Re: [RFC] x86,perf: Implement minimal P4 PMU driver v14

From: Cyrill Gorcunov
Date: Thu Mar 11 2010 - 16:15:51 EST


On Thu, Mar 11, 2010 at 07:39:21PM +0100, Ingo Molnar wrote:
>
> * Ingo Molnar <mingo@xxxxxxx> wrote:
>
> > * Cyrill Gorcunov <gorcunov@xxxxxxxxxx> wrote:
> >
> > > x86,perf: Implement minimal P4 PMU driver v15
> >
> > tried it on a Pentium-D dual core CPU, and it boots fine:
>
> an Athlon64 testbox was not as happy:
>
> [ 0.253338] calling spawn_nmi_watchdog_task+0x0/0x63 @ 1
> [ 0.256675] NMI watchdog enabled, takes one hw-pmu counter.
> [ 0.260013] nmi_watchdog: hardware not available, trying software events
> [ 0.263380] BUG: unable to handle kernel NULL pointer dereference at (null)
> [ 0.266666] IP: [<(null)>] (null)
> [ 0.266666] *pde = 00000000
> [ 0.266666] Oops: 0000 [#1] PREEMPT SMP DEBUG_PAGEALLOC
> [ 0.266666] last sysfs file:
> [ 0.266666]
> [ 0.266666] Pid: 1, comm: swapper Not tainted 2.6.34-rc1-tip+ #20943 /
> [ 0.266666] EIP: 0060:[<00000000>] EFLAGS: 00010046 CPU: 0
> [ 0.266666] EIP is at 0x0
> [ 0.266666] EAX: 434035b0 EBX: 00000000 ECX: 7f81fe08 EDX: 00000000
> [ 0.266666] ESI: 43406444 EDI: 7f82e004 EBP: 7f81ff14 ESP: 7f81fdf0
> [ 0.266666] DS: 007b ES: 007b FS: 00d8 GS: 0000 SS: 0068
> [ 0.266666] Process swapper (pid: 1, ti=7f81f000 task=7f824000 task.ti=7f81f000)
>
> config and full crashlog attached. I had to exclude tip:perf/x86 for now
> (reverting commit a072738e04 cured the crash), you can re-create that kernel
> by doing this:
>
> git checkout tip/master
> git merge tip/perf/x86
>
> (and fixes would be nice to have as delta patches against perf/x86 as well.)
>
> Thanks,
>
> Ingo

Perhaps something like the patch below (tested with kvm)? With this patch
we will actually waste ~4/8 bytes per PMU (intel,amd,p6) since this call
hits on p4 only, so I think perhaps better to use one x86 scheduler hook
instead of empty schedule_events() in PMU, hmm?
---

x86,perf: Fix NULL deref on not assigned x86_pmu

In case of not assigned x86_pmu and software events
NULL dereference may being hit via x86_pmu::schedule_events
method.

Fix it by calling x86_pmu::schedule_events only if we
have one. Otherwise use general scheduler.

Also the former x86_schedule_events calls restored.

Signed-off-by: Cyrill Gorcunov <gorcunov@xxxxxxxxxx>
---
arch/x86/kernel/cpu/perf_event.c | 10 +++++++---
arch/x86/kernel/cpu/perf_event_amd.c | 1 -
arch/x86/kernel/cpu/perf_event_intel.c | 2 --
arch/x86/kernel/cpu/perf_event_p6.c | 1 -
4 files changed, 7 insertions(+), 7 deletions(-)

Index: linux-2.6.git/arch/x86/kernel/cpu/perf_event.c
=====================================================================
--- linux-2.6.git.orig/arch/x86/kernel/cpu/perf_event.c
+++ linux-2.6.git/arch/x86/kernel/cpu/perf_event.c
@@ -604,6 +604,10 @@ static int x86_schedule_events(struct cp
int i, j, w, wmax, num = 0;
struct hw_perf_event *hwc;

+ /* the PMU has its own scheduler */
+ if (unlikely(x86_pmu.schedule_events))
+ return x86_pmu.schedule_events(cpuc, n, assign);
+
bitmap_zero(used_mask, X86_PMC_IDX_MAX);

for (i = 0; i < n; i++) {
@@ -936,7 +940,7 @@ static int x86_pmu_enable(struct perf_ev
if (n < 0)
return n;

- ret = x86_pmu.schedule_events(cpuc, n, assign);
+ ret = x86_schedule_events(cpuc, n, assign);
if (ret)
return ret;
/*
@@ -1268,7 +1272,7 @@ int hw_perf_group_sched_in(struct perf_e
if (n0 < 0)
return n0;

- ret = x86_pmu.schedule_events(cpuc, n0, assign);
+ ret = x86_schedule_events(cpuc, n0, assign);
if (ret)
return ret;

@@ -1521,7 +1525,7 @@ static int validate_group(struct perf_ev

fake_cpuc->n_events = n;

- ret = x86_pmu.schedule_events(fake_cpuc, n, NULL);
+ ret = x86_schedule_events(fake_cpuc, n, NULL);

out_free:
kfree(fake_cpuc);
Index: linux-2.6.git/arch/x86/kernel/cpu/perf_event_amd.c
=====================================================================
--- linux-2.6.git.orig/arch/x86/kernel/cpu/perf_event_amd.c
+++ linux-2.6.git/arch/x86/kernel/cpu/perf_event_amd.c
@@ -364,7 +364,6 @@ static __initconst struct x86_pmu amd_pm
.enable = x86_pmu_enable_event,
.disable = x86_pmu_disable_event,
.hw_config = x86_hw_config,
- .schedule_events = x86_schedule_events,
.eventsel = MSR_K7_EVNTSEL0,
.perfctr = MSR_K7_PERFCTR0,
.event_map = amd_pmu_event_map,
Index: linux-2.6.git/arch/x86/kernel/cpu/perf_event_intel.c
=====================================================================
--- linux-2.6.git.orig/arch/x86/kernel/cpu/perf_event_intel.c
+++ linux-2.6.git/arch/x86/kernel/cpu/perf_event_intel.c
@@ -750,7 +750,6 @@ static __initconst struct x86_pmu core_p
.enable = x86_pmu_enable_event,
.disable = x86_pmu_disable_event,
.hw_config = x86_hw_config,
- .schedule_events = x86_schedule_events,
.eventsel = MSR_ARCH_PERFMON_EVENTSEL0,
.perfctr = MSR_ARCH_PERFMON_PERFCTR0,
.event_map = intel_pmu_event_map,
@@ -789,7 +788,6 @@ static __initconst struct x86_pmu intel_
.enable = intel_pmu_enable_event,
.disable = intel_pmu_disable_event,
.hw_config = x86_hw_config,
- .schedule_events = x86_schedule_events,
.eventsel = MSR_ARCH_PERFMON_EVENTSEL0,
.perfctr = MSR_ARCH_PERFMON_PERFCTR0,
.event_map = intel_pmu_event_map,
Index: linux-2.6.git/arch/x86/kernel/cpu/perf_event_p6.c
=====================================================================
--- linux-2.6.git.orig/arch/x86/kernel/cpu/perf_event_p6.c
+++ linux-2.6.git/arch/x86/kernel/cpu/perf_event_p6.c
@@ -110,7 +110,6 @@ static __initconst struct x86_pmu p6_pmu
.enable = p6_pmu_enable_event,
.disable = p6_pmu_disable_event,
.hw_config = x86_hw_config,
- .schedule_events = x86_schedule_events,
.eventsel = MSR_P6_EVNTSEL0,
.perfctr = MSR_P6_PERFCTR0,
.event_map = p6_pmu_event_map,
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/