Perf Event support for ARMv7 (was: Re: [PATCH 5/5] arm/perfevents: implement perf event support for ARMv6)
Jean Pihet
jpihet at mvista.com
Tue Dec 29 08:58:20 EST 2009
Hi,
Here is the updated patch after testing on HW.
I will rebase it on Jamie's latest patch set as soon as they are out.
Feedback is welcome!
The comments and remarks are the same as in the previous version:
It now supports:
- Cortex-A8 and Cortex-A9 processors,
- The low level code has been completely redesigned to allow the dynamic
detection of the number of available counters, based on the PMCR value,
- runtime detection of the CPU arch (v6 or v7) and model (Cortex-A8 or
Cortex-A9)
The code is for review, it has been checked, compiled and tested on OMAP3
(Cortex-A8).
Some remarks and questions:
1) The number of available counters can reach 32 on ARMv7, so the macro
ARMPMU_MAX_HWEVENTS is now defined as 32. Is that correct?
2) Please note that the Cortex-A9 events do not easily map to the predefined
events. Cf. armv7_a9_perf_map and armv7_a9_perf_cache_map in the code.
- the PERF_COUNT_HW_INSTRUCTIONS event is not found. It looks like the number
of instructions is calculated by adding events numbers (events from 0x70 till
0x74: MAIN_UNIT_EXECUTED_INST, SECOND_UNIT_EXECUTED_INST,
LD_ST_UNIT_EXECUTED_INST, FP_EXECUTED_INST and NEON_EXECUTED_INST),
- the HW_BRANCH events are not found
- the global cache events 0x50 and 0x51 define the COHERENT_LINE_HIT and
COHERENT_LINE_MISS events, is that correct?
- L1 and L2 cache events are not found. Those could be available in separate
PL310 registers, TBC
- no TLB events excepted the ITLB_MISS event are found.
Any thoughts?
Regards,
Jean
---
From 9d559085417187870be2823e8a10fba6a513aae9 Mon Sep 17 00:00:00 2001
From: Jean Pihet <jpihet at mvista.com>
Date: Fri, 18 Dec 2009 17:46:21 +0100
Subject: [PATCH] arm/perfevents: add support for ARMv7
Adds the Performance Events support for ARMv7 processor, using
the PMNC unit in HW.
Supports the following:
- Cortex-A8 and Cortex-A9 processors,
- dynamic detection of the number of available counters,
based on the PMCR value,
- runtime detection of the CPU arch (v6 or v7)
and model (Cortex-A8 or Cortex-A9)
Tested on OMAP3530 (Cortex-A8) only.
Signed-off-by: Jean Pihet <jpihet at mvista.com>
---
arch/arm/Kconfig | 2 +-
arch/arm/kernel/perf_event.c | 923
+++++++++++++++++++++++++++++++++++++++++-
2 files changed, 907 insertions(+), 18 deletions(-)
diff --git a/arch/arm/Kconfig b/arch/arm/Kconfig
index e5bd97e..5c6afdb 100644
--- a/arch/arm/Kconfig
+++ b/arch/arm/Kconfig
@@ -1172,7 +1172,7 @@ config HIGHPTE
config HW_PERF_EVENTS
bool "Enable hardware performance counter support for perf events"
- depends on PERF_EVENTS && CPU_HAS_PMU && CPU_V6
+ depends on PERF_EVENTS && CPU_HAS_PMU && (CPU_V6 || CPU_V7)
default y
help
Enable hardware performance counter support for perf events. If
diff --git a/arch/arm/kernel/perf_event.c b/arch/arm/kernel/perf_event.c
index abb5267..d56c2cd 100644
--- a/arch/arm/kernel/perf_event.c
+++ b/arch/arm/kernel/perf_event.c
@@ -4,6 +4,7 @@
* ARM performance counter support.
*
* Copyright (C) 2009 picoChip Designs, Ltd., Jamie Iles
+ * ARMv7 support: Jean Pihet <jpihet at mvista.com>
*
* This code is based on the sparc64 perf event code, which is in turn based
* on the x86 code. Callchain code is based on the ARM OProfile backtrace
@@ -35,8 +36,12 @@ DEFINE_SPINLOCK(pmu_lock);
* ARMv6 supports a maximum of 3 events, starting from index 1. If we add
* another platform that supports more, we need to increase this to be the
* largest of all platforms.
+ *
+ * ARMv7 supports up to 32 events:
+ * cycle counter CCNT + 31 events counters CNT0..30.
+ * Cortex-A8 has 1+4 counters, Cortex-A9 has 1+6 counters
*/
-#define ARMPMU_MAX_HWEVENTS 4
+#define ARMPMU_MAX_HWEVENTS 32
/* The events for a given CPU. */
struct cpu_hw_events {
@@ -61,7 +66,7 @@ struct cpu_hw_events {
DEFINE_PER_CPU(struct cpu_hw_events, cpu_hw_events);
struct arm_pmu {
- const char *name;
+ char *name;
irqreturn_t (*handle_irq)(int irq_num, void *dev);
void (*enable)(struct hw_perf_event *evt, int idx);
void (*disable)(struct hw_perf_event *evt, int idx);
@@ -965,26 +970,910 @@ static struct arm_pmu armv6pmu = {
.max_period = (1LLU << 32) - 1,
};
+/*
+ * ARMv7 Cortex-A8 and Cortex-A9 Performance Events handling code.
+ *
+ * Copied from ARMv6 code, with the low level code inspired
+ * by the ARMv7 Oprofile code.
+ *
+ * Cortex-A8 has up to 4 configurable performance counters and
+ * a single cycle counter.
+ * Cortex-A9 has up to 31 configurable performance counters and
+ * a single cycle counter.
+ *
+ * All counters can be enabled/disabled and IRQ masked separately. The cycle
+ * counter and all 4 performance counters together can be reset separately.
+ */
+
+#define ARMV7_PMU_CORTEX_A8_NAME "ARMv7 Cortex-A8"
+
+#define ARMV7_PMU_CORTEX_A9_NAME "ARMv7 Cortex-A9"
+
+/* Common ARMv7 event types */
+enum armv7_perf_types {
+ ARMV7_PERFCTR_PMNC_SW_INCR = 0x00,
+ ARMV7_PERFCTR_IFETCH_MISS = 0x01,
+ ARMV7_PERFCTR_ITLB_MISS = 0x02,
+ ARMV7_PERFCTR_DCACHE_REFILL = 0x03,
+ ARMV7_PERFCTR_DCACHE_ACCESS = 0x04,
+ ARMV7_PERFCTR_DTLB_REFILL = 0x05,
+ ARMV7_PERFCTR_DREAD = 0x06,
+ ARMV7_PERFCTR_DWRITE = 0x07,
+
+ ARMV7_PERFCTR_EXC_TAKEN = 0x09,
+ ARMV7_PERFCTR_EXC_EXECUTED = 0x0A,
+ ARMV7_PERFCTR_CID_WRITE = 0x0B,
+ ARMV7_PERFCTR_PC_WRITE = 0x0C,
+ ARMV7_PERFCTR_PC_IMM_BRANCH = 0x0D,
+ ARMV7_PERFCTR_UNALIGNED_ACCESS = 0x0F,
+ ARMV7_PERFCTR_PC_BRANCH_MIS_PRED = 0x10,
+ ARMV7_PERFCTR_CLOCK_CYCLES = 0x11,
+
+ ARMV7_PERFCTR_PC_BRANCH_MIS_USED = 0x12,
+
+ ARMV7_PERFCTR_CPU_CYCLES = 0xFF
+};
+
+/* ARMv7 Cortex-A8 specific event types */
+enum armv7_a8_perf_types {
+ ARMV7_PERFCTR_INSTR_EXECUTED = 0x08,
+
+ ARMV7_PERFCTR_PC_PROC_RETURN = 0x0E,
+
+ ARMV7_PERFCTR_WRITE_BUFFER_FULL = 0x40,
+ ARMV7_PERFCTR_L2_STORE_MERGED = 0x41,
+ ARMV7_PERFCTR_L2_STORE_BUFF = 0x42,
+ ARMV7_PERFCTR_L2_ACCESS = 0x43,
+ ARMV7_PERFCTR_L2_CACH_MISS = 0x44,
+ ARMV7_PERFCTR_AXI_READ_CYCLES = 0x45,
+ ARMV7_PERFCTR_AXI_WRITE_CYCLES = 0x46,
+ ARMV7_PERFCTR_MEMORY_REPLAY = 0x47,
+ ARMV7_PERFCTR_UNALIGNED_ACCESS_REPLAY = 0x48,
+ ARMV7_PERFCTR_L1_DATA_MISS = 0x49,
+ ARMV7_PERFCTR_L1_INST_MISS = 0x4A,
+ ARMV7_PERFCTR_L1_DATA_COLORING = 0x4B,
+ ARMV7_PERFCTR_L1_NEON_DATA = 0x4C,
+ ARMV7_PERFCTR_L1_NEON_CACH_DATA = 0x4D,
+ ARMV7_PERFCTR_L2_NEON = 0x4E,
+ ARMV7_PERFCTR_L2_NEON_HIT = 0x4F,
+ ARMV7_PERFCTR_L1_INST = 0x50,
+ ARMV7_PERFCTR_PC_RETURN_MIS_PRED = 0x51,
+ ARMV7_PERFCTR_PC_BRANCH_FAILED = 0x52,
+ ARMV7_PERFCTR_PC_BRANCH_TAKEN = 0x53,
+ ARMV7_PERFCTR_PC_BRANCH_EXECUTED = 0x54,
+ ARMV7_PERFCTR_OP_EXECUTED = 0x55,
+ ARMV7_PERFCTR_CYCLES_INST_STALL = 0x56,
+ ARMV7_PERFCTR_CYCLES_INST = 0x57,
+ ARMV7_PERFCTR_CYCLES_NEON_DATA_STALL = 0x58,
+ ARMV7_PERFCTR_CYCLES_NEON_INST_STALL = 0x59,
+ ARMV7_PERFCTR_NEON_CYCLES = 0x5A,
+
+ ARMV7_PERFCTR_PMU0_EVENTS = 0x70,
+ ARMV7_PERFCTR_PMU1_EVENTS = 0x71,
+ ARMV7_PERFCTR_PMU_EVENTS = 0x72,
+};
+
+/* ARMv7 Cortex-A9 specific event types */
+enum armv7_a9_perf_types {
+ ARMV7_PERFCTR_JAVA_HW_BYTECODE_EXEC = 0x40,
+ ARMV7_PERFCTR_JAVA_SW_BYTECODE_EXEC = 0x41,
+ ARMV7_PERFCTR_JAZELLE_BRANCH_EXEC = 0x42,
+
+ ARMV7_PERFCTR_COHERENT_LINE_MISS = 0x50,
+ ARMV7_PERFCTR_COHERENT_LINE_HIT = 0x51,
+
+ ARMV7_PERFCTR_ICACHE_DEP_STALL_CYCLES = 0x60,
+ ARMV7_PERFCTR_DCACHE_DEP_STALL_CYCLES = 0x61,
+ ARMV7_PERFCTR_TLB_MISS_DEP_STALL_CYCLES = 0x62,
+ ARMV7_PERFCTR_STREX_EXECUTED_PASSED = 0x63,
+ ARMV7_PERFCTR_STREX_EXECUTED_FAILED = 0x64,
+ ARMV7_PERFCTR_DATA_EVICTION = 0x65,
+ ARMV7_PERFCTR_ISSUE_STAGE_NO_INST = 0x66,
+ ARMV7_PERFCTR_ISSUE_STAGE_EMPTY = 0x67,
+ ARMV7_PERFCTR_INST_OUT_OF_RENAME_STAGE = 0x68,
+
+ ARMV7_PERFCTR_PREDICTABLE_FUNCT_RETURNS = 0x6E,
+
+ ARMV7_PERFCTR_MAIN_UNIT_EXECUTED_INST = 0x70,
+ ARMV7_PERFCTR_SECOND_UNIT_EXECUTED_INST = 0x71,
+ ARMV7_PERFCTR_LD_ST_UNIT_EXECUTED_INST = 0x72,
+ ARMV7_PERFCTR_FP_EXECUTED_INST = 0x73,
+ ARMV7_PERFCTR_NEON_EXECUTED_INST = 0x74,
+
+ ARMV7_PERFCTR_PLD_FULL_DEP_STALL_CYCLES = 0x80,
+ ARMV7_PERFCTR_DATA_WR_DEP_STALL_CYCLES = 0x81,
+ ARMV7_PERFCTR_ITLB_MISS_DEP_STALL_CYCLES = 0x82,
+ ARMV7_PERFCTR_DTLB_MISS_DEP_STALL_CYCLES = 0x83,
+ ARMV7_PERFCTR_MICRO_ITLB_MISS_DEP_STALL_CYCLES = 0x84,
+ ARMV7_PERFCTR_MICRO_DTLB_MISS_DEP_STALL_CYCLES = 0x85,
+ ARMV7_PERFCTR_DMB_DEP_STALL_CYCLES = 0x86,
+
+ ARMV7_PERFCTR_INTGR_CLK_ENABLED_CYCLES = 0x8A,
+ ARMV7_PERFCTR_DATA_ENGINE_CLK_EN_CYCLES = 0x8B,
+
+ ARMV7_PERFCTR_ISB_INST = 0x90,
+ ARMV7_PERFCTR_DSB_INST = 0x91,
+ ARMV7_PERFCTR_DMB_INST = 0x92,
+ ARMV7_PERFCTR_EXT_INTERRUPTS = 0x93,
+
+ ARMV7_PERFCTR_PLE_CACHE_LINE_RQST_COMPLETED = 0xA0,
+ ARMV7_PERFCTR_PLE_CACHE_LINE_RQST_SKIPPED = 0xA1,
+ ARMV7_PERFCTR_PLE_FIFO_FLUSH = 0xA2,
+ ARMV7_PERFCTR_PLE_RQST_COMPLETED = 0xA3,
+ ARMV7_PERFCTR_PLE_FIFO_OVERFLOW = 0xA4,
+ ARMV7_PERFCTR_PLE_RQST_PROG = 0xA5
+};
+
+/*
+ * Cortex-A8 HW events mapping
+ *
+ * The hardware events that we support. We do support cache operations but
+ * we have harvard caches and no way to combine instruction and data
+ * accesses/misses in hardware.
+ */
+static const unsigned armv7_a8_perf_map[PERF_COUNT_HW_MAX] = {
+ [PERF_COUNT_HW_CPU_CYCLES] = ARMV7_PERFCTR_CPU_CYCLES,
+ [PERF_COUNT_HW_INSTRUCTIONS] = ARMV7_PERFCTR_INSTR_EXECUTED,
+ [PERF_COUNT_HW_CACHE_REFERENCES] = HW_OP_UNSUPPORTED,
+ [PERF_COUNT_HW_CACHE_MISSES] = HW_OP_UNSUPPORTED,
+ [PERF_COUNT_HW_BRANCH_INSTRUCTIONS] = ARMV7_PERFCTR_PC_BRANCH_TAKEN,
+ [PERF_COUNT_HW_BRANCH_MISSES] = ARMV7_PERFCTR_PC_BRANCH_FAILED,
+ [PERF_COUNT_HW_BUS_CYCLES] = ARMV7_PERFCTR_CLOCK_CYCLES,
+};
+
+static const unsigned armv7_a8_perf_cache_map[PERF_COUNT_HW_CACHE_MAX]
+ [PERF_COUNT_HW_CACHE_OP_MAX]
+ [PERF_COUNT_HW_CACHE_RESULT_MAX] = {
+ [C(L1D)] = {
+ /*
+ * The performance counters don't differentiate between read
+ * and write accesses/misses so this isn't strictly correct,
+ * but it's the best we can do. Writes and reads get
+ * combined.
+ */
+ [C(OP_READ)] = {
+ [C(RESULT_ACCESS)] = ARMV7_PERFCTR_DCACHE_ACCESS,
+ [C(RESULT_MISS)] = ARMV7_PERFCTR_L1_DATA_MISS,
+ },
+ [C(OP_WRITE)] = {
+ [C(RESULT_ACCESS)] = ARMV7_PERFCTR_DCACHE_ACCESS,
+ [C(RESULT_MISS)] = ARMV7_PERFCTR_L1_DATA_MISS,
+ },
+ [C(OP_PREFETCH)] = {
+ [C(RESULT_ACCESS)] = CACHE_OP_UNSUPPORTED,
+ [C(RESULT_MISS)] = CACHE_OP_UNSUPPORTED,
+ },
+ },
+ [C(L1I)] = {
+ [C(OP_READ)] = {
+ [C(RESULT_ACCESS)] = ARMV7_PERFCTR_L1_INST,
+ [C(RESULT_MISS)] = ARMV7_PERFCTR_L1_INST_MISS,
+ },
+ [C(OP_WRITE)] = {
+ [C(RESULT_ACCESS)] = ARMV7_PERFCTR_L1_INST,
+ [C(RESULT_MISS)] = ARMV7_PERFCTR_L1_INST_MISS,
+ },
+ [C(OP_PREFETCH)] = {
+ [C(RESULT_ACCESS)] = CACHE_OP_UNSUPPORTED,
+ [C(RESULT_MISS)] = CACHE_OP_UNSUPPORTED,
+ },
+ },
+ [C(LL)] = {
+ [C(OP_READ)] = {
+ [C(RESULT_ACCESS)] = ARMV7_PERFCTR_L2_ACCESS,
+ [C(RESULT_MISS)] = ARMV7_PERFCTR_L2_CACH_MISS,
+ },
+ [C(OP_WRITE)] = {
+ [C(RESULT_ACCESS)] = ARMV7_PERFCTR_L2_ACCESS,
+ [C(RESULT_MISS)] = ARMV7_PERFCTR_L2_CACH_MISS,
+ },
+ [C(OP_PREFETCH)] = {
+ [C(RESULT_ACCESS)] = CACHE_OP_UNSUPPORTED,
+ [C(RESULT_MISS)] = CACHE_OP_UNSUPPORTED,
+ },
+ },
+ [C(DTLB)] = {
+ /*
+ * Only ITLB misses and DTLB refills are supported.
+ * If users want the DTLB refills misses a raw counter
+ * must be used.
+ */
+ [C(OP_READ)] = {
+ [C(RESULT_ACCESS)] = CACHE_OP_UNSUPPORTED,
+ [C(RESULT_MISS)] = CACHE_OP_UNSUPPORTED,
+ },
+ [C(OP_WRITE)] = {
+ [C(RESULT_ACCESS)] = CACHE_OP_UNSUPPORTED,
+ [C(RESULT_MISS)] = CACHE_OP_UNSUPPORTED,
+ },
+ [C(OP_PREFETCH)] = {
+ [C(RESULT_ACCESS)] = CACHE_OP_UNSUPPORTED,
+ [C(RESULT_MISS)] = CACHE_OP_UNSUPPORTED,
+ },
+ },
+ [C(ITLB)] = {
+ [C(OP_READ)] = {
+ [C(RESULT_ACCESS)] = CACHE_OP_UNSUPPORTED,
+ [C(RESULT_MISS)] = ARMV7_PERFCTR_ITLB_MISS,
+ },
+ [C(OP_WRITE)] = {
+ [C(RESULT_ACCESS)] = CACHE_OP_UNSUPPORTED,
+ [C(RESULT_MISS)] = ARMV7_PERFCTR_ITLB_MISS,
+ },
+ [C(OP_PREFETCH)] = {
+ [C(RESULT_ACCESS)] = CACHE_OP_UNSUPPORTED,
+ [C(RESULT_MISS)] = CACHE_OP_UNSUPPORTED,
+ },
+ },
+ [C(BPU)] = {
+ [C(OP_READ)] = {
+ [C(RESULT_ACCESS)] = ARMV7_PERFCTR_PC_BRANCH_TAKEN,
+ [C(RESULT_MISS)]
+ = ARMV7_PERFCTR_PC_BRANCH_FAILED,
+ },
+ [C(OP_WRITE)] = {
+ [C(RESULT_ACCESS)] = ARMV7_PERFCTR_PC_BRANCH_TAKEN,
+ [C(RESULT_MISS)]
+ = ARMV7_PERFCTR_PC_BRANCH_FAILED,
+ },
+ [C(OP_PREFETCH)] = {
+ [C(RESULT_ACCESS)] = CACHE_OP_UNSUPPORTED,
+ [C(RESULT_MISS)] = CACHE_OP_UNSUPPORTED,
+ },
+ },
+};
+
+/*
+ * Cortex-A9 HW events mapping
+ */
+static const unsigned armv7_a9_perf_map[PERF_COUNT_HW_MAX] = {
+ [PERF_COUNT_HW_CPU_CYCLES] = ARMV7_PERFCTR_CPU_CYCLES,
+ [PERF_COUNT_HW_INSTRUCTIONS] = HW_OP_UNSUPPORTED,
+ [PERF_COUNT_HW_CACHE_REFERENCES] = ARMV7_PERFCTR_COHERENT_LINE_HIT,
+ [PERF_COUNT_HW_CACHE_MISSES] = ARMV7_PERFCTR_COHERENT_LINE_MISS,
+ [PERF_COUNT_HW_BRANCH_INSTRUCTIONS] = HW_OP_UNSUPPORTED,
+ [PERF_COUNT_HW_BRANCH_MISSES] = HW_OP_UNSUPPORTED,
+ [PERF_COUNT_HW_BUS_CYCLES] = ARMV7_PERFCTR_CLOCK_CYCLES,
+};
+
+static const unsigned armv7_a9_perf_cache_map[PERF_COUNT_HW_CACHE_MAX]
+ [PERF_COUNT_HW_CACHE_OP_MAX]
+ [PERF_COUNT_HW_CACHE_RESULT_MAX] = {
+ [C(L1D)] = {
+ /*
+ * The performance counters don't differentiate between read
+ * and write accesses/misses so this isn't strictly correct,
+ * but it's the best we can do. Writes and reads get
+ * combined.
+ */
+ [C(OP_READ)] = {
+ [C(RESULT_ACCESS)] = ARMV7_PERFCTR_DCACHE_ACCESS,
+ [C(RESULT_MISS)] = CACHE_OP_UNSUPPORTED,
+ },
+ [C(OP_WRITE)] = {
+ [C(RESULT_ACCESS)] = ARMV7_PERFCTR_DCACHE_ACCESS,
+ [C(RESULT_MISS)] = CACHE_OP_UNSUPPORTED,
+ },
+ [C(OP_PREFETCH)] = {
+ [C(RESULT_ACCESS)] = CACHE_OP_UNSUPPORTED,
+ [C(RESULT_MISS)] = CACHE_OP_UNSUPPORTED,
+ },
+ },
+ [C(L1I)] = {
+ [C(OP_READ)] = {
+ [C(RESULT_ACCESS)] = CACHE_OP_UNSUPPORTED,
+ [C(RESULT_MISS)] = CACHE_OP_UNSUPPORTED,
+ },
+ [C(OP_WRITE)] = {
+ [C(RESULT_ACCESS)] = CACHE_OP_UNSUPPORTED,
+ [C(RESULT_MISS)] = CACHE_OP_UNSUPPORTED,
+ },
+ [C(OP_PREFETCH)] = {
+ [C(RESULT_ACCESS)] = CACHE_OP_UNSUPPORTED,
+ [C(RESULT_MISS)] = CACHE_OP_UNSUPPORTED,
+ },
+ },
+ [C(LL)] = {
+ [C(OP_READ)] = {
+ [C(RESULT_ACCESS)] = CACHE_OP_UNSUPPORTED,
+ [C(RESULT_MISS)] = CACHE_OP_UNSUPPORTED,
+ },
+ [C(OP_WRITE)] = {
+ [C(RESULT_ACCESS)] = CACHE_OP_UNSUPPORTED,
+ [C(RESULT_MISS)] = CACHE_OP_UNSUPPORTED,
+ },
+ [C(OP_PREFETCH)] = {
+ [C(RESULT_ACCESS)] = CACHE_OP_UNSUPPORTED,
+ [C(RESULT_MISS)] = CACHE_OP_UNSUPPORTED,
+ },
+ },
+ [C(DTLB)] = {
+ /*
+ * Only ITLB misses and DTLB refills are supported.
+ * If users want the DTLB refills misses a raw counter
+ * must be used.
+ */
+ [C(OP_READ)] = {
+ [C(RESULT_ACCESS)] = CACHE_OP_UNSUPPORTED,
+ [C(RESULT_MISS)] = CACHE_OP_UNSUPPORTED,
+ },
+ [C(OP_WRITE)] = {
+ [C(RESULT_ACCESS)] = CACHE_OP_UNSUPPORTED,
+ [C(RESULT_MISS)] = CACHE_OP_UNSUPPORTED,
+ },
+ [C(OP_PREFETCH)] = {
+ [C(RESULT_ACCESS)] = CACHE_OP_UNSUPPORTED,
+ [C(RESULT_MISS)] = CACHE_OP_UNSUPPORTED,
+ },
+ },
+ [C(ITLB)] = {
+ [C(OP_READ)] = {
+ [C(RESULT_ACCESS)] = CACHE_OP_UNSUPPORTED,
+ [C(RESULT_MISS)] = ARMV7_PERFCTR_ITLB_MISS,
+ },
+ [C(OP_WRITE)] = {
+ [C(RESULT_ACCESS)] = CACHE_OP_UNSUPPORTED,
+ [C(RESULT_MISS)] = ARMV7_PERFCTR_ITLB_MISS,
+ },
+ [C(OP_PREFETCH)] = {
+ [C(RESULT_ACCESS)] = CACHE_OP_UNSUPPORTED,
+ [C(RESULT_MISS)] = CACHE_OP_UNSUPPORTED,
+ },
+ },
+ [C(BPU)] = {
+ [C(OP_READ)] = {
+ [C(RESULT_ACCESS)] = CACHE_OP_UNSUPPORTED,
+ [C(RESULT_MISS)] = CACHE_OP_UNSUPPORTED,
+ },
+ [C(OP_WRITE)] = {
+ [C(RESULT_ACCESS)] = CACHE_OP_UNSUPPORTED,
+ [C(RESULT_MISS)] = CACHE_OP_UNSUPPORTED,
+ },
+ [C(OP_PREFETCH)] = {
+ [C(RESULT_ACCESS)] = CACHE_OP_UNSUPPORTED,
+ [C(RESULT_MISS)] = CACHE_OP_UNSUPPORTED,
+ },
+ },
+};
+
+/*
+ * Perf Events counters
+ */
+enum armv7_counters {
+ ARMV7_CYCLE_COUNTER = 1, /* Cycle counter */
+ ARMV7_COUNTER0 = 2, /* First event counter */
+};
+
+/* The last event counter is (ARMV7_COUNTER0 + armpmu->num_events) */
+#define ARMV7_COUNTER_LAST (ARMV7_COUNTER0 + armpmu->num_events)
+
+/*
+ * ARMv7 low level PMNC access
+ */
+
+/*
+ * Per-CPU PMNC: config reg
+ */
+#define ARMV7_PMNC_E (1 << 0) /* Enable all counters */
+#define ARMV7_PMNC_P (1 << 1) /* Reset all counters */
+#define ARMV7_PMNC_C (1 << 2) /* Cycle counter reset */
+#define ARMV7_PMNC_D (1 << 3) /* CCNT counts every 64th cpu cycle */
+#define ARMV7_PMNC_X (1 << 4) /* Export to ETM */
+#define ARMV7_PMNC_DP (1 << 5) /* Disable CCNT if non-invasive debug*/
+#define ARMV7_PMNC_N_SHIFT 11 /* Number of counters supported */
+#define ARMV7_PMNC_N_MASK 0x1f
+#define ARMV7_PMNC_MASK 0x3f /* Mask for writable bits */
+
+/*
+ * Available counters
+ */
+#define ARMV7_CNT0 0 /* First event counter */
+#define ARMV7_CCNT 31 /* Cycle counter */
+
+#define ARMV7_A8_CNTMAX 5 /* Cortex-A8: up to 4 counters + CCNT */
+#define ARMV7_A9_CNTMAX 32 /* Cortex-A9: up to 31 counters + CCNT*/
+
+/* Perf Event to low level counters mapping */
+#define ARMV7_EVENT_CNT_TO_CNTx (ARMV7_COUNTER0 - ARMV7_CNT0)
+
+/*
+ * CNTENS: counters enable reg
+ */
+#define ARMV7_CNTENS_P(idx) (1 << (idx - ARMV7_EVENT_CNT_TO_CNTx))
+#define ARMV7_CNTENS_C (1 << ARMV7_CCNT)
+
+/*
+ * CNTENC: counters disable reg
+ */
+#define ARMV7_CNTENC_P(idx) (1 << (idx - ARMV7_EVENT_CNT_TO_CNTx))
+#define ARMV7_CNTENC_C (1 << ARMV7_CCNT)
+
+/*
+ * INTENS: counters overflow interrupt enable reg
+ */
+#define ARMV7_INTENS_P(idx) (1 << (idx - ARMV7_EVENT_CNT_TO_CNTx))
+#define ARMV7_INTENS_C (1 << ARMV7_CCNT)
+
+/*
+ * INTENC: counters overflow interrupt disable reg
+ */
+#define ARMV7_INTENC_P(idx) (1 << (idx - ARMV7_EVENT_CNT_TO_CNTx))
+#define ARMV7_INTENC_C (1 << ARMV7_CCNT)
+
+/*
+ * EVTSEL: Event selection reg
+ */
+#define ARMV7_EVTSEL_MASK 0x7f /* Mask for writable bits */
+
+/*
+ * SELECT: Counter selection reg
+ */
+#define ARMV7_SELECT_MASK 0x1f /* Mask for writable bits */
+
+/*
+ * FLAG: counters overflow flag status reg
+ */
+#define ARMV7_FLAG_P(idx) (1 << (idx - ARMV7_EVENT_CNT_TO_CNTx))
+#define ARMV7_FLAG_C (1 << ARMV7_CCNT)
+#define ARMV7_FLAG_MASK 0xffffffff /* Mask for writable bits */
+#define ARMV7_OVERFLOWED_MASK ARMV7_FLAG_MASK
+
+static inline unsigned long armv7_pmnc_read(void)
+{
+ u32 val;
+ asm volatile("mrc p15, 0, %0, c9, c12, 0" : "=r"(val));
+ return val;
+}
+
+static inline void armv7_pmnc_write(unsigned long val)
+{
+ val &= ARMV7_PMNC_MASK;
+ asm volatile("mcr p15, 0, %0, c9, c12, 0" : : "r"(val));
+}
+
+static inline int armv7_pmnc_has_overflowed(unsigned long pmnc)
+{
+ return pmnc & ARMV7_OVERFLOWED_MASK;
+}
+
+static inline int armv7_pmnc_counter_has_overflowed(unsigned long pmnc,
+ enum armv7_counters counter)
+{
+ int ret;
+
+ if (counter == ARMV7_CYCLE_COUNTER)
+ ret = pmnc & ARMV7_FLAG_C;
+ else if ((counter >= ARMV7_COUNTER0) && (counter <= ARMV7_COUNTER_LAST))
+ ret = pmnc & ARMV7_FLAG_P(counter);
+ else
+ BUG();
+
+ return ret;
+}
+
+static inline int armv7_pmnc_select_counter(unsigned int idx)
+{
+ u32 val;
+
+ if ((idx < ARMV7_COUNTER0) || (idx > ARMV7_COUNTER_LAST)) {
+ pr_err("CPU%u selecting wrong PMNC counter"
+ " %d\n", smp_processor_id(), idx);
+ return -1;
+ }
+
+ val = (idx - ARMV7_EVENT_CNT_TO_CNTx) & ARMV7_SELECT_MASK;
+ asm volatile("mcr p15, 0, %0, c9, c12, 5" : : "r" (val));
+
+ return idx;
+}
+
+static inline u32 armv7pmu_read_counter(int idx)
+{
+ unsigned long value = 0;
+
+ if (idx == ARMV7_CYCLE_COUNTER)
+ asm volatile("mrc p15, 0, %0, c9, c13, 0" : "=r" (value));
+ else if ((idx >= ARMV7_COUNTER0) && (idx <= ARMV7_COUNTER_LAST)) {
+ if (armv7_pmnc_select_counter(idx) == idx)
+ asm volatile("mrc p15, 0, %0, c9, c13, 2"
+ : "=r" (value));
+ } else
+ BUG();
+
+ return value;
+}
+
+static inline void armv7pmu_write_counter(int idx, u32 value)
+{
+ if (idx == ARMV7_CYCLE_COUNTER)
+ asm volatile("mcr p15, 0, %0, c9, c13, 0" : : "r" (value));
+ else if ((idx >= ARMV7_COUNTER0) && (idx <= ARMV7_COUNTER_LAST)) {
+ if (armv7_pmnc_select_counter(idx) == idx)
+ asm volatile("mcr p15, 0, %0, c9, c13, 2"
+ : : "r" (value));
+ } else
+ BUG();
+}
+
+static inline void armv7_pmnc_write_evtsel(unsigned int idx, u32 val)
+{
+ if (armv7_pmnc_select_counter(idx) == idx) {
+ val &= ARMV7_EVTSEL_MASK;
+ asm volatile("mcr p15, 0, %0, c9, c13, 1" : : "r" (val));
+ }
+}
+
+static inline u32 armv7_pmnc_enable_counter(unsigned int idx)
+{
+ u32 val;
+
+ if ((idx != ARMV7_CYCLE_COUNTER) &&
+ ((idx < ARMV7_COUNTER0) || (idx > ARMV7_COUNTER_LAST))) {
+ pr_err("CPU%u enabling wrong PMNC counter"
+ " %d\n", smp_processor_id(), idx);
+ return -1;
+ }
+
+ if (idx == ARMV7_CYCLE_COUNTER)
+ val = ARMV7_CNTENS_C;
+ else
+ val = ARMV7_CNTENS_P(idx);
+
+ asm volatile("mcr p15, 0, %0, c9, c12, 1" : : "r" (val));
+
+ return idx;
+}
+
+static inline u32 armv7_pmnc_disable_counter(unsigned int idx)
+{
+ u32 val;
+
+
+ if ((idx != ARMV7_CYCLE_COUNTER) &&
+ ((idx < ARMV7_COUNTER0) || (idx > ARMV7_COUNTER_LAST))) {
+ pr_err("CPU%u disabling wrong PMNC counter"
+ " %d\n", smp_processor_id(), idx);
+ return -1;
+ }
+
+ if (idx == ARMV7_CYCLE_COUNTER)
+ val = ARMV7_CNTENC_C;
+ else
+ val = ARMV7_CNTENC_P(idx);
+
+ asm volatile("mcr p15, 0, %0, c9, c12, 2" : : "r" (val));
+
+ return idx;
+}
+
+static inline u32 armv7_pmnc_enable_intens(unsigned int idx)
+{
+ u32 val;
+
+ if ((idx != ARMV7_CYCLE_COUNTER) &&
+ ((idx < ARMV7_COUNTER0) || (idx > ARMV7_COUNTER_LAST))) {
+ pr_err("CPU%u enabling wrong PMNC counter"
+ " interrupt enable %d\n", smp_processor_id(), idx);
+ return -1;
+ }
+
+ if (idx == ARMV7_CYCLE_COUNTER)
+ val = ARMV7_INTENS_C;
+ else
+ val = ARMV7_INTENS_P(idx);
+
+ asm volatile("mcr p15, 0, %0, c9, c14, 1" : : "r" (val));
+
+ return idx;
+}
+
+static inline u32 armv7_pmnc_disable_intens(unsigned int idx)
+{
+ u32 val;
+
+ if ((idx != ARMV7_CYCLE_COUNTER) &&
+ ((idx < ARMV7_COUNTER0) || (idx > ARMV7_COUNTER_LAST))) {
+ pr_err("CPU%u disabling wrong PMNC counter"
+ " interrupt enable %d\n", smp_processor_id(), idx);
+ return -1;
+ }
+
+ if (idx == ARMV7_CYCLE_COUNTER)
+ val = ARMV7_INTENC_C;
+ else
+ val = ARMV7_INTENC_P(idx);
+
+ asm volatile("mcr p15, 0, %0, c9, c14, 2" : : "r" (val));
+
+ return idx;
+}
+
+static inline u32 armv7_pmnc_getreset_flags(void)
+{
+ u32 val;
+
+ /* Read */
+ asm volatile("mrc p15, 0, %0, c9, c12, 3" : "=r" (val));
+
+ /* Write to clear flags */
+ val &= ARMV7_FLAG_MASK;
+ asm volatile("mcr p15, 0, %0, c9, c12, 3" : : "r" (val));
+
+ return val;
+}
+
+void armv7pmu_enable_event(struct hw_perf_event *hwc, int idx)
+{
+ unsigned long flags;
+
+ /*
+ * Enable counter and interrupt, and set the counter to count
+ * the event that we're interested in.
+ */
+ spin_lock_irqsave(&pmu_lock, flags);
+
+ /*
+ * Disable counter
+ */
+ armv7_pmnc_disable_counter(idx);
+
+ /*
+ * Set event (if destined for PMNx counters)
+ * We don't need to set the event if it's a cycle count
+ */
+ if (idx != ARMV7_CYCLE_COUNTER)
+ armv7_pmnc_write_evtsel(idx, hwc->config_base);
+
+ /*
+ * Enable interrupt for this counter
+ */
+ armv7_pmnc_enable_intens(idx);
+
+ /*
+ * Enable counter
+ */
+ armv7_pmnc_enable_counter(idx);
+
+ spin_unlock_irqrestore(&pmu_lock, flags);
+}
+
+static void armv7pmu_disable_event(struct hw_perf_event *hwc, int idx)
+{
+ unsigned long flags;
+
+ /*
+ * Disable counter and interrupt
+ */
+ spin_lock_irqsave(&pmu_lock, flags);
+
+ /*
+ * Disable counter
+ */
+ armv7_pmnc_disable_counter(idx);
+
+ /*
+ * Disable interrupt for this counter
+ */
+ armv7_pmnc_disable_intens(idx);
+
+ spin_unlock_irqrestore(&pmu_lock, flags);
+}
+
+static irqreturn_t armv7pmu_handle_irq(int irq_num, void *dev)
+{
+ unsigned long pmnc;
+ struct perf_sample_data data;
+ struct cpu_hw_events *cpuc;
+ struct pt_regs *regs;
+ int idx;
+
+ /*
+ * Get and reset the IRQ flags
+ */
+ pmnc = armv7_pmnc_getreset_flags();
+
+ /*
+ * Did an overflow occur?
+ */
+ if (!armv7_pmnc_has_overflowed(pmnc))
+ return IRQ_NONE;
+
+ /*
+ * Handle the counter(s) overflow(s)
+ */
+ regs = get_irq_regs();
+
+ data.addr = 0;
+
+ cpuc = &__get_cpu_var(cpu_hw_events);
+ for (idx = 0; idx <= armpmu->num_events; ++idx) {
+ struct perf_event *event = cpuc->events[idx];
+ struct hw_perf_event *hwc;
+
+ if (!test_bit(idx, cpuc->active_mask))
+ continue;
+
+ /*
+ * We have a single interrupt for all counters. Check that
+ * each counter has overflowed before we process it.
+ */
+ if (!armv7_pmnc_counter_has_overflowed(pmnc, idx))
+ continue;
+
+ hwc = &event->hw;
+ armpmu_event_update(event, hwc, idx);
+ data.period = event->hw.last_period;
+ if (!armpmu_event_set_period(event, hwc, idx))
+ continue;
+
+ if (perf_event_overflow(event, 0, &data, regs))
+ armpmu->disable(hwc, idx);
+ }
+
+ /*
+ * Handle the pending perf events.
+ *
+ * Note: this call *must* be run with interrupts enabled. For
+ * platforms that can have the PMU interrupts raised as a PMI, this
+ * will not work.
+ */
+ perf_event_do_pending();
+
+ return IRQ_HANDLED;
+}
+
+static void armv7pmu_start(void)
+{
+ unsigned long flags;
+
+ spin_lock_irqsave(&pmu_lock, flags);
+ /* Enable all counters */
+ armv7_pmnc_write(armv7_pmnc_read() | ARMV7_PMNC_E);
+ spin_unlock_irqrestore(&pmu_lock, flags);
+}
+
+static void armv7pmu_stop(void)
+{
+ unsigned long flags;
+
+ spin_lock_irqsave(&pmu_lock, flags);
+ /* Disable all counters */
+ armv7_pmnc_write(armv7_pmnc_read() & ~ARMV7_PMNC_E);
+ spin_unlock_irqrestore(&pmu_lock, flags);
+}
+
+static inline int armv7_a8_pmu_event_map(int config)
+{
+ int mapping = armv7_a8_perf_map[config];
+ if (HW_OP_UNSUPPORTED == mapping)
+ mapping = -EOPNOTSUPP;
+ return mapping;
+}
+
+static inline int armv7_a9_pmu_event_map(int config)
+{
+ int mapping = armv7_a9_perf_map[config];
+ if (HW_OP_UNSUPPORTED == mapping)
+ mapping = -EOPNOTSUPP;
+ return mapping;
+}
+
+static u64 armv7pmu_raw_event(u64 config)
+{
+ return config & 0xff;
+}
+
+static int armv7pmu_get_event_idx(struct cpu_hw_events *cpuc,
+ struct hw_perf_event *event)
+{
+ int idx;
+
+ /* Always place a cycle counter into the cycle counter. */
+ if (event->config_base == ARMV7_PERFCTR_CPU_CYCLES) {
+ if (test_and_set_bit(ARMV7_CYCLE_COUNTER, cpuc->used_mask))
+ return -EAGAIN;
+
+ return ARMV7_CYCLE_COUNTER;
+ } else {
+ /*
+ * For anything other than a cycle counter, try and use
+ * the events counters
+ */
+ for (idx = ARMV7_COUNTER0; idx <= armpmu->num_events; ++idx) {
+ if (!test_and_set_bit(idx, cpuc->used_mask))
+ return idx;
+ }
+
+ /* The counters are all in use. */
+ return -EAGAIN;
+ }
+}
+
+static struct arm_pmu armv7pmu = {
+ .handle_irq = armv7pmu_handle_irq,
+ .enable = armv7pmu_enable_event,
+ .disable = armv7pmu_disable_event,
+ .raw_event = armv7pmu_raw_event,
+ .read_counter = armv7pmu_read_counter,
+ .write_counter = armv7pmu_write_counter,
+ .get_event_idx = armv7pmu_get_event_idx,
+ .start = armv7pmu_start,
+ .stop = armv7pmu_stop,
+ .max_period = (1LLU << 32) - 1,
+};
+
static int __init
init_hw_perf_events(void)
{
-#define CPUID_V6_MASK 0x7F000
-#define CPUID_V6_BITS 0x7B000
- unsigned long cpuid = read_cpuid_id();
-
- if (CPUID_V6_BITS == (cpuid & CPUID_V6_MASK)) {
- armpmu = &armv6pmu;
- memcpy(armpmu_perf_cache_map, armv6_perf_cache_map,
- sizeof(armv6_perf_cache_map));
- perf_max_events = armv6pmu.num_events;
- } else {
- pr_info("no hardware support available\n");
- perf_max_events = -1;
- }
+#define CPUID_V6_MASK 0x7F000
+#define CPUID_V6_BITS 0x7B000
+
+#define CPUID_CORTEX_A8_BITS 0xC080
+#define CPUID_CORTEX_A8_MASK 0xFFF0
+
+#define CPUID_CORTEX_A9_BITS 0xC090
+#define CPUID_CORTEX_A9_MASK 0xFFF0
+
+ unsigned long cpuid = read_cpuid_id();
+
+ /*
+ * ARMv6 detection
+ */
+ if (CPUID_V6_BITS == (cpuid & CPUID_V6_MASK)) {
+ armpmu = &armv6pmu;
+ memcpy(armpmu_perf_cache_map, armv6_perf_cache_map,
+ sizeof(armv6_perf_cache_map));
+ perf_max_events = armv6pmu.num_events;
+ }
+ /*
+ * ARMv7 detection
+ */
+ else if (cpu_architecture() == CPU_ARCH_ARMv7) {
+ /*
+ * Cortex-A8 detection
+ */
+ if ((cpuid & CPUID_CORTEX_A8_MASK) == CPUID_CORTEX_A8_BITS) {
+ armv7pmu.name = ARMV7_PMU_CORTEX_A8_NAME;
+ memcpy(armpmu_perf_cache_map, armv7_a8_perf_cache_map,
+ sizeof(armv7_a8_perf_cache_map));
+ armv7pmu.event_map = armv7_a8_pmu_event_map;
+ armpmu = &armv7pmu;
+ } else
+ /*
+ * Cortex-A9 detection
+ */
+ if ((cpuid & CPUID_CORTEX_A9_MASK)
+ == CPUID_CORTEX_A9_BITS) {
+ armv7pmu.name = ARMV7_PMU_CORTEX_A9_NAME;
+ memcpy(armpmu_perf_cache_map,
+ armv7_a9_perf_cache_map,
+ sizeof(armv7_a9_perf_cache_map));
+ armv7pmu.event_map = armv7_a9_pmu_event_map;
+ armpmu = &armv7pmu;
+ } else
+ perf_max_events = -1;
+
+ if (armpmu) {
+ u32 nb_cnt;
+
+ /* Read the nb of CNTx counters supported from PMNC */
+ nb_cnt = (armv7_pmnc_read() >> ARMV7_PMNC_N_SHIFT)
+ & ARMV7_PMNC_N_MASK;
+ /* Add the CPU cycles counter */
+ armv7pmu.num_events = nb_cnt + 1;
+ perf_max_events = armv7pmu.num_events;
+
+ /* Initialize & Reset PMNC: C bit and P bit */
+ armv7_pmnc_write(ARMV7_PMNC_P | ARMV7_PMNC_C);
+ }
+ } else {
+ pr_info("no hardware support available\n");
+ perf_max_events = -1;
+ }
if (armpmu)
- pr_info("enabled with %s PMU driver\n",
- armpmu->name);
+ pr_info("enabled with %s PMU driver, %d counters available\n",
+ armpmu->name, armpmu->num_events);
return 0;
}
--
1.6.2.5.168.g3823
On Tuesday 22 December 2009 17:51:39 Jean Pihet wrote:
> Hi,
>
> Here is the updated patch. It now supports:
> - Cortex-A8 and Cortex-A9 processors,
> - The low level code has been completely redesigned to allow the dynamic
> detection of the number of available counters, based on the PMCR value,
> - runtime detection of the CPU arch (v6 or v7) and model (Cortex-A8 or
> Cortex-A9)
>
> The code is for review, it has been checked, compiled and boot tested on
> OMAP3 (Cortex-A8). Unfortunately I am still facing some cross compilation
> problems of the tools/perf utility.
>
> Some remarks and questions:
>
> 1) The number of available counters can reach 32 on ARMv7, so the macro
> ARMPMU_MAX_HWEVENTS is now defined as 32. Is that correct?
>
> 2) Please note that the Cortex-A9 events do not easily map to the
> predefined events. Cf. armv7_a9_perf_map and armv7_a9_perf_cache_map in the
> code. - the PERF_COUNT_HW_INSTRUCTIONS event is not found. It looks like
> the number of instructions is calculated by adding events numbers (events
> from 0x70 till 0x74: MAIN_UNIT_EXECUTED_INST, SECOND_UNIT_EXECUTED_INST,
> LD_ST_UNIT_EXECUTED_INST, FP_EXECUTED_INST and NEON_EXECUTED_INST),
> - the HW_BRANCH events are not found
> - the global cache events 0x50 and 0x51 define the COHERENT_LINE_HIT and
> COHERENT_LINE_MISS events, is that correct?
> - L1 and L2 cache events are not found. Those could be available in
> separate PL310 registers, TBC
> - no TLB events excepted the ITLB_MISS event are found.
>
> Any thoughts?
>
> Regards,
> Jean
>
More information about the linux-arm-kernel
mailing list