xref: /openbmc/linux/drivers/gpu/drm/vc4/vc4_drv.h (revision 0ea33321)
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/refcount.h>
10 #include <linux/uaccess.h>
11 
12 #include <drm/drm_atomic.h>
13 #include <drm/drm_debugfs.h>
14 #include <drm/drm_device.h>
15 #include <drm/drm_encoder.h>
16 #include <drm/drm_gem_cma_helper.h>
17 #include <drm/drm_managed.h>
18 #include <drm/drm_mm.h>
19 #include <drm/drm_modeset_lock.h>
20 
21 #include "uapi/drm/vc4_drm.h"
22 
23 struct drm_device;
24 struct drm_gem_object;
25 
26 /* Don't forget to update vc4_bo.c: bo_type_names[] when adding to
27  * this.
28  */
29 enum vc4_kernel_bo_type {
30 	/* Any kernel allocation (gem_create_object hook) before it
31 	 * gets another type set.
32 	 */
33 	VC4_BO_TYPE_KERNEL,
34 	VC4_BO_TYPE_V3D,
35 	VC4_BO_TYPE_V3D_SHADER,
36 	VC4_BO_TYPE_DUMB,
37 	VC4_BO_TYPE_BIN,
38 	VC4_BO_TYPE_RCL,
39 	VC4_BO_TYPE_BCL,
40 	VC4_BO_TYPE_KERNEL_CACHE,
41 	VC4_BO_TYPE_COUNT
42 };
43 
44 /* Performance monitor object. The perform lifetime is controlled by userspace
45  * using perfmon related ioctls. A perfmon can be attached to a submit_cl
46  * request, and when this is the case, HW perf counters will be activated just
47  * before the submit_cl is submitted to the GPU and disabled when the job is
48  * done. This way, only events related to a specific job will be counted.
49  */
50 struct vc4_perfmon {
51 	/* Tracks the number of users of the perfmon, when this counter reaches
52 	 * zero the perfmon is destroyed.
53 	 */
54 	refcount_t refcnt;
55 
56 	/* Number of counters activated in this perfmon instance
57 	 * (should be less than DRM_VC4_MAX_PERF_COUNTERS).
58 	 */
59 	u8 ncounters;
60 
61 	/* Events counted by the HW perf counters. */
62 	u8 events[DRM_VC4_MAX_PERF_COUNTERS];
63 
64 	/* Storage for counter values. Counters are incremented by the HW
65 	 * perf counter values every time the perfmon is attached to a GPU job.
66 	 * This way, perfmon users don't have to retrieve the results after
67 	 * each job if they want to track events covering several submissions.
68 	 * Note that counter values can't be reset, but you can fake a reset by
69 	 * destroying the perfmon and creating a new one.
70 	 */
71 	u64 counters[];
72 };
73 
74 struct vc4_dev {
75 	struct drm_device base;
76 
77 	struct vc4_hvs *hvs;
78 	struct vc4_v3d *v3d;
79 	struct vc4_dpi *dpi;
80 	struct vc4_dsi *dsi1;
81 	struct vc4_vec *vec;
82 	struct vc4_txp *txp;
83 
84 	struct vc4_hang_state *hang_state;
85 
86 	/* The kernel-space BO cache.  Tracks buffers that have been
87 	 * unreferenced by all other users (refcounts of 0!) but not
88 	 * yet freed, so we can do cheap allocations.
89 	 */
90 	struct vc4_bo_cache {
91 		/* Array of list heads for entries in the BO cache,
92 		 * based on number of pages, so we can do O(1) lookups
93 		 * in the cache when allocating.
94 		 */
95 		struct list_head *size_list;
96 		uint32_t size_list_size;
97 
98 		/* List of all BOs in the cache, ordered by age, so we
99 		 * can do O(1) lookups when trying to free old
100 		 * buffers.
101 		 */
102 		struct list_head time_list;
103 		struct work_struct time_work;
104 		struct timer_list time_timer;
105 	} bo_cache;
106 
107 	u32 num_labels;
108 	struct vc4_label {
109 		const char *name;
110 		u32 num_allocated;
111 		u32 size_allocated;
112 	} *bo_labels;
113 
114 	/* Protects bo_cache and bo_labels. */
115 	struct mutex bo_lock;
116 
117 	/* Purgeable BO pool. All BOs in this pool can have their memory
118 	 * reclaimed if the driver is unable to allocate new BOs. We also
119 	 * keep stats related to the purge mechanism here.
120 	 */
121 	struct {
122 		struct list_head list;
123 		unsigned int num;
124 		size_t size;
125 		unsigned int purged_num;
126 		size_t purged_size;
127 		struct mutex lock;
128 	} purgeable;
129 
130 	uint64_t dma_fence_context;
131 
132 	/* Sequence number for the last job queued in bin_job_list.
133 	 * Starts at 0 (no jobs emitted).
134 	 */
135 	uint64_t emit_seqno;
136 
137 	/* Sequence number for the last completed job on the GPU.
138 	 * Starts at 0 (no jobs completed).
139 	 */
140 	uint64_t finished_seqno;
141 
142 	/* List of all struct vc4_exec_info for jobs to be executed in
143 	 * the binner.  The first job in the list is the one currently
144 	 * programmed into ct0ca for execution.
145 	 */
146 	struct list_head bin_job_list;
147 
148 	/* List of all struct vc4_exec_info for jobs that have
149 	 * completed binning and are ready for rendering.  The first
150 	 * job in the list is the one currently programmed into ct1ca
151 	 * for execution.
152 	 */
153 	struct list_head render_job_list;
154 
155 	/* List of the finished vc4_exec_infos waiting to be freed by
156 	 * job_done_work.
157 	 */
158 	struct list_head job_done_list;
159 	/* Spinlock used to synchronize the job_list and seqno
160 	 * accesses between the IRQ handler and GEM ioctls.
161 	 */
162 	spinlock_t job_lock;
163 	wait_queue_head_t job_wait_queue;
164 	struct work_struct job_done_work;
165 
166 	/* Used to track the active perfmon if any. Access to this field is
167 	 * protected by job_lock.
168 	 */
169 	struct vc4_perfmon *active_perfmon;
170 
171 	/* List of struct vc4_seqno_cb for callbacks to be made from a
172 	 * workqueue when the given seqno is passed.
173 	 */
174 	struct list_head seqno_cb_list;
175 
176 	/* The memory used for storing binner tile alloc, tile state,
177 	 * and overflow memory allocations.  This is freed when V3D
178 	 * powers down.
179 	 */
180 	struct vc4_bo *bin_bo;
181 
182 	/* Size of blocks allocated within bin_bo. */
183 	uint32_t bin_alloc_size;
184 
185 	/* Bitmask of the bin_alloc_size chunks in bin_bo that are
186 	 * used.
187 	 */
188 	uint32_t bin_alloc_used;
189 
190 	/* Bitmask of the current bin_alloc used for overflow memory. */
191 	uint32_t bin_alloc_overflow;
192 
193 	/* Incremented when an underrun error happened after an atomic commit.
194 	 * This is particularly useful to detect when a specific modeset is too
195 	 * demanding in term of memory or HVS bandwidth which is hard to guess
196 	 * at atomic check time.
197 	 */
198 	atomic_t underrun;
199 
200 	struct work_struct overflow_mem_work;
201 
202 	int power_refcount;
203 
204 	/* Set to true when the load tracker is supported. */
205 	bool load_tracker_available;
206 
207 	/* Set to true when the load tracker is active. */
208 	bool load_tracker_enabled;
209 
210 	/* Mutex controlling the power refcount. */
211 	struct mutex power_lock;
212 
213 	struct {
214 		struct timer_list timer;
215 		struct work_struct reset_work;
216 	} hangcheck;
217 
218 	struct semaphore async_modeset;
219 
220 	struct drm_modeset_lock ctm_state_lock;
221 	struct drm_private_obj ctm_manager;
222 	struct drm_private_obj hvs_channels;
223 	struct drm_private_obj load_tracker;
224 
225 	/* List of vc4_debugfs_info_entry for adding to debugfs once
226 	 * the minor is available (after drm_dev_register()).
227 	 */
228 	struct list_head debugfs_list;
229 
230 	/* Mutex for binner bo allocation. */
231 	struct mutex bin_bo_lock;
232 	/* Reference count for our binner bo. */
233 	struct kref bin_bo_kref;
234 };
235 
236 static inline struct vc4_dev *
237 to_vc4_dev(struct drm_device *dev)
238 {
239 	return container_of(dev, struct vc4_dev, base);
240 }
241 
242 struct vc4_bo {
243 	struct drm_gem_cma_object base;
244 
245 	/* seqno of the last job to render using this BO. */
246 	uint64_t seqno;
247 
248 	/* seqno of the last job to use the RCL to write to this BO.
249 	 *
250 	 * Note that this doesn't include binner overflow memory
251 	 * writes.
252 	 */
253 	uint64_t write_seqno;
254 
255 	bool t_format;
256 
257 	/* List entry for the BO's position in either
258 	 * vc4_exec_info->unref_list or vc4_dev->bo_cache.time_list
259 	 */
260 	struct list_head unref_head;
261 
262 	/* Time in jiffies when the BO was put in vc4->bo_cache. */
263 	unsigned long free_time;
264 
265 	/* List entry for the BO's position in vc4_dev->bo_cache.size_list */
266 	struct list_head size_head;
267 
268 	/* Struct for shader validation state, if created by
269 	 * DRM_IOCTL_VC4_CREATE_SHADER_BO.
270 	 */
271 	struct vc4_validated_shader_info *validated_shader;
272 
273 	/* One of enum vc4_kernel_bo_type, or VC4_BO_TYPE_COUNT + i
274 	 * for user-allocated labels.
275 	 */
276 	int label;
277 
278 	/* Count the number of active users. This is needed to determine
279 	 * whether we can move the BO to the purgeable list or not (when the BO
280 	 * is used by the GPU or the display engine we can't purge it).
281 	 */
282 	refcount_t usecnt;
283 
284 	/* Store purgeable/purged state here */
285 	u32 madv;
286 	struct mutex madv_lock;
287 };
288 
289 static inline struct vc4_bo *
290 to_vc4_bo(struct drm_gem_object *bo)
291 {
292 	return container_of(to_drm_gem_cma_obj(bo), struct vc4_bo, base);
293 }
294 
295 struct vc4_fence {
296 	struct dma_fence base;
297 	struct drm_device *dev;
298 	/* vc4 seqno for signaled() test */
299 	uint64_t seqno;
300 };
301 
302 static inline struct vc4_fence *
303 to_vc4_fence(struct dma_fence *fence)
304 {
305 	return container_of(fence, struct vc4_fence, base);
306 }
307 
308 struct vc4_seqno_cb {
309 	struct work_struct work;
310 	uint64_t seqno;
311 	void (*func)(struct vc4_seqno_cb *cb);
312 };
313 
314 struct vc4_v3d {
315 	struct vc4_dev *vc4;
316 	struct platform_device *pdev;
317 	void __iomem *regs;
318 	struct clk *clk;
319 	struct debugfs_regset32 regset;
320 };
321 
322 struct vc4_hvs {
323 	struct platform_device *pdev;
324 	void __iomem *regs;
325 	u32 __iomem *dlist;
326 
327 	struct clk *core_clk;
328 
329 	/* Memory manager for CRTCs to allocate space in the display
330 	 * list.  Units are dwords.
331 	 */
332 	struct drm_mm dlist_mm;
333 	/* Memory manager for the LBM memory used by HVS scaling. */
334 	struct drm_mm lbm_mm;
335 	spinlock_t mm_lock;
336 
337 	struct drm_mm_node mitchell_netravali_filter;
338 
339 	struct debugfs_regset32 regset;
340 
341 	/* HVS version 5 flag, therefore requires updated dlist structures */
342 	bool hvs5;
343 };
344 
345 struct vc4_plane {
346 	struct drm_plane base;
347 };
348 
349 static inline struct vc4_plane *
350 to_vc4_plane(struct drm_plane *plane)
351 {
352 	return container_of(plane, struct vc4_plane, base);
353 }
354 
355 enum vc4_scaling_mode {
356 	VC4_SCALING_NONE,
357 	VC4_SCALING_TPZ,
358 	VC4_SCALING_PPF,
359 };
360 
361 struct vc4_plane_state {
362 	struct drm_plane_state base;
363 	/* System memory copy of the display list for this element, computed
364 	 * at atomic_check time.
365 	 */
366 	u32 *dlist;
367 	u32 dlist_size; /* Number of dwords allocated for the display list */
368 	u32 dlist_count; /* Number of used dwords in the display list. */
369 
370 	/* Offset in the dlist to various words, for pageflip or
371 	 * cursor updates.
372 	 */
373 	u32 pos0_offset;
374 	u32 pos2_offset;
375 	u32 ptr0_offset;
376 	u32 lbm_offset;
377 
378 	/* Offset where the plane's dlist was last stored in the
379 	 * hardware at vc4_crtc_atomic_flush() time.
380 	 */
381 	u32 __iomem *hw_dlist;
382 
383 	/* Clipped coordinates of the plane on the display. */
384 	int crtc_x, crtc_y, crtc_w, crtc_h;
385 	/* Clipped area being scanned from in the FB. */
386 	u32 src_x, src_y;
387 
388 	u32 src_w[2], src_h[2];
389 
390 	/* Scaling selection for the RGB/Y plane and the Cb/Cr planes. */
391 	enum vc4_scaling_mode x_scaling[2], y_scaling[2];
392 	bool is_unity;
393 	bool is_yuv;
394 
395 	/* Offset to start scanning out from the start of the plane's
396 	 * BO.
397 	 */
398 	u32 offsets[3];
399 
400 	/* Our allocation in LBM for temporary storage during scaling. */
401 	struct drm_mm_node lbm;
402 
403 	/* Set when the plane has per-pixel alpha content or does not cover
404 	 * the entire screen. This is a hint to the CRTC that it might need
405 	 * to enable background color fill.
406 	 */
407 	bool needs_bg_fill;
408 
409 	/* Mark the dlist as initialized. Useful to avoid initializing it twice
410 	 * when async update is not possible.
411 	 */
412 	bool dlist_initialized;
413 
414 	/* Load of this plane on the HVS block. The load is expressed in HVS
415 	 * cycles/sec.
416 	 */
417 	u64 hvs_load;
418 
419 	/* Memory bandwidth needed for this plane. This is expressed in
420 	 * bytes/sec.
421 	 */
422 	u64 membus_load;
423 };
424 
425 static inline struct vc4_plane_state *
426 to_vc4_plane_state(struct drm_plane_state *state)
427 {
428 	return container_of(state, struct vc4_plane_state, base);
429 }
430 
431 enum vc4_encoder_type {
432 	VC4_ENCODER_TYPE_NONE,
433 	VC4_ENCODER_TYPE_HDMI0,
434 	VC4_ENCODER_TYPE_HDMI1,
435 	VC4_ENCODER_TYPE_VEC,
436 	VC4_ENCODER_TYPE_DSI0,
437 	VC4_ENCODER_TYPE_DSI1,
438 	VC4_ENCODER_TYPE_SMI,
439 	VC4_ENCODER_TYPE_DPI,
440 };
441 
442 struct vc4_encoder {
443 	struct drm_encoder base;
444 	enum vc4_encoder_type type;
445 	u32 clock_select;
446 
447 	void (*pre_crtc_configure)(struct drm_encoder *encoder);
448 	void (*pre_crtc_enable)(struct drm_encoder *encoder);
449 	void (*post_crtc_enable)(struct drm_encoder *encoder);
450 
451 	void (*post_crtc_disable)(struct drm_encoder *encoder);
452 	void (*post_crtc_powerdown)(struct drm_encoder *encoder);
453 };
454 
455 static inline struct vc4_encoder *
456 to_vc4_encoder(struct drm_encoder *encoder)
457 {
458 	return container_of(encoder, struct vc4_encoder, base);
459 }
460 
461 struct vc4_crtc_data {
462 	/* Bitmask of channels (FIFOs) of the HVS that the output can source from */
463 	unsigned int hvs_available_channels;
464 
465 	/* Which output of the HVS this pixelvalve sources from. */
466 	int hvs_output;
467 };
468 
469 struct vc4_pv_data {
470 	struct vc4_crtc_data	base;
471 
472 	/* Depth of the PixelValve FIFO in bytes */
473 	unsigned int fifo_depth;
474 
475 	/* Number of pixels output per clock period */
476 	u8 pixels_per_clock;
477 
478 	enum vc4_encoder_type encoder_types[4];
479 	const char *debugfs_name;
480 
481 };
482 
483 struct vc4_crtc {
484 	struct drm_crtc base;
485 	struct platform_device *pdev;
486 	const struct vc4_crtc_data *data;
487 	void __iomem *regs;
488 
489 	/* Timestamp at start of vblank irq - unaffected by lock delays. */
490 	ktime_t t_vblank;
491 
492 	u8 lut_r[256];
493 	u8 lut_g[256];
494 	u8 lut_b[256];
495 
496 	struct drm_pending_vblank_event *event;
497 
498 	struct debugfs_regset32 regset;
499 };
500 
501 static inline struct vc4_crtc *
502 to_vc4_crtc(struct drm_crtc *crtc)
503 {
504 	return container_of(crtc, struct vc4_crtc, base);
505 }
506 
507 static inline const struct vc4_crtc_data *
508 vc4_crtc_to_vc4_crtc_data(const struct vc4_crtc *crtc)
509 {
510 	return crtc->data;
511 }
512 
513 static inline const struct vc4_pv_data *
514 vc4_crtc_to_vc4_pv_data(const struct vc4_crtc *crtc)
515 {
516 	const struct vc4_crtc_data *data = vc4_crtc_to_vc4_crtc_data(crtc);
517 
518 	return container_of(data, struct vc4_pv_data, base);
519 }
520 
521 struct vc4_crtc_state {
522 	struct drm_crtc_state base;
523 	/* Dlist area for this CRTC configuration. */
524 	struct drm_mm_node mm;
525 	bool feed_txp;
526 	bool txp_armed;
527 	unsigned int assigned_channel;
528 
529 	struct {
530 		unsigned int left;
531 		unsigned int right;
532 		unsigned int top;
533 		unsigned int bottom;
534 	} margins;
535 
536 	/* Transitional state below, only valid during atomic commits */
537 	bool update_muxing;
538 };
539 
540 #define VC4_HVS_CHANNEL_DISABLED ((unsigned int)-1)
541 
542 static inline struct vc4_crtc_state *
543 to_vc4_crtc_state(struct drm_crtc_state *crtc_state)
544 {
545 	return container_of(crtc_state, struct vc4_crtc_state, base);
546 }
547 
548 #define V3D_READ(offset) readl(vc4->v3d->regs + offset)
549 #define V3D_WRITE(offset, val) writel(val, vc4->v3d->regs + offset)
550 #define HVS_READ(offset) readl(vc4->hvs->regs + offset)
551 #define HVS_WRITE(offset, val) writel(val, vc4->hvs->regs + offset)
552 
553 #define VC4_REG32(reg) { .name = #reg, .offset = reg }
554 
555 struct vc4_exec_info {
556 	/* Sequence number for this bin/render job. */
557 	uint64_t seqno;
558 
559 	/* Latest write_seqno of any BO that binning depends on. */
560 	uint64_t bin_dep_seqno;
561 
562 	struct dma_fence *fence;
563 
564 	/* Last current addresses the hardware was processing when the
565 	 * hangcheck timer checked on us.
566 	 */
567 	uint32_t last_ct0ca, last_ct1ca;
568 
569 	/* Kernel-space copy of the ioctl arguments */
570 	struct drm_vc4_submit_cl *args;
571 
572 	/* This is the array of BOs that were looked up at the start of exec.
573 	 * Command validation will use indices into this array.
574 	 */
575 	struct drm_gem_cma_object **bo;
576 	uint32_t bo_count;
577 
578 	/* List of BOs that are being written by the RCL.  Other than
579 	 * the binner temporary storage, this is all the BOs written
580 	 * by the job.
581 	 */
582 	struct drm_gem_cma_object *rcl_write_bo[4];
583 	uint32_t rcl_write_bo_count;
584 
585 	/* Pointers for our position in vc4->job_list */
586 	struct list_head head;
587 
588 	/* List of other BOs used in the job that need to be released
589 	 * once the job is complete.
590 	 */
591 	struct list_head unref_list;
592 
593 	/* Current unvalidated indices into @bo loaded by the non-hardware
594 	 * VC4_PACKET_GEM_HANDLES.
595 	 */
596 	uint32_t bo_index[2];
597 
598 	/* This is the BO where we store the validated command lists, shader
599 	 * records, and uniforms.
600 	 */
601 	struct drm_gem_cma_object *exec_bo;
602 
603 	/**
604 	 * This tracks the per-shader-record state (packet 64) that
605 	 * determines the length of the shader record and the offset
606 	 * it's expected to be found at.  It gets read in from the
607 	 * command lists.
608 	 */
609 	struct vc4_shader_state {
610 		uint32_t addr;
611 		/* Maximum vertex index referenced by any primitive using this
612 		 * shader state.
613 		 */
614 		uint32_t max_index;
615 	} *shader_state;
616 
617 	/** How many shader states the user declared they were using. */
618 	uint32_t shader_state_size;
619 	/** How many shader state records the validator has seen. */
620 	uint32_t shader_state_count;
621 
622 	bool found_tile_binning_mode_config_packet;
623 	bool found_start_tile_binning_packet;
624 	bool found_increment_semaphore_packet;
625 	bool found_flush;
626 	uint8_t bin_tiles_x, bin_tiles_y;
627 	/* Physical address of the start of the tile alloc array
628 	 * (where each tile's binned CL will start)
629 	 */
630 	uint32_t tile_alloc_offset;
631 	/* Bitmask of which binner slots are freed when this job completes. */
632 	uint32_t bin_slots;
633 
634 	/**
635 	 * Computed addresses pointing into exec_bo where we start the
636 	 * bin thread (ct0) and render thread (ct1).
637 	 */
638 	uint32_t ct0ca, ct0ea;
639 	uint32_t ct1ca, ct1ea;
640 
641 	/* Pointer to the unvalidated bin CL (if present). */
642 	void *bin_u;
643 
644 	/* Pointers to the shader recs.  These paddr gets incremented as CL
645 	 * packets are relocated in validate_gl_shader_state, and the vaddrs
646 	 * (u and v) get incremented and size decremented as the shader recs
647 	 * themselves are validated.
648 	 */
649 	void *shader_rec_u;
650 	void *shader_rec_v;
651 	uint32_t shader_rec_p;
652 	uint32_t shader_rec_size;
653 
654 	/* Pointers to the uniform data.  These pointers are incremented, and
655 	 * size decremented, as each batch of uniforms is uploaded.
656 	 */
657 	void *uniforms_u;
658 	void *uniforms_v;
659 	uint32_t uniforms_p;
660 	uint32_t uniforms_size;
661 
662 	/* Pointer to a performance monitor object if the user requested it,
663 	 * NULL otherwise.
664 	 */
665 	struct vc4_perfmon *perfmon;
666 
667 	/* Whether the exec has taken a reference to the binner BO, which should
668 	 * happen with a VC4_PACKET_TILE_BINNING_MODE_CONFIG packet.
669 	 */
670 	bool bin_bo_used;
671 };
672 
673 /* Per-open file private data. Any driver-specific resource that has to be
674  * released when the DRM file is closed should be placed here.
675  */
676 struct vc4_file {
677 	struct {
678 		struct idr idr;
679 		struct mutex lock;
680 	} perfmon;
681 
682 	bool bin_bo_used;
683 };
684 
685 static inline struct vc4_exec_info *
686 vc4_first_bin_job(struct vc4_dev *vc4)
687 {
688 	return list_first_entry_or_null(&vc4->bin_job_list,
689 					struct vc4_exec_info, head);
690 }
691 
692 static inline struct vc4_exec_info *
693 vc4_first_render_job(struct vc4_dev *vc4)
694 {
695 	return list_first_entry_or_null(&vc4->render_job_list,
696 					struct vc4_exec_info, head);
697 }
698 
699 static inline struct vc4_exec_info *
700 vc4_last_render_job(struct vc4_dev *vc4)
701 {
702 	if (list_empty(&vc4->render_job_list))
703 		return NULL;
704 	return list_last_entry(&vc4->render_job_list,
705 			       struct vc4_exec_info, head);
706 }
707 
708 /**
709  * struct vc4_texture_sample_info - saves the offsets into the UBO for texture
710  * setup parameters.
711  *
712  * This will be used at draw time to relocate the reference to the texture
713  * contents in p0, and validate that the offset combined with
714  * width/height/stride/etc. from p1 and p2/p3 doesn't sample outside the BO.
715  * Note that the hardware treats unprovided config parameters as 0, so not all
716  * of them need to be set up for every texure sample, and we'll store ~0 as
717  * the offset to mark the unused ones.
718  *
719  * See the VC4 3D architecture guide page 41 ("Texture and Memory Lookup Unit
720  * Setup") for definitions of the texture parameters.
721  */
722 struct vc4_texture_sample_info {
723 	bool is_direct;
724 	uint32_t p_offset[4];
725 };
726 
727 /**
728  * struct vc4_validated_shader_info - information about validated shaders that
729  * needs to be used from command list validation.
730  *
731  * For a given shader, each time a shader state record references it, we need
732  * to verify that the shader doesn't read more uniforms than the shader state
733  * record's uniform BO pointer can provide, and we need to apply relocations
734  * and validate the shader state record's uniforms that define the texture
735  * samples.
736  */
737 struct vc4_validated_shader_info {
738 	uint32_t uniforms_size;
739 	uint32_t uniforms_src_size;
740 	uint32_t num_texture_samples;
741 	struct vc4_texture_sample_info *texture_samples;
742 
743 	uint32_t num_uniform_addr_offsets;
744 	uint32_t *uniform_addr_offsets;
745 
746 	bool is_threaded;
747 };
748 
749 /**
750  * __wait_for - magic wait macro
751  *
752  * Macro to help avoid open coding check/wait/timeout patterns. Note that it's
753  * important that we check the condition again after having timed out, since the
754  * timeout could be due to preemption or similar and we've never had a chance to
755  * check the condition before the timeout.
756  */
757 #define __wait_for(OP, COND, US, Wmin, Wmax) ({ \
758 	const ktime_t end__ = ktime_add_ns(ktime_get_raw(), 1000ll * (US)); \
759 	long wait__ = (Wmin); /* recommended min for usleep is 10 us */	\
760 	int ret__;							\
761 	might_sleep();							\
762 	for (;;) {							\
763 		const bool expired__ = ktime_after(ktime_get_raw(), end__); \
764 		OP;							\
765 		/* Guarantee COND check prior to timeout */		\
766 		barrier();						\
767 		if (COND) {						\
768 			ret__ = 0;					\
769 			break;						\
770 		}							\
771 		if (expired__) {					\
772 			ret__ = -ETIMEDOUT;				\
773 			break;						\
774 		}							\
775 		usleep_range(wait__, wait__ * 2);			\
776 		if (wait__ < (Wmax))					\
777 			wait__ <<= 1;					\
778 	}								\
779 	ret__;								\
780 })
781 
782 #define _wait_for(COND, US, Wmin, Wmax)	__wait_for(, (COND), (US), (Wmin), \
783 						   (Wmax))
784 #define wait_for(COND, MS)		_wait_for((COND), (MS) * 1000, 10, 1000)
785 
786 /* vc4_bo.c */
787 struct drm_gem_object *vc4_create_object(struct drm_device *dev, size_t size);
788 void vc4_free_object(struct drm_gem_object *gem_obj);
789 struct vc4_bo *vc4_bo_create(struct drm_device *dev, size_t size,
790 			     bool from_cache, enum vc4_kernel_bo_type type);
791 int vc4_dumb_create(struct drm_file *file_priv,
792 		    struct drm_device *dev,
793 		    struct drm_mode_create_dumb *args);
794 struct dma_buf *vc4_prime_export(struct drm_gem_object *obj, int flags);
795 int vc4_create_bo_ioctl(struct drm_device *dev, void *data,
796 			struct drm_file *file_priv);
797 int vc4_create_shader_bo_ioctl(struct drm_device *dev, void *data,
798 			       struct drm_file *file_priv);
799 int vc4_mmap_bo_ioctl(struct drm_device *dev, void *data,
800 		      struct drm_file *file_priv);
801 int vc4_set_tiling_ioctl(struct drm_device *dev, void *data,
802 			 struct drm_file *file_priv);
803 int vc4_get_tiling_ioctl(struct drm_device *dev, void *data,
804 			 struct drm_file *file_priv);
805 int vc4_get_hang_state_ioctl(struct drm_device *dev, void *data,
806 			     struct drm_file *file_priv);
807 int vc4_label_bo_ioctl(struct drm_device *dev, void *data,
808 		       struct drm_file *file_priv);
809 int vc4_mmap(struct file *filp, struct vm_area_struct *vma);
810 int vc4_prime_mmap(struct drm_gem_object *obj, struct vm_area_struct *vma);
811 struct drm_gem_object *vc4_prime_import_sg_table(struct drm_device *dev,
812 						 struct dma_buf_attachment *attach,
813 						 struct sg_table *sgt);
814 int vc4_prime_vmap(struct drm_gem_object *obj, struct dma_buf_map *map);
815 int vc4_bo_cache_init(struct drm_device *dev);
816 int vc4_bo_inc_usecnt(struct vc4_bo *bo);
817 void vc4_bo_dec_usecnt(struct vc4_bo *bo);
818 void vc4_bo_add_to_purgeable_pool(struct vc4_bo *bo);
819 void vc4_bo_remove_from_purgeable_pool(struct vc4_bo *bo);
820 
821 /* vc4_crtc.c */
822 extern struct platform_driver vc4_crtc_driver;
823 int vc4_crtc_disable_at_boot(struct drm_crtc *crtc);
824 int vc4_crtc_init(struct drm_device *drm, struct vc4_crtc *vc4_crtc,
825 		  const struct drm_crtc_funcs *crtc_funcs,
826 		  const struct drm_crtc_helper_funcs *crtc_helper_funcs);
827 void vc4_crtc_destroy(struct drm_crtc *crtc);
828 int vc4_page_flip(struct drm_crtc *crtc,
829 		  struct drm_framebuffer *fb,
830 		  struct drm_pending_vblank_event *event,
831 		  uint32_t flags,
832 		  struct drm_modeset_acquire_ctx *ctx);
833 struct drm_crtc_state *vc4_crtc_duplicate_state(struct drm_crtc *crtc);
834 void vc4_crtc_destroy_state(struct drm_crtc *crtc,
835 			    struct drm_crtc_state *state);
836 void vc4_crtc_reset(struct drm_crtc *crtc);
837 void vc4_crtc_handle_vblank(struct vc4_crtc *crtc);
838 void vc4_crtc_get_margins(struct drm_crtc_state *state,
839 			  unsigned int *right, unsigned int *left,
840 			  unsigned int *top, unsigned int *bottom);
841 
842 /* vc4_debugfs.c */
843 void vc4_debugfs_init(struct drm_minor *minor);
844 #ifdef CONFIG_DEBUG_FS
845 void vc4_debugfs_add_file(struct drm_device *drm,
846 			  const char *filename,
847 			  int (*show)(struct seq_file*, void*),
848 			  void *data);
849 void vc4_debugfs_add_regset32(struct drm_device *drm,
850 			      const char *filename,
851 			      struct debugfs_regset32 *regset);
852 #else
853 static inline void vc4_debugfs_add_file(struct drm_device *drm,
854 					const char *filename,
855 					int (*show)(struct seq_file*, void*),
856 					void *data)
857 {
858 }
859 
860 static inline void vc4_debugfs_add_regset32(struct drm_device *drm,
861 					    const char *filename,
862 					    struct debugfs_regset32 *regset)
863 {
864 }
865 #endif
866 
867 /* vc4_drv.c */
868 void __iomem *vc4_ioremap_regs(struct platform_device *dev, int index);
869 
870 /* vc4_dpi.c */
871 extern struct platform_driver vc4_dpi_driver;
872 
873 /* vc4_dsi.c */
874 extern struct platform_driver vc4_dsi_driver;
875 
876 /* vc4_fence.c */
877 extern const struct dma_fence_ops vc4_fence_ops;
878 
879 /* vc4_gem.c */
880 int vc4_gem_init(struct drm_device *dev);
881 int vc4_submit_cl_ioctl(struct drm_device *dev, void *data,
882 			struct drm_file *file_priv);
883 int vc4_wait_seqno_ioctl(struct drm_device *dev, void *data,
884 			 struct drm_file *file_priv);
885 int vc4_wait_bo_ioctl(struct drm_device *dev, void *data,
886 		      struct drm_file *file_priv);
887 void vc4_submit_next_bin_job(struct drm_device *dev);
888 void vc4_submit_next_render_job(struct drm_device *dev);
889 void vc4_move_job_to_render(struct drm_device *dev, struct vc4_exec_info *exec);
890 int vc4_wait_for_seqno(struct drm_device *dev, uint64_t seqno,
891 		       uint64_t timeout_ns, bool interruptible);
892 void vc4_job_handle_completed(struct vc4_dev *vc4);
893 int vc4_queue_seqno_cb(struct drm_device *dev,
894 		       struct vc4_seqno_cb *cb, uint64_t seqno,
895 		       void (*func)(struct vc4_seqno_cb *cb));
896 int vc4_gem_madvise_ioctl(struct drm_device *dev, void *data,
897 			  struct drm_file *file_priv);
898 
899 /* vc4_hdmi.c */
900 extern struct platform_driver vc4_hdmi_driver;
901 
902 /* vc4_vec.c */
903 extern struct platform_driver vc4_vec_driver;
904 
905 /* vc4_txp.c */
906 extern struct platform_driver vc4_txp_driver;
907 
908 /* vc4_irq.c */
909 irqreturn_t vc4_irq(int irq, void *arg);
910 void vc4_irq_preinstall(struct drm_device *dev);
911 int vc4_irq_postinstall(struct drm_device *dev);
912 void vc4_irq_uninstall(struct drm_device *dev);
913 void vc4_irq_reset(struct drm_device *dev);
914 
915 /* vc4_hvs.c */
916 extern struct platform_driver vc4_hvs_driver;
917 void vc4_hvs_stop_channel(struct drm_device *dev, unsigned int output);
918 int vc4_hvs_get_fifo_from_output(struct drm_device *dev, unsigned int output);
919 int vc4_hvs_atomic_check(struct drm_crtc *crtc, struct drm_crtc_state *state);
920 void vc4_hvs_atomic_enable(struct drm_crtc *crtc, struct drm_crtc_state *old_state);
921 void vc4_hvs_atomic_disable(struct drm_crtc *crtc, struct drm_crtc_state *old_state);
922 void vc4_hvs_atomic_flush(struct drm_crtc *crtc,
923 			  struct drm_atomic_state *state);
924 void vc4_hvs_dump_state(struct drm_device *dev);
925 void vc4_hvs_unmask_underrun(struct drm_device *dev, int channel);
926 void vc4_hvs_mask_underrun(struct drm_device *dev, int channel);
927 
928 /* vc4_kms.c */
929 int vc4_kms_load(struct drm_device *dev);
930 
931 /* vc4_plane.c */
932 struct drm_plane *vc4_plane_init(struct drm_device *dev,
933 				 enum drm_plane_type type);
934 int vc4_plane_create_additional_planes(struct drm_device *dev);
935 u32 vc4_plane_write_dlist(struct drm_plane *plane, u32 __iomem *dlist);
936 u32 vc4_plane_dlist_size(const struct drm_plane_state *state);
937 void vc4_plane_async_set_fb(struct drm_plane *plane,
938 			    struct drm_framebuffer *fb);
939 
940 /* vc4_v3d.c */
941 extern struct platform_driver vc4_v3d_driver;
942 extern const struct of_device_id vc4_v3d_dt_match[];
943 int vc4_v3d_get_bin_slot(struct vc4_dev *vc4);
944 int vc4_v3d_bin_bo_get(struct vc4_dev *vc4, bool *used);
945 void vc4_v3d_bin_bo_put(struct vc4_dev *vc4);
946 int vc4_v3d_pm_get(struct vc4_dev *vc4);
947 void vc4_v3d_pm_put(struct vc4_dev *vc4);
948 
949 /* vc4_validate.c */
950 int
951 vc4_validate_bin_cl(struct drm_device *dev,
952 		    void *validated,
953 		    void *unvalidated,
954 		    struct vc4_exec_info *exec);
955 
956 int
957 vc4_validate_shader_recs(struct drm_device *dev, struct vc4_exec_info *exec);
958 
959 struct drm_gem_cma_object *vc4_use_bo(struct vc4_exec_info *exec,
960 				      uint32_t hindex);
961 
962 int vc4_get_rcl(struct drm_device *dev, struct vc4_exec_info *exec);
963 
964 bool vc4_check_tex_size(struct vc4_exec_info *exec,
965 			struct drm_gem_cma_object *fbo,
966 			uint32_t offset, uint8_t tiling_format,
967 			uint32_t width, uint32_t height, uint8_t cpp);
968 
969 /* vc4_validate_shader.c */
970 struct vc4_validated_shader_info *
971 vc4_validate_shader(struct drm_gem_cma_object *shader_obj);
972 
973 /* vc4_perfmon.c */
974 void vc4_perfmon_get(struct vc4_perfmon *perfmon);
975 void vc4_perfmon_put(struct vc4_perfmon *perfmon);
976 void vc4_perfmon_start(struct vc4_dev *vc4, struct vc4_perfmon *perfmon);
977 void vc4_perfmon_stop(struct vc4_dev *vc4, struct vc4_perfmon *perfmon,
978 		      bool capture);
979 struct vc4_perfmon *vc4_perfmon_find(struct vc4_file *vc4file, int id);
980 void vc4_perfmon_open_file(struct vc4_file *vc4file);
981 void vc4_perfmon_close_file(struct vc4_file *vc4file);
982 int vc4_perfmon_create_ioctl(struct drm_device *dev, void *data,
983 			     struct drm_file *file_priv);
984 int vc4_perfmon_destroy_ioctl(struct drm_device *dev, void *data,
985 			      struct drm_file *file_priv);
986 int vc4_perfmon_get_values_ioctl(struct drm_device *dev, void *data,
987 				 struct drm_file *file_priv);
988 
989 #endif /* _VC4_DRV_H_ */
990