hw-breakpoint: Attribute authorship of hw-breakpoint related files
[linux-3.10.git] / kernel / hw_breakpoint.c
index 08f6d01..c166622 100644 (file)
  * Copyright (C) 2007 Alan Stern
  * Copyright (C) IBM Corporation, 2009
  * Copyright (C) 2009, Frederic Weisbecker <fweisbec@gmail.com>
+ *
+ * Thanks to Ingo Molnar for his many suggestions.
+ *
+ * Authors: Alan Stern <stern@rowland.harvard.edu>
+ *          K.Prasad <prasad@linux.vnet.ibm.com>
+ *          Frederic Weisbecker <fweisbec@gmail.com>
  */
 
 /*
 
 #include <linux/hw_breakpoint.h>
 
-#include <asm/processor.h>
+/*
+ * Constraints data
+ */
+
+/* Number of pinned cpu breakpoints in a cpu */
+static DEFINE_PER_CPU(unsigned int, nr_cpu_bp_pinned);
 
-#ifdef CONFIG_X86
-#include <asm/debugreg.h>
-#endif
+/* Number of pinned task breakpoints in a cpu */
+static DEFINE_PER_CPU(unsigned int, task_bp_pinned[HBP_NUM]);
 
-static atomic_t bp_slot;
+/* Number of non-pinned cpu/task breakpoints in a cpu */
+static DEFINE_PER_CPU(unsigned int, nr_bp_flexible);
 
-int reserve_bp_slot(struct perf_event *bp)
+/* Gather the number of total pinned and un-pinned bp in a cpuset */
+struct bp_busy_slots {
+       unsigned int pinned;
+       unsigned int flexible;
+};
+
+/* Serialize accesses to the above constraints */
+static DEFINE_MUTEX(nr_bp_mutex);
+
+/*
+ * Report the maximum number of pinned breakpoints a task
+ * have in this cpu
+ */
+static unsigned int max_task_bp_pinned(int cpu)
 {
-       if (atomic_inc_return(&bp_slot) == HBP_NUM) {
-               atomic_dec(&bp_slot);
+       int i;
+       unsigned int *tsk_pinned = per_cpu(task_bp_pinned, cpu);
 
-               return -ENOSPC;
+       for (i = HBP_NUM -1; i >= 0; i--) {
+               if (tsk_pinned[i] > 0)
+                       return i + 1;
        }
 
        return 0;
 }
 
+/*
+ * Report the number of pinned/un-pinned breakpoints we have in
+ * a given cpu (cpu > -1) or in all of them (cpu = -1).
+ */
+static void fetch_bp_busy_slots(struct bp_busy_slots *slots, int cpu)
+{
+       if (cpu >= 0) {
+               slots->pinned = per_cpu(nr_cpu_bp_pinned, cpu);
+               slots->pinned += max_task_bp_pinned(cpu);
+               slots->flexible = per_cpu(nr_bp_flexible, cpu);
+
+               return;
+       }
+
+       for_each_online_cpu(cpu) {
+               unsigned int nr;
+
+               nr = per_cpu(nr_cpu_bp_pinned, cpu);
+               nr += max_task_bp_pinned(cpu);
+
+               if (nr > slots->pinned)
+                       slots->pinned = nr;
+
+               nr = per_cpu(nr_bp_flexible, cpu);
+
+               if (nr > slots->flexible)
+                       slots->flexible = nr;
+       }
+}
+
+/*
+ * Add a pinned breakpoint for the given task in our constraint table
+ */
+static void toggle_bp_task_slot(struct task_struct *tsk, int cpu, bool enable)
+{
+       int count = 0;
+       struct perf_event *bp;
+       struct perf_event_context *ctx = tsk->perf_event_ctxp;
+       unsigned int *task_bp_pinned;
+       struct list_head *list;
+       unsigned long flags;
+
+       if (WARN_ONCE(!ctx, "No perf context for this task"))
+               return;
+
+       list = &ctx->event_list;
+
+       spin_lock_irqsave(&ctx->lock, flags);
+
+       /*
+        * The current breakpoint counter is not included in the list
+        * at the open() callback time
+        */
+       list_for_each_entry(bp, list, event_entry) {
+               if (bp->attr.type == PERF_TYPE_BREAKPOINT)
+                       count++;
+       }
+
+       spin_unlock_irqrestore(&ctx->lock, flags);
+
+       if (WARN_ONCE(count < 0, "No breakpoint counter found in the counter list"))
+               return;
+
+       task_bp_pinned = per_cpu(task_bp_pinned, cpu);
+       if (enable) {
+               task_bp_pinned[count]++;
+               if (count > 0)
+                       task_bp_pinned[count-1]--;
+       } else {
+               task_bp_pinned[count]--;
+               if (count > 0)
+                       task_bp_pinned[count-1]++;
+       }
+}
+
+/*
+ * Add/remove the given breakpoint in our constraint table
+ */
+static void toggle_bp_slot(struct perf_event *bp, bool enable)
+{
+       int cpu = bp->cpu;
+       struct task_struct *tsk = bp->ctx->task;
+
+       /* Pinned counter task profiling */
+       if (tsk) {
+               if (cpu >= 0) {
+                       toggle_bp_task_slot(tsk, cpu, enable);
+                       return;
+               }
+
+               for_each_online_cpu(cpu)
+                       toggle_bp_task_slot(tsk, cpu, enable);
+               return;
+       }
+
+       /* Pinned counter cpu profiling */
+       if (enable)
+               per_cpu(nr_cpu_bp_pinned, bp->cpu)++;
+       else
+               per_cpu(nr_cpu_bp_pinned, bp->cpu)--;
+}
+
+/*
+ * Contraints to check before allowing this new breakpoint counter:
+ *
+ *  == Non-pinned counter == (Considered as pinned for now)
+ *
+ *   - If attached to a single cpu, check:
+ *
+ *       (per_cpu(nr_bp_flexible, cpu) || (per_cpu(nr_cpu_bp_pinned, cpu)
+ *           + max(per_cpu(task_bp_pinned, cpu)))) < HBP_NUM
+ *
+ *       -> If there are already non-pinned counters in this cpu, it means
+ *          there is already a free slot for them.
+ *          Otherwise, we check that the maximum number of per task
+ *          breakpoints (for this cpu) plus the number of per cpu breakpoint
+ *          (for this cpu) doesn't cover every registers.
+ *
+ *   - If attached to every cpus, check:
+ *
+ *       (per_cpu(nr_bp_flexible, *) || (max(per_cpu(nr_cpu_bp_pinned, *))
+ *           + max(per_cpu(task_bp_pinned, *)))) < HBP_NUM
+ *
+ *       -> This is roughly the same, except we check the number of per cpu
+ *          bp for every cpu and we keep the max one. Same for the per tasks
+ *          breakpoints.
+ *
+ *
+ * == Pinned counter ==
+ *
+ *   - If attached to a single cpu, check:
+ *
+ *       ((per_cpu(nr_bp_flexible, cpu) > 1) + per_cpu(nr_cpu_bp_pinned, cpu)
+ *            + max(per_cpu(task_bp_pinned, cpu))) < HBP_NUM
+ *
+ *       -> Same checks as before. But now the nr_bp_flexible, if any, must keep
+ *          one register at least (or they will never be fed).
+ *
+ *   - If attached to every cpus, check:
+ *
+ *       ((per_cpu(nr_bp_flexible, *) > 1) + max(per_cpu(nr_cpu_bp_pinned, *))
+ *            + max(per_cpu(task_bp_pinned, *))) < HBP_NUM
+ */
+int reserve_bp_slot(struct perf_event *bp)
+{
+       struct bp_busy_slots slots = {0};
+       int ret = 0;
+
+       mutex_lock(&nr_bp_mutex);
+
+       fetch_bp_busy_slots(&slots, bp->cpu);
+
+       /* Flexible counters need to keep at least one slot */
+       if (slots.pinned + (!!slots.flexible) == HBP_NUM) {
+               ret = -ENOSPC;
+               goto end;
+       }
+
+       toggle_bp_slot(bp, true);
+
+end:
+       mutex_unlock(&nr_bp_mutex);
+
+       return ret;
+}
+
 void release_bp_slot(struct perf_event *bp)
 {
-       atomic_dec(&bp_slot);
+       mutex_lock(&nr_bp_mutex);
+
+       toggle_bp_slot(bp, false);
+
+       mutex_unlock(&nr_bp_mutex);
 }
 
+
 int __register_perf_hw_breakpoint(struct perf_event *bp)
 {
        int ret;
@@ -255,6 +452,7 @@ fail:
        /* return the error if any */
        return ERR_PTR(err);
 }
+EXPORT_SYMBOL_GPL(register_wide_hw_breakpoint);
 
 /**
  * unregister_wide_hw_breakpoint - unregister a wide breakpoint in the kernel
@@ -271,7 +469,7 @@ void unregister_wide_hw_breakpoint(struct perf_event **cpu_events)
        }
        free_percpu(cpu_events);
 }
-
+EXPORT_SYMBOL_GPL(unregister_wide_hw_breakpoint);
 
 static struct notifier_block hw_breakpoint_exceptions_nb = {
        .notifier_call = hw_breakpoint_exceptions_notify,