drivers/gpu/drm/vkms/vkms_composer.c

   1 // SPDX-License-Identifier: GPL-2.0+
   2
   3 #include <linux/crc32.h>
   4
   5 #include <drm/drm_atomic.h>
   6 #include <drm/drm_atomic_helper.h>
   7 #include <drm/drm_fourcc.h>
   8 #include <drm/drm_gem_framebuffer_helper.h>
   9 #include <drm/drm_gem_shmem_helper.h>
  10 #include <drm/drm_vblank.h>
  11
  12 #include "vkms_drv.h"
  13
  14 static u32 get_pixel_from_buffer(int x, int y, const u8 *buffer,
  15                                  const struct vkms_composer *composer)
  16 {
  17         u32 pixel;
  18         int src_offset = composer->offset + (y * composer->pitch)
  19                                       + (x * composer->cpp);
  20
  21         pixel = *(u32 *)&buffer[src_offset];
  22
  23         return pixel;
  24 }
  25
  26 /**
  27  * compute_crc - Compute CRC value on output frame
  28  *
  29  * @vaddr: address to final framebuffer
  30  * @composer: framebuffer's metadata
  31  *
  32  * returns CRC value computed using crc32 on the visible portion of
  33  * the final framebuffer at vaddr_out
  34  */
  35 static uint32_t compute_crc(const u8 *vaddr,
  36                             const struct vkms_composer *composer)
  37 {
  38         int x, y;
  39         u32 crc = 0, pixel = 0;
  40         int x_src = composer->src.x1 >> 16;
  41         int y_src = composer->src.y1 >> 16;
  42         int h_src = drm_rect_height(&composer->src) >> 16;
  43         int w_src = drm_rect_width(&composer->src) >> 16;
  44
  45         for (y = y_src; y < y_src + h_src; ++y) {
  46                 for (x = x_src; x < x_src + w_src; ++x) {
  47                         pixel = get_pixel_from_buffer(x, y, vaddr, composer);
  48                         crc = crc32_le(crc, (void *)&pixel, sizeof(u32));
  49                 }
  50         }
  51
  52         return crc;
  53 }
  54
  55 static u8 blend_channel(u8 src, u8 dst, u8 alpha)
  56 {
  57         u32 pre_blend;
  58         u8 new_color;
  59
  60         pre_blend = (src * 255 + dst * (255 - alpha));
  61
  62         /* Faster div by 255 */
  63         new_color = ((pre_blend + ((pre_blend + 257) >> 8)) >> 8);
  64
  65         return new_color;
  66 }
  67
  68 /**
  69  * alpha_blend - alpha blending equation
  70  * @argb_src: src pixel on premultiplied alpha mode
  71  * @argb_dst: dst pixel completely opaque
  72  *
  73  * blend pixels using premultiplied blend formula. The current DRM assumption
  74  * is that pixel color values have been already pre-multiplied with the alpha
  75  * channel values. See more drm_plane_create_blend_mode_property(). Also, this
  76  * formula assumes a completely opaque background.
  77  */
  78 static void alpha_blend(const u8 *argb_src, u8 *argb_dst)
  79 {
  80         u8 alpha;
  81
  82         alpha = argb_src[3];
  83         argb_dst[0] = blend_channel(argb_src[0], argb_dst[0], alpha);
  84         argb_dst[1] = blend_channel(argb_src[1], argb_dst[1], alpha);
  85         argb_dst[2] = blend_channel(argb_src[2], argb_dst[2], alpha);
  86 }
  87
  88 /**
  89  * x_blend - blending equation that ignores the pixel alpha
  90  *
  91  * overwrites RGB color value from src pixel to dst pixel.
  92  */
  93 static void x_blend(const u8 *xrgb_src, u8 *xrgb_dst)
  94 {
  95         memcpy(xrgb_dst, xrgb_src, sizeof(u8) * 3);
  96 }
  97
  98 /**
  99  * blend - blend value at vaddr_src with value at vaddr_dst
 100  * @vaddr_dst: destination address
 101  * @vaddr_src: source address
 102  * @dst_composer: destination framebuffer's metadata
 103  * @src_composer: source framebuffer's metadata
 104  * @pixel_blend: blending equation based on plane format
 105  *
 106  * Blend the vaddr_src value with the vaddr_dst value using a pixel blend
 107  * equation according to the supported plane formats DRM_FORMAT_(A/XRGB8888)
 108  * and clearing alpha channel to an completely opaque background. This function
 109  * uses buffer's metadata to locate the new composite values at vaddr_dst.
 110  *
 111  * TODO: completely clear the primary plane (a = 0xff) before starting to blend
 112  * pixel color values
 113  */
 114 static void blend(void *vaddr_dst, void *vaddr_src,
 115                   struct vkms_composer *dst_composer,
 116                   struct vkms_composer *src_composer,
 117                   void (*pixel_blend)(const u8 *, u8 *))
 118 {
 119         int i, j, j_dst, i_dst;
 120         int offset_src, offset_dst;
 121         u8 *pixel_dst, *pixel_src;
 122
 123         int x_src = src_composer->src.x1 >> 16;
 124         int y_src = src_composer->src.y1 >> 16;
 125
 126         int x_dst = src_composer->dst.x1;
 127         int y_dst = src_composer->dst.y1;
 128         int h_dst = drm_rect_height(&src_composer->dst);
 129         int w_dst = drm_rect_width(&src_composer->dst);
 130
 131         int y_limit = y_src + h_dst;
 132         int x_limit = x_src + w_dst;
 133
 134         for (i = y_src, i_dst = y_dst; i < y_limit; ++i) {
 135                 for (j = x_src, j_dst = x_dst; j < x_limit; ++j) {
 136                         offset_dst = dst_composer->offset
 137                                      + (i_dst * dst_composer->pitch)
 138                                      + (j_dst++ * dst_composer->cpp);
 139                         offset_src = src_composer->offset
 140                                      + (i * src_composer->pitch)
 141                                      + (j * src_composer->cpp);
 142
 143                         pixel_src = (u8 *)(vaddr_src + offset_src);
 144                         pixel_dst = (u8 *)(vaddr_dst + offset_dst);
 145                         pixel_blend(pixel_src, pixel_dst);
 146                         /* clearing alpha channel (0xff)*/
 147                         pixel_dst[3] = 0xff;
 148                 }
 149                 i_dst++;
 150         }
 151 }
 152
 153 static void compose_plane(struct vkms_composer *primary_composer,
 154                           struct vkms_composer *plane_composer,
 155                           void *vaddr_out)
 156 {
 157         struct drm_gem_object *plane_obj;
 158         struct drm_gem_shmem_object *plane_shmem_obj;
 159         struct drm_framebuffer *fb = &plane_composer->fb;
 160         void (*pixel_blend)(const u8 *p_src, u8 *p_dst);
 161
 162         plane_obj = drm_gem_fb_get_obj(&plane_composer->fb, 0);
 163         plane_shmem_obj = to_drm_gem_shmem_obj(plane_obj);
 164
 165         if (WARN_ON(!plane_shmem_obj->vaddr))
 166                 return;
 167
 168         if (fb->format->format == DRM_FORMAT_ARGB8888)
 169                 pixel_blend = &alpha_blend;
 170         else
 171                 pixel_blend = &x_blend;
 172
 173         blend(vaddr_out, plane_shmem_obj->vaddr, primary_composer,
 174               plane_composer, pixel_blend);
 175 }
 176
 177 static int compose_active_planes(void **vaddr_out,
 178                                  struct vkms_composer *primary_composer,
 179                                  struct vkms_crtc_state *crtc_state)
 180 {
 181         struct drm_framebuffer *fb = &primary_composer->fb;
 182         struct drm_gem_object *gem_obj = drm_gem_fb_get_obj(fb, 0);
 183         struct drm_gem_shmem_object *shmem_obj = to_drm_gem_shmem_obj(gem_obj);
 184         int i;
 185
 186         if (!*vaddr_out) {
 187                 *vaddr_out = kzalloc(shmem_obj->base.size, GFP_KERNEL);
 188                 if (!*vaddr_out) {
 189                         DRM_ERROR("Cannot allocate memory for output frame.");
 190                         return -ENOMEM;
 191                 }
 192         }
 193
 194         if (WARN_ON(!shmem_obj->vaddr))
 195                 return -EINVAL;
 196
 197         memcpy(*vaddr_out, shmem_obj->vaddr, shmem_obj->base.size);
 198
 199         /* If there are other planes besides primary, we consider the active
 200          * planes should be in z-order and compose them associatively:
 201          * ((primary <- overlay) <- cursor)
 202          */
 203         for (i = 1; i < crtc_state->num_active_planes; i++)
 204                 compose_plane(primary_composer,
 205                               crtc_state->active_planes[i]->composer,
 206                               *vaddr_out);
 207
 208         return 0;
 209 }
 210
 211 /**
 212  * vkms_composer_worker - ordered work_struct to compute CRC
 213  *
 214  * @work: work_struct
 215  *
 216  * Work handler for composing and computing CRCs. work_struct scheduled in
 217  * an ordered workqueue that's periodically scheduled to run by
 218  * _vblank_handle() and flushed at vkms_atomic_crtc_destroy_state().
 219  */
 220 void vkms_composer_worker(struct work_struct *work)
 221 {
 222         struct vkms_crtc_state *crtc_state = container_of(work,
 223                                                 struct vkms_crtc_state,
 224                                                 composer_work);
 225         struct drm_crtc *crtc = crtc_state->base.crtc;
 226         struct vkms_output *out = drm_crtc_to_vkms_output(crtc);
 227         struct vkms_composer *primary_composer = NULL;
 228         struct vkms_plane_state *act_plane = NULL;
 229         bool crc_pending, wb_pending;
 230         void *vaddr_out = NULL;
 231         u32 crc32 = 0;
 232         u64 frame_start, frame_end;
 233         int ret;
 234
 235         spin_lock_irq(&out->composer_lock);
 236         frame_start = crtc_state->frame_start;
 237         frame_end = crtc_state->frame_end;
 238         crc_pending = crtc_state->crc_pending;
 239         wb_pending = crtc_state->wb_pending;
 240         crtc_state->frame_start = 0;
 241         crtc_state->frame_end = 0;
 242         crtc_state->crc_pending = false;
 243         spin_unlock_irq(&out->composer_lock);
 244
 245         /*
 246          * We raced with the vblank hrtimer and previous work already computed
 247          * the crc, nothing to do.
 248          */
 249         if (!crc_pending)
 250                 return;
 251
 252         if (crtc_state->num_active_planes >= 1) {
 253                 act_plane = crtc_state->active_planes[0];
 254                 if (act_plane->base.plane->type == DRM_PLANE_TYPE_PRIMARY)
 255                         primary_composer = act_plane->composer;
 256         }
 257
 258         if (!primary_composer)
 259                 return;
 260
 261         if (wb_pending)
 262                 vaddr_out = crtc_state->active_writeback;
 263
 264         ret = compose_active_planes(&vaddr_out, primary_composer,
 265                                     crtc_state);
 266         if (ret) {
 267                 if (ret == -EINVAL && !wb_pending)
 268                         kfree(vaddr_out);
 269                 return;
 270         }
 271
 272         crc32 = compute_crc(vaddr_out, primary_composer);
 273
 274         if (wb_pending) {
 275                 drm_writeback_signal_completion(&out->wb_connector, 0);
 276                 spin_lock_irq(&out->composer_lock);
 277                 crtc_state->wb_pending = false;
 278                 spin_unlock_irq(&out->composer_lock);
 279         } else {
 280                 kfree(vaddr_out);
 281         }
 282
 283         /*
 284          * The worker can fall behind the vblank hrtimer, make sure we catch up.
 285          */
 286         while (frame_start <= frame_end)
 287                 drm_crtc_add_crc_entry(crtc, true, frame_start++, &crc32);
 288 }
 289
 290 static const char * const pipe_crc_sources[] = {"auto"};
 291
 292 const char *const *vkms_get_crc_sources(struct drm_crtc *crtc,
 293                                         size_t *count)
 294 {
 295         *count = ARRAY_SIZE(pipe_crc_sources);
 296         return pipe_crc_sources;
 297 }
 298
 299 static int vkms_crc_parse_source(const char *src_name, bool *enabled)
 300 {
 301         int ret = 0;
 302
 303         if (!src_name) {
 304                 *enabled = false;
 305         } else if (strcmp(src_name, "auto") == 0) {
 306                 *enabled = true;
 307         } else {
 308                 *enabled = false;
 309                 ret = -EINVAL;
 310         }
 311
 312         return ret;
 313 }
 314
 315 int vkms_verify_crc_source(struct drm_crtc *crtc, const char *src_name,
 316                            size_t *values_cnt)
 317 {
 318         bool enabled;
 319
 320         if (vkms_crc_parse_source(src_name, &enabled) < 0) {
 321                 DRM_DEBUG_DRIVER("unknown source %s\n", src_name);
 322                 return -EINVAL;
 323         }
 324
 325         *values_cnt = 1;
 326
 327         return 0;
 328 }
 329
 330 void vkms_set_composer(struct vkms_output *out, bool enabled)
 331 {
 332         bool old_enabled;
 333
 334         if (enabled)
 335                 drm_crtc_vblank_get(&out->crtc);
 336
 337         spin_lock_irq(&out->lock);
 338         old_enabled = out->composer_enabled;
 339         out->composer_enabled = enabled;
 340         spin_unlock_irq(&out->lock);
 341
 342         if (old_enabled)
 343                 drm_crtc_vblank_put(&out->crtc);
 344 }
 345
 346 int vkms_set_crc_source(struct drm_crtc *crtc, const char *src_name)
 347 {
 348         struct vkms_output *out = drm_crtc_to_vkms_output(crtc);
 349         bool enabled = false;
 350         int ret = 0;
 351
 352         ret = vkms_crc_parse_source(src_name, &enabled);
 353
 354         vkms_set_composer(out, enabled);
 355
 356         return ret;
 357 }