On Tue, Feb 11, 2014 at 5:50 PM, Peter Zijlstra <[email protected]> wrote:
> On Tue, Feb 11, 2014 at 05:25:39PM +0100, Stephane Eranian wrote:
>> On Tue, Feb 11, 2014 at 5:19 PM, Peter Zijlstra <[email protected]> wrote:
>> > On Tue, Feb 11, 2014 at 04:20:12PM +0100, Stephane Eranian wrote:
>> >> This patch adds a new uncore PMU for Intel SNB/IVB/HSW client
>> >
>> >
>> >> @@ -3501,6 +3844,28 @@ static int __init uncore_pci_init(void)
>> >> pci_uncores = ivt_pci_uncores;
>> >> uncore_pci_driver = &ivt_uncore_pci_driver;
>> >> break;
>> >> + case 42: /* Sandy Bridge */
>> >> + ret = snb_pci2phy_map_init(PCI_DEVICE_ID_INTEL_SNB_IMC);
>> >> + if (ret)
>> >> + return ret;
>> >> + pci_uncores = snb_pci_uncores;
>> >> + uncore_pci_driver = &snb_uncore_pci_driver;
>> >> + break;
>> >> + case 60: /* Haswell */
>> >> + case 69: /* Haswell Celeron */
>> >> + ret = snb_pci2phy_map_init(PCI_DEVICE_ID_INTEL_HSW_IMC);
>> >> + if (ret)
>> >> + return ret;
>> >> + pci_uncores = snb_pci_uncores;
>> >> + uncore_pci_driver = &hsw_uncore_pci_driver;
>> >> + break;
>> >> + case 58: /* Ivy Bridge */
>> >> + ret = snb_pci2phy_map_init(PCI_DEVICE_ID_INTEL_IVB_IMC);
>> >> + if (ret)
>> >> + return ret;
>> >> + pci_uncores = snb_pci_uncores;
>> >> + uncore_pci_driver = &ivb_uncore_pci_driver;
>> >> + break;
>> >> default:
>> >> return 0;
>> >> }
>> >
>> > I reorderd that list; but looking at perf_event_intel.c we have a lot
>> > more HSW clients listed there. Plz as to make it consistent.
>>
>> I don't have all of them, so no testing possible. I doubt they have so
>> many clients model numbers.
>
> Yeah, I don't have any of those chips.. last I have is WSM-EP.
>
> Anyway, perf_event_intel.c lists: 60,63,69,70,71 as being haswell
> clients. Andi did all that, so if its wrong its on Intel anyway.
>
What he has is okay for core. They all have the same core PMU at 99%.
But the uncore can be radically different. Look at IVB (58) vs IVB-EP (62).
So this is more risky there. That is why I did not add the other model numbers.
> I'm still thinking we ought to make a big Intel classification function;
> something that returns something like:
>
> struct intel_part {
> enum { client, ep, ex } type;
> enum { core, core2, nhm, wsm, snb, ivb, hsw } gen;
> };
>
> And we can do things like:
>
> if (ip.type == client && ip.gen >= snb)
>
> And then we only have to fix up the one classification function with all
> those stupid model numbers, instead of having them duplicated all over
> the stinking place.
Yeah, I am not opposed to that idea for PMU purposes.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
