gpu: nvgpu: restart timer instead of cancel

In gk20a_fifo_handle_sched_error(), we currently cancel
the timeout on all the channels
But this could cause us to miss one of stuck channel

hence, instead of cancelling, restart the timeout of channel
on which it is already active

Bug 200133289

Change-Id: I40e7e0e5394911fc110ab6fde39592b885dfaf7d
Signed-off-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-on: http://git-master/r/816133
Reviewed-by: Ishan Mittal <imittal@nvidia.com>
Tested-by: Ishan Mittal <imittal@nvidia.com>

3 files changed, 18 insertions, 5 deletions

diff --git a/drivers/gpu/nvgpu/gk20a/channel_gk20a.c b/drivers/gpu/nvgpu/gk20a/channel_gk20a.c
index 6f0d7375..34b62ac4 100644
--- a/drivers/gpu/nvgpu/gk20a/channel_gk20a.c
+++ b/drivers/gpu/nvgpu/gk20a/channel_gk20a.c
@@ -1574,7 +1574,7 @@ static void gk20a_channel_timeout_stop(struct channel_gk20a *ch)
         mutex_unlock(&ch->timeout.lock);
 }
 
-void gk20a_channel_timeout_stop_all_channels(struct gk20a *g)
+void gk20a_channel_timeout_restart_all_channels(struct gk20a *g)
 {
         u32 chid;
         struct fifo_gk20a *f = &g->fifo;
@@ -1583,7 +1583,20 @@ void gk20a_channel_timeout_stop_all_channels(struct gk20a *g)
                 struct channel_gk20a *ch = &f->channel[chid];
 
                 if (gk20a_channel_get(ch)) {
-                        gk20a_channel_timeout_stop(ch);
+                        mutex_lock(&ch->timeout.lock);
+                        if (!ch->timeout.initialized) {
+                                mutex_unlock(&ch->timeout.lock);
+                                gk20a_channel_put(ch);
+                                continue;
+                        }
+                        mutex_unlock(&ch->timeout.lock);
+
+                        cancel_delayed_work_sync(&ch->timeout.wq);
+                        if (!ch->has_timedout)
+                                schedule_delayed_work(&ch->timeout.wq,
+                                       msecs_to_jiffies(
+                                       gk20a_get_channel_watchdog_timeout(ch)));
+
                         gk20a_channel_put(ch);
                 }
         }
diff --git a/drivers/gpu/nvgpu/gk20a/channel_gk20a.h b/drivers/gpu/nvgpu/gk20a/channel_gk20a.h
index 280c50b1..3e18e053 100644
--- a/drivers/gpu/nvgpu/gk20a/channel_gk20a.h
+++ b/drivers/gpu/nvgpu/gk20a/channel_gk20a.h
@@ -256,5 +256,5 @@ void channel_gk20a_free_inst(struct gk20a *g, struct channel_gk20a *ch);
 int channel_gk20a_setup_ramfc(struct channel_gk20a *c,
                         u64 gpfifo_base, u32 gpfifo_entries, u32 flags);
 void channel_gk20a_enable(struct channel_gk20a *ch);
-void gk20a_channel_timeout_stop_all_channels(struct gk20a *g);
+void gk20a_channel_timeout_restart_all_channels(struct gk20a *g);
 #endif /* CHANNEL_GK20A_H */
diff --git a/drivers/gpu/nvgpu/gk20a/fifo_gk20a.c b/drivers/gpu/nvgpu/gk20a/fifo_gk20a.c
index ad7162fc..a035cd87 100644
--- a/drivers/gpu/nvgpu/gk20a/fifo_gk20a.c
+++ b/drivers/gpu/nvgpu/gk20a/fifo_gk20a.c
@@ -1450,7 +1450,7 @@ static bool gk20a_fifo_handle_sched_error(struct gk20a *g)
                 struct channel_gk20a *ch = &f->channel[id];
 
                 if (is_tsg) {
-                        gk20a_channel_timeout_stop_all_channels(g);
+                        gk20a_channel_timeout_restart_all_channels(g);
                         gk20a_fifo_recover(g, BIT(engine_id), id, true,
                                         true, true);
                         ret = true;
@@ -1472,7 +1472,7 @@ static bool gk20a_fifo_handle_sched_error(struct gk20a *g)
                          * Cancel all channels' timeout since SCHED error might
                          * trigger multiple watchdogs at a time
                          */
-                        gk20a_channel_timeout_stop_all_channels(g);
+                        gk20a_channel_timeout_restart_all_channels(g);
                         gk20a_fifo_recover(g, BIT(engine_id), id, false,
                                 true, ch->timeout_debug_dump);
                         ret = true;