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