drm/lima: use drm_sched for GPU scheduling

Signed-off-by: Qiang Yu 

Author: yuq

drivers/gpu/drm/lima/Kconfig

@@ -3,5 +3,6 @@ config DRM_LIMA
        tristate "LIMA (DRM support for ARM Mali 400/450 GPU)"
        depends on DRM
        depends on ARCH_SUNXI || ARCH_ROCKCHIP || ARCH_EXYNOS || ARCH_MESON
+       select DRM_SCHED
        help
          DRM driver for ARM Mali 400/450 GPUs.

drivers/gpu/drm/lima/lima.h

@@ -91,6 +91,8 @@ struct lima_pp {
 	atomic_t task;
 };
 
+#define LIMA_MAX_PIPE 2
+
 struct lima_device {
 	struct device *dev;
 	struct drm_device *ddev;
@@ -108,7 +110,7 @@ struct lima_device {
 
 	struct lima_l2_cache *l2_cache;
 
-	struct lima_sched_pipe *pipe[2];
+	struct lima_sched_pipe *pipe[LIMA_MAX_PIPE];
 
 	struct lima_gp *gp;
 	struct lima_pp *pp;
@@ -119,6 +121,7 @@ struct lima_device {
 
 struct lima_drm_priv {
 	struct lima_vm *vm;
+	struct lima_sched_context context[LIMA_MAX_PIPE];
 };
 
 struct lima_bo_va_mapping {
@@ -161,9 +164,9 @@ int lima_gem_mmap_offset(struct drm_file *file, u32 handle, u64 *offset);
 int lima_gem_mmap(struct file *filp, struct vm_area_struct *vma);
 int lima_gem_va_map(struct drm_file *file, u32 handle, u32 flags, u32 va);
 int lima_gem_va_unmap(struct drm_file *file, u32 handle, u32 va);
-int lima_gem_submit(struct drm_file *file, struct lima_sched_pipe *pipe,
-		    struct drm_lima_gem_submit_bo *bos, u32 nr_bos,
-		    void *frame, u32 *fence);
+int lima_gem_submit(struct drm_file *file, int pipe,
+		    struct drm_lima_gem_submit_bo *bos,
+		    u32 nr_bos, void *frame, u32 *fence);
 int lima_gem_wait(struct drm_file *file, u32 handle, u32 op, u64 timeout_ns);
 struct drm_gem_object *lima_gem_prime_import_sg_table(struct drm_device *dev,
 						      struct dma_buf_attachment *attach,

drivers/gpu/drm/lima/lima_drv.c

@@ -118,8 +118,7 @@ static int lima_ioctl_gem_submit(struct drm_device *dev, void *data, struct drm_
 		break;
 	}
 
-	err = lima_gem_submit(file, ldev->pipe[args->pipe], bos, args->nr_bos,
-			      frame, &args->fence);
+	err = lima_gem_submit(file, args->pipe, bos, args->nr_bos, frame, &args->fence);
 
 out1:
 	if (err)
@@ -132,12 +131,13 @@ static int lima_ioctl_gem_submit(struct drm_device *dev, void *data, struct drm_
 static int lima_ioctl_wait_fence(struct drm_device *dev, void *data, struct drm_file *file)
 {
 	struct drm_lima_wait_fence *args = data;
-	struct lima_device *ldev = to_lima_dev(dev);
+	struct lima_drm_priv *priv = file->driver_priv;
 
-	if (args->pipe >= ARRAY_SIZE(ldev->pipe))
+	if (args->pipe >= ARRAY_SIZE(priv->context))
 		return -EINVAL;
 
-	return lima_sched_pipe_wait_fence(ldev->pipe[args->pipe], args->fence, args->timeout_ns);
+	return lima_sched_context_wait_fence(priv->context + args->pipe,
+					     args->fence, args->timeout_ns);
 }
 
 static int lima_ioctl_gem_wait(struct drm_device *dev, void *data, struct drm_file *file)
@@ -152,7 +152,7 @@ static int lima_ioctl_gem_wait(struct drm_device *dev, void *data, struct drm_fi
 
 static int lima_drm_driver_open(struct drm_device *dev, struct drm_file *file)
 {
-	int err;
+	int err, i;
 	struct lima_drm_priv *priv;
 	struct lima_device *ldev = to_lima_dev(dev);
 
@@ -166,9 +166,19 @@ static int lima_drm_driver_open(struct drm_device *dev, struct drm_file *file)
 		goto err_out0;
 	}
 
+	for (i = 0; i < LIMA_MAX_PIPE; i++) {
+		err = lima_sched_context_init(ldev->pipe[i], priv->context + i);
+		if (err)
+			goto err_out1;
+	}
+
 	file->driver_priv = priv;
 	return 0;
 
+err_out1:
+	for (i--; i >= 0; i--)
+		lima_sched_context_fini(ldev->pipe[i], priv->context + i);
+	lima_vm_put(priv->vm);
 err_out0:
 	kfree(priv);
 	return err;
@@ -177,6 +187,11 @@ static int lima_drm_driver_open(struct drm_device *dev, struct drm_file *file)
 static void lima_drm_driver_postclose(struct drm_device *dev, struct drm_file *file)
 {
 	struct lima_drm_priv *priv = file->driver_priv;
+	struct lima_device *ldev = to_lima_dev(dev);
+	int i;
+
+	for (i = 0; i < LIMA_MAX_PIPE; i++)
+		lima_sched_context_fini(ldev->pipe[i], priv->context + i);
 
 	lima_vm_put(priv->vm);
 	kfree(priv);

drivers/gpu/drm/lima/lima_gem.c

@@ -374,11 +374,11 @@ static int lima_gem_lock_bos(struct lima_bo **bos, u32 nr_bos,
 	return ret;
 }
 
-static int lima_gem_sync_bo(struct lima_sched_task *task, u64 context,
-			    struct lima_bo *bo, bool write)
+static int lima_gem_sync_bo(struct lima_sched_task *task, struct lima_bo *bo, bool write)
 {
 	int i, err;
 	struct dma_fence *f;
+	u64 context = task->base.s_fence->finished.context;
 
 	if (write) {
 		struct reservation_object_list *fobj =
@@ -413,9 +413,9 @@ static int lima_gem_sync_bo(struct lima_sched_task *task, u64 context,
 	return 0;
 }
 
-int lima_gem_submit(struct drm_file *file, struct lima_sched_pipe *pipe,
-		    struct drm_lima_gem_submit_bo *bos, u32 nr_bos,
-		    void *frame, u32 *fence)
+int lima_gem_submit(struct drm_file *file, int pipe,
+		    struct drm_lima_gem_submit_bo *bos,
+		    u32 nr_bos, void *frame, u32 *fence)
 {
 	struct lima_bo **lbos;
 	int i, err = 0;
@@ -442,32 +442,28 @@ int lima_gem_submit(struct drm_file *file, struct lima_sched_pipe *pipe,
 	if (err)
 		goto out0;
 
-	task = lima_sched_task_create(priv->vm, frame);
+	task = lima_sched_task_create(priv->context + pipe, priv->vm, frame);
 	if (IS_ERR(task)) {
 		err = PTR_ERR(task);
 		goto out1;
 	}
 
 	for (i = 0; i < nr_bos; i++) {
-		err = lima_gem_sync_bo(task, pipe->fence_context, lbos[i],
-				       bos[i].flags & LIMA_SUBMIT_BO_WRITE);
+		err = lima_gem_sync_bo(task, lbos[i], bos[i].flags & LIMA_SUBMIT_BO_WRITE);
 		if (err)
 			goto out2;
 	}
 
-	err = lima_sched_task_queue(pipe, task);
-	if (err)
-		goto out2;
-
 	for (i = 0; i < nr_bos; i++) {
 		if (bos[i].flags & LIMA_SUBMIT_BO_WRITE)
-			reservation_object_add_excl_fence(lbos[i]->resv, task->fence);
+			reservation_object_add_excl_fence(
+				lbos[i]->resv, &task->base.s_fence->finished);
 		else
-			reservation_object_add_shared_fence(lbos[i]->resv, task->fence);
+			reservation_object_add_shared_fence(
+				lbos[i]->resv, &task->base.s_fence->finished);
 	}
-	dma_fence_put(task->fence);
 
-	*fence = task->fence->seqno;
+	*fence = lima_sched_context_queue_task(priv->context + pipe, task);
 
 out2:
 	if (err)

drivers/gpu/drm/lima/lima_gp.c

@@ -106,20 +106,22 @@ static irqreturn_t lima_gp_irq_handler(int irq, void *data)
 	struct lima_gp *gp = data;
 	struct lima_device *dev = gp->ip.dev;
 	u32 state = gp_read(INT_STAT);
+	bool task_done = false, fail = false;
 
 	/* for shared irq case */
 	if (!state)
 		return IRQ_NONE;
 
 	if (state & LIMA_GP_IRQ_MASK_ERROR) {
 		u32 status = gp_read(STATUS);
+
 		dev_info(dev->dev, "gp error irq state=%x status=%x\n",
 			 state, status);
-		lima_sched_pipe_task_done(&gp->pipe, true);
+
+		fail = true;
+		task_done = true;
 	}
 	else {
-		bool task_done = false;
-
 		if (state & LIMA_GP_IRQ_VS_END_CMD_LST) {
 			gp->task &= ~LIMA_GP_TASK_VS;
 			task_done = true;
@@ -129,12 +131,16 @@ static irqreturn_t lima_gp_irq_handler(int irq, void *data)
 			gp->task &= ~LIMA_GP_TASK_PLBU;
 			task_done = true;
 		}
-
-		if (task_done && !gp->task)
-			lima_sched_pipe_task_done(&gp->pipe, false);
 	}
 
 	gp_write(INT_CLEAR, state);
+
+	if (task_done) {
+		if (fail)
+			lima_sched_pipe_task_done(&gp->pipe, true);
+		else if (!gp->task)
+			lima_sched_pipe_task_done(&gp->pipe, false);
+	}
 	return IRQ_HANDLED;
 }
 

drivers/gpu/drm/lima/lima_pp.c

@@ -89,24 +89,32 @@ static irqreturn_t lima_pp_core_irq_handler(int irq, void *data)
 	struct lima_device *dev = core->ip.dev;
 	struct lima_pp *pp = dev->pp;
 	u32 state = pp_read(INT_STATUS);
+	bool task_done = false, fail = false;
 
 	/* for shared irq case */
 	if (!state)
 		return IRQ_NONE;
 
 	if (state & LIMA_PP_IRQ_MASK_ERROR) {
 		u32 status = pp_read(STATUS);
+
 		dev_info(dev->dev, "pp error irq state=%x status=%x\n",
 			 state, status);
-		lima_sched_pipe_task_done(&pp->pipe, true);
+
+		task_done = true;
+		fail = true;
 	}
 	else {
 		if ((state & LIMA_PP_IRQ_END_OF_FRAME) &&
 		    atomic_dec_and_test(&pp->task))
-			lima_sched_pipe_task_done(&pp->pipe, false);
+			task_done = true;
 	}
 
 	pp_write(INT_CLEAR, state);
+
+	if (task_done)
+		lima_sched_pipe_task_done(&pp->pipe, fail);
+
 	return IRQ_HANDLED;
 }