2013-03-11 19:15:22

by Konrad Rzeszutek Wilk

[permalink] [raw]
Subject: SNB-PEBS errate in perf code - omits stepping 2?

Hey,

I think quirk:

+static int intel_snb_pebs_broken(int cpu)
+{
+ u32 rev = UINT_MAX; /* default to broken for unknown models */
+
+ switch (cpu_data(cpu).x86_model) {
+ case 42: /* SNB */
+ rev = 0x28;
+ break;
+
+ case 45: /* SNB-EP */
+ switch (cpu_data(cpu).x86_mask) {
+ case 6: rev = 0x618; break;
+ case 7: rev = 0x70c; break;
+ }
+ }
+
+ return (cpu_data(cpu).microcode < rev);
+}
+


needs a bit of update? I've this CPU:

vendor_id : GenuineIntel
cpu family : 6
model : 45
model name : Genuine Intel(R) CPU @ 2.30GHz
stepping : 2
cpu MHz : 2294.534
cache size : 20480 KB
..
microcode : 0x8000020c

and since the stepping is '2' and there are no 'default' case
in the switch statement, it ends up failing b/c it is asking
for revision UINT_MAX.

Is there a specific microcode errate version for this type
of CPU? Or am I a lucky recipient of an early development CPU?


Perhaps that check should also have this:

diff --git a/arch/x86/kernel/cpu/perf_event_intel.c b/arch/x86/kernel/cpu/perf_event_intel.c
index 529c893..287d43a 100644
--- a/arch/x86/kernel/cpu/perf_event_intel.c
+++ b/arch/x86/kernel/cpu/perf_event_intel.c
@@ -1845,6 +1845,9 @@ static int intel_snb_pebs_broken(int cpu)
case 7: rev = 0x70c; break;
}
}
+ if (rev == UINT_MAX)
+ pr_warn("Please contact Intel to update quirk as you have"
+ "an unknown CPU stepping: %d\n", cpu_data(cpu).x86_mask);

return (cpu_data(cpu).microcode < rev);
}


2013-03-12 13:06:19

by Stephane Eranian

[permalink] [raw]
Subject: Re: SNB-PEBS errate in perf code - omits stepping 2?

On Mon, Mar 11, 2013 at 8:15 PM, Konrad Rzeszutek Wilk
<[email protected]> wrote:
> Hey,
>
> I think quirk:
>
> +static int intel_snb_pebs_broken(int cpu)
> +{
> + u32 rev = UINT_MAX; /* default to broken for unknown models */
> +
> + switch (cpu_data(cpu).x86_model) {
> + case 42: /* SNB */
> + rev = 0x28;
> + break;
> +
> + case 45: /* SNB-EP */
> + switch (cpu_data(cpu).x86_mask) {
> + case 6: rev = 0x618; break;
> + case 7: rev = 0x70c; break;
> + }
> + }
> +
> + return (cpu_data(cpu).microcode < rev);
> +}
> +
>
>
> needs a bit of update? I've this CPU:
>
> vendor_id : GenuineIntel
> cpu family : 6
> model : 45
> model name : Genuine Intel(R) CPU @ 2.30GHz
> stepping : 2
> cpu MHz : 2294.534
> cache size : 20480 KB
> ..
> microcode : 0x8000020c
>
> and since the stepping is '2' and there are no 'default' case
> in the switch statement, it ends up failing b/c it is asking
> for revision UINT_MAX.
>
> Is there a specific microcode errate version for this type
> of CPU? Or am I a lucky recipient of an early development CPU?
>
I believe you have an early sample CPU.
But I think you comment is still valid. I think the kernel should print
a message once for your situation.

>
> Perhaps that check should also have this:
>
> diff --git a/arch/x86/kernel/cpu/perf_event_intel.c b/arch/x86/kernel/cpu/perf_event_intel.c
> index 529c893..287d43a 100644
> --- a/arch/x86/kernel/cpu/perf_event_intel.c
> +++ b/arch/x86/kernel/cpu/perf_event_intel.c
> @@ -1845,6 +1845,9 @@ static int intel_snb_pebs_broken(int cpu)
> case 7: rev = 0x70c; break;
> }
> }
> + if (rev == UINT_MAX)
> + pr_warn("Please contact Intel to update quirk as you have"
> + "an unknown CPU stepping: %d\n", cpu_data(cpu).x86_mask);
>
> return (cpu_data(cpu).microcode < rev);
> }