[PATCH 5.12 346/700] mm: mmap_lock: use local locks instead of disabling preemption

[Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx>]

From: Nicolas Saenz Julienne <nsaenzju@xxxxxxxxxx>

[ Upstream commit 832b50725373e8c46781b7d4db104ec9cf564a6b ]

mmap_lock will explicitly disable/enable preemption upon manipulating its
local CPU variables.  This is to be expected, but in this case, it doesn't
play well with PREEMPT_RT.  The preemption disabled code section also
takes a spin-lock.  Spin-locks in RT systems will try to schedule, which
is exactly what we're trying to avoid.

To mitigate this, convert the explicit preemption handling to local_locks.
Which are RT aware, and will disable migration instead of preemption when
PREEMPT_RT=y.

The faulty call trace looks like the following:
    __mmap_lock_do_trace_*()
      preempt_disable()
      get_mm_memcg_path()
        cgroup_path()
          kernfs_path_from_node()
            spin_lock_irqsave() /* Scheduling while atomic! */

Link: https://lkml.kernel.org/r/20210604163506.2103900-1-nsaenzju@xxxxxxxxxx
Fixes: 2b5067a8143e3 ("mm: mmap_lock: add tracepoints around lock acquisition ")
Signed-off-by: Nicolas Saenz Julienne <nsaenzju@xxxxxxxxxx>
Tested-by: Axel Rasmussen <axelrasmussen@xxxxxxxxxx>
Reviewed-by: Axel Rasmussen <axelrasmussen@xxxxxxxxxx>
Cc: Vlastimil Babka <vbabka@xxxxxxx>
Cc: Steven Rostedt <rostedt@xxxxxxxxxxx>
Signed-off-by: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>
Signed-off-by: Linus Torvalds <torvalds@xxxxxxxxxxxxxxxxxxxx>
Signed-off-by: Sasha Levin <sashal@xxxxxxxxxx>
---
 mm/mmap_lock.c | 33 ++++++++++++++++++++++-----------
 1 file changed, 22 insertions(+), 11 deletions(-)

diff --git a/mm/mmap_lock.c b/mm/mmap_lock.c
index dcdde4f722a4..2ae3f33b85b1 100644
--- a/mm/mmap_lock.c
+++ b/mm/mmap_lock.c
@@ -11,6 +11,7 @@
 #include <linux/rcupdate.h>
 #include <linux/smp.h>
 #include <linux/trace_events.h>
+#include <linux/local_lock.h>
 
 EXPORT_TRACEPOINT_SYMBOL(mmap_lock_start_locking);
 EXPORT_TRACEPOINT_SYMBOL(mmap_lock_acquire_returned);
@@ -39,21 +40,30 @@ static int reg_refcount; /* Protected by reg_lock. */
  */
 #define CONTEXT_COUNT 4
 
-static DEFINE_PER_CPU(char __rcu *, memcg_path_buf);
+struct memcg_path {
+	local_lock_t lock;
+	char __rcu *buf;
+	local_t buf_idx;
+};
+static DEFINE_PER_CPU(struct memcg_path, memcg_paths) = {
+	.lock = INIT_LOCAL_LOCK(lock),
+	.buf_idx = LOCAL_INIT(0),
+};
+
 static char **tmp_bufs;
-static DEFINE_PER_CPU(int, memcg_path_buf_idx);
 
 /* Called with reg_lock held. */
 static void free_memcg_path_bufs(void)
 {
+	struct memcg_path *memcg_path;
 	int cpu;
 	char **old = tmp_bufs;
 
 	for_each_possible_cpu(cpu) {
-		*(old++) = rcu_dereference_protected(
-			per_cpu(memcg_path_buf, cpu),
+		memcg_path = per_cpu_ptr(&memcg_paths, cpu);
+		*(old++) = rcu_dereference_protected(memcg_path->buf,
 			lockdep_is_held(&reg_lock));
-		rcu_assign_pointer(per_cpu(memcg_path_buf, cpu), NULL);
+		rcu_assign_pointer(memcg_path->buf, NULL);
 	}
 
 	/* Wait for inflight memcg_path_buf users to finish. */
@@ -88,7 +98,7 @@ int trace_mmap_lock_reg(void)
 		new = kmalloc(MEMCG_PATH_BUF_SIZE * CONTEXT_COUNT, GFP_KERNEL);
 		if (new == NULL)
 			goto out_fail_free;
-		rcu_assign_pointer(per_cpu(memcg_path_buf, cpu), new);
+		rcu_assign_pointer(per_cpu_ptr(&memcg_paths, cpu)->buf, new);
 		/* Don't need to wait for inflights, they'd have gotten NULL. */
 	}
 
@@ -122,23 +132,24 @@ out:
 
 static inline char *get_memcg_path_buf(void)
 {
+	struct memcg_path *memcg_path = this_cpu_ptr(&memcg_paths);
 	char *buf;
 	int idx;
 
 	rcu_read_lock();
-	buf = rcu_dereference(*this_cpu_ptr(&memcg_path_buf));
+	buf = rcu_dereference(memcg_path->buf);
 	if (buf == NULL) {
 		rcu_read_unlock();
 		return NULL;
 	}
-	idx = this_cpu_add_return(memcg_path_buf_idx, MEMCG_PATH_BUF_SIZE) -
+	idx = local_add_return(MEMCG_PATH_BUF_SIZE, &memcg_path->buf_idx) -
 	      MEMCG_PATH_BUF_SIZE;
 	return &buf[idx];
 }
 
 static inline void put_memcg_path_buf(void)
 {
-	this_cpu_sub(memcg_path_buf_idx, MEMCG_PATH_BUF_SIZE);
+	local_sub(MEMCG_PATH_BUF_SIZE, &this_cpu_ptr(&memcg_paths)->buf_idx);
 	rcu_read_unlock();
 }
 
@@ -179,14 +190,14 @@ out:
 #define TRACE_MMAP_LOCK_EVENT(type, mm, ...)                                   \
 	do {                                                                   \
 		const char *memcg_path;                                        \
-		preempt_disable();                                             \
+		local_lock(&memcg_paths.lock);				       \
 		memcg_path = get_mm_memcg_path(mm);                            \
 		trace_mmap_lock_##type(mm,                                     \
 				       memcg_path != NULL ? memcg_path : "",   \
 				       ##__VA_ARGS__);                         \
 		if (likely(memcg_path != NULL))                                \
 			put_memcg_path_buf();                                  \
-		preempt_enable();                                              \
+		local_unlock(&memcg_paths.lock);			       \
 	} while (0)
 
 #else /* !CONFIG_MEMCG */
-- 
2.30.2