Re: [PATCH] perf/core: fast breakpoint modification via _IOC_MODIFY_BREAKPOINT

[Jiri Olsa]

On Wed, Nov 08, 2017 at 07:02:22AM -0800, Milind Chabbi wrote:
> On Wed, Nov 8, 2017 at 6:15 AM, Jiri Olsa <jolsa@xxxxxxxxxx> wrote:
> > On Mon, Nov 06, 2017 at 07:04:40AM -0800, Milind Chabbi wrote:
> >> Hi Jirka,
> >>
> >> I see the tabs in my sent email, do you have suggestions on how best to
> >> send this patch so that the tabs are preserved by the email client?
> >> Can anybody else also check if they received with/without tabs?
> >>
> >> release_bp_slot/reserve_bp_slot majic is not necessary since
> >> _IOC_MODIFY_BREAKPOINT ioctl modifies an already registered breakpoint
> >> without affecting the count of breakpoints active.
> >
> > but AFAICS you allow to change the breakpoint type (bp_type)
> > and slot counts are based on the breakpoint type
> >
> > jirka
> 
> Jirka,
> I am not able to fully understand your concern.
> Can you point to a code file and line related to your observation?
> The patch is modeled after the existing modify_user_hw_breakpoint() function
> present in events/hw_breakpoint.c; don't you see this problem in that code?

the reserve_bp_slot/release_bp_slot functions manage
counts for current breakpoints based on its type

those counts are cumulated in here:
  static DEFINE_PER_CPU(struct bp_cpuinfo, bp_cpuinfo[TYPE_MAX]);

you allow to change the breakpoint type, so I'd expect
to see some code that release slot count for old type
and take new one (if it's available)

jirka