hw breakpoints: Move to kernel/events/
authorBorislav Petkov <borislav.petkov@amd.com>
Tue, 3 May 2011 13:26:43 +0000 (15:26 +0200)
committerBorislav Petkov <borislav.petkov@amd.com>
Tue, 3 May 2011 13:26:43 +0000 (15:26 +0200)
As part of the events sybsystem unification, relocate hw_breakpoint.c
into its new destination.

Cc: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: Borislav Petkov <borislav.petkov@amd.com>
kernel/Makefile
kernel/events/Makefile
kernel/events/hw_breakpoint.c [new file with mode: 0644]
kernel/hw_breakpoint.c [deleted file]

index 79815306474f7f1f08a30d15189675425b2f61e5..e9cf19155b4620a061af103c9732fc5e7d7a8290 100644 (file)
@@ -105,7 +105,6 @@ obj-$(CONFIG_IRQ_WORK) += irq_work.o
 
 obj-$(CONFIG_PERF_EVENTS) += events/
 
-obj-$(CONFIG_HAVE_HW_BREAKPOINT) += hw_breakpoint.o
 obj-$(CONFIG_USER_RETURN_NOTIFIER) += user-return-notifier.o
 obj-$(CONFIG_PADATA) += padata.o
 obj-$(CONFIG_CRASH_DUMP) += crash_dump.o
index 26c00e4570e5d5cb041a85c594ba869f6fa4db08..1ce23d3d8394f3c1533b2e777b2da79ce6ce565d 100644 (file)
@@ -2,4 +2,5 @@ ifdef CONFIG_FUNCTION_TRACER
 CFLAGS_REMOVE_core.o = -pg
 endif
 
-obj-y += core.o
+obj-y := core.o
+obj-$(CONFIG_HAVE_HW_BREAKPOINT) += hw_breakpoint.o
diff --git a/kernel/events/hw_breakpoint.c b/kernel/events/hw_breakpoint.c
new file mode 100644 (file)
index 0000000..086adf2
--- /dev/null
@@ -0,0 +1,659 @@
+/*
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
+ *
+ * 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>
+ */
+
+/*
+ * HW_breakpoint: a unified kernel/user-space hardware breakpoint facility,
+ * using the CPU's debug registers.
+ * This file contains the arch-independent routines.
+ */
+
+#include <linux/irqflags.h>
+#include <linux/kallsyms.h>
+#include <linux/notifier.h>
+#include <linux/kprobes.h>
+#include <linux/kdebug.h>
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/percpu.h>
+#include <linux/sched.h>
+#include <linux/init.h>
+#include <linux/slab.h>
+#include <linux/list.h>
+#include <linux/cpu.h>
+#include <linux/smp.h>
+
+#include <linux/hw_breakpoint.h>
+
+
+/*
+ * Constraints data
+ */
+
+/* Number of pinned cpu breakpoints in a cpu */
+static DEFINE_PER_CPU(unsigned int, nr_cpu_bp_pinned[TYPE_MAX]);
+
+/* Number of pinned task breakpoints in a cpu */
+static DEFINE_PER_CPU(unsigned int *, nr_task_bp_pinned[TYPE_MAX]);
+
+/* Number of non-pinned cpu/task breakpoints in a cpu */
+static DEFINE_PER_CPU(unsigned int, nr_bp_flexible[TYPE_MAX]);
+
+static int nr_slots[TYPE_MAX];
+
+/* Keep track of the breakpoints attached to tasks */
+static LIST_HEAD(bp_task_head);
+
+static int constraints_initialized;
+
+/* 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);
+
+__weak int hw_breakpoint_weight(struct perf_event *bp)
+{
+       return 1;
+}
+
+static inline enum bp_type_idx find_slot_idx(struct perf_event *bp)
+{
+       if (bp->attr.bp_type & HW_BREAKPOINT_RW)
+               return TYPE_DATA;
+
+       return TYPE_INST;
+}
+
+/*
+ * Report the maximum number of pinned breakpoints a task
+ * have in this cpu
+ */
+static unsigned int max_task_bp_pinned(int cpu, enum bp_type_idx type)
+{
+       int i;
+       unsigned int *tsk_pinned = per_cpu(nr_task_bp_pinned[type], cpu);
+
+       for (i = nr_slots[type] - 1; i >= 0; i--) {
+               if (tsk_pinned[i] > 0)
+                       return i + 1;
+       }
+
+       return 0;
+}
+
+/*
+ * Count the number of breakpoints of the same type and same task.
+ * The given event must be not on the list.
+ */
+static int task_bp_pinned(struct perf_event *bp, enum bp_type_idx type)
+{
+       struct task_struct *tsk = bp->hw.bp_target;
+       struct perf_event *iter;
+       int count = 0;
+
+       list_for_each_entry(iter, &bp_task_head, hw.bp_list) {
+               if (iter->hw.bp_target == tsk && find_slot_idx(iter) == type)
+                       count += hw_breakpoint_weight(iter);
+       }
+
+       return count;
+}
+
+/*
+ * 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, struct perf_event *bp,
+                   enum bp_type_idx type)
+{
+       int cpu = bp->cpu;
+       struct task_struct *tsk = bp->hw.bp_target;
+
+       if (cpu >= 0) {
+               slots->pinned = per_cpu(nr_cpu_bp_pinned[type], cpu);
+               if (!tsk)
+                       slots->pinned += max_task_bp_pinned(cpu, type);
+               else
+                       slots->pinned += task_bp_pinned(bp, type);
+               slots->flexible = per_cpu(nr_bp_flexible[type], cpu);
+
+               return;
+       }
+
+       for_each_online_cpu(cpu) {
+               unsigned int nr;
+
+               nr = per_cpu(nr_cpu_bp_pinned[type], cpu);
+               if (!tsk)
+                       nr += max_task_bp_pinned(cpu, type);
+               else
+                       nr += task_bp_pinned(bp, type);
+
+               if (nr > slots->pinned)
+                       slots->pinned = nr;
+
+               nr = per_cpu(nr_bp_flexible[type], cpu);
+
+               if (nr > slots->flexible)
+                       slots->flexible = nr;
+       }
+}
+
+/*
+ * For now, continue to consider flexible as pinned, until we can
+ * ensure no flexible event can ever be scheduled before a pinned event
+ * in a same cpu.
+ */
+static void
+fetch_this_slot(struct bp_busy_slots *slots, int weight)
+{
+       slots->pinned += weight;
+}
+
+/*
+ * Add a pinned breakpoint for the given task in our constraint table
+ */
+static void toggle_bp_task_slot(struct perf_event *bp, int cpu, bool enable,
+                               enum bp_type_idx type, int weight)
+{
+       unsigned int *tsk_pinned;
+       int old_count = 0;
+       int old_idx = 0;
+       int idx = 0;
+
+       old_count = task_bp_pinned(bp, type);
+       old_idx = old_count - 1;
+       idx = old_idx + weight;
+
+       /* tsk_pinned[n] is the number of tasks having n breakpoints */
+       tsk_pinned = per_cpu(nr_task_bp_pinned[type], cpu);
+       if (enable) {
+               tsk_pinned[idx]++;
+               if (old_count > 0)
+                       tsk_pinned[old_idx]--;
+       } else {
+               tsk_pinned[idx]--;
+               if (old_count > 0)
+                       tsk_pinned[old_idx]++;
+       }
+}
+
+/*
+ * Add/remove the given breakpoint in our constraint table
+ */
+static void
+toggle_bp_slot(struct perf_event *bp, bool enable, enum bp_type_idx type,
+              int weight)
+{
+       int cpu = bp->cpu;
+       struct task_struct *tsk = bp->hw.bp_target;
+
+       /* Pinned counter cpu profiling */
+       if (!tsk) {
+
+               if (enable)
+                       per_cpu(nr_cpu_bp_pinned[type], bp->cpu) += weight;
+               else
+                       per_cpu(nr_cpu_bp_pinned[type], bp->cpu) -= weight;
+               return;
+       }
+
+       /* Pinned counter task profiling */
+
+       if (!enable)
+               list_del(&bp->hw.bp_list);
+
+       if (cpu >= 0) {
+               toggle_bp_task_slot(bp, cpu, enable, type, weight);
+       } else {
+               for_each_online_cpu(cpu)
+                       toggle_bp_task_slot(bp, cpu, enable, type, weight);
+       }
+
+       if (enable)
+               list_add_tail(&bp->hw.bp_list, &bp_task_head);
+}
+
+/*
+ * Function to perform processor-specific cleanup during unregistration
+ */
+__weak void arch_unregister_hw_breakpoint(struct perf_event *bp)
+{
+       /*
+        * A weak stub function here for those archs that don't define
+        * it inside arch/.../kernel/hw_breakpoint.c
+        */
+}
+
+/*
+ * 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(nr_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(nr_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(nr_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(nr_task_bp_pinned, *))) < HBP_NUM
+ */
+static int __reserve_bp_slot(struct perf_event *bp)
+{
+       struct bp_busy_slots slots = {0};
+       enum bp_type_idx type;
+       int weight;
+
+       /* We couldn't initialize breakpoint constraints on boot */
+       if (!constraints_initialized)
+               return -ENOMEM;
+
+       /* Basic checks */
+       if (bp->attr.bp_type == HW_BREAKPOINT_EMPTY ||
+           bp->attr.bp_type == HW_BREAKPOINT_INVALID)
+               return -EINVAL;
+
+       type = find_slot_idx(bp);
+       weight = hw_breakpoint_weight(bp);
+
+       fetch_bp_busy_slots(&slots, bp, type);
+       /*
+        * Simulate the addition of this breakpoint to the constraints
+        * and see the result.
+        */
+       fetch_this_slot(&slots, weight);
+
+       /* Flexible counters need to keep at least one slot */
+       if (slots.pinned + (!!slots.flexible) > nr_slots[type])
+               return -ENOSPC;
+
+       toggle_bp_slot(bp, true, type, weight);
+
+       return 0;
+}
+
+int reserve_bp_slot(struct perf_event *bp)
+{
+       int ret;
+
+       mutex_lock(&nr_bp_mutex);
+
+       ret = __reserve_bp_slot(bp);
+
+       mutex_unlock(&nr_bp_mutex);
+
+       return ret;
+}
+
+static void __release_bp_slot(struct perf_event *bp)
+{
+       enum bp_type_idx type;
+       int weight;
+
+       type = find_slot_idx(bp);
+       weight = hw_breakpoint_weight(bp);
+       toggle_bp_slot(bp, false, type, weight);
+}
+
+void release_bp_slot(struct perf_event *bp)
+{
+       mutex_lock(&nr_bp_mutex);
+
+       arch_unregister_hw_breakpoint(bp);
+       __release_bp_slot(bp);
+
+       mutex_unlock(&nr_bp_mutex);
+}
+
+/*
+ * Allow the kernel debugger to reserve breakpoint slots without
+ * taking a lock using the dbg_* variant of for the reserve and
+ * release breakpoint slots.
+ */
+int dbg_reserve_bp_slot(struct perf_event *bp)
+{
+       if (mutex_is_locked(&nr_bp_mutex))
+               return -1;
+
+       return __reserve_bp_slot(bp);
+}
+
+int dbg_release_bp_slot(struct perf_event *bp)
+{
+       if (mutex_is_locked(&nr_bp_mutex))
+               return -1;
+
+       __release_bp_slot(bp);
+
+       return 0;
+}
+
+static int validate_hw_breakpoint(struct perf_event *bp)
+{
+       int ret;
+
+       ret = arch_validate_hwbkpt_settings(bp);
+       if (ret)
+               return ret;
+
+       if (arch_check_bp_in_kernelspace(bp)) {
+               if (bp->attr.exclude_kernel)
+                       return -EINVAL;
+               /*
+                * Don't let unprivileged users set a breakpoint in the trap
+                * path to avoid trap recursion attacks.
+                */
+               if (!capable(CAP_SYS_ADMIN))
+                       return -EPERM;
+       }
+
+       return 0;
+}
+
+int register_perf_hw_breakpoint(struct perf_event *bp)
+{
+       int ret;
+
+       ret = reserve_bp_slot(bp);
+       if (ret)
+               return ret;
+
+       ret = validate_hw_breakpoint(bp);
+
+       /* if arch_validate_hwbkpt_settings() fails then release bp slot */
+       if (ret)
+               release_bp_slot(bp);
+
+       return ret;
+}
+
+/**
+ * register_user_hw_breakpoint - register a hardware breakpoint for user space
+ * @attr: breakpoint attributes
+ * @triggered: callback to trigger when we hit the breakpoint
+ * @tsk: pointer to 'task_struct' of the process to which the address belongs
+ */
+struct perf_event *
+register_user_hw_breakpoint(struct perf_event_attr *attr,
+                           perf_overflow_handler_t triggered,
+                           struct task_struct *tsk)
+{
+       return perf_event_create_kernel_counter(attr, -1, tsk, triggered);
+}
+EXPORT_SYMBOL_GPL(register_user_hw_breakpoint);
+
+/**
+ * modify_user_hw_breakpoint - modify a user-space hardware breakpoint
+ * @bp: the breakpoint structure to modify
+ * @attr: new breakpoint attributes
+ * @triggered: callback to trigger when we hit the breakpoint
+ * @tsk: pointer to 'task_struct' of the process to which the address belongs
+ */
+int modify_user_hw_breakpoint(struct perf_event *bp, struct perf_event_attr *attr)
+{
+       u64 old_addr = bp->attr.bp_addr;
+       u64 old_len = bp->attr.bp_len;
+       int old_type = bp->attr.bp_type;
+       int err = 0;
+
+       perf_event_disable(bp);
+
+       bp->attr.bp_addr = attr->bp_addr;
+       bp->attr.bp_type = attr->bp_type;
+       bp->attr.bp_len = attr->bp_len;
+
+       if (attr->disabled)
+               goto end;
+
+       err = validate_hw_breakpoint(bp);
+       if (!err)
+               perf_event_enable(bp);
+
+       if (err) {
+               bp->attr.bp_addr = old_addr;
+               bp->attr.bp_type = old_type;
+               bp->attr.bp_len = old_len;
+               if (!bp->attr.disabled)
+                       perf_event_enable(bp);
+
+               return err;
+       }
+
+end:
+       bp->attr.disabled = attr->disabled;
+
+       return 0;
+}
+EXPORT_SYMBOL_GPL(modify_user_hw_breakpoint);
+
+/**
+ * unregister_hw_breakpoint - unregister a user-space hardware breakpoint
+ * @bp: the breakpoint structure to unregister
+ */
+void unregister_hw_breakpoint(struct perf_event *bp)
+{
+       if (!bp)
+               return;
+       perf_event_release_kernel(bp);
+}
+EXPORT_SYMBOL_GPL(unregister_hw_breakpoint);
+
+/**
+ * register_wide_hw_breakpoint - register a wide breakpoint in the kernel
+ * @attr: breakpoint attributes
+ * @triggered: callback to trigger when we hit the breakpoint
+ *
+ * @return a set of per_cpu pointers to perf events
+ */
+struct perf_event * __percpu *
+register_wide_hw_breakpoint(struct perf_event_attr *attr,
+                           perf_overflow_handler_t triggered)
+{
+       struct perf_event * __percpu *cpu_events, **pevent, *bp;
+       long err;
+       int cpu;
+
+       cpu_events = alloc_percpu(typeof(*cpu_events));
+       if (!cpu_events)
+               return (void __percpu __force *)ERR_PTR(-ENOMEM);
+
+       get_online_cpus();
+       for_each_online_cpu(cpu) {
+               pevent = per_cpu_ptr(cpu_events, cpu);
+               bp = perf_event_create_kernel_counter(attr, cpu, NULL, triggered);
+
+               *pevent = bp;
+
+               if (IS_ERR(bp)) {
+                       err = PTR_ERR(bp);
+                       goto fail;
+               }
+       }
+       put_online_cpus();
+
+       return cpu_events;
+
+fail:
+       for_each_online_cpu(cpu) {
+               pevent = per_cpu_ptr(cpu_events, cpu);
+               if (IS_ERR(*pevent))
+                       break;
+               unregister_hw_breakpoint(*pevent);
+       }
+       put_online_cpus();
+
+       free_percpu(cpu_events);
+       return (void __percpu __force *)ERR_PTR(err);
+}
+EXPORT_SYMBOL_GPL(register_wide_hw_breakpoint);
+
+/**
+ * unregister_wide_hw_breakpoint - unregister a wide breakpoint in the kernel
+ * @cpu_events: the per cpu set of events to unregister
+ */
+void unregister_wide_hw_breakpoint(struct perf_event * __percpu *cpu_events)
+{
+       int cpu;
+       struct perf_event **pevent;
+
+       for_each_possible_cpu(cpu) {
+               pevent = per_cpu_ptr(cpu_events, cpu);
+               unregister_hw_breakpoint(*pevent);
+       }
+       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,
+       /* we need to be notified first */
+       .priority = 0x7fffffff
+};
+
+static void bp_perf_event_destroy(struct perf_event *event)
+{
+       release_bp_slot(event);
+}
+
+static int hw_breakpoint_event_init(struct perf_event *bp)
+{
+       int err;
+
+       if (bp->attr.type != PERF_TYPE_BREAKPOINT)
+               return -ENOENT;
+
+       err = register_perf_hw_breakpoint(bp);
+       if (err)
+               return err;
+
+       bp->destroy = bp_perf_event_destroy;
+
+       return 0;
+}
+
+static int hw_breakpoint_add(struct perf_event *bp, int flags)
+{
+       if (!(flags & PERF_EF_START))
+               bp->hw.state = PERF_HES_STOPPED;
+
+       return arch_install_hw_breakpoint(bp);
+}
+
+static void hw_breakpoint_del(struct perf_event *bp, int flags)
+{
+       arch_uninstall_hw_breakpoint(bp);
+}
+
+static void hw_breakpoint_start(struct perf_event *bp, int flags)
+{
+       bp->hw.state = 0;
+}
+
+static void hw_breakpoint_stop(struct perf_event *bp, int flags)
+{
+       bp->hw.state = PERF_HES_STOPPED;
+}
+
+static struct pmu perf_breakpoint = {
+       .task_ctx_nr    = perf_sw_context, /* could eventually get its own */
+
+       .event_init     = hw_breakpoint_event_init,
+       .add            = hw_breakpoint_add,
+       .del            = hw_breakpoint_del,
+       .start          = hw_breakpoint_start,
+       .stop           = hw_breakpoint_stop,
+       .read           = hw_breakpoint_pmu_read,
+};
+
+int __init init_hw_breakpoint(void)
+{
+       unsigned int **task_bp_pinned;
+       int cpu, err_cpu;
+       int i;
+
+       for (i = 0; i < TYPE_MAX; i++)
+               nr_slots[i] = hw_breakpoint_slots(i);
+
+       for_each_possible_cpu(cpu) {
+               for (i = 0; i < TYPE_MAX; i++) {
+                       task_bp_pinned = &per_cpu(nr_task_bp_pinned[i], cpu);
+                       *task_bp_pinned = kzalloc(sizeof(int) * nr_slots[i],
+                                                 GFP_KERNEL);
+                       if (!*task_bp_pinned)
+                               goto err_alloc;
+               }
+       }
+
+       constraints_initialized = 1;
+
+       perf_pmu_register(&perf_breakpoint, "breakpoint", PERF_TYPE_BREAKPOINT);
+
+       return register_die_notifier(&hw_breakpoint_exceptions_nb);
+
+ err_alloc:
+       for_each_possible_cpu(err_cpu) {
+               if (err_cpu == cpu)
+                       break;
+               for (i = 0; i < TYPE_MAX; i++)
+                       kfree(per_cpu(nr_task_bp_pinned[i], cpu));
+       }
+
+       return -ENOMEM;
+}
+
+
diff --git a/kernel/hw_breakpoint.c b/kernel/hw_breakpoint.c
deleted file mode 100644 (file)
index 086adf2..0000000
+++ /dev/null
@@ -1,659 +0,0 @@
-/*
- * This program is free software; you can redistribute it and/or modify
- * it under the terms of the GNU General Public License as published by
- * the Free Software Foundation; either version 2 of the License, or
- * (at your option) any later version.
- *
- * This program is distributed in the hope that it will be useful,
- * but WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
- * GNU General Public License for more details.
- *
- * You should have received a copy of the GNU General Public License
- * along with this program; if not, write to the Free Software
- * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
- *
- * 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>
- */
-
-/*
- * HW_breakpoint: a unified kernel/user-space hardware breakpoint facility,
- * using the CPU's debug registers.
- * This file contains the arch-independent routines.
- */
-
-#include <linux/irqflags.h>
-#include <linux/kallsyms.h>
-#include <linux/notifier.h>
-#include <linux/kprobes.h>
-#include <linux/kdebug.h>
-#include <linux/kernel.h>
-#include <linux/module.h>
-#include <linux/percpu.h>
-#include <linux/sched.h>
-#include <linux/init.h>
-#include <linux/slab.h>
-#include <linux/list.h>
-#include <linux/cpu.h>
-#include <linux/smp.h>
-
-#include <linux/hw_breakpoint.h>
-
-
-/*
- * Constraints data
- */
-
-/* Number of pinned cpu breakpoints in a cpu */
-static DEFINE_PER_CPU(unsigned int, nr_cpu_bp_pinned[TYPE_MAX]);
-
-/* Number of pinned task breakpoints in a cpu */
-static DEFINE_PER_CPU(unsigned int *, nr_task_bp_pinned[TYPE_MAX]);
-
-/* Number of non-pinned cpu/task breakpoints in a cpu */
-static DEFINE_PER_CPU(unsigned int, nr_bp_flexible[TYPE_MAX]);
-
-static int nr_slots[TYPE_MAX];
-
-/* Keep track of the breakpoints attached to tasks */
-static LIST_HEAD(bp_task_head);
-
-static int constraints_initialized;
-
-/* 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);
-
-__weak int hw_breakpoint_weight(struct perf_event *bp)
-{
-       return 1;
-}
-
-static inline enum bp_type_idx find_slot_idx(struct perf_event *bp)
-{
-       if (bp->attr.bp_type & HW_BREAKPOINT_RW)
-               return TYPE_DATA;
-
-       return TYPE_INST;
-}
-
-/*
- * Report the maximum number of pinned breakpoints a task
- * have in this cpu
- */
-static unsigned int max_task_bp_pinned(int cpu, enum bp_type_idx type)
-{
-       int i;
-       unsigned int *tsk_pinned = per_cpu(nr_task_bp_pinned[type], cpu);
-
-       for (i = nr_slots[type] - 1; i >= 0; i--) {
-               if (tsk_pinned[i] > 0)
-                       return i + 1;
-       }
-
-       return 0;
-}
-
-/*
- * Count the number of breakpoints of the same type and same task.
- * The given event must be not on the list.
- */
-static int task_bp_pinned(struct perf_event *bp, enum bp_type_idx type)
-{
-       struct task_struct *tsk = bp->hw.bp_target;
-       struct perf_event *iter;
-       int count = 0;
-
-       list_for_each_entry(iter, &bp_task_head, hw.bp_list) {
-               if (iter->hw.bp_target == tsk && find_slot_idx(iter) == type)
-                       count += hw_breakpoint_weight(iter);
-       }
-
-       return count;
-}
-
-/*
- * 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, struct perf_event *bp,
-                   enum bp_type_idx type)
-{
-       int cpu = bp->cpu;
-       struct task_struct *tsk = bp->hw.bp_target;
-
-       if (cpu >= 0) {
-               slots->pinned = per_cpu(nr_cpu_bp_pinned[type], cpu);
-               if (!tsk)
-                       slots->pinned += max_task_bp_pinned(cpu, type);
-               else
-                       slots->pinned += task_bp_pinned(bp, type);
-               slots->flexible = per_cpu(nr_bp_flexible[type], cpu);
-
-               return;
-       }
-
-       for_each_online_cpu(cpu) {
-               unsigned int nr;
-
-               nr = per_cpu(nr_cpu_bp_pinned[type], cpu);
-               if (!tsk)
-                       nr += max_task_bp_pinned(cpu, type);
-               else
-                       nr += task_bp_pinned(bp, type);
-
-               if (nr > slots->pinned)
-                       slots->pinned = nr;
-
-               nr = per_cpu(nr_bp_flexible[type], cpu);
-
-               if (nr > slots->flexible)
-                       slots->flexible = nr;
-       }
-}
-
-/*
- * For now, continue to consider flexible as pinned, until we can
- * ensure no flexible event can ever be scheduled before a pinned event
- * in a same cpu.
- */
-static void
-fetch_this_slot(struct bp_busy_slots *slots, int weight)
-{
-       slots->pinned += weight;
-}
-
-/*
- * Add a pinned breakpoint for the given task in our constraint table
- */
-static void toggle_bp_task_slot(struct perf_event *bp, int cpu, bool enable,
-                               enum bp_type_idx type, int weight)
-{
-       unsigned int *tsk_pinned;
-       int old_count = 0;
-       int old_idx = 0;
-       int idx = 0;
-
-       old_count = task_bp_pinned(bp, type);
-       old_idx = old_count - 1;
-       idx = old_idx + weight;
-
-       /* tsk_pinned[n] is the number of tasks having n breakpoints */
-       tsk_pinned = per_cpu(nr_task_bp_pinned[type], cpu);
-       if (enable) {
-               tsk_pinned[idx]++;
-               if (old_count > 0)
-                       tsk_pinned[old_idx]--;
-       } else {
-               tsk_pinned[idx]--;
-               if (old_count > 0)
-                       tsk_pinned[old_idx]++;
-       }
-}
-
-/*
- * Add/remove the given breakpoint in our constraint table
- */
-static void
-toggle_bp_slot(struct perf_event *bp, bool enable, enum bp_type_idx type,
-              int weight)
-{
-       int cpu = bp->cpu;
-       struct task_struct *tsk = bp->hw.bp_target;
-
-       /* Pinned counter cpu profiling */
-       if (!tsk) {
-
-               if (enable)
-                       per_cpu(nr_cpu_bp_pinned[type], bp->cpu) += weight;
-               else
-                       per_cpu(nr_cpu_bp_pinned[type], bp->cpu) -= weight;
-               return;
-       }
-
-       /* Pinned counter task profiling */
-
-       if (!enable)
-               list_del(&bp->hw.bp_list);
-
-       if (cpu >= 0) {
-               toggle_bp_task_slot(bp, cpu, enable, type, weight);
-       } else {
-               for_each_online_cpu(cpu)
-                       toggle_bp_task_slot(bp, cpu, enable, type, weight);
-       }
-
-       if (enable)
-               list_add_tail(&bp->hw.bp_list, &bp_task_head);
-}
-
-/*
- * Function to perform processor-specific cleanup during unregistration
- */
-__weak void arch_unregister_hw_breakpoint(struct perf_event *bp)
-{
-       /*
-        * A weak stub function here for those archs that don't define
-        * it inside arch/.../kernel/hw_breakpoint.c
-        */
-}
-
-/*
- * 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(nr_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(nr_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(nr_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(nr_task_bp_pinned, *))) < HBP_NUM
- */
-static int __reserve_bp_slot(struct perf_event *bp)
-{
-       struct bp_busy_slots slots = {0};
-       enum bp_type_idx type;
-       int weight;
-
-       /* We couldn't initialize breakpoint constraints on boot */
-       if (!constraints_initialized)
-               return -ENOMEM;
-
-       /* Basic checks */
-       if (bp->attr.bp_type == HW_BREAKPOINT_EMPTY ||
-           bp->attr.bp_type == HW_BREAKPOINT_INVALID)
-               return -EINVAL;
-
-       type = find_slot_idx(bp);
-       weight = hw_breakpoint_weight(bp);
-
-       fetch_bp_busy_slots(&slots, bp, type);
-       /*
-        * Simulate the addition of this breakpoint to the constraints
-        * and see the result.
-        */
-       fetch_this_slot(&slots, weight);
-
-       /* Flexible counters need to keep at least one slot */
-       if (slots.pinned + (!!slots.flexible) > nr_slots[type])
-               return -ENOSPC;
-
-       toggle_bp_slot(bp, true, type, weight);
-
-       return 0;
-}
-
-int reserve_bp_slot(struct perf_event *bp)
-{
-       int ret;
-
-       mutex_lock(&nr_bp_mutex);
-
-       ret = __reserve_bp_slot(bp);
-
-       mutex_unlock(&nr_bp_mutex);
-
-       return ret;
-}
-
-static void __release_bp_slot(struct perf_event *bp)
-{
-       enum bp_type_idx type;
-       int weight;
-
-       type = find_slot_idx(bp);
-       weight = hw_breakpoint_weight(bp);
-       toggle_bp_slot(bp, false, type, weight);
-}
-
-void release_bp_slot(struct perf_event *bp)
-{
-       mutex_lock(&nr_bp_mutex);
-
-       arch_unregister_hw_breakpoint(bp);
-       __release_bp_slot(bp);
-
-       mutex_unlock(&nr_bp_mutex);
-}
-
-/*
- * Allow the kernel debugger to reserve breakpoint slots without
- * taking a lock using the dbg_* variant of for the reserve and
- * release breakpoint slots.
- */
-int dbg_reserve_bp_slot(struct perf_event *bp)
-{
-       if (mutex_is_locked(&nr_bp_mutex))
-               return -1;
-
-       return __reserve_bp_slot(bp);
-}
-
-int dbg_release_bp_slot(struct perf_event *bp)
-{
-       if (mutex_is_locked(&nr_bp_mutex))
-               return -1;
-
-       __release_bp_slot(bp);
-
-       return 0;
-}
-
-static int validate_hw_breakpoint(struct perf_event *bp)
-{
-       int ret;
-
-       ret = arch_validate_hwbkpt_settings(bp);
-       if (ret)
-               return ret;
-
-       if (arch_check_bp_in_kernelspace(bp)) {
-               if (bp->attr.exclude_kernel)
-                       return -EINVAL;
-               /*
-                * Don't let unprivileged users set a breakpoint in the trap
-                * path to avoid trap recursion attacks.
-                */
-               if (!capable(CAP_SYS_ADMIN))
-                       return -EPERM;
-       }
-
-       return 0;
-}
-
-int register_perf_hw_breakpoint(struct perf_event *bp)
-{
-       int ret;
-
-       ret = reserve_bp_slot(bp);
-       if (ret)
-               return ret;
-
-       ret = validate_hw_breakpoint(bp);
-
-       /* if arch_validate_hwbkpt_settings() fails then release bp slot */
-       if (ret)
-               release_bp_slot(bp);
-
-       return ret;
-}
-
-/**
- * register_user_hw_breakpoint - register a hardware breakpoint for user space
- * @attr: breakpoint attributes
- * @triggered: callback to trigger when we hit the breakpoint
- * @tsk: pointer to 'task_struct' of the process to which the address belongs
- */
-struct perf_event *
-register_user_hw_breakpoint(struct perf_event_attr *attr,
-                           perf_overflow_handler_t triggered,
-                           struct task_struct *tsk)
-{
-       return perf_event_create_kernel_counter(attr, -1, tsk, triggered);
-}
-EXPORT_SYMBOL_GPL(register_user_hw_breakpoint);
-
-/**
- * modify_user_hw_breakpoint - modify a user-space hardware breakpoint
- * @bp: the breakpoint structure to modify
- * @attr: new breakpoint attributes
- * @triggered: callback to trigger when we hit the breakpoint
- * @tsk: pointer to 'task_struct' of the process to which the address belongs
- */
-int modify_user_hw_breakpoint(struct perf_event *bp, struct perf_event_attr *attr)
-{
-       u64 old_addr = bp->attr.bp_addr;
-       u64 old_len = bp->attr.bp_len;
-       int old_type = bp->attr.bp_type;
-       int err = 0;
-
-       perf_event_disable(bp);
-
-       bp->attr.bp_addr = attr->bp_addr;
-       bp->attr.bp_type = attr->bp_type;
-       bp->attr.bp_len = attr->bp_len;
-
-       if (attr->disabled)
-               goto end;
-
-       err = validate_hw_breakpoint(bp);
-       if (!err)
-               perf_event_enable(bp);
-
-       if (err) {
-               bp->attr.bp_addr = old_addr;
-               bp->attr.bp_type = old_type;
-               bp->attr.bp_len = old_len;
-               if (!bp->attr.disabled)
-                       perf_event_enable(bp);
-
-               return err;
-       }
-
-end:
-       bp->attr.disabled = attr->disabled;
-
-       return 0;
-}
-EXPORT_SYMBOL_GPL(modify_user_hw_breakpoint);
-
-/**
- * unregister_hw_breakpoint - unregister a user-space hardware breakpoint
- * @bp: the breakpoint structure to unregister
- */
-void unregister_hw_breakpoint(struct perf_event *bp)
-{
-       if (!bp)
-               return;
-       perf_event_release_kernel(bp);
-}
-EXPORT_SYMBOL_GPL(unregister_hw_breakpoint);
-
-/**
- * register_wide_hw_breakpoint - register a wide breakpoint in the kernel
- * @attr: breakpoint attributes
- * @triggered: callback to trigger when we hit the breakpoint
- *
- * @return a set of per_cpu pointers to perf events
- */
-struct perf_event * __percpu *
-register_wide_hw_breakpoint(struct perf_event_attr *attr,
-                           perf_overflow_handler_t triggered)
-{
-       struct perf_event * __percpu *cpu_events, **pevent, *bp;
-       long err;
-       int cpu;
-
-       cpu_events = alloc_percpu(typeof(*cpu_events));
-       if (!cpu_events)
-               return (void __percpu __force *)ERR_PTR(-ENOMEM);
-
-       get_online_cpus();
-       for_each_online_cpu(cpu) {
-               pevent = per_cpu_ptr(cpu_events, cpu);
-               bp = perf_event_create_kernel_counter(attr, cpu, NULL, triggered);
-
-               *pevent = bp;
-
-               if (IS_ERR(bp)) {
-                       err = PTR_ERR(bp);
-                       goto fail;
-               }
-       }
-       put_online_cpus();
-
-       return cpu_events;
-
-fail:
-       for_each_online_cpu(cpu) {
-               pevent = per_cpu_ptr(cpu_events, cpu);
-               if (IS_ERR(*pevent))
-                       break;
-               unregister_hw_breakpoint(*pevent);
-       }
-       put_online_cpus();
-
-       free_percpu(cpu_events);
-       return (void __percpu __force *)ERR_PTR(err);
-}
-EXPORT_SYMBOL_GPL(register_wide_hw_breakpoint);
-
-/**
- * unregister_wide_hw_breakpoint - unregister a wide breakpoint in the kernel
- * @cpu_events: the per cpu set of events to unregister
- */
-void unregister_wide_hw_breakpoint(struct perf_event * __percpu *cpu_events)
-{
-       int cpu;
-       struct perf_event **pevent;
-
-       for_each_possible_cpu(cpu) {
-               pevent = per_cpu_ptr(cpu_events, cpu);
-               unregister_hw_breakpoint(*pevent);
-       }
-       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,
-       /* we need to be notified first */
-       .priority = 0x7fffffff
-};
-
-static void bp_perf_event_destroy(struct perf_event *event)
-{
-       release_bp_slot(event);
-}
-
-static int hw_breakpoint_event_init(struct perf_event *bp)
-{
-       int err;
-
-       if (bp->attr.type != PERF_TYPE_BREAKPOINT)
-               return -ENOENT;
-
-       err = register_perf_hw_breakpoint(bp);
-       if (err)
-               return err;
-
-       bp->destroy = bp_perf_event_destroy;
-
-       return 0;
-}
-
-static int hw_breakpoint_add(struct perf_event *bp, int flags)
-{
-       if (!(flags & PERF_EF_START))
-               bp->hw.state = PERF_HES_STOPPED;
-
-       return arch_install_hw_breakpoint(bp);
-}
-
-static void hw_breakpoint_del(struct perf_event *bp, int flags)
-{
-       arch_uninstall_hw_breakpoint(bp);
-}
-
-static void hw_breakpoint_start(struct perf_event *bp, int flags)
-{
-       bp->hw.state = 0;
-}
-
-static void hw_breakpoint_stop(struct perf_event *bp, int flags)
-{
-       bp->hw.state = PERF_HES_STOPPED;
-}
-
-static struct pmu perf_breakpoint = {
-       .task_ctx_nr    = perf_sw_context, /* could eventually get its own */
-
-       .event_init     = hw_breakpoint_event_init,
-       .add            = hw_breakpoint_add,
-       .del            = hw_breakpoint_del,
-       .start          = hw_breakpoint_start,
-       .stop           = hw_breakpoint_stop,
-       .read           = hw_breakpoint_pmu_read,
-};
-
-int __init init_hw_breakpoint(void)
-{
-       unsigned int **task_bp_pinned;
-       int cpu, err_cpu;
-       int i;
-
-       for (i = 0; i < TYPE_MAX; i++)
-               nr_slots[i] = hw_breakpoint_slots(i);
-
-       for_each_possible_cpu(cpu) {
-               for (i = 0; i < TYPE_MAX; i++) {
-                       task_bp_pinned = &per_cpu(nr_task_bp_pinned[i], cpu);
-                       *task_bp_pinned = kzalloc(sizeof(int) * nr_slots[i],
-                                                 GFP_KERNEL);
-                       if (!*task_bp_pinned)
-                               goto err_alloc;
-               }
-       }
-
-       constraints_initialized = 1;
-
-       perf_pmu_register(&perf_breakpoint, "breakpoint", PERF_TYPE_BREAKPOINT);
-
-       return register_die_notifier(&hw_breakpoint_exceptions_nb);
-
- err_alloc:
-       for_each_possible_cpu(err_cpu) {
-               if (err_cpu == cpu)
-                       break;
-               for (i = 0; i < TYPE_MAX; i++)
-                       kfree(per_cpu(nr_task_bp_pinned[i], cpu));
-       }
-
-       return -ENOMEM;
-}
-
-