drm/msm: fix potential deadlock in gpu init

Somewhere along the way, the firmware loader sprouted another lock
dependency, resulting in possible deadlock scenario:

 &dev->struct_mutex --> &sb->s_type->i_mutex_key#2 --> &mm->mmap_sem

which is problematic vs things like gem mmap.

So introduce a separate mutex to synchronize gpu init.

Signed-off-by: Rob Clark <robdclark@gmail.com>
diff --git a/drivers/gpu/drm/msm/adreno/adreno_gpu.c b/drivers/gpu/drm/msm/adreno/adreno_gpu.c
index 76c1df7..655ce5b 100644
--- a/drivers/gpu/drm/msm/adreno/adreno_gpu.c
+++ b/drivers/gpu/drm/msm/adreno/adreno_gpu.c
@@ -95,7 +95,7 @@
 
 	DBG("%s", gpu->name);
 
-	ret = msm_gem_get_iova_locked(gpu->rb->bo, gpu->id, &gpu->rb_iova);
+	ret = msm_gem_get_iova(gpu->rb->bo, gpu->id, &gpu->rb_iova);
 	if (ret) {
 		gpu->rb_iova = 0;
 		dev_err(gpu->dev->dev, "could not map ringbuffer: %d\n", ret);
@@ -370,8 +370,10 @@
 			return ret;
 	}
 
+	mutex_lock(&drm->struct_mutex);
 	gpu->memptrs_bo = msm_gem_new(drm, sizeof(*gpu->memptrs),
 			MSM_BO_UNCACHED);
+	mutex_unlock(&drm->struct_mutex);
 	if (IS_ERR(gpu->memptrs_bo)) {
 		ret = PTR_ERR(gpu->memptrs_bo);
 		gpu->memptrs_bo = NULL;
@@ -379,13 +381,13 @@
 		return ret;
 	}
 
-	gpu->memptrs = msm_gem_vaddr_locked(gpu->memptrs_bo);
+	gpu->memptrs = msm_gem_vaddr(gpu->memptrs_bo);
 	if (!gpu->memptrs) {
 		dev_err(drm->dev, "could not vmap memptrs\n");
 		return -ENOMEM;
 	}
 
-	ret = msm_gem_get_iova_locked(gpu->memptrs_bo, gpu->base.id,
+	ret = msm_gem_get_iova(gpu->memptrs_bo, gpu->base.id,
 			&gpu->memptrs_iova);
 	if (ret) {
 		dev_err(drm->dev, "could not map memptrs: %d\n", ret);
diff --git a/drivers/gpu/drm/msm/msm_drv.c b/drivers/gpu/drm/msm/msm_drv.c
index a2f5bf6..b447c01 100644
--- a/drivers/gpu/drm/msm/msm_drv.c
+++ b/drivers/gpu/drm/msm/msm_drv.c
@@ -181,7 +181,6 @@
 	struct msm_kms *kms;
 	int ret;
 
-
 	priv = kzalloc(sizeof(*priv), GFP_KERNEL);
 	if (!priv) {
 		dev_err(dev->dev, "failed to allocate private data\n");
@@ -314,13 +313,15 @@
 
 static void load_gpu(struct drm_device *dev)
 {
+	static DEFINE_MUTEX(init_lock);
 	struct msm_drm_private *priv = dev->dev_private;
 	struct msm_gpu *gpu;
 
-	if (priv->gpu)
-		return;
+	mutex_lock(&init_lock);
 
-	mutex_lock(&dev->struct_mutex);
+	if (priv->gpu)
+		goto out;
+
 	gpu = a3xx_gpu_init(dev);
 	if (IS_ERR(gpu)) {
 		dev_warn(dev->dev, "failed to load a3xx gpu\n");
@@ -330,7 +331,9 @@
 
 	if (gpu) {
 		int ret;
+		mutex_lock(&dev->struct_mutex);
 		gpu->funcs->pm_resume(gpu);
+		mutex_unlock(&dev->struct_mutex);
 		ret = gpu->funcs->hw_init(gpu);
 		if (ret) {
 			dev_err(dev->dev, "gpu hw init failed: %d\n", ret);
@@ -340,12 +343,12 @@
 			/* give inactive pm a chance to kick in: */
 			msm_gpu_retire(gpu);
 		}
-
 	}
 
 	priv->gpu = gpu;
 
-	mutex_unlock(&dev->struct_mutex);
+out:
+	mutex_unlock(&init_lock);
 }
 
 static int msm_open(struct drm_device *dev, struct drm_file *file)
diff --git a/drivers/gpu/drm/msm/msm_gpu.c b/drivers/gpu/drm/msm/msm_gpu.c
index 915240b..4a0dce5 100644
--- a/drivers/gpu/drm/msm/msm_gpu.c
+++ b/drivers/gpu/drm/msm/msm_gpu.c
@@ -612,8 +612,11 @@
 	}
 	gpu->id = msm_register_mmu(drm, gpu->mmu);
 
+
 	/* Create ringbuffer: */
+	mutex_lock(&drm->struct_mutex);
 	gpu->rb = msm_ringbuffer_new(gpu, ringsz);
+	mutex_unlock(&drm->struct_mutex);
 	if (IS_ERR(gpu->rb)) {
 		ret = PTR_ERR(gpu->rb);
 		gpu->rb = NULL;