gpu: nvgpu: Protect sync by an own lock

Protect creation and deletion of sync by an own mutex. This prevents
deadlock in channel abort when abort is called from submit path.

Bug 200147887

Change-Id: I5d6308b773c1d1a6a89d4590e2e74c74d691f79d
Signed-off-by: Terje Bergstrom <tbergstrom@nvidia.com>
Reviewed-on: http://git-master/r/821127
(cherry picked from commit 4be2cc60de49d22a87986c4122accf3427e94412)
Reviewed-on: http://git-master/r/999489
(cherry picked from commit 39f4d7ad2928768e0f530ef2018233ef20548953)
Reviewed-on: http://git-master/r/1020291
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>

diff --git a/drivers/gpu/nvgpu/gk20a/channel_gk20a.c b/drivers/gpu/nvgpu/gk20a/channel_gk20a.c
index 4446597..167dd1c 100644 (file)
--- a/drivers/gpu/nvgpu/gk20a/channel_gk20a.c
+++ b/drivers/gpu/nvgpu/gk20a/channel_gk20a.c
@@ -325,10 +325,13 @@ void channel_gk20a_unbind(struct channel_gk20a *ch_gk20a)
         * resource at this point
         * if not, then it will be destroyed at channel_free()
         */
+       mutex_lock(&ch_gk20a->sync_lock);
        if (ch_gk20a->sync && platform->aggressive_sync_destroy) {
+
                ch_gk20a->sync->destroy(ch_gk20a->sync);
                ch_gk20a->sync = NULL;
        }
+       mutex_unlock(&ch_gk20a->sync_lock);
 }
 
 int channel_gk20a_alloc_inst(struct gk20a *g, struct channel_gk20a *ch)
@@ -391,10 +394,10 @@ void gk20a_channel_abort(struct channel_gk20a *ch, bool channel_preempt)
                gk20a_fifo_preempt(ch->g, ch);
 
        /* ensure no fences are pending */
-       mutex_lock(&ch->submit_lock);
+       mutex_lock(&ch->sync_lock);
        if (ch->sync)
                ch->sync->set_min_eq_max(ch->sync);
-       mutex_unlock(&ch->submit_lock);
+       mutex_unlock(&ch->sync_lock);
 
        /* release all job semaphores (applies only to jobs that use
           semaphore synchronization) */
@@ -808,10 +811,12 @@ static void gk20a_free_channel(struct channel_gk20a *ch)
        channel_gk20a_free_priv_cmdbuf(ch);
 
        /* sync must be destroyed before releasing channel vm */
+       mutex_lock(&ch->sync_lock);
        if (ch->sync) {
                ch->sync->destroy(ch->sync);
                ch->sync = NULL;
        }
+       mutex_unlock(&ch->sync_lock);
 
        /* release channel binding to the as_share */
        if (ch_vm->as_share)
@@ -1634,11 +1639,13 @@ static void gk20a_channel_clean_up_jobs(struct work_struct *work)
         * the sync resource
         */
        if (list_empty(&c->jobs)) {
+               mutex_lock(&c->sync_lock);
                if (c->sync && platform->aggressive_sync_destroy &&
                          gk20a_fence_is_expired(c->last_submit.post_fence)) {
                        c->sync->destroy(c->sync);
                        c->sync = NULL;
                }
+               mutex_unlock(&c->sync_lock);
        }
        mutex_unlock(&c->jobs_lock);
        mutex_unlock(&c->submit_lock);
@@ -1781,6 +1788,7 @@ int gk20a_submit_channel_gpfifo(struct channel_gk20a *c,
 
        mutex_lock(&c->submit_lock);
 
+       mutex_lock(&c->sync_lock);
        if (!c->sync) {
                c->sync = gk20a_channel_sync_create(c);
                if (!c->sync) {
@@ -1793,6 +1801,7 @@ int gk20a_submit_channel_gpfifo(struct channel_gk20a *c,
                if (err)
                        return err;
        }
+       mutex_unlock(&c->sync_lock);
 
        /*
         * optionally insert syncpt wait in the beginning of gpfifo submission
@@ -1999,6 +2008,7 @@ int gk20a_init_channel_support(struct gk20a *g, u32 chid)
        mutex_init(&c->submit_lock);
        INIT_DELAYED_WORK(&c->clean_up.wq, gk20a_channel_clean_up_jobs);
        mutex_init(&c->clean_up.lock);
+       mutex_init(&c->sync_lock);
        INIT_LIST_HEAD(&c->jobs);
 #if defined(CONFIG_GK20A_CYCLE_STATS)
        mutex_init(&c->cyclestate.cyclestate_buffer_mutex);

diff --git a/drivers/gpu/nvgpu/gk20a/channel_gk20a.h b/drivers/gpu/nvgpu/gk20a/channel_gk20a.h
index 2460672..6021c43 100644 (file)
--- a/drivers/gpu/nvgpu/gk20a/channel_gk20a.h
+++ b/drivers/gpu/nvgpu/gk20a/channel_gk20a.h
@@ -172,6 +172,7 @@ struct channel_gk20a {
        struct nvgpu_notification *error_notifier;
        void *error_notifier_va;
 
+       struct mutex sync_lock;
        struct gk20a_channel_sync *sync;
 
 #ifdef CONFIG_TEGRA_GR_VIRTUALIZATION