Thread (3 messages) 3 messages, 2 authors, 2013-06-25

Re: [PATCH] powerpc/perf: Freeze PMC5/6 if we're not using them on Power8

From: Anshuman Khandual <hidden>
Date: 2013-06-13 06:40:37
Subsystem: linux for powerpc (32-bit and 64-bit), the rest · Maintainers: Madhavan Srinivasan, Michael Ellerman, Linus Torvalds

On 06/13/2013 06:46 AM, Michael Ellerman wrote:
quoted hunk ↗ jump to hunk
On Power8 we can freeze PMC5 and 6 if we're not using them. Normally they
run all the time.

Signed-off-by: Michael Ellerman <redacted>
---
 arch/powerpc/include/asm/reg.h |    1 +
 arch/powerpc/perf/power8-pmu.c |    4 ++++
 2 files changed, 5 insertions(+)
diff --git a/arch/powerpc/include/asm/reg.h b/arch/powerpc/include/asm/reg.h
index 4a9e408..362142b 100644
--- a/arch/powerpc/include/asm/reg.h
+++ b/arch/powerpc/include/asm/reg.h
@@ -626,6 +626,7 @@
 #define   MMCR0_TRIGGER	0x00002000UL /* TRIGGER enable */
 #define   MMCR0_PMAO	0x00000080UL /* performance monitor alert has occurred, set to 0 after handling exception */
 #define   MMCR0_SHRFC	0x00000040UL /* SHRre freeze conditions between threads */
+#define   MMCR0_FC56	0x00000010UL /* freeze counters 5 and 6 */
 #define   MMCR0_FCTI	0x00000008UL /* freeze counters in tags inactive mode */
 #define   MMCR0_FCTA	0x00000004UL /* freeze counters in tags active mode */
 #define   MMCR0_FCWAIT	0x00000002UL /* freeze counter in WAIT state */
diff --git a/arch/powerpc/perf/power8-pmu.c b/arch/powerpc/perf/power8-pmu.c
index f7d1c4f..e791c68 100644
--- a/arch/powerpc/perf/power8-pmu.c
+++ b/arch/powerpc/perf/power8-pmu.c
@@ -378,6 +378,10 @@ static int power8_compute_mmcr(u64 event[], int n_ev,
 	if (pmc_inuse & 0x7c)
 		mmcr[0] |= MMCR0_PMCjCE;

+	/* If we're not using PMC 5 or 6, freeze them */
+	if (!(pmc_inuse & 0x60))
+		mmcr[0] |= MMCR0_FC56;
+
 	mmcr[1] = mmcr1;
 	mmcr[2] = mmcra;
Hey Michael,

This looks good. But we need to undo this changes when we terminate the perf session.
That way user would be able to continue reading PMC5 and PMC6 through /sys interface
as before (which may not be ideal). Adding the following changes along with this patch
would keep the status quo as it is.
diff --git a/arch/powerpc/perf/core-book3s.c b/arch/powerpc/perf/core-book3s.c
index 29c6482..141756a 100644
--- a/arch/powerpc/perf/core-book3s.c
+++ b/arch/powerpc/perf/core-book3s.c
@@ -881,6 +881,12 @@ static void power_pmu_disable(struct pmu *pmu)
 		}
 
 		/*
+ 		 * Undo PMC5/PMC6 freeze if already applied
+ 	 	 */
+		if (mfspr(SPRN_MMCR0) & MMCR0_FC56)
+			mtspr(SPRN_MMCR0, mfspr(SPRN_MMCR0) & ~PMCR0_FC56)
+
+		/*
 		 * Set the 'freeze counters' bit.
 		 * The barrier is to make sure the mtspr has been
 		 * executed and the PMU has frozen the events
Regards
Anshuman






 
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help