* [PATCH 0/5] perf/x86: Uncore updates
@ 2012-07-04 6:00 Yan, Zheng
2012-07-04 6:00 ` [PATCH 1/5] perf/x86: Use 0xff as pseudo code for fixed uncore event Yan, Zheng
` (4 more replies)
0 siblings, 5 replies; 21+ messages in thread
From: Yan, Zheng @ 2012-07-04 6:00 UTC (permalink / raw)
To: a.p.zijlstra, eranian, mingo, andi, linux-kernel
Hi, all
This series of patches consist of update/fix for the uncore driver.
The 1st, 2nd and the 4th patches are resending of old ones. The 3rd
patch fixes a bug that may cause GP fault on some types of CPU. The
5th patch adds Nehalem-EX uncore support.
Regards
Yan, Zheng
arch/x86/kernel/cpu/perf_event_intel_uncore.c | 1329 +++++++++++++++-
arch/x86/kernel/cpu/perf_event_intel_uncore.h | 219 ++-
2 files changed, 1471 insertions(+), 77 deletions(-)
^ permalink raw reply [flat|nested] 21+ messages in thread* [PATCH 1/5] perf/x86: Use 0xff as pseudo code for fixed uncore event 2012-07-04 6:00 [PATCH 0/5] perf/x86: Uncore updates Yan, Zheng @ 2012-07-04 6:00 ` Yan, Zheng 2012-07-04 6:00 ` [PATCH 2/5] perf/x86: Fix event constraint for SandyBridge-EP CBox Yan, Zheng ` (3 subsequent siblings) 4 siblings, 0 replies; 21+ messages in thread From: Yan, Zheng @ 2012-07-04 6:00 UTC (permalink / raw) To: a.p.zijlstra, eranian, mingo, andi, linux-kernel; +Cc: Yan, Zheng From: "Yan, Zheng" <zheng.z.yan@intel.com> Stephane Eranian suggestted using 0xff as pseudo code for fixed uncore event and using the umask value to determine which of the fixed events we want to map to. So far there is at most one fixed counter in a uncore PMU. So just change the definition of UNCORE_FIXED_EVENT to 0xff. Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com> --- arch/x86/kernel/cpu/perf_event_intel_uncore.c | 4 ++-- arch/x86/kernel/cpu/perf_event_intel_uncore.h | 2 +- 2 files changed, 3 insertions(+), 3 deletions(-) diff --git a/arch/x86/kernel/cpu/perf_event_intel_uncore.c b/arch/x86/kernel/cpu/perf_event_intel_uncore.c index 6f43f95..afa9970 100644 --- a/arch/x86/kernel/cpu/perf_event_intel_uncore.c +++ b/arch/x86/kernel/cpu/perf_event_intel_uncore.c @@ -179,7 +179,7 @@ static struct attribute *snbep_uncore_pcu_formats_attr[] = { }; static struct uncore_event_desc snbep_uncore_imc_events[] = { - INTEL_UNCORE_EVENT_DESC(clockticks, "event=0xff,umask=0xff"), + INTEL_UNCORE_EVENT_DESC(clockticks, "event=0xff,umask=0"), INTEL_UNCORE_EVENT_DESC(cas_count_read, "event=0x04,umask=0x03"), INTEL_UNCORE_EVENT_DESC(cas_count_write, "event=0x04,umask=0x0c"), { /* end: all zeroes */ }, @@ -616,7 +616,7 @@ static struct attribute_group nhm_uncore_format_group = { }; static struct uncore_event_desc nhm_uncore_events[] = { - INTEL_UNCORE_EVENT_DESC(clockticks, "event=0xff,umask=0xff"), + INTEL_UNCORE_EVENT_DESC(clockticks, "event=0xff,umask=0"), INTEL_UNCORE_EVENT_DESC(qmc_writes_full_any, "event=0x2f,umask=0x0f"), INTEL_UNCORE_EVENT_DESC(qmc_normal_reads_any, "event=0x2c,umask=0x0f"), INTEL_UNCORE_EVENT_DESC(qhl_request_ioh_reads, "event=0x20,umask=0x01"), diff --git a/arch/x86/kernel/cpu/perf_event_intel_uncore.h b/arch/x86/kernel/cpu/perf_event_intel_uncore.h index 4d52db0..88498c7 100644 --- a/arch/x86/kernel/cpu/perf_event_intel_uncore.h +++ b/arch/x86/kernel/cpu/perf_event_intel_uncore.h @@ -9,7 +9,7 @@ #define UNCORE_PMU_HRTIMER_INTERVAL (60 * NSEC_PER_SEC) -#define UNCORE_FIXED_EVENT 0xffff +#define UNCORE_FIXED_EVENT 0xff #define UNCORE_PMC_IDX_MAX_GENERIC 8 #define UNCORE_PMC_IDX_FIXED UNCORE_PMC_IDX_MAX_GENERIC #define UNCORE_PMC_IDX_MAX (UNCORE_PMC_IDX_FIXED + 1) -- 1.7.10.4 ^ permalink raw reply related [flat|nested] 21+ messages in thread
* [PATCH 2/5] perf/x86: Fix event constraint for SandyBridge-EP CBox 2012-07-04 6:00 [PATCH 0/5] perf/x86: Uncore updates Yan, Zheng 2012-07-04 6:00 ` [PATCH 1/5] perf/x86: Use 0xff as pseudo code for fixed uncore event Yan, Zheng @ 2012-07-04 6:00 ` Yan, Zheng 2012-07-04 6:00 ` [PATCH 3/5] perf/x86: Detect number of instances of uncore CBox Yan, Zheng ` (2 subsequent siblings) 4 siblings, 0 replies; 21+ messages in thread From: Yan, Zheng @ 2012-07-04 6:00 UTC (permalink / raw) To: a.p.zijlstra, eranian, mingo, andi, linux-kernel; +Cc: Yan, Zheng From: "Yan, Zheng" <zheng.z.yan@intel.com> The constraint for CBox event 0x1f should have overlap flag set. Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com> --- arch/x86/kernel/cpu/perf_event_intel_uncore.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/arch/x86/kernel/cpu/perf_event_intel_uncore.c b/arch/x86/kernel/cpu/perf_event_intel_uncore.c index afa9970..dbbf5a6 100644 --- a/arch/x86/kernel/cpu/perf_event_intel_uncore.c +++ b/arch/x86/kernel/cpu/perf_event_intel_uncore.c @@ -239,7 +239,7 @@ static struct event_constraint snbep_uncore_cbox_constraints[] = { UNCORE_EVENT_CONSTRAINT(0x1c, 0xc), UNCORE_EVENT_CONSTRAINT(0x1d, 0xc), UNCORE_EVENT_CONSTRAINT(0x1e, 0xc), - UNCORE_EVENT_CONSTRAINT(0x1f, 0xe), + EVENT_CONSTRAINT_OVERLAP(0x1f, 0xe, 0xff), UNCORE_EVENT_CONSTRAINT(0x21, 0x3), UNCORE_EVENT_CONSTRAINT(0x23, 0x3), UNCORE_EVENT_CONSTRAINT(0x31, 0x3), -- 1.7.10.4 ^ permalink raw reply related [flat|nested] 21+ messages in thread
* [PATCH 3/5] perf/x86: Detect number of instances of uncore CBox 2012-07-04 6:00 [PATCH 0/5] perf/x86: Uncore updates Yan, Zheng 2012-07-04 6:00 ` [PATCH 1/5] perf/x86: Use 0xff as pseudo code for fixed uncore event Yan, Zheng 2012-07-04 6:00 ` [PATCH 2/5] perf/x86: Fix event constraint for SandyBridge-EP CBox Yan, Zheng @ 2012-07-04 6:00 ` Yan, Zheng 2012-07-06 6:32 ` [tip:perf/core] " tip-bot for Yan, Zheng 2012-07-04 6:00 ` [PATCH 4/5] perf/x86: Uncore filter support for SandyBridge-EP Yan, Zheng 2012-07-04 6:00 ` [PATCH 5/5] perf/x86: Add Intel Nehalem-EX uncore support Yan, Zheng 4 siblings, 1 reply; 21+ messages in thread From: Yan, Zheng @ 2012-07-04 6:00 UTC (permalink / raw) To: a.p.zijlstra, eranian, mingo, andi, linux-kernel; +Cc: Yan, Zheng From: "Yan, Zheng" <zheng.z.yan@intel.com> The CBox manages the interface between the core and the LLC, so the instances of uncore CBox is equal to number of cores. Reported-by: Andrew Cooks <acooks@gmail.com> Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com> --- arch/x86/kernel/cpu/perf_event_intel_uncore.c | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-) diff --git a/arch/x86/kernel/cpu/perf_event_intel_uncore.c b/arch/x86/kernel/cpu/perf_event_intel_uncore.c index dbbf5a6..a394add 100644 --- a/arch/x86/kernel/cpu/perf_event_intel_uncore.c +++ b/arch/x86/kernel/cpu/perf_event_intel_uncore.c @@ -1605,8 +1605,9 @@ static void __init uncore_cpu_setup(void *dummy) static int __init uncore_cpu_init(void) { - int ret, cpu; + int ret, cpu, max_cores; + max_cores = boot_cpu_data.x86_max_cores; switch (boot_cpu_data.x86_model) { case 26: /* Nehalem */ case 30: @@ -1615,9 +1616,13 @@ static int __init uncore_cpu_init(void) msr_uncores = nhm_msr_uncores; break; case 42: /* Sandy Bridge */ + if (snb_uncore_cbox.num_boxes > max_cores) + snb_uncore_cbox.num_boxes = max_cores; msr_uncores = snb_msr_uncores; break; case 45: /* Sandy Birdge-EP */ + if (snbep_uncore_cbox.num_boxes > max_cores) + snbep_uncore_cbox.num_boxes = max_cores; msr_uncores = snbep_msr_uncores; break; default: -- 1.7.10.4 ^ permalink raw reply related [flat|nested] 21+ messages in thread
* [tip:perf/core] perf/x86: Detect number of instances of uncore CBox 2012-07-04 6:00 ` [PATCH 3/5] perf/x86: Detect number of instances of uncore CBox Yan, Zheng @ 2012-07-06 6:32 ` tip-bot for Yan, Zheng 0 siblings, 0 replies; 21+ messages in thread From: tip-bot for Yan, Zheng @ 2012-07-06 6:32 UTC (permalink / raw) To: linux-tip-commits Cc: linux-kernel, hpa, mingo, a.p.zijlstra, acooks, zheng.z.yan, tglx Commit-ID: 42089697244ba8e64fa43fb5e6d50d47a8e4cb00 Gitweb: http://git.kernel.org/tip/42089697244ba8e64fa43fb5e6d50d47a8e4cb00 Author: Yan, Zheng <zheng.z.yan@intel.com> AuthorDate: Wed, 4 Jul 2012 14:00:14 +0800 Committer: Ingo Molnar <mingo@kernel.org> CommitDate: Thu, 5 Jul 2012 21:56:00 +0200 perf/x86: Detect number of instances of uncore CBox The CBox manages the interface between the core and the LLC, so the instances of uncore CBox is equal to number of cores. Reported-by: Andrew Cooks <acooks@gmail.com> Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com> Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: http://lkml.kernel.org/r/1341381616-12229-4-git-send-email-zheng.z.yan@intel.com Signed-off-by: Ingo Molnar <mingo@kernel.org> --- arch/x86/kernel/cpu/perf_event_intel_uncore.c | 7 ++++++- 1 files changed, 6 insertions(+), 1 deletions(-) diff --git a/arch/x86/kernel/cpu/perf_event_intel_uncore.c b/arch/x86/kernel/cpu/perf_event_intel_uncore.c index 7d755d2..4fecbd0 100644 --- a/arch/x86/kernel/cpu/perf_event_intel_uncore.c +++ b/arch/x86/kernel/cpu/perf_event_intel_uncore.c @@ -1605,8 +1605,9 @@ static void __init uncore_cpu_setup(void *dummy) static int __init uncore_cpu_init(void) { - int ret, cpu; + int ret, cpu, max_cores; + max_cores = boot_cpu_data.x86_max_cores; switch (boot_cpu_data.x86_model) { case 26: /* Nehalem */ case 30: @@ -1615,9 +1616,13 @@ static int __init uncore_cpu_init(void) msr_uncores = nhm_msr_uncores; break; case 42: /* Sandy Bridge */ + if (snb_uncore_cbox.num_boxes > max_cores) + snb_uncore_cbox.num_boxes = max_cores; msr_uncores = snb_msr_uncores; break; case 45: /* Sandy Birdge-EP */ + if (snbep_uncore_cbox.num_boxes > max_cores) + snbep_uncore_cbox.num_boxes = max_cores; msr_uncores = snbep_msr_uncores; break; default: ^ permalink raw reply related [flat|nested] 21+ messages in thread
* [PATCH 4/5] perf/x86: Uncore filter support for SandyBridge-EP 2012-07-04 6:00 [PATCH 0/5] perf/x86: Uncore updates Yan, Zheng ` (2 preceding siblings ...) 2012-07-04 6:00 ` [PATCH 3/5] perf/x86: Detect number of instances of uncore CBox Yan, Zheng @ 2012-07-04 6:00 ` Yan, Zheng 2012-07-06 6:33 ` [tip:perf/core] " tip-bot for Yan, Zheng 2012-07-22 19:25 ` [PATCH 4/5] " Stephane Eranian 2012-07-04 6:00 ` [PATCH 5/5] perf/x86: Add Intel Nehalem-EX uncore support Yan, Zheng 4 siblings, 2 replies; 21+ messages in thread From: Yan, Zheng @ 2012-07-04 6:00 UTC (permalink / raw) To: a.p.zijlstra, eranian, mingo, andi, linux-kernel; +Cc: Yan, Zheng From: "Yan, Zheng" <zheng.z.yan@intel.com> This patch adds C-Box and PCU filter support for SandyBridge-EP uncore. We can filter C-Box events by thread/core ID and filter PCU events by frequency/voltage. Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com> --- arch/x86/kernel/cpu/perf_event_intel_uncore.c | 228 ++++++++++++++++++++----- arch/x86/kernel/cpu/perf_event_intel_uncore.h | 24 ++- 2 files changed, 206 insertions(+), 46 deletions(-) diff --git a/arch/x86/kernel/cpu/perf_event_intel_uncore.c b/arch/x86/kernel/cpu/perf_event_intel_uncore.c index a394add..c0a80ff 100644 --- a/arch/x86/kernel/cpu/perf_event_intel_uncore.c +++ b/arch/x86/kernel/cpu/perf_event_intel_uncore.c @@ -14,10 +14,13 @@ static cpumask_t uncore_cpu_mask; /* constraint for the fixed counter */ static struct event_constraint constraint_fixed = EVENT_CONSTRAINT(~0ULL, 1 << UNCORE_PMC_IDX_FIXED, ~0ULL); +static struct event_constraint constraint_empty = + EVENT_CONSTRAINT(0, 0, 0); DEFINE_UNCORE_FORMAT_ATTR(event, event, "config:0-7"); DEFINE_UNCORE_FORMAT_ATTR(umask, umask, "config:8-15"); DEFINE_UNCORE_FORMAT_ATTR(edge, edge, "config:18"); +DEFINE_UNCORE_FORMAT_ATTR(tid_en, tid_en, "config:19"); DEFINE_UNCORE_FORMAT_ATTR(inv, inv, "config:23"); DEFINE_UNCORE_FORMAT_ATTR(cmask5, cmask, "config:24-28"); DEFINE_UNCORE_FORMAT_ATTR(cmask8, cmask, "config:24-31"); @@ -26,8 +29,19 @@ DEFINE_UNCORE_FORMAT_ATTR(thresh5, thresh, "config:24-28"); DEFINE_UNCORE_FORMAT_ATTR(occ_sel, occ_sel, "config:14-15"); DEFINE_UNCORE_FORMAT_ATTR(occ_invert, occ_invert, "config:30"); DEFINE_UNCORE_FORMAT_ATTR(occ_edge, occ_edge, "config:14-51"); +DEFINE_UNCORE_FORMAT_ATTR(filter_tid, filter_tid, "config1:0-4"); +DEFINE_UNCORE_FORMAT_ATTR(filter_nid, filter_nid, "config1:10-17"); +DEFINE_UNCORE_FORMAT_ATTR(filter_state, filter_state, "config1:18-22"); +DEFINE_UNCORE_FORMAT_ATTR(filter_opc, filter_opc, "config1:23-31"); +DEFINE_UNCORE_FORMAT_ATTR(filter_brand0, filter_brand0, "config1:0-7"); +DEFINE_UNCORE_FORMAT_ATTR(filter_brand1, filter_brand1, "config1:8-15"); +DEFINE_UNCORE_FORMAT_ATTR(filter_brand2, filter_brand2, "config1:16-23"); +DEFINE_UNCORE_FORMAT_ATTR(filter_brand3, filter_brand3, "config1:24-31"); /* Sandy Bridge-EP uncore support */ +static struct intel_uncore_type snbep_uncore_cbox; +static struct intel_uncore_type snbep_uncore_pcu; + static void snbep_uncore_pci_disable_box(struct intel_uncore_box *box) { struct pci_dev *pdev = box->pci_dev; @@ -120,6 +134,10 @@ static void snbep_uncore_msr_enable_event(struct intel_uncore_box *box, struct perf_event *event) { struct hw_perf_event *hwc = &event->hw; + struct hw_perf_event_extra *reg1 = &hwc->extra_reg; + + if (reg1->idx != EXTRA_REG_NONE) + wrmsrl(reg1->reg, reg1->config); wrmsrl(hwc->config_base, hwc->config | SNBEP_PMON_CTL_EN); } @@ -149,6 +167,71 @@ static void snbep_uncore_msr_init_box(struct intel_uncore_box *box) wrmsrl(msr, SNBEP_PMON_BOX_CTL_INT); } +static struct event_constraint * +snbep_uncore_get_constraint(struct intel_uncore_box *box, + struct perf_event *event) +{ + struct intel_uncore_extra_reg *er; + struct hw_perf_event_extra *reg1 = &event->hw.extra_reg; + unsigned long flags; + bool ok = false; + + if (reg1->idx == EXTRA_REG_NONE || (box->phys_id >= 0 && reg1->alloc)) + return NULL; + + er = &box->shared_regs[reg1->idx]; + raw_spin_lock_irqsave(&er->lock, flags); + if (!atomic_read(&er->ref) || er->config1 == reg1->config) { + atomic_inc(&er->ref); + er->config1 = reg1->config; + ok = true; + } + raw_spin_unlock_irqrestore(&er->lock, flags); + + if (ok) { + if (box->phys_id >= 0) + reg1->alloc = 1; + return NULL; + } + return &constraint_empty; +} + +static void snbep_uncore_put_constraint(struct intel_uncore_box *box, + struct perf_event *event) +{ + struct intel_uncore_extra_reg *er; + struct hw_perf_event_extra *reg1 = &event->hw.extra_reg; + + if (box->phys_id < 0 || !reg1->alloc) + return; + + er = &box->shared_regs[reg1->idx]; + atomic_dec(&er->ref); + reg1->alloc = 0; +} + +static int snbep_uncore_hw_config(struct intel_uncore_box *box, + struct perf_event *event) +{ + struct hw_perf_event *hwc = &event->hw; + struct hw_perf_event_extra *reg1 = &hwc->extra_reg; + + if (box->pmu->type == &snbep_uncore_cbox) { + reg1->reg = SNBEP_C0_MSR_PMON_BOX_FILTER + + SNBEP_CBO_MSR_OFFSET * box->pmu->pmu_idx; + reg1->config = event->attr.config1 & + SNBEP_CB0_MSR_PMON_BOX_FILTER_MASK; + } else if (box->pmu->type == &snbep_uncore_pcu) { + reg1->reg = SNBEP_PCU_MSR_PMON_BOX_FILTER; + reg1->config = event->attr.config1 & + SNBEP_PCU_MSR_PMON_BOX_FILTER_MASK; + } else { + return 0; + } + reg1->idx = 0; + return 0; +} + static struct attribute *snbep_uncore_formats_attr[] = { &format_attr_event.attr, &format_attr_umask.attr, @@ -167,6 +250,20 @@ static struct attribute *snbep_uncore_ubox_formats_attr[] = { NULL, }; +static struct attribute *snbep_uncore_cbox_formats_attr[] = { + &format_attr_event.attr, + &format_attr_umask.attr, + &format_attr_edge.attr, + &format_attr_tid_en.attr, + &format_attr_inv.attr, + &format_attr_thresh8.attr, + &format_attr_filter_tid.attr, + &format_attr_filter_nid.attr, + &format_attr_filter_state.attr, + &format_attr_filter_opc.attr, + NULL, +}; + static struct attribute *snbep_uncore_pcu_formats_attr[] = { &format_attr_event.attr, &format_attr_occ_sel.attr, @@ -175,6 +272,10 @@ static struct attribute *snbep_uncore_pcu_formats_attr[] = { &format_attr_thresh5.attr, &format_attr_occ_invert.attr, &format_attr_occ_edge.attr, + &format_attr_filter_brand0.attr, + &format_attr_filter_brand1.attr, + &format_attr_filter_brand2.attr, + &format_attr_filter_brand3.attr, NULL, }; @@ -203,6 +304,11 @@ static struct attribute_group snbep_uncore_ubox_format_group = { .attrs = snbep_uncore_ubox_formats_attr, }; +static struct attribute_group snbep_uncore_cbox_format_group = { + .name = "format", + .attrs = snbep_uncore_cbox_formats_attr, +}; + static struct attribute_group snbep_uncore_pcu_format_group = { .name = "format", .attrs = snbep_uncore_pcu_formats_attr, @@ -215,6 +321,9 @@ static struct intel_uncore_ops snbep_uncore_msr_ops = { .disable_event = snbep_uncore_msr_disable_event, .enable_event = snbep_uncore_msr_enable_event, .read_counter = snbep_uncore_msr_read_counter, + .get_constraint = snbep_uncore_get_constraint, + .put_constraint = snbep_uncore_put_constraint, + .hw_config = snbep_uncore_hw_config, }; static struct intel_uncore_ops snbep_uncore_pci_ops = { @@ -307,31 +416,33 @@ static struct intel_uncore_type snbep_uncore_ubox = { }; static struct intel_uncore_type snbep_uncore_cbox = { - .name = "cbox", - .num_counters = 4, - .num_boxes = 8, - .perf_ctr_bits = 44, - .event_ctl = SNBEP_C0_MSR_PMON_CTL0, - .perf_ctr = SNBEP_C0_MSR_PMON_CTR0, - .event_mask = SNBEP_PMON_RAW_EVENT_MASK, - .box_ctl = SNBEP_C0_MSR_PMON_BOX_CTL, - .msr_offset = SNBEP_CBO_MSR_OFFSET, - .constraints = snbep_uncore_cbox_constraints, - .ops = &snbep_uncore_msr_ops, - .format_group = &snbep_uncore_format_group, + .name = "cbox", + .num_counters = 4, + .num_boxes = 8, + .perf_ctr_bits = 44, + .event_ctl = SNBEP_C0_MSR_PMON_CTL0, + .perf_ctr = SNBEP_C0_MSR_PMON_CTR0, + .event_mask = SNBEP_CBO_MSR_PMON_RAW_EVENT_MASK, + .box_ctl = SNBEP_C0_MSR_PMON_BOX_CTL, + .msr_offset = SNBEP_CBO_MSR_OFFSET, + .num_shared_regs = 1, + .constraints = snbep_uncore_cbox_constraints, + .ops = &snbep_uncore_msr_ops, + .format_group = &snbep_uncore_cbox_format_group, }; static struct intel_uncore_type snbep_uncore_pcu = { - .name = "pcu", - .num_counters = 4, - .num_boxes = 1, - .perf_ctr_bits = 48, - .perf_ctr = SNBEP_PCU_MSR_PMON_CTR0, - .event_ctl = SNBEP_PCU_MSR_PMON_CTL0, - .event_mask = SNBEP_PCU_MSR_PMON_RAW_EVENT_MASK, - .box_ctl = SNBEP_PCU_MSR_PMON_BOX_CTL, - .ops = &snbep_uncore_msr_ops, - .format_group = &snbep_uncore_pcu_format_group, + .name = "pcu", + .num_counters = 4, + .num_boxes = 1, + .perf_ctr_bits = 48, + .perf_ctr = SNBEP_PCU_MSR_PMON_CTR0, + .event_ctl = SNBEP_PCU_MSR_PMON_CTL0, + .event_mask = SNBEP_PCU_MSR_PMON_RAW_EVENT_MASK, + .box_ctl = SNBEP_PCU_MSR_PMON_BOX_CTL, + .num_shared_regs = 1, + .ops = &snbep_uncore_msr_ops, + .format_group = &snbep_uncore_pcu_format_group, }; static struct intel_uncore_type *snbep_msr_uncores[] = { @@ -747,15 +858,22 @@ static void uncore_pmu_init_hrtimer(struct intel_uncore_box *box) box->hrtimer.function = uncore_pmu_hrtimer; } -struct intel_uncore_box *uncore_alloc_box(int cpu) +struct intel_uncore_box *uncore_alloc_box(struct intel_uncore_type *type, + int cpu) { struct intel_uncore_box *box; + int i, size; - box = kmalloc_node(sizeof(*box), GFP_KERNEL | __GFP_ZERO, - cpu_to_node(cpu)); + size = sizeof(*box) + type->num_shared_regs * + sizeof(struct intel_uncore_extra_reg); + + box = kmalloc_node(size, GFP_KERNEL | __GFP_ZERO, cpu_to_node(cpu)); if (!box) return NULL; + for (i = 0; i < type->num_shared_regs; i++) + raw_spin_lock_init(&box->shared_regs[i].lock); + uncore_pmu_init_hrtimer(box); atomic_set(&box->refcnt, 1); box->cpu = -1; @@ -834,11 +952,18 @@ static int uncore_collect_events(struct intel_uncore_box *box, } static struct event_constraint * -uncore_event_constraint(struct intel_uncore_type *type, - struct perf_event *event) +uncore_get_event_constraint(struct intel_uncore_box *box, + struct perf_event *event) { + struct intel_uncore_type *type = box->pmu->type; struct event_constraint *c; + if (type->ops->get_constraint) { + c = type->ops->get_constraint(box, event); + if (c) + return c; + } + if (event->hw.config == ~0ULL) return &constraint_fixed; @@ -852,19 +977,25 @@ uncore_event_constraint(struct intel_uncore_type *type, return &type->unconstrainted; } +static void uncore_put_event_constraint(struct intel_uncore_box *box, + struct perf_event *event) +{ + if (box->pmu->type->ops->put_constraint) + box->pmu->type->ops->put_constraint(box, event); +} + static int uncore_assign_events(struct intel_uncore_box *box, int assign[], int n) { unsigned long used_mask[BITS_TO_LONGS(UNCORE_PMC_IDX_MAX)]; struct event_constraint *c, *constraints[UNCORE_PMC_IDX_MAX]; - int i, ret, wmin, wmax; + int i, wmin, wmax, ret = 0; struct hw_perf_event *hwc; bitmap_zero(used_mask, UNCORE_PMC_IDX_MAX); for (i = 0, wmin = UNCORE_PMC_IDX_MAX, wmax = 0; i < n; i++) { - c = uncore_event_constraint(box->pmu->type, - box->event_list[i]); + c = uncore_get_event_constraint(box, box->event_list[i]); constraints[i] = c; wmin = min(wmin, c->weight); wmax = max(wmax, c->weight); @@ -888,13 +1019,17 @@ static int uncore_assign_events(struct intel_uncore_box *box, break; __set_bit(hwc->idx, used_mask); - assign[i] = hwc->idx; + if (assign) + assign[i] = hwc->idx; } - if (i == n) - return 0; - /* slow path */ - ret = perf_assign_events(constraints, n, wmin, wmax, assign); + if (i != n) + ret = perf_assign_events(constraints, n, wmin, wmax, assign); + + if (!assign || ret) { + for (i = 0; i < n; i++) + uncore_put_event_constraint(box, box->event_list[i]); + } return ret ? -EINVAL : 0; } @@ -1021,6 +1156,8 @@ static void uncore_pmu_event_del(struct perf_event *event, int flags) for (i = 0; i < box->n_events; i++) { if (event == box->event_list[i]) { + uncore_put_event_constraint(box, event); + while (++i < box->n_events) box->event_list[i - 1] = box->event_list[i]; @@ -1048,10 +1185,9 @@ static int uncore_validate_group(struct intel_uncore_pmu *pmu, { struct perf_event *leader = event->group_leader; struct intel_uncore_box *fake_box; - int assign[UNCORE_PMC_IDX_MAX]; int ret = -EINVAL, n; - fake_box = uncore_alloc_box(smp_processor_id()); + fake_box = uncore_alloc_box(pmu->type, smp_processor_id()); if (!fake_box) return -ENOMEM; @@ -1073,7 +1209,7 @@ static int uncore_validate_group(struct intel_uncore_pmu *pmu, fake_box->n_events = n; - ret = uncore_assign_events(fake_box, assign, n); + ret = uncore_assign_events(fake_box, NULL, n); out: kfree(fake_box); return ret; @@ -1117,6 +1253,10 @@ int uncore_pmu_event_init(struct perf_event *event) return -EINVAL; event->cpu = box->cpu; + event->hw.idx = -1; + event->hw.last_tag = ~0ULL; + event->hw.extra_reg.idx = EXTRA_REG_NONE; + if (event->attr.config == UNCORE_FIXED_EVENT) { /* no fixed counter */ if (!pmu->type->fixed_ctl) @@ -1130,11 +1270,13 @@ int uncore_pmu_event_init(struct perf_event *event) hwc->config = ~0ULL; } else { hwc->config = event->attr.config & pmu->type->event_mask; + if (pmu->type->ops->hw_config) { + ret = pmu->type->ops->hw_config(box, event); + if (ret) + return ret; + } } - event->hw.idx = -1; - event->hw.last_tag = ~0ULL; - if (event->group_leader != event) ret = uncore_validate_group(pmu, event); else @@ -1276,7 +1418,7 @@ static int __devinit uncore_pci_add(struct intel_uncore_type *type, if (phys_id < 0) return -ENODEV; - box = uncore_alloc_box(0); + box = uncore_alloc_box(type, 0); if (!box) return -ENOMEM; @@ -1458,7 +1600,7 @@ static int __cpuinit uncore_cpu_prepare(int cpu, int phys_id) if (pmu->func_id < 0) pmu->func_id = j; - box = uncore_alloc_box(cpu); + box = uncore_alloc_box(type, cpu); if (!box) return -ENOMEM; diff --git a/arch/x86/kernel/cpu/perf_event_intel_uncore.h b/arch/x86/kernel/cpu/perf_event_intel_uncore.h index 88498c7..b13e9ea 100644 --- a/arch/x86/kernel/cpu/perf_event_intel_uncore.h +++ b/arch/x86/kernel/cpu/perf_event_intel_uncore.h @@ -97,6 +97,10 @@ SNBEP_PMON_CTL_INVERT | \ SNBEP_U_MSR_PMON_CTL_TRESH_MASK) +#define SNBEP_CBO_PMON_CTL_TID_EN (1 << 19) +#define SNBEP_CBO_MSR_PMON_RAW_EVENT_MASK (SNBEP_PMON_RAW_EVENT_MASK | \ + SNBEP_CBO_PMON_CTL_TID_EN) + /* SNB-EP PCU event control */ #define SNBEP_PCU_MSR_PMON_CTL_OCC_SEL_MASK 0x0000c000 #define SNBEP_PCU_MSR_PMON_CTL_TRESH_MASK 0x1f000000 @@ -140,15 +144,17 @@ /* SNB-EP Cbo register */ #define SNBEP_C0_MSR_PMON_CTR0 0xd16 #define SNBEP_C0_MSR_PMON_CTL0 0xd10 -#define SNBEP_C0_MSR_PMON_BOX_FILTER 0xd14 #define SNBEP_C0_MSR_PMON_BOX_CTL 0xd04 +#define SNBEP_C0_MSR_PMON_BOX_FILTER 0xd14 +#define SNBEP_CB0_MSR_PMON_BOX_FILTER_MASK 0xfffffc1f #define SNBEP_CBO_MSR_OFFSET 0x20 /* SNB-EP PCU register */ #define SNBEP_PCU_MSR_PMON_CTR0 0xc36 #define SNBEP_PCU_MSR_PMON_CTL0 0xc30 -#define SNBEP_PCU_MSR_PMON_BOX_FILTER 0xc34 #define SNBEP_PCU_MSR_PMON_BOX_CTL 0xc24 +#define SNBEP_PCU_MSR_PMON_BOX_FILTER 0xc34 +#define SNBEP_PCU_MSR_PMON_BOX_FILTER_MASK 0xffffffff #define SNBEP_PCU_MSR_CORE_C3_CTR 0x3fc #define SNBEP_PCU_MSR_CORE_C6_CTR 0x3fd @@ -163,7 +169,6 @@ struct intel_uncore_type { int num_boxes; int perf_ctr_bits; int fixed_ctr_bits; - int single_fixed; unsigned perf_ctr; unsigned event_ctl; unsigned event_mask; @@ -171,6 +176,8 @@ struct intel_uncore_type { unsigned fixed_ctl; unsigned box_ctl; unsigned msr_offset; + unsigned num_shared_regs:8; + unsigned single_fixed:1; struct event_constraint unconstrainted; struct event_constraint *constraints; struct intel_uncore_pmu *pmus; @@ -188,6 +195,10 @@ struct intel_uncore_ops { void (*disable_event)(struct intel_uncore_box *, struct perf_event *); void (*enable_event)(struct intel_uncore_box *, struct perf_event *); u64 (*read_counter)(struct intel_uncore_box *, struct perf_event *); + int (*hw_config)(struct intel_uncore_box *, struct perf_event *); + struct event_constraint *(*get_constraint)(struct intel_uncore_box *, + struct perf_event *); + void (*put_constraint)(struct intel_uncore_box *, struct perf_event *); }; struct intel_uncore_pmu { @@ -200,6 +211,12 @@ struct intel_uncore_pmu { struct list_head box_list; }; +struct intel_uncore_extra_reg { + raw_spinlock_t lock; + u64 config1; + atomic_t ref; +}; + struct intel_uncore_box { int phys_id; int n_active; /* number of active events */ @@ -215,6 +232,7 @@ struct intel_uncore_box { struct intel_uncore_pmu *pmu; struct hrtimer hrtimer; struct list_head list; + struct intel_uncore_extra_reg shared_regs[0]; }; #define UNCORE_BOX_FLAG_INITIATED 0 -- 1.7.10.4 ^ permalink raw reply related [flat|nested] 21+ messages in thread
* [tip:perf/core] perf/x86: Uncore filter support for SandyBridge-EP 2012-07-04 6:00 ` [PATCH 4/5] perf/x86: Uncore filter support for SandyBridge-EP Yan, Zheng @ 2012-07-06 6:33 ` tip-bot for Yan, Zheng 2012-07-22 19:25 ` [PATCH 4/5] " Stephane Eranian 1 sibling, 0 replies; 21+ messages in thread From: tip-bot for Yan, Zheng @ 2012-07-06 6:33 UTC (permalink / raw) To: linux-tip-commits Cc: linux-kernel, hpa, mingo, a.p.zijlstra, zheng.z.yan, tglx Commit-ID: 6a67943a18c264d5f3df436da38edb3e59adc905 Gitweb: http://git.kernel.org/tip/6a67943a18c264d5f3df436da38edb3e59adc905 Author: Yan, Zheng <zheng.z.yan@intel.com> AuthorDate: Wed, 4 Jul 2012 14:00:15 +0800 Committer: Ingo Molnar <mingo@kernel.org> CommitDate: Thu, 5 Jul 2012 21:56:01 +0200 perf/x86: Uncore filter support for SandyBridge-EP This patch adds C-Box and PCU filter support for SandyBridge-EP uncore. We can filter C-Box events by thread/core ID and filter PCU events by frequency/voltage. Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com> Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: http://lkml.kernel.org/r/1341381616-12229-5-git-send-email-zheng.z.yan@intel.com Signed-off-by: Ingo Molnar <mingo@kernel.org> --- arch/x86/kernel/cpu/perf_event_intel_uncore.c | 228 ++++++++++++++++++++----- arch/x86/kernel/cpu/perf_event_intel_uncore.h | 24 +++- 2 files changed, 206 insertions(+), 46 deletions(-) diff --git a/arch/x86/kernel/cpu/perf_event_intel_uncore.c b/arch/x86/kernel/cpu/perf_event_intel_uncore.c index 4fecbd0..19faffc 100644 --- a/arch/x86/kernel/cpu/perf_event_intel_uncore.c +++ b/arch/x86/kernel/cpu/perf_event_intel_uncore.c @@ -14,10 +14,13 @@ static cpumask_t uncore_cpu_mask; /* constraint for the fixed counter */ static struct event_constraint constraint_fixed = EVENT_CONSTRAINT(~0ULL, 1 << UNCORE_PMC_IDX_FIXED, ~0ULL); +static struct event_constraint constraint_empty = + EVENT_CONSTRAINT(0, 0, 0); DEFINE_UNCORE_FORMAT_ATTR(event, event, "config:0-7"); DEFINE_UNCORE_FORMAT_ATTR(umask, umask, "config:8-15"); DEFINE_UNCORE_FORMAT_ATTR(edge, edge, "config:18"); +DEFINE_UNCORE_FORMAT_ATTR(tid_en, tid_en, "config:19"); DEFINE_UNCORE_FORMAT_ATTR(inv, inv, "config:23"); DEFINE_UNCORE_FORMAT_ATTR(cmask5, cmask, "config:24-28"); DEFINE_UNCORE_FORMAT_ATTR(cmask8, cmask, "config:24-31"); @@ -26,8 +29,19 @@ DEFINE_UNCORE_FORMAT_ATTR(thresh5, thresh, "config:24-28"); DEFINE_UNCORE_FORMAT_ATTR(occ_sel, occ_sel, "config:14-15"); DEFINE_UNCORE_FORMAT_ATTR(occ_invert, occ_invert, "config:30"); DEFINE_UNCORE_FORMAT_ATTR(occ_edge, occ_edge, "config:14-51"); +DEFINE_UNCORE_FORMAT_ATTR(filter_tid, filter_tid, "config1:0-4"); +DEFINE_UNCORE_FORMAT_ATTR(filter_nid, filter_nid, "config1:10-17"); +DEFINE_UNCORE_FORMAT_ATTR(filter_state, filter_state, "config1:18-22"); +DEFINE_UNCORE_FORMAT_ATTR(filter_opc, filter_opc, "config1:23-31"); +DEFINE_UNCORE_FORMAT_ATTR(filter_brand0, filter_brand0, "config1:0-7"); +DEFINE_UNCORE_FORMAT_ATTR(filter_brand1, filter_brand1, "config1:8-15"); +DEFINE_UNCORE_FORMAT_ATTR(filter_brand2, filter_brand2, "config1:16-23"); +DEFINE_UNCORE_FORMAT_ATTR(filter_brand3, filter_brand3, "config1:24-31"); /* Sandy Bridge-EP uncore support */ +static struct intel_uncore_type snbep_uncore_cbox; +static struct intel_uncore_type snbep_uncore_pcu; + static void snbep_uncore_pci_disable_box(struct intel_uncore_box *box) { struct pci_dev *pdev = box->pci_dev; @@ -120,6 +134,10 @@ static void snbep_uncore_msr_enable_event(struct intel_uncore_box *box, struct perf_event *event) { struct hw_perf_event *hwc = &event->hw; + struct hw_perf_event_extra *reg1 = &hwc->extra_reg; + + if (reg1->idx != EXTRA_REG_NONE) + wrmsrl(reg1->reg, reg1->config); wrmsrl(hwc->config_base, hwc->config | SNBEP_PMON_CTL_EN); } @@ -149,6 +167,71 @@ static void snbep_uncore_msr_init_box(struct intel_uncore_box *box) wrmsrl(msr, SNBEP_PMON_BOX_CTL_INT); } +static struct event_constraint * +snbep_uncore_get_constraint(struct intel_uncore_box *box, + struct perf_event *event) +{ + struct intel_uncore_extra_reg *er; + struct hw_perf_event_extra *reg1 = &event->hw.extra_reg; + unsigned long flags; + bool ok = false; + + if (reg1->idx == EXTRA_REG_NONE || (box->phys_id >= 0 && reg1->alloc)) + return NULL; + + er = &box->shared_regs[reg1->idx]; + raw_spin_lock_irqsave(&er->lock, flags); + if (!atomic_read(&er->ref) || er->config1 == reg1->config) { + atomic_inc(&er->ref); + er->config1 = reg1->config; + ok = true; + } + raw_spin_unlock_irqrestore(&er->lock, flags); + + if (ok) { + if (box->phys_id >= 0) + reg1->alloc = 1; + return NULL; + } + return &constraint_empty; +} + +static void snbep_uncore_put_constraint(struct intel_uncore_box *box, + struct perf_event *event) +{ + struct intel_uncore_extra_reg *er; + struct hw_perf_event_extra *reg1 = &event->hw.extra_reg; + + if (box->phys_id < 0 || !reg1->alloc) + return; + + er = &box->shared_regs[reg1->idx]; + atomic_dec(&er->ref); + reg1->alloc = 0; +} + +static int snbep_uncore_hw_config(struct intel_uncore_box *box, + struct perf_event *event) +{ + struct hw_perf_event *hwc = &event->hw; + struct hw_perf_event_extra *reg1 = &hwc->extra_reg; + + if (box->pmu->type == &snbep_uncore_cbox) { + reg1->reg = SNBEP_C0_MSR_PMON_BOX_FILTER + + SNBEP_CBO_MSR_OFFSET * box->pmu->pmu_idx; + reg1->config = event->attr.config1 & + SNBEP_CB0_MSR_PMON_BOX_FILTER_MASK; + } else if (box->pmu->type == &snbep_uncore_pcu) { + reg1->reg = SNBEP_PCU_MSR_PMON_BOX_FILTER; + reg1->config = event->attr.config1 & + SNBEP_PCU_MSR_PMON_BOX_FILTER_MASK; + } else { + return 0; + } + reg1->idx = 0; + return 0; +} + static struct attribute *snbep_uncore_formats_attr[] = { &format_attr_event.attr, &format_attr_umask.attr, @@ -167,6 +250,20 @@ static struct attribute *snbep_uncore_ubox_formats_attr[] = { NULL, }; +static struct attribute *snbep_uncore_cbox_formats_attr[] = { + &format_attr_event.attr, + &format_attr_umask.attr, + &format_attr_edge.attr, + &format_attr_tid_en.attr, + &format_attr_inv.attr, + &format_attr_thresh8.attr, + &format_attr_filter_tid.attr, + &format_attr_filter_nid.attr, + &format_attr_filter_state.attr, + &format_attr_filter_opc.attr, + NULL, +}; + static struct attribute *snbep_uncore_pcu_formats_attr[] = { &format_attr_event.attr, &format_attr_occ_sel.attr, @@ -175,6 +272,10 @@ static struct attribute *snbep_uncore_pcu_formats_attr[] = { &format_attr_thresh5.attr, &format_attr_occ_invert.attr, &format_attr_occ_edge.attr, + &format_attr_filter_brand0.attr, + &format_attr_filter_brand1.attr, + &format_attr_filter_brand2.attr, + &format_attr_filter_brand3.attr, NULL, }; @@ -203,6 +304,11 @@ static struct attribute_group snbep_uncore_ubox_format_group = { .attrs = snbep_uncore_ubox_formats_attr, }; +static struct attribute_group snbep_uncore_cbox_format_group = { + .name = "format", + .attrs = snbep_uncore_cbox_formats_attr, +}; + static struct attribute_group snbep_uncore_pcu_format_group = { .name = "format", .attrs = snbep_uncore_pcu_formats_attr, @@ -215,6 +321,9 @@ static struct intel_uncore_ops snbep_uncore_msr_ops = { .disable_event = snbep_uncore_msr_disable_event, .enable_event = snbep_uncore_msr_enable_event, .read_counter = snbep_uncore_msr_read_counter, + .get_constraint = snbep_uncore_get_constraint, + .put_constraint = snbep_uncore_put_constraint, + .hw_config = snbep_uncore_hw_config, }; static struct intel_uncore_ops snbep_uncore_pci_ops = { @@ -307,31 +416,33 @@ static struct intel_uncore_type snbep_uncore_ubox = { }; static struct intel_uncore_type snbep_uncore_cbox = { - .name = "cbox", - .num_counters = 4, - .num_boxes = 8, - .perf_ctr_bits = 44, - .event_ctl = SNBEP_C0_MSR_PMON_CTL0, - .perf_ctr = SNBEP_C0_MSR_PMON_CTR0, - .event_mask = SNBEP_PMON_RAW_EVENT_MASK, - .box_ctl = SNBEP_C0_MSR_PMON_BOX_CTL, - .msr_offset = SNBEP_CBO_MSR_OFFSET, - .constraints = snbep_uncore_cbox_constraints, - .ops = &snbep_uncore_msr_ops, - .format_group = &snbep_uncore_format_group, + .name = "cbox", + .num_counters = 4, + .num_boxes = 8, + .perf_ctr_bits = 44, + .event_ctl = SNBEP_C0_MSR_PMON_CTL0, + .perf_ctr = SNBEP_C0_MSR_PMON_CTR0, + .event_mask = SNBEP_CBO_MSR_PMON_RAW_EVENT_MASK, + .box_ctl = SNBEP_C0_MSR_PMON_BOX_CTL, + .msr_offset = SNBEP_CBO_MSR_OFFSET, + .num_shared_regs = 1, + .constraints = snbep_uncore_cbox_constraints, + .ops = &snbep_uncore_msr_ops, + .format_group = &snbep_uncore_cbox_format_group, }; static struct intel_uncore_type snbep_uncore_pcu = { - .name = "pcu", - .num_counters = 4, - .num_boxes = 1, - .perf_ctr_bits = 48, - .perf_ctr = SNBEP_PCU_MSR_PMON_CTR0, - .event_ctl = SNBEP_PCU_MSR_PMON_CTL0, - .event_mask = SNBEP_PCU_MSR_PMON_RAW_EVENT_MASK, - .box_ctl = SNBEP_PCU_MSR_PMON_BOX_CTL, - .ops = &snbep_uncore_msr_ops, - .format_group = &snbep_uncore_pcu_format_group, + .name = "pcu", + .num_counters = 4, + .num_boxes = 1, + .perf_ctr_bits = 48, + .perf_ctr = SNBEP_PCU_MSR_PMON_CTR0, + .event_ctl = SNBEP_PCU_MSR_PMON_CTL0, + .event_mask = SNBEP_PCU_MSR_PMON_RAW_EVENT_MASK, + .box_ctl = SNBEP_PCU_MSR_PMON_BOX_CTL, + .num_shared_regs = 1, + .ops = &snbep_uncore_msr_ops, + .format_group = &snbep_uncore_pcu_format_group, }; static struct intel_uncore_type *snbep_msr_uncores[] = { @@ -747,15 +858,22 @@ static void uncore_pmu_init_hrtimer(struct intel_uncore_box *box) box->hrtimer.function = uncore_pmu_hrtimer; } -struct intel_uncore_box *uncore_alloc_box(int cpu) +struct intel_uncore_box *uncore_alloc_box(struct intel_uncore_type *type, + int cpu) { struct intel_uncore_box *box; + int i, size; - box = kmalloc_node(sizeof(*box), GFP_KERNEL | __GFP_ZERO, - cpu_to_node(cpu)); + size = sizeof(*box) + type->num_shared_regs * + sizeof(struct intel_uncore_extra_reg); + + box = kmalloc_node(size, GFP_KERNEL | __GFP_ZERO, cpu_to_node(cpu)); if (!box) return NULL; + for (i = 0; i < type->num_shared_regs; i++) + raw_spin_lock_init(&box->shared_regs[i].lock); + uncore_pmu_init_hrtimer(box); atomic_set(&box->refcnt, 1); box->cpu = -1; @@ -834,11 +952,18 @@ static int uncore_collect_events(struct intel_uncore_box *box, } static struct event_constraint * -uncore_event_constraint(struct intel_uncore_type *type, - struct perf_event *event) +uncore_get_event_constraint(struct intel_uncore_box *box, + struct perf_event *event) { + struct intel_uncore_type *type = box->pmu->type; struct event_constraint *c; + if (type->ops->get_constraint) { + c = type->ops->get_constraint(box, event); + if (c) + return c; + } + if (event->hw.config == ~0ULL) return &constraint_fixed; @@ -852,19 +977,25 @@ uncore_event_constraint(struct intel_uncore_type *type, return &type->unconstrainted; } +static void uncore_put_event_constraint(struct intel_uncore_box *box, + struct perf_event *event) +{ + if (box->pmu->type->ops->put_constraint) + box->pmu->type->ops->put_constraint(box, event); +} + static int uncore_assign_events(struct intel_uncore_box *box, int assign[], int n) { unsigned long used_mask[BITS_TO_LONGS(UNCORE_PMC_IDX_MAX)]; struct event_constraint *c, *constraints[UNCORE_PMC_IDX_MAX]; - int i, ret, wmin, wmax; + int i, wmin, wmax, ret = 0; struct hw_perf_event *hwc; bitmap_zero(used_mask, UNCORE_PMC_IDX_MAX); for (i = 0, wmin = UNCORE_PMC_IDX_MAX, wmax = 0; i < n; i++) { - c = uncore_event_constraint(box->pmu->type, - box->event_list[i]); + c = uncore_get_event_constraint(box, box->event_list[i]); constraints[i] = c; wmin = min(wmin, c->weight); wmax = max(wmax, c->weight); @@ -888,13 +1019,17 @@ static int uncore_assign_events(struct intel_uncore_box *box, break; __set_bit(hwc->idx, used_mask); - assign[i] = hwc->idx; + if (assign) + assign[i] = hwc->idx; } - if (i == n) - return 0; - /* slow path */ - ret = perf_assign_events(constraints, n, wmin, wmax, assign); + if (i != n) + ret = perf_assign_events(constraints, n, wmin, wmax, assign); + + if (!assign || ret) { + for (i = 0; i < n; i++) + uncore_put_event_constraint(box, box->event_list[i]); + } return ret ? -EINVAL : 0; } @@ -1021,6 +1156,8 @@ static void uncore_pmu_event_del(struct perf_event *event, int flags) for (i = 0; i < box->n_events; i++) { if (event == box->event_list[i]) { + uncore_put_event_constraint(box, event); + while (++i < box->n_events) box->event_list[i - 1] = box->event_list[i]; @@ -1048,10 +1185,9 @@ static int uncore_validate_group(struct intel_uncore_pmu *pmu, { struct perf_event *leader = event->group_leader; struct intel_uncore_box *fake_box; - int assign[UNCORE_PMC_IDX_MAX]; int ret = -EINVAL, n; - fake_box = uncore_alloc_box(smp_processor_id()); + fake_box = uncore_alloc_box(pmu->type, smp_processor_id()); if (!fake_box) return -ENOMEM; @@ -1073,7 +1209,7 @@ static int uncore_validate_group(struct intel_uncore_pmu *pmu, fake_box->n_events = n; - ret = uncore_assign_events(fake_box, assign, n); + ret = uncore_assign_events(fake_box, NULL, n); out: kfree(fake_box); return ret; @@ -1117,6 +1253,10 @@ int uncore_pmu_event_init(struct perf_event *event) return -EINVAL; event->cpu = box->cpu; + event->hw.idx = -1; + event->hw.last_tag = ~0ULL; + event->hw.extra_reg.idx = EXTRA_REG_NONE; + if (event->attr.config == UNCORE_FIXED_EVENT) { /* no fixed counter */ if (!pmu->type->fixed_ctl) @@ -1130,11 +1270,13 @@ int uncore_pmu_event_init(struct perf_event *event) hwc->config = ~0ULL; } else { hwc->config = event->attr.config & pmu->type->event_mask; + if (pmu->type->ops->hw_config) { + ret = pmu->type->ops->hw_config(box, event); + if (ret) + return ret; + } } - event->hw.idx = -1; - event->hw.last_tag = ~0ULL; - if (event->group_leader != event) ret = uncore_validate_group(pmu, event); else @@ -1276,7 +1418,7 @@ static int __devinit uncore_pci_add(struct intel_uncore_type *type, if (phys_id < 0) return -ENODEV; - box = uncore_alloc_box(0); + box = uncore_alloc_box(type, 0); if (!box) return -ENOMEM; @@ -1458,7 +1600,7 @@ static int __cpuinit uncore_cpu_prepare(int cpu, int phys_id) if (pmu->func_id < 0) pmu->func_id = j; - box = uncore_alloc_box(cpu); + box = uncore_alloc_box(type, cpu); if (!box) return -ENOMEM; diff --git a/arch/x86/kernel/cpu/perf_event_intel_uncore.h b/arch/x86/kernel/cpu/perf_event_intel_uncore.h index 88498c7..b13e9ea 100644 --- a/arch/x86/kernel/cpu/perf_event_intel_uncore.h +++ b/arch/x86/kernel/cpu/perf_event_intel_uncore.h @@ -97,6 +97,10 @@ SNBEP_PMON_CTL_INVERT | \ SNBEP_U_MSR_PMON_CTL_TRESH_MASK) +#define SNBEP_CBO_PMON_CTL_TID_EN (1 << 19) +#define SNBEP_CBO_MSR_PMON_RAW_EVENT_MASK (SNBEP_PMON_RAW_EVENT_MASK | \ + SNBEP_CBO_PMON_CTL_TID_EN) + /* SNB-EP PCU event control */ #define SNBEP_PCU_MSR_PMON_CTL_OCC_SEL_MASK 0x0000c000 #define SNBEP_PCU_MSR_PMON_CTL_TRESH_MASK 0x1f000000 @@ -140,15 +144,17 @@ /* SNB-EP Cbo register */ #define SNBEP_C0_MSR_PMON_CTR0 0xd16 #define SNBEP_C0_MSR_PMON_CTL0 0xd10 -#define SNBEP_C0_MSR_PMON_BOX_FILTER 0xd14 #define SNBEP_C0_MSR_PMON_BOX_CTL 0xd04 +#define SNBEP_C0_MSR_PMON_BOX_FILTER 0xd14 +#define SNBEP_CB0_MSR_PMON_BOX_FILTER_MASK 0xfffffc1f #define SNBEP_CBO_MSR_OFFSET 0x20 /* SNB-EP PCU register */ #define SNBEP_PCU_MSR_PMON_CTR0 0xc36 #define SNBEP_PCU_MSR_PMON_CTL0 0xc30 -#define SNBEP_PCU_MSR_PMON_BOX_FILTER 0xc34 #define SNBEP_PCU_MSR_PMON_BOX_CTL 0xc24 +#define SNBEP_PCU_MSR_PMON_BOX_FILTER 0xc34 +#define SNBEP_PCU_MSR_PMON_BOX_FILTER_MASK 0xffffffff #define SNBEP_PCU_MSR_CORE_C3_CTR 0x3fc #define SNBEP_PCU_MSR_CORE_C6_CTR 0x3fd @@ -163,7 +169,6 @@ struct intel_uncore_type { int num_boxes; int perf_ctr_bits; int fixed_ctr_bits; - int single_fixed; unsigned perf_ctr; unsigned event_ctl; unsigned event_mask; @@ -171,6 +176,8 @@ struct intel_uncore_type { unsigned fixed_ctl; unsigned box_ctl; unsigned msr_offset; + unsigned num_shared_regs:8; + unsigned single_fixed:1; struct event_constraint unconstrainted; struct event_constraint *constraints; struct intel_uncore_pmu *pmus; @@ -188,6 +195,10 @@ struct intel_uncore_ops { void (*disable_event)(struct intel_uncore_box *, struct perf_event *); void (*enable_event)(struct intel_uncore_box *, struct perf_event *); u64 (*read_counter)(struct intel_uncore_box *, struct perf_event *); + int (*hw_config)(struct intel_uncore_box *, struct perf_event *); + struct event_constraint *(*get_constraint)(struct intel_uncore_box *, + struct perf_event *); + void (*put_constraint)(struct intel_uncore_box *, struct perf_event *); }; struct intel_uncore_pmu { @@ -200,6 +211,12 @@ struct intel_uncore_pmu { struct list_head box_list; }; +struct intel_uncore_extra_reg { + raw_spinlock_t lock; + u64 config1; + atomic_t ref; +}; + struct intel_uncore_box { int phys_id; int n_active; /* number of active events */ @@ -215,6 +232,7 @@ struct intel_uncore_box { struct intel_uncore_pmu *pmu; struct hrtimer hrtimer; struct list_head list; + struct intel_uncore_extra_reg shared_regs[0]; }; #define UNCORE_BOX_FLAG_INITIATED 0 ^ permalink raw reply related [flat|nested] 21+ messages in thread
* Re: [PATCH 4/5] perf/x86: Uncore filter support for SandyBridge-EP 2012-07-04 6:00 ` [PATCH 4/5] perf/x86: Uncore filter support for SandyBridge-EP Yan, Zheng 2012-07-06 6:33 ` [tip:perf/core] " tip-bot for Yan, Zheng @ 2012-07-22 19:25 ` Stephane Eranian 1 sibling, 0 replies; 21+ messages in thread From: Stephane Eranian @ 2012-07-22 19:25 UTC (permalink / raw) To: Yan, Zheng; +Cc: a.p.zijlstra, mingo, andi, linux-kernel On Wed, Jul 4, 2012 at 8:00 AM, Yan, Zheng <zheng.z.yan@intel.com> wrote: > > From: "Yan, Zheng" <zheng.z.yan@intel.com> > > This patch adds C-Box and PCU filter support for SandyBridge-EP > uncore. We can filter C-Box events by thread/core ID and filter > PCU events by frequency/voltage. > > Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com> > --- > arch/x86/kernel/cpu/perf_event_intel_uncore.c | 228 ++++++++++++++++++++----- > arch/x86/kernel/cpu/perf_event_intel_uncore.h | 24 ++- > 2 files changed, 206 insertions(+), 46 deletions(-) > > diff --git a/arch/x86/kernel/cpu/perf_event_intel_uncore.c b/arch/x86/kernel/cpu/perf_event_intel_uncore.c > index a394add..c0a80ff 100644 > --- a/arch/x86/kernel/cpu/perf_event_intel_uncore.c > +++ b/arch/x86/kernel/cpu/perf_event_intel_uncore.c > @@ -14,10 +14,13 @@ static cpumask_t uncore_cpu_mask; > /* constraint for the fixed counter */ > static struct event_constraint constraint_fixed = > EVENT_CONSTRAINT(~0ULL, 1 << UNCORE_PMC_IDX_FIXED, ~0ULL); > +static struct event_constraint constraint_empty = > + EVENT_CONSTRAINT(0, 0, 0); > > DEFINE_UNCORE_FORMAT_ATTR(event, event, "config:0-7"); > DEFINE_UNCORE_FORMAT_ATTR(umask, umask, "config:8-15"); > DEFINE_UNCORE_FORMAT_ATTR(edge, edge, "config:18"); > +DEFINE_UNCORE_FORMAT_ATTR(tid_en, tid_en, "config:19"); > DEFINE_UNCORE_FORMAT_ATTR(inv, inv, "config:23"); > DEFINE_UNCORE_FORMAT_ATTR(cmask5, cmask, "config:24-28"); > DEFINE_UNCORE_FORMAT_ATTR(cmask8, cmask, "config:24-31"); > @@ -26,8 +29,19 @@ DEFINE_UNCORE_FORMAT_ATTR(thresh5, thresh, "config:24-28"); > DEFINE_UNCORE_FORMAT_ATTR(occ_sel, occ_sel, "config:14-15"); > DEFINE_UNCORE_FORMAT_ATTR(occ_invert, occ_invert, "config:30"); > DEFINE_UNCORE_FORMAT_ATTR(occ_edge, occ_edge, "config:14-51"); > +DEFINE_UNCORE_FORMAT_ATTR(filter_tid, filter_tid, "config1:0-4"); > +DEFINE_UNCORE_FORMAT_ATTR(filter_nid, filter_nid, "config1:10-17"); > +DEFINE_UNCORE_FORMAT_ATTR(filter_state, filter_state, "config1:18-22"); > +DEFINE_UNCORE_FORMAT_ATTR(filter_opc, filter_opc, "config1:23-31"); > +DEFINE_UNCORE_FORMAT_ATTR(filter_brand0, filter_brand0, "config1:0-7"); > +DEFINE_UNCORE_FORMAT_ATTR(filter_brand1, filter_brand1, "config1:8-15"); > +DEFINE_UNCORE_FORMAT_ATTR(filter_brand2, filter_brand2, "config1:16-23"); > +DEFINE_UNCORE_FORMAT_ATTR(filter_brand3, filter_brand3, "config1:24-31"); > That's not filter_brand*, that's filter_band*. > /* Sandy Bridge-EP uncore support */ > +static struct intel_uncore_type snbep_uncore_cbox; > +static struct intel_uncore_type snbep_uncore_pcu; > + > static void snbep_uncore_pci_disable_box(struct intel_uncore_box *box) > { > struct pci_dev *pdev = box->pci_dev; > @@ -120,6 +134,10 @@ static void snbep_uncore_msr_enable_event(struct intel_uncore_box *box, > struct perf_event *event) > { > struct hw_perf_event *hwc = &event->hw; > + struct hw_perf_event_extra *reg1 = &hwc->extra_reg; > + > + if (reg1->idx != EXTRA_REG_NONE) > + wrmsrl(reg1->reg, reg1->config); > > wrmsrl(hwc->config_base, hwc->config | SNBEP_PMON_CTL_EN); > } > @@ -149,6 +167,71 @@ static void snbep_uncore_msr_init_box(struct intel_uncore_box *box) > wrmsrl(msr, SNBEP_PMON_BOX_CTL_INT); > } > > +static struct event_constraint * > +snbep_uncore_get_constraint(struct intel_uncore_box *box, > + struct perf_event *event) > +{ > + struct intel_uncore_extra_reg *er; > + struct hw_perf_event_extra *reg1 = &event->hw.extra_reg; > + unsigned long flags; > + bool ok = false; > + > + if (reg1->idx == EXTRA_REG_NONE || (box->phys_id >= 0 && reg1->alloc)) > + return NULL; > + > + er = &box->shared_regs[reg1->idx]; > + raw_spin_lock_irqsave(&er->lock, flags); > + if (!atomic_read(&er->ref) || er->config1 == reg1->config) { > + atomic_inc(&er->ref); > + er->config1 = reg1->config; > + ok = true; > + } > + raw_spin_unlock_irqrestore(&er->lock, flags); > + > + if (ok) { > + if (box->phys_id >= 0) > + reg1->alloc = 1; > + return NULL; > + } > + return &constraint_empty; > +} > + > +static void snbep_uncore_put_constraint(struct intel_uncore_box *box, > + struct perf_event *event) > +{ > + struct intel_uncore_extra_reg *er; > + struct hw_perf_event_extra *reg1 = &event->hw.extra_reg; > + > + if (box->phys_id < 0 || !reg1->alloc) > + return; > + > + er = &box->shared_regs[reg1->idx]; > + atomic_dec(&er->ref); > + reg1->alloc = 0; > +} > + > +static int snbep_uncore_hw_config(struct intel_uncore_box *box, > + struct perf_event *event) > +{ > + struct hw_perf_event *hwc = &event->hw; > + struct hw_perf_event_extra *reg1 = &hwc->extra_reg; > + > + if (box->pmu->type == &snbep_uncore_cbox) { > + reg1->reg = SNBEP_C0_MSR_PMON_BOX_FILTER + > + SNBEP_CBO_MSR_OFFSET * box->pmu->pmu_idx; > + reg1->config = event->attr.config1 & > + SNBEP_CB0_MSR_PMON_BOX_FILTER_MASK; > + } else if (box->pmu->type == &snbep_uncore_pcu) { > + reg1->reg = SNBEP_PCU_MSR_PMON_BOX_FILTER; > + reg1->config = event->attr.config1 & > + SNBEP_PCU_MSR_PMON_BOX_FILTER_MASK; > + } else { > + return 0; > + } > + reg1->idx = 0; > + return 0; > +} > + > static struct attribute *snbep_uncore_formats_attr[] = { > &format_attr_event.attr, > &format_attr_umask.attr, > @@ -167,6 +250,20 @@ static struct attribute *snbep_uncore_ubox_formats_attr[] = { > NULL, > }; > > +static struct attribute *snbep_uncore_cbox_formats_attr[] = { > + &format_attr_event.attr, > + &format_attr_umask.attr, > + &format_attr_edge.attr, > + &format_attr_tid_en.attr, > + &format_attr_inv.attr, > + &format_attr_thresh8.attr, > + &format_attr_filter_tid.attr, > + &format_attr_filter_nid.attr, > + &format_attr_filter_state.attr, > + &format_attr_filter_opc.attr, > + NULL, > +}; > + > static struct attribute *snbep_uncore_pcu_formats_attr[] = { > &format_attr_event.attr, > &format_attr_occ_sel.attr, > @@ -175,6 +272,10 @@ static struct attribute *snbep_uncore_pcu_formats_attr[] = { > &format_attr_thresh5.attr, > &format_attr_occ_invert.attr, > &format_attr_occ_edge.attr, > + &format_attr_filter_brand0.attr, > + &format_attr_filter_brand1.attr, > + &format_attr_filter_brand2.attr, > + &format_attr_filter_brand3.attr, > NULL, > }; > > @@ -203,6 +304,11 @@ static struct attribute_group snbep_uncore_ubox_format_group = { > .attrs = snbep_uncore_ubox_formats_attr, > }; > > +static struct attribute_group snbep_uncore_cbox_format_group = { > + .name = "format", > + .attrs = snbep_uncore_cbox_formats_attr, > +}; > + > static struct attribute_group snbep_uncore_pcu_format_group = { > .name = "format", > .attrs = snbep_uncore_pcu_formats_attr, > @@ -215,6 +321,9 @@ static struct intel_uncore_ops snbep_uncore_msr_ops = { > .disable_event = snbep_uncore_msr_disable_event, > .enable_event = snbep_uncore_msr_enable_event, > .read_counter = snbep_uncore_msr_read_counter, > + .get_constraint = snbep_uncore_get_constraint, > + .put_constraint = snbep_uncore_put_constraint, > + .hw_config = snbep_uncore_hw_config, > }; > > static struct intel_uncore_ops snbep_uncore_pci_ops = { > @@ -307,31 +416,33 @@ static struct intel_uncore_type snbep_uncore_ubox = { > }; > > static struct intel_uncore_type snbep_uncore_cbox = { > - .name = "cbox", > - .num_counters = 4, > - .num_boxes = 8, > - .perf_ctr_bits = 44, > - .event_ctl = SNBEP_C0_MSR_PMON_CTL0, > - .perf_ctr = SNBEP_C0_MSR_PMON_CTR0, > - .event_mask = SNBEP_PMON_RAW_EVENT_MASK, > - .box_ctl = SNBEP_C0_MSR_PMON_BOX_CTL, > - .msr_offset = SNBEP_CBO_MSR_OFFSET, > - .constraints = snbep_uncore_cbox_constraints, > - .ops = &snbep_uncore_msr_ops, > - .format_group = &snbep_uncore_format_group, > + .name = "cbox", > + .num_counters = 4, > + .num_boxes = 8, > + .perf_ctr_bits = 44, > + .event_ctl = SNBEP_C0_MSR_PMON_CTL0, > + .perf_ctr = SNBEP_C0_MSR_PMON_CTR0, > + .event_mask = SNBEP_CBO_MSR_PMON_RAW_EVENT_MASK, > + .box_ctl = SNBEP_C0_MSR_PMON_BOX_CTL, > + .msr_offset = SNBEP_CBO_MSR_OFFSET, > + .num_shared_regs = 1, > + .constraints = snbep_uncore_cbox_constraints, > + .ops = &snbep_uncore_msr_ops, > + .format_group = &snbep_uncore_cbox_format_group, > }; > > static struct intel_uncore_type snbep_uncore_pcu = { > - .name = "pcu", > - .num_counters = 4, > - .num_boxes = 1, > - .perf_ctr_bits = 48, > - .perf_ctr = SNBEP_PCU_MSR_PMON_CTR0, > - .event_ctl = SNBEP_PCU_MSR_PMON_CTL0, > - .event_mask = SNBEP_PCU_MSR_PMON_RAW_EVENT_MASK, > - .box_ctl = SNBEP_PCU_MSR_PMON_BOX_CTL, > - .ops = &snbep_uncore_msr_ops, > - .format_group = &snbep_uncore_pcu_format_group, > + .name = "pcu", > + .num_counters = 4, > + .num_boxes = 1, > + .perf_ctr_bits = 48, > + .perf_ctr = SNBEP_PCU_MSR_PMON_CTR0, > + .event_ctl = SNBEP_PCU_MSR_PMON_CTL0, > + .event_mask = SNBEP_PCU_MSR_PMON_RAW_EVENT_MASK, > + .box_ctl = SNBEP_PCU_MSR_PMON_BOX_CTL, > + .num_shared_regs = 1, > + .ops = &snbep_uncore_msr_ops, > + .format_group = &snbep_uncore_pcu_format_group, > }; > > static struct intel_uncore_type *snbep_msr_uncores[] = { > @@ -747,15 +858,22 @@ static void uncore_pmu_init_hrtimer(struct intel_uncore_box *box) > box->hrtimer.function = uncore_pmu_hrtimer; > } > > -struct intel_uncore_box *uncore_alloc_box(int cpu) > +struct intel_uncore_box *uncore_alloc_box(struct intel_uncore_type *type, > + int cpu) > { > struct intel_uncore_box *box; > + int i, size; > > - box = kmalloc_node(sizeof(*box), GFP_KERNEL | __GFP_ZERO, > - cpu_to_node(cpu)); > + size = sizeof(*box) + type->num_shared_regs * > + sizeof(struct intel_uncore_extra_reg); > + > + box = kmalloc_node(size, GFP_KERNEL | __GFP_ZERO, cpu_to_node(cpu)); > if (!box) > return NULL; > > + for (i = 0; i < type->num_shared_regs; i++) > + raw_spin_lock_init(&box->shared_regs[i].lock); > + > uncore_pmu_init_hrtimer(box); > atomic_set(&box->refcnt, 1); > box->cpu = -1; > @@ -834,11 +952,18 @@ static int uncore_collect_events(struct intel_uncore_box *box, > } > > static struct event_constraint * > -uncore_event_constraint(struct intel_uncore_type *type, > - struct perf_event *event) > +uncore_get_event_constraint(struct intel_uncore_box *box, > + struct perf_event *event) > { > + struct intel_uncore_type *type = box->pmu->type; > struct event_constraint *c; > > + if (type->ops->get_constraint) { > + c = type->ops->get_constraint(box, event); > + if (c) > + return c; > + } > + > if (event->hw.config == ~0ULL) > return &constraint_fixed; > > @@ -852,19 +977,25 @@ uncore_event_constraint(struct intel_uncore_type *type, > return &type->unconstrainted; > } > > +static void uncore_put_event_constraint(struct intel_uncore_box *box, > + struct perf_event *event) > +{ > + if (box->pmu->type->ops->put_constraint) > + box->pmu->type->ops->put_constraint(box, event); > +} > + > static int uncore_assign_events(struct intel_uncore_box *box, > int assign[], int n) > { > unsigned long used_mask[BITS_TO_LONGS(UNCORE_PMC_IDX_MAX)]; > struct event_constraint *c, *constraints[UNCORE_PMC_IDX_MAX]; > - int i, ret, wmin, wmax; > + int i, wmin, wmax, ret = 0; > struct hw_perf_event *hwc; > > bitmap_zero(used_mask, UNCORE_PMC_IDX_MAX); > > for (i = 0, wmin = UNCORE_PMC_IDX_MAX, wmax = 0; i < n; i++) { > - c = uncore_event_constraint(box->pmu->type, > - box->event_list[i]); > + c = uncore_get_event_constraint(box, box->event_list[i]); > constraints[i] = c; > wmin = min(wmin, c->weight); > wmax = max(wmax, c->weight); > @@ -888,13 +1019,17 @@ static int uncore_assign_events(struct intel_uncore_box *box, > break; > > __set_bit(hwc->idx, used_mask); > - assign[i] = hwc->idx; > + if (assign) > + assign[i] = hwc->idx; > } > - if (i == n) > - return 0; > - > /* slow path */ > - ret = perf_assign_events(constraints, n, wmin, wmax, assign); > + if (i != n) > + ret = perf_assign_events(constraints, n, wmin, wmax, assign); > + > + if (!assign || ret) { > + for (i = 0; i < n; i++) > + uncore_put_event_constraint(box, box->event_list[i]); > + } > return ret ? -EINVAL : 0; > } > > @@ -1021,6 +1156,8 @@ static void uncore_pmu_event_del(struct perf_event *event, int flags) > > for (i = 0; i < box->n_events; i++) { > if (event == box->event_list[i]) { > + uncore_put_event_constraint(box, event); > + > while (++i < box->n_events) > box->event_list[i - 1] = box->event_list[i]; > > @@ -1048,10 +1185,9 @@ static int uncore_validate_group(struct intel_uncore_pmu *pmu, > { > struct perf_event *leader = event->group_leader; > struct intel_uncore_box *fake_box; > - int assign[UNCORE_PMC_IDX_MAX]; > int ret = -EINVAL, n; > > - fake_box = uncore_alloc_box(smp_processor_id()); > + fake_box = uncore_alloc_box(pmu->type, smp_processor_id()); > if (!fake_box) > return -ENOMEM; > > @@ -1073,7 +1209,7 @@ static int uncore_validate_group(struct intel_uncore_pmu *pmu, > > fake_box->n_events = n; > > - ret = uncore_assign_events(fake_box, assign, n); > + ret = uncore_assign_events(fake_box, NULL, n); > out: > kfree(fake_box); > return ret; > @@ -1117,6 +1253,10 @@ int uncore_pmu_event_init(struct perf_event *event) > return -EINVAL; > event->cpu = box->cpu; > > + event->hw.idx = -1; > + event->hw.last_tag = ~0ULL; > + event->hw.extra_reg.idx = EXTRA_REG_NONE; > + > if (event->attr.config == UNCORE_FIXED_EVENT) { > /* no fixed counter */ > if (!pmu->type->fixed_ctl) > @@ -1130,11 +1270,13 @@ int uncore_pmu_event_init(struct perf_event *event) > hwc->config = ~0ULL; > } else { > hwc->config = event->attr.config & pmu->type->event_mask; > + if (pmu->type->ops->hw_config) { > + ret = pmu->type->ops->hw_config(box, event); > + if (ret) > + return ret; > + } > } > > - event->hw.idx = -1; > - event->hw.last_tag = ~0ULL; > - > if (event->group_leader != event) > ret = uncore_validate_group(pmu, event); > else > @@ -1276,7 +1418,7 @@ static int __devinit uncore_pci_add(struct intel_uncore_type *type, > if (phys_id < 0) > return -ENODEV; > > - box = uncore_alloc_box(0); > + box = uncore_alloc_box(type, 0); > if (!box) > return -ENOMEM; > > @@ -1458,7 +1600,7 @@ static int __cpuinit uncore_cpu_prepare(int cpu, int phys_id) > if (pmu->func_id < 0) > pmu->func_id = j; > > - box = uncore_alloc_box(cpu); > + box = uncore_alloc_box(type, cpu); > if (!box) > return -ENOMEM; > > diff --git a/arch/x86/kernel/cpu/perf_event_intel_uncore.h b/arch/x86/kernel/cpu/perf_event_intel_uncore.h > index 88498c7..b13e9ea 100644 > --- a/arch/x86/kernel/cpu/perf_event_intel_uncore.h > +++ b/arch/x86/kernel/cpu/perf_event_intel_uncore.h > @@ -97,6 +97,10 @@ > SNBEP_PMON_CTL_INVERT | \ > SNBEP_U_MSR_PMON_CTL_TRESH_MASK) > > +#define SNBEP_CBO_PMON_CTL_TID_EN (1 << 19) > +#define SNBEP_CBO_MSR_PMON_RAW_EVENT_MASK (SNBEP_PMON_RAW_EVENT_MASK | \ > + SNBEP_CBO_PMON_CTL_TID_EN) > + > /* SNB-EP PCU event control */ > #define SNBEP_PCU_MSR_PMON_CTL_OCC_SEL_MASK 0x0000c000 > #define SNBEP_PCU_MSR_PMON_CTL_TRESH_MASK 0x1f000000 > @@ -140,15 +144,17 @@ > /* SNB-EP Cbo register */ > #define SNBEP_C0_MSR_PMON_CTR0 0xd16 > #define SNBEP_C0_MSR_PMON_CTL0 0xd10 > -#define SNBEP_C0_MSR_PMON_BOX_FILTER 0xd14 > #define SNBEP_C0_MSR_PMON_BOX_CTL 0xd04 > +#define SNBEP_C0_MSR_PMON_BOX_FILTER 0xd14 > +#define SNBEP_CB0_MSR_PMON_BOX_FILTER_MASK 0xfffffc1f > #define SNBEP_CBO_MSR_OFFSET 0x20 > > /* SNB-EP PCU register */ > #define SNBEP_PCU_MSR_PMON_CTR0 0xc36 > #define SNBEP_PCU_MSR_PMON_CTL0 0xc30 > -#define SNBEP_PCU_MSR_PMON_BOX_FILTER 0xc34 > #define SNBEP_PCU_MSR_PMON_BOX_CTL 0xc24 > +#define SNBEP_PCU_MSR_PMON_BOX_FILTER 0xc34 > +#define SNBEP_PCU_MSR_PMON_BOX_FILTER_MASK 0xffffffff > #define SNBEP_PCU_MSR_CORE_C3_CTR 0x3fc > #define SNBEP_PCU_MSR_CORE_C6_CTR 0x3fd > > @@ -163,7 +169,6 @@ struct intel_uncore_type { > int num_boxes; > int perf_ctr_bits; > int fixed_ctr_bits; > - int single_fixed; > unsigned perf_ctr; > unsigned event_ctl; > unsigned event_mask; > @@ -171,6 +176,8 @@ struct intel_uncore_type { > unsigned fixed_ctl; > unsigned box_ctl; > unsigned msr_offset; > + unsigned num_shared_regs:8; > + unsigned single_fixed:1; > struct event_constraint unconstrainted; > struct event_constraint *constraints; > struct intel_uncore_pmu *pmus; > @@ -188,6 +195,10 @@ struct intel_uncore_ops { > void (*disable_event)(struct intel_uncore_box *, struct perf_event *); > void (*enable_event)(struct intel_uncore_box *, struct perf_event *); > u64 (*read_counter)(struct intel_uncore_box *, struct perf_event *); > + int (*hw_config)(struct intel_uncore_box *, struct perf_event *); > + struct event_constraint *(*get_constraint)(struct intel_uncore_box *, > + struct perf_event *); > + void (*put_constraint)(struct intel_uncore_box *, struct perf_event *); > }; > > struct intel_uncore_pmu { > @@ -200,6 +211,12 @@ struct intel_uncore_pmu { > struct list_head box_list; > }; > > +struct intel_uncore_extra_reg { > + raw_spinlock_t lock; > + u64 config1; > + atomic_t ref; > +}; > + > struct intel_uncore_box { > int phys_id; > int n_active; /* number of active events */ > @@ -215,6 +232,7 @@ struct intel_uncore_box { > struct intel_uncore_pmu *pmu; > struct hrtimer hrtimer; > struct list_head list; > + struct intel_uncore_extra_reg shared_regs[0]; > }; > > #define UNCORE_BOX_FLAG_INITIATED 0 > -- > 1.7.10.4 > ^ permalink raw reply [flat|nested] 21+ messages in thread
* [PATCH 5/5] perf/x86: Add Intel Nehalem-EX uncore support 2012-07-04 6:00 [PATCH 0/5] perf/x86: Uncore updates Yan, Zheng ` (3 preceding siblings ...) 2012-07-04 6:00 ` [PATCH 4/5] perf/x86: Uncore filter support for SandyBridge-EP Yan, Zheng @ 2012-07-04 6:00 ` Yan, Zheng 2012-07-04 10:04 ` Peter Zijlstra 4 siblings, 1 reply; 21+ messages in thread From: Yan, Zheng @ 2012-07-04 6:00 UTC (permalink / raw) To: a.p.zijlstra, eranian, mingo, andi, linux-kernel; +Cc: Yan, Zheng From: "Yan, Zheng" <zheng.z.yan@intel.com> The uncore subsystem in Nehalem-EX consists of 7 components (U-Box, C-Box, B-Box, S-Box, R-Box, M-Box and W-Box). This patch is large because the way to program these boxes is diverse. Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com> --- arch/x86/kernel/cpu/perf_event_intel_uncore.c | 1178 +++++++++++++++++++++++-- arch/x86/kernel/cpu/perf_event_intel_uncore.h | 195 +++- 2 files changed, 1301 insertions(+), 72 deletions(-) diff --git a/arch/x86/kernel/cpu/perf_event_intel_uncore.c b/arch/x86/kernel/cpu/perf_event_intel_uncore.c index c0a80ff..ff4f40f 100644 --- a/arch/x86/kernel/cpu/perf_event_intel_uncore.c +++ b/arch/x86/kernel/cpu/perf_event_intel_uncore.c @@ -38,6 +38,63 @@ DEFINE_UNCORE_FORMAT_ATTR(filter_brand1, filter_brand1, "config1:8-15"); DEFINE_UNCORE_FORMAT_ATTR(filter_brand2, filter_brand2, "config1:16-23"); DEFINE_UNCORE_FORMAT_ATTR(filter_brand3, filter_brand3, "config1:24-31"); +static u64 uncore_msr_read_counter(struct intel_uncore_box *box, + struct perf_event *event) +{ + u64 count; + rdmsrl(event->hw.event_base, count); + return count; +} + +/* + * generic get constraint function for shared match/mask registers. + */ +static struct event_constraint * +uncore_get_constraint(struct intel_uncore_box *box, struct perf_event *event) +{ + struct intel_uncore_extra_reg *er; + struct hw_perf_event_extra *reg1 = &event->hw.extra_reg; + struct hw_perf_event_extra *reg2 = &event->hw.branch_reg; + unsigned long flags; + bool ok = false; + + if (reg1->idx == EXTRA_REG_NONE || (box->phys_id >= 0 && reg1->alloc)) + return NULL; + + er = &box->shared_regs[reg1->idx]; + raw_spin_lock_irqsave(&er->lock, flags); + if (!atomic_read(&er->ref) || + (er->config1 == reg1->config && er->config2 == reg2->config)) { + atomic_inc(&er->ref); + er->config1 = reg1->config; + er->config2 = reg2->config; + ok = true; + } + raw_spin_unlock_irqrestore(&er->lock, flags); + + if (ok) { + if (box->phys_id >= 0) + reg1->alloc = 1; + return NULL; + } + + return &constraint_empty; +} + +static void uncore_put_constraint(struct intel_uncore_box *box, + struct perf_event *event) +{ + struct intel_uncore_extra_reg *er; + struct hw_perf_event_extra *reg1 = &event->hw.extra_reg; + + if (box->phys_id < 0 || !reg1->alloc) + return; + + er = &box->shared_regs[reg1->idx]; + atomic_dec(&er->ref); + reg1->alloc = 0; +} + /* Sandy Bridge-EP uncore support */ static struct intel_uncore_type snbep_uncore_cbox; static struct intel_uncore_type snbep_uncore_pcu; @@ -150,16 +207,6 @@ static void snbep_uncore_msr_disable_event(struct intel_uncore_box *box, wrmsrl(hwc->config_base, hwc->config); } -static u64 snbep_uncore_msr_read_counter(struct intel_uncore_box *box, - struct perf_event *event) -{ - struct hw_perf_event *hwc = &event->hw; - u64 count; - - rdmsrl(hwc->event_base, count); - return count; -} - static void snbep_uncore_msr_init_box(struct intel_uncore_box *box) { unsigned msr = uncore_msr_box_ctl(box); @@ -167,49 +214,6 @@ static void snbep_uncore_msr_init_box(struct intel_uncore_box *box) wrmsrl(msr, SNBEP_PMON_BOX_CTL_INT); } -static struct event_constraint * -snbep_uncore_get_constraint(struct intel_uncore_box *box, - struct perf_event *event) -{ - struct intel_uncore_extra_reg *er; - struct hw_perf_event_extra *reg1 = &event->hw.extra_reg; - unsigned long flags; - bool ok = false; - - if (reg1->idx == EXTRA_REG_NONE || (box->phys_id >= 0 && reg1->alloc)) - return NULL; - - er = &box->shared_regs[reg1->idx]; - raw_spin_lock_irqsave(&er->lock, flags); - if (!atomic_read(&er->ref) || er->config1 == reg1->config) { - atomic_inc(&er->ref); - er->config1 = reg1->config; - ok = true; - } - raw_spin_unlock_irqrestore(&er->lock, flags); - - if (ok) { - if (box->phys_id >= 0) - reg1->alloc = 1; - return NULL; - } - return &constraint_empty; -} - -static void snbep_uncore_put_constraint(struct intel_uncore_box *box, - struct perf_event *event) -{ - struct intel_uncore_extra_reg *er; - struct hw_perf_event_extra *reg1 = &event->hw.extra_reg; - - if (box->phys_id < 0 || !reg1->alloc) - return; - - er = &box->shared_regs[reg1->idx]; - atomic_dec(&er->ref); - reg1->alloc = 0; -} - static int snbep_uncore_hw_config(struct intel_uncore_box *box, struct perf_event *event) { @@ -320,9 +324,9 @@ static struct intel_uncore_ops snbep_uncore_msr_ops = { .enable_box = snbep_uncore_msr_enable_box, .disable_event = snbep_uncore_msr_disable_event, .enable_event = snbep_uncore_msr_enable_event, - .read_counter = snbep_uncore_msr_read_counter, - .get_constraint = snbep_uncore_get_constraint, - .put_constraint = snbep_uncore_put_constraint, + .read_counter = uncore_msr_read_counter, + .get_constraint = uncore_get_constraint, + .put_constraint = uncore_put_constraint, .hw_config = snbep_uncore_hw_config, }; @@ -603,7 +607,6 @@ static void snbep_pci2phy_map_init(void) } /* end of Sandy Bridge-EP uncore support */ - /* Sandy Bridge uncore support */ static void snb_uncore_msr_enable_event(struct intel_uncore_box *box, struct perf_event *event) @@ -622,14 +625,6 @@ static void snb_uncore_msr_disable_event(struct intel_uncore_box *box, wrmsrl(event->hw.config_base, 0); } -static u64 snb_uncore_msr_read_counter(struct intel_uncore_box *box, - struct perf_event *event) -{ - u64 count; - rdmsrl(event->hw.event_base, count); - return count; -} - static void snb_uncore_msr_init_box(struct intel_uncore_box *box) { if (box->pmu->pmu_idx == 0) { @@ -656,7 +651,7 @@ static struct intel_uncore_ops snb_uncore_msr_ops = { .init_box = snb_uncore_msr_init_box, .disable_event = snb_uncore_msr_disable_event, .enable_event = snb_uncore_msr_enable_event, - .read_counter = snb_uncore_msr_read_counter, + .read_counter = uncore_msr_read_counter, }; static struct event_constraint snb_uncore_cbox_constraints[] = { @@ -744,7 +739,7 @@ static struct intel_uncore_ops nhm_uncore_msr_ops = { .enable_box = nhm_uncore_msr_enable_box, .disable_event = snb_uncore_msr_disable_event, .enable_event = nhm_uncore_msr_enable_event, - .read_counter = snb_uncore_msr_read_counter, + .read_counter = uncore_msr_read_counter, }; static struct intel_uncore_type nhm_uncore = { @@ -769,6 +764,1047 @@ static struct intel_uncore_type *nhm_msr_uncores[] = { }; /* end of Nehalem uncore support */ +/* Nehalem-EX uncore support */ +static struct intel_uncore_ops nhmex_uncore_ops; + +#define __BITS_VALUE(x, i, n) ((typeof(x))(((x) >> ((i) * (n))) & \ + ((1ULL << (n)) - 1))) + +DEFINE_UNCORE_FORMAT_ATTR(event5, event, "config:1-5"); +DEFINE_UNCORE_FORMAT_ATTR(counter, counter, "config:6-7"); +DEFINE_UNCORE_FORMAT_ATTR(mm_cfg, mm_cfg, "config:63"); +DEFINE_UNCORE_FORMAT_ATTR(match, match, "config1:0-63"); +DEFINE_UNCORE_FORMAT_ATTR(mask, mask, "config2:0-63"); + +static struct attribute *nhmex_uncore_ubox_formats_attr[] = { + &format_attr_event.attr, + &format_attr_edge.attr, + NULL, +}; + +static struct attribute_group nhmex_uncore_ubox_format_group = { + .name = "format", + .attrs = nhmex_uncore_ubox_formats_attr, +}; + +static struct intel_uncore_type nhmex_uncore_ubox = { + .name = "ubox", + .num_counters = 1, + .num_boxes = 1, + .perf_ctr_bits = 48, + .event_ctl = NHMEX_U_MSR_PMON_EV_SEL, + .perf_ctr = NHMEX_U_MSR_PMON_CTR, + .event_mask = NHMEX_U_PMON_RAW_EVENT_MASK, + .box_ctl = NHMEX_U_MSR_PMON_GLOBAL_CTL, + .ops = &nhmex_uncore_ops, + .format_group = &nhmex_uncore_ubox_format_group +}; + +static struct attribute *nhmex_uncore_cbox_formats_attr[] = { + &format_attr_event.attr, + &format_attr_umask.attr, + &format_attr_edge.attr, + &format_attr_inv.attr, + &format_attr_thresh8.attr, + NULL, +}; + +static struct attribute_group nhmex_uncore_cbox_format_group = { + .name = "format", + .attrs = nhmex_uncore_cbox_formats_attr, +}; + +static struct intel_uncore_type nhmex_uncore_cbox = { + .name = "cbox", + .num_counters = 6, + .num_boxes = 8, + .perf_ctr_bits = 48, + .event_ctl = NHMEX_C0_MSR_PMON_EV_SEL0, + .perf_ctr = NHMEX_C0_MSR_PMON_CTR0, + .event_mask = NHMEX_PMON_RAW_EVENT_MASK, + .box_ctl = NHMEX_C0_MSR_PMON_GLOBAL_CTL, + .msr_offset = NHMEX_C_MSR_OFFSET, + .pair_ctr_ctl = 1, + .ops = &nhmex_uncore_ops, + .format_group = &nhmex_uncore_cbox_format_group +}; + +static struct uncore_event_desc nhmex_uncore_wbox_events[] = { + INTEL_UNCORE_EVENT_DESC(clockticks, "event=0xff,umask=0"), + { /* end: all zeroes */ }, +}; + +static struct intel_uncore_type nhmex_uncore_wbox = { + .name = "wbox", + .num_counters = 4, + .num_boxes = 1, + .perf_ctr_bits = 48, + .event_ctl = NHMEX_W_MSR_PMON_CNT0, + .perf_ctr = NHMEX_W_MSR_PMON_EVT_SEL0, + .fixed_ctr = NHMEX_W_MSR_PMON_FIXED_CTR, + .fixed_ctl = NHMEX_W_MSR_PMON_FIXED_CTL, + .event_mask = NHMEX_PMON_RAW_EVENT_MASK, + .box_ctl = NHMEX_W_MSR_GLOBAL_CTL, + .pair_ctr_ctl = 1, + .event_descs = nhmex_uncore_wbox_events, + .ops = &nhmex_uncore_ops, + .format_group = &nhmex_uncore_cbox_format_group +}; + +static int nhmex_bbox_hw_config(struct intel_uncore_box *box, + struct perf_event *event) +{ + struct hw_perf_event *hwc = &event->hw; + struct hw_perf_event_extra *reg1 = &hwc->extra_reg; + struct hw_perf_event_extra *reg2 = &hwc->branch_reg; + int ctr, ev_sel; + + ctr = (hwc->config & NHMEX_B_PMON_CTR_MASK) >> + NHMEX_B_PMON_CTR_SHIFT; + ev_sel = (hwc->config & NHMEX_B_PMON_CTL_EV_SEL_MASK) >> + NHMEX_B_PMON_CTL_EV_SEL_SHIFT; + + /* events that do not use the match/mask registers */ + if ((ctr == 0 && ev_sel > 0x3) || (ctr == 1 && ev_sel > 0x6) || + (ctr == 2 && ev_sel != 0x4) || ctr == 3) + return 0; + + if (box->pmu->pmu_idx == 0) + reg1->reg = NHMEX_B0_MSR_MATCH; + else + reg1->reg = NHMEX_B1_MSR_MATCH; + reg1->idx = 0; + reg1->config = event->attr.config1; + reg2->config = event->attr.config2; + return 0; +} + +static void nhmex_bbox_msr_enable_event(struct intel_uncore_box *box, + struct perf_event *event) +{ + struct hw_perf_event *hwc = &event->hw; + struct hw_perf_event_extra *reg1 = &hwc->extra_reg; + struct hw_perf_event_extra *reg2 = &hwc->branch_reg; + + if (reg1->idx != EXTRA_REG_NONE) { + wrmsrl(reg1->reg, reg1->config); + wrmsrl(reg1->reg + 1, reg2->config); + } + wrmsrl(hwc->config_base, NHMEX_PMON_CTL_EN_BIT0 | + (hwc->config & NHMEX_B_PMON_CTL_EV_SEL_MASK)); +} + +/* + * The Bbox has 4 counters, but each counter monitors different events. + * Use bits 6-7 in event config to select counter. + */ +static struct event_constraint nhmex_uncore_bbox_constraints[] = { + EVENT_CONSTRAINT(0 , 1, 0xc0), + EVENT_CONSTRAINT(0x40, 2, 0xc0), + EVENT_CONSTRAINT(0x80, 4, 0xc0), + EVENT_CONSTRAINT(0xc0, 8, 0xc0), + EVENT_CONSTRAINT_END, +}; + +static struct attribute *nhmex_uncore_bbox_formats_attr[] = { + &format_attr_event5.attr, + &format_attr_counter.attr, + &format_attr_match.attr, + &format_attr_mask.attr, + NULL, +}; + +static struct attribute_group nhmex_uncore_bbox_format_group = { + .name = "format", + .attrs = nhmex_uncore_bbox_formats_attr, +}; + +static struct intel_uncore_type nhmex_uncore_bbox = { + .name = "bbox", + .num_counters = 4, + .num_boxes = 2, + .perf_ctr_bits = 48, + .event_ctl = NHMEX_B0_MSR_PMON_CTL0, + .perf_ctr = NHMEX_B0_MSR_PMON_CTR0, + .event_mask = NHMEX_B_PMON_RAW_EVENT_MASK, + .box_ctl = NHMEX_B0_MSR_PMON_GLOBAL_CTL, + .msr_offset = NHMEX_B_MSR_OFFSET, + .pair_ctr_ctl = 1, + .num_shared_regs = 1, + .constraints = nhmex_uncore_bbox_constraints, + .ops = &nhmex_uncore_ops, + .format_group = &nhmex_uncore_bbox_format_group +}; + +static int nhmex_sbox_hw_config(struct intel_uncore_box *box, + struct perf_event *event) +{ + struct hw_perf_event_extra *reg1 = &event->hw.extra_reg; + struct hw_perf_event_extra *reg2 = &event->hw.branch_reg; + + if (event->attr.config & NHMEX_S_PMON_MM_CFG_EN) { + reg1->config = event->attr.config1; + reg2->config = event->attr.config2; + } else { + reg1->config = ~0ULL; + reg2->config = ~0ULL; + } + if (box->pmu->pmu_idx == 0) + reg1->reg = NHMEX_S0_MSR_MM_CFG; + else + reg1->reg = NHMEX_S1_MSR_MM_CFG; + reg1->idx = 0; + return 0; +} + +static void nhmex_sbox_msr_enable_event(struct intel_uncore_box *box, + struct perf_event *event) +{ + struct hw_perf_event *hwc = &event->hw; + struct hw_perf_event_extra *reg1 = &hwc->extra_reg; + struct hw_perf_event_extra *reg2 = &hwc->branch_reg; + + wrmsrl(reg1->reg, 0); + if (reg1->config != ~0ULL || reg2->config != ~0ULL) { + wrmsrl(reg1->reg + 1, reg1->config); + wrmsrl(reg1->reg + 2, reg2->config); + wrmsrl(reg1->reg, NHMEX_S_PMON_MM_CFG_EN); + } + wrmsrl(hwc->config_base, hwc->config | NHMEX_PMON_CTL_EN_BIT22); +} + +static struct attribute *nhmex_uncore_sbox_formats_attr[] = { + &format_attr_event.attr, + &format_attr_umask.attr, + &format_attr_edge.attr, + &format_attr_inv.attr, + &format_attr_thresh8.attr, + &format_attr_mm_cfg.attr, + &format_attr_match.attr, + &format_attr_mask.attr, + NULL, +}; + +static struct attribute_group nhmex_uncore_sbox_format_group = { + .name = "format", + .attrs = nhmex_uncore_sbox_formats_attr, +}; + +static struct intel_uncore_type nhmex_uncore_sbox = { + .name = "sbox", + .num_counters = 4, + .num_boxes = 2, + .perf_ctr_bits = 48, + .event_ctl = NHMEX_S0_MSR_PMON_CTL0, + .perf_ctr = NHMEX_S0_MSR_PMON_CTR0, + .event_mask = NHMEX_PMON_RAW_EVENT_MASK, + .box_ctl = NHMEX_S0_MSR_PMON_GLOBAL_CTL, + .msr_offset = NHMEX_S_MSR_OFFSET, + .pair_ctr_ctl = 1, + .num_shared_regs = 1, + .ops = &nhmex_uncore_ops, + .format_group = &nhmex_uncore_sbox_format_group +}; + +enum { + EXTRA_REG_NHMEX_M_FILTER, + EXTRA_REG_NHMEX_M_DSP, + EXTRA_REG_NHMEX_M_ISS, + EXTRA_REG_NHMEX_M_MAP, + EXTRA_REG_NHMEX_M_MSC_THR, + EXTRA_REG_NHMEX_M_PGT, + EXTRA_REG_NHMEX_M_PLD, + EXTRA_REG_NHMEX_M_ZDP_CTL_FVC, +}; + +static struct extra_reg nhmex_uncore_mbox_extra_regs[] = { + MBOX_INC_SEL_EXTAR_REG(0x0, DSP), + MBOX_INC_SEL_EXTAR_REG(0x4, MSC_THR), + MBOX_INC_SEL_EXTAR_REG(0x5, MSC_THR), + MBOX_INC_SEL_EXTAR_REG(0x9, ISS), + /* event 0xa uses two extra registers */ + MBOX_INC_SEL_EXTAR_REG(0xa, ISS), + MBOX_INC_SEL_EXTAR_REG(0xa, PLD), + MBOX_INC_SEL_EXTAR_REG(0xb, PLD), + MBOX_INC_SEL_EXTAR_REG(0xd, ZDP_CTL_FVC), + MBOX_INC_SEL_EXTAR_REG(0xe, ZDP_CTL_FVC), + MBOX_INC_SEL_EXTAR_REG(0xf, ZDP_CTL_FVC), + MBOX_INC_SEL_EXTAR_REG(0x10, ZDP_CTL_FVC), + MBOX_INC_SEL_EXTAR_REG(0x16, PGT), + MBOX_SET_FLAG_SEL_EXTRA_REG(0x0, DSP), + MBOX_SET_FLAG_SEL_EXTRA_REG(0x1, ISS), + MBOX_SET_FLAG_SEL_EXTRA_REG(0x5, PGT), + MBOX_SET_FLAG_SEL_EXTRA_REG(0x6, MAP), + EVENT_EXTRA_END +}; + +static bool nhmex_mbox_get_shared_reg(struct intel_uncore_box *box, + int idx, u64 config) +{ + struct intel_uncore_extra_reg *er; + unsigned long flags; + bool ret = false; + u64 mask; + + if (idx < EXTRA_REG_NHMEX_M_ZDP_CTL_FVC) { + er = &box->shared_regs[idx]; + raw_spin_lock_irqsave(&er->lock, flags); + if (!atomic_read(&er->ref) || er->config == config) { + atomic_inc(&er->ref); + er->config = config; + ret = true; + } + raw_spin_unlock_irqrestore(&er->lock, flags); + return ret; + } + /* + * The ZDP_CTL_FVC MSR has 4 sub-fields which are used by + * event FVC0 ~ FVC4 respectively. + */ + idx -= EXTRA_REG_NHMEX_M_ZDP_CTL_FVC; + if (WARN_ON_ONCE(idx >= 4)) + return false; + + mask = NHMEX_M_PMON_ZDP_CTL_FVC_MASK; + er = &box->shared_regs[EXTRA_REG_NHMEX_M_ZDP_CTL_FVC]; + raw_spin_lock_irqsave(&er->lock, flags); + if (__BITS_VALUE(atomic_read(&er->ref), idx, 8)) + mask |= NHMEX_M_PMON_ZDP_CTL_FVC_EVENT_MASK(idx); + + if (!atomic_read(&er->ref) || !((er->config ^ config) & mask)) { + atomic_add(1 << (idx * 8), &er->ref); + mask = NHMEX_M_PMON_ZDP_CTL_FVC_MASK | + NHMEX_M_PMON_ZDP_CTL_FVC_EVENT_MASK(idx); + er->config &= ~mask; + er->config |= (config & mask); + ret = true; + } + raw_spin_unlock_irqrestore(&er->lock, flags); + return ret; +} + +static void nhmex_mbox_put_shared_reg(struct intel_uncore_box *box, int idx) +{ + struct intel_uncore_extra_reg *er; + + if (idx < EXTRA_REG_NHMEX_M_ZDP_CTL_FVC) { + er = &box->shared_regs[idx]; + atomic_dec(&er->ref); + return; + } + + idx -= EXTRA_REG_NHMEX_M_ZDP_CTL_FVC; + er = &box->shared_regs[EXTRA_REG_NHMEX_M_ZDP_CTL_FVC]; + atomic_sub(1 << (idx * 8), &er->ref); +} + +u64 nhmex_mbox_alter_er(struct perf_event *event, int new_idx, bool modify) +{ + struct hw_perf_event *hwc = &event->hw; + struct hw_perf_event_extra *reg1 = &hwc->extra_reg; + int idx, orig_idx = __BITS_VALUE(reg1->idx, 0, 8); + u64 config = reg1->config; + + idx = orig_idx - EXTRA_REG_NHMEX_M_ZDP_CTL_FVC; + config &= NHMEX_M_PMON_ZDP_CTL_FVC_EVENT_MASK(idx); + if (new_idx > orig_idx) { + idx = new_idx - orig_idx; + config <<= 3 * idx; + } else { + idx = orig_idx - new_idx; + config >>= 3 * idx; + } + config |= NHMEX_M_PMON_ZDP_CTL_FVC_MASK & reg1->config; + if (modify) { + if (new_idx > orig_idx) + hwc->config += idx << NHMEX_M_PMON_CTL_INC_SEL_SHIFT; + else + hwc->config -= idx << NHMEX_M_PMON_CTL_INC_SEL_SHIFT; + reg1->config = config; + reg1->idx = ~0xff | new_idx; + } + return config; +} + +static struct event_constraint * +nhmex_mbox_get_constraint(struct intel_uncore_box *box, + struct perf_event *event) +{ + struct hw_perf_event_extra *reg1 = &event->hw.extra_reg; + struct hw_perf_event_extra *reg2 = &event->hw.branch_reg; + int i, idx[2], alloc = 0; + u64 config1 = reg1->config; + + idx[0] = __BITS_VALUE(reg1->idx, 0, 8); + idx[1] = __BITS_VALUE(reg1->idx, 1, 8); +again: + for (i = 0; i < 2; i++) { + if (box->phys_id >= 0 && (reg1->alloc & (0x1 << i))) + idx[i] = 0xff; + + if (idx[i] == 0xff) + continue; + + if (!nhmex_mbox_get_shared_reg(box, idx[i], + __BITS_VALUE(config1, i, 32))) + goto fail; + alloc |= (0x1 << i); + } + + if (!(box->phys_id >= 0 && reg2->alloc) && + !nhmex_mbox_get_shared_reg(box, reg2->idx, reg2->config)) + goto fail; + + if (box->phys_id >= 0) { + if (idx[0] != 0xff && + idx[0] != __BITS_VALUE(reg1->idx, 0, 8)) + nhmex_mbox_alter_er(event, idx[0], true); + reg1->alloc |= alloc; + reg2->alloc = 1; + } + return NULL; +fail: + if (idx[0] != 0xff && !(alloc & 0x1) && + idx[0] >= EXTRA_REG_NHMEX_M_ZDP_CTL_FVC) { + /* + * There are four event selection fields in the ZDP_CTL_FVC + * register, they are for events 0xd ~ 0x10 respectively. + * Try the rest fields if we failed to get the current field. + */ + BUG_ON(__BITS_VALUE(reg1->idx, 1, 8) != 0xff); + idx[0] -= EXTRA_REG_NHMEX_M_ZDP_CTL_FVC; + idx[0] = (idx[0] + 1) % 4; + idx[0] += EXTRA_REG_NHMEX_M_ZDP_CTL_FVC; + if (idx[0] != __BITS_VALUE(reg1->idx, 0, 8)) { + config1 = nhmex_mbox_alter_er(event, idx[0], false); + goto again; + } + } + + if (alloc & 0x1) + nhmex_mbox_put_shared_reg(box, idx[0]); + if (alloc & 0x2) + nhmex_mbox_put_shared_reg(box, idx[1]); + return &constraint_empty; +} + +static void nhmex_mbox_put_constraint(struct intel_uncore_box *box, + struct perf_event *event) +{ + struct hw_perf_event_extra *reg1 = &event->hw.extra_reg; + struct hw_perf_event_extra *reg2 = &event->hw.branch_reg; + + if (box->phys_id < 0) + return; + + if (reg1->alloc & 0x1) + nhmex_mbox_put_shared_reg(box, __BITS_VALUE(reg1->idx, 0, 8)); + if (reg1->alloc & 0x2) + nhmex_mbox_put_shared_reg(box, __BITS_VALUE(reg1->idx, 1, 8)); + reg1->alloc = 0; + + if (reg2->alloc) { + nhmex_mbox_put_shared_reg(box, reg2->idx); + reg2->alloc = 0; + } +} + +static int nhmex_mbox_extra_reg_idx(struct extra_reg *er) +{ + + if (er->idx < EXTRA_REG_NHMEX_M_ZDP_CTL_FVC) + return er->idx; + return er->idx + (er->event >> NHMEX_M_PMON_CTL_INC_SEL_SHIFT) - 0xd; +} + +static int nhmex_mbox_hw_config(struct intel_uncore_box *box, + struct perf_event *event) +{ + struct intel_uncore_type *type = box->pmu->type; + struct hw_perf_event_extra *reg1 = &event->hw.extra_reg; + struct hw_perf_event_extra *reg2 = &event->hw.branch_reg; + struct extra_reg *er; + unsigned msr; + int reg_idx = 0; + + if (WARN_ON_ONCE(reg1->idx != -1)) + return -EINVAL; + /* + * The mbox events may require 2 extra MSRs at the most. But only + * the lower 32 bits in these MSRs are significant, so we can use + * config1 to pass two MSRs' config. + */ + for (er = nhmex_uncore_mbox_extra_regs; er->msr; er++) { + if (er->event != (event->hw.config & er->config_mask)) + continue; + if (event->attr.config1 & ~er->valid_mask) + return -EINVAL; + if (er->idx == __BITS_VALUE(reg1->idx, 0, 8) || + er->idx == __BITS_VALUE(reg1->idx, 1, 8)) + continue; + if (WARN_ON_ONCE(reg_idx >= 2)) + return -EINVAL; + + msr = er->msr + type->msr_offset * box->pmu->pmu_idx; + if (WARN_ON_ONCE(msr >= 0xffff || er->idx >= 0xff)) + return -EINVAL; + + /* always use the 32~63 bits to pass the PLD config */ + if (er->idx == EXTRA_REG_NHMEX_M_PLD) + reg_idx = 1; + + reg1->idx &= ~(0xff << (reg_idx * 8)); + reg1->reg &= ~(0xffff << (reg_idx * 16)); + reg1->idx |= nhmex_mbox_extra_reg_idx(er) << (reg_idx * 8); + reg1->reg |= msr << (reg_idx * 16); + reg1->config = event->attr.config1; + reg_idx++; + } + + reg2->idx = EXTRA_REG_NHMEX_M_FILTER; + if (event->attr.config2 & NHMEX_M_PMON_MM_CFG_EN) + reg2->config = event->attr.config2; + else + reg2->config = ~0ULL; + if (box->pmu->pmu_idx == 0) + reg2->reg = NHMEX_M0_MSR_PMU_MM_CFG; + else + reg2->reg = NHMEX_M1_MSR_PMU_MM_CFG; + return 0; +} + +static u64 nhmex_mbox_shared_reg_config(struct intel_uncore_box *box, int idx) +{ + struct intel_uncore_extra_reg *er; + unsigned long flags; + u64 config; + + if (idx < EXTRA_REG_NHMEX_M_ZDP_CTL_FVC) + er = &box->shared_regs[idx]; + else + er = &box->shared_regs[EXTRA_REG_NHMEX_M_ZDP_CTL_FVC]; + + raw_spin_lock_irqsave(&er->lock, flags); + config = er->config; + raw_spin_unlock_irqrestore(&er->lock, flags); + + return config; +} + +static void nhmex_mbox_msr_enable_event(struct intel_uncore_box *box, + struct perf_event *event) +{ + struct hw_perf_event *hwc = &event->hw; + struct hw_perf_event_extra *reg1 = &hwc->extra_reg; + struct hw_perf_event_extra *reg2 = &hwc->branch_reg; + int idx; + + idx = __BITS_VALUE(reg1->idx, 0, 8); + if (idx != 0xff) + wrmsrl(__BITS_VALUE(reg1->reg, 0, 16), + nhmex_mbox_shared_reg_config(box, idx)); + idx = __BITS_VALUE(reg1->idx, 1, 8); + if (idx != 0xff) + wrmsrl(__BITS_VALUE(reg1->reg, 1, 16), + nhmex_mbox_shared_reg_config(box, idx)); + + wrmsrl(reg2->reg, 0); + if (reg2->config != ~0ULL) { + wrmsrl(reg2->reg + 1, + reg2->config & NHMEX_M_PMON_ADDR_MATCH_MASK); + wrmsrl(reg2->reg + 2, NHMEX_M_PMON_ADDR_MASK_MASK & + (reg2->config >> NHMEX_M_PMON_ADDR_MASK_SHIFT)); + wrmsrl(reg2->reg, NHMEX_M_PMON_MM_CFG_EN); + } + + wrmsrl(hwc->config_base, hwc->config | NHMEX_PMON_CTL_EN_BIT0); +} + +DEFINE_UNCORE_FORMAT_ATTR(count_mode, count_mode, "config:2-3"); +DEFINE_UNCORE_FORMAT_ATTR(storage_mode, storage_mode, "config:4-5"); +DEFINE_UNCORE_FORMAT_ATTR(wrap_mode, wrap_mode, "config:6"); +DEFINE_UNCORE_FORMAT_ATTR(flag_mode, flag_mode, "config:7"); +DEFINE_UNCORE_FORMAT_ATTR(inc_sel, inc_sel, "config:9-13"); +DEFINE_UNCORE_FORMAT_ATTR(set_flag_sel, set_flag_sel, "config:19-21"); +DEFINE_UNCORE_FORMAT_ATTR(filter_cfg, filter_cfg, "config2:63"); +DEFINE_UNCORE_FORMAT_ATTR(filter_match, filter_match, "config2:0-33"); +DEFINE_UNCORE_FORMAT_ATTR(filter_mask, filter_mask, "config2:34-61"); +DEFINE_UNCORE_FORMAT_ATTR(dsp, dsp, "config1:0-31"); +DEFINE_UNCORE_FORMAT_ATTR(thr, thr, "config1:0-31"); +DEFINE_UNCORE_FORMAT_ATTR(fvc, fvc, "config1:0-31"); +DEFINE_UNCORE_FORMAT_ATTR(pgt, pgt, "config1:0-31"); +DEFINE_UNCORE_FORMAT_ATTR(map, map, "config1:0-31"); +DEFINE_UNCORE_FORMAT_ATTR(iss, iss, "config1:0-31"); +DEFINE_UNCORE_FORMAT_ATTR(pld, pld, "config1:32-63"); + +static struct attribute *nhmex_uncore_mbox_formats_attr[] = { + &format_attr_count_mode.attr, + &format_attr_storage_mode.attr, + &format_attr_wrap_mode.attr, + &format_attr_flag_mode.attr, + &format_attr_inc_sel.attr, + &format_attr_set_flag_sel.attr, + &format_attr_filter_cfg.attr, + &format_attr_filter_match.attr, + &format_attr_filter_mask.attr, + &format_attr_dsp.attr, + &format_attr_thr.attr, + &format_attr_fvc.attr, + &format_attr_pgt.attr, + &format_attr_map.attr, + &format_attr_iss.attr, + &format_attr_pld.attr, + NULL, +}; + +static struct attribute_group nhmex_uncore_mbox_format_group = { + .name = "format", + .attrs = nhmex_uncore_mbox_formats_attr, +}; + +static struct uncore_event_desc nhmex_uncore_mbox_events[] = { + INTEL_UNCORE_EVENT_DESC(bbox_cmds_read, "inc_sel=0xd,fvc=0x2800"), + INTEL_UNCORE_EVENT_DESC(bbox_cmds_write, "inc_sel=0xd,fvc=0x2820"), + { /* end: all zeroes */ }, +}; + +static struct intel_uncore_type nhmex_uncore_mbox = { + .name = "mbox", + .num_counters = 6, + .num_boxes = 2, + .perf_ctr_bits = 48, + .event_ctl = NHMEX_M0_MSR_PMU_CTL0, + .perf_ctr = NHMEX_M0_MSR_PMU_CNT0, + .event_mask = NHMEX_M_PMON_RAW_EVENT_MASK, + .box_ctl = NHMEX_M0_MSR_GLOBAL_CTL, + .msr_offset = NHMEX_M_MSR_OFFSET, + .pair_ctr_ctl = 1, + .num_shared_regs = 8, + .event_descs = nhmex_uncore_mbox_events, + .ops = &nhmex_uncore_ops, + .format_group = &nhmex_uncore_mbox_format_group, +}; + +void nhmex_rbox_alter_er(struct intel_uncore_box *box, + struct perf_event *event) +{ + struct hw_perf_event *hwc = &event->hw; + struct hw_perf_event_extra *reg1 = &hwc->extra_reg; + int port; + + if (reg1->idx % 2) { + reg1->idx--; + hwc->config -= 1 << NHMEX_R_PMON_CTL_EV_SEL_SHIFT; + } else { + reg1->idx++; + hwc->config += 1 << NHMEX_R_PMON_CTL_EV_SEL_SHIFT; + } + + port = reg1->idx / 6 + box->pmu->pmu_idx * 4; + switch (reg1->idx % 6) { + case 0: + reg1->reg = NHMEX_R_MSR_PORTN_IPERF_CFG0(port); + break; + case 1: + reg1->reg = NHMEX_R_MSR_PORTN_IPERF_CFG1(port); + break; + case 2: + reg1->config >>= 8; + break; + case 3: + reg1->config <<= 8; + break; + case 4: + reg1->reg = NHMEX_R_MSR_PORTN_XBR_SET1_MM_CFG(port); + break; + case 5: + reg1->reg = NHMEX_R_MSR_PORTN_XBR_SET2_MM_CFG(port); + break; + }; +} + +/* + * Each rbox has 4 event set which monitor PQI port 0~3 or 4~7. + * An event set consists of 6 events, the third and fourth events + * in an event set use the same extra register. + */ +static struct event_constraint * +nhmex_rbox_get_constraint(struct intel_uncore_box *box, + struct perf_event *event) +{ + struct hw_perf_event *hwc = &event->hw; + struct hw_perf_event_extra *reg1 = &hwc->extra_reg; + struct hw_perf_event_extra *reg2 = &hwc->branch_reg; + struct intel_uncore_extra_reg *er; + unsigned long flags; + int idx, er_idx; + u64 config1; + bool ok = false; + + if (box->phys_id >= 0 && reg1->alloc) + return NULL; + + idx = reg1->idx % 6; + config1 = reg1->config; +again: + er_idx = idx; + if (er_idx > 2) + er_idx--; + er_idx += (reg1->idx / 6) * 5; + + er = &box->shared_regs[er_idx]; + raw_spin_lock_irqsave(&er->lock, flags); + if (idx < 2) { + if (!atomic_read(&er->ref) || er->config == reg1->config) { + atomic_inc(&er->ref); + er->config = reg1->config; + ok = true; + } + } else if (idx == 2 || idx == 3) { + /* use the same extra register, but different fields */ + u64 mask = 0xff << ((idx - 2) * 8); + if (!__BITS_VALUE(atomic_read(&er->ref), idx - 2, 8) || + !((er->config ^ config1) & mask)) { + atomic_add(1 << ((idx - 2) * 8), &er->ref); + er->config &= ~mask; + er->config |= config1 & mask; + ok = true; + } + } else { + if (!atomic_read(&er->ref) || + (er->config == (hwc->config >> 32) && + er->config1 == reg1->config && + er->config2 == reg2->config)) { + atomic_inc(&er->ref); + er->config = (hwc->config >> 32); + er->config1 = reg1->config; + er->config2 = reg2->config; + ok = true; + } + } + raw_spin_unlock_irqrestore(&er->lock, flags); + + if (!ok) { + /* + * The Rbox events and shared registers are always in pairs. + * Check the other register if we failed to get the current + * one. + */ + if (idx % 2) + idx--; + else + idx++; + if (idx != reg1->idx % 6) { + if (idx == 2) + config1 >>= 8; + else if (idx == 3) + config1 <<= 8; + goto again; + } + } else { + if (box->phys_id >= 0) { + if (idx != reg1->idx % 6) + nhmex_rbox_alter_er(box, event); + reg1->alloc = 1; + } + return NULL; + } + return &constraint_empty; +} + +static void nhmex_rbox_put_constraint(struct intel_uncore_box *box, + struct perf_event *event) +{ + struct intel_uncore_extra_reg *er; + struct hw_perf_event_extra *reg1 = &event->hw.extra_reg; + int idx, er_idx; + + if (box->phys_id < 0 || !reg1->alloc) + return; + + idx = reg1->idx % 6; + er_idx = idx; + if (er_idx > 2) + er_idx--; + er_idx += (reg1->idx / 6) * 5; + + er = &box->shared_regs[er_idx]; + if (idx == 2 || idx == 3) + atomic_sub(1 << ((idx - 2) * 8), &er->ref); + else + atomic_dec(&er->ref); + + reg1->alloc = 0; +} + +static int nhmex_rbox_hw_config(struct intel_uncore_box *box, + struct perf_event *event) +{ + struct hw_perf_event *hwc = &event->hw; + struct hw_perf_event_extra *reg1 = &event->hw.extra_reg; + struct hw_perf_event_extra *reg2 = &event->hw.branch_reg; + int port, idx; + + idx = (event->hw.config & NHMEX_R_PMON_CTL_EV_SEL_MASK) >> + NHMEX_R_PMON_CTL_EV_SEL_SHIFT; + if (idx >= 0x18) + return -EINVAL; + + reg1->idx = idx; + reg1->config = event->attr.config1; + + port = idx / 6 + box->pmu->pmu_idx * 4; + idx %= 6; + switch (idx) { + case 0: + reg1->reg = NHMEX_R_MSR_PORTN_IPERF_CFG0(port); + break; + case 1: + reg1->reg = NHMEX_R_MSR_PORTN_IPERF_CFG1(port); + break; + case 2: + case 3: + reg1->reg = NHMEX_R_MSR_PORTN_QLX_CFG(port); + break; + case 4: + case 5: + if (idx == 4) + reg1->reg = NHMEX_R_MSR_PORTN_XBR_SET1_MM_CFG(port); + else + reg1->reg = NHMEX_R_MSR_PORTN_XBR_SET2_MM_CFG(port); + reg2->config = event->attr.config2; + hwc->config |= event->attr.config & (~0ULL << 32); + break; + }; + return 0; +} + +static u64 nhmex_rbox_shared_reg_config(struct intel_uncore_box *box, int idx) +{ + struct intel_uncore_extra_reg *er; + unsigned long flags; + u64 config; + + er = &box->shared_regs[idx]; + raw_spin_lock_irqsave(&er->lock, flags); + config = er->config; + raw_spin_unlock_irqrestore(&er->lock, flags); + return config; +} + +static void nhmex_rbox_msr_enable_event(struct intel_uncore_box *box, + struct perf_event *event) +{ + struct hw_perf_event *hwc = &event->hw; + struct hw_perf_event_extra *reg1 = &hwc->extra_reg; + struct hw_perf_event_extra *reg2 = &hwc->branch_reg; + int idx, er_idx; + + idx = reg1->idx % 6; + er_idx = idx; + if (er_idx > 2) + er_idx--; + er_idx += (reg1->idx / 6) * 5; + + switch (idx) { + case 0: + case 1: + wrmsrl(reg1->reg, reg1->config); + break; + case 2: + case 3: + wrmsrl(reg1->reg, nhmex_rbox_shared_reg_config(box, er_idx)); + break; + case 4: + case 5: + wrmsrl(reg1->reg, reg1->config); + wrmsrl(reg1->reg + 1, hwc->config >> 32); + wrmsrl(reg1->reg + 2, reg2->config); + break; + }; + + wrmsrl(hwc->config_base, NHMEX_PMON_CTL_EN_BIT0 | + (hwc->config & NHMEX_R_PMON_CTL_EV_SEL_MASK)); +} + +DEFINE_UNCORE_FORMAT_ATTR(xbr_match, xbr_match, "config:32-63"); +DEFINE_UNCORE_FORMAT_ATTR(xbr_mm_cfg, xbr_mm_cfg, "config1:0-63"); +DEFINE_UNCORE_FORMAT_ATTR(xbr_mask, xbr_mask, "config2:0-63"); +DEFINE_UNCORE_FORMAT_ATTR(qlx_cfg, qlx_cfg, "config1:0-15"); +DEFINE_UNCORE_FORMAT_ATTR(iperf_cfg, iperf_cfg, "config1:0-31"); + +static struct attribute *nhmex_uncore_rbox_formats_attr[] = { + &format_attr_event5.attr, + &format_attr_xbr_mm_cfg.attr, + &format_attr_xbr_match.attr, + &format_attr_xbr_mask.attr, + &format_attr_qlx_cfg.attr, + &format_attr_iperf_cfg.attr, + NULL, +}; + +static struct attribute_group nhmex_uncore_rbox_format_group = { + .name = "format", + .attrs = nhmex_uncore_rbox_formats_attr, +}; + +static struct uncore_event_desc nhmex_uncore_rbox_events[] = { + INTEL_UNCORE_EVENT_DESC(qpi0_flit_send, + "event=0x0,iperf_cfg=0x80000000"), + INTEL_UNCORE_EVENT_DESC(qpi1_filt_send, + "event=0x6,iperf_cfg=0x80000000"), + INTEL_UNCORE_EVENT_DESC(qpi0_idle_filt, + "event=0x0,iperf_cfg=0x40000000"), + INTEL_UNCORE_EVENT_DESC(qpi1_idle_filt, + "event=0x6,iperf_cfg=0x40000000"), + INTEL_UNCORE_EVENT_DESC(qpi0_date_response, + "event=0x0,iperf_cfg=0xc4"), + INTEL_UNCORE_EVENT_DESC(qpi1_date_response, + "event=0x6,iperf_cfg=0xc4"), + { /* end: all zeroes */ }, +}; + +static struct intel_uncore_type nhmex_uncore_rbox = { + .name = "rbox", + .num_counters = 8, + .num_boxes = 2, + .perf_ctr_bits = 48, + .event_ctl = NHMEX_R_MSR_PMON_CTL0, + .perf_ctr = NHMEX_R_MSR_PMON_CNT0, + .event_mask = NHMEX_R_PMON_RAW_EVENT_MASK, + .box_ctl = NHMEX_R_MSR_GLOBAL_CTL, + .msr_offset = NHMEX_R_MSR_OFFSET, + .pair_ctr_ctl = 1, + .num_shared_regs = 20, + .event_descs = nhmex_uncore_rbox_events, + .ops = &nhmex_uncore_ops, + .format_group = &nhmex_uncore_rbox_format_group +}; + +static void nhmex_uncore_msr_init_box(struct intel_uncore_box *box) +{ + wrmsrl(NHMEX_U_MSR_PMON_GLOBAL_CTL, NHMEX_U_PMON_GLOBAL_EN_ALL); +} + +static void nhmex_uncore_msr_disable_box(struct intel_uncore_box *box) +{ + unsigned msr = uncore_msr_box_ctl(box); + u64 config; + if (msr) { + rdmsrl(msr, config); + config &= ~((1ULL << uncore_num_counters(box)) - 1); + /* WBox has a fixed counter */ + if (uncore_msr_fixed_ctl(box)) + config &= ~NHMEX_W_PMON_GLOBAL_FIXED_EN; + wrmsrl(msr, config); + } +} + +static void nhmex_uncore_msr_enable_box(struct intel_uncore_box *box) +{ + unsigned msr = uncore_msr_box_ctl(box); + u64 config; + if (msr) { + rdmsrl(msr, config); + config |= (1ULL << uncore_num_counters(box)) - 1; + /* WBox has a fixed counter */ + if (uncore_msr_fixed_ctl(box)) + config |= NHMEX_W_PMON_GLOBAL_FIXED_EN; + wrmsrl(msr, config); + } +} + +static void nhmex_uncore_msr_disable_event(struct intel_uncore_box *box, + struct perf_event *event) +{ + wrmsrl(event->hw.config_base, 0); +} + +static void nhmex_uncore_msr_enable_event(struct intel_uncore_box *box, + struct perf_event *event) +{ + struct hw_perf_event *hwc = &event->hw; + + if (box->pmu->type == &nhmex_uncore_bbox) + nhmex_bbox_msr_enable_event(box, event); + else if (box->pmu->type == &nhmex_uncore_sbox) + nhmex_sbox_msr_enable_event(box, event); + else if (box->pmu->type == &nhmex_uncore_mbox) + nhmex_mbox_msr_enable_event(box, event); + else if (box->pmu->type == &nhmex_uncore_rbox) + nhmex_rbox_msr_enable_event(box, event); + else if (hwc->idx >= UNCORE_PMC_IDX_FIXED) + wrmsrl(hwc->config_base, NHMEX_PMON_CTL_EN_BIT0); + else if (box->pmu->type->event_mask & NHMEX_PMON_CTL_EN_BIT0) + wrmsrl(hwc->config_base, hwc->config | NHMEX_PMON_CTL_EN_BIT22); + else + wrmsrl(hwc->config_base, hwc->config | NHMEX_PMON_CTL_EN_BIT0); +} + +static struct event_constraint * +nhmex_uncore_get_constraint(struct intel_uncore_box *box, + struct perf_event *event) +{ + if (box->pmu->type == &nhmex_uncore_bbox || + box->pmu->type == &nhmex_uncore_sbox) + return uncore_get_constraint(box, event); + if (box->pmu->type == &nhmex_uncore_mbox) + return nhmex_mbox_get_constraint(box, event); + if (box->pmu->type == &nhmex_uncore_rbox) + return nhmex_rbox_get_constraint(box, event); + return NULL; +} + +static void nhmex_uncore_put_constraint(struct intel_uncore_box *box, + struct perf_event *event) +{ + if (box->pmu->type == &nhmex_uncore_bbox || + box->pmu->type == &nhmex_uncore_sbox) + uncore_put_constraint(box, event); + else if (box->pmu->type == &nhmex_uncore_mbox) + nhmex_mbox_put_constraint(box, event); + else if (box->pmu->type == &nhmex_uncore_rbox) + nhmex_rbox_put_constraint(box, event); +} + +static int nhmex_uncore_hw_config(struct intel_uncore_box *box, + struct perf_event *event) +{ + if (box->pmu->type == &nhmex_uncore_bbox) + return nhmex_bbox_hw_config(box, event); + if (box->pmu->type == &nhmex_uncore_sbox) + return nhmex_sbox_hw_config(box, event); + if (box->pmu->type == &nhmex_uncore_mbox) + return nhmex_mbox_hw_config(box, event); + if (box->pmu->type == &nhmex_uncore_rbox) + return nhmex_rbox_hw_config(box, event); + return 0; +} + +static struct intel_uncore_ops nhmex_uncore_ops = { + .init_box = nhmex_uncore_msr_init_box, + .disable_box = nhmex_uncore_msr_disable_box, + .enable_box = nhmex_uncore_msr_enable_box, + .disable_event = nhmex_uncore_msr_disable_event, + .enable_event = nhmex_uncore_msr_enable_event, + .read_counter = uncore_msr_read_counter, + .hw_config = nhmex_uncore_hw_config, + .get_constraint = nhmex_uncore_get_constraint, + .put_constraint = nhmex_uncore_put_constraint, +}; + +static struct intel_uncore_type *nhmex_msr_uncores[] = { + &nhmex_uncore_ubox, + &nhmex_uncore_cbox, + &nhmex_uncore_bbox, + &nhmex_uncore_sbox, + &nhmex_uncore_mbox, + &nhmex_uncore_rbox, + &nhmex_uncore_wbox, + NULL, +}; +/* end of Nehalem-EX uncore support */ + static void uncore_assign_hw_event(struct intel_uncore_box *box, struct perf_event *event, int idx) { @@ -1257,6 +2293,11 @@ int uncore_pmu_event_init(struct perf_event *event) event->hw.last_tag = ~0ULL; event->hw.extra_reg.idx = EXTRA_REG_NONE; + event->hw.idx = -1; + event->hw.last_tag = ~0ULL; + event->hw.extra_reg.idx = -1; + BUG_ON(event->hw.extra_reg.alloc); + if (event->attr.config == UNCORE_FIXED_EVENT) { /* no fixed counter */ if (!pmu->type->fixed_ctl) @@ -1767,6 +2808,9 @@ static int __init uncore_cpu_init(void) snbep_uncore_cbox.num_boxes = max_cores; msr_uncores = snbep_msr_uncores; break; + case 46: + msr_uncores = nhmex_msr_uncores; + break; default: return 0; } diff --git a/arch/x86/kernel/cpu/perf_event_intel_uncore.h b/arch/x86/kernel/cpu/perf_event_intel_uncore.h index b13e9ea..5757f56 100644 --- a/arch/x86/kernel/cpu/perf_event_intel_uncore.h +++ b/arch/x86/kernel/cpu/perf_event_intel_uncore.h @@ -5,8 +5,6 @@ #include "perf_event.h" #define UNCORE_PMU_NAME_LEN 32 -#define UNCORE_BOX_HASH_SIZE 8 - #define UNCORE_PMU_HRTIMER_INTERVAL (60 * NSEC_PER_SEC) #define UNCORE_FIXED_EVENT 0xff @@ -158,6 +156,190 @@ #define SNBEP_PCU_MSR_CORE_C3_CTR 0x3fc #define SNBEP_PCU_MSR_CORE_C6_CTR 0x3fd +/* NHM-EX event control */ +#define NHMEX_PMON_CTL_EV_SEL_MASK 0x000000ff +#define NHMEX_PMON_CTL_UMASK_MASK 0x0000ff00 +#define NHMEX_PMON_CTL_EN_BIT0 (1 << 0) +#define NHMEX_PMON_CTL_EDGE_DET (1 << 18) +#define NHMEX_PMON_CTL_PMI_EN (1 << 20) +#define NHMEX_PMON_CTL_EN_BIT22 (1 << 22) +#define NHMEX_PMON_CTL_INVERT (1 << 23) +#define NHMEX_PMON_CTL_TRESH_MASK 0xff000000 +#define NHMEX_PMON_RAW_EVENT_MASK (NHMEX_PMON_CTL_EV_SEL_MASK | \ + NHMEX_PMON_CTL_UMASK_MASK | \ + NHMEX_PMON_CTL_EDGE_DET | \ + NHMEX_PMON_CTL_INVERT | \ + NHMEX_PMON_CTL_TRESH_MASK) + +/* NHM-EX Ubox */ +#define NHMEX_U_MSR_PMON_GLOBAL_CTL 0xc00 +#define NHMEX_U_MSR_PMON_CTR 0xc11 +#define NHMEX_U_MSR_PMON_EV_SEL 0xc10 + +#define NHMEX_U_PMON_GLOBAL_EN (1 << 0) +#define NHMEX_U_PMON_GLOBAL_PMI_CORE_SEL 0x0000001e +#define NHMEX_U_PMON_GLOBAL_EN_ALL (1 << 28) +#define NHMEX_U_PMON_GLOBAL_RST_ALL (1 << 29) +#define NHMEX_U_PMON_GLOBAL_FRZ_ALL (1 << 31) + +#define NHMEX_U_PMON_RAW_EVENT_MASK \ + (NHMEX_PMON_CTL_EV_SEL_MASK | \ + NHMEX_PMON_CTL_EDGE_DET) + +/* NHM-EX Cbox */ +#define NHMEX_C0_MSR_PMON_GLOBAL_CTL 0xd00 +#define NHMEX_C0_MSR_PMON_CTR0 0xd11 +#define NHMEX_C0_MSR_PMON_EV_SEL0 0xd10 +#define NHMEX_C_MSR_OFFSET 0x20 + +/* NHM-EX Bbox */ +#define NHMEX_B0_MSR_PMON_GLOBAL_CTL 0xc20 +#define NHMEX_B0_MSR_PMON_CTR0 0xc31 +#define NHMEX_B0_MSR_PMON_CTL0 0xc30 +#define NHMEX_B_MSR_OFFSET 0x40 +#define NHMEX_B0_MSR_MATCH 0xe45 +#define NHMEX_B0_MSR_MASK 0xe46 +#define NHMEX_B1_MSR_MATCH 0xe4d +#define NHMEX_B1_MSR_MASK 0xe4e + +#define NHMEX_B_PMON_CTL_EN (1 << 0) +#define NHMEX_B_PMON_CTL_EV_SEL_SHIFT 1 +#define NHMEX_B_PMON_CTL_EV_SEL_MASK \ + (0x1f << NHMEX_B_PMON_CTL_EV_SEL_SHIFT) +#define NHMEX_B_PMON_CTR_SHIFT 6 +#define NHMEX_B_PMON_CTR_MASK \ + (0x3 << NHMEX_B_PMON_CTR_SHIFT) +#define NHMEX_B_PMON_RAW_EVENT_MASK \ + (NHMEX_B_PMON_CTL_EV_SEL_MASK | \ + NHMEX_B_PMON_CTR_MASK) + +/* NHM-EX Sbox */ +#define NHMEX_S0_MSR_PMON_GLOBAL_CTL 0xc40 +#define NHMEX_S0_MSR_PMON_CTR0 0xc51 +#define NHMEX_S0_MSR_PMON_CTL0 0xc50 +#define NHMEX_S_MSR_OFFSET 0x80 +#define NHMEX_S0_MSR_MM_CFG 0xe48 +#define NHMEX_S0_MSR_MATCH 0xe49 +#define NHMEX_S0_MSR_MASK 0xe4a +#define NHMEX_S1_MSR_MM_CFG 0xe58 +#define NHMEX_S1_MSR_MATCH 0xe59 +#define NHMEX_S1_MSR_MASK 0xe5a + +#define NHMEX_S_PMON_MM_CFG_EN (0x1ULL << 63) + +/* NHM-EX Mbox */ +#define NHMEX_M0_MSR_GLOBAL_CTL 0xca0 +#define NHMEX_M0_MSR_PMU_DSP 0xca5 +#define NHMEX_M0_MSR_PMU_ISS 0xca6 +#define NHMEX_M0_MSR_PMU_MAP 0xca7 +#define NHMEX_M0_MSR_PMU_MSC_THR 0xca8 +#define NHMEX_M0_MSR_PMU_PGT 0xca9 +#define NHMEX_M0_MSR_PMU_PLD 0xcaa +#define NHMEX_M0_MSR_PMU_ZDP_CTL_FVC 0xcab +#define NHMEX_M0_MSR_PMU_CTL0 0xcb0 +#define NHMEX_M0_MSR_PMU_CNT0 0xcb1 +#define NHMEX_M_MSR_OFFSET 0x40 +#define NHMEX_M0_MSR_PMU_MM_CFG 0xe54 +#define NHMEX_M1_MSR_PMU_MM_CFG 0xe5c + +#define NHMEX_M_PMON_MM_CFG_EN (1ULL << 63) +#define NHMEX_M_PMON_ADDR_MATCH_MASK 0x3ffffffffULL +#define NHMEX_M_PMON_ADDR_MASK_MASK 0x7ffffffULL +#define NHMEX_M_PMON_ADDR_MASK_SHIFT 34 + +#define NHMEX_M_PMON_CTL_EN (1 << 0) +#define NHMEX_M_PMON_CTL_PMI_EN (1 << 1) +#define NHMEX_M_PMON_CTL_COUNT_MODE_SHIFT 2 +#define NHMEX_M_PMON_CTL_COUNT_MODE_MASK \ + (0x3 << NHMEX_M_PMON_CTL_COUNT_MODE_SHIFT) +#define NHMEX_M_PMON_CTL_STORAGE_MODE_SHIFT 4 +#define NHMEX_M_PMON_CTL_STORAGE_MODE_MASK \ + (0x3 << NHMEX_M_PMON_CTL_STORAGE_MODE_SHIFT) +#define NHMEX_M_PMON_CTL_WRAP_MODE (1 << 6) +#define NHMEX_M_PMON_CTL_FLAG_MODE (1 << 7) +#define NHMEX_M_PMON_CTL_INC_SEL_SHIFT 9 +#define NHMEX_M_PMON_CTL_INC_SEL_MASK \ + (0x1f << NHMEX_M_PMON_CTL_INC_SEL_SHIFT) +#define NHMEX_M_PMON_CTL_SET_FLAG_SEL_SHIFT 19 +#define NHMEX_M_PMON_CTL_SET_FLAG_SEL_MASK \ + (0x7 << NHMEX_M_PMON_CTL_SET_FLAG_SEL_SHIFT) +#define NHMEX_M_PMON_RAW_EVENT_MASK \ + (NHMEX_M_PMON_CTL_COUNT_MODE_MASK | \ + NHMEX_M_PMON_CTL_STORAGE_MODE_MASK | \ + NHMEX_M_PMON_CTL_WRAP_MODE | \ + NHMEX_M_PMON_CTL_FLAG_MODE | \ + NHMEX_M_PMON_CTL_INC_SEL_MASK | \ + NHMEX_M_PMON_CTL_SET_FLAG_SEL_MASK) + + +#define NHMEX_M_PMON_ZDP_CTL_FVC_FVID_MASK 0x1f +#define NHMEX_M_PMON_ZDP_CTL_FVC_BCMD_MASK (0x7 << 5) +#define NHMEX_M_PMON_ZDP_CTL_FVC_RSP_MASK (0x7 << 8) +#define NHMEX_M_PMON_ZDP_CTL_FVC_PBOX_INIT_ERR (1 << 23) +#define NHMEX_M_PMON_ZDP_CTL_FVC_MASK \ + (NHMEX_M_PMON_ZDP_CTL_FVC_FVID_MASK | \ + NHMEX_M_PMON_ZDP_CTL_FVC_BCMD_MASK | \ + NHMEX_M_PMON_ZDP_CTL_FVC_RSP_MASK | \ + NHMEX_M_PMON_ZDP_CTL_FVC_PBOX_INIT_ERR) +#define NHMEX_M_PMON_ZDP_CTL_FVC_EVENT_MASK(n) (0x7 << (11 + 3 * (n))) + + +#define MBOX_INC_SEL(x) ((x) << NHMEX_M_PMON_CTL_INC_SEL_SHIFT) +#define MBOX_SET_FLAG_SEL(x) (((x) << NHMEX_M_PMON_CTL_SET_FLAG_SEL_SHIFT) | \ + NHMEX_M_PMON_CTL_FLAG_MODE) +#define MBOX_INC_SEL_MASK (NHMEX_M_PMON_CTL_INC_SEL_MASK | \ + NHMEX_M_PMON_CTL_FLAG_MODE) +#define MBOX_SET_FLAG_SEL_MASK (NHMEX_M_PMON_CTL_SET_FLAG_SEL_MASK | \ + NHMEX_M_PMON_CTL_FLAG_MODE) +#define MBOX_INC_SEL_EXTAR_REG(c, r) \ + EVENT_EXTRA_REG(MBOX_INC_SEL(c), NHMEX_M0_MSR_PMU_##r, \ + MBOX_INC_SEL_MASK, (u64)-1, NHMEX_M_##r) +#define MBOX_SET_FLAG_SEL_EXTRA_REG(c, r) \ + EVENT_EXTRA_REG(MBOX_SET_FLAG_SEL(c), NHMEX_M0_MSR_PMU_##r, \ + MBOX_SET_FLAG_SEL_MASK, \ + (u64)-1, NHMEX_M_##r) + +/* NHM-EX Rbox */ +#define NHMEX_R_MSR_GLOBAL_CTL 0xe00 +#define NHMEX_R_MSR_PMON_CTL0 0xe10 +#define NHMEX_R_MSR_PMON_CNT0 0xe11 +#define NHMEX_R_MSR_OFFSET 0x20 + +#define NHMEX_R_MSR_PORTN_QLX_CFG(n) \ + ((n) < 4 ? (0xe0c + (n)) : (0xe2c + (n) - 4)) +#define NHMEX_R_MSR_PORTN_IPERF_CFG0(n) (0xe04 + (n)) +#define NHMEX_R_MSR_PORTN_IPERF_CFG1(n) (0xe24 + (n)) +#define NHMEX_R_MSR_PORTN_XBR_OFFSET(n) \ + (((n) < 4 ? 0 : 0x10) + (n) * 4) +#define NHMEX_R_MSR_PORTN_XBR_SET1_MM_CFG(n) \ + (0xe60 + NHMEX_R_MSR_PORTN_XBR_OFFSET(n)) +#define NHMEX_R_MSR_PORTN_XBR_SET1_MATCH(n) \ + (NHMEX_R_MSR_PORTN_XBR_SET1_MM_CFG(n) + 1) +#define NHMEX_R_MSR_PORTN_XBR_SET1_MASK(n) \ + (NHMEX_R_MSR_PORTN_XBR_SET1_MM_CFG(n) + 2) +#define NHMEX_R_MSR_PORTN_XBR_SET2_MM_CFG(n) \ + (0xe70 + NHMEX_R_MSR_PORTN_XBR_OFFSET(n)) +#define NHMEX_R_MSR_PORTN_XBR_SET2_MATCH(n) \ + (NHMEX_R_MSR_PORTN_XBR_SET2_MM_CFG(n) + 1) +#define NHMEX_R_MSR_PORTN_XBR_SET2_MASK(n) \ + (NHMEX_R_MSR_PORTN_XBR_SET2_MM_CFG(n) + 2) + +#define NHMEX_R_PMON_CTL_EN (1 << 0) +#define NHMEX_R_PMON_CTL_EV_SEL_SHIFT 1 +#define NHMEX_R_PMON_CTL_EV_SEL_MASK \ + (0x1f << NHMEX_R_PMON_CTL_EV_SEL_SHIFT) +#define NHMEX_R_PMON_CTL_PMI_EN (1 << 6) +#define NHMEX_R_PMON_RAW_EVENT_MASK NHMEX_R_PMON_CTL_EV_SEL_MASK + +/* NHM-EX Wbox */ +#define NHMEX_W_MSR_GLOBAL_CTL 0xc80 +#define NHMEX_W_MSR_PMON_CNT0 0xc90 +#define NHMEX_W_MSR_PMON_EVT_SEL0 0xc91 +#define NHMEX_W_MSR_PMON_FIXED_CTR 0x394 +#define NHMEX_W_MSR_PMON_FIXED_CTL 0x395 + +#define NHMEX_W_PMON_GLOBAL_FIXED_EN (1ULL << 31) + struct intel_uncore_ops; struct intel_uncore_pmu; struct intel_uncore_box; @@ -178,6 +360,7 @@ struct intel_uncore_type { unsigned msr_offset; unsigned num_shared_regs:8; unsigned single_fixed:1; + unsigned pair_ctr_ctl:1; struct event_constraint unconstrainted; struct event_constraint *constraints; struct intel_uncore_pmu *pmus; @@ -213,7 +396,7 @@ struct intel_uncore_pmu { struct intel_uncore_extra_reg { raw_spinlock_t lock; - u64 config1; + u64 config, config1, config2; atomic_t ref; }; @@ -323,14 +506,16 @@ unsigned uncore_msr_fixed_ctr(struct intel_uncore_box *box) static inline unsigned uncore_msr_event_ctl(struct intel_uncore_box *box, int idx) { - return idx + box->pmu->type->event_ctl + + return box->pmu->type->event_ctl + + (box->pmu->type->pair_ctr_ctl ? 2 * idx : idx) + box->pmu->type->msr_offset * box->pmu->pmu_idx; } static inline unsigned uncore_msr_perf_ctr(struct intel_uncore_box *box, int idx) { - return idx + box->pmu->type->perf_ctr + + return box->pmu->type->perf_ctr + + (box->pmu->type->pair_ctr_ctl ? 2 * idx : idx) + box->pmu->type->msr_offset * box->pmu->pmu_idx; } -- 1.7.10.4 ^ permalink raw reply related [flat|nested] 21+ messages in thread
* Re: [PATCH 5/5] perf/x86: Add Intel Nehalem-EX uncore support 2012-07-04 6:00 ` [PATCH 5/5] perf/x86: Add Intel Nehalem-EX uncore support Yan, Zheng @ 2012-07-04 10:04 ` Peter Zijlstra 2012-07-04 16:00 ` Yan, Zheng 2012-07-05 6:32 ` [PATCH V2 " Yan, Zheng 0 siblings, 2 replies; 21+ messages in thread From: Peter Zijlstra @ 2012-07-04 10:04 UTC (permalink / raw) To: Yan, Zheng; +Cc: eranian, mingo, andi, linux-kernel On Wed, 2012-07-04 at 14:00 +0800, Yan, Zheng wrote: > +static void nhmex_uncore_msr_enable_event(struct intel_uncore_box *box, > + struct perf_event *event) > +{ > + struct hw_perf_event *hwc = &event->hw; > + > + if (box->pmu->type == &nhmex_uncore_bbox) > + nhmex_bbox_msr_enable_event(box, event); > + else if (box->pmu->type == &nhmex_uncore_sbox) > + nhmex_sbox_msr_enable_event(box, event); > + else if (box->pmu->type == &nhmex_uncore_mbox) > + nhmex_mbox_msr_enable_event(box, event); > + else if (box->pmu->type == &nhmex_uncore_rbox) > + nhmex_rbox_msr_enable_event(box, event); > + else if (hwc->idx >= UNCORE_PMC_IDX_FIXED) > + wrmsrl(hwc->config_base, NHMEX_PMON_CTL_EN_BIT0); > + else if (box->pmu->type->event_mask & NHMEX_PMON_CTL_EN_BIT0) > + wrmsrl(hwc->config_base, hwc->config | NHMEX_PMON_CTL_EN_BIT22); > + else > + wrmsrl(hwc->config_base, hwc->config | NHMEX_PMON_CTL_EN_BIT0); > +} wouldn't it be easier to do something like: box->pmu->type->enable_event(box, event); The same for these other functions that are massive ->type switches. Also, can you please add more comments, note all the face_cpuc stuff, and there's some rather dense code in all the alternative stuff. Also, how can a single extra register require 192 bits of config space? ^ permalink raw reply [flat|nested] 21+ messages in thread
* Re: [PATCH 5/5] perf/x86: Add Intel Nehalem-EX uncore support 2012-07-04 10:04 ` Peter Zijlstra @ 2012-07-04 16:00 ` Yan, Zheng 2012-07-04 16:53 ` Peter Zijlstra 2012-07-05 6:32 ` [PATCH V2 " Yan, Zheng 1 sibling, 1 reply; 21+ messages in thread From: Yan, Zheng @ 2012-07-04 16:00 UTC (permalink / raw) To: Peter Zijlstra; +Cc: eranian, mingo, andi, linux-kernel On 07/04/2012 06:04 PM, Peter Zijlstra wrote: > On Wed, 2012-07-04 at 14:00 +0800, Yan, Zheng wrote: >> +static void nhmex_uncore_msr_enable_event(struct intel_uncore_box *box, >> + struct perf_event *event) >> +{ >> + struct hw_perf_event *hwc = &event->hw; >> + >> + if (box->pmu->type == &nhmex_uncore_bbox) >> + nhmex_bbox_msr_enable_event(box, event); >> + else if (box->pmu->type == &nhmex_uncore_sbox) >> + nhmex_sbox_msr_enable_event(box, event); >> + else if (box->pmu->type == &nhmex_uncore_mbox) >> + nhmex_mbox_msr_enable_event(box, event); >> + else if (box->pmu->type == &nhmex_uncore_rbox) >> + nhmex_rbox_msr_enable_event(box, event); >> + else if (hwc->idx >= UNCORE_PMC_IDX_FIXED) >> + wrmsrl(hwc->config_base, NHMEX_PMON_CTL_EN_BIT0); >> + else if (box->pmu->type->event_mask & NHMEX_PMON_CTL_EN_BIT0) >> + wrmsrl(hwc->config_base, hwc->config | NHMEX_PMON_CTL_EN_BIT22); >> + else >> + wrmsrl(hwc->config_base, hwc->config | NHMEX_PMON_CTL_EN_BIT0); >> +} > > wouldn't it be easier to do something like: > > box->pmu->type->enable_event(box, event); > Ok, I will update the patch. > The same for these other functions that are massive ->type switches. > > Also, can you please add more comments, note all the face_cpuc stuff, > and there's some rather dense code in all the alternative stuff. > OK > Also, how can a single extra register require 192 bits of config space? > Some events in the M-Box and R-Box require programming up to 3 extra registers. For example, the event 0x5 in the R-Box requires programming extra registers XBR_SET1_MM_CFG, XBR_SET1_MATCH and XBR_SET1_MATCH. Regards Yan, Zheng ^ permalink raw reply [flat|nested] 21+ messages in thread
* Re: [PATCH 5/5] perf/x86: Add Intel Nehalem-EX uncore support 2012-07-04 16:00 ` Yan, Zheng @ 2012-07-04 16:53 ` Peter Zijlstra 2012-07-04 17:46 ` Yan, Zheng 0 siblings, 1 reply; 21+ messages in thread From: Peter Zijlstra @ 2012-07-04 16:53 UTC (permalink / raw) To: Yan, Zheng; +Cc: eranian, mingo, andi, linux-kernel On Thu, 2012-07-05 at 00:00 +0800, Yan, Zheng wrote: > > > Also, how can a single extra register require 192 bits of config space? > > > Some events in the M-Box and R-Box require programming up to 3 extra > registers. For example, the event 0x5 in the R-Box requires programming > extra registers XBR_SET1_MM_CFG, XBR_SET1_MATCH and XBR_SET1_MATCH. But shouldn't you then use 3 extra_reg instead of blowing up the one? ^ permalink raw reply [flat|nested] 21+ messages in thread
* Re: [PATCH 5/5] perf/x86: Add Intel Nehalem-EX uncore support 2012-07-04 16:53 ` Peter Zijlstra @ 2012-07-04 17:46 ` Yan, Zheng 0 siblings, 0 replies; 21+ messages in thread From: Yan, Zheng @ 2012-07-04 17:46 UTC (permalink / raw) To: Peter Zijlstra; +Cc: Yan, Zheng, eranian, mingo, andi, linux-kernel On Thu, Jul 5, 2012 at 12:53 AM, Peter Zijlstra <a.p.zijlstra@chello.nl> wrote: > On Thu, 2012-07-05 at 00:00 +0800, Yan, Zheng wrote: >> >> > Also, how can a single extra register require 192 bits of config space? >> > >> Some events in the M-Box and R-Box require programming up to 3 extra >> registers. For example, the event 0x5 in the R-Box requires programming >> extra registers XBR_SET1_MM_CFG, XBR_SET1_MATCH and XBR_SET1_MATCH. > > But shouldn't you then use 3 extra_reg instead of blowing up the one? > These 3 registers are within a logical set, are always used at the same time. I think it's natural to use a single data structure to track them. Regards Yan, Zheng ^ permalink raw reply [flat|nested] 21+ messages in thread
* [PATCH V2 5/5] perf/x86: Add Intel Nehalem-EX uncore support 2012-07-04 10:04 ` Peter Zijlstra 2012-07-04 16:00 ` Yan, Zheng @ 2012-07-05 6:32 ` Yan, Zheng 2012-07-05 14:51 ` Andi Kleen ` (2 more replies) 1 sibling, 3 replies; 21+ messages in thread From: Yan, Zheng @ 2012-07-05 6:32 UTC (permalink / raw) To: Peter Zijlstra, eranian, mingo, andi, linux-kernel The uncore subsystem in Nehalem-EX consists of 7 components (U-Box, C-Box, B-Box, S-Box, R-Box, M-Box and W-Box). This patch is large because the way to program these boxes is diverse. Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com> --- arch/x86/kernel/cpu/perf_event_intel_uncore.c | 1191 +++++++++++++++++++++++-- arch/x86/kernel/cpu/perf_event_intel_uncore.h | 203 ++++- 2 files changed, 1322 insertions(+), 72 deletions(-) diff --git a/arch/x86/kernel/cpu/perf_event_intel_uncore.c b/arch/x86/kernel/cpu/perf_event_intel_uncore.c index c0a80ff..f3b1e01 100644 --- a/arch/x86/kernel/cpu/perf_event_intel_uncore.c +++ b/arch/x86/kernel/cpu/perf_event_intel_uncore.c @@ -38,6 +38,77 @@ DEFINE_UNCORE_FORMAT_ATTR(filter_brand1, filter_brand1, "config1:8-15"); DEFINE_UNCORE_FORMAT_ATTR(filter_brand2, filter_brand2, "config1:16-23"); DEFINE_UNCORE_FORMAT_ATTR(filter_brand3, filter_brand3, "config1:24-31"); +static u64 uncore_msr_read_counter(struct intel_uncore_box *box, + struct perf_event *event) +{ + u64 count; + rdmsrl(event->hw.event_base, count); + return count; +} + +/* + * generic get constraint function for shared match/mask registers. + */ +static struct event_constraint * +uncore_get_constraint(struct intel_uncore_box *box, struct perf_event *event) +{ + struct intel_uncore_extra_reg *er; + struct hw_perf_event_extra *reg1 = &event->hw.extra_reg; + struct hw_perf_event_extra *reg2 = &event->hw.branch_reg; + unsigned long flags; + bool ok = false; + + /* + * reg->alloc can be set due to existing state, so for fake box we + * need to ignore this, otherwise we might fail to allocate proper + * fake state for this extra reg constraint. + */ + if (reg1->idx == EXTRA_REG_NONE || + (!uncore_box_is_fake(box) && reg1->alloc)) + return NULL; + + er = &box->shared_regs[reg1->idx]; + raw_spin_lock_irqsave(&er->lock, flags); + if (!atomic_read(&er->ref) || + (er->config1 == reg1->config && er->config2 == reg2->config)) { + atomic_inc(&er->ref); + er->config1 = reg1->config; + er->config2 = reg2->config; + ok = true; + } + raw_spin_unlock_irqrestore(&er->lock, flags); + + if (ok) { + if (!uncore_box_is_fake(box)) + reg1->alloc = 1; + return NULL; + } + + return &constraint_empty; +} + +static void uncore_put_constraint(struct intel_uncore_box *box, + struct perf_event *event) +{ + struct intel_uncore_extra_reg *er; + struct hw_perf_event_extra *reg1 = &event->hw.extra_reg; + + /* + * Only put constraint if extra reg was actually allocated. Also + * takes care of event which do not use an extra shared reg. + * + * Also, if this is a fake box we shouldn't touch any event state + * (reg->alloc) and we don't care about leaving inconsistent box + * state either since it will be thrown out. + */ + if (uncore_box_is_fake(box) || !reg1->alloc) + return; + + er = &box->shared_regs[reg1->idx]; + atomic_dec(&er->ref); + reg1->alloc = 0; +} + /* Sandy Bridge-EP uncore support */ static struct intel_uncore_type snbep_uncore_cbox; static struct intel_uncore_type snbep_uncore_pcu; @@ -150,16 +221,6 @@ static void snbep_uncore_msr_disable_event(struct intel_uncore_box *box, wrmsrl(hwc->config_base, hwc->config); } -static u64 snbep_uncore_msr_read_counter(struct intel_uncore_box *box, - struct perf_event *event) -{ - struct hw_perf_event *hwc = &event->hw; - u64 count; - - rdmsrl(hwc->event_base, count); - return count; -} - static void snbep_uncore_msr_init_box(struct intel_uncore_box *box) { unsigned msr = uncore_msr_box_ctl(box); @@ -167,49 +228,6 @@ static void snbep_uncore_msr_init_box(struct intel_uncore_box *box) wrmsrl(msr, SNBEP_PMON_BOX_CTL_INT); } -static struct event_constraint * -snbep_uncore_get_constraint(struct intel_uncore_box *box, - struct perf_event *event) -{ - struct intel_uncore_extra_reg *er; - struct hw_perf_event_extra *reg1 = &event->hw.extra_reg; - unsigned long flags; - bool ok = false; - - if (reg1->idx == EXTRA_REG_NONE || (box->phys_id >= 0 && reg1->alloc)) - return NULL; - - er = &box->shared_regs[reg1->idx]; - raw_spin_lock_irqsave(&er->lock, flags); - if (!atomic_read(&er->ref) || er->config1 == reg1->config) { - atomic_inc(&er->ref); - er->config1 = reg1->config; - ok = true; - } - raw_spin_unlock_irqrestore(&er->lock, flags); - - if (ok) { - if (box->phys_id >= 0) - reg1->alloc = 1; - return NULL; - } - return &constraint_empty; -} - -static void snbep_uncore_put_constraint(struct intel_uncore_box *box, - struct perf_event *event) -{ - struct intel_uncore_extra_reg *er; - struct hw_perf_event_extra *reg1 = &event->hw.extra_reg; - - if (box->phys_id < 0 || !reg1->alloc) - return; - - er = &box->shared_regs[reg1->idx]; - atomic_dec(&er->ref); - reg1->alloc = 0; -} - static int snbep_uncore_hw_config(struct intel_uncore_box *box, struct perf_event *event) { @@ -320,9 +338,9 @@ static struct intel_uncore_ops snbep_uncore_msr_ops = { .enable_box = snbep_uncore_msr_enable_box, .disable_event = snbep_uncore_msr_disable_event, .enable_event = snbep_uncore_msr_enable_event, - .read_counter = snbep_uncore_msr_read_counter, - .get_constraint = snbep_uncore_get_constraint, - .put_constraint = snbep_uncore_put_constraint, + .read_counter = uncore_msr_read_counter, + .get_constraint = uncore_get_constraint, + .put_constraint = uncore_put_constraint, .hw_config = snbep_uncore_hw_config, }; @@ -603,7 +621,6 @@ static void snbep_pci2phy_map_init(void) } /* end of Sandy Bridge-EP uncore support */ - /* Sandy Bridge uncore support */ static void snb_uncore_msr_enable_event(struct intel_uncore_box *box, struct perf_event *event) @@ -622,14 +639,6 @@ static void snb_uncore_msr_disable_event(struct intel_uncore_box *box, wrmsrl(event->hw.config_base, 0); } -static u64 snb_uncore_msr_read_counter(struct intel_uncore_box *box, - struct perf_event *event) -{ - u64 count; - rdmsrl(event->hw.event_base, count); - return count; -} - static void snb_uncore_msr_init_box(struct intel_uncore_box *box) { if (box->pmu->pmu_idx == 0) { @@ -656,7 +665,7 @@ static struct intel_uncore_ops snb_uncore_msr_ops = { .init_box = snb_uncore_msr_init_box, .disable_event = snb_uncore_msr_disable_event, .enable_event = snb_uncore_msr_enable_event, - .read_counter = snb_uncore_msr_read_counter, + .read_counter = uncore_msr_read_counter, }; static struct event_constraint snb_uncore_cbox_constraints[] = { @@ -744,7 +753,7 @@ static struct intel_uncore_ops nhm_uncore_msr_ops = { .enable_box = nhm_uncore_msr_enable_box, .disable_event = snb_uncore_msr_disable_event, .enable_event = nhm_uncore_msr_enable_event, - .read_counter = snb_uncore_msr_read_counter, + .read_counter = uncore_msr_read_counter, }; static struct intel_uncore_type nhm_uncore = { @@ -769,6 +778,1051 @@ static struct intel_uncore_type *nhm_msr_uncores[] = { }; /* end of Nehalem uncore support */ +/* Nehalem-EX uncore support */ +#define __BITS_VALUE(x, i, n) ((typeof(x))(((x) >> ((i) * (n))) & \ + ((1ULL << (n)) - 1))) + +DEFINE_UNCORE_FORMAT_ATTR(event5, event, "config:1-5"); +DEFINE_UNCORE_FORMAT_ATTR(counter, counter, "config:6-7"); +DEFINE_UNCORE_FORMAT_ATTR(mm_cfg, mm_cfg, "config:63"); +DEFINE_UNCORE_FORMAT_ATTR(match, match, "config1:0-63"); +DEFINE_UNCORE_FORMAT_ATTR(mask, mask, "config2:0-63"); + +static void nhmex_uncore_msr_init_box(struct intel_uncore_box *box) +{ + wrmsrl(NHMEX_U_MSR_PMON_GLOBAL_CTL, NHMEX_U_PMON_GLOBAL_EN_ALL); +} + +static void nhmex_uncore_msr_disable_box(struct intel_uncore_box *box) +{ + unsigned msr = uncore_msr_box_ctl(box); + u64 config; + if (msr) { + rdmsrl(msr, config); + config &= ~((1ULL << uncore_num_counters(box)) - 1); + /* WBox has a fixed counter */ + if (uncore_msr_fixed_ctl(box)) + config &= ~NHMEX_W_PMON_GLOBAL_FIXED_EN; + wrmsrl(msr, config); + } +} + +static void nhmex_uncore_msr_enable_box(struct intel_uncore_box *box) +{ + unsigned msr = uncore_msr_box_ctl(box); + u64 config; + if (msr) { + rdmsrl(msr, config); + config |= (1ULL << uncore_num_counters(box)) - 1; + /* WBox has a fixed counter */ + if (uncore_msr_fixed_ctl(box)) + config |= NHMEX_W_PMON_GLOBAL_FIXED_EN; + wrmsrl(msr, config); + } +} + +static void nhmex_uncore_msr_disable_event(struct intel_uncore_box *box, + struct perf_event *event) +{ + wrmsrl(event->hw.config_base, 0); +} + +static void nhmex_uncore_msr_enable_event(struct intel_uncore_box *box, + struct perf_event *event) +{ + struct hw_perf_event *hwc = &event->hw; + + if (hwc->idx >= UNCORE_PMC_IDX_FIXED) + wrmsrl(hwc->config_base, NHMEX_PMON_CTL_EN_BIT0); + else if (box->pmu->type->event_mask & NHMEX_PMON_CTL_EN_BIT0) + wrmsrl(hwc->config_base, hwc->config | NHMEX_PMON_CTL_EN_BIT22); + else + wrmsrl(hwc->config_base, hwc->config | NHMEX_PMON_CTL_EN_BIT0); +} + +#define NHMEX_UNCORE_OPS_COMMON_INIT() \ + .init_box = nhmex_uncore_msr_init_box, \ + .disable_box = nhmex_uncore_msr_disable_box, \ + .enable_box = nhmex_uncore_msr_enable_box, \ + .disable_event = nhmex_uncore_msr_disable_event, \ + .read_counter = uncore_msr_read_counter + +static struct intel_uncore_ops nhmex_uncore_ops = { + NHMEX_UNCORE_OPS_COMMON_INIT(), + .enable_event = nhmex_uncore_msr_enable_event, +}; + +static struct attribute *nhmex_uncore_ubox_formats_attr[] = { + &format_attr_event.attr, + &format_attr_edge.attr, + NULL, +}; + +static struct attribute_group nhmex_uncore_ubox_format_group = { + .name = "format", + .attrs = nhmex_uncore_ubox_formats_attr, +}; + +static struct intel_uncore_type nhmex_uncore_ubox = { + .name = "ubox", + .num_counters = 1, + .num_boxes = 1, + .perf_ctr_bits = 48, + .event_ctl = NHMEX_U_MSR_PMON_EV_SEL, + .perf_ctr = NHMEX_U_MSR_PMON_CTR, + .event_mask = NHMEX_U_PMON_RAW_EVENT_MASK, + .box_ctl = NHMEX_U_MSR_PMON_GLOBAL_CTL, + .ops = &nhmex_uncore_ops, + .format_group = &nhmex_uncore_ubox_format_group +}; + +static struct attribute *nhmex_uncore_cbox_formats_attr[] = { + &format_attr_event.attr, + &format_attr_umask.attr, + &format_attr_edge.attr, + &format_attr_inv.attr, + &format_attr_thresh8.attr, + NULL, +}; + +static struct attribute_group nhmex_uncore_cbox_format_group = { + .name = "format", + .attrs = nhmex_uncore_cbox_formats_attr, +}; + +static struct intel_uncore_type nhmex_uncore_cbox = { + .name = "cbox", + .num_counters = 6, + .num_boxes = 8, + .perf_ctr_bits = 48, + .event_ctl = NHMEX_C0_MSR_PMON_EV_SEL0, + .perf_ctr = NHMEX_C0_MSR_PMON_CTR0, + .event_mask = NHMEX_PMON_RAW_EVENT_MASK, + .box_ctl = NHMEX_C0_MSR_PMON_GLOBAL_CTL, + .msr_offset = NHMEX_C_MSR_OFFSET, + .pair_ctr_ctl = 1, + .ops = &nhmex_uncore_ops, + .format_group = &nhmex_uncore_cbox_format_group +}; + +static struct uncore_event_desc nhmex_uncore_wbox_events[] = { + INTEL_UNCORE_EVENT_DESC(clockticks, "event=0xff,umask=0"), + { /* end: all zeroes */ }, +}; + +static struct intel_uncore_type nhmex_uncore_wbox = { + .name = "wbox", + .num_counters = 4, + .num_boxes = 1, + .perf_ctr_bits = 48, + .event_ctl = NHMEX_W_MSR_PMON_CNT0, + .perf_ctr = NHMEX_W_MSR_PMON_EVT_SEL0, + .fixed_ctr = NHMEX_W_MSR_PMON_FIXED_CTR, + .fixed_ctl = NHMEX_W_MSR_PMON_FIXED_CTL, + .event_mask = NHMEX_PMON_RAW_EVENT_MASK, + .box_ctl = NHMEX_W_MSR_GLOBAL_CTL, + .pair_ctr_ctl = 1, + .event_descs = nhmex_uncore_wbox_events, + .ops = &nhmex_uncore_ops, + .format_group = &nhmex_uncore_cbox_format_group +}; + +static int nhmex_bbox_hw_config(struct intel_uncore_box *box, + struct perf_event *event) +{ + struct hw_perf_event *hwc = &event->hw; + struct hw_perf_event_extra *reg1 = &hwc->extra_reg; + struct hw_perf_event_extra *reg2 = &hwc->branch_reg; + int ctr, ev_sel; + + ctr = (hwc->config & NHMEX_B_PMON_CTR_MASK) >> + NHMEX_B_PMON_CTR_SHIFT; + ev_sel = (hwc->config & NHMEX_B_PMON_CTL_EV_SEL_MASK) >> + NHMEX_B_PMON_CTL_EV_SEL_SHIFT; + + /* events that do not use the match/mask registers */ + if ((ctr == 0 && ev_sel > 0x3) || (ctr == 1 && ev_sel > 0x6) || + (ctr == 2 && ev_sel != 0x4) || ctr == 3) + return 0; + + if (box->pmu->pmu_idx == 0) + reg1->reg = NHMEX_B0_MSR_MATCH; + else + reg1->reg = NHMEX_B1_MSR_MATCH; + reg1->idx = 0; + reg1->config = event->attr.config1; + reg2->config = event->attr.config2; + return 0; +} + +static void nhmex_bbox_msr_enable_event(struct intel_uncore_box *box, + struct perf_event *event) +{ + struct hw_perf_event *hwc = &event->hw; + struct hw_perf_event_extra *reg1 = &hwc->extra_reg; + struct hw_perf_event_extra *reg2 = &hwc->branch_reg; + + if (reg1->idx != EXTRA_REG_NONE) { + wrmsrl(reg1->reg, reg1->config); + wrmsrl(reg1->reg + 1, reg2->config); + } + wrmsrl(hwc->config_base, NHMEX_PMON_CTL_EN_BIT0 | + (hwc->config & NHMEX_B_PMON_CTL_EV_SEL_MASK)); +} + +/* + * The Bbox has 4 counters, but each counter monitors different events. + * Use bits 6-7 in the event config to select counter. + */ +static struct event_constraint nhmex_uncore_bbox_constraints[] = { + EVENT_CONSTRAINT(0 , 1, 0xc0), + EVENT_CONSTRAINT(0x40, 2, 0xc0), + EVENT_CONSTRAINT(0x80, 4, 0xc0), + EVENT_CONSTRAINT(0xc0, 8, 0xc0), + EVENT_CONSTRAINT_END, +}; + +static struct attribute *nhmex_uncore_bbox_formats_attr[] = { + &format_attr_event5.attr, + &format_attr_counter.attr, + &format_attr_match.attr, + &format_attr_mask.attr, + NULL, +}; + +static struct attribute_group nhmex_uncore_bbox_format_group = { + .name = "format", + .attrs = nhmex_uncore_bbox_formats_attr, +}; + +static struct intel_uncore_ops nhmex_uncore_bbox_ops = { + NHMEX_UNCORE_OPS_COMMON_INIT(), + .enable_event = nhmex_bbox_msr_enable_event, + .hw_config = nhmex_bbox_hw_config, + .get_constraint = uncore_get_constraint, + .put_constraint = uncore_put_constraint, +}; + +static struct intel_uncore_type nhmex_uncore_bbox = { + .name = "bbox", + .num_counters = 4, + .num_boxes = 2, + .perf_ctr_bits = 48, + .event_ctl = NHMEX_B0_MSR_PMON_CTL0, + .perf_ctr = NHMEX_B0_MSR_PMON_CTR0, + .event_mask = NHMEX_B_PMON_RAW_EVENT_MASK, + .box_ctl = NHMEX_B0_MSR_PMON_GLOBAL_CTL, + .msr_offset = NHMEX_B_MSR_OFFSET, + .pair_ctr_ctl = 1, + .num_shared_regs = 1, + .constraints = nhmex_uncore_bbox_constraints, + .ops = &nhmex_uncore_bbox_ops, + .format_group = &nhmex_uncore_bbox_format_group +}; + +static int nhmex_sbox_hw_config(struct intel_uncore_box *box, + struct perf_event *event) +{ + struct hw_perf_event_extra *reg1 = &event->hw.extra_reg; + struct hw_perf_event_extra *reg2 = &event->hw.branch_reg; + + if (event->attr.config & NHMEX_S_PMON_MM_CFG_EN) { + reg1->config = event->attr.config1; + reg2->config = event->attr.config2; + } else { + reg1->config = ~0ULL; + reg2->config = ~0ULL; + } + if (box->pmu->pmu_idx == 0) + reg1->reg = NHMEX_S0_MSR_MM_CFG; + else + reg1->reg = NHMEX_S1_MSR_MM_CFG; + reg1->idx = 0; + return 0; +} + +static void nhmex_sbox_msr_enable_event(struct intel_uncore_box *box, + struct perf_event *event) +{ + struct hw_perf_event *hwc = &event->hw; + struct hw_perf_event_extra *reg1 = &hwc->extra_reg; + struct hw_perf_event_extra *reg2 = &hwc->branch_reg; + + wrmsrl(reg1->reg, 0); + if (reg1->config != ~0ULL || reg2->config != ~0ULL) { + wrmsrl(reg1->reg + 1, reg1->config); + wrmsrl(reg1->reg + 2, reg2->config); + wrmsrl(reg1->reg, NHMEX_S_PMON_MM_CFG_EN); + } + wrmsrl(hwc->config_base, hwc->config | NHMEX_PMON_CTL_EN_BIT22); +} + +static struct attribute *nhmex_uncore_sbox_formats_attr[] = { + &format_attr_event.attr, + &format_attr_umask.attr, + &format_attr_edge.attr, + &format_attr_inv.attr, + &format_attr_thresh8.attr, + &format_attr_mm_cfg.attr, + &format_attr_match.attr, + &format_attr_mask.attr, + NULL, +}; + +static struct attribute_group nhmex_uncore_sbox_format_group = { + .name = "format", + .attrs = nhmex_uncore_sbox_formats_attr, +}; + +static struct intel_uncore_ops nhmex_uncore_sbox_ops = { + NHMEX_UNCORE_OPS_COMMON_INIT(), + .enable_event = nhmex_sbox_msr_enable_event, + .hw_config = nhmex_sbox_hw_config, + .get_constraint = uncore_get_constraint, + .put_constraint = uncore_put_constraint, +}; + +static struct intel_uncore_type nhmex_uncore_sbox = { + .name = "sbox", + .num_counters = 4, + .num_boxes = 2, + .perf_ctr_bits = 48, + .event_ctl = NHMEX_S0_MSR_PMON_CTL0, + .perf_ctr = NHMEX_S0_MSR_PMON_CTR0, + .event_mask = NHMEX_PMON_RAW_EVENT_MASK, + .box_ctl = NHMEX_S0_MSR_PMON_GLOBAL_CTL, + .msr_offset = NHMEX_S_MSR_OFFSET, + .pair_ctr_ctl = 1, + .num_shared_regs = 1, + .ops = &nhmex_uncore_sbox_ops, + .format_group = &nhmex_uncore_sbox_format_group +}; + +enum { + EXTRA_REG_NHMEX_M_FILTER, + EXTRA_REG_NHMEX_M_DSP, + EXTRA_REG_NHMEX_M_ISS, + EXTRA_REG_NHMEX_M_MAP, + EXTRA_REG_NHMEX_M_MSC_THR, + EXTRA_REG_NHMEX_M_PGT, + EXTRA_REG_NHMEX_M_PLD, + EXTRA_REG_NHMEX_M_ZDP_CTL_FVC, +}; + +static struct extra_reg nhmex_uncore_mbox_extra_regs[] = { + MBOX_INC_SEL_EXTAR_REG(0x0, DSP), + MBOX_INC_SEL_EXTAR_REG(0x4, MSC_THR), + MBOX_INC_SEL_EXTAR_REG(0x5, MSC_THR), + MBOX_INC_SEL_EXTAR_REG(0x9, ISS), + /* event 0xa uses two extra registers */ + MBOX_INC_SEL_EXTAR_REG(0xa, ISS), + MBOX_INC_SEL_EXTAR_REG(0xa, PLD), + MBOX_INC_SEL_EXTAR_REG(0xb, PLD), + /* events 0xd ~ 0x10 use the same extra register */ + MBOX_INC_SEL_EXTAR_REG(0xd, ZDP_CTL_FVC), + MBOX_INC_SEL_EXTAR_REG(0xe, ZDP_CTL_FVC), + MBOX_INC_SEL_EXTAR_REG(0xf, ZDP_CTL_FVC), + MBOX_INC_SEL_EXTAR_REG(0x10, ZDP_CTL_FVC), + MBOX_INC_SEL_EXTAR_REG(0x16, PGT), + MBOX_SET_FLAG_SEL_EXTRA_REG(0x0, DSP), + MBOX_SET_FLAG_SEL_EXTRA_REG(0x1, ISS), + MBOX_SET_FLAG_SEL_EXTRA_REG(0x5, PGT), + MBOX_SET_FLAG_SEL_EXTRA_REG(0x6, MAP), + EVENT_EXTRA_END +}; + +static bool nhmex_mbox_get_shared_reg(struct intel_uncore_box *box, + int idx, u64 config) +{ + struct intel_uncore_extra_reg *er; + unsigned long flags; + bool ret = false; + u64 mask; + + if (idx < EXTRA_REG_NHMEX_M_ZDP_CTL_FVC) { + er = &box->shared_regs[idx]; + raw_spin_lock_irqsave(&er->lock, flags); + if (!atomic_read(&er->ref) || er->config == config) { + atomic_inc(&er->ref); + er->config = config; + ret = true; + } + raw_spin_unlock_irqrestore(&er->lock, flags); + return ret; + } + /* + * The ZDP_CTL_FVC MSR has 4 fields which are used to control + * events 0xd ~ 0x10. Besides these 4 fields, there are additional + * fields which are shared. + */ + idx -= EXTRA_REG_NHMEX_M_ZDP_CTL_FVC; + if (WARN_ON_ONCE(idx >= 4)) + return false; + + /* mask of the shared fields */ + mask = NHMEX_M_PMON_ZDP_CTL_FVC_MASK; + er = &box->shared_regs[EXTRA_REG_NHMEX_M_ZDP_CTL_FVC]; + raw_spin_lock_irqsave(&er->lock, flags); + /* add mask of the non-shared field if it's in use */ + if (__BITS_VALUE(atomic_read(&er->ref), idx, 8)) + mask |= NHMEX_M_PMON_ZDP_CTL_FVC_EVENT_MASK(idx); + + if (!atomic_read(&er->ref) || !((er->config ^ config) & mask)) { + atomic_add(1 << (idx * 8), &er->ref); + mask = NHMEX_M_PMON_ZDP_CTL_FVC_MASK | + NHMEX_M_PMON_ZDP_CTL_FVC_EVENT_MASK(idx); + er->config &= ~mask; + er->config |= (config & mask); + ret = true; + } + raw_spin_unlock_irqrestore(&er->lock, flags); + return ret; +} + +static void nhmex_mbox_put_shared_reg(struct intel_uncore_box *box, int idx) +{ + struct intel_uncore_extra_reg *er; + + if (idx < EXTRA_REG_NHMEX_M_ZDP_CTL_FVC) { + er = &box->shared_regs[idx]; + atomic_dec(&er->ref); + return; + } + + idx -= EXTRA_REG_NHMEX_M_ZDP_CTL_FVC; + er = &box->shared_regs[EXTRA_REG_NHMEX_M_ZDP_CTL_FVC]; + atomic_sub(1 << (idx * 8), &er->ref); +} + +u64 nhmex_mbox_alter_er(struct perf_event *event, int new_idx, bool modify) +{ + struct hw_perf_event *hwc = &event->hw; + struct hw_perf_event_extra *reg1 = &hwc->extra_reg; + int idx, orig_idx = __BITS_VALUE(reg1->idx, 0, 8); + u64 config = reg1->config; + + /* get the non-shared control bits and shift them */ + idx = orig_idx - EXTRA_REG_NHMEX_M_ZDP_CTL_FVC; + config &= NHMEX_M_PMON_ZDP_CTL_FVC_EVENT_MASK(idx); + if (new_idx > orig_idx) { + idx = new_idx - orig_idx; + config <<= 3 * idx; + } else { + idx = orig_idx - new_idx; + config >>= 3 * idx; + } + /* add the shared control bits back */ + config |= NHMEX_M_PMON_ZDP_CTL_FVC_MASK & reg1->config; + if (modify) { + /* adjust the main event selector */ + if (new_idx > orig_idx) + hwc->config += idx << NHMEX_M_PMON_CTL_INC_SEL_SHIFT; + else + hwc->config -= idx << NHMEX_M_PMON_CTL_INC_SEL_SHIFT; + reg1->config = config; + reg1->idx = ~0xff | new_idx; + } + return config; +} + +static struct event_constraint * +nhmex_mbox_get_constraint(struct intel_uncore_box *box, + struct perf_event *event) +{ + struct hw_perf_event_extra *reg1 = &event->hw.extra_reg; + struct hw_perf_event_extra *reg2 = &event->hw.branch_reg; + int i, idx[2], alloc = 0; + u64 config1 = reg1->config; + + idx[0] = __BITS_VALUE(reg1->idx, 0, 8); + idx[1] = __BITS_VALUE(reg1->idx, 1, 8); +again: + for (i = 0; i < 2; i++) { + if (!uncore_box_is_fake(box) && (reg1->alloc & (0x1 << i))) + idx[i] = 0xff; + + if (idx[i] == 0xff) + continue; + + if (!nhmex_mbox_get_shared_reg(box, idx[i], + __BITS_VALUE(config1, i, 32))) + goto fail; + alloc |= (0x1 << i); + } + + /* for the match/mask registers */ + if ((uncore_box_is_fake(box) || !reg2->alloc) && + !nhmex_mbox_get_shared_reg(box, reg2->idx, reg2->config)) + goto fail; + + /* + * If it's a fake box -- as per validate_{group,event}() we + * shouldn't touch event state and we can avoid doing so + * since both will only call get_event_constraints() once + * on each event, this avoids the need for reg->alloc. + */ + if (!uncore_box_is_fake(box)) { + if (idx[0] != 0xff && idx[0] != __BITS_VALUE(reg1->idx, 0, 8)) + nhmex_mbox_alter_er(event, idx[0], true); + reg1->alloc |= alloc; + reg2->alloc = 1; + } + return NULL; +fail: + if (idx[0] != 0xff && !(alloc & 0x1) && + idx[0] >= EXTRA_REG_NHMEX_M_ZDP_CTL_FVC) { + /* + * events 0xd ~ 0x10 are functional identical, but are + * controlled by different fields in the ZDP_CTL_FVC + * register. If we failed to take one field, try the + * rest 3 choices. + */ + BUG_ON(__BITS_VALUE(reg1->idx, 1, 8) != 0xff); + idx[0] -= EXTRA_REG_NHMEX_M_ZDP_CTL_FVC; + idx[0] = (idx[0] + 1) % 4; + idx[0] += EXTRA_REG_NHMEX_M_ZDP_CTL_FVC; + if (idx[0] != __BITS_VALUE(reg1->idx, 0, 8)) { + config1 = nhmex_mbox_alter_er(event, idx[0], false); + goto again; + } + } + + if (alloc & 0x1) + nhmex_mbox_put_shared_reg(box, idx[0]); + if (alloc & 0x2) + nhmex_mbox_put_shared_reg(box, idx[1]); + return &constraint_empty; +} + +static void nhmex_mbox_put_constraint(struct intel_uncore_box *box, + struct perf_event *event) +{ + struct hw_perf_event_extra *reg1 = &event->hw.extra_reg; + struct hw_perf_event_extra *reg2 = &event->hw.branch_reg; + + if (uncore_box_is_fake(box)) + return; + + if (reg1->alloc & 0x1) + nhmex_mbox_put_shared_reg(box, __BITS_VALUE(reg1->idx, 0, 8)); + if (reg1->alloc & 0x2) + nhmex_mbox_put_shared_reg(box, __BITS_VALUE(reg1->idx, 1, 8)); + reg1->alloc = 0; + + if (reg2->alloc) { + nhmex_mbox_put_shared_reg(box, reg2->idx); + reg2->alloc = 0; + } +} + +static int nhmex_mbox_extra_reg_idx(struct extra_reg *er) +{ + + if (er->idx < EXTRA_REG_NHMEX_M_ZDP_CTL_FVC) + return er->idx; + return er->idx + (er->event >> NHMEX_M_PMON_CTL_INC_SEL_SHIFT) - 0xd; +} + +static int nhmex_mbox_hw_config(struct intel_uncore_box *box, + struct perf_event *event) +{ + struct intel_uncore_type *type = box->pmu->type; + struct hw_perf_event_extra *reg1 = &event->hw.extra_reg; + struct hw_perf_event_extra *reg2 = &event->hw.branch_reg; + struct extra_reg *er; + unsigned msr; + int reg_idx = 0; + + if (WARN_ON_ONCE(reg1->idx != -1)) + return -EINVAL; + /* + * The mbox events may require 2 extra MSRs at the most. But only + * the lower 32 bits in these MSRs are significant, so we can use + * config1 to pass two MSRs' config. + */ + for (er = nhmex_uncore_mbox_extra_regs; er->msr; er++) { + if (er->event != (event->hw.config & er->config_mask)) + continue; + if (event->attr.config1 & ~er->valid_mask) + return -EINVAL; + if (er->idx == __BITS_VALUE(reg1->idx, 0, 8) || + er->idx == __BITS_VALUE(reg1->idx, 1, 8)) + continue; + if (WARN_ON_ONCE(reg_idx >= 2)) + return -EINVAL; + + msr = er->msr + type->msr_offset * box->pmu->pmu_idx; + if (WARN_ON_ONCE(msr >= 0xffff || er->idx >= 0xff)) + return -EINVAL; + + /* always use the 32~63 bits to pass the PLD config */ + if (er->idx == EXTRA_REG_NHMEX_M_PLD) + reg_idx = 1; + + reg1->idx &= ~(0xff << (reg_idx * 8)); + reg1->reg &= ~(0xffff << (reg_idx * 16)); + reg1->idx |= nhmex_mbox_extra_reg_idx(er) << (reg_idx * 8); + reg1->reg |= msr << (reg_idx * 16); + reg1->config = event->attr.config1; + reg_idx++; + } + /* use config2 to pass the filter config */ + reg2->idx = EXTRA_REG_NHMEX_M_FILTER; + if (event->attr.config2 & NHMEX_M_PMON_MM_CFG_EN) + reg2->config = event->attr.config2; + else + reg2->config = ~0ULL; + if (box->pmu->pmu_idx == 0) + reg2->reg = NHMEX_M0_MSR_PMU_MM_CFG; + else + reg2->reg = NHMEX_M1_MSR_PMU_MM_CFG; + return 0; +} + +static u64 nhmex_mbox_shared_reg_config(struct intel_uncore_box *box, int idx) +{ + struct intel_uncore_extra_reg *er; + unsigned long flags; + u64 config; + + if (idx < EXTRA_REG_NHMEX_M_ZDP_CTL_FVC) + return box->shared_regs[idx].config; + + er = &box->shared_regs[EXTRA_REG_NHMEX_M_ZDP_CTL_FVC]; + raw_spin_lock_irqsave(&er->lock, flags); + config = er->config; + raw_spin_unlock_irqrestore(&er->lock, flags); + return config; +} + +static void nhmex_mbox_msr_enable_event(struct intel_uncore_box *box, + struct perf_event *event) +{ + struct hw_perf_event *hwc = &event->hw; + struct hw_perf_event_extra *reg1 = &hwc->extra_reg; + struct hw_perf_event_extra *reg2 = &hwc->branch_reg; + int idx; + + idx = __BITS_VALUE(reg1->idx, 0, 8); + if (idx != 0xff) + wrmsrl(__BITS_VALUE(reg1->reg, 0, 16), + nhmex_mbox_shared_reg_config(box, idx)); + idx = __BITS_VALUE(reg1->idx, 1, 8); + if (idx != 0xff) + wrmsrl(__BITS_VALUE(reg1->reg, 1, 16), + nhmex_mbox_shared_reg_config(box, idx)); + + wrmsrl(reg2->reg, 0); + if (reg2->config != ~0ULL) { + wrmsrl(reg2->reg + 1, + reg2->config & NHMEX_M_PMON_ADDR_MATCH_MASK); + wrmsrl(reg2->reg + 2, NHMEX_M_PMON_ADDR_MASK_MASK & + (reg2->config >> NHMEX_M_PMON_ADDR_MASK_SHIFT)); + wrmsrl(reg2->reg, NHMEX_M_PMON_MM_CFG_EN); + } + + wrmsrl(hwc->config_base, hwc->config | NHMEX_PMON_CTL_EN_BIT0); +} + +DEFINE_UNCORE_FORMAT_ATTR(count_mode, count_mode, "config:2-3"); +DEFINE_UNCORE_FORMAT_ATTR(storage_mode, storage_mode, "config:4-5"); +DEFINE_UNCORE_FORMAT_ATTR(wrap_mode, wrap_mode, "config:6"); +DEFINE_UNCORE_FORMAT_ATTR(flag_mode, flag_mode, "config:7"); +DEFINE_UNCORE_FORMAT_ATTR(inc_sel, inc_sel, "config:9-13"); +DEFINE_UNCORE_FORMAT_ATTR(set_flag_sel, set_flag_sel, "config:19-21"); +DEFINE_UNCORE_FORMAT_ATTR(filter_cfg, filter_cfg, "config2:63"); +DEFINE_UNCORE_FORMAT_ATTR(filter_match, filter_match, "config2:0-33"); +DEFINE_UNCORE_FORMAT_ATTR(filter_mask, filter_mask, "config2:34-61"); +DEFINE_UNCORE_FORMAT_ATTR(dsp, dsp, "config1:0-31"); +DEFINE_UNCORE_FORMAT_ATTR(thr, thr, "config1:0-31"); +DEFINE_UNCORE_FORMAT_ATTR(fvc, fvc, "config1:0-31"); +DEFINE_UNCORE_FORMAT_ATTR(pgt, pgt, "config1:0-31"); +DEFINE_UNCORE_FORMAT_ATTR(map, map, "config1:0-31"); +DEFINE_UNCORE_FORMAT_ATTR(iss, iss, "config1:0-31"); +DEFINE_UNCORE_FORMAT_ATTR(pld, pld, "config1:32-63"); + +static struct attribute *nhmex_uncore_mbox_formats_attr[] = { + &format_attr_count_mode.attr, + &format_attr_storage_mode.attr, + &format_attr_wrap_mode.attr, + &format_attr_flag_mode.attr, + &format_attr_inc_sel.attr, + &format_attr_set_flag_sel.attr, + &format_attr_filter_cfg.attr, + &format_attr_filter_match.attr, + &format_attr_filter_mask.attr, + &format_attr_dsp.attr, + &format_attr_thr.attr, + &format_attr_fvc.attr, + &format_attr_pgt.attr, + &format_attr_map.attr, + &format_attr_iss.attr, + &format_attr_pld.attr, + NULL, +}; + +static struct attribute_group nhmex_uncore_mbox_format_group = { + .name = "format", + .attrs = nhmex_uncore_mbox_formats_attr, +}; + +static struct uncore_event_desc nhmex_uncore_mbox_events[] = { + INTEL_UNCORE_EVENT_DESC(bbox_cmds_read, "inc_sel=0xd,fvc=0x2800"), + INTEL_UNCORE_EVENT_DESC(bbox_cmds_write, "inc_sel=0xd,fvc=0x2820"), + { /* end: all zeroes */ }, +}; + +static struct intel_uncore_ops nhmex_uncore_mbox_ops = { + NHMEX_UNCORE_OPS_COMMON_INIT(), + .enable_event = nhmex_mbox_msr_enable_event, + .hw_config = nhmex_mbox_hw_config, + .get_constraint = nhmex_mbox_get_constraint, + .put_constraint = nhmex_mbox_put_constraint, +}; + +static struct intel_uncore_type nhmex_uncore_mbox = { + .name = "mbox", + .num_counters = 6, + .num_boxes = 2, + .perf_ctr_bits = 48, + .event_ctl = NHMEX_M0_MSR_PMU_CTL0, + .perf_ctr = NHMEX_M0_MSR_PMU_CNT0, + .event_mask = NHMEX_M_PMON_RAW_EVENT_MASK, + .box_ctl = NHMEX_M0_MSR_GLOBAL_CTL, + .msr_offset = NHMEX_M_MSR_OFFSET, + .pair_ctr_ctl = 1, + .num_shared_regs = 8, + .event_descs = nhmex_uncore_mbox_events, + .ops = &nhmex_uncore_mbox_ops, + .format_group = &nhmex_uncore_mbox_format_group, +}; + +void nhmex_rbox_alter_er(struct intel_uncore_box *box, + struct perf_event *event) +{ + struct hw_perf_event *hwc = &event->hw; + struct hw_perf_event_extra *reg1 = &hwc->extra_reg; + int port; + + /* adjust the main event selector */ + if (reg1->idx % 2) { + reg1->idx--; + hwc->config -= 1 << NHMEX_R_PMON_CTL_EV_SEL_SHIFT; + } else { + reg1->idx++; + hwc->config += 1 << NHMEX_R_PMON_CTL_EV_SEL_SHIFT; + } + + /* adjust address or config of extra register */ + port = reg1->idx / 6 + box->pmu->pmu_idx * 4; + switch (reg1->idx % 6) { + case 0: + reg1->reg = NHMEX_R_MSR_PORTN_IPERF_CFG0(port); + break; + case 1: + reg1->reg = NHMEX_R_MSR_PORTN_IPERF_CFG1(port); + break; + case 2: + /* the 8~15 bits to the 0~7 bits */ + reg1->config >>= 8; + break; + case 3: + /* the 0~7 bits to the 8~15 bits */ + reg1->config <<= 8; + break; + case 4: + reg1->reg = NHMEX_R_MSR_PORTN_XBR_SET1_MM_CFG(port); + break; + case 5: + reg1->reg = NHMEX_R_MSR_PORTN_XBR_SET2_MM_CFG(port); + break; + }; +} + +/* + * Each rbox has 4 event set which monitor PQI port 0~3 or 4~7. + * An event set consists of 6 events, the 3rd and 4th events in + * an event set use the same extra register. So an event set uses + * 5 extra registers. + */ +static struct event_constraint * +nhmex_rbox_get_constraint(struct intel_uncore_box *box, + struct perf_event *event) +{ + struct hw_perf_event *hwc = &event->hw; + struct hw_perf_event_extra *reg1 = &hwc->extra_reg; + struct hw_perf_event_extra *reg2 = &hwc->branch_reg; + struct intel_uncore_extra_reg *er; + unsigned long flags; + int idx, er_idx; + u64 config1; + bool ok = false; + + if (!uncore_box_is_fake(box) && reg1->alloc) + return NULL; + + idx = reg1->idx % 6; + config1 = reg1->config; +again: + er_idx = idx; + /* the 3rd and 4th events use the same extra register */ + if (er_idx > 2) + er_idx--; + er_idx += (reg1->idx / 6) * 5; + + er = &box->shared_regs[er_idx]; + raw_spin_lock_irqsave(&er->lock, flags); + if (idx < 2) { + if (!atomic_read(&er->ref) || er->config == reg1->config) { + atomic_inc(&er->ref); + er->config = reg1->config; + ok = true; + } + } else if (idx == 2 || idx == 3) { + /* + * these two events use different fields in a extra register, + * the 0~7 bits and the 8~15 bits respectively. + */ + u64 mask = 0xff << ((idx - 2) * 8); + if (!__BITS_VALUE(atomic_read(&er->ref), idx - 2, 8) || + !((er->config ^ config1) & mask)) { + atomic_add(1 << ((idx - 2) * 8), &er->ref); + er->config &= ~mask; + er->config |= config1 & mask; + ok = true; + } + } else { + if (!atomic_read(&er->ref) || + (er->config == (hwc->config >> 32) && + er->config1 == reg1->config && + er->config2 == reg2->config)) { + atomic_inc(&er->ref); + er->config = (hwc->config >> 32); + er->config1 = reg1->config; + er->config2 = reg2->config; + ok = true; + } + } + raw_spin_unlock_irqrestore(&er->lock, flags); + + if (!ok) { + /* + * The Rbox events are always in pairs. The paired + * events are functional identical, but use different + * extra registers. If we failed to take an extra + * register, try the alternative. + */ + if (idx % 2) + idx--; + else + idx++; + if (idx != reg1->idx % 6) { + if (idx == 2) + config1 >>= 8; + else if (idx == 3) + config1 <<= 8; + goto again; + } + } else { + if (!uncore_box_is_fake(box)) { + if (idx != reg1->idx % 6) + nhmex_rbox_alter_er(box, event); + reg1->alloc = 1; + } + return NULL; + } + return &constraint_empty; +} + +static void nhmex_rbox_put_constraint(struct intel_uncore_box *box, + struct perf_event *event) +{ + struct intel_uncore_extra_reg *er; + struct hw_perf_event_extra *reg1 = &event->hw.extra_reg; + int idx, er_idx; + + if (uncore_box_is_fake(box) || !reg1->alloc) + return; + + idx = reg1->idx % 6; + er_idx = idx; + if (er_idx > 2) + er_idx--; + er_idx += (reg1->idx / 6) * 5; + + er = &box->shared_regs[er_idx]; + if (idx == 2 || idx == 3) + atomic_sub(1 << ((idx - 2) * 8), &er->ref); + else + atomic_dec(&er->ref); + + reg1->alloc = 0; +} + +static int nhmex_rbox_hw_config(struct intel_uncore_box *box, + struct perf_event *event) +{ + struct hw_perf_event *hwc = &event->hw; + struct hw_perf_event_extra *reg1 = &event->hw.extra_reg; + struct hw_perf_event_extra *reg2 = &event->hw.branch_reg; + int port, idx; + + idx = (event->hw.config & NHMEX_R_PMON_CTL_EV_SEL_MASK) >> + NHMEX_R_PMON_CTL_EV_SEL_SHIFT; + if (idx >= 0x18) + return -EINVAL; + + reg1->idx = idx; + reg1->config = event->attr.config1; + + port = idx / 6 + box->pmu->pmu_idx * 4; + idx %= 6; + switch (idx) { + case 0: + reg1->reg = NHMEX_R_MSR_PORTN_IPERF_CFG0(port); + break; + case 1: + reg1->reg = NHMEX_R_MSR_PORTN_IPERF_CFG1(port); + break; + case 2: + case 3: + reg1->reg = NHMEX_R_MSR_PORTN_QLX_CFG(port); + break; + case 4: + case 5: + if (idx == 4) + reg1->reg = NHMEX_R_MSR_PORTN_XBR_SET1_MM_CFG(port); + else + reg1->reg = NHMEX_R_MSR_PORTN_XBR_SET2_MM_CFG(port); + reg2->config = event->attr.config2; + hwc->config |= event->attr.config & (~0ULL << 32); + break; + }; + return 0; +} + +static u64 nhmex_rbox_shared_reg_config(struct intel_uncore_box *box, int idx) +{ + struct intel_uncore_extra_reg *er; + unsigned long flags; + u64 config; + + er = &box->shared_regs[idx]; + raw_spin_lock_irqsave(&er->lock, flags); + config = er->config; + raw_spin_unlock_irqrestore(&er->lock, flags); + return config; +} + +static void nhmex_rbox_msr_enable_event(struct intel_uncore_box *box, + struct perf_event *event) +{ + struct hw_perf_event *hwc = &event->hw; + struct hw_perf_event_extra *reg1 = &hwc->extra_reg; + struct hw_perf_event_extra *reg2 = &hwc->branch_reg; + int idx, er_idx; + + idx = reg1->idx % 6; + er_idx = idx; + if (er_idx > 2) + er_idx--; + er_idx += (reg1->idx / 6) * 5; + + switch (idx) { + case 0: + case 1: + wrmsrl(reg1->reg, reg1->config); + break; + case 2: + case 3: + wrmsrl(reg1->reg, nhmex_rbox_shared_reg_config(box, er_idx)); + break; + case 4: + case 5: + wrmsrl(reg1->reg, reg1->config); + wrmsrl(reg1->reg + 1, hwc->config >> 32); + wrmsrl(reg1->reg + 2, reg2->config); + break; + }; + + wrmsrl(hwc->config_base, NHMEX_PMON_CTL_EN_BIT0 | + (hwc->config & NHMEX_R_PMON_CTL_EV_SEL_MASK)); +} + +DEFINE_UNCORE_FORMAT_ATTR(xbr_match, xbr_match, "config:32-63"); +DEFINE_UNCORE_FORMAT_ATTR(xbr_mm_cfg, xbr_mm_cfg, "config1:0-63"); +DEFINE_UNCORE_FORMAT_ATTR(xbr_mask, xbr_mask, "config2:0-63"); +DEFINE_UNCORE_FORMAT_ATTR(qlx_cfg, qlx_cfg, "config1:0-15"); +DEFINE_UNCORE_FORMAT_ATTR(iperf_cfg, iperf_cfg, "config1:0-31"); + +static struct attribute *nhmex_uncore_rbox_formats_attr[] = { + &format_attr_event5.attr, + &format_attr_xbr_mm_cfg.attr, + &format_attr_xbr_match.attr, + &format_attr_xbr_mask.attr, + &format_attr_qlx_cfg.attr, + &format_attr_iperf_cfg.attr, + NULL, +}; + +static struct attribute_group nhmex_uncore_rbox_format_group = { + .name = "format", + .attrs = nhmex_uncore_rbox_formats_attr, +}; + +static struct uncore_event_desc nhmex_uncore_rbox_events[] = { + INTEL_UNCORE_EVENT_DESC(qpi0_flit_send, + "event=0x0,iperf_cfg=0x80000000"), + INTEL_UNCORE_EVENT_DESC(qpi1_filt_send, + "event=0x6,iperf_cfg=0x80000000"), + INTEL_UNCORE_EVENT_DESC(qpi0_idle_filt, + "event=0x0,iperf_cfg=0x40000000"), + INTEL_UNCORE_EVENT_DESC(qpi1_idle_filt, + "event=0x6,iperf_cfg=0x40000000"), + INTEL_UNCORE_EVENT_DESC(qpi0_date_response, + "event=0x0,iperf_cfg=0xc4"), + INTEL_UNCORE_EVENT_DESC(qpi1_date_response, + "event=0x6,iperf_cfg=0xc4"), + { /* end: all zeroes */ }, +}; + +static struct intel_uncore_ops nhmex_uncore_rbox_ops = { + NHMEX_UNCORE_OPS_COMMON_INIT(), + .enable_event = nhmex_rbox_msr_enable_event, + .hw_config = nhmex_rbox_hw_config, + .get_constraint = nhmex_rbox_get_constraint, + .put_constraint = nhmex_rbox_put_constraint, +}; + +static struct intel_uncore_type nhmex_uncore_rbox = { + .name = "rbox", + .num_counters = 8, + .num_boxes = 2, + .perf_ctr_bits = 48, + .event_ctl = NHMEX_R_MSR_PMON_CTL0, + .perf_ctr = NHMEX_R_MSR_PMON_CNT0, + .event_mask = NHMEX_R_PMON_RAW_EVENT_MASK, + .box_ctl = NHMEX_R_MSR_GLOBAL_CTL, + .msr_offset = NHMEX_R_MSR_OFFSET, + .pair_ctr_ctl = 1, + .num_shared_regs = 20, + .event_descs = nhmex_uncore_rbox_events, + .ops = &nhmex_uncore_rbox_ops, + .format_group = &nhmex_uncore_rbox_format_group +}; + +static struct intel_uncore_type *nhmex_msr_uncores[] = { + &nhmex_uncore_ubox, + &nhmex_uncore_cbox, + &nhmex_uncore_bbox, + &nhmex_uncore_sbox, + &nhmex_uncore_mbox, + &nhmex_uncore_rbox, + &nhmex_uncore_wbox, + NULL, +}; +/* end of Nehalem-EX uncore support */ + static void uncore_assign_hw_event(struct intel_uncore_box *box, struct perf_event *event, int idx) { @@ -1767,6 +2821,9 @@ static int __init uncore_cpu_init(void) snbep_uncore_cbox.num_boxes = max_cores; msr_uncores = snbep_msr_uncores; break; + case 46: + msr_uncores = nhmex_msr_uncores; + break; default: return 0; } diff --git a/arch/x86/kernel/cpu/perf_event_intel_uncore.h b/arch/x86/kernel/cpu/perf_event_intel_uncore.h index b13e9ea..47b1776 100644 --- a/arch/x86/kernel/cpu/perf_event_intel_uncore.h +++ b/arch/x86/kernel/cpu/perf_event_intel_uncore.h @@ -5,8 +5,6 @@ #include "perf_event.h" #define UNCORE_PMU_NAME_LEN 32 -#define UNCORE_BOX_HASH_SIZE 8 - #define UNCORE_PMU_HRTIMER_INTERVAL (60 * NSEC_PER_SEC) #define UNCORE_FIXED_EVENT 0xff @@ -158,6 +156,193 @@ #define SNBEP_PCU_MSR_CORE_C3_CTR 0x3fc #define SNBEP_PCU_MSR_CORE_C6_CTR 0x3fd +/* NHM-EX event control */ +#define NHMEX_PMON_CTL_EV_SEL_MASK 0x000000ff +#define NHMEX_PMON_CTL_UMASK_MASK 0x0000ff00 +#define NHMEX_PMON_CTL_EN_BIT0 (1 << 0) +#define NHMEX_PMON_CTL_EDGE_DET (1 << 18) +#define NHMEX_PMON_CTL_PMI_EN (1 << 20) +#define NHMEX_PMON_CTL_EN_BIT22 (1 << 22) +#define NHMEX_PMON_CTL_INVERT (1 << 23) +#define NHMEX_PMON_CTL_TRESH_MASK 0xff000000 +#define NHMEX_PMON_RAW_EVENT_MASK (NHMEX_PMON_CTL_EV_SEL_MASK | \ + NHMEX_PMON_CTL_UMASK_MASK | \ + NHMEX_PMON_CTL_EDGE_DET | \ + NHMEX_PMON_CTL_INVERT | \ + NHMEX_PMON_CTL_TRESH_MASK) + +/* NHM-EX Ubox */ +#define NHMEX_U_MSR_PMON_GLOBAL_CTL 0xc00 +#define NHMEX_U_MSR_PMON_CTR 0xc11 +#define NHMEX_U_MSR_PMON_EV_SEL 0xc10 + +#define NHMEX_U_PMON_GLOBAL_EN (1 << 0) +#define NHMEX_U_PMON_GLOBAL_PMI_CORE_SEL 0x0000001e +#define NHMEX_U_PMON_GLOBAL_EN_ALL (1 << 28) +#define NHMEX_U_PMON_GLOBAL_RST_ALL (1 << 29) +#define NHMEX_U_PMON_GLOBAL_FRZ_ALL (1 << 31) + +#define NHMEX_U_PMON_RAW_EVENT_MASK \ + (NHMEX_PMON_CTL_EV_SEL_MASK | \ + NHMEX_PMON_CTL_EDGE_DET) + +/* NHM-EX Cbox */ +#define NHMEX_C0_MSR_PMON_GLOBAL_CTL 0xd00 +#define NHMEX_C0_MSR_PMON_CTR0 0xd11 +#define NHMEX_C0_MSR_PMON_EV_SEL0 0xd10 +#define NHMEX_C_MSR_OFFSET 0x20 + +/* NHM-EX Bbox */ +#define NHMEX_B0_MSR_PMON_GLOBAL_CTL 0xc20 +#define NHMEX_B0_MSR_PMON_CTR0 0xc31 +#define NHMEX_B0_MSR_PMON_CTL0 0xc30 +#define NHMEX_B_MSR_OFFSET 0x40 +#define NHMEX_B0_MSR_MATCH 0xe45 +#define NHMEX_B0_MSR_MASK 0xe46 +#define NHMEX_B1_MSR_MATCH 0xe4d +#define NHMEX_B1_MSR_MASK 0xe4e + +#define NHMEX_B_PMON_CTL_EN (1 << 0) +#define NHMEX_B_PMON_CTL_EV_SEL_SHIFT 1 +#define NHMEX_B_PMON_CTL_EV_SEL_MASK \ + (0x1f << NHMEX_B_PMON_CTL_EV_SEL_SHIFT) +#define NHMEX_B_PMON_CTR_SHIFT 6 +#define NHMEX_B_PMON_CTR_MASK \ + (0x3 << NHMEX_B_PMON_CTR_SHIFT) +#define NHMEX_B_PMON_RAW_EVENT_MASK \ + (NHMEX_B_PMON_CTL_EV_SEL_MASK | \ + NHMEX_B_PMON_CTR_MASK) + +/* NHM-EX Sbox */ +#define NHMEX_S0_MSR_PMON_GLOBAL_CTL 0xc40 +#define NHMEX_S0_MSR_PMON_CTR0 0xc51 +#define NHMEX_S0_MSR_PMON_CTL0 0xc50 +#define NHMEX_S_MSR_OFFSET 0x80 +#define NHMEX_S0_MSR_MM_CFG 0xe48 +#define NHMEX_S0_MSR_MATCH 0xe49 +#define NHMEX_S0_MSR_MASK 0xe4a +#define NHMEX_S1_MSR_MM_CFG 0xe58 +#define NHMEX_S1_MSR_MATCH 0xe59 +#define NHMEX_S1_MSR_MASK 0xe5a + +#define NHMEX_S_PMON_MM_CFG_EN (0x1ULL << 63) + +/* NHM-EX Mbox */ +#define NHMEX_M0_MSR_GLOBAL_CTL 0xca0 +#define NHMEX_M0_MSR_PMU_DSP 0xca5 +#define NHMEX_M0_MSR_PMU_ISS 0xca6 +#define NHMEX_M0_MSR_PMU_MAP 0xca7 +#define NHMEX_M0_MSR_PMU_MSC_THR 0xca8 +#define NHMEX_M0_MSR_PMU_PGT 0xca9 +#define NHMEX_M0_MSR_PMU_PLD 0xcaa +#define NHMEX_M0_MSR_PMU_ZDP_CTL_FVC 0xcab +#define NHMEX_M0_MSR_PMU_CTL0 0xcb0 +#define NHMEX_M0_MSR_PMU_CNT0 0xcb1 +#define NHMEX_M_MSR_OFFSET 0x40 +#define NHMEX_M0_MSR_PMU_MM_CFG 0xe54 +#define NHMEX_M1_MSR_PMU_MM_CFG 0xe5c + +#define NHMEX_M_PMON_MM_CFG_EN (1ULL << 63) +#define NHMEX_M_PMON_ADDR_MATCH_MASK 0x3ffffffffULL +#define NHMEX_M_PMON_ADDR_MASK_MASK 0x7ffffffULL +#define NHMEX_M_PMON_ADDR_MASK_SHIFT 34 + +#define NHMEX_M_PMON_CTL_EN (1 << 0) +#define NHMEX_M_PMON_CTL_PMI_EN (1 << 1) +#define NHMEX_M_PMON_CTL_COUNT_MODE_SHIFT 2 +#define NHMEX_M_PMON_CTL_COUNT_MODE_MASK \ + (0x3 << NHMEX_M_PMON_CTL_COUNT_MODE_SHIFT) +#define NHMEX_M_PMON_CTL_STORAGE_MODE_SHIFT 4 +#define NHMEX_M_PMON_CTL_STORAGE_MODE_MASK \ + (0x3 << NHMEX_M_PMON_CTL_STORAGE_MODE_SHIFT) +#define NHMEX_M_PMON_CTL_WRAP_MODE (1 << 6) +#define NHMEX_M_PMON_CTL_FLAG_MODE (1 << 7) +#define NHMEX_M_PMON_CTL_INC_SEL_SHIFT 9 +#define NHMEX_M_PMON_CTL_INC_SEL_MASK \ + (0x1f << NHMEX_M_PMON_CTL_INC_SEL_SHIFT) +#define NHMEX_M_PMON_CTL_SET_FLAG_SEL_SHIFT 19 +#define NHMEX_M_PMON_CTL_SET_FLAG_SEL_MASK \ + (0x7 << NHMEX_M_PMON_CTL_SET_FLAG_SEL_SHIFT) +#define NHMEX_M_PMON_RAW_EVENT_MASK \ + (NHMEX_M_PMON_CTL_COUNT_MODE_MASK | \ + NHMEX_M_PMON_CTL_STORAGE_MODE_MASK | \ + NHMEX_M_PMON_CTL_WRAP_MODE | \ + NHMEX_M_PMON_CTL_FLAG_MODE | \ + NHMEX_M_PMON_CTL_INC_SEL_MASK | \ + NHMEX_M_PMON_CTL_SET_FLAG_SEL_MASK) + + +#define NHMEX_M_PMON_ZDP_CTL_FVC_FVID_MASK 0x1f +#define NHMEX_M_PMON_ZDP_CTL_FVC_BCMD_MASK (0x7 << 5) +#define NHMEX_M_PMON_ZDP_CTL_FVC_RSP_MASK (0x7 << 8) +#define NHMEX_M_PMON_ZDP_CTL_FVC_PBOX_INIT_ERR (1 << 23) +#define NHMEX_M_PMON_ZDP_CTL_FVC_MASK \ + (NHMEX_M_PMON_ZDP_CTL_FVC_FVID_MASK | \ + NHMEX_M_PMON_ZDP_CTL_FVC_BCMD_MASK | \ + NHMEX_M_PMON_ZDP_CTL_FVC_RSP_MASK | \ + NHMEX_M_PMON_ZDP_CTL_FVC_PBOX_INIT_ERR) +#define NHMEX_M_PMON_ZDP_CTL_FVC_EVENT_MASK(n) (0x7 << (11 + 3 * (n))) + +/* + * use the 9~13 bits to select event If the 7th bit is not set, + * otherwise use the 19~21 bits to select event. + */ +#define MBOX_INC_SEL(x) ((x) << NHMEX_M_PMON_CTL_INC_SEL_SHIFT) +#define MBOX_SET_FLAG_SEL(x) (((x) << NHMEX_M_PMON_CTL_SET_FLAG_SEL_SHIFT) | \ + NHMEX_M_PMON_CTL_FLAG_MODE) +#define MBOX_INC_SEL_MASK (NHMEX_M_PMON_CTL_INC_SEL_MASK | \ + NHMEX_M_PMON_CTL_FLAG_MODE) +#define MBOX_SET_FLAG_SEL_MASK (NHMEX_M_PMON_CTL_SET_FLAG_SEL_MASK | \ + NHMEX_M_PMON_CTL_FLAG_MODE) +#define MBOX_INC_SEL_EXTAR_REG(c, r) \ + EVENT_EXTRA_REG(MBOX_INC_SEL(c), NHMEX_M0_MSR_PMU_##r, \ + MBOX_INC_SEL_MASK, (u64)-1, NHMEX_M_##r) +#define MBOX_SET_FLAG_SEL_EXTRA_REG(c, r) \ + EVENT_EXTRA_REG(MBOX_SET_FLAG_SEL(c), NHMEX_M0_MSR_PMU_##r, \ + MBOX_SET_FLAG_SEL_MASK, \ + (u64)-1, NHMEX_M_##r) + +/* NHM-EX Rbox */ +#define NHMEX_R_MSR_GLOBAL_CTL 0xe00 +#define NHMEX_R_MSR_PMON_CTL0 0xe10 +#define NHMEX_R_MSR_PMON_CNT0 0xe11 +#define NHMEX_R_MSR_OFFSET 0x20 + +#define NHMEX_R_MSR_PORTN_QLX_CFG(n) \ + ((n) < 4 ? (0xe0c + (n)) : (0xe2c + (n) - 4)) +#define NHMEX_R_MSR_PORTN_IPERF_CFG0(n) (0xe04 + (n)) +#define NHMEX_R_MSR_PORTN_IPERF_CFG1(n) (0xe24 + (n)) +#define NHMEX_R_MSR_PORTN_XBR_OFFSET(n) \ + (((n) < 4 ? 0 : 0x10) + (n) * 4) +#define NHMEX_R_MSR_PORTN_XBR_SET1_MM_CFG(n) \ + (0xe60 + NHMEX_R_MSR_PORTN_XBR_OFFSET(n)) +#define NHMEX_R_MSR_PORTN_XBR_SET1_MATCH(n) \ + (NHMEX_R_MSR_PORTN_XBR_SET1_MM_CFG(n) + 1) +#define NHMEX_R_MSR_PORTN_XBR_SET1_MASK(n) \ + (NHMEX_R_MSR_PORTN_XBR_SET1_MM_CFG(n) + 2) +#define NHMEX_R_MSR_PORTN_XBR_SET2_MM_CFG(n) \ + (0xe70 + NHMEX_R_MSR_PORTN_XBR_OFFSET(n)) +#define NHMEX_R_MSR_PORTN_XBR_SET2_MATCH(n) \ + (NHMEX_R_MSR_PORTN_XBR_SET2_MM_CFG(n) + 1) +#define NHMEX_R_MSR_PORTN_XBR_SET2_MASK(n) \ + (NHMEX_R_MSR_PORTN_XBR_SET2_MM_CFG(n) + 2) + +#define NHMEX_R_PMON_CTL_EN (1 << 0) +#define NHMEX_R_PMON_CTL_EV_SEL_SHIFT 1 +#define NHMEX_R_PMON_CTL_EV_SEL_MASK \ + (0x1f << NHMEX_R_PMON_CTL_EV_SEL_SHIFT) +#define NHMEX_R_PMON_CTL_PMI_EN (1 << 6) +#define NHMEX_R_PMON_RAW_EVENT_MASK NHMEX_R_PMON_CTL_EV_SEL_MASK + +/* NHM-EX Wbox */ +#define NHMEX_W_MSR_GLOBAL_CTL 0xc80 +#define NHMEX_W_MSR_PMON_CNT0 0xc90 +#define NHMEX_W_MSR_PMON_EVT_SEL0 0xc91 +#define NHMEX_W_MSR_PMON_FIXED_CTR 0x394 +#define NHMEX_W_MSR_PMON_FIXED_CTL 0x395 + +#define NHMEX_W_PMON_GLOBAL_FIXED_EN (1ULL << 31) + struct intel_uncore_ops; struct intel_uncore_pmu; struct intel_uncore_box; @@ -178,6 +363,7 @@ struct intel_uncore_type { unsigned msr_offset; unsigned num_shared_regs:8; unsigned single_fixed:1; + unsigned pair_ctr_ctl:1; struct event_constraint unconstrainted; struct event_constraint *constraints; struct intel_uncore_pmu *pmus; @@ -213,7 +399,7 @@ struct intel_uncore_pmu { struct intel_uncore_extra_reg { raw_spinlock_t lock; - u64 config1; + u64 config, config1, config2; atomic_t ref; }; @@ -323,14 +509,16 @@ unsigned uncore_msr_fixed_ctr(struct intel_uncore_box *box) static inline unsigned uncore_msr_event_ctl(struct intel_uncore_box *box, int idx) { - return idx + box->pmu->type->event_ctl + + return box->pmu->type->event_ctl + + (box->pmu->type->pair_ctr_ctl ? 2 * idx : idx) + box->pmu->type->msr_offset * box->pmu->pmu_idx; } static inline unsigned uncore_msr_perf_ctr(struct intel_uncore_box *box, int idx) { - return idx + box->pmu->type->perf_ctr + + return box->pmu->type->perf_ctr + + (box->pmu->type->pair_ctr_ctl ? 2 * idx : idx) + box->pmu->type->msr_offset * box->pmu->pmu_idx; } @@ -422,3 +610,8 @@ static inline void uncore_box_init(struct intel_uncore_box *box) box->pmu->type->ops->init_box(box); } } + +static inline bool uncore_box_is_fake(struct intel_uncore_box *box) +{ + return (box->phys_id < 0); +} -- 1.7.10.4 ^ permalink raw reply related [flat|nested] 21+ messages in thread
* Re: [PATCH V2 5/5] perf/x86: Add Intel Nehalem-EX uncore support 2012-07-05 6:32 ` [PATCH V2 " Yan, Zheng @ 2012-07-05 14:51 ` Andi Kleen 2012-07-05 15:46 ` Yan, Zheng 2012-07-10 12:17 ` Peter Zijlstra 2012-07-26 15:11 ` [tip:perf/core] " tip-bot for Yan, Zheng 2 siblings, 1 reply; 21+ messages in thread From: Andi Kleen @ 2012-07-05 14:51 UTC (permalink / raw) To: Yan, Zheng; +Cc: Peter Zijlstra, eranian, mingo, andi, linux-kernel On Thu, Jul 05, 2012 at 02:32:17PM +0800, Yan, Zheng wrote: > The uncore subsystem in Nehalem-EX consists of 7 components > (U-Box, C-Box, B-Box, S-Box, R-Box, M-Box and W-Box). This > patch is large because the way to program these boxes is > diverse. Thanks for doing the driver. Lots of work. May be worth adding a CONFIG for the uncore code now? Maybe even a module, so that not every distro kernel has it always in memory. I don't think perf has support for tracking module counts, but I guess it would be ok to have the module be not unloadable once loaded by setting the count to -1. Also did you do some random testing by putting randomized values into all the exported registers and see if anything is crashable for unpriv. userspace? > + * events are functional identical, but use different > + * extra registers. If we failed to take an extra > + * register, try the alternative. > + */ > + if (idx % 2) > + idx--; > + else > + idx++; > + if (idx != reg1->idx % 6) { > + if (idx == 2) > + config1 >>= 8; > + else if (idx == 3) > + config1 <<= 8; > + goto again; Does this limit the retries? > @@ -1767,6 +2821,9 @@ static int __init uncore_cpu_init(void) > snbep_uncore_cbox.num_boxes = max_cores; > msr_uncores = snbep_msr_uncores; > break; > + case 46: > + msr_uncores = nhmex_msr_uncores; > + break; Ok need Westmere-EX too. I think that is quite similar. -Andi -- ak@linux.intel.com -- Speaking for myself only. ^ permalink raw reply [flat|nested] 21+ messages in thread
* Re: [PATCH V2 5/5] perf/x86: Add Intel Nehalem-EX uncore support 2012-07-05 14:51 ` Andi Kleen @ 2012-07-05 15:46 ` Yan, Zheng 0 siblings, 0 replies; 21+ messages in thread From: Yan, Zheng @ 2012-07-05 15:46 UTC (permalink / raw) To: Andi Kleen; +Cc: Peter Zijlstra, eranian, mingo, linux-kernel On 07/05/2012 10:51 PM, Andi Kleen wrote: > On Thu, Jul 05, 2012 at 02:32:17PM +0800, Yan, Zheng wrote: >> The uncore subsystem in Nehalem-EX consists of 7 components >> (U-Box, C-Box, B-Box, S-Box, R-Box, M-Box and W-Box). This >> patch is large because the way to program these boxes is >> diverse. > > Thanks for doing the driver. Lots of work. > > May be worth adding a CONFIG for the uncore code now? > Maybe even a module, so that not every distro kernel has it always > in memory. I don't think perf has support for tracking > module counts, but I guess it would be ok to have the module be not > unloadable once loaded by setting the count to -1. Yes, I think it's good to compile the uncore driver as a module. > > Also did you do some random testing by putting randomized values into > all the exported registers and see if anything is crashable for > unpriv. userspace? No, I just did functional tests for these registers. unpriv? I think perf is only available to root by default. > >> + * events are functional identical, but use different >> + * extra registers. If we failed to take an extra >> + * register, try the alternative. >> + */ >> + if (idx % 2) >> + idx--; >> + else >> + idx++; >> + if (idx != reg1->idx % 6) { >> + if (idx == 2) >> + config1 >>= 8; >> + else if (idx == 3) >> + config1 <<= 8; >> + goto again; > > Does this limit the retries? Yes, the (idx != reg1->idx % 6) check does that. Regards Yan, Zheng ^ permalink raw reply [flat|nested] 21+ messages in thread
* Re: [PATCH V2 5/5] perf/x86: Add Intel Nehalem-EX uncore support 2012-07-05 6:32 ` [PATCH V2 " Yan, Zheng 2012-07-05 14:51 ` Andi Kleen @ 2012-07-10 12:17 ` Peter Zijlstra 2012-07-11 2:06 ` Yan, Zheng 2012-07-26 15:11 ` [tip:perf/core] " tip-bot for Yan, Zheng 2 siblings, 1 reply; 21+ messages in thread From: Peter Zijlstra @ 2012-07-10 12:17 UTC (permalink / raw) To: Yan, Zheng; +Cc: eranian, mingo, andi, linux-kernel On Thu, 2012-07-05 at 14:32 +0800, Yan, Zheng wrote: > The uncore subsystem in Nehalem-EX consists of 7 components > (U-Box, C-Box, B-Box, S-Box, R-Box, M-Box and W-Box). This > patch is large because the way to program these boxes is > diverse. > Thanks, I added the below little delta. Did you manage to fix that SNB-EP crash already? Also, should we be adding 58 (IVB) to the SNB uncore? The SDM states it should all be the same programming wise. --- --- a/arch/x86/kernel/cpu/perf_event_intel_uncore.c +++ b/arch/x86/kernel/cpu/perf_event_intel_uncore.c @@ -893,7 +893,7 @@ static struct attribute_group nhmex_unco static struct intel_uncore_type nhmex_uncore_cbox = { .name = "cbox", .num_counters = 6, - .num_boxes = 8, + .num_boxes = 10, .perf_ctr_bits = 48, .event_ctl = NHMEX_C0_MSR_PMON_EV_SEL0, .perf_ctr = NHMEX_C0_MSR_PMON_CTR0, @@ -2806,9 +2806,9 @@ static int __init uncore_cpu_init(void) max_cores = boot_cpu_data.x86_max_cores; switch (boot_cpu_data.x86_model) { case 26: /* Nehalem */ - case 30: + case 30: /* Nehalem-EP */ case 37: /* Westmere */ - case 44: + case 44: /* Westmere-EP */ msr_uncores = nhm_msr_uncores; break; case 42: /* Sandy Bridge */ @@ -2821,7 +2821,10 @@ static int __init uncore_cpu_init(void) snbep_uncore_cbox.num_boxes = max_cores; msr_uncores = snbep_msr_uncores; break; - case 46: + case 46: /* Nehalem-EX */ + case 47: /* Westmere-EX aka. Xeon E7 */ + if (nhmex_uncore_cbox.num_boxes > max_cores) + nhmex_uncore_cbox.num_boxes = max_cores; msr_uncores = nhmex_msr_uncores; break; default: ^ permalink raw reply [flat|nested] 21+ messages in thread
* Re: [PATCH V2 5/5] perf/x86: Add Intel Nehalem-EX uncore support 2012-07-10 12:17 ` Peter Zijlstra @ 2012-07-11 2:06 ` Yan, Zheng 2012-07-11 9:57 ` Peter Zijlstra 0 siblings, 1 reply; 21+ messages in thread From: Yan, Zheng @ 2012-07-11 2:06 UTC (permalink / raw) To: Peter Zijlstra; +Cc: eranian, mingo, andi, linux-kernel On 07/10/2012 08:17 PM, Peter Zijlstra wrote: > On Thu, 2012-07-05 at 14:32 +0800, Yan, Zheng wrote: >> The uncore subsystem in Nehalem-EX consists of 7 components >> (U-Box, C-Box, B-Box, S-Box, R-Box, M-Box and W-Box). This >> patch is large because the way to program these boxes is >> diverse. >> > Thanks, I added the below little delta. > > Did you manage to fix that SNB-EP crash already? Yes, commit 42089697 (perf/x86: Detect number of instances of uncore CBox) should fix that. > > Also, should we be adding 58 (IVB) to the SNB uncore? The SDM states it > should all be the same programming wise. > It's completely untested, I think it's better not to do that now. > --- > --- a/arch/x86/kernel/cpu/perf_event_intel_uncore.c > +++ b/arch/x86/kernel/cpu/perf_event_intel_uncore.c > > @@ -893,7 +893,7 @@ static struct attribute_group nhmex_unco > static struct intel_uncore_type nhmex_uncore_cbox = { > .name = "cbox", > .num_counters = 6, > - .num_boxes = 8, > + .num_boxes = 10, > .perf_ctr_bits = 48, > .event_ctl = NHMEX_C0_MSR_PMON_EV_SEL0, > .perf_ctr = NHMEX_C0_MSR_PMON_CTR0, This doesn't work. The MSRs for Cbox 8/9 in WSM-EX aren't contiguous with Cbox 0~7. Furthermore the encoding for the FVC register in the Mbox are different between NHM-EX and WSM-EX. Please don't merge this patch, I will send a new one with WSM-EX support and some fixes soon. Regards Yan, Zheng ^ permalink raw reply [flat|nested] 21+ messages in thread
* Re: [PATCH V2 5/5] perf/x86: Add Intel Nehalem-EX uncore support 2012-07-11 2:06 ` Yan, Zheng @ 2012-07-11 9:57 ` Peter Zijlstra 2012-07-11 12:02 ` Yan, Zheng 0 siblings, 1 reply; 21+ messages in thread From: Peter Zijlstra @ 2012-07-11 9:57 UTC (permalink / raw) To: Yan, Zheng; +Cc: eranian, mingo, andi, linux-kernel On Wed, 2012-07-11 at 10:06 +0800, Yan, Zheng wrote: > This doesn't work. The MSRs for Cbox 8/9 in WSM-EX aren't contiguous with > Cbox 0~7. Argh! > Furthermore the encoding for the FVC register in the Mbox are > different between NHM-EX and WSM-EX. OK, 18.7.1 from Vol. 3B doesn't mention that.. where did you find this? > Please don't merge this patch, I will send a new one with WSM-EX support > and some fixes soon. Found it in the inbox, thanks! ^ permalink raw reply [flat|nested] 21+ messages in thread
* Re: [PATCH V2 5/5] perf/x86: Add Intel Nehalem-EX uncore support 2012-07-11 9:57 ` Peter Zijlstra @ 2012-07-11 12:02 ` Yan, Zheng 0 siblings, 0 replies; 21+ messages in thread From: Yan, Zheng @ 2012-07-11 12:02 UTC (permalink / raw) To: Peter Zijlstra; +Cc: Yan, Zheng, eranian, mingo, andi, linux-kernel On Wed, Jul 11, 2012 at 5:57 PM, Peter Zijlstra <a.p.zijlstra@chello.nl> wrote: > On Wed, 2012-07-11 at 10:06 +0800, Yan, Zheng wrote: >> This doesn't work. The MSRs for Cbox 8/9 in WSM-EX aren't contiguous with >> Cbox 0~7. > > Argh! > >> Furthermore the encoding for the FVC register in the Mbox are >> different between NHM-EX and WSM-EX. > > OK, 18.7.1 from Vol. 3B doesn't mention that.. where did you find this? > Table 2-80 in <<Intel Xeon Processor 7500 Series Uncore Programming Guide>> and <<Intel Xeon Processor E7 Family Uncore Performance Monitoring Programming Guide>>. The fvid field in NHM-EX is 5 bits, in WSM-EX is 6 bits. Thanks Yan, Zheng ^ permalink raw reply [flat|nested] 21+ messages in thread
* [tip:perf/core] perf/x86: Add Intel Nehalem-EX uncore support 2012-07-05 6:32 ` [PATCH V2 " Yan, Zheng 2012-07-05 14:51 ` Andi Kleen 2012-07-10 12:17 ` Peter Zijlstra @ 2012-07-26 15:11 ` tip-bot for Yan, Zheng 2 siblings, 0 replies; 21+ messages in thread From: tip-bot for Yan, Zheng @ 2012-07-26 15:11 UTC (permalink / raw) To: linux-tip-commits Cc: linux-kernel, hpa, mingo, a.p.zijlstra, zheng.z.yan, tglx Commit-ID: 254298c726b93bb8ed92774b4a209b479851fa6d Gitweb: http://git.kernel.org/tip/254298c726b93bb8ed92774b4a209b479851fa6d Author: Yan, Zheng <zheng.z.yan@intel.com> AuthorDate: Thu, 5 Jul 2012 14:32:17 +0800 Committer: Ingo Molnar <mingo@kernel.org> CommitDate: Thu, 26 Jul 2012 12:23:11 +0200 perf/x86: Add Intel Nehalem-EX uncore support The uncore subsystem in Nehalem-EX consists of 7 components (U-Box, C-Box, B-Box, S-Box, R-Box, M-Box and W-Box). This patch is large because the way to program these boxes is diverse. Signed-off-by: Yan, Zheng <zheng.z.yan@intel.com> Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: http://lkml.kernel.org/r/4FF534F1.3030307@intel.com [ Improved the code. ] Signed-off-by: Ingo Molnar <mingo@kernel.org> --- arch/x86/kernel/cpu/perf_event_intel_uncore.c | 1482 +++++++++++++++++++++---- arch/x86/kernel/cpu/perf_event_intel_uncore.h | 203 ++++- 2 files changed, 1454 insertions(+), 231 deletions(-) diff --git a/arch/x86/kernel/cpu/perf_event_intel_uncore.c b/arch/x86/kernel/cpu/perf_event_intel_uncore.c index a5de59f..d998170 100644 --- a/arch/x86/kernel/cpu/perf_event_intel_uncore.c +++ b/arch/x86/kernel/cpu/perf_event_intel_uncore.c @@ -38,6 +38,77 @@ DEFINE_UNCORE_FORMAT_ATTR(filter_band1, filter_band1, "config1:8-15"); DEFINE_UNCORE_FORMAT_ATTR(filter_band2, filter_band2, "config1:16-23"); DEFINE_UNCORE_FORMAT_ATTR(filter_band3, filter_band3, "config1:24-31"); +static u64 uncore_msr_read_counter(struct intel_uncore_box *box, struct perf_event *event) +{ + u64 count; + + rdmsrl(event->hw.event_base, count); + + return count; +} + +/* + * generic get constraint function for shared match/mask registers. + */ +static struct event_constraint * +uncore_get_constraint(struct intel_uncore_box *box, struct perf_event *event) +{ + struct intel_uncore_extra_reg *er; + struct hw_perf_event_extra *reg1 = &event->hw.extra_reg; + struct hw_perf_event_extra *reg2 = &event->hw.branch_reg; + unsigned long flags; + bool ok = false; + + /* + * reg->alloc can be set due to existing state, so for fake box we + * need to ignore this, otherwise we might fail to allocate proper + * fake state for this extra reg constraint. + */ + if (reg1->idx == EXTRA_REG_NONE || + (!uncore_box_is_fake(box) && reg1->alloc)) + return NULL; + + er = &box->shared_regs[reg1->idx]; + raw_spin_lock_irqsave(&er->lock, flags); + if (!atomic_read(&er->ref) || + (er->config1 == reg1->config && er->config2 == reg2->config)) { + atomic_inc(&er->ref); + er->config1 = reg1->config; + er->config2 = reg2->config; + ok = true; + } + raw_spin_unlock_irqrestore(&er->lock, flags); + + if (ok) { + if (!uncore_box_is_fake(box)) + reg1->alloc = 1; + return NULL; + } + + return &constraint_empty; +} + +static void uncore_put_constraint(struct intel_uncore_box *box, struct perf_event *event) +{ + struct intel_uncore_extra_reg *er; + struct hw_perf_event_extra *reg1 = &event->hw.extra_reg; + + /* + * Only put constraint if extra reg was actually allocated. Also + * takes care of event which do not use an extra shared reg. + * + * Also, if this is a fake box we shouldn't touch any event state + * (reg->alloc) and we don't care about leaving inconsistent box + * state either since it will be thrown out. + */ + if (uncore_box_is_fake(box) || !reg1->alloc) + return; + + er = &box->shared_regs[reg1->idx]; + atomic_dec(&er->ref); + reg1->alloc = 0; +} + /* Sandy Bridge-EP uncore support */ static struct intel_uncore_type snbep_uncore_cbox; static struct intel_uncore_type snbep_uncore_pcu; @@ -64,18 +135,15 @@ static void snbep_uncore_pci_enable_box(struct intel_uncore_box *box) pci_write_config_dword(pdev, box_ctl, config); } -static void snbep_uncore_pci_enable_event(struct intel_uncore_box *box, - struct perf_event *event) +static void snbep_uncore_pci_enable_event(struct intel_uncore_box *box, struct perf_event *event) { struct pci_dev *pdev = box->pci_dev; struct hw_perf_event *hwc = &event->hw; - pci_write_config_dword(pdev, hwc->config_base, hwc->config | - SNBEP_PMON_CTL_EN); + pci_write_config_dword(pdev, hwc->config_base, hwc->config | SNBEP_PMON_CTL_EN); } -static void snbep_uncore_pci_disable_event(struct intel_uncore_box *box, - struct perf_event *event) +static void snbep_uncore_pci_disable_event(struct intel_uncore_box *box, struct perf_event *event) { struct pci_dev *pdev = box->pci_dev; struct hw_perf_event *hwc = &event->hw; @@ -83,8 +151,7 @@ static void snbep_uncore_pci_disable_event(struct intel_uncore_box *box, pci_write_config_dword(pdev, hwc->config_base, hwc->config); } -static u64 snbep_uncore_pci_read_counter(struct intel_uncore_box *box, - struct perf_event *event) +static u64 snbep_uncore_pci_read_counter(struct intel_uncore_box *box, struct perf_event *event) { struct pci_dev *pdev = box->pci_dev; struct hw_perf_event *hwc = &event->hw; @@ -92,14 +159,15 @@ static u64 snbep_uncore_pci_read_counter(struct intel_uncore_box *box, pci_read_config_dword(pdev, hwc->event_base, (u32 *)&count); pci_read_config_dword(pdev, hwc->event_base + 4, (u32 *)&count + 1); + return count; } static void snbep_uncore_pci_init_box(struct intel_uncore_box *box) { struct pci_dev *pdev = box->pci_dev; - pci_write_config_dword(pdev, SNBEP_PCI_PMON_BOX_CTL, - SNBEP_PMON_BOX_CTL_INT); + + pci_write_config_dword(pdev, SNBEP_PCI_PMON_BOX_CTL, SNBEP_PMON_BOX_CTL_INT); } static void snbep_uncore_msr_disable_box(struct intel_uncore_box *box) @@ -112,7 +180,6 @@ static void snbep_uncore_msr_disable_box(struct intel_uncore_box *box) rdmsrl(msr, config); config |= SNBEP_PMON_BOX_CTL_FRZ; wrmsrl(msr, config); - return; } } @@ -126,12 +193,10 @@ static void snbep_uncore_msr_enable_box(struct intel_uncore_box *box) rdmsrl(msr, config); config &= ~SNBEP_PMON_BOX_CTL_FRZ; wrmsrl(msr, config); - return; } } -static void snbep_uncore_msr_enable_event(struct intel_uncore_box *box, - struct perf_event *event) +static void snbep_uncore_msr_enable_event(struct intel_uncore_box *box, struct perf_event *event) { struct hw_perf_event *hwc = &event->hw; struct hw_perf_event_extra *reg1 = &hwc->extra_reg; @@ -150,68 +215,15 @@ static void snbep_uncore_msr_disable_event(struct intel_uncore_box *box, wrmsrl(hwc->config_base, hwc->config); } -static u64 snbep_uncore_msr_read_counter(struct intel_uncore_box *box, - struct perf_event *event) -{ - struct hw_perf_event *hwc = &event->hw; - u64 count; - - rdmsrl(hwc->event_base, count); - return count; -} - static void snbep_uncore_msr_init_box(struct intel_uncore_box *box) { unsigned msr = uncore_msr_box_ctl(box); + if (msr) wrmsrl(msr, SNBEP_PMON_BOX_CTL_INT); } -static struct event_constraint * -snbep_uncore_get_constraint(struct intel_uncore_box *box, - struct perf_event *event) -{ - struct intel_uncore_extra_reg *er; - struct hw_perf_event_extra *reg1 = &event->hw.extra_reg; - unsigned long flags; - bool ok = false; - - if (reg1->idx == EXTRA_REG_NONE || (box->phys_id >= 0 && reg1->alloc)) - return NULL; - - er = &box->shared_regs[reg1->idx]; - raw_spin_lock_irqsave(&er->lock, flags); - if (!atomic_read(&er->ref) || er->config1 == reg1->config) { - atomic_inc(&er->ref); - er->config1 = reg1->config; - ok = true; - } - raw_spin_unlock_irqrestore(&er->lock, flags); - - if (ok) { - if (box->phys_id >= 0) - reg1->alloc = 1; - return NULL; - } - return &constraint_empty; -} - -static void snbep_uncore_put_constraint(struct intel_uncore_box *box, - struct perf_event *event) -{ - struct intel_uncore_extra_reg *er; - struct hw_perf_event_extra *reg1 = &event->hw.extra_reg; - - if (box->phys_id < 0 || !reg1->alloc) - return; - - er = &box->shared_regs[reg1->idx]; - atomic_dec(&er->ref); - reg1->alloc = 0; -} - -static int snbep_uncore_hw_config(struct intel_uncore_box *box, - struct perf_event *event) +static int snbep_uncore_hw_config(struct intel_uncore_box *box, struct perf_event *event) { struct hw_perf_event *hwc = &event->hw; struct hw_perf_event_extra *reg1 = &hwc->extra_reg; @@ -221,14 +233,16 @@ static int snbep_uncore_hw_config(struct intel_uncore_box *box, SNBEP_CBO_MSR_OFFSET * box->pmu->pmu_idx; reg1->config = event->attr.config1 & SNBEP_CB0_MSR_PMON_BOX_FILTER_MASK; - } else if (box->pmu->type == &snbep_uncore_pcu) { - reg1->reg = SNBEP_PCU_MSR_PMON_BOX_FILTER; - reg1->config = event->attr.config1 & - SNBEP_PCU_MSR_PMON_BOX_FILTER_MASK; } else { - return 0; + if (box->pmu->type == &snbep_uncore_pcu) { + reg1->reg = SNBEP_PCU_MSR_PMON_BOX_FILTER; + reg1->config = event->attr.config1 & SNBEP_PCU_MSR_PMON_BOX_FILTER_MASK; + } else { + return 0; + } } reg1->idx = 0; + return 0; } @@ -320,9 +334,9 @@ static struct intel_uncore_ops snbep_uncore_msr_ops = { .enable_box = snbep_uncore_msr_enable_box, .disable_event = snbep_uncore_msr_disable_event, .enable_event = snbep_uncore_msr_enable_event, - .read_counter = snbep_uncore_msr_read_counter, - .get_constraint = snbep_uncore_get_constraint, - .put_constraint = snbep_uncore_put_constraint, + .read_counter = uncore_msr_read_counter, + .get_constraint = uncore_get_constraint, + .put_constraint = uncore_put_constraint, .hw_config = snbep_uncore_hw_config, }; @@ -589,188 +603,1208 @@ static void snbep_pci2phy_map_init(void) /* get the Node ID mapping */ pci_read_config_dword(ubox_dev, 0x54, &config); /* - * every three bits in the Node ID mapping register maps - * to a particular node. + * every three bits in the Node ID mapping register maps + * to a particular node. + */ + for (i = 0; i < 8; i++) { + if (nodeid == ((config >> (3 * i)) & 0x7)) { + pcibus_to_physid[bus] = i; + break; + } + } + }; + return; +} +/* end of Sandy Bridge-EP uncore support */ + +/* Sandy Bridge uncore support */ +static void snb_uncore_msr_enable_event(struct intel_uncore_box *box, struct perf_event *event) +{ + struct hw_perf_event *hwc = &event->hw; + + if (hwc->idx < UNCORE_PMC_IDX_FIXED) + wrmsrl(hwc->config_base, hwc->config | SNB_UNC_CTL_EN); + else + wrmsrl(hwc->config_base, SNB_UNC_CTL_EN); +} + +static void snb_uncore_msr_disable_event(struct intel_uncore_box *box, struct perf_event *event) +{ + wrmsrl(event->hw.config_base, 0); +} + +static void snb_uncore_msr_init_box(struct intel_uncore_box *box) +{ + if (box->pmu->pmu_idx == 0) { + wrmsrl(SNB_UNC_PERF_GLOBAL_CTL, + SNB_UNC_GLOBAL_CTL_EN | SNB_UNC_GLOBAL_CTL_CORE_ALL); + } +} + +static struct attribute *snb_uncore_formats_attr[] = { + &format_attr_event.attr, + &format_attr_umask.attr, + &format_attr_edge.attr, + &format_attr_inv.attr, + &format_attr_cmask5.attr, + NULL, +}; + +static struct attribute_group snb_uncore_format_group = { + .name = "format", + .attrs = snb_uncore_formats_attr, +}; + +static struct intel_uncore_ops snb_uncore_msr_ops = { + .init_box = snb_uncore_msr_init_box, + .disable_event = snb_uncore_msr_disable_event, + .enable_event = snb_uncore_msr_enable_event, + .read_counter = uncore_msr_read_counter, +}; + +static struct event_constraint snb_uncore_cbox_constraints[] = { + UNCORE_EVENT_CONSTRAINT(0x80, 0x1), + UNCORE_EVENT_CONSTRAINT(0x83, 0x1), + EVENT_CONSTRAINT_END +}; + +static struct intel_uncore_type snb_uncore_cbox = { + .name = "cbox", + .num_counters = 2, + .num_boxes = 4, + .perf_ctr_bits = 44, + .fixed_ctr_bits = 48, + .perf_ctr = SNB_UNC_CBO_0_PER_CTR0, + .event_ctl = SNB_UNC_CBO_0_PERFEVTSEL0, + .fixed_ctr = SNB_UNC_FIXED_CTR, + .fixed_ctl = SNB_UNC_FIXED_CTR_CTRL, + .single_fixed = 1, + .event_mask = SNB_UNC_RAW_EVENT_MASK, + .msr_offset = SNB_UNC_CBO_MSR_OFFSET, + .constraints = snb_uncore_cbox_constraints, + .ops = &snb_uncore_msr_ops, + .format_group = &snb_uncore_format_group, +}; + +static struct intel_uncore_type *snb_msr_uncores[] = { + &snb_uncore_cbox, + NULL, +}; +/* end of Sandy Bridge uncore support */ + +/* Nehalem uncore support */ +static void nhm_uncore_msr_disable_box(struct intel_uncore_box *box) +{ + wrmsrl(NHM_UNC_PERF_GLOBAL_CTL, 0); +} + +static void nhm_uncore_msr_enable_box(struct intel_uncore_box *box) +{ + wrmsrl(NHM_UNC_PERF_GLOBAL_CTL, NHM_UNC_GLOBAL_CTL_EN_PC_ALL | NHM_UNC_GLOBAL_CTL_EN_FC); +} + +static void nhm_uncore_msr_enable_event(struct intel_uncore_box *box, struct perf_event *event) +{ + struct hw_perf_event *hwc = &event->hw; + + if (hwc->idx < UNCORE_PMC_IDX_FIXED) + wrmsrl(hwc->config_base, hwc->config | SNB_UNC_CTL_EN); + else + wrmsrl(hwc->config_base, NHM_UNC_FIXED_CTR_CTL_EN); +} + +static struct attribute *nhm_uncore_formats_attr[] = { + &format_attr_event.attr, + &format_attr_umask.attr, + &format_attr_edge.attr, + &format_attr_inv.attr, + &format_attr_cmask8.attr, + NULL, +}; + +static struct attribute_group nhm_uncore_format_group = { + .name = "format", + .attrs = nhm_uncore_formats_attr, +}; + +static struct uncore_event_desc nhm_uncore_events[] = { + INTEL_UNCORE_EVENT_DESC(clockticks, "event=0xff,umask=0x00"), + INTEL_UNCORE_EVENT_DESC(qmc_writes_full_any, "event=0x2f,umask=0x0f"), + INTEL_UNCORE_EVENT_DESC(qmc_normal_reads_any, "event=0x2c,umask=0x0f"), + INTEL_UNCORE_EVENT_DESC(qhl_request_ioh_reads, "event=0x20,umask=0x01"), + INTEL_UNCORE_EVENT_DESC(qhl_request_ioh_writes, "event=0x20,umask=0x02"), + INTEL_UNCORE_EVENT_DESC(qhl_request_remote_reads, "event=0x20,umask=0x04"), + INTEL_UNCORE_EVENT_DESC(qhl_request_remote_writes, "event=0x20,umask=0x08"), + INTEL_UNCORE_EVENT_DESC(qhl_request_local_reads, "event=0x20,umask=0x10"), + INTEL_UNCORE_EVENT_DESC(qhl_request_local_writes, "event=0x20,umask=0x20"), + { /* end: all zeroes */ }, +}; + +static struct intel_uncore_ops nhm_uncore_msr_ops = { + .disable_box = nhm_uncore_msr_disable_box, + .enable_box = nhm_uncore_msr_enable_box, + .disable_event = snb_uncore_msr_disable_event, + .enable_event = nhm_uncore_msr_enable_event, + .read_counter = uncore_msr_read_counter, +}; + +static struct intel_uncore_type nhm_uncore = { + .name = "", + .num_counters = 8, + .num_boxes = 1, + .perf_ctr_bits = 48, + .fixed_ctr_bits = 48, + .event_ctl = NHM_UNC_PERFEVTSEL0, + .perf_ctr = NHM_UNC_UNCORE_PMC0, + .fixed_ctr = NHM_UNC_FIXED_CTR, + .fixed_ctl = NHM_UNC_FIXED_CTR_CTRL, + .event_mask = NHM_UNC_RAW_EVENT_MASK, + .event_descs = nhm_uncore_events, + .ops = &nhm_uncore_msr_ops, + .format_group = &nhm_uncore_format_group, +}; + +static struct intel_uncore_type *nhm_msr_uncores[] = { + &nhm_uncore, + NULL, +}; +/* end of Nehalem uncore support */ + +/* Nehalem-EX uncore support */ +#define __BITS_VALUE(x, i, n) ((typeof(x))(((x) >> ((i) * (n))) & \ + ((1ULL << (n)) - 1))) + +DEFINE_UNCORE_FORMAT_ATTR(event5, event, "config:1-5"); +DEFINE_UNCORE_FORMAT_ATTR(counter, counter, "config:6-7"); +DEFINE_UNCORE_FORMAT_ATTR(mm_cfg, mm_cfg, "config:63"); +DEFINE_UNCORE_FORMAT_ATTR(match, match, "config1:0-63"); +DEFINE_UNCORE_FORMAT_ATTR(mask, mask, "config2:0-63"); + +static void nhmex_uncore_msr_init_box(struct intel_uncore_box *box) +{ + wrmsrl(NHMEX_U_MSR_PMON_GLOBAL_CTL, NHMEX_U_PMON_GLOBAL_EN_ALL); +} + +static void nhmex_uncore_msr_disable_box(struct intel_uncore_box *box) +{ + unsigned msr = uncore_msr_box_ctl(box); + u64 config; + + if (msr) { + rdmsrl(msr, config); + config &= ~((1ULL << uncore_num_counters(box)) - 1); + /* WBox has a fixed counter */ + if (uncore_msr_fixed_ctl(box)) + config &= ~NHMEX_W_PMON_GLOBAL_FIXED_EN; + wrmsrl(msr, config); + } +} + +static void nhmex_uncore_msr_enable_box(struct intel_uncore_box *box) +{ + unsigned msr = uncore_msr_box_ctl(box); + u64 config; + + if (msr) { + rdmsrl(msr, config); + config |= (1ULL << uncore_num_counters(box)) - 1; + /* WBox has a fixed counter */ + if (uncore_msr_fixed_ctl(box)) + config |= NHMEX_W_PMON_GLOBAL_FIXED_EN; + wrmsrl(msr, config); + } +} + +static void nhmex_uncore_msr_disable_event(struct intel_uncore_box *box, struct perf_event *event) +{ + wrmsrl(event->hw.config_base, 0); +} + +static void nhmex_uncore_msr_enable_event(struct intel_uncore_box *box, struct perf_event *event) +{ + struct hw_perf_event *hwc = &event->hw; + + if (hwc->idx >= UNCORE_PMC_IDX_FIXED) + wrmsrl(hwc->config_base, NHMEX_PMON_CTL_EN_BIT0); + else if (box->pmu->type->event_mask & NHMEX_PMON_CTL_EN_BIT0) + wrmsrl(hwc->config_base, hwc->config | NHMEX_PMON_CTL_EN_BIT22); + else + wrmsrl(hwc->config_base, hwc->config | NHMEX_PMON_CTL_EN_BIT0); +} + +#define NHMEX_UNCORE_OPS_COMMON_INIT() \ + .init_box = nhmex_uncore_msr_init_box, \ + .disable_box = nhmex_uncore_msr_disable_box, \ + .enable_box = nhmex_uncore_msr_enable_box, \ + .disable_event = nhmex_uncore_msr_disable_event, \ + .read_counter = uncore_msr_read_counter + +static struct intel_uncore_ops nhmex_uncore_ops = { + NHMEX_UNCORE_OPS_COMMON_INIT(), + .enable_event = nhmex_uncore_msr_enable_event, +}; + +static struct attribute *nhmex_uncore_ubox_formats_attr[] = { + &format_attr_event.attr, + &format_attr_edge.attr, + NULL, +}; + +static struct attribute_group nhmex_uncore_ubox_format_group = { + .name = "format", + .attrs = nhmex_uncore_ubox_formats_attr, +}; + +static struct intel_uncore_type nhmex_uncore_ubox = { + .name = "ubox", + .num_counters = 1, + .num_boxes = 1, + .perf_ctr_bits = 48, + .event_ctl = NHMEX_U_MSR_PMON_EV_SEL, + .perf_ctr = NHMEX_U_MSR_PMON_CTR, + .event_mask = NHMEX_U_PMON_RAW_EVENT_MASK, + .box_ctl = NHMEX_U_MSR_PMON_GLOBAL_CTL, + .ops = &nhmex_uncore_ops, + .format_group = &nhmex_uncore_ubox_format_group +}; + +static struct attribute *nhmex_uncore_cbox_formats_attr[] = { + &format_attr_event.attr, + &format_attr_umask.attr, + &format_attr_edge.attr, + &format_attr_inv.attr, + &format_attr_thresh8.attr, + NULL, +}; + +static struct attribute_group nhmex_uncore_cbox_format_group = { + .name = "format", + .attrs = nhmex_uncore_cbox_formats_attr, +}; + +static struct intel_uncore_type nhmex_uncore_cbox = { + .name = "cbox", + .num_counters = 6, + .num_boxes = 8, + .perf_ctr_bits = 48, + .event_ctl = NHMEX_C0_MSR_PMON_EV_SEL0, + .perf_ctr = NHMEX_C0_MSR_PMON_CTR0, + .event_mask = NHMEX_PMON_RAW_EVENT_MASK, + .box_ctl = NHMEX_C0_MSR_PMON_GLOBAL_CTL, + .msr_offset = NHMEX_C_MSR_OFFSET, + .pair_ctr_ctl = 1, + .ops = &nhmex_uncore_ops, + .format_group = &nhmex_uncore_cbox_format_group +}; + +static struct uncore_event_desc nhmex_uncore_wbox_events[] = { + INTEL_UNCORE_EVENT_DESC(clockticks, "event=0xff,umask=0"), + { /* end: all zeroes */ }, +}; + +static struct intel_uncore_type nhmex_uncore_wbox = { + .name = "wbox", + .num_counters = 4, + .num_boxes = 1, + .perf_ctr_bits = 48, + .event_ctl = NHMEX_W_MSR_PMON_CNT0, + .perf_ctr = NHMEX_W_MSR_PMON_EVT_SEL0, + .fixed_ctr = NHMEX_W_MSR_PMON_FIXED_CTR, + .fixed_ctl = NHMEX_W_MSR_PMON_FIXED_CTL, + .event_mask = NHMEX_PMON_RAW_EVENT_MASK, + .box_ctl = NHMEX_W_MSR_GLOBAL_CTL, + .pair_ctr_ctl = 1, + .event_descs = nhmex_uncore_wbox_events, + .ops = &nhmex_uncore_ops, + .format_group = &nhmex_uncore_cbox_format_group +}; + +static int nhmex_bbox_hw_config(struct intel_uncore_box *box, struct perf_event *event) +{ + struct hw_perf_event *hwc = &event->hw; + struct hw_perf_event_extra *reg1 = &hwc->extra_reg; + struct hw_perf_event_extra *reg2 = &hwc->branch_reg; + int ctr, ev_sel; + + ctr = (hwc->config & NHMEX_B_PMON_CTR_MASK) >> + NHMEX_B_PMON_CTR_SHIFT; + ev_sel = (hwc->config & NHMEX_B_PMON_CTL_EV_SEL_MASK) >> + NHMEX_B_PMON_CTL_EV_SEL_SHIFT; + + /* events that do not use the match/mask registers */ + if ((ctr == 0 && ev_sel > 0x3) || (ctr == 1 && ev_sel > 0x6) || + (ctr == 2 && ev_sel != 0x4) || ctr == 3) + return 0; + + if (box->pmu->pmu_idx == 0) + reg1->reg = NHMEX_B0_MSR_MATCH; + else + reg1->reg = NHMEX_B1_MSR_MATCH; + reg1->idx = 0; + reg1->config = event->attr.config1; + reg2->config = event->attr.config2; + return 0; +} + +static void nhmex_bbox_msr_enable_event(struct intel_uncore_box *box, struct perf_event *event) +{ + struct hw_perf_event *hwc = &event->hw; + struct hw_perf_event_extra *reg1 = &hwc->extra_reg; + struct hw_perf_event_extra *reg2 = &hwc->branch_reg; + + if (reg1->idx != EXTRA_REG_NONE) { + wrmsrl(reg1->reg, reg1->config); + wrmsrl(reg1->reg + 1, reg2->config); + } + wrmsrl(hwc->config_base, NHMEX_PMON_CTL_EN_BIT0 | + (hwc->config & NHMEX_B_PMON_CTL_EV_SEL_MASK)); +} + +/* + * The Bbox has 4 counters, but each counter monitors different events. + * Use bits 6-7 in the event config to select counter. + */ +static struct event_constraint nhmex_uncore_bbox_constraints[] = { + EVENT_CONSTRAINT(0 , 1, 0xc0), + EVENT_CONSTRAINT(0x40, 2, 0xc0), + EVENT_CONSTRAINT(0x80, 4, 0xc0), + EVENT_CONSTRAINT(0xc0, 8, 0xc0), + EVENT_CONSTRAINT_END, +}; + +static struct attribute *nhmex_uncore_bbox_formats_attr[] = { + &format_attr_event5.attr, + &format_attr_counter.attr, + &format_attr_match.attr, + &format_attr_mask.attr, + NULL, +}; + +static struct attribute_group nhmex_uncore_bbox_format_group = { + .name = "format", + .attrs = nhmex_uncore_bbox_formats_attr, +}; + +static struct intel_uncore_ops nhmex_uncore_bbox_ops = { + NHMEX_UNCORE_OPS_COMMON_INIT(), + .enable_event = nhmex_bbox_msr_enable_event, + .hw_config = nhmex_bbox_hw_config, + .get_constraint = uncore_get_constraint, + .put_constraint = uncore_put_constraint, +}; + +static struct intel_uncore_type nhmex_uncore_bbox = { + .name = "bbox", + .num_counters = 4, + .num_boxes = 2, + .perf_ctr_bits = 48, + .event_ctl = NHMEX_B0_MSR_PMON_CTL0, + .perf_ctr = NHMEX_B0_MSR_PMON_CTR0, + .event_mask = NHMEX_B_PMON_RAW_EVENT_MASK, + .box_ctl = NHMEX_B0_MSR_PMON_GLOBAL_CTL, + .msr_offset = NHMEX_B_MSR_OFFSET, + .pair_ctr_ctl = 1, + .num_shared_regs = 1, + .constraints = nhmex_uncore_bbox_constraints, + .ops = &nhmex_uncore_bbox_ops, + .format_group = &nhmex_uncore_bbox_format_group +}; + +static int nhmex_sbox_hw_config(struct intel_uncore_box *box, struct perf_event *event) +{ + struct hw_perf_event_extra *reg1 = &event->hw.extra_reg; + struct hw_perf_event_extra *reg2 = &event->hw.branch_reg; + + if (event->attr.config & NHMEX_S_PMON_MM_CFG_EN) { + reg1->config = event->attr.config1; + reg2->config = event->attr.config2; + } else { + reg1->config = ~0ULL; + reg2->config = ~0ULL; + } + + if (box->pmu->pmu_idx == 0) + reg1->reg = NHMEX_S0_MSR_MM_CFG; + else + reg1->reg = NHMEX_S1_MSR_MM_CFG; + + reg1->idx = 0; + + return 0; +} + +static void nhmex_sbox_msr_enable_event(struct intel_uncore_box *box, struct perf_event *event) +{ + struct hw_perf_event *hwc = &event->hw; + struct hw_perf_event_extra *reg1 = &hwc->extra_reg; + struct hw_perf_event_extra *reg2 = &hwc->branch_reg; + + wrmsrl(reg1->reg, 0); + if (reg1->config != ~0ULL || reg2->config != ~0ULL) { + wrmsrl(reg1->reg + 1, reg1->config); + wrmsrl(reg1->reg + 2, reg2->config); + wrmsrl(reg1->reg, NHMEX_S_PMON_MM_CFG_EN); + } + wrmsrl(hwc->config_base, hwc->config | NHMEX_PMON_CTL_EN_BIT22); +} + +static struct attribute *nhmex_uncore_sbox_formats_attr[] = { + &format_attr_event.attr, + &format_attr_umask.attr, + &format_attr_edge.attr, + &format_attr_inv.attr, + &format_attr_thresh8.attr, + &format_attr_mm_cfg.attr, + &format_attr_match.attr, + &format_attr_mask.attr, + NULL, +}; + +static struct attribute_group nhmex_uncore_sbox_format_group = { + .name = "format", + .attrs = nhmex_uncore_sbox_formats_attr, +}; + +static struct intel_uncore_ops nhmex_uncore_sbox_ops = { + NHMEX_UNCORE_OPS_COMMON_INIT(), + .enable_event = nhmex_sbox_msr_enable_event, + .hw_config = nhmex_sbox_hw_config, + .get_constraint = uncore_get_constraint, + .put_constraint = uncore_put_constraint, +}; + +static struct intel_uncore_type nhmex_uncore_sbox = { + .name = "sbox", + .num_counters = 4, + .num_boxes = 2, + .perf_ctr_bits = 48, + .event_ctl = NHMEX_S0_MSR_PMON_CTL0, + .perf_ctr = NHMEX_S0_MSR_PMON_CTR0, + .event_mask = NHMEX_PMON_RAW_EVENT_MASK, + .box_ctl = NHMEX_S0_MSR_PMON_GLOBAL_CTL, + .msr_offset = NHMEX_S_MSR_OFFSET, + .pair_ctr_ctl = 1, + .num_shared_regs = 1, + .ops = &nhmex_uncore_sbox_ops, + .format_group = &nhmex_uncore_sbox_format_group +}; + +enum { + EXTRA_REG_NHMEX_M_FILTER, + EXTRA_REG_NHMEX_M_DSP, + EXTRA_REG_NHMEX_M_ISS, + EXTRA_REG_NHMEX_M_MAP, + EXTRA_REG_NHMEX_M_MSC_THR, + EXTRA_REG_NHMEX_M_PGT, + EXTRA_REG_NHMEX_M_PLD, + EXTRA_REG_NHMEX_M_ZDP_CTL_FVC, +}; + +static struct extra_reg nhmex_uncore_mbox_extra_regs[] = { + MBOX_INC_SEL_EXTAR_REG(0x0, DSP), + MBOX_INC_SEL_EXTAR_REG(0x4, MSC_THR), + MBOX_INC_SEL_EXTAR_REG(0x5, MSC_THR), + MBOX_INC_SEL_EXTAR_REG(0x9, ISS), + /* event 0xa uses two extra registers */ + MBOX_INC_SEL_EXTAR_REG(0xa, ISS), + MBOX_INC_SEL_EXTAR_REG(0xa, PLD), + MBOX_INC_SEL_EXTAR_REG(0xb, PLD), + /* events 0xd ~ 0x10 use the same extra register */ + MBOX_INC_SEL_EXTAR_REG(0xd, ZDP_CTL_FVC), + MBOX_INC_SEL_EXTAR_REG(0xe, ZDP_CTL_FVC), + MBOX_INC_SEL_EXTAR_REG(0xf, ZDP_CTL_FVC), + MBOX_INC_SEL_EXTAR_REG(0x10, ZDP_CTL_FVC), + MBOX_INC_SEL_EXTAR_REG(0x16, PGT), + MBOX_SET_FLAG_SEL_EXTRA_REG(0x0, DSP), + MBOX_SET_FLAG_SEL_EXTRA_REG(0x1, ISS), + MBOX_SET_FLAG_SEL_EXTRA_REG(0x5, PGT), + MBOX_SET_FLAG_SEL_EXTRA_REG(0x6, MAP), + EVENT_EXTRA_END +}; + +static bool nhmex_mbox_get_shared_reg(struct intel_uncore_box *box, int idx, u64 config) +{ + struct intel_uncore_extra_reg *er; + unsigned long flags; + bool ret = false; + u64 mask; + + if (idx < EXTRA_REG_NHMEX_M_ZDP_CTL_FVC) { + er = &box->shared_regs[idx]; + raw_spin_lock_irqsave(&er->lock, flags); + if (!atomic_read(&er->ref) || er->config == config) { + atomic_inc(&er->ref); + er->config = config; + ret = true; + } + raw_spin_unlock_irqrestore(&er->lock, flags); + + return ret; + } + /* + * The ZDP_CTL_FVC MSR has 4 fields which are used to control + * events 0xd ~ 0x10. Besides these 4 fields, there are additional + * fields which are shared. + */ + idx -= EXTRA_REG_NHMEX_M_ZDP_CTL_FVC; + if (WARN_ON_ONCE(idx >= 4)) + return false; + + /* mask of the shared fields */ + mask = NHMEX_M_PMON_ZDP_CTL_FVC_MASK; + er = &box->shared_regs[EXTRA_REG_NHMEX_M_ZDP_CTL_FVC]; + + raw_spin_lock_irqsave(&er->lock, flags); + /* add mask of the non-shared field if it's in use */ + if (__BITS_VALUE(atomic_read(&er->ref), idx, 8)) + mask |= NHMEX_M_PMON_ZDP_CTL_FVC_EVENT_MASK(idx); + + if (!atomic_read(&er->ref) || !((er->config ^ config) & mask)) { + atomic_add(1 << (idx * 8), &er->ref); + mask = NHMEX_M_PMON_ZDP_CTL_FVC_MASK | + NHMEX_M_PMON_ZDP_CTL_FVC_EVENT_MASK(idx); + er->config &= ~mask; + er->config |= (config & mask); + ret = true; + } + raw_spin_unlock_irqrestore(&er->lock, flags); + + return ret; +} + +static void nhmex_mbox_put_shared_reg(struct intel_uncore_box *box, int idx) +{ + struct intel_uncore_extra_reg *er; + + if (idx < EXTRA_REG_NHMEX_M_ZDP_CTL_FVC) { + er = &box->shared_regs[idx]; + atomic_dec(&er->ref); + return; + } + + idx -= EXTRA_REG_NHMEX_M_ZDP_CTL_FVC; + er = &box->shared_regs[EXTRA_REG_NHMEX_M_ZDP_CTL_FVC]; + atomic_sub(1 << (idx * 8), &er->ref); +} + +u64 nhmex_mbox_alter_er(struct perf_event *event, int new_idx, bool modify) +{ + struct hw_perf_event *hwc = &event->hw; + struct hw_perf_event_extra *reg1 = &hwc->extra_reg; + int idx, orig_idx = __BITS_VALUE(reg1->idx, 0, 8); + u64 config = reg1->config; + + /* get the non-shared control bits and shift them */ + idx = orig_idx - EXTRA_REG_NHMEX_M_ZDP_CTL_FVC; + config &= NHMEX_M_PMON_ZDP_CTL_FVC_EVENT_MASK(idx); + if (new_idx > orig_idx) { + idx = new_idx - orig_idx; + config <<= 3 * idx; + } else { + idx = orig_idx - new_idx; + config >>= 3 * idx; + } + + /* add the shared control bits back */ + config |= NHMEX_M_PMON_ZDP_CTL_FVC_MASK & reg1->config; + if (modify) { + /* adjust the main event selector */ + if (new_idx > orig_idx) + hwc->config += idx << NHMEX_M_PMON_CTL_INC_SEL_SHIFT; + else + hwc->config -= idx << NHMEX_M_PMON_CTL_INC_SEL_SHIFT; + reg1->config = config; + reg1->idx = ~0xff | new_idx; + } + return config; +} + +static struct event_constraint * +nhmex_mbox_get_constraint(struct intel_uncore_box *box, struct perf_event *event) +{ + struct hw_perf_event_extra *reg1 = &event->hw.extra_reg; + struct hw_perf_event_extra *reg2 = &event->hw.branch_reg; + int i, idx[2], alloc = 0; + u64 config1 = reg1->config; + + idx[0] = __BITS_VALUE(reg1->idx, 0, 8); + idx[1] = __BITS_VALUE(reg1->idx, 1, 8); +again: + for (i = 0; i < 2; i++) { + if (!uncore_box_is_fake(box) && (reg1->alloc & (0x1 << i))) + idx[i] = 0xff; + + if (idx[i] == 0xff) + continue; + + if (!nhmex_mbox_get_shared_reg(box, idx[i], + __BITS_VALUE(config1, i, 32))) + goto fail; + alloc |= (0x1 << i); + } + + /* for the match/mask registers */ + if ((uncore_box_is_fake(box) || !reg2->alloc) && + !nhmex_mbox_get_shared_reg(box, reg2->idx, reg2->config)) + goto fail; + + /* + * If it's a fake box -- as per validate_{group,event}() we + * shouldn't touch event state and we can avoid doing so + * since both will only call get_event_constraints() once + * on each event, this avoids the need for reg->alloc. + */ + if (!uncore_box_is_fake(box)) { + if (idx[0] != 0xff && idx[0] != __BITS_VALUE(reg1->idx, 0, 8)) + nhmex_mbox_alter_er(event, idx[0], true); + reg1->alloc |= alloc; + reg2->alloc = 1; + } + return NULL; +fail: + if (idx[0] != 0xff && !(alloc & 0x1) && + idx[0] >= EXTRA_REG_NHMEX_M_ZDP_CTL_FVC) { + /* + * events 0xd ~ 0x10 are functional identical, but are + * controlled by different fields in the ZDP_CTL_FVC + * register. If we failed to take one field, try the + * rest 3 choices. */ - for (i = 0; i < 8; i++) { - if (nodeid == ((config >> (3 * i)) & 0x7)) { - pcibus_to_physid[bus] = i; - break; - } + BUG_ON(__BITS_VALUE(reg1->idx, 1, 8) != 0xff); + idx[0] -= EXTRA_REG_NHMEX_M_ZDP_CTL_FVC; + idx[0] = (idx[0] + 1) % 4; + idx[0] += EXTRA_REG_NHMEX_M_ZDP_CTL_FVC; + if (idx[0] != __BITS_VALUE(reg1->idx, 0, 8)) { + config1 = nhmex_mbox_alter_er(event, idx[0], false); + goto again; } - }; - return; -} -/* end of Sandy Bridge-EP uncore support */ + } + if (alloc & 0x1) + nhmex_mbox_put_shared_reg(box, idx[0]); + if (alloc & 0x2) + nhmex_mbox_put_shared_reg(box, idx[1]); + return &constraint_empty; +} -/* Sandy Bridge uncore support */ -static void snb_uncore_msr_enable_event(struct intel_uncore_box *box, - struct perf_event *event) +static void nhmex_mbox_put_constraint(struct intel_uncore_box *box, struct perf_event *event) { - struct hw_perf_event *hwc = &event->hw; + struct hw_perf_event_extra *reg1 = &event->hw.extra_reg; + struct hw_perf_event_extra *reg2 = &event->hw.branch_reg; - if (hwc->idx < UNCORE_PMC_IDX_FIXED) - wrmsrl(hwc->config_base, hwc->config | SNB_UNC_CTL_EN); - else - wrmsrl(hwc->config_base, SNB_UNC_CTL_EN); + if (uncore_box_is_fake(box)) + return; + + if (reg1->alloc & 0x1) + nhmex_mbox_put_shared_reg(box, __BITS_VALUE(reg1->idx, 0, 8)); + if (reg1->alloc & 0x2) + nhmex_mbox_put_shared_reg(box, __BITS_VALUE(reg1->idx, 1, 8)); + reg1->alloc = 0; + + if (reg2->alloc) { + nhmex_mbox_put_shared_reg(box, reg2->idx); + reg2->alloc = 0; + } } -static void snb_uncore_msr_disable_event(struct intel_uncore_box *box, - struct perf_event *event) +static int nhmex_mbox_extra_reg_idx(struct extra_reg *er) { - wrmsrl(event->hw.config_base, 0); + if (er->idx < EXTRA_REG_NHMEX_M_ZDP_CTL_FVC) + return er->idx; + return er->idx + (er->event >> NHMEX_M_PMON_CTL_INC_SEL_SHIFT) - 0xd; } -static u64 snb_uncore_msr_read_counter(struct intel_uncore_box *box, - struct perf_event *event) +static int nhmex_mbox_hw_config(struct intel_uncore_box *box, struct perf_event *event) { - u64 count; - rdmsrl(event->hw.event_base, count); - return count; + struct intel_uncore_type *type = box->pmu->type; + struct hw_perf_event_extra *reg1 = &event->hw.extra_reg; + struct hw_perf_event_extra *reg2 = &event->hw.branch_reg; + struct extra_reg *er; + unsigned msr; + int reg_idx = 0; + + if (WARN_ON_ONCE(reg1->idx != -1)) + return -EINVAL; + /* + * The mbox events may require 2 extra MSRs at the most. But only + * the lower 32 bits in these MSRs are significant, so we can use + * config1 to pass two MSRs' config. + */ + for (er = nhmex_uncore_mbox_extra_regs; er->msr; er++) { + if (er->event != (event->hw.config & er->config_mask)) + continue; + if (event->attr.config1 & ~er->valid_mask) + return -EINVAL; + if (er->idx == __BITS_VALUE(reg1->idx, 0, 8) || + er->idx == __BITS_VALUE(reg1->idx, 1, 8)) + continue; + if (WARN_ON_ONCE(reg_idx >= 2)) + return -EINVAL; + + msr = er->msr + type->msr_offset * box->pmu->pmu_idx; + if (WARN_ON_ONCE(msr >= 0xffff || er->idx >= 0xff)) + return -EINVAL; + + /* always use the 32~63 bits to pass the PLD config */ + if (er->idx == EXTRA_REG_NHMEX_M_PLD) + reg_idx = 1; + + reg1->idx &= ~(0xff << (reg_idx * 8)); + reg1->reg &= ~(0xffff << (reg_idx * 16)); + reg1->idx |= nhmex_mbox_extra_reg_idx(er) << (reg_idx * 8); + reg1->reg |= msr << (reg_idx * 16); + reg1->config = event->attr.config1; + reg_idx++; + } + /* use config2 to pass the filter config */ + reg2->idx = EXTRA_REG_NHMEX_M_FILTER; + if (event->attr.config2 & NHMEX_M_PMON_MM_CFG_EN) + reg2->config = event->attr.config2; + else + reg2->config = ~0ULL; + if (box->pmu->pmu_idx == 0) + reg2->reg = NHMEX_M0_MSR_PMU_MM_CFG; + else + reg2->reg = NHMEX_M1_MSR_PMU_MM_CFG; + + return 0; } -static void snb_uncore_msr_init_box(struct intel_uncore_box *box) +static u64 nhmex_mbox_shared_reg_config(struct intel_uncore_box *box, int idx) { - if (box->pmu->pmu_idx == 0) { - wrmsrl(SNB_UNC_PERF_GLOBAL_CTL, - SNB_UNC_GLOBAL_CTL_EN | SNB_UNC_GLOBAL_CTL_CORE_ALL); - } + struct intel_uncore_extra_reg *er; + unsigned long flags; + u64 config; + + if (idx < EXTRA_REG_NHMEX_M_ZDP_CTL_FVC) + return box->shared_regs[idx].config; + + er = &box->shared_regs[EXTRA_REG_NHMEX_M_ZDP_CTL_FVC]; + raw_spin_lock_irqsave(&er->lock, flags); + config = er->config; + raw_spin_unlock_irqrestore(&er->lock, flags); + return config; } -static struct attribute *snb_uncore_formats_attr[] = { - &format_attr_event.attr, - &format_attr_umask.attr, - &format_attr_edge.attr, - &format_attr_inv.attr, - &format_attr_cmask5.attr, +static void nhmex_mbox_msr_enable_event(struct intel_uncore_box *box, struct perf_event *event) +{ + struct hw_perf_event *hwc = &event->hw; + struct hw_perf_event_extra *reg1 = &hwc->extra_reg; + struct hw_perf_event_extra *reg2 = &hwc->branch_reg; + int idx; + + idx = __BITS_VALUE(reg1->idx, 0, 8); + if (idx != 0xff) + wrmsrl(__BITS_VALUE(reg1->reg, 0, 16), + nhmex_mbox_shared_reg_config(box, idx)); + idx = __BITS_VALUE(reg1->idx, 1, 8); + if (idx != 0xff) + wrmsrl(__BITS_VALUE(reg1->reg, 1, 16), + nhmex_mbox_shared_reg_config(box, idx)); + + wrmsrl(reg2->reg, 0); + if (reg2->config != ~0ULL) { + wrmsrl(reg2->reg + 1, + reg2->config & NHMEX_M_PMON_ADDR_MATCH_MASK); + wrmsrl(reg2->reg + 2, NHMEX_M_PMON_ADDR_MASK_MASK & + (reg2->config >> NHMEX_M_PMON_ADDR_MASK_SHIFT)); + wrmsrl(reg2->reg, NHMEX_M_PMON_MM_CFG_EN); + } + + wrmsrl(hwc->config_base, hwc->config | NHMEX_PMON_CTL_EN_BIT0); +} + +DEFINE_UNCORE_FORMAT_ATTR(count_mode, count_mode, "config:2-3"); +DEFINE_UNCORE_FORMAT_ATTR(storage_mode, storage_mode, "config:4-5"); +DEFINE_UNCORE_FORMAT_ATTR(wrap_mode, wrap_mode, "config:6"); +DEFINE_UNCORE_FORMAT_ATTR(flag_mode, flag_mode, "config:7"); +DEFINE_UNCORE_FORMAT_ATTR(inc_sel, inc_sel, "config:9-13"); +DEFINE_UNCORE_FORMAT_ATTR(set_flag_sel, set_flag_sel, "config:19-21"); +DEFINE_UNCORE_FORMAT_ATTR(filter_cfg, filter_cfg, "config2:63"); +DEFINE_UNCORE_FORMAT_ATTR(filter_match, filter_match, "config2:0-33"); +DEFINE_UNCORE_FORMAT_ATTR(filter_mask, filter_mask, "config2:34-61"); +DEFINE_UNCORE_FORMAT_ATTR(dsp, dsp, "config1:0-31"); +DEFINE_UNCORE_FORMAT_ATTR(thr, thr, "config1:0-31"); +DEFINE_UNCORE_FORMAT_ATTR(fvc, fvc, "config1:0-31"); +DEFINE_UNCORE_FORMAT_ATTR(pgt, pgt, "config1:0-31"); +DEFINE_UNCORE_FORMAT_ATTR(map, map, "config1:0-31"); +DEFINE_UNCORE_FORMAT_ATTR(iss, iss, "config1:0-31"); +DEFINE_UNCORE_FORMAT_ATTR(pld, pld, "config1:32-63"); + +static struct attribute *nhmex_uncore_mbox_formats_attr[] = { + &format_attr_count_mode.attr, + &format_attr_storage_mode.attr, + &format_attr_wrap_mode.attr, + &format_attr_flag_mode.attr, + &format_attr_inc_sel.attr, + &format_attr_set_flag_sel.attr, + &format_attr_filter_cfg.attr, + &format_attr_filter_match.attr, + &format_attr_filter_mask.attr, + &format_attr_dsp.attr, + &format_attr_thr.attr, + &format_attr_fvc.attr, + &format_attr_pgt.attr, + &format_attr_map.attr, + &format_attr_iss.attr, + &format_attr_pld.attr, NULL, }; -static struct attribute_group snb_uncore_format_group = { - .name = "format", - .attrs = snb_uncore_formats_attr, +static struct attribute_group nhmex_uncore_mbox_format_group = { + .name = "format", + .attrs = nhmex_uncore_mbox_formats_attr, }; -static struct intel_uncore_ops snb_uncore_msr_ops = { - .init_box = snb_uncore_msr_init_box, - .disable_event = snb_uncore_msr_disable_event, - .enable_event = snb_uncore_msr_enable_event, - .read_counter = snb_uncore_msr_read_counter, +static struct uncore_event_desc nhmex_uncore_mbox_events[] = { + INTEL_UNCORE_EVENT_DESC(bbox_cmds_read, "inc_sel=0xd,fvc=0x2800"), + INTEL_UNCORE_EVENT_DESC(bbox_cmds_write, "inc_sel=0xd,fvc=0x2820"), + { /* end: all zeroes */ }, }; -static struct event_constraint snb_uncore_cbox_constraints[] = { - UNCORE_EVENT_CONSTRAINT(0x80, 0x1), - UNCORE_EVENT_CONSTRAINT(0x83, 0x1), - EVENT_CONSTRAINT_END +static struct intel_uncore_ops nhmex_uncore_mbox_ops = { + NHMEX_UNCORE_OPS_COMMON_INIT(), + .enable_event = nhmex_mbox_msr_enable_event, + .hw_config = nhmex_mbox_hw_config, + .get_constraint = nhmex_mbox_get_constraint, + .put_constraint = nhmex_mbox_put_constraint, }; -static struct intel_uncore_type snb_uncore_cbox = { - .name = "cbox", - .num_counters = 2, - .num_boxes = 4, - .perf_ctr_bits = 44, - .fixed_ctr_bits = 48, - .perf_ctr = SNB_UNC_CBO_0_PER_CTR0, - .event_ctl = SNB_UNC_CBO_0_PERFEVTSEL0, - .fixed_ctr = SNB_UNC_FIXED_CTR, - .fixed_ctl = SNB_UNC_FIXED_CTR_CTRL, - .single_fixed = 1, - .event_mask = SNB_UNC_RAW_EVENT_MASK, - .msr_offset = SNB_UNC_CBO_MSR_OFFSET, - .constraints = snb_uncore_cbox_constraints, - .ops = &snb_uncore_msr_ops, - .format_group = &snb_uncore_format_group, +static struct intel_uncore_type nhmex_uncore_mbox = { + .name = "mbox", + .num_counters = 6, + .num_boxes = 2, + .perf_ctr_bits = 48, + .event_ctl = NHMEX_M0_MSR_PMU_CTL0, + .perf_ctr = NHMEX_M0_MSR_PMU_CNT0, + .event_mask = NHMEX_M_PMON_RAW_EVENT_MASK, + .box_ctl = NHMEX_M0_MSR_GLOBAL_CTL, + .msr_offset = NHMEX_M_MSR_OFFSET, + .pair_ctr_ctl = 1, + .num_shared_regs = 8, + .event_descs = nhmex_uncore_mbox_events, + .ops = &nhmex_uncore_mbox_ops, + .format_group = &nhmex_uncore_mbox_format_group, }; -static struct intel_uncore_type *snb_msr_uncores[] = { - &snb_uncore_cbox, - NULL, -}; -/* end of Sandy Bridge uncore support */ +void nhmex_rbox_alter_er(struct intel_uncore_box *box, struct perf_event *event) +{ + struct hw_perf_event *hwc = &event->hw; + struct hw_perf_event_extra *reg1 = &hwc->extra_reg; + int port; -/* Nehalem uncore support */ -static void nhm_uncore_msr_disable_box(struct intel_uncore_box *box) + /* adjust the main event selector */ + if (reg1->idx % 2) { + reg1->idx--; + hwc->config -= 1 << NHMEX_R_PMON_CTL_EV_SEL_SHIFT; + } else { + reg1->idx++; + hwc->config += 1 << NHMEX_R_PMON_CTL_EV_SEL_SHIFT; + } + + /* adjust address or config of extra register */ + port = reg1->idx / 6 + box->pmu->pmu_idx * 4; + switch (reg1->idx % 6) { + case 0: + reg1->reg = NHMEX_R_MSR_PORTN_IPERF_CFG0(port); + break; + case 1: + reg1->reg = NHMEX_R_MSR_PORTN_IPERF_CFG1(port); + break; + case 2: + /* the 8~15 bits to the 0~7 bits */ + reg1->config >>= 8; + break; + case 3: + /* the 0~7 bits to the 8~15 bits */ + reg1->config <<= 8; + break; + case 4: + reg1->reg = NHMEX_R_MSR_PORTN_XBR_SET1_MM_CFG(port); + break; + case 5: + reg1->reg = NHMEX_R_MSR_PORTN_XBR_SET2_MM_CFG(port); + break; + }; +} + +/* + * Each rbox has 4 event set which monitor PQI port 0~3 or 4~7. + * An event set consists of 6 events, the 3rd and 4th events in + * an event set use the same extra register. So an event set uses + * 5 extra registers. + */ +static struct event_constraint * +nhmex_rbox_get_constraint(struct intel_uncore_box *box, struct perf_event *event) { - wrmsrl(NHM_UNC_PERF_GLOBAL_CTL, 0); + struct hw_perf_event *hwc = &event->hw; + struct hw_perf_event_extra *reg1 = &hwc->extra_reg; + struct hw_perf_event_extra *reg2 = &hwc->branch_reg; + struct intel_uncore_extra_reg *er; + unsigned long flags; + int idx, er_idx; + u64 config1; + bool ok = false; + + if (!uncore_box_is_fake(box) && reg1->alloc) + return NULL; + + idx = reg1->idx % 6; + config1 = reg1->config; +again: + er_idx = idx; + /* the 3rd and 4th events use the same extra register */ + if (er_idx > 2) + er_idx--; + er_idx += (reg1->idx / 6) * 5; + + er = &box->shared_regs[er_idx]; + raw_spin_lock_irqsave(&er->lock, flags); + if (idx < 2) { + if (!atomic_read(&er->ref) || er->config == reg1->config) { + atomic_inc(&er->ref); + er->config = reg1->config; + ok = true; + } + } else if (idx == 2 || idx == 3) { + /* + * these two events use different fields in a extra register, + * the 0~7 bits and the 8~15 bits respectively. + */ + u64 mask = 0xff << ((idx - 2) * 8); + if (!__BITS_VALUE(atomic_read(&er->ref), idx - 2, 8) || + !((er->config ^ config1) & mask)) { + atomic_add(1 << ((idx - 2) * 8), &er->ref); + er->config &= ~mask; + er->config |= config1 & mask; + ok = true; + } + } else { + if (!atomic_read(&er->ref) || + (er->config == (hwc->config >> 32) && + er->config1 == reg1->config && + er->config2 == reg2->config)) { + atomic_inc(&er->ref); + er->config = (hwc->config >> 32); + er->config1 = reg1->config; + er->config2 = reg2->config; + ok = true; + } + } + raw_spin_unlock_irqrestore(&er->lock, flags); + + if (!ok) { + /* + * The Rbox events are always in pairs. The paired + * events are functional identical, but use different + * extra registers. If we failed to take an extra + * register, try the alternative. + */ + if (idx % 2) + idx--; + else + idx++; + if (idx != reg1->idx % 6) { + if (idx == 2) + config1 >>= 8; + else if (idx == 3) + config1 <<= 8; + goto again; + } + } else { + if (!uncore_box_is_fake(box)) { + if (idx != reg1->idx % 6) + nhmex_rbox_alter_er(box, event); + reg1->alloc = 1; + } + return NULL; + } + return &constraint_empty; } -static void nhm_uncore_msr_enable_box(struct intel_uncore_box *box) +static void nhmex_rbox_put_constraint(struct intel_uncore_box *box, struct perf_event *event) { - wrmsrl(NHM_UNC_PERF_GLOBAL_CTL, - NHM_UNC_GLOBAL_CTL_EN_PC_ALL | NHM_UNC_GLOBAL_CTL_EN_FC); + struct intel_uncore_extra_reg *er; + struct hw_perf_event_extra *reg1 = &event->hw.extra_reg; + int idx, er_idx; + + if (uncore_box_is_fake(box) || !reg1->alloc) + return; + + idx = reg1->idx % 6; + er_idx = idx; + if (er_idx > 2) + er_idx--; + er_idx += (reg1->idx / 6) * 5; + + er = &box->shared_regs[er_idx]; + if (idx == 2 || idx == 3) + atomic_sub(1 << ((idx - 2) * 8), &er->ref); + else + atomic_dec(&er->ref); + + reg1->alloc = 0; } -static void nhm_uncore_msr_enable_event(struct intel_uncore_box *box, - struct perf_event *event) +static int nhmex_rbox_hw_config(struct intel_uncore_box *box, struct perf_event *event) { struct hw_perf_event *hwc = &event->hw; + struct hw_perf_event_extra *reg1 = &event->hw.extra_reg; + struct hw_perf_event_extra *reg2 = &event->hw.branch_reg; + int port, idx; - if (hwc->idx < UNCORE_PMC_IDX_FIXED) - wrmsrl(hwc->config_base, hwc->config | SNB_UNC_CTL_EN); - else - wrmsrl(hwc->config_base, NHM_UNC_FIXED_CTR_CTL_EN); + idx = (event->hw.config & NHMEX_R_PMON_CTL_EV_SEL_MASK) >> + NHMEX_R_PMON_CTL_EV_SEL_SHIFT; + if (idx >= 0x18) + return -EINVAL; + + reg1->idx = idx; + reg1->config = event->attr.config1; + + port = idx / 6 + box->pmu->pmu_idx * 4; + idx %= 6; + switch (idx) { + case 0: + reg1->reg = NHMEX_R_MSR_PORTN_IPERF_CFG0(port); + break; + case 1: + reg1->reg = NHMEX_R_MSR_PORTN_IPERF_CFG1(port); + break; + case 2: + case 3: + reg1->reg = NHMEX_R_MSR_PORTN_QLX_CFG(port); + break; + case 4: + case 5: + if (idx == 4) + reg1->reg = NHMEX_R_MSR_PORTN_XBR_SET1_MM_CFG(port); + else + reg1->reg = NHMEX_R_MSR_PORTN_XBR_SET2_MM_CFG(port); + reg2->config = event->attr.config2; + hwc->config |= event->attr.config & (~0ULL << 32); + break; + }; + return 0; } -static struct attribute *nhm_uncore_formats_attr[] = { - &format_attr_event.attr, - &format_attr_umask.attr, - &format_attr_edge.attr, - &format_attr_inv.attr, - &format_attr_cmask8.attr, +static u64 nhmex_rbox_shared_reg_config(struct intel_uncore_box *box, int idx) +{ + struct intel_uncore_extra_reg *er; + unsigned long flags; + u64 config; + + er = &box->shared_regs[idx]; + + raw_spin_lock_irqsave(&er->lock, flags); + config = er->config; + raw_spin_unlock_irqrestore(&er->lock, flags); + + return config; +} + +static void nhmex_rbox_msr_enable_event(struct intel_uncore_box *box, struct perf_event *event) +{ + struct hw_perf_event *hwc = &event->hw; + struct hw_perf_event_extra *reg1 = &hwc->extra_reg; + struct hw_perf_event_extra *reg2 = &hwc->branch_reg; + int idx, er_idx; + + idx = reg1->idx % 6; + er_idx = idx; + if (er_idx > 2) + er_idx--; + er_idx += (reg1->idx / 6) * 5; + + switch (idx) { + case 0: + case 1: + wrmsrl(reg1->reg, reg1->config); + break; + case 2: + case 3: + wrmsrl(reg1->reg, nhmex_rbox_shared_reg_config(box, er_idx)); + break; + case 4: + case 5: + wrmsrl(reg1->reg, reg1->config); + wrmsrl(reg1->reg + 1, hwc->config >> 32); + wrmsrl(reg1->reg + 2, reg2->config); + break; + }; + + wrmsrl(hwc->config_base, NHMEX_PMON_CTL_EN_BIT0 | + (hwc->config & NHMEX_R_PMON_CTL_EV_SEL_MASK)); +} + +DEFINE_UNCORE_FORMAT_ATTR(xbr_match, xbr_match, "config:32-63"); +DEFINE_UNCORE_FORMAT_ATTR(xbr_mm_cfg, xbr_mm_cfg, "config1:0-63"); +DEFINE_UNCORE_FORMAT_ATTR(xbr_mask, xbr_mask, "config2:0-63"); +DEFINE_UNCORE_FORMAT_ATTR(qlx_cfg, qlx_cfg, "config1:0-15"); +DEFINE_UNCORE_FORMAT_ATTR(iperf_cfg, iperf_cfg, "config1:0-31"); + +static struct attribute *nhmex_uncore_rbox_formats_attr[] = { + &format_attr_event5.attr, + &format_attr_xbr_mm_cfg.attr, + &format_attr_xbr_match.attr, + &format_attr_xbr_mask.attr, + &format_attr_qlx_cfg.attr, + &format_attr_iperf_cfg.attr, NULL, }; -static struct attribute_group nhm_uncore_format_group = { +static struct attribute_group nhmex_uncore_rbox_format_group = { .name = "format", - .attrs = nhm_uncore_formats_attr, + .attrs = nhmex_uncore_rbox_formats_attr, }; -static struct uncore_event_desc nhm_uncore_events[] = { - INTEL_UNCORE_EVENT_DESC(clockticks, "event=0xff,umask=0x00"), - INTEL_UNCORE_EVENT_DESC(qmc_writes_full_any, "event=0x2f,umask=0x0f"), - INTEL_UNCORE_EVENT_DESC(qmc_normal_reads_any, "event=0x2c,umask=0x0f"), - INTEL_UNCORE_EVENT_DESC(qhl_request_ioh_reads, "event=0x20,umask=0x01"), - INTEL_UNCORE_EVENT_DESC(qhl_request_ioh_writes, "event=0x20,umask=0x02"), - INTEL_UNCORE_EVENT_DESC(qhl_request_remote_reads, "event=0x20,umask=0x04"), - INTEL_UNCORE_EVENT_DESC(qhl_request_remote_writes, "event=0x20,umask=0x08"), - INTEL_UNCORE_EVENT_DESC(qhl_request_local_reads, "event=0x20,umask=0x10"), - INTEL_UNCORE_EVENT_DESC(qhl_request_local_writes, "event=0x20,umask=0x20"), +static struct uncore_event_desc nhmex_uncore_rbox_events[] = { + INTEL_UNCORE_EVENT_DESC(qpi0_flit_send, "event=0x0,iperf_cfg=0x80000000"), + INTEL_UNCORE_EVENT_DESC(qpi1_filt_send, "event=0x6,iperf_cfg=0x80000000"), + INTEL_UNCORE_EVENT_DESC(qpi0_idle_filt, "event=0x0,iperf_cfg=0x40000000"), + INTEL_UNCORE_EVENT_DESC(qpi1_idle_filt, "event=0x6,iperf_cfg=0x40000000"), + INTEL_UNCORE_EVENT_DESC(qpi0_date_response, "event=0x0,iperf_cfg=0xc4"), + INTEL_UNCORE_EVENT_DESC(qpi1_date_response, "event=0x6,iperf_cfg=0xc4"), { /* end: all zeroes */ }, }; -static struct intel_uncore_ops nhm_uncore_msr_ops = { - .disable_box = nhm_uncore_msr_disable_box, - .enable_box = nhm_uncore_msr_enable_box, - .disable_event = snb_uncore_msr_disable_event, - .enable_event = nhm_uncore_msr_enable_event, - .read_counter = snb_uncore_msr_read_counter, +static struct intel_uncore_ops nhmex_uncore_rbox_ops = { + NHMEX_UNCORE_OPS_COMMON_INIT(), + .enable_event = nhmex_rbox_msr_enable_event, + .hw_config = nhmex_rbox_hw_config, + .get_constraint = nhmex_rbox_get_constraint, + .put_constraint = nhmex_rbox_put_constraint, }; -static struct intel_uncore_type nhm_uncore = { - .name = "", - .num_counters = 8, - .num_boxes = 1, - .perf_ctr_bits = 48, - .fixed_ctr_bits = 48, - .event_ctl = NHM_UNC_PERFEVTSEL0, - .perf_ctr = NHM_UNC_UNCORE_PMC0, - .fixed_ctr = NHM_UNC_FIXED_CTR, - .fixed_ctl = NHM_UNC_FIXED_CTR_CTRL, - .event_mask = NHM_UNC_RAW_EVENT_MASK, - .event_descs = nhm_uncore_events, - .ops = &nhm_uncore_msr_ops, - .format_group = &nhm_uncore_format_group, +static struct intel_uncore_type nhmex_uncore_rbox = { + .name = "rbox", + .num_counters = 8, + .num_boxes = 2, + .perf_ctr_bits = 48, + .event_ctl = NHMEX_R_MSR_PMON_CTL0, + .perf_ctr = NHMEX_R_MSR_PMON_CNT0, + .event_mask = NHMEX_R_PMON_RAW_EVENT_MASK, + .box_ctl = NHMEX_R_MSR_GLOBAL_CTL, + .msr_offset = NHMEX_R_MSR_OFFSET, + .pair_ctr_ctl = 1, + .num_shared_regs = 20, + .event_descs = nhmex_uncore_rbox_events, + .ops = &nhmex_uncore_rbox_ops, + .format_group = &nhmex_uncore_rbox_format_group }; -static struct intel_uncore_type *nhm_msr_uncores[] = { - &nhm_uncore, +static struct intel_uncore_type *nhmex_msr_uncores[] = { + &nhmex_uncore_ubox, + &nhmex_uncore_cbox, + &nhmex_uncore_bbox, + &nhmex_uncore_sbox, + &nhmex_uncore_mbox, + &nhmex_uncore_rbox, + &nhmex_uncore_wbox, NULL, }; -/* end of Nehalem uncore support */ +/* end of Nehalem-EX uncore support */ -static void uncore_assign_hw_event(struct intel_uncore_box *box, - struct perf_event *event, int idx) +static void uncore_assign_hw_event(struct intel_uncore_box *box, struct perf_event *event, int idx) { struct hw_perf_event *hwc = &event->hw; @@ -787,8 +1821,7 @@ static void uncore_assign_hw_event(struct intel_uncore_box *box, hwc->event_base = uncore_perf_ctr(box, hwc->idx); } -static void uncore_perf_event_update(struct intel_uncore_box *box, - struct perf_event *event) +static void uncore_perf_event_update(struct intel_uncore_box *box, struct perf_event *event) { u64 prev_count, new_count, delta; int shift; @@ -858,14 +1891,12 @@ static void uncore_pmu_init_hrtimer(struct intel_uncore_box *box) box->hrtimer.function = uncore_pmu_hrtimer; } -struct intel_uncore_box *uncore_alloc_box(struct intel_uncore_type *type, - int cpu) +struct intel_uncore_box *uncore_alloc_box(struct intel_uncore_type *type, int cpu) { struct intel_uncore_box *box; int i, size; - size = sizeof(*box) + type->num_shared_regs * - sizeof(struct intel_uncore_extra_reg); + size = sizeof(*box) + type->num_shared_regs * sizeof(struct intel_uncore_extra_reg); box = kmalloc_node(size, GFP_KERNEL | __GFP_ZERO, cpu_to_node(cpu)); if (!box) @@ -915,12 +1946,11 @@ static struct intel_uncore_box *uncore_event_to_box(struct perf_event *event) * perf core schedules event on the basis of cpu, uncore events are * collected by one of the cpus inside a physical package. */ - return uncore_pmu_to_box(uncore_event_to_pmu(event), - smp_processor_id()); + return uncore_pmu_to_box(uncore_event_to_pmu(event), smp_processor_id()); } -static int uncore_collect_events(struct intel_uncore_box *box, - struct perf_event *leader, bool dogrp) +static int +uncore_collect_events(struct intel_uncore_box *box, struct perf_event *leader, bool dogrp) { struct perf_event *event; int n, max_count; @@ -952,8 +1982,7 @@ static int uncore_collect_events(struct intel_uncore_box *box, } static struct event_constraint * -uncore_get_event_constraint(struct intel_uncore_box *box, - struct perf_event *event) +uncore_get_event_constraint(struct intel_uncore_box *box, struct perf_event *event) { struct intel_uncore_type *type = box->pmu->type; struct event_constraint *c; @@ -977,15 +2006,13 @@ uncore_get_event_constraint(struct intel_uncore_box *box, return &type->unconstrainted; } -static void uncore_put_event_constraint(struct intel_uncore_box *box, - struct perf_event *event) +static void uncore_put_event_constraint(struct intel_uncore_box *box, struct perf_event *event) { if (box->pmu->type->ops->put_constraint) box->pmu->type->ops->put_constraint(box, event); } -static int uncore_assign_events(struct intel_uncore_box *box, - int assign[], int n) +static int uncore_assign_events(struct intel_uncore_box *box, int assign[], int n) { unsigned long used_mask[BITS_TO_LONGS(UNCORE_PMC_IDX_MAX)]; struct event_constraint *c, *constraints[UNCORE_PMC_IDX_MAX]; @@ -1407,8 +2434,7 @@ static bool pcidrv_registered; /* * add a pci uncore device */ -static int __devinit uncore_pci_add(struct intel_uncore_type *type, - struct pci_dev *pdev) +static int __devinit uncore_pci_add(struct intel_uncore_type *type, struct pci_dev *pdev) { struct intel_uncore_pmu *pmu; struct intel_uncore_box *box; @@ -1485,6 +2511,7 @@ static int __devinit uncore_pci_probe(struct pci_dev *pdev, struct intel_uncore_type *type; type = (struct intel_uncore_type *)id->driver_data; + return uncore_pci_add(type, pdev); } @@ -1612,8 +2639,8 @@ static int __cpuinit uncore_cpu_prepare(int cpu, int phys_id) return 0; } -static void __cpuinit uncore_change_context(struct intel_uncore_type **uncores, - int old_cpu, int new_cpu) +static void __cpuinit +uncore_change_context(struct intel_uncore_type **uncores, int old_cpu, int new_cpu) { struct intel_uncore_type *type; struct intel_uncore_pmu *pmu; @@ -1694,8 +2721,8 @@ static void __cpuinit uncore_event_init_cpu(int cpu) uncore_change_context(pci_uncores, -1, cpu); } -static int __cpuinit uncore_cpu_notifier(struct notifier_block *self, - unsigned long action, void *hcpu) +static int + __cpuinit uncore_cpu_notifier(struct notifier_block *self, unsigned long action, void *hcpu) { unsigned int cpu = (long)hcpu; @@ -1732,12 +2759,12 @@ static int __cpuinit uncore_cpu_notifier(struct notifier_block *self, } static struct notifier_block uncore_cpu_nb __cpuinitdata = { - .notifier_call = uncore_cpu_notifier, + .notifier_call = uncore_cpu_notifier, /* * to migrate uncore events, our notifier should be executed * before perf core's notifier. */ - .priority = CPU_PRI_PERF + 1, + .priority = CPU_PRI_PERF + 1, }; static void __init uncore_cpu_setup(void *dummy) @@ -1767,6 +2794,9 @@ static int __init uncore_cpu_init(void) snbep_uncore_cbox.num_boxes = max_cores; msr_uncores = snbep_msr_uncores; break; + case 46: + msr_uncores = nhmex_msr_uncores; + break; default: return 0; } diff --git a/arch/x86/kernel/cpu/perf_event_intel_uncore.h b/arch/x86/kernel/cpu/perf_event_intel_uncore.h index b13e9ea..47b1776 100644 --- a/arch/x86/kernel/cpu/perf_event_intel_uncore.h +++ b/arch/x86/kernel/cpu/perf_event_intel_uncore.h @@ -5,8 +5,6 @@ #include "perf_event.h" #define UNCORE_PMU_NAME_LEN 32 -#define UNCORE_BOX_HASH_SIZE 8 - #define UNCORE_PMU_HRTIMER_INTERVAL (60 * NSEC_PER_SEC) #define UNCORE_FIXED_EVENT 0xff @@ -158,6 +156,193 @@ #define SNBEP_PCU_MSR_CORE_C3_CTR 0x3fc #define SNBEP_PCU_MSR_CORE_C6_CTR 0x3fd +/* NHM-EX event control */ +#define NHMEX_PMON_CTL_EV_SEL_MASK 0x000000ff +#define NHMEX_PMON_CTL_UMASK_MASK 0x0000ff00 +#define NHMEX_PMON_CTL_EN_BIT0 (1 << 0) +#define NHMEX_PMON_CTL_EDGE_DET (1 << 18) +#define NHMEX_PMON_CTL_PMI_EN (1 << 20) +#define NHMEX_PMON_CTL_EN_BIT22 (1 << 22) +#define NHMEX_PMON_CTL_INVERT (1 << 23) +#define NHMEX_PMON_CTL_TRESH_MASK 0xff000000 +#define NHMEX_PMON_RAW_EVENT_MASK (NHMEX_PMON_CTL_EV_SEL_MASK | \ + NHMEX_PMON_CTL_UMASK_MASK | \ + NHMEX_PMON_CTL_EDGE_DET | \ + NHMEX_PMON_CTL_INVERT | \ + NHMEX_PMON_CTL_TRESH_MASK) + +/* NHM-EX Ubox */ +#define NHMEX_U_MSR_PMON_GLOBAL_CTL 0xc00 +#define NHMEX_U_MSR_PMON_CTR 0xc11 +#define NHMEX_U_MSR_PMON_EV_SEL 0xc10 + +#define NHMEX_U_PMON_GLOBAL_EN (1 << 0) +#define NHMEX_U_PMON_GLOBAL_PMI_CORE_SEL 0x0000001e +#define NHMEX_U_PMON_GLOBAL_EN_ALL (1 << 28) +#define NHMEX_U_PMON_GLOBAL_RST_ALL (1 << 29) +#define NHMEX_U_PMON_GLOBAL_FRZ_ALL (1 << 31) + +#define NHMEX_U_PMON_RAW_EVENT_MASK \ + (NHMEX_PMON_CTL_EV_SEL_MASK | \ + NHMEX_PMON_CTL_EDGE_DET) + +/* NHM-EX Cbox */ +#define NHMEX_C0_MSR_PMON_GLOBAL_CTL 0xd00 +#define NHMEX_C0_MSR_PMON_CTR0 0xd11 +#define NHMEX_C0_MSR_PMON_EV_SEL0 0xd10 +#define NHMEX_C_MSR_OFFSET 0x20 + +/* NHM-EX Bbox */ +#define NHMEX_B0_MSR_PMON_GLOBAL_CTL 0xc20 +#define NHMEX_B0_MSR_PMON_CTR0 0xc31 +#define NHMEX_B0_MSR_PMON_CTL0 0xc30 +#define NHMEX_B_MSR_OFFSET 0x40 +#define NHMEX_B0_MSR_MATCH 0xe45 +#define NHMEX_B0_MSR_MASK 0xe46 +#define NHMEX_B1_MSR_MATCH 0xe4d +#define NHMEX_B1_MSR_MASK 0xe4e + +#define NHMEX_B_PMON_CTL_EN (1 << 0) +#define NHMEX_B_PMON_CTL_EV_SEL_SHIFT 1 +#define NHMEX_B_PMON_CTL_EV_SEL_MASK \ + (0x1f << NHMEX_B_PMON_CTL_EV_SEL_SHIFT) +#define NHMEX_B_PMON_CTR_SHIFT 6 +#define NHMEX_B_PMON_CTR_MASK \ + (0x3 << NHMEX_B_PMON_CTR_SHIFT) +#define NHMEX_B_PMON_RAW_EVENT_MASK \ + (NHMEX_B_PMON_CTL_EV_SEL_MASK | \ + NHMEX_B_PMON_CTR_MASK) + +/* NHM-EX Sbox */ +#define NHMEX_S0_MSR_PMON_GLOBAL_CTL 0xc40 +#define NHMEX_S0_MSR_PMON_CTR0 0xc51 +#define NHMEX_S0_MSR_PMON_CTL0 0xc50 +#define NHMEX_S_MSR_OFFSET 0x80 +#define NHMEX_S0_MSR_MM_CFG 0xe48 +#define NHMEX_S0_MSR_MATCH 0xe49 +#define NHMEX_S0_MSR_MASK 0xe4a +#define NHMEX_S1_MSR_MM_CFG 0xe58 +#define NHMEX_S1_MSR_MATCH 0xe59 +#define NHMEX_S1_MSR_MASK 0xe5a + +#define NHMEX_S_PMON_MM_CFG_EN (0x1ULL << 63) + +/* NHM-EX Mbox */ +#define NHMEX_M0_MSR_GLOBAL_CTL 0xca0 +#define NHMEX_M0_MSR_PMU_DSP 0xca5 +#define NHMEX_M0_MSR_PMU_ISS 0xca6 +#define NHMEX_M0_MSR_PMU_MAP 0xca7 +#define NHMEX_M0_MSR_PMU_MSC_THR 0xca8 +#define NHMEX_M0_MSR_PMU_PGT 0xca9 +#define NHMEX_M0_MSR_PMU_PLD 0xcaa +#define NHMEX_M0_MSR_PMU_ZDP_CTL_FVC 0xcab +#define NHMEX_M0_MSR_PMU_CTL0 0xcb0 +#define NHMEX_M0_MSR_PMU_CNT0 0xcb1 +#define NHMEX_M_MSR_OFFSET 0x40 +#define NHMEX_M0_MSR_PMU_MM_CFG 0xe54 +#define NHMEX_M1_MSR_PMU_MM_CFG 0xe5c + +#define NHMEX_M_PMON_MM_CFG_EN (1ULL << 63) +#define NHMEX_M_PMON_ADDR_MATCH_MASK 0x3ffffffffULL +#define NHMEX_M_PMON_ADDR_MASK_MASK 0x7ffffffULL +#define NHMEX_M_PMON_ADDR_MASK_SHIFT 34 + +#define NHMEX_M_PMON_CTL_EN (1 << 0) +#define NHMEX_M_PMON_CTL_PMI_EN (1 << 1) +#define NHMEX_M_PMON_CTL_COUNT_MODE_SHIFT 2 +#define NHMEX_M_PMON_CTL_COUNT_MODE_MASK \ + (0x3 << NHMEX_M_PMON_CTL_COUNT_MODE_SHIFT) +#define NHMEX_M_PMON_CTL_STORAGE_MODE_SHIFT 4 +#define NHMEX_M_PMON_CTL_STORAGE_MODE_MASK \ + (0x3 << NHMEX_M_PMON_CTL_STORAGE_MODE_SHIFT) +#define NHMEX_M_PMON_CTL_WRAP_MODE (1 << 6) +#define NHMEX_M_PMON_CTL_FLAG_MODE (1 << 7) +#define NHMEX_M_PMON_CTL_INC_SEL_SHIFT 9 +#define NHMEX_M_PMON_CTL_INC_SEL_MASK \ + (0x1f << NHMEX_M_PMON_CTL_INC_SEL_SHIFT) +#define NHMEX_M_PMON_CTL_SET_FLAG_SEL_SHIFT 19 +#define NHMEX_M_PMON_CTL_SET_FLAG_SEL_MASK \ + (0x7 << NHMEX_M_PMON_CTL_SET_FLAG_SEL_SHIFT) +#define NHMEX_M_PMON_RAW_EVENT_MASK \ + (NHMEX_M_PMON_CTL_COUNT_MODE_MASK | \ + NHMEX_M_PMON_CTL_STORAGE_MODE_MASK | \ + NHMEX_M_PMON_CTL_WRAP_MODE | \ + NHMEX_M_PMON_CTL_FLAG_MODE | \ + NHMEX_M_PMON_CTL_INC_SEL_MASK | \ + NHMEX_M_PMON_CTL_SET_FLAG_SEL_MASK) + + +#define NHMEX_M_PMON_ZDP_CTL_FVC_FVID_MASK 0x1f +#define NHMEX_M_PMON_ZDP_CTL_FVC_BCMD_MASK (0x7 << 5) +#define NHMEX_M_PMON_ZDP_CTL_FVC_RSP_MASK (0x7 << 8) +#define NHMEX_M_PMON_ZDP_CTL_FVC_PBOX_INIT_ERR (1 << 23) +#define NHMEX_M_PMON_ZDP_CTL_FVC_MASK \ + (NHMEX_M_PMON_ZDP_CTL_FVC_FVID_MASK | \ + NHMEX_M_PMON_ZDP_CTL_FVC_BCMD_MASK | \ + NHMEX_M_PMON_ZDP_CTL_FVC_RSP_MASK | \ + NHMEX_M_PMON_ZDP_CTL_FVC_PBOX_INIT_ERR) +#define NHMEX_M_PMON_ZDP_CTL_FVC_EVENT_MASK(n) (0x7 << (11 + 3 * (n))) + +/* + * use the 9~13 bits to select event If the 7th bit is not set, + * otherwise use the 19~21 bits to select event. + */ +#define MBOX_INC_SEL(x) ((x) << NHMEX_M_PMON_CTL_INC_SEL_SHIFT) +#define MBOX_SET_FLAG_SEL(x) (((x) << NHMEX_M_PMON_CTL_SET_FLAG_SEL_SHIFT) | \ + NHMEX_M_PMON_CTL_FLAG_MODE) +#define MBOX_INC_SEL_MASK (NHMEX_M_PMON_CTL_INC_SEL_MASK | \ + NHMEX_M_PMON_CTL_FLAG_MODE) +#define MBOX_SET_FLAG_SEL_MASK (NHMEX_M_PMON_CTL_SET_FLAG_SEL_MASK | \ + NHMEX_M_PMON_CTL_FLAG_MODE) +#define MBOX_INC_SEL_EXTAR_REG(c, r) \ + EVENT_EXTRA_REG(MBOX_INC_SEL(c), NHMEX_M0_MSR_PMU_##r, \ + MBOX_INC_SEL_MASK, (u64)-1, NHMEX_M_##r) +#define MBOX_SET_FLAG_SEL_EXTRA_REG(c, r) \ + EVENT_EXTRA_REG(MBOX_SET_FLAG_SEL(c), NHMEX_M0_MSR_PMU_##r, \ + MBOX_SET_FLAG_SEL_MASK, \ + (u64)-1, NHMEX_M_##r) + +/* NHM-EX Rbox */ +#define NHMEX_R_MSR_GLOBAL_CTL 0xe00 +#define NHMEX_R_MSR_PMON_CTL0 0xe10 +#define NHMEX_R_MSR_PMON_CNT0 0xe11 +#define NHMEX_R_MSR_OFFSET 0x20 + +#define NHMEX_R_MSR_PORTN_QLX_CFG(n) \ + ((n) < 4 ? (0xe0c + (n)) : (0xe2c + (n) - 4)) +#define NHMEX_R_MSR_PORTN_IPERF_CFG0(n) (0xe04 + (n)) +#define NHMEX_R_MSR_PORTN_IPERF_CFG1(n) (0xe24 + (n)) +#define NHMEX_R_MSR_PORTN_XBR_OFFSET(n) \ + (((n) < 4 ? 0 : 0x10) + (n) * 4) +#define NHMEX_R_MSR_PORTN_XBR_SET1_MM_CFG(n) \ + (0xe60 + NHMEX_R_MSR_PORTN_XBR_OFFSET(n)) +#define NHMEX_R_MSR_PORTN_XBR_SET1_MATCH(n) \ + (NHMEX_R_MSR_PORTN_XBR_SET1_MM_CFG(n) + 1) +#define NHMEX_R_MSR_PORTN_XBR_SET1_MASK(n) \ + (NHMEX_R_MSR_PORTN_XBR_SET1_MM_CFG(n) + 2) +#define NHMEX_R_MSR_PORTN_XBR_SET2_MM_CFG(n) \ + (0xe70 + NHMEX_R_MSR_PORTN_XBR_OFFSET(n)) +#define NHMEX_R_MSR_PORTN_XBR_SET2_MATCH(n) \ + (NHMEX_R_MSR_PORTN_XBR_SET2_MM_CFG(n) + 1) +#define NHMEX_R_MSR_PORTN_XBR_SET2_MASK(n) \ + (NHMEX_R_MSR_PORTN_XBR_SET2_MM_CFG(n) + 2) + +#define NHMEX_R_PMON_CTL_EN (1 << 0) +#define NHMEX_R_PMON_CTL_EV_SEL_SHIFT 1 +#define NHMEX_R_PMON_CTL_EV_SEL_MASK \ + (0x1f << NHMEX_R_PMON_CTL_EV_SEL_SHIFT) +#define NHMEX_R_PMON_CTL_PMI_EN (1 << 6) +#define NHMEX_R_PMON_RAW_EVENT_MASK NHMEX_R_PMON_CTL_EV_SEL_MASK + +/* NHM-EX Wbox */ +#define NHMEX_W_MSR_GLOBAL_CTL 0xc80 +#define NHMEX_W_MSR_PMON_CNT0 0xc90 +#define NHMEX_W_MSR_PMON_EVT_SEL0 0xc91 +#define NHMEX_W_MSR_PMON_FIXED_CTR 0x394 +#define NHMEX_W_MSR_PMON_FIXED_CTL 0x395 + +#define NHMEX_W_PMON_GLOBAL_FIXED_EN (1ULL << 31) + struct intel_uncore_ops; struct intel_uncore_pmu; struct intel_uncore_box; @@ -178,6 +363,7 @@ struct intel_uncore_type { unsigned msr_offset; unsigned num_shared_regs:8; unsigned single_fixed:1; + unsigned pair_ctr_ctl:1; struct event_constraint unconstrainted; struct event_constraint *constraints; struct intel_uncore_pmu *pmus; @@ -213,7 +399,7 @@ struct intel_uncore_pmu { struct intel_uncore_extra_reg { raw_spinlock_t lock; - u64 config1; + u64 config, config1, config2; atomic_t ref; }; @@ -323,14 +509,16 @@ unsigned uncore_msr_fixed_ctr(struct intel_uncore_box *box) static inline unsigned uncore_msr_event_ctl(struct intel_uncore_box *box, int idx) { - return idx + box->pmu->type->event_ctl + + return box->pmu->type->event_ctl + + (box->pmu->type->pair_ctr_ctl ? 2 * idx : idx) + box->pmu->type->msr_offset * box->pmu->pmu_idx; } static inline unsigned uncore_msr_perf_ctr(struct intel_uncore_box *box, int idx) { - return idx + box->pmu->type->perf_ctr + + return box->pmu->type->perf_ctr + + (box->pmu->type->pair_ctr_ctl ? 2 * idx : idx) + box->pmu->type->msr_offset * box->pmu->pmu_idx; } @@ -422,3 +610,8 @@ static inline void uncore_box_init(struct intel_uncore_box *box) box->pmu->type->ops->init_box(box); } } + +static inline bool uncore_box_is_fake(struct intel_uncore_box *box) +{ + return (box->phys_id < 0); +} ^ permalink raw reply related [flat|nested] 21+ messages in thread
end of thread, other threads:[~2012-07-26 15:12 UTC | newest] Thread overview: 21+ messages (download: mbox.gz follow: Atom feed -- links below jump to the message on this page -- 2012-07-04 6:00 [PATCH 0/5] perf/x86: Uncore updates Yan, Zheng 2012-07-04 6:00 ` [PATCH 1/5] perf/x86: Use 0xff as pseudo code for fixed uncore event Yan, Zheng 2012-07-04 6:00 ` [PATCH 2/5] perf/x86: Fix event constraint for SandyBridge-EP CBox Yan, Zheng 2012-07-04 6:00 ` [PATCH 3/5] perf/x86: Detect number of instances of uncore CBox Yan, Zheng 2012-07-06 6:32 ` [tip:perf/core] " tip-bot for Yan, Zheng 2012-07-04 6:00 ` [PATCH 4/5] perf/x86: Uncore filter support for SandyBridge-EP Yan, Zheng 2012-07-06 6:33 ` [tip:perf/core] " tip-bot for Yan, Zheng 2012-07-22 19:25 ` [PATCH 4/5] " Stephane Eranian 2012-07-04 6:00 ` [PATCH 5/5] perf/x86: Add Intel Nehalem-EX uncore support Yan, Zheng 2012-07-04 10:04 ` Peter Zijlstra 2012-07-04 16:00 ` Yan, Zheng 2012-07-04 16:53 ` Peter Zijlstra 2012-07-04 17:46 ` Yan, Zheng 2012-07-05 6:32 ` [PATCH V2 " Yan, Zheng 2012-07-05 14:51 ` Andi Kleen 2012-07-05 15:46 ` Yan, Zheng 2012-07-10 12:17 ` Peter Zijlstra 2012-07-11 2:06 ` Yan, Zheng 2012-07-11 9:57 ` Peter Zijlstra 2012-07-11 12:02 ` Yan, Zheng 2012-07-26 15:11 ` [tip:perf/core] " tip-bot for Yan, Zheng
This is a public inbox, see mirroring instructions for how to clone and mirror all data and code used for this inbox; as well as URLs for NNTP newsgroup(s).