Merge tag 'drm-intel-fixes-2017-12-07' of git://anongit.freedesktop.org/drm/drm-intel into drm-fixes

- Fix for fd.o bug #103997 CNL eDP + HDMI causing a machine hard hang (James)
- Fix to allow suspending with a wedged GPU to hopefully unwedge it (Chris)
- Fix for Gen2 vblank timestap/frame counter jumps (Ville)
- Revert of a W/A for enabling FBC on CNL/GLK for certain images
  and sizes (Rodrigo)
- Lockdep fix for i915 userptr code (Chris)

gvt-fixes-2017-12-06

- Fix invalid hw reg read value for vGPU (Xiong)
- Fix qemu warning on PCI ROM bar missing (Changbin)
- Workaround preemption regression (Zhenyu)

* tag 'drm-intel-fixes-2017-12-07' of git://anongit.freedesktop.org/drm/drm-intel:
  Revert "drm/i915: Display WA #1133 WaFbcSkipSegments:cnl, glk"
  drm/i915: Call i915_gem_init_userptr() before taking struct_mutex
  drm/i915/gvt: set max priority for gvt context
  drm/i915/gvt: Don't mark vgpu context as inactive when preempted
  drm/i915/gvt: Limit read hw reg to active vgpu
  drm/i915/gvt: Export intel_gvt_render_mmio_to_ring_id()
  drm/i915/gvt: Emulate PCI expansion ROM base address register
  drm/i915/cnl: Mask previous DDI - PLL mapping
  drm/i915: Fix vblank timestamp/frame counter jumps on gen2
  drm/i915: Skip switch-to-kernel-context on suspend when wedged

Author: airlied

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

@@ -208,6 +208,20 @@ static int emulate_pci_command_write(struct intel_vgpu *vgpu,
 	return 0;
 }
 
+static int emulate_pci_rom_bar_write(struct intel_vgpu *vgpu,
+	unsigned int offset, void *p_data, unsigned int bytes)
+{
+	u32 *pval = (u32 *)(vgpu_cfg_space(vgpu) + offset);
+	u32 new = *(u32 *)(p_data);
+
+	if ((new & PCI_ROM_ADDRESS_MASK) == PCI_ROM_ADDRESS_MASK)
+		/* We don't have rom, return size of 0. */
+		*pval = 0;
+	else
+		vgpu_pci_cfg_mem_write(vgpu, offset, p_data, bytes);
+	return 0;
+}
+
 static int emulate_pci_bar_write(struct intel_vgpu *vgpu, unsigned int offset,
 	void *p_data, unsigned int bytes)
 {
@@ -300,6 +314,11 @@ int intel_vgpu_emulate_cfg_write(struct intel_vgpu *vgpu, unsigned int offset,
 	}
 
 	switch (rounddown(offset, 4)) {
+	case PCI_ROM_ADDRESS:
+		if (WARN_ON(!IS_ALIGNED(offset, 4)))
+			return -EINVAL;
+		return emulate_pci_rom_bar_write(vgpu, offset, p_data, bytes);
+
 	case PCI_BASE_ADDRESS_0 ... PCI_BASE_ADDRESS_5:
 		if (WARN_ON(!IS_ALIGNED(offset, 4)))
 			return -EINVAL;
@@ -375,6 +394,8 @@ void intel_vgpu_init_cfg_space(struct intel_vgpu *vgpu,
 				pci_resource_len(gvt->dev_priv->drm.pdev, 0);
 	vgpu->cfg_space.bar[INTEL_GVT_PCI_BAR_APERTURE].size =
 				pci_resource_len(gvt->dev_priv->drm.pdev, 2);
+
+	memset(vgpu_cfg_space(vgpu) + PCI_ROM_ADDRESS, 0, 4);
 }
 
 /**

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

@@ -137,17 +137,26 @@ static int new_mmio_info(struct intel_gvt *gvt,
 	return 0;
 }
 
-static int render_mmio_to_ring_id(struct intel_gvt *gvt, unsigned int reg)
+/**
+ * intel_gvt_render_mmio_to_ring_id - convert a mmio offset into ring id
+ * @gvt: a GVT device
+ * @offset: register offset
+ *
+ * Returns:
+ * Ring ID on success, negative error code if failed.
+ */
+int intel_gvt_render_mmio_to_ring_id(struct intel_gvt *gvt,
+		unsigned int offset)
 {
 	enum intel_engine_id id;
 	struct intel_engine_cs *engine;
 
-	reg &= ~GENMASK(11, 0);
+	offset &= ~GENMASK(11, 0);
 	for_each_engine(engine, gvt->dev_priv, id) {
-		if (engine->mmio_base == reg)
+		if (engine->mmio_base == offset)
 			return id;
 	}
-	return -1;
+	return -ENODEV;
 }
 
 #define offset_to_fence_num(offset) \
@@ -1398,18 +1407,36 @@ static int skl_lcpll_write(struct intel_vgpu *vgpu, unsigned int offset,
 static int mmio_read_from_hw(struct intel_vgpu *vgpu,
 		unsigned int offset, void *p_data, unsigned int bytes)
 {
-	struct drm_i915_private *dev_priv = vgpu->gvt->dev_priv;
+	struct intel_gvt *gvt = vgpu->gvt;
+	struct drm_i915_private *dev_priv = gvt->dev_priv;
+	int ring_id;
+	u32 ring_base;
+
+	ring_id = intel_gvt_render_mmio_to_ring_id(gvt, offset);
+	/**
+	 * Read HW reg in following case
+	 * a. the offset isn't a ring mmio
+	 * b. the offset's ring is running on hw.
+	 * c. the offset is ring time stamp mmio
+	 */
+	if (ring_id >= 0)
+		ring_base = dev_priv->engine[ring_id]->mmio_base;
+
+	if (ring_id < 0 || vgpu  == gvt->scheduler.engine_owner[ring_id] ||
+	    offset == i915_mmio_reg_offset(RING_TIMESTAMP(ring_base)) ||
+	    offset == i915_mmio_reg_offset(RING_TIMESTAMP_UDW(ring_base))) {
+		mmio_hw_access_pre(dev_priv);
+		vgpu_vreg(vgpu, offset) = I915_READ(_MMIO(offset));
+		mmio_hw_access_post(dev_priv);
+	}
 
-	mmio_hw_access_pre(dev_priv);
-	vgpu_vreg(vgpu, offset) = I915_READ(_MMIO(offset));
-	mmio_hw_access_post(dev_priv);
 	return intel_vgpu_default_mmio_read(vgpu, offset, p_data, bytes);
 }
 
 static int elsp_mmio_write(struct intel_vgpu *vgpu, unsigned int offset,
 		void *p_data, unsigned int bytes)
 {
-	int ring_id = render_mmio_to_ring_id(vgpu->gvt, offset);
+	int ring_id = intel_gvt_render_mmio_to_ring_id(vgpu->gvt, offset);
 	struct intel_vgpu_execlist *execlist;
 	u32 data = *(u32 *)p_data;
 	int ret = 0;
@@ -1436,7 +1463,7 @@ static int ring_mode_mmio_write(struct intel_vgpu *vgpu, unsigned int offset,
 		void *p_data, unsigned int bytes)
 {
 	u32 data = *(u32 *)p_data;
-	int ring_id = render_mmio_to_ring_id(vgpu->gvt, offset);
+	int ring_id = intel_gvt_render_mmio_to_ring_id(vgpu->gvt, offset);
 	bool enable_execlist;
 
 	write_vreg(vgpu, offset, p_data, bytes);

drivers/gpu/drm/i915/gvt/mmio.h

@@ -65,6 +65,8 @@ struct intel_gvt_mmio_info {
 	struct hlist_node node;
 };
 
+int intel_gvt_render_mmio_to_ring_id(struct intel_gvt *gvt,
+		unsigned int reg);
 unsigned long intel_gvt_get_device_type(struct intel_gvt *gvt);
 bool intel_gvt_match_device(struct intel_gvt *gvt, unsigned long device);
 

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

@@ -131,6 +131,20 @@ static inline bool is_gvt_request(struct drm_i915_gem_request *req)
 	return i915_gem_context_force_single_submission(req->ctx);
 }
 
+static void save_ring_hw_state(struct intel_vgpu *vgpu, int ring_id)
+{
+	struct drm_i915_private *dev_priv = vgpu->gvt->dev_priv;
+	u32 ring_base = dev_priv->engine[ring_id]->mmio_base;
+	i915_reg_t reg;
+
+	reg = RING_INSTDONE(ring_base);
+	vgpu_vreg(vgpu, i915_mmio_reg_offset(reg)) = I915_READ_FW(reg);
+	reg = RING_ACTHD(ring_base);
+	vgpu_vreg(vgpu, i915_mmio_reg_offset(reg)) = I915_READ_FW(reg);
+	reg = RING_ACTHD_UDW(ring_base);
+	vgpu_vreg(vgpu, i915_mmio_reg_offset(reg)) = I915_READ_FW(reg);
+}
+
 static int shadow_context_status_change(struct notifier_block *nb,
 		unsigned long action, void *data)
 {
@@ -175,9 +189,12 @@ static int shadow_context_status_change(struct notifier_block *nb,
 		atomic_set(&workload->shadow_ctx_active, 1);
 		break;
 	case INTEL_CONTEXT_SCHEDULE_OUT:
-	case INTEL_CONTEXT_SCHEDULE_PREEMPTED:
+		save_ring_hw_state(workload->vgpu, ring_id);
 		atomic_set(&workload->shadow_ctx_active, 0);
 		break;
+	case INTEL_CONTEXT_SCHEDULE_PREEMPTED:
+		save_ring_hw_state(workload->vgpu, ring_id);
+		break;
 	default:
 		WARN_ON(1);
 		return NOTIFY_OK;
@@ -740,6 +757,9 @@ int intel_vgpu_init_gvt_context(struct intel_vgpu *vgpu)
 	if (IS_ERR(vgpu->shadow_ctx))
 		return PTR_ERR(vgpu->shadow_ctx);
 
+	if (INTEL_INFO(vgpu->gvt->dev_priv)->has_logical_ring_preemption)
+		vgpu->shadow_ctx->priority = INT_MAX;
+
 	vgpu->shadow_ctx->engine[RCS].initialised = true;
 
 	bitmap_zero(vgpu->shadow_ctx_desc_updated, I915_NUM_ENGINES);

drivers/gpu/drm/i915/i915_gem.c

@@ -4712,17 +4712,19 @@ int i915_gem_suspend(struct drm_i915_private *dev_priv)
 	 * state. Fortunately, the kernel_context is disposable and we do
 	 * not rely on its state.
 	 */
-	ret = i915_gem_switch_to_kernel_context(dev_priv);
-	if (ret)
-		goto err_unlock;
+	if (!i915_terminally_wedged(&dev_priv->gpu_error)) {
+		ret = i915_gem_switch_to_kernel_context(dev_priv);
+		if (ret)
+			goto err_unlock;
 
-	ret = i915_gem_wait_for_idle(dev_priv,
-				     I915_WAIT_INTERRUPTIBLE |
-				     I915_WAIT_LOCKED);
-	if (ret && ret != -EIO)
-		goto err_unlock;
+		ret = i915_gem_wait_for_idle(dev_priv,
+					     I915_WAIT_INTERRUPTIBLE |
+					     I915_WAIT_LOCKED);
+		if (ret && ret != -EIO)
+			goto err_unlock;
 
-	assert_kernel_context_is_current(dev_priv);
+		assert_kernel_context_is_current(dev_priv);
+	}
 	i915_gem_contexts_lost(dev_priv);
 	mutex_unlock(&dev->struct_mutex);
 
@@ -4946,8 +4948,6 @@ int i915_gem_init(struct drm_i915_private *dev_priv)
 {
 	int ret;
 
-	mutex_lock(&dev_priv->drm.struct_mutex);
-
 	/*
 	 * We need to fallback to 4K pages since gvt gtt handling doesn't
 	 * support huge page entries - we will need to check either hypervisor
@@ -4967,18 +4967,19 @@ int i915_gem_init(struct drm_i915_private *dev_priv)
 		dev_priv->gt.cleanup_engine = intel_logical_ring_cleanup;
 	}
 
+	ret = i915_gem_init_userptr(dev_priv);
+	if (ret)
+		return ret;
+
 	/* This is just a security blanket to placate dragons.
 	 * On some systems, we very sporadically observe that the first TLBs
 	 * used by the CS may be stale, despite us poking the TLB reset. If
 	 * we hold the forcewake during initialisation these problems
 	 * just magically go away.
 	 */
+	mutex_lock(&dev_priv->drm.struct_mutex);
 	intel_uncore_forcewake_get(dev_priv, FORCEWAKE_ALL);
 
-	ret = i915_gem_init_userptr(dev_priv);
-	if (ret)
-		goto out_unlock;
-
 	ret = i915_gem_init_ggtt(dev_priv);
 	if (ret)
 		goto out_unlock;

drivers/gpu/drm/i915/i915_reg.h

@@ -2951,9 +2951,6 @@ enum i915_power_well_id {
 #define ILK_DPFC_CHICKEN	_MMIO(0x43224)
 #define   ILK_DPFC_DISABLE_DUMMY0 (1<<8)
 #define   ILK_DPFC_NUKE_ON_ANY_MODIFICATION	(1<<23)
-#define   GLK_SKIP_SEG_EN		(1<<12)
-#define   GLK_SKIP_SEG_COUNT_MASK	(3<<10)
-#define   GLK_SKIP_SEG_COUNT(x)		((x)<<10)
 #define ILK_FBC_RT_BASE		_MMIO(0x2128)
 #define   ILK_FBC_RT_VALID	(1<<0)
 #define   SNB_FBC_FRONT_BUFFER	(1<<1)

drivers/gpu/drm/i915/intel_ddi.c

@@ -2131,6 +2131,7 @@ static void intel_ddi_clk_select(struct intel_encoder *encoder,
 	if (IS_CANNONLAKE(dev_priv)) {
 		/* Configure DPCLKA_CFGCR0 to map the DPLL to the DDI. */
 		val = I915_READ(DPCLKA_CFGCR0);
+		val &= ~DPCLKA_CFGCR0_DDI_CLK_SEL_MASK(port);
 		val |= DPCLKA_CFGCR0_DDI_CLK_SEL(pll->id, port);
 		I915_WRITE(DPCLKA_CFGCR0, val);
 

drivers/gpu/drm/i915/intel_display.c

@@ -1000,7 +1000,8 @@ enum transcoder intel_pipe_to_cpu_transcoder(struct drm_i915_private *dev_priv,
 	return crtc->config->cpu_transcoder;
 }
 
-static bool pipe_dsl_stopped(struct drm_i915_private *dev_priv, enum pipe pipe)
+static bool pipe_scanline_is_moving(struct drm_i915_private *dev_priv,
+				    enum pipe pipe)
 {
 	i915_reg_t reg = PIPEDSL(pipe);
 	u32 line1, line2;
@@ -1015,7 +1016,28 @@ static bool pipe_dsl_stopped(struct drm_i915_private *dev_priv, enum pipe pipe)
 	msleep(5);
 	line2 = I915_READ(reg) & line_mask;
 
-	return line1 == line2;
+	return line1 != line2;
+}
+
+static void wait_for_pipe_scanline_moving(struct intel_crtc *crtc, bool state)
+{
+	struct drm_i915_private *dev_priv = to_i915(crtc->base.dev);
+	enum pipe pipe = crtc->pipe;
+
+	/* Wait for the display line to settle/start moving */
+	if (wait_for(pipe_scanline_is_moving(dev_priv, pipe) == state, 100))
+		DRM_ERROR("pipe %c scanline %s wait timed out\n",
+			  pipe_name(pipe), onoff(state));
+}
+
+static void intel_wait_for_pipe_scanline_stopped(struct intel_crtc *crtc)
+{
+	wait_for_pipe_scanline_moving(crtc, false);
+}
+
+static void intel_wait_for_pipe_scanline_moving(struct intel_crtc *crtc)
+{
+	wait_for_pipe_scanline_moving(crtc, true);
 }
 
 /*
@@ -1038,7 +1060,6 @@ static void intel_wait_for_pipe_off(struct intel_crtc *crtc)
 {
 	struct drm_i915_private *dev_priv = to_i915(crtc->base.dev);
 	enum transcoder cpu_transcoder = crtc->config->cpu_transcoder;
-	enum pipe pipe = crtc->pipe;
 
 	if (INTEL_GEN(dev_priv) >= 4) {
 		i915_reg_t reg = PIPECONF(cpu_transcoder);
@@ -1049,9 +1070,7 @@ static void intel_wait_for_pipe_off(struct intel_crtc *crtc)
 					    100))
 			WARN(1, "pipe_off wait timed out\n");
 	} else {
-		/* Wait for the display line to settle */
-		if (wait_for(pipe_dsl_stopped(dev_priv, pipe), 100))
-			WARN(1, "pipe_off wait timed out\n");
+		intel_wait_for_pipe_scanline_stopped(crtc);
 	}
 }
 
@@ -1936,15 +1955,14 @@ static void intel_enable_pipe(struct intel_crtc *crtc)
 	POSTING_READ(reg);
 
 	/*
-	 * Until the pipe starts DSL will read as 0, which would cause
-	 * an apparent vblank timestamp jump, which messes up also the
-	 * frame count when it's derived from the timestamps. So let's
-	 * wait for the pipe to start properly before we call
-	 * drm_crtc_vblank_on()
+	 * Until the pipe starts PIPEDSL reads will return a stale value,
+	 * which causes an apparent vblank timestamp jump when PIPEDSL
+	 * resets to its proper value. That also messes up the frame count
+	 * when it's derived from the timestamps. So let's wait for the
+	 * pipe to start properly before we call drm_crtc_vblank_on()
 	 */
-	if (dev->max_vblank_count == 0 &&
-	    wait_for(intel_get_crtc_scanline(crtc) != crtc->scanline_offset, 50))
-		DRM_ERROR("pipe %c didn't start\n", pipe_name(pipe));
+	if (dev->max_vblank_count == 0)
+		intel_wait_for_pipe_scanline_moving(crtc);
 }
 
 /**
@@ -14643,6 +14661,8 @@ void i830_enable_pipe(struct drm_i915_private *dev_priv, enum pipe pipe)
 
 void i830_disable_pipe(struct drm_i915_private *dev_priv, enum pipe pipe)
 {
+	struct intel_crtc *crtc = intel_get_crtc_for_pipe(dev_priv, pipe);
+
 	DRM_DEBUG_KMS("disabling pipe %c due to force quirk\n",
 		      pipe_name(pipe));
 
@@ -14652,8 +14672,7 @@ void i830_disable_pipe(struct drm_i915_private *dev_priv, enum pipe pipe)
 	I915_WRITE(PIPECONF(pipe), 0);
 	POSTING_READ(PIPECONF(pipe));
 
-	if (wait_for(pipe_dsl_stopped(dev_priv, pipe), 100))
-		DRM_ERROR("pipe %c off wait timed out\n", pipe_name(pipe));
+	intel_wait_for_pipe_scanline_stopped(crtc);
 
 	I915_WRITE(DPLL(pipe), DPLL_VGA_MODE_DIS);
 	POSTING_READ(DPLL(pipe));

drivers/gpu/drm/i915/intel_pm.c

@@ -124,7 +124,6 @@ static void bxt_init_clock_gating(struct drm_i915_private *dev_priv)
 
 static void glk_init_clock_gating(struct drm_i915_private *dev_priv)
 {
-	u32 val;
 	gen9_init_clock_gating(dev_priv);
 
 	/*
@@ -144,11 +143,6 @@ static void glk_init_clock_gating(struct drm_i915_private *dev_priv)
 		I915_WRITE(CHICKEN_MISC_2, val);
 	}
 
-	/* Display WA #1133: WaFbcSkipSegments:glk */
-	val = I915_READ(ILK_DPFC_CHICKEN);
-	val &= ~GLK_SKIP_SEG_COUNT_MASK;
-	val |= GLK_SKIP_SEG_EN | GLK_SKIP_SEG_COUNT(1);
-	I915_WRITE(ILK_DPFC_CHICKEN, val);
 }
 
 static void i915_pineview_get_mem_freq(struct drm_i915_private *dev_priv)
@@ -8517,7 +8511,6 @@ static void cnp_init_clock_gating(struct drm_i915_private *dev_priv)
 
 static void cnl_init_clock_gating(struct drm_i915_private *dev_priv)
 {
-	u32 val;
 	cnp_init_clock_gating(dev_priv);
 
 	/* This is not an Wa. Enable for better image quality */
@@ -8537,12 +8530,6 @@ static void cnl_init_clock_gating(struct drm_i915_private *dev_priv)
 		I915_WRITE(SLICE_UNIT_LEVEL_CLKGATE,
 			   I915_READ(SLICE_UNIT_LEVEL_CLKGATE) |
 			   SARBUNIT_CLKGATE_DIS);
-
-	/* Display WA #1133: WaFbcSkipSegments:cnl */
-	val = I915_READ(ILK_DPFC_CHICKEN);
-	val &= ~GLK_SKIP_SEG_COUNT_MASK;
-	val |= GLK_SKIP_SEG_EN | GLK_SKIP_SEG_COUNT(1);
-	I915_WRITE(ILK_DPFC_CHICKEN, val);
 }
 
 static void cfl_init_clock_gating(struct drm_i915_private *dev_priv)