Message-Id: <20211201010217.886919-4-eranian@google.com>
Date:   Tue, 30 Nov 2021 17:02:07 -0800
From:   Stephane Eranian <eranian@...gle.com>
To:     linux-kernel@...r.kernel.org
Cc:     peterz@...radead.org, kim.phillips@....com, acme@...hat.com,
        jolsa@...hat.com, songliubraving@...com, mpe@...erman.id.au,
        maddy@...ux.ibm.com
Subject: [PATCH v3 03/13] perf/x86/amd: add AMD Fam19h Branch Sampling support

This patch adds support for the AMD Fam19h 16-deep branch sampling feature as
described in the AMD PPR Fam19h Model 01h Revision B1.  This is a model
specific extension. It is not an architected AMD feature.

The Branch Sampling (BRS) operates with a 16-deep saturating buffer in MSR
registers. There is no branch type filtering. All control flow changes are
captured. BRS relies on specific programming of the core PMU of Fam19h.  In
particular, the following requirements must be met:
 - the sampling period be greater than 16 (BRS depth)
 - the sampling period must use a fixed and not frequency mode

BRS interacts with the NMI interrupt as well. Because enabling BRS is
expensive, it is only activated after P event occurrences, where P is the
desired sampling period.  At P occurrences of the event, the counter
overflows, the CPU catches the interrupt, activates BRS for 16 branches until
it saturates, and then delivers the NMI to the kernel.  Between the overflow
and the time BRS activates more branches may be executed skewing the period.
All along, the sampling event keeps counting. The skid may be attenuated by
reducing the sampling period by 16 (subsequent patch).

BRS is integrated into perf_events seamlessly via the same
PERF_RECORD_BRANCH_STACK sample format. BRS generates perf_branch_entry
records in the sampling buffer. No prediction information is supported. The
branches are stored in reverse order of execution.  The most recent branch is
the first entry in each record.

No modification to the perf tool is necessary.

BRS can be used with any sampling event. However, it is recommended to use
the RETIRED_BRANCH_INSTRUCTIONS event because it matches what the BRS
captures.

$ perf record -b -c 1000037 -e cpu/event=0xc2,name=ret_br_instructions/ test

$ perf report -D
56531696056126 0x193c000 [0x1a8]: PERF_RECORD_SAMPLE(IP, 0x2): 18122/18230: 0x401d24 period: 1000037 addr: 0
... branch stack: nr:16
.....  0: 0000000000401d24 -> 0000000000401d5a 0 cycles      0
.....  1: 0000000000401d5c -> 0000000000401d24 0 cycles      0
.....  2: 0000000000401d22 -> 0000000000401d5c 0 cycles      0
.....  3: 0000000000401d5e -> 0000000000401d22 0 cycles      0
.....  4: 0000000000401d20 -> 0000000000401d5e 0 cycles      0
.....  5: 0000000000401d3e -> 0000000000401d20 0 cycles      0
.....  6: 0000000000401d42 -> 0000000000401d3e 0 cycles      0
.....  7: 0000000000401d3c -> 0000000000401d42 0 cycles      0
.....  8: 0000000000401d44 -> 0000000000401d3c 0 cycles      0
.....  9: 0000000000401d3a -> 0000000000401d44 0 cycles      0
..... 10: 0000000000401d46 -> 0000000000401d3a 0 cycles      0
..... 11: 0000000000401d38 -> 0000000000401d46 0 cycles      0
..... 12: 0000000000401d48 -> 0000000000401d38 0 cycles      0
..... 13: 0000000000401d36 -> 0000000000401d48 0 cycles      0
..... 14: 0000000000401d4a -> 0000000000401d36 0 cycles      0
..... 15: 0000000000401d34 -> 0000000000401d4a 0 cycles      0
 ... thread: test:18230
 ...... dso: test

Signed-off-by: Stephane Eranian <eranian@...gle.com>
---
 arch/x86/events/amd/Makefile     |   2 +-
 arch/x86/events/amd/brs.c        | 317 +++++++++++++++++++++++++++++++
 arch/x86/events/amd/core.c       | 197 ++++++++++++++++++-
 arch/x86/events/core.c           |  10 +-
 arch/x86/events/perf_event.h     | 101 ++++++++--
 arch/x86/include/asm/msr-index.h |   4 +
 6 files changed, 609 insertions(+), 22 deletions(-)
 create mode 100644 arch/x86/events/amd/brs.c

diff --git a/arch/x86/events/amd/Makefile b/arch/x86/events/amd/Makefile
index 6cbe38d5fd9d..cf323ffab5cd 100644
--- a/arch/x86/events/amd/Makefile
+++ b/arch/x86/events/amd/Makefile
@@ -1,5 +1,5 @@
 # SPDX-License-Identifier: GPL-2.0
-obj-$(CONFIG_CPU_SUP_AMD)		+= core.o
+obj-$(CONFIG_CPU_SUP_AMD)		+= core.o brs.o
 obj-$(CONFIG_PERF_EVENTS_AMD_POWER)	+= power.o
 obj-$(CONFIG_X86_LOCAL_APIC)		+= ibs.o
 obj-$(CONFIG_PERF_EVENTS_AMD_UNCORE)	+= amd-uncore.o
diff --git a/arch/x86/events/amd/brs.c b/arch/x86/events/amd/brs.c
new file mode 100644
index 000000000000..851b03bcbea2
--- /dev/null
+++ b/arch/x86/events/amd/brs.c
@@ -0,0 +1,317 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Implement support for AMD Fam19h Branch Sampling feature
+ * Based on specifications published in AMD PPR Fam19 Model 01
+ *
+ * Copyright 2021 Google LLC
+ * Contributed by Stephane Eranian <eranian@...gle.com>
+ */
+#include <linux/kernel.h>
+#include <asm/msr.h>
+#include <asm/cpufeature.h>
+
+#include "../perf_event.h"
+
+#define BRS_POISON	0xFFFFFFFFFFFFFFFEULL /* mark limit of valid entries */
+
+/* Debug Extension Configuration register layout */
+union amd_debug_extn_cfg {
+	__u64 val;
+	struct {
+		__u64	rsvd0:2,  /* reserved */
+			brsmen:1, /* branch sample enable */
+			rsvd4_3:2,/* reserved - must be 0x3 */
+			vb:1,     /* valid branches recorded */
+			rsvd2:10, /* reserved */
+			msroff:4, /* index of next entry to write */
+			rsvd3:4,  /* reserved */
+			pmc:3,    /* #PMC holding the sampling event */
+			rsvd4:37; /* reserved */
+	};
+};
+
+static inline unsigned int brs_from(int idx)
+{
+	return MSR_AMD_SAMP_BR_FROM + 2 * idx;
+}
+
+static inline unsigned int brs_to(int idx)
+{
+	return MSR_AMD_SAMP_BR_FROM + 2 * idx + 1;
+}
+
+static inline void set_debug_extn_cfg(u64 val)
+{
+	/* bits[4:3] must always be set to 11b */
+	wrmsrl(MSR_AMD_DBG_EXTN_CFG, val | 3ULL << 3);
+}
+
+static inline u64 get_debug_extn_cfg(void)
+{
+	u64 val;
+
+	rdmsrl(MSR_AMD_DBG_EXTN_CFG, val);
+	return val;
+}
+
+static bool __init amd_brs_detect(void)
+{
+	if (!boot_cpu_has(X86_FEATURE_AMD_BRS))
+		return false;
+
+	switch (boot_cpu_data.x86) {
+	case 0x19: /* AMD Fam19h (Zen3) */
+		x86_pmu.lbr_nr = 16;
+
+		/* No hardware filtering supported */
+		x86_pmu.lbr_sel_map = NULL;
+		x86_pmu.lbr_sel_mask = 0;
+		break;
+	default:
+		return false;
+	}
+
+	return true;
+}
+
+/*
+ * Current BRS implementation does not support branch type or privilege level
+ * filtering. Therefore, this function simply enforces these limitations. No need for
+ * a br_sel_map. Software filtering is not supported because it would not correlate well
+ * with a sampling period.
+ */
+int amd_brs_setup_filter(struct perf_event *event)
+{
+	u64 type = event->attr.branch_sample_type;
+
+	/* No BRS support */
+	if (!x86_pmu.lbr_nr)
+		return -EOPNOTSUPP;
+
+	/* Can only capture all branches, i.e., no filtering */
+	if ((type & ~PERF_SAMPLE_BRANCH_PLM_ALL) != PERF_SAMPLE_BRANCH_ANY)
+		return -EINVAL;
+
+	/* can only capture at all priv levels due to the way BRS works */
+	if ((type & PERF_SAMPLE_BRANCH_PLM_ALL) != PERF_SAMPLE_BRANCH_PLM_ALL)
+		return -EINVAL;
+
+	return 0;
+}
+
+/* tos = top of stack, i.e., last valid entry written */
+static inline int amd_brs_get_tos(union amd_debug_extn_cfg *cfg)
+{
+	/*
+	 * msroff: index of next entry to write so top-of-stack is one off
+	 * if BRS is full then msroff is set back to 0.
+	 */
+	return (cfg->msroff ? cfg->msroff : x86_pmu.lbr_nr) - 1;
+}
+
+/*
+ * make sure we have a sane BRS offset to begin with
+ * especially with kexec
+ */
+void amd_brs_reset(void)
+{
+	/*
+	 * Reset config
+	 */
+	set_debug_extn_cfg(0);
+
+	/*
+	 * Mark first entry as poisoned
+	 */
+	wrmsrl(brs_to(0), BRS_POISON);
+}
+
+int __init amd_brs_init(void)
+{
+	if (!amd_brs_detect())
+		return -EOPNOTSUPP;
+
+	pr_cont("%d-deep BRS, ", x86_pmu.lbr_nr);
+
+	return 0;
+}
+
+void amd_brs_enable(void)
+{
+	struct cpu_hw_events *cpuc = this_cpu_ptr(&cpu_hw_events);
+	union amd_debug_extn_cfg cfg;
+
+	/* Activate only on first user */
+	if (++cpuc->brs_active > 1)
+		return;
+
+	cfg.val    = 0; /* reset all fields */
+	cfg.brsmen = 1; /* enable branch sampling */
+
+	/* Set enable bit */
+	set_debug_extn_cfg(cfg.val);
+}
+
+void amd_brs_enable_all(void)
+{
+	struct cpu_hw_events *cpuc = this_cpu_ptr(&cpu_hw_events);
+	if (cpuc->lbr_users)
+		amd_brs_enable();
+}
+
+void amd_brs_disable(void)
+{
+	struct cpu_hw_events *cpuc = this_cpu_ptr(&cpu_hw_events);
+	union amd_debug_extn_cfg cfg;
+
+	/* Check if active (could be disabled via x86_pmu_disable_all()) */
+	if (!cpuc->brs_active)
+		return;
+
+	/* Only disable for last user */
+	if (--cpuc->brs_active)
+		return;
+
+	/*
+	 * Clear the brsmen bit but preserve the others as they contain
+	 * useful state such as vb and msroff
+	 */
+	cfg.val = get_debug_extn_cfg();
+
+	/*
+	 * When coming in on interrupt and BRS is full, then hw will have
+	 * already stopped BRS, no need to issue wrmsr again
+	 */
+	if (cfg.brsmen) {
+		cfg.brsmen = 0;
+		set_debug_extn_cfg(cfg.val);
+	}
+}
+
+void amd_brs_disable_all(void)
+{
+	struct cpu_hw_events *cpuc = this_cpu_ptr(&cpu_hw_events);
+	if (cpuc->lbr_users)
+		amd_brs_disable();
+}
+
+/*
+ * Caller must ensure amd_brs_inuse() is true before calling
+ * return:
+ */
+void amd_brs_drain(void)
+{
+	struct cpu_hw_events *cpuc = this_cpu_ptr(&cpu_hw_events);
+	struct perf_event *event = cpuc->events[0];
+	struct perf_branch_entry *br = cpuc->lbr_entries;
+	union amd_debug_extn_cfg cfg;
+	u32 i, nr = 0, num, tos, start;
+	u32 shift = 64 - boot_cpu_data.x86_virt_bits;
+
+	/*
+	 * BRS event forced on PMC0,
+	 * so check if there is an event.
+	 * It is possible to have lbr_users > 0 but the event
+	 * not yet scheduled due to long latency PMU irq
+	 */
+	if (!event)
+		goto empty;
+
+	cfg.val = get_debug_extn_cfg();
+
+	/* Sanity check [0-x86_pmu.lbr_nr] */
+	if (WARN_ON_ONCE(cfg.msroff >= x86_pmu.lbr_nr))
+		goto empty;
+
+	/* No valid branch */
+	if (cfg.vb == 0)
+		goto empty;
+
+	/*
+	 * msr.off points to next entry to be written
+	 * tos = most recent entry index = msr.off - 1
+	 * BRS register buffer saturates, so we know we have
+	 * start < tos and that we have to read from start to tos
+	 */
+	start = 0;
+	tos = amd_brs_get_tos(&cfg);
+
+	num = tos - start + 1;
+
+	/*
+	 * BRS is only one pass (saturation) from MSROFF to depth-1
+	 * MSROFF wraps to zero when buffer is full
+	 */
+	for (i = 0; i < num; i++) {
+		u32 brs_idx = tos - i;
+		u64 from, to;
+
+		rdmsrl(brs_to(brs_idx), to);
+
+		/* Entry does not belong to us (as marked by kernel) */
+		if (to == BRS_POISON)
+			break;
+
+		rdmsrl(brs_from(brs_idx), from);
+
+		/*
+		 * Sign-extend SAMP_BR_TO to 64 bits, bits 61-63 are reserved.
+		 * Necessary to generate proper virtual addresses suitable for
+		 * symbolization
+		 */
+		to = (u64)(((s64)to << shift) >> shift);
+
+		perf_clear_branch_entry_bitfields(br+nr);
+
+		br[nr].from = from;
+		br[nr].to   = to;
+
+		nr++;
+	}
+empty:
+	/* Record number of sampled branches */
+	cpuc->lbr_stack.nr = nr;
+}
+
+/*
+ * Poison most recent entry to prevent reuse by next task
+ * required because BRS entry are not tagged by PID
+ */
+static void amd_brs_poison_buffer(void)
+{
+	union amd_debug_extn_cfg cfg;
+	unsigned int idx;
+
+	/* Get current state */
+	cfg.val = get_debug_extn_cfg();
+
+	/* idx is most recently written entry */
+	idx = amd_brs_get_tos(&cfg);
+
+	/* Poison target of entry */
+	wrmsrl(brs_to(idx), BRS_POISON);
+}
+
+/*
+ * On context switch in, we need to make sure no samples from previous user
+ * are left in the BRS.
+ *
+ * On ctxswin, sched_in = true, called after the PMU has started
+ * On ctxswout, sched_in = false, called before the PMU is stopped
+ */
+void amd_pmu_brs_sched_task(struct perf_event_context *ctx, bool sched_in)
+{
+	struct cpu_hw_events *cpuc = this_cpu_ptr(&cpu_hw_events);
+
+	/* no active users */
+	if (!cpuc->lbr_users)
+		return;
+
+	/*
+	 * On context switch in, we need to ensure we do not use entries
+	 * from previous BRS user on that CPU, so we poison the buffer as
+	 * a faster way compared to resetting all entries.
+	 */
+	if (sched_in)
+		amd_brs_poison_buffer();
+}
diff --git a/arch/x86/events/amd/core.c b/arch/x86/events/amd/core.c
index 9687a8aef01c..44d8f618bb3e 100644
--- a/arch/x86/events/amd/core.c
+++ b/arch/x86/events/amd/core.c
@@ -327,6 +327,8 @@ static inline bool amd_is_pair_event_code(struct hw_perf_event *hwc)
 
 static int amd_core_hw_config(struct perf_event *event)
 {
+	int ret = 0;
+
 	if (event->attr.exclude_host && event->attr.exclude_guest)
 		/*
 		 * When HO == GO == 1 the hardware treats that as GO == HO == 0
@@ -343,7 +345,32 @@ static int amd_core_hw_config(struct perf_event *event)
 	if ((x86_pmu.flags & PMU_FL_PAIR) && amd_is_pair_event_code(&event->hw))
 		event->hw.flags |= PERF_X86_EVENT_PAIR;
 
-	return 0;
+	/*
+	 * if branch stack is requested
+	 */
+	if (has_branch_stack(event) && is_sampling_event(event)) {
+		/*
+		 * BRS implementation does not work with frequency mode
+		 * reprogramming of the period.
+		 */
+		if (event->attr.freq)
+			return -EINVAL;
+		/*
+		 * The kernel subtracts BRS depth from period, so it must be big enough
+		 */
+		if (event->attr.sample_period <= x86_pmu.lbr_nr)
+			return -EINVAL;
+
+		/*
+		 * Check if we can allow PERF_SAMPLE_BRANCH_STACK
+		 */
+		ret = amd_brs_setup_filter(event);
+
+		/* only set in case of success */
+		if (!ret)
+			event->hw.flags |= PERF_X86_EVENT_AMD_BRS;
+	}
+	return ret;
 }
 
 static inline int amd_is_nb_event(struct hw_perf_event *hwc)
@@ -366,7 +393,7 @@ static int amd_pmu_hw_config(struct perf_event *event)
 	if (event->attr.precise_ip && get_ibs_caps())
 		return -ENOENT;
 
-	if (has_branch_stack(event))
+	if (has_branch_stack(event) && !x86_pmu.lbr_nr)
 		return -EOPNOTSUPP;
 
 	ret = x86_pmu_hw_config(event);
@@ -555,6 +582,8 @@ static void amd_pmu_cpu_starting(int cpu)
 
 	cpuc->amd_nb->nb_id = nb_id;
 	cpuc->amd_nb->refcnt++;
+
+	amd_brs_reset();
 }
 
 static void amd_pmu_cpu_dead(int cpu)
@@ -610,6 +639,8 @@ static void amd_pmu_disable_all(void)
 	struct cpu_hw_events *cpuc = this_cpu_ptr(&cpu_hw_events);
 	int idx;
 
+	amd_brs_disable_all();
+
 	x86_pmu_disable_all();
 
 	/*
@@ -634,6 +665,30 @@ static void amd_pmu_disable_all(void)
 	}
 }
 
+static void amd_pmu_enable_event(struct perf_event *event)
+{
+	x86_pmu_enable_event(event);
+}
+
+static void amd_pmu_enable_all(int added)
+{
+	struct cpu_hw_events *cpuc = this_cpu_ptr(&cpu_hw_events);
+	struct hw_perf_event *hwc;
+	int idx;
+
+	amd_brs_enable_all();
+
+	for (idx = 0; idx < x86_pmu.num_counters; idx++) {
+		hwc = &cpuc->events[idx]->hw;
+
+		/* only activate events which are marked as active */
+		if (!test_bit(idx, cpuc->active_mask))
+			continue;
+
+		amd_pmu_enable_event(cpuc->events[idx]);
+	}
+}
+
 static void amd_pmu_disable_event(struct perf_event *event)
 {
 	x86_pmu_disable_event(event);
@@ -651,6 +706,18 @@ static void amd_pmu_disable_event(struct perf_event *event)
 	amd_pmu_wait_on_overflow(event->hw.idx);
 }
 
+static void amd_pmu_add_event(struct perf_event *event)
+{
+	if (needs_branch_stack(event))
+		amd_pmu_brs_add(event);
+}
+
+static void amd_pmu_del_event(struct perf_event *event)
+{
+	if (needs_branch_stack(event))
+		amd_pmu_brs_del(event);
+}
+
 /*
  * Because of NMI latency, if multiple PMC counters are active or other sources
  * of NMIs are received, the perf NMI handler can handle one or more overflowed
@@ -671,11 +738,31 @@ static void amd_pmu_disable_event(struct perf_event *event)
  */
 static int amd_pmu_handle_irq(struct pt_regs *regs)
 {
+	struct cpu_hw_events *cpuc = this_cpu_ptr(&cpu_hw_events);
 	int handled;
+	int pmu_enabled;
+
+	/*
+	 * Save the PMU state.
+	 * It needs to be restored when leaving the handler.
+	 */
+	pmu_enabled = cpuc->enabled;
+	cpuc->enabled = 0;
+
+	/* stop everything (includes BRS) */
+	amd_pmu_disable_all();
+
+	/* Drain BRS is in use (could be inactive) */
+	if (cpuc->lbr_users)
+		amd_brs_drain();
 
 	/* Process any counter overflows */
 	handled = x86_pmu_handle_irq(regs);
 
+	cpuc->enabled = pmu_enabled;
+	if (pmu_enabled)
+		amd_pmu_enable_all(0);
+
 	/*
 	 * If a counter was handled, record a timestamp such that un-handled
 	 * NMIs will be claimed if arriving within that window.
@@ -897,6 +984,51 @@ static void amd_put_event_constraints_f17h(struct cpu_hw_events *cpuc,
 		--cpuc->n_pair;
 }
 
+/*
+ * Because of the way BRS operates with an inactive and active phases, and
+ * the link to one counter, it is not possible to have two events using BRS
+ * scheduled at the same time. There would be an issue with enforcing the
+ * period of each one and given that the BRS saturates, it would not be possible
+ * to guarantee correlated content for all events. Therefore, in situations
+ * where multiple events want to use BRS, the kernel enforces mutual exclusion.
+ * Exclusion is enforced by chosing only one counter for events using BRS.
+ * The event scheduling logic will then automatically multiplex the
+ * events and ensure that at most one event is actively using BRS.
+ *
+ * The BRS counter could be any counter, but there is no constraint on Fam19h,
+ * therefore all counters are equal and thus we pick the first one: PMC0
+ */
+static struct event_constraint amd_fam19h_brs_cntr0_constraint =
+	EVENT_CONSTRAINT(0, 0x1, AMD64_RAW_EVENT_MASK);
+
+static struct event_constraint amd_fam19h_brs_pair_cntr0_constraint =
+	__EVENT_CONSTRAINT(0, 0x1, AMD64_RAW_EVENT_MASK, 1, 0, PERF_X86_EVENT_PAIR);
+
+static struct event_constraint *
+amd_get_event_constraints_f19h(struct cpu_hw_events *cpuc, int idx,
+			  struct perf_event *event)
+{
+	struct hw_perf_event *hwc = &event->hw;
+	bool has_brs = has_amd_brs(hwc);
+
+	/*
+	 * In case BRS is used with an event requiring a counter pair,
+	 * the kernel allows it but only on counter 0 & 1 to enforce
+	 * multiplexing requiring to protect BRS in case of multiple
+	 * BRS users
+	 */
+	if (amd_is_pair_event_code(hwc)) {
+		return has_brs ? &amd_fam19h_brs_pair_cntr0_constraint
+			       : &pair_constraint;
+	}
+
+	if (has_brs)
+		return &amd_fam19h_brs_cntr0_constraint;
+
+	return &unconstrained;
+}
+
+
 static ssize_t amd_event_sysfs_show(char *page, u64 config)
 {
 	u64 event = (config & ARCH_PERFMON_EVENTSEL_EVENT) |
@@ -905,12 +1037,19 @@ static ssize_t amd_event_sysfs_show(char *page, u64 config)
 	return x86_event_sysfs_show(page, config, event);
 }
 
+static void amd_pmu_sched_task(struct perf_event_context *ctx,
+				 bool sched_in)
+{
+	if (sched_in && x86_pmu.lbr_nr)
+		amd_pmu_brs_sched_task(ctx, sched_in);
+}
+
 static __initconst const struct x86_pmu amd_pmu = {
 	.name			= "AMD",
 	.handle_irq		= amd_pmu_handle_irq,
 	.disable_all		= amd_pmu_disable_all,
-	.enable_all		= x86_pmu_enable_all,
-	.enable			= x86_pmu_enable_event,
+	.enable_all		= amd_pmu_enable_all,
+	.enable			= amd_pmu_enable_event,
 	.disable		= amd_pmu_disable_event,
 	.hw_config		= amd_pmu_hw_config,
 	.schedule_events	= x86_schedule_events,
@@ -920,6 +1059,8 @@ static __initconst const struct x86_pmu amd_pmu = {
 	.event_map		= amd_pmu_event_map,
 	.max_events		= ARRAY_SIZE(amd_perfmon_event_map),
 	.num_counters		= AMD64_NUM_COUNTERS,
+	.add			= amd_pmu_add_event,
+	.del			= amd_pmu_del_event,
 	.cntval_bits		= 48,
 	.cntval_mask		= (1ULL << 48) - 1,
 	.apic			= 1,
@@ -938,6 +1079,37 @@ static __initconst const struct x86_pmu amd_pmu = {
 	.amd_nb_constraints	= 1,
 };
 
+static ssize_t branches_show(struct device *cdev,
+			      struct device_attribute *attr,
+			      char *buf)
+{
+	return snprintf(buf, PAGE_SIZE, "%d\n", x86_pmu.lbr_nr);
+}
+
+static DEVICE_ATTR_RO(branches);
+
+static struct attribute *amd_pmu_brs_attrs[] = {
+	&dev_attr_branches.attr,
+	NULL,
+};
+
+static umode_t
+amd_brs_is_visible(struct kobject *kobj, struct attribute *attr, int i)
+{
+	return x86_pmu.lbr_nr ? attr->mode : 0;
+}
+
+static struct attribute_group group_caps_amd_brs = {
+	.name  = "caps",
+	.attrs = amd_pmu_brs_attrs,
+	.is_visible = amd_brs_is_visible,
+};
+
+static const struct attribute_group *amd_attr_update[] = {
+	&group_caps_amd_brs,
+	NULL,
+};
+
 static int __init amd_core_pmu_init(void)
 {
 	u64 even_ctr_mask = 0ULL;
@@ -989,6 +1161,23 @@ static int __init amd_core_pmu_init(void)
 		x86_pmu.flags |= PMU_FL_PAIR;
 	}
 
+	if (boot_cpu_data.x86 >= 0x19) {
+		/*
+		 * On AMD, invoking pmu_disable_all() is very expensive and the function is
+		 * invoked on context-switch in via sched_task_in(), so enable only when necessary
+		 */
+		if (!amd_brs_init()) {
+			x86_pmu.get_event_constraints = amd_get_event_constraints_f19h;
+			x86_pmu.sched_task = amd_pmu_sched_task;
+			/*
+			 * The put_event_constraints callback is shared with
+			 * Fam17h, set above
+			 */
+		}
+	}
+
+	x86_pmu.attr_update = amd_attr_update;
+
 	pr_cont("core perfctr, ");
 	return 0;
 }
diff --git a/arch/x86/events/core.c b/arch/x86/events/core.c
index 620347398027..e294ff5d176e 100644
--- a/arch/x86/events/core.c
+++ b/arch/x86/events/core.c
@@ -1338,6 +1338,10 @@ static void x86_pmu_enable(struct pmu *pmu)
 			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;
@@ -1704,11 +1708,15 @@ int x86_pmu_handle_irq(struct pt_regs *regs)
 		 * event overflow
 		 */
 		handled++;
-		perf_sample_data_init(&data, 0, event->hw.last_period);
 
 		if (!x86_perf_event_set_period(event))
 			continue;
 
+		perf_sample_data_init(&data, 0, event->hw.last_period);
+
+		if (has_branch_stack(event))
+			data.br_stack = &cpuc->lbr_stack;
+
 		if (perf_event_overflow(event, &data, regs))
 			x86_pmu_stop(event, 0);
 	}
diff --git a/arch/x86/events/perf_event.h b/arch/x86/events/perf_event.h
index 5480db242083..e7c904a9515d 100644
--- a/arch/x86/events/perf_event.h
+++ b/arch/x86/events/perf_event.h
@@ -67,22 +67,23 @@ static inline bool constraint_match(struct event_constraint *c, u64 ecode)
 /*
  * struct hw_perf_event.flags flags
  */
-#define PERF_X86_EVENT_PEBS_LDLAT	0x0001 /* ld+ldlat data address sampling */
-#define PERF_X86_EVENT_PEBS_ST		0x0002 /* st data address sampling */
-#define PERF_X86_EVENT_PEBS_ST_HSW	0x0004 /* haswell style datala, store */
-#define PERF_X86_EVENT_PEBS_LD_HSW	0x0008 /* haswell style datala, load */
-#define PERF_X86_EVENT_PEBS_NA_HSW	0x0010 /* haswell style datala, unknown */
-#define PERF_X86_EVENT_EXCL		0x0020 /* HT exclusivity on counter */
-#define PERF_X86_EVENT_DYNAMIC		0x0040 /* dynamic alloc'd constraint */
-#define PERF_X86_EVENT_RDPMC_ALLOWED	0x0080 /* grant rdpmc permission */
-#define PERF_X86_EVENT_EXCL_ACCT	0x0100 /* accounted EXCL event */
-#define PERF_X86_EVENT_AUTO_RELOAD	0x0200 /* use PEBS auto-reload */
-#define PERF_X86_EVENT_LARGE_PEBS	0x0400 /* use large PEBS */
-#define PERF_X86_EVENT_PEBS_VIA_PT	0x0800 /* use PT buffer for PEBS */
-#define PERF_X86_EVENT_PAIR		0x1000 /* Large Increment per Cycle */
-#define PERF_X86_EVENT_LBR_SELECT	0x2000 /* Save/Restore MSR_LBR_SELECT */
-#define PERF_X86_EVENT_TOPDOWN		0x4000 /* Count Topdown slots/metrics events */
-#define PERF_X86_EVENT_PEBS_STLAT	0x8000 /* st+stlat data address sampling */
+#define PERF_X86_EVENT_PEBS_LDLAT	0x00001 /* ld+ldlat data address sampling */
+#define PERF_X86_EVENT_PEBS_ST		0x00002 /* st data address sampling */
+#define PERF_X86_EVENT_PEBS_ST_HSW	0x00004 /* haswell style datala, store */
+#define PERF_X86_EVENT_PEBS_LD_HSW	0x00008 /* haswell style datala, load */
+#define PERF_X86_EVENT_PEBS_NA_HSW	0x00010 /* haswell style datala, unknown */
+#define PERF_X86_EVENT_EXCL		0x00020 /* HT exclusivity on counter */
+#define PERF_X86_EVENT_DYNAMIC		0x00040 /* dynamic alloc'd constraint */
+#define PERF_X86_EVENT_RDPMC_ALLOWED	0x00080 /* grant rdpmc permission */
+#define PERF_X86_EVENT_EXCL_ACCT	0x00100 /* accounted EXCL event */
+#define PERF_X86_EVENT_AUTO_RELOAD	0x00200 /* use PEBS auto-reload */
+#define PERF_X86_EVENT_LARGE_PEBS	0x00400 /* use large PEBS */
+#define PERF_X86_EVENT_PEBS_VIA_PT	0x00800 /* use PT buffer for PEBS */
+#define PERF_X86_EVENT_PAIR		0x01000 /* Large Increment per Cycle */
+#define PERF_X86_EVENT_LBR_SELECT	0x02000 /* Save/Restore MSR_LBR_SELECT */
+#define PERF_X86_EVENT_TOPDOWN		0x04000 /* Count Topdown slots/metrics events */
+#define PERF_X86_EVENT_PEBS_STLAT	0x08000 /* st+stlat data address sampling */
+#define PERF_X86_EVENT_AMD_BRS		0x10000 /* AMD Branch Sampling */
 
 static inline bool is_topdown_count(struct perf_event *event)
 {
@@ -324,6 +325,8 @@ struct cpu_hw_events {
 	 * AMD specific bits
 	 */
 	struct amd_nb			*amd_nb;
+	int				brs_active; /* BRS is enabled */
+
 	/* Inverted mask of bits to clear in the perf_ctr ctrl registers */
 	u64				perf_ctr_virt_mask;
 	int				n_pair; /* Large increment events */
@@ -1099,6 +1102,11 @@ int x86_pmu_hw_config(struct perf_event *event);
 
 void x86_pmu_disable_all(void);
 
+static inline bool has_amd_brs(struct hw_perf_event *hwc)
+{
+	return hwc->flags & PERF_X86_EVENT_AMD_BRS;
+}
+
 static inline bool is_counter_pair(struct hw_perf_event *hwc)
 {
 	return hwc->flags & PERF_X86_EVENT_PAIR;
@@ -1204,6 +1212,50 @@ static inline bool fixed_counter_disabled(int i, struct pmu *pmu)
 #ifdef CONFIG_CPU_SUP_AMD
 
 int amd_pmu_init(void);
+int amd_brs_init(void);
+void amd_brs_disable(void);
+void amd_brs_enable(void);
+void amd_brs_enable_all(void);
+void amd_brs_disable_all(void);
+void amd_brs_drain(void);
+void amd_brs_disable_all(void);
+int amd_brs_setup_filter(struct perf_event *event);
+void amd_brs_reset(void);
+
+static inline void amd_pmu_brs_add(struct perf_event *event)
+{
+	struct cpu_hw_events *cpuc = this_cpu_ptr(&cpu_hw_events);
+
+	perf_sched_cb_inc(event->ctx->pmu);
+	cpuc->lbr_users++;
+	/*
+	 * No need to reset BRS because it is reset
+	 * on brs_enable() and it is saturating
+	 */
+}
+
+static inline void amd_pmu_brs_del(struct perf_event *event)
+{
+	struct cpu_hw_events *cpuc = this_cpu_ptr(&cpu_hw_events);
+
+	cpuc->lbr_users--;
+	WARN_ON_ONCE(cpuc->lbr_users < 0);
+
+	perf_sched_cb_dec(event->ctx->pmu);
+}
+
+void amd_pmu_brs_sched_task(struct perf_event_context *ctx, bool sched_in);
+
+/*
+ * check if BRS is activated on the CPU
+ * active defined as it has non-zero users and DBG_EXT_CFG.BRSEN=1
+ */
+static inline bool amd_brs_active(void)
+{
+	struct cpu_hw_events *cpuc = this_cpu_ptr(&cpu_hw_events);
+
+	return cpuc->brs_active;
+}
 
 #else /* CONFIG_CPU_SUP_AMD */
 
@@ -1212,6 +1264,23 @@ static inline int amd_pmu_init(void)
 	return 0;
 }
 
+static inline int amd_brs_init(void)
+{
+	return 0;
+}
+
+static inline void amd_brs_drain(void)
+{
+}
+
+static inline void amd_brs_enable_all(void)
+{
+}
+
+static inline void amd_brs_disable_all(void)
+{
+}
+
 #endif /* CONFIG_CPU_SUP_AMD */
 
 static inline int is_pebs_pt(struct perf_event *event)
diff --git a/arch/x86/include/asm/msr-index.h b/arch/x86/include/asm/msr-index.h
index 01e2650b9585..bc7481fd8513 100644
--- a/arch/x86/include/asm/msr-index.h
+++ b/arch/x86/include/asm/msr-index.h
@@ -650,6 +650,10 @@
 #define MSR_IA32_PERF_CTL		0x00000199
 #define INTEL_PERF_CTL_MASK		0xffff
 
+/* AMD Branch Sampling configuration */
+#define MSR_AMD_DBG_EXTN_CFG		0xc000010f
+#define MSR_AMD_SAMP_BR_FROM		0xc0010300
+
 #define MSR_IA32_MPERF			0x000000e7
 #define MSR_IA32_APERF			0x000000e8
 
-- 
2.34.0.rc2.393.gf8c9666880-goog

Powered by blists - more mailing lists