[PATCH] powerpc/perf: Power8 PMU support
Benjamin Herrenschmidt
benh at kernel.crashing.org
Mon Apr 15 17:31:26 EST 2013
On Mon, 2013-04-15 at 14:17 +1000, Michael Ellerman wrote:
> This patch adds preliminary support for the power8 PMU to perf.
Might be worthwhile to have a small blurb explaining roughly what you
mean by "preliminary" :-)
Cheers,
Ben.
> Signed-off-by: Michael Ellerman <michael at ellerman.id.au>
> ---
> arch/powerpc/perf/Makefile | 3 +-
> arch/powerpc/perf/power8-pmu.c | 454 ++++++++++++++++++++++++++++++++++++++++
> 2 files changed, 456 insertions(+), 1 deletion(-)
> create mode 100644 arch/powerpc/perf/power8-pmu.c
>
> diff --git a/arch/powerpc/perf/Makefile b/arch/powerpc/perf/Makefile
> index af3fac2..472db18 100644
> --- a/arch/powerpc/perf/Makefile
> +++ b/arch/powerpc/perf/Makefile
> @@ -4,7 +4,8 @@ obj-$(CONFIG_PERF_EVENTS) += callchain.o
>
> obj-$(CONFIG_PPC_PERF_CTRS) += core-book3s.o
> obj64-$(CONFIG_PPC_PERF_CTRS) += power4-pmu.o ppc970-pmu.o power5-pmu.o \
> - power5+-pmu.o power6-pmu.o power7-pmu.o
> + power5+-pmu.o power6-pmu.o power7-pmu.o \
> + power8-pmu.o
> obj32-$(CONFIG_PPC_PERF_CTRS) += mpc7450-pmu.o
>
> obj-$(CONFIG_FSL_EMB_PERF_EVENT) += core-fsl-emb.o
> diff --git a/arch/powerpc/perf/power8-pmu.c b/arch/powerpc/perf/power8-pmu.c
> new file mode 100644
> index 0000000..106ae0b
> --- /dev/null
> +++ b/arch/powerpc/perf/power8-pmu.c
> @@ -0,0 +1,454 @@
> +/*
> + * Performance counter support for POWER8 processors.
> + *
> + * Copyright 2009 Paul Mackerras, IBM Corporation.
> + * Copyright 2013 Michael Ellerman, IBM Corporation.
> + *
> + * This program is free software; you can redistribute it and/or
> + * modify it under the terms of the GNU General Public License
> + * as published by the Free Software Foundation; either version
> + * 2 of the License, or (at your option) any later version.
> + */
> +
> +#include <linux/kernel.h>
> +#include <linux/perf_event.h>
> +#include <asm/firmware.h>
> +
> +
> +/*
> + * Some power8 event codes.
> + */
> +#define PM_CYC 0x0001e
> +#define PM_GCT_NOSLOT_CYC 0x100f8
> +#define PM_CMPLU_STALL 0x4000a /* or 0x1e054 */
> +#define PM_INST_CMPL 0x00002
> +#define PM_BRU_FIN 0x10068
> +#define PM_BR_MPRED_CMPL 0x400f6
> +
> +
> +/*
> + * Raw event encoding for POWER8:
> + *
> + * 60 56 52 48 44 40 36 32
> + * | - - - - | - - - - | - - - - | - - - - | - - - - | - - - - | - - - - | - - - - |
> + * [ thresh_cmp ] [ thresh_ctl ]
> + * |
> + * thresh start/stop OR FAB match -*
> + *
> + * 28 24 20 16 12 8 4 0
> + * | - - - - | - - - - | - - - - | - - - - | - - - - | - - - - | - - - - | - - - - |
> + * [ ] [ sample ] [cache] [ pmc ] [unit ] c m [ pmcxsel ]
> + * | | | | |
> + * | | | | *- mark
> + * | | *- L1/L2/L3 cache_sel |
> + * | | |
> + * | *- sampling mode for marked events *- combine
> + * |
> + * *- thresh_sel
> + *
> + * Below uses IBM bit numbering.
> + *
> + * MMCR1[x:y] = unit (PMCxUNIT)
> + * MMCR1[x] = combine (PMCxCOMB)
> + *
> + * if pmc == 3 and unit == 0 and pmcxsel[0:6] == 0b0101011
> + * # PM_MRK_FAB_RSP_MATCH
> + * MMCR1[20:27] = thresh_ctl (FAB_CRESP_MATCH / FAB_TYPE_MATCH)
> + * else if pmc == 4 and unit == 0xf and pmcxsel[0:6] == 0b0101001
> + * # PM_MRK_FAB_RSP_MATCH_CYC
> + * MMCR1[20:27] = thresh_ctl (FAB_CRESP_MATCH / FAB_TYPE_MATCH)
> + * else
> + * MMCRA[48:55] = thresh_ctl (THRESH START/END)
> + *
> + * if thresh_sel:
> + * MMCRA[45:47] = thresh_sel
> + *
> + * if thresh_cmp:
> + * MMCRA[22:24] = thresh_cmp[0:2]
> + * MMCRA[25:31] = thresh_cmp[3:9]
> + *
> + * if unit == 6 or unit == 7
> + * MMCRC[53:55] = cache_sel[1:3] (L2EVENT_SEL)
> + * else if unit == 8 or unit == 9:
> + * if cache_sel[0] == 0: # L3 bank
> + * MMCRC[47:49] = cache_sel[1:3] (L3EVENT_SEL0)
> + * else if cache_sel[0] == 1:
> + * MMCRC[50:51] = cache_sel[2:3] (L3EVENT_SEL1)
> + * else if cache_sel[1]: # L1 event
> + * MMCR1[16] = cache_sel[2]
> + * MMCR1[17] = cache_sel[3]
> + *
> + * if mark:
> + * MMCRA[63] = 1 (SAMPLE_ENABLE)
> + * MMCRA[57:59] = sample[0:2] (RAND_SAMP_ELIG)
> + * MMCRA[61:62] = sample[3:4] (RAND_SAMP_MODE)
> + *
> + */
> +
> +#define EVENT_THR_CMP_SHIFT 40 /* Threshold CMP value */
> +#define EVENT_THR_CMP_MASK 0x3ff
> +#define EVENT_THR_CTL_SHIFT 32 /* Threshold control value (start/stop) */
> +#define EVENT_THR_CTL_MASK 0xffull
> +#define EVENT_THR_SEL_SHIFT 29 /* Threshold select value */
> +#define EVENT_THR_SEL_MASK 0x7
> +#define EVENT_THRESH_SHIFT 29 /* All threshold bits */
> +#define EVENT_THRESH_MASK 0x1fffffull
> +#define EVENT_SAMPLE_SHIFT 24 /* Sampling mode & eligibility */
> +#define EVENT_SAMPLE_MASK 0x1f
> +#define EVENT_CACHE_SEL_SHIFT 20 /* L2/L3 cache select */
> +#define EVENT_CACHE_SEL_MASK 0xf
> +#define EVENT_IS_L1 (4 << EVENT_CACHE_SEL_SHIFT)
> +#define EVENT_PMC_SHIFT 16 /* PMC number (1-based) */
> +#define EVENT_PMC_MASK 0xf
> +#define EVENT_UNIT_SHIFT 12 /* Unit */
> +#define EVENT_UNIT_MASK 0xf
> +#define EVENT_COMBINE_SHIFT 11 /* Combine bit */
> +#define EVENT_COMBINE_MASK 0x1
> +#define EVENT_MARKED_SHIFT 8 /* Marked bit */
> +#define EVENT_MARKED_MASK 0x1
> +#define EVENT_IS_MARKED (EVENT_MARKED_MASK << EVENT_MARKED_SHIFT)
> +#define EVENT_PSEL_MASK 0xff /* PMCxSEL value */
> +
> +/*
> + * Layout of constraint bits:
> + *
> + * 60 56 52 48 44 40 36 32
> + * | - - - - | - - - - | - - - - | - - - - | - - - - | - - - - | - - - - | - - - - |
> + * [ fab_match ] [ thresh_cmp ] [ thresh_ctl ] [ ]
> + * |
> + * thresh_sel -*
> + *
> + * 28 24 20 16 12 8 4 0
> + * | - - - - | - - - - | - - - - | - - - - | - - - - | - - - - | - - - - | - - - - |
> + * [ ] [ sample ] [ ] [6] [5] [4] [3] [2] [1]
> + * | |
> + * L1 I/D qualifier -* | Count of events for each PMC.
> + * | p1, p2, p3, p4, p5, p6.
> + * nc - number of counters -*
> + *
> + * The PMC fields P1..P6, and NC, are adder fields. As we accumulate constraints
> + * we want the low bit of each field to be added to any existing value.
> + *
> + * Everything else is a value field.
> + */
> +
> +#define CNST_FAB_MATCH_VAL(v) (((v) & EVENT_THR_CTL_MASK) << 56)
> +#define CNST_FAB_MATCH_MASK CNST_FAB_MATCH_VAL(EVENT_THR_CTL_MASK)
> +
> +/* We just throw all the threshold bits into the constraint */
> +#define CNST_THRESH_VAL(v) (((v) & EVENT_THRESH_MASK) << 32)
> +#define CNST_THRESH_MASK CNST_THRESH_VAL(EVENT_THRESH_MASK)
> +
> +#define CNST_L1_QUAL_VAL(v) (((v) & 3) << 22)
> +#define CNST_L1_QUAL_MASK CNST_L1_QUAL_VAL(3)
> +
> +#define CNST_SAMPLE_VAL(v) (((v) & EVENT_SAMPLE_MASK) << 16)
> +#define CNST_SAMPLE_MASK CNST_SAMPLE_VAL(EVENT_SAMPLE_MASK)
> +
> +/*
> + * For NC we are counting up to 4 events. This requires three bits, and we need
> + * the fifth event to overflow and set the 4th bit. To achieve that we bias the
> + * fields by 3 in test_adder.
> + */
> +#define CNST_NC_SHIFT 12
> +#define CNST_NC_VAL (1 << CNST_NC_SHIFT)
> +#define CNST_NC_MASK (8 << CNST_NC_SHIFT)
> +#define POWER8_TEST_ADDER (3 << CNST_NC_SHIFT)
> +
> +/*
> + * For the per-PMC fields we have two bits. The low bit is added, so if two
> + * events ask for the same PMC the sum will overflow, setting the high bit,
> + * indicating an error. So our mask sets the high bit.
> + */
> +#define CNST_PMC_SHIFT(pmc) ((pmc - 1) * 2)
> +#define CNST_PMC_VAL(pmc) (1 << CNST_PMC_SHIFT(pmc))
> +#define CNST_PMC_MASK(pmc) (2 << CNST_PMC_SHIFT(pmc))
> +
> +/* Our add_fields is defined as: */
> +#define POWER8_ADD_FIELDS \
> + CNST_PMC_VAL(1) | CNST_PMC_VAL(2) | CNST_PMC_VAL(3) | \
> + CNST_PMC_VAL(4) | CNST_PMC_VAL(5) | CNST_PMC_VAL(6) | CNST_NC_VAL
> +
> +
> +/* Bits in MMCR1 for POWER8 */
> +#define MMCR1_UNIT_SHIFT(pmc) (60 - (4 * ((pmc) - 1)))
> +#define MMCR1_COMBINE_SHIFT(pmc) (35 - ((pmc) - 1))
> +#define MMCR1_PMCSEL_SHIFT(pmc) (24 - (((pmc) - 1)) * 8)
> +#define MMCR1_DC_QUAL_SHIFT 47
> +#define MMCR1_IC_QUAL_SHIFT 46
> +
> +/* Bits in MMCRA for POWER8 */
> +#define MMCRA_SAMP_MODE_SHIFT 1
> +#define MMCRA_SAMP_ELIG_SHIFT 4
> +#define MMCRA_THR_CTL_SHIFT 8
> +#define MMCRA_THR_SEL_SHIFT 16
> +#define MMCRA_THR_CMP_SHIFT 32
> +#define MMCRA_SDAR_MODE_TLB (1ull << 42)
> +
> +
> +static inline bool event_is_fab_match(u64 event)
> +{
> + /* Only check pmc, unit and pmcxsel, ignore the edge bit (0) */
> + event &= 0xff0fe;
> +
> + /* PM_MRK_FAB_RSP_MATCH & PM_MRK_FAB_RSP_MATCH_CYC */
> + return (event == 0x30056 || event == 0x4f052);
> +}
> +
> +static int power8_get_constraint(u64 event, unsigned long *maskp, unsigned long *valp)
> +{
> + unsigned int unit, pmc, cache;
> + unsigned long mask, value;
> +
> + mask = value = 0;
> +
> + pmc = (event >> EVENT_PMC_SHIFT) & EVENT_PMC_MASK;
> + unit = (event >> EVENT_UNIT_SHIFT) & EVENT_UNIT_MASK;
> + cache = (event >> EVENT_CACHE_SEL_SHIFT) & EVENT_CACHE_SEL_MASK;
> +
> + if (pmc) {
> + if (pmc > 6)
> + return -1;
> +
> + mask |= CNST_PMC_MASK(pmc);
> + value |= CNST_PMC_VAL(pmc);
> +
> + if (pmc >= 5 && event != 0x500fa && event != 0x600f4)
> + return -1;
> + }
> +
> + if (pmc <= 4) {
> + /*
> + * Add to number of counters in use. Note this includes events with
> + * a PMC of 0 - they still need a PMC, it's just assigned later.
> + * Don't count events on PMC 5 & 6, there is only one valid event
> + * on each of those counters, and they are handled above.
> + */
> + mask |= CNST_NC_MASK;
> + value |= CNST_NC_VAL;
> + }
> +
> + if (unit >= 6 && unit <= 9) {
> + /*
> + * L2/L3 events contain a cache selector field, which is
> + * supposed to be programmed into MMCRC. However MMCRC is only
> + * HV writable, and there is no API for guest kernels to modify
> + * it. The solution is for the hypervisor to initialise the
> + * field to zeroes, and for us to only ever allow events that
> + * have a cache selector of zero.
> + */
> + if (cache)
> + return -1;
> +
> + } else if (event & EVENT_IS_L1) {
> + mask |= CNST_L1_QUAL_MASK;
> + value |= CNST_L1_QUAL_VAL(cache);
> + }
> +
> + if (event & EVENT_IS_MARKED) {
> + mask |= CNST_SAMPLE_MASK;
> + value |= CNST_SAMPLE_VAL(event >> EVENT_SAMPLE_SHIFT);
> + }
> +
> + /*
> + * Special case for PM_MRK_FAB_RSP_MATCH and PM_MRK_FAB_RSP_MATCH_CYC,
> + * the threshold control bits are used for the match value.
> + */
> + if (event_is_fab_match(event)) {
> + mask |= CNST_FAB_MATCH_MASK;
> + value |= CNST_FAB_MATCH_VAL(event >> EVENT_THR_CTL_SHIFT);
> + } else {
> + /*
> + * Check the mantissa upper two bits are not zero, unless the
> + * exponent is also zero. See the THRESH_CMP_MANTISSA doc.
> + */
> + unsigned int cmp, exp;
> +
> + cmp = (event >> EVENT_THR_CMP_SHIFT) & EVENT_THR_CMP_MASK;
> + exp = cmp >> 7;
> +
> + if (exp && (cmp & 0x60) == 0)
> + return -1;
> +
> + mask |= CNST_THRESH_MASK;
> + value |= CNST_THRESH_VAL(event >> EVENT_THRESH_SHIFT);
> + }
> +
> + *maskp = mask;
> + *valp = value;
> +
> + return 0;
> +}
> +
> +static int power8_compute_mmcr(u64 event[], int n_ev,
> + unsigned int hwc[], unsigned long mmcr[])
> +{
> + unsigned long mmcra, mmcr1, unit, combine, psel, cache, val;
> + unsigned int pmc, pmc_inuse;
> + int i;
> +
> + pmc_inuse = 0;
> +
> + /* First pass to count resource use */
> + for (i = 0; i < n_ev; ++i) {
> + pmc = (event[i] >> EVENT_PMC_SHIFT) & EVENT_PMC_MASK;
> + if (pmc)
> + pmc_inuse |= 1 << pmc;
> + }
> +
> + /* In continous sampling mode, update SDAR on TLB miss */
> + mmcra = MMCRA_SDAR_MODE_TLB;
> + mmcr1 = 0;
> +
> + /* Second pass: assign PMCs, set all MMCR1 fields */
> + for (i = 0; i < n_ev; ++i) {
> + pmc = (event[i] >> EVENT_PMC_SHIFT) & EVENT_PMC_MASK;
> + unit = (event[i] >> EVENT_UNIT_SHIFT) & EVENT_UNIT_MASK;
> + combine = (event[i] >> EVENT_COMBINE_SHIFT) & EVENT_COMBINE_MASK;
> + psel = event[i] & EVENT_PSEL_MASK;
> +
> + if (!pmc) {
> + for (pmc = 1; pmc <= 4; ++pmc) {
> + if (!(pmc_inuse & (1 << pmc)))
> + break;
> + }
> +
> + pmc_inuse |= 1 << pmc;
> + }
> +
> + if (pmc <= 4) {
> + mmcr1 |= unit << MMCR1_UNIT_SHIFT(pmc);
> + mmcr1 |= combine << MMCR1_COMBINE_SHIFT(pmc);
> + mmcr1 |= psel << MMCR1_PMCSEL_SHIFT(pmc);
> + }
> +
> + if (event[i] & EVENT_IS_L1) {
> + cache = event[i] >> EVENT_CACHE_SEL_SHIFT;
> + mmcr1 |= (cache & 1) << MMCR1_IC_QUAL_SHIFT;
> + cache >>= 1;
> + mmcr1 |= (cache & 1) << MMCR1_DC_QUAL_SHIFT;
> + }
> +
> + if (event[i] & EVENT_IS_MARKED) {
> + mmcra |= MMCRA_SAMPLE_ENABLE;
> +
> + val = (event[i] >> EVENT_SAMPLE_SHIFT) & EVENT_SAMPLE_MASK;
> + if (val) {
> + mmcra |= (val & 3) << MMCRA_SAMP_MODE_SHIFT;
> + mmcra |= (val >> 2) << MMCRA_SAMP_ELIG_SHIFT;
> + }
> + }
> +
> + /*
> + * PM_MRK_FAB_RSP_MATCH and PM_MRK_FAB_RSP_MATCH_CYC,
> + * the threshold bits are used for the match value.
> + */
> + if (event_is_fab_match(event[i])) {
> + mmcr1 |= (event[i] >> EVENT_THR_CTL_SHIFT) &
> + EVENT_THR_CTL_MASK;
> + } else {
> + val = (event[i] >> EVENT_THR_CTL_SHIFT) & EVENT_THR_CTL_MASK;
> + mmcra |= val << MMCRA_THR_CTL_SHIFT;
> + val = (event[i] >> EVENT_THR_SEL_SHIFT) & EVENT_THR_SEL_MASK;
> + mmcra |= val << MMCRA_THR_SEL_SHIFT;
> + val = (event[i] >> EVENT_THR_CMP_SHIFT) & EVENT_THR_CMP_MASK;
> + mmcra |= val << MMCRA_THR_CMP_SHIFT;
> + }
> +
> + hwc[i] = pmc - 1;
> + }
> +
> + /* Return MMCRx values */
> + mmcr[0] = 0;
> +
> + /* pmc_inuse is 1-based */
> + if (pmc_inuse & 2)
> + mmcr[0] = MMCR0_PMC1CE;
> +
> + if (pmc_inuse & 0x7c)
> + mmcr[0] |= MMCR0_PMCjCE;
> +
> + mmcr[1] = mmcr1;
> + mmcr[2] = mmcra;
> +
> + return 0;
> +}
> +
> +static void power8_disable_pmc(unsigned int pmc, unsigned long mmcr[])
> +{
> + if (pmc <= 3)
> + mmcr[1] &= ~(0xffUL << MMCR1_PMCSEL_SHIFT(pmc + 1));
> +}
> +
> +PMU_FORMAT_ATTR(event, "config:0-49");
> +PMU_FORMAT_ATTR(pmcxsel, "config:0-7");
> +PMU_FORMAT_ATTR(mark, "config:8");
> +PMU_FORMAT_ATTR(combine, "config:11");
> +PMU_FORMAT_ATTR(unit, "config:12-15");
> +PMU_FORMAT_ATTR(pmc, "config:16-19");
> +PMU_FORMAT_ATTR(cache_sel, "config:20-23");
> +PMU_FORMAT_ATTR(sample_mode, "config:24-28");
> +PMU_FORMAT_ATTR(thresh_sel, "config:29-31");
> +PMU_FORMAT_ATTR(thresh_stop, "config:32-35");
> +PMU_FORMAT_ATTR(thresh_start, "config:36-39");
> +PMU_FORMAT_ATTR(thresh_cmp, "config:40-49");
> +
> +static struct attribute *power8_pmu_format_attr[] = {
> + &format_attr_event.attr,
> + &format_attr_pmcxsel.attr,
> + &format_attr_mark.attr,
> + &format_attr_combine.attr,
> + &format_attr_unit.attr,
> + &format_attr_pmc.attr,
> + &format_attr_cache_sel.attr,
> + &format_attr_sample_mode.attr,
> + &format_attr_thresh_sel.attr,
> + &format_attr_thresh_stop.attr,
> + &format_attr_thresh_start.attr,
> + &format_attr_thresh_cmp.attr,
> + NULL,
> +};
> +
> +struct attribute_group power8_pmu_format_group = {
> + .name = "format",
> + .attrs = power8_pmu_format_attr,
> +};
> +
> +static const struct attribute_group *power8_pmu_attr_groups[] = {
> + &power8_pmu_format_group,
> + NULL,
> +};
> +
> +static int power8_generic_events[] = {
> + [PERF_COUNT_HW_CPU_CYCLES] = PM_CYC,
> + [PERF_COUNT_HW_STALLED_CYCLES_FRONTEND] = PM_GCT_NOSLOT_CYC,
> + [PERF_COUNT_HW_STALLED_CYCLES_BACKEND] = PM_CMPLU_STALL,
> + [PERF_COUNT_HW_INSTRUCTIONS] = PM_INST_CMPL,
> + [PERF_COUNT_HW_BRANCH_INSTRUCTIONS] = PM_BRU_FIN,
> + [PERF_COUNT_HW_BRANCH_MISSES] = PM_BR_MPRED_CMPL,
> +};
> +
> +static struct power_pmu power8_pmu = {
> + .name = "POWER8",
> + .n_counter = 6,
> + .max_alternatives = 0,
> + .add_fields = POWER8_ADD_FIELDS,
> + .test_adder = POWER8_TEST_ADDER,
> + .compute_mmcr = power8_compute_mmcr,
> + .get_constraint = power8_get_constraint,
> + .disable_pmc = power8_disable_pmc,
> + .flags = PPMU_HAS_SSLOT | PPMU_HAS_SIER,
> + .n_generic = ARRAY_SIZE(power8_generic_events),
> + .generic_events = power8_generic_events,
> + .attr_groups = power8_pmu_attr_groups,
> +};
> +
> +static int __init init_power8_pmu(void)
> +{
> + if (!cur_cpu_spec->oprofile_cpu_type ||
> + strcmp(cur_cpu_spec->oprofile_cpu_type, "ppc64/power8"))
> + return -ENODEV;
> +
> + return register_power_pmu(&power8_pmu);
> +}
> +early_initcall(init_power8_pmu);
More information about the Linuxppc-dev
mailing list