Merge tag 'drm-msm-next-2021-12-26' of ssh://gitlab.freedesktop.org/drm/msm into drm-next

* dpu plane state cleanup in prep for multirect
* dpu debugfs cleanup (and moving things to atomic_print_state) in prep
  for multirect
* dp support for sc7280
* struct_mutex removal
* include more GMU state in gpu devcore dumps
* add support for a506
* remove old eDP sub-driver (never was used in any upstream supported
  devices and modern things with eDP will use DP sub-driver instead)
* debugfs to disable hw gpu hang detect for (igt tests)
* debugfs for dumping display hw state
* and the usual assortment of cleanup and bug fixes

There still seems to be a timing issue with dpu, showing up on sc7180
devices, after the bridge probe-order change. Ie. things work great if
loglevel is high enough (or enough debug options are enabled, etc).
We'll continue to debug this in the new year.

Signed-off-by: Dave Airlie <airlied@redhat.com>
From: Rob Clark <robdclark@gmail.com>
Link: https://patchwork.freedesktop.org/patch/msgid/CAF6AEGs+vwr0nkwgYzuYAsCoHtypWpWav+yVvLZGsEJy8tJ56A@mail.gmail.com

1 files changed, 9 insertions, 7 deletions

diff --git a/drivers/gpu/drm/msm/msm_rd.c b/drivers/gpu/drm/msm/msm_rd.c
index b55398a34fa4..81432ec07012 100644
--- a/drivers/gpu/drm/msm/msm_rd.c
+++ b/drivers/gpu/drm/msm/msm_rd.c
@@ -86,7 +86,7 @@ struct msm_rd_state {
 	struct msm_gem_submit *submit;
 
 	/* fifo access is synchronized on the producer side by
-	 * struct_mutex held by submit code (otherwise we could
+	 * gpu->lock held by submit code (otherwise we could
 	 * end up w/ cmds logged in different order than they
 	 * were executed).  And read_lock synchronizes the reads
 	 */
@@ -181,9 +181,12 @@ static int rd_open(struct inode *inode, struct file *file)
 	uint32_t gpu_id;
 	int ret = 0;
 
-	mutex_lock(&dev->struct_mutex);
+	if (!gpu)
+		return -ENODEV;
 
-	if (rd->open || !gpu) {
+	mutex_lock(&gpu->lock);
+
+	if (rd->open) {
 		ret = -EBUSY;
 		goto out;
 	}
@@ -200,7 +203,7 @@ static int rd_open(struct inode *inode, struct file *file)
 	rd_write_section(rd, RD_GPU_ID, &gpu_id, sizeof(gpu_id));
 
 out:
-	mutex_unlock(&dev->struct_mutex);
+	mutex_unlock(&gpu->lock);
 	return ret;
 }
 
@@ -340,11 +343,10 @@ out_unlock:
 	msm_gem_unlock(&obj->base);
 }
 
-/* called under struct_mutex */
+/* called under gpu->lock */
 void msm_rd_dump_submit(struct msm_rd_state *rd, struct msm_gem_submit *submit,
 		const char *fmt, ...)
 {
-	struct drm_device *dev = submit->dev;
 	struct task_struct *task;
 	char msg[256];
 	int i, n;
@@ -355,7 +357,7 @@ void msm_rd_dump_submit(struct msm_rd_state *rd, struct msm_gem_submit *submit,
 	/* writing into fifo is serialized by caller, and
 	 * rd->read_lock is used to serialize the reads
 	 */
-	WARN_ON(!mutex_is_locked(&dev->struct_mutex));
+	WARN_ON(!mutex_is_locked(&submit->gpu->lock));
 
 	if (fmt) {
 		va_list args;