Merge tag 'drm-intel-fixes-2020-11-05' of...

Merge tag 'drm-intel-fixes-2020-11-05' of git://anongit.freedesktop.org/drm/drm-intel into drm-fixes

- GVT fixes including vGPU suspend/resume fixes and workaround for APL guest GPU hang.
- Fix set domain's cache coherency (Chris)
- Fixes around breadcrumbs (Chris)
- Fix encoder lookup during PSR atomic (Imre)
- Hold onto an explicit ref to i915_vma_work.pinned (Chris)
Signed-off-by: Dave Airlie <airlied@redhat.com>

From: Rodrigo Vivi <rodrigo.vivi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20201105173026.GA858446@intel.com

drivers/gpu/drm/i915/display/intel_psr.c

@@ -1754,7 +1754,7 @@ void intel_psr_atomic_check(struct drm_connector *connector,
 		return;
 
 	intel_connector = to_intel_connector(connector);
-	dig_port = enc_to_dig_port(intel_attached_encoder(intel_connector));
+	dig_port = enc_to_dig_port(to_intel_encoder(new_state->best_encoder));
 	if (dev_priv->psr.dp != &dig_port->dp)
 		return;
 

drivers/gpu/drm/i915/gem/i915_gem_domain.c

@@ -508,21 +508,6 @@ i915_gem_set_domain_ioctl(struct drm_device *dev, void *data,
 	if (!obj)
 		return -ENOENT;
 
-	/*
-	 * Already in the desired write domain? Nothing for us to do!
-	 *
-	 * We apply a little bit of cunning here to catch a broader set of
-	 * no-ops. If obj->write_domain is set, we must be in the same
-	 * obj->read_domains, and only that domain. Therefore, if that
-	 * obj->write_domain matches the request read_domains, we are
-	 * already in the same read/write domain and can skip the operation,
-	 * without having to further check the requested write_domain.
-	 */
-	if (READ_ONCE(obj->write_domain) == read_domains) {
-		err = 0;
-		goto out;
-	}
-
 	/*
 	 * Try to flush the object off the GPU without holding the lock.
 	 * We will repeat the flush holding the lock in the normal manner
@@ -560,6 +545,19 @@ i915_gem_set_domain_ioctl(struct drm_device *dev, void *data,
 	if (err)
 		goto out;
 
+	/*
+	 * Already in the desired write domain? Nothing for us to do!
+	 *
+	 * We apply a little bit of cunning here to catch a broader set of
+	 * no-ops. If obj->write_domain is set, we must be in the same
+	 * obj->read_domains, and only that domain. Therefore, if that
+	 * obj->write_domain matches the request read_domains, we are
+	 * already in the same read/write domain and can skip the operation,
+	 * without having to further check the requested write_domain.
+	 */
+	if (READ_ONCE(obj->write_domain) == read_domains)
+		goto out_unpin;
+
 	err = i915_gem_object_lock_interruptible(obj, NULL);
 	if (err)
 		goto out_unpin;

drivers/gpu/drm/i915/gt/intel_engine.h

@@ -245,22 +245,14 @@ static inline u32 *gen12_emit_pipe_control(u32 *batch, u32 flags0, u32 flags1, u
 }
 
 static inline u32 *
-__gen8_emit_ggtt_write_rcs(u32 *cs, u32 value, u32 gtt_offset, u32 flags0, u32 flags1)
+__gen8_emit_write_rcs(u32 *cs, u32 value, u32 offset, u32 flags0, u32 flags1)
 {
-	/* We're using qword write, offset should be aligned to 8 bytes. */
-	GEM_BUG_ON(!IS_ALIGNED(gtt_offset, 8));
-
-	/* w/a for post sync ops following a GPGPU operation we
-	 * need a prior CS_STALL, which is emitted by the flush
-	 * following the batch.
-	 */
 	*cs++ = GFX_OP_PIPE_CONTROL(6) | flags0;
-	*cs++ = flags1 | PIPE_CONTROL_QW_WRITE | PIPE_CONTROL_GLOBAL_GTT_IVB;
-	*cs++ = gtt_offset;
+	*cs++ = flags1 | PIPE_CONTROL_QW_WRITE;
+	*cs++ = offset;
 	*cs++ = 0;
 	*cs++ = value;
-	/* We're thrashing one dword of HWS. */
-	*cs++ = 0;
+	*cs++ = 0; /* We're thrashing one extra dword. */
 
 	return cs;
 }
@@ -268,13 +260,38 @@ __gen8_emit_ggtt_write_rcs(u32 *cs, u32 value, u32 gtt_offset, u32 flags0, u32 f
 static inline u32*
 gen8_emit_ggtt_write_rcs(u32 *cs, u32 value, u32 gtt_offset, u32 flags)
 {
-	return __gen8_emit_ggtt_write_rcs(cs, value, gtt_offset, 0, flags);
+	/* We're using qword write, offset should be aligned to 8 bytes. */
+	GEM_BUG_ON(!IS_ALIGNED(gtt_offset, 8));
+
+	return __gen8_emit_write_rcs(cs,
+				     value,
+				     gtt_offset,
+				     0,
+				     flags | PIPE_CONTROL_GLOBAL_GTT_IVB);
 }
 
 static inline u32*
 gen12_emit_ggtt_write_rcs(u32 *cs, u32 value, u32 gtt_offset, u32 flags0, u32 flags1)
 {
-	return __gen8_emit_ggtt_write_rcs(cs, value, gtt_offset, flags0, flags1);
+	/* We're using qword write, offset should be aligned to 8 bytes. */
+	GEM_BUG_ON(!IS_ALIGNED(gtt_offset, 8));
+
+	return __gen8_emit_write_rcs(cs,
+				     value,
+				     gtt_offset,
+				     flags0,
+				     flags1 | PIPE_CONTROL_GLOBAL_GTT_IVB);
+}
+
+static inline u32 *
+__gen8_emit_flush_dw(u32 *cs, u32 value, u32 gtt_offset, u32 flags)
+{
+	*cs++ = (MI_FLUSH_DW + 1) | flags;
+	*cs++ = gtt_offset;
+	*cs++ = 0;
+	*cs++ = value;
+
+	return cs;
 }
 
 static inline u32 *
@@ -285,12 +302,10 @@ gen8_emit_ggtt_write(u32 *cs, u32 value, u32 gtt_offset, u32 flags)
 	/* Offset should be aligned to 8 bytes for both (QW/DW) write types */
 	GEM_BUG_ON(!IS_ALIGNED(gtt_offset, 8));
 
-	*cs++ = (MI_FLUSH_DW + 1) | MI_FLUSH_DW_OP_STOREDW | flags;
-	*cs++ = gtt_offset | MI_FLUSH_DW_USE_GTT;
-	*cs++ = 0;
-	*cs++ = value;
-
-	return cs;
+	return __gen8_emit_flush_dw(cs,
+				    value,
+				    gtt_offset | MI_FLUSH_DW_USE_GTT,
+				    flags | MI_FLUSH_DW_OP_STOREDW);
 }
 
 static inline void __intel_engine_reset(struct intel_engine_cs *engine,

drivers/gpu/drm/i915/gt/intel_lrc.c

@@ -3547,6 +3547,19 @@ static const struct intel_context_ops execlists_context_ops = {
 	.destroy = execlists_context_destroy,
 };
 
+static u32 hwsp_offset(const struct i915_request *rq)
+{
+	const struct intel_timeline_cacheline *cl;
+
+	/* Before the request is executed, the timeline/cachline is fixed */
+
+	cl = rcu_dereference_protected(rq->hwsp_cacheline, 1);
+	if (cl)
+		return cl->ggtt_offset;
+
+	return rcu_dereference_protected(rq->timeline, 1)->hwsp_offset;
+}
+
 static int gen8_emit_init_breadcrumb(struct i915_request *rq)
 {
 	u32 *cs;
@@ -3569,7 +3582,7 @@ static int gen8_emit_init_breadcrumb(struct i915_request *rq)
 	*cs++ = MI_NOOP;
 
 	*cs++ = MI_STORE_DWORD_IMM_GEN4 | MI_USE_GGTT;
-	*cs++ = i915_request_timeline(rq)->hwsp_offset;
+	*cs++ = hwsp_offset(rq);
 	*cs++ = 0;
 	*cs++ = rq->fence.seqno - 1;
 
@@ -4886,11 +4899,9 @@ gen8_emit_fini_breadcrumb_tail(struct i915_request *request, u32 *cs)
 	return gen8_emit_wa_tail(request, cs);
 }
 
-static u32 *emit_xcs_breadcrumb(struct i915_request *request, u32 *cs)
+static u32 *emit_xcs_breadcrumb(struct i915_request *rq, u32 *cs)
 {
-	u32 addr = i915_request_active_timeline(request)->hwsp_offset;
-
-	return gen8_emit_ggtt_write(cs, request->fence.seqno, addr, 0);
+	return gen8_emit_ggtt_write(cs, rq->fence.seqno, hwsp_offset(rq), 0);
 }
 
 static u32 *gen8_emit_fini_breadcrumb(struct i915_request *rq, u32 *cs)
@@ -4909,7 +4920,7 @@ static u32 *gen8_emit_fini_breadcrumb_rcs(struct i915_request *request, u32 *cs)
 	/* XXX flush+write+CS_STALL all in one upsets gem_concurrent_blt:kbl */
 	cs = gen8_emit_ggtt_write_rcs(cs,
 				      request->fence.seqno,
-				      i915_request_active_timeline(request)->hwsp_offset,
+				      hwsp_offset(request),
 				      PIPE_CONTROL_FLUSH_ENABLE |
 				      PIPE_CONTROL_CS_STALL);
 
@@ -4921,7 +4932,7 @@ gen11_emit_fini_breadcrumb_rcs(struct i915_request *request, u32 *cs)
 {
 	cs = gen8_emit_ggtt_write_rcs(cs,
 				      request->fence.seqno,
-				      i915_request_active_timeline(request)->hwsp_offset,
+				      hwsp_offset(request),
 				      PIPE_CONTROL_CS_STALL |
 				      PIPE_CONTROL_TILE_CACHE_FLUSH |
 				      PIPE_CONTROL_RENDER_TARGET_CACHE_FLUSH |
@@ -4983,7 +4994,9 @@ gen12_emit_fini_breadcrumb_tail(struct i915_request *request, u32 *cs)
 
 static u32 *gen12_emit_fini_breadcrumb(struct i915_request *rq, u32 *cs)
 {
-	return gen12_emit_fini_breadcrumb_tail(rq, emit_xcs_breadcrumb(rq, cs));
+	/* XXX Stalling flush before seqno write; post-sync not */
+	cs = emit_xcs_breadcrumb(rq, __gen8_emit_flush_dw(cs, 0, 0, 0));
+	return gen12_emit_fini_breadcrumb_tail(rq, cs);
 }
 
 static u32 *
@@ -4991,7 +5004,7 @@ gen12_emit_fini_breadcrumb_rcs(struct i915_request *request, u32 *cs)
 {
 	cs = gen12_emit_ggtt_write_rcs(cs,
 				       request->fence.seqno,
-				       i915_request_active_timeline(request)->hwsp_offset,
+				       hwsp_offset(request),
 				       PIPE_CONTROL0_HDC_PIPELINE_FLUSH,
 				       PIPE_CONTROL_CS_STALL |
 				       PIPE_CONTROL_TILE_CACHE_FLUSH |

drivers/gpu/drm/i915/gt/intel_timeline.c

@@ -188,10 +188,14 @@ cacheline_alloc(struct intel_timeline_hwsp *hwsp, unsigned int cacheline)
 	return cl;
 }
 
-static void cacheline_acquire(struct intel_timeline_cacheline *cl)
+static void cacheline_acquire(struct intel_timeline_cacheline *cl,
+			      u32 ggtt_offset)
 {
-	if (cl)
-		i915_active_acquire(&cl->active);
+	if (!cl)
+		return;
+
+	cl->ggtt_offset = ggtt_offset;
+	i915_active_acquire(&cl->active);
 }
 
 static void cacheline_release(struct intel_timeline_cacheline *cl)
@@ -340,7 +344,7 @@ int intel_timeline_pin(struct intel_timeline *tl, struct i915_gem_ww_ctx *ww)
 	GT_TRACE(tl->gt, "timeline:%llx using HWSP offset:%x\n",
 		 tl->fence_context, tl->hwsp_offset);
 
-	cacheline_acquire(tl->hwsp_cacheline);
+	cacheline_acquire(tl->hwsp_cacheline, tl->hwsp_offset);
 	if (atomic_fetch_inc(&tl->pin_count)) {
 		cacheline_release(tl->hwsp_cacheline);
 		__i915_vma_unpin(tl->hwsp_ggtt);
@@ -515,7 +519,7 @@ __intel_timeline_get_seqno(struct intel_timeline *tl,
 	GT_TRACE(tl->gt, "timeline:%llx using HWSP offset:%x\n",
 		 tl->fence_context, tl->hwsp_offset);
 
-	cacheline_acquire(cl);
+	cacheline_acquire(cl, tl->hwsp_offset);
 	tl->hwsp_cacheline = cl;
 
 	*seqno = timeline_advance(tl);
@@ -573,9 +577,7 @@ int intel_timeline_read_hwsp(struct i915_request *from,
 	if (err)
 		goto out;
 
-	*hwsp = i915_ggtt_offset(cl->hwsp->vma) +
-		ptr_unmask_bits(cl->vaddr, CACHELINE_BITS) * CACHELINE_BYTES;
-
+	*hwsp = cl->ggtt_offset;
 out:
 	i915_active_release(&cl->active);
 	return err;

drivers/gpu/drm/i915/gt/intel_timeline_types.h

@@ -94,6 +94,8 @@ struct intel_timeline_cacheline {
 	struct intel_timeline_hwsp *hwsp;
 	void *vaddr;
 
+	u32 ggtt_offset;
+
 	struct rcu_head rcu;
 };
 

drivers/gpu/drm/i915/gvt/handlers.c

@@ -1489,7 +1489,8 @@ static int hws_pga_write(struct intel_vgpu *vgpu, unsigned int offset,
 	const struct intel_engine_cs *engine =
 		intel_gvt_render_mmio_to_engine(vgpu->gvt, offset);
 
-	if (!intel_gvt_ggtt_validate_range(vgpu, value, I915_GTT_PAGE_SIZE)) {
+	if (value != 0 &&
+	    !intel_gvt_ggtt_validate_range(vgpu, value, I915_GTT_PAGE_SIZE)) {
 		gvt_vgpu_err("write invalid HWSP address, reg:0x%x, value:0x%x\n",
 			      offset, value);
 		return -EINVAL;
@@ -1650,6 +1651,34 @@ static int edp_psr_imr_iir_write(struct intel_vgpu *vgpu,
 	return 0;
 }
 
+/**
+ * FixMe:
+ * If guest fills non-priv batch buffer on ApolloLake/Broxton as Mesa i965 did:
+ * 717e7539124d (i965: Use a WC map and memcpy for the batch instead of pwrite.)
+ * Due to the missing flush of bb filled by VM vCPU, host GPU hangs on executing
+ * these MI_BATCH_BUFFER.
+ * Temporarily workaround this by setting SNOOP bit for PAT3 used by PPGTT
+ * PML4 PTE: PAT(0) PCD(1) PWT(1).
+ * The performance is still expected to be low, will need further improvement.
+ */
+static int bxt_ppat_low_write(struct intel_vgpu *vgpu, unsigned int offset,
+			      void *p_data, unsigned int bytes)
+{
+	u64 pat =
+		GEN8_PPAT(0, CHV_PPAT_SNOOP) |
+		GEN8_PPAT(1, 0) |
+		GEN8_PPAT(2, 0) |
+		GEN8_PPAT(3, CHV_PPAT_SNOOP) |
+		GEN8_PPAT(4, CHV_PPAT_SNOOP) |
+		GEN8_PPAT(5, CHV_PPAT_SNOOP) |
+		GEN8_PPAT(6, CHV_PPAT_SNOOP) |
+		GEN8_PPAT(7, CHV_PPAT_SNOOP);
+
+	vgpu_vreg(vgpu, offset) = lower_32_bits(pat);
+
+	return 0;
+}
+
 static int guc_status_read(struct intel_vgpu *vgpu,
 			   unsigned int offset, void *p_data,
 			   unsigned int bytes)
@@ -2812,7 +2841,7 @@ static int init_bdw_mmio_info(struct intel_gvt *gvt)
 
 	MMIO_DH(GEN6_PCODE_MAILBOX, D_BDW_PLUS, NULL, mailbox_write);
 
-	MMIO_D(GEN8_PRIVATE_PAT_LO, D_BDW_PLUS);
+	MMIO_D(GEN8_PRIVATE_PAT_LO, D_BDW_PLUS & ~D_BXT);
 	MMIO_D(GEN8_PRIVATE_PAT_HI, D_BDW_PLUS);
 
 	MMIO_D(GAMTARBMODE, D_BDW_PLUS);
@@ -3139,7 +3168,7 @@ static int init_skl_mmio_info(struct intel_gvt *gvt)
 		 NULL, NULL);
 
 	MMIO_DFH(GAMT_CHKN_BIT_REG, D_KBL | D_CFL, F_CMD_ACCESS, NULL, NULL);
-	MMIO_D(GEN9_CTX_PREEMPT_REG, D_SKL_PLUS);
+	MMIO_D(GEN9_CTX_PREEMPT_REG, D_SKL_PLUS & ~D_BXT);
 
 	return 0;
 }
@@ -3313,9 +3342,21 @@ static int init_bxt_mmio_info(struct intel_gvt *gvt)
 	MMIO_D(GEN8_PUSHBUS_SHIFT, D_BXT);
 	MMIO_D(GEN6_GFXPAUSE, D_BXT);
 	MMIO_DFH(GEN8_L3SQCREG1, D_BXT, F_CMD_ACCESS, NULL, NULL);
+	MMIO_DFH(GEN8_L3CNTLREG, D_BXT, F_CMD_ACCESS, NULL, NULL);
+	MMIO_DFH(_MMIO(0x20D8), D_BXT, F_CMD_ACCESS, NULL, NULL);
+	MMIO_F(GEN8_RING_CS_GPR(RENDER_RING_BASE, 0), 0x40, F_CMD_ACCESS,
+	       0, 0, D_BXT, NULL, NULL);
+	MMIO_F(GEN8_RING_CS_GPR(GEN6_BSD_RING_BASE, 0), 0x40, F_CMD_ACCESS,
+	       0, 0, D_BXT, NULL, NULL);
+	MMIO_F(GEN8_RING_CS_GPR(BLT_RING_BASE, 0), 0x40, F_CMD_ACCESS,
+	       0, 0, D_BXT, NULL, NULL);
+	MMIO_F(GEN8_RING_CS_GPR(VEBOX_RING_BASE, 0), 0x40, F_CMD_ACCESS,
+	       0, 0, D_BXT, NULL, NULL);
 
 	MMIO_DFH(GEN9_CTX_PREEMPT_REG, D_BXT, F_CMD_ACCESS, NULL, NULL);
 
+	MMIO_DH(GEN8_PRIVATE_PAT_LO, D_BXT, NULL, bxt_ppat_low_write);
+
 	return 0;
 }
 

drivers/gpu/drm/i915/gvt/scheduler.c

@@ -1277,7 +1277,7 @@ void intel_vgpu_clean_submission(struct intel_vgpu *vgpu)
 
 	i915_context_ppgtt_root_restore(s, i915_vm_to_ppgtt(s->shadow[0]->vm));
 	for_each_engine(engine, vgpu->gvt->gt, id)
-		intel_context_unpin(s->shadow[id]);
+		intel_context_put(s->shadow[id]);
 
 	kmem_cache_destroy(s->workloads);
 }
@@ -1369,11 +1369,6 @@ int intel_vgpu_setup_submission(struct intel_vgpu *vgpu)
 			ce->ring = __intel_context_ring_size(ring_size);
 		}
 
-		ret = intel_context_pin(ce);
-		intel_context_put(ce);
-		if (ret)
-			goto out_shadow_ctx;
-
 		s->shadow[i] = ce;
 	}
 
@@ -1405,7 +1400,6 @@ int intel_vgpu_setup_submission(struct intel_vgpu *vgpu)
 		if (IS_ERR(s->shadow[i]))
 			break;
 
-		intel_context_unpin(s->shadow[i]);
 		intel_context_put(s->shadow[i]);
 	}
 	i915_vm_put(&ppgtt->vm);
@@ -1479,6 +1473,7 @@ void intel_vgpu_destroy_workload(struct intel_vgpu_workload *workload)
 {
 	struct intel_vgpu_submission *s = &workload->vgpu->submission;
 
+	intel_context_unpin(s->shadow[workload->engine->id]);
 	release_shadow_batch_buffer(workload);
 	release_shadow_wa_ctx(&workload->wa_ctx);
 
@@ -1724,6 +1719,12 @@ intel_vgpu_create_workload(struct intel_vgpu *vgpu,
 		return ERR_PTR(ret);
 	}
 
+	ret = intel_context_pin(s->shadow[engine->id]);
+	if (ret) {
+		intel_vgpu_destroy_workload(workload);
+		return ERR_PTR(ret);
+	}
+
 	return workload;
 }
 

drivers/gpu/drm/i915/i915_vma.c

@@ -314,8 +314,10 @@ static void __vma_release(struct dma_fence_work *work)
 {
 	struct i915_vma_work *vw = container_of(work, typeof(*vw), base);
 
-	if (vw->pinned)
+	if (vw->pinned) {
 		__i915_gem_object_unpin_pages(vw->pinned);
+		i915_gem_object_put(vw->pinned);
+	}
 
 	i915_vm_free_pt_stash(vw->vm, &vw->stash);
 	i915_vm_put(vw->vm);
@@ -431,7 +433,7 @@ int i915_vma_bind(struct i915_vma *vma,
 
 		if (vma->obj) {
 			__i915_gem_object_pin_pages(vma->obj);
-			work->pinned = vma->obj;
+			work->pinned = i915_gem_object_get(vma->obj);
 		}
 	} else {
 		vma->ops->bind_vma(vma->vm, NULL, vma, cache_level, bind_flags);