drm/i915: Refine tracepoints

A lot of minor tweaks to fix the tracepoints, improve the outputting for
ftrace, and to generally make the tracepoints useful again. It is a start
and enough to begin identifying performance issues and gaps in our
coverage.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

drivers/gpu/drm/i915/i915_dma.c

@@ -2004,7 +2004,6 @@ int i915_driver_load(struct drm_device *dev, unsigned long flags)
 
 	spin_lock_init(&dev_priv->irq_lock);
 	spin_lock_init(&dev_priv->error_lock);
-	dev_priv->trace_irq_seqno = 0;
 
 	ret = drm_vblank_init(dev, I915_NUM_PIPE);
 	if (ret)

drivers/gpu/drm/i915/i915_drv.h

@@ -76,10 +76,7 @@ enum plane {
 #define DRIVER_PATCHLEVEL	0
 
 #define WATCH_COHERENCY	0
-#define WATCH_EXEC	0
-#define WATCH_RELOC	0
 #define WATCH_LISTS	0
-#define WATCH_PWRITE	0
 
 #define I915_GEM_PHYS_CURSOR_0 1
 #define I915_GEM_PHYS_CURSOR_1 2
@@ -289,7 +286,6 @@ typedef struct drm_i915_private {
 	int page_flipping;
 
 	atomic_t irq_received;
-	u32 trace_irq_seqno;
 
 	/* protects the irq masks */
 	spinlock_t irq_lock;
@@ -1001,7 +997,6 @@ extern int i915_irq_emit(struct drm_device *dev, void *data,
 			 struct drm_file *file_priv);
 extern int i915_irq_wait(struct drm_device *dev, void *data,
 			 struct drm_file *file_priv);
-void i915_trace_irq_get(struct drm_device *dev, u32 seqno);
 
 extern irqreturn_t i915_driver_irq_handler(DRM_IRQ_ARGS);
 extern void i915_driver_irq_preinstall(struct drm_device * dev);
@@ -1095,8 +1090,7 @@ int i915_gem_get_aperture_ioctl(struct drm_device *dev, void *data,
 				struct drm_file *file_priv);
 void i915_gem_load(struct drm_device *dev);
 int i915_gem_init_object(struct drm_gem_object *obj);
-int __must_check i915_gem_flush_ring(struct drm_device *dev,
-				     struct intel_ring_buffer *ring,
+int __must_check i915_gem_flush_ring(struct intel_ring_buffer *ring,
 				     uint32_t invalidate_domains,
 				     uint32_t flush_domains);
 struct drm_i915_gem_object *i915_gem_alloc_object(struct drm_device *dev,
@@ -1127,10 +1121,9 @@ i915_seqno_passed(uint32_t seq1, uint32_t seq2)
 }
 
 static inline u32
-i915_gem_next_request_seqno(struct drm_device *dev,
-			    struct intel_ring_buffer *ring)
+i915_gem_next_request_seqno(struct intel_ring_buffer *ring)
 {
-	drm_i915_private_t *dev_priv = dev->dev_private;
+	drm_i915_private_t *dev_priv = ring->dev->dev_private;
 	return ring->outstanding_lazy_request = dev_priv->next_seqno;
 }
 
@@ -1155,14 +1148,12 @@ void i915_gem_do_init(struct drm_device *dev,
 		      unsigned long end);
 int __must_check i915_gpu_idle(struct drm_device *dev);
 int __must_check i915_gem_idle(struct drm_device *dev);
-int __must_check i915_add_request(struct drm_device *dev,
-				  struct drm_file *file_priv,
-				  struct drm_i915_gem_request *request,
-				  struct intel_ring_buffer *ring);
-int __must_check i915_do_wait_request(struct drm_device *dev,
-				      uint32_t seqno,
-				      bool interruptible,
-				      struct intel_ring_buffer *ring);
+int __must_check i915_add_request(struct intel_ring_buffer *ring,
+				  struct drm_file *file,
+				  struct drm_i915_gem_request *request);
+int __must_check i915_wait_request(struct intel_ring_buffer *ring,
+				   uint32_t seqno,
+				   bool interruptible);
 int i915_gem_fault(struct vm_area_struct *vma, struct vm_fault *vmf);
 int __must_check
 i915_gem_object_set_to_gtt_domain(struct drm_i915_gem_object *obj,
@@ -1311,7 +1302,7 @@ extern void intel_display_print_error_state(struct seq_file *m,
 #define __i915_read(x, y) \
 static inline u##x i915_read##x(struct drm_i915_private *dev_priv, u32 reg) { \
 	u##x val = read##y(dev_priv->regs + reg); \
-	trace_i915_reg_rw('R', reg, val, sizeof(val)); \
+	trace_i915_reg_rw(false, reg, val, sizeof(val)); \
 	return val; \
 }
 __i915_read(8, b)
@@ -1322,7 +1313,7 @@ __i915_read(64, q)
 
 #define __i915_write(x, y) \
 static inline void i915_write##x(struct drm_i915_private *dev_priv, u32 reg, u##x val) { \
-	trace_i915_reg_rw('W', reg, val, sizeof(val)); \
+	trace_i915_reg_rw(true, reg, val, sizeof(val)); \
 	write##y(val, dev_priv->regs + reg); \
 }
 __i915_write(8, b)
@@ -1371,25 +1362,4 @@ static inline u32 i915_safe_read(struct drm_i915_private *dev_priv, u32 reg)
 	return val;
 }
 
-static inline void
-i915_write(struct drm_i915_private *dev_priv, u32 reg, u64 val, int len)
-{
-       /* Trace down the write operation before the real write */
-       trace_i915_reg_rw('W', reg, val, len);
-       switch (len) {
-       case 8:
-               writeq(val, dev_priv->regs + reg);
-               break;
-       case 4:
-               writel(val, dev_priv->regs + reg);
-               break;
-       case 2:
-               writew(val, dev_priv->regs + reg);
-               break;
-       case 1:
-               writeb(val, dev_priv->regs + reg);
-               break;
-       }
-}
-
 #endif

drivers/gpu/drm/i915/i915_gem_debug.c

@@ -134,51 +134,6 @@ i915_verify_lists(struct drm_device *dev)
 }
 #endif /* WATCH_INACTIVE */
 
-
-#if WATCH_EXEC | WATCH_PWRITE
-static void
-i915_gem_dump_page(struct page *page, uint32_t start, uint32_t end,
-		   uint32_t bias, uint32_t mark)
-{
-	uint32_t *mem = kmap_atomic(page, KM_USER0);
-	int i;
-	for (i = start; i < end; i += 4)
-		DRM_INFO("%08x: %08x%s\n",
-			  (int) (bias + i), mem[i / 4],
-			  (bias + i == mark) ? " ********" : "");
-	kunmap_atomic(mem, KM_USER0);
-	/* give syslog time to catch up */
-	msleep(1);
-}
-
-void
-i915_gem_dump_object(struct drm_i915_gem_object *obj, int len,
-		     const char *where, uint32_t mark)
-{
-	int page;
-
-	DRM_INFO("%s: object at offset %08x\n", where, obj->gtt_offset);
-	for (page = 0; page < (len + PAGE_SIZE-1) / PAGE_SIZE; page++) {
-		int page_len, chunk, chunk_len;
-
-		page_len = len - page * PAGE_SIZE;
-		if (page_len > PAGE_SIZE)
-			page_len = PAGE_SIZE;
-
-		for (chunk = 0; chunk < page_len; chunk += 128) {
-			chunk_len = page_len - chunk;
-			if (chunk_len > 128)
-				chunk_len = 128;
-			i915_gem_dump_page(obj->pages[page],
-					   chunk, chunk + chunk_len,
-					   obj->gtt_offset +
-					   page * PAGE_SIZE,
-					   mark);
-		}
-	}
-}
-#endif
-
 #if WATCH_COHERENCY
 void
 i915_gem_object_check_coherency(struct drm_i915_gem_object *obj, int handle)

drivers/gpu/drm/i915/i915_gem_evict.c

@@ -30,6 +30,7 @@
 #include "drm.h"
 #include "i915_drv.h"
 #include "i915_drm.h"
+#include "i915_trace.h"
 
 static bool
 mark_free(struct drm_i915_gem_object *obj, struct list_head *unwind)
@@ -63,6 +64,8 @@ i915_gem_evict_something(struct drm_device *dev, int min_size,
 			return 0;
 	}
 
+	trace_i915_gem_evict(dev, min_size, alignment, mappable);
+
 	/*
 	 * The goal is to evict objects and amalgamate space in LRU order.
 	 * The oldest idle objects reside on the inactive list, which is in
@@ -189,6 +192,8 @@ i915_gem_evict_everything(struct drm_device *dev, bool purgeable_only)
 	if (lists_empty)
 		return -ENOSPC;
 
+	trace_i915_gem_evict_everything(dev, purgeable_only);
+
 	/* Flush everything (on to the inactive lists) and evict */
 	ret = i915_gpu_idle(dev);
 	if (ret)

drivers/gpu/drm/i915/i915_gem_execbuffer.c

@@ -282,21 +282,6 @@ i915_gem_execbuffer_relocate_entry(struct drm_i915_gem_object *obj,
 
 	target_offset = to_intel_bo(target_obj)->gtt_offset;
 
-#if WATCH_RELOC
-	DRM_INFO("%s: obj %p offset %08x target %d "
-		 "read %08x write %08x gtt %08x "
-		 "presumed %08x delta %08x\n",
-		 __func__,
-		 obj,
-		 (int) reloc->offset,
-		 (int) reloc->target_handle,
-		 (int) reloc->read_domains,
-		 (int) reloc->write_domain,
-		 (int) target_offset,
-		 (int) reloc->presumed_offset,
-		 reloc->delta);
-#endif
-
 	/* The target buffer should have appeared before us in the
 	 * exec_object list, so it should have a GTT space bound by now.
 	 */
@@ -747,8 +732,7 @@ i915_gem_execbuffer_flush(struct drm_device *dev,
 	if ((flush_domains | invalidate_domains) & I915_GEM_GPU_DOMAINS) {
 		for (i = 0; i < I915_NUM_RINGS; i++)
 			if (flush_rings & (1 << i)) {
-				ret = i915_gem_flush_ring(dev,
-							  &dev_priv->ring[i],
+				ret = i915_gem_flush_ring(&dev_priv->ring[i],
 							  invalidate_domains,
 							  flush_domains);
 				if (ret)
@@ -787,7 +771,7 @@ i915_gem_execbuffer_sync_rings(struct drm_i915_gem_object *obj,
 		if (request == NULL)
 			return -ENOMEM;
 
-		ret = i915_add_request(obj->base.dev, NULL, request, from);
+		ret = i915_add_request(from, NULL, request);
 		if (ret) {
 			kfree(request);
 			return ret;
@@ -815,12 +799,6 @@ i915_gem_execbuffer_move_to_gpu(struct intel_ring_buffer *ring,
 		i915_gem_object_set_to_gpu_domain(obj, ring, &cd);
 
 	if (cd.invalidate_domains | cd.flush_domains) {
-#if WATCH_EXEC
-		DRM_INFO("%s: invalidate_domains %08x flush_domains %08x\n",
-			  __func__,
-			 cd.invalidate_domains,
-			 cd.flush_domains);
-#endif
 		ret = i915_gem_execbuffer_flush(ring->dev,
 						cd.invalidate_domains,
 						cd.flush_domains,
@@ -924,6 +902,10 @@ i915_gem_execbuffer_move_to_active(struct list_head *objects,
 	struct drm_i915_gem_object *obj;
 
 	list_for_each_entry(obj, objects, exec_list) {
+		  u32 old_read = obj->base.read_domains;
+		  u32 old_write = obj->base.write_domain;
+
+
 		obj->base.read_domains = obj->base.pending_read_domains;
 		obj->base.write_domain = obj->base.pending_write_domain;
 		obj->fenced_gpu_access = obj->pending_fenced_gpu_access;
@@ -937,9 +919,7 @@ i915_gem_execbuffer_move_to_active(struct list_head *objects,
 			intel_mark_busy(ring->dev, obj);
 		}
 
-		trace_i915_gem_object_change_domain(obj,
-						    obj->base.read_domains,
-						    obj->base.write_domain);
+		trace_i915_gem_object_change_domain(obj, old_read, old_write);
 	}
 }
 
@@ -961,14 +941,14 @@ i915_gem_execbuffer_retire_commands(struct drm_device *dev,
 	if (INTEL_INFO(dev)->gen >= 4)
 		invalidate |= I915_GEM_DOMAIN_SAMPLER;
 	if (ring->flush(ring, invalidate, 0)) {
-		i915_gem_next_request_seqno(dev, ring);
+		i915_gem_next_request_seqno(ring);
 		return;
 	}
 
 	/* Add a breadcrumb for the completion of the batch buffer */
 	request = kzalloc(sizeof(*request), GFP_KERNEL);
-	if (request == NULL || i915_add_request(dev, file, request, ring)) {
-		i915_gem_next_request_seqno(dev, ring);
+	if (request == NULL || i915_add_request(ring, file, request)) {
+		i915_gem_next_request_seqno(ring);
 		kfree(request);
 	}
 }
@@ -998,10 +978,6 @@ i915_gem_do_execbuffer(struct drm_device *dev, void *data,
 	if (ret)
 		return ret;
 
-#if WATCH_EXEC
-	DRM_INFO("buffers_ptr %d buffer_count %d len %08x\n",
-		  (int) args->buffers_ptr, args->buffer_count, args->batch_len);
-#endif
 	switch (args->flags & I915_EXEC_RING_MASK) {
 	case I915_EXEC_DEFAULT:
 	case I915_EXEC_RENDER:
@@ -1172,7 +1148,7 @@ i915_gem_do_execbuffer(struct drm_device *dev, void *data,
 	if (ret)
 		goto err;
 
-	seqno = i915_gem_next_request_seqno(dev, ring);
+	seqno = i915_gem_next_request_seqno(ring);
 	for (i = 0; i < ARRAY_SIZE(ring->sync_seqno); i++) {
 		if (seqno < ring->sync_seqno[i]) {
 			/* The GPU can not handle its semaphore value wrapping,
@@ -1187,6 +1163,8 @@ i915_gem_do_execbuffer(struct drm_device *dev, void *data,
 		}
 	}
 
+	trace_i915_gem_ring_dispatch(ring, seqno);
+
 	exec_start = batch_obj->gtt_offset + args->batch_start_offset;
 	exec_len = args->batch_len;
 	if (cliprects) {
@@ -1243,11 +1221,6 @@ i915_gem_execbuffer(struct drm_device *dev, void *data,
 	struct drm_i915_gem_exec_object2 *exec2_list = NULL;
 	int ret, i;
 
-#if WATCH_EXEC
-	DRM_INFO("buffers_ptr %d buffer_count %d len %08x\n",
-		  (int) args->buffers_ptr, args->buffer_count, args->batch_len);
-#endif
-
 	if (args->buffer_count < 1) {
 		DRM_ERROR("execbuf with %d buffers\n", args->buffer_count);
 		return -EINVAL;
@@ -1328,11 +1301,6 @@ i915_gem_execbuffer2(struct drm_device *dev, void *data,
 	struct drm_i915_gem_exec_object2 *exec2_list = NULL;
 	int ret;
 
-#if WATCH_EXEC
-	DRM_INFO("buffers_ptr %d buffer_count %d len %08x\n",
-		  (int) args->buffers_ptr, args->buffer_count, args->batch_len);
-#endif
-
 	if (args->buffer_count < 1) {
 		DRM_ERROR("execbuf2 with %d buffers\n", args->buffer_count);
 		return -EINVAL;

drivers/gpu/drm/i915/i915_irq.c

@@ -365,7 +365,7 @@ static void notify_ring(struct drm_device *dev,
 		return;
 
 	seqno = ring->get_seqno(ring);
-	trace_i915_gem_request_complete(dev, seqno);
+	trace_i915_gem_request_complete(ring, seqno);
 
 	ring->irq_seqno = seqno;
 	wake_up_all(&ring->irq_queue);
@@ -1273,16 +1273,6 @@ static int i915_emit_irq(struct drm_device * dev)
 	return dev_priv->counter;
 }
 
-void i915_trace_irq_get(struct drm_device *dev, u32 seqno)
-{
-	drm_i915_private_t *dev_priv = (drm_i915_private_t *) dev->dev_private;
-	struct intel_ring_buffer *ring = LP_RING(dev_priv);
-
-	if (dev_priv->trace_irq_seqno == 0 &&
-	    ring->irq_get(ring))
-		dev_priv->trace_irq_seqno = seqno;
-}
-
 static int i915_wait_irq(struct drm_device * dev, int irq_nr)
 {
 	drm_i915_private_t *dev_priv = (drm_i915_private_t *) dev->dev_private;

drivers/gpu/drm/i915/intel_overlay.c

@@ -221,16 +221,15 @@ static int intel_overlay_do_wait_request(struct intel_overlay *overlay,
 	int ret;
 
 	BUG_ON(overlay->last_flip_req);
-	ret = i915_add_request(dev, NULL, request, LP_RING(dev_priv));
+	ret = i915_add_request(LP_RING(dev_priv), NULL, request);
 	if (ret) {
 	    kfree(request);
 	    return ret;
 	}
 	overlay->last_flip_req = request->seqno;
 	overlay->flip_tail = tail;
-	ret = i915_do_wait_request(dev,
-				   overlay->last_flip_req, true,
-				   LP_RING(dev_priv));
+	ret = i915_wait_request(LP_RING(dev_priv),
+				overlay->last_flip_req, true);
 	if (ret)
 		return ret;
 
@@ -364,7 +363,7 @@ static int intel_overlay_continue(struct intel_overlay *overlay,
 	OUT_RING(flip_addr);
         ADVANCE_LP_RING();
 
-	ret = i915_add_request(dev, NULL, request, LP_RING(dev_priv));
+	ret = i915_add_request(LP_RING(dev_priv), NULL, request);
 	if (ret) {
 		kfree(request);
 		return ret;
@@ -453,8 +452,8 @@ static int intel_overlay_recover_from_interrupt(struct intel_overlay *overlay,
 	if (overlay->last_flip_req == 0)
 		return 0;
 
-	ret = i915_do_wait_request(dev, overlay->last_flip_req,
-				   interruptible, LP_RING(dev_priv));
+	ret = i915_wait_request(LP_RING(dev_priv),
+				overlay->last_flip_req, interruptible);
 	if (ret)
 		return ret;
 

drivers/gpu/drm/i915/intel_ringbuffer.c

@@ -62,18 +62,9 @@ render_ring_flush(struct intel_ring_buffer *ring,
 		  u32	flush_domains)
 {
 	struct drm_device *dev = ring->dev;
-	drm_i915_private_t *dev_priv = dev->dev_private;
 	u32 cmd;
 	int ret;
 
-#if WATCH_EXEC
-	DRM_INFO("%s: invalidate %08x flush %08x\n", __func__,
-		  invalidate_domains, flush_domains);
-#endif
-
-	trace_i915_gem_request_flush(dev, dev_priv->next_seqno,
-				     invalidate_domains, flush_domains);
-
 	if ((invalidate_domains | flush_domains) & I915_GEM_GPU_DOMAINS) {
 		/*
 		 * read/write caches:
@@ -122,9 +113,6 @@ render_ring_flush(struct intel_ring_buffer *ring,
 		    (IS_G4X(dev) || IS_GEN5(dev)))
 			cmd |= MI_INVALIDATE_ISP;
 
-#if WATCH_EXEC
-		DRM_INFO("%s: queue flush %08x to ring\n", __func__, cmd);
-#endif
 		ret = intel_ring_begin(ring, 2);
 		if (ret)
 			return ret;
@@ -714,11 +702,8 @@ render_ring_dispatch_execbuffer(struct intel_ring_buffer *ring,
 				u32 offset, u32 len)
 {
 	struct drm_device *dev = ring->dev;
-	drm_i915_private_t *dev_priv = dev->dev_private;
 	int ret;
 
-	trace_i915_gem_request_submit(dev, dev_priv->next_seqno + 1);
-
 	if (IS_I830(dev) || IS_845G(dev)) {
 		ret = intel_ring_begin(ring, 4);
 		if (ret)
@@ -953,13 +938,13 @@ int intel_wait_ring_buffer(struct intel_ring_buffer *ring, int n)
 			return 0;
 	}
 
-	trace_i915_ring_wait_begin (dev);
+	trace_i915_ring_wait_begin(ring);
 	end = jiffies + 3 * HZ;
 	do {
 		ring->head = I915_READ_HEAD(ring);
 		ring->space = ring_space(ring);
 		if (ring->space >= n) {
-			trace_i915_ring_wait_end(dev);
+			trace_i915_ring_wait_end(ring);
 			return 0;
 		}
 
@@ -973,7 +958,7 @@ int intel_wait_ring_buffer(struct intel_ring_buffer *ring, int n)
 		if (atomic_read(&dev_priv->mm.wedged))
 			return -EAGAIN;
 	} while (!time_after(jiffies, end));
-	trace_i915_ring_wait_end (dev);
+	trace_i915_ring_wait_end(ring);
 	return -EBUSY;
 }
 

drivers/gpu/drm/i915/intel_ringbuffer.h

@@ -58,6 +58,7 @@ struct  intel_ring_buffer {
 	u32		irq_refcount;
 	u32		irq_mask;
 	u32		irq_seqno;		/* last seq seem at irq time */
+	u32		trace_irq_seqno;
 	u32		waiting_seqno;
 	u32		sync_seqno[I915_NUM_RINGS-1];
 	bool __must_check (*irq_get)(struct intel_ring_buffer *ring);
@@ -186,6 +187,12 @@ int intel_init_blt_ring_buffer(struct drm_device *dev);
 u32 intel_ring_get_active_head(struct intel_ring_buffer *ring);
 void intel_ring_setup_status_page(struct intel_ring_buffer *ring);
 
+static inline void i915_trace_irq_get(struct intel_ring_buffer *ring, u32 seqno)
+{
+	if (ring->trace_irq_seqno == 0 && ring->irq_get(ring))
+		ring->trace_irq_seqno = seqno;
+}
+
 /* DRI warts */
 int intel_render_ring_init_dri(struct drm_device *dev, u64 start, u32 size);