Merge branch 'perf/urgent' of git://github.com/acmel/linux into perf/urgent

diff --git a/arch/x86/kernel/cpu/perf_event_amd_ibs.c b/arch/x86/kernel/cpu/perf_event_amd_ibs.c
index ab6343d21825d7d9328fae81fd6faef3bf092a13..3b8a2d30d14e8ebeb2c58406212fbbd3c79ba935 100644 (file)
--- a/arch/x86/kernel/cpu/perf_event_amd_ibs.c
+++ b/arch/x86/kernel/cpu/perf_event_amd_ibs.c
@@ -199,8 +199,7 @@ static int force_ibs_eilvt_setup(void)
                goto out;
        }
 
-       pr_err(FW_BUG "using offset %d for IBS interrupts\n", offset);
-       pr_err(FW_BUG "workaround enabled for IBS LVT offset\n");
+       pr_info("IBS: LVT offset %d assigned\n", offset);
 
        return 0;
 out:
@@ -265,19 +264,23 @@ perf_ibs_cpu_notifier(struct notifier_block *self, unsigned long action, void *h
 static __init int amd_ibs_init(void)
 {
        u32 caps;
-       int ret;
+       int ret = -EINVAL;
 
        caps = __get_ibs_caps();
        if (!caps)
                return -ENODEV; /* ibs not supported by the cpu */
 
-       if (!ibs_eilvt_valid()) {
-               ret = force_ibs_eilvt_setup();
-               if (ret) {
-                       pr_err("Failed to setup IBS, %d\n", ret);
-                       return ret;
-               }
-       }
+       /*
+        * Force LVT offset assignment for family 10h: The offsets are
+        * not assigned by the BIOS for this family, so the OS is
+        * responsible for doing it. If the OS assignment fails, fall
+        * back to BIOS settings and try to setup this.
+        */
+       if (boot_cpu_data.x86 == 0x10)
+               force_ibs_eilvt_setup();
+
+       if (!ibs_eilvt_valid())
+               goto out;
 
        get_online_cpus();
        ibs_caps = caps;
@@ -287,7 +290,11 @@ static __init int amd_ibs_init(void)
        smp_call_function(setup_APIC_ibs, NULL, 1);
        put_online_cpus();
 
-       return perf_event_ibs_init();
+       ret = perf_event_ibs_init();
+out:
+       if (ret)
+               pr_err("Failed to setup IBS, %d\n", ret);
+       return ret;
 }
 
 /* Since we need the pci subsystem to init ibs we can't do this earlier: */

diff --git a/arch/x86/kernel/cpu/perf_event_intel.c b/arch/x86/kernel/cpu/perf_event_intel.c
index 2be5ebe9987209d41e76aede9e28ee9271070a34..8d601b18bf9f43688f2a6c5cb90994e00a4fa8d8 100644 (file)
--- a/arch/x86/kernel/cpu/perf_event_intel.c
+++ b/arch/x86/kernel/cpu/perf_event_intel.c
@@ -1545,6 +1545,13 @@ static void intel_clovertown_quirks(void)
        x86_pmu.pebs_constraints = NULL;
 }
 
+static void intel_sandybridge_quirks(void)
+{
+       printk(KERN_WARNING "PEBS disabled due to CPU errata.\n");
+       x86_pmu.pebs = 0;
+       x86_pmu.pebs_constraints = NULL;
+}
+
 __init int intel_pmu_init(void)
 {
        union cpuid10_edx edx;
@@ -1694,6 +1701,7 @@ __init int intel_pmu_init(void)
                break;
 
        case 42: /* SandyBridge */
+               x86_pmu.quirks = intel_sandybridge_quirks;
        case 45: /* SandyBridge, "Romely-EP" */
                memcpy(hw_cache_event_ids, snb_hw_cache_event_ids,
                       sizeof(hw_cache_event_ids));

diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h
index 1e9ebe5e0091e7fa1b1fabb72f72af979f35104c..b1f89122bf6a820102f43714fd42246d1dadd207 100644 (file)
--- a/include/linux/perf_event.h
+++ b/include/linux/perf_event.h
@@ -822,6 +822,7 @@ struct perf_event {
        int                             mmap_locked;
        struct user_struct              *mmap_user;
        struct ring_buffer              *rb;
+       struct list_head                rb_entry;
 
        /* poll related */
        wait_queue_head_t               waitq;

diff --git a/kernel/events/core.c b/kernel/events/core.c
index b0c1186fd97bcdaed7e8cdb537d9a61740bb86f3..600c1629b64db7c3fdb88c1e4c6d08412118ed1b 100644 (file)
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -185,6 +185,9 @@ static void cpu_ctx_sched_in(struct perf_cpu_context *cpuctx,
 static void update_context_time(struct perf_event_context *ctx);
 static u64 perf_event_time(struct perf_event *event);
 
+static void ring_buffer_attach(struct perf_event *event,
+                              struct ring_buffer *rb);
+
 void __weak perf_event_print_debug(void)       { }
 
 extern __weak const char *perf_pmu_name(void)
@@ -3191,12 +3194,33 @@ static unsigned int perf_poll(struct file *file, poll_table *wait)
        struct ring_buffer *rb;
        unsigned int events = POLL_HUP;
 
+       /*
+        * Race between perf_event_set_output() and perf_poll(): perf_poll()
+        * grabs the rb reference but perf_event_set_output() overrides it.
+        * Here is the timeline for two threads T1, T2:
+        * t0: T1, rb = rcu_dereference(event->rb)
+        * t1: T2, old_rb = event->rb
+        * t2: T2, event->rb = new rb
+        * t3: T2, ring_buffer_detach(old_rb)
+        * t4: T1, ring_buffer_attach(rb1)
+        * t5: T1, poll_wait(event->waitq)
+        *
+        * To avoid this problem, we grab mmap_mutex in perf_poll()
+        * thereby ensuring that the assignment of the new ring buffer
+        * and the detachment of the old buffer appear atomic to perf_poll()
+        */
+       mutex_lock(&event->mmap_mutex);
+
        rcu_read_lock();
        rb = rcu_dereference(event->rb);
-       if (rb)
+       if (rb) {
+               ring_buffer_attach(event, rb);
                events = atomic_xchg(&rb->poll, 0);
+       }
        rcu_read_unlock();
 
+       mutex_unlock(&event->mmap_mutex);
+
        poll_wait(file, &event->waitq, wait);
 
        return events;
@@ -3497,6 +3521,49 @@ unlock:
        return ret;
 }
 
+static void ring_buffer_attach(struct perf_event *event,
+                              struct ring_buffer *rb)
+{
+       unsigned long flags;
+
+       if (!list_empty(&event->rb_entry))
+               return;
+
+       spin_lock_irqsave(&rb->event_lock, flags);
+       if (!list_empty(&event->rb_entry))
+               goto unlock;
+
+       list_add(&event->rb_entry, &rb->event_list);
+unlock:
+       spin_unlock_irqrestore(&rb->event_lock, flags);
+}
+
+static void ring_buffer_detach(struct perf_event *event,
+                              struct ring_buffer *rb)
+{
+       unsigned long flags;
+
+       if (list_empty(&event->rb_entry))
+               return;
+
+       spin_lock_irqsave(&rb->event_lock, flags);
+       list_del_init(&event->rb_entry);
+       wake_up_all(&event->waitq);
+       spin_unlock_irqrestore(&rb->event_lock, flags);
+}
+
+static void ring_buffer_wakeup(struct perf_event *event)
+{
+       struct ring_buffer *rb;
+
+       rcu_read_lock();
+       rb = rcu_dereference(event->rb);
+       list_for_each_entry_rcu(event, &rb->event_list, rb_entry) {
+               wake_up_all(&event->waitq);
+       }
+       rcu_read_unlock();
+}
+
 static void rb_free_rcu(struct rcu_head *rcu_head)
 {
        struct ring_buffer *rb;
@@ -3522,9 +3589,19 @@ static struct ring_buffer *ring_buffer_get(struct perf_event *event)
 
 static void ring_buffer_put(struct ring_buffer *rb)
 {
+       struct perf_event *event, *n;
+       unsigned long flags;
+
        if (!atomic_dec_and_test(&rb->refcount))
                return;
 
+       spin_lock_irqsave(&rb->event_lock, flags);
+       list_for_each_entry_safe(event, n, &rb->event_list, rb_entry) {
+               list_del_init(&event->rb_entry);
+               wake_up_all(&event->waitq);
+       }
+       spin_unlock_irqrestore(&rb->event_lock, flags);
+
        call_rcu(&rb->rcu_head, rb_free_rcu);
 }
 
@@ -3547,6 +3624,7 @@ static void perf_mmap_close(struct vm_area_struct *vma)
                atomic_long_sub((size >> PAGE_SHIFT) + 1, &user->locked_vm);
                vma->vm_mm->pinned_vm -= event->mmap_locked;
                rcu_assign_pointer(event->rb, NULL);
+               ring_buffer_detach(event, rb);
                mutex_unlock(&event->mmap_mutex);
 
                ring_buffer_put(rb);
@@ -3701,7 +3779,7 @@ static const struct file_operations perf_fops = {
 
 void perf_event_wakeup(struct perf_event *event)
 {
-       wake_up_all(&event->waitq);
+       ring_buffer_wakeup(event);
 
        if (event->pending_kill) {
                kill_fasync(&event->fasync, SIGIO, event->pending_kill);
@@ -5823,6 +5901,8 @@ perf_event_alloc(struct perf_event_attr *attr, int cpu,
        INIT_LIST_HEAD(&event->group_entry);
        INIT_LIST_HEAD(&event->event_entry);
        INIT_LIST_HEAD(&event->sibling_list);
+       INIT_LIST_HEAD(&event->rb_entry);
+
        init_waitqueue_head(&event->waitq);
        init_irq_work(&event->pending, perf_pending_event);
 
@@ -6029,6 +6109,8 @@ set:
 
        old_rb = event->rb;
        rcu_assign_pointer(event->rb, rb);
+       if (old_rb)
+               ring_buffer_detach(event, old_rb);
        ret = 0;
 unlock:
        mutex_unlock(&event->mmap_mutex);

diff --git a/kernel/events/internal.h b/kernel/events/internal.h
index 09097dd8116c0e0bf5120d4da26c5f539a7f600a..64568a699375f105232eb588963da8707f926295 100644 (file)
--- a/kernel/events/internal.h
+++ b/kernel/events/internal.h
@@ -22,6 +22,9 @@ struct ring_buffer {
        local_t                         lost;           /* nr records lost   */
 
        long                            watermark;      /* wakeup watermark  */
+       /* poll crap */
+       spinlock_t                      event_lock;
+       struct list_head                event_list;
 
        struct perf_event_mmap_page     *user_page;
        void                            *data_pages[0];

diff --git a/kernel/events/ring_buffer.c b/kernel/events/ring_buffer.c
index a2a29205cc0fc10913277132162e0a515a944552..7f3011c6b57fa3288c7e4c46a555736763765a84 100644 (file)
--- a/kernel/events/ring_buffer.c
+++ b/kernel/events/ring_buffer.c
@@ -209,6 +209,9 @@ ring_buffer_init(struct ring_buffer *rb, long watermark, int flags)
                rb->writable = 1;
 
        atomic_set(&rb->refcount, 1);
+
+       INIT_LIST_HEAD(&rb->event_list);
+       spin_lock_init(&rb->event_lock);
 }
 
 #ifndef CONFIG_PERF_USE_VMALLOC