[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20230905154634.000075c5@huawei.com>
Date: Tue, 5 Sep 2023 15:46:34 +0100
From: Jonathan Cameron <Jonathan.Cameron@...wei.com>
To: Will Deacon <will@...nel.org>
CC: Jeongtae Park <jtp.park@...sung.com>, <linux-cxl@...r.kernel.org>,
<linux-arm-kernel@...ts.infradead.org>,
<linux-kernel@...r.kernel.org>,
"Mark Rutland" <mark.rutland@....com>,
Kyungsan Kim <ks0204.kim@...sung.com>,
Wonjae Lee <wj28.lee@...sung.com>,
Hojin Nam <hj96.nam@...sung.com>,
Junhyeok Im <junhyeok.im@...sung.com>,
Jehoon Park <jehoon.park@...sung.com>
Subject: Re: [PATCH] perf: CXL: fix mismatched number of counters mask
On Tue, 5 Sep 2023 15:28:54 +0100
Will Deacon <will@...nel.org> wrote:
> On Tue, Sep 05, 2023 at 09:33:09PM +0900, Jeongtae Park wrote:
> > The number of Count Units field is described as 6 bits long
> > in the CXL 3.0 specification. However, its mask value was
> > only declared as 5 bits long.
> >
> > Signed-off-by: Jeongtae Park <jtp.park@...sung.com>
> > ---
> > drivers/perf/cxl_pmu.c | 2 +-
> > 1 file changed, 1 insertion(+), 1 deletion(-)
> >
> > diff --git a/drivers/perf/cxl_pmu.c b/drivers/perf/cxl_pmu.c
> > index 0a8f597e695b..365d964b0f6a 100644
> > --- a/drivers/perf/cxl_pmu.c
> > +++ b/drivers/perf/cxl_pmu.c
> > @@ -25,7 +25,7 @@
> > #include "../cxl/pmu.h"
> >
> > #define CXL_PMU_CAP_REG 0x0
> > -#define CXL_PMU_CAP_NUM_COUNTERS_MSK GENMASK_ULL(4, 0)
> > +#define CXL_PMU_CAP_NUM_COUNTERS_MSK GENMASK_ULL(5, 0)
> > #define CXL_PMU_CAP_COUNTER_WIDTH_MSK GENMASK_ULL(15, 8)
> > #define CXL_PMU_CAP_NUM_EVN_CAP_REG_SUP_MSK GENMASK_ULL(24, 20)
> > #define CXL_PMU_CAP_FILTERS_SUP_MSK GENMASK_ULL(39, 32)
>
> I don't have access to the CXL spec, but widening this mask looks like
> it puts us out-of-whack with CXL_PMU_MAX_COUNTERS.
>
> Did v3.0 of the spec bump the number of counters? If so, can you please
> check that this is a backwards-compatible change?
CXL Performance monitors were only introduced in CXL 3.0 so not that.
The max value that register can take is 0x3f (0 based, so 64 counters ==
CXL_PMU_MAX_COUNTERS)
So it should be 6 bits wide. I did some history digging and this isn't
even a draft spec / final spec issue - simple typo I guess.
Fix seems correct to me.
Acked-by: Jonathan Cameron <Jonathan.Cameron@...wei.com>
>
> Will
Powered by blists - more mailing lists