Re: [PATCH 4/9] drivers/perf: hisi: Use ACPI driver_data to retrieve SLLC PMU information

[Jonathan Cameron]

On Sat, 1 Mar 2025 06:43:49 +0000
Will Deacon <will@xxxxxxxxxx> wrote:

> On Tue, Feb 18, 2025 at 05:19:55PM +0800, Yicong Yang wrote:
> > From: Junhao He <hejunhao3@xxxxxxxxxx>
> > 
> > Make use of struct acpi_device_id::driver_data for version specific
> > information rather than judge the version register. This will help
> > to simplify the probe process and also a bit easier for extension.
> > 
> > Factor out SLLC register definition to struct hisi_sllc_pmu_regs.
> > No functional changes intended.
> > 
> > Signed-off-by: Junhao He <hejunhao3@xxxxxxxxxx>
> > Signed-off-by: Yicong Yang <yangyicong@xxxxxxxxxxxxx>
> > ---
> >  drivers/perf/hisilicon/hisi_uncore_sllc_pmu.c | 191 ++++++++++++------
> >  1 file changed, 125 insertions(+), 66 deletions(-)
> > 
> > diff --git a/drivers/perf/hisilicon/hisi_uncore_sllc_pmu.c b/drivers/perf/hisilicon/hisi_uncore_sllc_pmu.c
> > index dbd079016fc4..c1fd60d397c3 100644
> > --- a/drivers/perf/hisilicon/hisi_uncore_sllc_pmu.c
> > +++ b/drivers/perf/hisilicon/hisi_uncore_sllc_pmu.c
> > @@ -36,11 +36,14 @@
> >  #define SLLC_SRCID_NONE			0x0
> >  #define SLLC_TGTID_EN			BIT(5)
> >  #define SLLC_TGTID_NONE			0x0
> > -#define SLLC_TGTID_MIN_SHIFT		1
> > -#define SLLC_TGTID_MAX_SHIFT		12
> > -#define SLLC_SRCID_CMD_SHIFT		1
> > -#define SLLC_SRCID_MSK_SHIFT		12
> > +#define SLLC_TGTID_MIN_MSK		GENMASK(11, 1)
> > +#define SLLC_TGTID_MAX_MSK		GENMASK(22, 12)
> > +#define SLLC_SRCID_CMD_MSK		GENMASK(11, 1)
> > +#define SLLC_SRCID_MSK_MSK		GENMASK(22, 12)
> >  #define SLLC_NR_EVENTS			0x80
> > +#define SLLC_EVENT_CNTn(cnt0, n)	((cnt0) + (n) * 8)
> > +#define SLLC_FIRST_BIT(_mask)		(find_first_bit((const unsigned long *)&(_mask), 32))
> > +#define SLLC_FIELD_PREP(_mask, _val)	(_mask & (_val << SLLC_FIRST_BIT(_mask)))  
> 
> It's a bit of a shame to have to compute this dynamically given that th
> input mask is constant for a given device. Is it not possible to use the
> generic FIELD_PREP macro in per-device code and then just dispatch to
> that, instead of funneling everything through hisi_sllc_pmu_regs?
There is yet another ongoing discussion of handling non const
field cases:
https://lore.kernel.org/all/cover.1739540679.git.geert+renesas@xxxxxxxxx/

Maybe that will resolve with a nicer solution but I doubt it.

Alternatively best plan may be a shift (and length) instead
of the neat solution of FIELD_PREP().

Jonathan

> 
> Will