Re: [PATCH] perf/x86/intel: Export mem events only if there's PEBs support

From: Jiri Olsa
Date: Thu Sep 06 2018 - 09:57:54 EST


On Tue, Aug 28, 2018 at 10:19:11AM +0200, Jiri Olsa wrote:
> On Tue, Aug 28, 2018 at 10:12:32AM +0200, Peter Zijlstra wrote:
> > On Mon, Aug 27, 2018 at 11:06:24AM +0200, Jiri Olsa wrote:
> > > +static __init struct attribute **get_hsw_events_attrs(bool *alloc)
> > > {
> > > + if (boot_cpu_has(X86_FEATURE_RTM)) {
> > > + *alloc = true;
> > > + return merge_attr(hsw_events_attrs, hsw_tsx_events_attrs);
> > > + }
> > > + return hsw_events_attrs;
> > > }
> >
> >
> >
> > > @@ -4357,6 +4374,15 @@ __init int intel_pmu_init(void)
> > > WARN_ON(!x86_pmu.format_attrs);
> > > }
> > >
> > > + if (x86_pmu.pebs && mem_attr) {
> > > + struct attribute **attr = x86_pmu.cpu_events;
> > > +
> > > + x86_pmu.cpu_events = merge_attr(x86_pmu.cpu_events, mem_attr);
> > > +
> > > + if (alloc_events)
> > > + kfree(attr);
> > > + }
> > > +
> > > if (x86_pmu.num_counters > INTEL_PMC_MAX_GENERIC) {
> > > WARN(1, KERN_ERR "hw perf events %d > max(%d), clipping!",
> > > x86_pmu.num_counters, INTEL_PMC_MAX_GENERIC);
> >
> >
> > static __init struct attribute **
> > get_events_attrs(struct attribute **base,
> > struct attribute **mem,
> > struct attribute **tsx)
> > {
> > struct attribute **attrs = base;
> > struct attribute **old;
> >
> > if (mem) {
> > old = attrs;
> > attrs = merge_attrs(attrs, mem);
> > if (old != base)
> > kfree(old);
> > }
> >
> > if (tsx && boot_cpu_has(X86_FEATURE_RTM)) {
> > old = attrs;
> > attrs = merge_attrs(attrs, tsx);
> > if (old != base)
> > kfree(old);
> > }
> >
> > return attrs;
> > }
> >
> > Would that not help to concentrate things a little more?
>
> looks like it would ;-) will check and repost

new version attached.. Michael tested on several machines,
but I couldn't find haswell with working tsx, to test
that those events are displayed

thanks,
jirka


---
Memory events depends on PEBs support and access to LDLAT msr,
but we display them in /sys/devices/cpu/events even if the cpu
does not provide those, like for KVM guests.

That brings the false assumption that those events should
be available, while they fail event to open.

Separating the mem-* events attributes and merging them
with cpu_events only if there's PEBs support detected.

We could also check if LDLAT msr is available, but the
PEBs check seems to cover the need now.

Suggested-by: Peter Zijlstra <peterz@xxxxxxxxxxxxx>
Signed-off-by: Jiri Olsa <jolsa@xxxxxxxxxx>
---
arch/x86/events/core.c | 8 ++---
arch/x86/events/intel/core.c | 69 +++++++++++++++++++++++++++---------
2 files changed, 56 insertions(+), 21 deletions(-)

diff --git a/arch/x86/events/core.c b/arch/x86/events/core.c
index 5f4829f10129..1a17004f6559 100644
--- a/arch/x86/events/core.c
+++ b/arch/x86/events/core.c
@@ -1631,9 +1631,9 @@ __init struct attribute **merge_attr(struct attribute **a, struct attribute **b)
struct attribute **new;
int j, i;

- for (j = 0; a[j]; j++)
+ for (j = 0; a && a[j]; j++)
;
- for (i = 0; b[i]; i++)
+ for (i = 0; b && b[i]; i++)
j++;
j++;

@@ -1642,9 +1642,9 @@ __init struct attribute **merge_attr(struct attribute **a, struct attribute **b)
return NULL;

j = 0;
- for (i = 0; a[i]; i++)
+ for (i = 0; a && a[i]; i++)
new[j++] = a[i];
- for (i = 0; b[i]; i++)
+ for (i = 0; b && b[i]; i++)
new[j++] = b[i];
new[j] = NULL;

diff --git a/arch/x86/events/intel/core.c b/arch/x86/events/intel/core.c
index 035c37481f57..a10f80f697b7 100644
--- a/arch/x86/events/intel/core.c
+++ b/arch/x86/events/intel/core.c
@@ -242,7 +242,7 @@ EVENT_ATTR_STR(mem-loads, mem_ld_nhm, "event=0x0b,umask=0x10,ldlat=3");
EVENT_ATTR_STR(mem-loads, mem_ld_snb, "event=0xcd,umask=0x1,ldlat=3");
EVENT_ATTR_STR(mem-stores, mem_st_snb, "event=0xcd,umask=0x2");

-static struct attribute *nhm_events_attrs[] = {
+static struct attribute *nhm_mem_events_attrs[] = {
EVENT_PTR(mem_ld_nhm),
NULL,
};
@@ -278,8 +278,6 @@ EVENT_ATTR_STR_HT(topdown-recovery-bubbles.scale, td_recovery_bubbles_scale,
"4", "2");

static struct attribute *snb_events_attrs[] = {
- EVENT_PTR(mem_ld_snb),
- EVENT_PTR(mem_st_snb),
EVENT_PTR(td_slots_issued),
EVENT_PTR(td_slots_retired),
EVENT_PTR(td_fetch_bubbles),
@@ -290,6 +288,12 @@ static struct attribute *snb_events_attrs[] = {
NULL,
};

+static struct attribute *snb_mem_events_attrs[] = {
+ EVENT_PTR(mem_ld_snb),
+ EVENT_PTR(mem_st_snb),
+ NULL,
+};
+
static struct event_constraint intel_hsw_event_constraints[] = {
FIXED_EVENT_CONSTRAINT(0x00c0, 0), /* INST_RETIRED.ANY */
FIXED_EVENT_CONSTRAINT(0x003c, 1), /* CPU_CLK_UNHALTED.CORE */
@@ -3764,8 +3768,6 @@ EVENT_ATTR_STR(cycles-t, cycles_t, "event=0x3c,in_tx=1");
EVENT_ATTR_STR(cycles-ct, cycles_ct, "event=0x3c,in_tx=1,in_tx_cp=1");

static struct attribute *hsw_events_attrs[] = {
- EVENT_PTR(mem_ld_hsw),
- EVENT_PTR(mem_st_hsw),
EVENT_PTR(td_slots_issued),
EVENT_PTR(td_slots_retired),
EVENT_PTR(td_fetch_bubbles),
@@ -3776,6 +3778,12 @@ static struct attribute *hsw_events_attrs[] = {
NULL
};

+static struct attribute *hsw_mem_events_attrs[] = {
+ EVENT_PTR(mem_ld_hsw),
+ EVENT_PTR(mem_st_hsw),
+ NULL,
+};
+
static struct attribute *hsw_tsx_events_attrs[] = {
EVENT_PTR(tx_start),
EVENT_PTR(tx_commit),
@@ -3792,13 +3800,6 @@ static struct attribute *hsw_tsx_events_attrs[] = {
NULL
};

-static __init struct attribute **get_hsw_events_attrs(void)
-{
- return boot_cpu_has(X86_FEATURE_RTM) ?
- merge_attr(hsw_events_attrs, hsw_tsx_events_attrs) :
- hsw_events_attrs;
-}
-
static ssize_t freeze_on_smi_show(struct device *cdev,
struct device_attribute *attr,
char *buf)
@@ -3875,9 +3876,32 @@ static struct attribute *intel_pmu_attrs[] = {
NULL,
};

+static __init struct attribute **
+get_events_attrs(struct attribute **base,
+ struct attribute **mem,
+ struct attribute **tsx)
+{
+ struct attribute **attrs = base;
+ struct attribute **old;
+
+ if (mem && x86_pmu.pebs)
+ attrs = merge_attr(attrs, mem);
+
+ if (tsx && boot_cpu_has(X86_FEATURE_RTM)) {
+ old = attrs;
+ attrs = merge_attr(attrs, tsx);
+ if (old != base)
+ kfree(old);
+ }
+
+ return attrs;
+}
+
__init int intel_pmu_init(void)
{
struct attribute **extra_attr = NULL;
+ struct attribute **mem_attr = NULL;
+ struct attribute **tsx_attr = NULL;
struct attribute **to_free = NULL;
union cpuid10_edx edx;
union cpuid10_eax eax;
@@ -3986,7 +4010,7 @@ __init int intel_pmu_init(void)
x86_pmu.enable_all = intel_pmu_nhm_enable_all;
x86_pmu.extra_regs = intel_nehalem_extra_regs;

- x86_pmu.cpu_events = nhm_events_attrs;
+ mem_attr = nhm_mem_events_attrs;

/* UOPS_ISSUED.STALLED_CYCLES */
intel_perfmon_event_map[PERF_COUNT_HW_STALLED_CYCLES_FRONTEND] =
@@ -4112,7 +4136,7 @@ __init int intel_pmu_init(void)
x86_pmu.extra_regs = intel_westmere_extra_regs;
x86_pmu.flags |= PMU_FL_HAS_RSP_1;

- x86_pmu.cpu_events = nhm_events_attrs;
+ mem_attr = nhm_mem_events_attrs;

/* UOPS_ISSUED.STALLED_CYCLES */
intel_perfmon_event_map[PERF_COUNT_HW_STALLED_CYCLES_FRONTEND] =
@@ -4152,6 +4176,7 @@ __init int intel_pmu_init(void)
x86_pmu.flags |= PMU_FL_NO_HT_SHARING;

x86_pmu.cpu_events = snb_events_attrs;
+ mem_attr = snb_mem_events_attrs;

/* UOPS_ISSUED.ANY,c=1,i=1 to count stall cycles */
intel_perfmon_event_map[PERF_COUNT_HW_STALLED_CYCLES_FRONTEND] =
@@ -4192,6 +4217,7 @@ __init int intel_pmu_init(void)
x86_pmu.flags |= PMU_FL_NO_HT_SHARING;

x86_pmu.cpu_events = snb_events_attrs;
+ mem_attr = snb_mem_events_attrs;

/* UOPS_ISSUED.ANY,c=1,i=1 to count stall cycles */
intel_perfmon_event_map[PERF_COUNT_HW_STALLED_CYCLES_FRONTEND] =
@@ -4226,10 +4252,12 @@ __init int intel_pmu_init(void)

x86_pmu.hw_config = hsw_hw_config;
x86_pmu.get_event_constraints = hsw_get_event_constraints;
- x86_pmu.cpu_events = get_hsw_events_attrs();
+ x86_pmu.cpu_events = hsw_events_attrs;
x86_pmu.lbr_double_abort = true;
extra_attr = boot_cpu_has(X86_FEATURE_RTM) ?
hsw_format_attr : nhm_format_attr;
+ mem_attr = hsw_mem_events_attrs;
+ tsx_attr = hsw_tsx_events_attrs;
pr_cont("Haswell events, ");
name = "haswell";
break;
@@ -4265,10 +4293,12 @@ __init int intel_pmu_init(void)

x86_pmu.hw_config = hsw_hw_config;
x86_pmu.get_event_constraints = hsw_get_event_constraints;
- x86_pmu.cpu_events = get_hsw_events_attrs();
+ x86_pmu.cpu_events = hsw_events_attrs;
x86_pmu.limit_period = bdw_limit_period;
extra_attr = boot_cpu_has(X86_FEATURE_RTM) ?
hsw_format_attr : nhm_format_attr;
+ mem_attr = hsw_mem_events_attrs;
+ tsx_attr = hsw_tsx_events_attrs;
pr_cont("Broadwell events, ");
name = "broadwell";
break;
@@ -4324,7 +4354,9 @@ __init int intel_pmu_init(void)
hsw_format_attr : nhm_format_attr;
extra_attr = merge_attr(extra_attr, skl_format_attr);
to_free = extra_attr;
- x86_pmu.cpu_events = get_hsw_events_attrs();
+ x86_pmu.cpu_events = hsw_events_attrs;
+ mem_attr = hsw_mem_events_attrs;
+ tsx_attr = hsw_tsx_events_attrs;
intel_pmu_pebs_data_source_skl(
boot_cpu_data.x86_model == INTEL_FAM6_SKYLAKE_X);
pr_cont("Skylake events, ");
@@ -4357,6 +4389,9 @@ __init int intel_pmu_init(void)
WARN_ON(!x86_pmu.format_attrs);
}

+ x86_pmu.cpu_events = get_events_attrs(x86_pmu.cpu_events,
+ mem_attr, tsx_attr);
+
if (x86_pmu.num_counters > INTEL_PMC_MAX_GENERIC) {
WARN(1, KERN_ERR "hw perf events %d > max(%d), clipping!",
x86_pmu.num_counters, INTEL_PMC_MAX_GENERIC);
--
2.17.1