gpu: nvgpu: acquire mutex for notifier read

We use &ch->error_notifier_mutex to protect
writes and free of error notifier
But we currently do not protect reading of
notifier in gk20a_fifo_set_ctx_mmu_error()
and vgpu_fifo_set_ctx_mmu_error()

Add new API gk20a_set_error_notifier_locked()
which is same as gk20a_set_error_notifier()
but without the locks.

In *_fifo_set_ctx_mmu_error() APIs, acquire
the mutex explicitly, and then use this new
API

gk20a_set_error_notifier() will now just call
gk20a_set_error_notifier_locked() within
a mutex

Bug 1824788
Bug 1844312

Change-Id: I1f3831dc63fe1daa761b2e17e4de3c155f505d6f
Signed-off-by: Deepak Nibade <dnibade@nvidia.com>
Reviewed-on: http://git-master/r/1273471
Reviewed-by: Automatic_Commit_Validation_User
Reviewed-by: Richard Zhao <rizhao@nvidia.com>
GVS: Gerrit_Virtual_Submit
Reviewed-by: Bharat Nihalani <bnihalani@nvidia.com>

4 files changed, 24 insertions, 13 deletions

diff --git a/drivers/gpu/nvgpu/gk20a/channel_gk20a.c b/drivers/gpu/nvgpu/gk20a/channel_gk20a.c
index 129921ad..38cb368c 100644
--- a/drivers/gpu/nvgpu/gk20a/channel_gk20a.c
+++ b/drivers/gpu/nvgpu/gk20a/channel_gk20a.c
@@ -827,11 +827,12 @@ static int gk20a_init_error_notifier(struct channel_gk20a *ch,
         return 0;
 }
 
-void gk20a_set_error_notifier(struct channel_gk20a *ch, __u32 error)
+/**
+ * gk20a_set_error_notifier_locked()
+ * Should be called with ch->error_notifier_mutex held
+ */
+void gk20a_set_error_notifier_locked(struct channel_gk20a *ch, __u32 error)
 {
-        bool notifier_set = false;
-
-        mutex_lock(&ch->error_notifier_mutex);
         if (ch->error_notifier_ref) {
                 struct timespec time_data;
                 u64 nsec;
@@ -845,13 +846,16 @@ void gk20a_set_error_notifier(struct channel_gk20a *ch, __u32 error)
                 ch->error_notifier->info32 = error;
                 ch->error_notifier->status = 0xffff;
 
-                notifier_set = true;
-        }
-        mutex_unlock(&ch->error_notifier_mutex);
-
-        if (notifier_set)
                 gk20a_err(dev_from_gk20a(ch->g),
                     "error notifier set to %d for ch %d", error, ch->hw_chid);
+        }
+}
+
+void gk20a_set_error_notifier(struct channel_gk20a *ch, __u32 error)
+{
+        mutex_lock(&ch->error_notifier_mutex);
+        gk20a_set_error_notifier_locked(ch, error);
+        mutex_unlock(&ch->error_notifier_mutex);
 }
 
 static void gk20a_free_error_notifiers(struct channel_gk20a *ch)
diff --git a/drivers/gpu/nvgpu/gk20a/channel_gk20a.h b/drivers/gpu/nvgpu/gk20a/channel_gk20a.h
index ed0b04ee..0a0d94b7 100644
--- a/drivers/gpu/nvgpu/gk20a/channel_gk20a.h
+++ b/drivers/gpu/nvgpu/gk20a/channel_gk20a.h
@@ -235,6 +235,7 @@ void gk20a_disable_channel(struct channel_gk20a *ch);
 void gk20a_channel_abort(struct channel_gk20a *ch, bool channel_preempt);
 void gk20a_channel_abort_clean_up(struct channel_gk20a *ch);
 void gk20a_set_error_notifier(struct channel_gk20a *ch, __u32 error);
+void gk20a_set_error_notifier_locked(struct channel_gk20a *ch, __u32 error);
 void gk20a_channel_semaphore_wakeup(struct gk20a *g, bool post_events);
 int gk20a_channel_alloc_priv_cmdbuf(struct channel_gk20a *c, u32 size,
                              struct priv_cmd_entry *entry);
diff --git a/drivers/gpu/nvgpu/gk20a/fifo_gk20a.c b/drivers/gpu/nvgpu/gk20a/fifo_gk20a.c
index b4fa8f0f..fb772ebd 100644
--- a/drivers/gpu/nvgpu/gk20a/fifo_gk20a.c
+++ b/drivers/gpu/nvgpu/gk20a/fifo_gk20a.c
@@ -1219,7 +1219,8 @@ static bool gk20a_fifo_set_ctx_mmu_error(struct gk20a *g,
         if (!ch)
                 return verbose;
 
-        if (ch->error_notifier) {
+        mutex_lock(&ch->error_notifier_mutex);
+        if (ch->error_notifier_ref) {
                 u32 err = ch->error_notifier->info32;
                 if (ch->error_notifier->status == 0xffff) {
                         /* If error code is already set, this mmu fault
@@ -1230,10 +1231,12 @@ static bool gk20a_fifo_set_ctx_mmu_error(struct gk20a *g,
                         if (err == NVGPU_CHANNEL_FIFO_ERROR_IDLE_TIMEOUT)
                                 verbose = ch->timeout_debug_dump;
                 } else {
-                        gk20a_set_error_notifier(ch,
+                        gk20a_set_error_notifier_locked(ch,
                                 NVGPU_CHANNEL_FIFO_ERROR_MMU_ERR_FLT);
                 }
         }
+        mutex_unlock(&ch->error_notifier_mutex);
+
         /* mark channel as faulted */
         ch->has_timedout = true;
         wmb();
diff --git a/drivers/gpu/nvgpu/vgpu/fifo_vgpu.c b/drivers/gpu/nvgpu/vgpu/fifo_vgpu.c
index 5ffc6a00..e9a2d582 100644
--- a/drivers/gpu/nvgpu/vgpu/fifo_vgpu.c
+++ b/drivers/gpu/nvgpu/vgpu/fifo_vgpu.c
@@ -678,17 +678,20 @@ static int vgpu_fifo_force_reset_ch(struct channel_gk20a *ch,
 static void vgpu_fifo_set_ctx_mmu_error(struct gk20a *g,
                 struct channel_gk20a *ch)
 {
-        if (ch->error_notifier) {
+        mutex_lock(&ch->error_notifier_mutex);
+        if (ch->error_notifier_ref) {
                 if (ch->error_notifier->status == 0xffff) {
                         /* If error code is already set, this mmu fault
                          * was triggered as part of recovery from other
                          * error condition.
                          * Don't overwrite error flag. */
                 } else {
-                        gk20a_set_error_notifier(ch,
+                        gk20a_set_error_notifier_locked(ch,
                                 NVGPU_CHANNEL_FIFO_ERROR_MMU_ERR_FLT);
                 }
         }
+        mutex_unlock(&ch->error_notifier_mutex);
+
         /* mark channel as faulted */
         ch->has_timedout = true;
         wmb();