linux-stable/arch/arm64/kernel/perf_event.c

1348 lines
40 KiB
C
Raw Normal View History

// SPDX-License-Identifier: GPL-2.0-only
/*
* ARMv8 PMUv3 Performance Events handling code.
*
* Copyright (C) 2012 ARM Limited
* Author: Will Deacon <will.deacon@arm.com>
*
* This code is based heavily on the ARMv7 perf event code.
*/
#include <asm/irq_regs.h>
#include <asm/perf_event.h>
#include <asm/sysreg.h>
#include <asm/virt.h>
#include <clocksource/arm_arch_timer.h>
#include <linux/acpi.h>
arm64: perf: Add cap_user_time aarch64 It is useful to get the running time of a thread. Doing so in an efficient manner can be important for performance of user applications. Avoiding system calls in `clock_gettime` when handling CLOCK_THREAD_CPUTIME_ID is important. Other clocks are handled in the VDSO, but CLOCK_THREAD_CPUTIME_ID falls back on the system call. CLOCK_THREAD_CPUTIME_ID is not handled in the VDSO since it would have costs associated with maintaining updated user space accessible time offsets. These offsets have to be updated everytime the a thread is scheduled/descheduled. However, for programs regularly checking the running time of a thread, this is a performance improvement. This patch takes a middle ground, and adds support for cap_user_time an optional feature of the perf_event API. This way costs are only incurred when the perf_event api is enabled. This is done the same way as it is in x86. Ultimately this allows calculating the thread running time in userspace on aarch64 as follows (adapted from perf_event_open manpage): u32 seq, time_mult, time_shift; u64 running, count, time_offset, quot, rem, delta; struct perf_event_mmap_page *pc; pc = buf; // buf is the perf event mmaped page as documented in the API. if (pc->cap_usr_time) { do { seq = pc->lock; barrier(); running = pc->time_running; count = readCNTVCT_EL0(); // Read ARM hardware clock. time_offset = pc->time_offset; time_mult = pc->time_mult; time_shift = pc->time_shift; barrier(); } while (pc->lock != seq); quot = (count >> time_shift); rem = count & (((u64)1 << time_shift) - 1); delta = time_offset + quot * time_mult + ((rem * time_mult) >> time_shift); running += delta; // running now has the current nanosecond level thread time. } Summary of changes in the patch: For aarch64 systems, make arch_perf_update_userpage update the timing information stored in the perf_event page. Requiring the following calculations: - Calculate the appropriate time_mult, and time_shift factors to convert ticks to nano seconds for the current clock frequency. - Adjust the mult and shift factors to avoid shift factors of 32 bits. (possibly unnecessary) - The time_offset userspace should apply when doing calculations: negative the current sched time (now), because time_running and time_enabled fields of the perf_event page have just been updated. Toggle bits to appropriate values: - Enable cap_user_time Signed-off-by: Michael O'Farrell <micpof@gmail.com> Signed-off-by: Will Deacon <will.deacon@arm.com>
2018-07-30 20:14:34 +00:00
#include <linux/clocksource.h>
#include <linux/kvm_host.h>
#include <linux/of.h>
#include <linux/perf/arm_pmu.h>
#include <linux/platform_device.h>
#include <linux/sched_clock.h>
#include <linux/smp.h>
/* ARMv8 Cortex-A53 specific event types. */
#define ARMV8_A53_PERFCTR_PREF_LINEFILL 0xC2
/* ARMv8 Cavium ThunderX specific event types. */
#define ARMV8_THUNDER_PERFCTR_L1D_CACHE_MISS_ST 0xE9
#define ARMV8_THUNDER_PERFCTR_L1D_CACHE_PREF_ACCESS 0xEA
#define ARMV8_THUNDER_PERFCTR_L1D_CACHE_PREF_MISS 0xEB
#define ARMV8_THUNDER_PERFCTR_L1I_CACHE_PREF_ACCESS 0xEC
#define ARMV8_THUNDER_PERFCTR_L1I_CACHE_PREF_MISS 0xED
/*
* ARMv8 Architectural defined events, not all of these may
* be supported on any given implementation. Unsupported events will
* be disabled at run-time based on the PMCEID registers.
*/
static const unsigned armv8_pmuv3_perf_map[PERF_COUNT_HW_MAX] = {
PERF_MAP_ALL_UNSUPPORTED,
[PERF_COUNT_HW_CPU_CYCLES] = ARMV8_PMUV3_PERFCTR_CPU_CYCLES,
[PERF_COUNT_HW_INSTRUCTIONS] = ARMV8_PMUV3_PERFCTR_INST_RETIRED,
[PERF_COUNT_HW_CACHE_REFERENCES] = ARMV8_PMUV3_PERFCTR_L1D_CACHE,
[PERF_COUNT_HW_CACHE_MISSES] = ARMV8_PMUV3_PERFCTR_L1D_CACHE_REFILL,
[PERF_COUNT_HW_BRANCH_INSTRUCTIONS] = ARMV8_PMUV3_PERFCTR_PC_WRITE_RETIRED,
[PERF_COUNT_HW_BRANCH_MISSES] = ARMV8_PMUV3_PERFCTR_BR_MIS_PRED,
[PERF_COUNT_HW_BUS_CYCLES] = ARMV8_PMUV3_PERFCTR_BUS_CYCLES,
[PERF_COUNT_HW_STALLED_CYCLES_FRONTEND] = ARMV8_PMUV3_PERFCTR_STALL_FRONTEND,
[PERF_COUNT_HW_STALLED_CYCLES_BACKEND] = ARMV8_PMUV3_PERFCTR_STALL_BACKEND,
};
static const unsigned armv8_pmuv3_perf_cache_map[PERF_COUNT_HW_CACHE_MAX]
[PERF_COUNT_HW_CACHE_OP_MAX]
[PERF_COUNT_HW_CACHE_RESULT_MAX] = {
PERF_CACHE_MAP_ALL_UNSUPPORTED,
[C(L1D)][C(OP_READ)][C(RESULT_ACCESS)] = ARMV8_PMUV3_PERFCTR_L1D_CACHE,
[C(L1D)][C(OP_READ)][C(RESULT_MISS)] = ARMV8_PMUV3_PERFCTR_L1D_CACHE_REFILL,
[C(L1I)][C(OP_READ)][C(RESULT_ACCESS)] = ARMV8_PMUV3_PERFCTR_L1I_CACHE,
[C(L1I)][C(OP_READ)][C(RESULT_MISS)] = ARMV8_PMUV3_PERFCTR_L1I_CACHE_REFILL,
[C(DTLB)][C(OP_READ)][C(RESULT_MISS)] = ARMV8_PMUV3_PERFCTR_L1D_TLB_REFILL,
[C(DTLB)][C(OP_READ)][C(RESULT_ACCESS)] = ARMV8_PMUV3_PERFCTR_L1D_TLB,
[C(ITLB)][C(OP_READ)][C(RESULT_MISS)] = ARMV8_PMUV3_PERFCTR_L1I_TLB_REFILL,
[C(ITLB)][C(OP_READ)][C(RESULT_ACCESS)] = ARMV8_PMUV3_PERFCTR_L1I_TLB,
[C(LL)][C(OP_READ)][C(RESULT_MISS)] = ARMV8_PMUV3_PERFCTR_LL_CACHE_MISS_RD,
[C(LL)][C(OP_READ)][C(RESULT_ACCESS)] = ARMV8_PMUV3_PERFCTR_LL_CACHE_RD,
[C(BPU)][C(OP_READ)][C(RESULT_ACCESS)] = ARMV8_PMUV3_PERFCTR_BR_PRED,
[C(BPU)][C(OP_READ)][C(RESULT_MISS)] = ARMV8_PMUV3_PERFCTR_BR_MIS_PRED,
};
static const unsigned armv8_a53_perf_cache_map[PERF_COUNT_HW_CACHE_MAX]
[PERF_COUNT_HW_CACHE_OP_MAX]
[PERF_COUNT_HW_CACHE_RESULT_MAX] = {
PERF_CACHE_MAP_ALL_UNSUPPORTED,
[C(L1D)][C(OP_PREFETCH)][C(RESULT_MISS)] = ARMV8_A53_PERFCTR_PREF_LINEFILL,
[C(NODE)][C(OP_READ)][C(RESULT_ACCESS)] = ARMV8_IMPDEF_PERFCTR_BUS_ACCESS_RD,
[C(NODE)][C(OP_WRITE)][C(RESULT_ACCESS)] = ARMV8_IMPDEF_PERFCTR_BUS_ACCESS_WR,
};
static const unsigned armv8_a57_perf_cache_map[PERF_COUNT_HW_CACHE_MAX]
[PERF_COUNT_HW_CACHE_OP_MAX]
[PERF_COUNT_HW_CACHE_RESULT_MAX] = {
PERF_CACHE_MAP_ALL_UNSUPPORTED,
[C(L1D)][C(OP_READ)][C(RESULT_ACCESS)] = ARMV8_IMPDEF_PERFCTR_L1D_CACHE_RD,
[C(L1D)][C(OP_READ)][C(RESULT_MISS)] = ARMV8_IMPDEF_PERFCTR_L1D_CACHE_REFILL_RD,
[C(L1D)][C(OP_WRITE)][C(RESULT_ACCESS)] = ARMV8_IMPDEF_PERFCTR_L1D_CACHE_WR,
[C(L1D)][C(OP_WRITE)][C(RESULT_MISS)] = ARMV8_IMPDEF_PERFCTR_L1D_CACHE_REFILL_WR,
[C(DTLB)][C(OP_READ)][C(RESULT_MISS)] = ARMV8_IMPDEF_PERFCTR_L1D_TLB_REFILL_RD,
[C(DTLB)][C(OP_WRITE)][C(RESULT_MISS)] = ARMV8_IMPDEF_PERFCTR_L1D_TLB_REFILL_WR,
[C(NODE)][C(OP_READ)][C(RESULT_ACCESS)] = ARMV8_IMPDEF_PERFCTR_BUS_ACCESS_RD,
[C(NODE)][C(OP_WRITE)][C(RESULT_ACCESS)] = ARMV8_IMPDEF_PERFCTR_BUS_ACCESS_WR,
};
static const unsigned armv8_a73_perf_cache_map[PERF_COUNT_HW_CACHE_MAX]
[PERF_COUNT_HW_CACHE_OP_MAX]
[PERF_COUNT_HW_CACHE_RESULT_MAX] = {
PERF_CACHE_MAP_ALL_UNSUPPORTED,
[C(L1D)][C(OP_READ)][C(RESULT_ACCESS)] = ARMV8_IMPDEF_PERFCTR_L1D_CACHE_RD,
[C(L1D)][C(OP_WRITE)][C(RESULT_ACCESS)] = ARMV8_IMPDEF_PERFCTR_L1D_CACHE_WR,
};
static const unsigned armv8_thunder_perf_cache_map[PERF_COUNT_HW_CACHE_MAX]
[PERF_COUNT_HW_CACHE_OP_MAX]
[PERF_COUNT_HW_CACHE_RESULT_MAX] = {
PERF_CACHE_MAP_ALL_UNSUPPORTED,
[C(L1D)][C(OP_READ)][C(RESULT_ACCESS)] = ARMV8_IMPDEF_PERFCTR_L1D_CACHE_RD,
[C(L1D)][C(OP_READ)][C(RESULT_MISS)] = ARMV8_IMPDEF_PERFCTR_L1D_CACHE_REFILL_RD,
[C(L1D)][C(OP_WRITE)][C(RESULT_ACCESS)] = ARMV8_IMPDEF_PERFCTR_L1D_CACHE_WR,
[C(L1D)][C(OP_WRITE)][C(RESULT_MISS)] = ARMV8_THUNDER_PERFCTR_L1D_CACHE_MISS_ST,
[C(L1D)][C(OP_PREFETCH)][C(RESULT_ACCESS)] = ARMV8_THUNDER_PERFCTR_L1D_CACHE_PREF_ACCESS,
[C(L1D)][C(OP_PREFETCH)][C(RESULT_MISS)] = ARMV8_THUNDER_PERFCTR_L1D_CACHE_PREF_MISS,
[C(L1I)][C(OP_PREFETCH)][C(RESULT_ACCESS)] = ARMV8_THUNDER_PERFCTR_L1I_CACHE_PREF_ACCESS,
[C(L1I)][C(OP_PREFETCH)][C(RESULT_MISS)] = ARMV8_THUNDER_PERFCTR_L1I_CACHE_PREF_MISS,
[C(DTLB)][C(OP_READ)][C(RESULT_ACCESS)] = ARMV8_IMPDEF_PERFCTR_L1D_TLB_RD,
[C(DTLB)][C(OP_READ)][C(RESULT_MISS)] = ARMV8_IMPDEF_PERFCTR_L1D_TLB_REFILL_RD,
[C(DTLB)][C(OP_WRITE)][C(RESULT_ACCESS)] = ARMV8_IMPDEF_PERFCTR_L1D_TLB_WR,
[C(DTLB)][C(OP_WRITE)][C(RESULT_MISS)] = ARMV8_IMPDEF_PERFCTR_L1D_TLB_REFILL_WR,
};
static const unsigned armv8_vulcan_perf_cache_map[PERF_COUNT_HW_CACHE_MAX]
[PERF_COUNT_HW_CACHE_OP_MAX]
[PERF_COUNT_HW_CACHE_RESULT_MAX] = {
PERF_CACHE_MAP_ALL_UNSUPPORTED,
[C(L1D)][C(OP_READ)][C(RESULT_ACCESS)] = ARMV8_IMPDEF_PERFCTR_L1D_CACHE_RD,
[C(L1D)][C(OP_READ)][C(RESULT_MISS)] = ARMV8_IMPDEF_PERFCTR_L1D_CACHE_REFILL_RD,
[C(L1D)][C(OP_WRITE)][C(RESULT_ACCESS)] = ARMV8_IMPDEF_PERFCTR_L1D_CACHE_WR,
[C(L1D)][C(OP_WRITE)][C(RESULT_MISS)] = ARMV8_IMPDEF_PERFCTR_L1D_CACHE_REFILL_WR,
[C(DTLB)][C(OP_READ)][C(RESULT_ACCESS)] = ARMV8_IMPDEF_PERFCTR_L1D_TLB_RD,
[C(DTLB)][C(OP_WRITE)][C(RESULT_ACCESS)] = ARMV8_IMPDEF_PERFCTR_L1D_TLB_WR,
[C(DTLB)][C(OP_READ)][C(RESULT_MISS)] = ARMV8_IMPDEF_PERFCTR_L1D_TLB_REFILL_RD,
[C(DTLB)][C(OP_WRITE)][C(RESULT_MISS)] = ARMV8_IMPDEF_PERFCTR_L1D_TLB_REFILL_WR,
[C(NODE)][C(OP_READ)][C(RESULT_ACCESS)] = ARMV8_IMPDEF_PERFCTR_BUS_ACCESS_RD,
[C(NODE)][C(OP_WRITE)][C(RESULT_ACCESS)] = ARMV8_IMPDEF_PERFCTR_BUS_ACCESS_WR,
};
static ssize_t
armv8pmu_events_sysfs_show(struct device *dev,
struct device_attribute *attr, char *page)
{
struct perf_pmu_events_attr *pmu_attr;
pmu_attr = container_of(attr, struct perf_pmu_events_attr, attr);
return sprintf(page, "event=0x%04llx\n", pmu_attr->id);
}
#define ARMV8_EVENT_ATTR(name, config) \
PMU_EVENT_ATTR_ID(name, armv8pmu_events_sysfs_show, config)
static struct attribute *armv8_pmuv3_event_attrs[] = {
ARMV8_EVENT_ATTR(sw_incr, ARMV8_PMUV3_PERFCTR_SW_INCR),
ARMV8_EVENT_ATTR(l1i_cache_refill, ARMV8_PMUV3_PERFCTR_L1I_CACHE_REFILL),
ARMV8_EVENT_ATTR(l1i_tlb_refill, ARMV8_PMUV3_PERFCTR_L1I_TLB_REFILL),
ARMV8_EVENT_ATTR(l1d_cache_refill, ARMV8_PMUV3_PERFCTR_L1D_CACHE_REFILL),
ARMV8_EVENT_ATTR(l1d_cache, ARMV8_PMUV3_PERFCTR_L1D_CACHE),
ARMV8_EVENT_ATTR(l1d_tlb_refill, ARMV8_PMUV3_PERFCTR_L1D_TLB_REFILL),
ARMV8_EVENT_ATTR(ld_retired, ARMV8_PMUV3_PERFCTR_LD_RETIRED),
ARMV8_EVENT_ATTR(st_retired, ARMV8_PMUV3_PERFCTR_ST_RETIRED),
ARMV8_EVENT_ATTR(inst_retired, ARMV8_PMUV3_PERFCTR_INST_RETIRED),
ARMV8_EVENT_ATTR(exc_taken, ARMV8_PMUV3_PERFCTR_EXC_TAKEN),
ARMV8_EVENT_ATTR(exc_return, ARMV8_PMUV3_PERFCTR_EXC_RETURN),
ARMV8_EVENT_ATTR(cid_write_retired, ARMV8_PMUV3_PERFCTR_CID_WRITE_RETIRED),
ARMV8_EVENT_ATTR(pc_write_retired, ARMV8_PMUV3_PERFCTR_PC_WRITE_RETIRED),
ARMV8_EVENT_ATTR(br_immed_retired, ARMV8_PMUV3_PERFCTR_BR_IMMED_RETIRED),
ARMV8_EVENT_ATTR(br_return_retired, ARMV8_PMUV3_PERFCTR_BR_RETURN_RETIRED),
ARMV8_EVENT_ATTR(unaligned_ldst_retired, ARMV8_PMUV3_PERFCTR_UNALIGNED_LDST_RETIRED),
ARMV8_EVENT_ATTR(br_mis_pred, ARMV8_PMUV3_PERFCTR_BR_MIS_PRED),
ARMV8_EVENT_ATTR(cpu_cycles, ARMV8_PMUV3_PERFCTR_CPU_CYCLES),
ARMV8_EVENT_ATTR(br_pred, ARMV8_PMUV3_PERFCTR_BR_PRED),
ARMV8_EVENT_ATTR(mem_access, ARMV8_PMUV3_PERFCTR_MEM_ACCESS),
ARMV8_EVENT_ATTR(l1i_cache, ARMV8_PMUV3_PERFCTR_L1I_CACHE),
ARMV8_EVENT_ATTR(l1d_cache_wb, ARMV8_PMUV3_PERFCTR_L1D_CACHE_WB),
ARMV8_EVENT_ATTR(l2d_cache, ARMV8_PMUV3_PERFCTR_L2D_CACHE),
ARMV8_EVENT_ATTR(l2d_cache_refill, ARMV8_PMUV3_PERFCTR_L2D_CACHE_REFILL),
ARMV8_EVENT_ATTR(l2d_cache_wb, ARMV8_PMUV3_PERFCTR_L2D_CACHE_WB),
ARMV8_EVENT_ATTR(bus_access, ARMV8_PMUV3_PERFCTR_BUS_ACCESS),
ARMV8_EVENT_ATTR(memory_error, ARMV8_PMUV3_PERFCTR_MEMORY_ERROR),
ARMV8_EVENT_ATTR(inst_spec, ARMV8_PMUV3_PERFCTR_INST_SPEC),
ARMV8_EVENT_ATTR(ttbr_write_retired, ARMV8_PMUV3_PERFCTR_TTBR_WRITE_RETIRED),
ARMV8_EVENT_ATTR(bus_cycles, ARMV8_PMUV3_PERFCTR_BUS_CYCLES),
/* Don't expose the chain event in /sys, since it's useless in isolation */
ARMV8_EVENT_ATTR(l1d_cache_allocate, ARMV8_PMUV3_PERFCTR_L1D_CACHE_ALLOCATE),
ARMV8_EVENT_ATTR(l2d_cache_allocate, ARMV8_PMUV3_PERFCTR_L2D_CACHE_ALLOCATE),
ARMV8_EVENT_ATTR(br_retired, ARMV8_PMUV3_PERFCTR_BR_RETIRED),
ARMV8_EVENT_ATTR(br_mis_pred_retired, ARMV8_PMUV3_PERFCTR_BR_MIS_PRED_RETIRED),
ARMV8_EVENT_ATTR(stall_frontend, ARMV8_PMUV3_PERFCTR_STALL_FRONTEND),
ARMV8_EVENT_ATTR(stall_backend, ARMV8_PMUV3_PERFCTR_STALL_BACKEND),
ARMV8_EVENT_ATTR(l1d_tlb, ARMV8_PMUV3_PERFCTR_L1D_TLB),
ARMV8_EVENT_ATTR(l1i_tlb, ARMV8_PMUV3_PERFCTR_L1I_TLB),
ARMV8_EVENT_ATTR(l2i_cache, ARMV8_PMUV3_PERFCTR_L2I_CACHE),
ARMV8_EVENT_ATTR(l2i_cache_refill, ARMV8_PMUV3_PERFCTR_L2I_CACHE_REFILL),
ARMV8_EVENT_ATTR(l3d_cache_allocate, ARMV8_PMUV3_PERFCTR_L3D_CACHE_ALLOCATE),
ARMV8_EVENT_ATTR(l3d_cache_refill, ARMV8_PMUV3_PERFCTR_L3D_CACHE_REFILL),
ARMV8_EVENT_ATTR(l3d_cache, ARMV8_PMUV3_PERFCTR_L3D_CACHE),
ARMV8_EVENT_ATTR(l3d_cache_wb, ARMV8_PMUV3_PERFCTR_L3D_CACHE_WB),
ARMV8_EVENT_ATTR(l2d_tlb_refill, ARMV8_PMUV3_PERFCTR_L2D_TLB_REFILL),
ARMV8_EVENT_ATTR(l2i_tlb_refill, ARMV8_PMUV3_PERFCTR_L2I_TLB_REFILL),
ARMV8_EVENT_ATTR(l2d_tlb, ARMV8_PMUV3_PERFCTR_L2D_TLB),
ARMV8_EVENT_ATTR(l2i_tlb, ARMV8_PMUV3_PERFCTR_L2I_TLB),
ARMV8_EVENT_ATTR(remote_access, ARMV8_PMUV3_PERFCTR_REMOTE_ACCESS),
ARMV8_EVENT_ATTR(ll_cache, ARMV8_PMUV3_PERFCTR_LL_CACHE),
ARMV8_EVENT_ATTR(ll_cache_miss, ARMV8_PMUV3_PERFCTR_LL_CACHE_MISS),
ARMV8_EVENT_ATTR(dtlb_walk, ARMV8_PMUV3_PERFCTR_DTLB_WALK),
ARMV8_EVENT_ATTR(itlb_walk, ARMV8_PMUV3_PERFCTR_ITLB_WALK),
ARMV8_EVENT_ATTR(ll_cache_rd, ARMV8_PMUV3_PERFCTR_LL_CACHE_RD),
ARMV8_EVENT_ATTR(ll_cache_miss_rd, ARMV8_PMUV3_PERFCTR_LL_CACHE_MISS_RD),
ARMV8_EVENT_ATTR(remote_access_rd, ARMV8_PMUV3_PERFCTR_REMOTE_ACCESS_RD),
ARMV8_EVENT_ATTR(l1d_cache_lmiss_rd, ARMV8_PMUV3_PERFCTR_L1D_CACHE_LMISS_RD),
ARMV8_EVENT_ATTR(op_retired, ARMV8_PMUV3_PERFCTR_OP_RETIRED),
ARMV8_EVENT_ATTR(op_spec, ARMV8_PMUV3_PERFCTR_OP_SPEC),
ARMV8_EVENT_ATTR(stall, ARMV8_PMUV3_PERFCTR_STALL),
ARMV8_EVENT_ATTR(stall_slot_backend, ARMV8_PMUV3_PERFCTR_STALL_SLOT_BACKEND),
ARMV8_EVENT_ATTR(stall_slot_frontend, ARMV8_PMUV3_PERFCTR_STALL_SLOT_FRONTEND),
ARMV8_EVENT_ATTR(stall_slot, ARMV8_PMUV3_PERFCTR_STALL_SLOT),
ARMV8_EVENT_ATTR(sample_pop, ARMV8_SPE_PERFCTR_SAMPLE_POP),
ARMV8_EVENT_ATTR(sample_feed, ARMV8_SPE_PERFCTR_SAMPLE_FEED),
ARMV8_EVENT_ATTR(sample_filtrate, ARMV8_SPE_PERFCTR_SAMPLE_FILTRATE),
ARMV8_EVENT_ATTR(sample_collision, ARMV8_SPE_PERFCTR_SAMPLE_COLLISION),
ARMV8_EVENT_ATTR(cnt_cycles, ARMV8_AMU_PERFCTR_CNT_CYCLES),
ARMV8_EVENT_ATTR(stall_backend_mem, ARMV8_AMU_PERFCTR_STALL_BACKEND_MEM),
ARMV8_EVENT_ATTR(l1i_cache_lmiss, ARMV8_PMUV3_PERFCTR_L1I_CACHE_LMISS),
ARMV8_EVENT_ATTR(l2d_cache_lmiss_rd, ARMV8_PMUV3_PERFCTR_L2D_CACHE_LMISS_RD),
ARMV8_EVENT_ATTR(l2i_cache_lmiss, ARMV8_PMUV3_PERFCTR_L2I_CACHE_LMISS),
ARMV8_EVENT_ATTR(l3d_cache_lmiss_rd, ARMV8_PMUV3_PERFCTR_L3D_CACHE_LMISS_RD),
ARMV8_EVENT_ATTR(ldst_align_lat, ARMV8_PMUV3_PERFCTR_LDST_ALIGN_LAT),
ARMV8_EVENT_ATTR(ld_align_lat, ARMV8_PMUV3_PERFCTR_LD_ALIGN_LAT),
ARMV8_EVENT_ATTR(st_align_lat, ARMV8_PMUV3_PERFCTR_ST_ALIGN_LAT),
ARMV8_EVENT_ATTR(mem_access_checked, ARMV8_MTE_PERFCTR_MEM_ACCESS_CHECKED),
ARMV8_EVENT_ATTR(mem_access_checked_rd, ARMV8_MTE_PERFCTR_MEM_ACCESS_CHECKED_RD),
ARMV8_EVENT_ATTR(mem_access_checked_wr, ARMV8_MTE_PERFCTR_MEM_ACCESS_CHECKED_WR),
NULL,
};
static umode_t
armv8pmu_event_attr_is_visible(struct kobject *kobj,
struct attribute *attr, int unused)
{
struct device *dev = kobj_to_dev(kobj);
struct pmu *pmu = dev_get_drvdata(dev);
struct arm_pmu *cpu_pmu = container_of(pmu, struct arm_pmu, pmu);
struct perf_pmu_events_attr *pmu_attr;
pmu_attr = container_of(attr, struct perf_pmu_events_attr, attr.attr);
if (pmu_attr->id < ARMV8_PMUV3_MAX_COMMON_EVENTS &&
test_bit(pmu_attr->id, cpu_pmu->pmceid_bitmap))
return attr->mode;
if (pmu_attr->id >= ARMV8_PMUV3_EXT_COMMON_EVENT_BASE) {
u64 id = pmu_attr->id - ARMV8_PMUV3_EXT_COMMON_EVENT_BASE;
if (id < ARMV8_PMUV3_MAX_COMMON_EVENTS &&
test_bit(id, cpu_pmu->pmceid_ext_bitmap))
return attr->mode;
}
return 0;
}
static const struct attribute_group armv8_pmuv3_events_attr_group = {
.name = "events",
.attrs = armv8_pmuv3_event_attrs,
.is_visible = armv8pmu_event_attr_is_visible,
};
PMU_FORMAT_ATTR(event, "config:0-15");
PMU_FORMAT_ATTR(long, "config1:0");
static inline bool armv8pmu_event_is_64bit(struct perf_event *event)
{
return event->attr.config1 & 0x1;
}
static struct attribute *armv8_pmuv3_format_attrs[] = {
&format_attr_event.attr,
&format_attr_long.attr,
NULL,
};
static const struct attribute_group armv8_pmuv3_format_attr_group = {
.name = "format",
.attrs = armv8_pmuv3_format_attrs,
};
static ssize_t slots_show(struct device *dev, struct device_attribute *attr,
char *page)
{
struct pmu *pmu = dev_get_drvdata(dev);
struct arm_pmu *cpu_pmu = container_of(pmu, struct arm_pmu, pmu);
u32 slots = cpu_pmu->reg_pmmir & ARMV8_PMU_SLOTS_MASK;
return sysfs_emit(page, "0x%08x\n", slots);
}
static DEVICE_ATTR_RO(slots);
static ssize_t bus_slots_show(struct device *dev, struct device_attribute *attr,
char *page)
{
struct pmu *pmu = dev_get_drvdata(dev);
struct arm_pmu *cpu_pmu = container_of(pmu, struct arm_pmu, pmu);
u32 bus_slots = (cpu_pmu->reg_pmmir >> ARMV8_PMU_BUS_SLOTS_SHIFT)
& ARMV8_PMU_BUS_SLOTS_MASK;
return sysfs_emit(page, "0x%08x\n", bus_slots);
}
static DEVICE_ATTR_RO(bus_slots);
static ssize_t bus_width_show(struct device *dev, struct device_attribute *attr,
char *page)
{
struct pmu *pmu = dev_get_drvdata(dev);
struct arm_pmu *cpu_pmu = container_of(pmu, struct arm_pmu, pmu);
u32 bus_width = (cpu_pmu->reg_pmmir >> ARMV8_PMU_BUS_WIDTH_SHIFT)
& ARMV8_PMU_BUS_WIDTH_MASK;
u32 val = 0;
/* Encoded as Log2(number of bytes), plus one */
if (bus_width > 2 && bus_width < 13)
val = 1 << (bus_width - 1);
return sysfs_emit(page, "0x%08x\n", val);
}
static DEVICE_ATTR_RO(bus_width);
static struct attribute *armv8_pmuv3_caps_attrs[] = {
&dev_attr_slots.attr,
&dev_attr_bus_slots.attr,
&dev_attr_bus_width.attr,
NULL,
};
static const struct attribute_group armv8_pmuv3_caps_attr_group = {
.name = "caps",
.attrs = armv8_pmuv3_caps_attrs,
};
/*
* Perf Events' indices
*/
#define ARMV8_IDX_CYCLE_COUNTER 0
#define ARMV8_IDX_COUNTER0 1
/*
* We unconditionally enable ARMv8.5-PMU long event counter support
* (64-bit events) where supported. Indicate if this arm_pmu has long
* event counter support.
*/
static bool armv8pmu_has_long_event(struct arm_pmu *cpu_pmu)
{
return (cpu_pmu->pmuver >= ID_AA64DFR0_PMUVER_8_5);
}
/*
* We must chain two programmable counters for 64 bit events,
* except when we have allocated the 64bit cycle counter (for CPU
* cycles event). This must be called only when the event has
* a counter allocated.
*/
static inline bool armv8pmu_event_is_chained(struct perf_event *event)
{
int idx = event->hw.idx;
struct arm_pmu *cpu_pmu = to_arm_pmu(event->pmu);
return !WARN_ON(idx < 0) &&
armv8pmu_event_is_64bit(event) &&
!armv8pmu_has_long_event(cpu_pmu) &&
(idx != ARMV8_IDX_CYCLE_COUNTER);
}
/*
* ARMv8 low level PMU access
*/
/*
* Perf Event to low level counters mapping
*/
#define ARMV8_IDX_TO_COUNTER(x) \
(((x) - ARMV8_IDX_COUNTER0) & ARMV8_PMU_COUNTER_MASK)
arm64: perf: Avoid PMXEV* indirection Currently we access the counter registers and their respective type registers indirectly. This requires us to write to PMSELR, issue an ISB, then access the relevant PMXEV* registers. This is unfortunate, because: * Under virtualization, accessing one register requires two traps to the hypervisor, even though we could access the register directly with a single trap. * We have to issue an ISB which we could otherwise avoid the cost of. * When we use NMIs, the NMI handler will have to save/restore the select register in case the code it preempted was attempting to access a counter or its type register. We can avoid these issues by directly accessing the relevant registers. This patch adds helpers to do so. In armv8pmu_enable_event() we still need the ISB to prevent the PE from reordering the write to PMINTENSET_EL1 register. If the interrupt is enabled before we disable the counter and the new event is configured, we might get an interrupt triggered by the previously programmed event overflowing, but which we wrongly attribute to the event that we are enabling. Execute an ISB after we disable the counter. In the process, remove the comment that refers to the ARMv7 PMU. [Julien T.: Don't inline read/write functions to avoid big code-size increase, remove unused read_pmevtypern function, fix counter index issue.] [Alexandru E.: Removed comment, removed trailing semicolons in macros, added ISB] Signed-off-by: Mark Rutland <mark.rutland@arm.com> Signed-off-by: Julien Thierry <julien.thierry@arm.com> Signed-off-by: Alexandru Elisei <alexandru.elisei@arm.com> Tested-by: Sumit Garg <sumit.garg@linaro.org> (Developerbox) Cc: Julien Thierry <julien.thierry.kdev@gmail.com> Cc: Will Deacon <will.deacon@arm.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Ingo Molnar <mingo@redhat.com> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Catalin Marinas <catalin.marinas@arm.com> Link: https://lore.kernel.org/r/20200924110706.254996-3-alexandru.elisei@arm.com Signed-off-by: Will Deacon <will@kernel.org>
2020-09-24 11:07:01 +00:00
/*
* This code is really good
*/
#define PMEVN_CASE(n, case_macro) \
case n: case_macro(n); break
#define PMEVN_SWITCH(x, case_macro) \
do { \
switch (x) { \
PMEVN_CASE(0, case_macro); \
PMEVN_CASE(1, case_macro); \
PMEVN_CASE(2, case_macro); \
PMEVN_CASE(3, case_macro); \
PMEVN_CASE(4, case_macro); \
PMEVN_CASE(5, case_macro); \
PMEVN_CASE(6, case_macro); \
PMEVN_CASE(7, case_macro); \
PMEVN_CASE(8, case_macro); \
PMEVN_CASE(9, case_macro); \
PMEVN_CASE(10, case_macro); \
PMEVN_CASE(11, case_macro); \
PMEVN_CASE(12, case_macro); \
PMEVN_CASE(13, case_macro); \
PMEVN_CASE(14, case_macro); \
PMEVN_CASE(15, case_macro); \
PMEVN_CASE(16, case_macro); \
PMEVN_CASE(17, case_macro); \
PMEVN_CASE(18, case_macro); \
PMEVN_CASE(19, case_macro); \
PMEVN_CASE(20, case_macro); \
PMEVN_CASE(21, case_macro); \
PMEVN_CASE(22, case_macro); \
PMEVN_CASE(23, case_macro); \
PMEVN_CASE(24, case_macro); \
PMEVN_CASE(25, case_macro); \
PMEVN_CASE(26, case_macro); \
PMEVN_CASE(27, case_macro); \
PMEVN_CASE(28, case_macro); \
PMEVN_CASE(29, case_macro); \
PMEVN_CASE(30, case_macro); \
default: WARN(1, "Invalid PMEV* index\n"); \
} \
} while (0)
#define RETURN_READ_PMEVCNTRN(n) \
return read_sysreg(pmevcntr##n##_el0)
static unsigned long read_pmevcntrn(int n)
{
PMEVN_SWITCH(n, RETURN_READ_PMEVCNTRN);
return 0;
}
#define WRITE_PMEVCNTRN(n) \
write_sysreg(val, pmevcntr##n##_el0)
static void write_pmevcntrn(int n, unsigned long val)
{
PMEVN_SWITCH(n, WRITE_PMEVCNTRN);
}
#define WRITE_PMEVTYPERN(n) \
write_sysreg(val, pmevtyper##n##_el0)
static void write_pmevtypern(int n, unsigned long val)
{
PMEVN_SWITCH(n, WRITE_PMEVTYPERN);
}
static inline u32 armv8pmu_pmcr_read(void)
{
return read_sysreg(pmcr_el0);
}
static inline void armv8pmu_pmcr_write(u32 val)
{
val &= ARMV8_PMU_PMCR_MASK;
isb();
write_sysreg(val, pmcr_el0);
}
static inline int armv8pmu_has_overflowed(u32 pmovsr)
{
return pmovsr & ARMV8_PMU_OVERFLOWED_MASK;
}
static inline int armv8pmu_counter_has_overflowed(u32 pmnc, int idx)
{
return pmnc & BIT(ARMV8_IDX_TO_COUNTER(idx));
}
static inline u64 armv8pmu_read_evcntr(int idx)
{
u32 counter = ARMV8_IDX_TO_COUNTER(idx);
arm64: perf: Avoid PMXEV* indirection Currently we access the counter registers and their respective type registers indirectly. This requires us to write to PMSELR, issue an ISB, then access the relevant PMXEV* registers. This is unfortunate, because: * Under virtualization, accessing one register requires two traps to the hypervisor, even though we could access the register directly with a single trap. * We have to issue an ISB which we could otherwise avoid the cost of. * When we use NMIs, the NMI handler will have to save/restore the select register in case the code it preempted was attempting to access a counter or its type register. We can avoid these issues by directly accessing the relevant registers. This patch adds helpers to do so. In armv8pmu_enable_event() we still need the ISB to prevent the PE from reordering the write to PMINTENSET_EL1 register. If the interrupt is enabled before we disable the counter and the new event is configured, we might get an interrupt triggered by the previously programmed event overflowing, but which we wrongly attribute to the event that we are enabling. Execute an ISB after we disable the counter. In the process, remove the comment that refers to the ARMv7 PMU. [Julien T.: Don't inline read/write functions to avoid big code-size increase, remove unused read_pmevtypern function, fix counter index issue.] [Alexandru E.: Removed comment, removed trailing semicolons in macros, added ISB] Signed-off-by: Mark Rutland <mark.rutland@arm.com> Signed-off-by: Julien Thierry <julien.thierry@arm.com> Signed-off-by: Alexandru Elisei <alexandru.elisei@arm.com> Tested-by: Sumit Garg <sumit.garg@linaro.org> (Developerbox) Cc: Julien Thierry <julien.thierry.kdev@gmail.com> Cc: Will Deacon <will.deacon@arm.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Ingo Molnar <mingo@redhat.com> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Catalin Marinas <catalin.marinas@arm.com> Link: https://lore.kernel.org/r/20200924110706.254996-3-alexandru.elisei@arm.com Signed-off-by: Will Deacon <will@kernel.org>
2020-09-24 11:07:01 +00:00
return read_pmevcntrn(counter);
}
static inline u64 armv8pmu_read_hw_counter(struct perf_event *event)
{
int idx = event->hw.idx;
u64 val = armv8pmu_read_evcntr(idx);
if (armv8pmu_event_is_chained(event))
val = (val << 32) | armv8pmu_read_evcntr(idx - 1);
return val;
}
/*
* The cycle counter is always a 64-bit counter. When ARMV8_PMU_PMCR_LP
* is set the event counters also become 64-bit counters. Unless the
* user has requested a long counter (attr.config1) then we want to
* interrupt upon 32-bit overflow - we achieve this by applying a bias.
*/
static bool armv8pmu_event_needs_bias(struct perf_event *event)
{
struct arm_pmu *cpu_pmu = to_arm_pmu(event->pmu);
struct hw_perf_event *hwc = &event->hw;
int idx = hwc->idx;
if (armv8pmu_event_is_64bit(event))
return false;
if (armv8pmu_has_long_event(cpu_pmu) ||
idx == ARMV8_IDX_CYCLE_COUNTER)
return true;
return false;
}
static u64 armv8pmu_bias_long_counter(struct perf_event *event, u64 value)
{
if (armv8pmu_event_needs_bias(event))
value |= GENMASK(63, 32);
return value;
}
static u64 armv8pmu_unbias_long_counter(struct perf_event *event, u64 value)
{
if (armv8pmu_event_needs_bias(event))
value &= ~GENMASK(63, 32);
return value;
}
static u64 armv8pmu_read_counter(struct perf_event *event)
{
struct hw_perf_event *hwc = &event->hw;
int idx = hwc->idx;
u64 value;
if (idx == ARMV8_IDX_CYCLE_COUNTER)
value = read_sysreg(pmccntr_el0);
else
value = armv8pmu_read_hw_counter(event);
return armv8pmu_unbias_long_counter(event, value);
}
static inline void armv8pmu_write_evcntr(int idx, u64 value)
{
arm64: perf: Avoid PMXEV* indirection Currently we access the counter registers and their respective type registers indirectly. This requires us to write to PMSELR, issue an ISB, then access the relevant PMXEV* registers. This is unfortunate, because: * Under virtualization, accessing one register requires two traps to the hypervisor, even though we could access the register directly with a single trap. * We have to issue an ISB which we could otherwise avoid the cost of. * When we use NMIs, the NMI handler will have to save/restore the select register in case the code it preempted was attempting to access a counter or its type register. We can avoid these issues by directly accessing the relevant registers. This patch adds helpers to do so. In armv8pmu_enable_event() we still need the ISB to prevent the PE from reordering the write to PMINTENSET_EL1 register. If the interrupt is enabled before we disable the counter and the new event is configured, we might get an interrupt triggered by the previously programmed event overflowing, but which we wrongly attribute to the event that we are enabling. Execute an ISB after we disable the counter. In the process, remove the comment that refers to the ARMv7 PMU. [Julien T.: Don't inline read/write functions to avoid big code-size increase, remove unused read_pmevtypern function, fix counter index issue.] [Alexandru E.: Removed comment, removed trailing semicolons in macros, added ISB] Signed-off-by: Mark Rutland <mark.rutland@arm.com> Signed-off-by: Julien Thierry <julien.thierry@arm.com> Signed-off-by: Alexandru Elisei <alexandru.elisei@arm.com> Tested-by: Sumit Garg <sumit.garg@linaro.org> (Developerbox) Cc: Julien Thierry <julien.thierry.kdev@gmail.com> Cc: Will Deacon <will.deacon@arm.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Ingo Molnar <mingo@redhat.com> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Catalin Marinas <catalin.marinas@arm.com> Link: https://lore.kernel.org/r/20200924110706.254996-3-alexandru.elisei@arm.com Signed-off-by: Will Deacon <will@kernel.org>
2020-09-24 11:07:01 +00:00
u32 counter = ARMV8_IDX_TO_COUNTER(idx);
write_pmevcntrn(counter, value);
}
static inline void armv8pmu_write_hw_counter(struct perf_event *event,
u64 value)
{
int idx = event->hw.idx;
if (armv8pmu_event_is_chained(event)) {
armv8pmu_write_evcntr(idx, upper_32_bits(value));
armv8pmu_write_evcntr(idx - 1, lower_32_bits(value));
} else {
armv8pmu_write_evcntr(idx, value);
}
}
static void armv8pmu_write_counter(struct perf_event *event, u64 value)
{
struct hw_perf_event *hwc = &event->hw;
int idx = hwc->idx;
value = armv8pmu_bias_long_counter(event, value);
if (idx == ARMV8_IDX_CYCLE_COUNTER)
write_sysreg(value, pmccntr_el0);
else
armv8pmu_write_hw_counter(event, value);
}
static inline void armv8pmu_write_evtype(int idx, u32 val)
{
arm64: perf: Avoid PMXEV* indirection Currently we access the counter registers and their respective type registers indirectly. This requires us to write to PMSELR, issue an ISB, then access the relevant PMXEV* registers. This is unfortunate, because: * Under virtualization, accessing one register requires two traps to the hypervisor, even though we could access the register directly with a single trap. * We have to issue an ISB which we could otherwise avoid the cost of. * When we use NMIs, the NMI handler will have to save/restore the select register in case the code it preempted was attempting to access a counter or its type register. We can avoid these issues by directly accessing the relevant registers. This patch adds helpers to do so. In armv8pmu_enable_event() we still need the ISB to prevent the PE from reordering the write to PMINTENSET_EL1 register. If the interrupt is enabled before we disable the counter and the new event is configured, we might get an interrupt triggered by the previously programmed event overflowing, but which we wrongly attribute to the event that we are enabling. Execute an ISB after we disable the counter. In the process, remove the comment that refers to the ARMv7 PMU. [Julien T.: Don't inline read/write functions to avoid big code-size increase, remove unused read_pmevtypern function, fix counter index issue.] [Alexandru E.: Removed comment, removed trailing semicolons in macros, added ISB] Signed-off-by: Mark Rutland <mark.rutland@arm.com> Signed-off-by: Julien Thierry <julien.thierry@arm.com> Signed-off-by: Alexandru Elisei <alexandru.elisei@arm.com> Tested-by: Sumit Garg <sumit.garg@linaro.org> (Developerbox) Cc: Julien Thierry <julien.thierry.kdev@gmail.com> Cc: Will Deacon <will.deacon@arm.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Ingo Molnar <mingo@redhat.com> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Catalin Marinas <catalin.marinas@arm.com> Link: https://lore.kernel.org/r/20200924110706.254996-3-alexandru.elisei@arm.com Signed-off-by: Will Deacon <will@kernel.org>
2020-09-24 11:07:01 +00:00
u32 counter = ARMV8_IDX_TO_COUNTER(idx);
val &= ARMV8_PMU_EVTYPE_MASK;
arm64: perf: Avoid PMXEV* indirection Currently we access the counter registers and their respective type registers indirectly. This requires us to write to PMSELR, issue an ISB, then access the relevant PMXEV* registers. This is unfortunate, because: * Under virtualization, accessing one register requires two traps to the hypervisor, even though we could access the register directly with a single trap. * We have to issue an ISB which we could otherwise avoid the cost of. * When we use NMIs, the NMI handler will have to save/restore the select register in case the code it preempted was attempting to access a counter or its type register. We can avoid these issues by directly accessing the relevant registers. This patch adds helpers to do so. In armv8pmu_enable_event() we still need the ISB to prevent the PE from reordering the write to PMINTENSET_EL1 register. If the interrupt is enabled before we disable the counter and the new event is configured, we might get an interrupt triggered by the previously programmed event overflowing, but which we wrongly attribute to the event that we are enabling. Execute an ISB after we disable the counter. In the process, remove the comment that refers to the ARMv7 PMU. [Julien T.: Don't inline read/write functions to avoid big code-size increase, remove unused read_pmevtypern function, fix counter index issue.] [Alexandru E.: Removed comment, removed trailing semicolons in macros, added ISB] Signed-off-by: Mark Rutland <mark.rutland@arm.com> Signed-off-by: Julien Thierry <julien.thierry@arm.com> Signed-off-by: Alexandru Elisei <alexandru.elisei@arm.com> Tested-by: Sumit Garg <sumit.garg@linaro.org> (Developerbox) Cc: Julien Thierry <julien.thierry.kdev@gmail.com> Cc: Will Deacon <will.deacon@arm.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Ingo Molnar <mingo@redhat.com> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Catalin Marinas <catalin.marinas@arm.com> Link: https://lore.kernel.org/r/20200924110706.254996-3-alexandru.elisei@arm.com Signed-off-by: Will Deacon <will@kernel.org>
2020-09-24 11:07:01 +00:00
write_pmevtypern(counter, val);
}
static inline void armv8pmu_write_event_type(struct perf_event *event)
{
struct hw_perf_event *hwc = &event->hw;
int idx = hwc->idx;
/*
* For chained events, the low counter is programmed to count
* the event of interest and the high counter is programmed
* with CHAIN event code with filters set to count at all ELs.
*/
if (armv8pmu_event_is_chained(event)) {
u32 chain_evt = ARMV8_PMUV3_PERFCTR_CHAIN |
ARMV8_PMU_INCLUDE_EL2;
armv8pmu_write_evtype(idx - 1, hwc->config_base);
armv8pmu_write_evtype(idx, chain_evt);
} else {
arm64: perf: Avoid PMXEV* indirection Currently we access the counter registers and their respective type registers indirectly. This requires us to write to PMSELR, issue an ISB, then access the relevant PMXEV* registers. This is unfortunate, because: * Under virtualization, accessing one register requires two traps to the hypervisor, even though we could access the register directly with a single trap. * We have to issue an ISB which we could otherwise avoid the cost of. * When we use NMIs, the NMI handler will have to save/restore the select register in case the code it preempted was attempting to access a counter or its type register. We can avoid these issues by directly accessing the relevant registers. This patch adds helpers to do so. In armv8pmu_enable_event() we still need the ISB to prevent the PE from reordering the write to PMINTENSET_EL1 register. If the interrupt is enabled before we disable the counter and the new event is configured, we might get an interrupt triggered by the previously programmed event overflowing, but which we wrongly attribute to the event that we are enabling. Execute an ISB after we disable the counter. In the process, remove the comment that refers to the ARMv7 PMU. [Julien T.: Don't inline read/write functions to avoid big code-size increase, remove unused read_pmevtypern function, fix counter index issue.] [Alexandru E.: Removed comment, removed trailing semicolons in macros, added ISB] Signed-off-by: Mark Rutland <mark.rutland@arm.com> Signed-off-by: Julien Thierry <julien.thierry@arm.com> Signed-off-by: Alexandru Elisei <alexandru.elisei@arm.com> Tested-by: Sumit Garg <sumit.garg@linaro.org> (Developerbox) Cc: Julien Thierry <julien.thierry.kdev@gmail.com> Cc: Will Deacon <will.deacon@arm.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Ingo Molnar <mingo@redhat.com> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Catalin Marinas <catalin.marinas@arm.com> Link: https://lore.kernel.org/r/20200924110706.254996-3-alexandru.elisei@arm.com Signed-off-by: Will Deacon <will@kernel.org>
2020-09-24 11:07:01 +00:00
if (idx == ARMV8_IDX_CYCLE_COUNTER)
write_sysreg(hwc->config_base, pmccfiltr_el0);
else
armv8pmu_write_evtype(idx, hwc->config_base);
}
}
static u32 armv8pmu_event_cnten_mask(struct perf_event *event)
{
int counter = ARMV8_IDX_TO_COUNTER(event->hw.idx);
u32 mask = BIT(counter);
if (armv8pmu_event_is_chained(event))
mask |= BIT(counter - 1);
return mask;
}
static inline void armv8pmu_enable_counter(u32 mask)
{
2020-09-24 11:07:00 +00:00
/*
* Make sure event configuration register writes are visible before we
* enable the counter.
* */
isb();
write_sysreg(mask, pmcntenset_el0);
}
static inline void armv8pmu_enable_event_counter(struct perf_event *event)
{
struct perf_event_attr *attr = &event->attr;
u32 mask = armv8pmu_event_cnten_mask(event);
kvm_set_pmu_events(mask, attr);
/* We rely on the hypervisor switch code to enable guest counters */
if (!kvm_pmu_counter_deferred(attr))
armv8pmu_enable_counter(mask);
}
static inline void armv8pmu_disable_counter(u32 mask)
{
write_sysreg(mask, pmcntenclr_el0);
arm64: perf: Avoid PMXEV* indirection Currently we access the counter registers and their respective type registers indirectly. This requires us to write to PMSELR, issue an ISB, then access the relevant PMXEV* registers. This is unfortunate, because: * Under virtualization, accessing one register requires two traps to the hypervisor, even though we could access the register directly with a single trap. * We have to issue an ISB which we could otherwise avoid the cost of. * When we use NMIs, the NMI handler will have to save/restore the select register in case the code it preempted was attempting to access a counter or its type register. We can avoid these issues by directly accessing the relevant registers. This patch adds helpers to do so. In armv8pmu_enable_event() we still need the ISB to prevent the PE from reordering the write to PMINTENSET_EL1 register. If the interrupt is enabled before we disable the counter and the new event is configured, we might get an interrupt triggered by the previously programmed event overflowing, but which we wrongly attribute to the event that we are enabling. Execute an ISB after we disable the counter. In the process, remove the comment that refers to the ARMv7 PMU. [Julien T.: Don't inline read/write functions to avoid big code-size increase, remove unused read_pmevtypern function, fix counter index issue.] [Alexandru E.: Removed comment, removed trailing semicolons in macros, added ISB] Signed-off-by: Mark Rutland <mark.rutland@arm.com> Signed-off-by: Julien Thierry <julien.thierry@arm.com> Signed-off-by: Alexandru Elisei <alexandru.elisei@arm.com> Tested-by: Sumit Garg <sumit.garg@linaro.org> (Developerbox) Cc: Julien Thierry <julien.thierry.kdev@gmail.com> Cc: Will Deacon <will.deacon@arm.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Ingo Molnar <mingo@redhat.com> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Catalin Marinas <catalin.marinas@arm.com> Link: https://lore.kernel.org/r/20200924110706.254996-3-alexandru.elisei@arm.com Signed-off-by: Will Deacon <will@kernel.org>
2020-09-24 11:07:01 +00:00
/*
* Make sure the effects of disabling the counter are visible before we
* start configuring the event.
*/
isb();
}
static inline void armv8pmu_disable_event_counter(struct perf_event *event)
{
struct perf_event_attr *attr = &event->attr;
u32 mask = armv8pmu_event_cnten_mask(event);
kvm_clr_pmu_events(mask);
/* We rely on the hypervisor switch code to disable guest counters */
if (!kvm_pmu_counter_deferred(attr))
armv8pmu_disable_counter(mask);
}
static inline void armv8pmu_enable_intens(u32 mask)
{
write_sysreg(mask, pmintenset_el1);
}
static inline void armv8pmu_enable_event_irq(struct perf_event *event)
{
u32 counter = ARMV8_IDX_TO_COUNTER(event->hw.idx);
armv8pmu_enable_intens(BIT(counter));
}
static inline void armv8pmu_disable_intens(u32 mask)
{
write_sysreg(mask, pmintenclr_el1);
isb();
/* Clear the overflow flag in case an interrupt is pending. */
write_sysreg(mask, pmovsclr_el0);
isb();
}
static inline void armv8pmu_disable_event_irq(struct perf_event *event)
{
u32 counter = ARMV8_IDX_TO_COUNTER(event->hw.idx);
armv8pmu_disable_intens(BIT(counter));
}
static inline u32 armv8pmu_getreset_flags(void)
{
u32 value;
/* Read */
value = read_sysreg(pmovsclr_el0);
/* Write to clear flags */
value &= ARMV8_PMU_OVSR_MASK;
write_sysreg(value, pmovsclr_el0);
return value;
}
static void armv8pmu_enable_event(struct perf_event *event)
{
/*
* Enable counter and interrupt, and set the counter to count
* the event that we're interested in.
*/
/*
* Disable counter
*/
armv8pmu_disable_event_counter(event);
/*
arm64: perf: Avoid PMXEV* indirection Currently we access the counter registers and their respective type registers indirectly. This requires us to write to PMSELR, issue an ISB, then access the relevant PMXEV* registers. This is unfortunate, because: * Under virtualization, accessing one register requires two traps to the hypervisor, even though we could access the register directly with a single trap. * We have to issue an ISB which we could otherwise avoid the cost of. * When we use NMIs, the NMI handler will have to save/restore the select register in case the code it preempted was attempting to access a counter or its type register. We can avoid these issues by directly accessing the relevant registers. This patch adds helpers to do so. In armv8pmu_enable_event() we still need the ISB to prevent the PE from reordering the write to PMINTENSET_EL1 register. If the interrupt is enabled before we disable the counter and the new event is configured, we might get an interrupt triggered by the previously programmed event overflowing, but which we wrongly attribute to the event that we are enabling. Execute an ISB after we disable the counter. In the process, remove the comment that refers to the ARMv7 PMU. [Julien T.: Don't inline read/write functions to avoid big code-size increase, remove unused read_pmevtypern function, fix counter index issue.] [Alexandru E.: Removed comment, removed trailing semicolons in macros, added ISB] Signed-off-by: Mark Rutland <mark.rutland@arm.com> Signed-off-by: Julien Thierry <julien.thierry@arm.com> Signed-off-by: Alexandru Elisei <alexandru.elisei@arm.com> Tested-by: Sumit Garg <sumit.garg@linaro.org> (Developerbox) Cc: Julien Thierry <julien.thierry.kdev@gmail.com> Cc: Will Deacon <will.deacon@arm.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Ingo Molnar <mingo@redhat.com> Cc: Arnaldo Carvalho de Melo <acme@kernel.org> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Catalin Marinas <catalin.marinas@arm.com> Link: https://lore.kernel.org/r/20200924110706.254996-3-alexandru.elisei@arm.com Signed-off-by: Will Deacon <will@kernel.org>
2020-09-24 11:07:01 +00:00
* Set event.
*/
armv8pmu_write_event_type(event);
/*
* Enable interrupt for this counter
*/
armv8pmu_enable_event_irq(event);
/*
* Enable counter
*/
armv8pmu_enable_event_counter(event);
}
static void armv8pmu_disable_event(struct perf_event *event)
{
/*
* Disable counter
*/
armv8pmu_disable_event_counter(event);
/*
* Disable interrupt for this counter
*/
armv8pmu_disable_event_irq(event);
}
static void armv8pmu_start(struct arm_pmu *cpu_pmu)
{
/* Enable all counters */
armv8pmu_pmcr_write(armv8pmu_pmcr_read() | ARMV8_PMU_PMCR_E);
}
static void armv8pmu_stop(struct arm_pmu *cpu_pmu)
{
/* Disable all counters */
armv8pmu_pmcr_write(armv8pmu_pmcr_read() & ~ARMV8_PMU_PMCR_E);
}
static irqreturn_t armv8pmu_handle_irq(struct arm_pmu *cpu_pmu)
{
u32 pmovsr;
struct perf_sample_data data;
struct pmu_hw_events *cpuc = this_cpu_ptr(cpu_pmu->hw_events);
struct pt_regs *regs;
int idx;
/*
* Get and reset the IRQ flags
*/
pmovsr = armv8pmu_getreset_flags();
/*
* Did an overflow occur?
*/
if (!armv8pmu_has_overflowed(pmovsr))
return IRQ_NONE;
/*
* Handle the counter(s) overflow(s)
*/
regs = get_irq_regs();
/*
* Stop the PMU while processing the counter overflows
* to prevent skews in group events.
*/
armv8pmu_stop(cpu_pmu);
for (idx = 0; idx < cpu_pmu->num_events; ++idx) {
struct perf_event *event = cpuc->events[idx];
struct hw_perf_event *hwc;
/* Ignore if we don't have an event. */
if (!event)
continue;
/*
* We have a single interrupt for all counters. Check that
* each counter has overflowed before we process it.
*/
if (!armv8pmu_counter_has_overflowed(pmovsr, idx))
continue;
hwc = &event->hw;
armpmu_event_update(event);
perf_sample_data_init(&data, 0, hwc->last_period);
if (!armpmu_event_set_period(event))
continue;
2020-09-24 11:07:03 +00:00
/*
* Perf event overflow will queue the processing of the event as
* an irq_work which will be taken care of in the handling of
* IPI_IRQ_WORK.
*/
if (perf_event_overflow(event, &data, regs))
cpu_pmu->disable(event);
}
armv8pmu_start(cpu_pmu);
return IRQ_HANDLED;
}
static int armv8pmu_get_single_idx(struct pmu_hw_events *cpuc,
struct arm_pmu *cpu_pmu)
{
int idx;
for (idx = ARMV8_IDX_COUNTER0; idx < cpu_pmu->num_events; idx++) {
if (!test_and_set_bit(idx, cpuc->used_mask))
return idx;
}
return -EAGAIN;
}
static int armv8pmu_get_chain_idx(struct pmu_hw_events *cpuc,
struct arm_pmu *cpu_pmu)
{
int idx;
/*
* Chaining requires two consecutive event counters, where
* the lower idx must be even.
*/
for (idx = ARMV8_IDX_COUNTER0 + 1; idx < cpu_pmu->num_events; idx += 2) {
if (!test_and_set_bit(idx, cpuc->used_mask)) {
/* Check if the preceding even counter is available */
if (!test_and_set_bit(idx - 1, cpuc->used_mask))
return idx;
/* Release the Odd counter */
clear_bit(idx, cpuc->used_mask);
}
}
return -EAGAIN;
}
static int armv8pmu_get_event_idx(struct pmu_hw_events *cpuc,
struct perf_event *event)
{
struct arm_pmu *cpu_pmu = to_arm_pmu(event->pmu);
struct hw_perf_event *hwc = &event->hw;
unsigned long evtype = hwc->config_base & ARMV8_PMU_EVTYPE_EVENT;
/* Always prefer to place a cycle counter into the cycle counter. */
if (evtype == ARMV8_PMUV3_PERFCTR_CPU_CYCLES) {
if (!test_and_set_bit(ARMV8_IDX_CYCLE_COUNTER, cpuc->used_mask))
return ARMV8_IDX_CYCLE_COUNTER;
}
/*
* Otherwise use events counters
*/
if (armv8pmu_event_is_64bit(event) &&
!armv8pmu_has_long_event(cpu_pmu))
return armv8pmu_get_chain_idx(cpuc, cpu_pmu);
else
return armv8pmu_get_single_idx(cpuc, cpu_pmu);
}
static void armv8pmu_clear_event_idx(struct pmu_hw_events *cpuc,
struct perf_event *event)
{
int idx = event->hw.idx;
clear_bit(idx, cpuc->used_mask);
if (armv8pmu_event_is_chained(event))
clear_bit(idx - 1, cpuc->used_mask);
}
/*
* Add an event filter to a given event.
*/
static int armv8pmu_set_event_filter(struct hw_perf_event *event,
struct perf_event_attr *attr)
{
unsigned long config_base = 0;
if (attr->exclude_idle)
return -EPERM;
/*
* If we're running in hyp mode, then we *are* the hypervisor.
* Therefore we ignore exclude_hv in this configuration, since
* there's no hypervisor to sample anyway. This is consistent
* with other architectures (x86 and Power).
*/
if (is_kernel_in_hyp_mode()) {
if (!attr->exclude_kernel && !attr->exclude_host)
config_base |= ARMV8_PMU_INCLUDE_EL2;
if (attr->exclude_guest)
config_base |= ARMV8_PMU_EXCLUDE_EL1;
if (attr->exclude_host)
config_base |= ARMV8_PMU_EXCLUDE_EL0;
} else {
if (!attr->exclude_hv && !attr->exclude_host)
config_base |= ARMV8_PMU_INCLUDE_EL2;
}
/*
* Filter out !VHE kernels and guest kernels
*/
if (attr->exclude_kernel)
config_base |= ARMV8_PMU_EXCLUDE_EL1;
if (attr->exclude_user)
config_base |= ARMV8_PMU_EXCLUDE_EL0;
/*
* Install the filter into config_base as this is used to
* construct the event type.
*/
event->config_base = config_base;
return 0;
}
static int armv8pmu_filter_match(struct perf_event *event)
{
unsigned long evtype = event->hw.config_base & ARMV8_PMU_EVTYPE_EVENT;
return evtype != ARMV8_PMUV3_PERFCTR_CHAIN;
}
static void armv8pmu_reset(void *info)
{
struct arm_pmu *cpu_pmu = (struct arm_pmu *)info;
u32 pmcr;
/* The counter and interrupt enable registers are unknown at reset. */
armv8pmu_disable_counter(U32_MAX);
armv8pmu_disable_intens(U32_MAX);
/* Clear the counters we flip at guest entry/exit */
kvm_clr_pmu_events(U32_MAX);
/*
* Initialize & Reset PMNC. Request overflow interrupt for
* 64 bit cycle counter but cheat in armv8pmu_write_counter().
*/
pmcr = ARMV8_PMU_PMCR_P | ARMV8_PMU_PMCR_C | ARMV8_PMU_PMCR_LC;
/* Enable long event counter support where available */
if (armv8pmu_has_long_event(cpu_pmu))
pmcr |= ARMV8_PMU_PMCR_LP;
armv8pmu_pmcr_write(pmcr);
}
static int __armv8_pmuv3_map_event(struct perf_event *event,
const unsigned (*extra_event_map)
[PERF_COUNT_HW_MAX],
const unsigned (*extra_cache_map)
[PERF_COUNT_HW_CACHE_MAX]
[PERF_COUNT_HW_CACHE_OP_MAX]
[PERF_COUNT_HW_CACHE_RESULT_MAX])
{
int hw_event_id;
struct arm_pmu *armpmu = to_arm_pmu(event->pmu);
hw_event_id = armpmu_map_event(event, &armv8_pmuv3_perf_map,
&armv8_pmuv3_perf_cache_map,
ARMV8_PMU_EVTYPE_EVENT);
if (armv8pmu_event_is_64bit(event))
event->hw.flags |= ARMPMU_EVT_64BIT;
/* Only expose micro/arch events supported by this PMU */
if ((hw_event_id > 0) && (hw_event_id < ARMV8_PMUV3_MAX_COMMON_EVENTS)
&& test_bit(hw_event_id, armpmu->pmceid_bitmap)) {
return hw_event_id;
}
return armpmu_map_event(event, extra_event_map, extra_cache_map,
ARMV8_PMU_EVTYPE_EVENT);
}
static int armv8_pmuv3_map_event(struct perf_event *event)
{
return __armv8_pmuv3_map_event(event, NULL, NULL);
}
static int armv8_a53_map_event(struct perf_event *event)
{
return __armv8_pmuv3_map_event(event, NULL, &armv8_a53_perf_cache_map);
}
static int armv8_a57_map_event(struct perf_event *event)
{
return __armv8_pmuv3_map_event(event, NULL, &armv8_a57_perf_cache_map);
}
static int armv8_a73_map_event(struct perf_event *event)
{
return __armv8_pmuv3_map_event(event, NULL, &armv8_a73_perf_cache_map);
}
static int armv8_thunder_map_event(struct perf_event *event)
{
return __armv8_pmuv3_map_event(event, NULL,
&armv8_thunder_perf_cache_map);
}
static int armv8_vulcan_map_event(struct perf_event *event)
{
return __armv8_pmuv3_map_event(event, NULL,
&armv8_vulcan_perf_cache_map);
}
struct armv8pmu_probe_info {
struct arm_pmu *pmu;
bool present;
};
static void __armv8pmu_probe_pmu(void *info)
{
struct armv8pmu_probe_info *probe = info;
struct arm_pmu *cpu_pmu = probe->pmu;
u64 dfr0;
u64 pmceid_raw[2];
u32 pmceid[2];
int pmuver;
dfr0 = read_sysreg(id_aa64dfr0_el1);
pmuver = cpuid_feature_extract_unsigned_field(dfr0,
ID_AA64DFR0_PMUVER_SHIFT);
if (pmuver == ID_AA64DFR0_PMUVER_IMP_DEF || pmuver == 0)
return;
cpu_pmu->pmuver = pmuver;
probe->present = true;
/* Read the nb of CNTx counters supported from PMNC */
cpu_pmu->num_events = (armv8pmu_pmcr_read() >> ARMV8_PMU_PMCR_N_SHIFT)
& ARMV8_PMU_PMCR_N_MASK;
/* Add the CPU cycles counter */
cpu_pmu->num_events += 1;
pmceid[0] = pmceid_raw[0] = read_sysreg(pmceid0_el0);
pmceid[1] = pmceid_raw[1] = read_sysreg(pmceid1_el0);
bitmap_from_arr32(cpu_pmu->pmceid_bitmap,
pmceid, ARMV8_PMUV3_MAX_COMMON_EVENTS);
pmceid[0] = pmceid_raw[0] >> 32;
pmceid[1] = pmceid_raw[1] >> 32;
bitmap_from_arr32(cpu_pmu->pmceid_ext_bitmap,
pmceid, ARMV8_PMUV3_MAX_COMMON_EVENTS);
/* store PMMIR_EL1 register for sysfs */
if (pmuver >= ID_AA64DFR0_PMUVER_8_4 && (pmceid_raw[1] & BIT(31)))
cpu_pmu->reg_pmmir = read_cpuid(PMMIR_EL1);
else
cpu_pmu->reg_pmmir = 0;
}
static int armv8pmu_probe_pmu(struct arm_pmu *cpu_pmu)
{
struct armv8pmu_probe_info probe = {
.pmu = cpu_pmu,
.present = false,
};
int ret;
ret = smp_call_function_any(&cpu_pmu->supported_cpus,
__armv8pmu_probe_pmu,
&probe, 1);
if (ret)
return ret;
return probe.present ? 0 : -ENODEV;
}
static int armv8_pmu_init(struct arm_pmu *cpu_pmu, char *name,
int (*map_event)(struct perf_event *event),
const struct attribute_group *events,
const struct attribute_group *format,
const struct attribute_group *caps)
{
int ret = armv8pmu_probe_pmu(cpu_pmu);
if (ret)
return ret;
cpu_pmu->handle_irq = armv8pmu_handle_irq;
cpu_pmu->enable = armv8pmu_enable_event;
cpu_pmu->disable = armv8pmu_disable_event;
cpu_pmu->read_counter = armv8pmu_read_counter;
cpu_pmu->write_counter = armv8pmu_write_counter;
cpu_pmu->get_event_idx = armv8pmu_get_event_idx;
cpu_pmu->clear_event_idx = armv8pmu_clear_event_idx;
cpu_pmu->start = armv8pmu_start;
cpu_pmu->stop = armv8pmu_stop;
cpu_pmu->reset = armv8pmu_reset;
cpu_pmu->set_event_filter = armv8pmu_set_event_filter;
cpu_pmu->filter_match = armv8pmu_filter_match;
cpu_pmu->name = name;
cpu_pmu->map_event = map_event;
cpu_pmu->attr_groups[ARMPMU_ATTR_GROUP_EVENTS] = events ?
events : &armv8_pmuv3_events_attr_group;
cpu_pmu->attr_groups[ARMPMU_ATTR_GROUP_FORMATS] = format ?
format : &armv8_pmuv3_format_attr_group;
cpu_pmu->attr_groups[ARMPMU_ATTR_GROUP_CAPS] = caps ?
caps : &armv8_pmuv3_caps_attr_group;
return 0;
}
static int armv8_pmu_init_nogroups(struct arm_pmu *cpu_pmu, char *name,
int (*map_event)(struct perf_event *event))
{
return armv8_pmu_init(cpu_pmu, name, map_event, NULL, NULL, NULL);
}
static int armv8_pmuv3_init(struct arm_pmu *cpu_pmu)
{
return armv8_pmu_init_nogroups(cpu_pmu, "armv8_pmuv3",
armv8_pmuv3_map_event);
}
static int armv8_a34_pmu_init(struct arm_pmu *cpu_pmu)
{
return armv8_pmu_init_nogroups(cpu_pmu, "armv8_cortex_a34",
armv8_pmuv3_map_event);
}
static int armv8_a35_pmu_init(struct arm_pmu *cpu_pmu)
{
return armv8_pmu_init_nogroups(cpu_pmu, "armv8_cortex_a35",
armv8_a53_map_event);
}
static int armv8_a53_pmu_init(struct arm_pmu *cpu_pmu)
{
return armv8_pmu_init_nogroups(cpu_pmu, "armv8_cortex_a53",
armv8_a53_map_event);
}
static int armv8_a55_pmu_init(struct arm_pmu *cpu_pmu)
{
return armv8_pmu_init_nogroups(cpu_pmu, "armv8_cortex_a55",
armv8_pmuv3_map_event);
}
static int armv8_a57_pmu_init(struct arm_pmu *cpu_pmu)
{
return armv8_pmu_init_nogroups(cpu_pmu, "armv8_cortex_a57",
armv8_a57_map_event);
}
static int armv8_a65_pmu_init(struct arm_pmu *cpu_pmu)
{
return armv8_pmu_init_nogroups(cpu_pmu, "armv8_cortex_a65",
armv8_pmuv3_map_event);
}
static int armv8_a72_pmu_init(struct arm_pmu *cpu_pmu)
{
return armv8_pmu_init_nogroups(cpu_pmu, "armv8_cortex_a72",
armv8_a57_map_event);
}
static int armv8_a73_pmu_init(struct arm_pmu *cpu_pmu)
{
return armv8_pmu_init_nogroups(cpu_pmu, "armv8_cortex_a73",
armv8_a73_map_event);
}
static int armv8_a75_pmu_init(struct arm_pmu *cpu_pmu)
{
return armv8_pmu_init_nogroups(cpu_pmu, "armv8_cortex_a75",
armv8_pmuv3_map_event);
}
static int armv8_a76_pmu_init(struct arm_pmu *cpu_pmu)
{
return armv8_pmu_init_nogroups(cpu_pmu, "armv8_cortex_a76",
armv8_pmuv3_map_event);
}
static int armv8_a77_pmu_init(struct arm_pmu *cpu_pmu)
{
return armv8_pmu_init_nogroups(cpu_pmu, "armv8_cortex_a77",
armv8_pmuv3_map_event);
}
static int armv8_a78_pmu_init(struct arm_pmu *cpu_pmu)
{
return armv8_pmu_init_nogroups(cpu_pmu, "armv8_cortex_a78",
armv8_pmuv3_map_event);
}
static int armv8_e1_pmu_init(struct arm_pmu *cpu_pmu)
{
return armv8_pmu_init_nogroups(cpu_pmu, "armv8_neoverse_e1",
armv8_pmuv3_map_event);
}
static int armv8_n1_pmu_init(struct arm_pmu *cpu_pmu)
{
return armv8_pmu_init_nogroups(cpu_pmu, "armv8_neoverse_n1",
armv8_pmuv3_map_event);
}
static int armv8_thunder_pmu_init(struct arm_pmu *cpu_pmu)
{
return armv8_pmu_init_nogroups(cpu_pmu, "armv8_cavium_thunder",
armv8_thunder_map_event);
}
static int armv8_vulcan_pmu_init(struct arm_pmu *cpu_pmu)
{
return armv8_pmu_init_nogroups(cpu_pmu, "armv8_brcm_vulcan",
armv8_vulcan_map_event);
}
static const struct of_device_id armv8_pmu_of_device_ids[] = {
{.compatible = "arm,armv8-pmuv3", .data = armv8_pmuv3_init},
{.compatible = "arm,cortex-a34-pmu", .data = armv8_a34_pmu_init},
{.compatible = "arm,cortex-a35-pmu", .data = armv8_a35_pmu_init},
{.compatible = "arm,cortex-a53-pmu", .data = armv8_a53_pmu_init},
{.compatible = "arm,cortex-a55-pmu", .data = armv8_a55_pmu_init},
{.compatible = "arm,cortex-a57-pmu", .data = armv8_a57_pmu_init},
{.compatible = "arm,cortex-a65-pmu", .data = armv8_a65_pmu_init},
{.compatible = "arm,cortex-a72-pmu", .data = armv8_a72_pmu_init},
{.compatible = "arm,cortex-a73-pmu", .data = armv8_a73_pmu_init},
{.compatible = "arm,cortex-a75-pmu", .data = armv8_a75_pmu_init},
{.compatible = "arm,cortex-a76-pmu", .data = armv8_a76_pmu_init},
{.compatible = "arm,cortex-a77-pmu", .data = armv8_a77_pmu_init},
{.compatible = "arm,cortex-a78-pmu", .data = armv8_a78_pmu_init},
{.compatible = "arm,neoverse-e1-pmu", .data = armv8_e1_pmu_init},
{.compatible = "arm,neoverse-n1-pmu", .data = armv8_n1_pmu_init},
{.compatible = "cavium,thunder-pmu", .data = armv8_thunder_pmu_init},
{.compatible = "brcm,vulcan-pmu", .data = armv8_vulcan_pmu_init},
{},
};
static int armv8_pmu_device_probe(struct platform_device *pdev)
{
return arm_pmu_device_probe(pdev, armv8_pmu_of_device_ids, NULL);
}
static struct platform_driver armv8_pmu_driver = {
.driver = {
.name = ARMV8_PMU_PDEV_NAME,
.of_match_table = armv8_pmu_of_device_ids,
arm64: perf: set suppress_bind_attrs flag to true The armv8_pmuv3 driver doesn't have a remove function, and when the test 'CONFIG_DEBUG_TEST_DRIVER_REMOVE=y' is enabled, the following Call trace can be seen. [ 1.424287] Failed to register pmu: armv8_pmuv3, reason -17 [ 1.424870] WARNING: CPU: 0 PID: 1 at ../kernel/events/core.c:11771 perf_event_sysfs_init+0x98/0xdc [ 1.425220] Modules linked in: [ 1.425531] CPU: 0 PID: 1 Comm: swapper/0 Tainted: G W 4.19.0-rc7-next-20181012-00003-ge7a97b1ad77b-dirty #35 [ 1.425951] Hardware name: linux,dummy-virt (DT) [ 1.426212] pstate: 80000005 (Nzcv daif -PAN -UAO) [ 1.426458] pc : perf_event_sysfs_init+0x98/0xdc [ 1.426720] lr : perf_event_sysfs_init+0x98/0xdc [ 1.426908] sp : ffff00000804bd50 [ 1.427077] x29: ffff00000804bd50 x28: ffff00000934e078 [ 1.427429] x27: ffff000009546000 x26: 0000000000000007 [ 1.427757] x25: ffff000009280710 x24: 00000000ffffffef [ 1.428086] x23: ffff000009408000 x22: 0000000000000000 [ 1.428415] x21: ffff000009136008 x20: ffff000009408730 [ 1.428744] x19: ffff80007b20b400 x18: 000000000000000a [ 1.429075] x17: 0000000000000000 x16: 0000000000000000 [ 1.429418] x15: 0000000000000400 x14: 2e79726f74636572 [ 1.429748] x13: 696420656d617320 x12: 656874206e692065 [ 1.430060] x11: 6d616e20656d6173 x10: 2065687420687469 [ 1.430335] x9 : ffff00000804bd50 x8 : 206e6f7361657220 [ 1.430610] x7 : 2c3376756d705f38 x6 : ffff00000954d7ce [ 1.430880] x5 : 0000000000000000 x4 : 0000000000000000 [ 1.431226] x3 : 0000000000000000 x2 : ffffffffffffffff [ 1.431554] x1 : 4d151327adc50b00 x0 : 0000000000000000 [ 1.431868] Call trace: [ 1.432102] perf_event_sysfs_init+0x98/0xdc [ 1.432382] do_one_initcall+0x6c/0x1a8 [ 1.432637] kernel_init_freeable+0x1bc/0x280 [ 1.432905] kernel_init+0x18/0x160 [ 1.433115] ret_from_fork+0x10/0x18 [ 1.433297] ---[ end trace 27fd415390eb9883 ]--- Rework to set suppress_bind_attrs flag to avoid removing the device when CONFIG_DEBUG_TEST_DRIVER_REMOVE=y, since there's no real reason to remove the armv8_pmuv3 driver. Cc: Arnd Bergmann <arnd@arndb.de> Co-developed-by: Arnd Bergmann <arnd@arndb.de> Signed-off-by: Anders Roxell <anders.roxell@linaro.org> Signed-off-by: Will Deacon <will.deacon@arm.com>
2018-10-17 15:26:22 +00:00
.suppress_bind_attrs = true,
},
.probe = armv8_pmu_device_probe,
};
static int __init armv8_pmu_driver_init(void)
{
if (acpi_disabled)
Revert "arm64: Enable perf events based hard lockup detector" This reverts commit 367c820ef08082e68df8a3bc12e62393af21e4b5. lockup_detector_init() makes heavy use of per-cpu variables and must be called with preemption disabled. Usually, it's handled early during boot in kernel_init_freeable(), before SMP has been initialised. Since we do not know whether or not our PMU interrupt can be signalled as an NMI until considerably later in the boot process, the Arm PMU driver attempts to re-initialise the lockup detector off the back of a device_initcall(). Unfortunately, this is called from preemptible context and results in the following splat: | BUG: using smp_processor_id() in preemptible [00000000] code: swapper/0/1 | caller is debug_smp_processor_id+0x20/0x2c | CPU: 2 PID: 1 Comm: swapper/0 Not tainted 5.10.0+ #276 | Hardware name: linux,dummy-virt (DT) | Call trace: | dump_backtrace+0x0/0x3c0 | show_stack+0x20/0x6c | dump_stack+0x2f0/0x42c | check_preemption_disabled+0x1cc/0x1dc | debug_smp_processor_id+0x20/0x2c | hardlockup_detector_event_create+0x34/0x18c | hardlockup_detector_perf_init+0x2c/0x134 | watchdog_nmi_probe+0x18/0x24 | lockup_detector_init+0x44/0xa8 | armv8_pmu_driver_init+0x54/0x78 | do_one_initcall+0x184/0x43c | kernel_init_freeable+0x368/0x380 | kernel_init+0x1c/0x1cc | ret_from_fork+0x10/0x30 Rather than bodge this with raw_smp_processor_id() or randomly disabling preemption, simply revert the culprit for now until we figure out how to do this properly. Reported-by: Lecopzer Chen <lecopzer.chen@mediatek.com> Signed-off-by: Will Deacon <will@kernel.org> Acked-by: Mark Rutland <mark.rutland@arm.com> Cc: Sumit Garg <sumit.garg@linaro.org> Cc: Alexandru Elisei <alexandru.elisei@arm.com> Link: https://lore.kernel.org/r/20201221162249.3119-1-lecopzer.chen@mediatek.com Link: https://lore.kernel.org/r/20210112221855.10666-1-will@kernel.org Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
2021-01-12 22:18:55 +00:00
return platform_driver_register(&armv8_pmu_driver);
else
Revert "arm64: Enable perf events based hard lockup detector" This reverts commit 367c820ef08082e68df8a3bc12e62393af21e4b5. lockup_detector_init() makes heavy use of per-cpu variables and must be called with preemption disabled. Usually, it's handled early during boot in kernel_init_freeable(), before SMP has been initialised. Since we do not know whether or not our PMU interrupt can be signalled as an NMI until considerably later in the boot process, the Arm PMU driver attempts to re-initialise the lockup detector off the back of a device_initcall(). Unfortunately, this is called from preemptible context and results in the following splat: | BUG: using smp_processor_id() in preemptible [00000000] code: swapper/0/1 | caller is debug_smp_processor_id+0x20/0x2c | CPU: 2 PID: 1 Comm: swapper/0 Not tainted 5.10.0+ #276 | Hardware name: linux,dummy-virt (DT) | Call trace: | dump_backtrace+0x0/0x3c0 | show_stack+0x20/0x6c | dump_stack+0x2f0/0x42c | check_preemption_disabled+0x1cc/0x1dc | debug_smp_processor_id+0x20/0x2c | hardlockup_detector_event_create+0x34/0x18c | hardlockup_detector_perf_init+0x2c/0x134 | watchdog_nmi_probe+0x18/0x24 | lockup_detector_init+0x44/0xa8 | armv8_pmu_driver_init+0x54/0x78 | do_one_initcall+0x184/0x43c | kernel_init_freeable+0x368/0x380 | kernel_init+0x1c/0x1cc | ret_from_fork+0x10/0x30 Rather than bodge this with raw_smp_processor_id() or randomly disabling preemption, simply revert the culprit for now until we figure out how to do this properly. Reported-by: Lecopzer Chen <lecopzer.chen@mediatek.com> Signed-off-by: Will Deacon <will@kernel.org> Acked-by: Mark Rutland <mark.rutland@arm.com> Cc: Sumit Garg <sumit.garg@linaro.org> Cc: Alexandru Elisei <alexandru.elisei@arm.com> Link: https://lore.kernel.org/r/20201221162249.3119-1-lecopzer.chen@mediatek.com Link: https://lore.kernel.org/r/20210112221855.10666-1-will@kernel.org Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
2021-01-12 22:18:55 +00:00
return arm_pmu_acpi_probe(armv8_pmuv3_init);
}
device_initcall(armv8_pmu_driver_init)
arm64: perf: Add cap_user_time aarch64 It is useful to get the running time of a thread. Doing so in an efficient manner can be important for performance of user applications. Avoiding system calls in `clock_gettime` when handling CLOCK_THREAD_CPUTIME_ID is important. Other clocks are handled in the VDSO, but CLOCK_THREAD_CPUTIME_ID falls back on the system call. CLOCK_THREAD_CPUTIME_ID is not handled in the VDSO since it would have costs associated with maintaining updated user space accessible time offsets. These offsets have to be updated everytime the a thread is scheduled/descheduled. However, for programs regularly checking the running time of a thread, this is a performance improvement. This patch takes a middle ground, and adds support for cap_user_time an optional feature of the perf_event API. This way costs are only incurred when the perf_event api is enabled. This is done the same way as it is in x86. Ultimately this allows calculating the thread running time in userspace on aarch64 as follows (adapted from perf_event_open manpage): u32 seq, time_mult, time_shift; u64 running, count, time_offset, quot, rem, delta; struct perf_event_mmap_page *pc; pc = buf; // buf is the perf event mmaped page as documented in the API. if (pc->cap_usr_time) { do { seq = pc->lock; barrier(); running = pc->time_running; count = readCNTVCT_EL0(); // Read ARM hardware clock. time_offset = pc->time_offset; time_mult = pc->time_mult; time_shift = pc->time_shift; barrier(); } while (pc->lock != seq); quot = (count >> time_shift); rem = count & (((u64)1 << time_shift) - 1); delta = time_offset + quot * time_mult + ((rem * time_mult) >> time_shift); running += delta; // running now has the current nanosecond level thread time. } Summary of changes in the patch: For aarch64 systems, make arch_perf_update_userpage update the timing information stored in the perf_event page. Requiring the following calculations: - Calculate the appropriate time_mult, and time_shift factors to convert ticks to nano seconds for the current clock frequency. - Adjust the mult and shift factors to avoid shift factors of 32 bits. (possibly unnecessary) - The time_offset userspace should apply when doing calculations: negative the current sched time (now), because time_running and time_enabled fields of the perf_event page have just been updated. Toggle bits to appropriate values: - Enable cap_user_time Signed-off-by: Michael O'Farrell <micpof@gmail.com> Signed-off-by: Will Deacon <will.deacon@arm.com>
2018-07-30 20:14:34 +00:00
void arch_perf_update_userpage(struct perf_event *event,
struct perf_event_mmap_page *userpg, u64 now)
{
struct clock_read_data *rd;
unsigned int seq;
u64 ns;
arm64: perf: Add cap_user_time aarch64 It is useful to get the running time of a thread. Doing so in an efficient manner can be important for performance of user applications. Avoiding system calls in `clock_gettime` when handling CLOCK_THREAD_CPUTIME_ID is important. Other clocks are handled in the VDSO, but CLOCK_THREAD_CPUTIME_ID falls back on the system call. CLOCK_THREAD_CPUTIME_ID is not handled in the VDSO since it would have costs associated with maintaining updated user space accessible time offsets. These offsets have to be updated everytime the a thread is scheduled/descheduled. However, for programs regularly checking the running time of a thread, this is a performance improvement. This patch takes a middle ground, and adds support for cap_user_time an optional feature of the perf_event API. This way costs are only incurred when the perf_event api is enabled. This is done the same way as it is in x86. Ultimately this allows calculating the thread running time in userspace on aarch64 as follows (adapted from perf_event_open manpage): u32 seq, time_mult, time_shift; u64 running, count, time_offset, quot, rem, delta; struct perf_event_mmap_page *pc; pc = buf; // buf is the perf event mmaped page as documented in the API. if (pc->cap_usr_time) { do { seq = pc->lock; barrier(); running = pc->time_running; count = readCNTVCT_EL0(); // Read ARM hardware clock. time_offset = pc->time_offset; time_mult = pc->time_mult; time_shift = pc->time_shift; barrier(); } while (pc->lock != seq); quot = (count >> time_shift); rem = count & (((u64)1 << time_shift) - 1); delta = time_offset + quot * time_mult + ((rem * time_mult) >> time_shift); running += delta; // running now has the current nanosecond level thread time. } Summary of changes in the patch: For aarch64 systems, make arch_perf_update_userpage update the timing information stored in the perf_event page. Requiring the following calculations: - Calculate the appropriate time_mult, and time_shift factors to convert ticks to nano seconds for the current clock frequency. - Adjust the mult and shift factors to avoid shift factors of 32 bits. (possibly unnecessary) - The time_offset userspace should apply when doing calculations: negative the current sched time (now), because time_running and time_enabled fields of the perf_event page have just been updated. Toggle bits to appropriate values: - Enable cap_user_time Signed-off-by: Michael O'Farrell <micpof@gmail.com> Signed-off-by: Will Deacon <will.deacon@arm.com>
2018-07-30 20:14:34 +00:00
userpg->cap_user_time = 0;
userpg->cap_user_time_zero = 0;
userpg->cap_user_time_short = 0;
do {
rd = sched_clock_read_begin(&seq);
if (rd->read_sched_clock != arch_timer_read_counter)
return;
userpg->time_mult = rd->mult;
userpg->time_shift = rd->shift;
userpg->time_zero = rd->epoch_ns;
userpg->time_cycles = rd->epoch_cyc;
userpg->time_mask = rd->sched_clock_mask;
/*
* Subtract the cycle base, such that software that
* doesn't know about cap_user_time_short still 'works'
* assuming no wraps.
*/
ns = mul_u64_u32_shr(rd->epoch_cyc, rd->mult, rd->shift);
userpg->time_zero -= ns;
} while (sched_clock_read_retry(seq));
userpg->time_offset = userpg->time_zero - now;
arm64: perf: Add cap_user_time aarch64 It is useful to get the running time of a thread. Doing so in an efficient manner can be important for performance of user applications. Avoiding system calls in `clock_gettime` when handling CLOCK_THREAD_CPUTIME_ID is important. Other clocks are handled in the VDSO, but CLOCK_THREAD_CPUTIME_ID falls back on the system call. CLOCK_THREAD_CPUTIME_ID is not handled in the VDSO since it would have costs associated with maintaining updated user space accessible time offsets. These offsets have to be updated everytime the a thread is scheduled/descheduled. However, for programs regularly checking the running time of a thread, this is a performance improvement. This patch takes a middle ground, and adds support for cap_user_time an optional feature of the perf_event API. This way costs are only incurred when the perf_event api is enabled. This is done the same way as it is in x86. Ultimately this allows calculating the thread running time in userspace on aarch64 as follows (adapted from perf_event_open manpage): u32 seq, time_mult, time_shift; u64 running, count, time_offset, quot, rem, delta; struct perf_event_mmap_page *pc; pc = buf; // buf is the perf event mmaped page as documented in the API. if (pc->cap_usr_time) { do { seq = pc->lock; barrier(); running = pc->time_running; count = readCNTVCT_EL0(); // Read ARM hardware clock. time_offset = pc->time_offset; time_mult = pc->time_mult; time_shift = pc->time_shift; barrier(); } while (pc->lock != seq); quot = (count >> time_shift); rem = count & (((u64)1 << time_shift) - 1); delta = time_offset + quot * time_mult + ((rem * time_mult) >> time_shift); running += delta; // running now has the current nanosecond level thread time. } Summary of changes in the patch: For aarch64 systems, make arch_perf_update_userpage update the timing information stored in the perf_event page. Requiring the following calculations: - Calculate the appropriate time_mult, and time_shift factors to convert ticks to nano seconds for the current clock frequency. - Adjust the mult and shift factors to avoid shift factors of 32 bits. (possibly unnecessary) - The time_offset userspace should apply when doing calculations: negative the current sched time (now), because time_running and time_enabled fields of the perf_event page have just been updated. Toggle bits to appropriate values: - Enable cap_user_time Signed-off-by: Michael O'Farrell <micpof@gmail.com> Signed-off-by: Will Deacon <will.deacon@arm.com>
2018-07-30 20:14:34 +00:00
/*
* time_shift is not expected to be greater than 31 due to
* the original published conversion algorithm shifting a
* 32-bit value (now specifies a 64-bit value) - refer
* perf_event_mmap_page documentation in perf_event.h.
*/
if (userpg->time_shift == 32) {
userpg->time_shift = 31;
arm64: perf: Add cap_user_time aarch64 It is useful to get the running time of a thread. Doing so in an efficient manner can be important for performance of user applications. Avoiding system calls in `clock_gettime` when handling CLOCK_THREAD_CPUTIME_ID is important. Other clocks are handled in the VDSO, but CLOCK_THREAD_CPUTIME_ID falls back on the system call. CLOCK_THREAD_CPUTIME_ID is not handled in the VDSO since it would have costs associated with maintaining updated user space accessible time offsets. These offsets have to be updated everytime the a thread is scheduled/descheduled. However, for programs regularly checking the running time of a thread, this is a performance improvement. This patch takes a middle ground, and adds support for cap_user_time an optional feature of the perf_event API. This way costs are only incurred when the perf_event api is enabled. This is done the same way as it is in x86. Ultimately this allows calculating the thread running time in userspace on aarch64 as follows (adapted from perf_event_open manpage): u32 seq, time_mult, time_shift; u64 running, count, time_offset, quot, rem, delta; struct perf_event_mmap_page *pc; pc = buf; // buf is the perf event mmaped page as documented in the API. if (pc->cap_usr_time) { do { seq = pc->lock; barrier(); running = pc->time_running; count = readCNTVCT_EL0(); // Read ARM hardware clock. time_offset = pc->time_offset; time_mult = pc->time_mult; time_shift = pc->time_shift; barrier(); } while (pc->lock != seq); quot = (count >> time_shift); rem = count & (((u64)1 << time_shift) - 1); delta = time_offset + quot * time_mult + ((rem * time_mult) >> time_shift); running += delta; // running now has the current nanosecond level thread time. } Summary of changes in the patch: For aarch64 systems, make arch_perf_update_userpage update the timing information stored in the perf_event page. Requiring the following calculations: - Calculate the appropriate time_mult, and time_shift factors to convert ticks to nano seconds for the current clock frequency. - Adjust the mult and shift factors to avoid shift factors of 32 bits. (possibly unnecessary) - The time_offset userspace should apply when doing calculations: negative the current sched time (now), because time_running and time_enabled fields of the perf_event page have just been updated. Toggle bits to appropriate values: - Enable cap_user_time Signed-off-by: Michael O'Farrell <micpof@gmail.com> Signed-off-by: Will Deacon <will.deacon@arm.com>
2018-07-30 20:14:34 +00:00
userpg->time_mult >>= 1;
}
/*
* Internal timekeeping for enabled/running/stopped times
* is always computed with the sched_clock.
*/
userpg->cap_user_time = 1;
userpg->cap_user_time_zero = 1;
userpg->cap_user_time_short = 1;
arm64: perf: Add cap_user_time aarch64 It is useful to get the running time of a thread. Doing so in an efficient manner can be important for performance of user applications. Avoiding system calls in `clock_gettime` when handling CLOCK_THREAD_CPUTIME_ID is important. Other clocks are handled in the VDSO, but CLOCK_THREAD_CPUTIME_ID falls back on the system call. CLOCK_THREAD_CPUTIME_ID is not handled in the VDSO since it would have costs associated with maintaining updated user space accessible time offsets. These offsets have to be updated everytime the a thread is scheduled/descheduled. However, for programs regularly checking the running time of a thread, this is a performance improvement. This patch takes a middle ground, and adds support for cap_user_time an optional feature of the perf_event API. This way costs are only incurred when the perf_event api is enabled. This is done the same way as it is in x86. Ultimately this allows calculating the thread running time in userspace on aarch64 as follows (adapted from perf_event_open manpage): u32 seq, time_mult, time_shift; u64 running, count, time_offset, quot, rem, delta; struct perf_event_mmap_page *pc; pc = buf; // buf is the perf event mmaped page as documented in the API. if (pc->cap_usr_time) { do { seq = pc->lock; barrier(); running = pc->time_running; count = readCNTVCT_EL0(); // Read ARM hardware clock. time_offset = pc->time_offset; time_mult = pc->time_mult; time_shift = pc->time_shift; barrier(); } while (pc->lock != seq); quot = (count >> time_shift); rem = count & (((u64)1 << time_shift) - 1); delta = time_offset + quot * time_mult + ((rem * time_mult) >> time_shift); running += delta; // running now has the current nanosecond level thread time. } Summary of changes in the patch: For aarch64 systems, make arch_perf_update_userpage update the timing information stored in the perf_event page. Requiring the following calculations: - Calculate the appropriate time_mult, and time_shift factors to convert ticks to nano seconds for the current clock frequency. - Adjust the mult and shift factors to avoid shift factors of 32 bits. (possibly unnecessary) - The time_offset userspace should apply when doing calculations: negative the current sched time (now), because time_running and time_enabled fields of the perf_event page have just been updated. Toggle bits to appropriate values: - Enable cap_user_time Signed-off-by: Michael O'Farrell <micpof@gmail.com> Signed-off-by: Will Deacon <will.deacon@arm.com>
2018-07-30 20:14:34 +00:00
}