[Intel-gfx] [CI] drm/i915/gen7: Re-enable full-ppgtt for ivb, byt, hsw

[Chris Wilson]

After much hair pulling, resort to preallocating the ppGTT entries on
init to circumvent the apparent lack of PD invalidate following the
write to PP_DCLV upon switching mm between contexts (and here the same
context after binding new objects). However, the details of that PP_DCLV
invalidate are still unknown, and it appears we need to reload the mm
twice to cover over a timing issue. Worrying.

Fixes: 3dc007fe9b2b ("drm/i915/gtt: Downgrade gen7 (ivb, byt, hsw) back to aliasing-ppgtt")
Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
---
 .../gpu/drm/i915/gt/intel_ring_submission.c   | 35 ++++++++++---------
 drivers/gpu/drm/i915/i915_gem_gtt.c           | 14 ++++----
 drivers/gpu/drm/i915/i915_pci.c               |  4 +--
 3 files changed, 28 insertions(+), 25 deletions(-)

diff --git a/drivers/gpu/drm/i915/gt/intel_ring_submission.c b/drivers/gpu/drm/i915/gt/intel_ring_submission.c
index f25ceccb335e..070127ebcc9b 100644
--- a/drivers/gpu/drm/i915/gt/intel_ring_submission.c
+++ b/drivers/gpu/drm/i915/gt/intel_ring_submission.c
@@ -1366,13 +1366,21 @@ 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, 6);
+	cs = intel_ring_begin(rq, 12);
 	if (IS_ERR(cs))
 		return PTR_ERR(cs);
 
-	*cs++ = MI_LOAD_REGISTER_IMM(1);
+	*cs++ = MI_LOAD_REGISTER_IMM(2);
 	*cs++ = i915_mmio_reg_offset(RING_PP_DIR_DCLV(engine->mmio_base));
 	*cs++ = PP_DIR_DCLV_2G;
+	*cs++ = i915_mmio_reg_offset(RING_PP_DIR_DCLV(engine->mmio_base)) + 4;
+	*cs++ = 0;
+
+	*cs++ = MI_STORE_REGISTER_MEM | MI_SRM_LRM_GLOBAL_GTT;
+	*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));
@@ -1579,30 +1587,25 @@ static int switch_context(struct i915_request *rq)
 {
 	struct intel_context *ce = rq->hw_context;
 	struct i915_address_space *vm = vm_alias(ce);
+	u32 hw_flags = 0;
 	int ret;
 
 	GEM_BUG_ON(HAS_EXECLISTS(rq->i915));
 
 	if (vm) {
-		ret = load_pd_dir(rq, i915_vm_to_ppgtt(vm));
-		if (ret)
-			return ret;
+		int loops = 8;
+
+		do {
+			ret = load_pd_dir(rq, i915_vm_to_ppgtt(vm));
+			if (ret)
+				return ret;
+		} while (--loops);
 	}
 
 	if (ce->state) {
-		u32 hw_flags;
-
 		GEM_BUG_ON(rq->engine->id != RCS0);
 
-		/*
-		 * The kernel context(s) is treated as pure scratch and is not
-		 * expected to retain any state (as we sacrifice it during
-		 * suspend and on resume it may be corrupted). This is ok,
-		 * as nothing actually executes using the kernel context; it
-		 * is purely used for flushing user contexts.
-		 */
-		hw_flags = 0;
-		if (i915_gem_context_is_kernel(rq->gem_context))
+		if (!rq->engine->default_state)
 			hw_flags = MI_RESTORE_INHIBIT;
 
 		ret = mi_set_context(rq, hw_flags);
diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.c b/drivers/gpu/drm/i915/i915_gem_gtt.c
index 6239a9adbf14..fb72ceeeb2b3 100644
--- a/drivers/gpu/drm/i915/i915_gem_gtt.c
+++ b/drivers/gpu/drm/i915/i915_gem_gtt.c
@@ -1692,7 +1692,6 @@ static int gen6_alloc_va_range(struct i915_address_space *vm,
 	intel_wakeref_t wakeref;
 	u64 from = start;
 	unsigned int pde;
-	bool flush = false;
 	int ret = 0;
 
 	wakeref = intel_runtime_pm_get(&vm->i915->runtime_pm);
@@ -1717,11 +1716,6 @@ static int gen6_alloc_va_range(struct i915_address_space *vm,
 			spin_lock(&pd->lock);
 			if (pd->entry[pde] == &vm->scratch[1]) {
 				pd->entry[pde] = pt;
-				if (i915_vma_is_bound(ppgtt->vma,
-						      I915_VMA_GLOBAL_BIND)) {
-					gen6_write_pde(ppgtt, pde, pt);
-					flush = true;
-				}
 			} else {
 				alloc = pt;
 				pt = pd->entry[pde];
@@ -1732,8 +1726,13 @@ static int gen6_alloc_va_range(struct i915_address_space *vm,
 	}
 	spin_unlock(&pd->lock);
 
-	if (flush)
+	if (i915_vma_is_bound(ppgtt->vma, I915_VMA_GLOBAL_BIND)) {
+		/* Rewrite them all! Anything less misses an invalidate. */
+		gen6_for_all_pdes(pt, pd, pde)
+			gen6_write_pde(ppgtt, pde, pt);
+		ioread32(ppgtt->pd_addr + pde - 1);
 		gen6_ggtt_invalidate(vm->gt->ggtt);
+	}
 
 	goto out;
 
@@ -1994,6 +1993,7 @@ static struct i915_ppgtt *gen6_ppgtt_create(struct drm_i915_private *i915)
 err_pd:
 	kfree(ppgtt->base.pd);
 err_free:
+	mutex_destroy(&ppgtt->pin_mutex);
 	kfree(ppgtt);
 	return ERR_PTR(err);
 }
diff --git a/drivers/gpu/drm/i915/i915_pci.c b/drivers/gpu/drm/i915/i915_pci.c
index da3e9b5752ac..bba6b50e6beb 100644
--- a/drivers/gpu/drm/i915/i915_pci.c
+++ b/drivers/gpu/drm/i915/i915_pci.c
@@ -436,7 +436,7 @@ static const struct intel_device_info intel_sandybridge_m_gt2_info = {
 	.has_rc6 = 1, \
 	.has_rc6p = 1, \
 	.has_rps = true, \
-	.ppgtt_type = INTEL_PPGTT_ALIASING, \
+	.ppgtt_type = INTEL_PPGTT_FULL, \
 	.ppgtt_size = 31, \
 	IVB_PIPE_OFFSETS, \
 	IVB_CURSOR_OFFSETS, \
@@ -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