Merge branch 'parisc-4.6-2' of git://git.kernel.org/pub/scm/linux/kernel/git/deller...
[cascardo/linux.git] / drivers / gpu / drm / rcar-du / rcar_du_plane.c
index c3ed952..8460ae1 100644 (file)
@@ -1,7 +1,7 @@
 /*
  * rcar_du_plane.c  --  R-Car Display Unit Planes
  *
- * Copyright (C) 2013-2014 Renesas Electronics Corporation
+ * Copyright (C) 2013-2015 Renesas Electronics Corporation
  *
  * Contact: Laurent Pinchart (laurent.pinchart@ideasonboard.com)
  *
@@ -12,6 +12,7 @@
  */
 
 #include <drm/drmP.h>
+#include <drm/drm_atomic.h>
 #include <drm/drm_atomic_helper.h>
 #include <drm/drm_crtc.h>
 #include <drm/drm_crtc_helper.h>
 #include <drm/drm_plane_helper.h>
 
 #include "rcar_du_drv.h"
+#include "rcar_du_group.h"
 #include "rcar_du_kms.h"
 #include "rcar_du_plane.h"
 #include "rcar_du_regs.h"
 
-#define RCAR_DU_COLORKEY_NONE          (0 << 24)
-#define RCAR_DU_COLORKEY_SOURCE                (1 << 24)
-#define RCAR_DU_COLORKEY_MASK          (1 << 24)
+/* -----------------------------------------------------------------------------
+ * Atomic hardware plane allocator
+ *
+ * The hardware plane allocator is solely based on the atomic plane states
+ * without keeping any external state to avoid races between .atomic_check()
+ * and .atomic_commit().
+ *
+ * The core idea is to avoid using a free planes bitmask that would need to be
+ * shared between check and commit handlers with a collective knowledge based on
+ * the allocated hardware plane(s) for each KMS plane. The allocator then loops
+ * over all plane states to compute the free planes bitmask, allocates hardware
+ * planes based on that bitmask, and stores the result back in the plane states.
+ *
+ * For this to work we need to access the current state of planes not touched by
+ * the atomic update. To ensure that it won't be modified, we need to lock all
+ * planes using drm_atomic_get_plane_state(). This effectively serializes atomic
+ * updates from .atomic_check() up to completion (when swapping the states if
+ * the check step has succeeded) or rollback (when freeing the states if the
+ * check step has failed).
+ *
+ * Allocation is performed in the .atomic_check() handler and applied
+ * automatically when the core swaps the old and new states.
+ */
+
+static bool rcar_du_plane_needs_realloc(struct rcar_du_plane *plane,
+                                       struct rcar_du_plane_state *new_state)
+{
+       struct rcar_du_plane_state *cur_state;
+
+       cur_state = to_rcar_plane_state(plane->plane.state);
+
+       /* Lowering the number of planes doesn't strictly require reallocation
+        * as the extra hardware plane will be freed when committing, but doing
+        * so could lead to more fragmentation.
+        */
+       if (!cur_state->format ||
+           cur_state->format->planes != new_state->format->planes)
+               return true;
+
+       /* Reallocate hardware planes if the source has changed. */
+       if (cur_state->source != new_state->source)
+               return true;
 
-static u32 rcar_du_plane_read(struct rcar_du_group *rgrp,
-                             unsigned int index, u32 reg)
+       return false;
+}
+
+static unsigned int rcar_du_plane_hwmask(struct rcar_du_plane_state *state)
+{
+       unsigned int mask;
+
+       if (state->hwindex == -1)
+               return 0;
+
+       mask = 1 << state->hwindex;
+       if (state->format->planes == 2)
+               mask |= 1 << ((state->hwindex + 1) % 8);
+
+       return mask;
+}
+
+/*
+ * The R8A7790 DU can source frames directly from the VSP1 devices VSPD0 and
+ * VSPD1. VSPD0 feeds DU0/1 plane 0, and VSPD1 feeds either DU2 plane 0 or
+ * DU0/1 plane 1.
+ *
+ * Allocate the correct fixed plane when sourcing frames from VSPD0 or VSPD1,
+ * and allocate planes in reverse index order otherwise to ensure maximum
+ * availability of planes 0 and 1.
+ *
+ * The caller is responsible for ensuring that the requested source is
+ * compatible with the DU revision.
+ */
+static int rcar_du_plane_hwalloc(struct rcar_du_plane *plane,
+                                struct rcar_du_plane_state *state,
+                                unsigned int free)
 {
-       return rcar_du_read(rgrp->dev,
-                           rgrp->mmio_offset + index * PLANE_OFF + reg);
+       unsigned int num_planes = state->format->planes;
+       int fixed = -1;
+       int i;
+
+       if (state->source == RCAR_DU_PLANE_VSPD0) {
+               /* VSPD0 feeds plane 0 on DU0/1. */
+               if (plane->group->index != 0)
+                       return -EINVAL;
+
+               fixed = 0;
+       } else if (state->source == RCAR_DU_PLANE_VSPD1) {
+               /* VSPD1 feeds plane 1 on DU0/1 or plane 0 on DU2. */
+               fixed = plane->group->index == 0 ? 1 : 0;
+       }
+
+       if (fixed >= 0)
+               return free & (1 << fixed) ? fixed : -EBUSY;
+
+       for (i = RCAR_DU_NUM_HW_PLANES - 1; i >= 0; --i) {
+               if (!(free & (1 << i)))
+                       continue;
+
+               if (num_planes == 1 || free & (1 << ((i + 1) % 8)))
+                       break;
+       }
+
+       return i < 0 ? -EBUSY : i;
 }
 
+int rcar_du_atomic_check_planes(struct drm_device *dev,
+                               struct drm_atomic_state *state)
+{
+       struct rcar_du_device *rcdu = dev->dev_private;
+       unsigned int group_freed_planes[RCAR_DU_MAX_GROUPS] = { 0, };
+       unsigned int group_free_planes[RCAR_DU_MAX_GROUPS] = { 0, };
+       bool needs_realloc = false;
+       unsigned int groups = 0;
+       unsigned int i;
+
+       /* Check if hardware planes need to be reallocated. */
+       for (i = 0; i < dev->mode_config.num_total_plane; ++i) {
+               struct rcar_du_plane_state *plane_state;
+               struct rcar_du_plane *plane;
+               unsigned int index;
+
+               if (!state->planes[i])
+                       continue;
+
+               plane = to_rcar_plane(state->planes[i]);
+               plane_state = to_rcar_plane_state(state->plane_states[i]);
+
+               dev_dbg(rcdu->dev, "%s: checking plane (%u,%tu)\n", __func__,
+                       plane->group->index, plane - plane->group->planes);
+
+               /* If the plane is being disabled we don't need to go through
+                * the full reallocation procedure. Just mark the hardware
+                * plane(s) as freed.
+                */
+               if (!plane_state->format) {
+                       dev_dbg(rcdu->dev, "%s: plane is being disabled\n",
+                               __func__);
+                       index = plane - plane->group->planes;
+                       group_freed_planes[plane->group->index] |= 1 << index;
+                       plane_state->hwindex = -1;
+                       continue;
+               }
+
+               /* If the plane needs to be reallocated mark it as such, and
+                * mark the hardware plane(s) as free.
+                */
+               if (rcar_du_plane_needs_realloc(plane, plane_state)) {
+                       dev_dbg(rcdu->dev, "%s: plane needs reallocation\n",
+                               __func__);
+                       groups |= 1 << plane->group->index;
+                       needs_realloc = true;
+
+                       index = plane - plane->group->planes;
+                       group_freed_planes[plane->group->index] |= 1 << index;
+                       plane_state->hwindex = -1;
+               }
+       }
+
+       if (!needs_realloc)
+               return 0;
+
+       /* Grab all plane states for the groups that need reallocation to ensure
+        * locking and avoid racy updates. This serializes the update operation,
+        * but there's not much we can do about it as that's the hardware
+        * design.
+        *
+        * Compute the used planes mask for each group at the same time to avoid
+        * looping over the planes separately later.
+        */
+       while (groups) {
+               unsigned int index = ffs(groups) - 1;
+               struct rcar_du_group *group = &rcdu->groups[index];
+               unsigned int used_planes = 0;
+
+               dev_dbg(rcdu->dev, "%s: finding free planes for group %u\n",
+                       __func__, index);
+
+               for (i = 0; i < group->num_planes; ++i) {
+                       struct rcar_du_plane *plane = &group->planes[i];
+                       struct rcar_du_plane_state *plane_state;
+                       struct drm_plane_state *s;
+
+                       s = drm_atomic_get_plane_state(state, &plane->plane);
+                       if (IS_ERR(s))
+                               return PTR_ERR(s);
+
+                       /* If the plane has been freed in the above loop its
+                        * hardware planes must not be added to the used planes
+                        * bitmask. However, the current state doesn't reflect
+                        * the free state yet, as we've modified the new state
+                        * above. Use the local freed planes list to check for
+                        * that condition instead.
+                        */
+                       if (group_freed_planes[index] & (1 << i)) {
+                               dev_dbg(rcdu->dev,
+                                       "%s: plane (%u,%tu) has been freed, skipping\n",
+                                       __func__, plane->group->index,
+                                       plane - plane->group->planes);
+                               continue;
+                       }
+
+                       plane_state = to_rcar_plane_state(plane->plane.state);
+                       used_planes |= rcar_du_plane_hwmask(plane_state);
+
+                       dev_dbg(rcdu->dev,
+                               "%s: plane (%u,%tu) uses %u hwplanes (index %d)\n",
+                               __func__, plane->group->index,
+                               plane - plane->group->planes,
+                               plane_state->format ?
+                               plane_state->format->planes : 0,
+                               plane_state->hwindex);
+               }
+
+               group_free_planes[index] = 0xff & ~used_planes;
+               groups &= ~(1 << index);
+
+               dev_dbg(rcdu->dev, "%s: group %u free planes mask 0x%02x\n",
+                       __func__, index, group_free_planes[index]);
+       }
+
+       /* Reallocate hardware planes for each plane that needs it. */
+       for (i = 0; i < dev->mode_config.num_total_plane; ++i) {
+               struct rcar_du_plane_state *plane_state;
+               struct rcar_du_plane *plane;
+               unsigned int crtc_planes;
+               unsigned int free;
+               int idx;
+
+               if (!state->planes[i])
+                       continue;
+
+               plane = to_rcar_plane(state->planes[i]);
+               plane_state = to_rcar_plane_state(state->plane_states[i]);
+
+               dev_dbg(rcdu->dev, "%s: allocating plane (%u,%tu)\n", __func__,
+                       plane->group->index, plane - plane->group->planes);
+
+               /* Skip planes that are being disabled or don't need to be
+                * reallocated.
+                */
+               if (!plane_state->format ||
+                   !rcar_du_plane_needs_realloc(plane, plane_state))
+                       continue;
+
+               /* Try to allocate the plane from the free planes currently
+                * associated with the target CRTC to avoid restarting the CRTC
+                * group and thus minimize flicker. If it fails fall back to
+                * allocating from all free planes.
+                */
+               crtc_planes = to_rcar_crtc(plane_state->state.crtc)->index % 2
+                           ? plane->group->dptsr_planes
+                           : ~plane->group->dptsr_planes;
+               free = group_free_planes[plane->group->index];
+
+               idx = rcar_du_plane_hwalloc(plane, plane_state,
+                                           free & crtc_planes);
+               if (idx < 0)
+                       idx = rcar_du_plane_hwalloc(plane, plane_state,
+                                                   free);
+               if (idx < 0) {
+                       dev_dbg(rcdu->dev, "%s: no available hardware plane\n",
+                               __func__);
+                       return idx;
+               }
+
+               dev_dbg(rcdu->dev, "%s: allocated %u hwplanes (index %u)\n",
+                       __func__, plane_state->format->planes, idx);
+
+               plane_state->hwindex = idx;
+
+               group_free_planes[plane->group->index] &=
+                       ~rcar_du_plane_hwmask(plane_state);
+
+               dev_dbg(rcdu->dev, "%s: group %u free planes mask 0x%02x\n",
+                       __func__, plane->group->index,
+                       group_free_planes[plane->group->index]);
+       }
+
+       return 0;
+}
+
+/* -----------------------------------------------------------------------------
+ * Plane Setup
+ */
+
+#define RCAR_DU_COLORKEY_NONE          (0 << 24)
+#define RCAR_DU_COLORKEY_SOURCE                (1 << 24)
+#define RCAR_DU_COLORKEY_MASK          (1 << 24)
+
 static void rcar_du_plane_write(struct rcar_du_group *rgrp,
                                unsigned int index, u32 reg, u32 data)
 {
@@ -42,34 +322,45 @@ static void rcar_du_plane_write(struct rcar_du_group *rgrp,
                      data);
 }
 
-static void rcar_du_plane_setup_fb(struct rcar_du_plane *plane)
+static void rcar_du_plane_setup_scanout(struct rcar_du_group *rgrp,
+                                       const struct rcar_du_plane_state *state)
 {
-       struct rcar_du_plane_state *state =
-               to_rcar_plane_state(plane->plane.state);
-       struct drm_framebuffer *fb = plane->plane.state->fb;
-       struct rcar_du_group *rgrp = plane->group;
        unsigned int src_x = state->state.src_x >> 16;
        unsigned int src_y = state->state.src_y >> 16;
        unsigned int index = state->hwindex;
-       struct drm_gem_cma_object *gem;
+       unsigned int pitch;
        bool interlaced;
-       u32 mwr;
+       u32 dma[2];
 
        interlaced = state->state.crtc->state->adjusted_mode.flags
                   & DRM_MODE_FLAG_INTERLACE;
 
+       if (state->source == RCAR_DU_PLANE_MEMORY) {
+               struct drm_framebuffer *fb = state->state.fb;
+               struct drm_gem_cma_object *gem;
+               unsigned int i;
+
+               if (state->format->planes == 2)
+                       pitch = fb->pitches[0];
+               else
+                       pitch = fb->pitches[0] * 8 / state->format->bpp;
+
+               for (i = 0; i < state->format->planes; ++i) {
+                       gem = drm_fb_cma_get_gem_obj(fb, i);
+                       dma[i] = gem->paddr + fb->offsets[i];
+               }
+       } else {
+               pitch = state->state.src_w >> 16;
+               dma[0] = 0;
+               dma[1] = 0;
+       }
+
        /* Memory pitch (expressed in pixels). Must be doubled for interlaced
         * operation with 32bpp formats.
         */
-       if (state->format->planes == 2)
-               mwr = fb->pitches[0];
-       else
-               mwr = fb->pitches[0] * 8 / state->format->bpp;
-
-       if (interlaced && state->format->bpp == 32)
-               mwr *= 2;
-
-       rcar_du_plane_write(rgrp, index, PnMWR, mwr);
+       rcar_du_plane_write(rgrp, index, PnMWR,
+                           (interlaced && state->format->bpp == 32) ?
+                           pitch * 2 : pitch);
 
        /* The Y position is expressed in raster line units and must be doubled
         * for 32bpp formats, according to the R8A7790 datasheet. No mention of
@@ -87,30 +378,25 @@ static void rcar_du_plane_setup_fb(struct rcar_du_plane *plane)
        rcar_du_plane_write(rgrp, index, PnSPYR, src_y *
                            (!interlaced && state->format->bpp == 32 ? 2 : 1));
 
-       gem = drm_fb_cma_get_gem_obj(fb, 0);
-       rcar_du_plane_write(rgrp, index, PnDSA0R, gem->paddr + fb->offsets[0]);
+       rcar_du_plane_write(rgrp, index, PnDSA0R, dma[0]);
 
        if (state->format->planes == 2) {
                index = (index + 1) % 8;
 
-               rcar_du_plane_write(rgrp, index, PnMWR, fb->pitches[0]);
+               rcar_du_plane_write(rgrp, index, PnMWR, pitch);
 
                rcar_du_plane_write(rgrp, index, PnSPXR, src_x);
                rcar_du_plane_write(rgrp, index, PnSPYR, src_y *
                                    (state->format->bpp == 16 ? 2 : 1) / 2);
 
-               gem = drm_fb_cma_get_gem_obj(fb, 1);
-               rcar_du_plane_write(rgrp, index, PnDSA0R,
-                                   gem->paddr + fb->offsets[1]);
+               rcar_du_plane_write(rgrp, index, PnDSA0R, dma[1]);
        }
 }
 
-static void rcar_du_plane_setup_mode(struct rcar_du_plane *plane,
-                                    unsigned int index)
+static void rcar_du_plane_setup_mode(struct rcar_du_group *rgrp,
+                                    unsigned int index,
+                                    const struct rcar_du_plane_state *state)
 {
-       struct rcar_du_plane_state *state =
-               to_rcar_plane_state(plane->plane.state);
-       struct rcar_du_group *rgrp = plane->group;
        u32 colorkey;
        u32 pnmr;
 
@@ -168,12 +454,10 @@ static void rcar_du_plane_setup_mode(struct rcar_du_plane *plane,
        }
 }
 
-static void __rcar_du_plane_setup(struct rcar_du_plane *plane,
-                                 unsigned int index)
+static void rcar_du_plane_setup_format_gen2(struct rcar_du_group *rgrp,
+                                           unsigned int index,
+                                           const struct rcar_du_plane_state *state)
 {
-       struct rcar_du_plane_state *state =
-               to_rcar_plane_state(plane->plane.state);
-       struct rcar_du_group *rgrp = plane->group;
        u32 ddcr2 = PnDDCR2_CODE;
        u32 ddcr4;
 
@@ -182,11 +466,8 @@ static void __rcar_du_plane_setup(struct rcar_du_plane *plane,
         * The data format is selected by the DDDF field in PnMR and the EDF
         * field in DDCR4.
         */
-       ddcr4 = rcar_du_plane_read(rgrp, index, PnDDCR4);
-       ddcr4 &= ~PnDDCR4_EDF_MASK;
-       ddcr4 |= state->format->edf | PnDDCR4_CODE;
 
-       rcar_du_plane_setup_mode(plane, index);
+       rcar_du_plane_setup_mode(rgrp, index, state);
 
        if (state->format->planes == 2) {
                if (state->hwindex != index) {
@@ -204,31 +485,72 @@ static void __rcar_du_plane_setup(struct rcar_du_plane *plane,
        }
 
        rcar_du_plane_write(rgrp, index, PnDDCR2, ddcr2);
+
+       ddcr4 = state->format->edf | PnDDCR4_CODE;
+       if (state->source != RCAR_DU_PLANE_MEMORY)
+               ddcr4 |= PnDDCR4_VSPS;
+
        rcar_du_plane_write(rgrp, index, PnDDCR4, ddcr4);
+}
+
+static void rcar_du_plane_setup_format_gen3(struct rcar_du_group *rgrp,
+                                           unsigned int index,
+                                           const struct rcar_du_plane_state *state)
+{
+       rcar_du_plane_write(rgrp, index, PnMR,
+                           PnMR_SPIM_TP_OFF | state->format->pnmr);
+
+       rcar_du_plane_write(rgrp, index, PnDDCR4,
+                           state->format->edf | PnDDCR4_CODE);
+}
+
+static void rcar_du_plane_setup_format(struct rcar_du_group *rgrp,
+                                      unsigned int index,
+                                      const struct rcar_du_plane_state *state)
+{
+       struct rcar_du_device *rcdu = rgrp->dev;
+
+       if (rcdu->info->gen < 3)
+               rcar_du_plane_setup_format_gen2(rgrp, index, state);
+       else
+               rcar_du_plane_setup_format_gen3(rgrp, index, state);
 
        /* Destination position and size */
-       rcar_du_plane_write(rgrp, index, PnDSXR, plane->plane.state->crtc_w);
-       rcar_du_plane_write(rgrp, index, PnDSYR, plane->plane.state->crtc_h);
-       rcar_du_plane_write(rgrp, index, PnDPXR, plane->plane.state->crtc_x);
-       rcar_du_plane_write(rgrp, index, PnDPYR, plane->plane.state->crtc_y);
-
-       /* Wrap-around and blinking, disabled */
-       rcar_du_plane_write(rgrp, index, PnWASPR, 0);
-       rcar_du_plane_write(rgrp, index, PnWAMWR, 4095);
-       rcar_du_plane_write(rgrp, index, PnBTR, 0);
-       rcar_du_plane_write(rgrp, index, PnMLR, 0);
+       rcar_du_plane_write(rgrp, index, PnDSXR, state->state.crtc_w);
+       rcar_du_plane_write(rgrp, index, PnDSYR, state->state.crtc_h);
+       rcar_du_plane_write(rgrp, index, PnDPXR, state->state.crtc_x);
+       rcar_du_plane_write(rgrp, index, PnDPYR, state->state.crtc_y);
+
+       if (rcdu->info->gen < 3) {
+               /* Wrap-around and blinking, disabled */
+               rcar_du_plane_write(rgrp, index, PnWASPR, 0);
+               rcar_du_plane_write(rgrp, index, PnWAMWR, 4095);
+               rcar_du_plane_write(rgrp, index, PnBTR, 0);
+               rcar_du_plane_write(rgrp, index, PnMLR, 0);
+       }
 }
 
-void rcar_du_plane_setup(struct rcar_du_plane *plane)
+void __rcar_du_plane_setup(struct rcar_du_group *rgrp,
+                          const struct rcar_du_plane_state *state)
 {
-       struct rcar_du_plane_state *state =
-               to_rcar_plane_state(plane->plane.state);
+       struct rcar_du_device *rcdu = rgrp->dev;
 
-       __rcar_du_plane_setup(plane, state->hwindex);
+       rcar_du_plane_setup_format(rgrp, state->hwindex, state);
        if (state->format->planes == 2)
-               __rcar_du_plane_setup(plane, (state->hwindex + 1) % 8);
+               rcar_du_plane_setup_format(rgrp, (state->hwindex + 1) % 8,
+                                          state);
 
-       rcar_du_plane_setup_fb(plane);
+       if (rcdu->info->gen < 3)
+               rcar_du_plane_setup_scanout(rgrp, state);
+
+       if (state->source == RCAR_DU_PLANE_VSPD1) {
+               unsigned int vspd1_sink = rgrp->index ? 2 : 0;
+
+               if (rcdu->vspd1_sink != vspd1_sink) {
+                       rcdu->vspd1_sink = vspd1_sink;
+                       rcar_du_set_dpad0_vsp1_routing(rcdu);
+               }
+       }
 }
 
 static int rcar_du_plane_atomic_check(struct drm_plane *plane,
@@ -263,9 +585,27 @@ static void rcar_du_plane_atomic_update(struct drm_plane *plane,
                                        struct drm_plane_state *old_state)
 {
        struct rcar_du_plane *rplane = to_rcar_plane(plane);
+       struct rcar_du_plane_state *old_rstate;
+       struct rcar_du_plane_state *new_rstate;
+
+       if (!plane->state->crtc)
+               return;
+
+       rcar_du_plane_setup(rplane);
+
+       /* Check whether the source has changed from memory to live source or
+        * from live source to memory. The source has been configured by the
+        * VSPS bit in the PnDDCR4 register. Although the datasheet states that
+        * the bit is updated during vertical blanking, it seems that updates
+        * only occur when the DU group is held in reset through the DSYSR.DRES
+        * bit. We thus need to restart the group if the source changes.
+        */
+       old_rstate = to_rcar_plane_state(old_state);
+       new_rstate = to_rcar_plane_state(plane->state);
 
-       if (plane->state->crtc)
-               rcar_du_plane_setup(rplane);
+       if ((old_rstate->source == RCAR_DU_PLANE_MEMORY) !=
+           (new_rstate->source == RCAR_DU_PLANE_MEMORY))
+               rplane->group->need_restart = true;
 }
 
 static const struct drm_plane_helper_funcs rcar_du_plane_helper_funcs = {
@@ -313,6 +653,7 @@ static void rcar_du_plane_reset(struct drm_plane *plane)
                return;
 
        state->hwindex = -1;
+       state->source = RCAR_DU_PLANE_MEMORY;
        state->alpha = 255;
        state->colorkey = RCAR_DU_COLORKEY_NONE;
        state->zpos = plane->type == DRM_PLANE_TYPE_PRIMARY ? 0 : 1;