Re: [PATCH v7 00/12] unwind_deferred: Implement sframe handling

[Steven Rostedt <rostedt@goodmis.org>]

On Wed, 2 Jul 2025 14:57:22 +0200
Jens Remus <jremus@linux.ibm.com> wrote:

> Hello Steve!
> 
> On 01.07.2025 20:49, Steven Rostedt wrote:
> > This code is based on top of:
> > 
> >  https://lore.kernel.org/linux-trace-kernel/20250701005321.942306427@goodmis.org/
> >  git://git.kernel.org/pub/scm/linux/kernel/git/trace/linux-trace.git unwind/core
> > 
> > This is the implementation of parsing the SFrame section in an ELF file.  
> 
> ...
> 
> > The code for this patch series can be found here:
> > 
> >   git://git.kernel.org/pub/scm/linux/kernel/git/trace/linux-trace.git unwind/sframe  
> 
> Wouldn't it make sense to include your related perf (tools) series [1]
> in that branch to ease testing?  Provided you also include the minor
> fix [2] to make perf tools work. :-)

I was planning on making a branch with perf and sframe merged (haven't
pushed it out yet). Since sframe doesn't technically rely on it (I use
tracing too test it too) I'm keeping them separate.

> 
> Additionally it would make sense to include the patches from Josh that
> add SFrame information to the vDSO on x86 [3].
> 
> [1]: [PATCH v12 00/11] perf: Support the deferred unwinding infrastructure,
> https://lore.kernel.org/linux-trace-kernel/20250701180410.755491417@goodmis.org/
> 
> [2]: https://lore.kernel.org/linux-trace-kernel/51903e66-56bc-42a4-b80c-9c3223e2a48a@linux.ibm.com/
> 
> [3]: [PATCH v6 0/6] x86/vdso: VDSO updates and fixes for sframes,
> https://lore.kernel.org/all/20250425023750.669174660@goodmis.org/

I have them in separate branches too and will be posting them
separately. Again, the "merged" branch will contain them all.

> 
> > Changes since v6: https://lore.kernel.org/linux-trace-kernel/20250617225009.233007152@goodmis.org/  
> 
> > - Moved the addition of the prctl(), that allows libraries to add the elf
> >   sections to the kernel, to the last patch and labeled it as "DO NOT APPLY".
> >   This should instead be a proper system call and work to make it robust and
> >   flexible still needs to be done. The prctl() patch is added for debugging
> >   purposes only.  
> 
> Does PR_SET_VMA [4] create a precedent case for the SFrame prctls?
> 
> [4]: https://man7.org/linux/man-pages/man2/pr_set_vma.2const.html
> 

At the last sframe meeting we came to the decision to make it a proper
system call. Just seems cleaner that way. If others feel prctl() is the
way to go, we can definitely do that too.

-- Steve