target/linux/brcm2708/patches-4.4/0281-drm-vc4-Add-support-for-scaling-of-display-planes.patch

   1 From 8d513c7a67cce0bf0ef312323753eccbd0f3f71a Mon Sep 17 00:00:00 2001
   2 From: Eric Anholt <eric@anholt.net>
   3 Date: Tue, 20 Oct 2015 16:06:57 +0100
   4 Subject: [PATCH 281/423] drm/vc4: Add support for scaling of display planes.
   5
   6 This implements a simple policy for choosing scaling modes
   7 (trapezoidal for decimation, PPF for magnification), and a single PPF
   8 filter (Mitchell/Netravali's recommendation).
   9
  10 Signed-off-by: Eric Anholt <eric@anholt.net>
  11 (cherry picked from commit 21af94cf1a4c2d3450ab7fead58e6e2291ab92a9)
  12 ---
  13  drivers/gpu/drm/vc4/vc4_drv.h   |   4 +
  14  drivers/gpu/drm/vc4/vc4_hvs.c   |  84 +++++++++++++
  15  drivers/gpu/drm/vc4/vc4_plane.c | 253 +++++++++++++++++++++++++++++++++++++---
  16  drivers/gpu/drm/vc4/vc4_regs.h  |  46 ++++++++
  17  4 files changed, 374 insertions(+), 13 deletions(-)
  18
  19 --- a/drivers/gpu/drm/vc4/vc4_drv.h
  20 +++ b/drivers/gpu/drm/vc4/vc4_drv.h
  21 @@ -156,7 +156,11 @@ struct vc4_hvs {
  22          * list.  Units are dwords.
  23          */
  24         struct drm_mm dlist_mm;
  25 +       /* Memory manager for the LBM memory used by HVS scaling. */
  26 +       struct drm_mm lbm_mm;
  27         spinlock_t mm_lock;
  28 +
  29 +       struct drm_mm_node mitchell_netravali_filter;
  30  };
  31
  32  struct vc4_plane {
  33 --- a/drivers/gpu/drm/vc4/vc4_hvs.c
  34 +++ b/drivers/gpu/drm/vc4/vc4_hvs.c
  35 @@ -100,12 +100,76 @@ int vc4_hvs_debugfs_regs(struct seq_file
  36  }
  37  #endif
  38
  39 +/* The filter kernel is composed of dwords each containing 3 9-bit
  40 + * signed integers packed next to each other.
  41 + */
  42 +#define VC4_INT_TO_COEFF(coeff) (coeff & 0x1ff)
  43 +#define VC4_PPF_FILTER_WORD(c0, c1, c2)                                \
  44 +       ((((c0) & 0x1ff) << 0) |                                \
  45 +        (((c1) & 0x1ff) << 9) |                                \
  46 +        (((c2) & 0x1ff) << 18))
  47 +
  48 +/* The whole filter kernel is arranged as the coefficients 0-16 going
  49 + * up, then a pad, then 17-31 going down and reversed within the
  50 + * dwords.  This means that a linear phase kernel (where it's
  51 + * symmetrical at the boundary between 15 and 16) has the last 5
  52 + * dwords matching the first 5, but reversed.
  53 + */
  54 +#define VC4_LINEAR_PHASE_KERNEL(c0, c1, c2, c3, c4, c5, c6, c7, c8,    \
  55 +                               c9, c10, c11, c12, c13, c14, c15)       \
  56 +       {VC4_PPF_FILTER_WORD(c0, c1, c2),                               \
  57 +        VC4_PPF_FILTER_WORD(c3, c4, c5),                               \
  58 +        VC4_PPF_FILTER_WORD(c6, c7, c8),                               \
  59 +        VC4_PPF_FILTER_WORD(c9, c10, c11),                             \
  60 +        VC4_PPF_FILTER_WORD(c12, c13, c14),                            \
  61 +        VC4_PPF_FILTER_WORD(c15, c15, 0)}
  62 +
  63 +#define VC4_LINEAR_PHASE_KERNEL_DWORDS 6
  64 +#define VC4_KERNEL_DWORDS (VC4_LINEAR_PHASE_KERNEL_DWORDS * 2 - 1)
  65 +
  66 +/* Recommended B=1/3, C=1/3 filter choice from Mitchell/Netravali.
  67 + * http://www.cs.utexas.edu/~fussell/courses/cs384g/lectures/mitchell/Mitchell.pdf
  68 + */
  69 +static const u32 mitchell_netravali_1_3_1_3_kernel[] =
  70 +       VC4_LINEAR_PHASE_KERNEL(0, -2, -6, -8, -10, -8, -3, 2, 18,
  71 +                               50, 82, 119, 155, 187, 213, 227);
  72 +
  73 +static int vc4_hvs_upload_linear_kernel(struct vc4_hvs *hvs,
  74 +                                       struct drm_mm_node *space,
  75 +                                       const u32 *kernel)
  76 +{
  77 +       int ret, i;
  78 +       u32 __iomem *dst_kernel;
  79 +
  80 +       ret = drm_mm_insert_node(&hvs->dlist_mm, space, VC4_KERNEL_DWORDS, 1,
  81 +                                0);
  82 +       if (ret) {
  83 +               DRM_ERROR("Failed to allocate space for filter kernel: %d\n",
  84 +                         ret);
  85 +               return ret;
  86 +       }
  87 +
  88 +       dst_kernel = hvs->dlist + space->start;
  89 +
  90 +       for (i = 0; i < VC4_KERNEL_DWORDS; i++) {
  91 +               if (i < VC4_LINEAR_PHASE_KERNEL_DWORDS)
  92 +                       writel(kernel[i], &dst_kernel[i]);
  93 +               else {
  94 +                       writel(kernel[VC4_KERNEL_DWORDS - i - 1],
  95 +                              &dst_kernel[i]);
  96 +               }
  97 +       }
  98 +
  99 +       return 0;
 100 +}
 101 +
 102  static int vc4_hvs_bind(struct device *dev, struct device *master, void *data)
 103  {
 104         struct platform_device *pdev = to_platform_device(dev);
 105         struct drm_device *drm = dev_get_drvdata(master);
 106         struct vc4_dev *vc4 = drm->dev_private;
 107         struct vc4_hvs *hvs = NULL;
 108 +       int ret;
 109
 110         hvs = devm_kzalloc(&pdev->dev, sizeof(*hvs), GFP_KERNEL);
 111         if (!hvs)
 112 @@ -130,6 +194,22 @@ static int vc4_hvs_bind(struct device *d
 113                     HVS_BOOTLOADER_DLIST_END,
 114                     (SCALER_DLIST_SIZE >> 2) - HVS_BOOTLOADER_DLIST_END);
 115
 116 +       /* Set up the HVS LBM memory manager.  We could have some more
 117 +        * complicated data structure that allowed reuse of LBM areas
 118 +        * between planes when they don't overlap on the screen, but
 119 +        * for now we just allocate globally.
 120 +        */
 121 +       drm_mm_init(&hvs->lbm_mm, 0, 96 * 1024);
 122 +
 123 +       /* Upload filter kernels.  We only have the one for now, so we
 124 +        * keep it around for the lifetime of the driver.
 125 +        */
 126 +       ret = vc4_hvs_upload_linear_kernel(hvs,
 127 +                                          &hvs->mitchell_netravali_filter,
 128 +                                          mitchell_netravali_1_3_1_3_kernel);
 129 +       if (ret)
 130 +               return ret;
 131 +
 132         vc4->hvs = hvs;
 133         return 0;
 134  }
 135 @@ -140,7 +220,11 @@ static void vc4_hvs_unbind(struct device
 136         struct drm_device *drm = dev_get_drvdata(master);
 137         struct vc4_dev *vc4 = drm->dev_private;
 138
 139 +       if (vc4->hvs->mitchell_netravali_filter.allocated)
 140 +               drm_mm_remove_node(&vc4->hvs->mitchell_netravali_filter);
 141 +
 142         drm_mm_takedown(&vc4->hvs->dlist_mm);
 143 +       drm_mm_takedown(&vc4->hvs->lbm_mm);
 144
 145         vc4->hvs = NULL;
 146  }
 147 --- a/drivers/gpu/drm/vc4/vc4_plane.c
 148 +++ b/drivers/gpu/drm/vc4/vc4_plane.c
 149 @@ -24,6 +24,12 @@
 150  #include "drm_fb_cma_helper.h"
 151  #include "drm_plane_helper.h"
 152
 153 +enum vc4_scaling_mode {
 154 +       VC4_SCALING_NONE,
 155 +       VC4_SCALING_TPZ,
 156 +       VC4_SCALING_PPF,
 157 +};
 158 +
 159  struct vc4_plane_state {
 160         struct drm_plane_state base;
 161         /* System memory copy of the display list for this element, computed
 162 @@ -47,13 +53,19 @@ struct vc4_plane_state {
 163
 164         /* Clipped coordinates of the plane on the display. */
 165         int crtc_x, crtc_y, crtc_w, crtc_h;
 166 -       /* Clipped size of the area scanned from in the FB. */
 167 -       u32 src_w, src_h;
 168 +       /* Clipped area being scanned from in the FB. */
 169 +       u32 src_x, src_y, src_w, src_h;
 170 +
 171 +       enum vc4_scaling_mode x_scaling, y_scaling;
 172 +       bool is_unity;
 173
 174         /* Offset to start scanning out from the start of the plane's
 175          * BO.
 176          */
 177         u32 offset;
 178 +
 179 +       /* Our allocation in LBM for temporary storage during scaling. */
 180 +       struct drm_mm_node lbm;
 181  };
 182
 183  static inline struct vc4_plane_state *
 184 @@ -106,6 +118,16 @@ static const struct hvs_format *vc4_get_
 185         return NULL;
 186  }
 187
 188 +static enum vc4_scaling_mode vc4_get_scaling_mode(u32 src, u32 dst)
 189 +{
 190 +       if (dst > src)
 191 +               return VC4_SCALING_PPF;
 192 +       else if (dst < src)
 193 +               return VC4_SCALING_TPZ;
 194 +       else
 195 +               return VC4_SCALING_NONE;
 196 +}
 197 +
 198  static bool plane_enabled(struct drm_plane_state *state)
 199  {
 200         return state->fb && state->crtc;
 201 @@ -122,6 +144,8 @@ static struct drm_plane_state *vc4_plane
 202         if (!vc4_state)
 203                 return NULL;
 204
 205 +       memset(&vc4_state->lbm, 0, sizeof(vc4_state->lbm));
 206 +
 207         __drm_atomic_helper_plane_duplicate_state(plane, &vc4_state->base);
 208
 209         if (vc4_state->dlist) {
 210 @@ -141,8 +165,17 @@ static struct drm_plane_state *vc4_plane
 211  static void vc4_plane_destroy_state(struct drm_plane *plane,
 212                                     struct drm_plane_state *state)
 213  {
 214 +       struct vc4_dev *vc4 = to_vc4_dev(plane->dev);
 215         struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
 216
 217 +       if (vc4_state->lbm.allocated) {
 218 +               unsigned long irqflags;
 219 +
 220 +               spin_lock_irqsave(&vc4->hvs->mm_lock, irqflags);
 221 +               drm_mm_remove_node(&vc4_state->lbm);
 222 +               spin_unlock_irqrestore(&vc4->hvs->mm_lock, irqflags);
 223 +       }
 224 +
 225         kfree(vc4_state->dlist);
 226         __drm_atomic_helper_plane_destroy_state(plane, &vc4_state->base);
 227         kfree(state);
 228 @@ -181,23 +214,60 @@ static void vc4_dlist_write(struct vc4_p
 229         vc4_state->dlist[vc4_state->dlist_count++] = val;
 230  }
 231
 232 +/* Returns the scl0/scl1 field based on whether the dimensions need to
 233 + * be up/down/non-scaled.
 234 + *
 235 + * This is a replication of a table from the spec.
 236 + */
 237 +static u32 vc4_get_scl_field(struct drm_plane_state *state)
 238 +{
 239 +       struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
 240 +
 241 +       switch (vc4_state->x_scaling << 2 | vc4_state->y_scaling) {
 242 +       case VC4_SCALING_PPF << 2 | VC4_SCALING_PPF:
 243 +               return SCALER_CTL0_SCL_H_PPF_V_PPF;
 244 +       case VC4_SCALING_TPZ << 2 | VC4_SCALING_PPF:
 245 +               return SCALER_CTL0_SCL_H_TPZ_V_PPF;
 246 +       case VC4_SCALING_PPF << 2 | VC4_SCALING_TPZ:
 247 +               return SCALER_CTL0_SCL_H_PPF_V_TPZ;
 248 +       case VC4_SCALING_TPZ << 2 | VC4_SCALING_TPZ:
 249 +               return SCALER_CTL0_SCL_H_TPZ_V_TPZ;
 250 +       case VC4_SCALING_PPF << 2 | VC4_SCALING_NONE:
 251 +               return SCALER_CTL0_SCL_H_PPF_V_NONE;
 252 +       case VC4_SCALING_NONE << 2 | VC4_SCALING_PPF:
 253 +               return SCALER_CTL0_SCL_H_NONE_V_PPF;
 254 +       case VC4_SCALING_NONE << 2 | VC4_SCALING_TPZ:
 255 +               return SCALER_CTL0_SCL_H_NONE_V_TPZ;
 256 +       case VC4_SCALING_TPZ << 2 | VC4_SCALING_NONE:
 257 +               return SCALER_CTL0_SCL_H_TPZ_V_NONE;
 258 +       default:
 259 +       case VC4_SCALING_NONE << 2 | VC4_SCALING_NONE:
 260 +               /* The unity case is independently handled by
 261 +                * SCALER_CTL0_UNITY.
 262 +                */
 263 +               return 0;
 264 +       }
 265 +}
 266 +
 267  static int vc4_plane_setup_clipping_and_scaling(struct drm_plane_state *state)
 268  {
 269 +       struct drm_plane *plane = state->plane;
 270         struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
 271         struct drm_framebuffer *fb = state->fb;
 272 +       u32 subpixel_src_mask = (1 << 16) - 1;
 273
 274         vc4_state->offset = fb->offsets[0];
 275
 276 -       if (state->crtc_w << 16 != state->src_w ||
 277 -           state->crtc_h << 16 != state->src_h) {
 278 -               /* We don't support scaling yet, which involves
 279 -                * allocating the LBM memory for scaling temporary
 280 -                * storage, and putting filter kernels in the HVS
 281 -                * context.
 282 -                */
 283 +       /* We don't support subpixel source positioning for scaling. */
 284 +       if ((state->src_x & subpixel_src_mask) ||
 285 +           (state->src_y & subpixel_src_mask) ||
 286 +           (state->src_w & subpixel_src_mask) ||
 287 +           (state->src_h & subpixel_src_mask)) {
 288                 return -EINVAL;
 289         }
 290
 291 +       vc4_state->src_x = state->src_x >> 16;
 292 +       vc4_state->src_y = state->src_y >> 16;
 293         vc4_state->src_w = state->src_w >> 16;
 294         vc4_state->src_h = state->src_h >> 16;
 295
 296 @@ -206,6 +276,23 @@ static int vc4_plane_setup_clipping_and_
 297         vc4_state->crtc_w = state->crtc_w;
 298         vc4_state->crtc_h = state->crtc_h;
 299
 300 +       vc4_state->x_scaling = vc4_get_scaling_mode(vc4_state->src_w,
 301 +                                                   vc4_state->crtc_w);
 302 +       vc4_state->y_scaling = vc4_get_scaling_mode(vc4_state->src_h,
 303 +                                                   vc4_state->crtc_h);
 304 +       vc4_state->is_unity = (vc4_state->x_scaling == VC4_SCALING_NONE &&
 305 +                              vc4_state->y_scaling == VC4_SCALING_NONE);
 306 +
 307 +       /* No configuring scaling on the cursor plane, since it gets
 308 +          non-vblank-synced updates, and scaling requires requires
 309 +          LBM changes which have to be vblank-synced.
 310 +        */
 311 +       if (plane->type == DRM_PLANE_TYPE_CURSOR && !vc4_state->is_unity)
 312 +               return -EINVAL;
 313 +
 314 +       /* Clamp the on-screen start x/y to 0.  The hardware doesn't
 315 +        * support negative y, and negative x wastes bandwidth.
 316 +        */
 317         if (vc4_state->crtc_x < 0) {
 318                 vc4_state->offset += (drm_format_plane_cpp(fb->pixel_format,
 319                                                            0) *
 320 @@ -223,6 +310,87 @@ static int vc4_plane_setup_clipping_and_
 321         return 0;
 322  }
 323
 324 +static void vc4_write_tpz(struct vc4_plane_state *vc4_state, u32 src, u32 dst)
 325 +{
 326 +       u32 scale, recip;
 327 +
 328 +       scale = (1 << 16) * src / dst;
 329 +
 330 +       /* The specs note that while the reciprocal would be defined
 331 +        * as (1<<32)/scale, ~0 is close enough.
 332 +        */
 333 +       recip = ~0 / scale;
 334 +
 335 +       vc4_dlist_write(vc4_state,
 336 +                       VC4_SET_FIELD(scale, SCALER_TPZ0_SCALE) |
 337 +                       VC4_SET_FIELD(0, SCALER_TPZ0_IPHASE));
 338 +       vc4_dlist_write(vc4_state,
 339 +                       VC4_SET_FIELD(recip, SCALER_TPZ1_RECIP));
 340 +}
 341 +
 342 +static void vc4_write_ppf(struct vc4_plane_state *vc4_state, u32 src, u32 dst)
 343 +{
 344 +       u32 scale = (1 << 16) * src / dst;
 345 +
 346 +       vc4_dlist_write(vc4_state,
 347 +                       SCALER_PPF_AGC |
 348 +                       VC4_SET_FIELD(scale, SCALER_PPF_SCALE) |
 349 +                       VC4_SET_FIELD(0, SCALER_PPF_IPHASE));
 350 +}
 351 +
 352 +static u32 vc4_lbm_size(struct drm_plane_state *state)
 353 +{
 354 +       struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
 355 +       /* This is the worst case number.  One of the two sizes will
 356 +        * be used depending on the scaling configuration.
 357 +        */
 358 +       u32 pix_per_line = max(vc4_state->src_w, (u32)vc4_state->crtc_w);
 359 +       u32 lbm;
 360 +
 361 +       if (vc4_state->is_unity)
 362 +               return 0;
 363 +       else if (vc4_state->y_scaling == VC4_SCALING_TPZ)
 364 +               lbm = pix_per_line * 8;
 365 +       else {
 366 +               /* In special cases, this multiplier might be 12. */
 367 +               lbm = pix_per_line * 16;
 368 +       }
 369 +
 370 +       lbm = roundup(lbm, 32);
 371 +
 372 +       return lbm;
 373 +}
 374 +
 375 +static void vc4_write_scaling_parameters(struct drm_plane_state *state)
 376 +{
 377 +       struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
 378 +
 379 +       /* Ch0 H-PPF Word 0: Scaling Parameters */
 380 +       if (vc4_state->x_scaling == VC4_SCALING_PPF) {
 381 +               vc4_write_ppf(vc4_state,
 382 +                             vc4_state->src_w, vc4_state->crtc_w);
 383 +       }
 384 +
 385 +       /* Ch0 V-PPF Words 0-1: Scaling Parameters, Context */
 386 +       if (vc4_state->y_scaling == VC4_SCALING_PPF) {
 387 +               vc4_write_ppf(vc4_state,
 388 +                             vc4_state->src_h, vc4_state->crtc_h);
 389 +               vc4_dlist_write(vc4_state, 0xc0c0c0c0);
 390 +       }
 391 +
 392 +       /* Ch0 H-TPZ Words 0-1: Scaling Parameters, Recip */
 393 +       if (vc4_state->x_scaling == VC4_SCALING_TPZ) {
 394 +               vc4_write_tpz(vc4_state,
 395 +                             vc4_state->src_w, vc4_state->crtc_w);
 396 +       }
 397 +
 398 +       /* Ch0 V-TPZ Words 0-2: Scaling Parameters, Recip, Context */
 399 +       if (vc4_state->y_scaling == VC4_SCALING_TPZ) {
 400 +               vc4_write_tpz(vc4_state,
 401 +                             vc4_state->src_h, vc4_state->crtc_h);
 402 +               vc4_dlist_write(vc4_state, 0xc0c0c0c0);
 403 +       }
 404 +}
 405
 406  /* Writes out a full display list for an active plane to the plane's
 407   * private dlist state.
 408 @@ -230,22 +398,50 @@ static int vc4_plane_setup_clipping_and_
 409  static int vc4_plane_mode_set(struct drm_plane *plane,
 410                               struct drm_plane_state *state)
 411  {
 412 +       struct vc4_dev *vc4 = to_vc4_dev(plane->dev);
 413         struct vc4_plane_state *vc4_state = to_vc4_plane_state(state);
 414         struct drm_framebuffer *fb = state->fb;
 415         struct drm_gem_cma_object *bo = drm_fb_cma_get_gem_obj(fb, 0);
 416         u32 ctl0_offset = vc4_state->dlist_count;
 417         const struct hvs_format *format = vc4_get_hvs_format(fb->pixel_format);
 418 +       u32 scl;
 419 +       u32 lbm_size;
 420 +       unsigned long irqflags;
 421         int ret;
 422
 423         ret = vc4_plane_setup_clipping_and_scaling(state);
 424         if (ret)
 425                 return ret;
 426
 427 +       /* Allocate the LBM memory that the HVS will use for temporary
 428 +        * storage due to our scaling/format conversion.
 429 +        */
 430 +       lbm_size = vc4_lbm_size(state);
 431 +       if (lbm_size) {
 432 +               if (!vc4_state->lbm.allocated) {
 433 +                       spin_lock_irqsave(&vc4->hvs->mm_lock, irqflags);
 434 +                       ret = drm_mm_insert_node(&vc4->hvs->lbm_mm,
 435 +                                                &vc4_state->lbm,
 436 +                                                lbm_size, 32, 0);
 437 +                       spin_unlock_irqrestore(&vc4->hvs->mm_lock, irqflags);
 438 +               } else {
 439 +                       WARN_ON_ONCE(lbm_size != vc4_state->lbm.size);
 440 +               }
 441 +       }
 442 +
 443 +       if (ret)
 444 +               return ret;
 445 +
 446 +       scl = vc4_get_scl_field(state);
 447 +
 448 +       /* Control word */
 449         vc4_dlist_write(vc4_state,
 450                         SCALER_CTL0_VALID |
 451                         (format->pixel_order << SCALER_CTL0_ORDER_SHIFT) |
 452                         (format->hvs << SCALER_CTL0_PIXEL_FORMAT_SHIFT) |
 453 -                       SCALER_CTL0_UNITY);
 454 +                       (vc4_state->is_unity ? SCALER_CTL0_UNITY : 0) |
 455 +                       VC4_SET_FIELD(scl, SCALER_CTL0_SCL0) |
 456 +                       VC4_SET_FIELD(scl, SCALER_CTL0_SCL1));
 457
 458         /* Position Word 0: Image Positions and Alpha Value */
 459         vc4_state->pos0_offset = vc4_state->dlist_count;
 460 @@ -254,9 +450,14 @@ static int vc4_plane_mode_set(struct drm
 461                         VC4_SET_FIELD(vc4_state->crtc_x, SCALER_POS0_START_X) |
 462                         VC4_SET_FIELD(vc4_state->crtc_y, SCALER_POS0_START_Y));
 463
 464 -       /* Position Word 1: Scaled Image Dimensions.
 465 -        * Skipped due to SCALER_CTL0_UNITY scaling.
 466 -        */
 467 +       /* Position Word 1: Scaled Image Dimensions. */
 468 +       if (!vc4_state->is_unity) {
 469 +               vc4_dlist_write(vc4_state,
 470 +                               VC4_SET_FIELD(vc4_state->crtc_w,
 471 +                                             SCALER_POS1_SCL_WIDTH) |
 472 +                               VC4_SET_FIELD(vc4_state->crtc_h,
 473 +                                             SCALER_POS1_SCL_HEIGHT));
 474 +       }
 475
 476         /* Position Word 2: Source Image Size, Alpha Mode */
 477         vc4_state->pos2_offset = vc4_state->dlist_count;
 478 @@ -282,6 +483,32 @@ static int vc4_plane_mode_set(struct drm
 479         vc4_dlist_write(vc4_state,
 480                         VC4_SET_FIELD(fb->pitches[0], SCALER_SRC_PITCH));
 481
 482 +       if (!vc4_state->is_unity) {
 483 +               /* LBM Base Address. */
 484 +               if (vc4_state->y_scaling != VC4_SCALING_NONE)
 485 +                       vc4_dlist_write(vc4_state, vc4_state->lbm.start);
 486 +
 487 +               vc4_write_scaling_parameters(state);
 488 +
 489 +               /* If any PPF setup was done, then all the kernel
 490 +                * pointers get uploaded.
 491 +                */
 492 +               if (vc4_state->x_scaling == VC4_SCALING_PPF ||
 493 +                   vc4_state->y_scaling == VC4_SCALING_PPF) {
 494 +                       u32 kernel = VC4_SET_FIELD(vc4->hvs->mitchell_netravali_filter.start,
 495 +                                                  SCALER_PPF_KERNEL_OFFSET);
 496 +
 497 +                       /* HPPF plane 0 */
 498 +                       vc4_dlist_write(vc4_state, kernel);
 499 +                       /* VPPF plane 0 */
 500 +                       vc4_dlist_write(vc4_state, kernel);
 501 +                       /* HPPF plane 1 */
 502 +                       vc4_dlist_write(vc4_state, kernel);
 503 +                       /* VPPF plane 1 */
 504 +                       vc4_dlist_write(vc4_state, kernel);
 505 +               }
 506 +       }
 507 +
 508         vc4_state->dlist[ctl0_offset] |=
 509                 VC4_SET_FIELD(vc4_state->dlist_count, SCALER_CTL0_SIZE);
 510
 511 --- a/drivers/gpu/drm/vc4/vc4_regs.h
 512 +++ b/drivers/gpu/drm/vc4/vc4_regs.h
 513 @@ -536,6 +536,21 @@ enum hvs_pixel_format {
 514  #define SCALER_CTL0_ORDER_MASK                 VC4_MASK(14, 13)
 515  #define SCALER_CTL0_ORDER_SHIFT                        13
 516
 517 +#define SCALER_CTL0_SCL1_MASK                  VC4_MASK(10, 8)
 518 +#define SCALER_CTL0_SCL1_SHIFT                 8
 519 +
 520 +#define SCALER_CTL0_SCL0_MASK                  VC4_MASK(7, 5)
 521 +#define SCALER_CTL0_SCL0_SHIFT                 5
 522 +
 523 +#define SCALER_CTL0_SCL_H_PPF_V_PPF            0
 524 +#define SCALER_CTL0_SCL_H_TPZ_V_PPF            1
 525 +#define SCALER_CTL0_SCL_H_PPF_V_TPZ            2
 526 +#define SCALER_CTL0_SCL_H_TPZ_V_TPZ            3
 527 +#define SCALER_CTL0_SCL_H_PPF_V_NONE           4
 528 +#define SCALER_CTL0_SCL_H_NONE_V_PPF           5
 529 +#define SCALER_CTL0_SCL_H_NONE_V_TPZ           6
 530 +#define SCALER_CTL0_SCL_H_TPZ_V_NONE           7
 531 +
 532  /* Set to indicate no scaling. */
 533  #define SCALER_CTL0_UNITY                      BIT(4)
 534
 535 @@ -551,6 +566,12 @@ enum hvs_pixel_format {
 536  #define SCALER_POS0_START_X_MASK               VC4_MASK(11, 0)
 537  #define SCALER_POS0_START_X_SHIFT              0
 538
 539 +#define SCALER_POS1_SCL_HEIGHT_MASK            VC4_MASK(27, 16)
 540 +#define SCALER_POS1_SCL_HEIGHT_SHIFT           16
 541 +
 542 +#define SCALER_POS1_SCL_WIDTH_MASK             VC4_MASK(11, 0)
 543 +#define SCALER_POS1_SCL_WIDTH_SHIFT            0
 544 +
 545  #define SCALER_POS2_ALPHA_MODE_MASK            VC4_MASK(31, 30)
 546  #define SCALER_POS2_ALPHA_MODE_SHIFT           30
 547  #define SCALER_POS2_ALPHA_MODE_PIPELINE                0
 548 @@ -564,6 +585,31 @@ enum hvs_pixel_format {
 549  #define SCALER_POS2_WIDTH_MASK                 VC4_MASK(11, 0)
 550  #define SCALER_POS2_WIDTH_SHIFT                        0
 551
 552 +#define SCALER_TPZ0_VERT_RECALC                        BIT(31)
 553 +#define SCALER_TPZ0_SCALE_MASK                 VC4_MASK(28, 8)
 554 +#define SCALER_TPZ0_SCALE_SHIFT                        8
 555 +#define SCALER_TPZ0_IPHASE_MASK                        VC4_MASK(7, 0)
 556 +#define SCALER_TPZ0_IPHASE_SHIFT               0
 557 +#define SCALER_TPZ1_RECIP_MASK                 VC4_MASK(15, 0)
 558 +#define SCALER_TPZ1_RECIP_SHIFT                        0
 559 +
 560 +/* Skips interpolating coefficients to 64 phases, so just 8 are used.
 561 + * Required for nearest neighbor.
 562 + */
 563 +#define SCALER_PPF_NOINTERP                    BIT(31)
 564 +/* Replaes the highest valued coefficient with one that makes all 4
 565 + * sum to unity.
 566 + */
 567 +#define SCALER_PPF_AGC                         BIT(30)
 568 +#define SCALER_PPF_SCALE_MASK                  VC4_MASK(24, 8)
 569 +#define SCALER_PPF_SCALE_SHIFT                 8
 570 +#define SCALER_PPF_IPHASE_MASK                 VC4_MASK(6, 0)
 571 +#define SCALER_PPF_IPHASE_SHIFT                        0
 572 +
 573 +#define SCALER_PPF_KERNEL_OFFSET_MASK          VC4_MASK(13, 0)
 574 +#define SCALER_PPF_KERNEL_OFFSET_SHIFT         0
 575 +#define SCALER_PPF_KERNEL_UNCACHED             BIT(31)
 576 +
 577  #define SCALER_SRC_PITCH_MASK                  VC4_MASK(15, 0)
 578  #define SCALER_SRC_PITCH_SHIFT                 0
 579