The following patchset enables 4 additional performance counters in
AMD family 15h processors that counts northbridge events -- such as
DRAM accesses.
This patchset is based on previous work done by Robert Richter
<[email protected]> :
https://lkml.org/lkml/2012/6/19/324
The main differences are:
- The northbridge counters are indexed contiguously right above the
core performance counters.
- MSR address offset calculations are moved to architecture specific
files.
- Interrups are set up to be delivered only to a single core.
Jacob Shin (3):
perf, amd: Refactor northbridge event constraints handler for code
sharing
perf, x86: Move MSR address offset calculation to architecture
specific files
perf, amd: Enable northbridge performance counters on AMD family 15h
Robert Richter (1):
perf, amd: Simplify northbridge event constraints handler
arch/x86/include/asm/cpufeature.h | 2 +
arch/x86/include/asm/msr-index.h | 2 +
arch/x86/include/asm/perf_event.h | 6 +
arch/x86/kernel/cpu/perf_event.h | 21 +--
arch/x86/kernel/cpu/perf_event_amd.c | 279 +++++++++++++++++++++++-----------
5 files changed, 207 insertions(+), 103 deletions(-)
--
1.7.9.5
From: Robert Richter <[email protected]>
Code simplification, there is no functional change.
Signed-off-by: Robert Richter <[email protected]>
Signed-off-by: Jacob Shin <[email protected]>
---
arch/x86/kernel/cpu/perf_event_amd.c | 68 +++++++++++++---------------------
1 file changed, 26 insertions(+), 42 deletions(-)
diff --git a/arch/x86/kernel/cpu/perf_event_amd.c b/arch/x86/kernel/cpu/perf_event_amd.c
index 4528ae7..d60c5c7 100644
--- a/arch/x86/kernel/cpu/perf_event_amd.c
+++ b/arch/x86/kernel/cpu/perf_event_amd.c
@@ -256,9 +256,8 @@ amd_get_event_constraints(struct cpu_hw_events *cpuc, struct perf_event *event)
{
struct hw_perf_event *hwc = &event->hw;
struct amd_nb *nb = cpuc->amd_nb;
- struct perf_event *old = NULL;
- int max = x86_pmu.num_counters;
- int i, j, k = -1;
+ struct perf_event *old;
+ int idx, new = -1;
/*
* if not NB event or no NB, then no constraints
@@ -276,48 +275,33 @@ amd_get_event_constraints(struct cpu_hw_events *cpuc, struct perf_event *event)
* because of successive calls to x86_schedule_events() from
* hw_perf_group_sched_in() without hw_perf_enable()
*/
- for (i = 0; i < max; i++) {
- /*
- * keep track of first free slot
- */
- if (k == -1 && !nb->owners[i])
- k = i;
+ for (idx = 0; idx < x86_pmu.num_counters; idx++) {
+ if (new == -1 || hwc->idx == idx)
+ /* assign free slot, prefer hwc->idx */
+ old = cmpxchg(nb->owners + idx, NULL, event);
+ else if (nb->owners[idx] == event)
+ /* event already present */
+ old = event;
+ else
+ continue;
+
+ if (old && old != event)
+ continue;
+
+ /* reassign to this slot */
+ if (new != -1)
+ cmpxchg(nb->owners + new, event, NULL);
+ new = idx;
/* already present, reuse */
- if (nb->owners[i] == event)
- goto done;
- }
- /*
- * not present, so grab a new slot
- * starting either at:
- */
- if (hwc->idx != -1) {
- /* previous assignment */
- i = hwc->idx;
- } else if (k != -1) {
- /* start from free slot found */
- i = k;
- } else {
- /*
- * event not found, no slot found in
- * first pass, try again from the
- * beginning
- */
- i = 0;
- }
- j = i;
- do {
- old = cmpxchg(nb->owners+i, NULL, event);
- if (!old)
+ if (old == event)
break;
- if (++i == max)
- i = 0;
- } while (i != j);
-done:
- if (!old)
- return &nb->event_constraints[i];
-
- return &emptyconstraint;
+ }
+
+ if (new == -1)
+ return &emptyconstraint;
+
+ return &nb->event_constraints[new];
}
static struct amd_nb *amd_alloc_nb(int cpu)
--
1.7.9.5
On AMD family 15h processors, there are 4 new performance counters
(in addition to 6 core performance counters) that can be used for
counting northbridge events (i.e. DRAM accesses). Their bit fields are
almost identical to the core performance counters. However, the same
set of MSRs are shared between multiple cores (that share the same
northbridge). We will reuse the same code path as existing family 10h
northbridge event constraints handler logic to enforce sharing.
Based on previous patch by Robert Richter <[email protected]>
Signed-off-by: Jacob Shin <[email protected]>
Signed-off-by: Robert Richter <[email protected]>
---
arch/x86/include/asm/cpufeature.h | 2 +
arch/x86/include/asm/msr-index.h | 2 +
arch/x86/include/asm/perf_event.h | 6 ++
arch/x86/kernel/cpu/perf_event_amd.c | 142 ++++++++++++++++++++++++++--------
4 files changed, 120 insertions(+), 32 deletions(-)
diff --git a/arch/x86/include/asm/cpufeature.h b/arch/x86/include/asm/cpufeature.h
index 8c297aa..17f75b8 100644
--- a/arch/x86/include/asm/cpufeature.h
+++ b/arch/x86/include/asm/cpufeature.h
@@ -167,6 +167,7 @@
#define X86_FEATURE_TBM (6*32+21) /* trailing bit manipulations */
#define X86_FEATURE_TOPOEXT (6*32+22) /* topology extensions CPUID leafs */
#define X86_FEATURE_PERFCTR_CORE (6*32+23) /* core performance counter extensions */
+#define X86_FEATURE_PERFCTR_NB (6*32+24) /* nb performance counter extensions */
/*
* Auxiliary flags: Linux defined - For features scattered in various
@@ -308,6 +309,7 @@ extern const char * const x86_power_flags[32];
#define cpu_has_hypervisor boot_cpu_has(X86_FEATURE_HYPERVISOR)
#define cpu_has_pclmulqdq boot_cpu_has(X86_FEATURE_PCLMULQDQ)
#define cpu_has_perfctr_core boot_cpu_has(X86_FEATURE_PERFCTR_CORE)
+#define cpu_has_perfctr_nb boot_cpu_has(X86_FEATURE_PERFCTR_NB)
#define cpu_has_cx8 boot_cpu_has(X86_FEATURE_CX8)
#define cpu_has_cx16 boot_cpu_has(X86_FEATURE_CX16)
#define cpu_has_eager_fpu boot_cpu_has(X86_FEATURE_EAGER_FPU)
diff --git a/arch/x86/include/asm/msr-index.h b/arch/x86/include/asm/msr-index.h
index 7f0edce..e67ff1e 100644
--- a/arch/x86/include/asm/msr-index.h
+++ b/arch/x86/include/asm/msr-index.h
@@ -157,6 +157,8 @@
/* Fam 15h MSRs */
#define MSR_F15H_PERF_CTL 0xc0010200
#define MSR_F15H_PERF_CTR 0xc0010201
+#define MSR_F15H_NB_PERF_CTL 0xc0010240
+#define MSR_F15H_NB_PERF_CTR 0xc0010241
/* Fam 10h MSRs */
#define MSR_FAM10H_MMIO_CONF_BASE 0xc0010058
diff --git a/arch/x86/include/asm/perf_event.h b/arch/x86/include/asm/perf_event.h
index 4fabcdf..75e039c 100644
--- a/arch/x86/include/asm/perf_event.h
+++ b/arch/x86/include/asm/perf_event.h
@@ -29,6 +29,8 @@
#define ARCH_PERFMON_EVENTSEL_INV (1ULL << 23)
#define ARCH_PERFMON_EVENTSEL_CMASK 0xFF000000ULL
+#define AMD_PERFMON_EVENTSEL_INT_CORE_ENABLE (1ULL << 36)
+#define AMD_PERFMON_EVENTSEL_INT_CORE_SEL_MASK (0x0FULL << 37)
#define AMD_PERFMON_EVENTSEL_GUESTONLY (1ULL << 40)
#define AMD_PERFMON_EVENTSEL_HOSTONLY (1ULL << 41)
@@ -46,8 +48,12 @@
#define AMD64_RAW_EVENT_MASK \
(X86_RAW_EVENT_MASK | \
AMD64_EVENTSEL_EVENT)
+#define AMD64_NB_EVENT_MASK \
+ (AMD64_EVENTSEL_EVENT | \
+ ARCH_PERFMON_EVENTSEL_UMASK)
#define AMD64_NUM_COUNTERS 4
#define AMD64_NUM_COUNTERS_CORE 6
+#define AMD64_NUM_COUNTERS_NB 4
#define ARCH_PERFMON_UNHALTED_CORE_CYCLES_SEL 0x3c
#define ARCH_PERFMON_UNHALTED_CORE_CYCLES_UMASK (0x00 << 8)
diff --git a/arch/x86/kernel/cpu/perf_event_amd.c b/arch/x86/kernel/cpu/perf_event_amd.c
index 078beb5..adf4026 100644
--- a/arch/x86/kernel/cpu/perf_event_amd.c
+++ b/arch/x86/kernel/cpu/perf_event_amd.c
@@ -143,10 +143,15 @@ static unsigned int addr_offsets[X86_PMC_IDX_MAX] __read_mostly;
*
* CPUs with core performance counter extensions:
* 6 counters starting at 0xc0010200 each offset by 2
+ *
+ * CPUs with north bridge performance counter extensions:
+ * 4 additional counters starting at 0xc0010240 each offset by 2
+ * (indexed right above either one of the above core counters)
*/
static inline int amd_pmu_addr_offset(int index)
{
int offset;
+ int ncore;
if (!index)
return index;
@@ -158,8 +163,17 @@ static inline int amd_pmu_addr_offset(int index)
if (!cpu_has_perfctr_core) {
offset = index;
+ ncore = AMD64_NUM_COUNTERS;
} else {
offset = index << 1;
+ ncore = AMD64_NUM_COUNTERS_CORE;
+ }
+
+ /* find offset of NB counters with respect to x86_pmu.eventsel */
+ if (cpu_has_perfctr_nb) {
+ if (index >= ncore && index < (ncore + AMD64_NUM_COUNTERS_NB))
+ offset = (MSR_F15H_NB_PERF_CTL - x86_pmu.eventsel) +
+ ((index - ncore) << 1);
}
addr_offsets[index] = offset;
@@ -167,6 +181,66 @@ static inline int amd_pmu_addr_offset(int index)
return offset;
}
+/*
+ * AMD64 events are detected based on their event codes.
+ */
+static inline unsigned int amd_get_event_code(struct hw_perf_event *hwc)
+{
+ return ((hwc->config >> 24) & 0x0f00) | (hwc->config & 0x00ff);
+}
+
+static inline int amd_is_nb_event(struct hw_perf_event *hwc)
+{
+ return (hwc->config & 0xe0) == 0xe0;
+}
+
+static inline int amd_is_nb_perfctr_event(struct hw_perf_event *hwc)
+{
+ return cpu_has_perfctr_nb && amd_is_nb_event(hwc);
+}
+
+static inline int amd_has_nb(struct cpu_hw_events *cpuc)
+{
+ struct amd_nb *nb = cpuc->amd_nb;
+
+ return nb && nb->nb_id != -1;
+}
+
+/*
+ * AMD NB counters (MSRs 0xc0010240 etc.) do not support the following
+ * flags:
+ *
+ * Host/Guest Only
+ * Counter Mask
+ * Invert Comparison
+ * Edge Detect
+ * Operating-System Mode
+ * User Mode
+ *
+ * Try to fix the config for default settings, otherwise fail.
+ */
+static int amd_nb_event_config(struct perf_event *event)
+{
+ if (!amd_is_nb_perfctr_event(&event->hw))
+ return 0;
+
+ if (event->attr.exclude_host || event->attr.exclude_guest
+ || event->attr.exclude_user || event->attr.exclude_kernel)
+ goto fail;
+
+ event->hw.config &= ~(ARCH_PERFMON_EVENTSEL_USR | ARCH_PERFMON_EVENTSEL_OS);
+
+ if (event->hw.config & ~(AMD64_NB_EVENT_MASK | ARCH_PERFMON_EVENTSEL_INT |
+ AMD_PERFMON_EVENTSEL_INT_CORE_ENABLE |
+ AMD_PERFMON_EVENTSEL_INT_CORE_SEL_MASK))
+ goto fail;
+
+ return 0;
+fail:
+ pr_debug("Invalid nb counter config value: %016Lx\n", event->hw.config);
+ return -EINVAL;
+}
+
static int amd_pmu_hw_config(struct perf_event *event)
{
int ret;
@@ -175,13 +249,13 @@ 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))
+ return -EOPNOTSUPP;
+
ret = x86_pmu_hw_config(event);
if (ret)
return ret;
- if (has_branch_stack(event))
- return -EOPNOTSUPP;
-
if (event->attr.exclude_host && event->attr.exclude_guest)
/*
* When HO == GO == 1 the hardware treats that as GO == HO == 0
@@ -195,32 +269,10 @@ static int amd_pmu_hw_config(struct perf_event *event)
else if (event->attr.exclude_guest)
event->hw.config |= AMD_PERFMON_EVENTSEL_HOSTONLY;
- if (event->attr.type != PERF_TYPE_RAW)
- return 0;
-
- event->hw.config |= event->attr.config & AMD64_RAW_EVENT_MASK;
-
- return 0;
-}
-
-/*
- * AMD64 events are detected based on their event codes.
- */
-static inline unsigned int amd_get_event_code(struct hw_perf_event *hwc)
-{
- return ((hwc->config >> 24) & 0x0f00) | (hwc->config & 0x00ff);
-}
-
-static inline int amd_is_nb_event(struct hw_perf_event *hwc)
-{
- return (hwc->config & 0xe0) == 0xe0;
-}
-
-static inline int amd_has_nb(struct cpu_hw_events *cpuc)
-{
- struct amd_nb *nb = cpuc->amd_nb;
+ if (event->attr.type == PERF_TYPE_RAW)
+ event->hw.config |= event->attr.config & AMD64_RAW_EVENT_MASK;
- return nb && nb->nb_id != -1;
+ return amd_nb_event_config(event);
}
static void __amd_put_nb_event_constraints(struct cpu_hw_events *cpuc,
@@ -324,6 +376,15 @@ __amd_get_nb_event_constraints(struct cpu_hw_events *cpuc, struct perf_event *ev
if (new == -1)
return &emptyconstraint;
+ /* set up interrupts to be delivered only to this core */
+ if (cpu_has_perfctr_nb) {
+ struct cpuinfo_x86 *c = &cpu_data(smp_processor_id());
+
+ hwc->config |= AMD_PERFMON_EVENTSEL_INT_CORE_ENABLE;
+ hwc->config &= ~AMD_PERFMON_EVENTSEL_INT_CORE_SEL_MASK;
+ hwc->config |= (0ULL | (c->cpu_core_id)) << 37;
+ }
+
return &nb->event_constraints[new];
}
@@ -420,7 +481,7 @@ amd_get_event_constraints(struct cpu_hw_events *cpuc, struct perf_event *event)
/*
* if not NB event or no NB, then no constraints
*/
- if ((amd_has_nb(cpuc) && amd_is_nb_event(&event->hw)))
+ if (!(amd_has_nb(cpuc) && amd_is_nb_event(&event->hw)))
return &unconstrained;
return __amd_get_nb_event_constraints(cpuc, event, &unconstrained);
@@ -521,6 +582,7 @@ static struct event_constraint amd_f15_PMC3 = EVENT_CONSTRAINT(0, 0x08, 0);
static struct event_constraint amd_f15_PMC30 = EVENT_CONSTRAINT_OVERLAP(0, 0x09, 0);
static struct event_constraint amd_f15_PMC50 = EVENT_CONSTRAINT(0, 0x3F, 0);
static struct event_constraint amd_f15_PMC53 = EVENT_CONSTRAINT(0, 0x38, 0);
+static struct event_constraint amd_f15_NBPMC30 = EVENT_CONSTRAINT(0, 0x3C0, 0);
static struct event_constraint *
amd_get_event_constraints_f15h(struct cpu_hw_events *cpuc, struct perf_event *event)
@@ -587,8 +649,11 @@ amd_get_event_constraints_f15h(struct cpu_hw_events *cpuc, struct perf_event *ev
return &amd_f15_PMC20;
}
case AMD_EVENT_NB:
- /* not yet implemented */
- return &emptyconstraint;
+ if (cpuc->is_fake)
+ return &amd_f15_NBPMC30;
+
+ return __amd_get_nb_event_constraints(cpuc, event,
+ &amd_f15_NBPMC30);
default:
return &emptyconstraint;
}
@@ -626,7 +691,7 @@ static __initconst const struct x86_pmu amd_pmu = {
static int setup_event_constraints(void)
{
- if (boot_cpu_data.x86 >= 0x15)
+ if (boot_cpu_data.x86 == 0x15)
x86_pmu.get_event_constraints = amd_get_event_constraints_f15h;
return 0;
}
@@ -656,6 +721,18 @@ static int setup_perfctr_core(void)
return 0;
}
+static int setup_perfctr_nb(void)
+{
+ if (!cpu_has_perfctr_nb)
+ return -ENODEV;
+
+ x86_pmu.num_counters += AMD64_NUM_COUNTERS_NB;
+
+ printk(KERN_INFO "perf: AMD northbridge performance counters detected\n");
+
+ return 0;
+}
+
__init int amd_pmu_init(void)
{
/* Performance-monitoring supported from K7 and later: */
@@ -666,6 +743,7 @@ __init int amd_pmu_init(void)
setup_event_constraints();
setup_perfctr_core();
+ setup_perfctr_nb();
/* Events are common for all AMDs */
memcpy(hw_cache_event_ids, amd_hw_cache_event_ids,
--
1.7.9.5
Move counter index to MSR address offset calculation to architecture
specific files. This prepares the way for perf_event_amd to enable
counter addresses that are not contiguous -- for example AMD Family
15h processors have 6 core performance counters starting at 0xc0010200
and 4 northbridge performance counters starting at 0xc0010240.
Signed-off-by: Jacob Shin <[email protected]>
---
arch/x86/kernel/cpu/perf_event.h | 21 +++++---------------
arch/x86/kernel/cpu/perf_event_amd.c | 36 ++++++++++++++++++++++++++++++++++
2 files changed, 41 insertions(+), 16 deletions(-)
diff --git a/arch/x86/kernel/cpu/perf_event.h b/arch/x86/kernel/cpu/perf_event.h
index 271d257..aacf025 100644
--- a/arch/x86/kernel/cpu/perf_event.h
+++ b/arch/x86/kernel/cpu/perf_event.h
@@ -325,6 +325,7 @@ struct x86_pmu {
int (*schedule_events)(struct cpu_hw_events *cpuc, int n, int *assign);
unsigned eventsel;
unsigned perfctr;
+ int (*addr_offset)(int index);
u64 (*event_map)(int);
int max_events;
int num_counters;
@@ -444,28 +445,16 @@ extern u64 __read_mostly hw_cache_extra_regs
u64 x86_perf_event_update(struct perf_event *event);
-static inline int x86_pmu_addr_offset(int index)
-{
- int offset;
-
- /* offset = X86_FEATURE_PERFCTR_CORE ? index << 1 : index */
- alternative_io(ASM_NOP2,
- "shll $1, %%eax",
- X86_FEATURE_PERFCTR_CORE,
- "=a" (offset),
- "a" (index));
-
- return offset;
-}
-
static inline unsigned int x86_pmu_config_addr(int index)
{
- return x86_pmu.eventsel + x86_pmu_addr_offset(index);
+ return x86_pmu.eventsel +
+ (x86_pmu.addr_offset ? x86_pmu.addr_offset(index) : index);
}
static inline unsigned int x86_pmu_event_addr(int index)
{
- return x86_pmu.perfctr + x86_pmu_addr_offset(index);
+ return x86_pmu.perfctr +
+ (x86_pmu.addr_offset ? x86_pmu.addr_offset(index) : index);
}
int x86_setup_perfctr(struct perf_event *event);
diff --git a/arch/x86/kernel/cpu/perf_event_amd.c b/arch/x86/kernel/cpu/perf_event_amd.c
index d17debd..078beb5 100644
--- a/arch/x86/kernel/cpu/perf_event_amd.c
+++ b/arch/x86/kernel/cpu/perf_event_amd.c
@@ -132,6 +132,41 @@ static u64 amd_pmu_event_map(int hw_event)
return amd_perfmon_event_map[hw_event];
}
+/*
+ * Previously calculated offsets
+ */
+static unsigned int addr_offsets[X86_PMC_IDX_MAX] __read_mostly;
+
+/*
+ * Legacy CPUs:
+ * 4 counters starting at 0xc0010000 each offset by 1
+ *
+ * CPUs with core performance counter extensions:
+ * 6 counters starting at 0xc0010200 each offset by 2
+ */
+static inline int amd_pmu_addr_offset(int index)
+{
+ int offset;
+
+ if (!index)
+ return index;
+
+ offset = addr_offsets[index];
+
+ if (offset)
+ return offset;
+
+ if (!cpu_has_perfctr_core) {
+ offset = index;
+ } else {
+ offset = index << 1;
+ }
+
+ addr_offsets[index] = offset;
+
+ return offset;
+}
+
static int amd_pmu_hw_config(struct perf_event *event)
{
int ret;
@@ -570,6 +605,7 @@ static __initconst const struct x86_pmu amd_pmu = {
.schedule_events = x86_schedule_events,
.eventsel = MSR_K7_EVNTSEL0,
.perfctr = MSR_K7_PERFCTR0,
+ .addr_offset = amd_pmu_addr_offset,
.event_map = amd_pmu_event_map,
.max_events = ARRAY_SIZE(amd_perfmon_event_map),
.num_counters = AMD64_NUM_COUNTERS,
--
1.7.9.5
Breakout and generalize family 10h northbridge event contraints code
so that later we can reuse the same code path with other AMD processor
families that have the same northbridge event constraints.
Based on previous patch by Robert Richter <[email protected]>
Signed-off-by: Jacob Shin <[email protected]>
Signed-off-by: Robert Richter <[email protected]>
---
arch/x86/kernel/cpu/perf_event_amd.c | 43 ++++++++++++++++++++--------------
1 file changed, 25 insertions(+), 18 deletions(-)
diff --git a/arch/x86/kernel/cpu/perf_event_amd.c b/arch/x86/kernel/cpu/perf_event_amd.c
index d60c5c7..d17debd 100644
--- a/arch/x86/kernel/cpu/perf_event_amd.c
+++ b/arch/x86/kernel/cpu/perf_event_amd.c
@@ -188,20 +188,13 @@ static inline int amd_has_nb(struct cpu_hw_events *cpuc)
return nb && nb->nb_id != -1;
}
-static void amd_put_event_constraints(struct cpu_hw_events *cpuc,
- struct perf_event *event)
+static void __amd_put_nb_event_constraints(struct cpu_hw_events *cpuc,
+ struct perf_event *event)
{
- struct hw_perf_event *hwc = &event->hw;
struct amd_nb *nb = cpuc->amd_nb;
int i;
/*
- * only care about NB events
- */
- if (!(amd_has_nb(cpuc) && amd_is_nb_event(hwc)))
- return;
-
- /*
* need to scan whole list because event may not have
* been assigned during scheduling
*
@@ -247,12 +240,13 @@ static void amd_put_event_constraints(struct cpu_hw_events *cpuc,
*
* Given that resources are allocated (cmpxchg), they must be
* eventually freed for others to use. This is accomplished by
- * calling amd_put_event_constraints().
+ * calling __amd_put_nb_event_constraints()
*
* Non NB events are not impacted by this restriction.
*/
static struct event_constraint *
-amd_get_event_constraints(struct cpu_hw_events *cpuc, struct perf_event *event)
+__amd_get_nb_event_constraints(struct cpu_hw_events *cpuc, struct perf_event *event,
+ struct event_constraint *c)
{
struct hw_perf_event *hwc = &event->hw;
struct amd_nb *nb = cpuc->amd_nb;
@@ -260,12 +254,6 @@ amd_get_event_constraints(struct cpu_hw_events *cpuc, struct perf_event *event)
int idx, new = -1;
/*
- * if not NB event or no NB, then no constraints
- */
- if (!(amd_has_nb(cpuc) && amd_is_nb_event(hwc)))
- return &unconstrained;
-
- /*
* detect if already present, if so reuse
*
* cannot merge with actual allocation
@@ -275,7 +263,7 @@ amd_get_event_constraints(struct cpu_hw_events *cpuc, struct perf_event *event)
* because of successive calls to x86_schedule_events() from
* hw_perf_group_sched_in() without hw_perf_enable()
*/
- for (idx = 0; idx < x86_pmu.num_counters; idx++) {
+ for_each_set_bit(idx, c->idxmsk, X86_PMC_IDX_MAX) {
if (new == -1 || hwc->idx == idx)
/* assign free slot, prefer hwc->idx */
old = cmpxchg(nb->owners + idx, NULL, event);
@@ -391,6 +379,25 @@ static void amd_pmu_cpu_dead(int cpu)
}
}
+static struct event_constraint *
+amd_get_event_constraints(struct cpu_hw_events *cpuc, struct perf_event *event)
+{
+ /*
+ * if not NB event or no NB, then no constraints
+ */
+ if ((amd_has_nb(cpuc) && amd_is_nb_event(&event->hw)))
+ return &unconstrained;
+
+ return __amd_get_nb_event_constraints(cpuc, event, &unconstrained);
+}
+
+static void amd_put_event_constraints(struct cpu_hw_events *cpuc,
+ struct perf_event *event)
+{
+ if (amd_has_nb(cpuc) && amd_is_nb_event(&event->hw))
+ __amd_put_nb_event_constraints(cpuc, event);
+}
+
PMU_FORMAT_ATTR(event, "config:0-7,32-35");
PMU_FORMAT_ATTR(umask, "config:8-15" );
PMU_FORMAT_ATTR(edge, "config:18" );
--
1.7.9.5
On 09.11.12 19:01:34, Jacob Shin wrote:
> The following patchset enables 4 additional performance counters in
> AMD family 15h processors that counts northbridge events -- such as
> DRAM accesses.
>
> This patchset is based on previous work done by Robert Richter
> <[email protected]> :
>
> https://lkml.org/lkml/2012/6/19/324
The original patch set of this is here (a rebased version):
http://git.kernel.org/?p=linux/kernel/git/rric/oprofile.git;a=shortlog;h=refs/heads/perf-nb
This code was tested in detail.
> The main differences are:
>
> - The northbridge counters are indexed contiguously right above the
> core performance counters.
>
> - MSR address offset calculations are moved to architecture specific
> files.
>
> - Interrups are set up to be delivered only to a single core.
So I rather suggest to make delta patches on top of my patches.
Peter's main concerns were that my patch set is not in the
Intel-uncore style. I started reworking this but was not able to
finish my work. This concerns still exist.
Due to the current situation I would rather prefer to create a
tip:perf/amd-nb branch that includes my patches and then add all
further necessary steps for mainline acceptance on top of it.
Thanks,
-Robert
On Sat, Nov 10, 2012 at 12:50 PM, Robert Richter <[email protected]> wrote:
> On 09.11.12 19:01:34, Jacob Shin wrote:
>> The following patchset enables 4 additional performance counters in
>> AMD family 15h processors that counts northbridge events -- such as
>> DRAM accesses.
>>
>> This patchset is based on previous work done by Robert Richter
>> <[email protected]> :
>>
>> https://lkml.org/lkml/2012/6/19/324
>
> The original patch set of this is here (a rebased version):
>
> http://git.kernel.org/?p=linux/kernel/git/rric/oprofile.git;a=shortlog;h=refs/heads/perf-nb
>
> This code was tested in detail.
>
>> The main differences are:
>>
>> - The northbridge counters are indexed contiguously right above the
>> core performance counters.
>>
>> - MSR address offset calculations are moved to architecture specific
>> files.
>>
>> - Interrups are set up to be delivered only to a single core.
>
> So I rather suggest to make delta patches on top of my patches.
>
> Peter's main concerns were that my patch set is not in the
> Intel-uncore style. I started reworking this but was not able to
> finish my work. This concerns still exist.
>
That was my concern too. I don't recall exactly why it could not
be totally disconnected from the core PMU. I think hardware-wise,
it was possible. Could you refresh my memory?
On Sat, Nov 10, 2012 at 12:50:27PM +0100, Robert Richter wrote:
> On 09.11.12 19:01:34, Jacob Shin wrote:
> > The following patchset enables 4 additional performance counters in
> > AMD family 15h processors that counts northbridge events -- such as
> > DRAM accesses.
> >
> > This patchset is based on previous work done by Robert Richter
> > <[email protected]> :
> >
> > https://lkml.org/lkml/2012/6/19/324
>
> The original patch set of this is here (a rebased version):
>
> http://git.kernel.org/?p=linux/kernel/git/rric/oprofile.git;a=shortlog;h=refs/heads/perf-nb
>
> This code was tested in detail.
>
> > The main differences are:
> >
> > - The northbridge counters are indexed contiguously right above the
> > core performance counters.
> >
> > - MSR address offset calculations are moved to architecture specific
> > files.
> >
> > - Interrups are set up to be delivered only to a single core.
>
> So I rather suggest to make delta patches on top of my patches.
Okay, if we have to, I can rework my patches on top of that, as long
as the end result looks something like I'm suggesting above. Because
in an upcoming processor family, there is no core performance counter
extensions, but we do have northbridge performance counters. Meaning
the counter address base would be c0010000 and northbridge counters
live in c0010240, being 0x240 apart, we could make counter masks work
but that testng awful alot of 0's for every address offset calculation
.
>
> Peter's main concerns were that my patch set is not in the
> Intel-uncore style. I started reworking this but was not able to
> finish my work. This concerns still exist.
Right, I considered this too, and still, I agree with you Robert that
it makes more sense to just extend AMD's x86 PMU.
1. Because the hardware interface -- register bit fields, are alost
identical
2. Because the interrupt delivery mechanism is also identical --
delivered via same APIC interrupt vector.
I think my proposed patchset on top of current Linus's tree is pretty
minimal, and is isolated to AMD so it should be easier to swallow.
Peter, could you take a look at the patchset and if you still prefer
a intel uncore like implementation?
>
> Due to the current situation I would rather prefer to create a
> tip:perf/amd-nb branch that includes my patches and then add all
> further necessary steps for mainline acceptance on top of it.
Okay, Peter, let me know if this is a route to go, and I'll generate
my patchset on top of that.
Thanks,
-Jacob
>
> Thanks,
>
> -Robert
>
Hi,
Anybody from AMD or formerly @ AMD care to submit a libpfm4 patch
to add the Fam15th NB events?
I'd like to avoid having to type them in manually.
Thanks.
On Sun, Nov 11, 2012 at 7:44 PM, Jacob Shin <[email protected]> wrote:
> On Sat, Nov 10, 2012 at 12:50:27PM +0100, Robert Richter wrote:
>> On 09.11.12 19:01:34, Jacob Shin wrote:
>> > The following patchset enables 4 additional performance counters in
>> > AMD family 15h processors that counts northbridge events -- such as
>> > DRAM accesses.
>> >
>> > This patchset is based on previous work done by Robert Richter
>> > <[email protected]> :
>> >
>> > https://lkml.org/lkml/2012/6/19/324
>>
>> The original patch set of this is here (a rebased version):
>>
>> http://git.kernel.org/?p=linux/kernel/git/rric/oprofile.git;a=shortlog;h=refs/heads/perf-nb
>>
>> This code was tested in detail.
>>
>> > The main differences are:
>> >
>> > - The northbridge counters are indexed contiguously right above the
>> > core performance counters.
>> >
>> > - MSR address offset calculations are moved to architecture specific
>> > files.
>> >
>> > - Interrups are set up to be delivered only to a single core.
>>
>> So I rather suggest to make delta patches on top of my patches.
>
> Okay, if we have to, I can rework my patches on top of that, as long
> as the end result looks something like I'm suggesting above. Because
> in an upcoming processor family, there is no core performance counter
> extensions, but we do have northbridge performance counters. Meaning
> the counter address base would be c0010000 and northbridge counters
> live in c0010240, being 0x240 apart, we could make counter masks work
> but that testng awful alot of 0's for every address offset calculation
> .
>
>>
>> Peter's main concerns were that my patch set is not in the
>> Intel-uncore style. I started reworking this but was not able to
>> finish my work. This concerns still exist.
>
> Right, I considered this too, and still, I agree with you Robert that
> it makes more sense to just extend AMD's x86 PMU.
>
> 1. Because the hardware interface -- register bit fields, are alost
> identical
>
> 2. Because the interrupt delivery mechanism is also identical --
> delivered via same APIC interrupt vector.
>
> I think my proposed patchset on top of current Linus's tree is pretty
> minimal, and is isolated to AMD so it should be easier to swallow.
>
> Peter, could you take a look at the patchset and if you still prefer
> a intel uncore like implementation?
>
>>
>> Due to the current situation I would rather prefer to create a
>> tip:perf/amd-nb branch that includes my patches and then add all
>> further necessary steps for mainline acceptance on top of it.
>
> Okay, Peter, let me know if this is a route to go, and I'll generate
> my patchset on top of that.
>
> Thanks,
>
> -Jacob
>
>>
>> Thanks,
>>
>> -Robert
>>
>
Stephane,
On 12.11.12 13:24:38, Stephane Eranian wrote:
> Anybody from AMD or formerly @ AMD care to submit a libpfm4 patch
> to add the Fam15th NB events?
>
> I'd like to avoid having to type them in manually.
Suravee may probably help you here.
HTH
-Robert
On 11.11.12 19:17:07, Stephane Eranian wrote:
> On Sat, Nov 10, 2012 at 12:50 PM, Robert Richter <[email protected]> wrote:
> > Peter's main concerns were that my patch set is not in the
> > Intel-uncore style. I started reworking this but was not able to
> > finish my work. This concerns still exist.
> >
> That was my concern too. I don't recall exactly why it could not
> be totally disconnected from the core PMU. I think hardware-wise,
> it was possible. Could you refresh my memory?
Current implementation only allows the use of a single x86 pmu.
Multiple instances of x86 pmus in a system like an additional pmu for
nb counters requires a complete rework. You need to remove global
variables and extend function interfaces by arguments pointing to a
struct x86_pmu. All this without adding overhead (e.g. pointer
chasing) compared to the existing code.
This is also the reason why the intel-uncore implemenation is
basically a copy-and-paste of generic x86 pmu code plus uncore
specific changes.
Avoiding code cuplication and a complex rework were the reasons for
simply extending the family 10h implementation to also support family
15h nb counters. This seemed to me the best approach that time.
-Robert
Jacob,
On 11.11.12 12:44:26, Jacob Shin wrote:
> Because
> in an upcoming processor family, there is no core performance counter
> extensions, but we do have northbridge performance counters. Meaning
> the counter address base would be c0010000 and northbridge counters
> live in c0010240, being 0x240 apart, we could make counter masks work
> but that testng awful alot of 0's for every address offset calculation
> .
I agree with you that my imlementation of counter masks does not fit
perfectly for this reason and also since it requires too much changes
in generic code (for_each_generic_counter()) which Peter did not like
either. So it is ok to me to replace your solution with the patch that
implenents counter bitmasks.
-Robert
On Mon, Nov 12, 2012 at 03:22:33PM +0100, Robert Richter wrote:
> Stephane,
>
> On 12.11.12 13:24:38, Stephane Eranian wrote:
> > Anybody from AMD or formerly @ AMD care to submit a libpfm4 patch
> > to add the Fam15th NB events?
> >
> > I'd like to avoid having to type them in manually.
>
> Suravee may probably help you here.
Suravee is out of the office until the end of the month,
Is this necessary ASAP for this perf patchset to go upstream? If so,
I'll commit to getting a patch out for libpfm4 soon.
Otherwise, I'll ask Suravee to work on it when he gets back.
Thanks,
-Jacob
>
> HTH
>
> -Robert
>
On Mon, Nov 12, 2012 at 5:13 PM, Jacob Shin <[email protected]> wrote:
> On Mon, Nov 12, 2012 at 03:22:33PM +0100, Robert Richter wrote:
>> Stephane,
>>
>> On 12.11.12 13:24:38, Stephane Eranian wrote:
>> > Anybody from AMD or formerly @ AMD care to submit a libpfm4 patch
>> > to add the Fam15th NB events?
>> >
>> > I'd like to avoid having to type them in manually.
>>
>> Suravee may probably help you here.
>
> Suravee is out of the office until the end of the month,
>
> Is this necessary ASAP for this perf patchset to go upstream? If so,
> I'll commit to getting a patch out for libpfm4 soon.
>
> Otherwise, I'll ask Suravee to work on it when he gets back.
>
Not critical for the kernel part. USeful for user level tools beyond
perf.
> Thanks,
>
> -Jacob
>
>>
>> HTH
>>
>> -Robert
>>
>