Messages in this thread | | | Date | Thu, 21 May 2015 06:18:15 -0700 | Subject | Re: [PATCH 01/10] perf,x86: Fix event/group validation | From | Stephane Eranian <> |
| |
On Thu, May 21, 2015 at 6:09 AM, Peter Zijlstra <peterz@infradead.org> wrote: > On Thu, May 21, 2015 at 06:07:20AM -0700, Stephane Eranian wrote: >> On Thu, May 21, 2015 at 5:56 AM, Peter Zijlstra <peterz@infradead.org> wrote: >> > On Thu, May 21, 2015 at 05:35:02AM -0700, Stephane Eranian wrote: >> >> > Commit e979121b1b15 ("perf/x86/intel: Implement cross-HT corruption >> >> > bug workaround") made the situation much worse by actually setting the >> >> > event->hw.constraint value to NULL, so when validation and actual >> >> > scheduling interact we get NULL ptr derefs. >> >> > >> >> >> >> But x86_schedule_events() does reset the hw.constraint for each invocation: >> >> >> >> c = x86_pmu.get_event_constraints(cpuc, i, cpuc->event_list[i]); >> >> hwc->constraint = c; >> > >> > Yes, so if you have: >> > >> > validate_group() >> > >> > hwc->constraint = c; >> > >> Ok, you get that because validate_group() invokes x6_schedule_events() but >> on the fake_cpuc. This on fake_cpuc->event_list[]->hwc. >> >> > <context switch> >> > >> > c = hwc->constraint; >> > >> > The second c might not be the first. >> And where does this assignment come from? > > That's a read. The <context switch> can include a call to > x86_schedule_events(). Yes, but x86_schedule_events() never reads the constraint without setting it again before.
> >> For actual scheduling, we are using the actual cpuc, not fake_cpuc. >> Validate_group() does not modify global cpuc state. Or am I missing >> something? > > No, but x86_schedule_event() can modify event state, which is the fail. > Yes, it does modify the cpuc->event_list[]->hwc, because it is used as a cache for *EACH* invocation of the function. It is irrelevant outside the function.
>
| |