Re: [PATCH 4/6] trace: trace syscall in its handler not from ptrace handler

[Vaibhav Nagarnaik]

On Wed, Mar 28, 2012 at 11:20 PM, Ingo Molnar <mingo@xxxxxxxxxx> wrote:
>
> * Vaibhav Nagarnaik <vnagarnaik@xxxxxxxxxx> wrote:
>
>> On Wed, Mar 28, 2012 at 7:43 PM, H. Peter Anvin <hpa@xxxxxxxxx> wrote:
>> > But instead you add a penalty for every syscall, even if tracing is
>> > disabled.  Not cool.
>>
>> I just ran a small test binary which calls syscall(SYS_getuid) in a
>> tight loop and calculates the latency per syscall.
>>
>> Without my patch: it is 70 ns/call
>> With my patch: it is 83 ns/call
>>
>> So yes, it does add a bit of latency to the syscall even if
>> tracing is disabled. I wonder if I can change the redirection
>> function so that it doesn't add so much latency.
>
> There's a really simple rule for anything tracing/debugging
> related: and syscalls don't add *ANY* kind of latency to the
> non-tracing case. That is true of the current syscall tracing
> bits, they work via a TIF flag and don't add any latency.
>

Thanks for the feedback. I had missed this added latency due to this
patch when tracing is disabled.

To fix that, instead of a TIF flag, I am using a flag in the
current->trace bitmap. I check that flag before jumping to the tracing
function. That reduces the latency from 83 ns/call to 74 ns/call.


Thanks

Vaibhav Nagarnaik
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/