Re: [PATCH v6 0/5] Add ftrace direct call for arm64

[Mark Rutland]

On Tue, Apr 11, 2023 at 06:54:24PM +0100, Will Deacon wrote:
> On Tue, Apr 11, 2023 at 01:44:56PM -0400, Steven Rostedt wrote:
> > On Tue, 11 Apr 2023 18:08:08 +0100
> > Will Deacon <will@xxxxxxxxxx> wrote:
> > 
> > > On Tue, Apr 11, 2023 at 12:47:49PM -0400, Steven Rostedt wrote:
> > > > On Tue, 11 Apr 2023 16:56:45 +0100
> > > > Mark Rutland <mark.rutland@xxxxxxx> wrote:
> > > >   
> > > > > IIUC Steve was hoping to take the FUNCTION_GRAPH_RETVAL series through the
> > > > > trace tree, and if that's still the plan, maybe both should go that way?  
> > > > 
> > > > The conflict is minor, and I think I prefer to still have the ARM64 bits go
> > > > through the arm64 tree, as it will get better testing, and I don't like to
> > > > merge branches ;-)
> > > > 
> > > > I've added Linus to the Cc so he knows that there will be conflicts, but as
> > > > long as we mention it in our pull request, with a branch that includes the
> > > > solution, it should be fine going through two different trees.  
> > > 
> > > If it's just the simple asm-offsets conflict that Mark mentioned, then that
> > > sounds fine to me. However, patches 3-5 don't seem to have anything to do
> > 
> > I guess 3 and 5 are not, but patch 4 adds arm64 code to the samples (as
> > it requires arch specific asm to handle the direct trampolines).
> 
> Sorry, yes, I was thinking of arch/arm64/ and then failed spectacularly
> at communicating :)
> 
> > > with arm64 at all and I'd prefer those to go via other trees (esp. as patch
> > > 3 is an independent -stable candidate and the last one is a bpf selftest
> > > change which conflicts in -next).
> > > 
> > > So I'll queue the first two in arm64 on a branch (or-next/ftrace) based
> > > on trace-direct-v6.3-rc3.
> > 
> > Are 3-5 dependent on those changes? If not, I can pull them into my tree.
> 
> Good question. Florent?

Patch 3 (the fix to the ftrace test) does not depend upon patches 1 and 2. It
probably would've been better to queue that as a preparatory fix before the
other changes.

Patch 4 (adding arm64 support to the samples) depends on patch 3. The arm64
parts depends upon patch 1 to be selectable, and without patch 1 the samples
will behave the same as before. It could be queued independently of patch 1,
but won't have any effect until merged with patch 1.

Patch 5 (the bpf selftest list changes) depends on patch 1 alone.

Perhaps we could queue 1 and 2 via the arm64 tree, 3 and 4 via the ftrace tree,
and follow up with patch 5 via the bpf tree after -rc1?

Thanks,
Mark.