Move the assignment of the ref out-pointer in host1x_intr_add_action to happen within the spinlock. With the current arrangement, it is possible for the waiter to complete before the assignment has happened, which breaks horribly if the waiter completion callback tries to use the reference.
In practice, there is currently no situation where this issue can manifest -- it was first noticed with the upcoming DMA fence implementation patches. As such this doesn't need to be backported.
Signed-off-by: Mikko Perttunen mperttunen@nvidia.com --- drivers/gpu/host1x/intr.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)
diff --git a/drivers/gpu/host1x/intr.c b/drivers/gpu/host1x/intr.c index 69b0e8e41466..6d1f3c0fdbe7 100644 --- a/drivers/gpu/host1x/intr.c +++ b/drivers/gpu/host1x/intr.c @@ -235,10 +235,11 @@ int host1x_intr_add_action(struct host1x *host, struct host1x_syncpt *syncpt, host1x_hw_intr_enable_syncpt_intr(host, syncpt->id); }
- spin_unlock(&syncpt->intr.lock); - if (ref) *ref = waiter; + + spin_unlock(&syncpt->intr.lock); + return 0; }