drm/vc4: add tracepoints for CL submissions

Trace submit_cl_ioctl and related IRQs for CL submission and bin/render
jobs execution. It might be helpful to get a rendering timeline and
track job throttling.

Signed-off-by: Melissa Wen <mwen@igalia.com>
Reviewed-by: Jose Maria Casanova Crespo <jmcasanova@igalia.com>
Signed-off-by: Maxime Ripard <maxime@cerno.tech>
Link: https://patchwork.freedesktop.org/patch/msgid/20220201212651.zhltjmaokisffq3x@mail.igalia.com

drivers/gpu/drm/vc4/vc4_gem.c

@@ -485,6 +485,8 @@ again:
 	 * immediately move it to the to-be-rendered queue.
 	 */
 	if (exec->ct0ca != exec->ct0ea) {
+		trace_vc4_submit_cl(dev, false, exec->seqno, exec->ct0ca,
+				    exec->ct0ea);
 		submit_cl(dev, 0, exec->ct0ca, exec->ct0ea);
 	} else {
 		struct vc4_exec_info *next;
@@ -519,6 +521,7 @@ vc4_submit_next_render_job(struct drm_device *dev)
 	 */
 	vc4_flush_texture_caches(dev);
 
+	trace_vc4_submit_cl(dev, true, exec->seqno, exec->ct1ca, exec->ct1ea);
 	submit_cl(dev, 1, exec->ct1ca, exec->ct1ea);
 }
 
@@ -1135,6 +1138,10 @@ vc4_submit_cl_ioctl(struct drm_device *dev, void *data,
 	struct dma_fence *in_fence;
 	int ret = 0;
 
+	trace_vc4_submit_cl_ioctl(dev, args->bin_cl_size,
+				  args->shader_rec_size,
+				  args->bo_handle_count);
+
 	if (!vc4->v3d) {
 		DRM_DEBUG("VC4_SUBMIT_CL with no VC4 V3D probed\n");
 		return -ENODEV;

drivers/gpu/drm/vc4/vc4_irq.c

@@ -51,6 +51,7 @@
 
 #include "vc4_drv.h"
 #include "vc4_regs.h"
+#include "vc4_trace.h"
 
 #define V3D_DRIVER_IRQS (V3D_INT_OUTOMEM | \
 			 V3D_INT_FLDONE | \
@@ -123,6 +124,8 @@ vc4_irq_finish_bin_job(struct drm_device *dev)
 	if (!exec)
 		return;
 
+	trace_vc4_bcl_end_irq(dev, exec->seqno);
+
 	vc4_move_job_to_render(dev, exec);
 	next = vc4_first_bin_job(vc4);
 
@@ -161,6 +164,8 @@ vc4_irq_finish_render_job(struct drm_device *dev)
 	if (!exec)
 		return;
 
+	trace_vc4_rcl_end_irq(dev, exec->seqno);
+
 	vc4->finished_seqno++;
 	list_move_tail(&exec->head, &vc4->job_done_list);
 

drivers/gpu/drm/vc4/vc4_trace.h

@@ -52,6 +52,101 @@ TRACE_EVENT(vc4_wait_for_seqno_end,
 		      __entry->dev, __entry->seqno)
 );
 
+TRACE_EVENT(vc4_submit_cl_ioctl,
+	    TP_PROTO(struct drm_device *dev, u32 bin_cl_size, u32 shader_rec_size, u32 bo_count),
+	    TP_ARGS(dev, bin_cl_size, shader_rec_size, bo_count),
+
+	    TP_STRUCT__entry(
+			     __field(u32, dev)
+			     __field(u32, bin_cl_size)
+			     __field(u32, shader_rec_size)
+			     __field(u32, bo_count)
+			     ),
+
+	    TP_fast_assign(
+			   __entry->dev = dev->primary->index;
+			   __entry->bin_cl_size = bin_cl_size;
+			   __entry->shader_rec_size = shader_rec_size;
+			   __entry->bo_count = bo_count;
+			   ),
+
+	    TP_printk("dev=%u, bin_cl_size=%u, shader_rec_size=%u, bo_count=%u",
+		      __entry->dev,
+		      __entry->bin_cl_size,
+		      __entry->shader_rec_size,
+		      __entry->bo_count)
+);
+
+TRACE_EVENT(vc4_submit_cl,
+	    TP_PROTO(struct drm_device *dev, bool is_render,
+		     uint64_t seqno,
+		     u32 ctnqba, u32 ctnqea),
+	    TP_ARGS(dev, is_render, seqno, ctnqba, ctnqea),
+
+	    TP_STRUCT__entry(
+			     __field(u32, dev)
+			     __field(bool, is_render)
+			     __field(u64, seqno)
+			     __field(u32, ctnqba)
+			     __field(u32, ctnqea)
+			     ),
+
+	    TP_fast_assign(
+			   __entry->dev = dev->primary->index;
+			   __entry->is_render = is_render;
+			   __entry->seqno = seqno;
+			   __entry->ctnqba = ctnqba;
+			   __entry->ctnqea = ctnqea;
+			   ),
+
+	    TP_printk("dev=%u, %s, seqno=%llu, 0x%08x..0x%08x",
+		      __entry->dev,
+		      __entry->is_render ? "RCL" : "BCL",
+		      __entry->seqno,
+		      __entry->ctnqba,
+		      __entry->ctnqea)
+);
+
+TRACE_EVENT(vc4_bcl_end_irq,
+	    TP_PROTO(struct drm_device *dev,
+		     uint64_t seqno),
+	    TP_ARGS(dev, seqno),
+
+	    TP_STRUCT__entry(
+			     __field(u32, dev)
+			     __field(u64, seqno)
+			     ),
+
+	    TP_fast_assign(
+			   __entry->dev = dev->primary->index;
+			   __entry->seqno = seqno;
+			   ),
+
+	    TP_printk("dev=%u, seqno=%llu",
+		      __entry->dev,
+		      __entry->seqno)
+);
+
+TRACE_EVENT(vc4_rcl_end_irq,
+	    TP_PROTO(struct drm_device *dev,
+		     uint64_t seqno),
+	    TP_ARGS(dev, seqno),
+
+	    TP_STRUCT__entry(
+			     __field(u32, dev)
+			     __field(u64, seqno)
+			     ),
+
+	    TP_fast_assign(
+			   __entry->dev = dev->primary->index;
+			   __entry->seqno = seqno;
+			   ),
+
+	    TP_printk("dev=%u, seqno=%llu",
+		      __entry->dev,
+		      __entry->seqno)
+);
+
 #endif /* _VC4_TRACE_H_ */
 
 /* This part must be outside protection */