Messages in this thread Patch in this message |  | | | Subject | Re: [PATCH] perf: fix possible divide-by-zero in perf_swevent_overflow() | | From | Peter Zijlstra <> | | Date | Thu, 26 Aug 2010 14:58:06 +0200 |
| |
On Thu, 2010-08-26 at 20:36 +0800, DDD wrote: > Peter Zijlstra wrote: > > On Thu, 2010-08-26 at 20:07 +0800, Dongdong Deng wrote: > >> The event->hw.last_period is possible to zero, thus it will > >> cause divide_by_zero later in perf_swevent_set_period(). > > > > How can it be zero? > > When I am running the kgdbts to test the hw_breakpoint_layer with kgdb, > I get a call trace as following and this problem is hardly to reproduce. > > Maybe the root cause was from kgdb/hw_breakpoint_layer,
Yeah, I think there's a bug in the hw_breakpoint stuff, does something like the below fix it?
> but add a checking is good to us and harmless. :-)
Except that code path is already too bloated and should be reduced not added to and conditionals are expensive.
--- kernel/hw_breakpoint.c | 13 ++++++++++++- 1 files changed, 12 insertions(+), 1 deletions(-) diff --git a/kernel/hw_breakpoint.c b/kernel/hw_breakpoint.c index d71a987..f57ebee 100644 --- a/kernel/hw_breakpoint.c +++ b/kernel/hw_breakpoint.c @@ -600,9 +600,20 @@ static int __init init_hw_breakpoint(void) } core_initcall(init_hw_breakpoint); +static int hw_breakpoint_enable(struct perf_event *event) +{ + struct hw_perf_event *hwc = &event->hw; + + if (hwc->sample_period) { + hwc->last_period = hwc->sample_period; + perf_swevent_set_period(event); + } + + return arch_install_hw_breakpoint(event); +} struct pmu perf_ops_bp = { - .enable = arch_install_hw_breakpoint, + .enable = hw_breakpoint_enable, .disable = arch_uninstall_hw_breakpoint, .read = hw_breakpoint_pmu_read, };
|  |