Skip to contentSkip to conversation

drm/i915: Dumb down the semaphore logic

PATCH In review Ben Widawsky 1 participant 3 messages Sep 15, 2011

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH v4 1/2 drm/i915: Dumb down the semaphore logic

While I think the previous code is correct, it was hard to follow and hard to debug. Since we already have a ring abstraction, might as well use it to handle the semaphore updates and compares.

I don't expect this code to make semaphores better or worse, but you never know...

v2:
Remove magic per Keith's suggestions.
Ran Daniel's gem_ring_sync_loop test on this.

v3:
Ignored one of Keith's suggestions.

v4:
Removed some bloat per Daniel's recommendation.

Cc Daniel Vetter <[email protected]>
Cc Keith Packard <[email protected]>
---
 drivers/gpu/drm/i915/i915_gem_execbuffer.c |    3 +-
 drivers/gpu/drm/i915/i915_reg.h            |   13 +++
 drivers/gpu/drm/i915/intel_ringbuffer.c    |  143 ++++++++++++++++++++--------
 drivers/gpu/drm/i915/intel_ringbuffer.h    |    8 +-
 4 files changed, 123 insertions(+), 44 deletions(-)
drivers/gpu/drm/i915/i915_gem_execbuffer.c +2 -1
index 4934cf8..3693e83 100644
@@ -784,7 +784,8 @@ i915_gem_execbuffer_sync_rings(struct drm_i915_gem_object *obj,
 	}
 
 	from->sync_seqno[idx] = seqno;
-	return intel_ring_sync(to, from, seqno - 1);
+
+	return to->sync_to(to, from, seqno - 1);
 }
 
 static int
drivers/gpu/drm/i915/i915_reg.h +13 -0
index 542453f..8ace3d3 100644
@@ -194,6 +194,13 @@
 #define  MI_SEMAPHORE_UPDATE	    (1<<21)
 #define  MI_SEMAPHORE_COMPARE	    (1<<20)
 #define  MI_SEMAPHORE_REGISTER	    (1<<18)
+#define  MI_SEMAPHORE_SYNC_RV	    (2<<16)
+#define  MI_SEMAPHORE_SYNC_RB	    (0<<16)
+#define  MI_SEMAPHORE_SYNC_VR	    (0<<16)
+#define  MI_SEMAPHORE_SYNC_VB	    (2<<16)
+#define  MI_SEMAPHORE_SYNC_BR	    (2<<16)
+#define  MI_SEMAPHORE_SYNC_BV	    (0<<16)
+#define  MI_SEMAPHORE_SYNC_INVALID  (1<<0)
 /*
  * 3D instructions used by the kernel
  */
@@ -296,6 +303,12 @@
 #define RING_CTL(base)		((base)+0x3c)
 #define RING_SYNC_0(base)	((base)+0x40)
 #define RING_SYNC_1(base)	((base)+0x44)
+#define GEN6_RVSYNC (RING_SYNC_0(RENDER_RING_BASE))
+#define GEN6_RBSYNC (RING_SYNC_1(RENDER_RING_BASE))
+#define GEN6_VRSYNC (RING_SYNC_1(GEN6_BSD_RING_BASE))
+#define GEN6_VBSYNC (RING_SYNC_0(GEN6_BSD_RING_BASE))
+#define GEN6_BRSYNC (RING_SYNC_0(BLT_RING_BASE))
+#define GEN6_BVSYNC (RING_SYNC_1(BLT_RING_BASE))
 #define RING_MAX_IDLE(base)	((base)+0x54)
 #define RING_HWS_PGA(base)	((base)+0x80)
 #define RING_HWS_PGA_GEN6(base)	((base)+0x2080)
drivers/gpu/drm/i915/intel_ringbuffer.c +103 -40

200 lines not on this page · raw message

drivers/gpu/drm/i915/intel_ringbuffer.h +5 -3
index 39ac2b6..2ae1c4a 100644
@@ -75,7 +75,12 @@ struct  intel_ring_buffer {
 	int		(*dispatch_execbuffer)(struct intel_ring_buffer *ring,
 					       u32 offset, u32 length);
 	void		(*cleanup)(struct intel_ring_buffer *ring);
+	int		(*sync_to)(struct intel_ring_buffer *ring,
+				   struct intel_ring_buffer *to,
+				   u32 seqno);
 
+	u32		semaphore_register[3]; /*our mbox written by others */
+	u32		signal_mbox[2]; /* mboxes this ring signals to */
 	/**
 	 * List of objects currently involved in rendering from the
 	 * ringbuffer.
@@ -180,9 +185,6 @@ static inline void intel_ring_emit(struct intel_ring_buffer *ring,
 void intel_ring_advance(struct intel_ring_buffer *ring);
 
 u32 intel_ring_get_seqno(struct intel_ring_buffer *ring);
-int intel_ring_sync(struct intel_ring_buffer *ring,
-		    struct intel_ring_buffer *to,
-		    u32 seqno);
 
 int intel_init_render_ring_buffer(struct drm_device *dev);
 int intel_init_bsd_ring_buffer(struct drm_device *dev);
Signature · 1 line
1.7.6.1
#1 lore raw

PATCH v4 2/2 drm/i915: tracepoints for semaphores

The tracepoints give enough info to figure the updates and compares (document terminology for signals and waits) and dependencies therein of the semaphore mailboxes.

Here are arguments to perf to get interesting info (mostly copied from
Chris):
record -f -g -c 1 -e i915:intel_ringbuffer_add_request -e i915:intel_ringbuffer_consume_semaphore -a

---
 drivers/gpu/drm/i915/i915_trace.h       |   61 +++++++++++++++++++++++++++++++
 drivers/gpu/drm/i915/intel_ringbuffer.c |    4 ++
 2 files changed, 65 insertions(+), 0 deletions(-)
drivers/gpu/drm/i915/i915_trace.h +61 -0
index d623fef..47d55fd 100644
@@ -410,6 +410,67 @@ TRACE_EVENT(i915_reg_rw,
 		     (u32)(__entry->val >> 32))
 );
 
+#define SEMAPHORE_REG_NAME(x) (\
+	(x == RENDER_RING_BASE + 0x40) ? ("RVSYNC") : \
+	((x == RENDER_RING_BASE + 0x44) ? ("RBSYNC") : \
+	((x == GEN6_BSD_RING_BASE + 0x40) ? ("VRSYNC") : \
+	((x == GEN6_BSD_RING_BASE + 0x44) ? ("VBSYNC") : \
+	((x == BLT_RING_BASE + 0x40) ? ("BVSYNC") : \
+	((x == BLT_RING_BASE + 0x44) ? ("BRSYNC") : \
+	("UNKNOWN")))))))
+
+#define RING_NAME(x) (\
+	(x == 1) ? ("rcs") : \
+	((x == 2) ? ("vcs") : \
+	((x == 4) ? ("bcs") : \
+	("unk"))))
+
+TRACE_EVENT(intel_ringbuffer_add_request,
+	    TP_PROTO(struct intel_ring_buffer *ring,
+		     u32 seqno,
+		     u32 reg1,
+		     u32 reg2),
+	    TP_ARGS(ring, seqno, reg1, reg2),
+	    TP_STRUCT__entry(
+			     __field(int, id)
+			     __field(u32, seqno)
+			     __field(u32, reg1)
+			     __field(u32, reg2)
+			    ),
+	    TP_fast_assign(
+			   __entry->id = ring->id;
+			   __entry->seqno = seqno;
+			   __entry->reg1 = reg1;
+			   __entry->reg2 = reg2;
+			  ),
+	    TP_printk("ring = %s, seqno = %u, signal mbox 1 = %d, signal mbox 2 = %d",
+		      RING_NAME(__entry->id), __entry->seqno,
+		      SEMAPHORE_REG_NAME(__entry->reg1),
+		      SEMAPHORE_REG_NAME(__entry->reg2))
+);
+
+TRACE_EVENT(intel_ringbuffer_consume_semaphore,
+	    TP_PROTO(struct intel_ring_buffer *updater,
+		     struct intel_ring_buffer *comparer,
+		     u32 seqno),
+	    TP_ARGS(updater, comparer, seqno),
+	    TP_STRUCT__entry(
+			     __field(int, src)
+			     __field(int, dest)
+			     __field(u32, seqno)
+			    ),
+	    TP_fast_assign(
+			   __entry->src = updater->id;
+			   __entry->dest = comparer->id;
+			   __entry->seqno = seqno;
+			  ),
+	    /* It's not easy to macro the consuming mbox register */
+	    TP_printk("signaller = %s, waiter = %s, seqno = %u",
+		      RING_NAME(__entry->src),
+		      RING_NAME(__entry->dest),
+		      __entry->seqno)
+);
+
 #endif /* _I915_TRACE_H_ */
 
 /* This part must be outside protection */
drivers/gpu/drm/i915/intel_ringbuffer.c +4 -0
index f25d296..38e9ca4 100644
@@ -361,6 +361,8 @@ gen6_add_request(struct intel_ring_buffer *ring,
 	intel_ring_emit(ring, MI_USER_INTERRUPT);
 	intel_ring_advance(ring);
 
+	trace_intel_ringbuffer_add_request(ring, seqno, mbox1_reg, mbox2_reg);
+
 	return 0;
 }
 
@@ -392,6 +394,8 @@ intel_ring_sync(struct intel_ring_buffer *waiter,
 	intel_ring_emit(waiter, MI_NOOP);
 	intel_ring_advance(waiter);
 
+	trace_intel_ringbuffer_consume_semaphore(signaller, waiter, seqno);
+
 	return 0;
 }
 
Signature · 1 line
1.7.6.1
#2 lore raw
In reply to Ben Widawsky
Quoted 11 lines · Ben Widawsky

While I think the previous code is correct, it was hard to follow and hard to debug. Since we already have a ring abstraction, might as well use it to handle the semaphore updates and compares.

I don't expect this code to make semaphores better or worse, but you never know...

v2:
Remove magic per Keith's suggestions.
Ran Daniel's gem_ring_sync_loop test on this.

Quoted 5 lines

v3:
Ignored one of Keith's suggestions.

v4:
Removed some bloat per Daniel's recommendation.

Just to clarify, I think I like the existing code a bit better than what the code looked like with Daniel's other recommendation to cut out the per ring sync_to function. If he shows me something good, there will be a v5, I guess.