Merge tag 'perf-core-for-mingo' of git://git.kernel.org/pub/scm/linux/kernel/git...
authorIngo Molnar <mingo@kernel.org>
Sat, 22 Feb 2014 16:26:24 +0000 (17:26 +0100)
committerIngo Molnar <mingo@kernel.org>
Sat, 22 Feb 2014 16:26:24 +0000 (17:26 +0100)
Pull perf/core improvements and fixes from Arnaldo Carvalho de Melo:

User Visible:

  * Allow setting preferred callchain method in .perfconfig (Jiri Olsa)

  * Show in what binaries/modules 'perf probe's are set (Masami Hiramatsu)

  * Support distro-style debuginfo for uprobe in 'perf probe' (Masami Hiramatsu)

Developer stuff:

  * Move some hashing and fs related code from tools/perf/util/
    to tools/lib/ so that it can be used by more tools/ living
    utilities (Borislav Petkov)

  * Prepare DWARF unwinding code for using an elfutils alternative
    unwinding library (Jiri Olsa)

  * Fix DWARF unwind max_stack processing (Jiri Olsa)

  * Add dwarf unwind 'perf test' entry (Jiri Olsa)

  * 'perf probe' improvements including memory leak fixes,
    sharing the intlist class with other tools, uprobes/kprobes
    code sharing and use of ref_reloc_sym (Masami Hiramatsu)

  * Shorten sample symbol resolving by adding cpumode to
    struct addr_location (Arnaldo Carvalho de Melo)

Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
arch/x86/kernel/cpu/perf_event_intel_uncore.c
arch/x86/kernel/cpu/perf_event_intel_uncore.h
include/linux/irq_work.h
include/linux/pci_ids.h
kernel/events/core.c
kernel/irq_work.c

index 29c2487..de5c568 100644 (file)
@@ -66,6 +66,47 @@ DEFINE_UNCORE_FORMAT_ATTR(mask_vnw, mask_vnw, "config2:3-4");
 DEFINE_UNCORE_FORMAT_ATTR(mask0, mask0, "config2:0-31");
 DEFINE_UNCORE_FORMAT_ATTR(mask1, mask1, "config2:32-63");
 
+static void uncore_pmu_start_hrtimer(struct intel_uncore_box *box);
+static void uncore_pmu_cancel_hrtimer(struct intel_uncore_box *box);
+static void uncore_perf_event_update(struct intel_uncore_box *box, struct perf_event *event);
+static void uncore_pmu_event_read(struct perf_event *event);
+
+static struct intel_uncore_pmu *uncore_event_to_pmu(struct perf_event *event)
+{
+       return container_of(event->pmu, struct intel_uncore_pmu, pmu);
+}
+
+static struct intel_uncore_box *
+uncore_pmu_to_box(struct intel_uncore_pmu *pmu, int cpu)
+{
+       struct intel_uncore_box *box;
+
+       box = *per_cpu_ptr(pmu->box, cpu);
+       if (box)
+               return box;
+
+       raw_spin_lock(&uncore_box_lock);
+       list_for_each_entry(box, &pmu->box_list, list) {
+               if (box->phys_id == topology_physical_package_id(cpu)) {
+                       atomic_inc(&box->refcnt);
+                       *per_cpu_ptr(pmu->box, cpu) = box;
+                       break;
+               }
+       }
+       raw_spin_unlock(&uncore_box_lock);
+
+       return *per_cpu_ptr(pmu->box, cpu);
+}
+
+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());
+}
+
 static u64 uncore_msr_read_counter(struct intel_uncore_box *box, struct perf_event *event)
 {
        u64 count;
@@ -1631,6 +1672,345 @@ static struct intel_uncore_type *snb_msr_uncores[] = {
        &snb_uncore_cbox,
        NULL,
 };
+
+enum {
+       SNB_PCI_UNCORE_IMC,
+};
+
+static struct uncore_event_desc snb_uncore_imc_events[] = {
+       INTEL_UNCORE_EVENT_DESC(data_reads,  "event=0x01"),
+       INTEL_UNCORE_EVENT_DESC(data_reads.scale, "6.103515625e-5"),
+       INTEL_UNCORE_EVENT_DESC(data_reads.unit, "MiB"),
+
+       INTEL_UNCORE_EVENT_DESC(data_writes, "event=0x02"),
+       INTEL_UNCORE_EVENT_DESC(data_writes.scale, "6.103515625e-5"),
+       INTEL_UNCORE_EVENT_DESC(data_writes.unit, "MiB"),
+
+       { /* end: all zeroes */ },
+};
+
+#define SNB_UNCORE_PCI_IMC_EVENT_MASK          0xff
+#define SNB_UNCORE_PCI_IMC_BAR_OFFSET          0x48
+
+/* page size multiple covering all config regs */
+#define SNB_UNCORE_PCI_IMC_MAP_SIZE            0x6000
+
+#define SNB_UNCORE_PCI_IMC_DATA_READS          0x1
+#define SNB_UNCORE_PCI_IMC_DATA_READS_BASE     0x5050
+#define SNB_UNCORE_PCI_IMC_DATA_WRITES         0x2
+#define SNB_UNCORE_PCI_IMC_DATA_WRITES_BASE    0x5054
+#define SNB_UNCORE_PCI_IMC_CTR_BASE            SNB_UNCORE_PCI_IMC_DATA_READS_BASE
+
+static struct attribute *snb_uncore_imc_formats_attr[] = {
+       &format_attr_event.attr,
+       NULL,
+};
+
+static struct attribute_group snb_uncore_imc_format_group = {
+       .name = "format",
+       .attrs = snb_uncore_imc_formats_attr,
+};
+
+static void snb_uncore_imc_init_box(struct intel_uncore_box *box)
+{
+       struct pci_dev *pdev = box->pci_dev;
+       u32 addr_lo, addr_hi;
+       resource_size_t addr;
+
+       pci_read_config_dword(pdev, SNB_UNCORE_PCI_IMC_BAR_OFFSET, &addr_lo);
+       addr = addr_lo;
+
+#ifdef CONFIG_PHYS_ADDR_T_64BIT
+       pci_read_config_dword(pdev, SNB_UNCORE_PCI_IMC_BAR_OFFSET+4, &addr_hi);
+       addr = ((resource_size_t)addr_hi << 32) | addr_lo;
+#endif
+
+       addr &= ~(PAGE_SIZE - 1);
+
+       box->io_addr = ioremap(addr, SNB_UNCORE_PCI_IMC_MAP_SIZE);
+       box->hrtimer_duration = UNCORE_SNB_IMC_HRTIMER_INTERVAL;
+}
+
+static void snb_uncore_imc_enable_box(struct intel_uncore_box *box)
+{}
+
+static void snb_uncore_imc_disable_box(struct intel_uncore_box *box)
+{}
+
+static void snb_uncore_imc_enable_event(struct intel_uncore_box *box, struct perf_event *event)
+{}
+
+static void snb_uncore_imc_disable_event(struct intel_uncore_box *box, struct perf_event *event)
+{}
+
+static u64 snb_uncore_imc_read_counter(struct intel_uncore_box *box, struct perf_event *event)
+{
+       struct hw_perf_event *hwc = &event->hw;
+
+       return (u64)*(unsigned int *)(box->io_addr + hwc->event_base);
+}
+
+/*
+ * custom event_init() function because we define our own fixed, free
+ * running counters, so we do not want to conflict with generic uncore
+ * logic. Also simplifies processing
+ */
+static int snb_uncore_imc_event_init(struct perf_event *event)
+{
+       struct intel_uncore_pmu *pmu;
+       struct intel_uncore_box *box;
+       struct hw_perf_event *hwc = &event->hw;
+       u64 cfg = event->attr.config & SNB_UNCORE_PCI_IMC_EVENT_MASK;
+       int idx, base;
+
+       if (event->attr.type != event->pmu->type)
+               return -ENOENT;
+
+       pmu = uncore_event_to_pmu(event);
+       /* no device found for this pmu */
+       if (pmu->func_id < 0)
+               return -ENOENT;
+
+       /* Sampling not supported yet */
+       if (hwc->sample_period)
+               return -EINVAL;
+
+       /* unsupported modes and filters */
+       if (event->attr.exclude_user   ||
+           event->attr.exclude_kernel ||
+           event->attr.exclude_hv     ||
+           event->attr.exclude_idle   ||
+           event->attr.exclude_host   ||
+           event->attr.exclude_guest  ||
+           event->attr.sample_period) /* no sampling */
+               return -EINVAL;
+
+       /*
+        * Place all uncore events for a particular physical package
+        * onto a single cpu
+        */
+       if (event->cpu < 0)
+               return -EINVAL;
+
+       /* check only supported bits are set */
+       if (event->attr.config & ~SNB_UNCORE_PCI_IMC_EVENT_MASK)
+               return -EINVAL;
+
+       box = uncore_pmu_to_box(pmu, event->cpu);
+       if (!box || box->cpu < 0)
+               return -EINVAL;
+
+       event->cpu = box->cpu;
+
+       event->hw.idx = -1;
+       event->hw.last_tag = ~0ULL;
+       event->hw.extra_reg.idx = EXTRA_REG_NONE;
+       event->hw.branch_reg.idx = EXTRA_REG_NONE;
+       /*
+        * check event is known (whitelist, determines counter)
+        */
+       switch (cfg) {
+       case SNB_UNCORE_PCI_IMC_DATA_READS:
+               base = SNB_UNCORE_PCI_IMC_DATA_READS_BASE;
+               idx = UNCORE_PMC_IDX_FIXED;
+               break;
+       case SNB_UNCORE_PCI_IMC_DATA_WRITES:
+               base = SNB_UNCORE_PCI_IMC_DATA_WRITES_BASE;
+               idx = UNCORE_PMC_IDX_FIXED + 1;
+               break;
+       default:
+               return -EINVAL;
+       }
+
+       /* must be done before validate_group */
+       event->hw.event_base = base;
+       event->hw.config = cfg;
+       event->hw.idx = idx;
+
+       /* no group validation needed, we have free running counters */
+
+       return 0;
+}
+
+static int snb_uncore_imc_hw_config(struct intel_uncore_box *box, struct perf_event *event)
+{
+       return 0;
+}
+
+static void snb_uncore_imc_event_start(struct perf_event *event, int flags)
+{
+       struct intel_uncore_box *box = uncore_event_to_box(event);
+       u64 count;
+
+       if (WARN_ON_ONCE(!(event->hw.state & PERF_HES_STOPPED)))
+               return;
+
+       event->hw.state = 0;
+       box->n_active++;
+
+       list_add_tail(&event->active_entry, &box->active_list);
+
+       count = snb_uncore_imc_read_counter(box, event);
+       local64_set(&event->hw.prev_count, count);
+
+       if (box->n_active == 1)
+               uncore_pmu_start_hrtimer(box);
+}
+
+static void snb_uncore_imc_event_stop(struct perf_event *event, int flags)
+{
+       struct intel_uncore_box *box = uncore_event_to_box(event);
+       struct hw_perf_event *hwc = &event->hw;
+
+       if (!(hwc->state & PERF_HES_STOPPED)) {
+               box->n_active--;
+
+               WARN_ON_ONCE(hwc->state & PERF_HES_STOPPED);
+               hwc->state |= PERF_HES_STOPPED;
+
+               list_del(&event->active_entry);
+
+               if (box->n_active == 0)
+                       uncore_pmu_cancel_hrtimer(box);
+       }
+
+       if ((flags & PERF_EF_UPDATE) && !(hwc->state & PERF_HES_UPTODATE)) {
+               /*
+                * Drain the remaining delta count out of a event
+                * that we are disabling:
+                */
+               uncore_perf_event_update(box, event);
+               hwc->state |= PERF_HES_UPTODATE;
+       }
+}
+
+static int snb_uncore_imc_event_add(struct perf_event *event, int flags)
+{
+       struct intel_uncore_box *box = uncore_event_to_box(event);
+       struct hw_perf_event *hwc = &event->hw;
+
+       if (!box)
+               return -ENODEV;
+
+       hwc->state = PERF_HES_UPTODATE | PERF_HES_STOPPED;
+       if (!(flags & PERF_EF_START))
+               hwc->state |= PERF_HES_ARCH;
+
+       snb_uncore_imc_event_start(event, 0);
+
+       box->n_events++;
+
+       return 0;
+}
+
+static void snb_uncore_imc_event_del(struct perf_event *event, int flags)
+{
+       struct intel_uncore_box *box = uncore_event_to_box(event);
+       int i;
+
+       snb_uncore_imc_event_stop(event, PERF_EF_UPDATE);
+
+       for (i = 0; i < box->n_events; i++) {
+               if (event == box->event_list[i]) {
+                       --box->n_events;
+                       break;
+               }
+       }
+}
+
+static int snb_pci2phy_map_init(int devid)
+{
+       struct pci_dev *dev = NULL;
+       int bus;
+
+       dev = pci_get_device(PCI_VENDOR_ID_INTEL, devid, dev);
+       if (!dev)
+               return -ENOTTY;
+
+       bus = dev->bus->number;
+
+       pcibus_to_physid[bus] = 0;
+
+       pci_dev_put(dev);
+
+       return 0;
+}
+
+static struct pmu snb_uncore_imc_pmu = {
+       .task_ctx_nr    = perf_invalid_context,
+       .event_init     = snb_uncore_imc_event_init,
+       .add            = snb_uncore_imc_event_add,
+       .del            = snb_uncore_imc_event_del,
+       .start          = snb_uncore_imc_event_start,
+       .stop           = snb_uncore_imc_event_stop,
+       .read           = uncore_pmu_event_read,
+};
+
+static struct intel_uncore_ops snb_uncore_imc_ops = {
+       .init_box       = snb_uncore_imc_init_box,
+       .enable_box     = snb_uncore_imc_enable_box,
+       .disable_box    = snb_uncore_imc_disable_box,
+       .disable_event  = snb_uncore_imc_disable_event,
+       .enable_event   = snb_uncore_imc_enable_event,
+       .hw_config      = snb_uncore_imc_hw_config,
+       .read_counter   = snb_uncore_imc_read_counter,
+};
+
+static struct intel_uncore_type snb_uncore_imc = {
+       .name           = "imc",
+       .num_counters   = 2,
+       .num_boxes      = 1,
+       .fixed_ctr_bits = 32,
+       .fixed_ctr      = SNB_UNCORE_PCI_IMC_CTR_BASE,
+       .event_descs    = snb_uncore_imc_events,
+       .format_group   = &snb_uncore_imc_format_group,
+       .perf_ctr       = SNB_UNCORE_PCI_IMC_DATA_READS_BASE,
+       .event_mask     = SNB_UNCORE_PCI_IMC_EVENT_MASK,
+       .ops            = &snb_uncore_imc_ops,
+       .pmu            = &snb_uncore_imc_pmu,
+};
+
+static struct intel_uncore_type *snb_pci_uncores[] = {
+       [SNB_PCI_UNCORE_IMC]    = &snb_uncore_imc,
+       NULL,
+};
+
+static DEFINE_PCI_DEVICE_TABLE(snb_uncore_pci_ids) = {
+       { /* IMC */
+               PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_SNB_IMC),
+               .driver_data = UNCORE_PCI_DEV_DATA(SNB_PCI_UNCORE_IMC, 0),
+       },
+};
+
+static DEFINE_PCI_DEVICE_TABLE(ivb_uncore_pci_ids) = {
+       { /* IMC */
+               PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_IVB_IMC),
+               .driver_data = UNCORE_PCI_DEV_DATA(SNB_PCI_UNCORE_IMC, 0),
+       },
+};
+
+static DEFINE_PCI_DEVICE_TABLE(hsw_uncore_pci_ids) = {
+       { /* IMC */
+               PCI_DEVICE(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_HSW_IMC),
+               .driver_data = UNCORE_PCI_DEV_DATA(SNB_PCI_UNCORE_IMC, 0),
+       },
+};
+
+static struct pci_driver snb_uncore_pci_driver = {
+       .name           = "snb_uncore",
+       .id_table       = snb_uncore_pci_ids,
+};
+
+static struct pci_driver ivb_uncore_pci_driver = {
+       .name           = "ivb_uncore",
+       .id_table       = ivb_uncore_pci_ids,
+};
+
+static struct pci_driver hsw_uncore_pci_driver = {
+       .name           = "hsw_uncore",
+       .id_table       = hsw_uncore_pci_ids,
+};
+
 /* end of Sandy Bridge uncore support */
 
 /* Nehalem uncore support */
@@ -2781,6 +3161,7 @@ again:
 static enum hrtimer_restart uncore_pmu_hrtimer(struct hrtimer *hrtimer)
 {
        struct intel_uncore_box *box;
+       struct perf_event *event;
        unsigned long flags;
        int bit;
 
@@ -2793,19 +3174,27 @@ static enum hrtimer_restart uncore_pmu_hrtimer(struct hrtimer *hrtimer)
         */
        local_irq_save(flags);
 
+       /*
+        * handle boxes with an active event list as opposed to active
+        * counters
+        */
+       list_for_each_entry(event, &box->active_list, active_entry) {
+               uncore_perf_event_update(box, event);
+       }
+
        for_each_set_bit(bit, box->active_mask, UNCORE_PMC_IDX_MAX)
                uncore_perf_event_update(box, box->events[bit]);
 
        local_irq_restore(flags);
 
-       hrtimer_forward_now(hrtimer, ns_to_ktime(UNCORE_PMU_HRTIMER_INTERVAL));
+       hrtimer_forward_now(hrtimer, ns_to_ktime(box->hrtimer_duration));
        return HRTIMER_RESTART;
 }
 
 static void uncore_pmu_start_hrtimer(struct intel_uncore_box *box)
 {
        __hrtimer_start_range_ns(&box->hrtimer,
-                       ns_to_ktime(UNCORE_PMU_HRTIMER_INTERVAL), 0,
+                       ns_to_ktime(box->hrtimer_duration), 0,
                        HRTIMER_MODE_REL_PINNED, 0);
 }
 
@@ -2839,43 +3228,12 @@ static struct intel_uncore_box *uncore_alloc_box(struct intel_uncore_type *type,
        box->cpu = -1;
        box->phys_id = -1;
 
-       return box;
-}
+       /* set default hrtimer timeout */
+       box->hrtimer_duration = UNCORE_PMU_HRTIMER_INTERVAL;
 
-static struct intel_uncore_box *
-uncore_pmu_to_box(struct intel_uncore_pmu *pmu, int cpu)
-{
-       struct intel_uncore_box *box;
+       INIT_LIST_HEAD(&box->active_list);
 
-       box = *per_cpu_ptr(pmu->box, cpu);
-       if (box)
-               return box;
-
-       raw_spin_lock(&uncore_box_lock);
-       list_for_each_entry(box, &pmu->box_list, list) {
-               if (box->phys_id == topology_physical_package_id(cpu)) {
-                       atomic_inc(&box->refcnt);
-                       *per_cpu_ptr(pmu->box, cpu) = box;
-                       break;
-               }
-       }
-       raw_spin_unlock(&uncore_box_lock);
-
-       return *per_cpu_ptr(pmu->box, cpu);
-}
-
-static struct intel_uncore_pmu *uncore_event_to_pmu(struct perf_event *event)
-{
-       return container_of(event->pmu, struct intel_uncore_pmu, pmu);
-}
-
-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 box;
 }
 
 static int
@@ -3271,16 +3629,21 @@ static int __init uncore_pmu_register(struct intel_uncore_pmu *pmu)
 {
        int ret;
 
-       pmu->pmu = (struct pmu) {
-               .attr_groups    = pmu->type->attr_groups,
-               .task_ctx_nr    = perf_invalid_context,
-               .event_init     = uncore_pmu_event_init,
-               .add            = uncore_pmu_event_add,
-               .del            = uncore_pmu_event_del,
-               .start          = uncore_pmu_event_start,
-               .stop           = uncore_pmu_event_stop,
-               .read           = uncore_pmu_event_read,
-       };
+       if (!pmu->type->pmu) {
+               pmu->pmu = (struct pmu) {
+                       .attr_groups    = pmu->type->attr_groups,
+                       .task_ctx_nr    = perf_invalid_context,
+                       .event_init     = uncore_pmu_event_init,
+                       .add            = uncore_pmu_event_add,
+                       .del            = uncore_pmu_event_del,
+                       .start          = uncore_pmu_event_start,
+                       .stop           = uncore_pmu_event_stop,
+                       .read           = uncore_pmu_event_read,
+               };
+       } else {
+               pmu->pmu = *pmu->type->pmu;
+               pmu->pmu.attr_groups = pmu->type->attr_groups;
+       }
 
        if (pmu->type->num_boxes == 1) {
                if (strlen(pmu->type->name) > 0)
@@ -3493,6 +3856,28 @@ static int __init uncore_pci_init(void)
                pci_uncores = ivt_pci_uncores;
                uncore_pci_driver = &ivt_uncore_pci_driver;
                break;
+       case 42: /* Sandy Bridge */
+               ret = snb_pci2phy_map_init(PCI_DEVICE_ID_INTEL_SNB_IMC);
+               if (ret)
+                       return ret;
+               pci_uncores = snb_pci_uncores;
+               uncore_pci_driver = &snb_uncore_pci_driver;
+               break;
+       case 58: /* Ivy Bridge */
+               ret = snb_pci2phy_map_init(PCI_DEVICE_ID_INTEL_IVB_IMC);
+               if (ret)
+                       return ret;
+               pci_uncores = snb_pci_uncores;
+               uncore_pci_driver = &ivb_uncore_pci_driver;
+               break;
+       case 60: /* Haswell */
+       case 69: /* Haswell Celeron */
+               ret = snb_pci2phy_map_init(PCI_DEVICE_ID_INTEL_HSW_IMC);
+               if (ret)
+                       return ret;
+               pci_uncores = snb_pci_uncores;
+               uncore_pci_driver = &hsw_uncore_pci_driver;
+               break;
        default:
                return 0;
        }
@@ -3764,7 +4149,7 @@ static void __init uncore_cpu_setup(void *dummy)
 
 static int __init uncore_cpu_init(void)
 {
-       int ret, cpu, max_cores;
+       int ret, max_cores;
 
        max_cores = boot_cpu_data.x86_max_cores;
        switch (boot_cpu_data.x86_model) {
@@ -3808,29 +4193,6 @@ static int __init uncore_cpu_init(void)
        if (ret)
                return ret;
 
-       get_online_cpus();
-
-       for_each_online_cpu(cpu) {
-               int i, phys_id = topology_physical_package_id(cpu);
-
-               for_each_cpu(i, &uncore_cpu_mask) {
-                       if (phys_id == topology_physical_package_id(i)) {
-                               phys_id = -1;
-                               break;
-                       }
-               }
-               if (phys_id < 0)
-                       continue;
-
-               uncore_cpu_prepare(cpu, phys_id);
-               uncore_event_init_cpu(cpu);
-       }
-       on_each_cpu(uncore_cpu_setup, NULL, 1);
-
-       register_cpu_notifier(&uncore_cpu_nb);
-
-       put_online_cpus();
-
        return 0;
 }
 
@@ -3859,6 +4221,41 @@ static int __init uncore_pmus_register(void)
        return 0;
 }
 
+static void uncore_cpumask_init(void)
+{
+       int cpu;
+
+       /*
+        * ony invoke once from msr or pci init code
+        */
+       if (!cpumask_empty(&uncore_cpu_mask))
+               return;
+
+       get_online_cpus();
+
+       for_each_online_cpu(cpu) {
+               int i, phys_id = topology_physical_package_id(cpu);
+
+               for_each_cpu(i, &uncore_cpu_mask) {
+                       if (phys_id == topology_physical_package_id(i)) {
+                               phys_id = -1;
+                               break;
+                       }
+               }
+               if (phys_id < 0)
+                       continue;
+
+               uncore_cpu_prepare(cpu, phys_id);
+               uncore_event_init_cpu(cpu);
+       }
+       on_each_cpu(uncore_cpu_setup, NULL, 1);
+
+       register_cpu_notifier(&uncore_cpu_nb);
+
+       put_online_cpus();
+}
+
+
 static int __init intel_uncore_init(void)
 {
        int ret;
@@ -3877,6 +4274,7 @@ static int __init intel_uncore_init(void)
                uncore_pci_exit();
                goto fail;
        }
+       uncore_cpumask_init();
 
        uncore_pmus_register();
        return 0;
index a80ab71..90236f0 100644 (file)
@@ -6,6 +6,7 @@
 
 #define UNCORE_PMU_NAME_LEN            32
 #define UNCORE_PMU_HRTIMER_INTERVAL    (60LL * NSEC_PER_SEC)
+#define UNCORE_SNB_IMC_HRTIMER_INTERVAL (5ULL * NSEC_PER_SEC)
 
 #define UNCORE_FIXED_EVENT             0xff
 #define UNCORE_PMC_IDX_MAX_GENERIC     8
@@ -440,6 +441,7 @@ struct intel_uncore_type {
        struct intel_uncore_ops *ops;
        struct uncore_event_desc *event_descs;
        const struct attribute_group *attr_groups[4];
+       struct pmu *pmu; /* for custom pmu ops */
 };
 
 #define pmu_group attr_groups[0]
@@ -488,8 +490,11 @@ struct intel_uncore_box {
        u64 tags[UNCORE_PMC_IDX_MAX];
        struct pci_dev *pci_dev;
        struct intel_uncore_pmu *pmu;
+       u64 hrtimer_duration; /* hrtimer timeout for this box */
        struct hrtimer hrtimer;
        struct list_head list;
+       struct list_head active_list;
+       void *io_addr;
        struct intel_uncore_extra_reg shared_regs[0];
 };
 
index add13c8..19ae05d 100644 (file)
@@ -32,7 +32,7 @@ void init_irq_work(struct irq_work *work, void (*func)(struct irq_work *))
 
 #define DEFINE_IRQ_WORK(name, _f) struct irq_work name = { .func = (_f), }
 
-void irq_work_queue(struct irq_work *work);
+bool irq_work_queue(struct irq_work *work);
 void irq_work_run(void);
 void irq_work_sync(struct irq_work *work);
 
index 97fbecd..7399e6a 100644 (file)
 
 #define PCI_VENDOR_ID_INTEL            0x8086
 #define PCI_DEVICE_ID_INTEL_EESSC      0x0008
+#define PCI_DEVICE_ID_INTEL_SNB_IMC    0x0100
+#define PCI_DEVICE_ID_INTEL_IVB_IMC    0x0154
+#define PCI_DEVICE_ID_INTEL_HSW_IMC    0x0c00
 #define PCI_DEVICE_ID_INTEL_PXHD_0     0x0320
 #define PCI_DEVICE_ID_INTEL_PXHD_1     0x0321
 #define PCI_DEVICE_ID_INTEL_PXH_0      0x0329
index 2067cbb..45e5543 100644 (file)
@@ -243,7 +243,7 @@ static void perf_duration_warn(struct irq_work *w)
        printk_ratelimited(KERN_WARNING
                        "perf interrupt took too long (%lld > %lld), lowering "
                        "kernel.perf_event_max_sample_rate to %d\n",
-                       avg_local_sample_len, allowed_ns,
+                       avg_local_sample_len, allowed_ns >> 1,
                        sysctl_perf_event_sample_rate);
 }
 
@@ -283,7 +283,12 @@ void perf_sample_event_took(u64 sample_len_ns)
 
        update_perf_cpu_limits();
 
-       irq_work_queue(&perf_duration_work);
+       if (!irq_work_queue(&perf_duration_work)) {
+               early_printk("perf interrupt took too long (%lld > %lld), lowering "
+                            "kernel.perf_event_max_sample_rate to %d\n",
+                            avg_local_sample_len, allowed_ns >> 1,
+                            sysctl_perf_event_sample_rate);
+       }
 }
 
 static atomic64_t perf_event_id;
index 55fcce6..a82170e 100644 (file)
@@ -61,11 +61,11 @@ void __weak arch_irq_work_raise(void)
  *
  * Can be re-enqueued while the callback is still in progress.
  */
-void irq_work_queue(struct irq_work *work)
+bool irq_work_queue(struct irq_work *work)
 {
        /* Only queue if not already pending */
        if (!irq_work_claim(work))
-               return;
+               return false;
 
        /* Queue the entry and raise the IPI if needed. */
        preempt_disable();
@@ -83,6 +83,8 @@ void irq_work_queue(struct irq_work *work)
        }
 
        preempt_enable();
+
+       return true;
 }
 EXPORT_SYMBOL_GPL(irq_work_queue);