Merge tag 'drm-misc-next-2017-09-20' of git://anongit.freedesktop.org/git/drm-misc...
[linux-2.6-block.git] / drivers / gpu / drm / i915 / intel_display.c
index 3f505682963fc15083ea724c25f9efab4b0c9e4a..4b34fa6954f9997b01aa5953171dbd7d0a49dde2 100644 (file)
 #include <linux/dma_remapping.h>
 #include <linux/reservation.h>
 
-static bool is_mmio_work(struct intel_flip_work *work)
-{
-       return work->mmio_work.func;
-}
-
 /* Primary plane formats for gen <= 3 */
 static const uint32_t i8xx_primary_formats[] = {
        DRM_FORMAT_C8,
@@ -72,6 +67,12 @@ static const uint32_t i965_primary_formats[] = {
        DRM_FORMAT_XBGR2101010,
 };
 
+static const uint64_t i9xx_format_modifiers[] = {
+       I915_FORMAT_MOD_X_TILED,
+       DRM_FORMAT_MOD_LINEAR,
+       DRM_FORMAT_MOD_INVALID
+};
+
 static const uint32_t skl_primary_formats[] = {
        DRM_FORMAT_C8,
        DRM_FORMAT_RGB565,
@@ -87,11 +88,34 @@ static const uint32_t skl_primary_formats[] = {
        DRM_FORMAT_VYUY,
 };
 
+static const uint64_t skl_format_modifiers_noccs[] = {
+       I915_FORMAT_MOD_Yf_TILED,
+       I915_FORMAT_MOD_Y_TILED,
+       I915_FORMAT_MOD_X_TILED,
+       DRM_FORMAT_MOD_LINEAR,
+       DRM_FORMAT_MOD_INVALID
+};
+
+static const uint64_t skl_format_modifiers_ccs[] = {
+       I915_FORMAT_MOD_Yf_TILED_CCS,
+       I915_FORMAT_MOD_Y_TILED_CCS,
+       I915_FORMAT_MOD_Yf_TILED,
+       I915_FORMAT_MOD_Y_TILED,
+       I915_FORMAT_MOD_X_TILED,
+       DRM_FORMAT_MOD_LINEAR,
+       DRM_FORMAT_MOD_INVALID
+};
+
 /* Cursor formats */
 static const uint32_t intel_cursor_formats[] = {
        DRM_FORMAT_ARGB8888,
 };
 
+static const uint64_t cursor_format_modifiers[] = {
+       DRM_FORMAT_MOD_LINEAR,
+       DRM_FORMAT_MOD_INVALID
+};
+
 static void i9xx_crtc_clock_get(struct intel_crtc *crtc,
                                struct intel_crtc_state *pipe_config);
 static void ironlake_pch_clock_get(struct intel_crtc *crtc,
@@ -1777,7 +1801,7 @@ static void lpt_enable_pch_transcoder(struct drm_i915_private *dev_priv,
 
        /* FDI must be feeding us bits for PCH ports */
        assert_fdi_tx_enabled(dev_priv, (enum pipe) cpu_transcoder);
-       assert_fdi_rx_enabled(dev_priv, TRANSCODER_A);
+       assert_fdi_rx_enabled(dev_priv, PIPE_A);
 
        /* Workaround: set timing override bit. */
        val = I915_READ(TRANS_CHICKEN2(PIPE_A));
@@ -1853,16 +1877,16 @@ void lpt_disable_pch_transcoder(struct drm_i915_private *dev_priv)
        I915_WRITE(TRANS_CHICKEN2(PIPE_A), val);
 }
 
-enum transcoder intel_crtc_pch_transcoder(struct intel_crtc *crtc)
+enum pipe intel_crtc_pch_transcoder(struct intel_crtc *crtc)
 {
        struct drm_i915_private *dev_priv = to_i915(crtc->base.dev);
 
        WARN_ON(!crtc->config->has_pch_encoder);
 
        if (HAS_PCH_LPT(dev_priv))
-               return TRANSCODER_A;
+               return PIPE_A;
        else
-               return (enum transcoder) crtc->pipe;
+               return crtc->pipe;
 }
 
 /**
@@ -1901,7 +1925,7 @@ static void intel_enable_pipe(struct intel_crtc *crtc)
                if (crtc->config->has_pch_encoder) {
                        /* if driving the PCH, we need FDI enabled */
                        assert_fdi_rx_pll_enabled(dev_priv,
-                                                 (enum pipe) intel_crtc_pch_transcoder(crtc));
+                                                 intel_crtc_pch_transcoder(crtc));
                        assert_fdi_tx_pll_enabled(dev_priv,
                                                  (enum pipe) cpu_transcoder);
                }
@@ -1999,11 +2023,19 @@ intel_tile_width_bytes(const struct drm_framebuffer *fb, int plane)
                        return 128;
                else
                        return 512;
+       case I915_FORMAT_MOD_Y_TILED_CCS:
+               if (plane == 1)
+                       return 128;
+               /* fall through */
        case I915_FORMAT_MOD_Y_TILED:
                if (IS_GEN2(dev_priv) || HAS_128_BYTE_Y_TILING(dev_priv))
                        return 128;
                else
                        return 512;
+       case I915_FORMAT_MOD_Yf_TILED_CCS:
+               if (plane == 1)
+                       return 128;
+               /* fall through */
        case I915_FORMAT_MOD_Yf_TILED:
                switch (cpp) {
                case 1:
@@ -2110,7 +2142,7 @@ static unsigned int intel_surf_alignment(const struct drm_framebuffer *fb,
        struct drm_i915_private *dev_priv = to_i915(fb->dev);
 
        /* AUX_DIST needs only 4K alignment */
-       if (fb->format->format == DRM_FORMAT_NV12 && plane == 1)
+       if (plane == 1)
                return 4096;
 
        switch (fb->modifier) {
@@ -2120,6 +2152,8 @@ static unsigned int intel_surf_alignment(const struct drm_framebuffer *fb,
                if (INTEL_GEN(dev_priv) >= 9)
                        return 256 * 1024;
                return 0;
+       case I915_FORMAT_MOD_Y_TILED_CCS:
+       case I915_FORMAT_MOD_Yf_TILED_CCS:
        case I915_FORMAT_MOD_Y_TILED:
        case I915_FORMAT_MOD_Yf_TILED:
                return 1 * 1024 * 1024;
@@ -2162,6 +2196,8 @@ intel_pin_and_fence_fb_obj(struct drm_framebuffer *fb, unsigned int rotation)
         */
        intel_runtime_pm_get(dev_priv);
 
+       atomic_inc(&dev_priv->gpu_error.pending_fb_pin);
+
        vma = i915_gem_object_pin_to_display_plane(obj, alignment, &view);
        if (IS_ERR(vma))
                goto err;
@@ -2189,6 +2225,8 @@ intel_pin_and_fence_fb_obj(struct drm_framebuffer *fb, unsigned int rotation)
 
        i915_vma_get(vma);
 err:
+       atomic_dec(&dev_priv->gpu_error.pending_fb_pin);
+
        intel_runtime_pm_put(dev_priv);
        return vma;
 }
@@ -2250,17 +2288,13 @@ void intel_add_fb_offsets(int *x, int *y,
        }
 }
 
-/*
- * Input tile dimensions and pitch must already be
- * rotated to match x and y, and in pixel units.
- */
-static u32 _intel_adjust_tile_offset(int *x, int *y,
-                                    unsigned int tile_width,
-                                    unsigned int tile_height,
-                                    unsigned int tile_size,
-                                    unsigned int pitch_tiles,
-                                    u32 old_offset,
-                                    u32 new_offset)
+static u32 __intel_adjust_tile_offset(int *x, int *y,
+                                     unsigned int tile_width,
+                                     unsigned int tile_height,
+                                     unsigned int tile_size,
+                                     unsigned int pitch_tiles,
+                                     u32 old_offset,
+                                     u32 new_offset)
 {
        unsigned int pitch_pixels = pitch_tiles * tile_width;
        unsigned int tiles;
@@ -2281,18 +2315,13 @@ static u32 _intel_adjust_tile_offset(int *x, int *y,
        return new_offset;
 }
 
-/*
- * Adjust the tile offset by moving the difference into
- * the x/y offsets.
- */
-static u32 intel_adjust_tile_offset(int *x, int *y,
-                                   const struct intel_plane_state *state, int plane,
-                                   u32 old_offset, u32 new_offset)
+static u32 _intel_adjust_tile_offset(int *x, int *y,
+                                    const struct drm_framebuffer *fb, int plane,
+                                    unsigned int rotation,
+                                    u32 old_offset, u32 new_offset)
 {
-       const struct drm_i915_private *dev_priv = to_i915(state->base.plane->dev);
-       const struct drm_framebuffer *fb = state->base.fb;
+       const struct drm_i915_private *dev_priv = to_i915(fb->dev);
        unsigned int cpp = fb->format->cpp[plane];
-       unsigned int rotation = state->base.rotation;
        unsigned int pitch = intel_fb_pitch(fb, plane, rotation);
 
        WARN_ON(new_offset > old_offset);
@@ -2311,9 +2340,9 @@ static u32 intel_adjust_tile_offset(int *x, int *y,
                        pitch_tiles = pitch / (tile_width * cpp);
                }
 
-               _intel_adjust_tile_offset(x, y, tile_width, tile_height,
-                                         tile_size, pitch_tiles,
-                                         old_offset, new_offset);
+               __intel_adjust_tile_offset(x, y, tile_width, tile_height,
+                                          tile_size, pitch_tiles,
+                                          old_offset, new_offset);
        } else {
                old_offset += *y * pitch + *x * cpp;
 
@@ -2324,6 +2353,19 @@ static u32 intel_adjust_tile_offset(int *x, int *y,
        return new_offset;
 }
 
+/*
+ * Adjust the tile offset by moving the difference into
+ * the x/y offsets.
+ */
+static u32 intel_adjust_tile_offset(int *x, int *y,
+                                   const struct intel_plane_state *state, int plane,
+                                   u32 old_offset, u32 new_offset)
+{
+       return _intel_adjust_tile_offset(x, y, state->base.fb, plane,
+                                        state->base.rotation,
+                                        old_offset, new_offset);
+}
+
 /*
  * Computes the linear offset to the base tile and adjusts
  * x, y. bytes per pixel is assumed to be a power-of-two.
@@ -2375,9 +2417,9 @@ static u32 _intel_compute_tile_offset(const struct drm_i915_private *dev_priv,
                offset = (tile_rows * pitch_tiles + tiles) * tile_size;
                offset_aligned = offset & ~alignment;
 
-               _intel_adjust_tile_offset(x, y, tile_width, tile_height,
-                                         tile_size, pitch_tiles,
-                                         offset, offset_aligned);
+               __intel_adjust_tile_offset(x, y, tile_width, tile_height,
+                                          tile_size, pitch_tiles,
+                                          offset, offset_aligned);
        } else {
                offset = *y * pitch + *x * cpp;
                offset_aligned = offset & ~alignment;
@@ -2409,16 +2451,24 @@ u32 intel_compute_tile_offset(int *x, int *y,
                                          rotation, alignment);
 }
 
-/* Convert the fb->offset[] linear offset into x/y offsets */
-static void intel_fb_offset_to_xy(int *x, int *y,
-                                 const struct drm_framebuffer *fb, int plane)
+/* Convert the fb->offset[] into x/y offsets */
+static int intel_fb_offset_to_xy(int *x, int *y,
+                                const struct drm_framebuffer *fb, int plane)
 {
-       unsigned int cpp = fb->format->cpp[plane];
-       unsigned int pitch = fb->pitches[plane];
-       u32 linear_offset = fb->offsets[plane];
+       struct drm_i915_private *dev_priv = to_i915(fb->dev);
+
+       if (fb->modifier != DRM_FORMAT_MOD_LINEAR &&
+           fb->offsets[plane] % intel_tile_size(dev_priv))
+               return -EINVAL;
+
+       *x = 0;
+       *y = 0;
 
-       *y = linear_offset / pitch;
-       *x = linear_offset % pitch / cpp;
+       _intel_adjust_tile_offset(x, y,
+                                 fb, plane, DRM_MODE_ROTATE_0,
+                                 fb->offsets[plane], 0);
+
+       return 0;
 }
 
 static unsigned int intel_fb_modifier_to_tiling(uint64_t fb_modifier)
@@ -2427,12 +2477,48 @@ static unsigned int intel_fb_modifier_to_tiling(uint64_t fb_modifier)
        case I915_FORMAT_MOD_X_TILED:
                return I915_TILING_X;
        case I915_FORMAT_MOD_Y_TILED:
+       case I915_FORMAT_MOD_Y_TILED_CCS:
                return I915_TILING_Y;
        default:
                return I915_TILING_NONE;
        }
 }
 
+static const struct drm_format_info ccs_formats[] = {
+       { .format = DRM_FORMAT_XRGB8888, .depth = 24, .num_planes = 2, .cpp = { 4, 1, }, .hsub = 8, .vsub = 16, },
+       { .format = DRM_FORMAT_XBGR8888, .depth = 24, .num_planes = 2, .cpp = { 4, 1, }, .hsub = 8, .vsub = 16, },
+       { .format = DRM_FORMAT_ARGB8888, .depth = 32, .num_planes = 2, .cpp = { 4, 1, }, .hsub = 8, .vsub = 16, },
+       { .format = DRM_FORMAT_ABGR8888, .depth = 32, .num_planes = 2, .cpp = { 4, 1, }, .hsub = 8, .vsub = 16, },
+};
+
+static const struct drm_format_info *
+lookup_format_info(const struct drm_format_info formats[],
+                  int num_formats, u32 format)
+{
+       int i;
+
+       for (i = 0; i < num_formats; i++) {
+               if (formats[i].format == format)
+                       return &formats[i];
+       }
+
+       return NULL;
+}
+
+static const struct drm_format_info *
+intel_get_format_info(const struct drm_mode_fb_cmd2 *cmd)
+{
+       switch (cmd->modifier[0]) {
+       case I915_FORMAT_MOD_Y_TILED_CCS:
+       case I915_FORMAT_MOD_Yf_TILED_CCS:
+               return lookup_format_info(ccs_formats,
+                                         ARRAY_SIZE(ccs_formats),
+                                         cmd->pixel_format);
+       default:
+               return NULL;
+       }
+}
+
 static int
 intel_fill_fb_info(struct drm_i915_private *dev_priv,
                   struct drm_framebuffer *fb)
@@ -2449,12 +2535,50 @@ intel_fill_fb_info(struct drm_i915_private *dev_priv,
                unsigned int cpp, size;
                u32 offset;
                int x, y;
+               int ret;
 
                cpp = fb->format->cpp[i];
                width = drm_framebuffer_plane_width(fb->width, fb, i);
                height = drm_framebuffer_plane_height(fb->height, fb, i);
 
-               intel_fb_offset_to_xy(&x, &y, fb, i);
+               ret = intel_fb_offset_to_xy(&x, &y, fb, i);
+               if (ret) {
+                       DRM_DEBUG_KMS("bad fb plane %d offset: 0x%x\n",
+                                     i, fb->offsets[i]);
+                       return ret;
+               }
+
+               if ((fb->modifier == I915_FORMAT_MOD_Y_TILED_CCS ||
+                    fb->modifier == I915_FORMAT_MOD_Yf_TILED_CCS) && i == 1) {
+                       int hsub = fb->format->hsub;
+                       int vsub = fb->format->vsub;
+                       int tile_width, tile_height;
+                       int main_x, main_y;
+                       int ccs_x, ccs_y;
+
+                       intel_tile_dims(fb, i, &tile_width, &tile_height);
+                       tile_width *= hsub;
+                       tile_height *= vsub;
+
+                       ccs_x = (x * hsub) % tile_width;
+                       ccs_y = (y * vsub) % tile_height;
+                       main_x = intel_fb->normal[0].x % tile_width;
+                       main_y = intel_fb->normal[0].y % tile_height;
+
+                       /*
+                        * CCS doesn't have its own x/y offset register, so the intra CCS tile
+                        * x/y offsets must match between CCS and the main surface.
+                        */
+                       if (main_x != ccs_x || main_y != ccs_y) {
+                               DRM_DEBUG_KMS("Bad CCS x/y (main %d,%d ccs %d,%d) full (main %d,%d ccs %d,%d)\n",
+                                             main_x, main_y,
+                                             ccs_x, ccs_y,
+                                             intel_fb->normal[0].x,
+                                             intel_fb->normal[0].y,
+                                             x, y);
+                               return -EINVAL;
+                       }
+               }
 
                /*
                 * The fence (if used) is aligned to the start of the object
@@ -2465,7 +2589,7 @@ intel_fill_fb_info(struct drm_i915_private *dev_priv,
                 * fb layout agrees with the fence layout. We already check that the
                 * fb stride matches the fence stride elsewhere.
                 */
-               if (i915_gem_object_is_tiled(intel_fb->obj) &&
+               if (i == 0 && i915_gem_object_is_tiled(intel_fb->obj) &&
                    (x + width) * cpp > fb->pitches[i]) {
                        DRM_DEBUG_KMS("bad fb plane %d offset: 0x%x\n",
                                      i, fb->offsets[i]);
@@ -2528,10 +2652,10 @@ intel_fill_fb_info(struct drm_i915_private *dev_priv,
                         * We only keep the x/y offsets, so push all of the
                         * gtt offset into the x/y offsets.
                         */
-                       _intel_adjust_tile_offset(&x, &y,
-                                                 tile_width, tile_height,
-                                                 tile_size, pitch_tiles,
-                                                 gtt_offset_rotated * tile_size, 0);
+                       __intel_adjust_tile_offset(&x, &y,
+                                                  tile_width, tile_height,
+                                                  tile_size, pitch_tiles,
+                                                  gtt_offset_rotated * tile_size, 0);
 
                        gtt_offset_rotated += rot_info->plane[i].width * rot_info->plane[i].height;
 
@@ -2664,20 +2788,6 @@ out_unref_obj:
        return false;
 }
 
-/* Update plane->state->fb to match plane->fb after driver-internal updates */
-static void
-update_state_fb(struct drm_plane *plane)
-{
-       if (plane->fb == plane->state->fb)
-               return;
-
-       if (plane->state->fb)
-               drm_framebuffer_unreference(plane->state->fb);
-       plane->state->fb = plane->fb;
-       if (plane->state->fb)
-               drm_framebuffer_reference(plane->state->fb);
-}
-
 static void
 intel_set_plane_visible(struct intel_crtc_state *crtc_state,
                        struct intel_plane_state *plane_state,
@@ -2830,6 +2940,9 @@ static int skl_max_plane_width(const struct drm_framebuffer *fb, int plane,
                        break;
                }
                break;
+       case I915_FORMAT_MOD_Y_TILED_CCS:
+       case I915_FORMAT_MOD_Yf_TILED_CCS:
+               /* FIXME AUX plane? */
        case I915_FORMAT_MOD_Y_TILED:
        case I915_FORMAT_MOD_Yf_TILED:
                switch (cpp) {
@@ -2852,6 +2965,44 @@ static int skl_max_plane_width(const struct drm_framebuffer *fb, int plane,
        return 2048;
 }
 
+static bool skl_check_main_ccs_coordinates(struct intel_plane_state *plane_state,
+                                          int main_x, int main_y, u32 main_offset)
+{
+       const struct drm_framebuffer *fb = plane_state->base.fb;
+       int hsub = fb->format->hsub;
+       int vsub = fb->format->vsub;
+       int aux_x = plane_state->aux.x;
+       int aux_y = plane_state->aux.y;
+       u32 aux_offset = plane_state->aux.offset;
+       u32 alignment = intel_surf_alignment(fb, 1);
+
+       while (aux_offset >= main_offset && aux_y <= main_y) {
+               int x, y;
+
+               if (aux_x == main_x && aux_y == main_y)
+                       break;
+
+               if (aux_offset == 0)
+                       break;
+
+               x = aux_x / hsub;
+               y = aux_y / vsub;
+               aux_offset = intel_adjust_tile_offset(&x, &y, plane_state, 1,
+                                                     aux_offset, aux_offset - alignment);
+               aux_x = x * hsub + aux_x % hsub;
+               aux_y = y * vsub + aux_y % vsub;
+       }
+
+       if (aux_x != main_x || aux_y != main_y)
+               return false;
+
+       plane_state->aux.offset = aux_offset;
+       plane_state->aux.x = aux_x;
+       plane_state->aux.y = aux_y;
+
+       return true;
+}
+
 static int skl_check_main_surface(struct intel_plane_state *plane_state)
 {
        const struct drm_framebuffer *fb = plane_state->base.fb;
@@ -2894,7 +3045,7 @@ static int skl_check_main_surface(struct intel_plane_state *plane_state)
 
                while ((x + w) * cpp > fb->pitches[0]) {
                        if (offset == 0) {
-                               DRM_DEBUG_KMS("Unable to find suitable display surface offset\n");
+                               DRM_DEBUG_KMS("Unable to find suitable display surface offset due to X-tiling\n");
                                return -EINVAL;
                        }
 
@@ -2903,6 +3054,26 @@ static int skl_check_main_surface(struct intel_plane_state *plane_state)
                }
        }
 
+       /*
+        * CCS AUX surface doesn't have its own x/y offsets, we must make sure
+        * they match with the main surface x/y offsets.
+        */
+       if (fb->modifier == I915_FORMAT_MOD_Y_TILED_CCS ||
+           fb->modifier == I915_FORMAT_MOD_Yf_TILED_CCS) {
+               while (!skl_check_main_ccs_coordinates(plane_state, x, y, offset)) {
+                       if (offset == 0)
+                               break;
+
+                       offset = intel_adjust_tile_offset(&x, &y, plane_state, 0,
+                                                         offset, offset - alignment);
+               }
+
+               if (x != plane_state->aux.x || y != plane_state->aux.y) {
+                       DRM_DEBUG_KMS("Unable to find suitable display surface offset due to CCS\n");
+                       return -EINVAL;
+               }
+       }
+
        plane_state->main.offset = offset;
        plane_state->main.x = x;
        plane_state->main.y = y;
@@ -2939,6 +3110,49 @@ static int skl_check_nv12_aux_surface(struct intel_plane_state *plane_state)
        return 0;
 }
 
+static int skl_check_ccs_aux_surface(struct intel_plane_state *plane_state)
+{
+       struct intel_plane *plane = to_intel_plane(plane_state->base.plane);
+       struct intel_crtc *crtc = to_intel_crtc(plane_state->base.crtc);
+       const struct drm_framebuffer *fb = plane_state->base.fb;
+       int src_x = plane_state->base.src.x1 >> 16;
+       int src_y = plane_state->base.src.y1 >> 16;
+       int hsub = fb->format->hsub;
+       int vsub = fb->format->vsub;
+       int x = src_x / hsub;
+       int y = src_y / vsub;
+       u32 offset;
+
+       switch (plane->id) {
+       case PLANE_PRIMARY:
+       case PLANE_SPRITE0:
+               break;
+       default:
+               DRM_DEBUG_KMS("RC support only on plane 1 and 2\n");
+               return -EINVAL;
+       }
+
+       if (crtc->pipe == PIPE_C) {
+               DRM_DEBUG_KMS("No RC support on pipe C\n");
+               return -EINVAL;
+       }
+
+       if (plane_state->base.rotation & ~(DRM_MODE_ROTATE_0 | DRM_MODE_ROTATE_180)) {
+               DRM_DEBUG_KMS("RC support only with 0/180 degree rotation %x\n",
+                             plane_state->base.rotation);
+               return -EINVAL;
+       }
+
+       intel_add_fb_offsets(&x, &y, plane_state, 1);
+       offset = intel_compute_tile_offset(&x, &y, plane_state, 1);
+
+       plane_state->aux.offset = offset;
+       plane_state->aux.x = x * hsub + src_x % hsub;
+       plane_state->aux.y = y * vsub + src_y % vsub;
+
+       return 0;
+}
+
 int skl_check_plane_surface(struct intel_plane_state *plane_state)
 {
        const struct drm_framebuffer *fb = plane_state->base.fb;
@@ -2962,6 +3176,11 @@ int skl_check_plane_surface(struct intel_plane_state *plane_state)
                ret = skl_check_nv12_aux_surface(plane_state);
                if (ret)
                        return ret;
+       } else if (fb->modifier == I915_FORMAT_MOD_Y_TILED_CCS ||
+                  fb->modifier == I915_FORMAT_MOD_Yf_TILED_CCS) {
+               ret = skl_check_ccs_aux_surface(plane_state);
+               if (ret)
+                       return ret;
        } else {
                plane_state->aux.offset = ~0xfff;
                plane_state->aux.x = 0;
@@ -3268,8 +3487,12 @@ static u32 skl_plane_ctl_tiling(uint64_t fb_modifier)
                return PLANE_CTL_TILED_X;
        case I915_FORMAT_MOD_Y_TILED:
                return PLANE_CTL_TILED_Y;
+       case I915_FORMAT_MOD_Y_TILED_CCS:
+               return PLANE_CTL_TILED_Y | PLANE_CTL_DECOMPRESSION_ENABLE;
        case I915_FORMAT_MOD_Yf_TILED:
                return PLANE_CTL_TILED_YF;
+       case I915_FORMAT_MOD_Yf_TILED_CCS:
+               return PLANE_CTL_TILED_YF | PLANE_CTL_DECOMPRESSION_ENABLE;
        default:
                MISSING_CASE(fb_modifier);
        }
@@ -3342,6 +3565,7 @@ static void skylake_update_primary_plane(struct intel_plane *plane,
        u32 plane_ctl = plane_state->ctl;
        unsigned int rotation = plane_state->base.rotation;
        u32 stride = skl_plane_stride(fb, 0, rotation);
+       u32 aux_stride = skl_plane_stride(fb, 1, rotation);
        u32 surf_addr = plane_state->main.offset;
        int scaler_id = plane_state->scaler_id;
        int src_x = plane_state->main.x;
@@ -3378,6 +3602,10 @@ static void skylake_update_primary_plane(struct intel_plane *plane,
        I915_WRITE_FW(PLANE_OFFSET(pipe, plane_id), (src_y << 16) | src_x);
        I915_WRITE_FW(PLANE_STRIDE(pipe, plane_id), stride);
        I915_WRITE_FW(PLANE_SIZE(pipe, plane_id), (src_h << 16) | src_w);
+       I915_WRITE_FW(PLANE_AUX_DIST(pipe, plane_id),
+                     (plane_state->aux.offset - surf_addr) | aux_stride);
+       I915_WRITE_FW(PLANE_AUX_OFFSET(pipe, plane_id),
+                     (plane_state->aux.y << 16) | plane_state->aux.x);
 
        if (scaler_id >= 0) {
                uint32_t ps_ctrl = 0;
@@ -3419,14 +3647,6 @@ static void skylake_disable_primary_plane(struct intel_plane *primary,
        spin_unlock_irqrestore(&dev_priv->uncore.lock, irqflags);
 }
 
-static void intel_complete_page_flips(struct drm_i915_private *dev_priv)
-{
-       struct intel_crtc *crtc;
-
-       for_each_intel_crtc(&dev_priv->drm, crtc)
-               intel_finish_page_flip_cs(dev_priv, crtc->pipe);
-}
-
 static int
 __intel_display_resume(struct drm_device *dev,
                       struct drm_atomic_state *state,
@@ -3485,6 +3705,15 @@ void intel_prepare_reset(struct drm_i915_private *dev_priv)
            !gpu_reset_clobbers_display(dev_priv))
                return;
 
+       /* We have a modeset vs reset deadlock, defensively unbreak it. */
+       set_bit(I915_RESET_MODESET, &dev_priv->gpu_error.flags);
+       wake_up_all(&dev_priv->gpu_error.wait_queue);
+
+       if (atomic_read(&dev_priv->gpu_error.pending_fb_pin)) {
+               DRM_DEBUG_KMS("Modeset potentially stuck, unbreaking through wedging\n");
+               i915_gem_set_wedged(dev_priv);
+       }
+
        /*
         * Need mode_config.mutex so that we don't
         * trample ongoing ->detect() and whatnot.
@@ -3535,13 +3764,6 @@ void intel_finish_reset(struct drm_i915_private *dev_priv)
        if (!state)
                goto unlock;
 
-       /*
-        * Flips in the rings will be nuked by the reset,
-        * so complete all pending flips so that user space
-        * will get its events and not get stuck.
-        */
-       intel_complete_page_flips(dev_priv);
-
        dev_priv->modeset_restore_state = NULL;
 
        /* reset doesn't touch the display */
@@ -3578,35 +3800,8 @@ unlock:
        drm_modeset_drop_locks(ctx);
        drm_modeset_acquire_fini(ctx);
        mutex_unlock(&dev->mode_config.mutex);
-}
-
-static bool abort_flip_on_reset(struct intel_crtc *crtc)
-{
-       struct i915_gpu_error *error = &to_i915(crtc->base.dev)->gpu_error;
-
-       if (i915_reset_backoff(error))
-               return true;
-
-       if (crtc->reset_count != i915_reset_count(error))
-               return true;
 
-       return false;
-}
-
-static bool intel_crtc_has_pending_flip(struct drm_crtc *crtc)
-{
-       struct drm_device *dev = crtc->dev;
-       struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
-       bool pending;
-
-       if (abort_flip_on_reset(intel_crtc))
-               return false;
-
-       spin_lock_irq(&dev->event_lock);
-       pending = to_intel_crtc(crtc)->flip_work != NULL;
-       spin_unlock_irq(&dev->event_lock);
-
-       return pending;
+       clear_bit(I915_RESET_MODESET, &dev_priv->gpu_error.flags);
 }
 
 static void intel_update_pipe_config(struct intel_crtc *crtc,
@@ -4163,21 +4358,22 @@ static void ironlake_fdi_disable(struct drm_crtc *crtc)
 
 bool intel_has_pending_fb_unpin(struct drm_i915_private *dev_priv)
 {
-       struct intel_crtc *crtc;
-
-       /* Note that we don't need to be called with mode_config.lock here
-        * as our list of CRTC objects is static for the lifetime of the
-        * device and so cannot disappear as we iterate. Similarly, we can
-        * happily treat the predicates as racy, atomic checks as userspace
-        * cannot claim and pin a new fb without at least acquring the
-        * struct_mutex and so serialising with us.
-        */
-       for_each_intel_crtc(&dev_priv->drm, crtc) {
-               if (atomic_read(&crtc->unpin_work_count) == 0)
+       struct drm_crtc *crtc;
+       bool cleanup_done;
+
+       drm_for_each_crtc(crtc, &dev_priv->drm) {
+               struct drm_crtc_commit *commit;
+               spin_lock(&crtc->commit_lock);
+               commit = list_first_entry_or_null(&crtc->commit_list,
+                                                 struct drm_crtc_commit, commit_entry);
+               cleanup_done = commit ?
+                       try_wait_for_completion(&commit->cleanup_done) : true;
+               spin_unlock(&crtc->commit_lock);
+
+               if (cleanup_done)
                        continue;
 
-               if (crtc->flip_work)
-                       intel_wait_for_vblank(dev_priv, crtc->pipe);
+               drm_crtc_wait_one_vblank(crtc);
 
                return true;
        }
@@ -4185,57 +4381,6 @@ bool intel_has_pending_fb_unpin(struct drm_i915_private *dev_priv)
        return false;
 }
 
-static void page_flip_completed(struct intel_crtc *intel_crtc)
-{
-       struct drm_i915_private *dev_priv = to_i915(intel_crtc->base.dev);
-       struct intel_flip_work *work = intel_crtc->flip_work;
-
-       intel_crtc->flip_work = NULL;
-
-       if (work->event)
-               drm_crtc_send_vblank_event(&intel_crtc->base, work->event);
-
-       drm_crtc_vblank_put(&intel_crtc->base);
-
-       wake_up_all(&dev_priv->pending_flip_queue);
-       trace_i915_flip_complete(intel_crtc->plane,
-                                work->pending_flip_obj);
-
-       queue_work(dev_priv->wq, &work->unpin_work);
-}
-
-static int intel_crtc_wait_for_pending_flips(struct drm_crtc *crtc)
-{
-       struct drm_device *dev = crtc->dev;
-       struct drm_i915_private *dev_priv = to_i915(dev);
-       long ret;
-
-       WARN_ON(waitqueue_active(&dev_priv->pending_flip_queue));
-
-       ret = wait_event_interruptible_timeout(
-                                       dev_priv->pending_flip_queue,
-                                       !intel_crtc_has_pending_flip(crtc),
-                                       60*HZ);
-
-       if (ret < 0)
-               return ret;
-
-       if (ret == 0) {
-               struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
-               struct intel_flip_work *work;
-
-               spin_lock_irq(&dev->event_lock);
-               work = intel_crtc->flip_work;
-               if (work && !is_mmio_work(work)) {
-                       WARN_ONCE(1, "Removing stuck page flip\n");
-                       page_flip_completed(intel_crtc);
-               }
-               spin_unlock_irq(&dev->event_lock);
-       }
-
-       return 0;
-}
-
 void lpt_disable_iclkip(struct drm_i915_private *dev_priv)
 {
        u32 temp;
@@ -4555,7 +4700,7 @@ static void lpt_pch_enable(const struct intel_crtc_state *crtc_state)
        struct drm_i915_private *dev_priv = to_i915(crtc->base.dev);
        enum transcoder cpu_transcoder = crtc_state->cpu_transcoder;
 
-       assert_pch_transcoder_disabled(dev_priv, TRANSCODER_A);
+       assert_pch_transcoder_disabled(dev_priv, PIPE_A);
 
        lpt_program_iclkip(crtc);
 
@@ -4600,6 +4745,9 @@ skl_update_scaler(struct intel_crtc_state *crtc_state, bool force_detach,
         */
        need_scaling = src_w != dst_w || src_h != dst_h;
 
+       if (crtc_state->ycbcr420 && scaler_user == SKL_CRTC_INDEX)
+               need_scaling = true;
+
        /*
         * Scaling/fitting not supported in IF-ID mode in GEN9+
         * TODO: Interlace fetch mode doesn't support YUV420 planar formats.
@@ -5323,8 +5471,7 @@ static void haswell_crtc_enable(struct intel_crtc_state *pipe_config,
                return;
 
        if (intel_crtc->config->has_pch_encoder)
-               intel_set_pch_fifo_underrun_reporting(dev_priv, TRANSCODER_A,
-                                                     false);
+               intel_set_pch_fifo_underrun_reporting(dev_priv, PIPE_A, false);
 
        intel_encoders_pre_pll_enable(crtc, pipe_config, old_state);
 
@@ -5409,8 +5556,7 @@ static void haswell_crtc_enable(struct intel_crtc_state *pipe_config,
                intel_wait_for_vblank(dev_priv, pipe);
                intel_wait_for_vblank(dev_priv, pipe);
                intel_set_cpu_fifo_underrun_reporting(dev_priv, pipe, true);
-               intel_set_pch_fifo_underrun_reporting(dev_priv, TRANSCODER_A,
-                                                     true);
+               intel_set_pch_fifo_underrun_reporting(dev_priv, PIPE_A, true);
        }
 
        /* If we change the relative order between pipe/planes enabling, we need
@@ -5507,8 +5653,7 @@ static void haswell_crtc_disable(struct intel_crtc_state *old_crtc_state,
        enum transcoder cpu_transcoder = intel_crtc->config->cpu_transcoder;
 
        if (intel_crtc->config->has_pch_encoder)
-               intel_set_pch_fifo_underrun_reporting(dev_priv, TRANSCODER_A,
-                                                     false);
+               intel_set_pch_fifo_underrun_reporting(dev_priv, PIPE_A, false);
 
        intel_encoders_disable(crtc, old_crtc_state, old_state);
 
@@ -5536,8 +5681,7 @@ static void haswell_crtc_disable(struct intel_crtc_state *old_crtc_state,
        intel_encoders_post_disable(crtc, old_crtc_state, old_state);
 
        if (old_crtc_state->has_pch_encoder)
-               intel_set_pch_fifo_underrun_reporting(dev_priv, TRANSCODER_A,
-                                                     true);
+               intel_set_pch_fifo_underrun_reporting(dev_priv, PIPE_A, true);
 }
 
 static void i9xx_pfit_enable(struct intel_crtc *crtc)
@@ -5846,8 +5990,6 @@ static void intel_crtc_disable_noatomic(struct drm_crtc *crtc,
                return;
 
        if (crtc->primary->state->visible) {
-               WARN_ON(intel_crtc->flip_work);
-
                intel_pre_disable_primary_noatomic(crtc);
 
                intel_crtc_disable_planes(crtc, 1 << drm_plane_index(crtc->primary));
@@ -6256,6 +6398,16 @@ static int intel_crtc_compute_config(struct intel_crtc *crtc,
                return -EINVAL;
        }
 
+       if (pipe_config->ycbcr420 && pipe_config->base.ctm) {
+               /*
+                * There is only one pipe CSC unit per pipe, and we need that
+                * for output conversion from RGB->YCBCR. So if CTM is already
+                * applied we can't support YCBCR420 output.
+                */
+               DRM_DEBUG_KMS("YCBCR420 and CTM together are not possible\n");
+               return -EINVAL;
+       }
+
        /*
         * Pipe horizontal size must be even in:
         * - DVO ganged mode
@@ -8049,6 +8201,7 @@ static void haswell_set_pipemisc(struct drm_crtc *crtc)
 {
        struct drm_i915_private *dev_priv = to_i915(crtc->dev);
        struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
+       struct intel_crtc_state *config = intel_crtc->config;
 
        if (IS_BROADWELL(dev_priv) || INTEL_INFO(dev_priv)->gen >= 9) {
                u32 val = 0;
@@ -8074,6 +8227,12 @@ static void haswell_set_pipemisc(struct drm_crtc *crtc)
                if (intel_crtc->config->dither)
                        val |= PIPEMISC_DITHER_ENABLE | PIPEMISC_DITHER_TYPE_SP;
 
+               if (config->ycbcr420) {
+                       val |= PIPEMISC_OUTPUT_COLORSPACE_YUV |
+                               PIPEMISC_YUV420_ENABLE |
+                               PIPEMISC_YUV420_MODE_FULL_BLEND;
+               }
+
                I915_WRITE(PIPEMISC(intel_crtc->pipe), val);
        }
 }
@@ -8401,10 +8560,16 @@ skylake_get_initial_plane_config(struct intel_crtc *crtc,
                fb->modifier = I915_FORMAT_MOD_X_TILED;
                break;
        case PLANE_CTL_TILED_Y:
-               fb->modifier = I915_FORMAT_MOD_Y_TILED;
+               if (val & PLANE_CTL_DECOMPRESSION_ENABLE)
+                       fb->modifier = I915_FORMAT_MOD_Y_TILED_CCS;
+               else
+                       fb->modifier = I915_FORMAT_MOD_Y_TILED;
                break;
        case PLANE_CTL_TILED_YF:
-               fb->modifier = I915_FORMAT_MOD_Yf_TILED;
+               if (val & PLANE_CTL_DECOMPRESSION_ENABLE)
+                       fb->modifier = I915_FORMAT_MOD_Yf_TILED_CCS;
+               else
+                       fb->modifier = I915_FORMAT_MOD_Yf_TILED;
                break;
        default:
                MISSING_CASE(tiling);
@@ -8638,7 +8803,8 @@ static void assert_can_disable_lcpll(struct drm_i915_private *dev_priv)
                I915_STATE_WARN(crtc->active, "CRTC for pipe %c enabled\n",
                     pipe_name(crtc->pipe));
 
-       I915_STATE_WARN(I915_READ(HSW_PWR_WELL_DRIVER), "Power well on\n");
+       I915_STATE_WARN(I915_READ(HSW_PWR_WELL_CTL_DRIVER(HSW_DISP_PW_GLOBAL)),
+                       "Display power well on\n");
        I915_STATE_WARN(I915_READ(SPLL_CTL) & SPLL_PLL_ENABLE, "SPLL enabled\n");
        I915_STATE_WARN(I915_READ(WRPLL_CTL(0)) & WRPLL_PLL_ENABLE, "WRPLL1 enabled\n");
        I915_STATE_WARN(I915_READ(WRPLL_CTL(1)) & WRPLL_PLL_ENABLE, "WRPLL2 enabled\n");
@@ -9108,12 +9274,7 @@ static bool haswell_get_pipe_config(struct intel_crtc *crtc,
        u64 power_domain_mask;
        bool active;
 
-       if (INTEL_GEN(dev_priv) >= 9) {
-               intel_crtc_init_scalers(crtc, pipe_config);
-
-               pipe_config->scaler_state.scaler_id = -1;
-               pipe_config->scaler_state.scaler_users &= ~(1 << SKL_CRTC_INDEX);
-       }
+       intel_crtc_init_scalers(crtc, pipe_config);
 
        power_domain = POWER_DOMAIN_PIPE(crtc->pipe);
        if (!intel_display_power_get_if_enabled(dev_priv, power_domain))
@@ -9143,6 +9304,23 @@ static bool haswell_get_pipe_config(struct intel_crtc *crtc,
        pipe_config->gamma_mode =
                I915_READ(GAMMA_MODE(crtc->pipe)) & GAMMA_MODE_MODE_MASK;
 
+       if (IS_BROADWELL(dev_priv) || dev_priv->info.gen >= 9) {
+               u32 tmp = I915_READ(PIPEMISC(crtc->pipe));
+               bool clrspace_yuv = tmp & PIPEMISC_OUTPUT_COLORSPACE_YUV;
+
+               if (IS_GEMINILAKE(dev_priv) || dev_priv->info.gen >= 10) {
+                       bool blend_mode_420 = tmp &
+                                             PIPEMISC_YUV420_MODE_FULL_BLEND;
+
+                       pipe_config->ycbcr420 = tmp & PIPEMISC_YUV420_ENABLE;
+                       if (pipe_config->ycbcr420 != clrspace_yuv ||
+                           pipe_config->ycbcr420 != blend_mode_420)
+                               DRM_DEBUG_KMS("Bad 4:2:0 mode (%08x)\n", tmp);
+               } else if (clrspace_yuv) {
+                       DRM_DEBUG_KMS("YCbCr 4:2:0 Unsupported\n");
+               }
+       }
+
        power_domain = POWER_DOMAIN_PIPE_PANEL_FITTER(crtc->pipe);
        if (intel_display_power_get_if_enabled(dev_priv, power_domain)) {
                power_domain_mask |= BIT_ULL(power_domain);
@@ -10120,882 +10298,44 @@ struct drm_display_mode *intel_crtc_mode_get(struct drm_device *dev,
 static void intel_crtc_destroy(struct drm_crtc *crtc)
 {
        struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
-       struct drm_device *dev = crtc->dev;
-       struct intel_flip_work *work;
-
-       spin_lock_irq(&dev->event_lock);
-       work = intel_crtc->flip_work;
-       intel_crtc->flip_work = NULL;
-       spin_unlock_irq(&dev->event_lock);
-
-       if (work) {
-               cancel_work_sync(&work->mmio_work);
-               cancel_work_sync(&work->unpin_work);
-               kfree(work);
-       }
 
        drm_crtc_cleanup(crtc);
-
        kfree(intel_crtc);
 }
 
-static void intel_unpin_work_fn(struct work_struct *__work)
+/**
+ * intel_wm_need_update - Check whether watermarks need updating
+ * @plane: drm plane
+ * @state: new plane state
+ *
+ * Check current plane state versus the new one to determine whether
+ * watermarks need to be recalculated.
+ *
+ * Returns true or false.
+ */
+static bool intel_wm_need_update(struct drm_plane *plane,
+                                struct drm_plane_state *state)
 {
-       struct intel_flip_work *work =
-               container_of(__work, struct intel_flip_work, unpin_work);
-       struct intel_crtc *crtc = to_intel_crtc(work->crtc);
-       struct drm_device *dev = crtc->base.dev;
-       struct drm_plane *primary = crtc->base.primary;
+       struct intel_plane_state *new = to_intel_plane_state(state);
+       struct intel_plane_state *cur = to_intel_plane_state(plane->state);
 
-       if (is_mmio_work(work))
-               flush_work(&work->mmio_work);
+       /* Update watermarks on tiling or size changes. */
+       if (new->base.visible != cur->base.visible)
+               return true;
 
-       mutex_lock(&dev->struct_mutex);
-       intel_unpin_fb_vma(work->old_vma);
-       i915_gem_object_put(work->pending_flip_obj);
-       mutex_unlock(&dev->struct_mutex);
+       if (!cur->base.fb || !new->base.fb)
+               return false;
 
-       i915_gem_request_put(work->flip_queued_req);
+       if (cur->base.fb->modifier != new->base.fb->modifier ||
+           cur->base.rotation != new->base.rotation ||
+           drm_rect_width(&new->base.src) != drm_rect_width(&cur->base.src) ||
+           drm_rect_height(&new->base.src) != drm_rect_height(&cur->base.src) ||
+           drm_rect_width(&new->base.dst) != drm_rect_width(&cur->base.dst) ||
+           drm_rect_height(&new->base.dst) != drm_rect_height(&cur->base.dst))
+               return true;
 
-       intel_frontbuffer_flip_complete(to_i915(dev),
-                                       to_intel_plane(primary)->frontbuffer_bit);
-       intel_fbc_post_update(crtc);
-       drm_framebuffer_unreference(work->old_fb);
-
-       BUG_ON(atomic_read(&crtc->unpin_work_count) == 0);
-       atomic_dec(&crtc->unpin_work_count);
-
-       kfree(work);
-}
-
-/* Is 'a' after or equal to 'b'? */
-static bool g4x_flip_count_after_eq(u32 a, u32 b)
-{
-       return !((a - b) & 0x80000000);
-}
-
-static bool __pageflip_finished_cs(struct intel_crtc *crtc,
-                                  struct intel_flip_work *work)
-{
-       struct drm_device *dev = crtc->base.dev;
-       struct drm_i915_private *dev_priv = to_i915(dev);
-
-       if (abort_flip_on_reset(crtc))
-               return true;
-
-       /*
-        * The relevant registers doen't exist on pre-ctg.
-        * As the flip done interrupt doesn't trigger for mmio
-        * flips on gmch platforms, a flip count check isn't
-        * really needed there. But since ctg has the registers,
-        * include it in the check anyway.
-        */
-       if (INTEL_GEN(dev_priv) < 5 && !IS_G4X(dev_priv))
-               return true;
-
-       /*
-        * BDW signals flip done immediately if the plane
-        * is disabled, even if the plane enable is already
-        * armed to occur at the next vblank :(
-        */
-
-       /*
-        * A DSPSURFLIVE check isn't enough in case the mmio and CS flips
-        * used the same base address. In that case the mmio flip might
-        * have completed, but the CS hasn't even executed the flip yet.
-        *
-        * A flip count check isn't enough as the CS might have updated
-        * the base address just after start of vblank, but before we
-        * managed to process the interrupt. This means we'd complete the
-        * CS flip too soon.
-        *
-        * Combining both checks should get us a good enough result. It may
-        * still happen that the CS flip has been executed, but has not
-        * yet actually completed. But in case the base address is the same
-        * anyway, we don't really care.
-        */
-       return (I915_READ(DSPSURFLIVE(crtc->plane)) & ~0xfff) ==
-               crtc->flip_work->gtt_offset &&
-               g4x_flip_count_after_eq(I915_READ(PIPE_FLIPCOUNT_G4X(crtc->pipe)),
-                                   crtc->flip_work->flip_count);
-}
-
-static bool
-__pageflip_finished_mmio(struct intel_crtc *crtc,
-                              struct intel_flip_work *work)
-{
-       /*
-        * MMIO work completes when vblank is different from
-        * flip_queued_vblank.
-        *
-        * Reset counter value doesn't matter, this is handled by
-        * i915_wait_request finishing early, so no need to handle
-        * reset here.
-        */
-       return intel_crtc_get_vblank_counter(crtc) != work->flip_queued_vblank;
-}
-
-
-static bool pageflip_finished(struct intel_crtc *crtc,
-                             struct intel_flip_work *work)
-{
-       if (!atomic_read(&work->pending))
-               return false;
-
-       smp_rmb();
-
-       if (is_mmio_work(work))
-               return __pageflip_finished_mmio(crtc, work);
-       else
-               return __pageflip_finished_cs(crtc, work);
-}
-
-void intel_finish_page_flip_cs(struct drm_i915_private *dev_priv, int pipe)
-{
-       struct drm_device *dev = &dev_priv->drm;
-       struct intel_crtc *crtc = intel_get_crtc_for_pipe(dev_priv, pipe);
-       struct intel_flip_work *work;
-       unsigned long flags;
-
-       /* Ignore early vblank irqs */
-       if (!crtc)
-               return;
-
-       /*
-        * This is called both by irq handlers and the reset code (to complete
-        * lost pageflips) so needs the full irqsave spinlocks.
-        */
-       spin_lock_irqsave(&dev->event_lock, flags);
-       work = crtc->flip_work;
-
-       if (work != NULL &&
-           !is_mmio_work(work) &&
-           pageflip_finished(crtc, work))
-               page_flip_completed(crtc);
-
-       spin_unlock_irqrestore(&dev->event_lock, flags);
-}
-
-void intel_finish_page_flip_mmio(struct drm_i915_private *dev_priv, int pipe)
-{
-       struct drm_device *dev = &dev_priv->drm;
-       struct intel_crtc *crtc = intel_get_crtc_for_pipe(dev_priv, pipe);
-       struct intel_flip_work *work;
-       unsigned long flags;
-
-       /* Ignore early vblank irqs */
-       if (!crtc)
-               return;
-
-       /*
-        * This is called both by irq handlers and the reset code (to complete
-        * lost pageflips) so needs the full irqsave spinlocks.
-        */
-       spin_lock_irqsave(&dev->event_lock, flags);
-       work = crtc->flip_work;
-
-       if (work != NULL &&
-           is_mmio_work(work) &&
-           pageflip_finished(crtc, work))
-               page_flip_completed(crtc);
-
-       spin_unlock_irqrestore(&dev->event_lock, flags);
-}
-
-static inline void intel_mark_page_flip_active(struct intel_crtc *crtc,
-                                              struct intel_flip_work *work)
-{
-       work->flip_queued_vblank = intel_crtc_get_vblank_counter(crtc);
-
-       /* Ensure that the work item is consistent when activating it ... */
-       smp_mb__before_atomic();
-       atomic_set(&work->pending, 1);
-}
-
-static int intel_gen2_queue_flip(struct drm_device *dev,
-                                struct drm_crtc *crtc,
-                                struct drm_framebuffer *fb,
-                                struct drm_i915_gem_object *obj,
-                                struct drm_i915_gem_request *req,
-                                uint32_t flags)
-{
-       struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
-       u32 flip_mask, *cs;
-
-       cs = intel_ring_begin(req, 6);
-       if (IS_ERR(cs))
-               return PTR_ERR(cs);
-
-       /* Can't queue multiple flips, so wait for the previous
-        * one to finish before executing the next.
-        */
-       if (intel_crtc->plane)
-               flip_mask = MI_WAIT_FOR_PLANE_B_FLIP;
-       else
-               flip_mask = MI_WAIT_FOR_PLANE_A_FLIP;
-       *cs++ = MI_WAIT_FOR_EVENT | flip_mask;
-       *cs++ = MI_NOOP;
-       *cs++ = MI_DISPLAY_FLIP | MI_DISPLAY_FLIP_PLANE(intel_crtc->plane);
-       *cs++ = fb->pitches[0];
-       *cs++ = intel_crtc->flip_work->gtt_offset;
-       *cs++ = 0; /* aux display base address, unused */
-
-       return 0;
-}
-
-static int intel_gen3_queue_flip(struct drm_device *dev,
-                                struct drm_crtc *crtc,
-                                struct drm_framebuffer *fb,
-                                struct drm_i915_gem_object *obj,
-                                struct drm_i915_gem_request *req,
-                                uint32_t flags)
-{
-       struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
-       u32 flip_mask, *cs;
-
-       cs = intel_ring_begin(req, 6);
-       if (IS_ERR(cs))
-               return PTR_ERR(cs);
-
-       if (intel_crtc->plane)
-               flip_mask = MI_WAIT_FOR_PLANE_B_FLIP;
-       else
-               flip_mask = MI_WAIT_FOR_PLANE_A_FLIP;
-       *cs++ = MI_WAIT_FOR_EVENT | flip_mask;
-       *cs++ = MI_NOOP;
-       *cs++ = MI_DISPLAY_FLIP_I915 | MI_DISPLAY_FLIP_PLANE(intel_crtc->plane);
-       *cs++ = fb->pitches[0];
-       *cs++ = intel_crtc->flip_work->gtt_offset;
-       *cs++ = MI_NOOP;
-
-       return 0;
-}
-
-static int intel_gen4_queue_flip(struct drm_device *dev,
-                                struct drm_crtc *crtc,
-                                struct drm_framebuffer *fb,
-                                struct drm_i915_gem_object *obj,
-                                struct drm_i915_gem_request *req,
-                                uint32_t flags)
-{
-       struct drm_i915_private *dev_priv = to_i915(dev);
-       struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
-       u32 pf, pipesrc, *cs;
-
-       cs = intel_ring_begin(req, 4);
-       if (IS_ERR(cs))
-               return PTR_ERR(cs);
-
-       /* i965+ uses the linear or tiled offsets from the
-        * Display Registers (which do not change across a page-flip)
-        * so we need only reprogram the base address.
-        */
-       *cs++ = MI_DISPLAY_FLIP | MI_DISPLAY_FLIP_PLANE(intel_crtc->plane);
-       *cs++ = fb->pitches[0];
-       *cs++ = intel_crtc->flip_work->gtt_offset |
-               intel_fb_modifier_to_tiling(fb->modifier);
-
-       /* XXX Enabling the panel-fitter across page-flip is so far
-        * untested on non-native modes, so ignore it for now.
-        * pf = I915_READ(pipe == 0 ? PFA_CTL_1 : PFB_CTL_1) & PF_ENABLE;
-        */
-       pf = 0;
-       pipesrc = I915_READ(PIPESRC(intel_crtc->pipe)) & 0x0fff0fff;
-       *cs++ = pf | pipesrc;
-
-       return 0;
-}
-
-static int intel_gen6_queue_flip(struct drm_device *dev,
-                                struct drm_crtc *crtc,
-                                struct drm_framebuffer *fb,
-                                struct drm_i915_gem_object *obj,
-                                struct drm_i915_gem_request *req,
-                                uint32_t flags)
-{
-       struct drm_i915_private *dev_priv = to_i915(dev);
-       struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
-       u32 pf, pipesrc, *cs;
-
-       cs = intel_ring_begin(req, 4);
-       if (IS_ERR(cs))
-               return PTR_ERR(cs);
-
-       *cs++ = MI_DISPLAY_FLIP | MI_DISPLAY_FLIP_PLANE(intel_crtc->plane);
-       *cs++ = fb->pitches[0] | intel_fb_modifier_to_tiling(fb->modifier);
-       *cs++ = intel_crtc->flip_work->gtt_offset;
-
-       /* Contrary to the suggestions in the documentation,
-        * "Enable Panel Fitter" does not seem to be required when page
-        * flipping with a non-native mode, and worse causes a normal
-        * modeset to fail.
-        * pf = I915_READ(PF_CTL(intel_crtc->pipe)) & PF_ENABLE;
-        */
-       pf = 0;
-       pipesrc = I915_READ(PIPESRC(intel_crtc->pipe)) & 0x0fff0fff;
-       *cs++ = pf | pipesrc;
-
-       return 0;
-}
-
-static int intel_gen7_queue_flip(struct drm_device *dev,
-                                struct drm_crtc *crtc,
-                                struct drm_framebuffer *fb,
-                                struct drm_i915_gem_object *obj,
-                                struct drm_i915_gem_request *req,
-                                uint32_t flags)
-{
-       struct drm_i915_private *dev_priv = to_i915(dev);
-       struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
-       u32 *cs, plane_bit = 0;
-       int len, ret;
-
-       switch (intel_crtc->plane) {
-       case PLANE_A:
-               plane_bit = MI_DISPLAY_FLIP_IVB_PLANE_A;
-               break;
-       case PLANE_B:
-               plane_bit = MI_DISPLAY_FLIP_IVB_PLANE_B;
-               break;
-       case PLANE_C:
-               plane_bit = MI_DISPLAY_FLIP_IVB_PLANE_C;
-               break;
-       default:
-               WARN_ONCE(1, "unknown plane in flip command\n");
-               return -ENODEV;
-       }
-
-       len = 4;
-       if (req->engine->id == RCS) {
-               len += 6;
-               /*
-                * On Gen 8, SRM is now taking an extra dword to accommodate
-                * 48bits addresses, and we need a NOOP for the batch size to
-                * stay even.
-                */
-               if (IS_GEN8(dev_priv))
-                       len += 2;
-       }
-
-       /*
-        * BSpec MI_DISPLAY_FLIP for IVB:
-        * "The full packet must be contained within the same cache line."
-        *
-        * Currently the LRI+SRM+MI_DISPLAY_FLIP all fit within the same
-        * cacheline, if we ever start emitting more commands before
-        * the MI_DISPLAY_FLIP we may need to first emit everything else,
-        * then do the cacheline alignment, and finally emit the
-        * MI_DISPLAY_FLIP.
-        */
-       ret = intel_ring_cacheline_align(req);
-       if (ret)
-               return ret;
-
-       cs = intel_ring_begin(req, len);
-       if (IS_ERR(cs))
-               return PTR_ERR(cs);
-
-       /* Unmask the flip-done completion message. Note that the bspec says that
-        * we should do this for both the BCS and RCS, and that we must not unmask
-        * more than one flip event at any time (or ensure that one flip message
-        * can be sent by waiting for flip-done prior to queueing new flips).
-        * Experimentation says that BCS works despite DERRMR masking all
-        * flip-done completion events and that unmasking all planes at once
-        * for the RCS also doesn't appear to drop events. Setting the DERRMR
-        * to zero does lead to lockups within MI_DISPLAY_FLIP.
-        */
-       if (req->engine->id == RCS) {
-               *cs++ = MI_LOAD_REGISTER_IMM(1);
-               *cs++ = i915_mmio_reg_offset(DERRMR);
-               *cs++ = ~(DERRMR_PIPEA_PRI_FLIP_DONE |
-                         DERRMR_PIPEB_PRI_FLIP_DONE |
-                         DERRMR_PIPEC_PRI_FLIP_DONE);
-               if (IS_GEN8(dev_priv))
-                       *cs++ = MI_STORE_REGISTER_MEM_GEN8 |
-                               MI_SRM_LRM_GLOBAL_GTT;
-               else
-                       *cs++ = MI_STORE_REGISTER_MEM | MI_SRM_LRM_GLOBAL_GTT;
-               *cs++ = i915_mmio_reg_offset(DERRMR);
-               *cs++ = i915_ggtt_offset(req->engine->scratch) + 256;
-               if (IS_GEN8(dev_priv)) {
-                       *cs++ = 0;
-                       *cs++ = MI_NOOP;
-               }
-       }
-
-       *cs++ = MI_DISPLAY_FLIP_I915 | plane_bit;
-       *cs++ = fb->pitches[0] | intel_fb_modifier_to_tiling(fb->modifier);
-       *cs++ = intel_crtc->flip_work->gtt_offset;
-       *cs++ = MI_NOOP;
-
-       return 0;
-}
-
-static bool use_mmio_flip(struct intel_engine_cs *engine,
-                         struct drm_i915_gem_object *obj)
-{
-       /*
-        * This is not being used for older platforms, because
-        * non-availability of flip done interrupt forces us to use
-        * CS flips. Older platforms derive flip done using some clever
-        * tricks involving the flip_pending status bits and vblank irqs.
-        * So using MMIO flips there would disrupt this mechanism.
-        */
-
-       if (engine == NULL)
-               return true;
-
-       if (INTEL_GEN(engine->i915) < 5)
-               return false;
-
-       if (i915.use_mmio_flip < 0)
-               return false;
-       else if (i915.use_mmio_flip > 0)
-               return true;
-       else if (i915.enable_execlists)
-               return true;
-
-       return engine != i915_gem_object_last_write_engine(obj);
-}
-
-static void skl_do_mmio_flip(struct intel_crtc *intel_crtc,
-                            unsigned int rotation,
-                            struct intel_flip_work *work)
-{
-       struct drm_device *dev = intel_crtc->base.dev;
-       struct drm_i915_private *dev_priv = to_i915(dev);
-       struct drm_framebuffer *fb = intel_crtc->base.primary->fb;
-       const enum pipe pipe = intel_crtc->pipe;
-       u32 ctl, stride = skl_plane_stride(fb, 0, rotation);
-
-       ctl = I915_READ(PLANE_CTL(pipe, 0));
-       ctl &= ~PLANE_CTL_TILED_MASK;
-       switch (fb->modifier) {
-       case DRM_FORMAT_MOD_LINEAR:
-               break;
-       case I915_FORMAT_MOD_X_TILED:
-               ctl |= PLANE_CTL_TILED_X;
-               break;
-       case I915_FORMAT_MOD_Y_TILED:
-               ctl |= PLANE_CTL_TILED_Y;
-               break;
-       case I915_FORMAT_MOD_Yf_TILED:
-               ctl |= PLANE_CTL_TILED_YF;
-               break;
-       default:
-               MISSING_CASE(fb->modifier);
-       }
-
-       /*
-        * Both PLANE_CTL and PLANE_STRIDE are not updated on vblank but on
-        * PLANE_SURF updates, the update is then guaranteed to be atomic.
-        */
-       I915_WRITE(PLANE_CTL(pipe, 0), ctl);
-       I915_WRITE(PLANE_STRIDE(pipe, 0), stride);
-
-       I915_WRITE(PLANE_SURF(pipe, 0), work->gtt_offset);
-       POSTING_READ(PLANE_SURF(pipe, 0));
-}
-
-static void ilk_do_mmio_flip(struct intel_crtc *intel_crtc,
-                            struct intel_flip_work *work)
-{
-       struct drm_device *dev = intel_crtc->base.dev;
-       struct drm_i915_private *dev_priv = to_i915(dev);
-       struct drm_framebuffer *fb = intel_crtc->base.primary->fb;
-       i915_reg_t reg = DSPCNTR(intel_crtc->plane);
-       u32 dspcntr;
-
-       dspcntr = I915_READ(reg);
-
-       if (fb->modifier == I915_FORMAT_MOD_X_TILED)
-               dspcntr |= DISPPLANE_TILED;
-       else
-               dspcntr &= ~DISPPLANE_TILED;
-
-       I915_WRITE(reg, dspcntr);
-
-       I915_WRITE(DSPSURF(intel_crtc->plane), work->gtt_offset);
-       POSTING_READ(DSPSURF(intel_crtc->plane));
-}
-
-static void intel_mmio_flip_work_func(struct work_struct *w)
-{
-       struct intel_flip_work *work =
-               container_of(w, struct intel_flip_work, mmio_work);
-       struct intel_crtc *crtc = to_intel_crtc(work->crtc);
-       struct drm_i915_private *dev_priv = to_i915(crtc->base.dev);
-       struct intel_framebuffer *intel_fb =
-               to_intel_framebuffer(crtc->base.primary->fb);
-       struct drm_i915_gem_object *obj = intel_fb->obj;
-
-       WARN_ON(i915_gem_object_wait(obj, 0, MAX_SCHEDULE_TIMEOUT, NULL) < 0);
-
-       intel_pipe_update_start(crtc);
-
-       if (INTEL_GEN(dev_priv) >= 9)
-               skl_do_mmio_flip(crtc, work->rotation, work);
-       else
-               /* use_mmio_flip() retricts MMIO flips to ilk+ */
-               ilk_do_mmio_flip(crtc, work);
-
-       intel_pipe_update_end(crtc, work);
-}
-
-static int intel_default_queue_flip(struct drm_device *dev,
-                                   struct drm_crtc *crtc,
-                                   struct drm_framebuffer *fb,
-                                   struct drm_i915_gem_object *obj,
-                                   struct drm_i915_gem_request *req,
-                                   uint32_t flags)
-{
-       return -ENODEV;
-}
-
-static bool __pageflip_stall_check_cs(struct drm_i915_private *dev_priv,
-                                     struct intel_crtc *intel_crtc,
-                                     struct intel_flip_work *work)
-{
-       u32 addr, vblank;
-
-       if (!atomic_read(&work->pending))
-               return false;
-
-       smp_rmb();
-
-       vblank = intel_crtc_get_vblank_counter(intel_crtc);
-       if (work->flip_ready_vblank == 0) {
-               if (work->flip_queued_req &&
-                   !i915_gem_request_completed(work->flip_queued_req))
-                       return false;
-
-               work->flip_ready_vblank = vblank;
-       }
-
-       if (vblank - work->flip_ready_vblank < 3)
-               return false;
-
-       /* Potential stall - if we see that the flip has happened,
-        * assume a missed interrupt. */
-       if (INTEL_GEN(dev_priv) >= 4)
-               addr = I915_HI_DISPBASE(I915_READ(DSPSURF(intel_crtc->plane)));
-       else
-               addr = I915_READ(DSPADDR(intel_crtc->plane));
-
-       /* There is a potential issue here with a false positive after a flip
-        * to the same address. We could address this by checking for a
-        * non-incrementing frame counter.
-        */
-       return addr == work->gtt_offset;
-}
-
-void intel_check_page_flip(struct drm_i915_private *dev_priv, int pipe)
-{
-       struct drm_device *dev = &dev_priv->drm;
-       struct intel_crtc *crtc = intel_get_crtc_for_pipe(dev_priv, pipe);
-       struct intel_flip_work *work;
-
-       WARN_ON(!in_interrupt());
-
-       if (crtc == NULL)
-               return;
-
-       spin_lock(&dev->event_lock);
-       work = crtc->flip_work;
-
-       if (work != NULL && !is_mmio_work(work) &&
-           __pageflip_stall_check_cs(dev_priv, crtc, work)) {
-               WARN_ONCE(1,
-                         "Kicking stuck page flip: queued at %d, now %d\n",
-                       work->flip_queued_vblank, intel_crtc_get_vblank_counter(crtc));
-               page_flip_completed(crtc);
-               work = NULL;
-       }
-
-       if (work != NULL && !is_mmio_work(work) &&
-           intel_crtc_get_vblank_counter(crtc) - work->flip_queued_vblank > 1)
-               intel_queue_rps_boost_for_request(work->flip_queued_req);
-       spin_unlock(&dev->event_lock);
-}
-
-__maybe_unused
-static int intel_crtc_page_flip(struct drm_crtc *crtc,
-                               struct drm_framebuffer *fb,
-                               struct drm_pending_vblank_event *event,
-                               uint32_t page_flip_flags)
-{
-       struct drm_device *dev = crtc->dev;
-       struct drm_i915_private *dev_priv = to_i915(dev);
-       struct drm_framebuffer *old_fb = crtc->primary->fb;
-       struct drm_i915_gem_object *obj = intel_fb_obj(fb);
-       struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
-       struct drm_plane *primary = crtc->primary;
-       enum pipe pipe = intel_crtc->pipe;
-       struct intel_flip_work *work;
-       struct intel_engine_cs *engine;
-       bool mmio_flip;
-       struct drm_i915_gem_request *request;
-       struct i915_vma *vma;
-       int ret;
-
-       /*
-        * drm_mode_page_flip_ioctl() should already catch this, but double
-        * check to be safe.  In the future we may enable pageflipping from
-        * a disabled primary plane.
-        */
-       if (WARN_ON(intel_fb_obj(old_fb) == NULL))
-               return -EBUSY;
-
-       /* Can't change pixel format via MI display flips. */
-       if (fb->format != crtc->primary->fb->format)
-               return -EINVAL;
-
-       /*
-        * TILEOFF/LINOFF registers can't be changed via MI display flips.
-        * Note that pitch changes could also affect these register.
-        */
-       if (INTEL_GEN(dev_priv) > 3 &&
-           (fb->offsets[0] != crtc->primary->fb->offsets[0] ||
-            fb->pitches[0] != crtc->primary->fb->pitches[0]))
-               return -EINVAL;
-
-       if (i915_terminally_wedged(&dev_priv->gpu_error))
-               goto out_hang;
-
-       work = kzalloc(sizeof(*work), GFP_KERNEL);
-       if (work == NULL)
-               return -ENOMEM;
-
-       work->event = event;
-       work->crtc = crtc;
-       work->old_fb = old_fb;
-       INIT_WORK(&work->unpin_work, intel_unpin_work_fn);
-
-       ret = drm_crtc_vblank_get(crtc);
-       if (ret)
-               goto free_work;
-
-       /* We borrow the event spin lock for protecting flip_work */
-       spin_lock_irq(&dev->event_lock);
-       if (intel_crtc->flip_work) {
-               /* Before declaring the flip queue wedged, check if
-                * the hardware completed the operation behind our backs.
-                */
-               if (pageflip_finished(intel_crtc, intel_crtc->flip_work)) {
-                       DRM_DEBUG_DRIVER("flip queue: previous flip completed, continuing\n");
-                       page_flip_completed(intel_crtc);
-               } else {
-                       DRM_DEBUG_DRIVER("flip queue: crtc already busy\n");
-                       spin_unlock_irq(&dev->event_lock);
-
-                       drm_crtc_vblank_put(crtc);
-                       kfree(work);
-                       return -EBUSY;
-               }
-       }
-       intel_crtc->flip_work = work;
-       spin_unlock_irq(&dev->event_lock);
-
-       if (atomic_read(&intel_crtc->unpin_work_count) >= 2)
-               flush_workqueue(dev_priv->wq);
-
-       /* Reference the objects for the scheduled work. */
-       drm_framebuffer_reference(work->old_fb);
-
-       crtc->primary->fb = fb;
-       update_state_fb(crtc->primary);
-
-       work->pending_flip_obj = i915_gem_object_get(obj);
-
-       ret = i915_mutex_lock_interruptible(dev);
-       if (ret)
-               goto cleanup;
-
-       intel_crtc->reset_count = i915_reset_count(&dev_priv->gpu_error);
-       if (i915_reset_backoff_or_wedged(&dev_priv->gpu_error)) {
-               ret = -EIO;
-               goto unlock;
-       }
-
-       atomic_inc(&intel_crtc->unpin_work_count);
-
-       if (INTEL_GEN(dev_priv) >= 5 || IS_G4X(dev_priv))
-               work->flip_count = I915_READ(PIPE_FLIPCOUNT_G4X(pipe)) + 1;
-
-       if (IS_VALLEYVIEW(dev_priv) || IS_CHERRYVIEW(dev_priv)) {
-               engine = dev_priv->engine[BCS];
-               if (fb->modifier != old_fb->modifier)
-                       /* vlv: DISPLAY_FLIP fails to change tiling */
-                       engine = NULL;
-       } else if (IS_IVYBRIDGE(dev_priv) || IS_HASWELL(dev_priv)) {
-               engine = dev_priv->engine[BCS];
-       } else if (INTEL_GEN(dev_priv) >= 7) {
-               engine = i915_gem_object_last_write_engine(obj);
-               if (engine == NULL || engine->id != RCS)
-                       engine = dev_priv->engine[BCS];
-       } else {
-               engine = dev_priv->engine[RCS];
-       }
-
-       mmio_flip = use_mmio_flip(engine, obj);
-
-       vma = intel_pin_and_fence_fb_obj(fb, primary->state->rotation);
-       if (IS_ERR(vma)) {
-               ret = PTR_ERR(vma);
-               goto cleanup_pending;
-       }
-
-       work->old_vma = to_intel_plane_state(primary->state)->vma;
-       to_intel_plane_state(primary->state)->vma = vma;
-
-       work->gtt_offset = i915_ggtt_offset(vma) + intel_crtc->dspaddr_offset;
-       work->rotation = crtc->primary->state->rotation;
-
-       /*
-        * There's the potential that the next frame will not be compatible with
-        * FBC, so we want to call pre_update() before the actual page flip.
-        * The problem is that pre_update() caches some information about the fb
-        * object, so we want to do this only after the object is pinned. Let's
-        * be on the safe side and do this immediately before scheduling the
-        * flip.
-        */
-       intel_fbc_pre_update(intel_crtc, intel_crtc->config,
-                            to_intel_plane_state(primary->state));
-
-       if (mmio_flip) {
-               INIT_WORK(&work->mmio_work, intel_mmio_flip_work_func);
-               queue_work(system_unbound_wq, &work->mmio_work);
-       } else {
-               request = i915_gem_request_alloc(engine,
-                                                dev_priv->kernel_context);
-               if (IS_ERR(request)) {
-                       ret = PTR_ERR(request);
-                       goto cleanup_unpin;
-               }
-
-               ret = i915_gem_request_await_object(request, obj, false);
-               if (ret)
-                       goto cleanup_request;
-
-               ret = dev_priv->display.queue_flip(dev, crtc, fb, obj, request,
-                                                  page_flip_flags);
-               if (ret)
-                       goto cleanup_request;
-
-               intel_mark_page_flip_active(intel_crtc, work);
-
-               work->flip_queued_req = i915_gem_request_get(request);
-               i915_add_request(request);
-       }
-
-       i915_gem_object_wait_priority(obj, 0, I915_PRIORITY_DISPLAY);
-       i915_gem_track_fb(intel_fb_obj(old_fb), obj,
-                         to_intel_plane(primary)->frontbuffer_bit);
-       mutex_unlock(&dev->struct_mutex);
-
-       intel_frontbuffer_flip_prepare(to_i915(dev),
-                                      to_intel_plane(primary)->frontbuffer_bit);
-
-       trace_i915_flip_request(intel_crtc->plane, obj);
-
-       return 0;
-
-cleanup_request:
-       i915_add_request(request);
-cleanup_unpin:
-       to_intel_plane_state(primary->state)->vma = work->old_vma;
-       intel_unpin_fb_vma(vma);
-cleanup_pending:
-       atomic_dec(&intel_crtc->unpin_work_count);
-unlock:
-       mutex_unlock(&dev->struct_mutex);
-cleanup:
-       crtc->primary->fb = old_fb;
-       update_state_fb(crtc->primary);
-
-       i915_gem_object_put(obj);
-       drm_framebuffer_unreference(work->old_fb);
-
-       spin_lock_irq(&dev->event_lock);
-       intel_crtc->flip_work = NULL;
-       spin_unlock_irq(&dev->event_lock);
-
-       drm_crtc_vblank_put(crtc);
-free_work:
-       kfree(work);
-
-       if (ret == -EIO) {
-               struct drm_atomic_state *state;
-               struct drm_plane_state *plane_state;
-
-out_hang:
-               state = drm_atomic_state_alloc(dev);
-               if (!state)
-                       return -ENOMEM;
-               state->acquire_ctx = dev->mode_config.acquire_ctx;
-
-retry:
-               plane_state = drm_atomic_get_plane_state(state, primary);
-               ret = PTR_ERR_OR_ZERO(plane_state);
-               if (!ret) {
-                       drm_atomic_set_fb_for_plane(plane_state, fb);
-
-                       ret = drm_atomic_set_crtc_for_plane(plane_state, crtc);
-                       if (!ret)
-                               ret = drm_atomic_commit(state);
-               }
-
-               if (ret == -EDEADLK) {
-                       drm_modeset_backoff(state->acquire_ctx);
-                       drm_atomic_state_clear(state);
-                       goto retry;
-               }
-
-               drm_atomic_state_put(state);
-
-               if (ret == 0 && event) {
-                       spin_lock_irq(&dev->event_lock);
-                       drm_crtc_send_vblank_event(crtc, event);
-                       spin_unlock_irq(&dev->event_lock);
-               }
-       }
-       return ret;
-}
-
-
-/**
- * intel_wm_need_update - Check whether watermarks need updating
- * @plane: drm plane
- * @state: new plane state
- *
- * Check current plane state versus the new one to determine whether
- * watermarks need to be recalculated.
- *
- * Returns true or false.
- */
-static bool intel_wm_need_update(struct drm_plane *plane,
-                                struct drm_plane_state *state)
-{
-       struct intel_plane_state *new = to_intel_plane_state(state);
-       struct intel_plane_state *cur = to_intel_plane_state(plane->state);
-
-       /* Update watermarks on tiling or size changes. */
-       if (new->base.visible != cur->base.visible)
-               return true;
-
-       if (!cur->base.fb || !new->base.fb)
-               return false;
-
-       if (cur->base.fb->modifier != new->base.fb->modifier ||
-           cur->base.rotation != new->base.rotation ||
-           drm_rect_width(&new->base.src) != drm_rect_width(&cur->base.src) ||
-           drm_rect_height(&new->base.src) != drm_rect_height(&cur->base.src) ||
-           drm_rect_width(&new->base.dst) != drm_rect_width(&cur->base.dst) ||
-           drm_rect_height(&new->base.dst) != drm_rect_height(&cur->base.dst))
-               return true;
-
-       return false;
-}
+       return false;
+}
 
 static bool needs_scaling(struct intel_plane_state *state)
 {
@@ -11360,6 +10700,9 @@ static void intel_dump_pipe_config(struct intel_crtc *crtc,
                                      pipe_config->fdi_lanes,
                                      &pipe_config->fdi_m_n);
 
+       if (pipe_config->ycbcr420)
+               DRM_DEBUG_KMS("YCbCr 4:2:0 output enabled\n");
+
        if (intel_crtc_has_dp_encoder(pipe_config)) {
                intel_dump_m_n_config(pipe_config, "dp m_n",
                                pipe_config->lane_count, &pipe_config->dp_m_n);
@@ -11931,6 +11274,7 @@ intel_pipe_config_compare(struct drm_i915_private *dev_priv,
        PIPE_CONF_CHECK_I(hdmi_scrambling);
        PIPE_CONF_CHECK_I(hdmi_high_tmds_clock_ratio);
        PIPE_CONF_CHECK_I(has_infoframe);
+       PIPE_CONF_CHECK_I(ycbcr420);
 
        PIPE_CONF_CHECK_I(has_audio);
 
@@ -12772,31 +12116,7 @@ static int intel_atomic_check(struct drm_device *dev,
 static int intel_atomic_prepare_commit(struct drm_device *dev,
                                       struct drm_atomic_state *state)
 {
-       struct drm_i915_private *dev_priv = to_i915(dev);
-       struct drm_crtc_state *crtc_state;
-       struct drm_crtc *crtc;
-       int i, ret;
-
-       for_each_new_crtc_in_state(state, crtc, crtc_state, i) {
-               if (state->legacy_cursor_update)
-                       continue;
-
-               ret = intel_crtc_wait_for_pending_flips(crtc);
-               if (ret)
-                       return ret;
-
-               if (atomic_read(&to_intel_crtc(crtc)->unpin_work_count) >= 2)
-                       flush_workqueue(dev_priv->wq);
-       }
-
-       ret = mutex_lock_interruptible(&dev->struct_mutex);
-       if (ret)
-               return ret;
-
-       ret = drm_atomic_helper_prepare_planes(dev, state);
-       mutex_unlock(&dev->struct_mutex);
-
-       return ret;
+       return drm_atomic_helper_prepare_planes(dev, state);
 }
 
 u32 intel_crtc_get_vblank_counter(struct intel_crtc *crtc)
@@ -12939,6 +12259,30 @@ static void intel_atomic_helper_free_state_worker(struct work_struct *work)
        intel_atomic_helper_free_state(dev_priv);
 }
 
+static void intel_atomic_commit_fence_wait(struct intel_atomic_state *intel_state)
+{
+       struct wait_queue_entry wait_fence, wait_reset;
+       struct drm_i915_private *dev_priv = to_i915(intel_state->base.dev);
+
+       init_wait_entry(&wait_fence, 0);
+       init_wait_entry(&wait_reset, 0);
+       for (;;) {
+               prepare_to_wait(&intel_state->commit_ready.wait,
+                               &wait_fence, TASK_UNINTERRUPTIBLE);
+               prepare_to_wait(&dev_priv->gpu_error.wait_queue,
+                               &wait_reset, TASK_UNINTERRUPTIBLE);
+
+
+               if (i915_sw_fence_done(&intel_state->commit_ready)
+                   || test_bit(I915_RESET_MODESET, &dev_priv->gpu_error.flags))
+                       break;
+
+               schedule();
+       }
+       finish_wait(&intel_state->commit_ready.wait, &wait_fence);
+       finish_wait(&dev_priv->gpu_error.wait_queue, &wait_reset);
+}
+
 static void intel_atomic_commit_tail(struct drm_atomic_state *state)
 {
        struct drm_device *dev = state->dev;
@@ -12951,6 +12295,8 @@ static void intel_atomic_commit_tail(struct drm_atomic_state *state)
        u64 put_domains[I915_MAX_PIPES] = {};
        int i;
 
+       intel_atomic_commit_fence_wait(intel_state);
+
        drm_atomic_helper_wait_for_dependencies(state);
 
        if (intel_state->modeset)
@@ -13089,9 +12435,7 @@ static void intel_atomic_commit_tail(struct drm_atomic_state *state)
                intel_display_power_put(dev_priv, POWER_DOMAIN_MODESET);
        }
 
-       mutex_lock(&dev->struct_mutex);
        drm_atomic_helper_cleanup_planes(dev, state);
-       mutex_unlock(&dev->struct_mutex);
 
        drm_atomic_helper_commit_cleanup_done(state);
 
@@ -13117,10 +12461,8 @@ intel_atomic_commit_ready(struct i915_sw_fence *fence,
 
        switch (notify) {
        case FENCE_COMPLETE:
-               if (state->base.commit_work.func)
-                       queue_work(system_unbound_wq, &state->base.commit_work);
+               /* we do blocking waits in the worker, nothing to do here */
                break;
-
        case FENCE_FREE:
                {
                        struct intel_atomic_helper *helper =
@@ -13206,9 +12548,7 @@ static int intel_atomic_commit(struct drm_device *dev,
        if (ret) {
                i915_sw_fence_commit(&intel_state->commit_ready);
 
-               mutex_lock(&dev->struct_mutex);
                drm_atomic_helper_cleanup_planes(dev, state);
-               mutex_unlock(&dev->struct_mutex);
                return ret;
        }
        dev_priv->wm.distrust_bios_wm = false;
@@ -13224,14 +12564,14 @@ static int intel_atomic_commit(struct drm_device *dev,
        }
 
        drm_atomic_state_get(state);
-       INIT_WORK(&state->commit_work,
-                 nonblock ? intel_atomic_commit_work : NULL);
+       INIT_WORK(&state->commit_work, intel_atomic_commit_work);
 
        i915_sw_fence_commit(&intel_state->commit_ready);
-       if (!nonblock) {
-               i915_sw_fence_wait(&intel_state->commit_ready);
+       if (nonblock)
+               queue_work(system_unbound_wq, &state->commit_work);
+       else
                intel_atomic_commit_tail(state);
-       }
+
 
        return 0;
 }
@@ -13272,32 +12612,6 @@ intel_prepare_plane_fb(struct drm_plane *plane,
        struct drm_i915_gem_object *old_obj = intel_fb_obj(plane->state->fb);
        int ret;
 
-       if (obj) {
-               if (plane->type == DRM_PLANE_TYPE_CURSOR &&
-                   INTEL_INFO(dev_priv)->cursor_needs_physical) {
-                       const int align = intel_cursor_alignment(dev_priv);
-
-                       ret = i915_gem_object_attach_phys(obj, align);
-                       if (ret) {
-                               DRM_DEBUG_KMS("failed to attach phys object\n");
-                               return ret;
-                       }
-               } else {
-                       struct i915_vma *vma;
-
-                       vma = intel_pin_and_fence_fb_obj(fb, new_state->rotation);
-                       if (IS_ERR(vma)) {
-                               DRM_DEBUG_KMS("failed to pin object\n");
-                               return PTR_ERR(vma);
-                       }
-
-                       to_intel_plane_state(new_state)->vma = vma;
-               }
-       }
-
-       if (!obj && !old_obj)
-               return 0;
-
        if (old_obj) {
                struct drm_crtc_state *crtc_state =
                        drm_atomic_get_existing_crtc_state(new_state->state,
@@ -13336,6 +12650,38 @@ intel_prepare_plane_fb(struct drm_plane *plane,
        if (!obj)
                return 0;
 
+       ret = i915_gem_object_pin_pages(obj);
+       if (ret)
+               return ret;
+
+       ret = mutex_lock_interruptible(&dev_priv->drm.struct_mutex);
+       if (ret) {
+               i915_gem_object_unpin_pages(obj);
+               return ret;
+       }
+
+       if (plane->type == DRM_PLANE_TYPE_CURSOR &&
+           INTEL_INFO(dev_priv)->cursor_needs_physical) {
+               const int align = intel_cursor_alignment(dev_priv);
+
+               ret = i915_gem_object_attach_phys(obj, align);
+       } else {
+               struct i915_vma *vma;
+
+               vma = intel_pin_and_fence_fb_obj(fb, new_state->rotation);
+               if (!IS_ERR(vma))
+                       to_intel_plane_state(new_state)->vma = vma;
+               else
+                       ret =  PTR_ERR(vma);
+       }
+
+       i915_gem_object_wait_priority(obj, 0, I915_PRIORITY_DISPLAY);
+
+       mutex_unlock(&dev_priv->drm.struct_mutex);
+       i915_gem_object_unpin_pages(obj);
+       if (ret)
+               return ret;
+
        if (!new_state->fence) { /* implicit fencing */
                ret = i915_sw_fence_await_reservation(&intel_state->commit_ready,
                                                      obj->resv, NULL,
@@ -13343,8 +12689,6 @@ intel_prepare_plane_fb(struct drm_plane *plane,
                                                      GFP_KERNEL);
                if (ret < 0)
                        return ret;
-
-               i915_gem_object_wait_priority(obj, 0, I915_PRIORITY_DISPLAY);
        }
 
        return 0;
@@ -13367,8 +12711,11 @@ intel_cleanup_plane_fb(struct drm_plane *plane,
 
        /* Should only be called after a successful intel_prepare_plane_fb()! */
        vma = fetch_and_zero(&to_intel_plane_state(old_state)->vma);
-       if (vma)
+       if (vma) {
+               mutex_lock(&plane->dev->struct_mutex);
                intel_unpin_fb_vma(vma);
+               mutex_unlock(&plane->dev->struct_mutex);
+       }
 }
 
 int
@@ -13495,7 +12842,7 @@ static void intel_finish_crtc_commit(struct drm_crtc *crtc,
 {
        struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
 
-       intel_pipe_update_end(intel_crtc, NULL);
+       intel_pipe_update_end(intel_crtc);
 }
 
 /**
@@ -13511,7 +12858,102 @@ void intel_plane_destroy(struct drm_plane *plane)
        kfree(to_intel_plane(plane));
 }
 
-const struct drm_plane_funcs intel_plane_funcs = {
+static bool i8xx_mod_supported(uint32_t format, uint64_t modifier)
+{
+       switch (format) {
+       case DRM_FORMAT_C8:
+       case DRM_FORMAT_RGB565:
+       case DRM_FORMAT_XRGB1555:
+       case DRM_FORMAT_XRGB8888:
+               return modifier == DRM_FORMAT_MOD_LINEAR ||
+                       modifier == I915_FORMAT_MOD_X_TILED;
+       default:
+               return false;
+       }
+}
+
+static bool i965_mod_supported(uint32_t format, uint64_t modifier)
+{
+       switch (format) {
+       case DRM_FORMAT_C8:
+       case DRM_FORMAT_RGB565:
+       case DRM_FORMAT_XRGB8888:
+       case DRM_FORMAT_XBGR8888:
+       case DRM_FORMAT_XRGB2101010:
+       case DRM_FORMAT_XBGR2101010:
+               return modifier == DRM_FORMAT_MOD_LINEAR ||
+                       modifier == I915_FORMAT_MOD_X_TILED;
+       default:
+               return false;
+       }
+}
+
+static bool skl_mod_supported(uint32_t format, uint64_t modifier)
+{
+       switch (format) {
+       case DRM_FORMAT_XRGB8888:
+       case DRM_FORMAT_XBGR8888:
+       case DRM_FORMAT_ARGB8888:
+       case DRM_FORMAT_ABGR8888:
+               if (modifier == I915_FORMAT_MOD_Yf_TILED_CCS ||
+                   modifier == I915_FORMAT_MOD_Y_TILED_CCS)
+                       return true;
+               /* fall through */
+       case DRM_FORMAT_RGB565:
+       case DRM_FORMAT_XRGB2101010:
+       case DRM_FORMAT_XBGR2101010:
+       case DRM_FORMAT_YUYV:
+       case DRM_FORMAT_YVYU:
+       case DRM_FORMAT_UYVY:
+       case DRM_FORMAT_VYUY:
+               if (modifier == I915_FORMAT_MOD_Yf_TILED)
+                       return true;
+               /* fall through */
+       case DRM_FORMAT_C8:
+               if (modifier == DRM_FORMAT_MOD_LINEAR ||
+                   modifier == I915_FORMAT_MOD_X_TILED ||
+                   modifier == I915_FORMAT_MOD_Y_TILED)
+                       return true;
+               /* fall through */
+       default:
+               return false;
+       }
+}
+
+static bool intel_primary_plane_format_mod_supported(struct drm_plane *plane,
+                                                    uint32_t format,
+                                                    uint64_t modifier)
+{
+       struct drm_i915_private *dev_priv = to_i915(plane->dev);
+
+       if (WARN_ON(modifier == DRM_FORMAT_MOD_INVALID))
+               return false;
+
+       if ((modifier >> 56) != DRM_FORMAT_MOD_VENDOR_INTEL &&
+           modifier != DRM_FORMAT_MOD_LINEAR)
+               return false;
+
+       if (INTEL_GEN(dev_priv) >= 9)
+               return skl_mod_supported(format, modifier);
+       else if (INTEL_GEN(dev_priv) >= 4)
+               return i965_mod_supported(format, modifier);
+       else
+               return i8xx_mod_supported(format, modifier);
+
+       unreachable();
+}
+
+static bool intel_cursor_plane_format_mod_supported(struct drm_plane *plane,
+                                                   uint32_t format,
+                                                   uint64_t modifier)
+{
+       if (WARN_ON(modifier == DRM_FORMAT_MOD_INVALID))
+               return false;
+
+       return modifier == DRM_FORMAT_MOD_LINEAR && format == DRM_FORMAT_ARGB8888;
+}
+
+static struct drm_plane_funcs intel_plane_funcs = {
        .update_plane = drm_atomic_helper_update_plane,
        .disable_plane = drm_atomic_helper_disable_plane,
        .destroy = intel_plane_destroy,
@@ -13519,6 +12961,7 @@ const struct drm_plane_funcs intel_plane_funcs = {
        .atomic_set_property = intel_plane_atomic_set_property,
        .atomic_duplicate_state = intel_plane_duplicate_state,
        .atomic_destroy_state = intel_plane_destroy_state,
+       .format_mod_supported = intel_primary_plane_format_mod_supported,
 };
 
 static int
@@ -13537,7 +12980,7 @@ intel_legacy_cursor_update(struct drm_plane *plane,
        struct intel_plane *intel_plane = to_intel_plane(plane);
        struct drm_framebuffer *old_fb;
        struct drm_crtc_state *crtc_state = crtc->state;
-       struct i915_vma *old_vma;
+       struct i915_vma *old_vma, *vma;
 
        /*
         * When crtc is inactive or there is a modeset pending,
@@ -13603,8 +13046,6 @@ intel_legacy_cursor_update(struct drm_plane *plane,
                        goto out_unlock;
                }
        } else {
-               struct i915_vma *vma;
-
                vma = intel_pin_and_fence_fb_obj(fb, new_plane_state->rotation);
                if (IS_ERR(vma)) {
                        DRM_DEBUG_KMS("failed to pin object\n");
@@ -13661,6 +13102,7 @@ static const struct drm_plane_funcs intel_cursor_plane_funcs = {
        .atomic_set_property = intel_plane_atomic_set_property,
        .atomic_duplicate_state = intel_plane_duplicate_state,
        .atomic_destroy_state = intel_plane_destroy_state,
+       .format_mod_supported = intel_cursor_plane_format_mod_supported,
 };
 
 static struct intel_plane *
@@ -13671,6 +13113,7 @@ intel_primary_plane_create(struct drm_i915_private *dev_priv, enum pipe pipe)
        const uint32_t *intel_primary_formats;
        unsigned int supported_rotations;
        unsigned int num_formats;
+       const uint64_t *modifiers;
        int ret;
 
        primary = kzalloc(sizeof(*primary), GFP_KERNEL);
@@ -13706,21 +13149,34 @@ intel_primary_plane_create(struct drm_i915_private *dev_priv, enum pipe pipe)
        primary->frontbuffer_bit = INTEL_FRONTBUFFER_PRIMARY(pipe);
        primary->check_plane = intel_check_primary_plane;
 
-       if (INTEL_GEN(dev_priv) >= 9) {
+       if (INTEL_GEN(dev_priv) >= 10 || IS_GEMINILAKE(dev_priv)) {
+               intel_primary_formats = skl_primary_formats;
+               num_formats = ARRAY_SIZE(skl_primary_formats);
+               modifiers = skl_format_modifiers_ccs;
+
+               primary->update_plane = skylake_update_primary_plane;
+               primary->disable_plane = skylake_disable_primary_plane;
+       } else if (INTEL_GEN(dev_priv) >= 9) {
                intel_primary_formats = skl_primary_formats;
                num_formats = ARRAY_SIZE(skl_primary_formats);
+               if (pipe < PIPE_C)
+                       modifiers = skl_format_modifiers_ccs;
+               else
+                       modifiers = skl_format_modifiers_noccs;
 
                primary->update_plane = skylake_update_primary_plane;
                primary->disable_plane = skylake_disable_primary_plane;
        } else if (INTEL_GEN(dev_priv) >= 4) {
                intel_primary_formats = i965_primary_formats;
                num_formats = ARRAY_SIZE(i965_primary_formats);
+               modifiers = i9xx_format_modifiers;
 
                primary->update_plane = i9xx_update_primary_plane;
                primary->disable_plane = i9xx_disable_primary_plane;
        } else {
                intel_primary_formats = i8xx_primary_formats;
                num_formats = ARRAY_SIZE(i8xx_primary_formats);
+               modifiers = i9xx_format_modifiers;
 
                primary->update_plane = i9xx_update_primary_plane;
                primary->disable_plane = i9xx_disable_primary_plane;
@@ -13730,21 +13186,21 @@ intel_primary_plane_create(struct drm_i915_private *dev_priv, enum pipe pipe)
                ret = drm_universal_plane_init(&dev_priv->drm, &primary->base,
                                               0, &intel_plane_funcs,
                                               intel_primary_formats, num_formats,
-                                              NULL,
+                                              modifiers,
                                               DRM_PLANE_TYPE_PRIMARY,
                                               "plane 1%c", pipe_name(pipe));
        else if (INTEL_GEN(dev_priv) >= 5 || IS_G4X(dev_priv))
                ret = drm_universal_plane_init(&dev_priv->drm, &primary->base,
                                               0, &intel_plane_funcs,
                                               intel_primary_formats, num_formats,
-                                              NULL,
+                                              modifiers,
                                               DRM_PLANE_TYPE_PRIMARY,
                                               "primary %c", pipe_name(pipe));
        else
                ret = drm_universal_plane_init(&dev_priv->drm, &primary->base,
                                               0, &intel_plane_funcs,
                                               intel_primary_formats, num_formats,
-                                              NULL,
+                                              modifiers,
                                               DRM_PLANE_TYPE_PRIMARY,
                                               "plane %c", plane_name(primary->plane));
        if (ret)
@@ -13830,7 +13286,8 @@ intel_cursor_plane_create(struct drm_i915_private *dev_priv,
                                       0, &intel_cursor_plane_funcs,
                                       intel_cursor_formats,
                                       ARRAY_SIZE(intel_cursor_formats),
-                                      NULL, DRM_PLANE_TYPE_CURSOR,
+                                      cursor_format_modifiers,
+                                      DRM_PLANE_TYPE_CURSOR,
                                       "cursor %c", pipe_name(pipe));
        if (ret)
                goto fail;
@@ -14352,10 +13809,12 @@ static int intel_framebuffer_init(struct intel_framebuffer *intel_fb,
                                  struct drm_mode_fb_cmd2 *mode_cmd)
 {
        struct drm_i915_private *dev_priv = to_i915(obj->base.dev);
+       struct drm_framebuffer *fb = &intel_fb->base;
        struct drm_format_name_buf format_name;
-       u32 pitch_limit, stride_alignment;
+       u32 pitch_limit;
        unsigned int tiling, stride;
        int ret = -EINVAL;
+       int i;
 
        i915_gem_object_lock(obj);
        obj->framebuffer_references++;
@@ -14384,6 +13843,19 @@ static int intel_framebuffer_init(struct intel_framebuffer *intel_fb,
 
        /* Passed in modifier sanity checking. */
        switch (mode_cmd->modifier[0]) {
+       case I915_FORMAT_MOD_Y_TILED_CCS:
+       case I915_FORMAT_MOD_Yf_TILED_CCS:
+               switch (mode_cmd->pixel_format) {
+               case DRM_FORMAT_XBGR8888:
+               case DRM_FORMAT_ABGR8888:
+               case DRM_FORMAT_XRGB8888:
+               case DRM_FORMAT_ARGB8888:
+                       break;
+               default:
+                       DRM_DEBUG_KMS("RC supported only with RGB8888 formats\n");
+                       goto err;
+               }
+               /* fall through */
        case I915_FORMAT_MOD_Y_TILED:
        case I915_FORMAT_MOD_Yf_TILED:
                if (INTEL_GEN(dev_priv) < 9) {
@@ -14488,25 +13960,46 @@ static int intel_framebuffer_init(struct intel_framebuffer *intel_fb,
        if (mode_cmd->offsets[0] != 0)
                goto err;
 
-       drm_helper_mode_fill_fb_struct(&dev_priv->drm,
-                                      &intel_fb->base, mode_cmd);
+       drm_helper_mode_fill_fb_struct(&dev_priv->drm, fb, mode_cmd);
 
-       stride_alignment = intel_fb_stride_alignment(&intel_fb->base, 0);
-       if (mode_cmd->pitches[0] & (stride_alignment - 1)) {
-               DRM_DEBUG_KMS("pitch (%d) must be at least %u byte aligned\n",
-                             mode_cmd->pitches[0], stride_alignment);
-               goto err;
+       for (i = 0; i < fb->format->num_planes; i++) {
+               u32 stride_alignment;
+
+               if (mode_cmd->handles[i] != mode_cmd->handles[0]) {
+                       DRM_DEBUG_KMS("bad plane %d handle\n", i);
+                       goto err;
+               }
+
+               stride_alignment = intel_fb_stride_alignment(fb, i);
+
+               /*
+                * Display WA #0531: skl,bxt,kbl,glk
+                *
+                * Render decompression and plane width > 3840
+                * combined with horizontal panning requires the
+                * plane stride to be a multiple of 4. We'll just
+                * require the entire fb to accommodate that to avoid
+                * potential runtime errors at plane configuration time.
+                */
+               if (IS_GEN9(dev_priv) && i == 0 && fb->width > 3840 &&
+                   (fb->modifier == I915_FORMAT_MOD_Y_TILED_CCS ||
+                    fb->modifier == I915_FORMAT_MOD_Yf_TILED_CCS))
+                       stride_alignment *= 4;
+
+               if (fb->pitches[i] & (stride_alignment - 1)) {
+                       DRM_DEBUG_KMS("plane %d pitch (%d) must be at least %u byte aligned\n",
+                                     i, fb->pitches[i], stride_alignment);
+                       goto err;
+               }
        }
 
        intel_fb->obj = obj;
 
-       ret = intel_fill_fb_info(dev_priv, &intel_fb->base);
+       ret = intel_fill_fb_info(dev_priv, fb);
        if (ret)
                goto err;
 
-       ret = drm_framebuffer_init(obj->base.dev,
-                                  &intel_fb->base,
-                                  &intel_fb_funcs);
+       ret = drm_framebuffer_init(&dev_priv->drm, fb, &intel_fb_funcs);
        if (ret) {
                DRM_ERROR("framebuffer init failed %d\n", ret);
                goto err;
@@ -14554,6 +14047,7 @@ static void intel_atomic_state_free(struct drm_atomic_state *state)
 
 static const struct drm_mode_config_funcs intel_mode_funcs = {
        .fb_create = intel_user_framebuffer_create,
+       .get_format_info = intel_get_format_info,
        .output_poll_changed = intel_fbdev_output_poll_changed,
        .atomic_check = intel_atomic_check,
        .atomic_commit = intel_atomic_commit,
@@ -14653,34 +14147,6 @@ void intel_init_display_hooks(struct drm_i915_private *dev_priv)
                dev_priv->display.update_crtcs = skl_update_crtcs;
        else
                dev_priv->display.update_crtcs = intel_update_crtcs;
-
-       switch (INTEL_INFO(dev_priv)->gen) {
-       case 2:
-               dev_priv->display.queue_flip = intel_gen2_queue_flip;
-               break;
-
-       case 3:
-               dev_priv->display.queue_flip = intel_gen3_queue_flip;
-               break;
-
-       case 4:
-       case 5:
-               dev_priv->display.queue_flip = intel_gen4_queue_flip;
-               break;
-
-       case 6:
-               dev_priv->display.queue_flip = intel_gen6_queue_flip;
-               break;
-       case 7:
-       case 8: /* FIXME(BDW): Check that the gen8 RCS flip works. */
-               dev_priv->display.queue_flip = intel_gen7_queue_flip;
-               break;
-       case 9:
-               /* Drop through - unsupported since execlist only. */
-       default:
-               /* Default just returns -ENODEV to indicate unsupported */
-               dev_priv->display.queue_flip = intel_default_queue_flip;
-       }
 }
 
 /*
@@ -15611,7 +15077,7 @@ intel_modeset_setup_hw_state(struct drm_device *dev,
        } else if (IS_VALLEYVIEW(dev_priv) || IS_CHERRYVIEW(dev_priv)) {
                vlv_wm_get_hw_state(dev);
                vlv_wm_sanitize(dev_priv);
-       } else if (IS_GEN9(dev_priv)) {
+       } else if (INTEL_GEN(dev_priv) >= 9) {
                skl_wm_get_hw_state(dev);
        } else if (HAS_PCH_SPLIT(dev_priv)) {
                ilk_wm_get_hw_state(dev);
@@ -15718,6 +15184,9 @@ void intel_modeset_cleanup(struct drm_device *dev)
         */
        drm_kms_helper_poll_fini(dev);
 
+       /* poll work can call into fbdev, hence clean that up afterwards */
+       intel_fbdev_fini(dev_priv);
+
        intel_unregister_dsm_handler();
 
        intel_fbc_global_disable(dev_priv);
@@ -15837,7 +15306,8 @@ intel_display_capture_error_state(struct drm_i915_private *dev_priv)
                return NULL;
 
        if (IS_HASWELL(dev_priv) || IS_BROADWELL(dev_priv))
-               error->power_well_driver = I915_READ(HSW_PWR_WELL_DRIVER);
+               error->power_well_driver =
+                       I915_READ(HSW_PWR_WELL_CTL_DRIVER(HSW_DISP_PW_GLOBAL));
 
        for_each_pipe(dev_priv, i) {
                error->pipe[i].power_domain_on =