These patches add support for the Scorpion PMU found on devices
such as msm8660, qsd8x50, etc. The first patch is some groundwork
to make functions more "generic". Even then we end up copying quite
a bit of code from the Krait part into the Scorpion part with only
subtle tweaks because two things happen:
1. We gain another "region" register, i.e. L2LPM
2. The way to access the region registers are with different cp15 instructions
I'm not sure how to make this better, suggestions are welcome.
Stephen Boyd (2):
ARM: perf: Preparatory work for Scorpion PMU support
ARM: perf: Add support for Scorpion PMUs
Documentation/devicetree/bindings/arm/pmu.txt | 2 +
arch/arm/kernel/perf_event_cpu.c | 2 +
arch/arm/kernel/perf_event_v7.c | 482 +++++++++++++++++++++++---
3 files changed, 445 insertions(+), 41 deletions(-)
--
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
a Linux Foundation Collaborative Project
Do some things to make the Krait PMU support code generic enough
to be used by the Scorpion PMU support code.
* Rename the venum register functions to be venum instead of krait
specific because the same registers exist on Scorpion
* Add a krait_decode_event() function to decode our Krait specific
event encoding that's the same on Scorpion (modulo an extra region).
* Drop krait from krait_clear_pmresrn_group() so it can be used
by Scorpion code
* Drop a redundant check in krait_pmu_get_event_idx()
Signed-off-by: Stephen Boyd <[email protected]>
---
arch/arm/kernel/perf_event_v7.c | 87 ++++++++++++++++++++++-------------------
1 file changed, 46 insertions(+), 41 deletions(-)
diff --git a/arch/arm/kernel/perf_event_v7.c b/arch/arm/kernel/perf_event_v7.c
index 8993770c47de..84a3ec3bc592 100644
--- a/arch/arm/kernel/perf_event_v7.c
+++ b/arch/arm/kernel/perf_event_v7.c
@@ -1141,19 +1141,19 @@ static void krait_write_pmresrn(int n, u32 val)
}
}
-static u32 krait_read_vpmresr0(void)
+static u32 venum_read_pmresr(void)
{
u32 val;
asm volatile("mrc p10, 7, %0, c11, c0, 0" : "=r" (val));
return val;
}
-static void krait_write_vpmresr0(u32 val)
+static void venum_write_pmresr(u32 val)
{
asm volatile("mcr p10, 7, %0, c11, c0, 0" : : "r" (val));
}
-static void krait_pre_vpmresr0(u32 *venum_orig_val, u32 *fp_orig_val)
+static void venum_pre_pmresr(u32 *venum_orig_val, u32 *fp_orig_val)
{
u32 venum_new_val;
u32 fp_new_val;
@@ -1170,7 +1170,7 @@ static void krait_pre_vpmresr0(u32 *venum_orig_val, u32 *fp_orig_val)
fmxr(FPEXC, fp_new_val);
}
-static void krait_post_vpmresr0(u32 venum_orig_val, u32 fp_orig_val)
+static void venum_post_pmresr(u32 venum_orig_val, u32 fp_orig_val)
{
BUG_ON(preemptible());
/* Restore FPEXC */
@@ -1188,6 +1188,21 @@ static u32 krait_get_pmresrn_event(unsigned int region)
return pmresrn_table[region];
}
+static void krait_decode_event(u32 event, unsigned int *region,
+ unsigned int *group, unsigned int *code, bool *venum, bool *cpu)
+{
+ if (region)
+ *region = (event >> 12) & 0xf;
+ if (group)
+ *group = (event >> 0) & 0xf;
+ if (code)
+ *code = (event >> 4) & 0xff;
+ if (venum)
+ *venum = !!(event & VENUM_EVENT);
+ if (cpu)
+ *cpu = !!(event & KRAIT_EVENT);
+}
+
static void krait_evt_setup(int idx, u32 config_base)
{
u32 val;
@@ -1199,10 +1214,8 @@ static void krait_evt_setup(int idx, u32 config_base)
unsigned int group_shift;
bool venum_event;
- venum_event = !!(config_base & VENUM_EVENT);
- region = (config_base >> 12) & 0xf;
- code = (config_base >> 4) & 0xff;
- group = (config_base >> 0) & 0xf;
+ krait_decode_event(config_base, ®ion, &group, &code, &venum_event,
+ NULL);
group_shift = group * 8;
mask = 0xff << group_shift;
@@ -1220,13 +1233,13 @@ static void krait_evt_setup(int idx, u32 config_base)
asm volatile("mcr p15, 0, %0, c9, c15, 0" : : "r" (0));
if (venum_event) {
- krait_pre_vpmresr0(&vval, &fval);
- val = krait_read_vpmresr0();
+ venum_pre_pmresr(&vval, &fval);
+ val = venum_read_pmresr();
val &= ~mask;
val |= code << group_shift;
val |= PMRESRn_EN;
- krait_write_vpmresr0(val);
- krait_post_vpmresr0(vval, fval);
+ venum_write_pmresr(val);
+ venum_post_pmresr(vval, fval);
} else {
val = krait_read_pmresrn(region);
val &= ~mask;
@@ -1236,7 +1249,7 @@ static void krait_evt_setup(int idx, u32 config_base)
}
}
-static u32 krait_clear_pmresrn_group(u32 val, int group)
+static u32 clear_pmresrn_group(u32 val, int group)
{
u32 mask;
int group_shift;
@@ -1260,19 +1273,18 @@ static void krait_clearpmu(u32 config_base)
unsigned int group;
bool venum_event;
- venum_event = !!(config_base & VENUM_EVENT);
- region = (config_base >> 12) & 0xf;
- group = (config_base >> 0) & 0xf;
+ krait_decode_event(config_base, ®ion, &group, NULL, &venum_event,
+ NULL);
if (venum_event) {
- krait_pre_vpmresr0(&vval, &fval);
- val = krait_read_vpmresr0();
- val = krait_clear_pmresrn_group(val, group);
- krait_write_vpmresr0(val);
- krait_post_vpmresr0(vval, fval);
+ venum_pre_pmresr(&vval, &fval);
+ val = venum_read_pmresr();
+ val = clear_pmresrn_group(val, group);
+ venum_write_pmresr(val);
+ venum_post_pmresr(vval, fval);
} else {
val = krait_read_pmresrn(region);
- val = krait_clear_pmresrn_group(val, group);
+ val = clear_pmresrn_group(val, group);
krait_write_pmresrn(region, val);
}
}
@@ -1350,9 +1362,9 @@ static void krait_pmu_reset(void *info)
krait_write_pmresrn(1, 0);
krait_write_pmresrn(2, 0);
- krait_pre_vpmresr0(&vval, &fval);
- krait_write_vpmresr0(0);
- krait_post_vpmresr0(vval, fval);
+ venum_pre_pmresr(&vval, &fval);
+ venum_write_pmresr(0);
+ venum_post_pmresr(vval, fval);
}
static int krait_event_to_bit(struct perf_event *event, unsigned int region,
@@ -1386,26 +1398,20 @@ static int krait_pmu_get_event_idx(struct pmu_hw_events *cpuc,
{
int idx;
int bit = -1;
- unsigned int prefix;
unsigned int region;
unsigned int code;
unsigned int group;
- bool krait_event;
+ bool venum_event, krait_event;
struct hw_perf_event *hwc = &event->hw;
- region = (hwc->config_base >> 12) & 0xf;
- code = (hwc->config_base >> 4) & 0xff;
- group = (hwc->config_base >> 0) & 0xf;
- krait_event = !!(hwc->config_base & KRAIT_EVENT_MASK);
+ krait_decode_event(hwc->config_base, ®ion, &group, &code,
+ &venum_event, &krait_event);
- if (krait_event) {
+ if (venum_event || krait_event) {
/* Ignore invalid events */
if (group > 3 || region > 2)
return -EINVAL;
- prefix = hwc->config_base & KRAIT_EVENT_MASK;
- if (prefix != KRAIT_EVENT && prefix != VENUM_EVENT)
- return -EINVAL;
- if (prefix == VENUM_EVENT && (code & 0xe0))
+ if (venum_event && (code & 0xe0))
return -EINVAL;
bit = krait_event_to_bit(event, region, group);
@@ -1427,13 +1433,12 @@ static void krait_pmu_clear_event_idx(struct pmu_hw_events *cpuc,
struct hw_perf_event *hwc = &event->hw;
unsigned int region;
unsigned int group;
- bool krait_event;
+ bool venum_event, krait_event;
- region = (hwc->config_base >> 12) & 0xf;
- group = (hwc->config_base >> 0) & 0xf;
- krait_event = !!(hwc->config_base & KRAIT_EVENT_MASK);
+ krait_decode_event(hwc->config_base, ®ion, &group, NULL,
+ &venum_event, &krait_event);
- if (krait_event) {
+ if (venum_event || krait_event) {
bit = krait_event_to_bit(event, region, group);
clear_bit(bit, cpuc->used_mask);
}
--
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
a Linux Foundation Collaborative Project
Scorpion supports a set of local performance monitor event
selection registers (LPM) sitting behind a cp15 based interface
that extend the architected PMU events to include Scorpion CPU
and Venum VFP specific events. To use these events the user is
expected to program the lpm register with the event code shifted
into the group they care about and then point the PMNx event at
that region+group combo by writing a LPMn_GROUPx event. Add
support for this hardware.
Note: the raw event number is a pure software construct that
allows us to map the multi-dimensional number space of regions,
groups, and event codes into a flat event number space suitable
for use by the perf framework.
This is based on code originally written by Ashwin Chaugule and
Neil Leeder [1] massed to become similar to the Krait PMU support
code.
[1] https://www.codeaurora.org/cgit/quic/la/kernel/msm/tree/arch/arm/kernel/perf_event_msm.c?h=msm-3.4
Cc: Neil Leeder <[email protected]>
Cc: Ashwin Chaugule <[email protected]>
Cc: <[email protected]>
Signed-off-by: Stephen Boyd <[email protected]>
---
Documentation/devicetree/bindings/arm/pmu.txt | 2 +
arch/arm/kernel/perf_event_cpu.c | 2 +
arch/arm/kernel/perf_event_v7.c | 395 ++++++++++++++++++++++++++
3 files changed, 399 insertions(+)
diff --git a/Documentation/devicetree/bindings/arm/pmu.txt b/Documentation/devicetree/bindings/arm/pmu.txt
index 75ef91d08f3b..6e54a9d88b7a 100644
--- a/Documentation/devicetree/bindings/arm/pmu.txt
+++ b/Documentation/devicetree/bindings/arm/pmu.txt
@@ -18,6 +18,8 @@ Required properties:
"arm,arm11mpcore-pmu"
"arm,arm1176-pmu"
"arm,arm1136-pmu"
+ "qcom,scorpion-pmu"
+ "qcom,scorpion-mp-pmu"
"qcom,krait-pmu"
- interrupts : 1 combined interrupt or 1 per core. If the interrupt is a per-cpu
interrupt (PPI) then 1 interrupt should be specified.
diff --git a/arch/arm/kernel/perf_event_cpu.c b/arch/arm/kernel/perf_event_cpu.c
index dd9acc95ebc0..010ffd241434 100644
--- a/arch/arm/kernel/perf_event_cpu.c
+++ b/arch/arm/kernel/perf_event_cpu.c
@@ -242,6 +242,8 @@ static struct of_device_id cpu_pmu_of_device_ids[] = {
{.compatible = "arm,arm11mpcore-pmu", .data = armv6mpcore_pmu_init},
{.compatible = "arm,arm1176-pmu", .data = armv6_1176_pmu_init},
{.compatible = "arm,arm1136-pmu", .data = armv6_1136_pmu_init},
+ {.compatible = "qcom,scorpion-pmu", .data = scorpion_pmu_init},
+ {.compatible = "qcom,scorpion-mp-pmu", .data = scorpion_pmu_init},
{.compatible = "qcom,krait-pmu", .data = krait_pmu_init},
{},
};
diff --git a/arch/arm/kernel/perf_event_v7.c b/arch/arm/kernel/perf_event_v7.c
index 84a3ec3bc592..14bc8726f554 100644
--- a/arch/arm/kernel/perf_event_v7.c
+++ b/arch/arm/kernel/perf_event_v7.c
@@ -140,6 +140,23 @@ enum krait_perf_types {
KRAIT_PERFCTR_L1_DTLB_ACCESS = 0x12210,
};
+/* ARMv7 Scorpion specific event types */
+enum scorpion_perf_types {
+ SCORPION_LPM0_GROUP0 = 0x4c,
+ SCORPION_LPM1_GROUP0 = 0x50,
+ SCORPION_LPM2_GROUP0 = 0x54,
+ SCORPION_L2LPM_GROUP0 = 0x58,
+ SCORPION_VLPM_GROUP0 = 0x5c,
+
+ SCORPION_ICACHE_ACCESS = 0x10053,
+ SCORPION_ICACHE_MISS = 0x10052,
+
+ SCORPION_DTLB_ACCESS = 0x12013,
+ SCORPION_DTLB_MISS = 0x12012,
+
+ SCORPION_ITLB_MISS = 0x12021,
+};
+
/*
* Cortex-A8 HW events mapping
*
@@ -482,6 +499,51 @@ static const unsigned krait_perf_cache_map[PERF_COUNT_HW_CACHE_MAX]
};
/*
+ * Scorpion HW events mapping
+ */
+static const unsigned scorpion_perf_map[PERF_COUNT_HW_MAX] = {
+ PERF_MAP_ALL_UNSUPPORTED,
+ [PERF_COUNT_HW_CPU_CYCLES] = ARMV7_PERFCTR_CPU_CYCLES,
+ [PERF_COUNT_HW_INSTRUCTIONS] = ARMV7_PERFCTR_INSTR_EXECUTED,
+ [PERF_COUNT_HW_BRANCH_INSTRUCTIONS] = ARMV7_PERFCTR_PC_WRITE,
+ [PERF_COUNT_HW_BRANCH_MISSES] = ARMV7_PERFCTR_PC_BRANCH_MIS_PRED,
+ [PERF_COUNT_HW_BUS_CYCLES] = ARMV7_PERFCTR_CLOCK_CYCLES,
+};
+
+static const unsigned scorpion_perf_cache_map[PERF_COUNT_HW_CACHE_MAX]
+ [PERF_COUNT_HW_CACHE_OP_MAX]
+ [PERF_COUNT_HW_CACHE_RESULT_MAX] = {
+ PERF_CACHE_MAP_ALL_UNSUPPORTED,
+ /*
+ * The performance counters don't differentiate between read and write
+ * accesses/misses so this isn't strictly correct, but it's the best we
+ * can do. Writes and reads get combined.
+ */
+ [C(L1D)][C(OP_READ)][C(RESULT_ACCESS)] = ARMV7_PERFCTR_L1_DCACHE_ACCESS,
+ [C(L1D)][C(OP_READ)][C(RESULT_MISS)] = ARMV7_PERFCTR_L1_DCACHE_REFILL,
+ [C(L1D)][C(OP_WRITE)][C(RESULT_ACCESS)] = ARMV7_PERFCTR_L1_DCACHE_ACCESS,
+ [C(L1D)][C(OP_WRITE)][C(RESULT_MISS)] = ARMV7_PERFCTR_L1_DCACHE_REFILL,
+ [C(L1I)][C(OP_READ)][C(RESULT_ACCESS)] = SCORPION_ICACHE_ACCESS,
+ [C(L1I)][C(OP_READ)][C(RESULT_MISS)] = SCORPION_ICACHE_MISS,
+ [C(L1I)][C(OP_WRITE)][C(RESULT_ACCESS)] = SCORPION_ICACHE_ACCESS,
+ [C(L1I)][C(OP_WRITE)][C(RESULT_MISS)] = SCORPION_ICACHE_MISS,
+ /*
+ * Only ITLB misses and DTLB refills are supported. If users want the
+ * DTLB refills misses a raw counter must be used.
+ */
+ [C(DTLB)][C(OP_READ)][C(RESULT_ACCESS)] = SCORPION_DTLB_ACCESS,
+ [C(DTLB)][C(OP_READ)][C(RESULT_MISS)] = SCORPION_DTLB_MISS,
+ [C(DTLB)][C(OP_WRITE)][C(RESULT_ACCESS)] = SCORPION_DTLB_ACCESS,
+ [C(DTLB)][C(OP_WRITE)][C(RESULT_MISS)] = SCORPION_DTLB_MISS,
+ [C(ITLB)][C(OP_READ)][C(RESULT_MISS)] = SCORPION_ITLB_MISS,
+ [C(ITLB)][C(OP_WRITE)][C(RESULT_MISS)] = SCORPION_ITLB_MISS,
+ [C(BPU)][C(OP_READ)][C(RESULT_ACCESS)] = ARMV7_PERFCTR_PC_BRANCH_PRED,
+ [C(BPU)][C(OP_READ)][C(RESULT_MISS)] = ARMV7_PERFCTR_PC_BRANCH_PRED,
+ [C(BPU)][C(OP_WRITE)][C(RESULT_ACCESS)] = ARMV7_PERFCTR_PC_BRANCH_PRED,
+ [C(BPU)][C(OP_WRITE)][C(RESULT_MISS)] = ARMV7_PERFCTR_PC_BRANCH_PRED,
+};
+
+/*
* Perf Events' indices
*/
#define ARMV7_IDX_CYCLE_COUNTER 0
@@ -976,6 +1038,12 @@ static int krait_map_event_no_branch(struct perf_event *event)
&krait_perf_cache_map, 0xFFFFF);
}
+static int scorpion_map_event(struct perf_event *event)
+{
+ return armpmu_map_event(event, &scorpion_perf_map,
+ &scorpion_perf_cache_map, 0xFFFFF);
+}
+
static void armv7pmu_init(struct arm_pmu *cpu_pmu)
{
cpu_pmu->handle_irq = armv7pmu_handle_irq;
@@ -1463,6 +1531,333 @@ static int krait_pmu_init(struct arm_pmu *cpu_pmu)
cpu_pmu->clear_event_idx = krait_pmu_clear_event_idx;
return 0;
}
+
+/*
+ * Scorpion Local Performance Monitor Register (LPMn)
+ *
+ * 31 30 24 16 8 0
+ * +--------------------------------+
+ * LPM0 | EN | CC | CC | CC | CC | N = 1, R = 0
+ * +--------------------------------+
+ * LPM1 | EN | CC | CC | CC | CC | N = 1, R = 1
+ * +--------------------------------+
+ * LPM2 | EN | CC | CC | CC | CC | N = 1, R = 2
+ * +--------------------------------+
+ * L2LPM | EN | CC | CC | CC | CC | N = 1, R = 3
+ * +--------------------------------+
+ * VLPM | EN | CC | CC | CC | CC | N = 2, R = ?
+ * +--------------------------------+
+ * EN | G=3 | G=2 | G=1 | G=0
+ *
+ *
+ * Event Encoding:
+ *
+ * hwc->config_base = 0xNRCCG
+ *
+ * N = prefix, 1 for Scorpion CPU (LPMn/L2LPM), 2 for Venum VFP (VLPM)
+ * R = region register
+ * CC = class of events the group G is choosing from
+ * G = group or particular event
+ *
+ * Example: 0x12021 is a Scorpion CPU event in LPM2's group 1 with code 2
+ *
+ * A region (R) corresponds to a piece of the CPU (execution unit, instruction
+ * unit, etc.) while the event code (CC) corresponds to a particular class of
+ * events (interrupts for example). An event code is broken down into
+ * groups (G) that can be mapped into the PMU (irq, fiqs, and irq+fiqs for
+ * example).
+ */
+
+static u32 scorpion_read_pmresrn(int n)
+{
+ u32 val;
+
+ switch (n) {
+ case 0:
+ asm volatile("mrc p15, 0, %0, c15, c0, 0" : "=r" (val));
+ break;
+ case 1:
+ asm volatile("mrc p15, 1, %0, c15, c0, 0" : "=r" (val));
+ break;
+ case 2:
+ asm volatile("mrc p15, 2, %0, c15, c0, 0" : "=r" (val));
+ break;
+ case 3:
+ asm volatile("mrc p15, 3, %0, c15, c2, 0" : "=r" (val));
+ break;
+ default:
+ BUG(); /* Should be validated in scorpion_pmu_get_event_idx() */
+ }
+
+ return val;
+}
+
+static void scorpion_write_pmresrn(int n, u32 val)
+{
+ switch (n) {
+ case 0:
+ asm volatile("mcr p15, 0, %0, c15, c0, 0" : : "r" (val));
+ break;
+ case 1:
+ asm volatile("mcr p15, 1, %0, c15, c0, 0" : : "r" (val));
+ break;
+ case 2:
+ asm volatile("mcr p15, 2, %0, c15, c0, 0" : : "r" (val));
+ break;
+ case 3:
+ asm volatile("mcr p15, 3, %0, c15, c0, 0" : : "r" (val));
+ break;
+ default:
+ BUG(); /* Should be validated in scorpion_pmu_get_event_idx() */
+ }
+}
+
+static u32 scorpion_get_pmresrn_event(unsigned int region)
+{
+ static const u32 pmresrn_table[] = { SCORPION_LPM0_GROUP0,
+ SCORPION_LPM1_GROUP0,
+ SCORPION_LPM2_GROUP0,
+ SCORPION_L2LPM_GROUP0 };
+ return pmresrn_table[region];
+}
+
+static void scorpion_evt_setup(int idx, u32 config_base)
+{
+ u32 val;
+ u32 mask;
+ u32 vval, fval;
+ unsigned int region;
+ unsigned int group;
+ unsigned int code;
+ unsigned int group_shift;
+ bool venum_event;
+
+ krait_decode_event(config_base, ®ion, &group, &code, &venum_event,
+ NULL);
+
+ group_shift = group * 8;
+ mask = 0xff << group_shift;
+
+ /* Configure evtsel for the region and group */
+ if (venum_event)
+ val = SCORPION_VLPM_GROUP0;
+ else
+ val = scorpion_get_pmresrn_event(region);
+ val += group;
+ /* Mix in mode-exclusion bits */
+ val |= config_base & (ARMV7_EXCLUDE_USER | ARMV7_EXCLUDE_PL1);
+ armv7_pmnc_write_evtsel(idx, val);
+
+ asm volatile("mcr p15, 0, %0, c9, c15, 0" : : "r" (0));
+
+ if (venum_event) {
+ venum_pre_pmresr(&vval, &fval);
+ val = venum_read_pmresr();
+ val &= ~mask;
+ val |= code << group_shift;
+ val |= PMRESRn_EN;
+ venum_write_pmresr(val);
+ venum_post_pmresr(vval, fval);
+ } else {
+ val = scorpion_read_pmresrn(region);
+ val &= ~mask;
+ val |= code << group_shift;
+ val |= PMRESRn_EN;
+ scorpion_write_pmresrn(region, val);
+ }
+}
+
+static void scorpion_clearpmu(u32 config_base)
+{
+ u32 val;
+ u32 vval, fval;
+ unsigned int region;
+ unsigned int group;
+ bool venum_event;
+
+ krait_decode_event(config_base, ®ion, &group, NULL, &venum_event,
+ NULL);
+
+ if (venum_event) {
+ venum_pre_pmresr(&vval, &fval);
+ val = venum_read_pmresr();
+ val = clear_pmresrn_group(val, group);
+ venum_write_pmresr(val);
+ venum_post_pmresr(vval, fval);
+ } else {
+ val = scorpion_read_pmresrn(region);
+ val = clear_pmresrn_group(val, group);
+ scorpion_write_pmresrn(region, val);
+ }
+}
+
+static void scorpion_pmu_disable_event(struct perf_event *event)
+{
+ unsigned long flags;
+ struct hw_perf_event *hwc = &event->hw;
+ int idx = hwc->idx;
+ struct arm_pmu *cpu_pmu = to_arm_pmu(event->pmu);
+ struct pmu_hw_events *events = this_cpu_ptr(cpu_pmu->hw_events);
+
+ /* Disable counter and interrupt */
+ raw_spin_lock_irqsave(&events->pmu_lock, flags);
+
+ /* Disable counter */
+ armv7_pmnc_disable_counter(idx);
+
+ /*
+ * Clear pmresr code (if destined for PMNx counters)
+ */
+ if (hwc->config_base & KRAIT_EVENT_MASK)
+ scorpion_clearpmu(hwc->config_base);
+
+ /* Disable interrupt for this counter */
+ armv7_pmnc_disable_intens(idx);
+
+ raw_spin_unlock_irqrestore(&events->pmu_lock, flags);
+}
+
+static void scorpion_pmu_enable_event(struct perf_event *event)
+{
+ unsigned long flags;
+ struct hw_perf_event *hwc = &event->hw;
+ int idx = hwc->idx;
+ struct arm_pmu *cpu_pmu = to_arm_pmu(event->pmu);
+ struct pmu_hw_events *events = this_cpu_ptr(cpu_pmu->hw_events);
+
+ /*
+ * Enable counter and interrupt, and set the counter to count
+ * the event that we're interested in.
+ */
+ raw_spin_lock_irqsave(&events->pmu_lock, flags);
+
+ /* Disable counter */
+ armv7_pmnc_disable_counter(idx);
+
+ /*
+ * Set event (if destined for PMNx counters)
+ * We don't set the event for the cycle counter because we
+ * don't have the ability to perform event filtering.
+ */
+ if (hwc->config_base & KRAIT_EVENT_MASK)
+ scorpion_evt_setup(idx, hwc->config_base);
+ else if (idx != ARMV7_IDX_CYCLE_COUNTER)
+ armv7_pmnc_write_evtsel(idx, hwc->config_base);
+
+ /* Enable interrupt for this counter */
+ armv7_pmnc_enable_intens(idx);
+
+ /* Enable counter */
+ armv7_pmnc_enable_counter(idx);
+
+ raw_spin_unlock_irqrestore(&events->pmu_lock, flags);
+}
+
+static void scorpion_pmu_reset(void *info)
+{
+ u32 vval, fval;
+
+ armv7pmu_reset(info);
+
+ /* Clear all pmresrs */
+ scorpion_write_pmresrn(0, 0);
+ scorpion_write_pmresrn(1, 0);
+ scorpion_write_pmresrn(2, 0);
+ scorpion_write_pmresrn(3, 0);
+
+ venum_pre_pmresr(&vval, &fval);
+ venum_write_pmresr(0);
+ venum_post_pmresr(vval, fval);
+}
+
+static int scorpion_event_to_bit(struct perf_event *event, unsigned int region,
+ unsigned int group)
+{
+ int bit;
+ struct hw_perf_event *hwc = &event->hw;
+ struct arm_pmu *cpu_pmu = to_arm_pmu(event->pmu);
+
+ if (hwc->config_base & VENUM_EVENT)
+ bit = SCORPION_VLPM_GROUP0;
+ else
+ bit = scorpion_get_pmresrn_event(region);
+ bit -= scorpion_get_pmresrn_event(0);
+ bit += group;
+ /*
+ * Lower bits are reserved for use by the counters (see
+ * armv7pmu_get_event_idx() for more info)
+ */
+ bit += ARMV7_IDX_COUNTER_LAST(cpu_pmu) + 1;
+
+ return bit;
+}
+
+/*
+ * We check for column exclusion constraints here.
+ * Two events cant use the same group within a pmresr register.
+ */
+static int scorpion_pmu_get_event_idx(struct pmu_hw_events *cpuc,
+ struct perf_event *event)
+{
+ int idx;
+ int bit = -1;
+ unsigned int region;
+ unsigned int code;
+ unsigned int group;
+ bool venum_event, scorpion_event;
+ struct hw_perf_event *hwc = &event->hw;
+
+ krait_decode_event(hwc->config_base, ®ion, &group, &code,
+ &venum_event, &scorpion_event);
+
+ if (venum_event || scorpion_event) {
+ /* Ignore invalid events */
+ if (group > 3 || region > 3)
+ return -EINVAL;
+
+ bit = scorpion_event_to_bit(event, region, group);
+ if (test_and_set_bit(bit, cpuc->used_mask))
+ return -EAGAIN;
+ }
+
+ idx = armv7pmu_get_event_idx(cpuc, event);
+ if (idx < 0 && bit >= 0)
+ clear_bit(bit, cpuc->used_mask);
+
+ return idx;
+}
+
+static void scorpion_pmu_clear_event_idx(struct pmu_hw_events *cpuc,
+ struct perf_event *event)
+{
+ int bit;
+ struct hw_perf_event *hwc = &event->hw;
+ unsigned int region;
+ unsigned int group;
+ bool venum_event, scorpion_event;
+
+ krait_decode_event(hwc->config_base, ®ion, &group, NULL,
+ &venum_event, &scorpion_event);
+
+ if (venum_event || scorpion_event) {
+ bit = scorpion_event_to_bit(event, region, group);
+ clear_bit(bit, cpuc->used_mask);
+ }
+}
+
+static int scorpion_pmu_init(struct arm_pmu *cpu_pmu)
+{
+ armv7pmu_init(cpu_pmu);
+ cpu_pmu->name = "armv7_scorpion";
+ cpu_pmu->map_event = scorpion_map_event;
+ cpu_pmu->num_events = armv7_read_num_pmnc_events();
+ cpu_pmu->reset = scorpion_pmu_reset;
+ cpu_pmu->enable = scorpion_pmu_enable_event;
+ cpu_pmu->disable = scorpion_pmu_disable_event;
+ cpu_pmu->get_event_idx = scorpion_pmu_get_event_idx;
+ cpu_pmu->clear_event_idx = scorpion_pmu_clear_event_idx;
+ return 0;
+}
#else
static inline int armv7_a8_pmu_init(struct arm_pmu *cpu_pmu)
{
--
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
a Linux Foundation Collaborative Project
Hi Stephen,
On 10 February 2015 at 20:05, Stephen Boyd <[email protected]> wrote:
> Scorpion supports a set of local performance monitor event
> selection registers (LPM) sitting behind a cp15 based interface
> that extend the architected PMU events to include Scorpion CPU
> and Venum VFP specific events. To use these events the user is
> expected to program the lpm register with the event code shifted
> into the group they care about and then point the PMNx event at
> that region+group combo by writing a LPMn_GROUPx event. Add
> support for this hardware.
>
> Note: the raw event number is a pure software construct that
> allows us to map the multi-dimensional number space of regions,
> groups, and event codes into a flat event number space suitable
> for use by the perf framework.
>
> This is based on code originally written by Ashwin Chaugule and
> Neil Leeder [1] massed to become similar to the Krait PMU support
> code.
Thanks for taking this up!
Overall this series looks good to me, but from what I faintly
recollect, doesn't this (and the Krait pmu code) get affected by
powercollapse issues anymore?
e.g.
https://www.codeaurora.org/cgit/quic/la/kernel/msm/commit/arch/arm/kernel/perf_event_msm.c?h=msm-3.4&id=b5ca687960f0fea2f4735e83ca5c9543474c19de
Thanks,
Ashwin.
>
> [1] https://www.codeaurora.org/cgit/quic/la/kernel/msm/tree/arch/arm/kernel/perf_event_msm.c?h=msm-3.4
>
> Cc: Neil Leeder <[email protected]>
> Cc: Ashwin Chaugule <[email protected]>
> Cc: <[email protected]>
> Signed-off-by: Stephen Boyd <[email protected]>
> ---
> Documentation/devicetree/bindings/arm/pmu.txt | 2 +
> arch/arm/kernel/perf_event_cpu.c | 2 +
> arch/arm/kernel/perf_event_v7.c | 395 ++++++++++++++++++++++++++
> 3 files changed, 399 insertions(+)
>
> diff --git a/Documentation/devicetree/bindings/arm/pmu.txt b/Documentation/devicetree/bindings/arm/pmu.txt
> index 75ef91d08f3b..6e54a9d88b7a 100644
> --- a/Documentation/devicetree/bindings/arm/pmu.txt
> +++ b/Documentation/devicetree/bindings/arm/pmu.txt
> @@ -18,6 +18,8 @@ Required properties:
> "arm,arm11mpcore-pmu"
> "arm,arm1176-pmu"
> "arm,arm1136-pmu"
> + "qcom,scorpion-pmu"
> + "qcom,scorpion-mp-pmu"
> "qcom,krait-pmu"
> - interrupts : 1 combined interrupt or 1 per core. If the interrupt is a per-cpu
> interrupt (PPI) then 1 interrupt should be specified.
> diff --git a/arch/arm/kernel/perf_event_cpu.c b/arch/arm/kernel/perf_event_cpu.c
> index dd9acc95ebc0..010ffd241434 100644
> --- a/arch/arm/kernel/perf_event_cpu.c
> +++ b/arch/arm/kernel/perf_event_cpu.c
> @@ -242,6 +242,8 @@ static struct of_device_id cpu_pmu_of_device_ids[] = {
> {.compatible = "arm,arm11mpcore-pmu", .data = armv6mpcore_pmu_init},
> {.compatible = "arm,arm1176-pmu", .data = armv6_1176_pmu_init},
> {.compatible = "arm,arm1136-pmu", .data = armv6_1136_pmu_init},
> + {.compatible = "qcom,scorpion-pmu", .data = scorpion_pmu_init},
> + {.compatible = "qcom,scorpion-mp-pmu", .data = scorpion_pmu_init},
> {.compatible = "qcom,krait-pmu", .data = krait_pmu_init},
> {},
> };
> diff --git a/arch/arm/kernel/perf_event_v7.c b/arch/arm/kernel/perf_event_v7.c
> index 84a3ec3bc592..14bc8726f554 100644
> --- a/arch/arm/kernel/perf_event_v7.c
> +++ b/arch/arm/kernel/perf_event_v7.c
> @@ -140,6 +140,23 @@ enum krait_perf_types {
> KRAIT_PERFCTR_L1_DTLB_ACCESS = 0x12210,
> };
>
> +/* ARMv7 Scorpion specific event types */
> +enum scorpion_perf_types {
> + SCORPION_LPM0_GROUP0 = 0x4c,
> + SCORPION_LPM1_GROUP0 = 0x50,
> + SCORPION_LPM2_GROUP0 = 0x54,
> + SCORPION_L2LPM_GROUP0 = 0x58,
> + SCORPION_VLPM_GROUP0 = 0x5c,
> +
> + SCORPION_ICACHE_ACCESS = 0x10053,
> + SCORPION_ICACHE_MISS = 0x10052,
> +
> + SCORPION_DTLB_ACCESS = 0x12013,
> + SCORPION_DTLB_MISS = 0x12012,
> +
> + SCORPION_ITLB_MISS = 0x12021,
> +};
> +
> /*
> * Cortex-A8 HW events mapping
> *
> @@ -482,6 +499,51 @@ static const unsigned krait_perf_cache_map[PERF_COUNT_HW_CACHE_MAX]
> };
>
> /*
> + * Scorpion HW events mapping
> + */
> +static const unsigned scorpion_perf_map[PERF_COUNT_HW_MAX] = {
> + PERF_MAP_ALL_UNSUPPORTED,
> + [PERF_COUNT_HW_CPU_CYCLES] = ARMV7_PERFCTR_CPU_CYCLES,
> + [PERF_COUNT_HW_INSTRUCTIONS] = ARMV7_PERFCTR_INSTR_EXECUTED,
> + [PERF_COUNT_HW_BRANCH_INSTRUCTIONS] = ARMV7_PERFCTR_PC_WRITE,
> + [PERF_COUNT_HW_BRANCH_MISSES] = ARMV7_PERFCTR_PC_BRANCH_MIS_PRED,
> + [PERF_COUNT_HW_BUS_CYCLES] = ARMV7_PERFCTR_CLOCK_CYCLES,
> +};
> +
> +static const unsigned scorpion_perf_cache_map[PERF_COUNT_HW_CACHE_MAX]
> + [PERF_COUNT_HW_CACHE_OP_MAX]
> + [PERF_COUNT_HW_CACHE_RESULT_MAX] = {
> + PERF_CACHE_MAP_ALL_UNSUPPORTED,
> + /*
> + * The performance counters don't differentiate between read and write
> + * accesses/misses so this isn't strictly correct, but it's the best we
> + * can do. Writes and reads get combined.
> + */
> + [C(L1D)][C(OP_READ)][C(RESULT_ACCESS)] = ARMV7_PERFCTR_L1_DCACHE_ACCESS,
> + [C(L1D)][C(OP_READ)][C(RESULT_MISS)] = ARMV7_PERFCTR_L1_DCACHE_REFILL,
> + [C(L1D)][C(OP_WRITE)][C(RESULT_ACCESS)] = ARMV7_PERFCTR_L1_DCACHE_ACCESS,
> + [C(L1D)][C(OP_WRITE)][C(RESULT_MISS)] = ARMV7_PERFCTR_L1_DCACHE_REFILL,
> + [C(L1I)][C(OP_READ)][C(RESULT_ACCESS)] = SCORPION_ICACHE_ACCESS,
> + [C(L1I)][C(OP_READ)][C(RESULT_MISS)] = SCORPION_ICACHE_MISS,
> + [C(L1I)][C(OP_WRITE)][C(RESULT_ACCESS)] = SCORPION_ICACHE_ACCESS,
> + [C(L1I)][C(OP_WRITE)][C(RESULT_MISS)] = SCORPION_ICACHE_MISS,
> + /*
> + * Only ITLB misses and DTLB refills are supported. If users want the
> + * DTLB refills misses a raw counter must be used.
> + */
> + [C(DTLB)][C(OP_READ)][C(RESULT_ACCESS)] = SCORPION_DTLB_ACCESS,
> + [C(DTLB)][C(OP_READ)][C(RESULT_MISS)] = SCORPION_DTLB_MISS,
> + [C(DTLB)][C(OP_WRITE)][C(RESULT_ACCESS)] = SCORPION_DTLB_ACCESS,
> + [C(DTLB)][C(OP_WRITE)][C(RESULT_MISS)] = SCORPION_DTLB_MISS,
> + [C(ITLB)][C(OP_READ)][C(RESULT_MISS)] = SCORPION_ITLB_MISS,
> + [C(ITLB)][C(OP_WRITE)][C(RESULT_MISS)] = SCORPION_ITLB_MISS,
> + [C(BPU)][C(OP_READ)][C(RESULT_ACCESS)] = ARMV7_PERFCTR_PC_BRANCH_PRED,
> + [C(BPU)][C(OP_READ)][C(RESULT_MISS)] = ARMV7_PERFCTR_PC_BRANCH_PRED,
> + [C(BPU)][C(OP_WRITE)][C(RESULT_ACCESS)] = ARMV7_PERFCTR_PC_BRANCH_PRED,
> + [C(BPU)][C(OP_WRITE)][C(RESULT_MISS)] = ARMV7_PERFCTR_PC_BRANCH_PRED,
> +};
> +
> +/*
> * Perf Events' indices
> */
> #define ARMV7_IDX_CYCLE_COUNTER 0
> @@ -976,6 +1038,12 @@ static int krait_map_event_no_branch(struct perf_event *event)
> &krait_perf_cache_map, 0xFFFFF);
> }
>
> +static int scorpion_map_event(struct perf_event *event)
> +{
> + return armpmu_map_event(event, &scorpion_perf_map,
> + &scorpion_perf_cache_map, 0xFFFFF);
> +}
> +
> static void armv7pmu_init(struct arm_pmu *cpu_pmu)
> {
> cpu_pmu->handle_irq = armv7pmu_handle_irq;
> @@ -1463,6 +1531,333 @@ static int krait_pmu_init(struct arm_pmu *cpu_pmu)
> cpu_pmu->clear_event_idx = krait_pmu_clear_event_idx;
> return 0;
> }
> +
> +/*
> + * Scorpion Local Performance Monitor Register (LPMn)
> + *
> + * 31 30 24 16 8 0
> + * +--------------------------------+
> + * LPM0 | EN | CC | CC | CC | CC | N = 1, R = 0
> + * +--------------------------------+
> + * LPM1 | EN | CC | CC | CC | CC | N = 1, R = 1
> + * +--------------------------------+
> + * LPM2 | EN | CC | CC | CC | CC | N = 1, R = 2
> + * +--------------------------------+
> + * L2LPM | EN | CC | CC | CC | CC | N = 1, R = 3
> + * +--------------------------------+
> + * VLPM | EN | CC | CC | CC | CC | N = 2, R = ?
> + * +--------------------------------+
> + * EN | G=3 | G=2 | G=1 | G=0
> + *
> + *
> + * Event Encoding:
> + *
> + * hwc->config_base = 0xNRCCG
> + *
> + * N = prefix, 1 for Scorpion CPU (LPMn/L2LPM), 2 for Venum VFP (VLPM)
> + * R = region register
> + * CC = class of events the group G is choosing from
> + * G = group or particular event
> + *
> + * Example: 0x12021 is a Scorpion CPU event in LPM2's group 1 with code 2
> + *
> + * A region (R) corresponds to a piece of the CPU (execution unit, instruction
> + * unit, etc.) while the event code (CC) corresponds to a particular class of
> + * events (interrupts for example). An event code is broken down into
> + * groups (G) that can be mapped into the PMU (irq, fiqs, and irq+fiqs for
> + * example).
> + */
> +
> +static u32 scorpion_read_pmresrn(int n)
> +{
> + u32 val;
> +
> + switch (n) {
> + case 0:
> + asm volatile("mrc p15, 0, %0, c15, c0, 0" : "=r" (val));
> + break;
> + case 1:
> + asm volatile("mrc p15, 1, %0, c15, c0, 0" : "=r" (val));
> + break;
> + case 2:
> + asm volatile("mrc p15, 2, %0, c15, c0, 0" : "=r" (val));
> + break;
> + case 3:
> + asm volatile("mrc p15, 3, %0, c15, c2, 0" : "=r" (val));
> + break;
> + default:
> + BUG(); /* Should be validated in scorpion_pmu_get_event_idx() */
> + }
> +
> + return val;
> +}
> +
> +static void scorpion_write_pmresrn(int n, u32 val)
> +{
> + switch (n) {
> + case 0:
> + asm volatile("mcr p15, 0, %0, c15, c0, 0" : : "r" (val));
> + break;
> + case 1:
> + asm volatile("mcr p15, 1, %0, c15, c0, 0" : : "r" (val));
> + break;
> + case 2:
> + asm volatile("mcr p15, 2, %0, c15, c0, 0" : : "r" (val));
> + break;
> + case 3:
> + asm volatile("mcr p15, 3, %0, c15, c0, 0" : : "r" (val));
> + break;
> + default:
> + BUG(); /* Should be validated in scorpion_pmu_get_event_idx() */
> + }
> +}
> +
> +static u32 scorpion_get_pmresrn_event(unsigned int region)
> +{
> + static const u32 pmresrn_table[] = { SCORPION_LPM0_GROUP0,
> + SCORPION_LPM1_GROUP0,
> + SCORPION_LPM2_GROUP0,
> + SCORPION_L2LPM_GROUP0 };
> + return pmresrn_table[region];
> +}
> +
> +static void scorpion_evt_setup(int idx, u32 config_base)
> +{
> + u32 val;
> + u32 mask;
> + u32 vval, fval;
> + unsigned int region;
> + unsigned int group;
> + unsigned int code;
> + unsigned int group_shift;
> + bool venum_event;
> +
> + krait_decode_event(config_base, ®ion, &group, &code, &venum_event,
> + NULL);
> +
> + group_shift = group * 8;
> + mask = 0xff << group_shift;
> +
> + /* Configure evtsel for the region and group */
> + if (venum_event)
> + val = SCORPION_VLPM_GROUP0;
> + else
> + val = scorpion_get_pmresrn_event(region);
> + val += group;
> + /* Mix in mode-exclusion bits */
> + val |= config_base & (ARMV7_EXCLUDE_USER | ARMV7_EXCLUDE_PL1);
> + armv7_pmnc_write_evtsel(idx, val);
> +
> + asm volatile("mcr p15, 0, %0, c9, c15, 0" : : "r" (0));
> +
> + if (venum_event) {
> + venum_pre_pmresr(&vval, &fval);
> + val = venum_read_pmresr();
> + val &= ~mask;
> + val |= code << group_shift;
> + val |= PMRESRn_EN;
> + venum_write_pmresr(val);
> + venum_post_pmresr(vval, fval);
> + } else {
> + val = scorpion_read_pmresrn(region);
> + val &= ~mask;
> + val |= code << group_shift;
> + val |= PMRESRn_EN;
> + scorpion_write_pmresrn(region, val);
> + }
> +}
> +
> +static void scorpion_clearpmu(u32 config_base)
> +{
> + u32 val;
> + u32 vval, fval;
> + unsigned int region;
> + unsigned int group;
> + bool venum_event;
> +
> + krait_decode_event(config_base, ®ion, &group, NULL, &venum_event,
> + NULL);
> +
> + if (venum_event) {
> + venum_pre_pmresr(&vval, &fval);
> + val = venum_read_pmresr();
> + val = clear_pmresrn_group(val, group);
> + venum_write_pmresr(val);
> + venum_post_pmresr(vval, fval);
> + } else {
> + val = scorpion_read_pmresrn(region);
> + val = clear_pmresrn_group(val, group);
> + scorpion_write_pmresrn(region, val);
> + }
> +}
> +
> +static void scorpion_pmu_disable_event(struct perf_event *event)
> +{
> + unsigned long flags;
> + struct hw_perf_event *hwc = &event->hw;
> + int idx = hwc->idx;
> + struct arm_pmu *cpu_pmu = to_arm_pmu(event->pmu);
> + struct pmu_hw_events *events = this_cpu_ptr(cpu_pmu->hw_events);
> +
> + /* Disable counter and interrupt */
> + raw_spin_lock_irqsave(&events->pmu_lock, flags);
> +
> + /* Disable counter */
> + armv7_pmnc_disable_counter(idx);
> +
> + /*
> + * Clear pmresr code (if destined for PMNx counters)
> + */
> + if (hwc->config_base & KRAIT_EVENT_MASK)
> + scorpion_clearpmu(hwc->config_base);
> +
> + /* Disable interrupt for this counter */
> + armv7_pmnc_disable_intens(idx);
> +
> + raw_spin_unlock_irqrestore(&events->pmu_lock, flags);
> +}
> +
> +static void scorpion_pmu_enable_event(struct perf_event *event)
> +{
> + unsigned long flags;
> + struct hw_perf_event *hwc = &event->hw;
> + int idx = hwc->idx;
> + struct arm_pmu *cpu_pmu = to_arm_pmu(event->pmu);
> + struct pmu_hw_events *events = this_cpu_ptr(cpu_pmu->hw_events);
> +
> + /*
> + * Enable counter and interrupt, and set the counter to count
> + * the event that we're interested in.
> + */
> + raw_spin_lock_irqsave(&events->pmu_lock, flags);
> +
> + /* Disable counter */
> + armv7_pmnc_disable_counter(idx);
> +
> + /*
> + * Set event (if destined for PMNx counters)
> + * We don't set the event for the cycle counter because we
> + * don't have the ability to perform event filtering.
> + */
> + if (hwc->config_base & KRAIT_EVENT_MASK)
> + scorpion_evt_setup(idx, hwc->config_base);
> + else if (idx != ARMV7_IDX_CYCLE_COUNTER)
> + armv7_pmnc_write_evtsel(idx, hwc->config_base);
> +
> + /* Enable interrupt for this counter */
> + armv7_pmnc_enable_intens(idx);
> +
> + /* Enable counter */
> + armv7_pmnc_enable_counter(idx);
> +
> + raw_spin_unlock_irqrestore(&events->pmu_lock, flags);
> +}
> +
> +static void scorpion_pmu_reset(void *info)
> +{
> + u32 vval, fval;
> +
> + armv7pmu_reset(info);
> +
> + /* Clear all pmresrs */
> + scorpion_write_pmresrn(0, 0);
> + scorpion_write_pmresrn(1, 0);
> + scorpion_write_pmresrn(2, 0);
> + scorpion_write_pmresrn(3, 0);
> +
> + venum_pre_pmresr(&vval, &fval);
> + venum_write_pmresr(0);
> + venum_post_pmresr(vval, fval);
> +}
> +
> +static int scorpion_event_to_bit(struct perf_event *event, unsigned int region,
> + unsigned int group)
> +{
> + int bit;
> + struct hw_perf_event *hwc = &event->hw;
> + struct arm_pmu *cpu_pmu = to_arm_pmu(event->pmu);
> +
> + if (hwc->config_base & VENUM_EVENT)
> + bit = SCORPION_VLPM_GROUP0;
> + else
> + bit = scorpion_get_pmresrn_event(region);
> + bit -= scorpion_get_pmresrn_event(0);
> + bit += group;
> + /*
> + * Lower bits are reserved for use by the counters (see
> + * armv7pmu_get_event_idx() for more info)
> + */
> + bit += ARMV7_IDX_COUNTER_LAST(cpu_pmu) + 1;
> +
> + return bit;
> +}
> +
> +/*
> + * We check for column exclusion constraints here.
> + * Two events cant use the same group within a pmresr register.
> + */
> +static int scorpion_pmu_get_event_idx(struct pmu_hw_events *cpuc,
> + struct perf_event *event)
> +{
> + int idx;
> + int bit = -1;
> + unsigned int region;
> + unsigned int code;
> + unsigned int group;
> + bool venum_event, scorpion_event;
> + struct hw_perf_event *hwc = &event->hw;
> +
> + krait_decode_event(hwc->config_base, ®ion, &group, &code,
> + &venum_event, &scorpion_event);
> +
> + if (venum_event || scorpion_event) {
> + /* Ignore invalid events */
> + if (group > 3 || region > 3)
> + return -EINVAL;
> +
> + bit = scorpion_event_to_bit(event, region, group);
> + if (test_and_set_bit(bit, cpuc->used_mask))
> + return -EAGAIN;
> + }
> +
> + idx = armv7pmu_get_event_idx(cpuc, event);
> + if (idx < 0 && bit >= 0)
> + clear_bit(bit, cpuc->used_mask);
> +
> + return idx;
> +}
> +
> +static void scorpion_pmu_clear_event_idx(struct pmu_hw_events *cpuc,
> + struct perf_event *event)
> +{
> + int bit;
> + struct hw_perf_event *hwc = &event->hw;
> + unsigned int region;
> + unsigned int group;
> + bool venum_event, scorpion_event;
> +
> + krait_decode_event(hwc->config_base, ®ion, &group, NULL,
> + &venum_event, &scorpion_event);
> +
> + if (venum_event || scorpion_event) {
> + bit = scorpion_event_to_bit(event, region, group);
> + clear_bit(bit, cpuc->used_mask);
> + }
> +}
> +
> +static int scorpion_pmu_init(struct arm_pmu *cpu_pmu)
> +{
> + armv7pmu_init(cpu_pmu);
> + cpu_pmu->name = "armv7_scorpion";
> + cpu_pmu->map_event = scorpion_map_event;
> + cpu_pmu->num_events = armv7_read_num_pmnc_events();
> + cpu_pmu->reset = scorpion_pmu_reset;
> + cpu_pmu->enable = scorpion_pmu_enable_event;
> + cpu_pmu->disable = scorpion_pmu_disable_event;
> + cpu_pmu->get_event_idx = scorpion_pmu_get_event_idx;
> + cpu_pmu->clear_event_idx = scorpion_pmu_clear_event_idx;
> + return 0;
> +}
> #else
> static inline int armv7_a8_pmu_init(struct arm_pmu *cpu_pmu)
> {
> --
> The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
> a Linux Foundation Collaborative Project
>
>
> _______________________________________________
> linux-arm-kernel mailing list
> [email protected]
> http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
On 02/10, Ashwin Chaugule wrote:
> Hi Stephen,
>
> On 10 February 2015 at 20:05, Stephen Boyd <[email protected]> wrote:
> > Scorpion supports a set of local performance monitor event
> > selection registers (LPM) sitting behind a cp15 based interface
> > that extend the architected PMU events to include Scorpion CPU
> > and Venum VFP specific events. To use these events the user is
> > expected to program the lpm register with the event code shifted
> > into the group they care about and then point the PMNx event at
> > that region+group combo by writing a LPMn_GROUPx event. Add
> > support for this hardware.
> >
> > Note: the raw event number is a pure software construct that
> > allows us to map the multi-dimensional number space of regions,
> > groups, and event codes into a flat event number space suitable
> > for use by the perf framework.
> >
> > This is based on code originally written by Ashwin Chaugule and
> > Neil Leeder [1] massed to become similar to the Krait PMU support
> > code.
>
> Thanks for taking this up!
> Overall this series looks good to me, but from what I faintly
> recollect, doesn't this (and the Krait pmu code) get affected by
> powercollapse issues anymore?
> e.g.
> https://www.codeaurora.org/cgit/quic/la/kernel/msm/commit/arch/arm/kernel/perf_event_msm.c?h=msm-3.4&id=b5ca687960f0fea2f4735e83ca5c9543474c19de
>
Right now there isn't any power collapse support in mainline so
there's no immediate problem. Once we add power collapse support
(i.e. cpuidle) to the Scorpion and Krait platforms we'll need to
do something in the perf event code to properly maintain the
counts across idle. I imagine it would be done by registering for
cpu_pm notifications and then doing the save/restore on
CPU_PM_ENTER and CPU_PM_EXIT. At least, that's what you started
doing in this patch[1]. And then it seems the patch you mention
came after that and actually did the save/restore of the counts.
[1] https://www.codeaurora.org/cgit/quic/la/kernel/msm/commit/?h=msm-3.4&id=464983a7e991a484cac0bc0885cee4fee318d659
--
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
a Linux Foundation Collaborative Project
On 02/10, Stephen Boyd wrote:
> Scorpion supports a set of local performance monitor event
> selection registers (LPM) sitting behind a cp15 based interface
> that extend the architected PMU events to include Scorpion CPU
> and Venum VFP specific events. To use these events the user is
> expected to program the lpm register with the event code shifted
> into the group they care about and then point the PMNx event at
> that region+group combo by writing a LPMn_GROUPx event. Add
> support for this hardware.
>
> Note: the raw event number is a pure software construct that
> allows us to map the multi-dimensional number space of regions,
> groups, and event codes into a flat event number space suitable
> for use by the perf framework.
>
> This is based on code originally written by Ashwin Chaugule and
> Neil Leeder [1] massed to become similar to the Krait PMU support
> code.
>
> [1] https://www.codeaurora.org/cgit/quic/la/kernel/msm/tree/arch/arm/kernel/perf_event_msm.c?h=msm-3.4
>
> Cc: Neil Leeder <[email protected]>
> Cc: Ashwin Chaugule <[email protected]>
> Cc: <[email protected]>
> Signed-off-by: Stephen Boyd <[email protected]>
> ---
> Documentation/devicetree/bindings/arm/pmu.txt | 2 +
> arch/arm/kernel/perf_event_cpu.c | 2 +
> arch/arm/kernel/perf_event_v7.c | 395 ++++++++++++++++++++++++++
> 3 files changed, 399 insertions(+)
>
> diff --git a/Documentation/devicetree/bindings/arm/pmu.txt b/Documentation/devicetree/bindings/arm/pmu.txt
> index 75ef91d08f3b..6e54a9d88b7a 100644
> --- a/Documentation/devicetree/bindings/arm/pmu.txt
> +++ b/Documentation/devicetree/bindings/arm/pmu.txt
> @@ -18,6 +18,8 @@ Required properties:
> "arm,arm11mpcore-pmu"
> "arm,arm1176-pmu"
> "arm,arm1136-pmu"
> + "qcom,scorpion-pmu"
> + "qcom,scorpion-mp-pmu"
> "qcom,krait-pmu"
> - interrupts : 1 combined interrupt or 1 per core. If the interrupt is a per-cpu
> interrupt (PPI) then 1 interrupt should be specified.
> diff --git a/arch/arm/kernel/perf_event_cpu.c b/arch/arm/kernel/perf_event_cpu.c
> index dd9acc95ebc0..010ffd241434 100644
> --- a/arch/arm/kernel/perf_event_cpu.c
> +++ b/arch/arm/kernel/perf_event_cpu.c
> @@ -242,6 +242,8 @@ static struct of_device_id cpu_pmu_of_device_ids[] = {
> {.compatible = "arm,arm11mpcore-pmu", .data = armv6mpcore_pmu_init},
> {.compatible = "arm,arm1176-pmu", .data = armv6_1176_pmu_init},
> {.compatible = "arm,arm1136-pmu", .data = armv6_1136_pmu_init},
> + {.compatible = "qcom,scorpion-pmu", .data = scorpion_pmu_init},
> + {.compatible = "qcom,scorpion-mp-pmu", .data = scorpion_pmu_init},
> {.compatible = "qcom,krait-pmu", .data = krait_pmu_init},
> {},
> };
> diff --git a/arch/arm/kernel/perf_event_v7.c b/arch/arm/kernel/perf_event_v7.c
> index 84a3ec3bc592..14bc8726f554 100644
> --- a/arch/arm/kernel/perf_event_v7.c
> +++ b/arch/arm/kernel/perf_event_v7.c
> @@ -140,6 +140,23 @@ enum krait_perf_types {
> KRAIT_PERFCTR_L1_DTLB_ACCESS = 0x12210,
> };
>
> +/* ARMv7 Scorpion specific event types */
> +enum scorpion_perf_types {
> + SCORPION_LPM0_GROUP0 = 0x4c,
> + SCORPION_LPM1_GROUP0 = 0x50,
> + SCORPION_LPM2_GROUP0 = 0x54,
> + SCORPION_L2LPM_GROUP0 = 0x58,
> + SCORPION_VLPM_GROUP0 = 0x5c,
> +
> + SCORPION_ICACHE_ACCESS = 0x10053,
> + SCORPION_ICACHE_MISS = 0x10052,
> +
> + SCORPION_DTLB_ACCESS = 0x12013,
> + SCORPION_DTLB_MISS = 0x12012,
> +
> + SCORPION_ITLB_MISS = 0x12021,
> +};
> +
> /*
> * Cortex-A8 HW events mapping
> *
> @@ -482,6 +499,51 @@ static const unsigned krait_perf_cache_map[PERF_COUNT_HW_CACHE_MAX]
> };
>
> /*
> + * Scorpion HW events mapping
> + */
> +static const unsigned scorpion_perf_map[PERF_COUNT_HW_MAX] = {
> + PERF_MAP_ALL_UNSUPPORTED,
> + [PERF_COUNT_HW_CPU_CYCLES] = ARMV7_PERFCTR_CPU_CYCLES,
> + [PERF_COUNT_HW_INSTRUCTIONS] = ARMV7_PERFCTR_INSTR_EXECUTED,
> + [PERF_COUNT_HW_BRANCH_INSTRUCTIONS] = ARMV7_PERFCTR_PC_WRITE,
> + [PERF_COUNT_HW_BRANCH_MISSES] = ARMV7_PERFCTR_PC_BRANCH_MIS_PRED,
> + [PERF_COUNT_HW_BUS_CYCLES] = ARMV7_PERFCTR_CLOCK_CYCLES,
> +};
> +
> +static const unsigned scorpion_perf_cache_map[PERF_COUNT_HW_CACHE_MAX]
> + [PERF_COUNT_HW_CACHE_OP_MAX]
> + [PERF_COUNT_HW_CACHE_RESULT_MAX] = {
> + PERF_CACHE_MAP_ALL_UNSUPPORTED,
> + /*
> + * The performance counters don't differentiate between read and write
> + * accesses/misses so this isn't strictly correct, but it's the best we
> + * can do. Writes and reads get combined.
> + */
> + [C(L1D)][C(OP_READ)][C(RESULT_ACCESS)] = ARMV7_PERFCTR_L1_DCACHE_ACCESS,
> + [C(L1D)][C(OP_READ)][C(RESULT_MISS)] = ARMV7_PERFCTR_L1_DCACHE_REFILL,
> + [C(L1D)][C(OP_WRITE)][C(RESULT_ACCESS)] = ARMV7_PERFCTR_L1_DCACHE_ACCESS,
> + [C(L1D)][C(OP_WRITE)][C(RESULT_MISS)] = ARMV7_PERFCTR_L1_DCACHE_REFILL,
> + [C(L1I)][C(OP_READ)][C(RESULT_ACCESS)] = SCORPION_ICACHE_ACCESS,
> + [C(L1I)][C(OP_READ)][C(RESULT_MISS)] = SCORPION_ICACHE_MISS,
> + [C(L1I)][C(OP_WRITE)][C(RESULT_ACCESS)] = SCORPION_ICACHE_ACCESS,
> + [C(L1I)][C(OP_WRITE)][C(RESULT_MISS)] = SCORPION_ICACHE_MISS,
> + /*
> + * Only ITLB misses and DTLB refills are supported. If users want the
> + * DTLB refills misses a raw counter must be used.
> + */
> + [C(DTLB)][C(OP_READ)][C(RESULT_ACCESS)] = SCORPION_DTLB_ACCESS,
> + [C(DTLB)][C(OP_READ)][C(RESULT_MISS)] = SCORPION_DTLB_MISS,
> + [C(DTLB)][C(OP_WRITE)][C(RESULT_ACCESS)] = SCORPION_DTLB_ACCESS,
> + [C(DTLB)][C(OP_WRITE)][C(RESULT_MISS)] = SCORPION_DTLB_MISS,
> + [C(ITLB)][C(OP_READ)][C(RESULT_MISS)] = SCORPION_ITLB_MISS,
> + [C(ITLB)][C(OP_WRITE)][C(RESULT_MISS)] = SCORPION_ITLB_MISS,
> + [C(BPU)][C(OP_READ)][C(RESULT_ACCESS)] = ARMV7_PERFCTR_PC_BRANCH_PRED,
> + [C(BPU)][C(OP_READ)][C(RESULT_MISS)] = ARMV7_PERFCTR_PC_BRANCH_PRED,
> + [C(BPU)][C(OP_WRITE)][C(RESULT_ACCESS)] = ARMV7_PERFCTR_PC_BRANCH_PRED,
> + [C(BPU)][C(OP_WRITE)][C(RESULT_MISS)] = ARMV7_PERFCTR_PC_BRANCH_PRED,
> +};
> +
> +/*
> * Perf Events' indices
> */
> #define ARMV7_IDX_CYCLE_COUNTER 0
> @@ -976,6 +1038,12 @@ static int krait_map_event_no_branch(struct perf_event *event)
> &krait_perf_cache_map, 0xFFFFF);
> }
>
> +static int scorpion_map_event(struct perf_event *event)
> +{
> + return armpmu_map_event(event, &scorpion_perf_map,
> + &scorpion_perf_cache_map, 0xFFFFF);
> +}
> +
> static void armv7pmu_init(struct arm_pmu *cpu_pmu)
> {
> cpu_pmu->handle_irq = armv7pmu_handle_irq;
> @@ -1463,6 +1531,333 @@ static int krait_pmu_init(struct arm_pmu *cpu_pmu)
> cpu_pmu->clear_event_idx = krait_pmu_clear_event_idx;
> return 0;
> }
> +
> +/*
> + * Scorpion Local Performance Monitor Register (LPMn)
> + *
> + * 31 30 24 16 8 0
> + * +--------------------------------+
> + * LPM0 | EN | CC | CC | CC | CC | N = 1, R = 0
> + * +--------------------------------+
> + * LPM1 | EN | CC | CC | CC | CC | N = 1, R = 1
> + * +--------------------------------+
> + * LPM2 | EN | CC | CC | CC | CC | N = 1, R = 2
> + * +--------------------------------+
> + * L2LPM | EN | CC | CC | CC | CC | N = 1, R = 3
> + * +--------------------------------+
> + * VLPM | EN | CC | CC | CC | CC | N = 2, R = ?
> + * +--------------------------------+
> + * EN | G=3 | G=2 | G=1 | G=0
> + *
> + *
> + * Event Encoding:
> + *
> + * hwc->config_base = 0xNRCCG
> + *
> + * N = prefix, 1 for Scorpion CPU (LPMn/L2LPM), 2 for Venum VFP (VLPM)
> + * R = region register
> + * CC = class of events the group G is choosing from
> + * G = group or particular event
> + *
> + * Example: 0x12021 is a Scorpion CPU event in LPM2's group 1 with code 2
> + *
> + * A region (R) corresponds to a piece of the CPU (execution unit, instruction
> + * unit, etc.) while the event code (CC) corresponds to a particular class of
> + * events (interrupts for example). An event code is broken down into
> + * groups (G) that can be mapped into the PMU (irq, fiqs, and irq+fiqs for
> + * example).
> + */
> +
> +static u32 scorpion_read_pmresrn(int n)
> +{
> + u32 val;
> +
> + switch (n) {
> + case 0:
> + asm volatile("mrc p15, 0, %0, c15, c0, 0" : "=r" (val));
> + break;
> + case 1:
> + asm volatile("mrc p15, 1, %0, c15, c0, 0" : "=r" (val));
> + break;
> + case 2:
> + asm volatile("mrc p15, 2, %0, c15, c0, 0" : "=r" (val));
> + break;
> + case 3:
> + asm volatile("mrc p15, 3, %0, c15, c2, 0" : "=r" (val));
> + break;
> + default:
> + BUG(); /* Should be validated in scorpion_pmu_get_event_idx() */
> + }
> +
> + return val;
> +}
> +
> +static void scorpion_write_pmresrn(int n, u32 val)
> +{
> + switch (n) {
> + case 0:
> + asm volatile("mcr p15, 0, %0, c15, c0, 0" : : "r" (val));
> + break;
> + case 1:
> + asm volatile("mcr p15, 1, %0, c15, c0, 0" : : "r" (val));
> + break;
> + case 2:
> + asm volatile("mcr p15, 2, %0, c15, c0, 0" : : "r" (val));
> + break;
> + case 3:
> + asm volatile("mcr p15, 3, %0, c15, c0, 0" : : "r" (val));
> + break;
> + default:
> + BUG(); /* Should be validated in scorpion_pmu_get_event_idx() */
> + }
> +}
> +
> +static u32 scorpion_get_pmresrn_event(unsigned int region)
> +{
> + static const u32 pmresrn_table[] = { SCORPION_LPM0_GROUP0,
> + SCORPION_LPM1_GROUP0,
> + SCORPION_LPM2_GROUP0,
> + SCORPION_L2LPM_GROUP0 };
> + return pmresrn_table[region];
> +}
> +
> +static void scorpion_evt_setup(int idx, u32 config_base)
> +{
> + u32 val;
> + u32 mask;
> + u32 vval, fval;
> + unsigned int region;
> + unsigned int group;
> + unsigned int code;
> + unsigned int group_shift;
> + bool venum_event;
> +
> + krait_decode_event(config_base, ®ion, &group, &code, &venum_event,
> + NULL);
> +
> + group_shift = group * 8;
> + mask = 0xff << group_shift;
> +
> + /* Configure evtsel for the region and group */
> + if (venum_event)
> + val = SCORPION_VLPM_GROUP0;
> + else
> + val = scorpion_get_pmresrn_event(region);
> + val += group;
> + /* Mix in mode-exclusion bits */
> + val |= config_base & (ARMV7_EXCLUDE_USER | ARMV7_EXCLUDE_PL1);
> + armv7_pmnc_write_evtsel(idx, val);
> +
> + asm volatile("mcr p15, 0, %0, c9, c15, 0" : : "r" (0));
> +
> + if (venum_event) {
> + venum_pre_pmresr(&vval, &fval);
> + val = venum_read_pmresr();
> + val &= ~mask;
> + val |= code << group_shift;
> + val |= PMRESRn_EN;
> + venum_write_pmresr(val);
> + venum_post_pmresr(vval, fval);
> + } else {
> + val = scorpion_read_pmresrn(region);
> + val &= ~mask;
> + val |= code << group_shift;
> + val |= PMRESRn_EN;
> + scorpion_write_pmresrn(region, val);
> + }
> +}
> +
> +static void scorpion_clearpmu(u32 config_base)
> +{
> + u32 val;
> + u32 vval, fval;
> + unsigned int region;
> + unsigned int group;
> + bool venum_event;
> +
> + krait_decode_event(config_base, ®ion, &group, NULL, &venum_event,
> + NULL);
> +
> + if (venum_event) {
> + venum_pre_pmresr(&vval, &fval);
> + val = venum_read_pmresr();
> + val = clear_pmresrn_group(val, group);
> + venum_write_pmresr(val);
> + venum_post_pmresr(vval, fval);
> + } else {
> + val = scorpion_read_pmresrn(region);
> + val = clear_pmresrn_group(val, group);
> + scorpion_write_pmresrn(region, val);
> + }
> +}
> +
> +static void scorpion_pmu_disable_event(struct perf_event *event)
> +{
> + unsigned long flags;
> + struct hw_perf_event *hwc = &event->hw;
> + int idx = hwc->idx;
> + struct arm_pmu *cpu_pmu = to_arm_pmu(event->pmu);
> + struct pmu_hw_events *events = this_cpu_ptr(cpu_pmu->hw_events);
> +
> + /* Disable counter and interrupt */
> + raw_spin_lock_irqsave(&events->pmu_lock, flags);
> +
> + /* Disable counter */
> + armv7_pmnc_disable_counter(idx);
> +
> + /*
> + * Clear pmresr code (if destined for PMNx counters)
> + */
> + if (hwc->config_base & KRAIT_EVENT_MASK)
> + scorpion_clearpmu(hwc->config_base);
> +
> + /* Disable interrupt for this counter */
> + armv7_pmnc_disable_intens(idx);
> +
> + raw_spin_unlock_irqrestore(&events->pmu_lock, flags);
> +}
> +
> +static void scorpion_pmu_enable_event(struct perf_event *event)
> +{
> + unsigned long flags;
> + struct hw_perf_event *hwc = &event->hw;
> + int idx = hwc->idx;
> + struct arm_pmu *cpu_pmu = to_arm_pmu(event->pmu);
> + struct pmu_hw_events *events = this_cpu_ptr(cpu_pmu->hw_events);
> +
> + /*
> + * Enable counter and interrupt, and set the counter to count
> + * the event that we're interested in.
> + */
> + raw_spin_lock_irqsave(&events->pmu_lock, flags);
> +
> + /* Disable counter */
> + armv7_pmnc_disable_counter(idx);
> +
> + /*
> + * Set event (if destined for PMNx counters)
> + * We don't set the event for the cycle counter because we
> + * don't have the ability to perform event filtering.
> + */
> + if (hwc->config_base & KRAIT_EVENT_MASK)
> + scorpion_evt_setup(idx, hwc->config_base);
> + else if (idx != ARMV7_IDX_CYCLE_COUNTER)
> + armv7_pmnc_write_evtsel(idx, hwc->config_base);
> +
> + /* Enable interrupt for this counter */
> + armv7_pmnc_enable_intens(idx);
> +
> + /* Enable counter */
> + armv7_pmnc_enable_counter(idx);
> +
> + raw_spin_unlock_irqrestore(&events->pmu_lock, flags);
> +}
> +
> +static void scorpion_pmu_reset(void *info)
> +{
> + u32 vval, fval;
> +
> + armv7pmu_reset(info);
> +
> + /* Clear all pmresrs */
> + scorpion_write_pmresrn(0, 0);
> + scorpion_write_pmresrn(1, 0);
> + scorpion_write_pmresrn(2, 0);
> + scorpion_write_pmresrn(3, 0);
> +
> + venum_pre_pmresr(&vval, &fval);
> + venum_write_pmresr(0);
> + venum_post_pmresr(vval, fval);
> +}
> +
> +static int scorpion_event_to_bit(struct perf_event *event, unsigned int region,
> + unsigned int group)
> +{
> + int bit;
> + struct hw_perf_event *hwc = &event->hw;
> + struct arm_pmu *cpu_pmu = to_arm_pmu(event->pmu);
> +
> + if (hwc->config_base & VENUM_EVENT)
> + bit = SCORPION_VLPM_GROUP0;
> + else
> + bit = scorpion_get_pmresrn_event(region);
> + bit -= scorpion_get_pmresrn_event(0);
> + bit += group;
> + /*
> + * Lower bits are reserved for use by the counters (see
> + * armv7pmu_get_event_idx() for more info)
> + */
> + bit += ARMV7_IDX_COUNTER_LAST(cpu_pmu) + 1;
> +
> + return bit;
> +}
> +
> +/*
> + * We check for column exclusion constraints here.
> + * Two events cant use the same group within a pmresr register.
> + */
> +static int scorpion_pmu_get_event_idx(struct pmu_hw_events *cpuc,
> + struct perf_event *event)
> +{
> + int idx;
> + int bit = -1;
> + unsigned int region;
> + unsigned int code;
> + unsigned int group;
> + bool venum_event, scorpion_event;
> + struct hw_perf_event *hwc = &event->hw;
> +
> + krait_decode_event(hwc->config_base, ®ion, &group, &code,
> + &venum_event, &scorpion_event);
> +
> + if (venum_event || scorpion_event) {
> + /* Ignore invalid events */
> + if (group > 3 || region > 3)
> + return -EINVAL;
> +
> + bit = scorpion_event_to_bit(event, region, group);
> + if (test_and_set_bit(bit, cpuc->used_mask))
> + return -EAGAIN;
> + }
> +
> + idx = armv7pmu_get_event_idx(cpuc, event);
> + if (idx < 0 && bit >= 0)
> + clear_bit(bit, cpuc->used_mask);
> +
> + return idx;
> +}
> +
> +static void scorpion_pmu_clear_event_idx(struct pmu_hw_events *cpuc,
> + struct perf_event *event)
> +{
> + int bit;
> + struct hw_perf_event *hwc = &event->hw;
> + unsigned int region;
> + unsigned int group;
> + bool venum_event, scorpion_event;
> +
> + krait_decode_event(hwc->config_base, ®ion, &group, NULL,
> + &venum_event, &scorpion_event);
> +
> + if (venum_event || scorpion_event) {
> + bit = scorpion_event_to_bit(event, region, group);
> + clear_bit(bit, cpuc->used_mask);
> + }
> +}
> +
> +static int scorpion_pmu_init(struct arm_pmu *cpu_pmu)
> +{
> + armv7pmu_init(cpu_pmu);
> + cpu_pmu->name = "armv7_scorpion";
> + cpu_pmu->map_event = scorpion_map_event;
> + cpu_pmu->num_events = armv7_read_num_pmnc_events();
> + cpu_pmu->reset = scorpion_pmu_reset;
> + cpu_pmu->enable = scorpion_pmu_enable_event;
> + cpu_pmu->disable = scorpion_pmu_disable_event;
> + cpu_pmu->get_event_idx = scorpion_pmu_get_event_idx;
> + cpu_pmu->clear_event_idx = scorpion_pmu_clear_event_idx;
> + return 0;
> +}
> #else
> static inline int armv7_a8_pmu_init(struct arm_pmu *cpu_pmu)
I forgot to add the empty scorpion_pmu_init() when
CONFIG_CPU_V7=n. If there's no other comments by the end of the
week I'll send a v2.
--
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
a Linux Foundation Collaborative Project
On 11 February 2015 at 13:27, Stephen Boyd <[email protected]> wrote:
> On 02/10, Ashwin Chaugule wrote:
>> Hi Stephen,
>>
>> On 10 February 2015 at 20:05, Stephen Boyd <[email protected]> wrote:
>> > Scorpion supports a set of local performance monitor event
>> > selection registers (LPM) sitting behind a cp15 based interface
>> > that extend the architected PMU events to include Scorpion CPU
>> > and Venum VFP specific events. To use these events the user is
>> > expected to program the lpm register with the event code shifted
>> > into the group they care about and then point the PMNx event at
>> > that region+group combo by writing a LPMn_GROUPx event. Add
>> > support for this hardware.
>> >
>> > Note: the raw event number is a pure software construct that
>> > allows us to map the multi-dimensional number space of regions,
>> > groups, and event codes into a flat event number space suitable
>> > for use by the perf framework.
>> >
>> > This is based on code originally written by Ashwin Chaugule and
>> > Neil Leeder [1] massed to become similar to the Krait PMU support
>> > code.
>>
>> Thanks for taking this up!
>> Overall this series looks good to me, but from what I faintly
>> recollect, doesn't this (and the Krait pmu code) get affected by
>> powercollapse issues anymore?
>> e.g.
>> https://www.codeaurora.org/cgit/quic/la/kernel/msm/commit/arch/arm/kernel/perf_event_msm.c?h=msm-3.4&id=b5ca687960f0fea2f4735e83ca5c9543474c19de
>>
>
> Right now there isn't any power collapse support in mainline so
> there's no immediate problem. Once we add power collapse support
> (i.e. cpuidle) to the Scorpion and Krait platforms we'll need to
> do something in the perf event code to properly maintain the
> counts across idle. I imagine it would be done by registering for
> cpu_pm notifications and then doing the save/restore on
> CPU_PM_ENTER and CPU_PM_EXIT. At least, that's what you started
> doing in this patch[1]. And then it seems the patch you mention
> came after that and actually did the save/restore of the counts.
>
> [1] https://www.codeaurora.org/cgit/quic/la/kernel/msm/commit/?h=msm-3.4&id=464983a7e991a484cac0bc0885cee4fee318d659
Right. Thats essential whenever the power collapse stuff goes in.
Thanks,
Ashwin.
Hi,
I haven't given this a thorough review, but I spotted a couple of items
below.
On Wed, Feb 11, 2015 at 01:05:24AM +0000, Stephen Boyd wrote:
> Scorpion supports a set of local performance monitor event
> selection registers (LPM) sitting behind a cp15 based interface
> that extend the architected PMU events to include Scorpion CPU
> and Venum VFP specific events. To use these events the user is
> expected to program the lpm register with the event code shifted
> into the group they care about and then point the PMNx event at
> that region+group combo by writing a LPMn_GROUPx event. Add
> support for this hardware.
>
> Note: the raw event number is a pure software construct that
> allows us to map the multi-dimensional number space of regions,
> groups, and event codes into a flat event number space suitable
> for use by the perf framework.
>
> This is based on code originally written by Ashwin Chaugule and
> Neil Leeder [1] massed to become similar to the Krait PMU support
> code.
>
> [1] https://www.codeaurora.org/cgit/quic/la/kernel/msm/tree/arch/arm/kernel/perf_event_msm.c?h=msm-3.4
>
> Cc: Neil Leeder <[email protected]>
> Cc: Ashwin Chaugule <[email protected]>
> Cc: <[email protected]>
> Signed-off-by: Stephen Boyd <[email protected]>
> ---
> Documentation/devicetree/bindings/arm/pmu.txt | 2 +
> arch/arm/kernel/perf_event_cpu.c | 2 +
> arch/arm/kernel/perf_event_v7.c | 395 ++++++++++++++++++++++++++
> 3 files changed, 399 insertions(+)
>
> diff --git a/Documentation/devicetree/bindings/arm/pmu.txt b/Documentation/devicetree/bindings/arm/pmu.txt
> index 75ef91d08f3b..6e54a9d88b7a 100644
> --- a/Documentation/devicetree/bindings/arm/pmu.txt
> +++ b/Documentation/devicetree/bindings/arm/pmu.txt
> @@ -18,6 +18,8 @@ Required properties:
> "arm,arm11mpcore-pmu"
> "arm,arm1176-pmu"
> "arm,arm1136-pmu"
> + "qcom,scorpion-pmu"
> + "qcom,scorpion-mp-pmu"
Is the PMU any different in the MP and !MP variants? The code doesn't
seem to handle the two any differently and will pass either to userspace
as "armv7_scorpion".
If there is some difference that we don't handle right now, that's fine,
it just looks a little odd.
[...]
> +static const unsigned scorpion_perf_cache_map[PERF_COUNT_HW_CACHE_MAX]
> + [PERF_COUNT_HW_CACHE_OP_MAX]
> + [PERF_COUNT_HW_CACHE_RESULT_MAX] = {
> + PERF_CACHE_MAP_ALL_UNSUPPORTED,
> + /*
> + * The performance counters don't differentiate between read and write
> + * accesses/misses so this isn't strictly correct, but it's the best we
> + * can do. Writes and reads get combined.
> + */
> + [C(L1D)][C(OP_READ)][C(RESULT_ACCESS)] = ARMV7_PERFCTR_L1_DCACHE_ACCESS,
> + [C(L1D)][C(OP_READ)][C(RESULT_MISS)] = ARMV7_PERFCTR_L1_DCACHE_REFILL,
> + [C(L1D)][C(OP_WRITE)][C(RESULT_ACCESS)] = ARMV7_PERFCTR_L1_DCACHE_ACCESS,
> + [C(L1D)][C(OP_WRITE)][C(RESULT_MISS)] = ARMV7_PERFCTR_L1_DCACHE_REFILL,
> + [C(L1I)][C(OP_READ)][C(RESULT_ACCESS)] = SCORPION_ICACHE_ACCESS,
> + [C(L1I)][C(OP_READ)][C(RESULT_MISS)] = SCORPION_ICACHE_MISS,
> + [C(L1I)][C(OP_WRITE)][C(RESULT_ACCESS)] = SCORPION_ICACHE_ACCESS,
> + [C(L1I)][C(OP_WRITE)][C(RESULT_MISS)] = SCORPION_ICACHE_MISS,
These last two entries go against the policy we set in commit
40c390c768f89849: "ARM: perf: don't pretend to support counting of L1I
writes", so I think they should be dropped.
> + /*
> + * Only ITLB misses and DTLB refills are supported. If users want the
> + * DTLB refills misses a raw counter must be used.
> + */
> + [C(DTLB)][C(OP_READ)][C(RESULT_ACCESS)] = SCORPION_DTLB_ACCESS,
> + [C(DTLB)][C(OP_READ)][C(RESULT_MISS)] = SCORPION_DTLB_MISS,
> + [C(DTLB)][C(OP_WRITE)][C(RESULT_ACCESS)] = SCORPION_DTLB_ACCESS,
> + [C(DTLB)][C(OP_WRITE)][C(RESULT_MISS)] = SCORPION_DTLB_MISS,
> + [C(ITLB)][C(OP_READ)][C(RESULT_MISS)] = SCORPION_ITLB_MISS,
> + [C(ITLB)][C(OP_WRITE)][C(RESULT_MISS)] = SCORPION_ITLB_MISS,
> + [C(BPU)][C(OP_READ)][C(RESULT_ACCESS)] = ARMV7_PERFCTR_PC_BRANCH_PRED,
> + [C(BPU)][C(OP_READ)][C(RESULT_MISS)] = ARMV7_PERFCTR_PC_BRANCH_PRED,
> + [C(BPU)][C(OP_WRITE)][C(RESULT_ACCESS)] = ARMV7_PERFCTR_PC_BRANCH_PRED,
> + [C(BPU)][C(OP_WRITE)][C(RESULT_MISS)] = ARMV7_PERFCTR_PC_BRANCH_PRED,
> +};
Not ARMV7_PERFCTR_PC_BRANCH_MIS_PRED for the RESULT_MISS cases as with
all other ARMv7 instances (Krait included)?
Thanks,
Mark.
On 02/12/15 04:49, Mark Rutland wrote:
> Hi,
>
> I haven't given this a thorough review, but I spotted a couple of items
> below.
>
> On Wed, Feb 11, 2015 at 01:05:24AM +0000, Stephen Boyd wrote:
>> Scorpion supports a set of local performance monitor event
>> selection registers (LPM) sitting behind a cp15 based interface
>> that extend the architected PMU events to include Scorpion CPU
>> and Venum VFP specific events. To use these events the user is
>> expected to program the lpm register with the event code shifted
>> into the group they care about and then point the PMNx event at
>> that region+group combo by writing a LPMn_GROUPx event. Add
>> support for this hardware.
>>
>> Note: the raw event number is a pure software construct that
>> allows us to map the multi-dimensional number space of regions,
>> groups, and event codes into a flat event number space suitable
>> for use by the perf framework.
>>
>> This is based on code originally written by Ashwin Chaugule and
>> Neil Leeder [1] massed to become similar to the Krait PMU support
>> code.
>>
>> [1] https://www.codeaurora.org/cgit/quic/la/kernel/msm/tree/arch/arm/kernel/perf_event_msm.c?h=msm-3.4
>>
>> Cc: Neil Leeder <[email protected]>
>> Cc: Ashwin Chaugule <[email protected]>
>> Cc: <[email protected]>
>> Signed-off-by: Stephen Boyd <[email protected]>
>> ---
>> Documentation/devicetree/bindings/arm/pmu.txt | 2 +
>> arch/arm/kernel/perf_event_cpu.c | 2 +
>> arch/arm/kernel/perf_event_v7.c | 395 ++++++++++++++++++++++++++
>> 3 files changed, 399 insertions(+)
>>
>> diff --git a/Documentation/devicetree/bindings/arm/pmu.txt b/Documentation/devicetree/bindings/arm/pmu.txt
>> index 75ef91d08f3b..6e54a9d88b7a 100644
>> --- a/Documentation/devicetree/bindings/arm/pmu.txt
>> +++ b/Documentation/devicetree/bindings/arm/pmu.txt
>> @@ -18,6 +18,8 @@ Required properties:
>> "arm,arm11mpcore-pmu"
>> "arm,arm1176-pmu"
>> "arm,arm1136-pmu"
>> + "qcom,scorpion-pmu"
>> + "qcom,scorpion-mp-pmu"
> Is the PMU any different in the MP and !MP variants? The code doesn't
> seem to handle the two any differently and will pass either to userspace
> as "armv7_scorpion".
>
> If there is some difference that we don't handle right now, that's fine,
> it just looks a little odd.
It seems that on MP there are two event encodings on MP that aren't
there on !MP and vice versa[1]. So I made two compatibles to reflect
that. I'll make two names that go to userspace to clarify this.
>> +static const unsigned scorpion_perf_cache_map[PERF_COUNT_HW_CACHE_MAX]
>> + [PERF_COUNT_HW_CACHE_OP_MAX]
>> + [PERF_COUNT_HW_CACHE_RESULT_MAX] = {
>> + PERF_CACHE_MAP_ALL_UNSUPPORTED,
>> + /*
>> + * The performance counters don't differentiate between read and write
>> + * accesses/misses so this isn't strictly correct, but it's the best we
>> + * can do. Writes and reads get combined.
>> + */
>> + [C(L1D)][C(OP_READ)][C(RESULT_ACCESS)] = ARMV7_PERFCTR_L1_DCACHE_ACCESS,
>> + [C(L1D)][C(OP_READ)][C(RESULT_MISS)] = ARMV7_PERFCTR_L1_DCACHE_REFILL,
>> + [C(L1D)][C(OP_WRITE)][C(RESULT_ACCESS)] = ARMV7_PERFCTR_L1_DCACHE_ACCESS,
>> + [C(L1D)][C(OP_WRITE)][C(RESULT_MISS)] = ARMV7_PERFCTR_L1_DCACHE_REFILL,
>> + [C(L1I)][C(OP_READ)][C(RESULT_ACCESS)] = SCORPION_ICACHE_ACCESS,
>> + [C(L1I)][C(OP_READ)][C(RESULT_MISS)] = SCORPION_ICACHE_MISS,
>> + [C(L1I)][C(OP_WRITE)][C(RESULT_ACCESS)] = SCORPION_ICACHE_ACCESS,
>> + [C(L1I)][C(OP_WRITE)][C(RESULT_MISS)] = SCORPION_ICACHE_MISS,
> These last two entries go against the policy we set in commit
> 40c390c768f89849: "ARM: perf: don't pretend to support counting of L1I
> writes", so I think they should be dropped.
Fair enough. Thanks for the pointer.
>
>> + /*
>> + * Only ITLB misses and DTLB refills are supported. If users want the
>> + * DTLB refills misses a raw counter must be used.
>> + */
>> + [C(DTLB)][C(OP_READ)][C(RESULT_ACCESS)] = SCORPION_DTLB_ACCESS,
>> + [C(DTLB)][C(OP_READ)][C(RESULT_MISS)] = SCORPION_DTLB_MISS,
>> + [C(DTLB)][C(OP_WRITE)][C(RESULT_ACCESS)] = SCORPION_DTLB_ACCESS,
>> + [C(DTLB)][C(OP_WRITE)][C(RESULT_MISS)] = SCORPION_DTLB_MISS,
>> + [C(ITLB)][C(OP_READ)][C(RESULT_MISS)] = SCORPION_ITLB_MISS,
>> + [C(ITLB)][C(OP_WRITE)][C(RESULT_MISS)] = SCORPION_ITLB_MISS,
>> + [C(BPU)][C(OP_READ)][C(RESULT_ACCESS)] = ARMV7_PERFCTR_PC_BRANCH_PRED,
>> + [C(BPU)][C(OP_READ)][C(RESULT_MISS)] = ARMV7_PERFCTR_PC_BRANCH_PRED,
>> + [C(BPU)][C(OP_WRITE)][C(RESULT_ACCESS)] = ARMV7_PERFCTR_PC_BRANCH_PRED,
>> + [C(BPU)][C(OP_WRITE)][C(RESULT_MISS)] = ARMV7_PERFCTR_PC_BRANCH_PRED,
>> +};
> Not ARMV7_PERFCTR_PC_BRANCH_MIS_PRED for the RESULT_MISS cases as with
> all other ARMv7 instances (Krait included)?
I was just copying the stuff from downstream. I think it's a bug that
nobody noticed because the same problem was there on Krait and I fixed
it before sending upstream. Thanks for catching it.
--
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
a Linux Foundation Collaborative Project