powerpc/watchpoint: Don't allow concurrent perf and ptrace events

With Book3s DAWR, ptrace and perf watchpoints on powerpc behaves
differently. Ptrace watchpoint works in one-shot mode and generates
signal before executing instruction. It's ptrace user's job to
single-step the instruction and re-enable the watchpoint. OTOH, in
case of perf watchpoint, kernel emulates/single-steps the instruction
and then generates event. If perf and ptrace creates two events with
same or overlapping address ranges, it's ambiguous to decide who
should single-step the instruction. Because of this issue, don't
allow perf and ptrace watchpoint at the same time if their address
range overlaps.

Signed-off-by: Ravi Bangoria <ravi.bangoria@linux.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Reviewed-by: Michael Neuling <mikey@neuling.org>
Link: https://lore.kernel.org/r/20200514111741.97993-15-ravi.bangoria@linux.ibm.com

arch/powerpc/include/asm/hw_breakpoint.h

@@ -70,6 +70,8 @@ extern int hw_breakpoint_exceptions_notify(struct notifier_block *unused,
 						unsigned long val, void *data);
 int arch_install_hw_breakpoint(struct perf_event *bp);
 void arch_uninstall_hw_breakpoint(struct perf_event *bp);
+int arch_reserve_bp_slot(struct perf_event *bp);
+void arch_release_bp_slot(struct perf_event *bp);
 void arch_unregister_hw_breakpoint(struct perf_event *bp);
 void hw_breakpoint_pmu_read(struct perf_event *bp);
 extern void flush_ptrace_hw_breakpoint(struct task_struct *tsk);

arch/powerpc/kernel/hw_breakpoint.c

@@ -124,6 +124,227 @@ static bool is_ptrace_bp(struct perf_event *bp)
 	return bp->overflow_handler == ptrace_triggered;
 }
 
+struct breakpoint {
+	struct list_head list;
+	struct perf_event *bp;
+	bool ptrace_bp;
+};
+
+static DEFINE_PER_CPU(struct breakpoint *, cpu_bps[HBP_NUM_MAX]);
+static LIST_HEAD(task_bps);
+
+static struct breakpoint *alloc_breakpoint(struct perf_event *bp)
+{
+	struct breakpoint *tmp;
+
+	tmp = kzalloc(sizeof(*tmp), GFP_KERNEL);
+	if (!tmp)
+		return ERR_PTR(-ENOMEM);
+	tmp->bp = bp;
+	tmp->ptrace_bp = is_ptrace_bp(bp);
+	return tmp;
+}
+
+static bool bp_addr_range_overlap(struct perf_event *bp1, struct perf_event *bp2)
+{
+	__u64 bp1_saddr, bp1_eaddr, bp2_saddr, bp2_eaddr;
+
+	bp1_saddr = ALIGN_DOWN(bp1->attr.bp_addr, HW_BREAKPOINT_SIZE);
+	bp1_eaddr = ALIGN(bp1->attr.bp_addr + bp1->attr.bp_len, HW_BREAKPOINT_SIZE);
+	bp2_saddr = ALIGN_DOWN(bp2->attr.bp_addr, HW_BREAKPOINT_SIZE);
+	bp2_eaddr = ALIGN(bp2->attr.bp_addr + bp2->attr.bp_len, HW_BREAKPOINT_SIZE);
+
+	return (bp1_saddr < bp2_eaddr && bp1_eaddr > bp2_saddr);
+}
+
+static bool alternate_infra_bp(struct breakpoint *b, struct perf_event *bp)
+{
+	return is_ptrace_bp(bp) ? !b->ptrace_bp : b->ptrace_bp;
+}
+
+static bool can_co_exist(struct breakpoint *b, struct perf_event *bp)
+{
+	return !(alternate_infra_bp(b, bp) && bp_addr_range_overlap(b->bp, bp));
+}
+
+static int task_bps_add(struct perf_event *bp)
+{
+	struct breakpoint *tmp;
+
+	tmp = alloc_breakpoint(bp);
+	if (IS_ERR(tmp))
+		return PTR_ERR(tmp);
+
+	list_add(&tmp->list, &task_bps);
+	return 0;
+}
+
+static void task_bps_remove(struct perf_event *bp)
+{
+	struct list_head *pos, *q;
+
+	list_for_each_safe(pos, q, &task_bps) {
+		struct breakpoint *tmp = list_entry(pos, struct breakpoint, list);
+
+		if (tmp->bp == bp) {
+			list_del(&tmp->list);
+			kfree(tmp);
+			break;
+		}
+	}
+}
+
+/*
+ * If any task has breakpoint from alternate infrastructure,
+ * return true. Otherwise return false.
+ */
+static bool all_task_bps_check(struct perf_event *bp)
+{
+	struct breakpoint *tmp;
+
+	list_for_each_entry(tmp, &task_bps, list) {
+		if (!can_co_exist(tmp, bp))
+			return true;
+	}
+	return false;
+}
+
+/*
+ * If same task has breakpoint from alternate infrastructure,
+ * return true. Otherwise return false.
+ */
+static bool same_task_bps_check(struct perf_event *bp)
+{
+	struct breakpoint *tmp;
+
+	list_for_each_entry(tmp, &task_bps, list) {
+		if (tmp->bp->hw.target == bp->hw.target &&
+		    !can_co_exist(tmp, bp))
+			return true;
+	}
+	return false;
+}
+
+static int cpu_bps_add(struct perf_event *bp)
+{
+	struct breakpoint **cpu_bp;
+	struct breakpoint *tmp;
+	int i = 0;
+
+	tmp = alloc_breakpoint(bp);
+	if (IS_ERR(tmp))
+		return PTR_ERR(tmp);
+
+	cpu_bp = per_cpu_ptr(cpu_bps, bp->cpu);
+	for (i = 0; i < nr_wp_slots(); i++) {
+		if (!cpu_bp[i]) {
+			cpu_bp[i] = tmp;
+			break;
+		}
+	}
+	return 0;
+}
+
+static void cpu_bps_remove(struct perf_event *bp)
+{
+	struct breakpoint **cpu_bp;
+	int i = 0;
+
+	cpu_bp = per_cpu_ptr(cpu_bps, bp->cpu);
+	for (i = 0; i < nr_wp_slots(); i++) {
+		if (!cpu_bp[i])
+			continue;
+
+		if (cpu_bp[i]->bp == bp) {
+			kfree(cpu_bp[i]);
+			cpu_bp[i] = NULL;
+			break;
+		}
+	}
+}
+
+static bool cpu_bps_check(int cpu, struct perf_event *bp)
+{
+	struct breakpoint **cpu_bp;
+	int i;
+
+	cpu_bp = per_cpu_ptr(cpu_bps, cpu);
+	for (i = 0; i < nr_wp_slots(); i++) {
+		if (cpu_bp[i] && !can_co_exist(cpu_bp[i], bp))
+			return true;
+	}
+	return false;
+}
+
+static bool all_cpu_bps_check(struct perf_event *bp)
+{
+	int cpu;
+
+	for_each_online_cpu(cpu) {
+		if (cpu_bps_check(cpu, bp))
+			return true;
+	}
+	return false;
+}
+
+/*
+ * We don't use any locks to serialize accesses to cpu_bps or task_bps
+ * because are already inside nr_bp_mutex.
+ */
+int arch_reserve_bp_slot(struct perf_event *bp)
+{
+	int ret;
+
+	/* ptrace breakpoint */
+	if (is_ptrace_bp(bp)) {
+		if (all_cpu_bps_check(bp))
+			return -ENOSPC;
+
+		if (same_task_bps_check(bp))
+			return -ENOSPC;
+
+		return task_bps_add(bp);
+	}
+
+	/* perf breakpoint */
+	if (is_kernel_addr(bp->attr.bp_addr))
+		return 0;
+
+	if (bp->hw.target && bp->cpu == -1) {
+		if (same_task_bps_check(bp))
+			return -ENOSPC;
+
+		return task_bps_add(bp);
+	} else if (!bp->hw.target && bp->cpu != -1) {
+		if (all_task_bps_check(bp))
+			return -ENOSPC;
+
+		return cpu_bps_add(bp);
+	}
+
+	if (same_task_bps_check(bp))
+		return -ENOSPC;
+
+	ret = cpu_bps_add(bp);
+	if (ret)
+		return ret;
+	ret = task_bps_add(bp);
+	if (ret)
+		cpu_bps_remove(bp);
+
+	return ret;
+}
+
+void arch_release_bp_slot(struct perf_event *bp)
+{
+	if (!is_kernel_addr(bp->attr.bp_addr)) {
+		if (bp->hw.target)
+			task_bps_remove(bp);
+		if (bp->cpu != -1)
+			cpu_bps_remove(bp);
+	}
+}
+
 /*
  * Perform cleanup of arch-specific counters during unregistration
  * of the perf-event

kernel/events/hw_breakpoint.c

@@ -213,6 +213,15 @@ toggle_bp_slot(struct perf_event *bp, bool enable, enum bp_type_idx type,
 		list_del(&bp->hw.bp_list);
 }
 
+__weak int arch_reserve_bp_slot(struct perf_event *bp)
+{
+	return 0;
+}
+
+__weak void arch_release_bp_slot(struct perf_event *bp)
+{
+}
+
 /*
  * Function to perform processor-specific cleanup during unregistration
  */
@@ -270,6 +279,7 @@ static int __reserve_bp_slot(struct perf_event *bp, u64 bp_type)
 	struct bp_busy_slots slots = {0};
 	enum bp_type_idx type;
 	int weight;
+	int ret;
 
 	/* We couldn't initialize breakpoint constraints on boot */
 	if (!constraints_initialized)
@@ -294,6 +304,10 @@ static int __reserve_bp_slot(struct perf_event *bp, u64 bp_type)
 	if (slots.pinned + (!!slots.flexible) > nr_slots[type])
 		return -ENOSPC;
 
+	ret = arch_reserve_bp_slot(bp);
+	if (ret)
+		return ret;
+
 	toggle_bp_slot(bp, true, type, weight);
 
 	return 0;
@@ -317,6 +331,8 @@ static void __release_bp_slot(struct perf_event *bp, u64 bp_type)
 	enum bp_type_idx type;
 	int weight;
 
+	arch_release_bp_slot(bp);
+
 	type = find_slot_idx(bp_type);
 	weight = hw_breakpoint_weight(bp);
 	toggle_bp_slot(bp, false, type, weight);