[Intel-gfx] [PATCH] drm/i915/gt: Set the PD again for Haswell

Chris Wilson chris at chris-wilson.co.uk
Tue Dec 3 10:43:17 UTC 2019


And Haswell still occasionally forgets it is meant to be using a new
page directory, so repeat ourselves a little louder.

<7> [509.919864] heartbeat rcs0 heartbeat {prio:-2147483645} not ticking
<7> [509.919895] heartbeat 	Awake? 8
<7> [509.919903] heartbeat 	Barriers?: no
<7> [509.919912] heartbeat 	Heartbeat: 3008 ms ago
<7> [509.919930] heartbeat 	Reset count: 0 (global 0)
<7> [509.919937] heartbeat 	Requests:
<7> [509.921008] heartbeat 		active  a7eb:56e1*  @ 5847ms:
<7> [509.921157] heartbeat 		ring->start:  0x00001000
<7> [509.921164] heartbeat 		ring->head:   0x00001610
<7> [509.921170] heartbeat 		ring->tail:   0x000023d8
<7> [509.921176] heartbeat 		ring->emit:   0x000023d8
<7> [509.921182] heartbeat 		ring->space:  0x00002570
<7> [509.921189] heartbeat 		ring->hwsp:   0x7fffe100
<7> [509.921197] heartbeat [head 1628, postfix 1738, tail 1750, batch 0xffffffff_ffffffff]:
<7> [509.921289] heartbeat [0000] 7a000002 00100002 00000000 00000000 7a000002 01154c1e 7ffff080 00000000
<7> [509.921299] heartbeat [0020] 11000001 00002220 ffffffff 12400001 00002220 7ffff000 00000000 11000001
<7> [509.921308] heartbeat [0040] 00002228 6e900000 7a000002 00100002 00000000 00000000 7a000002 01154c1e
<7> [509.921317] heartbeat [0060] 7ffff080 00000000 12400001 00002228 7ffff000 00000000 7a000002 00100002
<7> [509.921326] heartbeat [0080] 00000000 00000000 7a000002 01154c1e 7ffff080 00000000 7a000002 001010a1
<7> [509.921335] heartbeat [00a0] 7ffff080 00000000 04000000 11000005 00022050 00010001 00012050 00010001
<7> [509.921345] heartbeat [00c0] 0001a050 00010001 00000000 0c000000 459a110c 00000000 11000005 00022050
<7> [509.921354] heartbeat [00e0] 00010000 00012050 00010000 0001a050 00010000 12400001 0001a050 7ffff000
<7> [509.921363] heartbeat [0100] 00000000 04000001 18802100 00000000 7a000002 011050a1 7fffe100 000056e1
<7> [509.921370] heartbeat [0120] 01000000 00000000
<7> [509.921538] heartbeat 	MMIO base:  0x00002000
<7> [509.921682] heartbeat 	CCID: 0x3fa0110d
<7> [509.922342] heartbeat 	RING_START: 0x00001000
<7> [509.922353] heartbeat 	RING_HEAD:  0x00001628
<7> [509.922366] heartbeat 	RING_TAIL:  0x000023d8
<7> [509.922381] heartbeat 	RING_CTL:   0x00003001
<7> [509.922396] heartbeat 	RING_MODE:  0x00004000
<7> [509.922408] heartbeat 	RING_IMR: ffffffde
<7> [509.922421] heartbeat 	ACTHD:  0x00000000_30e01628
<7> [509.922434] heartbeat 	BBADDR: 0x00000000_00004004
<7> [509.922446] heartbeat 	DMA_FADDR: 0x00000000_00002800
<7> [509.922458] heartbeat 	IPEIR: 0x00000000
<7> [509.922470] heartbeat 	IPEHR: 0x780c0000
<7> [509.922642] heartbeat 	PP_DIR_BASE: 0x6e700000
<7> [509.922652] heartbeat 	PP_DIR_BASE_READ: 0x00000000
<7> [509.922662] heartbeat 	PP_DIR_DCLV: 0xffffffff
<7> [509.922678] heartbeat 		E  a7eb:56e1*  @ 5849ms:
<7> [509.922689] heartbeat 		E  a7eb:56e2-  @ 5849ms:
<7> [509.922698] heartbeat 		E  a7eb:56e3  @ 5848ms:
<7> [509.922707] heartbeat 		E  a7eb:56e4  @ 5848ms:
<7> [509.922715] heartbeat 		E  a7eb:56e5  @ 5847ms:
<7> [509.922724] heartbeat 		E  a7eb:56e6  @ 5846ms:
<7> [509.922735] heartbeat 		E  a7eb:56e7  @ 5846ms:
<7> [509.922744] heartbeat 		...skipping 4 executing requests...
<7> [509.922754] heartbeat 		E  a7eb:56ec  @ 3010ms:
<7> [509.922796] heartbeat HWSP:
<7> [509.922807] heartbeat [0000] 00000001 00000000 00000000 00000000 00000000 00000000 00000000 00000000
<7> [509.922817] heartbeat [0020] 00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000000
<7> [509.922826] heartbeat *
<7> [509.922836] heartbeat [0100] 000056e0 00000000 00000000 00000000 00000000 00000000 00000000 00000000
<7> [509.922845] heartbeat [0120] 00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000000
<7> [509.922851] heartbeat *
<7> [509.922870] heartbeat Idle? no
<7> [509.922878] heartbeat Signals:
<7> [509.923000] heartbeat 	[a7eb:56e2] @ 5850ms

Here, we have a failed context restore after the PD switch, but note
that the PP_DIR_BASE register does not match the LRI in the ring.

Bump it to 8 loops, and with that Baytrail starts passing the sanity
checks.

Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
---
 .../gpu/drm/i915/gt/intel_ring_submission.c   | 51 +++++--------------
 drivers/gpu/drm/i915/i915_pci.c               |  2 +-
 2 files changed, 15 insertions(+), 38 deletions(-)

diff --git a/drivers/gpu/drm/i915/gt/intel_ring_submission.c b/drivers/gpu/drm/i915/gt/intel_ring_submission.c
index dcdeef0a776f..a5c1a1bd2f35 100644
--- a/drivers/gpu/drm/i915/gt/intel_ring_submission.c
+++ b/drivers/gpu/drm/i915/gt/intel_ring_submission.c
@@ -1366,7 +1366,7 @@ static int load_pd_dir(struct i915_request *rq, const struct i915_ppgtt *ppgtt)
 	const struct intel_engine_cs * const engine = rq->engine;
 	u32 *cs;
 
-	cs = intel_ring_begin(rq, 10);
+	cs = intel_ring_begin(rq, 12);
 	if (IS_ERR(cs))
 		return PTR_ERR(cs);
 
@@ -1378,34 +1378,19 @@ static int load_pd_dir(struct i915_request *rq, const struct i915_ppgtt *ppgtt)
 	*cs++ = i915_mmio_reg_offset(RING_PP_DIR_DCLV(engine->mmio_base));
 	*cs++ = intel_gt_scratch_offset(rq->engine->gt,
 					INTEL_GT_SCRATCH_FIELD_DEFAULT);
-	*cs++ = MI_NOOP;
 
 	*cs++ = MI_LOAD_REGISTER_IMM(1);
 	*cs++ = i915_mmio_reg_offset(RING_PP_DIR_BASE(engine->mmio_base));
 	*cs++ = px_base(ppgtt->pd)->ggtt_offset << 10;
 
-	intel_ring_advance(rq, cs);
-
-	return 0;
-}
-
-static int flush_pd_dir(struct i915_request *rq)
-{
-	const struct intel_engine_cs * const engine = rq->engine;
-	u32 *cs;
-
-	cs = intel_ring_begin(rq, 4);
-	if (IS_ERR(cs))
-		return PTR_ERR(cs);
-
-	/* Stall until the page table load is complete */
+	/* Stall until the page table load is complete? */
 	*cs++ = MI_STORE_REGISTER_MEM | MI_SRM_LRM_GLOBAL_GTT;
 	*cs++ = i915_mmio_reg_offset(RING_PP_DIR_BASE(engine->mmio_base));
 	*cs++ = intel_gt_scratch_offset(rq->engine->gt,
 					INTEL_GT_SCRATCH_FIELD_DEFAULT);
-	*cs++ = MI_NOOP;
 
 	intel_ring_advance(rq, cs);
+
 	return 0;
 }
 
@@ -1592,18 +1577,7 @@ static int switch_context(struct i915_request *rq)
 
 	if (vm) {
 		struct intel_engine_cs *engine = rq->engine;
-
-		ret = load_pd_dir(rq, i915_vm_to_ppgtt(vm));
-		if (ret)
-			return ret;
-
-		ret = engine->emit_flush(rq, EMIT_INVALIDATE);
-		if (ret)
-			return ret;
-
-		ret = flush_pd_dir(rq);
-		if (ret)
-			return ret;
+		int loops = 8; /* 2 for Haswell? 8 for Baytrail! */
 
 		/*
 		 * Not only do we need a full barrier (post-sync write) after
@@ -1613,13 +1587,16 @@ static int switch_context(struct i915_request *rq)
 		 * post-sync op, this extra pass appears vital before a
 		 * mm switch!
 		 */
-		ret = engine->emit_flush(rq, EMIT_INVALIDATE);
-		if (ret)
-			return ret;
-
-		ret = engine->emit_flush(rq, EMIT_FLUSH);
-		if (ret)
-			return ret;
+		do {
+			ret = load_pd_dir(rq, i915_vm_to_ppgtt(vm));
+			if (ret)
+				return ret;
+
+			ret = engine->emit_flush(rq,
+						 EMIT_FLUSH | EMIT_INVALIDATE);
+			if (ret)
+				return ret;
+		} while (--loops);
 	}
 
 	if (ce->state) {
diff --git a/drivers/gpu/drm/i915/i915_pci.c b/drivers/gpu/drm/i915/i915_pci.c
index 583e0cd94a6a..bba6b50e6beb 100644
--- a/drivers/gpu/drm/i915/i915_pci.c
+++ b/drivers/gpu/drm/i915/i915_pci.c
@@ -493,7 +493,7 @@ static const struct intel_device_info intel_valleyview_info = {
 	.has_rps = true,
 	.display.has_gmch = 1,
 	.display.has_hotplug = 1,
-	.ppgtt_type = INTEL_PPGTT_ALIASING,
+	.ppgtt_type = INTEL_PPGTT_FULL,
 	.ppgtt_size = 31,
 	.has_snoop = true,
 	.has_coherent_ggtt = false,
-- 
2.24.0



More information about the Intel-gfx mailing list