Quelle  power10-pmu.c   Sprache: C

// SPDX-License-Identifier: GPL-2.0-or-later
/*
 * Performance counter support for POWER10 processors.
 *
 * Copyright 2020 Madhavan Srinivasan, IBM Corporation.
 * Copyright 2020 Athira Rajeev, IBM Corporation.
 */

#define pr_fmt(fmt) "power10-pmu: " fmt

#include "isa207-common.h"

/*
 * Raw event encoding for Power10:
 *
 *        60        56        52        48        44        40        36        32
 * | - - - - | - - - - | - - - - | - - - - | - - - - | - - - - | - - - - | - - - - |
 *   | | [ ]   [ src_match ] [  src_mask ]   | [ ] [ l2l3_sel ]  [  thresh_ctl   ]
 *   | |  |                                  |  |                         |
 *   | |  *- IFM (Linux)                     |  |        thresh start/stop -*
 *   | *- BHRB (Linux)                       |  src_sel
 *   *- EBB (Linux)                          *invert_bit
 *
 *        28        24        20        16        12         8         4         0
 * | - - - - | - - - - | - - - - | - - - - | - - - - | - - - - | - - - - | - - - - |
 *   [   ] [  sample ]   [ ] [ ]   [ pmc ]   [unit ]   [ ] |  m   [    pmcxsel    ]
 *     |        |        |    |                        |   |  |
 *     |        |        |    |                        |   |  *- mark
 *     |        |        |    *- L1/L2/L3 cache_sel    |   |*-radix_scope_qual
 *     |        |        sdar_mode                     |
 *     |        *- sampling mode for marked events     *- combine
 *     |
 *     *- thresh_sel
 *
 * Below uses IBM bit numbering.
 *
 * MMCR1[x:y] = unit    (PMCxUNIT)
 * MMCR1[24]   = pmc1combine[0]
 * MMCR1[25]   = pmc1combine[1]
 * MMCR1[26]   = pmc2combine[0]
 * MMCR1[27]   = pmc2combine[1]
 * MMCR1[28]   = pmc3combine[0]
 * MMCR1[29]   = pmc3combine[1]
 * MMCR1[30]   = pmc4combine[0]
 * MMCR1[31]   = pmc4combine[1]
 *
 * if pmc == 3 and unit == 0 and pmcxsel[0:6] == 0b0101011
 * MMCR1[20:27] = thresh_ctl
 * else if pmc == 4 and unit == 0xf and pmcxsel[0:6] == 0b0101001
 * MMCR1[20:27] = thresh_ctl
 * else
 * MMCRA[48:55] = thresh_ctl   (THRESH START/END)
 *
 * if thresh_sel:
 * MMCRA[45:47] = thresh_sel
 *
 * if l2l3_sel:
 * MMCR2[56:60] = l2l3_sel[0:4]
 *
 * MMCR1[16] = cache_sel[0]
 * MMCR1[17] = cache_sel[1]
 * MMCR1[18] = radix_scope_qual
 *
 * 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)
 *
 * if EBB and BHRB:
 * MMCRA[32:33] = IFM
 *
 * MMCRA[SDAR_MODE]  = sdar_mode[0:1]
 */

/*
 * Some power10 event codes.
 */
#define EVENT(_name, _code)     enum{_name = _code}

#include "power10-events-list.h"

#undef EVENT

/* MMCRA IFM bits - POWER10 */
#define POWER10_MMCRA_IFM1  0x0000000040000000UL
#define POWER10_MMCRA_IFM2  0x0000000080000000UL
#define POWER10_MMCRA_IFM3  0x00000000C0000000UL
#define POWER10_MMCRA_BHRB_MASK  0x00000000C0000000UL

extern u64 PERF_REG_EXTENDED_MASK;

/* Table of alternatives, sorted by column 0 */
static const unsigned int power10_event_alternatives[][MAX_ALT] = {
 { PM_INST_CMPL_ALT,  PM_INST_CMPL },
 { PM_CYC_ALT,   PM_CYC },
};

static int power10_get_alternatives(u64 event, unsigned int flags, u64 alt[])
{
 int num_alt = 0;

 num_alt = isa207_get_alternatives(event, alt,
       ARRAY_SIZE(power10_event_alternatives), flags,
       power10_event_alternatives);

 return num_alt;
}

static int power10_check_attr_config(struct perf_event *ev)
{
 u64 val;
 u64 event = ev->attr.config;

 val = (event >> EVENT_SAMPLE_SHIFT) & EVENT_SAMPLE_MASK;
 if (val == 0x10 || isa3XX_check_attr_config(ev))
  return -EINVAL;

 return 0;
}

GENERIC_EVENT_ATTR(cpu-cycles,   PM_CYC);
GENERIC_EVENT_ATTR(instructions,  PM_INST_CMPL);
GENERIC_EVENT_ATTR(branch-instructions,  PM_BR_CMPL);
GENERIC_EVENT_ATTR(branch-misses,  PM_BR_MPRED_CMPL);
GENERIC_EVENT_ATTR(cache-references,  PM_LD_REF_L1);
GENERIC_EVENT_ATTR(cache-misses,  PM_LD_MISS_L1);
GENERIC_EVENT_ATTR(mem-loads,   MEM_LOADS);
GENERIC_EVENT_ATTR(mem-stores,   MEM_STORES);
GENERIC_EVENT_ATTR(branch-instructions,  PM_BR_FIN);
GENERIC_EVENT_ATTR(branch-misses,  PM_MPRED_BR_FIN);
GENERIC_EVENT_ATTR(cache-misses,  PM_LD_DEMAND_MISS_L1_FIN);

CACHE_EVENT_ATTR(L1-dcache-load-misses,  PM_LD_MISS_L1);
CACHE_EVENT_ATTR(L1-dcache-loads,  PM_LD_REF_L1);
CACHE_EVENT_ATTR(L1-dcache-prefetches,  PM_LD_PREFETCH_CACHE_LINE_MISS);
CACHE_EVENT_ATTR(L1-dcache-store-misses, PM_ST_MISS_L1);
CACHE_EVENT_ATTR(L1-icache-load-misses,  PM_L1_ICACHE_MISS);
CACHE_EVENT_ATTR(L1-icache-loads,  PM_INST_FROM_L1);
CACHE_EVENT_ATTR(L1-icache-prefetches,  PM_IC_PREF_REQ);
CACHE_EVENT_ATTR(LLC-load-misses,  PM_DATA_FROM_L3MISS);
CACHE_EVENT_ATTR(LLC-loads,   PM_DATA_FROM_L3);
CACHE_EVENT_ATTR(LLC-prefetches,  PM_L3_PF_MISS_L3);
CACHE_EVENT_ATTR(LLC-store-misses,  PM_L2_ST_MISS);
CACHE_EVENT_ATTR(LLC-stores,   PM_L2_ST);
CACHE_EVENT_ATTR(branch-load-misses,  PM_BR_MPRED_CMPL);
CACHE_EVENT_ATTR(branch-loads,   PM_BR_CMPL);
CACHE_EVENT_ATTR(dTLB-load-misses,  PM_DTLB_MISS);
CACHE_EVENT_ATTR(iTLB-load-misses,  PM_ITLB_MISS);

static struct attribute *power10_events_attr_dd1[] = {
 GENERIC_EVENT_PTR(PM_CYC),
 GENERIC_EVENT_PTR(PM_INST_CMPL),
 GENERIC_EVENT_PTR(PM_BR_CMPL),
 GENERIC_EVENT_PTR(PM_BR_MPRED_CMPL),
 GENERIC_EVENT_PTR(PM_LD_REF_L1),
 GENERIC_EVENT_PTR(PM_LD_MISS_L1),
 GENERIC_EVENT_PTR(MEM_LOADS),
 GENERIC_EVENT_PTR(MEM_STORES),
 CACHE_EVENT_PTR(PM_LD_MISS_L1),
 CACHE_EVENT_PTR(PM_LD_REF_L1),
 CACHE_EVENT_PTR(PM_LD_PREFETCH_CACHE_LINE_MISS),
 CACHE_EVENT_PTR(PM_ST_MISS_L1),
 CACHE_EVENT_PTR(PM_L1_ICACHE_MISS),
 CACHE_EVENT_PTR(PM_INST_FROM_L1),
 CACHE_EVENT_PTR(PM_IC_PREF_REQ),
 CACHE_EVENT_PTR(PM_DATA_FROM_L3MISS),
 CACHE_EVENT_PTR(PM_DATA_FROM_L3),
 CACHE_EVENT_PTR(PM_BR_MPRED_CMPL),
 CACHE_EVENT_PTR(PM_BR_CMPL),
 CACHE_EVENT_PTR(PM_DTLB_MISS),
 CACHE_EVENT_PTR(PM_ITLB_MISS),
 NULL
};

static struct attribute *power10_events_attr[] = {
 GENERIC_EVENT_PTR(PM_CYC),
 GENERIC_EVENT_PTR(PM_INST_CMPL),
 GENERIC_EVENT_PTR(PM_BR_FIN),
 GENERIC_EVENT_PTR(PM_MPRED_BR_FIN),
 GENERIC_EVENT_PTR(PM_LD_REF_L1),
 GENERIC_EVENT_PTR(PM_LD_DEMAND_MISS_L1_FIN),
 GENERIC_EVENT_PTR(MEM_LOADS),
 GENERIC_EVENT_PTR(MEM_STORES),
 CACHE_EVENT_PTR(PM_LD_MISS_L1),
 CACHE_EVENT_PTR(PM_LD_REF_L1),
 CACHE_EVENT_PTR(PM_LD_PREFETCH_CACHE_LINE_MISS),
 CACHE_EVENT_PTR(PM_ST_MISS_L1),
 CACHE_EVENT_PTR(PM_L1_ICACHE_MISS),
 CACHE_EVENT_PTR(PM_INST_FROM_L1),
 CACHE_EVENT_PTR(PM_IC_PREF_REQ),
 CACHE_EVENT_PTR(PM_DATA_FROM_L3MISS),
 CACHE_EVENT_PTR(PM_DATA_FROM_L3),
 CACHE_EVENT_PTR(PM_L3_PF_MISS_L3),
 CACHE_EVENT_PTR(PM_L2_ST_MISS),
 CACHE_EVENT_PTR(PM_L2_ST),
 CACHE_EVENT_PTR(PM_BR_MPRED_CMPL),
 CACHE_EVENT_PTR(PM_BR_CMPL),
 CACHE_EVENT_PTR(PM_DTLB_MISS),
 CACHE_EVENT_PTR(PM_ITLB_MISS),
 NULL
};

static const struct attribute_group power10_pmu_events_group_dd1 = {
 .name = "events",
 .attrs = power10_events_attr_dd1,
};

static const struct attribute_group power10_pmu_events_group = {
 .name = "events",
 .attrs = power10_events_attr,
};

PMU_FORMAT_ATTR(event,          "config:0-59");
PMU_FORMAT_ATTR(pmcxsel,        "config:0-7");
PMU_FORMAT_ATTR(mark,           "config:8");
PMU_FORMAT_ATTR(combine,        "config:10-11");
PMU_FORMAT_ATTR(unit,           "config:12-15");
PMU_FORMAT_ATTR(pmc,            "config:16-19");
PMU_FORMAT_ATTR(cache_sel,      "config:20-21");
PMU_FORMAT_ATTR(sdar_mode,      "config:22-23");
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(l2l3_sel,       "config:40-44");
PMU_FORMAT_ATTR(src_sel,        "config:45-46");
PMU_FORMAT_ATTR(invert_bit,     "config:47");
PMU_FORMAT_ATTR(src_mask,       "config:48-53");
PMU_FORMAT_ATTR(src_match,      "config:54-59");
PMU_FORMAT_ATTR(radix_scope, "config:9");
PMU_FORMAT_ATTR(thresh_cmp,     "config1:0-17");

static struct attribute *power10_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_sdar_mode.attr,
 &format_attr_sample_mode.attr,
 &format_attr_thresh_sel.attr,
 &format_attr_thresh_stop.attr,
 &format_attr_thresh_start.attr,
 &format_attr_l2l3_sel.attr,
 &format_attr_src_sel.attr,
 &format_attr_invert_bit.attr,
 &format_attr_src_mask.attr,
 &format_attr_src_match.attr,
 &format_attr_radix_scope.attr,
 &format_attr_thresh_cmp.attr,
 NULL,
};

static const struct attribute_group power10_pmu_format_group = {
 .name = "format",
 .attrs = power10_pmu_format_attr,
};

static struct attribute *power10_pmu_caps_attrs[] = {
 NULL
};

static struct attribute_group power10_pmu_caps_group = {
 .name  = "caps",
 .attrs = power10_pmu_caps_attrs,
};

static const struct attribute_group *power10_pmu_attr_groups_dd1[] = {
 &power10_pmu_format_group,
 &power10_pmu_events_group_dd1,
 &power10_pmu_caps_group,
 NULL,
};

static const struct attribute_group *power10_pmu_attr_groups[] = {
 &power10_pmu_format_group,
 &power10_pmu_events_group,
 &power10_pmu_caps_group,
 NULL,
};

static int power10_generic_events_dd1[] = {
 [PERF_COUNT_HW_CPU_CYCLES] =   PM_CYC,
 [PERF_COUNT_HW_INSTRUCTIONS] =   PM_INST_CMPL,
 [PERF_COUNT_HW_BRANCH_INSTRUCTIONS] =  PM_BR_CMPL,
 [PERF_COUNT_HW_BRANCH_MISSES] =   PM_BR_MPRED_CMPL,
 [PERF_COUNT_HW_CACHE_REFERENCES] =  PM_LD_REF_L1,
 [PERF_COUNT_HW_CACHE_MISSES] =   PM_LD_MISS_L1,
};

static int power10_generic_events[] = {
 [PERF_COUNT_HW_CPU_CYCLES] =   PM_CYC,
 [PERF_COUNT_HW_INSTRUCTIONS] =   PM_INST_CMPL,
 [PERF_COUNT_HW_BRANCH_INSTRUCTIONS] =  PM_BR_FIN,
 [PERF_COUNT_HW_BRANCH_MISSES] =   PM_MPRED_BR_FIN,
 [PERF_COUNT_HW_CACHE_REFERENCES] =  PM_LD_REF_L1,
 [PERF_COUNT_HW_CACHE_MISSES] =   PM_LD_DEMAND_MISS_L1_FIN,
};

static u64 power10_bhrb_filter_map(u64 branch_sample_type)
{
 u64 pmu_bhrb_filter = 0;

 /* BHRB and regular PMU events share the same privilege state
 * filter configuration. BHRB is always recorded along with a
 * regular PMU event. As the privilege state filter is handled
 * in the basic PMC configuration of the accompanying regular
 * PMU event, we ignore any separate BHRB specific request.
 */

 /* No branch filter requested */
 if (branch_sample_type & PERF_SAMPLE_BRANCH_ANY)
  return pmu_bhrb_filter;

 /* Invalid branch filter options - HW does not support */
 if (branch_sample_type & PERF_SAMPLE_BRANCH_ANY_RETURN)
  return -1;

 if (branch_sample_type & PERF_SAMPLE_BRANCH_IND_CALL) {
  pmu_bhrb_filter |= POWER10_MMCRA_IFM2;
  return pmu_bhrb_filter;
 }

 if (branch_sample_type & PERF_SAMPLE_BRANCH_COND) {
  pmu_bhrb_filter |= POWER10_MMCRA_IFM3;
  return pmu_bhrb_filter;
 }

 if (branch_sample_type & PERF_SAMPLE_BRANCH_CALL)
  return -1;

 if (branch_sample_type & PERF_SAMPLE_BRANCH_ANY_CALL) {
  pmu_bhrb_filter |= POWER10_MMCRA_IFM1;
  return pmu_bhrb_filter;
 }

 /* Every thing else is unsupported */
 return -1;
}

static void power10_config_bhrb(u64 pmu_bhrb_filter)
{
 pmu_bhrb_filter &= POWER10_MMCRA_BHRB_MASK;

 /* Enable BHRB filter in PMU */
 mtspr(SPRN_MMCRA, (mfspr(SPRN_MMCRA) | pmu_bhrb_filter));
}

#define C(x) PERF_COUNT_HW_CACHE_##x

/*
 * Table of generalized cache-related events.
 * 0 means not supported, -1 means nonsensical, other values
 * are event codes.
 */
static u64 power10_cache_events_dd1[C(MAX)][C(OP_MAX)][C(RESULT_MAX)] = {
 [C(L1D)] = {
  [C(OP_READ)] = {
   [C(RESULT_ACCESS)] = PM_LD_REF_L1,
   [C(RESULT_MISS)] = PM_LD_MISS_L1,
  },
  [C(OP_WRITE)] = {
   [C(RESULT_ACCESS)] = 0,
   [C(RESULT_MISS)] = PM_ST_MISS_L1,
  },
  [C(OP_PREFETCH)] = {
   [C(RESULT_ACCESS)] = PM_LD_PREFETCH_CACHE_LINE_MISS,
   [C(RESULT_MISS)] = 0,
  },
 },
 [C(L1I)] = {
  [C(OP_READ)] = {
   [C(RESULT_ACCESS)] = PM_INST_FROM_L1,
   [C(RESULT_MISS)] = PM_L1_ICACHE_MISS,
  },
  [C(OP_WRITE)] = {
   [C(RESULT_ACCESS)] = PM_INST_FROM_L1MISS,
   [C(RESULT_MISS)] = -1,
  },
  [C(OP_PREFETCH)] = {
   [C(RESULT_ACCESS)] = PM_IC_PREF_REQ,
   [C(RESULT_MISS)] = 0,
  },
 },
 [C(LL)] = {
  [C(OP_READ)] = {
   [C(RESULT_ACCESS)] = PM_DATA_FROM_L3,
   [C(RESULT_MISS)] = PM_DATA_FROM_L3MISS,
  },
  [C(OP_WRITE)] = {
   [C(RESULT_ACCESS)] = -1,
   [C(RESULT_MISS)] = -1,
  },
  [C(OP_PREFETCH)] = {
   [C(RESULT_ACCESS)] = -1,
   [C(RESULT_MISS)] = 0,
  },
 },
  [C(DTLB)] = {
  [C(OP_READ)] = {
   [C(RESULT_ACCESS)] = 0,
   [C(RESULT_MISS)] = PM_DTLB_MISS,
  },
  [C(OP_WRITE)] = {
   [C(RESULT_ACCESS)] = -1,
   [C(RESULT_MISS)] = -1,
  },
  [C(OP_PREFETCH)] = {
   [C(RESULT_ACCESS)] = -1,
   [C(RESULT_MISS)] = -1,
  },
 },
 [C(ITLB)] = {
  [C(OP_READ)] = {
   [C(RESULT_ACCESS)] = 0,
   [C(RESULT_MISS)] = PM_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)] = PM_BR_CMPL,
   [C(RESULT_MISS)] = PM_BR_MPRED_CMPL,
  },
  [C(OP_WRITE)] = {
   [C(RESULT_ACCESS)] = -1,
   [C(RESULT_MISS)] = -1,
  },
  [C(OP_PREFETCH)] = {
   [C(RESULT_ACCESS)] = -1,
   [C(RESULT_MISS)] = -1,
  },
 },
 [C(NODE)] = {
  [C(OP_READ)] = {
   [C(RESULT_ACCESS)] = -1,
   [C(RESULT_MISS)] = -1,
  },
  [C(OP_WRITE)] = {
   [C(RESULT_ACCESS)] = -1,
   [C(RESULT_MISS)] = -1,
  },
  [C(OP_PREFETCH)] = {
   [C(RESULT_ACCESS)] = -1,
   [C(RESULT_MISS)] = -1,
  },
 },
};

static u64 power10_cache_events[C(MAX)][C(OP_MAX)][C(RESULT_MAX)] = {
 [C(L1D)] = {
  [C(OP_READ)] = {
   [C(RESULT_ACCESS)] = PM_LD_REF_L1,
   [C(RESULT_MISS)] = PM_LD_MISS_L1,
  },
  [C(OP_WRITE)] = {
   [C(RESULT_ACCESS)] = 0,
   [C(RESULT_MISS)] = PM_ST_MISS_L1,
  },
  [C(OP_PREFETCH)] = {
   [C(RESULT_ACCESS)] = PM_LD_PREFETCH_CACHE_LINE_MISS,
   [C(RESULT_MISS)] = 0,
  },
 },
 [C(L1I)] = {
  [C(OP_READ)] = {
   [C(RESULT_ACCESS)] = PM_INST_FROM_L1,
   [C(RESULT_MISS)] = PM_L1_ICACHE_MISS,
  },
  [C(OP_WRITE)] = {
   [C(RESULT_ACCESS)] = PM_INST_FROM_L1MISS,
   [C(RESULT_MISS)] = -1,
  },
  [C(OP_PREFETCH)] = {
   [C(RESULT_ACCESS)] = PM_IC_PREF_REQ,
   [C(RESULT_MISS)] = 0,
  },
 },
 [C(LL)] = {
  [C(OP_READ)] = {
   [C(RESULT_ACCESS)] = PM_DATA_FROM_L3,
   [C(RESULT_MISS)] = PM_DATA_FROM_L3MISS,
  },
  [C(OP_WRITE)] = {
   [C(RESULT_ACCESS)] = PM_L2_ST,
   [C(RESULT_MISS)] = PM_L2_ST_MISS,
  },
  [C(OP_PREFETCH)] = {
   [C(RESULT_ACCESS)] = PM_L3_PF_MISS_L3,
   [C(RESULT_MISS)] = 0,
  },
 },
  [C(DTLB)] = {
  [C(OP_READ)] = {
   [C(RESULT_ACCESS)] = 0,
   [C(RESULT_MISS)] = PM_DTLB_MISS,
  },
  [C(OP_WRITE)] = {
   [C(RESULT_ACCESS)] = -1,
   [C(RESULT_MISS)] = -1,
  },
  [C(OP_PREFETCH)] = {
   [C(RESULT_ACCESS)] = -1,
   [C(RESULT_MISS)] = -1,
  },
 },
 [C(ITLB)] = {
  [C(OP_READ)] = {
   [C(RESULT_ACCESS)] = 0,
   [C(RESULT_MISS)] = PM_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)] = PM_BR_CMPL,
   [C(RESULT_MISS)] = PM_BR_MPRED_CMPL,
  },
  [C(OP_WRITE)] = {
   [C(RESULT_ACCESS)] = -1,
   [C(RESULT_MISS)] = -1,
  },
  [C(OP_PREFETCH)] = {
   [C(RESULT_ACCESS)] = -1,
   [C(RESULT_MISS)] = -1,
  },
 },
 [C(NODE)] = {
  [C(OP_READ)] = {
   [C(RESULT_ACCESS)] = -1,
   [C(RESULT_MISS)] = -1,
  },
  [C(OP_WRITE)] = {
   [C(RESULT_ACCESS)] = -1,
   [C(RESULT_MISS)] = -1,
  },
  [C(OP_PREFETCH)] = {
   [C(RESULT_ACCESS)] = -1,
   [C(RESULT_MISS)] = -1,
  },
 },
};

#undef C

/*
 * Set the MMCR0[CC56RUN] bit to enable counting for
 * PMC5 and PMC6 regardless of the state of CTRL[RUN],
 * so that we can use counters 5 and 6 as PM_INST_CMPL and
 * PM_CYC.
 */
static int power10_compute_mmcr(u64 event[], int n_ev,
    unsigned int hwc[], struct mmcr_regs *mmcr,
    struct perf_event *pevents[], u32 flags)
{
 int ret;

 ret = isa207_compute_mmcr(event, n_ev, hwc, mmcr, pevents, flags);
 if (!ret)
  mmcr->mmcr0 |= MMCR0_C56RUN;
 return ret;
}

static struct power_pmu power10_pmu = {
 .name   = "POWER10",
 .n_counter  = MAX_PMU_COUNTERS,
 .add_fields  = ISA207_ADD_FIELDS,
 .test_adder  = ISA207_TEST_ADDER,
 .group_constraint_mask = CNST_CACHE_PMC4_MASK,
 .group_constraint_val = CNST_CACHE_PMC4_VAL,
 .compute_mmcr  = power10_compute_mmcr,
 .config_bhrb  = power10_config_bhrb,
 .bhrb_filter_map = power10_bhrb_filter_map,
 .get_constraint  = isa207_get_constraint,
 .get_alternatives = power10_get_alternatives,
 .get_mem_data_src = isa207_get_mem_data_src,
 .get_mem_weight  = isa207_get_mem_weight,
 .disable_pmc  = isa207_disable_pmc,
 .flags   = PPMU_HAS_SIER | PPMU_ARCH_207S |
      PPMU_ARCH_31 | PPMU_HAS_ATTR_CONFIG1 |
      PPMU_P10,
 .n_generic  = ARRAY_SIZE(power10_generic_events),
 .generic_events  = power10_generic_events,
 .cache_events  = &power10_cache_events,
 .attr_groups  = power10_pmu_attr_groups,
 .bhrb_nr  = 32,
 .capabilities           = PERF_PMU_CAP_EXTENDED_REGS,
 .check_attr_config = power10_check_attr_config,
};

int __init init_power10_pmu(void)
{
 unsigned int pvr;
 int rc;

 pvr = mfspr(SPRN_PVR);
 if (PVR_VER(pvr) != PVR_POWER10)
  return -ENODEV;

 /* Add the ppmu flag for power10 DD1 */
 if ((PVR_CFG(pvr) == 1))
  power10_pmu.flags |= PPMU_P10_DD1;

 /* Set the PERF_REG_EXTENDED_MASK here */
 PERF_REG_EXTENDED_MASK = PERF_REG_PMU_MASK_31;

 if ((PVR_CFG(pvr) == 1)) {
  power10_pmu.generic_events = power10_generic_events_dd1;
  power10_pmu.attr_groups = power10_pmu_attr_groups_dd1;
  power10_pmu.cache_events = &power10_cache_events_dd1;
 }

 rc = register_power_pmu(&power10_pmu);
 if (rc)
  return rc;

 /* Tell userspace that EBB is supported */
 cur_cpu_spec->cpu_user_features2 |= PPC_FEATURE2_EBB;

 return 0;
}

static struct power_pmu power11_pmu;

int __init init_power11_pmu(void)
{
 unsigned int pvr;
 int rc;

 pvr = mfspr(SPRN_PVR);
 if (PVR_VER(pvr) != PVR_POWER11)
  return -ENODEV;

 /* Set the PERF_REG_EXTENDED_MASK here */
 PERF_REG_EXTENDED_MASK = PERF_REG_PMU_MASK_31;

 power11_pmu = power10_pmu;
 power11_pmu.name = "Power11";

 rc = register_power_pmu(&power11_pmu);
 if (rc)
  return rc;

 /* Tell userspace that EBB is supported */
 cur_cpu_spec->cpu_user_features2 |= PPC_FEATURE2_EBB;

 return 0;
}