Hello
The following patches include some fixes for the p6 PMU driver.
I noticed these when working on the KNC driver which is roughly based
on the p6 driver.
patch1: Fixes a wrong event constraint
patch2: Expands the number of generic events supported by p6
patch3: Cleans up p6_pmu_enable_event() and p6_pmu_disable_event()
Changes around the 2.6.33 to 2.6.34 transition left some
confusing dead code around that was no longer necessary.
I've tested all of these on a Pentium II machine I brought out of
retirement.
Vince Weaver
[email protected]
--
char x,_,*O="yyyyl1f41mm65mm5OU5m5_55m5555mmGUml1ED1ooG_ol\\lDClN:eml9niiobTQO"
"m8dlYoo=HUmmaNEnCDY[l1G@7oo975l1>UImmSG3m5`@Om5<OGm5__5mm4S_l1DTiyyyy";void X(
int x){while(x>-1)printf("\033[4%dm ",x%2*7),x-=2;}int main(){while(*O){for(x=
30;x--;)_=*O-48,X(_>>6?59:(_>>x%6)&1),x*=_<64,O+=!(x%6);X(5),X(0),puts("");}}
According to Intel SDM Volume 3B, FP_ASSIST is limited to Counter 1 only,
not Counter 0.
Tested on a Pentium II.
Signed-off-by: Vince Weaver <[email protected]>
diff -ur linux-3.7-rc1.orig/arch/x86/kernel/cpu/perf_event_p6.c linux-3.7-rc1/arch/x86/kernel/cpu/perf_event_p6.c
--- linux-3.7-rc1.orig/arch/x86/kernel/cpu/perf_event_p6.c 2012-10-14 17:41:04.000000000 -0400
+++ linux-3.7-rc1/arch/x86/kernel/cpu/perf_event_p6.c 2012-10-19 14:50:41.212241569 -0400
@@ -34,7 +34,7 @@
{
INTEL_EVENT_CONSTRAINT(0xc1, 0x1), /* FLOPS */
INTEL_EVENT_CONSTRAINT(0x10, 0x1), /* FP_COMP_OPS_EXE */
- INTEL_EVENT_CONSTRAINT(0x11, 0x1), /* FP_ASSIST */
+ INTEL_EVENT_CONSTRAINT(0x11, 0x2), /* FP_ASSIST */
INTEL_EVENT_CONSTRAINT(0x12, 0x2), /* MUL */
INTEL_EVENT_CONSTRAINT(0x13, 0x2), /* DIV */
INTEL_EVENT_CONSTRAINT(0x14, 0x1), /* CYCLES_DIV_BUSY */
This patch updates the generic events on p6, including some new
extended cache events.
Values for these events were taken from the equivelant PAPI predefined
events.
Tested on a Pentium II.
Signed-off-by: Vince Weaver <[email protected]>
diff -ur linux-3.7-rc1.orig/arch/x86/kernel/cpu/perf_event_p6.c linux-3.7-rc1/arch/x86/kernel/cpu/perf_event_p6.c
--- linux-3.7-rc1.orig/arch/x86/kernel/cpu/perf_event_p6.c 2012-10-14 17:41:04.000000000 -0400
+++ linux-3.7-rc1/arch/x86/kernel/cpu/perf_event_p6.c 2012-10-19 15:34:17.728239395 -0400
@@ -8,13 +8,106 @@
*/
static const u64 p6_perfmon_event_map[] =
{
- [PERF_COUNT_HW_CPU_CYCLES] = 0x0079,
- [PERF_COUNT_HW_INSTRUCTIONS] = 0x00c0,
- [PERF_COUNT_HW_CACHE_REFERENCES] = 0x0f2e,
- [PERF_COUNT_HW_CACHE_MISSES] = 0x012e,
- [PERF_COUNT_HW_BRANCH_INSTRUCTIONS] = 0x00c4,
- [PERF_COUNT_HW_BRANCH_MISSES] = 0x00c5,
- [PERF_COUNT_HW_BUS_CYCLES] = 0x0062,
+ [PERF_COUNT_HW_CPU_CYCLES] = 0x0079, /* CPU_CLK_UNHALTED */
+ [PERF_COUNT_HW_INSTRUCTIONS] = 0x00c0, /* INST_RETIRED */
+ [PERF_COUNT_HW_CACHE_REFERENCES] = 0x0f2e, /* L2_RQSTS:M:E:S:I */
+ [PERF_COUNT_HW_CACHE_MISSES] = 0x012e, /* L2_RQSTS:I */
+ [PERF_COUNT_HW_BRANCH_INSTRUCTIONS] = 0x00c4, /* BR_INST_RETIRED */
+ [PERF_COUNT_HW_BRANCH_MISSES] = 0x00c5, /* BR_MISS_PRED_RETIRED */
+ [PERF_COUNT_HW_BUS_CYCLES] = 0x0062, /* BUS_DRDY_CLOCKS */
+ [PERF_COUNT_HW_STALLED_CYCLES_FRONTEND] = 0x00a2, /* RESOURCE_STALLS */
+
+};
+
+static __initconst u64 p6_hw_cache_event_ids
+ [PERF_COUNT_HW_CACHE_MAX]
+ [PERF_COUNT_HW_CACHE_OP_MAX]
+ [PERF_COUNT_HW_CACHE_RESULT_MAX] =
+{
+ [ C(L1D) ] = {
+ [ C(OP_READ) ] = {
+ [ C(RESULT_ACCESS) ] = 0x0043, /* DATA_MEM_REFS */
+ [ C(RESULT_MISS) ] = 0x0045, /* DCU_LINES_IN */
+ },
+ [ C(OP_WRITE) ] = {
+ [ C(RESULT_ACCESS) ] = 0,
+ [ C(RESULT_MISS) ] = 0x0f29, /* L2_LD:M:E:S:I */
+ },
+ [ C(OP_PREFETCH) ] = {
+ [ C(RESULT_ACCESS) ] = 0,
+ [ C(RESULT_MISS) ] = 0,
+ },
+ },
+ [ C(L1I ) ] = {
+ [ C(OP_READ) ] = {
+ [ C(RESULT_ACCESS) ] = 0x0080, /* IFU_IFETCH */
+ [ C(RESULT_MISS) ] = 0x0f28, /* L2_IFETCH:M:E:S:I */
+ },
+ [ C(OP_WRITE) ] = {
+ [ C(RESULT_ACCESS) ] = -1,
+ [ C(RESULT_MISS) ] = -1,
+ },
+ [ C(OP_PREFETCH) ] = {
+ [ C(RESULT_ACCESS) ] = 0,
+ [ C(RESULT_MISS) ] = 0,
+ },
+ },
+ [ C(LL ) ] = {
+ [ C(OP_READ) ] = {
+ [ C(RESULT_ACCESS) ] = 0,
+ [ C(RESULT_MISS) ] = 0,
+ },
+ [ C(OP_WRITE) ] = {
+ [ C(RESULT_ACCESS) ] = 0,
+ [ C(RESULT_MISS) ] = 0x0025, /* L2_M_LINES_INM */
+ },
+ [ C(OP_PREFETCH) ] = {
+ [ C(RESULT_ACCESS) ] = 0,
+ [ C(RESULT_MISS) ] = 0,
+ },
+ },
+ [ C(DTLB) ] = {
+ [ C(OP_READ) ] = {
+ [ C(RESULT_ACCESS) ] = 0x0043, /* DATA_MEM_REFS */
+ [ C(RESULT_MISS) ] = 0,
+ },
+ [ C(OP_WRITE) ] = {
+ [ C(RESULT_ACCESS) ] = 0,
+ [ C(RESULT_MISS) ] = 0,
+ },
+ [ C(OP_PREFETCH) ] = {
+ [ C(RESULT_ACCESS) ] = 0,
+ [ C(RESULT_MISS) ] = 0,
+ },
+ },
+ [ C(ITLB) ] = {
+ [ C(OP_READ) ] = {
+ [ C(RESULT_ACCESS) ] = 0x0080, /* IFU_IFETCH */
+ [ C(RESULT_MISS) ] = 0x0085, /* ITLB_MISS */
+ },
+ [ C(OP_WRITE) ] = {
+ [ C(RESULT_ACCESS) ] = -1,
+ [ C(RESULT_MISS) ] = -1,
+ },
+ [ C(OP_PREFETCH) ] = {
+ [ C(RESULT_ACCESS) ] = -1,
+ [ C(RESULT_MISS) ] = -1,
+ },
+ },
+ [ C(BPU ) ] = {
+ [ C(OP_READ) ] = {
+ [ C(RESULT_ACCESS) ] = 0x00c4, /* BR_INST_RETIRED */
+ [ C(RESULT_MISS) ] = 0x00c5, /* BR_MISS_PRED_RETIRED */
+ },
+ [ C(OP_WRITE) ] = {
+ [ C(RESULT_ACCESS) ] = -1,
+ [ C(RESULT_MISS) ] = -1,
+ },
+ [ C(OP_PREFETCH) ] = {
+ [ C(RESULT_ACCESS) ] = -1,
+ [ C(RESULT_MISS) ] = -1,
+ },
+ },
};
static u64 p6_pmu_event_map(int hw_event)
@@ -158,5 +251,9 @@
x86_pmu = p6_pmu;
+ memcpy(hw_cache_event_ids, p6_hw_cache_event_ids,
+ sizeof(hw_cache_event_ids));
+
+
return 0;
}
Between 2.6.33 and 2.6.34 the PMU code was made modular.
The x86_pmu_enable() call was extended to disable cpuc->enabled
and iterate the counters, enabling one at a time, before calling
enable_all() at the end, followed by re-enabling cpuc->enabled.
Since cpuc->enabled was set to 0, that change effectively caused
the "val |= ARCH_PERFMON_EVENTSEL_ENABLE;" code in p6_pmu_enable_event()
and p6_pmu_disable_event() to be dead code that was never called.
This change removes this code (which was confusing) and adds some
extra commentary to make it more clear what is going on.
Signed-off-by: Vince Weaver <[email protected]>
diff -ur linux-3.7-rc1.orig/arch/x86/kernel/cpu/perf_event_p6.c linux-3.7-rc1/arch/x86/kernel/cpu/perf_event_p6.c
--- linux-3.7-rc1.orig/arch/x86/kernel/cpu/perf_event_p6.c 2012-10-14 17:41:04.000000000 -0400
+++ linux-3.7-rc1/arch/x86/kernel/cpu/perf_event_p6.c 2012-10-19 15:58:50.416238172 -0400
@@ -64,25 +64,25 @@
static inline void
p6_pmu_disable_event(struct perf_event *event)
{
- struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
struct hw_perf_event *hwc = &event->hw;
u64 val = P6_NOP_EVENT;
- if (cpuc->enabled)
- val |= ARCH_PERFMON_EVENTSEL_ENABLE;
-
(void)wrmsrl_safe(hwc->config_base, val);
}
static void p6_pmu_enable_event(struct perf_event *event)
{
- struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
struct hw_perf_event *hwc = &event->hw;
u64 val;
val = hwc->config;
- if (cpuc->enabled)
- val |= ARCH_PERFMON_EVENTSEL_ENABLE;
+
+ /*
+ * p6 only has a global event enable, set on PerfEvtSel0
+ * We "disable" events by programming P6_NOP_EVENT
+ * and we rely on p6_pmu_enable_all() being called
+ * to actually enable the events.
+ */
(void)wrmsrl_safe(hwc->config_base, val);
}
Commit-ID: 7991c9ca40d3127dd2ffa3a9c1e33f7d4005495a
Gitweb: http://git.kernel.org/tip/7991c9ca40d3127dd2ffa3a9c1e33f7d4005495a
Author: Vince Weaver <[email protected]>
AuthorDate: Fri, 19 Oct 2012 17:30:01 -0400
Committer: Ingo Molnar <[email protected]>
CommitDate: Wed, 24 Oct 2012 10:31:57 +0200
perf/x86: Fix P6 FP_ASSIST event constraint
According to Intel SDM Volume 3B, FP_ASSIST is limited to Counter 1 only,
not Counter 0.
Tested on a Pentium II.
Signed-off-by: Vince Weaver <[email protected]>
Signed-off-by: Peter Zijlstra <[email protected]>
Link: http://lkml.kernel.org/r/alpine.DEB.2.02.1210191728570.14552@vincent-weaver-1.um.maine.edu
Signed-off-by: Ingo Molnar <[email protected]>
---
arch/x86/kernel/cpu/perf_event_p6.c | 2 +-
1 files changed, 1 insertions(+), 1 deletions(-)
diff --git a/arch/x86/kernel/cpu/perf_event_p6.c b/arch/x86/kernel/cpu/perf_event_p6.c
index e4dd0f7..0ff5f7f 100644
--- a/arch/x86/kernel/cpu/perf_event_p6.c
+++ b/arch/x86/kernel/cpu/perf_event_p6.c
@@ -34,7 +34,7 @@ static struct event_constraint p6_event_constraints[] =
{
INTEL_EVENT_CONSTRAINT(0xc1, 0x1), /* FLOPS */
INTEL_EVENT_CONSTRAINT(0x10, 0x1), /* FP_COMP_OPS_EXE */
- INTEL_EVENT_CONSTRAINT(0x11, 0x1), /* FP_ASSIST */
+ INTEL_EVENT_CONSTRAINT(0x11, 0x2), /* FP_ASSIST */
INTEL_EVENT_CONSTRAINT(0x12, 0x2), /* MUL */
INTEL_EVENT_CONSTRAINT(0x13, 0x2), /* DIV */
INTEL_EVENT_CONSTRAINT(0x14, 0x1), /* CYCLES_DIV_BUSY */
Commit-ID: e09df47885d767e418902067ce1885aafa3b27db
Gitweb: http://git.kernel.org/tip/e09df47885d767e418902067ce1885aafa3b27db
Author: Vince Weaver <[email protected]>
AuthorDate: Fri, 19 Oct 2012 17:31:54 -0400
Committer: Ingo Molnar <[email protected]>
CommitDate: Wed, 24 Oct 2012 10:31:58 +0200
perf/x86: Update/fix generic events on P6 PMU
This patch updates the generic events on p6, including some new
extended cache events.
Values for these events were taken from the equivelant PAPI
predefined events.
Tested on a Pentium II.
Signed-off-by: Vince Weaver <[email protected]>
Signed-off-by: Peter Zijlstra <[email protected]>
Link: http://lkml.kernel.org/r/alpine.DEB.2.02.1210191730080.14552@vincent-weaver-1.um.maine.edu
Signed-off-by: Ingo Molnar <[email protected]>
---
arch/x86/kernel/cpu/perf_event_p6.c | 111 ++++++++++++++++++++++++++++++++--
1 files changed, 104 insertions(+), 7 deletions(-)
diff --git a/arch/x86/kernel/cpu/perf_event_p6.c b/arch/x86/kernel/cpu/perf_event_p6.c
index 0ff5f7f..9582fcb 100644
--- a/arch/x86/kernel/cpu/perf_event_p6.c
+++ b/arch/x86/kernel/cpu/perf_event_p6.c
@@ -8,13 +8,106 @@
*/
static const u64 p6_perfmon_event_map[] =
{
- [PERF_COUNT_HW_CPU_CYCLES] = 0x0079,
- [PERF_COUNT_HW_INSTRUCTIONS] = 0x00c0,
- [PERF_COUNT_HW_CACHE_REFERENCES] = 0x0f2e,
- [PERF_COUNT_HW_CACHE_MISSES] = 0x012e,
- [PERF_COUNT_HW_BRANCH_INSTRUCTIONS] = 0x00c4,
- [PERF_COUNT_HW_BRANCH_MISSES] = 0x00c5,
- [PERF_COUNT_HW_BUS_CYCLES] = 0x0062,
+ [PERF_COUNT_HW_CPU_CYCLES] = 0x0079, /* CPU_CLK_UNHALTED */
+ [PERF_COUNT_HW_INSTRUCTIONS] = 0x00c0, /* INST_RETIRED */
+ [PERF_COUNT_HW_CACHE_REFERENCES] = 0x0f2e, /* L2_RQSTS:M:E:S:I */
+ [PERF_COUNT_HW_CACHE_MISSES] = 0x012e, /* L2_RQSTS:I */
+ [PERF_COUNT_HW_BRANCH_INSTRUCTIONS] = 0x00c4, /* BR_INST_RETIRED */
+ [PERF_COUNT_HW_BRANCH_MISSES] = 0x00c5, /* BR_MISS_PRED_RETIRED */
+ [PERF_COUNT_HW_BUS_CYCLES] = 0x0062, /* BUS_DRDY_CLOCKS */
+ [PERF_COUNT_HW_STALLED_CYCLES_FRONTEND] = 0x00a2, /* RESOURCE_STALLS */
+
+};
+
+static __initconst u64 p6_hw_cache_event_ids
+ [PERF_COUNT_HW_CACHE_MAX]
+ [PERF_COUNT_HW_CACHE_OP_MAX]
+ [PERF_COUNT_HW_CACHE_RESULT_MAX] =
+{
+ [ C(L1D) ] = {
+ [ C(OP_READ) ] = {
+ [ C(RESULT_ACCESS) ] = 0x0043, /* DATA_MEM_REFS */
+ [ C(RESULT_MISS) ] = 0x0045, /* DCU_LINES_IN */
+ },
+ [ C(OP_WRITE) ] = {
+ [ C(RESULT_ACCESS) ] = 0,
+ [ C(RESULT_MISS) ] = 0x0f29, /* L2_LD:M:E:S:I */
+ },
+ [ C(OP_PREFETCH) ] = {
+ [ C(RESULT_ACCESS) ] = 0,
+ [ C(RESULT_MISS) ] = 0,
+ },
+ },
+ [ C(L1I ) ] = {
+ [ C(OP_READ) ] = {
+ [ C(RESULT_ACCESS) ] = 0x0080, /* IFU_IFETCH */
+ [ C(RESULT_MISS) ] = 0x0f28, /* L2_IFETCH:M:E:S:I */
+ },
+ [ C(OP_WRITE) ] = {
+ [ C(RESULT_ACCESS) ] = -1,
+ [ C(RESULT_MISS) ] = -1,
+ },
+ [ C(OP_PREFETCH) ] = {
+ [ C(RESULT_ACCESS) ] = 0,
+ [ C(RESULT_MISS) ] = 0,
+ },
+ },
+ [ C(LL ) ] = {
+ [ C(OP_READ) ] = {
+ [ C(RESULT_ACCESS) ] = 0,
+ [ C(RESULT_MISS) ] = 0,
+ },
+ [ C(OP_WRITE) ] = {
+ [ C(RESULT_ACCESS) ] = 0,
+ [ C(RESULT_MISS) ] = 0x0025, /* L2_M_LINES_INM */
+ },
+ [ C(OP_PREFETCH) ] = {
+ [ C(RESULT_ACCESS) ] = 0,
+ [ C(RESULT_MISS) ] = 0,
+ },
+ },
+ [ C(DTLB) ] = {
+ [ C(OP_READ) ] = {
+ [ C(RESULT_ACCESS) ] = 0x0043, /* DATA_MEM_REFS */
+ [ C(RESULT_MISS) ] = 0,
+ },
+ [ C(OP_WRITE) ] = {
+ [ C(RESULT_ACCESS) ] = 0,
+ [ C(RESULT_MISS) ] = 0,
+ },
+ [ C(OP_PREFETCH) ] = {
+ [ C(RESULT_ACCESS) ] = 0,
+ [ C(RESULT_MISS) ] = 0,
+ },
+ },
+ [ C(ITLB) ] = {
+ [ C(OP_READ) ] = {
+ [ C(RESULT_ACCESS) ] = 0x0080, /* IFU_IFETCH */
+ [ C(RESULT_MISS) ] = 0x0085, /* ITLB_MISS */
+ },
+ [ C(OP_WRITE) ] = {
+ [ C(RESULT_ACCESS) ] = -1,
+ [ C(RESULT_MISS) ] = -1,
+ },
+ [ C(OP_PREFETCH) ] = {
+ [ C(RESULT_ACCESS) ] = -1,
+ [ C(RESULT_MISS) ] = -1,
+ },
+ },
+ [ C(BPU ) ] = {
+ [ C(OP_READ) ] = {
+ [ C(RESULT_ACCESS) ] = 0x00c4, /* BR_INST_RETIRED */
+ [ C(RESULT_MISS) ] = 0x00c5, /* BR_MISS_PRED_RETIRED */
+ },
+ [ C(OP_WRITE) ] = {
+ [ C(RESULT_ACCESS) ] = -1,
+ [ C(RESULT_MISS) ] = -1,
+ },
+ [ C(OP_PREFETCH) ] = {
+ [ C(RESULT_ACCESS) ] = -1,
+ [ C(RESULT_MISS) ] = -1,
+ },
+ },
};
static u64 p6_pmu_event_map(int hw_event)
@@ -158,5 +251,9 @@ __init int p6_pmu_init(void)
x86_pmu = p6_pmu;
+ memcpy(hw_cache_event_ids, p6_hw_cache_event_ids,
+ sizeof(hw_cache_event_ids));
+
+
return 0;
}
Commit-ID: 58e9eaf06f5476cb2192ec1d012674ce5e79dd21
Gitweb: http://git.kernel.org/tip/58e9eaf06f5476cb2192ec1d012674ce5e79dd21
Author: Vince Weaver <[email protected]>
AuthorDate: Fri, 19 Oct 2012 17:33:38 -0400
Committer: Ingo Molnar <[email protected]>
CommitDate: Wed, 24 Oct 2012 10:32:00 +0200
perf/x86: Remove P6 cpuc->enabled check
Between 2.6.33 and 2.6.34 the PMU code was made modular.
The x86_pmu_enable() call was extended to disable cpuc->enabled
and iterate the counters, enabling one at a time, before calling
enable_all() at the end, followed by re-enabling cpuc->enabled.
Since cpuc->enabled was set to 0, that change effectively caused
the "val |= ARCH_PERFMON_EVENTSEL_ENABLE;" code in p6_pmu_enable_event()
and p6_pmu_disable_event() to be dead code that was never called.
This change removes this code (which was confusing) and adds some
extra commentary to make it more clear what is going on.
Signed-off-by: Vince Weaver <[email protected]>
Signed-off-by: Peter Zijlstra <[email protected]>
Link: http://lkml.kernel.org/r/alpine.DEB.2.02.1210191732000.14552@vincent-weaver-1.um.maine.edu
Signed-off-by: Ingo Molnar <[email protected]>
---
arch/x86/kernel/cpu/perf_event_p6.c | 14 +++++++-------
1 files changed, 7 insertions(+), 7 deletions(-)
diff --git a/arch/x86/kernel/cpu/perf_event_p6.c b/arch/x86/kernel/cpu/perf_event_p6.c
index 9582fcb..7d0270b 100644
--- a/arch/x86/kernel/cpu/perf_event_p6.c
+++ b/arch/x86/kernel/cpu/perf_event_p6.c
@@ -157,25 +157,25 @@ static void p6_pmu_enable_all(int added)
static inline void
p6_pmu_disable_event(struct perf_event *event)
{
- struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
struct hw_perf_event *hwc = &event->hw;
u64 val = P6_NOP_EVENT;
- if (cpuc->enabled)
- val |= ARCH_PERFMON_EVENTSEL_ENABLE;
-
(void)wrmsrl_safe(hwc->config_base, val);
}
static void p6_pmu_enable_event(struct perf_event *event)
{
- struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events);
struct hw_perf_event *hwc = &event->hw;
u64 val;
val = hwc->config;
- if (cpuc->enabled)
- val |= ARCH_PERFMON_EVENTSEL_ENABLE;
+
+ /*
+ * p6 only has a global event enable, set on PerfEvtSel0
+ * We "disable" events by programming P6_NOP_EVENT
+ * and we rely on p6_pmu_enable_all() being called
+ * to actually enable the events.
+ */
(void)wrmsrl_safe(hwc->config_base, val);
}