Merge branch 'drm-intel-next' of git://anongit.freedesktop.org/drm-intel into drm...
[cascardo/linux.git] / drivers / gpu / drm / i915 / intel_display.c
index 2113f40..3f8987b 100644 (file)
 #include <linux/reservation.h>
 #include <linux/dma-buf.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,
@@ -117,20 +122,17 @@ static void ironlake_pfit_disable(struct intel_crtc *crtc, bool force);
 static void ironlake_pfit_enable(struct intel_crtc *crtc);
 static void intel_modeset_setup_hw_state(struct drm_device *dev);
 static void intel_pre_disable_primary_noatomic(struct drm_crtc *crtc);
+static int ilk_max_pixel_rate(struct drm_atomic_state *state);
 
-typedef struct {
-       int     min, max;
-} intel_range_t;
-
-typedef struct {
-       int     dot_limit;
-       int     p2_slow, p2_fast;
-} intel_p2_t;
-
-typedef struct intel_limit intel_limit_t;
 struct intel_limit {
-       intel_range_t   dot, vco, n, m, m1, m2, p, p1;
-       intel_p2_t          p2;
+       struct {
+               int min, max;
+       } dot, vco, n, m, m1, m2, p, p1;
+
+       struct {
+               int dot_limit;
+               int p2_slow, p2_fast;
+       } p2;
 };
 
 /* returns HPLL frequency in kHz */
@@ -185,6 +187,7 @@ intel_pch_rawclk(struct drm_i915_private *dev_priv)
 static int
 intel_vlv_hrawclk(struct drm_i915_private *dev_priv)
 {
+       /* RAWCLK_FREQ_VLV register updated from power well code */
        return vlv_get_cck_clock_hpll(dev_priv, "hrawclk",
                                      CCK_DISPLAY_REF_CLOCK_CONTROL);
 }
@@ -218,7 +221,7 @@ intel_g4x_hrawclk(struct drm_i915_private *dev_priv)
        }
 }
 
-static void intel_update_rawclk(struct drm_i915_private *dev_priv)
+void intel_update_rawclk(struct drm_i915_private *dev_priv)
 {
        if (HAS_PCH_SPLIT(dev_priv))
                dev_priv->rawclk_freq = intel_pch_rawclk(dev_priv);
@@ -255,7 +258,7 @@ intel_fdi_link_freq(struct drm_i915_private *dev_priv,
                return 270000;
 }
 
-static const intel_limit_t intel_limits_i8xx_dac = {
+static const struct intel_limit intel_limits_i8xx_dac = {
        .dot = { .min = 25000, .max = 350000 },
        .vco = { .min = 908000, .max = 1512000 },
        .n = { .min = 2, .max = 16 },
@@ -268,7 +271,7 @@ static const intel_limit_t intel_limits_i8xx_dac = {
                .p2_slow = 4, .p2_fast = 2 },
 };
 
-static const intel_limit_t intel_limits_i8xx_dvo = {
+static const struct intel_limit intel_limits_i8xx_dvo = {
        .dot = { .min = 25000, .max = 350000 },
        .vco = { .min = 908000, .max = 1512000 },
        .n = { .min = 2, .max = 16 },
@@ -281,7 +284,7 @@ static const intel_limit_t intel_limits_i8xx_dvo = {
                .p2_slow = 4, .p2_fast = 4 },
 };
 
-static const intel_limit_t intel_limits_i8xx_lvds = {
+static const struct intel_limit intel_limits_i8xx_lvds = {
        .dot = { .min = 25000, .max = 350000 },
        .vco = { .min = 908000, .max = 1512000 },
        .n = { .min = 2, .max = 16 },
@@ -294,7 +297,7 @@ static const intel_limit_t intel_limits_i8xx_lvds = {
                .p2_slow = 14, .p2_fast = 7 },
 };
 
-static const intel_limit_t intel_limits_i9xx_sdvo = {
+static const struct intel_limit intel_limits_i9xx_sdvo = {
        .dot = { .min = 20000, .max = 400000 },
        .vco = { .min = 1400000, .max = 2800000 },
        .n = { .min = 1, .max = 6 },
@@ -307,7 +310,7 @@ static const intel_limit_t intel_limits_i9xx_sdvo = {
                .p2_slow = 10, .p2_fast = 5 },
 };
 
-static const intel_limit_t intel_limits_i9xx_lvds = {
+static const struct intel_limit intel_limits_i9xx_lvds = {
        .dot = { .min = 20000, .max = 400000 },
        .vco = { .min = 1400000, .max = 2800000 },
        .n = { .min = 1, .max = 6 },
@@ -321,7 +324,7 @@ static const intel_limit_t intel_limits_i9xx_lvds = {
 };
 
 
-static const intel_limit_t intel_limits_g4x_sdvo = {
+static const struct intel_limit intel_limits_g4x_sdvo = {
        .dot = { .min = 25000, .max = 270000 },
        .vco = { .min = 1750000, .max = 3500000},
        .n = { .min = 1, .max = 4 },
@@ -336,7 +339,7 @@ static const intel_limit_t intel_limits_g4x_sdvo = {
        },
 };
 
-static const intel_limit_t intel_limits_g4x_hdmi = {
+static const struct intel_limit intel_limits_g4x_hdmi = {
        .dot = { .min = 22000, .max = 400000 },
        .vco = { .min = 1750000, .max = 3500000},
        .n = { .min = 1, .max = 4 },
@@ -349,7 +352,7 @@ static const intel_limit_t intel_limits_g4x_hdmi = {
                .p2_slow = 10, .p2_fast = 5 },
 };
 
-static const intel_limit_t intel_limits_g4x_single_channel_lvds = {
+static const struct intel_limit intel_limits_g4x_single_channel_lvds = {
        .dot = { .min = 20000, .max = 115000 },
        .vco = { .min = 1750000, .max = 3500000 },
        .n = { .min = 1, .max = 3 },
@@ -363,7 +366,7 @@ static const intel_limit_t intel_limits_g4x_single_channel_lvds = {
        },
 };
 
-static const intel_limit_t intel_limits_g4x_dual_channel_lvds = {
+static const struct intel_limit intel_limits_g4x_dual_channel_lvds = {
        .dot = { .min = 80000, .max = 224000 },
        .vco = { .min = 1750000, .max = 3500000 },
        .n = { .min = 1, .max = 3 },
@@ -377,7 +380,7 @@ static const intel_limit_t intel_limits_g4x_dual_channel_lvds = {
        },
 };
 
-static const intel_limit_t intel_limits_pineview_sdvo = {
+static const struct intel_limit intel_limits_pineview_sdvo = {
        .dot = { .min = 20000, .max = 400000},
        .vco = { .min = 1700000, .max = 3500000 },
        /* Pineview's Ncounter is a ring counter */
@@ -392,7 +395,7 @@ static const intel_limit_t intel_limits_pineview_sdvo = {
                .p2_slow = 10, .p2_fast = 5 },
 };
 
-static const intel_limit_t intel_limits_pineview_lvds = {
+static const struct intel_limit intel_limits_pineview_lvds = {
        .dot = { .min = 20000, .max = 400000 },
        .vco = { .min = 1700000, .max = 3500000 },
        .n = { .min = 3, .max = 6 },
@@ -410,7 +413,7 @@ static const intel_limit_t intel_limits_pineview_lvds = {
  * We calculate clock using (register_value + 2) for N/M1/M2, so here
  * the range value for them is (actual_value - 2).
  */
-static const intel_limit_t intel_limits_ironlake_dac = {
+static const struct intel_limit intel_limits_ironlake_dac = {
        .dot = { .min = 25000, .max = 350000 },
        .vco = { .min = 1760000, .max = 3510000 },
        .n = { .min = 1, .max = 5 },
@@ -423,7 +426,7 @@ static const intel_limit_t intel_limits_ironlake_dac = {
                .p2_slow = 10, .p2_fast = 5 },
 };
 
-static const intel_limit_t intel_limits_ironlake_single_lvds = {
+static const struct intel_limit intel_limits_ironlake_single_lvds = {
        .dot = { .min = 25000, .max = 350000 },
        .vco = { .min = 1760000, .max = 3510000 },
        .n = { .min = 1, .max = 3 },
@@ -436,7 +439,7 @@ static const intel_limit_t intel_limits_ironlake_single_lvds = {
                .p2_slow = 14, .p2_fast = 14 },
 };
 
-static const intel_limit_t intel_limits_ironlake_dual_lvds = {
+static const struct intel_limit intel_limits_ironlake_dual_lvds = {
        .dot = { .min = 25000, .max = 350000 },
        .vco = { .min = 1760000, .max = 3510000 },
        .n = { .min = 1, .max = 3 },
@@ -450,7 +453,7 @@ static const intel_limit_t intel_limits_ironlake_dual_lvds = {
 };
 
 /* LVDS 100mhz refclk limits. */
-static const intel_limit_t intel_limits_ironlake_single_lvds_100m = {
+static const struct intel_limit intel_limits_ironlake_single_lvds_100m = {
        .dot = { .min = 25000, .max = 350000 },
        .vco = { .min = 1760000, .max = 3510000 },
        .n = { .min = 1, .max = 2 },
@@ -463,7 +466,7 @@ static const intel_limit_t intel_limits_ironlake_single_lvds_100m = {
                .p2_slow = 14, .p2_fast = 14 },
 };
 
-static const intel_limit_t intel_limits_ironlake_dual_lvds_100m = {
+static const struct intel_limit intel_limits_ironlake_dual_lvds_100m = {
        .dot = { .min = 25000, .max = 350000 },
        .vco = { .min = 1760000, .max = 3510000 },
        .n = { .min = 1, .max = 3 },
@@ -476,7 +479,7 @@ static const intel_limit_t intel_limits_ironlake_dual_lvds_100m = {
                .p2_slow = 7, .p2_fast = 7 },
 };
 
-static const intel_limit_t intel_limits_vlv = {
+static const struct intel_limit intel_limits_vlv = {
         /*
          * These are the data rate limits (measured in fast clocks)
          * since those are the strictest limits we have. The fast
@@ -492,7 +495,7 @@ static const intel_limit_t intel_limits_vlv = {
        .p2 = { .p2_slow = 2, .p2_fast = 20 }, /* slow=min, fast=max */
 };
 
-static const intel_limit_t intel_limits_chv = {
+static const struct intel_limit intel_limits_chv = {
        /*
         * These are the data rate limits (measured in fast clocks)
         * since those are the strictest limits we have.  The fast
@@ -508,7 +511,7 @@ static const intel_limit_t intel_limits_chv = {
        .p2 = { .p2_slow = 1, .p2_fast = 14 },
 };
 
-static const intel_limit_t intel_limits_bxt = {
+static const struct intel_limit intel_limits_bxt = {
        /* FIXME: find real dot limits */
        .dot = { .min = 0, .max = INT_MAX },
        .vco = { .min = 4800000, .max = 6700000 },
@@ -581,7 +584,7 @@ static bool intel_pipe_will_have_type(const struct intel_crtc_state *crtc_state,
  * divided-down version of it.
  */
 /* m1 is reserved as 0 in Pineview, n is a ring counter */
-static int pnv_calc_dpll_params(int refclk, intel_clock_t *clock)
+static int pnv_calc_dpll_params(int refclk, struct dpll *clock)
 {
        clock->m = clock->m2 + 2;
        clock->p = clock->p1 * clock->p2;
@@ -598,7 +601,7 @@ static uint32_t i9xx_dpll_compute_m(struct dpll *dpll)
        return 5 * (dpll->m1 + 2) + (dpll->m2 + 2);
 }
 
-static int i9xx_calc_dpll_params(int refclk, intel_clock_t *clock)
+static int i9xx_calc_dpll_params(int refclk, struct dpll *clock)
 {
        clock->m = i9xx_dpll_compute_m(clock);
        clock->p = clock->p1 * clock->p2;
@@ -610,7 +613,7 @@ static int i9xx_calc_dpll_params(int refclk, intel_clock_t *clock)
        return clock->dot;
 }
 
-static int vlv_calc_dpll_params(int refclk, intel_clock_t *clock)
+static int vlv_calc_dpll_params(int refclk, struct dpll *clock)
 {
        clock->m = clock->m1 * clock->m2;
        clock->p = clock->p1 * clock->p2;
@@ -622,7 +625,7 @@ static int vlv_calc_dpll_params(int refclk, intel_clock_t *clock)
        return clock->dot / 5;
 }
 
-int chv_calc_dpll_params(int refclk, intel_clock_t *clock)
+int chv_calc_dpll_params(int refclk, struct dpll *clock)
 {
        clock->m = clock->m1 * clock->m2;
        clock->p = clock->p1 * clock->p2;
@@ -642,8 +645,8 @@ int chv_calc_dpll_params(int refclk, intel_clock_t *clock)
  */
 
 static bool intel_PLL_is_valid(struct drm_device *dev,
-                              const intel_limit_t *limit,
-                              const intel_clock_t *clock)
+                              const struct intel_limit *limit,
+                              const struct dpll *clock)
 {
        if (clock->n   < limit->n.min   || limit->n.max   < clock->n)
                INTELPllInvalid("n out of range\n");
@@ -678,7 +681,7 @@ static bool intel_PLL_is_valid(struct drm_device *dev,
 }
 
 static int
-i9xx_select_p2_div(const intel_limit_t *limit,
+i9xx_select_p2_div(const struct intel_limit *limit,
                   const struct intel_crtc_state *crtc_state,
                   int target)
 {
@@ -713,13 +716,13 @@ i9xx_select_p2_div(const intel_limit_t *limit,
  * divider from @match_clock used for LVDS downclocking.
  */
 static bool
-i9xx_find_best_dpll(const intel_limit_t *limit,
+i9xx_find_best_dpll(const struct intel_limit *limit,
                    struct intel_crtc_state *crtc_state,
-                   int target, int refclk, intel_clock_t *match_clock,
-                   intel_clock_t *best_clock)
+                   int target, int refclk, struct dpll *match_clock,
+                   struct dpll *best_clock)
 {
        struct drm_device *dev = crtc_state->base.crtc->dev;
-       intel_clock_t clock;
+       struct dpll clock;
        int err = target;
 
        memset(best_clock, 0, sizeof(*best_clock));
@@ -770,13 +773,13 @@ i9xx_find_best_dpll(const intel_limit_t *limit,
  * divider from @match_clock used for LVDS downclocking.
  */
 static bool
-pnv_find_best_dpll(const intel_limit_t *limit,
+pnv_find_best_dpll(const struct intel_limit *limit,
                   struct intel_crtc_state *crtc_state,
-                  int target, int refclk, intel_clock_t *match_clock,
-                  intel_clock_t *best_clock)
+                  int target, int refclk, struct dpll *match_clock,
+                  struct dpll *best_clock)
 {
        struct drm_device *dev = crtc_state->base.crtc->dev;
-       intel_clock_t clock;
+       struct dpll clock;
        int err = target;
 
        memset(best_clock, 0, sizeof(*best_clock));
@@ -825,13 +828,13 @@ pnv_find_best_dpll(const intel_limit_t *limit,
  * divider from @match_clock used for LVDS downclocking.
  */
 static bool
-g4x_find_best_dpll(const intel_limit_t *limit,
+g4x_find_best_dpll(const struct intel_limit *limit,
                   struct intel_crtc_state *crtc_state,
-                  int target, int refclk, intel_clock_t *match_clock,
-                  intel_clock_t *best_clock)
+                  int target, int refclk, struct dpll *match_clock,
+                  struct dpll *best_clock)
 {
        struct drm_device *dev = crtc_state->base.crtc->dev;
-       intel_clock_t clock;
+       struct dpll clock;
        int max_n;
        bool found = false;
        /* approximately equals target * 0.00585 */
@@ -877,8 +880,8 @@ g4x_find_best_dpll(const intel_limit_t *limit,
  * best configuration and error found so far. Return the calculated error.
  */
 static bool vlv_PLL_is_optimal(struct drm_device *dev, int target_freq,
-                              const intel_clock_t *calculated_clock,
-                              const intel_clock_t *best_clock,
+                              const struct dpll *calculated_clock,
+                              const struct dpll *best_clock,
                               unsigned int best_error_ppm,
                               unsigned int *error_ppm)
 {
@@ -918,14 +921,14 @@ static bool vlv_PLL_is_optimal(struct drm_device *dev, int target_freq,
  * reflck * (5 * (m1 + 2) + (m2 + 2)) / (n + 2) / p1 / p2.
  */
 static bool
-vlv_find_best_dpll(const intel_limit_t *limit,
+vlv_find_best_dpll(const struct intel_limit *limit,
                   struct intel_crtc_state *crtc_state,
-                  int target, int refclk, intel_clock_t *match_clock,
-                  intel_clock_t *best_clock)
+                  int target, int refclk, struct dpll *match_clock,
+                  struct dpll *best_clock)
 {
        struct intel_crtc *crtc = to_intel_crtc(crtc_state->base.crtc);
        struct drm_device *dev = crtc->base.dev;
-       intel_clock_t clock;
+       struct dpll clock;
        unsigned int bestppm = 1000000;
        /* min update 19.2 MHz */
        int max_n = min(limit->n.max, refclk / 19200);
@@ -977,15 +980,15 @@ vlv_find_best_dpll(const intel_limit_t *limit,
  * reflck * (5 * (m1 + 2) + (m2 + 2)) / (n + 2) / p1 / p2.
  */
 static bool
-chv_find_best_dpll(const intel_limit_t *limit,
+chv_find_best_dpll(const struct intel_limit *limit,
                   struct intel_crtc_state *crtc_state,
-                  int target, int refclk, intel_clock_t *match_clock,
-                  intel_clock_t *best_clock)
+                  int target, int refclk, struct dpll *match_clock,
+                  struct dpll *best_clock)
 {
        struct intel_crtc *crtc = to_intel_crtc(crtc_state->base.crtc);
        struct drm_device *dev = crtc->base.dev;
        unsigned int best_error_ppm;
-       intel_clock_t clock;
+       struct dpll clock;
        uint64_t m2;
        int found = false;
 
@@ -1035,10 +1038,10 @@ chv_find_best_dpll(const intel_limit_t *limit,
 }
 
 bool bxt_find_best_dpll(struct intel_crtc_state *crtc_state, int target_clock,
-                       intel_clock_t *best_clock)
+                       struct dpll *best_clock)
 {
        int refclk = 100000;
-       const intel_limit_t *limit = &intel_limits_bxt;
+       const struct intel_limit *limit = &intel_limits_bxt;
 
        return chv_find_best_dpll(limit, crtc_state,
                                  target_clock, refclk, NULL, best_clock);
@@ -1203,7 +1206,7 @@ static void assert_fdi_tx_pll_enabled(struct drm_i915_private *dev_priv,
        u32 val;
 
        /* ILK FDI PLL is always enabled */
-       if (INTEL_INFO(dev_priv)->gen == 5)
+       if (IS_GEN5(dev_priv))
                return;
 
        /* On Haswell, DDI ports are responsible for the FDI PLL setup */
@@ -2309,7 +2312,7 @@ err_pm:
        return ret;
 }
 
-static void intel_unpin_fb_obj(struct drm_framebuffer *fb, unsigned int rotation)
+void intel_unpin_fb_obj(struct drm_framebuffer *fb, unsigned int rotation)
 {
        struct drm_i915_gem_object *obj = intel_fb_obj(fb);
        struct i915_ggtt_view view;
@@ -3110,17 +3113,12 @@ intel_pipe_set_base_atomic(struct drm_crtc *crtc, struct drm_framebuffer *fb,
        return -ENODEV;
 }
 
-static void intel_complete_page_flips(struct drm_device *dev)
+static void intel_complete_page_flips(struct drm_i915_private *dev_priv)
 {
-       struct drm_crtc *crtc;
-
-       for_each_crtc(dev, crtc) {
-               struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
-               enum plane plane = intel_crtc->plane;
+       struct intel_crtc *crtc;
 
-               intel_prepare_page_flip(dev, plane);
-               intel_finish_page_flip_plane(dev, plane);
-       }
+       for_each_intel_crtc(dev_priv->dev, crtc)
+               intel_finish_page_flip_cs(dev_priv, crtc->pipe);
 }
 
 static void intel_update_primary_planes(struct drm_device *dev)
@@ -3143,41 +3141,39 @@ static void intel_update_primary_planes(struct drm_device *dev)
        }
 }
 
-void intel_prepare_reset(struct drm_device *dev)
+void intel_prepare_reset(struct drm_i915_private *dev_priv)
 {
        /* no reset support for gen2 */
-       if (IS_GEN2(dev))
+       if (IS_GEN2(dev_priv))
                return;
 
        /* reset doesn't touch the display */
-       if (INTEL_INFO(dev)->gen >= 5 || IS_G4X(dev))
+       if (INTEL_GEN(dev_priv) >= 5 || IS_G4X(dev_priv))
                return;
 
-       drm_modeset_lock_all(dev);
+       drm_modeset_lock_all(dev_priv->dev);
        /*
         * Disabling the crtcs gracefully seems nicer. Also the
         * g33 docs say we should at least disable all the planes.
         */
-       intel_display_suspend(dev);
+       intel_display_suspend(dev_priv->dev);
 }
 
-void intel_finish_reset(struct drm_device *dev)
+void intel_finish_reset(struct drm_i915_private *dev_priv)
 {
-       struct drm_i915_private *dev_priv = to_i915(dev);
-
        /*
         * 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);
+       intel_complete_page_flips(dev_priv);
 
        /* no reset support for gen2 */
-       if (IS_GEN2(dev))
+       if (IS_GEN2(dev_priv))
                return;
 
        /* reset doesn't touch the display */
-       if (INTEL_INFO(dev)->gen >= 5 || IS_G4X(dev)) {
+       if (INTEL_GEN(dev_priv) >= 5 || IS_G4X(dev_priv)) {
                /*
                 * Flips in the rings have been nuked by the reset,
                 * so update the base address of all primary
@@ -3187,7 +3183,7 @@ void intel_finish_reset(struct drm_device *dev)
                 * FIXME: Atomic will make this obsolete since we won't schedule
                 * CS-based flips (which might get lost in gpu resets) any more.
                 */
-               intel_update_primary_planes(dev);
+               intel_update_primary_planes(dev_priv->dev);
                return;
        }
 
@@ -3198,18 +3194,18 @@ void intel_finish_reset(struct drm_device *dev)
        intel_runtime_pm_disable_interrupts(dev_priv);
        intel_runtime_pm_enable_interrupts(dev_priv);
 
-       intel_modeset_init_hw(dev);
+       intel_modeset_init_hw(dev_priv->dev);
 
        spin_lock_irq(&dev_priv->irq_lock);
        if (dev_priv->display.hpd_irq_setup)
-               dev_priv->display.hpd_irq_setup(dev);
+               dev_priv->display.hpd_irq_setup(dev_priv);
        spin_unlock_irq(&dev_priv->irq_lock);
 
-       intel_display_resume(dev);
+       intel_display_resume(dev_priv->dev);
 
        intel_hpd_init(dev_priv);
 
-       drm_modeset_unlock_all(dev);
+       drm_modeset_unlock_all(dev_priv->dev);
 }
 
 static bool intel_crtc_has_pending_flip(struct drm_crtc *crtc)
@@ -3224,7 +3220,7 @@ static bool intel_crtc_has_pending_flip(struct drm_crtc *crtc)
                return false;
 
        spin_lock_irq(&dev->event_lock);
-       pending = to_intel_crtc(crtc)->unpin_work != NULL;
+       pending = to_intel_crtc(crtc)->flip_work != NULL;
        spin_unlock_irq(&dev->event_lock);
 
        return pending;
@@ -3803,7 +3799,7 @@ bool intel_has_pending_fb_unpin(struct drm_device *dev)
                if (atomic_read(&crtc->unpin_work_count) == 0)
                        continue;
 
-               if (crtc->unpin_work)
+               if (crtc->flip_work)
                        intel_wait_for_vblank(dev, crtc->pipe);
 
                return true;
@@ -3815,11 +3811,9 @@ bool intel_has_pending_fb_unpin(struct drm_device *dev)
 static void page_flip_completed(struct intel_crtc *intel_crtc)
 {
        struct drm_i915_private *dev_priv = to_i915(intel_crtc->base.dev);
-       struct intel_unpin_work *work = intel_crtc->unpin_work;
+       struct intel_flip_work *work = intel_crtc->flip_work;
 
-       /* ensure that the unpin work is consistent wrt ->pending. */
-       smp_rmb();
-       intel_crtc->unpin_work = NULL;
+       intel_crtc->flip_work = NULL;
 
        if (work->event)
                drm_crtc_send_vblank_event(&intel_crtc->base, work->event);
@@ -3827,7 +3821,7 @@ static void page_flip_completed(struct intel_crtc *intel_crtc)
        drm_crtc_vblank_put(&intel_crtc->base);
 
        wake_up_all(&dev_priv->pending_flip_queue);
-       queue_work(dev_priv->wq, &work->work);
+       queue_work(dev_priv->wq, &work->unpin_work);
 
        trace_i915_flip_complete(intel_crtc->plane,
                                 work->pending_flip_obj);
@@ -3851,9 +3845,11 @@ static int intel_crtc_wait_for_pending_flips(struct drm_crtc *crtc)
 
        if (ret == 0) {
                struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
+               struct intel_flip_work *work;
 
                spin_lock_irq(&dev->event_lock);
-               if (intel_crtc->unpin_work) {
+               work = intel_crtc->flip_work;
+               if (work && !is_mmio_work(work)) {
                        WARN_ONCE(1, "Removing stuck page flip\n");
                        page_flip_completed(intel_crtc);
                }
@@ -5328,32 +5324,33 @@ static void intel_update_cdclk(struct drm_device *dev)
                         dev_priv->cdclk_freq);
 
        /*
-        * Program the gmbus_freq based on the cdclk frequency.
-        * BSpec erroneously claims we should aim for 4MHz, but
-        * in fact 1MHz is the correct frequency.
+        * 9:0 CMBUS [sic] CDCLK frequency (cdfreq):
+        * Programmng [sic] note: bit[9:2] should be programmed to the number
+        * of cdclk that generates 4MHz reference clock freq which is used to
+        * generate GMBus clock. This will vary with the cdclk freq.
         */
-       if (IS_VALLEYVIEW(dev) || IS_CHERRYVIEW(dev)) {
-               /*
-                * Program the gmbus_freq based on the cdclk frequency.
-                * BSpec erroneously claims we should aim for 4MHz, but
-                * in fact 1MHz is the correct frequency.
-                */
+       if (IS_VALLEYVIEW(dev_priv) || IS_CHERRYVIEW(dev_priv))
                I915_WRITE(GMBUSFREQ_VLV, DIV_ROUND_UP(dev_priv->cdclk_freq, 1000));
-       }
 
        if (dev_priv->max_cdclk_freq == 0)
                intel_update_max_cdclk(dev);
 }
 
-static void broxton_set_cdclk(struct drm_i915_private *dev_priv, int frequency)
+/* convert from kHz to .1 fixpoint MHz with -1MHz offset */
+static int skl_cdclk_decimal(int cdclk)
+{
+       return DIV_ROUND_CLOSEST(cdclk - 1000, 500);
+}
+
+static void broxton_set_cdclk(struct drm_i915_private *dev_priv, int cdclk)
 {
        uint32_t divider;
        uint32_t ratio;
-       uint32_t current_freq;
+       uint32_t current_cdclk;
        int ret;
 
        /* frequency = 19.2MHz * ratio / 2 / div{1,1.5,2,4} */
-       switch (frequency) {
+       switch (cdclk) {
        case 144000:
                divider = BXT_CDCLK_CD2X_DIV_SEL_4;
                ratio = BXT_DE_PLL_RATIO(60);
@@ -5383,7 +5380,7 @@ static void broxton_set_cdclk(struct drm_i915_private *dev_priv, int frequency)
                divider = 0;
                break;
        default:
-               DRM_ERROR("unsupported CDCLK freq %d", frequency);
+               DRM_ERROR("unsupported CDCLK freq %d", cdclk);
 
                return;
        }
@@ -5396,13 +5393,13 @@ static void broxton_set_cdclk(struct drm_i915_private *dev_priv, int frequency)
 
        if (ret) {
                DRM_ERROR("PCode CDCLK freq change notify failed (err %d, freq %d)\n",
-                         ret, frequency);
+                         ret, cdclk);
                return;
        }
 
-       current_freq = I915_READ(CDCLK_CTL) & CDCLK_FREQ_DECIMAL_MASK;
+       current_cdclk = I915_READ(CDCLK_CTL) & CDCLK_FREQ_DECIMAL_MASK;
        /* convert from .1 fixpoint MHz with -1MHz offset to kHz */
-       current_freq = current_freq * 500 + 1000;
+       current_cdclk = current_cdclk * 500 + 1000;
 
        /*
         * DE PLL has to be disabled when
@@ -5410,8 +5407,8 @@ static void broxton_set_cdclk(struct drm_i915_private *dev_priv, int frequency)
         * - before setting to 624MHz (PLL needs toggling)
         * - before setting to any frequency from 624MHz (PLL needs toggling)
         */
-       if (frequency == 19200 || frequency == 624000 ||
-           current_freq == 624000) {
+       if (cdclk == 19200 || cdclk == 624000 ||
+           current_cdclk == 624000) {
                I915_WRITE(BXT_DE_PLL_ENABLE, ~BXT_DE_PLL_PLL_ENABLE);
                /* Timeout 200us */
                if (wait_for(!(I915_READ(BXT_DE_PLL_ENABLE) & BXT_DE_PLL_LOCK),
@@ -5419,7 +5416,7 @@ static void broxton_set_cdclk(struct drm_i915_private *dev_priv, int frequency)
                        DRM_ERROR("timout waiting for DE PLL unlock\n");
        }
 
-       if (frequency != 19200) {
+       if (cdclk != 19200) {
                uint32_t val;
 
                val = I915_READ(BXT_DE_PLL_CTL);
@@ -5432,31 +5429,29 @@ static void broxton_set_cdclk(struct drm_i915_private *dev_priv, int frequency)
                if (wait_for(I915_READ(BXT_DE_PLL_ENABLE) & BXT_DE_PLL_LOCK, 1))
                        DRM_ERROR("timeout waiting for DE PLL lock\n");
 
-               val = I915_READ(CDCLK_CTL);
-               val &= ~BXT_CDCLK_CD2X_DIV_SEL_MASK;
-               val |= divider;
+               val = divider | skl_cdclk_decimal(cdclk);
+               /*
+                * FIXME if only the cd2x divider needs changing, it could be done
+                * without shutting off the pipe (if only one pipe is active).
+                */
+               val |= BXT_CDCLK_CD2X_PIPE_NONE;
                /*
                 * Disable SSA Precharge when CD clock frequency < 500 MHz,
                 * enable otherwise.
                 */
-               val &= ~BXT_CDCLK_SSA_PRECHARGE_ENABLE;
-               if (frequency >= 500000)
+               if (cdclk >= 500000)
                        val |= BXT_CDCLK_SSA_PRECHARGE_ENABLE;
-
-               val &= ~CDCLK_FREQ_DECIMAL_MASK;
-               /* convert from kHz to .1 fixpoint MHz with -1MHz offset */
-               val |= (frequency - 1000) / 500;
                I915_WRITE(CDCLK_CTL, val);
        }
 
        mutex_lock(&dev_priv->rps.hw_lock);
        ret = sandybridge_pcode_write(dev_priv, HSW_PCODE_DE_WRITE_FREQ_REQ,
-                                     DIV_ROUND_UP(frequency, 25000));
+                                     DIV_ROUND_UP(cdclk, 25000));
        mutex_unlock(&dev_priv->rps.hw_lock);
 
        if (ret) {
                DRM_ERROR("PCode CDCLK freq set failed, (err %d, freq %d)\n",
-                         ret, frequency);
+                         ret, cdclk);
                return;
        }
 
@@ -5545,11 +5540,6 @@ static const struct skl_cdclk_entry {
        { .freq = 675000, .vco = 8100 },
 };
 
-static unsigned int skl_cdclk_decimal(unsigned int freq)
-{
-       return (freq - 1000) / 500;
-}
-
 static unsigned int skl_cdclk_get_vco(unsigned int freq)
 {
        unsigned int i;
@@ -5565,23 +5555,18 @@ static unsigned int skl_cdclk_get_vco(unsigned int freq)
 }
 
 static void
-skl_dpll0_enable(struct drm_i915_private *dev_priv, unsigned int required_vco)
+skl_dpll0_enable(struct drm_i915_private *dev_priv, int vco)
 {
-       unsigned int min_freq;
+       int min_cdclk;
        u32 val;
 
        /* select the minimum CDCLK before enabling DPLL 0 */
-       val = I915_READ(CDCLK_CTL);
-       val &= ~CDCLK_FREQ_SEL_MASK | ~CDCLK_FREQ_DECIMAL_MASK;
-       val |= CDCLK_FREQ_337_308;
-
-       if (required_vco == 8640)
-               min_freq = 308570;
+       if (vco == 8640)
+               min_cdclk = 308570;
        else
-               min_freq = 337500;
-
-       val = CDCLK_FREQ_337_308 | skl_cdclk_decimal(min_freq);
+               min_cdclk = 337500;
 
+       val = CDCLK_FREQ_337_308 | skl_cdclk_decimal(min_cdclk);
        I915_WRITE(CDCLK_CTL, val);
        POSTING_READ(CDCLK_CTL);
 
@@ -5599,7 +5584,7 @@ skl_dpll0_enable(struct drm_i915_private *dev_priv, unsigned int required_vco)
        val &= ~(DPLL_CTRL1_HDMI_MODE(SKL_DPLL0) | DPLL_CTRL1_SSC(SKL_DPLL0) |
                 DPLL_CTRL1_LINK_RATE_MASK(SKL_DPLL0));
        val |= DPLL_CTRL1_OVERRIDE(SKL_DPLL0);
-       if (required_vco == 8640)
+       if (vco == 8640)
                val |= DPLL_CTRL1_LINK_RATE(DPLL_CTRL1_LINK_RATE_1080,
                                            SKL_DPLL0);
        else
@@ -5615,6 +5600,14 @@ skl_dpll0_enable(struct drm_i915_private *dev_priv, unsigned int required_vco)
                DRM_ERROR("DPLL0 not locked\n");
 }
 
+static void
+skl_dpll0_disable(struct drm_i915_private *dev_priv)
+{
+       I915_WRITE(LCPLL1_CTL, I915_READ(LCPLL1_CTL) & ~LCPLL_PLL_ENABLE);
+       if (wait_for(!(I915_READ(LCPLL1_CTL) & LCPLL_PLL_LOCK), 1))
+               DRM_ERROR("Couldn't disable DPLL0\n");
+}
+
 static bool skl_cdclk_pcu_ready(struct drm_i915_private *dev_priv)
 {
        int ret;
@@ -5642,12 +5635,12 @@ static bool skl_cdclk_wait_for_pcu_ready(struct drm_i915_private *dev_priv)
        return false;
 }
 
-static void skl_set_cdclk(struct drm_i915_private *dev_priv, unsigned int freq)
+static void skl_set_cdclk(struct drm_i915_private *dev_priv, int cdclk)
 {
        struct drm_device *dev = dev_priv->dev;
        u32 freq_select, pcu_ack;
 
-       DRM_DEBUG_DRIVER("Changing CDCLK to %dKHz\n", freq);
+       DRM_DEBUG_DRIVER("Changing CDCLK to %dKHz\n", cdclk);
 
        if (!skl_cdclk_wait_for_pcu_ready(dev_priv)) {
                DRM_ERROR("failed to inform PCU about cdclk change\n");
@@ -5655,7 +5648,7 @@ static void skl_set_cdclk(struct drm_i915_private *dev_priv, unsigned int freq)
        }
 
        /* set CDCLK_CTL */
-       switch(freq) {
+       switch (cdclk) {
        case 450000:
        case 432000:
                freq_select = CDCLK_FREQ_450_432;
@@ -5678,7 +5671,7 @@ static void skl_set_cdclk(struct drm_i915_private *dev_priv, unsigned int freq)
                break;
        }
 
-       I915_WRITE(CDCLK_CTL, freq_select | skl_cdclk_decimal(freq));
+       I915_WRITE(CDCLK_CTL, freq_select | skl_cdclk_decimal(cdclk));
        POSTING_READ(CDCLK_CTL);
 
        /* inform PCU of the change */
@@ -5700,21 +5693,18 @@ void skl_uninit_cdclk(struct drm_i915_private *dev_priv)
        if (I915_READ(DBUF_CTL) & DBUF_POWER_STATE)
                DRM_ERROR("DBuf power disable timeout\n");
 
-       /* disable DPLL0 */
-       I915_WRITE(LCPLL1_CTL, I915_READ(LCPLL1_CTL) & ~LCPLL_PLL_ENABLE);
-       if (wait_for(!(I915_READ(LCPLL1_CTL) & LCPLL_PLL_LOCK), 1))
-               DRM_ERROR("Couldn't disable DPLL0\n");
+       skl_dpll0_disable(dev_priv);
 }
 
 void skl_init_cdclk(struct drm_i915_private *dev_priv)
 {
-       unsigned int required_vco;
+       unsigned int vco;
 
        /* DPLL0 not enabled (happens on early BIOS versions) */
        if (!(I915_READ(LCPLL1_CTL) & LCPLL_PLL_ENABLE)) {
                /* enable DPLL0 */
-               required_vco = skl_cdclk_get_vco(dev_priv->skl_boot_cdclk);
-               skl_dpll0_enable(dev_priv, required_vco);
+               vco = skl_cdclk_get_vco(dev_priv->skl_boot_cdclk);
+               skl_dpll0_enable(dev_priv, vco);
        }
 
        /* set CDCLK to the frequency the BIOS chose */
@@ -5906,21 +5896,19 @@ static int valleyview_calc_cdclk(struct drm_i915_private *dev_priv,
                return 200000;
 }
 
-static int broxton_calc_cdclk(struct drm_i915_private *dev_priv,
-                             int max_pixclk)
+static int broxton_calc_cdclk(int max_pixclk)
 {
        /*
         * FIXME:
-        * - remove the guardband, it's not needed on BXT
         * - set 19.2MHz bypass frequency if there are no active pipes
         */
-       if (max_pixclk > 576000*9/10)
+       if (max_pixclk > 576000)
                return 624000;
-       else if (max_pixclk > 384000*9/10)
+       else if (max_pixclk > 384000)
                return 576000;
-       else if (max_pixclk > 288000*9/10)
+       else if (max_pixclk > 288000)
                return 384000;
-       else if (max_pixclk > 144000*9/10)
+       else if (max_pixclk > 144000)
                return 288000;
        else
                return 144000;
@@ -5963,9 +5951,6 @@ static int valleyview_modeset_calc_cdclk(struct drm_atomic_state *state)
        struct intel_atomic_state *intel_state =
                to_intel_atomic_state(state);
 
-       if (max_pixclk < 0)
-               return max_pixclk;
-
        intel_state->cdclk = intel_state->dev_cdclk =
                valleyview_calc_cdclk(dev_priv, max_pixclk);
 
@@ -5977,20 +5962,15 @@ static int valleyview_modeset_calc_cdclk(struct drm_atomic_state *state)
 
 static int broxton_modeset_calc_cdclk(struct drm_atomic_state *state)
 {
-       struct drm_device *dev = state->dev;
-       struct drm_i915_private *dev_priv = dev->dev_private;
-       int max_pixclk = intel_mode_max_pixclk(dev, state);
+       int max_pixclk = ilk_max_pixel_rate(state);
        struct intel_atomic_state *intel_state =
                to_intel_atomic_state(state);
 
-       if (max_pixclk < 0)
-               return max_pixclk;
-
        intel_state->cdclk = intel_state->dev_cdclk =
-               broxton_calc_cdclk(dev_priv, max_pixclk);
+               broxton_calc_cdclk(max_pixclk);
 
        if (!intel_state->active_crtcs)
-               intel_state->dev_cdclk = broxton_calc_cdclk(dev_priv, 0);
+               intel_state->dev_cdclk = broxton_calc_cdclk(0);
 
        return 0;
 }
@@ -6252,7 +6232,7 @@ static void intel_crtc_disable_noatomic(struct drm_crtc *crtc)
                return;
 
        if (to_intel_plane_state(crtc->primary->state)->visible) {
-               WARN_ON(intel_crtc->unpin_work);
+               WARN_ON(intel_crtc->flip_work);
 
                intel_pre_disable_primary_noatomic(crtc);
 
@@ -7063,7 +7043,7 @@ static uint32_t i9xx_dpll_compute_fp(struct dpll *dpll)
 
 static void i9xx_update_pll_dividers(struct intel_crtc *crtc,
                                     struct intel_crtc_state *crtc_state,
-                                    intel_clock_t *reduced_clock)
+                                    struct dpll *reduced_clock)
 {
        struct drm_device *dev = crtc->base.dev;
        u32 fp, fp2 = 0;
@@ -7487,7 +7467,7 @@ void vlv_force_pll_off(struct drm_device *dev, enum pipe pipe)
 
 static void i9xx_compute_dpll(struct intel_crtc *crtc,
                              struct intel_crtc_state *crtc_state,
-                             intel_clock_t *reduced_clock)
+                             struct dpll *reduced_clock)
 {
        struct drm_device *dev = crtc->base.dev;
        struct drm_i915_private *dev_priv = dev->dev_private;
@@ -7563,7 +7543,7 @@ static void i9xx_compute_dpll(struct intel_crtc *crtc,
 
 static void i8xx_compute_dpll(struct intel_crtc *crtc,
                              struct intel_crtc_state *crtc_state,
-                             intel_clock_t *reduced_clock)
+                             struct dpll *reduced_clock)
 {
        struct drm_device *dev = crtc->base.dev;
        struct drm_i915_private *dev_priv = dev->dev_private;
@@ -7817,7 +7797,7 @@ static int i8xx_crtc_compute_clock(struct intel_crtc *crtc,
 {
        struct drm_device *dev = crtc->base.dev;
        struct drm_i915_private *dev_priv = dev->dev_private;
-       const intel_limit_t *limit;
+       const struct intel_limit *limit;
        int refclk = 48000;
 
        memset(&crtc_state->dpll_hw_state, 0,
@@ -7853,7 +7833,7 @@ static int g4x_crtc_compute_clock(struct intel_crtc *crtc,
 {
        struct drm_device *dev = crtc->base.dev;
        struct drm_i915_private *dev_priv = dev->dev_private;
-       const intel_limit_t *limit;
+       const struct intel_limit *limit;
        int refclk = 96000;
 
        memset(&crtc_state->dpll_hw_state, 0,
@@ -7896,7 +7876,7 @@ static int pnv_crtc_compute_clock(struct intel_crtc *crtc,
 {
        struct drm_device *dev = crtc->base.dev;
        struct drm_i915_private *dev_priv = dev->dev_private;
-       const intel_limit_t *limit;
+       const struct intel_limit *limit;
        int refclk = 96000;
 
        memset(&crtc_state->dpll_hw_state, 0,
@@ -7930,7 +7910,7 @@ static int i9xx_crtc_compute_clock(struct intel_crtc *crtc,
 {
        struct drm_device *dev = crtc->base.dev;
        struct drm_i915_private *dev_priv = dev->dev_private;
-       const intel_limit_t *limit;
+       const struct intel_limit *limit;
        int refclk = 96000;
 
        memset(&crtc_state->dpll_hw_state, 0,
@@ -7963,7 +7943,7 @@ static int chv_crtc_compute_clock(struct intel_crtc *crtc,
                                  struct intel_crtc_state *crtc_state)
 {
        int refclk = 100000;
-       const intel_limit_t *limit = &intel_limits_chv;
+       const struct intel_limit *limit = &intel_limits_chv;
 
        memset(&crtc_state->dpll_hw_state, 0,
               sizeof(crtc_state->dpll_hw_state));
@@ -7984,7 +7964,7 @@ static int vlv_crtc_compute_clock(struct intel_crtc *crtc,
                                  struct intel_crtc_state *crtc_state)
 {
        int refclk = 100000;
-       const intel_limit_t *limit = &intel_limits_vlv;
+       const struct intel_limit *limit = &intel_limits_vlv;
 
        memset(&crtc_state->dpll_hw_state, 0,
               sizeof(crtc_state->dpll_hw_state));
@@ -8034,7 +8014,7 @@ static void vlv_crtc_clock_get(struct intel_crtc *crtc,
        struct drm_device *dev = crtc->base.dev;
        struct drm_i915_private *dev_priv = dev->dev_private;
        int pipe = pipe_config->cpu_transcoder;
-       intel_clock_t clock;
+       struct dpll clock;
        u32 mdiv;
        int refclk = 100000;
 
@@ -8131,7 +8111,7 @@ static void chv_crtc_clock_get(struct intel_crtc *crtc,
        struct drm_i915_private *dev_priv = dev->dev_private;
        int pipe = pipe_config->cpu_transcoder;
        enum dpio_channel port = vlv_pipe_to_channel(pipe);
-       intel_clock_t clock;
+       struct dpll clock;
        u32 cmn_dw13, pll_dw0, pll_dw1, pll_dw2, pll_dw3;
        int refclk = 100000;
 
@@ -8794,7 +8774,7 @@ static bool ironlake_needs_fb_cb_tune(struct dpll *dpll, int factor)
 
 static void ironlake_compute_dpll(struct intel_crtc *intel_crtc,
                                  struct intel_crtc_state *crtc_state,
-                                 intel_clock_t *reduced_clock)
+                                 struct dpll *reduced_clock)
 {
        struct drm_crtc *crtc = &intel_crtc->base;
        struct drm_device *dev = crtc->dev;
@@ -8902,10 +8882,10 @@ static int ironlake_crtc_compute_clock(struct intel_crtc *crtc,
 {
        struct drm_device *dev = crtc->base.dev;
        struct drm_i915_private *dev_priv = dev->dev_private;
-       intel_clock_t reduced_clock;
+       struct dpll reduced_clock;
        bool has_reduced_clock = false;
        struct intel_shared_dpll *pll;
-       const intel_limit_t *limit;
+       const struct intel_limit *limit;
        int refclk = 120000;
 
        memset(&crtc_state->dpll_hw_state, 0,
@@ -9300,6 +9280,10 @@ static bool ironlake_get_pipe_config(struct intel_crtc *crtc,
                ironlake_get_fdi_m_n_config(crtc, pipe_config);
 
                if (HAS_PCH_IBX(dev_priv)) {
+                       /*
+                        * The pipe->pch transcoder and pch transcoder->pll
+                        * mapping is fixed.
+                        */
                        pll_id = (enum intel_dpll_id) crtc->pipe;
                } else {
                        tmp = I915_READ(PCH_DPLL_SEL);
@@ -9687,6 +9671,18 @@ static void broadwell_set_cdclk(struct drm_device *dev, int cdclk)
             cdclk, dev_priv->cdclk_freq);
 }
 
+static int broadwell_calc_cdclk(int max_pixclk)
+{
+       if (max_pixclk > 540000)
+               return 675000;
+       else if (max_pixclk > 450000)
+               return 540000;
+       else if (max_pixclk > 337500)
+               return 450000;
+       else
+               return 337500;
+}
+
 static int broadwell_modeset_calc_cdclk(struct drm_atomic_state *state)
 {
        struct drm_i915_private *dev_priv = to_i915(state->dev);
@@ -9698,14 +9694,7 @@ static int broadwell_modeset_calc_cdclk(struct drm_atomic_state *state)
         * FIXME should also account for plane ratio
         * once 64bpp pixel formats are supported.
         */
-       if (max_pixclk > 540000)
-               cdclk = 675000;
-       else if (max_pixclk > 450000)
-               cdclk = 540000;
-       else if (max_pixclk > 337500)
-               cdclk = 450000;
-       else
-               cdclk = 337500;
+       cdclk = broadwell_calc_cdclk(max_pixclk);
 
        if (cdclk > dev_priv->max_cdclk_freq) {
                DRM_DEBUG_KMS("requested cdclk (%d kHz) exceeds max (%d kHz)\n",
@@ -9715,7 +9704,7 @@ static int broadwell_modeset_calc_cdclk(struct drm_atomic_state *state)
 
        intel_state->cdclk = intel_state->dev_cdclk = cdclk;
        if (!intel_state->active_crtcs)
-               intel_state->dev_cdclk = 337500;
+               intel_state->dev_cdclk = broadwell_calc_cdclk(0);
 
        return 0;
 }
@@ -9850,6 +9839,10 @@ static bool hsw_get_transcoder_state(struct intel_crtc *crtc,
        enum intel_display_power_domain power_domain;
        u32 tmp;
 
+       /*
+        * The pipe->transcoder mapping is fixed with the exception of the eDP
+        * transcoder handled below.
+        */
        pipe_config->cpu_transcoder = (enum transcoder) crtc->pipe;
 
        /*
@@ -10317,10 +10310,10 @@ intel_framebuffer_create_for_mode(struct drm_device *dev,
        struct drm_i915_gem_object *obj;
        struct drm_mode_fb_cmd2 mode_cmd = { 0 };
 
-       obj = i915_gem_alloc_object(dev,
+       obj = i915_gem_object_create(dev,
                                    intel_framebuffer_size_for_mode(mode, bpp));
-       if (obj == NULL)
-               return ERR_PTR(-ENOMEM);
+       if (IS_ERR(obj))
+               return ERR_CAST(obj);
 
        mode_cmd.width = mode->hdisplay;
        mode_cmd.height = mode->vdisplay;
@@ -10632,7 +10625,7 @@ static void i9xx_crtc_clock_get(struct intel_crtc *crtc,
        int pipe = pipe_config->cpu_transcoder;
        u32 dpll = pipe_config->dpll_hw_state.dpll;
        u32 fp;
-       intel_clock_t clock;
+       struct dpll clock;
        int port_clock;
        int refclk = i9xx_pll_refclk(dev, pipe_config);
 
@@ -10806,31 +10799,27 @@ struct drm_display_mode *intel_crtc_mode_get(struct drm_device *dev,
        return mode;
 }
 
-void intel_mark_busy(struct drm_device *dev)
+void intel_mark_busy(struct drm_i915_private *dev_priv)
 {
-       struct drm_i915_private *dev_priv = dev->dev_private;
-
        if (dev_priv->mm.busy)
                return;
 
        intel_runtime_pm_get(dev_priv);
        i915_update_gfx_val(dev_priv);
-       if (INTEL_INFO(dev)->gen >= 6)
+       if (INTEL_GEN(dev_priv) >= 6)
                gen6_rps_busy(dev_priv);
        dev_priv->mm.busy = true;
 }
 
-void intel_mark_idle(struct drm_device *dev)
+void intel_mark_idle(struct drm_i915_private *dev_priv)
 {
-       struct drm_i915_private *dev_priv = dev->dev_private;
-
        if (!dev_priv->mm.busy)
                return;
 
        dev_priv->mm.busy = false;
 
-       if (INTEL_INFO(dev)->gen >= 6)
-               gen6_rps_idle(dev->dev_private);
+       if (INTEL_GEN(dev_priv) >= 6)
+               gen6_rps_idle(dev_priv);
 
        intel_runtime_pm_put(dev_priv);
 }
@@ -10839,15 +10828,16 @@ 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_unpin_work *work;
+       struct intel_flip_work *work;
 
        spin_lock_irq(&dev->event_lock);
-       work = intel_crtc->unpin_work;
-       intel_crtc->unpin_work = NULL;
+       work = intel_crtc->flip_work;
+       intel_crtc->flip_work = NULL;
        spin_unlock_irq(&dev->event_lock);
 
        if (work) {
-               cancel_work_sync(&work->work);
+               cancel_work_sync(&work->mmio_work);
+               cancel_work_sync(&work->unpin_work);
                kfree(work);
        }
 
@@ -10858,12 +10848,15 @@ static void intel_crtc_destroy(struct drm_crtc *crtc)
 
 static void intel_unpin_work_fn(struct work_struct *__work)
 {
-       struct intel_unpin_work *work =
-               container_of(__work, struct intel_unpin_work, work);
+       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;
 
+       if (is_mmio_work(work))
+               flush_work(&work->mmio_work);
+
        mutex_lock(&dev->struct_mutex);
        intel_unpin_fb_obj(work->old_fb, primary->state->rotation);
        drm_gem_object_unreference(&work->pending_flip_obj->base);
@@ -10882,60 +10875,14 @@ static void intel_unpin_work_fn(struct work_struct *__work)
        kfree(work);
 }
 
-static void do_intel_finish_page_flip(struct drm_device *dev,
-                                     struct drm_crtc *crtc)
-{
-       struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
-       struct intel_unpin_work *work;
-       unsigned long flags;
-
-       /* Ignore early vblank irqs */
-       if (intel_crtc == NULL)
-               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 = intel_crtc->unpin_work;
-
-       /* Ensure we don't miss a work->pending update ... */
-       smp_rmb();
-
-       if (work == NULL || atomic_read(&work->pending) < INTEL_FLIP_COMPLETE) {
-               spin_unlock_irqrestore(&dev->event_lock, flags);
-               return;
-       }
-
-       page_flip_completed(intel_crtc);
-
-       spin_unlock_irqrestore(&dev->event_lock, flags);
-}
-
-void intel_finish_page_flip(struct drm_device *dev, int pipe)
-{
-       struct drm_i915_private *dev_priv = dev->dev_private;
-       struct drm_crtc *crtc = dev_priv->pipe_to_crtc_mapping[pipe];
-
-       do_intel_finish_page_flip(dev, crtc);
-}
-
-void intel_finish_page_flip_plane(struct drm_device *dev, int plane)
-{
-       struct drm_i915_private *dev_priv = dev->dev_private;
-       struct drm_crtc *crtc = dev_priv->plane_to_crtc_mapping[plane];
-
-       do_intel_finish_page_flip(dev, crtc);
-}
-
 /* Is 'a' after or equal to 'b'? */
 static bool g4x_flip_count_after_eq(u32 a, u32 b)
 {
        return !((a - b) & 0x80000000);
 }
 
-static bool page_flip_finished(struct intel_crtc *crtc)
+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 = dev->dev_private;
@@ -10977,40 +10924,103 @@ static bool page_flip_finished(struct intel_crtc *crtc)
         * anyway, we don't really care.
         */
        return (I915_READ(DSPSURFLIVE(crtc->plane)) & ~0xfff) ==
-               crtc->unpin_work->gtt_offset &&
+               crtc->flip_work->gtt_offset &&
                g4x_flip_count_after_eq(I915_READ(PIPE_FLIPCOUNT_G4X(crtc->pipe)),
-                                   crtc->unpin_work->flip_count);
+                                   crtc->flip_work->flip_count);
 }
 
-void intel_prepare_page_flip(struct drm_device *dev, int plane)
+static bool
+__pageflip_finished_mmio(struct intel_crtc *crtc,
+                              struct intel_flip_work *work)
 {
-       struct drm_i915_private *dev_priv = dev->dev_private;
-       struct intel_crtc *intel_crtc =
-               to_intel_crtc(dev_priv->plane_to_crtc_mapping[plane]);
+       /*
+        * 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->dev;
+       struct drm_crtc *crtc = dev_priv->pipe_to_crtc_mapping[pipe];
+       struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
+       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 = intel_crtc->flip_work;
+
+       if (work != NULL &&
+           !is_mmio_work(work) &&
+           pageflip_finished(intel_crtc, work))
+               page_flip_completed(intel_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->dev;
+       struct drm_crtc *crtc = dev_priv->pipe_to_crtc_mapping[pipe];
+       struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
+       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.
-        *
-        * NB: An MMIO update of the plane base pointer will also
-        * generate a page-flip completion irq, i.e. every modeset
-        * is also accompanied by a spurious intel_prepare_page_flip().
         */
        spin_lock_irqsave(&dev->event_lock, flags);
-       if (intel_crtc->unpin_work && page_flip_finished(intel_crtc))
-               atomic_inc_not_zero(&intel_crtc->unpin_work->pending);
+       work = intel_crtc->flip_work;
+
+       if (work != NULL &&
+           is_mmio_work(work) &&
+           pageflip_finished(intel_crtc, work))
+               page_flip_completed(intel_crtc);
+
        spin_unlock_irqrestore(&dev->event_lock, flags);
 }
 
-static inline void intel_mark_page_flip_active(struct intel_unpin_work *work)
+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_wmb();
-       atomic_set(&work->pending, INTEL_FLIP_PENDING);
-       /* and that it is marked active as soon as the irq could fire. */
-       smp_wmb();
+       smp_mb__before_atomic();
+       atomic_set(&work->pending, 1);
 }
 
 static int intel_gen2_queue_flip(struct drm_device *dev,
@@ -11041,10 +11051,9 @@ static int intel_gen2_queue_flip(struct drm_device *dev,
        intel_ring_emit(engine, MI_DISPLAY_FLIP |
                        MI_DISPLAY_FLIP_PLANE(intel_crtc->plane));
        intel_ring_emit(engine, fb->pitches[0]);
-       intel_ring_emit(engine, intel_crtc->unpin_work->gtt_offset);
+       intel_ring_emit(engine, intel_crtc->flip_work->gtt_offset);
        intel_ring_emit(engine, 0); /* aux display base address, unused */
 
-       intel_mark_page_flip_active(intel_crtc->unpin_work);
        return 0;
 }
 
@@ -11073,10 +11082,9 @@ static int intel_gen3_queue_flip(struct drm_device *dev,
        intel_ring_emit(engine, MI_DISPLAY_FLIP_I915 |
                        MI_DISPLAY_FLIP_PLANE(intel_crtc->plane));
        intel_ring_emit(engine, fb->pitches[0]);
-       intel_ring_emit(engine, intel_crtc->unpin_work->gtt_offset);
+       intel_ring_emit(engine, intel_crtc->flip_work->gtt_offset);
        intel_ring_emit(engine, MI_NOOP);
 
-       intel_mark_page_flip_active(intel_crtc->unpin_work);
        return 0;
 }
 
@@ -11104,7 +11112,7 @@ static int intel_gen4_queue_flip(struct drm_device *dev,
        intel_ring_emit(engine, MI_DISPLAY_FLIP |
                        MI_DISPLAY_FLIP_PLANE(intel_crtc->plane));
        intel_ring_emit(engine, fb->pitches[0]);
-       intel_ring_emit(engine, intel_crtc->unpin_work->gtt_offset |
+       intel_ring_emit(engine, intel_crtc->flip_work->gtt_offset |
                        obj->tiling_mode);
 
        /* XXX Enabling the panel-fitter across page-flip is so far
@@ -11115,7 +11123,6 @@ static int intel_gen4_queue_flip(struct drm_device *dev,
        pipesrc = I915_READ(PIPESRC(intel_crtc->pipe)) & 0x0fff0fff;
        intel_ring_emit(engine, pf | pipesrc);
 
-       intel_mark_page_flip_active(intel_crtc->unpin_work);
        return 0;
 }
 
@@ -11139,7 +11146,7 @@ static int intel_gen6_queue_flip(struct drm_device *dev,
        intel_ring_emit(engine, MI_DISPLAY_FLIP |
                        MI_DISPLAY_FLIP_PLANE(intel_crtc->plane));
        intel_ring_emit(engine, fb->pitches[0] | obj->tiling_mode);
-       intel_ring_emit(engine, intel_crtc->unpin_work->gtt_offset);
+       intel_ring_emit(engine, intel_crtc->flip_work->gtt_offset);
 
        /* Contrary to the suggestions in the documentation,
         * "Enable Panel Fitter" does not seem to be required when page
@@ -11151,7 +11158,6 @@ static int intel_gen6_queue_flip(struct drm_device *dev,
        pipesrc = I915_READ(PIPESRC(intel_crtc->pipe)) & 0x0fff0fff;
        intel_ring_emit(engine, pf | pipesrc);
 
-       intel_mark_page_flip_active(intel_crtc->unpin_work);
        return 0;
 }
 
@@ -11243,10 +11249,9 @@ static int intel_gen7_queue_flip(struct drm_device *dev,
 
        intel_ring_emit(engine, MI_DISPLAY_FLIP_I915 | plane_bit);
        intel_ring_emit(engine, (fb->pitches[0] | obj->tiling_mode));
-       intel_ring_emit(engine, intel_crtc->unpin_work->gtt_offset);
+       intel_ring_emit(engine, intel_crtc->flip_work->gtt_offset);
        intel_ring_emit(engine, (MI_NOOP));
 
-       intel_mark_page_flip_active(intel_crtc->unpin_work);
        return 0;
 }
 
@@ -11264,7 +11269,7 @@ static bool use_mmio_flip(struct intel_engine_cs *engine,
        if (engine == NULL)
                return true;
 
-       if (INTEL_INFO(engine->dev)->gen < 5)
+       if (INTEL_GEN(engine->i915) < 5)
                return false;
 
        if (i915.use_mmio_flip < 0)
@@ -11283,7 +11288,7 @@ static bool use_mmio_flip(struct intel_engine_cs *engine,
 
 static void skl_do_mmio_flip(struct intel_crtc *intel_crtc,
                             unsigned int rotation,
-                            struct intel_unpin_work *work)
+                            struct intel_flip_work *work)
 {
        struct drm_device *dev = intel_crtc->base.dev;
        struct drm_i915_private *dev_priv = dev->dev_private;
@@ -11335,7 +11340,7 @@ static void skl_do_mmio_flip(struct intel_crtc *intel_crtc,
 }
 
 static void ilk_do_mmio_flip(struct intel_crtc *intel_crtc,
-                            struct intel_unpin_work *work)
+                            struct intel_flip_work *work)
 {
        struct drm_device *dev = intel_crtc->base.dev;
        struct drm_i915_private *dev_priv = dev->dev_private;
@@ -11358,48 +11363,20 @@ static void ilk_do_mmio_flip(struct intel_crtc *intel_crtc,
        POSTING_READ(DSPSURF(intel_crtc->plane));
 }
 
-/*
- * XXX: This is the temporary way to update the plane registers until we get
- * around to using the usual plane update functions for MMIO flips
- */
-static void intel_do_mmio_flip(struct intel_mmio_flip *mmio_flip)
+static void intel_mmio_flip_work_func(struct work_struct *w)
 {
-       struct intel_crtc *crtc = mmio_flip->crtc;
-       struct intel_unpin_work *work;
-
-       spin_lock_irq(&crtc->base.dev->event_lock);
-       work = crtc->unpin_work;
-       spin_unlock_irq(&crtc->base.dev->event_lock);
-       if (work == NULL)
-               return;
-
-       intel_mark_page_flip_active(work);
-
-       intel_pipe_update_start(crtc);
-
-       if (INTEL_INFO(mmio_flip->i915)->gen >= 9)
-               skl_do_mmio_flip(crtc, mmio_flip->rotation, work);
-       else
-               /* use_mmio_flip() retricts MMIO flips to ilk+ */
-               ilk_do_mmio_flip(crtc, work);
-
-       intel_pipe_update_end(crtc);
-}
-
-static void intel_mmio_flip_work_func(struct work_struct *work)
-{
-       struct intel_mmio_flip *mmio_flip =
-               container_of(work, struct intel_mmio_flip, work);
+       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(mmio_flip->crtc->base.primary->fb);
+               to_intel_framebuffer(crtc->base.primary->fb);
        struct drm_i915_gem_object *obj = intel_fb->obj;
 
-       if (mmio_flip->req) {
-               WARN_ON(__i915_wait_request(mmio_flip->req,
+       if (work->flip_queued_req)
+               WARN_ON(__i915_wait_request(work->flip_queued_req,
                                            false, NULL,
-                                           &mmio_flip->i915->rps.mmioflips));
-               i915_gem_request_unreference__unlocked(mmio_flip->req);
-       }
+                                           &dev_priv->rps.mmioflips));
 
        /* For framebuffer backed by dmabuf, wait for fence */
        if (obj->base.dma_buf)
@@ -11407,29 +11384,15 @@ static void intel_mmio_flip_work_func(struct work_struct *work)
                                                            false, false,
                                                            MAX_SCHEDULE_TIMEOUT) < 0);
 
-       intel_do_mmio_flip(mmio_flip);
-       kfree(mmio_flip);
-}
-
-static int intel_queue_mmio_flip(struct drm_device *dev,
-                                struct drm_crtc *crtc,
-                                struct drm_i915_gem_object *obj)
-{
-       struct intel_mmio_flip *mmio_flip;
-
-       mmio_flip = kmalloc(sizeof(*mmio_flip), GFP_KERNEL);
-       if (mmio_flip == NULL)
-               return -ENOMEM;
-
-       mmio_flip->i915 = to_i915(dev);
-       mmio_flip->req = i915_gem_request_reference(obj->last_write_req);
-       mmio_flip->crtc = to_intel_crtc(crtc);
-       mmio_flip->rotation = crtc->primary->state->rotation;
+       intel_pipe_update_start(crtc);
 
-       INIT_WORK(&mmio_flip->work, intel_mmio_flip_work_func);
-       schedule_work(&mmio_flip->work);
+       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);
 
-       return 0;
+       intel_pipe_update_end(crtc, work);
 }
 
 static int intel_default_queue_flip(struct drm_device *dev,
@@ -11442,37 +11405,32 @@ static int intel_default_queue_flip(struct drm_device *dev,
        return -ENODEV;
 }
 
-static bool __intel_pageflip_stall_check(struct drm_device *dev,
-                                        struct drm_crtc *crtc)
+static bool __pageflip_stall_check_cs(struct drm_i915_private *dev_priv,
+                                     struct intel_crtc *intel_crtc,
+                                     struct intel_flip_work *work)
 {
-       struct drm_i915_private *dev_priv = dev->dev_private;
-       struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
-       struct intel_unpin_work *work = intel_crtc->unpin_work;
-       u32 addr;
+       u32 addr, vblank;
 
-       if (atomic_read(&work->pending) >= INTEL_FLIP_COMPLETE)
-               return true;
-
-       if (atomic_read(&work->pending) < INTEL_FLIP_PENDING)
+       if (!atomic_read(&work->pending))
                return false;
 
-       if (!work->enable_stall_check)
-               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, true))
                        return false;
 
-               work->flip_ready_vblank = drm_crtc_vblank_count(crtc);
+               work->flip_ready_vblank = vblank;
        }
 
-       if (drm_crtc_vblank_count(crtc) - work->flip_ready_vblank < 3)
+       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_INFO(dev)->gen >= 4)
+       if (INTEL_GEN(dev_priv) >= 4)
                addr = I915_HI_DISPBASE(I915_READ(DSPSURF(intel_crtc->plane)));
        else
                addr = I915_READ(DSPADDR(intel_crtc->plane));
@@ -11484,12 +11442,12 @@ static bool __intel_pageflip_stall_check(struct drm_device *dev,
        return addr == work->gtt_offset;
 }
 
-void intel_check_page_flip(struct drm_device *dev, int pipe)
+void intel_check_page_flip(struct drm_i915_private *dev_priv, int pipe)
 {
-       struct drm_i915_private *dev_priv = dev->dev_private;
+       struct drm_device *dev = dev_priv->dev;
        struct drm_crtc *crtc = dev_priv->pipe_to_crtc_mapping[pipe];
        struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
-       struct intel_unpin_work *work;
+       struct intel_flip_work *work;
 
        WARN_ON(!in_interrupt());
 
@@ -11497,16 +11455,20 @@ void intel_check_page_flip(struct drm_device *dev, int pipe)
                return;
 
        spin_lock(&dev->event_lock);
-       work = intel_crtc->unpin_work;
-       if (work != NULL && __intel_pageflip_stall_check(dev, crtc)) {
-               WARN_ONCE(1, "Kicking stuck page flip: queued at %d, now %d\n",
-                        work->flip_queued_vblank, drm_vblank_count(dev, pipe));
+       work = intel_crtc->flip_work;
+
+       if (work != NULL && !is_mmio_work(work) &&
+           __pageflip_stall_check_cs(dev_priv, intel_crtc, work)) {
+               WARN_ONCE(1,
+                         "Kicking stuck page flip: queued at %d, now %d\n",
+                       work->flip_queued_vblank, intel_crtc_get_vblank_counter(intel_crtc));
                page_flip_completed(intel_crtc);
                work = NULL;
        }
-       if (work != NULL &&
-           drm_vblank_count(dev, pipe) - work->flip_queued_vblank > 1)
-               intel_queue_rps_boost_for_request(dev, work->flip_queued_req);
+
+       if (work != NULL && !is_mmio_work(work) &&
+           intel_crtc_get_vblank_counter(intel_crtc) - work->flip_queued_vblank > 1)
+               intel_queue_rps_boost_for_request(work->flip_queued_req);
        spin_unlock(&dev->event_lock);
 }
 
@@ -11522,7 +11484,7 @@ static int intel_crtc_page_flip(struct drm_crtc *crtc,
        struct intel_crtc *intel_crtc = to_intel_crtc(crtc);
        struct drm_plane *primary = crtc->primary;
        enum pipe pipe = intel_crtc->pipe;
-       struct intel_unpin_work *work;
+       struct intel_flip_work *work;
        struct intel_engine_cs *engine;
        bool mmio_flip;
        struct drm_i915_gem_request *request = NULL;
@@ -11559,19 +11521,19 @@ static int intel_crtc_page_flip(struct drm_crtc *crtc,
        work->event = event;
        work->crtc = crtc;
        work->old_fb = old_fb;
-       INIT_WORK(&work->work, intel_unpin_work_fn);
+       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 unpin_work */
+       /* We borrow the event spin lock for protecting flip_work */
        spin_lock_irq(&dev->event_lock);
-       if (intel_crtc->unpin_work) {
+       if (intel_crtc->flip_work) {
                /* Before declaring the flip queue wedged, check if
                 * the hardware completed the operation behind our backs.
                 */
-               if (__intel_pageflip_stall_check(dev, crtc)) {
+               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 {
@@ -11583,7 +11545,7 @@ static int intel_crtc_page_flip(struct drm_crtc *crtc,
                        return -EBUSY;
                }
        }
-       intel_crtc->unpin_work = work;
+       intel_crtc->flip_work = work;
        spin_unlock_irq(&dev->event_lock);
 
        if (atomic_read(&intel_crtc->unpin_work_count) >= 2)
@@ -11638,6 +11600,11 @@ static int intel_crtc_page_flip(struct drm_crtc *crtc,
         */
        if (!mmio_flip) {
                ret = i915_gem_object_sync(obj, engine, &request);
+               if (!ret && !request) {
+                       request = i915_gem_request_alloc(engine, NULL);
+                       ret = PTR_ERR_OR_ZERO(request);
+               }
+
                if (ret)
                        goto cleanup_pending;
        }
@@ -11649,38 +11616,28 @@ static int intel_crtc_page_flip(struct drm_crtc *crtc,
        work->gtt_offset = intel_plane_obj_offset(to_intel_plane(primary),
                                                  obj, 0);
        work->gtt_offset += intel_crtc->dspaddr_offset;
+       work->rotation = crtc->primary->state->rotation;
 
        if (mmio_flip) {
-               ret = intel_queue_mmio_flip(dev, crtc, obj);
-               if (ret)
-                       goto cleanup_unpin;
+               INIT_WORK(&work->mmio_work, intel_mmio_flip_work_func);
 
                i915_gem_request_assign(&work->flip_queued_req,
                                        obj->last_write_req);
-       } else {
-               if (!request) {
-                       request = i915_gem_request_alloc(engine, NULL);
-                       if (IS_ERR(request)) {
-                               ret = PTR_ERR(request);
-                               goto cleanup_unpin;
-                       }
-               }
 
+               schedule_work(&work->mmio_work);
+       } else {
+               i915_gem_request_assign(&work->flip_queued_req, request);
                ret = dev_priv->display.queue_flip(dev, crtc, fb, obj, request,
                                                   page_flip_flags);
                if (ret)
                        goto cleanup_unpin;
 
-               i915_gem_request_assign(&work->flip_queued_req, request);
-       }
+               intel_mark_page_flip_active(intel_crtc, work);
 
-       if (request)
                i915_add_request_no_flush(request);
+       }
 
-       work->flip_queued_vblank = drm_crtc_vblank_count(crtc);
-       work->enable_stall_check = true;
-
-       i915_gem_track_fb(intel_fb_obj(work->old_fb), obj,
+       i915_gem_track_fb(intel_fb_obj(old_fb), obj,
                          to_intel_plane(primary)->frontbuffer_bit);
        mutex_unlock(&dev->struct_mutex);
 
@@ -11706,7 +11663,7 @@ cleanup:
        drm_framebuffer_unreference(work->old_fb);
 
        spin_lock_irq(&dev->event_lock);
-       intel_crtc->unpin_work = NULL;
+       intel_crtc->flip_work = NULL;
        spin_unlock_irq(&dev->event_lock);
 
        drm_crtc_vblank_put(crtc);
@@ -11834,6 +11791,11 @@ int intel_plane_atomic_calc_changes(struct drm_crtc_state *crtc_state,
         * Visibility is calculated as if the crtc was on, but
         * after scaler setup everything depends on it being off
         * when the crtc isn't active.
+        *
+        * FIXME this is wrong for watermarks. Watermarks should also
+        * be computed as if the pipe would be active. Perhaps move
+        * per-plane wm computation to the .check_plane() hook, and
+        * only combine the results from all planes in the current place?
         */
        if (!is_crtc_enabled)
                to_intel_plane_state(plane_state)->visible = visible = false;
@@ -12007,7 +11969,7 @@ static int intel_crtc_atomic_check(struct drm_crtc *crtc,
                }
        } else if (dev_priv->display.compute_intermediate_wm) {
                if (HAS_PCH_SPLIT(dev_priv) && INTEL_GEN(dev_priv) < 9)
-                       pipe_config->wm.intermediate = pipe_config->wm.optimal.ilk;
+                       pipe_config->wm.ilk.intermediate = pipe_config->wm.ilk.optimal;
        }
 
        if (INTEL_INFO(dev)->gen >= 9) {
@@ -13280,6 +13242,9 @@ static int intel_modeset_checks(struct drm_atomic_state *state)
                        intel_state->active_crtcs |= 1 << i;
                else
                        intel_state->active_crtcs &= ~(1 << i);
+
+               if (crtc_state->active != crtc->state->active)
+                       intel_state->active_pipe_changes |= drm_crtc_mask(crtc);
        }
 
        /*
@@ -13316,38 +13281,16 @@ static int intel_modeset_checks(struct drm_atomic_state *state)
  * phase.  The code here should be run after the per-crtc and per-plane 'check'
  * handlers to ensure that all derived state has been updated.
  */
-static void calc_watermark_data(struct drm_atomic_state *state)
+static int calc_watermark_data(struct drm_atomic_state *state)
 {
        struct drm_device *dev = state->dev;
-       struct intel_atomic_state *intel_state = to_intel_atomic_state(state);
-       struct drm_crtc *crtc;
-       struct drm_crtc_state *cstate;
-       struct drm_plane *plane;
-       struct drm_plane_state *pstate;
-
-       /*
-        * Calculate watermark configuration details now that derived
-        * plane/crtc state is all properly updated.
-        */
-       drm_for_each_crtc(crtc, dev) {
-               cstate = drm_atomic_get_existing_crtc_state(state, crtc) ?:
-                       crtc->state;
-
-               if (cstate->active)
-                       intel_state->wm_config.num_pipes_active++;
-       }
-       drm_for_each_legacy_plane(plane, dev) {
-               pstate = drm_atomic_get_existing_plane_state(state, plane) ?:
-                       plane->state;
+       struct drm_i915_private *dev_priv = to_i915(dev);
 
-               if (!to_intel_plane_state(pstate)->visible)
-                       continue;
+       /* Is there platform-specific watermark information to calculate? */
+       if (dev_priv->display.compute_global_watermarks)
+               return dev_priv->display.compute_global_watermarks(state);
 
-               intel_state->wm_config.sprites_enabled = true;
-               if (pstate->crtc_w != pstate->src_w >> 16 ||
-                   pstate->crtc_h != pstate->src_h >> 16)
-                       intel_state->wm_config.sprites_scaled = true;
-       }
+       return 0;
 }
 
 /**
@@ -13377,14 +13320,13 @@ static int intel_atomic_check(struct drm_device *dev,
                if (crtc_state->mode.private_flags != crtc->state->mode.private_flags)
                        crtc_state->mode_changed = true;
 
-               if (!crtc_state->enable) {
-                       if (needs_modeset(crtc_state))
-                               any_ms = true;
+               if (!needs_modeset(crtc_state))
                        continue;
-               }
 
-               if (!needs_modeset(crtc_state))
+               if (!crtc_state->enable) {
+                       any_ms = true;
                        continue;
+               }
 
                /* FIXME: For only active_changed we shouldn't need to do any
                 * state recomputation at all. */
@@ -13394,8 +13336,11 @@ static int intel_atomic_check(struct drm_device *dev,
                        return ret;
 
                ret = intel_modeset_pipe_config(crtc, pipe_config);
-               if (ret)
+               if (ret) {
+                       intel_dump_pipe_config(to_intel_crtc(crtc),
+                                              pipe_config, "[failed]");
                        return ret;
+               }
 
                if (i915.fastboot &&
                    intel_pipe_config_compare(dev,
@@ -13405,13 +13350,12 @@ static int intel_atomic_check(struct drm_device *dev,
                        to_intel_crtc_state(crtc_state)->update_pipe = true;
                }
 
-               if (needs_modeset(crtc_state)) {
+               if (needs_modeset(crtc_state))
                        any_ms = true;
 
-                       ret = drm_atomic_add_affected_planes(state, crtc);
-                       if (ret)
-                               return ret;
-               }
+               ret = drm_atomic_add_affected_planes(state, crtc);
+               if (ret)
+                       return ret;
 
                intel_dump_pipe_config(to_intel_crtc(crtc), pipe_config,
                                       needs_modeset(crtc_state) ?
@@ -13431,9 +13375,7 @@ static int intel_atomic_check(struct drm_device *dev,
                return ret;
 
        intel_fbc_choose_crtc(dev_priv, state);
-       calc_watermark_data(state);
-
-       return 0;
+       return calc_watermark_data(state);
 }
 
 static int intel_atomic_prepare_commit(struct drm_device *dev,
@@ -13495,6 +13437,16 @@ static int intel_atomic_prepare_commit(struct drm_device *dev,
        return ret;
 }
 
+u32 intel_crtc_get_vblank_counter(struct intel_crtc *crtc)
+{
+       struct drm_device *dev = crtc->base.dev;
+
+       if (!dev->max_vblank_count)
+               return drm_accurate_vblank_count(&crtc->base);
+
+       return dev->driver->get_vblank_counter(dev, crtc->pipe);
+}
+
 static void intel_atomic_wait_for_vblanks(struct drm_device *dev,
                                          struct drm_i915_private *dev_priv,
                                          unsigned crtc_mask)
@@ -13597,7 +13549,8 @@ static int intel_atomic_commit(struct drm_device *dev,
        }
 
        drm_atomic_helper_swap_state(dev, state);
-       dev_priv->wm.config = intel_state->wm_config;
+       dev_priv->wm.distrust_bios_wm = false;
+       dev_priv->wm.skl_results = intel_state->wm_results;
        intel_shared_dpll_commit(state);
 
        if (intel_state->modeset) {
@@ -14006,7 +13959,7 @@ 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);
+       intel_pipe_update_end(intel_crtc, NULL);
 }
 
 /**
@@ -15050,12 +15003,13 @@ void intel_init_display_hooks(struct drm_i915_private *dev_priv)
                dev_priv->display.fdi_link_train = ivb_manual_fdi_link_train;
        } else if (IS_HASWELL(dev_priv) || IS_BROADWELL(dev_priv)) {
                dev_priv->display.fdi_link_train = hsw_fdi_link_train;
-               if (IS_BROADWELL(dev_priv)) {
-                       dev_priv->display.modeset_commit_cdclk =
-                               broadwell_modeset_commit_cdclk;
-                       dev_priv->display.modeset_calc_cdclk =
-                               broadwell_modeset_calc_cdclk;
-               }
+       }
+
+       if (IS_BROADWELL(dev_priv)) {
+               dev_priv->display.modeset_commit_cdclk =
+                       broadwell_modeset_commit_cdclk;
+               dev_priv->display.modeset_calc_cdclk =
+                       broadwell_modeset_calc_cdclk;
        } else if (IS_VALLEYVIEW(dev_priv) || IS_CHERRYVIEW(dev_priv)) {
                dev_priv->display.modeset_commit_cdclk =
                        valleyview_modeset_commit_cdclk;
@@ -15293,7 +15247,7 @@ void intel_modeset_init_hw(struct drm_device *dev)
        dev_priv->atomic_cdclk_freq = dev_priv->cdclk_freq;
 
        intel_init_clock_gating(dev);
-       intel_enable_gt_powersave(dev);
+       intel_enable_gt_powersave(dev_priv);
 }
 
 /*
@@ -15363,7 +15317,6 @@ retry:
        }
 
        /* Write calculated watermark values back */
-       to_i915(dev)->wm.config = to_intel_atomic_state(state)->wm_config;
        for_each_crtc_in_state(state, crtc, cstate, i) {
                struct intel_crtc_state *cs = to_intel_crtc_state(cstate);
 
@@ -15461,7 +15414,6 @@ void intel_modeset_init(struct drm_device *dev)
        }
 
        intel_update_czclk(dev_priv);
-       intel_update_rawclk(dev_priv);
        intel_update_cdclk(dev);
 
        intel_shared_dpll_init(dev);
@@ -16025,15 +15977,16 @@ retry:
 
 void intel_modeset_gem_init(struct drm_device *dev)
 {
+       struct drm_i915_private *dev_priv = to_i915(dev);
        struct drm_crtc *c;
        struct drm_i915_gem_object *obj;
        int ret;
 
-       intel_init_gt_powersave(dev);
+       intel_init_gt_powersave(dev_priv);
 
        intel_modeset_init_hw(dev);
 
-       intel_setup_overlay(dev);
+       intel_setup_overlay(dev_priv);
 
        /*
         * Make sure any fbs we allocated at startup are properly
@@ -16076,7 +16029,7 @@ void intel_modeset_cleanup(struct drm_device *dev)
        struct drm_i915_private *dev_priv = dev->dev_private;
        struct intel_connector *connector;
 
-       intel_disable_gt_powersave(dev);
+       intel_disable_gt_powersave(dev_priv);
 
        intel_backlight_unregister(dev);
 
@@ -16106,9 +16059,9 @@ void intel_modeset_cleanup(struct drm_device *dev)
 
        drm_mode_config_cleanup(dev);
 
-       intel_cleanup_overlay(dev);
+       intel_cleanup_overlay(dev_priv);
 
-       intel_cleanup_gt_powersave(dev);
+       intel_cleanup_gt_powersave(dev_priv);
 
        intel_teardown_gmbus(dev);
 }
@@ -16204,9 +16157,8 @@ struct intel_display_error_state {
 };
 
 struct intel_display_error_state *
-intel_display_capture_error_state(struct drm_device *dev)
+intel_display_capture_error_state(struct drm_i915_private *dev_priv)
 {
-       struct drm_i915_private *dev_priv = dev->dev_private;
        struct intel_display_error_state *error;
        int transcoders[] = {
                TRANSCODER_A,
@@ -16216,14 +16168,14 @@ intel_display_capture_error_state(struct drm_device *dev)
        };
        int i;
 
-       if (INTEL_INFO(dev)->num_pipes == 0)
+       if (INTEL_INFO(dev_priv)->num_pipes == 0)
                return NULL;
 
        error = kzalloc(sizeof(*error), GFP_ATOMIC);
        if (error == NULL)
                return NULL;
 
-       if (IS_HASWELL(dev) || IS_BROADWELL(dev))
+       if (IS_HASWELL(dev_priv) || IS_BROADWELL(dev_priv))
                error->power_well_driver = I915_READ(HSW_PWR_WELL_DRIVER);
 
        for_each_pipe(dev_priv, i) {
@@ -16239,25 +16191,25 @@ intel_display_capture_error_state(struct drm_device *dev)
 
                error->plane[i].control = I915_READ(DSPCNTR(i));
                error->plane[i].stride = I915_READ(DSPSTRIDE(i));
-               if (INTEL_INFO(dev)->gen <= 3) {
+               if (INTEL_GEN(dev_priv) <= 3) {
                        error->plane[i].size = I915_READ(DSPSIZE(i));
                        error->plane[i].pos = I915_READ(DSPPOS(i));
                }
-               if (INTEL_INFO(dev)->gen <= 7 && !IS_HASWELL(dev))
+               if (INTEL_GEN(dev_priv) <= 7 && !IS_HASWELL(dev_priv))
                        error->plane[i].addr = I915_READ(DSPADDR(i));
-               if (INTEL_INFO(dev)->gen >= 4) {
+               if (INTEL_GEN(dev_priv) >= 4) {
                        error->plane[i].surface = I915_READ(DSPSURF(i));
                        error->plane[i].tile_offset = I915_READ(DSPTILEOFF(i));
                }
 
                error->pipe[i].source = I915_READ(PIPESRC(i));
 
-               if (HAS_GMCH_DISPLAY(dev))
+               if (HAS_GMCH_DISPLAY(dev_priv))
                        error->pipe[i].stat = I915_READ(PIPESTAT(i));
        }
 
        /* Note: this does not include DSI transcoders. */
-       error->num_transcoders = INTEL_INFO(dev)->num_pipes;
+       error->num_transcoders = INTEL_INFO(dev_priv)->num_pipes;
        if (HAS_DDI(dev_priv))
                error->num_transcoders++; /* Account for eDP. */