1/* 2 * Performance counter support for e500 family processors. 3 * 4 * Copyright 2008-2009 Paul Mackerras, IBM Corporation. 5 * Copyright 2010 Freescale Semiconductor, Inc. 6 * 7 * This program is free software; you can redistribute it and/or 8 * modify it under the terms of the GNU General Public License 9 * as published by the Free Software Foundation; either version 10 * 2 of the License, or (at your option) any later version. 11 */ 12#include <linux/string.h> 13#include <linux/perf_event.h> 14#include <asm/reg.h> 15#include <asm/cputable.h> 16 17/* 18 * Map of generic hardware event types to hardware events 19 * Zero if unsupported 20 */ 21static int e500_generic_events[] = { 22 [PERF_COUNT_HW_CPU_CYCLES] = 1, 23 [PERF_COUNT_HW_INSTRUCTIONS] = 2, 24 [PERF_COUNT_HW_CACHE_MISSES] = 41, /* Data L1 cache reloads */ 25 [PERF_COUNT_HW_BRANCH_INSTRUCTIONS] = 12, 26 [PERF_COUNT_HW_BRANCH_MISSES] = 15, 27}; 28 29#define C(x) PERF_COUNT_HW_CACHE_##x 30 31/* 32 * Table of generalized cache-related events. 33 * 0 means not supported, -1 means nonsensical, other values 34 * are event codes. 35 */ 36static int e500_cache_events[C(MAX)][C(OP_MAX)][C(RESULT_MAX)] = { 37 /* 38 * D-cache misses are not split into read/write/prefetch; 39 * use raw event 41. 40 */ 41 [C(L1D)] = { /* RESULT_ACCESS RESULT_MISS */ 42 [C(OP_READ)] = { 27, 0 }, 43 [C(OP_WRITE)] = { 28, 0 }, 44 [C(OP_PREFETCH)] = { 29, 0 }, 45 }, 46 [C(L1I)] = { /* RESULT_ACCESS RESULT_MISS */ 47 [C(OP_READ)] = { 2, 60 }, 48 [C(OP_WRITE)] = { -1, -1 }, 49 [C(OP_PREFETCH)] = { 0, 0 }, 50 }, 51 /* 52 * Assuming LL means L2, it's not a good match for this model. 53 * It allocates only on L1 castout or explicit prefetch, and 54 * does not have separate read/write events (but it does have 55 * separate instruction/data events). 56 */ 57 [C(LL)] = { /* RESULT_ACCESS RESULT_MISS */ 58 [C(OP_READ)] = { 0, 0 }, 59 [C(OP_WRITE)] = { 0, 0 }, 60 [C(OP_PREFETCH)] = { 0, 0 }, 61 }, 62 /* 63 * There are data/instruction MMU misses, but that's a miss on 64 * the chip's internal level-one TLB which is probably not 65 * what the user wants. Instead, unified level-two TLB misses 66 * are reported here. 67 */ 68 [C(DTLB)] = { /* RESULT_ACCESS RESULT_MISS */ 69 [C(OP_READ)] = { 26, 66 }, 70 [C(OP_WRITE)] = { -1, -1 }, 71 [C(OP_PREFETCH)] = { -1, -1 }, 72 }, 73 [C(BPU)] = { /* RESULT_ACCESS RESULT_MISS */ 74 [C(OP_READ)] = { 12, 15 }, 75 [C(OP_WRITE)] = { -1, -1 }, 76 [C(OP_PREFETCH)] = { -1, -1 }, 77 }, 78}; 79 80static int num_events = 128; 81 82/* Upper half of event id is PMLCb, for threshold events */ 83static u64 e500_xlate_event(u64 event_id) 84{ 85 u32 event_low = (u32)event_id; 86 u64 ret; 87 88 if (event_low >= num_events) 89 return 0; 90 91 ret = FSL_EMB_EVENT_VALID; 92 93 if (event_low >= 76 && event_low <= 81) { 94 ret |= FSL_EMB_EVENT_RESTRICTED; 95 ret |= event_id & 96 (FSL_EMB_EVENT_THRESHMUL | FSL_EMB_EVENT_THRESH); 97 } else if (event_id & 98 (FSL_EMB_EVENT_THRESHMUL | FSL_EMB_EVENT_THRESH)) { 99 /* Threshold requested on non-threshold event */ 100 return 0; 101 } 102 103 return ret; 104} 105 106static struct fsl_emb_pmu e500_pmu = { 107 .name = "e500 family", 108 .n_counter = 4, 109 .n_restricted = 2, 110 .xlate_event = e500_xlate_event, 111 .n_generic = ARRAY_SIZE(e500_generic_events), 112 .generic_events = e500_generic_events, 113 .cache_events = &e500_cache_events, 114}; 115 116static int init_e500_pmu(void) 117{ 118 if (!cur_cpu_spec->oprofile_cpu_type) 119 return -ENODEV; 120 121 if (!strcmp(cur_cpu_spec->oprofile_cpu_type, "ppc/e500mc")) 122 num_events = 256; 123 else if (strcmp(cur_cpu_spec->oprofile_cpu_type, "ppc/e500")) 124 return -ENODEV; 125 126 return register_fsl_emb_pmu(&e500_pmu); 127} 128 129arch_initcall(init_e500_pmu); 130