On 9/10/20 3:50 AM, peterz@xxxxxxxxxxxxx wrote:
> On Thu, Sep 10, 2020 at 10:32:23AM +0200, peterz@xxxxxxxxxxxxx wrote:
>>> @@ -363,7 +363,14 @@ perf_ibs_event_update(struct perf_ibs *perf_ibs, struct perf_event *event,
>>> static inline void perf_ibs_enable_event(struct perf_ibs *perf_ibs,
>>> struct hw_perf_event *hwc, u64 config)
>>> {
>>> - wrmsrl(hwc->config_base, hwc->config | config | perf_ibs->enable_mask);
>>> + u64 _config = (hwc->config | config) & ~perf_ibs->enable_mask;
>>> +
>>> + /* On Fam17h, the periodic fetch counter is set when IbsFetchEn is changed from 0 to 1 */
>>> + if (perf_ibs == &perf_ibs_fetch && boot_cpu_data.x86 >= 0x16 && boot_cpu_data.x86 <= 0x18)
>>> + wrmsrl(hwc->config_base, _config);
>
>> A better option would be to use hwc->flags, you're loading from that
>> line already, so it's guaranteed hot and then you only have a single
>> branch. Or stick it in perf_ibs near enable_mask, same difference.
>
> I fixed it for you.
> @@ -370,7 +371,13 @@ perf_ibs_event_update(struct perf_ibs *p
> static inline void perf_ibs_enable_event(struct perf_ibs *perf_ibs,
> struct hw_perf_event *hwc, u64 config)
> {
> - wrmsrl(hwc->config_base, hwc->config | config | perf_ibs->enable_mask);
> + u64 _config = (hwc->config | config) & ~perf_ibs->enable_mask;
> +
> + if (perf_ibs->fetch_count_reset_broken)
Nice, we don't even need the perf_ibs == &perf_ibs_fetch check
here because fetch_count_reset_broken is guaranteed to be 0 in
perf_ibs_op.
Thanks!
Kim
