1 // SPDX-License-Identifier: MIT 2 /* 3 * Copyright © 2020 Intel Corporation 4 */ 5 #include "intel_de.h" 6 #include "intel_display_types.h" 7 #include "intel_fb.h" 8 #include "skl_scaler.h" 9 #include "skl_universal_plane.h" 10 11 /* 12 * The hardware phase 0.0 refers to the center of the pixel. 13 * We want to start from the top/left edge which is phase 14 * -0.5. That matches how the hardware calculates the scaling 15 * factors (from top-left of the first pixel to bottom-right 16 * of the last pixel, as opposed to the pixel centers). 17 * 18 * For 4:2:0 subsampled chroma planes we obviously have to 19 * adjust that so that the chroma sample position lands in 20 * the right spot. 21 * 22 * Note that for packed YCbCr 4:2:2 formats there is no way to 23 * control chroma siting. The hardware simply replicates the 24 * chroma samples for both of the luma samples, and thus we don't 25 * actually get the expected MPEG2 chroma siting convention :( 26 * The same behaviour is observed on pre-SKL platforms as well. 27 * 28 * Theory behind the formula (note that we ignore sub-pixel 29 * source coordinates): 30 * s = source sample position 31 * d = destination sample position 32 * 33 * Downscaling 4:1: 34 * -0.5 35 * | 0.0 36 * | | 1.5 (initial phase) 37 * | | | 38 * v v v 39 * | s | s | s | s | 40 * | d | 41 * 42 * Upscaling 1:4: 43 * -0.5 44 * | -0.375 (initial phase) 45 * | | 0.0 46 * | | | 47 * v v v 48 * | s | 49 * | d | d | d | d | 50 */ 51 static u16 skl_scaler_calc_phase(int sub, int scale, bool chroma_cosited) 52 { 53 int phase = -0x8000; 54 u16 trip = 0; 55 56 if (chroma_cosited) 57 phase += (sub - 1) * 0x8000 / sub; 58 59 phase += scale / (2 * sub); 60 61 /* 62 * Hardware initial phase limited to [-0.5:1.5]. 63 * Since the max hardware scale factor is 3.0, we 64 * should never actually excdeed 1.0 here. 65 */ 66 WARN_ON(phase < -0x8000 || phase > 0x18000); 67 68 if (phase < 0) 69 phase = 0x10000 + phase; 70 else 71 trip = PS_PHASE_TRIP; 72 73 return ((phase >> 2) & PS_PHASE_MASK) | trip; 74 } 75 76 #define SKL_MIN_SRC_W 8 77 #define SKL_MAX_SRC_W 4096 78 #define SKL_MIN_SRC_H 8 79 #define SKL_MAX_SRC_H 4096 80 #define SKL_MIN_DST_W 8 81 #define SKL_MAX_DST_W 4096 82 #define SKL_MIN_DST_H 8 83 #define SKL_MAX_DST_H 4096 84 #define ICL_MAX_SRC_W 5120 85 #define ICL_MAX_SRC_H 4096 86 #define ICL_MAX_DST_W 5120 87 #define ICL_MAX_DST_H 4096 88 #define SKL_MIN_YUV_420_SRC_W 16 89 #define SKL_MIN_YUV_420_SRC_H 16 90 91 static int 92 skl_update_scaler(struct intel_crtc_state *crtc_state, bool force_detach, 93 unsigned int scaler_user, int *scaler_id, 94 int src_w, int src_h, int dst_w, int dst_h, 95 const struct drm_format_info *format, 96 u64 modifier, bool need_scaler) 97 { 98 struct intel_crtc_scaler_state *scaler_state = 99 &crtc_state->scaler_state; 100 struct intel_crtc *crtc = to_intel_crtc(crtc_state->uapi.crtc); 101 struct drm_i915_private *dev_priv = to_i915(crtc->base.dev); 102 const struct drm_display_mode *adjusted_mode = 103 &crtc_state->hw.adjusted_mode; 104 105 /* 106 * Src coordinates are already rotated by 270 degrees for 107 * the 90/270 degree plane rotation cases (to match the 108 * GTT mapping), hence no need to account for rotation here. 109 */ 110 if (src_w != dst_w || src_h != dst_h) 111 need_scaler = true; 112 113 /* 114 * Scaling/fitting not supported in IF-ID mode in GEN9+ 115 * TODO: Interlace fetch mode doesn't support YUV420 planar formats. 116 * Once NV12 is enabled, handle it here while allocating scaler 117 * for NV12. 118 */ 119 if (DISPLAY_VER(dev_priv) >= 9 && crtc_state->hw.enable && 120 need_scaler && adjusted_mode->flags & DRM_MODE_FLAG_INTERLACE) { 121 drm_dbg_kms(&dev_priv->drm, 122 "Pipe/Plane scaling not supported with IF-ID mode\n"); 123 return -EINVAL; 124 } 125 126 /* 127 * if plane is being disabled or scaler is no more required or force detach 128 * - free scaler binded to this plane/crtc 129 * - in order to do this, update crtc->scaler_usage 130 * 131 * Here scaler state in crtc_state is set free so that 132 * scaler can be assigned to other user. Actual register 133 * update to free the scaler is done in plane/panel-fit programming. 134 * For this purpose crtc/plane_state->scaler_id isn't reset here. 135 */ 136 if (force_detach || !need_scaler) { 137 if (*scaler_id >= 0) { 138 scaler_state->scaler_users &= ~(1 << scaler_user); 139 scaler_state->scalers[*scaler_id].in_use = 0; 140 141 drm_dbg_kms(&dev_priv->drm, 142 "scaler_user index %u.%u: " 143 "Staged freeing scaler id %d scaler_users = 0x%x\n", 144 crtc->pipe, scaler_user, *scaler_id, 145 scaler_state->scaler_users); 146 *scaler_id = -1; 147 } 148 return 0; 149 } 150 151 if (format && intel_format_info_is_yuv_semiplanar(format, modifier) && 152 (src_h < SKL_MIN_YUV_420_SRC_H || src_w < SKL_MIN_YUV_420_SRC_W)) { 153 drm_dbg_kms(&dev_priv->drm, 154 "Planar YUV: src dimensions not met\n"); 155 return -EINVAL; 156 } 157 158 /* range checks */ 159 if (src_w < SKL_MIN_SRC_W || src_h < SKL_MIN_SRC_H || 160 dst_w < SKL_MIN_DST_W || dst_h < SKL_MIN_DST_H || 161 (DISPLAY_VER(dev_priv) >= 11 && 162 (src_w > ICL_MAX_SRC_W || src_h > ICL_MAX_SRC_H || 163 dst_w > ICL_MAX_DST_W || dst_h > ICL_MAX_DST_H)) || 164 (DISPLAY_VER(dev_priv) < 11 && 165 (src_w > SKL_MAX_SRC_W || src_h > SKL_MAX_SRC_H || 166 dst_w > SKL_MAX_DST_W || dst_h > SKL_MAX_DST_H))) { 167 drm_dbg_kms(&dev_priv->drm, 168 "scaler_user index %u.%u: src %ux%u dst %ux%u " 169 "size is out of scaler range\n", 170 crtc->pipe, scaler_user, src_w, src_h, 171 dst_w, dst_h); 172 return -EINVAL; 173 } 174 175 /* mark this plane as a scaler user in crtc_state */ 176 scaler_state->scaler_users |= (1 << scaler_user); 177 drm_dbg_kms(&dev_priv->drm, "scaler_user index %u.%u: " 178 "staged scaling request for %ux%u->%ux%u scaler_users = 0x%x\n", 179 crtc->pipe, scaler_user, src_w, src_h, dst_w, dst_h, 180 scaler_state->scaler_users); 181 182 return 0; 183 } 184 185 int skl_update_scaler_crtc(struct intel_crtc_state *crtc_state) 186 { 187 const struct drm_display_mode *pipe_mode = &crtc_state->hw.pipe_mode; 188 int width, height; 189 190 if (crtc_state->pch_pfit.enabled) { 191 width = drm_rect_width(&crtc_state->pch_pfit.dst); 192 height = drm_rect_height(&crtc_state->pch_pfit.dst); 193 } else { 194 width = pipe_mode->crtc_hdisplay; 195 height = pipe_mode->crtc_vdisplay; 196 } 197 return skl_update_scaler(crtc_state, !crtc_state->hw.active, 198 SKL_CRTC_INDEX, 199 &crtc_state->scaler_state.scaler_id, 200 crtc_state->pipe_src_w, crtc_state->pipe_src_h, 201 width, height, NULL, 0, 202 crtc_state->pch_pfit.enabled); 203 } 204 205 /** 206 * skl_update_scaler_plane - Stages update to scaler state for a given plane. 207 * @crtc_state: crtc's scaler state 208 * @plane_state: atomic plane state to update 209 * 210 * Return 211 * 0 - scaler_usage updated successfully 212 * error - requested scaling cannot be supported or other error condition 213 */ 214 int skl_update_scaler_plane(struct intel_crtc_state *crtc_state, 215 struct intel_plane_state *plane_state) 216 { 217 struct intel_plane *intel_plane = 218 to_intel_plane(plane_state->uapi.plane); 219 struct drm_i915_private *dev_priv = to_i915(intel_plane->base.dev); 220 struct drm_framebuffer *fb = plane_state->hw.fb; 221 int ret; 222 bool force_detach = !fb || !plane_state->uapi.visible; 223 bool need_scaler = false; 224 225 /* Pre-gen11 and SDR planes always need a scaler for planar formats. */ 226 if (!icl_is_hdr_plane(dev_priv, intel_plane->id) && 227 fb && intel_format_info_is_yuv_semiplanar(fb->format, fb->modifier)) 228 need_scaler = true; 229 230 ret = skl_update_scaler(crtc_state, force_detach, 231 drm_plane_index(&intel_plane->base), 232 &plane_state->scaler_id, 233 drm_rect_width(&plane_state->uapi.src) >> 16, 234 drm_rect_height(&plane_state->uapi.src) >> 16, 235 drm_rect_width(&plane_state->uapi.dst), 236 drm_rect_height(&plane_state->uapi.dst), 237 fb ? fb->format : NULL, 238 fb ? fb->modifier : 0, 239 need_scaler); 240 241 if (ret || plane_state->scaler_id < 0) 242 return ret; 243 244 /* check colorkey */ 245 if (plane_state->ckey.flags) { 246 drm_dbg_kms(&dev_priv->drm, 247 "[PLANE:%d:%s] scaling with color key not allowed", 248 intel_plane->base.base.id, 249 intel_plane->base.name); 250 return -EINVAL; 251 } 252 253 /* Check src format */ 254 switch (fb->format->format) { 255 case DRM_FORMAT_RGB565: 256 case DRM_FORMAT_XBGR8888: 257 case DRM_FORMAT_XRGB8888: 258 case DRM_FORMAT_ABGR8888: 259 case DRM_FORMAT_ARGB8888: 260 case DRM_FORMAT_XRGB2101010: 261 case DRM_FORMAT_XBGR2101010: 262 case DRM_FORMAT_ARGB2101010: 263 case DRM_FORMAT_ABGR2101010: 264 case DRM_FORMAT_YUYV: 265 case DRM_FORMAT_YVYU: 266 case DRM_FORMAT_UYVY: 267 case DRM_FORMAT_VYUY: 268 case DRM_FORMAT_NV12: 269 case DRM_FORMAT_XYUV8888: 270 case DRM_FORMAT_P010: 271 case DRM_FORMAT_P012: 272 case DRM_FORMAT_P016: 273 case DRM_FORMAT_Y210: 274 case DRM_FORMAT_Y212: 275 case DRM_FORMAT_Y216: 276 case DRM_FORMAT_XVYU2101010: 277 case DRM_FORMAT_XVYU12_16161616: 278 case DRM_FORMAT_XVYU16161616: 279 break; 280 case DRM_FORMAT_XBGR16161616F: 281 case DRM_FORMAT_ABGR16161616F: 282 case DRM_FORMAT_XRGB16161616F: 283 case DRM_FORMAT_ARGB16161616F: 284 if (DISPLAY_VER(dev_priv) >= 11) 285 break; 286 fallthrough; 287 default: 288 drm_dbg_kms(&dev_priv->drm, 289 "[PLANE:%d:%s] FB:%d unsupported scaling format 0x%x\n", 290 intel_plane->base.base.id, intel_plane->base.name, 291 fb->base.id, fb->format->format); 292 return -EINVAL; 293 } 294 295 return 0; 296 } 297 298 static int glk_coef_tap(int i) 299 { 300 return i % 7; 301 } 302 303 static u16 glk_nearest_filter_coef(int t) 304 { 305 return t == 3 ? 0x0800 : 0x3000; 306 } 307 308 /* 309 * Theory behind setting nearest-neighbor integer scaling: 310 * 311 * 17 phase of 7 taps requires 119 coefficients in 60 dwords per set. 312 * The letter represents the filter tap (D is the center tap) and the number 313 * represents the coefficient set for a phase (0-16). 314 * 315 * +------------+------------------------+------------------------+ 316 * |Index value | Data value coeffient 1 | Data value coeffient 2 | 317 * +------------+------------------------+------------------------+ 318 * | 00h | B0 | A0 | 319 * +------------+------------------------+------------------------+ 320 * | 01h | D0 | C0 | 321 * +------------+------------------------+------------------------+ 322 * | 02h | F0 | E0 | 323 * +------------+------------------------+------------------------+ 324 * | 03h | A1 | G0 | 325 * +------------+------------------------+------------------------+ 326 * | 04h | C1 | B1 | 327 * +------------+------------------------+------------------------+ 328 * | ... | ... | ... | 329 * +------------+------------------------+------------------------+ 330 * | 38h | B16 | A16 | 331 * +------------+------------------------+------------------------+ 332 * | 39h | D16 | C16 | 333 * +------------+------------------------+------------------------+ 334 * | 3Ah | F16 | C16 | 335 * +------------+------------------------+------------------------+ 336 * | 3Bh | Reserved | G16 | 337 * +------------+------------------------+------------------------+ 338 * 339 * To enable nearest-neighbor scaling: program scaler coefficents with 340 * the center tap (Dxx) values set to 1 and all other values set to 0 as per 341 * SCALER_COEFFICIENT_FORMAT 342 * 343 */ 344 345 static void glk_program_nearest_filter_coefs(struct drm_i915_private *dev_priv, 346 enum pipe pipe, int id, int set) 347 { 348 int i; 349 350 intel_de_write_fw(dev_priv, GLK_PS_COEF_INDEX_SET(pipe, id, set), 351 PS_COEE_INDEX_AUTO_INC); 352 353 for (i = 0; i < 17 * 7; i += 2) { 354 u32 tmp; 355 int t; 356 357 t = glk_coef_tap(i); 358 tmp = glk_nearest_filter_coef(t); 359 360 t = glk_coef_tap(i + 1); 361 tmp |= glk_nearest_filter_coef(t) << 16; 362 363 intel_de_write_fw(dev_priv, GLK_PS_COEF_DATA_SET(pipe, id, set), 364 tmp); 365 } 366 367 intel_de_write_fw(dev_priv, GLK_PS_COEF_INDEX_SET(pipe, id, set), 0); 368 } 369 370 static u32 skl_scaler_get_filter_select(enum drm_scaling_filter filter, int set) 371 { 372 if (filter == DRM_SCALING_FILTER_NEAREST_NEIGHBOR) { 373 return (PS_FILTER_PROGRAMMED | 374 PS_Y_VERT_FILTER_SELECT(set) | 375 PS_Y_HORZ_FILTER_SELECT(set) | 376 PS_UV_VERT_FILTER_SELECT(set) | 377 PS_UV_HORZ_FILTER_SELECT(set)); 378 } 379 380 return PS_FILTER_MEDIUM; 381 } 382 383 static void skl_scaler_setup_filter(struct drm_i915_private *dev_priv, enum pipe pipe, 384 int id, int set, enum drm_scaling_filter filter) 385 { 386 switch (filter) { 387 case DRM_SCALING_FILTER_DEFAULT: 388 break; 389 case DRM_SCALING_FILTER_NEAREST_NEIGHBOR: 390 glk_program_nearest_filter_coefs(dev_priv, pipe, id, set); 391 break; 392 default: 393 MISSING_CASE(filter); 394 } 395 } 396 397 void skl_pfit_enable(const struct intel_crtc_state *crtc_state) 398 { 399 struct intel_crtc *crtc = to_intel_crtc(crtc_state->uapi.crtc); 400 struct drm_i915_private *dev_priv = to_i915(crtc->base.dev); 401 const struct intel_crtc_scaler_state *scaler_state = 402 &crtc_state->scaler_state; 403 struct drm_rect src = { 404 .x2 = crtc_state->pipe_src_w << 16, 405 .y2 = crtc_state->pipe_src_h << 16, 406 }; 407 const struct drm_rect *dst = &crtc_state->pch_pfit.dst; 408 u16 uv_rgb_hphase, uv_rgb_vphase; 409 enum pipe pipe = crtc->pipe; 410 int width = drm_rect_width(dst); 411 int height = drm_rect_height(dst); 412 int x = dst->x1; 413 int y = dst->y1; 414 int hscale, vscale; 415 unsigned long irqflags; 416 int id; 417 u32 ps_ctrl; 418 419 if (!crtc_state->pch_pfit.enabled) 420 return; 421 422 if (drm_WARN_ON(&dev_priv->drm, 423 crtc_state->scaler_state.scaler_id < 0)) 424 return; 425 426 hscale = drm_rect_calc_hscale(&src, dst, 0, INT_MAX); 427 vscale = drm_rect_calc_vscale(&src, dst, 0, INT_MAX); 428 429 uv_rgb_hphase = skl_scaler_calc_phase(1, hscale, false); 430 uv_rgb_vphase = skl_scaler_calc_phase(1, vscale, false); 431 432 id = scaler_state->scaler_id; 433 434 ps_ctrl = skl_scaler_get_filter_select(crtc_state->hw.scaling_filter, 0); 435 ps_ctrl |= PS_SCALER_EN | scaler_state->scalers[id].mode; 436 437 spin_lock_irqsave(&dev_priv->uncore.lock, irqflags); 438 439 skl_scaler_setup_filter(dev_priv, pipe, id, 0, 440 crtc_state->hw.scaling_filter); 441 442 intel_de_write_fw(dev_priv, SKL_PS_CTRL(pipe, id), ps_ctrl); 443 444 intel_de_write_fw(dev_priv, SKL_PS_VPHASE(pipe, id), 445 PS_Y_PHASE(0) | PS_UV_RGB_PHASE(uv_rgb_vphase)); 446 intel_de_write_fw(dev_priv, SKL_PS_HPHASE(pipe, id), 447 PS_Y_PHASE(0) | PS_UV_RGB_PHASE(uv_rgb_hphase)); 448 intel_de_write_fw(dev_priv, SKL_PS_WIN_POS(pipe, id), 449 x << 16 | y); 450 intel_de_write_fw(dev_priv, SKL_PS_WIN_SZ(pipe, id), 451 width << 16 | height); 452 453 spin_unlock_irqrestore(&dev_priv->uncore.lock, irqflags); 454 } 455 456 void 457 skl_program_plane_scaler(struct intel_plane *plane, 458 const struct intel_crtc_state *crtc_state, 459 const struct intel_plane_state *plane_state) 460 { 461 struct drm_i915_private *dev_priv = to_i915(plane->base.dev); 462 const struct drm_framebuffer *fb = plane_state->hw.fb; 463 enum pipe pipe = plane->pipe; 464 int scaler_id = plane_state->scaler_id; 465 const struct intel_scaler *scaler = 466 &crtc_state->scaler_state.scalers[scaler_id]; 467 int crtc_x = plane_state->uapi.dst.x1; 468 int crtc_y = plane_state->uapi.dst.y1; 469 u32 crtc_w = drm_rect_width(&plane_state->uapi.dst); 470 u32 crtc_h = drm_rect_height(&plane_state->uapi.dst); 471 u16 y_hphase, uv_rgb_hphase; 472 u16 y_vphase, uv_rgb_vphase; 473 int hscale, vscale; 474 u32 ps_ctrl; 475 476 hscale = drm_rect_calc_hscale(&plane_state->uapi.src, 477 &plane_state->uapi.dst, 478 0, INT_MAX); 479 vscale = drm_rect_calc_vscale(&plane_state->uapi.src, 480 &plane_state->uapi.dst, 481 0, INT_MAX); 482 483 /* TODO: handle sub-pixel coordinates */ 484 if (intel_format_info_is_yuv_semiplanar(fb->format, fb->modifier) && 485 !icl_is_hdr_plane(dev_priv, plane->id)) { 486 y_hphase = skl_scaler_calc_phase(1, hscale, false); 487 y_vphase = skl_scaler_calc_phase(1, vscale, false); 488 489 /* MPEG2 chroma siting convention */ 490 uv_rgb_hphase = skl_scaler_calc_phase(2, hscale, true); 491 uv_rgb_vphase = skl_scaler_calc_phase(2, vscale, false); 492 } else { 493 /* not used */ 494 y_hphase = 0; 495 y_vphase = 0; 496 497 uv_rgb_hphase = skl_scaler_calc_phase(1, hscale, false); 498 uv_rgb_vphase = skl_scaler_calc_phase(1, vscale, false); 499 } 500 501 ps_ctrl = skl_scaler_get_filter_select(plane_state->hw.scaling_filter, 0); 502 ps_ctrl |= PS_SCALER_EN | PS_PLANE_SEL(plane->id) | scaler->mode; 503 504 skl_scaler_setup_filter(dev_priv, pipe, scaler_id, 0, 505 plane_state->hw.scaling_filter); 506 507 intel_de_write_fw(dev_priv, SKL_PS_CTRL(pipe, scaler_id), ps_ctrl); 508 intel_de_write_fw(dev_priv, SKL_PS_VPHASE(pipe, scaler_id), 509 PS_Y_PHASE(y_vphase) | PS_UV_RGB_PHASE(uv_rgb_vphase)); 510 intel_de_write_fw(dev_priv, SKL_PS_HPHASE(pipe, scaler_id), 511 PS_Y_PHASE(y_hphase) | PS_UV_RGB_PHASE(uv_rgb_hphase)); 512 intel_de_write_fw(dev_priv, SKL_PS_WIN_POS(pipe, scaler_id), 513 (crtc_x << 16) | crtc_y); 514 intel_de_write_fw(dev_priv, SKL_PS_WIN_SZ(pipe, scaler_id), 515 (crtc_w << 16) | crtc_h); 516 } 517 518 static void skl_detach_scaler(struct intel_crtc *crtc, int id) 519 { 520 struct drm_device *dev = crtc->base.dev; 521 struct drm_i915_private *dev_priv = to_i915(dev); 522 unsigned long irqflags; 523 524 spin_lock_irqsave(&dev_priv->uncore.lock, irqflags); 525 526 intel_de_write_fw(dev_priv, SKL_PS_CTRL(crtc->pipe, id), 0); 527 intel_de_write_fw(dev_priv, SKL_PS_WIN_POS(crtc->pipe, id), 0); 528 intel_de_write_fw(dev_priv, SKL_PS_WIN_SZ(crtc->pipe, id), 0); 529 530 spin_unlock_irqrestore(&dev_priv->uncore.lock, irqflags); 531 } 532 533 /* 534 * This function detaches (aka. unbinds) unused scalers in hardware 535 */ 536 void skl_detach_scalers(const struct intel_crtc_state *crtc_state) 537 { 538 struct intel_crtc *crtc = to_intel_crtc(crtc_state->uapi.crtc); 539 const struct intel_crtc_scaler_state *scaler_state = 540 &crtc_state->scaler_state; 541 int i; 542 543 /* loop through and disable scalers that aren't in use */ 544 for (i = 0; i < crtc->num_scalers; i++) { 545 if (!scaler_state->scalers[i].in_use) 546 skl_detach_scaler(crtc, i); 547 } 548 } 549 550 void skl_scaler_disable(const struct intel_crtc_state *old_crtc_state) 551 { 552 struct intel_crtc *crtc = to_intel_crtc(old_crtc_state->uapi.crtc); 553 int i; 554 555 for (i = 0; i < crtc->num_scalers; i++) 556 skl_detach_scaler(crtc, i); 557 } 558