Merge tag 'gvt-fixes-2020-10-30' of https://github.com/intel/gvt-linux into drm-intel-fixes

gvt-fixes-2020-10-30

- Fix HWSP reset handling during vGPU suspend/resume (Colin)
- Apply flush workaround on APL now for possible guest hang (Colin)
- Fix vGPU context pin/unpin also for host suspend regression with
  vGPU created (Colin)
- more BXT/APL mmio cmd access fixes (Colin)

Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
From: Zhenyu Wang <zhenyuw@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20201030052117.GC27141@zhen-hp.sh.intel.com

2 files changed, 52 insertions, 10 deletions

diff --git a/drivers/gpu/drm/i915/gvt/handlers.c b/drivers/gpu/drm/i915/gvt/handlers.c
index 3be37e6fe33d..eb342a759943 100644
--- a/drivers/gpu/drm/i915/gvt/handlers.c
+++ b/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;
 }
 
diff --git a/drivers/gpu/drm/i915/gvt/scheduler.c b/drivers/gpu/drm/i915/gvt/scheduler.c
index 1570eb8aa978..aed2ef6466a2 100644
--- a/drivers/gpu/drm/i915/gvt/scheduler.c
+++ b/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 @@ out_shadow_ctx:
 		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;
 }