[PATCH v2 1/6] powerpc/perf: Factor of event_alternative function

[PATCH v2 1/6] powerpc/perf: Factor of event_alternative function

[Madhavan Srinivasan]

Factor out the power8 event_alternative function to share
the code with power9.

Signed-off-by: Madhavan Srinivasan <maddy@linux.vnet.ibm.com>
---
Changelog v1:
No changes to this patch, just a rebase

 arch/powerpc/perf/isa207-common.c | 36 ++++++++++++++++++++++++++++++++++++
 arch/powerpc/perf/isa207-common.h |  3 +++
 arch/powerpc/perf/power8-pmu.c    | 35 ++---------------------------------
 3 files changed, 41 insertions(+), 33 deletions(-)

diff --git a/arch/powerpc/perf/isa207-common.c b/arch/powerpc/perf/isa207-common.c
index 50e598cf644b..a86fadee352b 100644
--- a/arch/powerpc/perf/isa207-common.c
+++ b/arch/powerpc/perf/isa207-common.c
@@ -338,3 +338,39 @@ void isa207_disable_pmc(unsigned int pmc, unsigned long mmcr[])
 	if (pmc <= 3)
 		mmcr[1] &= ~(0xffUL << MMCR1_PMCSEL_SHIFT(pmc + 1));
 }
+
+static int find_alternative(u64 event, const unsigned int ev_alt[][MAX_ALT], int size)
+{
+	int i, j;
+
+	for (i = 0; i < size; ++i) {
+		if (event < ev_alt[i][0])
+			break;
+
+		for (j = 0; j < MAX_ALT && ev_alt[i][j]; ++j)
+			if (event == ev_alt[i][j])
+				return i;
+	}
+
+	return -1;
+}
+
+int isa207_get_alternatives(u64 event, u64 alt[],
+				const unsigned int ev_alt[][MAX_ALT], int size)
+{
+	int i, j, num_alt = 0;
+	u64 alt_event;
+
+	alt[num_alt++] = event;
+	i = find_alternative(event, ev_alt, size);
+	if (i >= 0) {
+		/* Filter out the original event, it's already in alt[0] */
+		for (j = 0; j < MAX_ALT; ++j) {
+			alt_event = ev_alt[i][j];
+			if (alt_event && alt_event != event)
+				alt[num_alt++] = alt_event;
+		}
+	}
+
+	return num_alt;
+}
diff --git a/arch/powerpc/perf/isa207-common.h b/arch/powerpc/perf/isa207-common.h
index 90495f1580c7..3e9150f6690a 100644
--- a/arch/powerpc/perf/isa207-common.h
+++ b/arch/powerpc/perf/isa207-common.h
@@ -260,5 +260,8 @@ int isa207_compute_mmcr(u64 event[], int n_ev,
 				unsigned int hwc[], unsigned long mmcr[],
 				struct perf_event *pevents[]);
 void isa207_disable_pmc(unsigned int pmc, unsigned long mmcr[]);
+int isa207_get_alternatives(u64 event, u64 alt[],
+				const unsigned int ev_alt[][MAX_ALT], int size);
+
 
 #endif
diff --git a/arch/powerpc/perf/power8-pmu.c b/arch/powerpc/perf/power8-pmu.c
index d07186382f3a..ce15b19a7962 100644
--- a/arch/powerpc/perf/power8-pmu.c
+++ b/arch/powerpc/perf/power8-pmu.c
@@ -48,43 +48,12 @@ static const unsigned int event_alternatives[][MAX_ALT] = {
 	{ PM_RUN_INST_CMPL_ALT,		PM_RUN_INST_CMPL },
 };
 
-/*
- * Scan the alternatives table for a match and return the
- * index into the alternatives table if found, else -1.
- */
-static int find_alternative(u64 event)
-{
-	int i, j;
-
-	for (i = 0; i < ARRAY_SIZE(event_alternatives); ++i) {
-		if (event < event_alternatives[i][0])
-			break;
-
-		for (j = 0; j < MAX_ALT && event_alternatives[i][j]; ++j)
-			if (event == event_alternatives[i][j])
-				return i;
-	}
-
-	return -1;
-}
-
 static int power8_get_alternatives(u64 event, unsigned int flags, u64 alt[])
 {
 	int i, j, num_alt = 0;
-	u64 alt_event;
-
-	alt[num_alt++] = event;
-
-	i = find_alternative(event);
-	if (i >= 0) {
-		/* Filter out the original event, it's already in alt[0] */
-		for (j = 0; j < MAX_ALT; ++j) {
-			alt_event = event_alternatives[i][j];
-			if (alt_event && alt_event != event)
-				alt[num_alt++] = alt_event;
-		}
-	}
 
+	num_alt = isa207_get_alternatives(event, alt, event_alternatives,
+					(int)ARRAY_SIZE(event_alternatives));
 	if (flags & PPMU_ONLY_COUNT_RUN) {
 		/*
 		 * We're only counting in RUN state, so PM_CYC is equivalent to
-- 
2.7.4

[PATCH v2 2/6] powerpc/perf: Add PM_INST_DISP event to Power9 event list

[Madhavan Srinivasan]

Signed-off-by: Madhavan Srinivasan <maddy@linux.vnet.ibm.com>
---
Changelog v1:
Fix the event code.

 arch/powerpc/perf/power9-events-list.h | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/arch/powerpc/perf/power9-events-list.h b/arch/powerpc/perf/power9-events-list.h
index 929b56d47ad9..71a6bfee5c02 100644
--- a/arch/powerpc/perf/power9-events-list.h
+++ b/arch/powerpc/perf/power9-events-list.h
@@ -53,3 +53,6 @@ EVENT(PM_ITLB_MISS,				0x400fc)
 EVENT(PM_RUN_INST_CMPL,				0x500fa)
 /* Run_cycles */
 EVENT(PM_RUN_CYC,				0x600f4)
+/* Instruction Dispatched */
+EVENT(PM_INST_DISP,				0x200f2)
+EVENT(PM_INST_DISP_ALT,				0x300f2)
-- 
2.7.4

[PATCH v2 3/6] powerpc/perf: Add alternative event table and function for power9

[Madhavan Srinivasan]

Signed-off-by: Madhavan Srinivasan <maddy@linux.vnet.ibm.com>
---
Change v1:
No changes, just a rebase

 arch/powerpc/perf/power9-pmu.c | 17 +++++++++++++++++
 1 file changed, 17 insertions(+)

diff --git a/arch/powerpc/perf/power9-pmu.c b/arch/powerpc/perf/power9-pmu.c
index 7332634e18c9..b38acff8a791 100644
--- a/arch/powerpc/perf/power9-pmu.c
+++ b/arch/powerpc/perf/power9-pmu.c
@@ -106,6 +106,21 @@ enum {
 /* PowerISA v2.07 format attribute structure*/
 extern struct attribute_group isa207_pmu_format_group;
 
+/* Table of alternatives, sorted by column 0 */
+static const unsigned int power9_event_alternatives[][MAX_ALT] = {
+	{ PM_INST_DISP,			PM_INST_DISP_ALT },
+};
+
+static int power9_get_alternatives(u64 event, unsigned int flags, u64 alt[])
+{
+	int num_alt = 0;
+
+	num_alt = isa207_get_alternatives(event, alt, power9_event_alternatives,
+				(int)ARRAY_SIZE(power9_event_alternatives));
+
+	return num_alt;
+}
+
 GENERIC_EVENT_ATTR(cpu-cycles,			PM_CYC);
 GENERIC_EVENT_ATTR(stalled-cycles-frontend,	PM_ICT_NOSLOT_CYC);
 GENERIC_EVENT_ATTR(stalled-cycles-backend,	PM_CMPLU_STALL);
@@ -383,6 +398,7 @@ static struct power_pmu power9_isa207_pmu = {
 	.config_bhrb		= power9_config_bhrb,
 	.bhrb_filter_map	= power9_bhrb_filter_map,
 	.get_constraint		= isa207_get_constraint,
+	.get_alternatives	= power9_get_alternatives,
 	.disable_pmc		= isa207_disable_pmc,
 	.flags			= PPMU_NO_SIAR | PPMU_ARCH_207S,
 	.n_generic		= ARRAY_SIZE(power9_generic_events),
@@ -401,6 +417,7 @@ static struct power_pmu power9_pmu = {
 	.config_bhrb		= power9_config_bhrb,
 	.bhrb_filter_map	= power9_bhrb_filter_map,
 	.get_constraint		= isa207_get_constraint,
+	.get_alternatives	= power9_get_alternatives,
 	.disable_pmc		= isa207_disable_pmc,
 	.flags			= PPMU_HAS_SIER | PPMU_ARCH_207S,
 	.n_generic		= ARRAY_SIZE(power9_generic_events),
-- 
2.7.4

[PATCH v2 4/6] powerpc/perf: Use PM_INST_DISP for generic instructions sample

[Madhavan Srinivasan]

Since PM_INST_CMPL may not provide right counts in all
sampling scenarios in power9 DD1, instead use PM_INST_DISP.
Patch also update generic instruction sampling with the same.

Signed-off-by: Madhavan Srinivasan <maddy@linux.vnet.ibm.com>
---
Changelog v1:
Based on DD1 check, modified the event code to use for "instructions"

 arch/powerpc/perf/power9-pmu.c | 20 ++++++++++++++++++--
 1 file changed, 18 insertions(+), 2 deletions(-)

diff --git a/arch/powerpc/perf/power9-pmu.c b/arch/powerpc/perf/power9-pmu.c
index b38acff8a791..454e9f70894f 100644
--- a/arch/powerpc/perf/power9-pmu.c
+++ b/arch/powerpc/perf/power9-pmu.c
@@ -228,6 +228,17 @@ static const struct attribute_group *power9_pmu_attr_groups[] = {
 	NULL,
 };
 
+static int power9_generic_events_dd1[] = {
+	[PERF_COUNT_HW_CPU_CYCLES] =			PM_CYC,
+	[PERF_COUNT_HW_STALLED_CYCLES_FRONTEND] =	PM_ICT_NOSLOT_CYC,
+	[PERF_COUNT_HW_STALLED_CYCLES_BACKEND] =	PM_CMPLU_STALL,
+	[PERF_COUNT_HW_INSTRUCTIONS] =			PM_INST_DISP,
+	[PERF_COUNT_HW_BRANCH_INSTRUCTIONS] =		PM_BRU_CMPL,
+	[PERF_COUNT_HW_BRANCH_MISSES] =			PM_BR_MPRED_CMPL,
+	[PERF_COUNT_HW_CACHE_REFERENCES] =		PM_LD_REF_L1,
+	[PERF_COUNT_HW_CACHE_MISSES] =			PM_LD_MISS_L1_FIN,
+};
+
 static int power9_generic_events[] = {
 	[PERF_COUNT_HW_CPU_CYCLES] =			PM_CYC,
 	[PERF_COUNT_HW_STALLED_CYCLES_FRONTEND] =	PM_ICT_NOSLOT_CYC,
@@ -401,8 +412,8 @@ static struct power_pmu power9_isa207_pmu = {
 	.get_alternatives	= power9_get_alternatives,
 	.disable_pmc		= isa207_disable_pmc,
 	.flags			= PPMU_NO_SIAR | PPMU_ARCH_207S,
-	.n_generic		= ARRAY_SIZE(power9_generic_events),
-	.generic_events		= power9_generic_events,
+	.n_generic		= ARRAY_SIZE(power9_generic_events_dd1),
+	.generic_events		= power9_generic_events_dd1,
 	.cache_events		= &power9_cache_events,
 	.attr_groups		= power9_isa207_pmu_attr_groups,
 	.bhrb_nr		= 32,
@@ -437,6 +448,11 @@ static int __init init_power9_pmu(void)
 		return -ENODEV;
 
 	if (cpu_has_feature(CPU_FTR_POWER9_DD1)) {
+		/*
+		 * Since PM_INST_CMPL may not provide right counts in all
+		 * sampling scenarios in power9 DD1, instead use PM_INST_DISP.
+		 */
+		EVENT_VAR(PM_INST_CMPL, _g).id = PM_INST_DISP;
 		rc = register_power_pmu(&power9_isa207_pmu);
 	} else {
 		rc = register_power_pmu(&power9_pmu);
-- 
2.7.4

[PATCH v2 5/6] powerpc/perf: Use Instruction Counter value

[Madhavan Srinivasan]

Since PM_INST_DISP include speculative instruction,
based on the workload the dispatch count could vary
considerably. Hence as an alternative, for completed
instruction counting, program the PM_INST_DISP event
to the MMCR* but use Instruction Counter register value.

Signed-off-by: Madhavan Srinivasan <maddy@linux.vnet.ibm.com>
---
Changelog v1:
1)Removed the #ifdef and added changes for EBB count updates
 
 arch/powerpc/perf/core-book3s.c | 36 ++++++++++++++++++++++++++++++++++++
 1 file changed, 36 insertions(+)

diff --git a/arch/powerpc/perf/core-book3s.c b/arch/powerpc/perf/core-book3s.c
index 270eb9b74e2e..87d17a1f7168 100644
--- a/arch/powerpc/perf/core-book3s.c
+++ b/arch/powerpc/perf/core-book3s.c
@@ -57,6 +57,7 @@ struct cpu_hw_events {
 	void				*bhrb_context;
 	struct	perf_branch_stack	bhrb_stack;
 	struct	perf_branch_entry	bhrb_entries[BHRB_MAX_ENTRIES];
+	u64				ic_init;
 };
 
 static DEFINE_PER_CPU(struct cpu_hw_events, cpu_hw_events);
@@ -127,6 +128,10 @@ static inline void power_pmu_bhrb_disable(struct perf_event *event) {}
 static void power_pmu_sched_task(struct perf_event_context *ctx, bool sched_in) {}
 static inline void power_pmu_bhrb_read(struct cpu_hw_events *cpuhw) {}
 static void pmao_restore_workaround(bool ebb) { }
+static bool use_ic(u64 event)
+{
+	return false;
+}
 #endif /* CONFIG_PPC32 */
 
 static bool regs_use_siar(struct pt_regs *regs)
@@ -688,6 +693,15 @@ static void pmao_restore_workaround(bool ebb)
 	mtspr(SPRN_PMC5, pmcs[4]);
 	mtspr(SPRN_PMC6, pmcs[5]);
 }
+
+static bool use_ic(u64 event)
+{
+	if (cpu_has_feature(CPU_FTR_POWER9_DD1) &&
+			(event == 0x200f2 || event == 0x300f2))
+		return true;
+
+	return false;
+}
 #endif /* CONFIG_PPC64 */
 
 static void perf_event_interrupt(struct pt_regs *regs);
@@ -1007,6 +1021,7 @@ static u64 check_and_compute_delta(u64 prev, u64 val)
 static void power_pmu_read(struct perf_event *event)
 {
 	s64 val, delta, prev;
+	struct cpu_hw_events *cpuhw = this_cpu_ptr(&cpu_hw_events);
 
 	if (event->hw.state & PERF_HES_STOPPED)
 		return;
@@ -1016,6 +1031,13 @@ static void power_pmu_read(struct perf_event *event)
 
 	if (is_ebb_event(event)) {
 		val = read_pmc(event->hw.idx);
+		if (use_ic(event->attr.config)) {
+			val = mfspr(SPRN_IC);
+			if (val > cpuhw->ic_init)
+				val = val - cpuhw->ic_init;
+			else
+				val = val + (0 - cpuhw->ic_init);
+		}
 		local64_set(&event->hw.prev_count, val);
 		return;
 	}
@@ -1029,6 +1051,13 @@ static void power_pmu_read(struct perf_event *event)
 		prev = local64_read(&event->hw.prev_count);
 		barrier();
 		val = read_pmc(event->hw.idx);
+		if (use_ic(event->attr.config)) {
+			val = mfspr(SPRN_IC);
+			if (val > cpuhw->ic_init)
+				val = val - cpuhw->ic_init;
+			else
+				val = val + (0 - cpuhw->ic_init);
+		}
 		delta = check_and_compute_delta(prev, val);
 		if (!delta)
 			return;
@@ -1466,6 +1495,13 @@ static int power_pmu_add(struct perf_event *event, int ef_flags)
 					event->attr.branch_sample_type);
 	}
 
+	/*
+	 * Workaround for POWER9 DD1 to use the Instruction Counter
+	 * register value for instruction counting
+	 */
+	if (use_ic(event->attr.config))
+		cpuhw->ic_init = mfspr(SPRN_IC);
+
 	perf_pmu_enable(event->pmu);
 	local_irq_restore(flags);
 	return ret;
-- 
2.7.4

[PATCH v2 6/6] powerpc/perf: Add restrictions to PMC5 in power9 DD1

[Madhavan Srinivasan]

PMC5 on POWER9 DD1 may not provide right counts in all
sampling scenarios, hence use PM_INST_DISP event instead
in PMC2 or PMC3 in preference.

Signed-off-by: Madhavan Srinivasan <maddy@linux.vnet.ibm.com>
---
Changelog v1:
No changes, just a rebase

 arch/powerpc/perf/isa207-common.h | 4 ++++
 arch/powerpc/perf/power9-pmu.c    | 2 +-
 2 files changed, 5 insertions(+), 1 deletion(-)

diff --git a/arch/powerpc/perf/isa207-common.h b/arch/powerpc/perf/isa207-common.h
index 3e9150f6690a..cf9bd8990159 100644
--- a/arch/powerpc/perf/isa207-common.h
+++ b/arch/powerpc/perf/isa207-common.h
@@ -222,6 +222,10 @@
 	CNST_PMC_VAL(1) | CNST_PMC_VAL(2) | CNST_PMC_VAL(3) | \
 	CNST_PMC_VAL(4) | CNST_PMC_VAL(5) | CNST_PMC_VAL(6) | CNST_NC_VAL
 
+/*
+ * Lets restrict use of PMC5 for instruction counting.
+ */
+#define P9_DD1_TEST_ADDER	(ISA207_TEST_ADDER | CNST_PMC_VAL(5))
 
 /* Bits in MMCR1 for PowerISA v2.07 */
 #define MMCR1_UNIT_SHIFT(pmc)		(60 - (4 * ((pmc) - 1)))
diff --git a/arch/powerpc/perf/power9-pmu.c b/arch/powerpc/perf/power9-pmu.c
index 454e9f70894f..5fe9cb1dc3b6 100644
--- a/arch/powerpc/perf/power9-pmu.c
+++ b/arch/powerpc/perf/power9-pmu.c
@@ -423,7 +423,7 @@ static struct power_pmu power9_pmu = {
 	.name			= "POWER9",
 	.n_counter		= MAX_PMU_COUNTERS,
 	.add_fields		= ISA207_ADD_FIELDS,
-	.test_adder		= ISA207_TEST_ADDER,
+	.test_adder		= P9_DD1_TEST_ADDER,
 	.compute_mmcr		= isa207_compute_mmcr,
 	.config_bhrb		= power9_config_bhrb,
 	.bhrb_filter_map	= power9_bhrb_filter_map,
-- 
2.7.4

Re: [v2,1/6] powerpc/perf: Factor of event_alternative function

[Michael Ellerman]

On Sun, 2017-02-12 at 17:03:10 UTC, Madhavan Srinivasan wrote:
> Factor out the power8 event_alternative function to share
> the code with power9.
> 
> Signed-off-by: Madhavan Srinivasan <maddy@linux.vnet.ibm.com>

Series applied to powerpc next, thanks.

https://git.kernel.org/powerpc/c/efe881afdd9996ccbcd2a09c93b724

cheers