From: duchangbin <changbin.du@huawei.com>
To: Adrian Hunter <adrian.hunter@intel.com>
Cc: duchangbin <changbin.du@huawei.com>,
Peter Zijlstra <peterz@infradead.org>,
Ingo Molnar <mingo@redhat.com>,
"Arnaldo Carvalho de Melo" <acme@kernel.org>,
Namhyung Kim <namhyung@kernel.org>,
Nathan Chancellor <nathan@kernel.org>,
Mark Rutland <mark.rutland@arm.com>,
Alexander Shishkin <alexander.shishkin@linux.intel.com>,
Jiri Olsa <jolsa@kernel.org>, "Ian Rogers" <irogers@google.com>,
"Liang, Kan" <kan.liang@linux.intel.com>,
"Nick Desaulniers" <ndesaulniers@google.com>,
Bill Wendling <morbo@google.com>,
Justin Stitt <justinstitt@google.com>,
"linux-perf-users@vger.kernel.org"
<linux-perf-users@vger.kernel.org>,
"linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>,
"llvm@lists.linux.dev" <llvm@lists.linux.dev>,
"Wanghui (OS Kernel Lab, Beijing)" <hw.huiwang@huawei.com>
Subject: Re: [PATCH v5 1/8] perf: support specify vdso path in cmdline
Date: Tue, 23 Jul 2024 01:50:24 +0000 [thread overview]
Message-ID: <04f65d430ccc4f88bd8c65e0adb463ad@huawei.com> (raw)
In-Reply-To: <02930c55-6423-4db4-b6f3-b783c572c121@intel.com>
On Mon, Jul 22, 2024 at 04:15:56PM +0300, Adrian Hunter wrote:
> On 22/07/24 15:19, duchangbin wrote:
> > On Thu, Jul 18, 2024 at 08:02:16PM +0300, Adrian Hunter wrote:
> >> On 2/07/24 07:18, Changbin Du wrote:
> >>> The vdso dumped from process memory (in buildid-cache) lacks debugging
> >>> info. To annotate vdso symbols with source lines we need specify a
> >>> debugging version.
> >>>
> >>> For x86, we can find them from your local build as
> >>> arch/x86/entry/vdso/vdso{32,64}.so.dbg. Or they may reside in
> >>> /lib/modules/<version>/vdso/vdso{32,64}.so on Ubuntu. But notice that
> >>> the buildid has to match.
> >>>
> >>> $ sudo perf record -a
> >>> $ sudo perf report --objdump=llvm-objdump \
> >>> --vdso arch/x86/entry/vdso/vdso64.so.dbg,arch/x86/entry/vdso/vdso32.so.dbg
> >>>
> >>> Samples: 17K of event 'cycles:P', 4000 Hz, Event count (approx.): 1760
> >>> __vdso_clock_gettime /work/linux-host/arch/x86/entry/vdso/vdso64.so.d
> >>> Percent│ movq -48(%rbp),%rsi
> >>> │ testq %rax,%rax
> >>> │ ; return vread_hvclock();
> >>> │ movq %rax,%rdx
> >>> │ ; if (unlikely(!vdso_cycles_ok(cycles)))
> >>> │ ↑ js eb
> >>> │ ↑ jmp 74
> >>> │ ; ts->tv_sec = vdso_ts->sec;
> >>> 0.02 │147: leaq 2(%rbx),%rax
> >>> │ shlq $4, %rax
> >>> │ addq %r10,%rax
> >>> │ ; while ((seq = READ_ONCE(vd->seq)) & 1) {
> >>> 9.38 │152: movl (%r10),%ecx
> >>>
> >>> When doing cross platform analysis, we also need specify the vdso path if
> >>> we are interested in its symbols.
> >>>
> >>
> >> Just realized this is about objdump. Sorry for not getting that
> >> earlier.
> >>
> >> Like perf tools, objdump follows the paradigm of attempting to
> >> locate and use debug info transparently. However, objdump looks
> >> at the installed debug info in /usr/lib/debug/.build-id/
> >>
> >> For example, if the debug file is copied (or linked) there, then
> >> objdump or llvm-objdump will find it:
> >>
> >> $ llvm-objdump-18 -dS ~/.debug/\[vdso\]/cf702469f4637840fd6ba1a8d8a628ff83253d04/vdso | head -20
> >>
> >> ~/.debug/[vdso]/cf702469f4637840fd6ba1a8d8a628ff83253d04/vdso: file format elf64-x86-64
> >>
> >> Disassembly of section .text:
> >>
> >> 00000000000006d0 <.text>:
> >> 6d0: 48 8d 3d 29 d9 ff ff leaq -0x26d7(%rip), %rdi # 0xffffffffffffe000
> >> 6d7: eb 19 jmp 0x6f2 <.text+0x22>
> >> 6d9: 4c 8b 0d 28 d9 ff ff movq -0x26d8(%rip), %r9 # 0xffffffffffffe008
> >> 6e0: 4c 8b 05 29 d9 ff ff movq -0x26d7(%rip), %r8 # 0xffffffffffffe010
> >> 6e7: 0f 01 f9 rdtscp
> >> 6ea: 66 90 nop
> >> 6ec: 8b 0f movl (%rdi), %ecx
> >> 6ee: 39 ce cmpl %ecx, %esi
> >> 6f0: 74 0e je 0x700 <.text+0x30>
> >> 6f2: 8b 37 movl (%rdi), %esi
> >> 6f4: 85 f6 testl %esi, %esi
> >> 6f6: 75 e1 jne 0x6d9 <.text+0x9>
> >> 6f8: 48 c7 c0 ff ff ff ff movq $-0x1, %rax
> >> 6ff: c3 retq
> >> $ sudo ln -s /lib/modules/6.9.2-local/build/arch/x86/entry/vdso/vdso64.so.dbg /usr/lib/debug/.build-id/cf/702469f4637840fd6ba1a8d8a628ff83253d04.debug
> >> $ llvm-objdump-18 -dS ~/.debug/\[vdso\]/cf702469f4637840fd6ba1a8d8a628ff83253d04/vdso | head -20
> >>
> >> ~/.debug/[vdso]/cf702469f4637840fd6ba1a8d8a628ff83253d04/vdso: file format elf64-x86-64
> >>
> >> Disassembly of section .text:
> >>
> >> 00000000000006d0 <vread_hvclock>:
> >> ; hv_read_tsc_page_tsc(const struct ms_hyperv_tsc_page *tsc_pg,
> >> 6d0: 48 8d 3d 29 d9 ff ff leaq -0x26d7(%rip), %rdi # 0xffffffffffffe000 <hvclock_page>
> >> 6d7: eb 19 jmp 0x6f2 <vread_hvclock+0x22>
> >> ; scale = READ_ONCE(tsc_pg->tsc_scale);
> >> 6d9: 4c 8b 0d 28 d9 ff ff movq -0x26d8(%rip), %r9 # 0xffffffffffffe008 <hvclock_page+0x8>
> >> ; offset = READ_ONCE(tsc_pg->tsc_offset);
> >> 6e0: 4c 8b 05 29 d9 ff ff movq -0x26d7(%rip), %r8 # 0xffffffffffffe010 <hvclock_page+0x10>
> >> ; asm volatile(ALTERNATIVE_2("rdtsc",
> >> 6e7: 0f 31 rdtsc
> >> 6e9: 90 nop
> >> 6ea: 90 nop
> >> 6eb: 90 nop
> >> ; } while (READ_ONCE(tsc_pg->tsc_sequence) != sequence);
> >> 6ec: 8b 0f movl (%rdi), %ecx
> >> $
> >>
> >> Would that solve your problem?
> >>
> >> Notably, later versions of llvm-objdump have an option
> >> --debug-file-directory which makes it possible to
> >> have the debug info in a directory owned by the user,
> >> for example:
> >>
> >> $ sudo rm /usr/lib/debug/.build-id/cf/702469f4637840fd6ba1a8d8a628ff83253d04.debug
> >> $ llvm-objdump-18 -dS ~/.debug/\[vdso\]/cf702469f4637840fd6ba1a8d8a628ff83253d04/vdso | head -20
> >>
> >> ~/.debug/[vdso]/cf702469f4637840fd6ba1a8d8a628ff83253d04/vdso: file format elf64-x86-64
> >>
> >> Disassembly of section .text:
> >>
> >> 00000000000006d0 <.text>:
> >> 6d0: 48 8d 3d 29 d9 ff ff leaq -0x26d7(%rip), %rdi # 0xffffffffffffe000
> >> 6d7: eb 19 jmp 0x6f2 <.text+0x22>
> >> 6d9: 4c 8b 0d 28 d9 ff ff movq -0x26d8(%rip), %r9 # 0xffffffffffffe008
> >> 6e0: 4c 8b 05 29 d9 ff ff movq -0x26d7(%rip), %r8 # 0xffffffffffffe010
> >> 6e7: 0f 01 f9 rdtscp
> >> 6ea: 66 90 nop
> >> 6ec: 8b 0f movl (%rdi), %ecx
> >> 6ee: 39 ce cmpl %ecx, %esi
> >> 6f0: 74 0e je 0x700 <.text+0x30>
> >> 6f2: 8b 37 movl (%rdi), %esi
> >> 6f4: 85 f6 testl %esi, %esi
> >> 6f6: 75 e1 jne 0x6d9 <.text+0x9>
> >> 6f8: 48 c7 c0 ff ff ff ff movq $-0x1, %rax
> >> 6ff: c3 retq
> >> $ mkdir -p /tmp/debug/.build-id/cf/
> >> $ ln -s /lib/modules/6.9.2-local/build/arch/x86/entry/vdso/vdso64.so.dbg /tmp/debug/.build-id/cf/702469f4637840fd6ba1a8d8a628ff83253d04.debug
> >> $ llvm-objdump-18 --debug-file-directory /tmp/debug -dS ~/.debug/\[vdso\]/cf702469f4637840fd6ba1a8d8a628ff83253d04/vdso | head -20
> >>
> >> ~/.debug/[vdso]/cf702469f4637840fd6ba1a8d8a628ff83253d04/vdso: file format elf64-x86-64
> >>
> >> Disassembly of section .text:
> >>
> >> 00000000000006d0 <vread_hvclock>:
> >> ; hv_read_tsc_page_tsc(const struct ms_hyperv_tsc_page *tsc_pg,
> >> 6d0: 48 8d 3d 29 d9 ff ff leaq -0x26d7(%rip), %rdi # 0xffffffffffffe000 <hvclock_page>
> >> 6d7: eb 19 jmp 0x6f2 <vread_hvclock+0x22>
> >> ; scale = READ_ONCE(tsc_pg->tsc_scale);
> >> 6d9: 4c 8b 0d 28 d9 ff ff movq -0x26d8(%rip), %r9 # 0xffffffffffffe008 <hvclock_page+0x8>
> >> ; offset = READ_ONCE(tsc_pg->tsc_offset);
> >> 6e0: 4c 8b 05 29 d9 ff ff movq -0x26d7(%rip), %r8 # 0xffffffffffffe010 <hvclock_page+0x10>
> >> ; asm volatile(ALTERNATIVE_2("rdtsc",
> >> 6e7: 0f 31 rdtsc
> >> 6e9: 90 nop
> >> 6ea: 90 nop
> >> 6eb: 90 nop
> >> ; } while (READ_ONCE(tsc_pg->tsc_sequence) != sequence);
> >> 6ec: 8b 0f movl (%rdi), %ecx
> >> $
> >>
> > Yes, not only llvm-objdump. I can see other tools also can handle this and even
> > can download from online by Debuginfod.
> > - perf can search debugging info there if HAVE_DEBUGINFOD_SUPPORT.
> > - gdb can search debugging info there and download via Debuginfod if not exist.
> >
> > While above requires debugging files are placed correctly in
> > /usr/lib/debug/.build-id/. For vdso mostly you need to copy it manully.
> >
> > The target of this series is to handle it transparently, espacially for locally
> > built kernels.
>
> Passing the file name to perf tool doesn't seem transparent.
>
This is an alternative way to find debugging file if the file is not in
pre-defined paths. This is how we handle vmlinux with '--vmlinux' option.
> For locally built kernels, maybe there should be a make target for it.
> There seems to be "vdso_install" for the stripped binary.
>
I don't know whether installing debugging files to /usr/lib/debug/.build-id/ is
a good scheme. For me, I really do not want too many debugging files occupy my
disk space. (If so, why only install vdso there, right?)
>
>
--
Cheers,
Changbin Du
next prev parent reply other threads:[~2024-07-23 1:50 UTC|newest]
Thread overview: 21+ messages / expand[flat|nested] mbox.gz Atom feed top
2024-07-02 4:18 [PATCH v5 0/8] perf: support specify vdso path in cmdline Changbin Du
2024-07-02 4:18 ` [PATCH v5 1/8] " Changbin Du
2024-07-18 17:02 ` Adrian Hunter
2024-07-22 12:19 ` duchangbin
2024-07-22 13:15 ` Adrian Hunter
2024-07-23 1:50 ` duchangbin [this message]
2024-07-02 4:18 ` [PATCH v5 2/8] perf: disasm: refactor function dso__disassemble_filename Changbin Du
2024-07-02 4:18 ` [PATCH v5 3/8] perf: disasm: use build_id_path if fallback failed Changbin Du
2024-07-02 4:18 ` [PATCH v5 4/8] perf: build-id: name debugging vdso as "debug" Changbin Du
2024-07-18 17:03 ` Adrian Hunter
2024-07-22 12:08 ` duchangbin
2024-07-22 13:21 ` Adrian Hunter
2024-07-23 1:51 ` duchangbin
2024-07-02 4:18 ` [PATCH v5 5/8] perf: symbol: generalize vmlinux path searching Changbin Du
2024-07-02 4:18 ` [PATCH v5 6/8] perf: build-id: try to search debugging vdso and add to cache Changbin Du
2024-07-02 4:18 ` [PATCH v5 7/8] perf: disasm: prefer debugging files in build-id cache Changbin Du
2024-07-02 4:18 ` [PATCH v5 8/8] perf buildid-cache: recognize vdso when adding files Changbin Du
2024-07-03 23:18 ` Namhyung Kim
2024-07-04 1:43 ` duchangbin
2024-07-24 14:55 ` kernel test robot
2024-07-15 11:38 ` [PATCH v5 0/8] perf: support specify vdso path in cmdline duchangbin
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=04f65d430ccc4f88bd8c65e0adb463ad@huawei.com \
--to=changbin.du@huawei.com \
--cc=acme@kernel.org \
--cc=adrian.hunter@intel.com \
--cc=alexander.shishkin@linux.intel.com \
--cc=hw.huiwang@huawei.com \
--cc=irogers@google.com \
--cc=jolsa@kernel.org \
--cc=justinstitt@google.com \
--cc=kan.liang@linux.intel.com \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-perf-users@vger.kernel.org \
--cc=llvm@lists.linux.dev \
--cc=mark.rutland@arm.com \
--cc=mingo@redhat.com \
--cc=morbo@google.com \
--cc=namhyung@kernel.org \
--cc=nathan@kernel.org \
--cc=ndesaulniers@google.com \
--cc=peterz@infradead.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox