Re: [PATCH] alpha: add simd_enabled argument to perf_reg_validate()

[Thomas Weißschuh]

On Thu, Oct 22, 2026 at 07:53:22AM +0800, Mi, Dapeng wrote:
> On 10/21/2026 11:38 PM, Peter Zijlstra wrote:
> > On Wed, Oct 21, 2026 at 01:49:35PM +0200, Thomas Weißschuh wrote:
> >> On Wed, Oct 21, 2026 at 12:23:55PM +0200, Magnus Lindholm wrote:
> >>> On Wed, Oct 21, 2026 at 10:18 AM Thomas Weißschuh
> >>> <thomas.weissschuh@xxxxxxxxxxxxx> wrote:
> >> (...)
> >>
> >>> Thanks for catching this. I reproduced the build failure on current
> >>> mainline and confirmed your fix with GENERIC SMP and DP264 UP builds.
> >>> Both also passed perf register validation and sampling tests in QEMU.
> >>>
> >>> Reviewed-by: Magnus Lindholm <linmag7@xxxxxxxxx>
> >>> Tested-by: Magnus Lindholm <linmag7@xxxxxxxxx>
> >> Thanks.
> >>
> >>> If you have a route for this patch upstream, please go ahead and take it,
> >>> unless Linus would prefer to pick it up directly.
> >> I have no route upstream. I hope for Linus to pick it up.
> >> But if that does not happen, maybe you could send a PR.
> > I can stick it in tip/perf/urgent if that helps :-) I'm the one merging
> > the broken patch in the first place I suppose :/
> >
> >> How this issue made it past -next would also be interesting to know.
> > Yeah this; normally I get robots screaming at me for things like this.
> 
> It looks the issue is caused by the merging order. When the patch series
> "Support SIMD/eGPRs/SSP registers sampling for perf" gets queued, the alpha
> patches which adds alpha specific perf_reg_validate() variant is still not
> merged, so the robot didn't find this issue. Thanks.

The robot should have tested the post-merge state in -next.
It turns out that the alpha tree was not part of -next this cycle [0].
It is being rectified for the next cycle, so everything is fine.

Also thanks to Peter for queuing the fix.

[0] https://lore.kernel.org/lkml/20261021185857.GB2608581@ax162/


Thomas