[PATCH v3] ARM, ARM64, LONGARCH, XTENSA: Delay HW BP notification to task_work()

[Sebastian Andrzej Siewior]

On 2026-10-06 18:17:52 [+0100], Ada Couprie Diaz wrote:
> Hi Sebastian,
Hi Ada,

> I find the function names a bit counter-intuitive, compared to the other
> arch-specific perf functions.
> Given the name `perf_arch_...`, I would have expected them to be defined
> in arch code, rather than in the generic perf code.
> From what I can see, usually perf functions calling an arch-specific
> function lack the `_arch_` infix of their `arch_` counterpart.
> 
> I do not know very well what we expect in perf, so I might be off-base,
> but would calling them `perf_hwbp_send_sig()` and `perf_hwbp_notify()`
> make sense ?

Thanks for looking at this. I renamed them as:
- perf_arch_hwbp_notify() -> perf_hwbp_notify() (generic)
- arch_hwbp_send_sig() -> perf_arch_hwbp_send_sig() (arch specific)
- perf_arch_hwbp_send_sig() -> perf_hwbp_send_sig() (generic)

> Otherwise, it looks good to me on the arm64 side !
> I had a look on the arm side as well, given the debug handling architecture
> is similar, and I think it is OK on there as well, though I wouldn't mind
> a more experienced arm review :)
> 
> Reviewed-by: Ada Couprie Diaz <ada.coupriediaz at arm.com>
> 
> I also tested the patch with pNMI and CONFIG_PREEMPT_RT on arm64 : I can
> confirm that the atomic sleep warning is gone and everything works
> as expected !
> 
> Tested-by: Ada Couprie Diaz <ada.coupriediaz at arm.com> (arm64)
> 
> Thanks a lot for looking into this, combined with[0] the hardware debug
> handling should be much cleaner ! :)

great.

> Kind regards,
> Ada

Sebastian