Igt-dev Archive on lore.kernel.org
 help / color / mirror / Atom feed
* [igt-dev] [PATCH i-g-t] lib/xe_query: Prevent cache corruption on multiple access
@ 2023-03-24 12:06 Zbigniew Kempczyński
  2023-03-24 13:08 ` [igt-dev] ✗ Fi.CI.BAT: failure for " Patchwork
  2023-03-24 20:59 ` [igt-dev] [PATCH i-g-t] " Niranjana Vishwanathapura
  0 siblings, 2 replies; 3+ messages in thread
From: Zbigniew Kempczyński @ 2023-03-24 12:06 UTC (permalink / raw)
  To: igt-dev

If multiple threads will enter getting the xe_device and cache is
empty we may encounter situation all of them will try to add config
data to the cache. As cache is resizable map and wasn't protected
by the mutex at the moment of insertion it's possible to corrupt the
map.

Lets add protecting mutex at the moment of adding to the cache. Due
to assertions around querying the kernel with ioctls mutex doesn't
protect gathering the config data so some threads may do this in
parallel. It's not a big deal though as all threads will see same
configuration so we add first (winning) thread xe_device data.

Signed-off-by: Zbigniew Kempczyński <zbigniew.kempczynski@intel.com>
---
 lib/xe/xe_query.c | 33 +++++++++++++++++++++------------
 1 file changed, 21 insertions(+), 12 deletions(-)

diff --git a/lib/xe/xe_query.c b/lib/xe/xe_query.c
index 1835232804..82a7e36e30 100644
--- a/lib/xe/xe_query.c
+++ b/lib/xe/xe_query.c
@@ -223,6 +223,16 @@ static struct xe_device *find_in_cache(int fd)
 	return xe_dev;
 }
 
+static void xe_device_free(struct xe_device *xe_dev)
+{
+	free(xe_dev->config);
+	free(xe_dev->gts);
+	free(xe_dev->hw_engines);
+	free(xe_dev->mem_usage);
+	free(xe_dev->vram_size);
+	free(xe_dev);
+}
+
 /**
  * xe_device_get:
  * @fd: xe device fd
@@ -234,7 +244,7 @@ static struct xe_device *find_in_cache(int fd)
  */
 struct xe_device *xe_device_get(int fd)
 {
-	struct xe_device *xe_dev;
+	struct xe_device *xe_dev, *prev;
 
 	xe_dev = find_in_cache(fd);
 	if (xe_dev)
@@ -260,21 +270,20 @@ struct xe_device *xe_device_get(int fd)
 	xe_dev->has_vram = __mem_has_vram(xe_dev->mem_usage);
 	xe_dev->supports_faults = xe_check_supports_faults(fd);
 
-	igt_map_insert(cache.map, &xe_dev->fd, xe_dev);
+	/* We may get here from multiple threads, use first cached xe_dev */
+	pthread_mutex_lock(&cache.cache_mutex);
+	prev = find_in_cache_unlocked(fd);
+	if (!prev) {
+		igt_map_insert(cache.map, &xe_dev->fd, xe_dev);
+	} else {
+		xe_device_free(xe_dev);
+		xe_dev = prev;
+	}
+	pthread_mutex_unlock(&cache.cache_mutex);
 
 	return xe_dev;
 }
 
-static void xe_device_free(struct xe_device *xe_dev)
-{
-	free(xe_dev->config);
-	free(xe_dev->gts);
-	free(xe_dev->hw_engines);
-	free(xe_dev->mem_usage);
-	free(xe_dev->vram_size);
-	free(xe_dev);
-}
-
 static void delete_in_cache(struct igt_map_entry *entry)
 {
 	xe_device_free((struct xe_device *)entry->data);
-- 
2.34.1

^ permalink raw reply related	[flat|nested] 3+ messages in thread
end of thread, other threads:[~2023-03-24 21:03 UTC | newest]

Thread overview: 3+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2023-03-24 12:06 [igt-dev] [PATCH i-g-t] lib/xe_query: Prevent cache corruption on multiple access Zbigniew Kempczyński
2023-03-24 13:08 ` [igt-dev] ✗ Fi.CI.BAT: failure for " Patchwork
2023-03-24 20:59 ` [igt-dev] [PATCH i-g-t] " Niranjana Vishwanathapura

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox