xref: /linux/drivers/gpu/drm/vc4/vc4_drv.h (revision dec1c62e91ba268ab2a6e339d4d7a59287d5eba1)
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 	unsigned int irq;
78 
79 	struct vc4_hvs *hvs;
80 	struct vc4_v3d *v3d;
81 	struct vc4_dpi *dpi;
82 	struct vc4_vec *vec;
83 	struct vc4_txp *txp;
84 
85 	struct vc4_hang_state *hang_state;
86 
87 	/* The kernel-space BO cache.  Tracks buffers that have been
88 	 * unreferenced by all other users (refcounts of 0!) but not
89 	 * yet freed, so we can do cheap allocations.
90 	 */
91 	struct vc4_bo_cache {
92 		/* Array of list heads for entries in the BO cache,
93 		 * based on number of pages, so we can do O(1) lookups
94 		 * in the cache when allocating.
95 		 */
96 		struct list_head *size_list;
97 		uint32_t size_list_size;
98 
99 		/* List of all BOs in the cache, ordered by age, so we
100 		 * can do O(1) lookups when trying to free old
101 		 * buffers.
102 		 */
103 		struct list_head time_list;
104 		struct work_struct time_work;
105 		struct timer_list time_timer;
106 	} bo_cache;
107 
108 	u32 num_labels;
109 	struct vc4_label {
110 		const char *name;
111 		u32 num_allocated;
112 		u32 size_allocated;
113 	} *bo_labels;
114 
115 	/* Protects bo_cache and bo_labels. */
116 	struct mutex bo_lock;
117 
118 	/* Purgeable BO pool. All BOs in this pool can have their memory
119 	 * reclaimed if the driver is unable to allocate new BOs. We also
120 	 * keep stats related to the purge mechanism here.
121 	 */
122 	struct {
123 		struct list_head list;
124 		unsigned int num;
125 		size_t size;
126 		unsigned int purged_num;
127 		size_t purged_size;
128 		struct mutex lock;
129 	} purgeable;
130 
131 	uint64_t dma_fence_context;
132 
133 	/* Sequence number for the last job queued in bin_job_list.
134 	 * Starts at 0 (no jobs emitted).
135 	 */
136 	uint64_t emit_seqno;
137 
138 	/* Sequence number for the last completed job on the GPU.
139 	 * Starts at 0 (no jobs completed).
140 	 */
141 	uint64_t finished_seqno;
142 
143 	/* List of all struct vc4_exec_info for jobs to be executed in
144 	 * the binner.  The first job in the list is the one currently
145 	 * programmed into ct0ca for execution.
146 	 */
147 	struct list_head bin_job_list;
148 
149 	/* List of all struct vc4_exec_info for jobs that have
150 	 * completed binning and are ready for rendering.  The first
151 	 * job in the list is the one currently programmed into ct1ca
152 	 * for execution.
153 	 */
154 	struct list_head render_job_list;
155 
156 	/* List of the finished vc4_exec_infos waiting to be freed by
157 	 * job_done_work.
158 	 */
159 	struct list_head job_done_list;
160 	/* Spinlock used to synchronize the job_list and seqno
161 	 * accesses between the IRQ handler and GEM ioctls.
162 	 */
163 	spinlock_t job_lock;
164 	wait_queue_head_t job_wait_queue;
165 	struct work_struct job_done_work;
166 
167 	/* Used to track the active perfmon if any. Access to this field is
168 	 * protected by job_lock.
169 	 */
170 	struct vc4_perfmon *active_perfmon;
171 
172 	/* List of struct vc4_seqno_cb for callbacks to be made from a
173 	 * workqueue when the given seqno is passed.
174 	 */
175 	struct list_head seqno_cb_list;
176 
177 	/* The memory used for storing binner tile alloc, tile state,
178 	 * and overflow memory allocations.  This is freed when V3D
179 	 * powers down.
180 	 */
181 	struct vc4_bo *bin_bo;
182 
183 	/* Size of blocks allocated within bin_bo. */
184 	uint32_t bin_alloc_size;
185 
186 	/* Bitmask of the bin_alloc_size chunks in bin_bo that are
187 	 * used.
188 	 */
189 	uint32_t bin_alloc_used;
190 
191 	/* Bitmask of the current bin_alloc used for overflow memory. */
192 	uint32_t bin_alloc_overflow;
193 
194 	/* Incremented when an underrun error happened after an atomic commit.
195 	 * This is particularly useful to detect when a specific modeset is too
196 	 * demanding in term of memory or HVS bandwidth which is hard to guess
197 	 * at atomic check time.
198 	 */
199 	atomic_t underrun;
200 
201 	struct work_struct overflow_mem_work;
202 
203 	int power_refcount;
204 
205 	/* Set to true when the load tracker is active. */
206 	bool load_tracker_enabled;
207 
208 	/* Mutex controlling the power refcount. */
209 	struct mutex power_lock;
210 
211 	struct {
212 		struct timer_list timer;
213 		struct work_struct reset_work;
214 	} hangcheck;
215 
216 	struct drm_modeset_lock ctm_state_lock;
217 	struct drm_private_obj ctm_manager;
218 	struct drm_private_obj hvs_channels;
219 	struct drm_private_obj load_tracker;
220 
221 	/* List of vc4_debugfs_info_entry for adding to debugfs once
222 	 * the minor is available (after drm_dev_register()).
223 	 */
224 	struct list_head debugfs_list;
225 
226 	/* Mutex for binner bo allocation. */
227 	struct mutex bin_bo_lock;
228 	/* Reference count for our binner bo. */
229 	struct kref bin_bo_kref;
230 };
231 
232 static inline struct vc4_dev *
233 to_vc4_dev(struct drm_device *dev)
234 {
235 	return container_of(dev, struct vc4_dev, base);
236 }
237 
238 struct vc4_bo {
239 	struct drm_gem_cma_object base;
240 
241 	/* seqno of the last job to render using this BO. */
242 	uint64_t seqno;
243 
244 	/* seqno of the last job to use the RCL to write to this BO.
245 	 *
246 	 * Note that this doesn't include binner overflow memory
247 	 * writes.
248 	 */
249 	uint64_t write_seqno;
250 
251 	bool t_format;
252 
253 	/* List entry for the BO's position in either
254 	 * vc4_exec_info->unref_list or vc4_dev->bo_cache.time_list
255 	 */
256 	struct list_head unref_head;
257 
258 	/* Time in jiffies when the BO was put in vc4->bo_cache. */
259 	unsigned long free_time;
260 
261 	/* List entry for the BO's position in vc4_dev->bo_cache.size_list */
262 	struct list_head size_head;
263 
264 	/* Struct for shader validation state, if created by
265 	 * DRM_IOCTL_VC4_CREATE_SHADER_BO.
266 	 */
267 	struct vc4_validated_shader_info *validated_shader;
268 
269 	/* One of enum vc4_kernel_bo_type, or VC4_BO_TYPE_COUNT + i
270 	 * for user-allocated labels.
271 	 */
272 	int label;
273 
274 	/* Count the number of active users. This is needed to determine
275 	 * whether we can move the BO to the purgeable list or not (when the BO
276 	 * is used by the GPU or the display engine we can't purge it).
277 	 */
278 	refcount_t usecnt;
279 
280 	/* Store purgeable/purged state here */
281 	u32 madv;
282 	struct mutex madv_lock;
283 };
284 
285 static inline struct vc4_bo *
286 to_vc4_bo(struct drm_gem_object *bo)
287 {
288 	return container_of(to_drm_gem_cma_obj(bo), struct vc4_bo, base);
289 }
290 
291 struct vc4_fence {
292 	struct dma_fence base;
293 	struct drm_device *dev;
294 	/* vc4 seqno for signaled() test */
295 	uint64_t seqno;
296 };
297 
298 static inline struct vc4_fence *
299 to_vc4_fence(struct dma_fence *fence)
300 {
301 	return container_of(fence, struct vc4_fence, base);
302 }
303 
304 struct vc4_seqno_cb {
305 	struct work_struct work;
306 	uint64_t seqno;
307 	void (*func)(struct vc4_seqno_cb *cb);
308 };
309 
310 struct vc4_v3d {
311 	struct vc4_dev *vc4;
312 	struct platform_device *pdev;
313 	void __iomem *regs;
314 	struct clk *clk;
315 	struct debugfs_regset32 regset;
316 };
317 
318 struct vc4_hvs {
319 	struct platform_device *pdev;
320 	void __iomem *regs;
321 	u32 __iomem *dlist;
322 
323 	struct clk *core_clk;
324 
325 	/* Memory manager for CRTCs to allocate space in the display
326 	 * list.  Units are dwords.
327 	 */
328 	struct drm_mm dlist_mm;
329 	/* Memory manager for the LBM memory used by HVS scaling. */
330 	struct drm_mm lbm_mm;
331 	spinlock_t mm_lock;
332 
333 	struct drm_mm_node mitchell_netravali_filter;
334 
335 	struct debugfs_regset32 regset;
336 
337 	/* HVS version 5 flag, therefore requires updated dlist structures */
338 	bool hvs5;
339 };
340 
341 struct vc4_plane {
342 	struct drm_plane base;
343 };
344 
345 static inline struct vc4_plane *
346 to_vc4_plane(struct drm_plane *plane)
347 {
348 	return container_of(plane, struct vc4_plane, base);
349 }
350 
351 enum vc4_scaling_mode {
352 	VC4_SCALING_NONE,
353 	VC4_SCALING_TPZ,
354 	VC4_SCALING_PPF,
355 };
356 
357 struct vc4_plane_state {
358 	struct drm_plane_state base;
359 	/* System memory copy of the display list for this element, computed
360 	 * at atomic_check time.
361 	 */
362 	u32 *dlist;
363 	u32 dlist_size; /* Number of dwords allocated for the display list */
364 	u32 dlist_count; /* Number of used dwords in the display list. */
365 
366 	/* Offset in the dlist to various words, for pageflip or
367 	 * cursor updates.
368 	 */
369 	u32 pos0_offset;
370 	u32 pos2_offset;
371 	u32 ptr0_offset;
372 	u32 lbm_offset;
373 
374 	/* Offset where the plane's dlist was last stored in the
375 	 * hardware at vc4_crtc_atomic_flush() time.
376 	 */
377 	u32 __iomem *hw_dlist;
378 
379 	/* Clipped coordinates of the plane on the display. */
380 	int crtc_x, crtc_y, crtc_w, crtc_h;
381 	/* Clipped area being scanned from in the FB. */
382 	u32 src_x, src_y;
383 
384 	u32 src_w[2], src_h[2];
385 
386 	/* Scaling selection for the RGB/Y plane and the Cb/Cr planes. */
387 	enum vc4_scaling_mode x_scaling[2], y_scaling[2];
388 	bool is_unity;
389 	bool is_yuv;
390 
391 	/* Offset to start scanning out from the start of the plane's
392 	 * BO.
393 	 */
394 	u32 offsets[3];
395 
396 	/* Our allocation in LBM for temporary storage during scaling. */
397 	struct drm_mm_node lbm;
398 
399 	/* Set when the plane has per-pixel alpha content or does not cover
400 	 * the entire screen. This is a hint to the CRTC that it might need
401 	 * to enable background color fill.
402 	 */
403 	bool needs_bg_fill;
404 
405 	/* Mark the dlist as initialized. Useful to avoid initializing it twice
406 	 * when async update is not possible.
407 	 */
408 	bool dlist_initialized;
409 
410 	/* Load of this plane on the HVS block. The load is expressed in HVS
411 	 * cycles/sec.
412 	 */
413 	u64 hvs_load;
414 
415 	/* Memory bandwidth needed for this plane. This is expressed in
416 	 * bytes/sec.
417 	 */
418 	u64 membus_load;
419 };
420 
421 static inline struct vc4_plane_state *
422 to_vc4_plane_state(struct drm_plane_state *state)
423 {
424 	return container_of(state, struct vc4_plane_state, base);
425 }
426 
427 enum vc4_encoder_type {
428 	VC4_ENCODER_TYPE_NONE,
429 	VC4_ENCODER_TYPE_HDMI0,
430 	VC4_ENCODER_TYPE_HDMI1,
431 	VC4_ENCODER_TYPE_VEC,
432 	VC4_ENCODER_TYPE_DSI0,
433 	VC4_ENCODER_TYPE_DSI1,
434 	VC4_ENCODER_TYPE_SMI,
435 	VC4_ENCODER_TYPE_DPI,
436 };
437 
438 struct vc4_encoder {
439 	struct drm_encoder base;
440 	enum vc4_encoder_type type;
441 	u32 clock_select;
442 
443 	void (*pre_crtc_configure)(struct drm_encoder *encoder, struct drm_atomic_state *state);
444 	void (*pre_crtc_enable)(struct drm_encoder *encoder, struct drm_atomic_state *state);
445 	void (*post_crtc_enable)(struct drm_encoder *encoder, struct drm_atomic_state *state);
446 
447 	void (*post_crtc_disable)(struct drm_encoder *encoder, struct drm_atomic_state *state);
448 	void (*post_crtc_powerdown)(struct drm_encoder *encoder, struct drm_atomic_state *state);
449 };
450 
451 static inline struct vc4_encoder *
452 to_vc4_encoder(struct drm_encoder *encoder)
453 {
454 	return container_of(encoder, struct vc4_encoder, base);
455 }
456 
457 struct vc4_crtc_data {
458 	/* Bitmask of channels (FIFOs) of the HVS that the output can source from */
459 	unsigned int hvs_available_channels;
460 
461 	/* Which output of the HVS this pixelvalve sources from. */
462 	int hvs_output;
463 };
464 
465 struct vc4_pv_data {
466 	struct vc4_crtc_data	base;
467 
468 	/* Depth of the PixelValve FIFO in bytes */
469 	unsigned int fifo_depth;
470 
471 	/* Number of pixels output per clock period */
472 	u8 pixels_per_clock;
473 
474 	enum vc4_encoder_type encoder_types[4];
475 	const char *debugfs_name;
476 
477 };
478 
479 struct vc4_crtc {
480 	struct drm_crtc base;
481 	struct platform_device *pdev;
482 	const struct vc4_crtc_data *data;
483 	void __iomem *regs;
484 
485 	/* Timestamp at start of vblank irq - unaffected by lock delays. */
486 	ktime_t t_vblank;
487 
488 	u8 lut_r[256];
489 	u8 lut_g[256];
490 	u8 lut_b[256];
491 
492 	struct drm_pending_vblank_event *event;
493 
494 	struct debugfs_regset32 regset;
495 
496 	/**
497 	 * @feeds_txp: True if the CRTC feeds our writeback controller.
498 	 */
499 	bool feeds_txp;
500 
501 	/**
502 	 * @irq_lock: Spinlock protecting the resources shared between
503 	 * the atomic code and our vblank handler.
504 	 */
505 	spinlock_t irq_lock;
506 
507 	/**
508 	 * @current_dlist: Start offset of the display list currently
509 	 * set in the HVS for that CRTC. Protected by @irq_lock, and
510 	 * copied in vc4_hvs_update_dlist() for the CRTC interrupt
511 	 * handler to have access to that value.
512 	 */
513 	unsigned int current_dlist;
514 
515 	/**
516 	 * @current_hvs_channel: HVS channel currently assigned to the
517 	 * CRTC. Protected by @irq_lock, and copied in
518 	 * vc4_hvs_atomic_begin() for the CRTC interrupt handler to have
519 	 * access to that value.
520 	 */
521 	unsigned int current_hvs_channel;
522 };
523 
524 static inline struct vc4_crtc *
525 to_vc4_crtc(struct drm_crtc *crtc)
526 {
527 	return container_of(crtc, struct vc4_crtc, base);
528 }
529 
530 static inline const struct vc4_crtc_data *
531 vc4_crtc_to_vc4_crtc_data(const struct vc4_crtc *crtc)
532 {
533 	return crtc->data;
534 }
535 
536 static inline const struct vc4_pv_data *
537 vc4_crtc_to_vc4_pv_data(const struct vc4_crtc *crtc)
538 {
539 	const struct vc4_crtc_data *data = vc4_crtc_to_vc4_crtc_data(crtc);
540 
541 	return container_of(data, struct vc4_pv_data, base);
542 }
543 
544 struct drm_encoder *vc4_get_crtc_encoder(struct drm_crtc *crtc,
545 					 struct drm_crtc_state *state);
546 
547 struct vc4_crtc_state {
548 	struct drm_crtc_state base;
549 	/* Dlist area for this CRTC configuration. */
550 	struct drm_mm_node mm;
551 	bool txp_armed;
552 	unsigned int assigned_channel;
553 
554 	struct {
555 		unsigned int left;
556 		unsigned int right;
557 		unsigned int top;
558 		unsigned int bottom;
559 	} margins;
560 
561 	unsigned long hvs_load;
562 
563 	/* Transitional state below, only valid during atomic commits */
564 	bool update_muxing;
565 };
566 
567 #define VC4_HVS_CHANNEL_DISABLED ((unsigned int)-1)
568 
569 static inline struct vc4_crtc_state *
570 to_vc4_crtc_state(struct drm_crtc_state *crtc_state)
571 {
572 	return container_of(crtc_state, struct vc4_crtc_state, base);
573 }
574 
575 #define V3D_READ(offset) readl(vc4->v3d->regs + offset)
576 #define V3D_WRITE(offset, val) writel(val, vc4->v3d->regs + offset)
577 #define HVS_READ(offset) readl(hvs->regs + offset)
578 #define HVS_WRITE(offset, val) writel(val, hvs->regs + offset)
579 
580 #define VC4_REG32(reg) { .name = #reg, .offset = reg }
581 
582 struct vc4_exec_info {
583 	/* Sequence number for this bin/render job. */
584 	uint64_t seqno;
585 
586 	/* Latest write_seqno of any BO that binning depends on. */
587 	uint64_t bin_dep_seqno;
588 
589 	struct dma_fence *fence;
590 
591 	/* Last current addresses the hardware was processing when the
592 	 * hangcheck timer checked on us.
593 	 */
594 	uint32_t last_ct0ca, last_ct1ca;
595 
596 	/* Kernel-space copy of the ioctl arguments */
597 	struct drm_vc4_submit_cl *args;
598 
599 	/* This is the array of BOs that were looked up at the start of exec.
600 	 * Command validation will use indices into this array.
601 	 */
602 	struct drm_gem_cma_object **bo;
603 	uint32_t bo_count;
604 
605 	/* List of BOs that are being written by the RCL.  Other than
606 	 * the binner temporary storage, this is all the BOs written
607 	 * by the job.
608 	 */
609 	struct drm_gem_cma_object *rcl_write_bo[4];
610 	uint32_t rcl_write_bo_count;
611 
612 	/* Pointers for our position in vc4->job_list */
613 	struct list_head head;
614 
615 	/* List of other BOs used in the job that need to be released
616 	 * once the job is complete.
617 	 */
618 	struct list_head unref_list;
619 
620 	/* Current unvalidated indices into @bo loaded by the non-hardware
621 	 * VC4_PACKET_GEM_HANDLES.
622 	 */
623 	uint32_t bo_index[2];
624 
625 	/* This is the BO where we store the validated command lists, shader
626 	 * records, and uniforms.
627 	 */
628 	struct drm_gem_cma_object *exec_bo;
629 
630 	/**
631 	 * This tracks the per-shader-record state (packet 64) that
632 	 * determines the length of the shader record and the offset
633 	 * it's expected to be found at.  It gets read in from the
634 	 * command lists.
635 	 */
636 	struct vc4_shader_state {
637 		uint32_t addr;
638 		/* Maximum vertex index referenced by any primitive using this
639 		 * shader state.
640 		 */
641 		uint32_t max_index;
642 	} *shader_state;
643 
644 	/** How many shader states the user declared they were using. */
645 	uint32_t shader_state_size;
646 	/** How many shader state records the validator has seen. */
647 	uint32_t shader_state_count;
648 
649 	bool found_tile_binning_mode_config_packet;
650 	bool found_start_tile_binning_packet;
651 	bool found_increment_semaphore_packet;
652 	bool found_flush;
653 	uint8_t bin_tiles_x, bin_tiles_y;
654 	/* Physical address of the start of the tile alloc array
655 	 * (where each tile's binned CL will start)
656 	 */
657 	uint32_t tile_alloc_offset;
658 	/* Bitmask of which binner slots are freed when this job completes. */
659 	uint32_t bin_slots;
660 
661 	/**
662 	 * Computed addresses pointing into exec_bo where we start the
663 	 * bin thread (ct0) and render thread (ct1).
664 	 */
665 	uint32_t ct0ca, ct0ea;
666 	uint32_t ct1ca, ct1ea;
667 
668 	/* Pointer to the unvalidated bin CL (if present). */
669 	void *bin_u;
670 
671 	/* Pointers to the shader recs.  These paddr gets incremented as CL
672 	 * packets are relocated in validate_gl_shader_state, and the vaddrs
673 	 * (u and v) get incremented and size decremented as the shader recs
674 	 * themselves are validated.
675 	 */
676 	void *shader_rec_u;
677 	void *shader_rec_v;
678 	uint32_t shader_rec_p;
679 	uint32_t shader_rec_size;
680 
681 	/* Pointers to the uniform data.  These pointers are incremented, and
682 	 * size decremented, as each batch of uniforms is uploaded.
683 	 */
684 	void *uniforms_u;
685 	void *uniforms_v;
686 	uint32_t uniforms_p;
687 	uint32_t uniforms_size;
688 
689 	/* Pointer to a performance monitor object if the user requested it,
690 	 * NULL otherwise.
691 	 */
692 	struct vc4_perfmon *perfmon;
693 
694 	/* Whether the exec has taken a reference to the binner BO, which should
695 	 * happen with a VC4_PACKET_TILE_BINNING_MODE_CONFIG packet.
696 	 */
697 	bool bin_bo_used;
698 };
699 
700 /* Per-open file private data. Any driver-specific resource that has to be
701  * released when the DRM file is closed should be placed here.
702  */
703 struct vc4_file {
704 	struct {
705 		struct idr idr;
706 		struct mutex lock;
707 	} perfmon;
708 
709 	bool bin_bo_used;
710 };
711 
712 static inline struct vc4_exec_info *
713 vc4_first_bin_job(struct vc4_dev *vc4)
714 {
715 	return list_first_entry_or_null(&vc4->bin_job_list,
716 					struct vc4_exec_info, head);
717 }
718 
719 static inline struct vc4_exec_info *
720 vc4_first_render_job(struct vc4_dev *vc4)
721 {
722 	return list_first_entry_or_null(&vc4->render_job_list,
723 					struct vc4_exec_info, head);
724 }
725 
726 static inline struct vc4_exec_info *
727 vc4_last_render_job(struct vc4_dev *vc4)
728 {
729 	if (list_empty(&vc4->render_job_list))
730 		return NULL;
731 	return list_last_entry(&vc4->render_job_list,
732 			       struct vc4_exec_info, head);
733 }
734 
735 /**
736  * struct vc4_texture_sample_info - saves the offsets into the UBO for texture
737  * setup parameters.
738  *
739  * This will be used at draw time to relocate the reference to the texture
740  * contents in p0, and validate that the offset combined with
741  * width/height/stride/etc. from p1 and p2/p3 doesn't sample outside the BO.
742  * Note that the hardware treats unprovided config parameters as 0, so not all
743  * of them need to be set up for every texure sample, and we'll store ~0 as
744  * the offset to mark the unused ones.
745  *
746  * See the VC4 3D architecture guide page 41 ("Texture and Memory Lookup Unit
747  * Setup") for definitions of the texture parameters.
748  */
749 struct vc4_texture_sample_info {
750 	bool is_direct;
751 	uint32_t p_offset[4];
752 };
753 
754 /**
755  * struct vc4_validated_shader_info - information about validated shaders that
756  * needs to be used from command list validation.
757  *
758  * For a given shader, each time a shader state record references it, we need
759  * to verify that the shader doesn't read more uniforms than the shader state
760  * record's uniform BO pointer can provide, and we need to apply relocations
761  * and validate the shader state record's uniforms that define the texture
762  * samples.
763  */
764 struct vc4_validated_shader_info {
765 	uint32_t uniforms_size;
766 	uint32_t uniforms_src_size;
767 	uint32_t num_texture_samples;
768 	struct vc4_texture_sample_info *texture_samples;
769 
770 	uint32_t num_uniform_addr_offsets;
771 	uint32_t *uniform_addr_offsets;
772 
773 	bool is_threaded;
774 };
775 
776 /**
777  * __wait_for - magic wait macro
778  *
779  * Macro to help avoid open coding check/wait/timeout patterns. Note that it's
780  * important that we check the condition again after having timed out, since the
781  * timeout could be due to preemption or similar and we've never had a chance to
782  * check the condition before the timeout.
783  */
784 #define __wait_for(OP, COND, US, Wmin, Wmax) ({ \
785 	const ktime_t end__ = ktime_add_ns(ktime_get_raw(), 1000ll * (US)); \
786 	long wait__ = (Wmin); /* recommended min for usleep is 10 us */	\
787 	int ret__;							\
788 	might_sleep();							\
789 	for (;;) {							\
790 		const bool expired__ = ktime_after(ktime_get_raw(), end__); \
791 		OP;							\
792 		/* Guarantee COND check prior to timeout */		\
793 		barrier();						\
794 		if (COND) {						\
795 			ret__ = 0;					\
796 			break;						\
797 		}							\
798 		if (expired__) {					\
799 			ret__ = -ETIMEDOUT;				\
800 			break;						\
801 		}							\
802 		usleep_range(wait__, wait__ * 2);			\
803 		if (wait__ < (Wmax))					\
804 			wait__ <<= 1;					\
805 	}								\
806 	ret__;								\
807 })
808 
809 #define _wait_for(COND, US, Wmin, Wmax)	__wait_for(, (COND), (US), (Wmin), \
810 						   (Wmax))
811 #define wait_for(COND, MS)		_wait_for((COND), (MS) * 1000, 10, 1000)
812 
813 /* vc4_bo.c */
814 struct drm_gem_object *vc4_create_object(struct drm_device *dev, size_t size);
815 struct vc4_bo *vc4_bo_create(struct drm_device *dev, size_t size,
816 			     bool from_cache, enum vc4_kernel_bo_type type);
817 int vc4_dumb_create(struct drm_file *file_priv,
818 		    struct drm_device *dev,
819 		    struct drm_mode_create_dumb *args);
820 int vc4_create_bo_ioctl(struct drm_device *dev, void *data,
821 			struct drm_file *file_priv);
822 int vc4_create_shader_bo_ioctl(struct drm_device *dev, void *data,
823 			       struct drm_file *file_priv);
824 int vc4_mmap_bo_ioctl(struct drm_device *dev, void *data,
825 		      struct drm_file *file_priv);
826 int vc4_set_tiling_ioctl(struct drm_device *dev, void *data,
827 			 struct drm_file *file_priv);
828 int vc4_get_tiling_ioctl(struct drm_device *dev, void *data,
829 			 struct drm_file *file_priv);
830 int vc4_get_hang_state_ioctl(struct drm_device *dev, void *data,
831 			     struct drm_file *file_priv);
832 int vc4_label_bo_ioctl(struct drm_device *dev, void *data,
833 		       struct drm_file *file_priv);
834 int vc4_bo_cache_init(struct drm_device *dev);
835 int vc4_bo_inc_usecnt(struct vc4_bo *bo);
836 void vc4_bo_dec_usecnt(struct vc4_bo *bo);
837 void vc4_bo_add_to_purgeable_pool(struct vc4_bo *bo);
838 void vc4_bo_remove_from_purgeable_pool(struct vc4_bo *bo);
839 
840 /* vc4_crtc.c */
841 extern struct platform_driver vc4_crtc_driver;
842 int vc4_crtc_disable_at_boot(struct drm_crtc *crtc);
843 int vc4_crtc_init(struct drm_device *drm, struct vc4_crtc *vc4_crtc,
844 		  const struct drm_crtc_funcs *crtc_funcs,
845 		  const struct drm_crtc_helper_funcs *crtc_helper_funcs);
846 void vc4_crtc_destroy(struct drm_crtc *crtc);
847 int vc4_page_flip(struct drm_crtc *crtc,
848 		  struct drm_framebuffer *fb,
849 		  struct drm_pending_vblank_event *event,
850 		  uint32_t flags,
851 		  struct drm_modeset_acquire_ctx *ctx);
852 struct drm_crtc_state *vc4_crtc_duplicate_state(struct drm_crtc *crtc);
853 void vc4_crtc_destroy_state(struct drm_crtc *crtc,
854 			    struct drm_crtc_state *state);
855 void vc4_crtc_reset(struct drm_crtc *crtc);
856 void vc4_crtc_handle_vblank(struct vc4_crtc *crtc);
857 void vc4_crtc_get_margins(struct drm_crtc_state *state,
858 			  unsigned int *left, unsigned int *right,
859 			  unsigned int *top, unsigned int *bottom);
860 
861 /* vc4_debugfs.c */
862 void vc4_debugfs_init(struct drm_minor *minor);
863 #ifdef CONFIG_DEBUG_FS
864 void vc4_debugfs_add_file(struct drm_device *drm,
865 			  const char *filename,
866 			  int (*show)(struct seq_file*, void*),
867 			  void *data);
868 void vc4_debugfs_add_regset32(struct drm_device *drm,
869 			      const char *filename,
870 			      struct debugfs_regset32 *regset);
871 #else
872 static inline void vc4_debugfs_add_file(struct drm_device *drm,
873 					const char *filename,
874 					int (*show)(struct seq_file*, void*),
875 					void *data)
876 {
877 }
878 
879 static inline void vc4_debugfs_add_regset32(struct drm_device *drm,
880 					    const char *filename,
881 					    struct debugfs_regset32 *regset)
882 {
883 }
884 #endif
885 
886 /* vc4_drv.c */
887 void __iomem *vc4_ioremap_regs(struct platform_device *dev, int index);
888 
889 /* vc4_dpi.c */
890 extern struct platform_driver vc4_dpi_driver;
891 
892 /* vc4_dsi.c */
893 extern struct platform_driver vc4_dsi_driver;
894 
895 /* vc4_fence.c */
896 extern const struct dma_fence_ops vc4_fence_ops;
897 
898 /* vc4_gem.c */
899 int vc4_gem_init(struct drm_device *dev);
900 int vc4_submit_cl_ioctl(struct drm_device *dev, void *data,
901 			struct drm_file *file_priv);
902 int vc4_wait_seqno_ioctl(struct drm_device *dev, void *data,
903 			 struct drm_file *file_priv);
904 int vc4_wait_bo_ioctl(struct drm_device *dev, void *data,
905 		      struct drm_file *file_priv);
906 void vc4_submit_next_bin_job(struct drm_device *dev);
907 void vc4_submit_next_render_job(struct drm_device *dev);
908 void vc4_move_job_to_render(struct drm_device *dev, struct vc4_exec_info *exec);
909 int vc4_wait_for_seqno(struct drm_device *dev, uint64_t seqno,
910 		       uint64_t timeout_ns, bool interruptible);
911 void vc4_job_handle_completed(struct vc4_dev *vc4);
912 int vc4_queue_seqno_cb(struct drm_device *dev,
913 		       struct vc4_seqno_cb *cb, uint64_t seqno,
914 		       void (*func)(struct vc4_seqno_cb *cb));
915 int vc4_gem_madvise_ioctl(struct drm_device *dev, void *data,
916 			  struct drm_file *file_priv);
917 
918 /* vc4_hdmi.c */
919 extern struct platform_driver vc4_hdmi_driver;
920 
921 /* vc4_vec.c */
922 extern struct platform_driver vc4_vec_driver;
923 
924 /* vc4_txp.c */
925 extern struct platform_driver vc4_txp_driver;
926 
927 /* vc4_irq.c */
928 void vc4_irq_enable(struct drm_device *dev);
929 void vc4_irq_disable(struct drm_device *dev);
930 int vc4_irq_install(struct drm_device *dev, int irq);
931 void vc4_irq_uninstall(struct drm_device *dev);
932 void vc4_irq_reset(struct drm_device *dev);
933 
934 /* vc4_hvs.c */
935 extern struct platform_driver vc4_hvs_driver;
936 void vc4_hvs_stop_channel(struct vc4_hvs *hvs, unsigned int output);
937 int vc4_hvs_get_fifo_from_output(struct vc4_hvs *hvs, unsigned int output);
938 u8 vc4_hvs_get_fifo_frame_count(struct vc4_hvs *hvs, unsigned int fifo);
939 int vc4_hvs_atomic_check(struct drm_crtc *crtc, struct drm_atomic_state *state);
940 void vc4_hvs_atomic_begin(struct drm_crtc *crtc, struct drm_atomic_state *state);
941 void vc4_hvs_atomic_enable(struct drm_crtc *crtc, struct drm_atomic_state *state);
942 void vc4_hvs_atomic_disable(struct drm_crtc *crtc, struct drm_atomic_state *state);
943 void vc4_hvs_atomic_flush(struct drm_crtc *crtc, struct drm_atomic_state *state);
944 void vc4_hvs_dump_state(struct vc4_hvs *hvs);
945 void vc4_hvs_unmask_underrun(struct vc4_hvs *hvs, int channel);
946 void vc4_hvs_mask_underrun(struct vc4_hvs *hvs, int channel);
947 
948 /* vc4_kms.c */
949 int vc4_kms_load(struct drm_device *dev);
950 
951 /* vc4_plane.c */
952 struct drm_plane *vc4_plane_init(struct drm_device *dev,
953 				 enum drm_plane_type type);
954 int vc4_plane_create_additional_planes(struct drm_device *dev);
955 u32 vc4_plane_write_dlist(struct drm_plane *plane, u32 __iomem *dlist);
956 u32 vc4_plane_dlist_size(const struct drm_plane_state *state);
957 void vc4_plane_async_set_fb(struct drm_plane *plane,
958 			    struct drm_framebuffer *fb);
959 
960 /* vc4_v3d.c */
961 extern struct platform_driver vc4_v3d_driver;
962 extern const struct of_device_id vc4_v3d_dt_match[];
963 int vc4_v3d_get_bin_slot(struct vc4_dev *vc4);
964 int vc4_v3d_bin_bo_get(struct vc4_dev *vc4, bool *used);
965 void vc4_v3d_bin_bo_put(struct vc4_dev *vc4);
966 int vc4_v3d_pm_get(struct vc4_dev *vc4);
967 void vc4_v3d_pm_put(struct vc4_dev *vc4);
968 
969 /* vc4_validate.c */
970 int
971 vc4_validate_bin_cl(struct drm_device *dev,
972 		    void *validated,
973 		    void *unvalidated,
974 		    struct vc4_exec_info *exec);
975 
976 int
977 vc4_validate_shader_recs(struct drm_device *dev, struct vc4_exec_info *exec);
978 
979 struct drm_gem_cma_object *vc4_use_bo(struct vc4_exec_info *exec,
980 				      uint32_t hindex);
981 
982 int vc4_get_rcl(struct drm_device *dev, struct vc4_exec_info *exec);
983 
984 bool vc4_check_tex_size(struct vc4_exec_info *exec,
985 			struct drm_gem_cma_object *fbo,
986 			uint32_t offset, uint8_t tiling_format,
987 			uint32_t width, uint32_t height, uint8_t cpp);
988 
989 /* vc4_validate_shader.c */
990 struct vc4_validated_shader_info *
991 vc4_validate_shader(struct drm_gem_cma_object *shader_obj);
992 
993 /* vc4_perfmon.c */
994 void vc4_perfmon_get(struct vc4_perfmon *perfmon);
995 void vc4_perfmon_put(struct vc4_perfmon *perfmon);
996 void vc4_perfmon_start(struct vc4_dev *vc4, struct vc4_perfmon *perfmon);
997 void vc4_perfmon_stop(struct vc4_dev *vc4, struct vc4_perfmon *perfmon,
998 		      bool capture);
999 struct vc4_perfmon *vc4_perfmon_find(struct vc4_file *vc4file, int id);
1000 void vc4_perfmon_open_file(struct vc4_file *vc4file);
1001 void vc4_perfmon_close_file(struct vc4_file *vc4file);
1002 int vc4_perfmon_create_ioctl(struct drm_device *dev, void *data,
1003 			     struct drm_file *file_priv);
1004 int vc4_perfmon_destroy_ioctl(struct drm_device *dev, void *data,
1005 			      struct drm_file *file_priv);
1006 int vc4_perfmon_get_values_ioctl(struct drm_device *dev, void *data,
1007 				 struct drm_file *file_priv);
1008 
1009 #endif /* _VC4_DRV_H_ */
1010