intel: Add support for kernel multi-ringbuffer API.

This introduces a new API to exec on BSD ring buffer, for H.264 VLD
decoding.

Signed-off-by: Xiang Hai hao <haihao.xiang@intel.com>
Signed-off-by: Zou Nan hai <nanhai.zou@intel.com>

include/drm/i915_drm.h

@@ -275,6 +275,7 @@ typedef struct drm_i915_irq_wait {
 #define I915_PARAM_HAS_OVERLAY           7
 #define I915_PARAM_HAS_PAGEFLIPPING	 8
 #define I915_PARAM_HAS_EXECBUF2          9
+#define I915_PARAM_HAS_BSD		 10
 
 typedef struct drm_i915_getparam {
 	int param;
@@ -616,7 +617,9 @@ struct drm_i915_gem_execbuffer2 {
 	__u32 num_cliprects;
 	/** This is a struct drm_clip_rect *cliprects */
 	__u64 cliprects_ptr;
-	__u64 flags; /* currently unused */
+#define I915_EXEC_RENDER		(1 << 0)
+#define I915_EXEC_BSD			(1 << 1)
+	__u64 flags;
 	__u64 rsvd1;
 	__u64 rsvd2;
 };

intel/intel_bufmgr.c

@@ -145,6 +145,19 @@ drm_intel_bo_exec(drm_intel_bo *bo, int used,
 	return bo->bufmgr->bo_exec(bo, used, cliprects, num_cliprects, DR4);
 }
 
+int
+drm_intel_bo_mrb_exec(drm_intel_bo *bo, int used,
+		drm_clip_rect_t *cliprects, int num_cliprects, int DR4,
+		int ring_flag)
+{
+	if (bo->bufmgr->bo_mrb_exec)
+		return bo->bufmgr->bo_mrb_exec(bo, used,
+					cliprects, num_cliprects, DR4,
+					ring_flag);
+
+	return -ENODEV;
+}
+
 void drm_intel_bufmgr_set_debug(drm_intel_bufmgr *bufmgr, int enable_debug)
 {
 	bufmgr->debug = enable_debug;

intel/intel_bufmgr.h

@@ -106,6 +106,9 @@ void drm_intel_bufmgr_set_debug(drm_intel_bufmgr *bufmgr, int enable_debug);
 void drm_intel_bufmgr_destroy(drm_intel_bufmgr *bufmgr);
 int drm_intel_bo_exec(drm_intel_bo *bo, int used,
 		      drm_clip_rect_t * cliprects, int num_cliprects, int DR4);
+int drm_intel_bo_mrb_exec(drm_intel_bo *bo, int used,
+			drm_clip_rect_t *cliprects, int num_cliprects, int DR4,
+			int ring_flag);
 int drm_intel_bufmgr_check_aperture_space(drm_intel_bo ** bo_array, int count);
 
 int drm_intel_bo_emit_reloc(drm_intel_bo *bo, uint32_t offset,

intel/intel_bufmgr_gem.c

@@ -1530,14 +1530,17 @@ drm_intel_gem_bo_exec(drm_intel_bo *bo, int used,
 }
 
 static int
-drm_intel_gem_bo_exec2(drm_intel_bo *bo, int used,
-		       drm_clip_rect_t *cliprects, int num_cliprects,
-		       int DR4)
+drm_intel_gem_bo_mrb_exec2(drm_intel_bo *bo, int used,
+			drm_clip_rect_t *cliprects, int num_cliprects, int DR4,
+			int ring_flag)
 {
 	drm_intel_bufmgr_gem *bufmgr_gem = (drm_intel_bufmgr_gem *)bo->bufmgr;
 	struct drm_i915_gem_execbuffer2 execbuf;
 	int ret, i;
 
+	if ((ring_flag != I915_EXEC_RENDER) && (ring_flag != I915_EXEC_BSD))
+		return -EINVAL;
+
 	pthread_mutex_lock(&bufmgr_gem->lock);
 	/* Update indices and set up the validate list. */
 	drm_intel_gem_bo_process_reloc2(bo);
@@ -1555,7 +1558,7 @@ drm_intel_gem_bo_exec2(drm_intel_bo *bo, int used,
 	execbuf.num_cliprects = num_cliprects;
 	execbuf.DR1 = 0;
 	execbuf.DR4 = DR4;
-	execbuf.flags = 0;
+	execbuf.flags = ring_flag;
 	execbuf.rsvd1 = 0;
 	execbuf.rsvd2 = 0;
 
@@ -1596,6 +1599,16 @@ drm_intel_gem_bo_exec2(drm_intel_bo *bo, int used,
 	return ret;
 }
 
+static int
+drm_intel_gem_bo_exec2(drm_intel_bo *bo, int used,
+		       drm_clip_rect_t *cliprects, int num_cliprects,
+		       int DR4)
+{
+	return drm_intel_gem_bo_mrb_exec2(bo, used,
+					cliprects, num_cliprects, DR4,
+					I915_EXEC_RENDER);
+}
+
 static int
 drm_intel_gem_bo_pin(drm_intel_bo *bo, uint32_t alignment)
 {
@@ -1974,7 +1987,7 @@ drm_intel_bufmgr_gem_init(int fd, int batch_size)
 	drm_i915_getparam_t gp;
 	int ret, i;
 	unsigned long size;
-	int exec2 = 0;
+	int exec2 = 0, has_bsd = 0;
 
 	bufmgr_gem = calloc(1, sizeof(*bufmgr_gem));
 	if (bufmgr_gem == NULL)
@@ -2023,6 +2036,11 @@ drm_intel_bufmgr_gem_init(int fd, int batch_size)
 	if (!ret)
 		exec2 = 1;
 
+	gp.param = I915_PARAM_HAS_BSD;
+	ret = ioctl(bufmgr_gem->fd, DRM_IOCTL_I915_GETPARAM, &gp);
+	if (!ret)
+		has_bsd = 1;
+
 	if (bufmgr_gem->gen < 4) {
 		gp.param = I915_PARAM_NUM_FENCES_AVAIL;
 		gp.value = &bufmgr_gem->available_fences;
@@ -2076,9 +2094,11 @@ drm_intel_bufmgr_gem_init(int fd, int batch_size)
 	bufmgr_gem->bufmgr.bo_set_tiling = drm_intel_gem_bo_set_tiling;
 	bufmgr_gem->bufmgr.bo_flink = drm_intel_gem_bo_flink;
 	/* Use the new one if available */
-	if (exec2)
+	if (exec2) {
 		bufmgr_gem->bufmgr.bo_exec = drm_intel_gem_bo_exec2;
-	else
+		if (has_bsd)
+			bufmgr_gem->bufmgr.bo_mrb_exec = drm_intel_gem_bo_mrb_exec2;
+	} else
 		bufmgr_gem->bufmgr.bo_exec = drm_intel_gem_bo_exec;
 	bufmgr_gem->bufmgr.bo_busy = drm_intel_gem_bo_busy;
 	bufmgr_gem->bufmgr.bo_madvise = drm_intel_gem_bo_madvise;

intel/intel_bufmgr_priv.h

@@ -173,6 +173,13 @@ struct _drm_intel_bufmgr {
 			drm_clip_rect_t *cliprects, int num_cliprects,
 			int DR4);
 
+	/** Executes the command buffer pointed to by bo on the selected
+	 * ring buffer
+	 */
+	int (*bo_mrb_exec) (drm_intel_bo *bo, int used,
+			drm_clip_rect_t *cliprects, int num_cliprects,
+			int DR4, int ring_flag);
+
 	/**
 	 * Pin a buffer to the aperture and fix the offset until unpinned
 	 *