On Thu, Sep 04, 2025 at 08:44:48AM +0200, Peter Zijlstra wrote:
> On Thu, Sep 04, 2025 at 08:21:03AM +0800, Jinchao Wang wrote:
> > Introduce hw_breakpoint_modify_local() as a generic helper to modify an
> > existing hardware breakpoint. The function invokes
> > hw_breakpoint_arch_parse() and delegates the reinstall step to the
> > architecture via arch_reinstall_hw_breakpoint().
> >
> > A weak default implementation of arch_reinstall_hw_breakpoint() is
> > provided, returning -EOPNOTSUPP on architectures without support.
> >
> > This makes the interface arch-independent while allowing x86 (and others)
> > to provide their own implementation.
> >
> > Signed-off-by: Jinchao Wang <wangjinchao...@gmail.com>
> > ---
> > include/linux/hw_breakpoint.h | 1 +
> > kernel/events/hw_breakpoint.c | 18 ++++++++++++++++++
> > 2 files changed, 19 insertions(+)
> >
> > diff --git a/include/linux/hw_breakpoint.h b/include/linux/hw_breakpoint.h
> > index db199d653dd1..9453b5bdb443 100644
> > --- a/include/linux/hw_breakpoint.h
> > +++ b/include/linux/hw_breakpoint.h
> > @@ -67,6 +67,7 @@ extern int
> > modify_user_hw_breakpoint_check(struct perf_event *bp, struct
> > perf_event_attr *attr,
> > bool check);
> >
> > +int hw_breakpoint_modify_local(struct perf_event *bp, struct
> > perf_event_attr *attr);
> > /*
> > * Kernel breakpoints are not associated with any particular thread.
> > */
> > diff --git a/kernel/events/hw_breakpoint.c b/kernel/events/hw_breakpoint.c
> > index 8ec2cb688903..ff428739f71e 100644
> > --- a/kernel/events/hw_breakpoint.c
> > +++ b/kernel/events/hw_breakpoint.c
> > @@ -983,6 +983,24 @@ static void hw_breakpoint_del(struct perf_event *bp,
> > int flags)
> > arch_uninstall_hw_breakpoint(bp);
> > }
> >
> > +int hw_breakpoint_modify_local(struct perf_event *bp, struct
> > perf_event_attr *attr)
> > +{
> > + int err;
> > +
> > + err = hw_breakpoint_arch_parse(bp, attr, counter_arch_bp(bp));
> > + if (err)
> > + return err;
> > +
> > + return arch_reinstall_hw_breakpoint(bp);
> > +}
> > +EXPORT_SYMBOL(hw_breakpoint_modify_local);
> > +
> > +/* weak fallback for arches without support */
> > +__weak int arch_reinstall_hw_breakpoint(struct perf_event *bp)
> > +{
> > + return -EOPNOTSUPP;
> > +}
>
> Again, so much fail :/
>
> So we have:
>
> {register,modify,unregister}_user_hw_breakpoint()
>
> and
>
> {register,unregister}_wide_hw_breakpoint()
>
> And you choose to extend this latter with hw_breakpoint_modify_local()
> instead of sticking with the naming scheme and say adding:
>
> modify_wide_hw_breakpoint_local().
>
> Also, again, that EXPORT is a fail, these other interfaces are all
> EXPORT_SYMBOL_GPL().
>
Thanks for your patience.
I was misled by another family:
- hw_breakpoint_add
- hw_breakpoint_del
- hw_breakpoint_start
- hw_breakpoint_stop
Since this logic was also added in the wprobe series by Masami, I will
adopt his version instead.
> Also note that modify_user_hw_breakpoint() doesn't seem to need new arch
> hooks. Yet you fail to explain why you think you do.
Thanks for feedback, I will study existing code to better handle the arch
dependencies.
