bpf: Make BPF and PREEMPT_RT co-exist

@@ -8143,26 +8143,43 @@ static bool is_tracing_prog_type(enum bp
 	}
 }
 
+static bool is_preallocated_map(struct bpf_map *map)
+{
+	if (!check_map_prealloc(map))
+		return false;
+	if (map->inner_map_meta && !check_map_prealloc(map->inner_map_meta))
+		return false;
+	return true;
+}
+
 static int check_map_prog_compatibility(struct bpf_verifier_env *env,
 					struct bpf_map *map,
 					struct bpf_prog *prog)
 
 {
-	/* Make sure that BPF_PROG_TYPE_PERF_EVENT programs only use
-	 * preallocated hash maps, since doing memory allocation
-	 * in overflow_handler can crash depending on where nmi got
-	 * triggered.
+	/*
+	 * Validate that trace type programs use preallocated hash maps.
+	 *
+	 * For programs attached to PERF events this is mandatory as the
+	 * perf NMI can hit any arbitrary code sequence.
+	 *
+	 * All other trace types using preallocated hash maps are unsafe as
+	 * well because tracepoint or kprobes can be inside locked regions
+	 * of the memory allocator or at a place where a recursion into the
+	 * memory allocator would see inconsistent state.
+	 *
+	 * For now running such programs is allowed for backwards
+	 * compatibility reasons, but warnings are emitted so developers are
+	 * made aware of the unsafety and can fix their programs before this
+	 * is enforced.
 	 */
-	if (prog->type == BPF_PROG_TYPE_PERF_EVENT) {
-		if (!check_map_prealloc(map)) {
+	if (is_tracing_prog_type(prog->type) && !is_preallocated_map(map)) {
+		if (prog->type == BPF_PROG_TYPE_PERF_EVENT) {
 			verbose(env, "perf_event programs can only use preallocated hash map\n");
 			return -EINVAL;
 		}
-		if (map->inner_map_meta &&
-		    !check_map_prealloc(map->inner_map_meta)) {
-			verbose(env, "perf_event programs can only use preallocated inner hash map\n");
-			return -EINVAL;
-		}
+		WARN_ONCE(1, "trace type BPF program uses run-time allocation\n");
+		verbose(env, "trace type programs with run-time allocated hash maps are unsafe. Switch to preallocated hash maps.\n");
 	}
 
 	if ((is_tracing_prog_type(prog->type) ||

@@ -8168,16 +8168,21 @@ static int check_map_prog_compatibility(
 	 * of the memory allocator or at a place where a recursion into the
 	 * memory allocator would see inconsistent state.
 	 *
-	 * For now running such programs is allowed for backwards
-	 * compatibility reasons, but warnings are emitted so developers are
-	 * made aware of the unsafety and can fix their programs before this
-	 * is enforced.
+	 * On RT enabled kernels run-time allocation of all trace type
+	 * programs is strictly prohibited due to lock type constraints. On
+	 * !RT kernels it is allowed for backwards compatibility reasons for
+	 * now, but warnings are emitted so developers are made aware of
+	 * the unsafety and can fix their programs before this is enforced.
 	 */
 	if (is_tracing_prog_type(prog->type) && !is_preallocated_map(map)) {
 		if (prog->type == BPF_PROG_TYPE_PERF_EVENT) {
 			verbose(env, "perf_event programs can only use preallocated hash map\n");
 			return -EINVAL;
 		}
+		if (IS_ENABLED(CONFIG_PREEMPT_RT)) {
+			verbose(env, "trace type programs can only use preallocated hash map\n");
+			return -EINVAL;
+		}
 		WARN_ONCE(1, "trace type BPF program uses run-time allocation\n");
 		verbose(env, "trace type programs with run-time allocated hash maps are unsafe. Switch to preallocated hash maps.\n");
 	}

@@ -27,6 +27,26 @@
 	.map_delete_batch =			\
 	generic_map_delete_batch
 
+/*
+ * The bucket lock has two protection scopes:
+ *
+ * 1) Serializing concurrent operations from BPF programs on differrent
+ *    CPUs
+ *
+ * 2) Serializing concurrent operations from BPF programs and sys_bpf()
+ *
+ * BPF programs can execute in any context including perf, kprobes and
+ * tracing. As there are almost no limits where perf, kprobes and tracing
+ * can be invoked from the lock operations need to be protected against
+ * deadlocks. Deadlocks can be caused by recursion and by an invocation in
+ * the lock held section when functions which acquire this lock are invoked
+ * from sys_bpf(). BPF recursion is prevented by incrementing the per CPU
+ * variable bpf_prog_active, which prevents BPF programs attached to perf
+ * events, kprobes and tracing to be invoked before the prior invocation
+ * from one of these contexts completed. sys_bpf() uses the same mechanism
+ * by pinning the task to the current CPU and incrementing the recursion
+ * protection accross the map operation.
+ */
 struct bucket {
 	struct hlist_nulls_head head;
 	raw_spinlock_t lock;

@@ -884,7 +904,6 @@ static int htab_map_update_elem(struct b
 		 */
 	}
 
-	/* bpf_map_update_elem() can be called in_irq() */
 	raw_spin_lock_irqsave(&b->lock, flags);
 
 	l_old = lookup_elem_raw(head, hash, key, key_size);

@@ -964,7 +983,6 @@ static int htab_lru_map_update_elem(stru
 		return -ENOMEM;
 	memcpy(l_new->key + round_up(map->key_size, 8), value, map->value_size);
 
-	/* bpf_map_update_elem() can be called in_irq() */
 	raw_spin_lock_irqsave(&b->lock, flags);
 
 	l_old = lookup_elem_raw(head, hash, key, key_size);

@@ -1019,7 +1037,6 @@ static int __htab_percpu_map_update_elem
 	b = __select_bucket(htab, hash);
 	head = &b->head;
 
-	/* bpf_map_update_elem() can be called in_irq() */
 	raw_spin_lock_irqsave(&b->lock, flags);
 
 	l_old = lookup_elem_raw(head, hash, key, key_size);

@@ -1083,7 +1100,6 @@ static int __htab_lru_percpu_map_update_
 			return -ENOMEM;
 	}
 
-	/* bpf_map_update_elem() can be called in_irq() */
 	raw_spin_lock_irqsave(&b->lock, flags);
 
 	l_old = lookup_elem_raw(head, hash, key, key_size);

@@ -1516,10 +1516,9 @@ void bpf_put_raw_tracepoint(struct bpf_r
 static __always_inline
 void __bpf_trace_run(struct bpf_prog *prog, u64 *args)
 {
+	cant_sleep();
 	rcu_read_lock();
-	preempt_disable();
 	(void) BPF_PROG_RUN(prog, args);
-	preempt_enable();
 	rcu_read_unlock();
 }
 

@@ -119,7 +119,6 @@ unsigned int trace_call_bpf(struct trace
 
 	return ret;
 }
-EXPORT_SYMBOL_GPL(trace_call_bpf);
 
 #ifdef CONFIG_BPF_KPROBE_OVERRIDE
 BPF_CALL_2(bpf_override_return, struct pt_regs *, regs, unsigned long, rc)

@@ -83,7 +83,7 @@ unsigned int trace_call_bpf(struct trace
 	if (in_nmi()) /* not supported yet */
 		return 1;
 
-	preempt_disable();
+	cant_sleep();
 
 	if (unlikely(__this_cpu_inc_return(bpf_prog_active) != 1)) {
 		/*

@@ -115,7 +115,6 @@ unsigned int trace_call_bpf(struct trace
 
  out:
 	__this_cpu_dec(bpf_prog_active);
-	preempt_enable();
 
 	return ret;
 }

@@ -9206,7 +9206,6 @@ static void bpf_overflow_handler(struct
 	int ret = 0;
 
 	ctx.regs = perf_arch_bpf_user_pt_regs(regs);
-	preempt_disable();
 	if (unlikely(__this_cpu_inc_return(bpf_prog_active) != 1))
 		goto out;
 	rcu_read_lock();

@@ -9214,7 +9213,6 @@ static void bpf_overflow_handler(struct
 	rcu_read_unlock();
 out:
 	__this_cpu_dec(bpf_prog_active);
-	preempt_enable();
 	if (!ret)
 		return;
 

@@ -706,15 +706,7 @@ static void htab_elem_free_rcu(struct rc
 	struct htab_elem *l = container_of(head, struct htab_elem, rcu);
 	struct bpf_htab *htab = l->htab;
 
-	/* must increment bpf_prog_active to avoid kprobe+bpf triggering while
-	 * we're calling kfree, otherwise deadlock is possible if kprobes
-	 * are placed somewhere inside of slub
-	 */
-	preempt_disable();
-	__this_cpu_inc(bpf_prog_active);
 	htab_elem_free(htab, l);
-	__this_cpu_dec(bpf_prog_active);
-	preempt_enable();
 }
 
 static void free_htab_elem(struct bpf_htab *htab, struct htab_elem *l)

@@ -25,12 +25,18 @@ void pcpu_freelist_destroy(struct pcpu_f
 	free_percpu(s->freelist);
 }
 
+static inline void pcpu_freelist_push_node(struct pcpu_freelist_head *head,
+					   struct pcpu_freelist_node *node)
+{
+	node->next = head->first;
+	head->first = node;
+}
+
 static inline void ___pcpu_freelist_push(struct pcpu_freelist_head *head,
 					 struct pcpu_freelist_node *node)
 {
 	raw_spin_lock(&head->lock);
-	node->next = head->first;
-	head->first = node;
+	pcpu_freelist_push_node(head, node);
 	raw_spin_unlock(&head->lock);
 }
 

@@ -56,21 +62,16 @@ void pcpu_freelist_populate(struct pcpu_
 			    u32 nr_elems)
 {
 	struct pcpu_freelist_head *head;
-	unsigned long flags;
 	int i, cpu, pcpu_entries;
 
 	pcpu_entries = nr_elems / num_possible_cpus() + 1;
 	i = 0;
 
-	/* disable irq to workaround lockdep false positive
-	 * in bpf usage pcpu_freelist_populate() will never race
-	 * with pcpu_freelist_push()
-	 */
-	local_irq_save(flags);
 	for_each_possible_cpu(cpu) {
 again:
 		head = per_cpu_ptr(s->freelist, cpu);
-		___pcpu_freelist_push(head, buf);
+		/* No locking required as this is not visible yet. */
+		pcpu_freelist_push_node(head, buf);
 		i++;
 		buf += elem_size;
 		if (i == nr_elems)

@@ -78,7 +79,6 @@ void pcpu_freelist_populate(struct pcpu_
 		if (i % pcpu_entries)
 			goto again;
 	}
-	local_irq_restore(flags);
 }
 
 struct pcpu_freelist_node *__pcpu_freelist_pop(struct pcpu_freelist *s)

@@ -576,8 +576,30 @@ DECLARE_STATIC_KEY_FALSE(bpf_stats_enabl
 	}								\
 	ret; })
 
-#define BPF_PROG_RUN(prog, ctx) __BPF_PROG_RUN(prog, ctx,		\
-					       bpf_dispatcher_nopfunc)
+#define BPF_PROG_RUN(prog, ctx)						\
+	__BPF_PROG_RUN(prog, ctx, bpf_dispatcher_nopfunc)
+
+/*
+ * Use in preemptible and therefore migratable context to make sure that
+ * the execution of the BPF program runs on one CPU.
+ *
+ * This uses migrate_disable/enable() explicitly to document that the
+ * invocation of a BPF program does not require reentrancy protection
+ * against a BPF program which is invoked from a preempting task.
+ *
+ * For non RT enabled kernels migrate_disable/enable() maps to
+ * preempt_disable/enable(), i.e. it disables also preemption.
+ */
+static inline u32 bpf_prog_run_pin_on_cpu(const struct bpf_prog *prog,
+					  void *ctx)
+{
+	u32 ret;
+
+	migrate_disable();
+	ret = __BPF_PROG_RUN(prog, ctx, bpf_dispatcher_nopfunc);
+	migrate_enable();
+	return ret;
+}
 
 #define BPF_SKB_CB_LEN QDISC_CB_PRIV_LEN
 

@@ -576,8 +576,30 @@ DECLARE_STATIC_KEY_FALSE(bpf_stats_enabl
 	}								\
 	ret; })
 
-#define BPF_PROG_RUN(prog, ctx) __BPF_PROG_RUN(prog, ctx,		\
-					       bpf_dispatcher_nopfunc)
+#define BPF_PROG_RUN(prog, ctx)						\
+	__BPF_PROG_RUN(prog, ctx, bpf_dispatcher_nopfunc)
+
+/*
+ * Use in preemptible and therefore migratable context to make sure that
+ * the execution of the BPF program runs on one CPU.
+ *
+ * This uses migrate_disable/enable() explicitly to document that the
+ * invocation of a BPF program does not require reentrancy protection
+ * against a BPF program which is invoked from a preempting task.
+ *
+ * For non RT enabled kernels migrate_disable/enable() maps to
+ * preempt_disable/enable(), i.e. it disables also preemption.
+ */
+static inline u32 bpf_prog_run_pin_on_cpu(const struct bpf_prog *prog,
+					  const void *ctx)
+{
+	u32 ret;
+
+	migrate_disable();
+	ret = __BPF_PROG_RUN(prog, ctx, bpf_dispatcher_nopfunc);
+	migrate_enable();
+	return ret;
+}
 
 #define BPF_SKB_CB_LEN QDISC_CB_PRIV_LEN
 

@@ -561,7 +561,7 @@ DECLARE_STATIC_KEY_FALSE(bpf_stats_enabl
 
 #define __BPF_PROG_RUN(prog, ctx, dfunc)	({			\
 	u32 ret;							\
-	cant_sleep();							\
+	cant_migrate();							\
 	if (static_branch_unlikely(&bpf_stats_enabled_key)) {		\
 		struct bpf_prog_stats *stats;				\
 		u64 start = sched_clock();				\

@@ -717,9 +717,7 @@ static inline u32 bpf_prog_run_clear_cb(
 	if (unlikely(prog->cb_access))
 		memset(cb_data, 0, BPF_SKB_CB_LEN);
 
-	preempt_disable();
-	res = BPF_PROG_RUN(prog, skb);
-	preempt_enable();
+	res = bpf_prog_run_pin_on_cpu(prog, skb);
 	return res;
 }
 

@@ -268,16 +268,14 @@ static u32 seccomp_run_filters(const str
 	 * All filters in the list are evaluated and the lowest BPF return
 	 * value always takes priority (ignoring the DATA).
 	 */
-	preempt_disable();
 	for (; f; f = f->prev) {
-		u32 cur_ret = BPF_PROG_RUN(f->prog, sd);
+		u32 cur_ret = bpf_prog_run_pin_on_cpu(f->prog, sd);
 
 		if (ACTION_ONLY(cur_ret) < ACTION_ONLY(ret)) {
 			ret = cur_ret;
 			*match = f;
 		}
 	}
-	preempt_enable();
 	return ret;
 }
 #endif /* CONFIG_SECCOMP_FILTER */

@@ -920,9 +920,7 @@ bool bpf_flow_dissect(struct bpf_prog *p
 		     (int)FLOW_DISSECTOR_F_STOP_AT_ENCAP);
 	flow_keys->flags = flags;
 
-	preempt_disable();
-	result = BPF_PROG_RUN(prog, ctx);
-	preempt_enable();
+	result = bpf_prog_run_pin_on_cpu(prog, ctx);
 
 	flow_keys->nhoff = clamp_t(u16, flow_keys->nhoff, nhoff, hlen);
 	flow_keys->thoff = clamp_t(u16, flow_keys->thoff,

@@ -628,7 +628,6 @@ int sk_psock_msg_verdict(struct sock *sk
 	struct bpf_prog *prog;
 	int ret;
 
-	preempt_disable();
 	rcu_read_lock();
 	prog = READ_ONCE(psock->progs.msg_parser);
 	if (unlikely(!prog)) {

@@ -638,7 +637,7 @@ int sk_psock_msg_verdict(struct sock *sk
 
 	sk_msg_compute_data_pointers(msg);
 	msg->sk = sk;
-	ret = BPF_PROG_RUN(prog, msg);
+	ret = bpf_prog_run_pin_on_cpu(prog, msg);
 	ret = sk_psock_map_verd(ret, msg->sk_redir);
 	psock->apply_bytes = msg->apply_bytes;
 	if (ret == __SK_REDIRECT) {

@@ -653,7 +652,6 @@ int sk_psock_msg_verdict(struct sock *sk
 	}
 out:
 	rcu_read_unlock();
-	preempt_enable();
 	return ret;
 }
 EXPORT_SYMBOL_GPL(sk_psock_msg_verdict);

@@ -665,9 +663,7 @@ static int sk_psock_bpf_run(struct sk_ps
 
 	skb->sk = psock->sk;
 	bpf_compute_data_end_sk_skb(skb);
-	preempt_disable();
-	ret = BPF_PROG_RUN(prog, skb);
-	preempt_enable();
+	ret = bpf_prog_run_pin_on_cpu(prog, skb);
 	/* strparser clones the skb before handing it to a upper layer,
 	 * meaning skb_orphan has been called. We NULL sk on the way out
 	 * to ensure we don't trigger a BUG_ON() in skb/sk operations

@@ -380,9 +380,7 @@ static int kcm_parse_func_strparser(stru
 	struct bpf_prog *prog = psock->bpf_prog;
 	int res;
 
-	preempt_disable();
-	res = BPF_PROG_RUN(prog, skb);
-	preempt_enable();
+	res = bpf_prog_run_pin_on_cpu(prog, skb);
 	return res;
 }
 

@@ -6660,14 +6660,14 @@ static int __run_one(const struct bpf_pr
 	u64 start, finish;
 	int ret = 0, i;
 
-	preempt_disable();
+	migrate_disable();
 	start = ktime_get_ns();
 
 	for (i = 0; i < runs; i++)
 		ret = BPF_PROG_RUN(fp, data);
 
 	finish = ktime_get_ns();
-	preempt_enable();
+	migrate_enable();
 
 	*duration = finish - start;
 	do_div(*duration, runs);

@@ -37,7 +37,7 @@ static int bpf_test_run(struct bpf_prog
 		repeat = 1;
 
 	rcu_read_lock();
-	preempt_disable();
+	migrate_disable();
 	time_start = ktime_get_ns();
 	for (i = 0; i < repeat; i++) {
 		bpf_cgroup_storage_set(storage);

@@ -54,18 +54,18 @@ static int bpf_test_run(struct bpf_prog
 
 		if (need_resched()) {
 			time_spent += ktime_get_ns() - time_start;
-			preempt_enable();
+			migrate_enable();
 			rcu_read_unlock();
 
 			cond_resched();
 
 			rcu_read_lock();
-			preempt_disable();
+			migrate_disable();
 			time_start = ktime_get_ns();
 		}
 	}
 	time_spent += ktime_get_ns() - time_start;
-	preempt_enable();
+	migrate_enable();
 	rcu_read_unlock();
 
 	do_div(time_spent, repeat);

@@ -367,8 +367,9 @@ void bpf_trampoline_put(struct bpf_tramp
 	mutex_unlock(&trampoline_mutex);
 }
 
-/* The logic is similar to BPF_PROG_RUN, but with explicit rcu and preempt that
- * are needed for trampoline. The macro is split into
+/* The logic is similar to BPF_PROG_RUN, but with an explicit
+ * rcu_read_lock() and migrate_disable() which are required
+ * for the trampoline. The macro is split into
  * call _bpf_prog_enter
  * call prog->bpf_func
  * call __bpf_prog_exit

@@ -378,7 +379,7 @@ u64 notrace __bpf_prog_enter(void)
 	u64 start = 0;
 
 	rcu_read_lock();
-	preempt_disable();
+	migrate_disable();
 	if (static_branch_unlikely(&bpf_stats_enabled_key))
 		start = sched_clock();
 	return start;

@@ -401,7 +402,7 @@ void notrace __bpf_prog_exit(struct bpf_
 		stats->nsecs += sched_clock() - start;
 		u64_stats_update_end(&stats->syncp);
 	}
-	preempt_enable();
+	migrate_enable();
 	rcu_read_unlock();
 }
 

@@ -885,7 +885,7 @@ int bpf_prog_array_copy(struct bpf_prog_
 		struct bpf_prog *_prog;			\
 		struct bpf_prog_array *_array;		\
 		u32 _ret = 1;				\
-		preempt_disable();			\
+		migrate_disable();			\
 		rcu_read_lock();			\
 		_array = rcu_dereference(array);	\
 		if (unlikely(check_non_null && !_array))\

@@ -898,7 +898,7 @@ int bpf_prog_array_copy(struct bpf_prog_
 		}					\
 _out:							\
 		rcu_read_unlock();			\
-		preempt_enable();			\
+		migrate_enable();			\
 		_ret;					\
 	 })
 

@@ -932,7 +932,7 @@ int bpf_prog_array_copy(struct bpf_prog_
 		u32 ret;				\
 		u32 _ret = 1;				\
 		u32 _cn = 0;				\
-		preempt_disable();			\
+		migrate_disable();			\
 		rcu_read_lock();			\
 		_array = rcu_dereference(array);	\
 		_item = &_array->items[0];		\

@@ -944,7 +944,7 @@ int bpf_prog_array_copy(struct bpf_prog_
 			_item++;			\
 		}					\
 		rcu_read_unlock();			\
-		preempt_enable();			\
+		migrate_enable();			\
 		if (_ret)				\
 			_ret = (_cn ? NET_XMIT_CN : NET_XMIT_SUCCESS);	\
 		else					\

@@ -677,6 +677,7 @@ static inline u8 *bpf_skb_cb(struct sk_b
 	return qdisc_skb_cb(skb)->data;
 }
 
+/* Must be invoked with migration disabled */
 static inline u32 __bpf_prog_run_save_cb(const struct bpf_prog *prog,
 					 struct sk_buff *skb)
 {

@@ -702,9 +703,9 @@ static inline u32 bpf_prog_run_save_cb(c
 {
 	u32 res;
 
-	preempt_disable();
+	migrate_disable();
 	res = __bpf_prog_run_save_cb(prog, skb);
-	preempt_enable();
+	migrate_enable();
 	return res;
 }
 

@@ -961,6 +961,36 @@ int bpf_prog_array_copy(struct bpf_prog_
 #ifdef CONFIG_BPF_SYSCALL
 DECLARE_PER_CPU(int, bpf_prog_active);
 
+/*
+ * Block execution of BPF programs attached to instrumentation (perf,
+ * kprobes, tracepoints) to prevent deadlocks on map operations as any of
+ * these events can happen inside a region which holds a map bucket lock
+ * and can deadlock on it.
+ *
+ * Use the preemption safe inc/dec variants on RT because migrate disable
+ * is preemptible on RT and preemption in the middle of the RMW operation
+ * might lead to inconsistent state. Use the raw variants for non RT
+ * kernels as migrate_disable() maps to preempt_disable() so the slightly
+ * more expensive save operation can be avoided.
+ */
+static inline void bpf_disable_instrumentation(void)
+{
+	migrate_disable();
+	if (IS_ENABLED(CONFIG_PREEMPT_RT))
+		this_cpu_inc(bpf_prog_active);
+	else
+		__this_cpu_inc(bpf_prog_active);
+}
+
+static inline void bpf_enable_instrumentation(void)
+{
+	if (IS_ENABLED(CONFIG_PREEMPT_RT))
+		this_cpu_dec(bpf_prog_active);
+	else
+		__this_cpu_dec(bpf_prog_active);
+	migrate_enable();
+}
+
 extern const struct file_operations bpf_map_fops;
 extern const struct file_operations bpf_prog_fops;
 

@@ -1333,8 +1333,7 @@ static int
 	}
 
 again:
-	preempt_disable();
-	this_cpu_inc(bpf_prog_active);
+	bpf_disable_instrumentation();
 	rcu_read_lock();
 again_nocopy:
 	dst_key = keys;

@@ -1362,8 +1361,7 @@ static int
 		 */
 		raw_spin_unlock_irqrestore(&b->lock, flags);
 		rcu_read_unlock();
-		this_cpu_dec(bpf_prog_active);
-		preempt_enable();
+		bpf_enable_instrumentation();
 		goto after_loop;
 	}
 

@@ -1374,8 +1372,7 @@ static int
 		 */
 		raw_spin_unlock_irqrestore(&b->lock, flags);
 		rcu_read_unlock();
-		this_cpu_dec(bpf_prog_active);
-		preempt_enable();
+		bpf_enable_instrumentation();
 		kvfree(keys);
 		kvfree(values);
 		goto alloc;

@@ -1445,8 +1442,7 @@ static int
 	}
 
 	rcu_read_unlock();
-	this_cpu_dec(bpf_prog_active);
-	preempt_enable();
+	bpf_enable_instrumentation();
 	if (bucket_cnt && (copy_to_user(ukeys + total * key_size, keys,
 	    key_size * bucket_cnt) ||
 	    copy_to_user(uvalues + total * value_size, values,

@@ -171,11 +171,7 @@ static int bpf_map_update_value(struct b
 						    flags);
 	}
 
-	/* must increment bpf_prog_active to avoid kprobe+bpf triggering from
-	 * inside bpf map update or delete otherwise deadlocks are possible
-	 */
-	preempt_disable();
-	__this_cpu_inc(bpf_prog_active);
+	bpf_disable_instrumentation();
 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
 		err = bpf_percpu_hash_update(map, key, value, flags);

@@ -206,8 +202,7 @@ static int bpf_map_update_value(struct b
 		err = map->ops->map_update_elem(map, key, value, flags);
 		rcu_read_unlock();
 	}
-	__this_cpu_dec(bpf_prog_active);
-	preempt_enable();
+	bpf_enable_instrumentation();
 	maybe_wait_bpf_programs(map);
 
 	return err;

@@ -222,8 +217,7 @@ static int bpf_map_copy_value(struct bpf
 	if (bpf_map_is_dev_bound(map))
 		return bpf_map_offload_lookup_elem(map, key, value);
 
-	preempt_disable();
-	this_cpu_inc(bpf_prog_active);
+	bpf_disable_instrumentation();
 	if (map->map_type == BPF_MAP_TYPE_PERCPU_HASH ||
 	    map->map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH) {
 		err = bpf_percpu_hash_copy(map, key, value);

@@ -268,8 +262,7 @@ static int bpf_map_copy_value(struct bpf
 		rcu_read_unlock();
 	}
 
-	this_cpu_dec(bpf_prog_active);
-	preempt_enable();
+	bpf_enable_instrumentation();
 	maybe_wait_bpf_programs(map);
 
 	return err;

@@ -1136,13 +1129,11 @@ static int map_delete_elem(union bpf_att
 		goto out;
 	}
 
-	preempt_disable();
-	__this_cpu_inc(bpf_prog_active);
+	bpf_disable_instrumentation();
 	rcu_read_lock();
 	err = map->ops->map_delete_elem(map, key);
 	rcu_read_unlock();
-	__this_cpu_dec(bpf_prog_active);
-	preempt_enable();
+	bpf_enable_instrumentation();
 	maybe_wait_bpf_programs(map);
 out:
 	kfree(key);

@@ -1254,13 +1245,11 @@ int generic_map_delete_batch(struct bpf_
 			break;
 		}
 
-		preempt_disable();
-		__this_cpu_inc(bpf_prog_active);
+		bpf_disable_instrumentation();
 		rcu_read_lock();
 		err = map->ops->map_delete_elem(map, key);
 		rcu_read_unlock();
-		__this_cpu_dec(bpf_prog_active);
-		preempt_enable();
+		bpf_enable_instrumentation();
 		maybe_wait_bpf_programs(map);
 		if (err)
 			break;

@@ -88,6 +88,32 @@ struct htab_elem {
 	char key[0] __aligned(8);
 };
 
+static void htab_init_buckets(struct bpf_htab *htab)
+{
+	unsigned i;
+
+	for (i = 0; i < htab->n_buckets; i++) {
+		INIT_HLIST_NULLS_HEAD(&htab->buckets[i].head, i);
+		raw_spin_lock_init(&htab->buckets[i].lock);
+	}
+}
+
+static inline unsigned long htab_lock_bucket(const struct bpf_htab *htab,
+					     struct bucket *b)
+{
+	unsigned long flags;
+
+	raw_spin_lock_irqsave(&b->lock, flags);
+	return flags;
+}
+
+static inline void htab_unlock_bucket(const struct bpf_htab *htab,
+				      struct bucket *b,
+				      unsigned long flags)
+{
+	raw_spin_unlock_irqrestore(&b->lock, flags);
+}
+
 static bool htab_lru_map_delete_node(void *arg, struct bpf_lru_node *node);
 
 static bool htab_is_lru(const struct bpf_htab *htab)

@@ -348,8 +374,8 @@ static struct bpf_map *htab_map_alloc(un
 	bool percpu_lru = (attr->map_flags & BPF_F_NO_COMMON_LRU);
 	bool prealloc = !(attr->map_flags & BPF_F_NO_PREALLOC);
 	struct bpf_htab *htab;
-	int err, i;
 	u64 cost;
+	int err;
 
 	htab = kzalloc(sizeof(*htab), GFP_USER);
 	if (!htab)

@@ -411,10 +437,7 @@ static struct bpf_map *htab_map_alloc(un
 	else
 		htab->hashrnd = get_random_int();
 
-	for (i = 0; i < htab->n_buckets; i++) {
-		INIT_HLIST_NULLS_HEAD(&htab->buckets[i].head, i);
-		raw_spin_lock_init(&htab->buckets[i].lock);
-	}
+	htab_init_buckets(htab);
 
 	if (prealloc) {
 		err = prealloc_init(htab);

@@ -622,7 +645,7 @@ static bool htab_lru_map_delete_node(voi
 	b = __select_bucket(htab, tgt_l->hash);
 	head = &b->head;
 
-	raw_spin_lock_irqsave(&b->lock, flags);
+	flags = htab_lock_bucket(htab, b);
 
 	hlist_nulls_for_each_entry_rcu(l, n, head, hash_node)
 		if (l == tgt_l) {

@@ -630,7 +653,7 @@ static bool htab_lru_map_delete_node(voi
 			break;
 		}
 
-	raw_spin_unlock_irqrestore(&b->lock, flags);
+	htab_unlock_bucket(htab, b, flags);
 
 	return l == tgt_l;
 }

@@ -896,7 +919,7 @@ static int htab_map_update_elem(struct b
 		 */
 	}
 
-	raw_spin_lock_irqsave(&b->lock, flags);
+	flags = htab_lock_bucket(htab, b);
 
 	l_old = lookup_elem_raw(head, hash, key, key_size);
 

@@ -937,7 +960,7 @@ static int htab_map_update_elem(struct b
 	}
 	ret = 0;
 err:
-	raw_spin_unlock_irqrestore(&b->lock, flags);
+	htab_unlock_bucket(htab, b, flags);
 	return ret;
 }
 

@@ -975,7 +998,7 @@ static int htab_lru_map_update_elem(stru
 		return -ENOMEM;
 	memcpy(l_new->key + round_up(map->key_size, 8), value, map->value_size);
 
-	raw_spin_lock_irqsave(&b->lock, flags);
+	flags = htab_lock_bucket(htab, b);
 
 	l_old = lookup_elem_raw(head, hash, key, key_size);
 

@@ -994,7 +1017,7 @@ static int htab_lru_map_update_elem(stru
 	ret = 0;
 
 err:
-	raw_spin_unlock_irqrestore(&b->lock, flags);
+	htab_unlock_bucket(htab, b, flags);
 
 	if (ret)
 		bpf_lru_push_free(&htab->lru, &l_new->lru_node);

@@ -1029,7 +1052,7 @@ static int __htab_percpu_map_update_elem
 	b = __select_bucket(htab, hash);
 	head = &b->head;
 
-	raw_spin_lock_irqsave(&b->lock, flags);
+	flags = htab_lock_bucket(htab, b);
 
 	l_old = lookup_elem_raw(head, hash, key, key_size);
 

@@ -1052,7 +1075,7 @@ static int __htab_percpu_map_update_elem
 	}
 	ret = 0;
 err:
-	raw_spin_unlock_irqrestore(&b->lock, flags);
+	htab_unlock_bucket(htab, b, flags);
 	return ret;
 }
 

@@ -1092,7 +1115,7 @@ static int __htab_lru_percpu_map_update_
 			return -ENOMEM;
 	}
 
-	raw_spin_lock_irqsave(&b->lock, flags);
+	flags = htab_lock_bucket(htab, b);
 
 	l_old = lookup_elem_raw(head, hash, key, key_size);
 

@@ -1114,7 +1137,7 @@ static int __htab_lru_percpu_map_update_
 	}
 	ret = 0;
 err:
-	raw_spin_unlock_irqrestore(&b->lock, flags);
+	htab_unlock_bucket(htab, b, flags);
 	if (l_new)
 		bpf_lru_push_free(&htab->lru, &l_new->lru_node);
 	return ret;

@@ -1152,7 +1175,7 @@ static int htab_map_delete_elem(struct b
 	b = __select_bucket(htab, hash);
 	head = &b->head;
 
-	raw_spin_lock_irqsave(&b->lock, flags);
+	flags = htab_lock_bucket(htab, b);
 
 	l = lookup_elem_raw(head, hash, key, key_size);
 

@@ -1162,7 +1185,7 @@ static int htab_map_delete_elem(struct b
 		ret = 0;
 	}
 
-	raw_spin_unlock_irqrestore(&b->lock, flags);
+	htab_unlock_bucket(htab, b, flags);
 	return ret;
 }
 

@@ -1184,7 +1207,7 @@ static int htab_lru_map_delete_elem(stru
 	b = __select_bucket(htab, hash);
 	head = &b->head;
 
-	raw_spin_lock_irqsave(&b->lock, flags);
+	flags = htab_lock_bucket(htab, b);
 
 	l = lookup_elem_raw(head, hash, key, key_size);
 

@@ -1193,7 +1216,7 @@ static int htab_lru_map_delete_elem(stru
 		ret = 0;
 	}
 
-	raw_spin_unlock_irqrestore(&b->lock, flags);
+	htab_unlock_bucket(htab, b, flags);
 	if (l)
 		bpf_lru_push_free(&htab->lru, &l->lru_node);
 	return ret;

@@ -1342,7 +1365,7 @@ static int
 	head = &b->head;
 	/* do not grab the lock unless need it (bucket_cnt > 0). */
 	if (locked)
-		raw_spin_lock_irqsave(&b->lock, flags);
+		flags = htab_lock_bucket(htab, b);
 
 	bucket_cnt = 0;
 	hlist_nulls_for_each_entry_rcu(l, n, head, hash_node)

@@ -1359,7 +1382,7 @@ static int
 		/* Note that since bucket_cnt > 0 here, it is implicit
 		 * that the locked was grabbed, so release it.
 		 */
-		raw_spin_unlock_irqrestore(&b->lock, flags);
+		htab_unlock_bucket(htab, b, flags);
 		rcu_read_unlock();
 		bpf_enable_instrumentation();
 		goto after_loop;

@@ -1370,7 +1393,7 @@ static int
 		/* Note that since bucket_cnt > 0 here, it is implicit
 		 * that the locked was grabbed, so release it.
 		 */
-		raw_spin_unlock_irqrestore(&b->lock, flags);
+		htab_unlock_bucket(htab, b, flags);
 		rcu_read_unlock();
 		bpf_enable_instrumentation();
 		kvfree(keys);

@@ -1423,7 +1446,7 @@ static int
 		dst_val += value_size;
 	}
 
-	raw_spin_unlock_irqrestore(&b->lock, flags);
+	htab_unlock_bucket(htab, b, flags);
 	locked = false;
 
 	while (node_to_free) {

@@ -46,10 +46,43 @@
  * from one of these contexts completed. sys_bpf() uses the same mechanism
  * by pinning the task to the current CPU and incrementing the recursion
  * protection accross the map operation.
+ *
+ * This has subtle implications on PREEMPT_RT. PREEMPT_RT forbids certain
+ * operations like memory allocations (even with GFP_ATOMIC) from atomic
+ * contexts. This is required because even with GFP_ATOMIC the memory
+ * allocator calls into code pathes which acquire locks with long held lock
+ * sections. To ensure the deterministic behaviour these locks are regular
+ * spinlocks, which are converted to 'sleepable' spinlocks on RT. The only
+ * true atomic contexts on an RT kernel are the low level hardware
+ * handling, scheduling, low level interrupt handling, NMIs etc. None of
+ * these contexts should ever do memory allocations.
+ *
+ * As regular device interrupt handlers and soft interrupts are forced into
+ * thread context, the existing code which does
+ *   spin_lock*(); alloc(GPF_ATOMIC); spin_unlock*();
+ * just works.
+ *
+ * In theory the BPF locks could be converted to regular spinlocks as well,
+ * but the bucket locks and percpu_freelist locks can be taken from
+ * arbitrary contexts (perf, kprobes, tracepoints) which are required to be
+ * atomic contexts even on RT. These mechanisms require preallocated maps,
+ * so there is no need to invoke memory allocations within the lock held
+ * sections.
+ *
+ * BPF maps which need dynamic allocation are only used from (forced)
+ * thread context on RT and can therefore use regular spinlocks which in
+ * turn allows to invoke memory allocations from the lock held section.
+ *
+ * On a non RT kernel this distinction is neither possible nor required.
+ * spinlock maps to raw_spinlock and the extra code is optimized out by the
+ * compiler.
  */
 struct bucket {
 	struct hlist_nulls_head head;
-	raw_spinlock_t lock;
+	union {
+		raw_spinlock_t raw_lock;
+		spinlock_t     lock;
+	};
 };
 
 struct bpf_htab {

@@ -88,13 +121,26 @@ struct htab_elem {
 	char key[0] __aligned(8);
 };
 
+static inline bool htab_is_prealloc(const struct bpf_htab *htab)
+{
+	return !(htab->map.map_flags & BPF_F_NO_PREALLOC);
+}
+
+static inline bool htab_use_raw_lock(const struct bpf_htab *htab)
+{
+	return (!IS_ENABLED(CONFIG_PREEMPT_RT) || htab_is_prealloc(htab));
+}
+
 static void htab_init_buckets(struct bpf_htab *htab)
 {
 	unsigned i;
 
 	for (i = 0; i < htab->n_buckets; i++) {
 		INIT_HLIST_NULLS_HEAD(&htab->buckets[i].head, i);
-		raw_spin_lock_init(&htab->buckets[i].lock);
+		if (htab_use_raw_lock(htab))
+			raw_spin_lock_init(&htab->buckets[i].raw_lock);
+		else
+			spin_lock_init(&htab->buckets[i].lock);
 	}
 }
 

@@ -103,7 +149,10 @@ static inline unsigned long htab_lock_bu
 {
 	unsigned long flags;
 
-	raw_spin_lock_irqsave(&b->lock, flags);
+	if (htab_use_raw_lock(htab))
+		raw_spin_lock_irqsave(&b->raw_lock, flags);
+	else
+		spin_lock_irqsave(&b->lock, flags);
 	return flags;
 }
 

@@ -111,7 +160,10 @@ static inline void htab_unlock_bucket(co
 				      struct bucket *b,
 				      unsigned long flags)
 {
-	raw_spin_unlock_irqrestore(&b->lock, flags);
+	if (htab_use_raw_lock(htab))
+		raw_spin_unlock_irqrestore(&b->raw_lock, flags);
+	else
+		spin_unlock_irqrestore(&b->lock, flags);
 }
 
 static bool htab_lru_map_delete_node(void *arg, struct bpf_lru_node *node);

@@ -128,11 +180,6 @@ static bool htab_is_percpu(const struct
 		htab->map.map_type == BPF_MAP_TYPE_LRU_PERCPU_HASH;
 }
 
-static bool htab_is_prealloc(const struct bpf_htab *htab)
-{
-	return !(htab->map.map_flags & BPF_F_NO_PREALLOC);
-}
-
 static inline void htab_elem_set_ptr(struct htab_elem *l, u32 key_size,
 				     void __percpu *pptr)
 {

@@ -34,7 +34,7 @@ struct lpm_trie {
 	size_t				n_entries;
 	size_t				max_prefixlen;
 	size_t				data_size;
-	raw_spinlock_t			lock;
+	spinlock_t			lock;
 };
 
 /* This trie implements a longest prefix match algorithm that can be used to

@@ -315,7 +315,7 @@ static int trie_update_elem(struct bpf_m
 	if (key->prefixlen > trie->max_prefixlen)
 		return -EINVAL;
 
-	raw_spin_lock_irqsave(&trie->lock, irq_flags);
+	spin_lock_irqsave(&trie->lock, irq_flags);
 
 	/* Allocate and fill a new node */
 

@@ -422,7 +422,7 @@ static int trie_update_elem(struct bpf_m
 		kfree(im_node);
 	}
 
-	raw_spin_unlock_irqrestore(&trie->lock, irq_flags);
+	spin_unlock_irqrestore(&trie->lock, irq_flags);
 
 	return ret;
 }

@@ -442,7 +442,7 @@ static int trie_delete_elem(struct bpf_m
 	if (key->prefixlen > trie->max_prefixlen)
 		return -EINVAL;
 
-	raw_spin_lock_irqsave(&trie->lock, irq_flags);
+	spin_lock_irqsave(&trie->lock, irq_flags);
 
 	/* Walk the tree looking for an exact key/length match and keeping
 	 * track of the path we traverse.  We will need to know the node

@@ -518,7 +518,7 @@ static int trie_delete_elem(struct bpf_m
 	kfree_rcu(node, rcu);
 
 out:
-	raw_spin_unlock_irqrestore(&trie->lock, irq_flags);
+	spin_unlock_irqrestore(&trie->lock, irq_flags);
 
 	return ret;
 }

@@ -575,7 +575,7 @@ static struct bpf_map *trie_alloc(union
 	if (ret)
 		goto out_err;
 
-	raw_spin_lock_init(&trie->lock);
+	spin_lock_init(&trie->lock);
 
 	return &trie->map;
 out_err:

@@ -40,6 +40,9 @@ static void do_up_read(struct irq_work *
 {
 	struct stack_map_irq_work *work;
 
+	if (WARN_ON_ONCE(IS_ENABLED(CONFIG_PREEMPT_RT)))
+		return;
+
 	work = container_of(entry, struct stack_map_irq_work, irq_work);
 	up_read_non_owner(work->sem);
 	work->sem = NULL;

@@ -288,10 +291,19 @@ static void stack_map_get_build_id_offse
 	struct stack_map_irq_work *work = NULL;
 
 	if (irqs_disabled()) {
-		work = this_cpu_ptr(&up_read_work);
-		if (atomic_read(&work->irq_work.flags) & IRQ_WORK_BUSY)
-			/* cannot queue more up_read, fallback */
+		if (!IS_ENABLED(CONFIG_PREEMPT_RT)) {
+			work = this_cpu_ptr(&up_read_work);
+			if (atomic_read(&work->irq_work.flags) & IRQ_WORK_BUSY) {
+				/* cannot queue more up_read, fallback */
+				irq_work_busy = true;
+			}
+		} else {
+			/*
+			 * PREEMPT_RT does not allow to trylock mmap sem in
+			 * interrupt disabled context. Force the fallback code.
+			 */
 			irq_work_busy = true;
+		}
 	}
 
 	/*