Re: [PATCH 2/4] perf/arm-cmn: Add CMN-650 support

[Ilkka Koskinen <ilkka@os.amperecomputing.com>]

[-- Attachment #1: Type: text/plain, Size: 3330 bytes --]



On Thu, 21 Apr 2022, Robin Murphy wrote:

> On 2022-04-21 08:25, Ilkka Koskinen wrote:
>> 
>> I still have a couple tiny comments. Otherwise the patch set looks good to 
>> me.
>> 
>> On Mon, 18 Apr 2022, Robin Murphy wrote:
>>> Add the identifiers and events for CMN-650, which slots into its
>>> evolutionary position between CMN-600 and the 700-series products.
>>> Imagine CMN-600 made bigger, and with most of the rough edges smoothed
>>> off, but that then balanced out by some bonkers PMU functionality for
>>> the new HN-P enhancement in CMN-650r2.
>>> 
>>> Most of the CXG events are actually common to newer revisions of CMN-600
>>> too, so they're arguably a little late; oh well.
>>> 
>>> Signed-off-by: Robin Murphy <robin.murphy@arm.com>
>>> ---
>>> drivers/perf/arm-cmn.c | 222 ++++++++++++++++++++++++++++++++---------
>>> 1 file changed, 176 insertions(+), 46 deletions(-)
>>> 
>>> diff --git a/drivers/perf/arm-cmn.c b/drivers/perf/arm-cmn.c
>>> index 9c1d82be7a2f..cce8516d465c 100644
>>> --- a/drivers/perf/arm-cmn.c
>>> +++ b/drivers/perf/arm-cmn.c


>>> @@ -607,12 +624,29 @@ static umode_t arm_cmn_event_attr_is_visible(struct 
>>> kobject *kobj,
>>>     }
>>> 
>>>     /* Revision-specific differences */
>>> -    if (cmn->model == CMN600 && cmn->rev < CMN600_R1P2) {
>>> -        if (eattr->type == CMN_TYPE_HNF && eattr->eventid == 0x1b)
>>> -            return 0;
>>> +    if (cmn->model == CMN600) {
>>> +        if (cmn->rev < CMN600_R1P3) {
>>> +            if (type == CMN_TYPE_CXRA && eventid > 0x10)
>>> +                return 0;
>>> +        }
>>> +        if (cmn->rev < CMN600_R1P2) {
>>> +            if (type == CMN_TYPE_HNF && eventid == 0x1b)
>>> +                return 0;
>>> +            if (type == CMN_TYPE_CXRA || type == CMN_TYPE_CXHA)
>>> +                return 0;
>>> +        }
>>> +    } else if (cmn->model == CMN650) {
>>> +        if (cmn->rev < CMN650_R2P0 || cmn->rev == CMN650_R1P2) {
>>> +            if (type == CMN_TYPE_HNF && eventid > 0x22)
>>> +                return 0;
>>> +            if (type == CMN_TYPE_SBSX && eventid == 0x17)
>>> +                return 0;
>>> +            if (type == CMN_TYPE_RNI && eventid > 0x10)
>>> +                return 0;
>>> +        }
>> 
>> What's the plan with cmn-650 r2p0 event settings? There seem to be a few 
>> extra ones made visible now. I'm fine with updating this patch or taking 
>> care of them in a separate one, which ever makes more sense.
>
> arm_cmn_event_attrs *should* contain all the events supported by r2p0, with 
> the ones not present on earlier revisions then being filtered out here. Have 
> I missed anything? Note that HN-Ps don't need per-event filtering since that 
> node type simply doesn't exist prior to CMN-650 r2, so is already filtered by 
> arm_cmn_node().

Ah, I was comparing the code and perf output to HN-F and RN-I event 
summary tables in TRM. The tables are missing some of the events where as 
the event_sel register descriptions have all of them listed. So, that was 
my mistake and everything looks good.

Cheers, Ilkka

>
> Thanks,
> Robin.
>

[-- Attachment #2: Type: text/plain, Size: 176 bytes --]

_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel