perf_event_p6.c 6.3 KB
Newer Older
1 2 3 4
#include <linux/perf_event.h>
#include <linux/types.h>

#include "perf_event.h"
5 6 7 8 9 10

/*
 * Not sure about some of these
 */
static const u64 p6_perfmon_event_map[] =
{
11 12 13 14 15 16 17 18 19 20 21
  [PERF_COUNT_HW_CPU_CYCLES]		= 0x0079,	/* CPU_CLK_UNHALTED */
  [PERF_COUNT_HW_INSTRUCTIONS]		= 0x00c0,	/* INST_RETIRED     */
  [PERF_COUNT_HW_CACHE_REFERENCES]	= 0x0f2e,	/* L2_RQSTS:M:E:S:I */
  [PERF_COUNT_HW_CACHE_MISSES]		= 0x012e,	/* L2_RQSTS:I       */
  [PERF_COUNT_HW_BRANCH_INSTRUCTIONS]	= 0x00c4,	/* BR_INST_RETIRED  */
  [PERF_COUNT_HW_BRANCH_MISSES]		= 0x00c5,	/* BR_MISS_PRED_RETIRED */
  [PERF_COUNT_HW_BUS_CYCLES]		= 0x0062,	/* BUS_DRDY_CLOCKS  */
  [PERF_COUNT_HW_STALLED_CYCLES_FRONTEND] = 0x00a2,	/* RESOURCE_STALLS  */

};

22
static u64 p6_hw_cache_event_ids
23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110
				[PERF_COUNT_HW_CACHE_MAX]
				[PERF_COUNT_HW_CACHE_OP_MAX]
				[PERF_COUNT_HW_CACHE_RESULT_MAX] =
{
 [ C(L1D) ] = {
	[ C(OP_READ) ] = {
		[ C(RESULT_ACCESS) ] = 0x0043,	/* DATA_MEM_REFS       */
                [ C(RESULT_MISS)   ] = 0x0045,	/* DCU_LINES_IN        */
	},
	[ C(OP_WRITE) ] = {
		[ C(RESULT_ACCESS) ] = 0,
		[ C(RESULT_MISS)   ] = 0x0f29,	/* L2_LD:M:E:S:I       */
	},
        [ C(OP_PREFETCH) ] = {
		[ C(RESULT_ACCESS) ] = 0,
		[ C(RESULT_MISS)   ] = 0,
        },
 },
 [ C(L1I ) ] = {
	[ C(OP_READ) ] = {
		[ C(RESULT_ACCESS) ] = 0x0080,	/* IFU_IFETCH         */
		[ C(RESULT_MISS)   ] = 0x0f28,	/* L2_IFETCH:M:E:S:I  */
	},
	[ C(OP_WRITE) ] = {
		[ C(RESULT_ACCESS) ] = -1,
		[ C(RESULT_MISS)   ] = -1,
	},
	[ C(OP_PREFETCH) ] = {
		[ C(RESULT_ACCESS) ] = 0,
		[ C(RESULT_MISS)   ] = 0,
	},
 },
 [ C(LL  ) ] = {
	[ C(OP_READ) ] = {
		[ C(RESULT_ACCESS) ] = 0,
		[ C(RESULT_MISS)   ] = 0,
	},
	[ C(OP_WRITE) ] = {
		[ C(RESULT_ACCESS) ] = 0,
		[ C(RESULT_MISS)   ] = 0x0025,	/* L2_M_LINES_INM     */
	},
	[ C(OP_PREFETCH) ] = {
		[ C(RESULT_ACCESS) ] = 0,
		[ C(RESULT_MISS)   ] = 0,
	},
 },
 [ C(DTLB) ] = {
	[ C(OP_READ) ] = {
		[ C(RESULT_ACCESS) ] = 0x0043,	/* DATA_MEM_REFS      */
		[ C(RESULT_MISS)   ] = 0,
	},
	[ C(OP_WRITE) ] = {
		[ C(RESULT_ACCESS) ] = 0,
		[ C(RESULT_MISS)   ] = 0,
	},
	[ C(OP_PREFETCH) ] = {
		[ C(RESULT_ACCESS) ] = 0,
		[ C(RESULT_MISS)   ] = 0,
	},
 },
 [ C(ITLB) ] = {
	[ C(OP_READ) ] = {
		[ C(RESULT_ACCESS) ] = 0x0080,	/* IFU_IFETCH         */
		[ C(RESULT_MISS)   ] = 0x0085,	/* ITLB_MISS          */
	},
	[ C(OP_WRITE) ] = {
		[ C(RESULT_ACCESS) ] = -1,
		[ C(RESULT_MISS)   ] = -1,
	},
	[ C(OP_PREFETCH) ] = {
		[ C(RESULT_ACCESS) ] = -1,
		[ C(RESULT_MISS)   ] = -1,
	},
 },
 [ C(BPU ) ] = {
	[ C(OP_READ) ] = {
		[ C(RESULT_ACCESS) ] = 0x00c4,	/* BR_INST_RETIRED      */
		[ C(RESULT_MISS)   ] = 0x00c5,	/* BR_MISS_PRED_RETIRED */
        },
	[ C(OP_WRITE) ] = {
		[ C(RESULT_ACCESS) ] = -1,
		[ C(RESULT_MISS)   ] = -1,
	},
	[ C(OP_PREFETCH) ] = {
		[ C(RESULT_ACCESS) ] = -1,
		[ C(RESULT_MISS)   ] = -1,
	},
 },
111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129
};

static u64 p6_pmu_event_map(int hw_event)
{
	return p6_perfmon_event_map[hw_event];
}

/*
 * Event setting that is specified not to count anything.
 * We use this to effectively disable a counter.
 *
 * L2_RQSTS with 0 MESI unit mask.
 */
#define P6_NOP_EVENT			0x0000002EULL

static struct event_constraint p6_event_constraints[] =
{
	INTEL_EVENT_CONSTRAINT(0xc1, 0x1),	/* FLOPS */
	INTEL_EVENT_CONSTRAINT(0x10, 0x1),	/* FP_COMP_OPS_EXE */
130
	INTEL_EVENT_CONSTRAINT(0x11, 0x2),	/* FP_ASSIST */
131 132 133 134 135 136 137 138 139 140 141 142
	INTEL_EVENT_CONSTRAINT(0x12, 0x2),	/* MUL */
	INTEL_EVENT_CONSTRAINT(0x13, 0x2),	/* DIV */
	INTEL_EVENT_CONSTRAINT(0x14, 0x1),	/* CYCLES_DIV_BUSY */
	EVENT_CONSTRAINT_END
};

static void p6_pmu_disable_all(void)
{
	u64 val;

	/* p6 only has one enable register */
	rdmsrl(MSR_P6_EVNTSEL0, val);
143
	val &= ~ARCH_PERFMON_EVENTSEL_ENABLE;
144 145 146
	wrmsrl(MSR_P6_EVNTSEL0, val);
}

147
static void p6_pmu_enable_all(int added)
148 149 150 151 152
{
	unsigned long val;

	/* p6 only has one enable register */
	rdmsrl(MSR_P6_EVNTSEL0, val);
153
	val |= ARCH_PERFMON_EVENTSEL_ENABLE;
154 155 156 157
	wrmsrl(MSR_P6_EVNTSEL0, val);
}

static inline void
158
p6_pmu_disable_event(struct perf_event *event)
159
{
160
	struct hw_perf_event *hwc = &event->hw;
161 162
	u64 val = P6_NOP_EVENT;

163
	(void)wrmsrl_safe(hwc->config_base, val);
164 165
}

166
static void p6_pmu_enable_event(struct perf_event *event)
167
{
168
	struct hw_perf_event *hwc = &event->hw;
169 170 171
	u64 val;

	val = hwc->config;
172 173 174 175 176 177 178

	/*
	 * p6 only has a global event enable, set on PerfEvtSel0
	 * We "disable" events by programming P6_NOP_EVENT
	 * and we rely on p6_pmu_enable_all() being called
	 * to actually enable the events.
	 */
179

180
	(void)wrmsrl_safe(hwc->config_base, val);
181 182
}

183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199
PMU_FORMAT_ATTR(event,	"config:0-7"	);
PMU_FORMAT_ATTR(umask,	"config:8-15"	);
PMU_FORMAT_ATTR(edge,	"config:18"	);
PMU_FORMAT_ATTR(pc,	"config:19"	);
PMU_FORMAT_ATTR(inv,	"config:23"	);
PMU_FORMAT_ATTR(cmask,	"config:24-31"	);

static struct attribute *intel_p6_formats_attr[] = {
	&format_attr_event.attr,
	&format_attr_umask.attr,
	&format_attr_edge.attr,
	&format_attr_pc.attr,
	&format_attr_inv.attr,
	&format_attr_cmask.attr,
	NULL,
};

200
static __initconst const struct x86_pmu p6_pmu = {
201 202 203 204 205 206
	.name			= "p6",
	.handle_irq		= x86_pmu_handle_irq,
	.disable_all		= p6_pmu_disable_all,
	.enable_all		= p6_pmu_enable_all,
	.enable			= p6_pmu_enable_event,
	.disable		= p6_pmu_disable_event,
207
	.hw_config		= x86_pmu_hw_config,
208
	.schedule_events	= x86_schedule_events,
209 210 211 212 213 214 215
	.eventsel		= MSR_P6_EVNTSEL0,
	.perfctr		= MSR_P6_PERFCTR0,
	.event_map		= p6_pmu_event_map,
	.max_events		= ARRAY_SIZE(p6_perfmon_event_map),
	.apic			= 1,
	.max_period		= (1ULL << 31) - 1,
	.version		= 0,
216
	.num_counters		= 2,
217 218 219 220 221 222 223
	/*
	 * Events have 40 bits implemented. However they are designed such
	 * that bits [32-39] are sign extensions of bit 31. As such the
	 * effective width of a event for P6-like PMU is 32 bits only.
	 *
	 * See IA-32 Intel Architecture Software developer manual Vol 3B
	 */
224 225
	.cntval_bits		= 32,
	.cntval_mask		= (1ULL << 32) - 1,
226 227
	.get_event_constraints	= x86_get_event_constraints,
	.event_constraints	= p6_event_constraints,
228 229

	.format_attrs		= intel_p6_formats_attr,
230 231
	.events_sysfs_show	= intel_event_sysfs_show,

232 233
};

234
__init int p6_pmu_init(void)
235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255
{
	switch (boot_cpu_data.x86_model) {
	case 1:
	case 3:  /* Pentium Pro */
	case 5:
	case 6:  /* Pentium II */
	case 7:
	case 8:
	case 11: /* Pentium III */
	case 9:
	case 13:
		/* Pentium M */
		break;
	default:
		pr_cont("unsupported p6 CPU model %d ",
			boot_cpu_data.x86_model);
		return -ENODEV;
	}

	x86_pmu = p6_pmu;

256 257 258 259
	memcpy(hw_cache_event_ids, p6_hw_cache_event_ids,
		sizeof(hw_cache_event_ids));


260 261
	return 0;
}