Re: [V2 PATCH 0/6] x86, NMI: give NMI handler a face-lift

[Jason Wessel <jason.wessel@windriver.com>]

On 11/18/2010 04:15 PM, Cyrill Gorcunov wrote:
> On Fri, Nov 19, 2010 at 12:56:50AM +0300, Cyrill Gorcunov wrote:
> ...
>   
>> ---
>>  arch/x86/kernel/cpu/perf_event_p4.c |   21 ++++++++++++---------
>>  1 file changed, 12 insertions(+), 9 deletions(-)
>>
>> Index: linux-2.6.git/arch/x86/kernel/cpu/perf_event_p4.c
>> =====================================================================
>> --- linux-2.6.git.orig/arch/x86/kernel/cpu/perf_event_p4.c
>> +++ linux-2.6.git/arch/x86/kernel/cpu/perf_event_p4.c
>> @@ -753,19 +753,22 @@ out:
>>  
>>  static inline int p4_pmu_clear_cccr_ovf(struct hw_perf_event *hwc)
>>  {
>> -	int overflow = 0;
>> -	u32 low, high;
>> +	u32 overflow = 0;
>> +	u32 low, low_cccr, high;
>>  
>> -	rdmsr(hwc->config_base + hwc->idx, low, high);
>> +	/* an official way for overflow indication */
>> +	rdmsr(hwc->config_base + hwc->idx, low_cccr, high);
>> +	overflow |= (low_cccr & P4_CCCR_OVF);
>> +
>> +	/* unflagged overflows */
>> +	rdmsr(hwc->event_base + hwc->idx, low, high);
>> +	overflow |= high & 0x80000000;
>>  
>> -	/* we need to check high bit for unflagged overflows */
>> -	if ((low & P4_CCCR_OVF) || !(high & (1 << 31))) {
>> -		overflow = 1;
>> +	if (overflow)
>>     
>
> this should be rather 'if (low_cccr & P4_CCCR_OVF)' otherwise
> redundant checking_wrmsrl called, updated patch below
>
> /me seems should not touch code at all today 
> ---
>
> perf, x86: P4 PMU - Fix unflagged overflows handling
>
> Jason pointed out that kgdb no longer works with new
> nmi-watchdog. Don found the reason -- P4 PMU reads CCCR
> register instead of counter itself, it forces NMIs to
> be eaten by perf subsystem.
>
> Fix it by reading a proper register.
>
> v2: Call checking_wrmsrl only if needed
>
> Reported-by: Jason Wessel <jason.wessel@windriver.com>
> Reported-by: Don Zickus <dzickus@redhat.com>
> Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
> ---
>
> Jason I've removed your Tested-by because the patch differ a bit
>
>   

It is retested and confirmed to work on real HW.   This no longer works
on qemu/kvm likely because the emulation is incorrect.

Thanks,
Jason.