| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-ID: <20200818135805.338001015@infradead.org>
Date: Tue, 18 Aug 2020 15:57:53 +0200
From: Peter Zijlstra <peterz@...radead.org>
To: x86@...nel.org
Cc: peterz@...radead.org, linux-kernel@...r.kernel.org,
rostedt@...dmis.org, mhiramat@...nel.org, bristot@...hat.com,
jbaron@...mai.com, torvalds@...ux-foundation.org,
tglx@...utronix.de, mingo@...nel.org, namit@...are.com,
hpa@...or.com, luto@...nel.org, ard.biesheuvel@...aro.org,
jpoimboe@...hat.com, pbonzini@...hat.com,
mathieu.desnoyers@...icios.com, linux@...musvillemoes.dk
Subject: [PATCH v7 18/18] x86/perf, static_call: Optimize x86_pmu methods
Replace many of the indirect calls with static_call().
The average PMI time, as measured by perf_sample_event_took()*:
PRE: 3283.03 [ns]
POST: 3145.12 [ns]
Which is a ~138 [ns] win per PMI, or a ~4.2% decrease.
[*] on an IVB-EP, using: 'perf record -a -e cycles -- make O=defconfig-build/ -j80'
Signed-off-by: Peter Zijlstra (Intel) <peterz@...radead.org>
---
arch/x86/events/core.c | 140 +++++++++++++++++++++++++++++++++++--------------
1 file changed, 100 insertions(+), 40 deletions(-)
--- a/arch/x86/events/core.c
+++ b/arch/x86/events/core.c
@@ -28,6 +28,7 @@
#include <linux/bitops.h>
#include <linux/device.h>
#include <linux/nospec.h>
+#include <linux/static_call.h>
#include <asm/apic.h>
#include <asm/stacktrace.h>
@@ -52,6 +53,34 @@ DEFINE_PER_CPU(struct cpu_hw_events, cpu
DEFINE_STATIC_KEY_FALSE(rdpmc_never_available_key);
DEFINE_STATIC_KEY_FALSE(rdpmc_always_available_key);
+/*
+ * 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);
+
+DEFINE_STATIC_CALL_NULL(x86_pmu_add, *x86_pmu.add);
+DEFINE_STATIC_CALL_NULL(x86_pmu_del, *x86_pmu.del);
+DEFINE_STATIC_CALL_NULL(x86_pmu_read, *x86_pmu.read);
+
+DEFINE_STATIC_CALL_NULL(x86_pmu_schedule_events, *x86_pmu.schedule_events);
+DEFINE_STATIC_CALL_NULL(x86_pmu_get_event_constraints, *x86_pmu.get_event_constraints);
+DEFINE_STATIC_CALL_NULL(x86_pmu_put_event_constraints, *x86_pmu.put_event_constraints);
+
+DEFINE_STATIC_CALL_NULL(x86_pmu_start_scheduling, *x86_pmu.start_scheduling);
+DEFINE_STATIC_CALL_NULL(x86_pmu_commit_scheduling, *x86_pmu.commit_scheduling);
+DEFINE_STATIC_CALL_NULL(x86_pmu_stop_scheduling, *x86_pmu.stop_scheduling);
+
+DEFINE_STATIC_CALL_NULL(x86_pmu_sched_task, *x86_pmu.sched_task);
+DEFINE_STATIC_CALL_NULL(x86_pmu_swap_task_ctx, *x86_pmu.swap_task_ctx);
+
+DEFINE_STATIC_CALL_NULL(x86_pmu_drain_pebs, *x86_pmu.drain_pebs);
+DEFINE_STATIC_CALL_NULL(x86_pmu_pebs_aliases, *x86_pmu.pebs_aliases);
+
u64 __read_mostly hw_cache_event_ids
[PERF_COUNT_HW_CACHE_MAX]
[PERF_COUNT_HW_CACHE_OP_MAX]
@@ -660,7 +695,7 @@ static void x86_pmu_disable(struct pmu *
cpuc->enabled = 0;
barrier();
- x86_pmu.disable_all();
+ static_call(x86_pmu_disable_all)();
}
void x86_pmu_enable_all(int added)
@@ -907,8 +942,7 @@ int x86_schedule_events(struct cpu_hw_ev
if (cpuc->txn_flags & PERF_PMU_TXN_ADD)
n0 -= cpuc->n_txn;
- if (x86_pmu.start_scheduling)
- x86_pmu.start_scheduling(cpuc);
+ 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];
@@ -925,7 +959,7 @@ int x86_schedule_events(struct cpu_hw_ev
* change due to external factors (sibling state, allow_tfa).
*/
if (!c || (c->flags & PERF_X86_EVENT_DYNAMIC)) {
- c = x86_pmu.get_event_constraints(cpuc, i, cpuc->event_list[i]);
+ c = static_call(x86_pmu_get_event_constraints)(cpuc, i, cpuc->event_list[i]);
cpuc->event_constraint[i] = c;
}
@@ -1008,8 +1042,7 @@ int x86_schedule_events(struct cpu_hw_ev
if (!unsched && assign) {
for (i = 0; i < n; i++) {
e = cpuc->event_list[i];
- if (x86_pmu.commit_scheduling)
- x86_pmu.commit_scheduling(cpuc, i, assign[i]);
+ static_call_cond(x86_pmu_commit_scheduling)(cpuc, i, assign[i]);
}
} else {
for (i = n0; i < n; i++) {
@@ -1018,15 +1051,13 @@ int x86_schedule_events(struct cpu_hw_ev
/*
* release events that failed scheduling
*/
- if (x86_pmu.put_event_constraints)
- x86_pmu.put_event_constraints(cpuc, e);
+ static_call_cond(x86_pmu_put_event_constraints)(cpuc, e);
cpuc->event_constraint[i] = NULL;
}
}
- if (x86_pmu.stop_scheduling)
- x86_pmu.stop_scheduling(cpuc);
+ static_call_cond(x86_pmu_stop_scheduling)(cpuc);
return unsched ? -EINVAL : 0;
}
@@ -1217,7 +1248,7 @@ static void x86_pmu_enable(struct pmu *p
cpuc->enabled = 1;
barrier();
- x86_pmu.enable_all(added);
+ static_call(x86_pmu_enable_all)(added);
}
static DEFINE_PER_CPU(u64 [X86_PMC_IDX_MAX], pmc_prev_left);
@@ -1338,7 +1369,7 @@ static int x86_pmu_add(struct perf_event
if (cpuc->txn_flags & PERF_PMU_TXN_ADD)
goto done_collect;
- ret = x86_pmu.schedule_events(cpuc, n, assign);
+ ret = static_call(x86_pmu_schedule_events)(cpuc, n, assign);
if (ret)
goto out;
/*
@@ -1356,13 +1387,11 @@ static int x86_pmu_add(struct perf_event
cpuc->n_added += n - n0;
cpuc->n_txn += n - n0;
- if (x86_pmu.add) {
- /*
- * This is before x86_pmu_enable() will call x86_pmu_start(),
- * so we enable LBRs before an event needs them etc..
- */
- x86_pmu.add(event);
- }
+ /*
+ * 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:
@@ -1390,7 +1419,7 @@ static void x86_pmu_start(struct perf_ev
cpuc->events[idx] = event;
__set_bit(idx, cpuc->active_mask);
__set_bit(idx, cpuc->running);
- x86_pmu.enable(event);
+ static_call(x86_pmu_enable)(event);
perf_event_update_userpage(event);
}
@@ -1460,7 +1489,7 @@ void x86_pmu_stop(struct perf_event *eve
struct hw_perf_event *hwc = &event->hw;
if (test_bit(hwc->idx, cpuc->active_mask)) {
- x86_pmu.disable(event);
+ static_call(x86_pmu_disable)(event);
__clear_bit(hwc->idx, cpuc->active_mask);
cpuc->events[hwc->idx] = NULL;
WARN_ON_ONCE(hwc->state & PERF_HES_STOPPED);
@@ -1510,8 +1539,7 @@ static void x86_pmu_del(struct perf_even
if (i >= cpuc->n_events - cpuc->n_added)
--cpuc->n_added;
- if (x86_pmu.put_event_constraints)
- x86_pmu.put_event_constraints(cpuc, event);
+ static_call_cond(x86_pmu_put_event_constraints)(cpuc, event);
/* Delete the array entry. */
while (++i < cpuc->n_events) {
@@ -1524,13 +1552,12 @@ static void x86_pmu_del(struct perf_even
perf_event_update_userpage(event);
do_del:
- if (x86_pmu.del) {
- /*
- * This is after x86_pmu_stop(); so we disable LBRs after any
- * event can need them etc..
- */
- x86_pmu.del(event);
- }
+
+ /*
+ * This is after x86_pmu_stop(); so we disable LBRs after any
+ * event can need them etc..
+ */
+ static_call_cond(x86_pmu_del)(event);
}
int x86_pmu_handle_irq(struct pt_regs *regs)
@@ -1608,7 +1635,7 @@ perf_event_nmi_handler(unsigned int cmd,
return NMI_DONE;
start_clock = sched_clock();
- ret = x86_pmu.handle_irq(regs);
+ ret = static_call(x86_pmu_handle_irq)(regs);
finish_clock = sched_clock();
perf_sample_event_took(finish_clock - start_clock);
@@ -1821,6 +1848,38 @@ ssize_t x86_event_sysfs_show(char *page,
static struct attribute_group x86_pmu_attr_group;
static struct attribute_group x86_pmu_caps_group;
+static void x86_pmu_static_call_update(void)
+{
+ static_call_update(x86_pmu_handle_irq, x86_pmu.handle_irq);
+ static_call_update(x86_pmu_disable_all, x86_pmu.disable_all);
+ static_call_update(x86_pmu_enable_all, x86_pmu.enable_all);
+ static_call_update(x86_pmu_enable, x86_pmu.enable);
+ static_call_update(x86_pmu_disable, x86_pmu.disable);
+
+ static_call_update(x86_pmu_add, x86_pmu.add);
+ static_call_update(x86_pmu_del, x86_pmu.del);
+ static_call_update(x86_pmu_read, x86_pmu.read);
+
+ static_call_update(x86_pmu_schedule_events, x86_pmu.schedule_events);
+ static_call_update(x86_pmu_get_event_constraints, x86_pmu.get_event_constraints);
+ static_call_update(x86_pmu_put_event_constraints, x86_pmu.put_event_constraints);
+
+ static_call_update(x86_pmu_start_scheduling, x86_pmu.start_scheduling);
+ static_call_update(x86_pmu_commit_scheduling, x86_pmu.commit_scheduling);
+ static_call_update(x86_pmu_stop_scheduling, x86_pmu.stop_scheduling);
+
+ static_call_update(x86_pmu_sched_task, x86_pmu.sched_task);
+ static_call_update(x86_pmu_swap_task_ctx, x86_pmu.swap_task_ctx);
+
+ static_call_update(x86_pmu_drain_pebs, x86_pmu.drain_pebs);
+ static_call_update(x86_pmu_pebs_aliases, x86_pmu.pebs_aliases);
+}
+
+static void _x86_pmu_read(struct perf_event *event)
+{
+ x86_perf_event_update(event);
+}
+
static int __init init_hw_perf_events(void)
{
struct x86_pmu_quirk *quirk;
@@ -1889,6 +1948,11 @@ static int __init init_hw_perf_events(vo
pr_info("... fixed-purpose events: %d\n", x86_pmu.num_counters_fixed);
pr_info("... event mask: %016Lx\n", x86_pmu.intel_ctrl);
+ if (!x86_pmu.read)
+ x86_pmu.read = _x86_pmu_read;
+
+ x86_pmu_static_call_update();
+
/*
* Install callbacks. Core will call them for each online
* cpu.
@@ -1925,11 +1989,9 @@ static int __init init_hw_perf_events(vo
}
early_initcall(init_hw_perf_events);
-static inline void x86_pmu_read(struct perf_event *event)
+static void x86_pmu_read(struct perf_event *event)
{
- if (x86_pmu.read)
- return x86_pmu.read(event);
- x86_perf_event_update(event);
+ static_call(x86_pmu_read)(event);
}
/*
@@ -2006,7 +2068,7 @@ static int x86_pmu_commit_txn(struct pmu
if (!x86_pmu_initialized())
return -EAGAIN;
- ret = x86_pmu.schedule_events(cpuc, n, assign);
+ ret = static_call(x86_pmu_schedule_events)(cpuc, n, assign);
if (ret)
return ret;
@@ -2299,15 +2361,13 @@ static const struct attribute_group *x86
static void x86_pmu_sched_task(struct perf_event_context *ctx, bool sched_in)
{
- if (x86_pmu.sched_task)
- x86_pmu.sched_task(ctx, sched_in);
+ static_call_cond(x86_pmu_sched_task)(ctx, sched_in);
}
static void x86_pmu_swap_task_ctx(struct perf_event_context *prev,
struct perf_event_context *next)
{
- if (x86_pmu.swap_task_ctx)
- x86_pmu.swap_task_ctx(prev, next);
+ static_call_cond(x86_pmu_swap_task_ctx)(prev, next);
}
void perf_check_microcode(void)
Powered by blists - more mailing lists