Re: [PATCH v5 2/6] perf arm64: Allow version comparisons of CPU IDs

[John Garry <john.g.garry@oracle.com>]

On 16/08/2023 10:12, James Clark wrote:
> 
> 
> On 15/08/2023 10:35, John Garry wrote:
>> On 11/08/2023 15:39, James Clark wrote:
>>> Currently variant and revision fields are masked out of the MIDR so
>>> it's not possible to compare different versions of the same CPU.
>>> In a later commit a workaround will be removed just for N2 r0p3, so
>>> enable comparisons on version.
>>>
>>> This has the side effect of changing the MIDR stored in the header of
>>> the perf.data file to no longer have masked version fields. It also
>>> affects the lookups in mapfile.csv, but as that currently only has
>>> zeroed version fields, it has no actual effect. The mapfile.csv
>>> documentation also states to zero the version fields, so unless this
>>> isn't done it will continue to have no effect.
>>>
>>
>> This looks ok apart from a couple of comments, below.
>>
>> Thanks,
>> John
>>
>>> Signed-off-by: James Clark <james.clark@arm.com>
>>> ---
>>>    tools/perf/arch/arm64/util/header.c | 64 ++++++++++++++++++++++-------
>>>    1 file changed, 50 insertions(+), 14 deletions(-)
>>>
>>> diff --git a/tools/perf/arch/arm64/util/header.c
>>> b/tools/perf/arch/arm64/util/header.c
>>> index 80b9f6287fe2..8f74e801e1ab 100644
>>> --- a/tools/perf/arch/arm64/util/header.c
>>> +++ b/tools/perf/arch/arm64/util/header.c
>>> @@ -1,3 +1,6 @@
>>> +#include <linux/kernel.h>
>>> +#include <linux/bits.h>
>>> +#include <linux/bitfield.h>
>>>    #include <stdio.h>
>>>    #include <stdlib.h>
>>>    #include <perf/cpumap.h>
>>> @@ -10,14 +13,12 @@
>>>      #define MIDR "/regs/identification/midr_el1"
>>>    #define MIDR_SIZE 19
>>> -#define MIDR_REVISION_MASK      0xf
>>> -#define MIDR_VARIANT_SHIFT      20
>>> -#define MIDR_VARIANT_MASK       (0xf << MIDR_VARIANT_SHIFT)
>>> +#define MIDR_REVISION_MASK      GENMASK(3, 0)
>>> +#define MIDR_VARIANT_MASK    GENMASK(23, 20)
>>>      static int _get_cpuid(char *buf, size_t sz, struct perf_cpu_map
>>> *cpus)
>>>    {
>>>        const char *sysfs = sysfs__mountpoint();
>>> -    u64 midr = 0;
>>>        int cpu;
>>>          if (!sysfs || sz < MIDR_SIZE)
>>> @@ -44,21 +45,11 @@ static int _get_cpuid(char *buf, size_t sz, struct
>>> perf_cpu_map *cpus)
>>>            }
>>>            fclose(file);
>>>    -        /* Ignore/clear Variant[23:20] and
>>> -         * Revision[3:0] of MIDR
>>> -         */
>>> -        midr = strtoul(buf, NULL, 16);
>>> -        midr &= (~(MIDR_VARIANT_MASK | MIDR_REVISION_MASK));
>>> -        scnprintf(buf, MIDR_SIZE, "0x%016lx", midr);
>>>            /* got midr break loop */
>>>            break;
>>>        }
>>>          perf_cpu_map__put(cpus);
>>> -
>>> -    if (!midr)
>>> -        return EINVAL;
>>
>> Is there a reason to drop this check?
>>
>> As I see, it is still checked in perf_pmu__getcpudid() ->
>> get_cpuid_str() -> _get_cpuid(), and we don't zero the buf allocated in
>> _get_cpuid()
>>
> 
> Ah yes, now if all the files fail to open or read then buf will be
> uninitialized. I make it so that it will return EINVAL unless the
> fgets() succeeds, but I don't think we need to add the strtoul() back in?

Right, I don't think that the strtoul() is required.

> 
>>> -
>>>        return 0;
>>>    }
>>>    @@ -99,3 +90,48 @@ char *get_cpuid_str(struct perf_pmu *pmu)
>>>          return buf;
>>>    }
>>> +
>>> +/*
>>> + * Return 0 if idstr is a higher or equal to version of the same part as
>>> + * mapcpuid.
>>
>> And what other values may be returned? If just 0/1, then can we have a
>> bool return value?
>>
> 
> I don't think that's best for consistency. All the other CPU ID
> comparison functions return the strcmp style return values which is the
> reverse of booleans. We could change them all to bool, but it would be a
> big change, and they'd still have strcmp in the name which suggests
> -1/0/1 return values (although -1 is never used here).
> 
> I will add to the comment that 1 is returned for a comparison failure
> thought. That is missing.

ok, fine.

> 
>>> + *
>>> + * Therefore, if mapcpuid has 0 for revision and variant then any
>>> version of
>>> + * idstr will match as long as it's the same CPU type.
>>> + */
>>> +int strcmp_cpuid_str(const char *mapcpuid, const char *idstr)
>>> +{
>>> +    u64 map_id = strtoull(mapcpuid, NULL, 16);
>>> +    char map_id_variant = FIELD_GET(MIDR_VARIANT_MASK, map_id);
>>> +    char map_id_revision = FIELD_GET(MIDR_REVISION_MASK, map_id);
>>> +    u64 id = strtoull(idstr, NULL, 16);
>>> +    char id_variant = FIELD_GET(MIDR_VARIANT_MASK, id);
>>> +    char id_revision = FIELD_GET(MIDR_REVISION_MASK, id);
>>> +    u64 id_fields = ~(MIDR_VARIANT_MASK | MIDR_REVISION_MASK);
>>> +
>>> +    /* Compare without version first */
>>> +    if ((map_id & id_fields) != (id & id_fields))
>>> +        return 1;
>>> +
>>> +    /*
>>> +     * ID matches, now compare version.
>>> +     *
>>> +     * Arm revisions (like r0p0) are compared here like two digit semver
>>> +     * values eg. 1.3 < 2.0 < 2.1 < 2.2. The events json file with the
>>> +     * highest matching version is used.
>>> +     *
>>> +     *  r = high value = 'Variant' field in MIDR
>>> +     *  p = low value  = 'Revision' field in MIDR
>>> +     *
>>> +     */
>>> +    if (id_variant > map_id_variant)
>>> +        return 0;
>>> +
>>> +    if (id_variant == map_id_variant && id_revision >= map_id_revision)
>>> +        return 0;
>>> +
>>> +    /*
>>> +     * variant is less than mapfile variant or variants are the same but
>>> +     * the revision doesn't match. Return no match.
>>> +     */
>>> +    return 1;
>>> +}
>>