| Message ID | 1592487344-30555-1-git-send-email-zhangshaokun@hisilicon.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | [v3,1/3] arm64: perf: Add support caps in sysfs | expand |
Hi Will/Mark, Sorry for my noise, any more comments please? Thanks, Shaokun 在 2020/6/18 21:35, Shaokun Zhang 写道: > ARMv8.4-PMU introduces the PMMIR_EL1 registers and some new PMU events, > like STALL_SLOT etc, are related to it. Let's add a caps directory to > /sys/bus/event_source/devices/armv8_pmuv3_0/ and support slots from > PMMIR_EL1 registers in this entry. The user programs can get the slots > from sysfs directly. > > Cc: Will Deacon <will@kernel.org> > Cc: Mark Rutland <mark.rutland@arm.com> > Signed-off-by: Shaokun Zhang <zhangshaokun@hisilicon.com> > --- > ChangeLog in v3: > * Fix one typo in patch3 > > ChangeLog in v2: > * Add caps entry in sysfs > * Fix the PMU events typos > * Add one new patch to correct event ID in sysfs > > arch/arm64/include/asm/sysreg.h | 2 + > arch/arm64/kernel/perf_event.c | 87 +++++++++++++++++++++++++++++++---------- > include/linux/perf/arm_pmu.h | 1 + > 3 files changed, 69 insertions(+), 21 deletions(-) > > diff --git a/arch/arm64/include/asm/sysreg.h b/arch/arm64/include/asm/sysreg.h > index 463175f80341..56c45a9207c7 100644 > --- a/arch/arm64/include/asm/sysreg.h > +++ b/arch/arm64/include/asm/sysreg.h > @@ -321,6 +321,8 @@ > #define SYS_PMINTENSET_EL1 sys_reg(3, 0, 9, 14, 1) > #define SYS_PMINTENCLR_EL1 sys_reg(3, 0, 9, 14, 2) > > +#define SYS_PMMIR_EL1 sys_reg(3, 0, 9, 14, 6) > + > #define SYS_MAIR_EL1 sys_reg(3, 0, 10, 2, 0) > #define SYS_AMAIR_EL1 sys_reg(3, 0, 10, 3, 0) > > diff --git a/arch/arm64/kernel/perf_event.c b/arch/arm64/kernel/perf_event.c > index 4d7879484cec..5f2ac87e4b91 100644 > --- a/arch/arm64/kernel/perf_event.c > +++ b/arch/arm64/kernel/perf_event.c > @@ -277,6 +277,51 @@ static struct attribute_group armv8_pmuv3_format_attr_group = { > .attrs = armv8_pmuv3_format_attrs, > }; > > +static inline int armv8pmu_get_pmu_version(void) > +{ > + int pmuver; > + u64 dfr0; > + > + dfr0 = read_sysreg(id_aa64dfr0_el1); > + pmuver = cpuid_feature_extract_unsigned_field(dfr0, > + ID_AA64DFR0_PMUVER_SHIFT); > + > + return pmuver; > +} > + > +static umode_t > +armv8pmu_caps_attr_is_visible(struct kobject *kobj, struct attribute *attr, > + int unused) > +{ > + int pmuver = armv8pmu_get_pmu_version(); > + > + if (pmuver >= ID_AA64DFR0_PMUVER_8_4) > + return attr->mode; > + > + return 0; > +} > + > +static ssize_t slots_show(struct device *dev, struct device_attribute *attr, > + char *buf) > +{ > + int slots = read_sysreg_s(SYS_PMMIR_EL1) & 0xFF; > + > + return snprintf(buf, PAGE_SIZE, "%d\n", slots); > +} > + > +static DEVICE_ATTR_RO(slots); > + > +static struct attribute *armv8_pmuv3_caps_attrs[] = { > + &dev_attr_slots.attr, > + NULL, > +}; > + > +static struct attribute_group armv8_pmuv3_caps_attr_group = { > + .name = "caps", > + .attrs = armv8_pmuv3_caps_attrs, > + .is_visible = armv8pmu_caps_attr_is_visible, > +}; > + > /* > * Perf Events' indices > */ > @@ -940,14 +985,11 @@ static void __armv8pmu_probe_pmu(void *info) > { > struct armv8pmu_probe_info *probe = info; > struct arm_pmu *cpu_pmu = probe->pmu; > - u64 dfr0; > u64 pmceid_raw[2]; > u32 pmceid[2]; > int pmuver; > > - dfr0 = read_sysreg(id_aa64dfr0_el1); > - pmuver = cpuid_feature_extract_unsigned_field(dfr0, > - ID_AA64DFR0_PMUVER_SHIFT); > + pmuver = armv8pmu_get_pmu_version(); > if (pmuver == 0xf || pmuver == 0) > return; > > @@ -994,7 +1036,8 @@ static int armv8pmu_probe_pmu(struct arm_pmu *cpu_pmu) > static int armv8_pmu_init(struct arm_pmu *cpu_pmu, char *name, > int (*map_event)(struct perf_event *event), > const struct attribute_group *events, > - const struct attribute_group *format) > + const struct attribute_group *format, > + const struct attribute_group *caps) > { > int ret = armv8pmu_probe_pmu(cpu_pmu); > if (ret) > @@ -1019,6 +1062,8 @@ static int armv8_pmu_init(struct arm_pmu *cpu_pmu, char *name, > events : &armv8_pmuv3_events_attr_group; > cpu_pmu->attr_groups[ARMPMU_ATTR_GROUP_FORMATS] = format ? > format : &armv8_pmuv3_format_attr_group; > + cpu_pmu->attr_groups[ARMPMU_ATTR_GROUP_CAPS] = caps ? > + caps : &armv8_pmuv3_caps_attr_group; > > return 0; > } > @@ -1026,97 +1071,97 @@ static int armv8_pmu_init(struct arm_pmu *cpu_pmu, char *name, > static int armv8_pmuv3_init(struct arm_pmu *cpu_pmu) > { > return armv8_pmu_init(cpu_pmu, "armv8_pmuv3", > - armv8_pmuv3_map_event, NULL, NULL); > + armv8_pmuv3_map_event, NULL, NULL, NULL); > } > > static int armv8_a34_pmu_init(struct arm_pmu *cpu_pmu) > { > return armv8_pmu_init(cpu_pmu, "armv8_cortex_a34", > - armv8_pmuv3_map_event, NULL, NULL); > + armv8_pmuv3_map_event, NULL, NULL, NULL); > } > > static int armv8_a35_pmu_init(struct arm_pmu *cpu_pmu) > { > return armv8_pmu_init(cpu_pmu, "armv8_cortex_a35", > - armv8_a53_map_event, NULL, NULL); > + armv8_a53_map_event, NULL, NULL, NULL); > } > > static int armv8_a53_pmu_init(struct arm_pmu *cpu_pmu) > { > return armv8_pmu_init(cpu_pmu, "armv8_cortex_a53", > - armv8_a53_map_event, NULL, NULL); > + armv8_a53_map_event, NULL, NULL, NULL); > } > > static int armv8_a55_pmu_init(struct arm_pmu *cpu_pmu) > { > return armv8_pmu_init(cpu_pmu, "armv8_cortex_a55", > - armv8_pmuv3_map_event, NULL, NULL); > + armv8_pmuv3_map_event, NULL, NULL, NULL); > } > > static int armv8_a57_pmu_init(struct arm_pmu *cpu_pmu) > { > return armv8_pmu_init(cpu_pmu, "armv8_cortex_a57", > - armv8_a57_map_event, NULL, NULL); > + armv8_a57_map_event, NULL, NULL, NULL); > } > > static int armv8_a65_pmu_init(struct arm_pmu *cpu_pmu) > { > return armv8_pmu_init(cpu_pmu, "armv8_cortex_a65", > - armv8_pmuv3_map_event, NULL, NULL); > + armv8_pmuv3_map_event, NULL, NULL, NULL); > } > > static int armv8_a72_pmu_init(struct arm_pmu *cpu_pmu) > { > return armv8_pmu_init(cpu_pmu, "armv8_cortex_a72", > - armv8_a57_map_event, NULL, NULL); > + armv8_a57_map_event, NULL, NULL, NULL); > } > > static int armv8_a73_pmu_init(struct arm_pmu *cpu_pmu) > { > return armv8_pmu_init(cpu_pmu, "armv8_cortex_a73", > - armv8_a73_map_event, NULL, NULL); > + armv8_a73_map_event, NULL, NULL, NULL); > } > > static int armv8_a75_pmu_init(struct arm_pmu *cpu_pmu) > { > return armv8_pmu_init(cpu_pmu, "armv8_cortex_a75", > - armv8_pmuv3_map_event, NULL, NULL); > + armv8_pmuv3_map_event, NULL, NULL, NULL); > } > > static int armv8_a76_pmu_init(struct arm_pmu *cpu_pmu) > { > return armv8_pmu_init(cpu_pmu, "armv8_cortex_a76", > - armv8_pmuv3_map_event, NULL, NULL); > + armv8_pmuv3_map_event, NULL, NULL, NULL); > } > > static int armv8_a77_pmu_init(struct arm_pmu *cpu_pmu) > { > return armv8_pmu_init(cpu_pmu, "armv8_cortex_a77", > - armv8_pmuv3_map_event, NULL, NULL); > + armv8_pmuv3_map_event, NULL, NULL, NULL); > } > > static int armv8_e1_pmu_init(struct arm_pmu *cpu_pmu) > { > return armv8_pmu_init(cpu_pmu, "armv8_neoverse_e1", > - armv8_pmuv3_map_event, NULL, NULL); > + armv8_pmuv3_map_event, NULL, NULL, NULL); > } > > static int armv8_n1_pmu_init(struct arm_pmu *cpu_pmu) > { > return armv8_pmu_init(cpu_pmu, "armv8_neoverse_n1", > - armv8_pmuv3_map_event, NULL, NULL); > + armv8_pmuv3_map_event, NULL, NULL, NULL); > } > > static int armv8_thunder_pmu_init(struct arm_pmu *cpu_pmu) > { > return armv8_pmu_init(cpu_pmu, "armv8_cavium_thunder", > - armv8_thunder_map_event, NULL, NULL); > + armv8_thunder_map_event, NULL, NULL, NULL); > } > > static int armv8_vulcan_pmu_init(struct arm_pmu *cpu_pmu) > { > return armv8_pmu_init(cpu_pmu, "armv8_brcm_vulcan", > - armv8_vulcan_map_event, NULL, NULL); > + armv8_vulcan_map_event, NULL, NULL, NULL); > } > > static const struct of_device_id armv8_pmu_of_device_ids[] = { > diff --git a/include/linux/perf/arm_pmu.h b/include/linux/perf/arm_pmu.h > index 5b616dde9a4c..1e129b57d51a 100644 > --- a/include/linux/perf/arm_pmu.h > +++ b/include/linux/perf/arm_pmu.h > @@ -73,6 +73,7 @@ enum armpmu_attr_groups { > ARMPMU_ATTR_GROUP_COMMON, > ARMPMU_ATTR_GROUP_EVENTS, > ARMPMU_ATTR_GROUP_FORMATS, > + ARMPMU_ATTR_GROUP_CAPS, > ARMPMU_NR_ATTR_GROUPS > }; > >
On Thu, Jun 18, 2020 at 09:35:42PM +0800, Shaokun Zhang wrote: > ARMv8.4-PMU introduces the PMMIR_EL1 registers and some new PMU events, > like STALL_SLOT etc, are related to it. Let's add a caps directory to > /sys/bus/event_source/devices/armv8_pmuv3_0/ and support slots from > PMMIR_EL1 registers in this entry. The user programs can get the slots > from sysfs directly. > > Cc: Will Deacon <will@kernel.org> > Cc: Mark Rutland <mark.rutland@arm.com> > Signed-off-by: Shaokun Zhang <zhangshaokun@hisilicon.com> > --- > ChangeLog in v3: > * Fix one typo in patch3 > > ChangeLog in v2: > * Add caps entry in sysfs > * Fix the PMU events typos > * Add one new patch to correct event ID in sysfs > > arch/arm64/include/asm/sysreg.h | 2 + > arch/arm64/kernel/perf_event.c | 87 +++++++++++++++++++++++++++++++---------- > include/linux/perf/arm_pmu.h | 1 + > 3 files changed, 69 insertions(+), 21 deletions(-) > > diff --git a/arch/arm64/include/asm/sysreg.h b/arch/arm64/include/asm/sysreg.h > index 463175f80341..56c45a9207c7 100644 > --- a/arch/arm64/include/asm/sysreg.h > +++ b/arch/arm64/include/asm/sysreg.h > @@ -321,6 +321,8 @@ > #define SYS_PMINTENSET_EL1 sys_reg(3, 0, 9, 14, 1) > #define SYS_PMINTENCLR_EL1 sys_reg(3, 0, 9, 14, 2) > > +#define SYS_PMMIR_EL1 sys_reg(3, 0, 9, 14, 6) > + > #define SYS_MAIR_EL1 sys_reg(3, 0, 10, 2, 0) > #define SYS_AMAIR_EL1 sys_reg(3, 0, 10, 3, 0) > > diff --git a/arch/arm64/kernel/perf_event.c b/arch/arm64/kernel/perf_event.c > index 4d7879484cec..5f2ac87e4b91 100644 > --- a/arch/arm64/kernel/perf_event.c > +++ b/arch/arm64/kernel/perf_event.c > @@ -277,6 +277,51 @@ static struct attribute_group armv8_pmuv3_format_attr_group = { > .attrs = armv8_pmuv3_format_attrs, > }; > > +static inline int armv8pmu_get_pmu_version(void) > +{ No need for 'inline' here. > + int pmuver; > + u64 dfr0; > + > + dfr0 = read_sysreg(id_aa64dfr0_el1); > + pmuver = cpuid_feature_extract_unsigned_field(dfr0, > + ID_AA64DFR0_PMUVER_SHIFT); > + > + return pmuver; > +} > + > +static umode_t > +armv8pmu_caps_attr_is_visible(struct kobject *kobj, struct attribute *attr, > + int unused) > +{ > + int pmuver = armv8pmu_get_pmu_version(); > + > + if (pmuver >= ID_AA64DFR0_PMUVER_8_4) > + return attr->mode; Is this sufficient? I'm a bit confused by the text in the Arm ARM that says: | If ARMv8.4-PMU is implemented: | * If STALL_SLOT is not implemented, it is IMPLEMENTATION DEFINED whether | the PMMIR System registers are implemented. | * If STALL_SLOT is implemented, then the PMMIR System registers are | implemented. whereas the register description for PMMIR_EL1 says: | This register is present only when ARMv8.4-PMU is implemented. Mark -- please could you clarify whether or not we need to check STALL_SLOT as well as the PMUVer? > + > + return 0; > +} > + > +static ssize_t slots_show(struct device *dev, struct device_attribute *attr, > + char *buf) > +{ > + int slots = read_sysreg_s(SYS_PMMIR_EL1) & 0xFF; > + > + return snprintf(buf, PAGE_SIZE, "%d\n", slots); > +} > + > +static DEVICE_ATTR_RO(slots); > + > +static struct attribute *armv8_pmuv3_caps_attrs[] = { > + &dev_attr_slots.attr, > + NULL, > +}; > + > +static struct attribute_group armv8_pmuv3_caps_attr_group = { > + .name = "caps", > + .attrs = armv8_pmuv3_caps_attrs, > + .is_visible = armv8pmu_caps_attr_is_visible, > +}; > + > /* > * Perf Events' indices > */ > @@ -940,14 +985,11 @@ static void __armv8pmu_probe_pmu(void *info) > { > struct armv8pmu_probe_info *probe = info; > struct arm_pmu *cpu_pmu = probe->pmu; > - u64 dfr0; > u64 pmceid_raw[2]; > u32 pmceid[2]; > int pmuver; > > - dfr0 = read_sysreg(id_aa64dfr0_el1); > - pmuver = cpuid_feature_extract_unsigned_field(dfr0, > - ID_AA64DFR0_PMUVER_SHIFT); > + pmuver = armv8pmu_get_pmu_version(); > if (pmuver == 0xf || pmuver == 0) > return; > > @@ -994,7 +1036,8 @@ static int armv8pmu_probe_pmu(struct arm_pmu *cpu_pmu) > static int armv8_pmu_init(struct arm_pmu *cpu_pmu, char *name, > int (*map_event)(struct perf_event *event), > const struct attribute_group *events, > - const struct attribute_group *format) > + const struct attribute_group *format, > + const struct attribute_group *caps) > { > int ret = armv8pmu_probe_pmu(cpu_pmu); > if (ret) > @@ -1019,6 +1062,8 @@ static int armv8_pmu_init(struct arm_pmu *cpu_pmu, char *name, > events : &armv8_pmuv3_events_attr_group; > cpu_pmu->attr_groups[ARMPMU_ATTR_GROUP_FORMATS] = format ? > format : &armv8_pmuv3_format_attr_group; > + cpu_pmu->attr_groups[ARMPMU_ATTR_GROUP_CAPS] = caps ? > + caps : &armv8_pmuv3_caps_attr_group; > > return 0; > } > @@ -1026,97 +1071,97 @@ static int armv8_pmu_init(struct arm_pmu *cpu_pmu, char *name, > static int armv8_pmuv3_init(struct arm_pmu *cpu_pmu) > { > return armv8_pmu_init(cpu_pmu, "armv8_pmuv3", > - armv8_pmuv3_map_event, NULL, NULL); > + armv8_pmuv3_map_event, NULL, NULL, NULL); Maybe we should add: static int armv8_pmu_init_nogroups(struct arm_pmu *cpu_pmu, char *name, int (*map_event)(struct perf_event *event)) { return armv8_pmu_init(cpu_pmu, name, map_event, NULL, NULL, NULL); } and then update all these CPU initialisers to use that instead? Will
On Mon, Jul 20, 2020 at 11:15:19AM +0100, Will Deacon wrote: > On Thu, Jun 18, 2020 at 09:35:42PM +0800, Shaokun Zhang wrote: > > +static umode_t > > +armv8pmu_caps_attr_is_visible(struct kobject *kobj, struct attribute *attr, > > + int unused) > > +{ > > + int pmuver = armv8pmu_get_pmu_version(); > > + > > + if (pmuver >= ID_AA64DFR0_PMUVER_8_4) > > + return attr->mode; > > Is this sufficient? I'm a bit confused by the text in the Arm ARM that says: > > | If ARMv8.4-PMU is implemented: > | * If STALL_SLOT is not implemented, it is IMPLEMENTATION DEFINED whether > | the PMMIR System registers are implemented. > | * If STALL_SLOT is implemented, then the PMMIR System registers are > | implemented. > > whereas the register description for PMMIR_EL1 says: > > | This register is present only when ARMv8.4-PMU is implemented. I think this is trying to say that when ARMv8.4-PMU is not implemented, PMMIR definitely isn't implemented (i.e. the the presence of PMMIR_EL1 implies the presence of ARMv8.4-PMU). > Mark -- please could you clarify whether or not we need to check STALL_SLOT > as well as the PMUVer? Given the explciit wording that it's IMP DEF, I suspect that we need to check both. I'll go chase this up. Mark.
On Mon, Jul 20, 2020 at 11:50:19AM +0100, Mark Rutland wrote: > On Mon, Jul 20, 2020 at 11:15:19AM +0100, Will Deacon wrote: > > On Thu, Jun 18, 2020 at 09:35:42PM +0800, Shaokun Zhang wrote: > > > > +static umode_t > > > +armv8pmu_caps_attr_is_visible(struct kobject *kobj, struct attribute *attr, > > > + int unused) > > > +{ > > > + int pmuver = armv8pmu_get_pmu_version(); > > > + > > > + if (pmuver >= ID_AA64DFR0_PMUVER_8_4) > > > + return attr->mode; > > > > Is this sufficient? I'm a bit confused by the text in the Arm ARM that says: > > > > | If ARMv8.4-PMU is implemented: > > | * If STALL_SLOT is not implemented, it is IMPLEMENTATION DEFINED whether > > | the PMMIR System registers are implemented. > > | * If STALL_SLOT is implemented, then the PMMIR System registers are > > | implemented. > > > > whereas the register description for PMMIR_EL1 says: > > > > | This register is present only when ARMv8.4-PMU is implemented. > > I think this is trying to say that when ARMv8.4-PMU is not implemented, > PMMIR definitely isn't implemented (i.e. the the presence of PMMIR_EL1 > implies the presence of ARMv8.4-PMU). > > > Mark -- please could you clarify whether or not we need to check STALL_SLOT > > as well as the PMUVer? > > Given the explciit wording that it's IMP DEF, I suspect that we need to > check both. > > I'll go chase this up. Thanks. In the meantime, Shaokun, can you send updated versions of the first two patches, please? We'll play it safe and check STALL_SLOT as well pending Mark's findings. Will
Hi Will, 在 2020/7/20 18:54, Will Deacon 写道: > On Mon, Jul 20, 2020 at 11:50:19AM +0100, Mark Rutland wrote: >> On Mon, Jul 20, 2020 at 11:15:19AM +0100, Will Deacon wrote: >>> On Thu, Jun 18, 2020 at 09:35:42PM +0800, Shaokun Zhang wrote: >> >>>> +static umode_t >>>> +armv8pmu_caps_attr_is_visible(struct kobject *kobj, struct attribute *attr, >>>> + int unused) >>>> +{ >>>> + int pmuver = armv8pmu_get_pmu_version(); >>>> + >>>> + if (pmuver >= ID_AA64DFR0_PMUVER_8_4) >>>> + return attr->mode; >>> >>> Is this sufficient? I'm a bit confused by the text in the Arm ARM that says: >>> >>> | If ARMv8.4-PMU is implemented: >>> | * If STALL_SLOT is not implemented, it is IMPLEMENTATION DEFINED whether >>> | the PMMIR System registers are implemented. >>> | * If STALL_SLOT is implemented, then the PMMIR System registers are >>> | implemented. >>> >>> whereas the register description for PMMIR_EL1 says: >>> >>> | This register is present only when ARMv8.4-PMU is implemented. >> >> I think this is trying to say that when ARMv8.4-PMU is not implemented, >> PMMIR definitely isn't implemented (i.e. the the presence of PMMIR_EL1 >> implies the presence of ARMv8.4-PMU). >> >>> Mark -- please could you clarify whether or not we need to check STALL_SLOT >>> as well as the PMUVer? >> >> Given the explciit wording that it's IMP DEF, I suspect that we need to >> check both. >> >> I'll go chase this up. > > Thanks. In the meantime, Shaokun, can you send updated versions of the first > two patches, please? We'll play it safe and check STALL_SLOT as well pending Ok, I will submit them soon addressed your other comments together. Thanks, Shaokun > Mark's findings. > > Will > > . >
Hi Will/Mark, 在 2020/7/20 18:54, Will Deacon 写道: > On Mon, Jul 20, 2020 at 11:50:19AM +0100, Mark Rutland wrote: >> On Mon, Jul 20, 2020 at 11:15:19AM +0100, Will Deacon wrote: >>> On Thu, Jun 18, 2020 at 09:35:42PM +0800, Shaokun Zhang wrote: >> >>>> +static umode_t >>>> +armv8pmu_caps_attr_is_visible(struct kobject *kobj, struct attribute *attr, >>>> + int unused) >>>> +{ >>>> + int pmuver = armv8pmu_get_pmu_version(); >>>> + >>>> + if (pmuver >= ID_AA64DFR0_PMUVER_8_4) >>>> + return attr->mode; >>> >>> Is this sufficient? I'm a bit confused by the text in the Arm ARM that says: >>> >>> | If ARMv8.4-PMU is implemented: >>> | * If STALL_SLOT is not implemented, it is IMPLEMENTATION DEFINED whether >>> | the PMMIR System registers are implemented. >>> | * If STALL_SLOT is implemented, then the PMMIR System registers are >>> | implemented. >>> >>> whereas the register description for PMMIR_EL1 says: >>> >>> | This register is present only when ARMv8.4-PMU is implemented. >> >> I think this is trying to say that when ARMv8.4-PMU is not implemented, >> PMMIR definitely isn't implemented (i.e. the the presence of PMMIR_EL1 >> implies the presence of ARMv8.4-PMU). >> >>> Mark -- please could you clarify whether or not we need to check STALL_SLOT >>> as well as the PMUVer? >> >> Given the explciit wording that it's IMP DEF, I suspect that we need to >> check both. >> >> I'll go chase this up. > > Thanks. In the meantime, Shaokun, can you send updated versions of the first > two patches, please? We'll play it safe and check STALL_SLOT as well pending In Arm ARM for PMMIR_EL1: SLOTS, bits [7:0] Operation width. The largest value by which the STALL_SLOT event might increment by in a single cycle. If the STALL_SLOT event is not implemented, this field might read as zero. And it's IMP DEF, even if the STALL_SLOT event is not implemented, and it can take the SLOTS value and we can display it in sysfs if I follow it correctly. Thanks, Shaokun > Mark's findings. > > Will > > . >
diff --git a/arch/arm64/include/asm/sysreg.h b/arch/arm64/include/asm/sysreg.h index 463175f80341..56c45a9207c7 100644 --- a/arch/arm64/include/asm/sysreg.h +++ b/arch/arm64/include/asm/sysreg.h @@ -321,6 +321,8 @@ #define SYS_PMINTENSET_EL1 sys_reg(3, 0, 9, 14, 1) #define SYS_PMINTENCLR_EL1 sys_reg(3, 0, 9, 14, 2) +#define SYS_PMMIR_EL1 sys_reg(3, 0, 9, 14, 6) + #define SYS_MAIR_EL1 sys_reg(3, 0, 10, 2, 0) #define SYS_AMAIR_EL1 sys_reg(3, 0, 10, 3, 0) diff --git a/arch/arm64/kernel/perf_event.c b/arch/arm64/kernel/perf_event.c index 4d7879484cec..5f2ac87e4b91 100644 --- a/arch/arm64/kernel/perf_event.c +++ b/arch/arm64/kernel/perf_event.c @@ -277,6 +277,51 @@ static struct attribute_group armv8_pmuv3_format_attr_group = { .attrs = armv8_pmuv3_format_attrs, }; +static inline int armv8pmu_get_pmu_version(void) +{ + int pmuver; + u64 dfr0; + + dfr0 = read_sysreg(id_aa64dfr0_el1); + pmuver = cpuid_feature_extract_unsigned_field(dfr0, + ID_AA64DFR0_PMUVER_SHIFT); + + return pmuver; +} + +static umode_t +armv8pmu_caps_attr_is_visible(struct kobject *kobj, struct attribute *attr, + int unused) +{ + int pmuver = armv8pmu_get_pmu_version(); + + if (pmuver >= ID_AA64DFR0_PMUVER_8_4) + return attr->mode; + + return 0; +} + +static ssize_t slots_show(struct device *dev, struct device_attribute *attr, + char *buf) +{ + int slots = read_sysreg_s(SYS_PMMIR_EL1) & 0xFF; + + return snprintf(buf, PAGE_SIZE, "%d\n", slots); +} + +static DEVICE_ATTR_RO(slots); + +static struct attribute *armv8_pmuv3_caps_attrs[] = { + &dev_attr_slots.attr, + NULL, +}; + +static struct attribute_group armv8_pmuv3_caps_attr_group = { + .name = "caps", + .attrs = armv8_pmuv3_caps_attrs, + .is_visible = armv8pmu_caps_attr_is_visible, +}; + /* * Perf Events' indices */ @@ -940,14 +985,11 @@ static void __armv8pmu_probe_pmu(void *info) { struct armv8pmu_probe_info *probe = info; struct arm_pmu *cpu_pmu = probe->pmu; - u64 dfr0; u64 pmceid_raw[2]; u32 pmceid[2]; int pmuver; - dfr0 = read_sysreg(id_aa64dfr0_el1); - pmuver = cpuid_feature_extract_unsigned_field(dfr0, - ID_AA64DFR0_PMUVER_SHIFT); + pmuver = armv8pmu_get_pmu_version(); if (pmuver == 0xf || pmuver == 0) return; @@ -994,7 +1036,8 @@ static int armv8pmu_probe_pmu(struct arm_pmu *cpu_pmu) static int armv8_pmu_init(struct arm_pmu *cpu_pmu, char *name, int (*map_event)(struct perf_event *event), const struct attribute_group *events, - const struct attribute_group *format) + const struct attribute_group *format, + const struct attribute_group *caps) { int ret = armv8pmu_probe_pmu(cpu_pmu); if (ret) @@ -1019,6 +1062,8 @@ static int armv8_pmu_init(struct arm_pmu *cpu_pmu, char *name, events : &armv8_pmuv3_events_attr_group; cpu_pmu->attr_groups[ARMPMU_ATTR_GROUP_FORMATS] = format ? format : &armv8_pmuv3_format_attr_group; + cpu_pmu->attr_groups[ARMPMU_ATTR_GROUP_CAPS] = caps ? + caps : &armv8_pmuv3_caps_attr_group; return 0; } @@ -1026,97 +1071,97 @@ static int armv8_pmu_init(struct arm_pmu *cpu_pmu, char *name, static int armv8_pmuv3_init(struct arm_pmu *cpu_pmu) { return armv8_pmu_init(cpu_pmu, "armv8_pmuv3", - armv8_pmuv3_map_event, NULL, NULL); + armv8_pmuv3_map_event, NULL, NULL, NULL); } static int armv8_a34_pmu_init(struct arm_pmu *cpu_pmu) { return armv8_pmu_init(cpu_pmu, "armv8_cortex_a34", - armv8_pmuv3_map_event, NULL, NULL); + armv8_pmuv3_map_event, NULL, NULL, NULL); } static int armv8_a35_pmu_init(struct arm_pmu *cpu_pmu) { return armv8_pmu_init(cpu_pmu, "armv8_cortex_a35", - armv8_a53_map_event, NULL, NULL); + armv8_a53_map_event, NULL, NULL, NULL); } static int armv8_a53_pmu_init(struct arm_pmu *cpu_pmu) { return armv8_pmu_init(cpu_pmu, "armv8_cortex_a53", - armv8_a53_map_event, NULL, NULL); + armv8_a53_map_event, NULL, NULL, NULL); } static int armv8_a55_pmu_init(struct arm_pmu *cpu_pmu) { return armv8_pmu_init(cpu_pmu, "armv8_cortex_a55", - armv8_pmuv3_map_event, NULL, NULL); + armv8_pmuv3_map_event, NULL, NULL, NULL); } static int armv8_a57_pmu_init(struct arm_pmu *cpu_pmu) { return armv8_pmu_init(cpu_pmu, "armv8_cortex_a57", - armv8_a57_map_event, NULL, NULL); + armv8_a57_map_event, NULL, NULL, NULL); } static int armv8_a65_pmu_init(struct arm_pmu *cpu_pmu) { return armv8_pmu_init(cpu_pmu, "armv8_cortex_a65", - armv8_pmuv3_map_event, NULL, NULL); + armv8_pmuv3_map_event, NULL, NULL, NULL); } static int armv8_a72_pmu_init(struct arm_pmu *cpu_pmu) { return armv8_pmu_init(cpu_pmu, "armv8_cortex_a72", - armv8_a57_map_event, NULL, NULL); + armv8_a57_map_event, NULL, NULL, NULL); } static int armv8_a73_pmu_init(struct arm_pmu *cpu_pmu) { return armv8_pmu_init(cpu_pmu, "armv8_cortex_a73", - armv8_a73_map_event, NULL, NULL); + armv8_a73_map_event, NULL, NULL, NULL); } static int armv8_a75_pmu_init(struct arm_pmu *cpu_pmu) { return armv8_pmu_init(cpu_pmu, "armv8_cortex_a75", - armv8_pmuv3_map_event, NULL, NULL); + armv8_pmuv3_map_event, NULL, NULL, NULL); } static int armv8_a76_pmu_init(struct arm_pmu *cpu_pmu) { return armv8_pmu_init(cpu_pmu, "armv8_cortex_a76", - armv8_pmuv3_map_event, NULL, NULL); + armv8_pmuv3_map_event, NULL, NULL, NULL); } static int armv8_a77_pmu_init(struct arm_pmu *cpu_pmu) { return armv8_pmu_init(cpu_pmu, "armv8_cortex_a77", - armv8_pmuv3_map_event, NULL, NULL); + armv8_pmuv3_map_event, NULL, NULL, NULL); } static int armv8_e1_pmu_init(struct arm_pmu *cpu_pmu) { return armv8_pmu_init(cpu_pmu, "armv8_neoverse_e1", - armv8_pmuv3_map_event, NULL, NULL); + armv8_pmuv3_map_event, NULL, NULL, NULL); } static int armv8_n1_pmu_init(struct arm_pmu *cpu_pmu) { return armv8_pmu_init(cpu_pmu, "armv8_neoverse_n1", - armv8_pmuv3_map_event, NULL, NULL); + armv8_pmuv3_map_event, NULL, NULL, NULL); } static int armv8_thunder_pmu_init(struct arm_pmu *cpu_pmu) { return armv8_pmu_init(cpu_pmu, "armv8_cavium_thunder", - armv8_thunder_map_event, NULL, NULL); + armv8_thunder_map_event, NULL, NULL, NULL); } static int armv8_vulcan_pmu_init(struct arm_pmu *cpu_pmu) { return armv8_pmu_init(cpu_pmu, "armv8_brcm_vulcan", - armv8_vulcan_map_event, NULL, NULL); + armv8_vulcan_map_event, NULL, NULL, NULL); } static const struct of_device_id armv8_pmu_of_device_ids[] = { diff --git a/include/linux/perf/arm_pmu.h b/include/linux/perf/arm_pmu.h index 5b616dde9a4c..1e129b57d51a 100644 --- a/include/linux/perf/arm_pmu.h +++ b/include/linux/perf/arm_pmu.h @@ -73,6 +73,7 @@ enum armpmu_attr_groups { ARMPMU_ATTR_GROUP_COMMON, ARMPMU_ATTR_GROUP_EVENTS, ARMPMU_ATTR_GROUP_FORMATS, + ARMPMU_ATTR_GROUP_CAPS, ARMPMU_NR_ATTR_GROUPS };
ARMv8.4-PMU introduces the PMMIR_EL1 registers and some new PMU events, like STALL_SLOT etc, are related to it. Let's add a caps directory to /sys/bus/event_source/devices/armv8_pmuv3_0/ and support slots from PMMIR_EL1 registers in this entry. The user programs can get the slots from sysfs directly. Cc: Will Deacon <will@kernel.org> Cc: Mark Rutland <mark.rutland@arm.com> Signed-off-by: Shaokun Zhang <zhangshaokun@hisilicon.com> --- ChangeLog in v3: * Fix one typo in patch3 ChangeLog in v2: * Add caps entry in sysfs * Fix the PMU events typos * Add one new patch to correct event ID in sysfs arch/arm64/include/asm/sysreg.h | 2 + arch/arm64/kernel/perf_event.c | 87 +++++++++++++++++++++++++++++++---------- include/linux/perf/arm_pmu.h | 1 + 3 files changed, 69 insertions(+), 21 deletions(-)