Re: [RFC 3/4] trace_uprobe: Support SDT markers having semaphore

From: Ravi Bangoria
Date: Wed Mar 07 2018 - 03:44:07 EST




On 03/06/2018 05:29 PM, Peter Zijlstra wrote:
> On Wed, Feb 28, 2018 at 01:23:44PM +0530, Ravi Bangoria wrote:
>> Userspace Statically Defined Tracepoints[1] are dtrace style markers
>> inside userspace applications. These markers are added by developer at
>> important places in the code. Each marker source expands to a single
>> nop instruction in the compiled code but there may be additional
>> overhead for computing the marker arguments which expands to couple of
>> instructions. If this computaion is quite more, execution of it can be
>> ommited by runtime if() condition when no one is tracing on the marker:
>>
>> if (semaphore > 0) {
>> Execute marker instructions;
>> }
>>
>> Default value of semaphore is 0. Tracer has to increment the semaphore
>> before recording on a marker and decrement it at the end of tracing.
>>
>> Implement the semaphore flip logic in trace_uprobe, leaving core uprobe
>> infrastructure as is, except one new callback from uprobe_mmap() to
>> trace_uprobe.
> W.T.H. is that called a semaphore? afaict its just a usage-counter.

I totally agree with you. But it's not me who named it semaphore :)

Please refer to "Semaphore Handling" section at:
https://sourceware.org/systemtap/wiki/UserSpaceProbeImplementation

We can surly name it differently in the kernel code and document it
properly in the Documents/tracing/

> There is no blocking, no releasing, nothing that would make it an actual
> semaphore.
>
> So please, remove all mention of semaphore from this code, because it,
> most emphatically, is not one.
>
> Also, would it not be much better to do userspace jump-labels for this?
> That completely avoids the dynamic branch at the SDT site.
>

Userspace jump-label is a good idea but...

Semaphore logic has already became a kinda ABI now. Tools like bcc,
gdb, systemtap etc. flip the semaphore while probing the marker.

Thanks,
Ravi