Skip to contentSkip to conversation

drivers/perf: Add support for ARMv8.3-SPE

PATCH Superseded Wei Li 2 participants 4 messages Sep 30, 2020 – Nov 25, 2020

Subsystems (2)

v2 of 4 Superseded Latest is v4 →

lore
Share & tools
lore

Wei Li

author
#0 lore raw

PATCH v2 1/1 drivers/perf: Add support for ARMv8.3-SPE

Armv8.3 extends the SPE by adding:
- Alignment field in the Events packet, and filtering on this event using PMSEVFR_EL1.
- Support for the Scalable Vector Extension (SVE).

The main additions for SVE are:
- Recording the vector length for SVE operations in the Operation Type packet. It is not possible to filter on vector length.
- Incomplete predicate and empty predicate fields in the Events packet, and filtering on these events using PMSEVFR_EL1.

Update the check of pmsevfr for empty/partial predicated SVE and alignment event in SPE driver. For adaption by the version of SPE, expose 'pmsver' as cap attribute to userspace.

---
v1 -> v2:
- Rename 'pmuver' to 'pmsver', change it's type to 'u16' from 'int'. (Suggested by Will and Leo.)
- Expose 'pmsver' as cap attribute through sysfs, instead of printing. (Suggested by Will.)

---
 arch/arm64/include/asm/sysreg.h |  4 +++-
 drivers/perf/arm_spe_pmu.c      | 18 ++++++++++++++++--
 2 files changed, 19 insertions(+), 3 deletions(-)
arch/arm64/include/asm/sysreg.h +3 -1
index 554a7e8ecb07..f4f9c1fc6398 100644
@@ -281,7 +281,6 @@
 #define SYS_PMSFCR_EL1_ST_SHIFT		18
 
 #define SYS_PMSEVFR_EL1			sys_reg(3, 0, 9, 9, 5)
-#define SYS_PMSEVFR_EL1_RES0		0x0000ffff00ff0f55UL
 
 #define SYS_PMSLATFR_EL1		sys_reg(3, 0, 9, 9, 6)
 #define SYS_PMSLATFR_EL1_MINLAT_SHIFT	0
@@ -787,6 +786,9 @@
 #define ID_AA64DFR0_PMUVER_8_5		0x6
 #define ID_AA64DFR0_PMUVER_IMP_DEF	0xf
 
+#define ID_AA64DFR0_PMSVER_8_2		0x1
+#define ID_AA64DFR0_PMSVER_8_3		0x2
+
 #define ID_DFR0_PERFMON_SHIFT		24
 
 #define ID_DFR0_PERFMON_8_1		0x4
drivers/perf/arm_spe_pmu.c +16 -2
index cc00915ad6d1..52e7869f5621 100644
@@ -54,7 +54,7 @@ struct arm_spe_pmu {
 	struct hlist_node			hotplug_node;
 
 	int					irq; /* PPI */
-
+	u16					pmsver;
 	u16					min_period;
 	u16					counter_sz;
 
@@ -80,6 +80,15 @@ struct arm_spe_pmu {
 /* Keep track of our dynamic hotplug state */
 static enum cpuhp_state arm_spe_pmu_online;
 
+static u64 sys_pmsevfr_el1_mask[] = {
+	[ID_AA64DFR0_PMSVER_8_2] = GENMASK_ULL(63, 48) | GENMASK_ULL(31, 24) |
+		GENMASK_ULL(15, 12) | BIT_ULL(7) | BIT_ULL(5) | BIT_ULL(3) |
+		BIT_ULL(1),
+	[ID_AA64DFR0_PMSVER_8_3] = GENMASK_ULL(63, 48) | GENMASK_ULL(31, 24) |
+		GENMASK_ULL(18, 17) | GENMASK_ULL(15, 11) | BIT_ULL(7) |
+		BIT_ULL(5) | BIT_ULL(3) | BIT_ULL(1),
+};
+
 enum arm_spe_pmu_buf_fault_action {
 	SPE_PMU_BUF_FAULT_ACT_SPURIOUS,
 	SPE_PMU_BUF_FAULT_ACT_FATAL,
@@ -93,6 +102,7 @@ enum arm_spe_pmu_capabilities {
 	SPE_PMU_CAP_FEAT_MAX,
 	SPE_PMU_CAP_CNT_SZ = SPE_PMU_CAP_FEAT_MAX,
 	SPE_PMU_CAP_MIN_IVAL,
+	SPE_PMU_CAP_PMSVER,
 };
 
 static int arm_spe_pmu_feat_caps[SPE_PMU_CAP_FEAT_MAX] = {
@@ -110,6 +120,8 @@ static u32 arm_spe_pmu_cap_get(struct arm_spe_pmu *spe_pmu, int cap)
 		return spe_pmu->counter_sz;
 	case SPE_PMU_CAP_MIN_IVAL:
 		return spe_pmu->min_period;
+	case SPE_PMU_CAP_PMSVER:
+		return spe_pmu->pmsver;
 	default:
 		WARN(1, "unknown cap %d\n", cap);
 	}
@@ -143,6 +155,7 @@ static struct attribute *arm_spe_pmu_cap_attr[] = {
 	SPE_CAP_EXT_ATTR_ENTRY(ernd, SPE_PMU_CAP_ERND),
 	SPE_CAP_EXT_ATTR_ENTRY(count_size, SPE_PMU_CAP_CNT_SZ),
 	SPE_CAP_EXT_ATTR_ENTRY(min_interval, SPE_PMU_CAP_MIN_IVAL),
+	SPE_CAP_EXT_ATTR_ENTRY(pmsver, SPE_PMU_CAP_PMSVER),
 	NULL,
 };
 
@@ -670,7 +683,7 @@ static int arm_spe_pmu_event_init(struct perf_event *event)
 	    !cpumask_test_cpu(event->cpu, &spe_pmu->supported_cpus))
 		return -ENOENT;
 
-	if (arm_spe_event_to_pmsevfr(event) & SYS_PMSEVFR_EL1_RES0)
+	if (arm_spe_event_to_pmsevfr(event) & ~sys_pmsevfr_el1_mask[spe_pmu->pmsver])
 		return -EOPNOTSUPP;
 
 	if (attr->exclude_idle)
@@ -937,6 +950,7 @@ static void __arm_spe_pmu_dev_probe(void *info)
 			fld, smp_processor_id());
 		return;
 	}
+	spe_pmu->pmsver = (u16)fld;
 
 	/* Read PMBIDR first to determine whether or not we have access */
 	reg = read_sysreg_s(SYS_PMBIDR_EL1);
Signature · 1 line
2.17.1
maintainer
#1 lore raw
Quoted 64 lines · Wei Li

Armv8.3 extends the SPE by adding:
- Alignment field in the Events packet, and filtering on this event using PMSEVFR_EL1.
- Support for the Scalable Vector Extension (SVE).

The main additions for SVE are:
- Recording the vector length for SVE operations in the Operation Type packet. It is not possible to filter on vector length.
- Incomplete predicate and empty predicate fields in the Events packet, and filtering on these events using PMSEVFR_EL1.

Update the check of pmsevfr for empty/partial predicated SVE and alignment event in SPE driver. For adaption by the version of SPE, expose 'pmsver' as cap attribute to userspace.

---
v1 -> v2:
- Rename 'pmuver' to 'pmsver', change it's type to 'u16' from 'int'. (Suggested by Will and Leo.)
- Expose 'pmsver' as cap attribute through sysfs, instead of printing. (Suggested by Will.)

---
 arch/arm64/include/asm/sysreg.h |  4 +++-
 drivers/perf/arm_spe_pmu.c      | 18 ++++++++++++++++--
 2 files changed, 19 insertions(+), 3 deletions(-)
arch/arm64/include/asm/sysreg.h +3 -1
index 554a7e8ecb07..f4f9c1fc6398 100644
@@ -281,7 +281,6 @@
 #define SYS_PMSFCR_EL1_ST_SHIFT		18
 
 #define SYS_PMSEVFR_EL1			sys_reg(3, 0, 9, 9, 5)
-#define SYS_PMSEVFR_EL1_RES0		0x0000ffff00ff0f55UL
 
 #define SYS_PMSLATFR_EL1		sys_reg(3, 0, 9, 9, 6)
 #define SYS_PMSLATFR_EL1_MINLAT_SHIFT	0
@@ -787,6 +786,9 @@
 #define ID_AA64DFR0_PMUVER_8_5		0x6
 #define ID_AA64DFR0_PMUVER_IMP_DEF	0xf
 
+#define ID_AA64DFR0_PMSVER_8_2		0x1
+#define ID_AA64DFR0_PMSVER_8_3		0x2
+
 #define ID_DFR0_PERFMON_SHIFT		24
 
 #define ID_DFR0_PERFMON_8_1		0x4
drivers/perf/arm_spe_pmu.c +1 -1
index cc00915ad6d1..52e7869f5621 100644
@@ -54,7 +54,7 @@ struct arm_spe_pmu {
 	struct hlist_node			hotplug_node;
 
 	int					irq; /* PPI */
-
+	u16					pmsver;
 	u16					min_period;
 	u16					counter_sz;
 
@@ -80,6 +80,15 @@ struct arm_spe_pmu {
 /* Keep track of our dynamic hotplug state */

drivers/perf/arm_spe_pmu.c struct arm_spe_pmu { · line 90

Quoted 10 lines
drivers/perf/arm_spe_pmu.c +8 -0
 static enum cpuhp_state arm_spe_pmu_online;
 
+static u64 sys_pmsevfr_el1_mask[] = {
+	[ID_AA64DFR0_PMSVER_8_2] = GENMASK_ULL(63, 48) | GENMASK_ULL(31, 24) |
+		GENMASK_ULL(15, 12) | BIT_ULL(7) | BIT_ULL(5) | BIT_ULL(3) |
+		BIT_ULL(1),
+	[ID_AA64DFR0_PMSVER_8_3] = GENMASK_ULL(63, 48) | GENMASK_ULL(31, 24) |
+		GENMASK_ULL(18, 17) | GENMASK_ULL(15, 11) | BIT_ULL(7) |
+		BIT_ULL(5) | BIT_ULL(3) | BIT_ULL(1),
+};

Ok, so I finally figured out what I don't like about this: it's the fact that the RES0 mask only ever reduces, but we have no way of ensuring that by construction with this approach. In other words, it's a bit brittle to keep all of these things defined entirely separately from one another.

How about a small change so that we define things like:

#define SYS_PMSEVFR_EL1_RES0_8_2	SYS_PMSEVFR_EL1_RES0 &
					~(...)
#define SYS_PMSEVFR_EL1_RES0_8_3	SYS_PMSEVFR_EL1_RES0_8_2 &
					~(...)

where the '...' parts identify the bits that are no longer RES0 for that version of the architecture?

What do you think?

Will

author
#2 lore raw

Hi Will,

Quoted 83 lines · Will Deacon
Quoted 64 lines · Wei Li

Armv8.3 extends the SPE by adding:
- Alignment field in the Events packet, and filtering on this event using PMSEVFR_EL1.
- Support for the Scalable Vector Extension (SVE).

The main additions for SVE are:
- Recording the vector length for SVE operations in the Operation Type packet. It is not possible to filter on vector length.
- Incomplete predicate and empty predicate fields in the Events packet, and filtering on these events using PMSEVFR_EL1.

Update the check of pmsevfr for empty/partial predicated SVE and alignment event in SPE driver. For adaption by the version of SPE, expose 'pmsver' as cap attribute to userspace.

---
v1 -> v2:
- Rename 'pmuver' to 'pmsver', change it's type to 'u16' from 'int'. (Suggested by Will and Leo.)
- Expose 'pmsver' as cap attribute through sysfs, instead of printing. (Suggested by Will.)

---
 arch/arm64/include/asm/sysreg.h |  4 +++-
 drivers/perf/arm_spe_pmu.c      | 18 ++++++++++++++++--
 2 files changed, 19 insertions(+), 3 deletions(-)
arch/arm64/include/asm/sysreg.h +3 -1
index 554a7e8ecb07..f4f9c1fc6398 100644
@@ -281,7 +281,6 @@
 #define SYS_PMSFCR_EL1_ST_SHIFT		18
 
 #define SYS_PMSEVFR_EL1			sys_reg(3, 0, 9, 9, 5)
-#define SYS_PMSEVFR_EL1_RES0		0x0000ffff00ff0f55UL
 
 #define SYS_PMSLATFR_EL1		sys_reg(3, 0, 9, 9, 6)
 #define SYS_PMSLATFR_EL1_MINLAT_SHIFT	0
@@ -787,6 +786,9 @@
 #define ID_AA64DFR0_PMUVER_8_5		0x6
 #define ID_AA64DFR0_PMUVER_IMP_DEF	0xf
 
+#define ID_AA64DFR0_PMSVER_8_2		0x1
+#define ID_AA64DFR0_PMSVER_8_3		0x2
+
 #define ID_DFR0_PERFMON_SHIFT		24
 
 #define ID_DFR0_PERFMON_8_1		0x4
drivers/perf/arm_spe_pmu.c +1 -1
index cc00915ad6d1..52e7869f5621 100644
@@ -54,7 +54,7 @@ struct arm_spe_pmu {
 	struct hlist_node			hotplug_node;
 
 	int					irq; /* PPI */
-
+	u16					pmsver;
 	u16					min_period;
 	u16					counter_sz;
 
@@ -80,6 +80,15 @@ struct arm_spe_pmu {
 /* Keep track of our dynamic hotplug state */

drivers/perf/arm_spe_pmu.c struct arm_spe_pmu { · line 90

Quoted 10 lines
drivers/perf/arm_spe_pmu.c +8 -0
 static enum cpuhp_state arm_spe_pmu_online;
 
+static u64 sys_pmsevfr_el1_mask[] = {
+	[ID_AA64DFR0_PMSVER_8_2] = GENMASK_ULL(63, 48) | GENMASK_ULL(31, 24) |
+		GENMASK_ULL(15, 12) | BIT_ULL(7) | BIT_ULL(5) | BIT_ULL(3) |
+		BIT_ULL(1),
+	[ID_AA64DFR0_PMSVER_8_3] = GENMASK_ULL(63, 48) | GENMASK_ULL(31, 24) |
+		GENMASK_ULL(18, 17) | GENMASK_ULL(15, 11) | BIT_ULL(7) |
+		BIT_ULL(5) | BIT_ULL(3) | BIT_ULL(1),
+};

Ok, so I finally figured out what I don't like about this: it's the fact that the RES0 mask only ever reduces, but we have no way of ensuring that by construction with this approach. In other words, it's a bit brittle to keep all of these things defined entirely separately from one another.

How about a small change so that we define things like:

Quoted 9 lines
#define SYS_PMSEVFR_EL1_RES0_8_2	SYS_PMSEVFR_EL1_RES0 &
					~(...)
#define SYS_PMSEVFR_EL1_RES0_8_3	SYS_PMSEVFR_EL1_RES0_8_2 &
					~(...)

where the '...' parts identify the bits that are no longer RES0 for that version of the architecture?

Sorry for the long delay.

These is also an array-index-out-of-bounds issue when accessing 'sys_pmsevfr_el1_mask', if the pmsver read in the future is bigger than what the driver supports.

So how about change to:

static u64 arm_spe_pmsevfr_mask(u16 pmsver)

{
	u64 mask = 0;

	if (pmsver >= ID_AA64DFR0_PMSVER_8_3)
		mask |= BIT_ULL(18) | BIT_ULL(17) | BIT_ULL(11);
	if (pmsver >= ID_AA64DFR0_PMSVER_8_2)
		mask |= GENMASK_ULL(63, 48) | GENMASK_ULL(31, 24) |
			GENMASK_ULL(15, 12) | BIT_ULL(7) | BIT_ULL(5) | BIT_ULL(3) |
			BIT_ULL(1);
	return mask;
}

Thus, the driver can try its best to support, and the definition is clear enough to show the difference between versions of SPE.

Or should i still define them as what you advised and add a check of pmsver to just serve the versions what the driver support?

Thanks,
Wei

maintainer
#3 lore raw
Quoted 52 lines · liwei (GF)
Quoted 21 lines · Will Deacon

drivers/perf/arm_spe_pmu.c struct arm_spe_pmu { · line 90

Quoted 12 lines · Wei Li
drivers/perf/arm_spe_pmu.c +8 -0
@@ -80,6 +80,15 @@ struct arm_spe_pmu {
 /* Keep track of our dynamic hotplug state */
 static enum cpuhp_state arm_spe_pmu_online;
 
+static u64 sys_pmsevfr_el1_mask[] = {
+	[ID_AA64DFR0_PMSVER_8_2] = GENMASK_ULL(63, 48) | GENMASK_ULL(31, 24) |
+		GENMASK_ULL(15, 12) | BIT_ULL(7) | BIT_ULL(5) | BIT_ULL(3) |
+		BIT_ULL(1),
+	[ID_AA64DFR0_PMSVER_8_3] = GENMASK_ULL(63, 48) | GENMASK_ULL(31, 24) |
+		GENMASK_ULL(18, 17) | GENMASK_ULL(15, 11) | BIT_ULL(7) |
+		BIT_ULL(5) | BIT_ULL(3) | BIT_ULL(1),
+};

Ok, so I finally figured out what I don't like about this: it's the fact that the RES0 mask only ever reduces, but we have no way of ensuring that by construction with this approach. In other words, it's a bit brittle to keep all of these things defined entirely separately from one another.

How about a small change so that we define things like:

Quoted 9 lines
#define SYS_PMSEVFR_EL1_RES0_8_2	SYS_PMSEVFR_EL1_RES0 &
					~(...)
#define SYS_PMSEVFR_EL1_RES0_8_3	SYS_PMSEVFR_EL1_RES0_8_2 &
					~(...)

where the '...' parts identify the bits that are no longer RES0 for that version of the architecture?

Sorry for the long delay.

These is also an array-index-out-of-bounds issue when accessing 'sys_pmsevfr_el1_mask', if the pmsver read in the future is bigger than what the driver supports.

So how about change to:

static u64 arm_spe_pmsevfr_mask(u16 pmsver)

{
	u64 mask = 0;

	if (pmsver >= ID_AA64DFR0_PMSVER_8_3)
		mask |= BIT_ULL(18) | BIT_ULL(17) | BIT_ULL(11);
	if (pmsver >= ID_AA64DFR0_PMSVER_8_2)
		mask |= GENMASK_ULL(63, 48) | GENMASK_ULL(31, 24) |
			GENMASK_ULL(15, 12) | BIT_ULL(7) | BIT_ULL(5) | BIT_ULL(3) |
			BIT_ULL(1);
	return mask;
}
Quoted 5 lines

Thus, the driver can try its best to support, and the definition is clear enough to show the difference between versions of SPE.

Or should i still define them as what you advised and add a check of pmsver to just serve the versions what the driver support?

I think I'd prefer that, yes.

Will