAG_ADDR32; > altmcount = "_mcount_wrapper";
On Wed, Oct 28, 2015 at 04:16:19AM -0400, Steven Rostedt wrote: > On Wed, 28 Oct 2015 16:04:05 +0800 > Li Bin <huawei.libin@huawei.com> wrote: > > > By now, the recordmcount only records the function that in > > following sections: > > .text/.ref.text/.sched.text/.spinlock.text/.irqentry.text/ > > .kprobes.text/.text.unlikely > > > > For the function that not in these sections, the call mcount > > will be in place and not be replaced when kernel boot up. And > > it will bring performance overhead, such as do_mem_abort (in > > .exception.text section). This patch make the call mcount to > > nop for this case in recordmcount. > > I can take this in my tree with an Ack from the ARM64 maintainers. Looks like it does what it says on the tin: Acked-by: Will Deacon <will.deacon@arm.com> Will
On 10/28/2015 08:58 PM, Will Deacon wrote: > On Wed, Oct 28, 2015 at 04:16:19AM -0400, Steven Rostedt wrote: >> On Wed, 28 Oct 2015 16:04:05 +0800 >> Li Bin <huawei.libin@huawei.com> wrote: >> >>> By now, the recordmcount only records the function that in >>> following sections: >>> .text/.ref.text/.sched.text/.spinlock.text/.irqentry.text/ >>> .kprobes.text/.text.unlikely >>> >>> For the function that not in these sections, the call mcount >>> will be in place and not be replaced when kernel boot up. And >>> it will bring performance overhead, such as do_mem_abort (in >>> .exception.text section). This patch make the call mcount to >>> nop for this case in recordmcount. The patch itself is fine (it might be a good idea to describe what 0x94000000 means though), but I'm wondering whether we have any reason to exclude .exception.text from tracing. In my understandings, this attribute (.exception.text) is utilized solely for pretty printing in dump_backtrace(). Thanks, -Takahiro AKASHI >> I can take this in my tree with an Ack from the ARM64 maintainers. > > Looks like it does what it says on the tin: > > Acked-by: Will Deacon <will.deacon@arm.com> > > Will >
Hi Li,
[auto build test ERROR on arm64/for-next/core -- if it's inappropriate base, please suggest rules for selecting the more suitable base]
url: https://github.com/0day-ci/linux/commits/Li-Bin/recordmcount-arm64-replace-the-ignored-mcount-call-into-nop/20151028-160846
config: arm64-allmodconfig (attached as .config)
reproduce:
wget https://git.kernel.org/cgit/linux/kernel/git/wfg/lkp-tests.git/plain/sbin/make.cross -O ~/bin/make.cross
chmod +x ~/bin/make.cross
# save the attached .config to linux build tree
make.cross ARCH=arm64
All errors (new ones prefixed by >>):
>> /bin/bash: line 1: 117734 Segmentation fault ./scripts/recordmcount "arch/arm64/kernel/traps.o"
--
>> /bin/bash: line 1: 113824 Segmentation fault ./scripts/recordmcount "arch/arm64/mm/fault.o"
---
0-DAY kernel test infrastructure Open Source Technology Center
https://lists.01.org/pipermail/kbuild-all Intel Corporation
On Thu, Oct 29, 2015 at 09:35:42PM +0800, kbuild test robot wrote: > [auto build test ERROR on arm64/for-next/core -- if it's inappropriate base, please suggest rules for selecting the more suitable base] > > url: https://github.com/0day-ci/linux/commits/Li-Bin/recordmcount-arm64-replace-the-ignored-mcount-call-into-nop/20151028-160846 > config: arm64-allmodconfig (attached as .config) > reproduce: > wget https://git.kernel.org/cgit/linux/kernel/git/wfg/lkp-tests.git/plain/sbin/make.cross -O ~/bin/make.cross > chmod +x ~/bin/make.cross > # save the attached .config to linux build tree > make.cross ARCH=arm64 > > All errors (new ones prefixed by >>): > > >> /bin/bash: line 1: 117734 Segmentation fault ./scripts/recordmcount "arch/arm64/kernel/traps.o" > -- > >> /bin/bash: line 1: 113824 Segmentation fault ./scripts/recordmcount "arch/arm64/mm/fault.o" I can reproduce this locally with an allmodconfig build. The offset parameter to make_nop_arm64 is nuts: Program received signal SIGSEGV, Segmentation fault. make_nop_arm64 (map=0x7ffff7fef000, offset=2600547309829750784) at scripts/recordmcount.c:170 170 if (*ptr != 0x94000000) Both shdr->sh_offset (0x17000000000000) and relp->r_offset (0x2400000000000000) look dodgy in the caller. Li, can you take a look please? Will
? 2015/10/29 23:09, Will Deacon ??: > On Thu, Oct 29, 2015 at 09:35:42PM +0800, kbuild test robot wrote: >> [auto build test ERROR on arm64/for-next/core -- if it's inappropriate base, please suggest rules for selecting the more suitable base] >> >> url: https://github.com/0day-ci/linux/commits/Li-Bin/recordmcount-arm64-replace-the-ignored-mcount-call-into-nop/20151028-160846 >> config: arm64-allmodconfig (attached as .config) >> reproduce: >> wget https://git.kernel.org/cgit/linux/kernel/git/wfg/lkp-tests.git/plain/sbin/make.cross -O ~/bin/make.cross >> chmod +x ~/bin/make.cross >> # save the attached .config to linux build tree >> make.cross ARCH=arm64 >> >> All errors (new ones prefixed by >>): >> >>>> /bin/bash: line 1: 117734 Segmentation fault ./scripts/recordmcount "arch/arm64/kernel/traps.o" >> -- >>>> /bin/bash: line 1: 113824 Segmentation fault ./scripts/recordmcount "arch/arm64/mm/fault.o" > > I can reproduce this locally with an allmodconfig build. The offset > parameter to make_nop_arm64 is nuts: > > Program received signal SIGSEGV, Segmentation fault. > make_nop_arm64 (map=0x7ffff7fef000, offset=2600547309829750784) at scripts/recordmcount.c:170 > 170 if (*ptr != 0x94000000) > > Both shdr->sh_offset (0x17000000000000) and relp->r_offset (0x2400000000000000) > look dodgy in the caller. > The allmodconfig configs the CONFIG_CPU_BIG_ENDIAN, and the test system is little eddian, so it triggers the error. It is a bug in nop_mcount: ... if (mcountsym == Elf_r_sym(relp) && !is_fake_mcount(relp)) { if (make_nop) { ret = make_nop((void *)ehdr, shdr->sh_offset + relp->r_offset); } ... shdr->sh_offset and welp->r_offset should handle endianness properly as _w(shdr->sh_offset)/ _w(relp->r_offset). I will post a patchset including this patch soon. Thanks, Li Bin > Li, can you take a look please? > > Will > > . >
On Fri, 30 Oct 2015 14:00:28 +0800 libin <huawei.libin@huawei.com> wrote: > It is a bug in nop_mcount: > ... > if (mcountsym == Elf_r_sym(relp) && !is_fake_mcount(relp)) { > if (make_nop) { > ret = make_nop((void *)ehdr, shdr->sh_offset + relp->r_offset); > } > ... > > shdr->sh_offset and welp->r_offset should handle endianness properly as _w(shdr->sh_offset)/ > _w(relp->r_offset). I will post a patchset including this patch soon. > Has this been fixed yet? -- Steve
On Mon, Nov 02, 2015 at 01:26:08PM -0500, Steven Rostedt wrote: > On Fri, 30 Oct 2015 14:00:28 +0800 > libin <huawei.libin@huawei.com> wrote: > > > It is a bug in nop_mcount: > > ... > > if (mcountsym == Elf_r_sym(relp) && !is_fake_mcount(relp)) { > > if (make_nop) { > > ret = make_nop((void *)ehdr, shdr->sh_offset + relp->r_offset); > > } > > ... > > > > shdr->sh_offset and welp->r_offset should handle endianness properly as _w(shdr->sh_offset)/ > > _w(relp->r_offset). I will post a patchset including this patch soon. > > > > Has this been fixed yet? http://lkml.kernel.org/r/1446193864-24593-1-git-send-email-huawei.libin@huawei.com I see you're playing "chase the patches" at the moment ;) Will
On Mon, 2 Nov 2015 18:28:37 +0000 Will Deacon <will.deacon@arm.com> wrote: > I see you're playing "chase the patches" at the moment ;) > Why yes I am! Trying to grab last minute patches from my INBOX (or they could be old patches that I just missed), before kicking off my 12 hour test suite. -- Steve
diff --git a/scripts/recordmcount.c b/scripts/recordmcount.c index 3d1984e..f697226 100644 --- a/scripts/recordmcount.c +++ b/scripts/recordmcount.c @@ -42,6 +42,7 @@ #ifndef EM_AARCH64 #define EM_AARCH64 183 +#define R_AARCH64_NONE 0 #define R_AARCH64_ABS64 257 #endif @@ -160,6 +161,21 @@ static int make_nop_x86(void *map, size_t const offset) return 0; } +static unsigned char ideal_nop4_arm64[4] = {0x1f, 0x20, 0x03, 0xd5}; +static int make_nop_arm64(void *map, size_t const offset) +{ + uint32_t *ptr; + + ptr = map + offset; + if (*ptr != 0x94000000) + return -1; + + /* Convert to nop */ + ulseek(fd_map, offset, SEEK_SET); + uwrite(fd_map, ideal_nop, 4); + return 0; +} + /* * Get the whole file as a programming convenience in order to avoid * malloc+lseek+read+free of many pieces. If successful, then mmap @@ -353,7 +369,12 @@ do_file(char const *const fname) altmcount = "__gnu_mcount_nc"; break; case EM_AARCH64: - reltype = R_AARCH64_ABS64; gpfx = '_'; break; + reltype = R_AARCH64_ABS64; + make_nop = make_nop_arm64; + rel_type_nop = R_AARCH64_NONE; + ideal_nop = ideal_nop4_arm64; + gpfx = '_'; + break; case EM_IA_64: reltype = R_IA64_IMM64; gpfx = '_'; break; case EM_METAG: reltype = R_METAG_ADDR32; altmcount = "_mcount_wrapper";
By now, the recordmcount only records the function that in following sections: .text/.ref.text/.sched.text/.spinlock.text/.irqentry.text/ .kprobes.text/.text.unlikely For the function that not in these sections, the call mcount will be in place and not be replaced when kernel boot up. And it will bring performance overhead, such as do_mem_abort (in .exception.text section). This patch make the call mcount to nop for this case in recordmcount. Signed-off-by: Li Bin <huawei.libin@huawei.com> --- scripts/recordmcount.c | 23 ++++++++++++++++++++++- 1 file changed, 22 insertions(+), 1 deletion(-)