Implement hwpmc(4) for Freescale e500 core.

This supports e500v1, e500v2, and e500mc. Tested only on e500v2, but the
performance counters are identical across all, with e500mc having some
additional events.

Relnotes:	Yes

1 files changed, 167 insertions, 0 deletions

diff --git a/sys/dev/hwpmc/pmc_events.h b/sys/dev/hwpmc/pmc_events.h
index 83182bf..7f878cb 100644
--- a/sys/dev/hwpmc/pmc_events.h
+++ b/sys/dev/hwpmc/pmc_events.h
@@ -5379,6 +5379,168 @@ __PMC_EV_ALIAS("IMPC_C0H_TRK_REQUEST.ALL", UCP_EVENT_84H_01H)
 
 #define PMC_EV_PPC970_FIRST		PMC_EV_PPC970_INSTR_COMPLETED
 #define PMC_EV_PPC970_LAST		PMC_EV_PPC970_ADDER
+
+#define __PMC_EV_E500() \
+	__PMC_EV(E500, CYCLES) \
+	__PMC_EV(E500, INSTR_COMPLETED) \
+	__PMC_EV(E500, UOPS_COMPLETED) \
+	__PMC_EV(E500, INSTR_FETCHED) \
+	__PMC_EV(E500, UOPS_DECODED) \
+	__PMC_EV(E500, PM_EVENT_TRANSITIONS) \
+	__PMC_EV(E500, PM_EVENT_CYCLES) \
+	__PMC_EV(E500, BRANCH_INSTRS_COMPLETED) \
+	__PMC_EV(E500, LOAD_UOPS_COMPLETED) \
+	__PMC_EV(E500, STORE_UOPS_COMPLETED) \
+	__PMC_EV(E500, CQ_REDIRECTS) \
+	__PMC_EV(E500, BRANCHES_FINISHED) \
+	__PMC_EV(E500, TAKEN_BRANCHES_FINISHED) \
+	__PMC_EV(E500, FINISHED_UNCOND_BRANCHES_MISS_BTB) \
+	__PMC_EV(E500, BRANCH_MISPRED) \
+	__PMC_EV(E500, BTB_BRANCH_MISPRED_FROM_DIRECTION) \
+	__PMC_EV(E500, BTB_HITS_PSEUDO_HITS) \
+	__PMC_EV(E500, CYCLES_DECODE_STALLED) \
+	__PMC_EV(E500, CYCLES_ISSUE_STALLED) \
+	__PMC_EV(E500, CYCLES_BRANCH_ISSUE_STALLED) \
+	__PMC_EV(E500, CYCLES_SU1_SCHED_STALLED) \
+	__PMC_EV(E500, CYCLES_SU2_SCHED_STALLED) \
+	__PMC_EV(E500, CYCLES_MU_SCHED_STALLED) \
+	__PMC_EV(E500, CYCLES_LRU_SCHED_STALLED) \
+	__PMC_EV(E500, CYCLES_BU_SCHED_STALLED) \
+	__PMC_EV(E500, TOTAL_TRANSLATED) \
+	__PMC_EV(E500, LOADS_TRANSLATED) \
+	__PMC_EV(E500, STORES_TRANSLATED) \
+	__PMC_EV(E500, TOUCHES_TRANSLATED) \
+	__PMC_EV(E500, CACHEOPS_TRANSLATED) \
+	__PMC_EV(E500, CACHE_INHIBITED_ACCESS_TRANSLATED) \
+	__PMC_EV(E500, GUARDED_LOADS_TRANSLATED) \
+	__PMC_EV(E500, WRITE_THROUGH_STORES_TRANSLATED) \
+	__PMC_EV(E500, MISALIGNED_LOAD_STORE_ACCESS_TRANSLATED) \
+	__PMC_EV(E500, TOTAL_ALLOCATED_TO_DLFB) \
+	__PMC_EV(E500, LOADS_TRANSLATED_ALLOCATED_TO_DLFB) \
+	__PMC_EV(E500, STORES_COMPLETED_ALLOCATED_TO_DLFB) \
+	__PMC_EV(E500, TOUCHES_TRANSLATED_ALLOCATED_TO_DLFB) \
+	__PMC_EV(E500, STORES_COMPLETED) \
+	__PMC_EV(E500, DATA_L1_CACHE_LOCKS) \
+	__PMC_EV(E500, DATA_L1_CACHE_RELOADS) \
+	__PMC_EV(E500, DATA_L1_CACHE_CASTOUTS) \
+	__PMC_EV(E500, LOAD_MISS_DLFB_FULL) \
+	__PMC_EV(E500, LOAD_MISS_LDQ_FULL) \
+	__PMC_EV(E500, LOAD_GUARDED_MISS) \
+	__PMC_EV(E500, STORE_TRANSLATE_WHEN_QUEUE_FULL) \
+	__PMC_EV(E500, ADDRESS_COLLISION) \
+	__PMC_EV(E500, DATA_MMU_MISS) \
+	__PMC_EV(E500, DATA_MMU_BUSY) \
+	__PMC_EV(E500, PART2_MISALIGNED_CACHE_ACCESS) \
+	__PMC_EV(E500, LOAD_MISS_DLFB_FULL_CYCLES) \
+	__PMC_EV(E500, LOAD_MISS_LDQ_FULL_CYCLES) \
+	__PMC_EV(E500, LOAD_GUARDED_MISS_CYCLES) \
+	__PMC_EV(E500, STORE_TRANSLATE_WHEN_QUEUE_FULL_CYCLES) \
+	__PMC_EV(E500, ADDRESS_COLLISION_CYCLES) \
+	__PMC_EV(E500, DATA_MMU_MISS_CYCLES) \
+	__PMC_EV(E500, DATA_MMU_BUSY_CYCLES) \
+	__PMC_EV(E500, PART2_MISALIGNED_CACHE_ACCESS_CYCLES) \
+	__PMC_EV(E500, INSTR_L1_CACHE_LOCKS) \
+	__PMC_EV(E500, INSTR_L1_CACHE_RELOADS) \
+	__PMC_EV(E500, INSTR_L1_CACHE_FETCHES) \
+	__PMC_EV(E500, INSTR_MMU_TLB4K_RELOADS) \
+	__PMC_EV(E500, INSTR_MMU_VSP_RELOADS) \
+	__PMC_EV(E500, DATA_MMU_TLB4K_RELOADS) \
+	__PMC_EV(E500, DATA_MMU_VSP_RELOADS) \
+	__PMC_EV(E500, L2MMU_MISSES) \
+	__PMC_EV(E500, BIU_MASTER_REQUESTS) \
+	__PMC_EV(E500, BIU_MASTER_INSTR_SIDE_REQUESTS) \
+	__PMC_EV(E500, BIU_MASTER_DATA_SIDE_REQUESTS) \
+	__PMC_EV(E500, BIU_MASTER_DATA_SIDE_CASTOUT_REQUESTS) \
+	__PMC_EV(E500, BIU_MASTER_RETRIES) \
+	__PMC_EV(E500, SNOOP_REQUESTS) \
+	__PMC_EV(E500, SNOOP_HITS) \
+	__PMC_EV(E500, SNOOP_PUSHES) \
+	__PMC_EV(E500, SNOOP_RETRIES) \
+	__PMC_EV(E500, DLFB_LOAD_MISS_CYCLES) \
+	__PMC_EV(E500, ILFB_FETCH_MISS_CYCLES) \
+	__PMC_EV(E500, EXT_INPU_INTR_LATENCY_CYCLES) \
+	__PMC_EV(E500, CRIT_INPUT_INTR_LATENCY_CYCLES) \
+	__PMC_EV(E500, EXT_INPUT_INTR_PENDING_LATENCY_CYCLES) \
+	__PMC_EV(E500, CRIT_INPUT_INTR_PENDING_LATENCY_CYCLES) \
+	__PMC_EV(E500, PMC0_OVERFLOW) \
+	__PMC_EV(E500, PMC1_OVERFLOW) \
+	__PMC_EV(E500, PMC2_OVERFLOW) \
+	__PMC_EV(E500, PMC3_OVERFLOW) \
+	__PMC_EV(E500, INTERRUPTS_TAKEN) \
+	__PMC_EV(E500, EXT_INPUT_INTR_TAKEN) \
+	__PMC_EV(E500, CRIT_INPUT_INTR_TAKEN) \
+	__PMC_EV(E500, SYSCALL_TRAP_INTR) \
+	__PMC_EV(E500, TLB_BIT_TRANSITIONS) \
+	__PMC_EV(E500, L2_LINEFILL_BUFFER) \
+	__PMC_EV(E500, LV2_VS) \
+	__PMC_EV(E500, CASTOUTS_RELEASED) \
+	__PMC_EV(E500, INTV_ALLOCATIONS) \
+	__PMC_EV(E500, DLFB_RETRIES_TO_MBAR) \
+	__PMC_EV(E500, STORE_RETRIES) \
+	__PMC_EV(E500, STASH_L1_HITS) \
+	__PMC_EV(E500, STASH_L2_HITS) \
+	__PMC_EV(E500, STASH_BUSY_1) \
+	__PMC_EV(E500, STASH_BUSY_2) \
+	__PMC_EV(E500, STASH_BUSY_3) \
+	__PMC_EV(E500, STASH_HITS) \
+	__PMC_EV(E500, STASH_HIT_DLFB) \
+	__PMC_EV(E500, STASH_REQUESTS) \
+	__PMC_EV(E500, STASH_REQUESTS_L1) \
+	__PMC_EV(E500, STASH_REQUESTS_L2) \
+	__PMC_EV(E500, STALLS_NO_CAQ_OR_COB) \
+	__PMC_EV(E500, L2_CACHE_ACCESSES) \
+	__PMC_EV(E500, L2_HIT_CACHE_ACCESSES) \
+	__PMC_EV(E500, L2_CACHE_DATA_ACCESSES) \
+	__PMC_EV(E500, L2_CACHE_DATA_HITS) \
+	__PMC_EV(E500, L2_CACHE_INSTR_ACCESSES) \
+	__PMC_EV(E500, L2_CACHE_INSTR_HITS) \
+	__PMC_EV(E500, L2_CACHE_ALLOCATIONS) \
+	__PMC_EV(E500, L2_CACHE_DATA_ALLOCATIONS) \
+	__PMC_EV(E500, L2_CACHE_DIRTY_DATA_ALLOCATIONS) \
+	__PMC_EV(E500, L2_CACHE_INSTR_ALLOCATIONS) \
+	__PMC_EV(E500, L2_CACHE_UPDATES) \
+	__PMC_EV(E500, L2_CACHE_CLEAN_UPDATES) \
+	__PMC_EV(E500, L2_CACHE_DIRTY_UPDATES) \
+	__PMC_EV(E500, L2_CACHE_CLEAN_REDUNDANT_UPDATES) \
+	__PMC_EV(E500, L2_CACHE_DIRTY_REDUNDANT_UPDATES) \
+	__PMC_EV(E500, L2_CACHE_LOCKS) \
+	__PMC_EV(E500, L2_CACHE_CASTOUTS) \
+	__PMC_EV(E500, L2_CACHE_DATA_DIRTY_HITS) \
+	__PMC_EV(E500, INSTR_LFB_WENT_HIGH_PRIORITY) \
+	__PMC_EV(E500, SNOOP_THROTTLING_TURNED_ON) \
+	__PMC_EV(E500, L2_CLEAN_LINE_INVALIDATIONS) \
+	__PMC_EV(E500, L2_INCOHERENT_LINE_INVALIDATIONS) \
+	__PMC_EV(E500, L2_COHERENT_LINE_INVALIDATIONS) \
+	__PMC_EV(E500, COHERENT_LOOKUP_MISS_DUE_TO_VALID_BUT_INCOHERENT_MATCHES) \
+	__PMC_EV(E500, IAC1S_DETECTED) \
+	__PMC_EV(E500, IAC2S_DETECTED) \
+	__PMC_EV(E500, DAC1S_DTECTED) \
+	__PMC_EV(E500, DAC2S_DTECTED) \
+	__PMC_EV(E500, DVT0_DETECTED) \
+	__PMC_EV(E500, DVT1_DETECTED) \
+	__PMC_EV(E500, DVT2_DETECTED) \
+	__PMC_EV(E500, DVT3_DETECTED) \
+	__PMC_EV(E500, DVT4_DETECTED) \
+	__PMC_EV(E500, DVT5_DETECTED) \
+	__PMC_EV(E500, DVT6_DETECTED) \
+	__PMC_EV(E500, DVT7_DETECTED) \
+	__PMC_EV(E500, CYCLES_COMPLETION_STALLED_NEXUS_FIFO_FULL) \
+	__PMC_EV(E500, FPU_DOUBLE_PUMP) \
+	__PMC_EV(E500, FPU_FINISH) \
+	__PMC_EV(E500, FPU_DIVIDE_CYCLES) \
+	__PMC_EV(E500, FPU_DENORM_INPUT_CYCLES) \
+	__PMC_EV(E500, FPU_RESULT_STALL_CYCLES) \
+	__PMC_EV(E500, FPU_FPSCR_FULL_STALL) \
+	__PMC_EV(E500, FPU_PIPE_SYNC_STALLS) \
+	__PMC_EV(E500, FPU_INPUT_DATA_STALLS) \
+	__PMC_EV(E500, DECORATED_LOADS) \
+	__PMC_EV(E500, DECORATED_STORES) \
+	__PMC_EV(E500, LOAD_RETRIES) \
+	__PMC_EV(E500, STWCX_SUCCESSES) \
+	__PMC_EV(E500, STWCX_FAILURES) \
+
+#define PMC_EV_E500_FIRST		PMC_EV_E500_CYCLES
+#define PMC_EV_E500_LAST		PMC_EV_E500_STWCX_FAILURES
 /*
  * All known PMC events.
  *
@@ -5400,6 +5562,9 @@ __PMC_EV_ALIAS("IMPC_C0H_TRK_REQUEST.ALL", UCP_EVENT_84H_01H)
  * 0x11300	0x00FF		MIPS 24K events
  * 0x11400	0x00FF		Octeon events
  * 0x11500	0x00FF		MIPS 74K events
+ * 0x13000	0x00FF		MPC7450 events
+ * 0x13100	0x00FF		IBM PPC970 events
+ * 0x13300	0x00FF		Freescale e500 events
  * 0x14000	0x0100		ARMv7 events
  * 0x20000	0x1000		Software events
  */
@@ -5436,6 +5601,8 @@ __PMC_EV_ALIAS("IMPC_C0H_TRK_REQUEST.ALL", UCP_EVENT_84H_01H)
 	__PMC_EV_PPC7450()			\
 	__PMC_EV_BLOCK(PPC970,	0x13100)	\
 	__PMC_EV_PPC970()			\
+	__PMC_EV_BLOCK(E500,	0x13300)	\
+	__PMC_EV_E500()				\
 	__PMC_EV_BLOCK(ARMV7,	0x14000)	\
 	__PMC_EV_ARMV7()			\