Re: [PATCH 2/4] tools headers x86: Sync msr-index.h with kernel sources

[Arnaldo Carvalho de Melo]

Em Fri, Aug 12, 2022 at 02:33:46PM +0530, Sandipan Das escreveu:
> Hi Jiri,
> 
> On 8/12/2022 2:03 PM, Jiri Olsa wrote:
> > On Thu, Aug 11, 2022 at 06:16:47PM +0530, Sandipan Das wrote:
> >> Sync msr-index.h with the kernel sources by adding the new AMD Last Branch
> >> Record Extension Version 2 (LbrExtV2) MSRs.
> >>
> >> Signed-off-by: Sandipan Das <sandipan.das@xxxxxxx>
> >> ---
> >>  tools/arch/x86/include/asm/msr-index.h | 5 +++++
> >>  1 file changed, 5 insertions(+)
> >>
> >> diff --git a/tools/arch/x86/include/asm/msr-index.h b/tools/arch/x86/include/asm/msr-index.h
> >> index cc615be27a54..7f9eaf497947 100644
> >> --- a/tools/arch/x86/include/asm/msr-index.h
> >> +++ b/tools/arch/x86/include/asm/msr-index.h
> >> @@ -574,6 +574,9 @@
> >>  #define MSR_AMD64_PERF_CNTR_GLOBAL_CTL		0xc0000301
> >>  #define MSR_AMD64_PERF_CNTR_GLOBAL_STATUS_CLR	0xc0000302
> >>  
> >> +/* AMD Last Branch Record MSRs */
> >> +#define MSR_AMD64_LBR_SELECT			0xc000010e
> > 
> > curious do we actualy use this in tools somewhere?
> > 
> > jirka
> > 
> 
> Commit 9dde6cadb92b ("tools arch x86: Sync the msr-index.h copy with the kernel sources")
> from Arnaldo says that adding these new MSR definitions in the tools headers allows the
> beautification scripts to pick up new entries which can be used for filtering MSR access
> traces.

Right, in this specific case that header should be moved to
tools/perf/trace/beauty/include/, as it is only used to harvest new MSRs
by:

⬢[acme@toolbox perf]$ tools/perf/trace/beauty/tracepoints/x86_msr.sh  | wc -l
328
⬢[acme@toolbox perf]$ tools/perf/trace/beauty/tracepoints/x86_msr.sh  | head
static const char *x86_MSRs[] = {
	[0x00000000] = "IA32_P5_MC_ADDR",
	[0x00000001] = "IA32_P5_MC_TYPE",
	[0x00000010] = "IA32_TSC",
	[0x00000017] = "IA32_PLATFORM_ID",
	[0x0000001b] = "IA32_APICBASE",
	[0x00000020] = "KNC_PERFCTR0",
	[0x00000021] = "KNC_PERFCTR1",
	[0x00000028] = "KNC_EVNTSEL0",
	[0x00000029] = "KNC_EVNTSEL1",
⬢[acme@toolbox perf]$

Other headers we have copies from the kernel may be used both for
harvesting non-enum enumerations like those to pretty print and use in
tracepoint filter expressions in 'perf trace', and in building perf on
older systems.

The preferred way is for kernel developers not to update the copies when
they add new stuff to the original file, so that perf developers have
the opportunity of checking if the new original file in the kernel don't
break the scripts in tools/perf/, or if the new additions are note
matched by regexps in the perf/tools/trace/beauty/ extraction scripts.

But kernel developers are welcome to update it iif they do these checks
themselves, at the very least do a 'make -C tools/perf' to check if it
builds before/after the update.

- Arnaldo
 
> E.g. one can trace the hardware LBR branch filter bits getting written to the LBR_SELECT
> MSR while recording branches.
> 
> $ perf record -j any,u true
> $ perf record -j any_call,u true
> 
> $ sudo perf trace -e msr:write_msr/max-stack=32/ --filter="msr == AMD64_LBR_SELECT"
> 
> [...]
> 224568.130 perf/9093 msr:write_msr(msr: AMD64_LBR_SELECT, val: 1)
>                                        do_trace_write_msr ([kernel.kallsyms])
>                                        do_trace_write_msr ([kernel.kallsyms])
>                                        native_write_msr ([kernel.kallsyms])
>                                        amd_pmu_lbr_enable_all ([kernel.kallsyms])
>                                        amd_pmu_v2_enable_all ([kernel.kallsyms])
>                                        x86_pmu_enable ([kernel.kallsyms])
>                                        ctx_resched ([kernel.kallsyms])
>                                        perf_event_exec ([kernel.kallsyms])
>                                        begin_new_exec ([kernel.kallsyms])
>                                        load_elf_binary ([kernel.kallsyms])
>                                        bprm_execve ([kernel.kallsyms])
>                                        do_execveat_common.isra.0 ([kernel.kallsyms])
>                                        __x64_sys_execve ([kernel.kallsyms])
>                                        do_syscall_64 ([kernel.kallsyms])
>                                        entry_SYSCALL_64 ([kernel.kallsyms])
>                                        execve (/usr/lib/x86_64-linux-gnu/libc.so.6)
> [...]
> 302748.439 perf/9126 msr:write_msr(msr: AMD64_LBR_SELECT, val: 229)
>                                        do_trace_write_msr ([kernel.kallsyms])
>                                        do_trace_write_msr ([kernel.kallsyms])
>                                        native_write_msr ([kernel.kallsyms])
>                                        amd_pmu_lbr_enable_all ([kernel.kallsyms])
>                                        amd_pmu_v2_enable_all ([kernel.kallsyms])
>                                        x86_pmu_enable ([kernel.kallsyms])
>                                        ctx_resched ([kernel.kallsyms])
>                                        perf_event_exec ([kernel.kallsyms])
>                                        begin_new_exec ([kernel.kallsyms])
>                                        load_elf_binary ([kernel.kallsyms])
>                                        bprm_execve ([kernel.kallsyms])
>                                        do_execveat_common.isra.0 ([kernel.kallsyms])
>                                        __x64_sys_execve ([kernel.kallsyms])
>                                        do_syscall_64 ([kernel.kallsyms])
>                                        entry_SYSCALL_64 ([kernel.kallsyms])
>                                        execve (/usr/lib/x86_64-linux-gnu/libc.so.6)
> [...]
> 
> I can add this example to the commit message in the next revision.
> 
> 
> - Sandipan

-- 

- Arnaldo