/* * This here uses DEFINE_STATIC_CALL_NULL() to get a static_call defined * from just a typename, as opposed to an actual function.
*/
DEFINE_STATIC_CALL_NULL(x86_pmu_handle_irq, *x86_pmu.handle_irq);
DEFINE_STATIC_CALL_NULL(x86_pmu_disable_all, *x86_pmu.disable_all);
DEFINE_STATIC_CALL_NULL(x86_pmu_enable_all, *x86_pmu.enable_all);
DEFINE_STATIC_CALL_NULL(x86_pmu_enable, *x86_pmu.enable);
DEFINE_STATIC_CALL_NULL(x86_pmu_disable, *x86_pmu.disable);
/* * This one is magic, it will get called even when PMU init fails (because * there is no PMU), in which case it should simply return NULL.
*/
DEFINE_STATIC_CALL_RET0(x86_pmu_guest_get_msrs, *x86_pmu.guest_get_msrs);
/* * Propagate event elapsed time into the generic event. * Can only be executed on the CPU where the event is active. * Returns the delta events processed.
*/
u64 x86_perf_event_update(struct perf_event *event)
{ struct hw_perf_event *hwc = &event->hw; int shift = 64 - x86_pmu.cntval_bits;
u64 prev_raw_count, new_raw_count;
u64 delta;
if (unlikely(!hwc->event_base)) return 0;
/* * Careful: an NMI might modify the previous event value. * * Our tactic to handle this is to first atomically read and * exchange a new raw count - then add that new-prev delta * count to the generic event atomically:
*/
prev_raw_count = local64_read(&hwc->prev_count); do {
new_raw_count = rdpmc(hwc->event_base_rdpmc);
} while (!local64_try_cmpxchg(&hwc->prev_count,
&prev_raw_count, new_raw_count));
/* * Now we have the new raw value and have updated the prev * timestamp already. We can now calculate the elapsed delta * (event-)time and add that to the generic event. * * Careful, not all hw sign-extends above the physical width * of the count.
*/
delta = (new_raw_count << shift) - (prev_raw_count << shift);
delta >>= shift;
/* * Find and validate any extra registers to set up.
*/ staticint x86_pmu_extra_regs(u64 config, struct perf_event *event)
{ struct extra_reg *extra_regs = hybrid(event->pmu, extra_regs); struct hw_perf_event_extra *reg; struct extra_reg *er;
reg = &event->hw.extra_reg;
if (!extra_regs) return 0;
for (er = extra_regs; er->msr; er++) { if (er->event != (config & er->config_mask)) continue; if (event->attr.config1 & ~er->valid_mask) return -EINVAL; /* Check if the extra msrs can be safely accessed*/ if (!er->extra_msr_access) return -ENXIO;
bool check_hw_exists(struct pmu *pmu, unsignedlong *cntr_mask, unsignedlong *fixed_cntr_mask)
{
u64 val, val_fail = -1, val_new= ~0; int i, reg, reg_fail = -1, ret = 0; int bios_fail = 0; int reg_safe = -1;
/* * Check to see if the BIOS enabled any of the counters, if so * complain and bail.
*/
for_each_set_bit(i, cntr_mask, X86_PMC_IDX_MAX) {
reg = x86_pmu_config_addr(i);
ret = rdmsrq_safe(reg, &val); if (ret) goto msr_fail; if (val & ARCH_PERFMON_EVENTSEL_ENABLE) {
bios_fail = 1;
val_fail = val;
reg_fail = reg;
} else {
reg_safe = i;
}
}
if (*(u64 *)fixed_cntr_mask) {
reg = MSR_ARCH_PERFMON_FIXED_CTR_CTRL;
ret = rdmsrq_safe(reg, &val); if (ret) goto msr_fail;
for_each_set_bit(i, fixed_cntr_mask, X86_PMC_IDX_MAX) { if (fixed_counter_disabled(i, pmu)) continue; if (val & (0x03ULL << i*4)) {
bios_fail = 1;
val_fail = val;
reg_fail = reg;
}
}
}
/* * If all the counters are enabled, the below test will always * fail. The tools will also become useless in this scenario. * Just fail and disable the hardware counters.
*/
/* * Read the current value, change it and read it back to see if it * matches, this is needed to detect certain hardware emulators * (qemu/kvm) that don't trap on the MSR access and always return 0s.
*/
reg = x86_pmu_event_addr(reg_safe); if (rdmsrq_safe(reg, &val)) goto msr_fail;
val ^= 0xffffUL;
ret = wrmsrq_safe(reg, val);
ret |= rdmsrq_safe(reg, &val_new); if (ret || val != val_new) goto msr_fail;
/* * We still allow the PMU driver to operate:
*/ if (bios_fail) {
pr_cont("Broken BIOS detected, complain to your hardware vendor.\n");
pr_err(FW_BUG "the BIOS has corrupted hw-PMU resources (MSR %x is %Lx)\n",
reg_fail, val_fail);
}
returntrue;
msr_fail: if (boot_cpu_has(X86_FEATURE_HYPERVISOR)) {
pr_cont("PMU not available due to virtualization, using software events only.\n");
} else {
pr_cont("Broken PMU hardware detected, using software events only.\n");
pr_err("Failed to access perfctr msr (MSR %x is %Lx)\n",
reg, val_new);
}
/* * Check if we can create event of a certain type (that no conflicting events * are present).
*/ int x86_add_exclusive(unsignedint what)
{ int i;
/* * When lbr_pt_coexist we allow PT to coexist with either LBR or BTS. * LBR and BTS are still mutually exclusive.
*/ if (x86_pmu.lbr_pt_coexist && what == x86_lbr_exclusive_pt) goto out;
if (!atomic_inc_not_zero(&x86_pmu.lbr_exclusive[what])) {
mutex_lock(&pmc_reserve_mutex); for (i = 0; i < ARRAY_SIZE(x86_pmu.lbr_exclusive); i++) { if (i != what && atomic_read(&x86_pmu.lbr_exclusive[i])) goto fail_unlock;
}
atomic_inc(&x86_pmu.lbr_exclusive[what]);
mutex_unlock(&pmc_reserve_mutex);
}
/* * The generic map:
*/
config = x86_pmu.event_map(attr->config);
if (config == 0) return -ENOENT;
if (config == -1LL) return -EINVAL;
hwc->config |= config;
return 0;
}
/* * check that branch_sample_type is compatible with * settings needed for precise_ip > 1 which implies * using the LBR to capture ALL taken branches at the * priv levels of the measurement
*/ staticinlineint precise_br_compat(struct perf_event *event)
{
u64 m = event->attr.branch_sample_type;
u64 b = 0;
/* must capture all branches */ if (!(m & PERF_SAMPLE_BRANCH_ANY)) return 0;
m &= PERF_SAMPLE_BRANCH_KERNEL | PERF_SAMPLE_BRANCH_USER;
if (!event->attr.exclude_user)
b |= PERF_SAMPLE_BRANCH_USER;
if (!event->attr.exclude_kernel)
b |= PERF_SAMPLE_BRANCH_KERNEL;
/* * ignore PERF_SAMPLE_BRANCH_HV, not supported on x86
*/
return m == b;
}
int x86_pmu_max_precise(void)
{ int precise = 0;
/* Support for constant skid */ if (x86_pmu.pebs_active && !x86_pmu.pebs_broken) {
precise++;
/* Support for IP fixup */ if (x86_pmu.lbr_nr || x86_pmu.intel_cap.pebs_format >= 2)
precise++;
if (x86_pmu.pebs_prec_dist)
precise++;
} return precise;
}
int x86_pmu_hw_config(struct perf_event *event)
{ if (event->attr.precise_ip) { int precise = x86_pmu_max_precise();
if (event->attr.precise_ip > precise) return -EOPNOTSUPP;
/* There's no sense in having PEBS for non sampling events: */ if (!is_sampling_event(event)) return -EINVAL;
} /* * check that PEBS LBR correction does not conflict with * whatever the user is asking with attr->branch_sample_type
*/ if (event->attr.precise_ip > 1 && x86_pmu.intel_cap.pebs_format < 2) {
u64 *br_type = &event->attr.branch_sample_type;
if (has_branch_stack(event)) { if (!precise_br_compat(event)) return -EOPNOTSUPP;
/* branch_sample_type is compatible */
} else { /* * user did not specify branch_sample_type * * For PEBS fixups, we capture all * the branches at the priv level of the * event.
*/
*br_type = PERF_SAMPLE_BRANCH_ANY;
if (!event->attr.exclude_user)
*br_type |= PERF_SAMPLE_BRANCH_USER;
if (!event->attr.exclude_kernel)
*br_type |= PERF_SAMPLE_BRANCH_KERNEL;
}
}
if (branch_sample_call_stack(event))
event->attach_state |= PERF_ATTACH_TASK_DATA;
/* * Generate PMC IRQs: * (keep 'enabled' bit clear for now)
*/
event->hw.config = ARCH_PERFMON_EVENTSEL_INT;
/* * Count user and OS events unless requested not to
*/ if (!event->attr.exclude_user)
event->hw.config |= ARCH_PERFMON_EVENTSEL_USR; if (!event->attr.exclude_kernel)
event->hw.config |= ARCH_PERFMON_EVENTSEL_OS;
if (event->attr.type == event->pmu->type)
event->hw.config |= x86_pmu_get_event_config(event);
if (is_sampling_event(event) && !event->attr.freq && x86_pmu.limit_period) {
s64 left = event->attr.sample_period;
x86_pmu.limit_period(event, &left); if (left > event->attr.sample_period) return -EINVAL;
}
/* sample_regs_user never support XMM registers */ if (unlikely(event->attr.sample_regs_user & PERF_REG_EXTENDED_MASK)) return -EINVAL; /* * Besides the general purpose registers, XMM registers may * be collected in PEBS on some platforms, e.g. Icelake
*/ if (unlikely(event->attr.sample_regs_intr & PERF_REG_EXTENDED_MASK)) { if (!(event->pmu->capabilities & PERF_PMU_CAP_EXTENDED_REGS)) return -EINVAL;
if (!event->attr.precise_ip) return -EINVAL;
}
return x86_setup_perfctr(event);
}
/* * Setup the hardware configuration for a given attr_type
*/ staticint __x86_pmu_event_init(struct perf_event *event)
{ int err;
if (!x86_pmu_initialized()) return -ENODEV;
err = x86_reserve_hardware(); if (err) return err;
/* * There may be PMI landing after enabled=0. The PMI hitting could be before or * after disable_all. * * If PMI hits before disable_all, the PMU will be disabled in the NMI handler. * It will not be re-enabled in the NMI handler again, because enabled=0. After * handling the NMI, disable_all will be called, which will not change the * state either. If PMI hits after disable_all, the PMU is already disabled * before entering NMI handler. The NMI handler will not change the state * either. * * So either situation is harmless.
*/ staticvoid x86_pmu_disable(struct pmu *pmu)
{ struct cpu_hw_events *cpuc = this_cpu_ptr(&cpu_hw_events);
int is_x86_event(struct perf_event *event)
{ /* * For a non-hybrid platforms, the type of X86 pmu is * always PERF_TYPE_RAW. * For a hybrid platform, the PERF_PMU_CAP_EXTENDED_HW_TYPE * is a unique capability for the X86 PMU. * Use them to detect a X86 event.
*/ if (event->pmu->type == PERF_TYPE_RAW ||
event->pmu->capabilities & PERF_PMU_CAP_EXTENDED_HW_TYPE) returntrue;
/* * All CPUs of the hybrid type have been offline. * The x86_get_pmu() should not be invoked.
*/ if (WARN_ON_ONCE(!cpuc->pmu)) return &pmu;
return cpuc->pmu;
} /* * Event scheduler state: * * Assign events iterating over all events and counters, beginning * with events with least weights first. Keep the current iterator * state in struct sched_state.
*/ struct sched_state { int weight; int event; /* event index */ int counter; /* counter index */ int unassigned; /* number of events to be assigned left */ int nr_gp; /* number of GP counters used */
u64 used;
};
/* Total max is X86_PMC_IDX_MAX, but we are O(n!) limited */ #define SCHED_STATES_MAX 2
struct perf_sched { int max_weight; int max_events; int max_gp; int saved_states; struct event_constraint **constraints; struct sched_state state; struct sched_state saved[SCHED_STATES_MAX];
};
/* * Initialize iterator that runs through all events and counters.
*/ staticvoid perf_sched_init(struct perf_sched *sched, struct event_constraint **constraints, int num, int wmin, int wmax, int gpmax)
{ int idx;
/* this assignment didn't work out */ /* XXX broken vs EVENT_PAIR */
sched->state.used &= ~BIT_ULL(sched->state.counter);
/* try the next one */
sched->state.counter++;
returntrue;
}
/* * Select a counter for the current event to schedule. Return true on * success.
*/ staticbool __perf_sched_find_counter(struct perf_sched *sched)
{ struct event_constraint *c; int idx;
if (!sched->state.unassigned) returnfalse;
if (sched->state.event >= sched->max_events) returnfalse;
/* Grab the first unused counter starting with idx */
idx = sched->state.counter;
for_each_set_bit_from(idx, c->idxmsk, INTEL_PMC_IDX_FIXED) {
u64 mask = BIT_ULL(idx);
if (c->flags & PERF_X86_EVENT_PAIR)
mask |= mask << 1;
if (sched->state.used & mask) continue;
if (sched->state.nr_gp++ >= sched->max_gp) returnfalse;
sched->state.used |= mask; goto done;
}
returnfalse;
done:
sched->state.counter = idx;
if (c->overlap)
perf_sched_save_state(sched);
returntrue;
}
staticbool perf_sched_find_counter(struct perf_sched *sched)
{ while (!__perf_sched_find_counter(sched)) { if (!perf_sched_restore_state(sched)) returnfalse;
}
returntrue;
}
/* * Go through all unassigned events and find the next one to schedule. * Take events with the least weight first. Return true on success.
*/ staticbool perf_sched_next_event(struct perf_sched *sched)
{ struct event_constraint *c;
if (!sched->state.unassigned || !--sched->state.unassigned) returnfalse;
do { /* next event */
sched->state.event++; if (sched->state.event >= sched->max_events) { /* next weight */
sched->state.event = 0;
sched->state.weight++; if (sched->state.weight > sched->max_weight) returnfalse;
}
c = sched->constraints[sched->state.event];
} while (c->weight != sched->state.weight);
sched->state.counter = 0; /* start with first counter */
returntrue;
}
/* * Assign a counter for each event.
*/ int perf_assign_events(struct event_constraint **constraints, int n, int wmin, int wmax, int gpmax, int *assign)
{ struct perf_sched sched;
perf_sched_init(&sched, constraints, n, wmin, wmax, gpmax);
do { if (!perf_sched_find_counter(&sched)) break; /* failed */ if (assign)
assign[sched.state.event] = sched.state.counter;
} while (perf_sched_next_event(&sched));
int x86_schedule_events(struct cpu_hw_events *cpuc, int n, int *assign)
{ struct event_constraint *c; struct perf_event *e; int n0, i, wmin, wmax, unsched = 0; struct hw_perf_event *hwc;
u64 used_mask = 0;
/* * Compute the number of events already present; see x86_pmu_add(), * validate_group() and x86_pmu_commit_txn(). For the former two * cpuc->n_events hasn't been updated yet, while for the latter * cpuc->n_txn contains the number of events added in the current * transaction.
*/
n0 = cpuc->n_events; if (cpuc->txn_flags & PERF_PMU_TXN_ADD)
n0 -= cpuc->n_txn;
static_call_cond(x86_pmu_start_scheduling)(cpuc);
for (i = 0, wmin = X86_PMC_IDX_MAX, wmax = 0; i < n; i++) {
c = cpuc->event_constraint[i];
/* * Previously scheduled events should have a cached constraint, * while new events should not have one.
*/
WARN_ON_ONCE((c && i >= n0) || (!c && i < n0));
/* * Request constraints for new events; or for those events that * have a dynamic constraint -- for those the constraint can * change due to external factors (sibling state, allow_tfa).
*/ if (!c || (c->flags & PERF_X86_EVENT_DYNAMIC)) {
c = static_call(x86_pmu_get_event_constraints)(cpuc, i, cpuc->event_list[i]);
cpuc->event_constraint[i] = c;
}
/* * fastpath, try to reuse previous register
*/ for (i = 0; i < n; i++) {
u64 mask;
hwc = &cpuc->event_list[i]->hw;
c = cpuc->event_constraint[i];
/* never assigned */ if (hwc->idx == -1) break;
/* constraint still honored */ if (!test_bit(hwc->idx, c->idxmsk)) break;
mask = BIT_ULL(hwc->idx); if (is_counter_pair(hwc))
mask |= mask << 1;
/* not already used */ if (used_mask & mask) break;
used_mask |= mask;
if (assign)
assign[i] = hwc->idx;
}
/* slow path */ if (i != n) { int gpmax = x86_pmu_max_num_counters(cpuc->pmu);
/* * Do not allow scheduling of more than half the available * generic counters. * * This helps avoid counter starvation of sibling thread by * ensuring at most half the counters cannot be in exclusive * mode. There is no designated counters for the limits. Any * N/2 counters can be used. This helps with events with * specific counter constraints.
*/ if (is_ht_workaround_enabled() && !cpuc->is_fake &&
READ_ONCE(cpuc->excl_cntrs->exclusive_present))
gpmax /= 2;
/* * Reduce the amount of available counters to allow fitting * the extra Merge events needed by large increment events.
*/ if (x86_pmu.flags & PMU_FL_PAIR) {
gpmax -= cpuc->n_pair;
WARN_ON(gpmax <= 0);
}
unsched = perf_assign_events(cpuc->event_constraint, n, wmin,
wmax, gpmax, assign);
}
/* * In case of success (unsched = 0), mark events as committed, * so we do not put_constraint() in case new events are added * and fail to be scheduled * * We invoke the lower level commit callback to lock the resource * * We do not need to do all of this in case we are called to * validate an event group (assign == NULL)
*/ if (!unsched && assign) { for (i = 0; i < n; i++)
static_call_cond(x86_pmu_commit_scheduling)(cpuc, i, assign[i]);
} else { for (i = n0; i < n; i++) {
e = cpuc->event_list[i];
/* * release events that failed scheduling
*/
static_call_cond(x86_pmu_put_event_constraints)(cpuc, e);
cpuc->event_constraint[i] = NULL;
}
}
static_call_cond(x86_pmu_stop_scheduling)(cpuc);
return unsched ? -EINVAL : 0;
}
staticint add_nr_metric_event(struct cpu_hw_events *cpuc, struct perf_event *event)
{ if (is_metric_event(event)) { if (cpuc->n_metric == INTEL_TD_METRIC_NUM) return -EINVAL;
cpuc->n_metric++;
cpuc->n_txn_metric++;
}
staticint collect_event(struct cpu_hw_events *cpuc, struct perf_event *event, int max_count, int n)
{ union perf_capabilities intel_cap = hybrid(cpuc->pmu, intel_cap);
if (intel_cap.perf_metrics && add_nr_metric_event(cpuc, event)) return -EINVAL;
if (n >= max_count + cpuc->n_metric) return -EINVAL;
cpuc->event_list[n] = event; if (is_counter_pair(&event->hw)) {
cpuc->n_pair++;
cpuc->n_txn_pair++;
}
return 0;
}
/* * dogrp: true if must collect siblings events (group) * returns total number of events and error code
*/ staticint collect_events(struct cpu_hw_events *cpuc, struct perf_event *leader, bool dogrp)
{ struct perf_event *event; int n, max_count;
/* current number of events already accepted */
n = cpuc->n_events; if (!cpuc->n_events)
cpuc->pebs_output = 0;
if (!cpuc->is_fake && leader->attr.precise_ip) { /* * For PEBS->PT, if !aux_event, the group leader (PT) went * away, the group was broken down and this singleton event * can't schedule any more.
*/ if (is_pebs_pt(leader) && !leader->aux_event) return -EINVAL;
/* * pebs_output: 0: no PEBS so far, 1: PT, 2: DS
*/ if (cpuc->pebs_output &&
cpuc->pebs_output != is_pebs_pt(leader) + 1) return -EINVAL;
cpuc->pebs_output = is_pebs_pt(leader) + 1;
}
if (is_x86_event(leader)) { if (collect_event(cpuc, leader, max_count, n)) return -EINVAL;
n++;
}
if (!dogrp) return n;
for_each_sibling_event(event, leader) { if (!is_x86_event(event) || event->state <= PERF_EVENT_STATE_OFF) continue;
if (collect_event(cpuc, event, max_count, n)) return -EINVAL;
n++;
} return n;
}
staticinlinevoid x86_assign_hw_event(struct perf_event *event, struct cpu_hw_events *cpuc, int i)
{ struct hw_perf_event *hwc = &event->hw; int idx;
/** * x86_perf_rdpmc_index - Return PMC counter used for event * @event: the perf_event to which the PMC counter was assigned * * The counter assigned to this performance event may change if interrupts * are enabled. This counter should thus never be used while interrupts are * enabled. Before this function is used to obtain the assigned counter the * event should be checked for validity using, for example, * perf_event_read_local(), within the same interrupt disabled section in * which this counter is planned to be used. * * Return: The index of the performance monitoring counter assigned to * @perf_event.
*/ int x86_perf_rdpmc_index(struct perf_event *event)
{
lockdep_assert_irqs_disabled();
if (cpuc->n_added) { int n_running = cpuc->n_events - cpuc->n_added;
/* * The late setup (after counters are scheduled) * is required for some cases, e.g., PEBS counters * snapshotting. Because an accurate counter index * is needed.
*/
static_call_cond(x86_pmu_late_setup)();
/* * apply assignment obtained either from * hw_perf_group_sched_in() or x86_pmu_enable() * * step1: save events moving to new counters
*/ for (i = 0; i < n_running; i++) {
event = cpuc->event_list[i];
hwc = &event->hw;
/* * we can avoid reprogramming counter if: * - assigned same counter as last time * - running on same CPU as last time * - no other event has used the counter since
*/ if (hwc->idx == -1 ||
match_prev_assignment(hwc, cpuc, i)) continue;
/* * Ensure we don't accidentally enable a stopped * counter simply because we rescheduled.
*/ if (hwc->state & PERF_HES_STOPPED)
hwc->state |= PERF_HES_ARCH;
x86_pmu_stop(event, PERF_EF_UPDATE);
}
/* * step2: reprogram moved events into new counters
*/ for (i = 0; i < cpuc->n_events; i++) {
event = cpuc->event_list[i];
hwc = &event->hw;
if (!match_prev_assignment(hwc, cpuc, i))
x86_assign_hw_event(event, cpuc, i); elseif (i < n_running) continue;
if (hwc->state & PERF_HES_ARCH) continue;
/* * if cpuc->enabled = 0, then no wrmsr as * per x86_pmu_enable_event()
*/
x86_pmu_start(event, PERF_EF_RELOAD);
}
cpuc->n_added = 0;
perf_events_lapic_init();
}
/* * Set the next IRQ period, based on the hwc->period_left value. * To be called with the event disabled in hw:
*/ int x86_perf_event_set_period(struct perf_event *event)
{ struct hw_perf_event *hwc = &event->hw;
s64 left = local64_read(&hwc->period_left);
s64 period = hwc->sample_period; int ret = 0, idx = hwc->idx;
if (unlikely(!hwc->event_base)) return 0;
/* * If we are way outside a reasonable range then just skip forward:
*/ if (unlikely(left <= -period)) {
left = period;
local64_set(&hwc->period_left, left);
hwc->last_period = period;
ret = 1;
}
if (unlikely(left <= 0)) {
left += period;
local64_set(&hwc->period_left, left);
hwc->last_period = period;
ret = 1;
} /* * Quirk: certain CPUs dont like it if just 1 hw_event is left:
*/ if (unlikely(left < 2))
left = 2;
if (left > x86_pmu.max_period)
left = x86_pmu.max_period;
/* * Sign extend the Merge event counter's upper 16 bits since * we currently declare a 48-bit counter width
*/ if (is_counter_pair(hwc))
wrmsrq(x86_pmu_event_addr(idx + 1), 0xffff);
perf_event_update_userpage(event);
return ret;
}
void x86_pmu_enable_event(struct perf_event *event)
{ if (__this_cpu_read(cpu_hw_events.enabled))
__x86_pmu_enable_event(&event->hw,
ARCH_PERFMON_EVENTSEL_ENABLE);
}
/* * Add a single event to the PMU. * * The event is added to the group of enabled events * but only if it can be scheduled with existing events.
*/ staticint x86_pmu_add(struct perf_event *event, int flags)
{ struct cpu_hw_events *cpuc = this_cpu_ptr(&cpu_hw_events); struct hw_perf_event *hwc; int assign[X86_PMC_IDX_MAX]; int n, n0, ret;
hwc = &event->hw;
n0 = cpuc->n_events;
ret = n = collect_events(cpuc, event, false); if (ret < 0) goto out;
/* * If group events scheduling transaction was started, * skip the schedulability test here, it will be performed * at commit time (->commit_txn) as a whole. * * If commit fails, we'll call ->del() on all events * for which ->add() was called.
*/ if (cpuc->txn_flags & PERF_PMU_TXN_ADD) goto done_collect;
ret = static_call(x86_pmu_schedule_events)(cpuc, n, assign); if (ret) goto out; /* * copy new assignment, now we know it is possible * will be used by hw_perf_enable()
*/
memcpy(cpuc->assign, assign, n*sizeof(int));
done_collect: /* * Commit the collect_events() state. See x86_pmu_del() and * x86_pmu_*_txn().
*/
cpuc->n_events = n;
cpuc->n_added += n - n0;
cpuc->n_txn += n - n0;
/* * This is before x86_pmu_enable() will call x86_pmu_start(), * so we enable LBRs before an event needs them etc..
*/
static_call_cond(x86_pmu_add)(event);
ret = 0;
out: return ret;
}
staticvoid x86_pmu_start(struct perf_event *event, int flags)
{ struct cpu_hw_events *cpuc = this_cpu_ptr(&cpu_hw_events); int idx = event->hw.idx;
if (WARN_ON_ONCE(!(event->hw.state & PERF_HES_STOPPED))) return;
if (WARN_ON_ONCE(idx == -1)) return;
if (flags & PERF_EF_RELOAD) {
WARN_ON_ONCE(!(event->hw.state & PERF_HES_UPTODATE));
static_call(x86_pmu_set_period)(event);
}
if ((flags & PERF_EF_UPDATE) && !(hwc->state & PERF_HES_UPTODATE)) { /* * Drain the remaining delta count out of a event * that we are disabling:
*/
static_call(x86_pmu_update)(event);
hwc->state |= PERF_HES_UPTODATE;
}
}
staticvoid x86_pmu_del(struct perf_event *event, int flags)
{ struct cpu_hw_events *cpuc = this_cpu_ptr(&cpu_hw_events); union perf_capabilities intel_cap = hybrid(cpuc->pmu, intel_cap); int i;
/* * If we're called during a txn, we only need to undo x86_pmu.add. * The events never got scheduled and ->cancel_txn will truncate * the event_list. * * XXX assumes any ->del() called during a TXN will only be on * an event added during that same TXN.
*/ if (cpuc->txn_flags & PERF_PMU_TXN_ADD) goto do_del;
__set_bit(event->hw.idx, cpuc->dirty);
/* * Not a TXN, therefore cleanup properly.
*/
x86_pmu_stop(event, PERF_EF_UPDATE);
for (i = 0; i < cpuc->n_events; i++) { if (event == cpuc->event_list[i]) break;
}
if (WARN_ON_ONCE(i == cpuc->n_events)) /* called ->del() without ->add() ? */ return;
/* If we have a newly added event; make sure to decrease n_added. */ if (i >= cpuc->n_events - cpuc->n_added)
--cpuc->n_added;
/* * Some chipsets need to unmask the LVTPC in a particular spot * inside the nmi handler. As a result, the unmasking was pushed * into all the nmi handlers. * * This generic handler doesn't seem to have any issues where the * unmasking occurs so it was left at the top.
*/
apic_write(APIC_LVTPC, APIC_DM_NMI);
for_each_set_bit(idx, x86_pmu.cntr_mask, X86_PMC_IDX_MAX) { if (!test_bit(idx, cpuc->active_mask)) continue;
/* * All PMUs/events that share this PMI handler should make sure to * increment active_events for their events.
*/ if (!atomic_read(&active_events)) return NMI_DONE;
start_clock = sched_clock();
ret = static_call(x86_pmu_handle_irq)(regs);
finish_clock = sched_clock();
for (i = 0 ; i < X86_PERF_KFREE_MAX; i++) {
kfree(cpuc->kfree_on_online[i]);
cpuc->kfree_on_online[i] = NULL;
} return 0;
}
staticint x86_pmu_starting_cpu(unsignedint cpu)
{ if (x86_pmu.cpu_starting)
x86_pmu.cpu_starting(cpu); return 0;
}
staticint x86_pmu_dying_cpu(unsignedint cpu)
{ if (x86_pmu.cpu_dying)
x86_pmu.cpu_dying(cpu); return 0;
}
staticvoid __init pmu_check_apic(void)
{ if (boot_cpu_has(X86_FEATURE_APIC)) return;
x86_pmu.apic = 0;
pr_info("no APIC, boot with the \"lapic\" boot parameter to force-enable it.\n");
pr_info("no hardware sampling interrupt available.\n");
/* * If we have a PMU initialized but no APIC * interrupts, we cannot sample hardware * events (user-space has to fall back and * sample via a hrtimer based software event):
*/
pmu.capabilities |= PERF_PMU_CAP_NO_INTERRUPT;
/* * Report conditional events depending on Hyper-Threading. * * This is overly conservative as usually the HT special * handling is not needed if the other CPU thread is idle. * * Note this does not (and cannot) handle the case when thread * siblings are invisible, for example with virtualization * if they are owned by some other guest. The user tool * has to re-read when a thread sibling gets onlined later.
*/ return sprintf(page, "%s",
topology_max_smt_threads() > 1 ?
pmu_attr->event_str_ht :
pmu_attr->event_str_noht);
}
if (hweight64(pmu_attr->pmu_type) == 1) return sprintf(page, "%s", pmu_attr->event_str);
/* * Hybrid PMUs may support the same event name, but with different * event encoding, e.g., the mem-loads event on an Atom PMU has * different event encoding from a Core PMU. * * The event_str includes all event encodings. Each event encoding * is divided by ";". The order of the event encodings must follow * the order of the hybrid PMU index.
*/
pmu = container_of(dev_get_drvdata(dev), struct x86_hybrid_pmu, pmu);
str = pmu_attr->event_str; for (i = 0; i < x86_pmu.num_hybrid_pmus; i++) { if (!(x86_pmu.hybrid_pmu[i].pmu_type & pmu_attr->pmu_type)) continue; if (x86_pmu.hybrid_pmu[i].pmu_type & pmu->pmu_type) {
next_str = strchr(str, ';'); if (next_str) return snprintf(page, next_str - str + 1, "%s", str); else return sprintf(page, "%s", str);
}
str = strchr(str, ';');
str++;
}
if (!x86_pmu.events_sysfs_show)
x86_pmu_events_group.attrs = &empty_attrs;
pmu.attr_update = x86_pmu.attr_update;
if (!is_hybrid())
x86_pmu_show_pmu_cap(NULL);
if (!x86_pmu.read)
x86_pmu.read = _x86_pmu_read;
if (!x86_pmu.guest_get_msrs)
x86_pmu.guest_get_msrs = (void *)&__static_call_return0;
if (!x86_pmu.set_period)
x86_pmu.set_period = x86_perf_event_set_period;
if (!x86_pmu.update)
x86_pmu.update = x86_perf_event_update;
x86_pmu_static_call_update();
/* * Install callbacks. Core will call them for each online * cpu.
*/
err = cpuhp_setup_state(CPUHP_PERF_X86_PREPARE, "perf/x86:prepare",
x86_pmu_prepare_cpu, x86_pmu_dead_cpu); if (err) return err;
err = cpuhp_setup_state(CPUHP_AP_PERF_X86_STARTING, "perf/x86:starting", x86_pmu_starting_cpu,
x86_pmu_dying_cpu); if (err) goto out;
err = cpuhp_setup_state(CPUHP_AP_PERF_X86_ONLINE, "perf/x86:online",
x86_pmu_online_cpu, NULL); if (err) goto out1;
if (!is_hybrid()) {
err = perf_pmu_register(&pmu, "cpu", PERF_TYPE_RAW); if (err) goto out2;
} else { struct x86_hybrid_pmu *hybrid_pmu; int i, j;
for (i = 0; i < x86_pmu.num_hybrid_pmus; i++) {
hybrid_pmu = &x86_pmu.hybrid_pmu[i];
/* * Start group events scheduling transaction * Set the flag to make pmu::enable() not perform the * schedulability test, it will be performed at commit time * * We only support PERF_PMU_TXN_ADD transactions. Save the * transaction flags but otherwise ignore non-PERF_PMU_TXN_ADD * transactions.
*/ staticvoid x86_pmu_start_txn(struct pmu *pmu, unsignedint txn_flags)
{ struct cpu_hw_events *cpuc = this_cpu_ptr(&cpu_hw_events);
WARN_ON_ONCE(cpuc->txn_flags); /* txn already in flight */
cpuc->txn_flags = txn_flags; if (txn_flags & ~PERF_PMU_TXN_ADD) return;
/* * Truncate collected array by the number of events added in this * transaction. See x86_pmu_add() and x86_pmu_*_txn().
*/
__this_cpu_sub(cpu_hw_events.n_added, __this_cpu_read(cpu_hw_events.n_txn));
__this_cpu_sub(cpu_hw_events.n_events, __this_cpu_read(cpu_hw_events.n_txn));
__this_cpu_sub(cpu_hw_events.n_pair, __this_cpu_read(cpu_hw_events.n_txn_pair));
__this_cpu_sub(cpu_hw_events.n_metric, __this_cpu_read(cpu_hw_events.n_txn_metric));
perf_pmu_enable(pmu);
}
/* * Commit group events scheduling transaction * Perform the group schedulability test as a whole * Return 0 if success * * Does not cancel the transaction on failure; expects the caller to do this.
*/ staticint x86_pmu_commit_txn(struct pmu *pmu)
{ struct cpu_hw_events *cpuc = this_cpu_ptr(&cpu_hw_events); int assign[X86_PMC_IDX_MAX]; int n, ret;
WARN_ON_ONCE(!cpuc->txn_flags); /* no txn in flight */
ret = static_call(x86_pmu_schedule_events)(cpuc, n, assign); if (ret) return ret;
/* * copy new assignment, now we know it is possible * will be used by hw_perf_enable()
*/
memcpy(cpuc->assign, assign, n*sizeof(int));
cpuc->txn_flags = 0;
perf_pmu_enable(pmu); return 0;
} /* * a fake_cpuc is used to validate event groups. Due to * the extra reg logic, we need to also allocate a fake * per_core and per_cpu structure. Otherwise, group events * using extra reg may conflict without the kernel being * able to catch this when the last event gets added to * the group.
*/ staticvoid free_fake_cpuc(struct cpu_hw_events *cpuc)
{
intel_cpuc_finish(cpuc);
kfree(cpuc);
}
/* * validate that we can schedule this event
*/ staticint validate_event(struct perf_event *event)
{ struct cpu_hw_events *fake_cpuc; struct event_constraint *c; int ret = 0;
fake_cpuc = allocate_fake_cpuc(event->pmu); if (IS_ERR(fake_cpuc)) return PTR_ERR(fake_cpuc);
c = x86_pmu.get_event_constraints(fake_cpuc, 0, event);
if (!c || !c->weight)
ret = -EINVAL;
if (x86_pmu.put_event_constraints)
x86_pmu.put_event_constraints(fake_cpuc, event);
free_fake_cpuc(fake_cpuc);
return ret;
}
/* * validate a single event group * * validation include: * - check events are compatible which each other * - events do not compete for the same counter * - number of events <= number of counters * * validation ensures the group can be loaded onto the * PMU if it was the only group available.
*/ staticint validate_group(struct perf_event *event)
{ struct perf_event *leader = event->group_leader; struct cpu_hw_events *fake_cpuc; int ret = -EINVAL, n;
/* * Reject events from different hybrid PMUs.
*/ if (is_hybrid()) { struct perf_event *sibling; struct pmu *pmu = NULL;
if (is_x86_event(leader))
pmu = leader->pmu;
for_each_sibling_event(sibling, leader) { if (!is_x86_event(sibling)) continue; if (!pmu)
pmu = sibling->pmu; elseif (pmu != sibling->pmu) return ret;
}
}
fake_cpuc = allocate_fake_cpuc(event->pmu); if (IS_ERR(fake_cpuc)) return PTR_ERR(fake_cpuc); /* * the event is not yet connected with its * siblings therefore we must first collect * existing siblings, then add the new event * before we can simulate the scheduling
*/
n = collect_events(fake_cpuc, leader, true); if (n < 0) goto out;
fake_cpuc->n_events = n;
n = collect_events(fake_cpuc, event, false); if (n < 0) goto out;
fake_cpuc->n_events = 0;
ret = x86_pmu.schedule_events(fake_cpuc, n, NULL);
/* * This function relies on not being called concurrently in two * tasks in the same mm. Otherwise one task could observe * perf_rdpmc_allowed > 1 and return all the way back to * userspace with CR4.PCE clear while another task is still * doing on_each_cpu_mask() to propagate CR4.PCE. * * For now, this can't happen because all callers hold mmap_lock * for write. If this changes, we'll need a different solution.
*/
mmap_assert_write_locked(mm);
if (atomic_inc_return(&mm->context.perf_rdpmc_allowed) == 1)
on_each_cpu_mask(mm_cpumask(mm), cr4_update_pce, NULL, 1);
}
Die Informationen auf dieser Webseite wurden
nach bestem Wissen sorgfältig zusammengestellt. Es wird jedoch weder Vollständigkeit, noch Richtigkeit,
noch Qualität der bereit gestellten Informationen zugesichert.
Bemerkung:
Die farbliche Syntaxdarstellung und die Messung sind noch experimentell.