openwrt_lede/target/linux/brcm2708/patches-4.4/0113-drm-vc4-Add-an-interface-for-capturing-the-GPU-state.patch

From 4deea5f5ad38d8b398355a636b9113150ca9da9f Mon Sep 17 00:00:00 2001
From: Eric Anholt <eric@anholt.net>
Date: Fri, 30 Oct 2015 10:09:02 -0700
Subject: [PATCH] drm/vc4: Add an interface for capturing the GPU state after a
 hang.

This can be parsed with vc4-gpu-tools tools for trying to figure out
what was going on.

Signed-off-by: Eric Anholt <eric@anholt.net>
---
 drivers/gpu/drm/vc4/vc4_bo.c  |   4 +-
 drivers/gpu/drm/vc4/vc4_drv.c |   1 +
 drivers/gpu/drm/vc4/vc4_drv.h |   4 +
 drivers/gpu/drm/vc4/vc4_gem.c | 185 ++++++++++++++++++++++++++++++++++++++++++
 include/uapi/drm/vc4_drm.h    |  45 ++++++++++
 5 files changed, 237 insertions(+), 2 deletions(-)

--- a/drivers/gpu/drm/vc4/vc4_bo.c
+++ b/drivers/gpu/drm/vc4/vc4_bo.c
@@ -415,8 +415,8 @@ int vc4_mmap(struct file *filp, struct v
 	gem_obj = vma->vm_private_data;
 	bo = to_vc4_bo(gem_obj);
 
-	if (bo->validated_shader) {
-		DRM_ERROR("mmaping of shader BOs not allowed.\n");
+	if (bo->validated_shader && (vma->vm_flags & VM_WRITE)) {
+		DRM_ERROR("mmaping of shader BOs for writing not allowed.\n");
 		return -EINVAL;
 	}
 
--- a/drivers/gpu/drm/vc4/vc4_drv.c
+++ b/drivers/gpu/drm/vc4/vc4_drv.c
@@ -81,6 +81,7 @@ static const struct drm_ioctl_desc vc4_d
 	DRM_IOCTL_DEF_DRV(VC4_CREATE_BO, vc4_create_bo_ioctl, 0),
 	DRM_IOCTL_DEF_DRV(VC4_MMAP_BO, vc4_mmap_bo_ioctl, 0),
 	DRM_IOCTL_DEF_DRV(VC4_CREATE_SHADER_BO, vc4_create_shader_bo_ioctl, 0),
+	DRM_IOCTL_DEF_DRV(VC4_GET_HANG_STATE, vc4_get_hang_state_ioctl, DRM_ROOT_ONLY),
 };
 
 static struct drm_driver vc4_drm_driver = {
--- a/drivers/gpu/drm/vc4/vc4_drv.h
+++ b/drivers/gpu/drm/vc4/vc4_drv.h
@@ -20,6 +20,8 @@ struct vc4_dev {
 	struct drm_fbdev_cma *fbdev;
 	struct rpi_firmware *firmware;
 
+	struct vc4_hang_state *hang_state;
+
 	/* The kernel-space BO cache.  Tracks buffers that have been
 	 * unreferenced by all other users (refcounts of 0!) but not
 	 * yet freed, so we can do cheap allocations.
@@ -366,6 +368,8 @@ int vc4_create_shader_bo_ioctl(struct dr
 			       struct drm_file *file_priv);
 int vc4_mmap_bo_ioctl(struct drm_device *dev, void *data,
 		      struct drm_file *file_priv);
+int vc4_get_hang_state_ioctl(struct drm_device *dev, void *data,
+			     struct drm_file *file_priv);
 int vc4_mmap(struct file *filp, struct vm_area_struct *vma);
 int vc4_prime_mmap(struct drm_gem_object *obj, struct vm_area_struct *vma);
 void *vc4_prime_vmap(struct drm_gem_object *obj);
--- a/drivers/gpu/drm/vc4/vc4_gem.c
+++ b/drivers/gpu/drm/vc4/vc4_gem.c
@@ -40,6 +40,186 @@ vc4_queue_hangcheck(struct drm_device *d
 		  round_jiffies_up(jiffies + msecs_to_jiffies(100)));
 }
 
+struct vc4_hang_state {
+	struct drm_vc4_get_hang_state user_state;
+
+	u32 bo_count;
+	struct drm_gem_object **bo;
+};
+
+static void
+vc4_free_hang_state(struct drm_device *dev, struct vc4_hang_state *state)
+{
+	unsigned int i;
+
+	mutex_lock(&dev->struct_mutex);
+	for (i = 0; i < state->user_state.bo_count; i++) {
+		drm_gem_object_unreference(state->bo[i]);
+	}
+	mutex_unlock(&dev->struct_mutex);
+
+	kfree(state);
+}
+
+int
+vc4_get_hang_state_ioctl(struct drm_device *dev, void *data,
+			 struct drm_file *file_priv)
+{
+ 	struct drm_vc4_get_hang_state *get_state = data;
+	struct drm_vc4_get_hang_state_bo *bo_state;
+	struct vc4_hang_state *kernel_state;
+ 	struct drm_vc4_get_hang_state *state;
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+	unsigned long irqflags;
+	u32 i;
+	int ret;
+
+	spin_lock_irqsave(&vc4->job_lock, irqflags);
+	kernel_state = vc4->hang_state;
+	if (!kernel_state) {
+		spin_unlock_irqrestore(&vc4->job_lock, irqflags);
+		return -ENOENT;
+	}
+	state = &kernel_state->user_state;
+
+	/* If the user's array isn't big enough, just return the
+	 * required array size.
+	 */
+	if (get_state->bo_count < state->bo_count) {
+		get_state->bo_count = state->bo_count;
+		spin_unlock_irqrestore(&vc4->job_lock, irqflags);
+		return 0;
+	}
+
+	vc4->hang_state = NULL;
+	spin_unlock_irqrestore(&vc4->job_lock, irqflags);
+
+	/* Save the user's BO pointer, so we don't stomp it with the memcpy. */
+	state->bo = get_state->bo;
+	memcpy(get_state, state, sizeof(*state));
+
+	bo_state = kcalloc(state->bo_count, sizeof(*bo_state), GFP_KERNEL);
+	if (!bo_state) {
+		ret = -ENOMEM;
+		goto err_free;
+	}
+
+	for (i = 0; i < state->bo_count; i++) {
+		struct vc4_bo *vc4_bo = to_vc4_bo(kernel_state->bo[i]);
+		u32 handle;
+		ret = drm_gem_handle_create(file_priv, kernel_state->bo[i],
+					    &handle);
+
+		if (ret) {
+			state->bo_count = i - 1;
+			goto err;
+		}
+		bo_state[i].handle = handle;
+		bo_state[i].paddr = vc4_bo->base.paddr;
+		bo_state[i].size = vc4_bo->base.base.size;
+	}
+
+	ret = copy_to_user((void __user *)(uintptr_t)get_state->bo,
+			   bo_state,
+			   state->bo_count * sizeof(*bo_state));
+	kfree(bo_state);
+
+ err_free:
+
+	vc4_free_hang_state(dev, kernel_state);
+
+err:
+	return ret;
+}
+
+static void
+vc4_save_hang_state(struct drm_device *dev)
+{
+	struct vc4_dev *vc4 = to_vc4_dev(dev);
+	struct drm_vc4_get_hang_state *state;
+	struct vc4_hang_state *kernel_state;
+	struct vc4_exec_info *exec;
+	struct vc4_bo *bo;
+	unsigned long irqflags;
+	unsigned int i, unref_list_count;
+
+	kernel_state = kcalloc(1, sizeof(*state), GFP_KERNEL);
+	if (!kernel_state)
+		return;
+
+	state = &kernel_state->user_state;
+
+	spin_lock_irqsave(&vc4->job_lock, irqflags);
+	exec = vc4_first_job(vc4);
+	if (!exec) {
+		spin_unlock_irqrestore(&vc4->job_lock, irqflags);
+		return;
+	}
+
+	unref_list_count = 0;
+	list_for_each_entry(bo, &exec->unref_list, unref_head)
+		unref_list_count++;
+
+	state->bo_count = exec->bo_count + unref_list_count;
+	kernel_state->bo = kcalloc(state->bo_count, sizeof(*kernel_state->bo),
+				   GFP_ATOMIC);
+	if (!kernel_state->bo) {
+		spin_unlock_irqrestore(&vc4->job_lock, irqflags);
+		return;
+	}
+
+	for (i = 0; i < exec->bo_count; i++) {
+		drm_gem_object_reference(&exec->bo[i].bo->base);
+		kernel_state->bo[i] = &exec->bo[i].bo->base;
+	}
+
+	list_for_each_entry(bo, &exec->unref_list, unref_head) {
+		drm_gem_object_reference(&bo->base.base);
+		kernel_state->bo[i] = &bo->base.base;
+		i++;
+	}
+
+	state->start_bin = exec->ct0ca;
+	state->start_render = exec->ct1ca;
+
+	spin_unlock_irqrestore(&vc4->job_lock, irqflags);
+
+	state->ct0ca = V3D_READ(V3D_CTNCA(0));
+	state->ct0ea = V3D_READ(V3D_CTNEA(0));
+
+	state->ct1ca = V3D_READ(V3D_CTNCA(1));
+	state->ct1ea = V3D_READ(V3D_CTNEA(1));
+
+	state->ct0cs = V3D_READ(V3D_CTNCS(0));
+	state->ct1cs = V3D_READ(V3D_CTNCS(1));
+
+	state->ct0ra0 = V3D_READ(V3D_CT00RA0);
+	state->ct1ra0 = V3D_READ(V3D_CT01RA0);
+
+	state->bpca = V3D_READ(V3D_BPCA);
+	state->bpcs = V3D_READ(V3D_BPCS);
+	state->bpoa = V3D_READ(V3D_BPOA);
+	state->bpos = V3D_READ(V3D_BPOS);
+
+	state->vpmbase = V3D_READ(V3D_VPMBASE);
+
+	state->dbge = V3D_READ(V3D_DBGE);
+	state->fdbgo = V3D_READ(V3D_FDBGO);
+	state->fdbgb = V3D_READ(V3D_FDBGB);
+	state->fdbgr = V3D_READ(V3D_FDBGR);
+	state->fdbgs = V3D_READ(V3D_FDBGS);
+	state->errstat = V3D_READ(V3D_ERRSTAT);
+
+	spin_lock_irqsave(&vc4->job_lock, irqflags);
+	if (vc4->hang_state) {
+		spin_unlock_irqrestore(&vc4->job_lock, irqflags);
+		vc4_free_hang_state(dev, kernel_state);
+	} else {
+		vc4->hang_state = kernel_state;
+		spin_unlock_irqrestore(&vc4->job_lock, irqflags);
+	}
+}
+
 static void
 vc4_reset(struct drm_device *dev)
 {
@@ -64,6 +244,8 @@ vc4_reset_work(struct work_struct *work)
 	struct vc4_dev *vc4 =
 		container_of(work, struct vc4_dev, hangcheck.reset_work);
 
+	vc4_save_hang_state(vc4->dev);
+
 	vc4_reset(vc4->dev);
 }
 
@@ -673,4 +855,7 @@ vc4_gem_destroy(struct drm_device *dev)
 	}
 
 	vc4_bo_cache_destroy(dev);
+
+	if (vc4->hang_state)
+		vc4_free_hang_state(dev, vc4->hang_state);
 }
--- a/include/uapi/drm/vc4_drm.h
+++ b/include/uapi/drm/vc4_drm.h
@@ -32,6 +32,7 @@
 #define DRM_VC4_CREATE_BO                         0x03
 #define DRM_VC4_MMAP_BO                           0x04
 #define DRM_VC4_CREATE_SHADER_BO                  0x05
+#define DRM_VC4_GET_HANG_STATE                    0x06
 
 #define DRM_IOCTL_VC4_SUBMIT_CL           DRM_IOWR( DRM_COMMAND_BASE + DRM_VC4_SUBMIT_CL, struct drm_vc4_submit_cl)
 #define DRM_IOCTL_VC4_WAIT_SEQNO          DRM_IOWR( DRM_COMMAND_BASE + DRM_VC4_WAIT_SEQNO, struct drm_vc4_wait_seqno)
@@ -39,6 +40,7 @@
 #define DRM_IOCTL_VC4_CREATE_BO           DRM_IOWR( DRM_COMMAND_BASE + DRM_VC4_CREATE_BO, struct drm_vc4_create_bo)
 #define DRM_IOCTL_VC4_MMAP_BO             DRM_IOWR( DRM_COMMAND_BASE + DRM_VC4_MMAP_BO, struct drm_vc4_mmap_bo)
 #define DRM_IOCTL_VC4_CREATE_SHADER_BO    DRM_IOWR( DRM_COMMAND_BASE + DRM_VC4_CREATE_SHADER_BO, struct drm_vc4_create_shader_bo)
+#define DRM_IOCTL_VC4_GET_HANG_STATE      DRM_IOWR( DRM_COMMAND_BASE + DRM_VC4_GET_HANG_STATE, struct drm_vc4_get_hang_state)
 
 struct drm_vc4_submit_rcl_surface {
 	uint32_t hindex; /* Handle index, or ~0 if not present. */
@@ -226,4 +228,47 @@ struct drm_vc4_mmap_bo {
 	uint64_t offset;
 };
 
+struct drm_vc4_get_hang_state_bo {
+	uint32_t handle;
+	uint32_t paddr;
+	uint32_t size;
+	uint32_t pad;
+};
+
+/**
+ * struct drm_vc4_hang_state - ioctl argument for collecting state
+ * from a GPU hang for analysis.
+*/
+struct drm_vc4_get_hang_state {
+	/** Pointer to array of struct drm_vc4_get_hang_state_bo. */
+	uint64_t bo;
+	/**
+	 * On input, the size of the bo array.  Output is the number
+	 * of bos to be returned.
+	 */
+	uint32_t bo_count;
+
+	uint32_t start_bin, start_render;
+
+	uint32_t ct0ca, ct0ea;
+	uint32_t ct1ca, ct1ea;
+	uint32_t ct0cs, ct1cs;
+	uint32_t ct0ra0, ct1ra0;
+
+	uint32_t bpca, bpcs;
+	uint32_t bpoa, bpos;
+
+	uint32_t vpmbase;
+
+	uint32_t dbge;
+	uint32_t fdbgo;
+	uint32_t fdbgb;
+	uint32_t fdbgr;
+	uint32_t fdbgs;
+	uint32_t errstat;
+
+	/* Pad that we may save more registers into in the future. */
+	uint32_t pad[16];
+};
+
 #endif /* _UAPI_VC4_DRM_H_ */