Skip to contentSkip to conversation

drm, i915: Fix memory leak in i915_gem_busy_ioctl().

PATCH In review Rakib Mullick 4 participants 8 messages Nov 15, 2011 – Nov 28, 2011

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 drm, i915: Fix memory leak in i915_gem_busy_ioctl().

A call to i915_add_request() has been made in function i915_gem_busy_ioctl(). i915_add_request can fail, so in it's exit path previously allocated memory needs to be freed.

---

drivers/gpu/drm/i915/i915_gem.c +4 -2
index d18b07a..2dc24ab 100644
@@ -3512,9 +3512,11 @@ i915_gem_busy_ioctl(struct drm_device *dev, void *data,
 			 * so emit a request to do so.
 			 */
 			request = kzalloc(sizeof(*request), GFP_KERNEL);
-			if (request)
+			if (request) {
 				ret = i915_add_request(obj->ring, NULL, request);
-			else
+				if (ret)
+					kfree(request);
+			} else
 				ret = -ENOMEM;
 		}
 
Reviewed-by on v1 1/1
#1 lore raw
Quoted 5 lines · Rakib Mullick

A call to i915_add_request() has been made in function i915_gem_busy_ioctl(). i915_add_request can fail, so in it's exit path previously allocated memory needs to be freed.

Indeed, nice catch (albeit totally unlikely to be hit, because the error only happens when the gpu ceases to progress in the ring, so imo not stable material). Keith, please pick this up for fixes, thanks.

Signature · 3 lines
Daniel Vetter
Mail: [email protected]
Mobile: +41 (0)79 365 57 48
#2 lore raw
Quoted 3 lines · Daniel Vetter

Indeed, nice catch (albeit totally unlikely to be hit, because the error only happens when the gpu ceases to progress in the ring, so imo not stable material). Keith, please pick this up for fixes, thanks.

It's already there and queued for airlied :-)

Signature · 1 line
[email protected]
#3 lore raw
Quoted 8 lines · Keith Packard
Quoted 3 lines · Daniel Vetter

Indeed, nice catch (albeit totally unlikely to be hit, because the error only happens when the gpu ceases to progress in the ring, so imo not stable material). Keith, please pick this up for fixes, thanks.

It's already there and queued for airlied :-)

Thank you guys for reviewing and taking the patch.

Now, while I was looking at the uses of i915_add_request(), I found the following code :

                        ret = i915_gem_flush_ring(ring, 0,
I915_GEM_GPU_DOMAINS);
                        request = kzalloc(sizeof(*request), GFP_KERNEL);
                        if (ret || request == NULL ||
                            i915_add_request(ring, NULL, request))
                            kfree(request);
Quoted 1 line

From above code, we might ended up by calling kfree(request) without

allocating request. Though, it's unlikely cause if request is NULL then BUG_ON will be hit in i915_add_request(). So, to unify the callee uses of i915_add_request(), I'm proposing the following patch. Please let me know what do you guys think. If you guys agree, I can sent a formal patch.

modi.linux-3.1.1/drivers/gpu/drm/i915/i915_gem.c +5 -5
Index: work/modi.linux-3.1.1/drivers/gpu/drm/i915/i915_gem.c
===================================================================
@@ -1736,8 +1736,6 @@ i915_add_request(struct intel_ring_buffe
 	int was_empty;
 	int ret;
 
-	BUG_ON(request == NULL);
-
 	ret = ring->add_request(ring, &seqno);
 	if (ret)
 	    return ret;
@@ -2002,9 +2000,11 @@ i915_gem_retire_work_handler(struct work
 			ret = i915_gem_flush_ring(ring,
 						  0, I915_GEM_GPU_DOMAINS);
 			request = kzalloc(sizeof(*request), GFP_KERNEL);
-			if (ret || request == NULL ||
-			    i915_add_request(ring, NULL, request))
-			    kfree(request);
+			if (request) {
+				ret = i915_add_request(ring, NULL, request);
+				if (ret)
+					kfree(request);
+			}
 		}
 
 		idle &= list_empty(&ring->request_list);

Thanks,
Rakib

#4 lore raw
Quoted 21 lines · Rakib Mullick
Quoted 8 lines · Keith Packard
Quoted 3 lines · Daniel Vetter

Indeed, nice catch (albeit totally unlikely to be hit, because the error only happens when the gpu ceases to progress in the ring, so imo not stable material). Keith, please pick this up for fixes, thanks.

It's already there and queued for airlied :-)

Thank you guys for reviewing and taking the patch.

Now, while I was looking at the uses of i915_add_request(), I found the following code :

                        ret = i915_gem_flush_ring(ring, 0,
I915_GEM_GPU_DOMAINS);
                        request = kzalloc(sizeof(*request), GFP_KERNEL);
                        if (ret || request == NULL ||
                            i915_add_request(ring, NULL, request))
                            kfree(request);
Quoted 6 lines

From above code, we might ended up by calling kfree(request) without allocating request. Though, it's unlikely cause if request is NULL then BUG_ON will be hit in i915_add_request(). So, to unify the callee uses of i915_add_request(), I'm proposing the following patch. Please let me know what do you guys think. If you guys agree, I can sent a formal patch.

kfree(NULL) is permitted and taken advantage of here along with the short-circuiting behaviour of '||'.
-Chris

Signature · 1 line
Chris Wilson, Intel Open Source Technology Centre
#5 lore raw
Quoted 29 lines · Chris Wilson
Quoted 21 lines · Rakib Mullick
Quoted 8 lines · Keith Packard
Quoted 3 lines · Daniel Vetter

Indeed, nice catch (albeit totally unlikely to be hit, because the error only happens when the gpu ceases to progress in the ring, so imo not stable material). Keith, please pick this up for fixes, thanks.

It's already there and queued for airlied :-)

Thank you guys for reviewing and taking the patch.

Now, while I was looking at the uses of i915_add_request(), I found the following code :

                        ret = i915_gem_flush_ring(ring, 0, I915_GEM_GPU_DOMAINS);
                        request = kzalloc(sizeof(*request), GFP_KERNEL); if (ret || request == NULL || i915_add_request(ring, NULL, request)) kfree(request);

Quoted 6 lines

From above code, we might ended up by calling kfree(request) without allocating request. Though, it's unlikely cause if request is NULL then BUG_ON will be hit in i915_add_request(). So, to unify the callee uses of i915_add_request(), I'm proposing the following patch. Please let me know what do you guys think. If you guys agree, I can sent a formal patch.

Quoted 2 lines

kfree(NULL) is permitted and taken advantage of here along with the short-circuiting behaviour of '||'.

Yes, no real problem with current code. I was just thinking from code cleanup's pov. Is BUG_ON really needed in i915_add_request() ?

thanks,
Rakib

#6 lore raw
Quoted 2 lines · Rakib Mullick

Yes, no real problem with current code. I was just thinking from code cleanup's pov. Is BUG_ON really needed in i915_add_request() ?

No, just documentation as a reminder that the request should be preallocated, ideally so that we can fail gracefully without touching hardware and leaving it in an inconsistent state wrt to our bookkeeping. (This is more apparent in the overlay code which could hang the chip/driver if we hit a malloc error too late.)

The BUG_ON has certainly outlived its usefulness.
-Chris

Signature · 1 line
Chris Wilson, Intel Open Source Technology Centre
#7 lore raw
Quoted 11 lines · Chris Wilson
Quoted 2 lines · Rakib Mullick

Yes, no real problem with current code. I was just thinking from code cleanup's pov. Is BUG_ON really needed in i915_add_request() ?

No, just documentation as a reminder that the request should be preallocated, ideally so that we can fail gracefully without touching hardware and leaving it in an inconsistent state wrt to our bookkeeping. (This is more apparent in the overlay code which could hang the chip/driver if we hit a malloc error too late.)

The BUG_ON has certainly outlived its usefulness.

Actually, I'm not seeing how BUG_ON could trigger (though, I've wrongly mentioned in previous thread, if request == NULL, BUG_ON could trigger), it's usefulness will never come into action. Other callers of i915_add_request also makes sure that, it gets called only if (request).

Although, kfree(NULL) is permitted, we shouldn't use it unnecessarily. Anyway, since the issue is not a big deal and no real bug, it could be droped.

Thanks,
Rakib