[RFCv1 4/4] perf: arm_spe: Dynamically switch PID tracing to contextidr
James Clark
james.clark at arm.com
Fri Oct 22 09:23:23 PDT 2021
On 22/10/2021 16:36, James Clark wrote:
>
>
> On 21/10/2021 14:45, Leo Yan wrote:
>> Now Arm64 provides API for enabling and disable PID tracing, Arm SPE
>> driver invokes these functions to dynamically enable it during
>> profiling when the program runs in root PID name space, and disable PID
>> tracing when the perf event is stopped.
>>
>> Device drivers should not depend on CONFIG_PID_IN_CONTEXTIDR for PID
>> tracing, so this patch uses the consistent condition for setting bit
>> EL1_CX for PMSCR.
>
> Hi Leo,
>
> I've been testing this change, but I'm seeing something strange. Not sure
> if it's a problem on my side or not yet. With this command:
>
> sudo ./perf record -vvv -e arm_spe//u -- taskset --cpu-list 1 bash -c ls
>
> I'm only seeing 0 values for context:
>
> sudo ./perf report -D | grep CONTEXT
>
> . 00038dce: 65 00 00 00 00 CONTEXT 0x0 el2
> . 00038e0e: 65 00 00 00 00 CONTEXT 0x0 el2
>
> I added a printk to the function, and I see it print non zero values, although
> there are some zero ones mixed in there too:
>
> diff --git a/arch/arm64/include/asm/mmu_context.h b/arch/arm64/include/asm/mmu_context.h
> index 0c1669db19a1..8f0fb43a5fac 100644
> --- a/arch/arm64/include/asm/mmu_context.h
> +++ b/arch/arm64/include/asm/mmu_context.h
> @@ -33,7 +33,8 @@ static inline void contextidr_thread_switch(struct task_struct *next)
> if (!static_branch_unlikely(&contextidr_in_use))
> return;
>
> - write_sysreg(task_pid_nr(next), contextidr_el1);
> + printk("Set %d\n", task_pid_nr(next));
> + write_sysreg(task_pid_nr(next), contextidr_el2);
> isb();
> }
>
>
> Results in this:
>
> [ 53.257905] Set 77
> [ 53.257909] Set 0
> [ 53.258180] Set 77
> [ 53.258183] Set 0
> [ 53.258385] Set 309
> [ 53.258385] Set 172
> [ 53.258425] Set 77
> [ 53.258443] Set 990
> [ 53.258449] Set 77
> [ 53.258455] Set 990
> [ 53.258467] Set 310
> [ 53.258719] Set 7
> [ 53.258728] Set 77
> [ 53.258731] Set 0
> [ 53.258733] Set 0
> [ 53.258738] Set 7
>
>
> Without your patchset I don't get 0 values in the SPE trace anymore:
>
> . 0000050e: 65 b1 01 00 00 CONTEXT 0x1b1 el2
> . 0000054e: 65 b1 01 00 00 CONTEXT 0x1b1 el2
> . 0000058e: 65 ac 01 00 00 CONTEXT 0x1ac el2
> . 000005ce: 65 ac 01 00 00 CONTEXT 0x1ac el2
>
Is it an issue with building with CONTEXTIDR disabled? Seems like this change results
in context packets set to 0 when it's disabled rather than having the packets disabled
like they used to be:
zcat /proc/config.gz | grep CONTEXTIDR
# CONFIG_PID_IN_CONTEXTIDR is not set
sudo ./perf report -D | grep CONTEXT
. 00045b4e: 65 00 00 00 00 CONTEXT 0x0 el2
When I build with CONFIG_PID_IN_CONTEXTIDR=y the contexts are non zero so it seems to
be working that way. But ./perf record -e arm_spe//u -a does have context IDs even
when CONFIG_PID_IN_CONTEXTIDR=n. So I'm still a bit confused.
More information about the linux-arm-kernel
mailing list