Merge tag 'gvt-fixes-2017-12-06' of https://github.com/intel/gvt-linux into drm-intel-fixes

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)

Signed-off-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20171206075105.wlh2ojubjczlstox@zhen-hp.sh.intel.com

4 files changed, 81 insertions, 11 deletions

diff --git a/drivers/gpu/drm/i915/gvt/cfg_space.c b/drivers/gpu/drm/i915/gvt/cfg_space.c
index ab19545d59a1..4ce2e6bd0680 100644
--- a/drivers/gpu/drm/i915/gvt/cfg_space.c
+++ b/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);
 }
 
 /**
diff --git a/drivers/gpu/drm/i915/gvt/handlers.c b/drivers/gpu/drm/i915/gvt/handlers.c
index 44cd5ff5e97d..1f840f6b81bb 100644
--- a/drivers/gpu/drm/i915/gvt/handlers.c
+++ b/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);
diff --git a/drivers/gpu/drm/i915/gvt/mmio.h b/drivers/gpu/drm/i915/gvt/mmio.h
index 32cd64ddad26..dbc04ad2c7a1 100644
--- a/drivers/gpu/drm/i915/gvt/mmio.h
+++ b/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);
 
diff --git a/drivers/gpu/drm/i915/gvt/scheduler.c b/drivers/gpu/drm/i915/gvt/scheduler.c
index 3ac1dc97a7a0..69f8f0d155b9 100644
--- a/drivers/gpu/drm/i915/gvt/scheduler.c
+++ b/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);