1 /* SPDX-License-Identifier: GPL-2.0-only */ 2 /* 3 * Copyright (C) 2015 Broadcom 4 */ 5 #ifndef _VC4_DRV_H_ 6 #define _VC4_DRV_H_ 7 8 #include <linux/delay.h> 9 #include <linux/of.h> 10 #include <linux/refcount.h> 11 #include <linux/uaccess.h> 12 13 #include <drm/drm_atomic.h> 14 #include <drm/drm_debugfs.h> 15 #include <drm/drm_device.h> 16 #include <drm/drm_encoder.h> 17 #include <drm/drm_gem_dma_helper.h> 18 #include <drm/drm_managed.h> 19 #include <drm/drm_mm.h> 20 #include <drm/drm_modeset_lock.h> 21 22 #include "uapi/drm/vc4_drm.h" 23 24 struct drm_device; 25 struct drm_gem_object; 26 27 /* Don't forget to update vc4_bo.c: bo_type_names[] when adding to 28 * this. 29 */ 30 enum vc4_kernel_bo_type { 31 /* Any kernel allocation (gem_create_object hook) before it 32 * gets another type set. 33 */ 34 VC4_BO_TYPE_KERNEL, 35 VC4_BO_TYPE_V3D, 36 VC4_BO_TYPE_V3D_SHADER, 37 VC4_BO_TYPE_DUMB, 38 VC4_BO_TYPE_BIN, 39 VC4_BO_TYPE_RCL, 40 VC4_BO_TYPE_BCL, 41 VC4_BO_TYPE_KERNEL_CACHE, 42 VC4_BO_TYPE_COUNT 43 }; 44 45 /* Performance monitor object. The perform lifetime is controlled by userspace 46 * using perfmon related ioctls. A perfmon can be attached to a submit_cl 47 * request, and when this is the case, HW perf counters will be activated just 48 * before the submit_cl is submitted to the GPU and disabled when the job is 49 * done. This way, only events related to a specific job will be counted. 50 */ 51 struct vc4_perfmon { 52 struct vc4_dev *dev; 53 54 /* Tracks the number of users of the perfmon, when this counter reaches 55 * zero the perfmon is destroyed. 56 */ 57 refcount_t refcnt; 58 59 /* Number of counters activated in this perfmon instance 60 * (should be less than DRM_VC4_MAX_PERF_COUNTERS). 61 */ 62 u8 ncounters; 63 64 /* Events counted by the HW perf counters. */ 65 u8 events[DRM_VC4_MAX_PERF_COUNTERS]; 66 67 /* Storage for counter values. Counters are incremented by the HW 68 * perf counter values every time the perfmon is attached to a GPU job. 69 * This way, perfmon users don't have to retrieve the results after 70 * each job if they want to track events covering several submissions. 71 * Note that counter values can't be reset, but you can fake a reset by 72 * destroying the perfmon and creating a new one. 73 */ 74 u64 counters[]; 75 }; 76 77 struct vc4_dev { 78 struct drm_device base; 79 struct device *dev; 80 81 bool is_vc5; 82 83 unsigned int irq; 84 85 struct vc4_hvs *hvs; 86 struct vc4_v3d *v3d; 87 88 struct vc4_hang_state *hang_state; 89 90 /* The kernel-space BO cache. Tracks buffers that have been 91 * unreferenced by all other users (refcounts of 0!) but not 92 * yet freed, so we can do cheap allocations. 93 */ 94 struct vc4_bo_cache { 95 /* Array of list heads for entries in the BO cache, 96 * based on number of pages, so we can do O(1) lookups 97 * in the cache when allocating. 98 */ 99 struct list_head *size_list; 100 uint32_t size_list_size; 101 102 /* List of all BOs in the cache, ordered by age, so we 103 * can do O(1) lookups when trying to free old 104 * buffers. 105 */ 106 struct list_head time_list; 107 struct work_struct time_work; 108 struct timer_list time_timer; 109 } bo_cache; 110 111 u32 num_labels; 112 struct vc4_label { 113 const char *name; 114 u32 num_allocated; 115 u32 size_allocated; 116 } *bo_labels; 117 118 /* Protects bo_cache and bo_labels. */ 119 struct mutex bo_lock; 120 121 /* Purgeable BO pool. All BOs in this pool can have their memory 122 * reclaimed if the driver is unable to allocate new BOs. We also 123 * keep stats related to the purge mechanism here. 124 */ 125 struct { 126 struct list_head list; 127 unsigned int num; 128 size_t size; 129 unsigned int purged_num; 130 size_t purged_size; 131 struct mutex lock; 132 } purgeable; 133 134 uint64_t dma_fence_context; 135 136 /* Sequence number for the last job queued in bin_job_list. 137 * Starts at 0 (no jobs emitted). 138 */ 139 uint64_t emit_seqno; 140 141 /* Sequence number for the last completed job on the GPU. 142 * Starts at 0 (no jobs completed). 143 */ 144 uint64_t finished_seqno; 145 146 /* List of all struct vc4_exec_info for jobs to be executed in 147 * the binner. The first job in the list is the one currently 148 * programmed into ct0ca for execution. 149 */ 150 struct list_head bin_job_list; 151 152 /* List of all struct vc4_exec_info for jobs that have 153 * completed binning and are ready for rendering. The first 154 * job in the list is the one currently programmed into ct1ca 155 * for execution. 156 */ 157 struct list_head render_job_list; 158 159 /* List of the finished vc4_exec_infos waiting to be freed by 160 * job_done_work. 161 */ 162 struct list_head job_done_list; 163 /* Spinlock used to synchronize the job_list and seqno 164 * accesses between the IRQ handler and GEM ioctls. 165 */ 166 spinlock_t job_lock; 167 wait_queue_head_t job_wait_queue; 168 struct work_struct job_done_work; 169 170 /* Used to track the active perfmon if any. Access to this field is 171 * protected by job_lock. 172 */ 173 struct vc4_perfmon *active_perfmon; 174 175 /* List of struct vc4_seqno_cb for callbacks to be made from a 176 * workqueue when the given seqno is passed. 177 */ 178 struct list_head seqno_cb_list; 179 180 /* The memory used for storing binner tile alloc, tile state, 181 * and overflow memory allocations. This is freed when V3D 182 * powers down. 183 */ 184 struct vc4_bo *bin_bo; 185 186 /* Size of blocks allocated within bin_bo. */ 187 uint32_t bin_alloc_size; 188 189 /* Bitmask of the bin_alloc_size chunks in bin_bo that are 190 * used. 191 */ 192 uint32_t bin_alloc_used; 193 194 /* Bitmask of the current bin_alloc used for overflow memory. */ 195 uint32_t bin_alloc_overflow; 196 197 /* Incremented when an underrun error happened after an atomic commit. 198 * This is particularly useful to detect when a specific modeset is too 199 * demanding in term of memory or HVS bandwidth which is hard to guess 200 * at atomic check time. 201 */ 202 atomic_t underrun; 203 204 struct work_struct overflow_mem_work; 205 206 int power_refcount; 207 208 /* Set to true when the load tracker is active. */ 209 bool load_tracker_enabled; 210 211 /* Mutex controlling the power refcount. */ 212 struct mutex power_lock; 213 214 struct { 215 struct timer_list timer; 216 struct work_struct reset_work; 217 } hangcheck; 218 219 struct drm_modeset_lock ctm_state_lock; 220 struct drm_private_obj ctm_manager; 221 struct drm_private_obj hvs_channels; 222 struct drm_private_obj load_tracker; 223 224 /* List of vc4_debugfs_info_entry for adding to debugfs once 225 * the minor is available (after drm_dev_register()). 226 */ 227 struct list_head debugfs_list; 228 229 /* Mutex for binner bo allocation. */ 230 struct mutex bin_bo_lock; 231 /* Reference count for our binner bo. */ 232 struct kref bin_bo_kref; 233 }; 234 235 static inline struct vc4_dev * 236 to_vc4_dev(const struct drm_device *dev) 237 { 238 return container_of(dev, struct vc4_dev, base); 239 } 240 241 struct vc4_bo { 242 struct drm_gem_dma_object base; 243 244 /* seqno of the last job to render using this BO. */ 245 uint64_t seqno; 246 247 /* seqno of the last job to use the RCL to write to this BO. 248 * 249 * Note that this doesn't include binner overflow memory 250 * writes. 251 */ 252 uint64_t write_seqno; 253 254 bool t_format; 255 256 /* List entry for the BO's position in either 257 * vc4_exec_info->unref_list or vc4_dev->bo_cache.time_list 258 */ 259 struct list_head unref_head; 260 261 /* Time in jiffies when the BO was put in vc4->bo_cache. */ 262 unsigned long free_time; 263 264 /* List entry for the BO's position in vc4_dev->bo_cache.size_list */ 265 struct list_head size_head; 266 267 /* Struct for shader validation state, if created by 268 * DRM_IOCTL_VC4_CREATE_SHADER_BO. 269 */ 270 struct vc4_validated_shader_info *validated_shader; 271 272 /* One of enum vc4_kernel_bo_type, or VC4_BO_TYPE_COUNT + i 273 * for user-allocated labels. 274 */ 275 int label; 276 277 /* Count the number of active users. This is needed to determine 278 * whether we can move the BO to the purgeable list or not (when the BO 279 * is used by the GPU or the display engine we can't purge it). 280 */ 281 refcount_t usecnt; 282 283 /* Store purgeable/purged state here */ 284 u32 madv; 285 struct mutex madv_lock; 286 }; 287 288 static inline struct vc4_bo * 289 to_vc4_bo(const struct drm_gem_object *bo) 290 { 291 return container_of(to_drm_gem_dma_obj(bo), struct vc4_bo, base); 292 } 293 294 struct vc4_fence { 295 struct dma_fence base; 296 struct drm_device *dev; 297 /* vc4 seqno for signaled() test */ 298 uint64_t seqno; 299 }; 300 301 static inline struct vc4_fence * 302 to_vc4_fence(const struct dma_fence *fence) 303 { 304 return container_of(fence, struct vc4_fence, base); 305 } 306 307 struct vc4_seqno_cb { 308 struct work_struct work; 309 uint64_t seqno; 310 void (*func)(struct vc4_seqno_cb *cb); 311 }; 312 313 struct vc4_v3d { 314 struct vc4_dev *vc4; 315 struct platform_device *pdev; 316 void __iomem *regs; 317 struct clk *clk; 318 struct debugfs_regset32 regset; 319 }; 320 321 struct vc4_hvs { 322 struct vc4_dev *vc4; 323 struct platform_device *pdev; 324 void __iomem *regs; 325 u32 __iomem *dlist; 326 327 struct clk *core_clk; 328 329 unsigned long max_core_rate; 330 331 /* Memory manager for CRTCs to allocate space in the display 332 * list. Units are dwords. 333 */ 334 struct drm_mm dlist_mm; 335 /* Memory manager for the LBM memory used by HVS scaling. */ 336 struct drm_mm lbm_mm; 337 spinlock_t mm_lock; 338 339 struct drm_mm_node mitchell_netravali_filter; 340 341 struct debugfs_regset32 regset; 342 343 /* 344 * Even if HDMI0 on the RPi4 can output modes requiring a pixel 345 * rate higher than 297MHz, it needs some adjustments in the 346 * config.txt file to be able to do so and thus won't always be 347 * available. 348 */ 349 bool vc5_hdmi_enable_hdmi_20; 350 351 /* 352 * 4096x2160@60 requires a core overclock to work, so register 353 * whether that is sufficient. 354 */ 355 bool vc5_hdmi_enable_4096by2160; 356 }; 357 358 #define HVS_NUM_CHANNELS 3 359 360 struct vc4_hvs_state { 361 struct drm_private_state base; 362 unsigned long core_clock_rate; 363 364 struct { 365 unsigned in_use: 1; 366 unsigned long fifo_load; 367 struct drm_crtc_commit *pending_commit; 368 } fifo_state[HVS_NUM_CHANNELS]; 369 }; 370 371 static inline struct vc4_hvs_state * 372 to_vc4_hvs_state(const struct drm_private_state *priv) 373 { 374 return container_of(priv, struct vc4_hvs_state, base); 375 } 376 377 struct vc4_hvs_state *vc4_hvs_get_global_state(struct drm_atomic_state *state); 378 struct vc4_hvs_state *vc4_hvs_get_old_global_state(const struct drm_atomic_state *state); 379 struct vc4_hvs_state *vc4_hvs_get_new_global_state(const struct drm_atomic_state *state); 380 381 struct vc4_plane { 382 struct drm_plane base; 383 }; 384 385 static inline struct vc4_plane * 386 to_vc4_plane(const struct drm_plane *plane) 387 { 388 return container_of(plane, struct vc4_plane, base); 389 } 390 391 enum vc4_scaling_mode { 392 VC4_SCALING_NONE, 393 VC4_SCALING_TPZ, 394 VC4_SCALING_PPF, 395 }; 396 397 struct vc4_plane_state { 398 struct drm_plane_state base; 399 /* System memory copy of the display list for this element, computed 400 * at atomic_check time. 401 */ 402 u32 *dlist; 403 u32 dlist_size; /* Number of dwords allocated for the display list */ 404 u32 dlist_count; /* Number of used dwords in the display list. */ 405 406 /* Offset in the dlist to various words, for pageflip or 407 * cursor updates. 408 */ 409 u32 pos0_offset; 410 u32 pos2_offset; 411 u32 ptr0_offset; 412 u32 lbm_offset; 413 414 /* Offset where the plane's dlist was last stored in the 415 * hardware at vc4_crtc_atomic_flush() time. 416 */ 417 u32 __iomem *hw_dlist; 418 419 /* Clipped coordinates of the plane on the display. */ 420 int crtc_x, crtc_y, crtc_w, crtc_h; 421 /* Clipped area being scanned from in the FB. */ 422 u32 src_x, src_y; 423 424 u32 src_w[2], src_h[2]; 425 426 /* Scaling selection for the RGB/Y plane and the Cb/Cr planes. */ 427 enum vc4_scaling_mode x_scaling[2], y_scaling[2]; 428 bool is_unity; 429 bool is_yuv; 430 431 /* Offset to start scanning out from the start of the plane's 432 * BO. 433 */ 434 u32 offsets[3]; 435 436 /* Our allocation in LBM for temporary storage during scaling. */ 437 struct drm_mm_node lbm; 438 439 /* Set when the plane has per-pixel alpha content or does not cover 440 * the entire screen. This is a hint to the CRTC that it might need 441 * to enable background color fill. 442 */ 443 bool needs_bg_fill; 444 445 /* Mark the dlist as initialized. Useful to avoid initializing it twice 446 * when async update is not possible. 447 */ 448 bool dlist_initialized; 449 450 /* Load of this plane on the HVS block. The load is expressed in HVS 451 * cycles/sec. 452 */ 453 u64 hvs_load; 454 455 /* Memory bandwidth needed for this plane. This is expressed in 456 * bytes/sec. 457 */ 458 u64 membus_load; 459 }; 460 461 static inline struct vc4_plane_state * 462 to_vc4_plane_state(const struct drm_plane_state *state) 463 { 464 return container_of(state, struct vc4_plane_state, base); 465 } 466 467 enum vc4_encoder_type { 468 VC4_ENCODER_TYPE_NONE, 469 VC4_ENCODER_TYPE_HDMI0, 470 VC4_ENCODER_TYPE_HDMI1, 471 VC4_ENCODER_TYPE_VEC, 472 VC4_ENCODER_TYPE_DSI0, 473 VC4_ENCODER_TYPE_DSI1, 474 VC4_ENCODER_TYPE_SMI, 475 VC4_ENCODER_TYPE_DPI, 476 VC4_ENCODER_TYPE_TXP, 477 }; 478 479 struct vc4_encoder { 480 struct drm_encoder base; 481 enum vc4_encoder_type type; 482 u32 clock_select; 483 484 void (*pre_crtc_configure)(struct drm_encoder *encoder, struct drm_atomic_state *state); 485 void (*pre_crtc_enable)(struct drm_encoder *encoder, struct drm_atomic_state *state); 486 void (*post_crtc_enable)(struct drm_encoder *encoder, struct drm_atomic_state *state); 487 488 void (*post_crtc_disable)(struct drm_encoder *encoder, struct drm_atomic_state *state); 489 void (*post_crtc_powerdown)(struct drm_encoder *encoder, struct drm_atomic_state *state); 490 }; 491 492 static inline struct vc4_encoder * 493 to_vc4_encoder(const struct drm_encoder *encoder) 494 { 495 return container_of(encoder, struct vc4_encoder, base); 496 } 497 498 struct vc4_crtc_data { 499 const char *name; 500 501 const char *debugfs_name; 502 503 /* Bitmask of channels (FIFOs) of the HVS that the output can source from */ 504 unsigned int hvs_available_channels; 505 506 /* Which output of the HVS this pixelvalve sources from. */ 507 int hvs_output; 508 }; 509 510 struct vc4_pv_data { 511 struct vc4_crtc_data base; 512 513 /* Depth of the PixelValve FIFO in bytes */ 514 unsigned int fifo_depth; 515 516 /* Number of pixels output per clock period */ 517 u8 pixels_per_clock; 518 519 enum vc4_encoder_type encoder_types[4]; 520 }; 521 522 struct vc4_crtc { 523 struct drm_crtc base; 524 struct platform_device *pdev; 525 const struct vc4_crtc_data *data; 526 void __iomem *regs; 527 528 /* Timestamp at start of vblank irq - unaffected by lock delays. */ 529 ktime_t t_vblank; 530 531 u8 lut_r[256]; 532 u8 lut_g[256]; 533 u8 lut_b[256]; 534 535 struct drm_pending_vblank_event *event; 536 537 struct debugfs_regset32 regset; 538 539 /** 540 * @feeds_txp: True if the CRTC feeds our writeback controller. 541 */ 542 bool feeds_txp; 543 544 /** 545 * @irq_lock: Spinlock protecting the resources shared between 546 * the atomic code and our vblank handler. 547 */ 548 spinlock_t irq_lock; 549 550 /** 551 * @current_dlist: Start offset of the display list currently 552 * set in the HVS for that CRTC. Protected by @irq_lock, and 553 * copied in vc4_hvs_update_dlist() for the CRTC interrupt 554 * handler to have access to that value. 555 */ 556 unsigned int current_dlist; 557 558 /** 559 * @current_hvs_channel: HVS channel currently assigned to the 560 * CRTC. Protected by @irq_lock, and copied in 561 * vc4_hvs_atomic_begin() for the CRTC interrupt handler to have 562 * access to that value. 563 */ 564 unsigned int current_hvs_channel; 565 }; 566 567 static inline struct vc4_crtc * 568 to_vc4_crtc(const struct drm_crtc *crtc) 569 { 570 return container_of(crtc, struct vc4_crtc, base); 571 } 572 573 static inline const struct vc4_crtc_data * 574 vc4_crtc_to_vc4_crtc_data(const struct vc4_crtc *crtc) 575 { 576 return crtc->data; 577 } 578 579 static inline const struct vc4_pv_data * 580 vc4_crtc_to_vc4_pv_data(const struct vc4_crtc *crtc) 581 { 582 const struct vc4_crtc_data *data = vc4_crtc_to_vc4_crtc_data(crtc); 583 584 return container_of(data, struct vc4_pv_data, base); 585 } 586 587 struct drm_encoder *vc4_get_crtc_encoder(struct drm_crtc *crtc, 588 struct drm_crtc_state *state); 589 590 struct vc4_crtc_state { 591 struct drm_crtc_state base; 592 /* Dlist area for this CRTC configuration. */ 593 struct drm_mm_node mm; 594 bool txp_armed; 595 unsigned int assigned_channel; 596 597 struct { 598 unsigned int left; 599 unsigned int right; 600 unsigned int top; 601 unsigned int bottom; 602 } margins; 603 604 unsigned long hvs_load; 605 606 /* Transitional state below, only valid during atomic commits */ 607 bool update_muxing; 608 }; 609 610 #define VC4_HVS_CHANNEL_DISABLED ((unsigned int)-1) 611 612 static inline struct vc4_crtc_state * 613 to_vc4_crtc_state(const struct drm_crtc_state *crtc_state) 614 { 615 return container_of(crtc_state, struct vc4_crtc_state, base); 616 } 617 618 #define V3D_READ(offset) readl(vc4->v3d->regs + offset) 619 #define V3D_WRITE(offset, val) writel(val, vc4->v3d->regs + offset) 620 #define HVS_READ(offset) readl(hvs->regs + offset) 621 #define HVS_WRITE(offset, val) writel(val, hvs->regs + offset) 622 623 #define VC4_REG32(reg) { .name = #reg, .offset = reg } 624 625 struct vc4_exec_info { 626 struct vc4_dev *dev; 627 628 /* Sequence number for this bin/render job. */ 629 uint64_t seqno; 630 631 /* Latest write_seqno of any BO that binning depends on. */ 632 uint64_t bin_dep_seqno; 633 634 struct dma_fence *fence; 635 636 /* Last current addresses the hardware was processing when the 637 * hangcheck timer checked on us. 638 */ 639 uint32_t last_ct0ca, last_ct1ca; 640 641 /* Kernel-space copy of the ioctl arguments */ 642 struct drm_vc4_submit_cl *args; 643 644 /* This is the array of BOs that were looked up at the start of exec. 645 * Command validation will use indices into this array. 646 */ 647 struct drm_gem_dma_object **bo; 648 uint32_t bo_count; 649 650 /* List of BOs that are being written by the RCL. Other than 651 * the binner temporary storage, this is all the BOs written 652 * by the job. 653 */ 654 struct drm_gem_dma_object *rcl_write_bo[4]; 655 uint32_t rcl_write_bo_count; 656 657 /* Pointers for our position in vc4->job_list */ 658 struct list_head head; 659 660 /* List of other BOs used in the job that need to be released 661 * once the job is complete. 662 */ 663 struct list_head unref_list; 664 665 /* Current unvalidated indices into @bo loaded by the non-hardware 666 * VC4_PACKET_GEM_HANDLES. 667 */ 668 uint32_t bo_index[2]; 669 670 /* This is the BO where we store the validated command lists, shader 671 * records, and uniforms. 672 */ 673 struct drm_gem_dma_object *exec_bo; 674 675 /** 676 * This tracks the per-shader-record state (packet 64) that 677 * determines the length of the shader record and the offset 678 * it's expected to be found at. It gets read in from the 679 * command lists. 680 */ 681 struct vc4_shader_state { 682 uint32_t addr; 683 /* Maximum vertex index referenced by any primitive using this 684 * shader state. 685 */ 686 uint32_t max_index; 687 } *shader_state; 688 689 /** How many shader states the user declared they were using. */ 690 uint32_t shader_state_size; 691 /** How many shader state records the validator has seen. */ 692 uint32_t shader_state_count; 693 694 bool found_tile_binning_mode_config_packet; 695 bool found_start_tile_binning_packet; 696 bool found_increment_semaphore_packet; 697 bool found_flush; 698 uint8_t bin_tiles_x, bin_tiles_y; 699 /* Physical address of the start of the tile alloc array 700 * (where each tile's binned CL will start) 701 */ 702 uint32_t tile_alloc_offset; 703 /* Bitmask of which binner slots are freed when this job completes. */ 704 uint32_t bin_slots; 705 706 /** 707 * Computed addresses pointing into exec_bo where we start the 708 * bin thread (ct0) and render thread (ct1). 709 */ 710 uint32_t ct0ca, ct0ea; 711 uint32_t ct1ca, ct1ea; 712 713 /* Pointer to the unvalidated bin CL (if present). */ 714 void *bin_u; 715 716 /* Pointers to the shader recs. These paddr gets incremented as CL 717 * packets are relocated in validate_gl_shader_state, and the vaddrs 718 * (u and v) get incremented and size decremented as the shader recs 719 * themselves are validated. 720 */ 721 void *shader_rec_u; 722 void *shader_rec_v; 723 uint32_t shader_rec_p; 724 uint32_t shader_rec_size; 725 726 /* Pointers to the uniform data. These pointers are incremented, and 727 * size decremented, as each batch of uniforms is uploaded. 728 */ 729 void *uniforms_u; 730 void *uniforms_v; 731 uint32_t uniforms_p; 732 uint32_t uniforms_size; 733 734 /* Pointer to a performance monitor object if the user requested it, 735 * NULL otherwise. 736 */ 737 struct vc4_perfmon *perfmon; 738 739 /* Whether the exec has taken a reference to the binner BO, which should 740 * happen with a VC4_PACKET_TILE_BINNING_MODE_CONFIG packet. 741 */ 742 bool bin_bo_used; 743 }; 744 745 /* Per-open file private data. Any driver-specific resource that has to be 746 * released when the DRM file is closed should be placed here. 747 */ 748 struct vc4_file { 749 struct vc4_dev *dev; 750 751 struct { 752 struct idr idr; 753 struct mutex lock; 754 } perfmon; 755 756 bool bin_bo_used; 757 }; 758 759 static inline struct vc4_exec_info * 760 vc4_first_bin_job(struct vc4_dev *vc4) 761 { 762 return list_first_entry_or_null(&vc4->bin_job_list, 763 struct vc4_exec_info, head); 764 } 765 766 static inline struct vc4_exec_info * 767 vc4_first_render_job(struct vc4_dev *vc4) 768 { 769 return list_first_entry_or_null(&vc4->render_job_list, 770 struct vc4_exec_info, head); 771 } 772 773 static inline struct vc4_exec_info * 774 vc4_last_render_job(struct vc4_dev *vc4) 775 { 776 if (list_empty(&vc4->render_job_list)) 777 return NULL; 778 return list_last_entry(&vc4->render_job_list, 779 struct vc4_exec_info, head); 780 } 781 782 /** 783 * struct vc4_texture_sample_info - saves the offsets into the UBO for texture 784 * setup parameters. 785 * 786 * This will be used at draw time to relocate the reference to the texture 787 * contents in p0, and validate that the offset combined with 788 * width/height/stride/etc. from p1 and p2/p3 doesn't sample outside the BO. 789 * Note that the hardware treats unprovided config parameters as 0, so not all 790 * of them need to be set up for every texure sample, and we'll store ~0 as 791 * the offset to mark the unused ones. 792 * 793 * See the VC4 3D architecture guide page 41 ("Texture and Memory Lookup Unit 794 * Setup") for definitions of the texture parameters. 795 */ 796 struct vc4_texture_sample_info { 797 bool is_direct; 798 uint32_t p_offset[4]; 799 }; 800 801 /** 802 * struct vc4_validated_shader_info - information about validated shaders that 803 * needs to be used from command list validation. 804 * 805 * For a given shader, each time a shader state record references it, we need 806 * to verify that the shader doesn't read more uniforms than the shader state 807 * record's uniform BO pointer can provide, and we need to apply relocations 808 * and validate the shader state record's uniforms that define the texture 809 * samples. 810 */ 811 struct vc4_validated_shader_info { 812 uint32_t uniforms_size; 813 uint32_t uniforms_src_size; 814 uint32_t num_texture_samples; 815 struct vc4_texture_sample_info *texture_samples; 816 817 uint32_t num_uniform_addr_offsets; 818 uint32_t *uniform_addr_offsets; 819 820 bool is_threaded; 821 }; 822 823 /** 824 * __wait_for - magic wait macro 825 * 826 * Macro to help avoid open coding check/wait/timeout patterns. Note that it's 827 * important that we check the condition again after having timed out, since the 828 * timeout could be due to preemption or similar and we've never had a chance to 829 * check the condition before the timeout. 830 */ 831 #define __wait_for(OP, COND, US, Wmin, Wmax) ({ \ 832 const ktime_t end__ = ktime_add_ns(ktime_get_raw(), 1000ll * (US)); \ 833 long wait__ = (Wmin); /* recommended min for usleep is 10 us */ \ 834 int ret__; \ 835 might_sleep(); \ 836 for (;;) { \ 837 const bool expired__ = ktime_after(ktime_get_raw(), end__); \ 838 OP; \ 839 /* Guarantee COND check prior to timeout */ \ 840 barrier(); \ 841 if (COND) { \ 842 ret__ = 0; \ 843 break; \ 844 } \ 845 if (expired__) { \ 846 ret__ = -ETIMEDOUT; \ 847 break; \ 848 } \ 849 usleep_range(wait__, wait__ * 2); \ 850 if (wait__ < (Wmax)) \ 851 wait__ <<= 1; \ 852 } \ 853 ret__; \ 854 }) 855 856 #define _wait_for(COND, US, Wmin, Wmax) __wait_for(, (COND), (US), (Wmin), \ 857 (Wmax)) 858 #define wait_for(COND, MS) _wait_for((COND), (MS) * 1000, 10, 1000) 859 860 /* vc4_bo.c */ 861 struct drm_gem_object *vc4_create_object(struct drm_device *dev, size_t size); 862 struct vc4_bo *vc4_bo_create(struct drm_device *dev, size_t size, 863 bool from_cache, enum vc4_kernel_bo_type type); 864 int vc4_bo_dumb_create(struct drm_file *file_priv, 865 struct drm_device *dev, 866 struct drm_mode_create_dumb *args); 867 int vc4_create_bo_ioctl(struct drm_device *dev, void *data, 868 struct drm_file *file_priv); 869 int vc4_create_shader_bo_ioctl(struct drm_device *dev, void *data, 870 struct drm_file *file_priv); 871 int vc4_mmap_bo_ioctl(struct drm_device *dev, void *data, 872 struct drm_file *file_priv); 873 int vc4_set_tiling_ioctl(struct drm_device *dev, void *data, 874 struct drm_file *file_priv); 875 int vc4_get_tiling_ioctl(struct drm_device *dev, void *data, 876 struct drm_file *file_priv); 877 int vc4_get_hang_state_ioctl(struct drm_device *dev, void *data, 878 struct drm_file *file_priv); 879 int vc4_label_bo_ioctl(struct drm_device *dev, void *data, 880 struct drm_file *file_priv); 881 int vc4_bo_cache_init(struct drm_device *dev); 882 int vc4_bo_inc_usecnt(struct vc4_bo *bo); 883 void vc4_bo_dec_usecnt(struct vc4_bo *bo); 884 void vc4_bo_add_to_purgeable_pool(struct vc4_bo *bo); 885 void vc4_bo_remove_from_purgeable_pool(struct vc4_bo *bo); 886 int vc4_bo_debugfs_init(struct drm_minor *minor); 887 888 /* vc4_crtc.c */ 889 extern struct platform_driver vc4_crtc_driver; 890 int vc4_crtc_disable_at_boot(struct drm_crtc *crtc); 891 int vc4_crtc_init(struct drm_device *drm, struct platform_device *pdev, 892 struct vc4_crtc *vc4_crtc, const struct vc4_crtc_data *data, 893 const struct drm_crtc_funcs *crtc_funcs, 894 const struct drm_crtc_helper_funcs *crtc_helper_funcs, 895 bool feeds_txp); 896 int vc4_page_flip(struct drm_crtc *crtc, 897 struct drm_framebuffer *fb, 898 struct drm_pending_vblank_event *event, 899 uint32_t flags, 900 struct drm_modeset_acquire_ctx *ctx); 901 struct drm_crtc_state *vc4_crtc_duplicate_state(struct drm_crtc *crtc); 902 void vc4_crtc_destroy_state(struct drm_crtc *crtc, 903 struct drm_crtc_state *state); 904 void vc4_crtc_reset(struct drm_crtc *crtc); 905 void vc4_crtc_handle_vblank(struct vc4_crtc *crtc); 906 void vc4_crtc_send_vblank(struct drm_crtc *crtc); 907 int vc4_crtc_late_register(struct drm_crtc *crtc); 908 void vc4_crtc_get_margins(struct drm_crtc_state *state, 909 unsigned int *left, unsigned int *right, 910 unsigned int *top, unsigned int *bottom); 911 912 /* vc4_debugfs.c */ 913 void vc4_debugfs_init(struct drm_minor *minor); 914 #ifdef CONFIG_DEBUG_FS 915 int vc4_debugfs_add_file(struct drm_minor *minor, 916 const char *filename, 917 int (*show)(struct seq_file*, void*), 918 void *data); 919 int vc4_debugfs_add_regset32(struct drm_minor *minor, 920 const char *filename, 921 struct debugfs_regset32 *regset); 922 #else 923 static inline int vc4_debugfs_add_file(struct drm_minor *minor, 924 const char *filename, 925 int (*show)(struct seq_file*, void*), 926 void *data) 927 { 928 return 0; 929 } 930 931 static inline int vc4_debugfs_add_regset32(struct drm_minor *minor, 932 const char *filename, 933 struct debugfs_regset32 *regset) 934 { 935 return 0; 936 } 937 #endif 938 939 /* vc4_drv.c */ 940 void __iomem *vc4_ioremap_regs(struct platform_device *dev, int index); 941 int vc4_dumb_fixup_args(struct drm_mode_create_dumb *args); 942 943 /* vc4_dpi.c */ 944 extern struct platform_driver vc4_dpi_driver; 945 946 /* vc4_dsi.c */ 947 extern struct platform_driver vc4_dsi_driver; 948 949 /* vc4_fence.c */ 950 extern const struct dma_fence_ops vc4_fence_ops; 951 952 /* vc4_gem.c */ 953 int vc4_gem_init(struct drm_device *dev); 954 int vc4_submit_cl_ioctl(struct drm_device *dev, void *data, 955 struct drm_file *file_priv); 956 int vc4_wait_seqno_ioctl(struct drm_device *dev, void *data, 957 struct drm_file *file_priv); 958 int vc4_wait_bo_ioctl(struct drm_device *dev, void *data, 959 struct drm_file *file_priv); 960 void vc4_submit_next_bin_job(struct drm_device *dev); 961 void vc4_submit_next_render_job(struct drm_device *dev); 962 void vc4_move_job_to_render(struct drm_device *dev, struct vc4_exec_info *exec); 963 int vc4_wait_for_seqno(struct drm_device *dev, uint64_t seqno, 964 uint64_t timeout_ns, bool interruptible); 965 void vc4_job_handle_completed(struct vc4_dev *vc4); 966 int vc4_queue_seqno_cb(struct drm_device *dev, 967 struct vc4_seqno_cb *cb, uint64_t seqno, 968 void (*func)(struct vc4_seqno_cb *cb)); 969 int vc4_gem_madvise_ioctl(struct drm_device *dev, void *data, 970 struct drm_file *file_priv); 971 972 /* vc4_hdmi.c */ 973 extern struct platform_driver vc4_hdmi_driver; 974 975 /* vc4_vec.c */ 976 extern struct platform_driver vc4_vec_driver; 977 978 /* vc4_txp.c */ 979 extern struct platform_driver vc4_txp_driver; 980 981 /* vc4_irq.c */ 982 void vc4_irq_enable(struct drm_device *dev); 983 void vc4_irq_disable(struct drm_device *dev); 984 int vc4_irq_install(struct drm_device *dev, int irq); 985 void vc4_irq_uninstall(struct drm_device *dev); 986 void vc4_irq_reset(struct drm_device *dev); 987 988 /* vc4_hvs.c */ 989 extern struct platform_driver vc4_hvs_driver; 990 void vc4_hvs_stop_channel(struct vc4_hvs *hvs, unsigned int output); 991 int vc4_hvs_get_fifo_from_output(struct vc4_hvs *hvs, unsigned int output); 992 u8 vc4_hvs_get_fifo_frame_count(struct vc4_hvs *hvs, unsigned int fifo); 993 int vc4_hvs_atomic_check(struct drm_crtc *crtc, struct drm_atomic_state *state); 994 void vc4_hvs_atomic_begin(struct drm_crtc *crtc, struct drm_atomic_state *state); 995 void vc4_hvs_atomic_enable(struct drm_crtc *crtc, struct drm_atomic_state *state); 996 void vc4_hvs_atomic_disable(struct drm_crtc *crtc, struct drm_atomic_state *state); 997 void vc4_hvs_atomic_flush(struct drm_crtc *crtc, struct drm_atomic_state *state); 998 void vc4_hvs_dump_state(struct vc4_hvs *hvs); 999 void vc4_hvs_unmask_underrun(struct vc4_hvs *hvs, int channel); 1000 void vc4_hvs_mask_underrun(struct vc4_hvs *hvs, int channel); 1001 int vc4_hvs_debugfs_init(struct drm_minor *minor); 1002 1003 /* vc4_kms.c */ 1004 int vc4_kms_load(struct drm_device *dev); 1005 1006 /* vc4_plane.c */ 1007 struct drm_plane *vc4_plane_init(struct drm_device *dev, 1008 enum drm_plane_type type, 1009 uint32_t possible_crtcs); 1010 int vc4_plane_create_additional_planes(struct drm_device *dev); 1011 u32 vc4_plane_write_dlist(struct drm_plane *plane, u32 __iomem *dlist); 1012 u32 vc4_plane_dlist_size(const struct drm_plane_state *state); 1013 void vc4_plane_async_set_fb(struct drm_plane *plane, 1014 struct drm_framebuffer *fb); 1015 1016 /* vc4_v3d.c */ 1017 extern struct platform_driver vc4_v3d_driver; 1018 extern const struct of_device_id vc4_v3d_dt_match[]; 1019 int vc4_v3d_get_bin_slot(struct vc4_dev *vc4); 1020 int vc4_v3d_bin_bo_get(struct vc4_dev *vc4, bool *used); 1021 void vc4_v3d_bin_bo_put(struct vc4_dev *vc4); 1022 int vc4_v3d_pm_get(struct vc4_dev *vc4); 1023 void vc4_v3d_pm_put(struct vc4_dev *vc4); 1024 int vc4_v3d_debugfs_init(struct drm_minor *minor); 1025 1026 /* vc4_validate.c */ 1027 int 1028 vc4_validate_bin_cl(struct drm_device *dev, 1029 void *validated, 1030 void *unvalidated, 1031 struct vc4_exec_info *exec); 1032 1033 int 1034 vc4_validate_shader_recs(struct drm_device *dev, struct vc4_exec_info *exec); 1035 1036 struct drm_gem_dma_object *vc4_use_bo(struct vc4_exec_info *exec, 1037 uint32_t hindex); 1038 1039 int vc4_get_rcl(struct drm_device *dev, struct vc4_exec_info *exec); 1040 1041 bool vc4_check_tex_size(struct vc4_exec_info *exec, 1042 struct drm_gem_dma_object *fbo, 1043 uint32_t offset, uint8_t tiling_format, 1044 uint32_t width, uint32_t height, uint8_t cpp); 1045 1046 /* vc4_validate_shader.c */ 1047 struct vc4_validated_shader_info * 1048 vc4_validate_shader(struct drm_gem_dma_object *shader_obj); 1049 1050 /* vc4_perfmon.c */ 1051 void vc4_perfmon_get(struct vc4_perfmon *perfmon); 1052 void vc4_perfmon_put(struct vc4_perfmon *perfmon); 1053 void vc4_perfmon_start(struct vc4_dev *vc4, struct vc4_perfmon *perfmon); 1054 void vc4_perfmon_stop(struct vc4_dev *vc4, struct vc4_perfmon *perfmon, 1055 bool capture); 1056 struct vc4_perfmon *vc4_perfmon_find(struct vc4_file *vc4file, int id); 1057 void vc4_perfmon_open_file(struct vc4_file *vc4file); 1058 void vc4_perfmon_close_file(struct vc4_file *vc4file); 1059 int vc4_perfmon_create_ioctl(struct drm_device *dev, void *data, 1060 struct drm_file *file_priv); 1061 int vc4_perfmon_destroy_ioctl(struct drm_device *dev, void *data, 1062 struct drm_file *file_priv); 1063 int vc4_perfmon_get_values_ioctl(struct drm_device *dev, void *data, 1064 struct drm_file *file_priv); 1065 1066 #endif /* _VC4_DRV_H_ */ 1067