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
23struct drm_device;
24struct drm_gem_object;
25
26/* Don't forget to update vc4_bo.c: bo_type_names[] when adding to
27 * this.
28 */
29enum 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 */
50struct 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
74struct 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_vec *vec;
81	struct vc4_txp *txp;
82
83	struct vc4_hang_state *hang_state;
84
85	/* The kernel-space BO cache.  Tracks buffers that have been
86	 * unreferenced by all other users (refcounts of 0!) but not
87	 * yet freed, so we can do cheap allocations.
88	 */
89	struct vc4_bo_cache {
90		/* Array of list heads for entries in the BO cache,
91		 * based on number of pages, so we can do O(1) lookups
92		 * in the cache when allocating.
93		 */
94		struct list_head *size_list;
95		uint32_t size_list_size;
96
97		/* List of all BOs in the cache, ordered by age, so we
98		 * can do O(1) lookups when trying to free old
99		 * buffers.
100		 */
101		struct list_head time_list;
102		struct work_struct time_work;
103		struct timer_list time_timer;
104	} bo_cache;
105
106	u32 num_labels;
107	struct vc4_label {
108		const char *name;
109		u32 num_allocated;
110		u32 size_allocated;
111	} *bo_labels;
112
113	/* Protects bo_cache and bo_labels. */
114	struct mutex bo_lock;
115
116	/* Purgeable BO pool. All BOs in this pool can have their memory
117	 * reclaimed if the driver is unable to allocate new BOs. We also
118	 * keep stats related to the purge mechanism here.
119	 */
120	struct {
121		struct list_head list;
122		unsigned int num;
123		size_t size;
124		unsigned int purged_num;
125		size_t purged_size;
126		struct mutex lock;
127	} purgeable;
128
129	uint64_t dma_fence_context;
130
131	/* Sequence number for the last job queued in bin_job_list.
132	 * Starts at 0 (no jobs emitted).
133	 */
134	uint64_t emit_seqno;
135
136	/* Sequence number for the last completed job on the GPU.
137	 * Starts at 0 (no jobs completed).
138	 */
139	uint64_t finished_seqno;
140
141	/* List of all struct vc4_exec_info for jobs to be executed in
142	 * the binner.  The first job in the list is the one currently
143	 * programmed into ct0ca for execution.
144	 */
145	struct list_head bin_job_list;
146
147	/* List of all struct vc4_exec_info for jobs that have
148	 * completed binning and are ready for rendering.  The first
149	 * job in the list is the one currently programmed into ct1ca
150	 * for execution.
151	 */
152	struct list_head render_job_list;
153
154	/* List of the finished vc4_exec_infos waiting to be freed by
155	 * job_done_work.
156	 */
157	struct list_head job_done_list;
158	/* Spinlock used to synchronize the job_list and seqno
159	 * accesses between the IRQ handler and GEM ioctls.
160	 */
161	spinlock_t job_lock;
162	wait_queue_head_t job_wait_queue;
163	struct work_struct job_done_work;
164
165	/* Used to track the active perfmon if any. Access to this field is
166	 * protected by job_lock.
167	 */
168	struct vc4_perfmon *active_perfmon;
169
170	/* List of struct vc4_seqno_cb for callbacks to be made from a
171	 * workqueue when the given seqno is passed.
172	 */
173	struct list_head seqno_cb_list;
174
175	/* The memory used for storing binner tile alloc, tile state,
176	 * and overflow memory allocations.  This is freed when V3D
177	 * powers down.
178	 */
179	struct vc4_bo *bin_bo;
180
181	/* Size of blocks allocated within bin_bo. */
182	uint32_t bin_alloc_size;
183
184	/* Bitmask of the bin_alloc_size chunks in bin_bo that are
185	 * used.
186	 */
187	uint32_t bin_alloc_used;
188
189	/* Bitmask of the current bin_alloc used for overflow memory. */
190	uint32_t bin_alloc_overflow;
191
192	/* Incremented when an underrun error happened after an atomic commit.
193	 * This is particularly useful to detect when a specific modeset is too
194	 * demanding in term of memory or HVS bandwidth which is hard to guess
195	 * at atomic check time.
196	 */
197	atomic_t underrun;
198
199	struct work_struct overflow_mem_work;
200
201	int power_refcount;
202
203	/* Set to true when the load tracker is supported. */
204	bool load_tracker_available;
205
206	/* Set to true when the load tracker is active. */
207	bool load_tracker_enabled;
208
209	/* Mutex controlling the power refcount. */
210	struct mutex power_lock;
211
212	struct {
213		struct timer_list timer;
214		struct work_struct reset_work;
215	} hangcheck;
216
217	struct semaphore async_modeset;
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
235static inline struct vc4_dev *
236to_vc4_dev(struct drm_device *dev)
237{
238	return container_of(dev, struct vc4_dev, base);
239}
240
241struct vc4_bo {
242	struct drm_gem_cma_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
288static inline struct vc4_bo *
289to_vc4_bo(struct drm_gem_object *bo)
290{
291	return container_of(to_drm_gem_cma_obj(bo), struct vc4_bo, base);
292}
293
294struct vc4_fence {
295	struct dma_fence base;
296	struct drm_device *dev;
297	/* vc4 seqno for signaled() test */
298	uint64_t seqno;
299};
300
301static inline struct vc4_fence *
302to_vc4_fence(struct dma_fence *fence)
303{
304	return container_of(fence, struct vc4_fence, base);
305}
306
307struct vc4_seqno_cb {
308	struct work_struct work;
309	uint64_t seqno;
310	void (*func)(struct vc4_seqno_cb *cb);
311};
312
313struct 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
321struct vc4_hvs {
322	struct platform_device *pdev;
323	void __iomem *regs;
324	u32 __iomem *dlist;
325
326	struct clk *core_clk;
327
328	/* Memory manager for CRTCs to allocate space in the display
329	 * list.  Units are dwords.
330	 */
331	struct drm_mm dlist_mm;
332	/* Memory manager for the LBM memory used by HVS scaling. */
333	struct drm_mm lbm_mm;
334	spinlock_t mm_lock;
335
336	struct drm_mm_node mitchell_netravali_filter;
337
338	struct debugfs_regset32 regset;
339
340	/* HVS version 5 flag, therefore requires updated dlist structures */
341	bool hvs5;
342};
343
344struct vc4_plane {
345	struct drm_plane base;
346};
347
348static inline struct vc4_plane *
349to_vc4_plane(struct drm_plane *plane)
350{
351	return container_of(plane, struct vc4_plane, base);
352}
353
354enum vc4_scaling_mode {
355	VC4_SCALING_NONE,
356	VC4_SCALING_TPZ,
357	VC4_SCALING_PPF,
358};
359
360struct vc4_plane_state {
361	struct drm_plane_state base;
362	/* System memory copy of the display list for this element, computed
363	 * at atomic_check time.
364	 */
365	u32 *dlist;
366	u32 dlist_size; /* Number of dwords allocated for the display list */
367	u32 dlist_count; /* Number of used dwords in the display list. */
368
369	/* Offset in the dlist to various words, for pageflip or
370	 * cursor updates.
371	 */
372	u32 pos0_offset;
373	u32 pos2_offset;
374	u32 ptr0_offset;
375	u32 lbm_offset;
376
377	/* Offset where the plane's dlist was last stored in the
378	 * hardware at vc4_crtc_atomic_flush() time.
379	 */
380	u32 __iomem *hw_dlist;
381
382	/* Clipped coordinates of the plane on the display. */
383	int crtc_x, crtc_y, crtc_w, crtc_h;
384	/* Clipped area being scanned from in the FB. */
385	u32 src_x, src_y;
386
387	u32 src_w[2], src_h[2];
388
389	/* Scaling selection for the RGB/Y plane and the Cb/Cr planes. */
390	enum vc4_scaling_mode x_scaling[2], y_scaling[2];
391	bool is_unity;
392	bool is_yuv;
393
394	/* Offset to start scanning out from the start of the plane's
395	 * BO.
396	 */
397	u32 offsets[3];
398
399	/* Our allocation in LBM for temporary storage during scaling. */
400	struct drm_mm_node lbm;
401
402	/* Set when the plane has per-pixel alpha content or does not cover
403	 * the entire screen. This is a hint to the CRTC that it might need
404	 * to enable background color fill.
405	 */
406	bool needs_bg_fill;
407
408	/* Mark the dlist as initialized. Useful to avoid initializing it twice
409	 * when async update is not possible.
410	 */
411	bool dlist_initialized;
412
413	/* Load of this plane on the HVS block. The load is expressed in HVS
414	 * cycles/sec.
415	 */
416	u64 hvs_load;
417
418	/* Memory bandwidth needed for this plane. This is expressed in
419	 * bytes/sec.
420	 */
421	u64 membus_load;
422};
423
424static inline struct vc4_plane_state *
425to_vc4_plane_state(struct drm_plane_state *state)
426{
427	return container_of(state, struct vc4_plane_state, base);
428}
429
430enum vc4_encoder_type {
431	VC4_ENCODER_TYPE_NONE,
432	VC4_ENCODER_TYPE_HDMI0,
433	VC4_ENCODER_TYPE_HDMI1,
434	VC4_ENCODER_TYPE_VEC,
435	VC4_ENCODER_TYPE_DSI0,
436	VC4_ENCODER_TYPE_DSI1,
437	VC4_ENCODER_TYPE_SMI,
438	VC4_ENCODER_TYPE_DPI,
439};
440
441struct vc4_encoder {
442	struct drm_encoder base;
443	enum vc4_encoder_type type;
444	u32 clock_select;
445
446	void (*pre_crtc_configure)(struct drm_encoder *encoder);
447	void (*pre_crtc_enable)(struct drm_encoder *encoder);
448	void (*post_crtc_enable)(struct drm_encoder *encoder);
449
450	void (*post_crtc_disable)(struct drm_encoder *encoder);
451	void (*post_crtc_powerdown)(struct drm_encoder *encoder);
452};
453
454static inline struct vc4_encoder *
455to_vc4_encoder(struct drm_encoder *encoder)
456{
457	return container_of(encoder, struct vc4_encoder, base);
458}
459
460struct vc4_crtc_data {
461	/* Bitmask of channels (FIFOs) of the HVS that the output can source from */
462	unsigned int hvs_available_channels;
463
464	/* Which output of the HVS this pixelvalve sources from. */
465	int hvs_output;
466};
467
468struct vc4_pv_data {
469	struct vc4_crtc_data	base;
470
471	/* Depth of the PixelValve FIFO in bytes */
472	unsigned int fifo_depth;
473
474	/* Number of pixels output per clock period */
475	u8 pixels_per_clock;
476
477	enum vc4_encoder_type encoder_types[4];
478	const char *debugfs_name;
479
480};
481
482struct vc4_crtc {
483	struct drm_crtc base;
484	struct platform_device *pdev;
485	const struct vc4_crtc_data *data;
486	void __iomem *regs;
487
488	/* Timestamp at start of vblank irq - unaffected by lock delays. */
489	ktime_t t_vblank;
490
491	u8 lut_r[256];
492	u8 lut_g[256];
493	u8 lut_b[256];
494
495	struct drm_pending_vblank_event *event;
496
497	struct debugfs_regset32 regset;
498};
499
500static inline struct vc4_crtc *
501to_vc4_crtc(struct drm_crtc *crtc)
502{
503	return container_of(crtc, struct vc4_crtc, base);
504}
505
506static inline const struct vc4_crtc_data *
507vc4_crtc_to_vc4_crtc_data(const struct vc4_crtc *crtc)
508{
509	return crtc->data;
510}
511
512static inline const struct vc4_pv_data *
513vc4_crtc_to_vc4_pv_data(const struct vc4_crtc *crtc)
514{
515	const struct vc4_crtc_data *data = vc4_crtc_to_vc4_crtc_data(crtc);
516
517	return container_of(data, struct vc4_pv_data, base);
518}
519
520struct vc4_crtc_state {
521	struct drm_crtc_state base;
522	/* Dlist area for this CRTC configuration. */
523	struct drm_mm_node mm;
524	bool feed_txp;
525	bool txp_armed;
526	unsigned int assigned_channel;
527
528	struct {
529		unsigned int left;
530		unsigned int right;
531		unsigned int top;
532		unsigned int bottom;
533	} margins;
534
535	/* Transitional state below, only valid during atomic commits */
536	bool update_muxing;
537};
538
539#define VC4_HVS_CHANNEL_DISABLED ((unsigned int)-1)
540
541static inline struct vc4_crtc_state *
542to_vc4_crtc_state(struct drm_crtc_state *crtc_state)
543{
544	return container_of(crtc_state, struct vc4_crtc_state, base);
545}
546
547#define V3D_READ(offset) readl(vc4->v3d->regs + offset)
548#define V3D_WRITE(offset, val) writel(val, vc4->v3d->regs + offset)
549#define HVS_READ(offset) readl(vc4->hvs->regs + offset)
550#define HVS_WRITE(offset, val) writel(val, vc4->hvs->regs + offset)
551
552#define VC4_REG32(reg) { .name = #reg, .offset = reg }
553
554struct vc4_exec_info {
555	/* Sequence number for this bin/render job. */
556	uint64_t seqno;
557
558	/* Latest write_seqno of any BO that binning depends on. */
559	uint64_t bin_dep_seqno;
560
561	struct dma_fence *fence;
562
563	/* Last current addresses the hardware was processing when the
564	 * hangcheck timer checked on us.
565	 */
566	uint32_t last_ct0ca, last_ct1ca;
567
568	/* Kernel-space copy of the ioctl arguments */
569	struct drm_vc4_submit_cl *args;
570
571	/* This is the array of BOs that were looked up at the start of exec.
572	 * Command validation will use indices into this array.
573	 */
574	struct drm_gem_cma_object **bo;
575	uint32_t bo_count;
576
577	/* List of BOs that are being written by the RCL.  Other than
578	 * the binner temporary storage, this is all the BOs written
579	 * by the job.
580	 */
581	struct drm_gem_cma_object *rcl_write_bo[4];
582	uint32_t rcl_write_bo_count;
583
584	/* Pointers for our position in vc4->job_list */
585	struct list_head head;
586
587	/* List of other BOs used in the job that need to be released
588	 * once the job is complete.
589	 */
590	struct list_head unref_list;
591
592	/* Current unvalidated indices into @bo loaded by the non-hardware
593	 * VC4_PACKET_GEM_HANDLES.
594	 */
595	uint32_t bo_index[2];
596
597	/* This is the BO where we store the validated command lists, shader
598	 * records, and uniforms.
599	 */
600	struct drm_gem_cma_object *exec_bo;
601
602	/**
603	 * This tracks the per-shader-record state (packet 64) that
604	 * determines the length of the shader record and the offset
605	 * it's expected to be found at.  It gets read in from the
606	 * command lists.
607	 */
608	struct vc4_shader_state {
609		uint32_t addr;
610		/* Maximum vertex index referenced by any primitive using this
611		 * shader state.
612		 */
613		uint32_t max_index;
614	} *shader_state;
615
616	/** How many shader states the user declared they were using. */
617	uint32_t shader_state_size;
618	/** How many shader state records the validator has seen. */
619	uint32_t shader_state_count;
620
621	bool found_tile_binning_mode_config_packet;
622	bool found_start_tile_binning_packet;
623	bool found_increment_semaphore_packet;
624	bool found_flush;
625	uint8_t bin_tiles_x, bin_tiles_y;
626	/* Physical address of the start of the tile alloc array
627	 * (where each tile's binned CL will start)
628	 */
629	uint32_t tile_alloc_offset;
630	/* Bitmask of which binner slots are freed when this job completes. */
631	uint32_t bin_slots;
632
633	/**
634	 * Computed addresses pointing into exec_bo where we start the
635	 * bin thread (ct0) and render thread (ct1).
636	 */
637	uint32_t ct0ca, ct0ea;
638	uint32_t ct1ca, ct1ea;
639
640	/* Pointer to the unvalidated bin CL (if present). */
641	void *bin_u;
642
643	/* Pointers to the shader recs.  These paddr gets incremented as CL
644	 * packets are relocated in validate_gl_shader_state, and the vaddrs
645	 * (u and v) get incremented and size decremented as the shader recs
646	 * themselves are validated.
647	 */
648	void *shader_rec_u;
649	void *shader_rec_v;
650	uint32_t shader_rec_p;
651	uint32_t shader_rec_size;
652
653	/* Pointers to the uniform data.  These pointers are incremented, and
654	 * size decremented, as each batch of uniforms is uploaded.
655	 */
656	void *uniforms_u;
657	void *uniforms_v;
658	uint32_t uniforms_p;
659	uint32_t uniforms_size;
660
661	/* Pointer to a performance monitor object if the user requested it,
662	 * NULL otherwise.
663	 */
664	struct vc4_perfmon *perfmon;
665
666	/* Whether the exec has taken a reference to the binner BO, which should
667	 * happen with a VC4_PACKET_TILE_BINNING_MODE_CONFIG packet.
668	 */
669	bool bin_bo_used;
670};
671
672/* Per-open file private data. Any driver-specific resource that has to be
673 * released when the DRM file is closed should be placed here.
674 */
675struct vc4_file {
676	struct {
677		struct idr idr;
678		struct mutex lock;
679	} perfmon;
680
681	bool bin_bo_used;
682};
683
684static inline struct vc4_exec_info *
685vc4_first_bin_job(struct vc4_dev *vc4)
686{
687	return list_first_entry_or_null(&vc4->bin_job_list,
688					struct vc4_exec_info, head);
689}
690
691static inline struct vc4_exec_info *
692vc4_first_render_job(struct vc4_dev *vc4)
693{
694	return list_first_entry_or_null(&vc4->render_job_list,
695					struct vc4_exec_info, head);
696}
697
698static inline struct vc4_exec_info *
699vc4_last_render_job(struct vc4_dev *vc4)
700{
701	if (list_empty(&vc4->render_job_list))
702		return NULL;
703	return list_last_entry(&vc4->render_job_list,
704			       struct vc4_exec_info, head);
705}
706
707/**
708 * struct vc4_texture_sample_info - saves the offsets into the UBO for texture
709 * setup parameters.
710 *
711 * This will be used at draw time to relocate the reference to the texture
712 * contents in p0, and validate that the offset combined with
713 * width/height/stride/etc. from p1 and p2/p3 doesn't sample outside the BO.
714 * Note that the hardware treats unprovided config parameters as 0, so not all
715 * of them need to be set up for every texure sample, and we'll store ~0 as
716 * the offset to mark the unused ones.
717 *
718 * See the VC4 3D architecture guide page 41 ("Texture and Memory Lookup Unit
719 * Setup") for definitions of the texture parameters.
720 */
721struct vc4_texture_sample_info {
722	bool is_direct;
723	uint32_t p_offset[4];
724};
725
726/**
727 * struct vc4_validated_shader_info - information about validated shaders that
728 * needs to be used from command list validation.
729 *
730 * For a given shader, each time a shader state record references it, we need
731 * to verify that the shader doesn't read more uniforms than the shader state
732 * record's uniform BO pointer can provide, and we need to apply relocations
733 * and validate the shader state record's uniforms that define the texture
734 * samples.
735 */
736struct vc4_validated_shader_info {
737	uint32_t uniforms_size;
738	uint32_t uniforms_src_size;
739	uint32_t num_texture_samples;
740	struct vc4_texture_sample_info *texture_samples;
741
742	uint32_t num_uniform_addr_offsets;
743	uint32_t *uniform_addr_offsets;
744
745	bool is_threaded;
746};
747
748/**
749 * __wait_for - magic wait macro
750 *
751 * Macro to help avoid open coding check/wait/timeout patterns. Note that it's
752 * important that we check the condition again after having timed out, since the
753 * timeout could be due to preemption or similar and we've never had a chance to
754 * check the condition before the timeout.
755 */
756#define __wait_for(OP, COND, US, Wmin, Wmax) ({ \
757	const ktime_t end__ = ktime_add_ns(ktime_get_raw(), 1000ll * (US)); \
758	long wait__ = (Wmin); /* recommended min for usleep is 10 us */	\
759	int ret__;							\
760	might_sleep();							\
761	for (;;) {							\
762		const bool expired__ = ktime_after(ktime_get_raw(), end__); \
763		OP;							\
764		/* Guarantee COND check prior to timeout */		\
765		barrier();						\
766		if (COND) {						\
767			ret__ = 0;					\
768			break;						\
769		}							\
770		if (expired__) {					\
771			ret__ = -ETIMEDOUT;				\
772			break;						\
773		}							\
774		usleep_range(wait__, wait__ * 2);			\
775		if (wait__ < (Wmax))					\
776			wait__ <<= 1;					\
777	}								\
778	ret__;								\
779})
780
781#define _wait_for(COND, US, Wmin, Wmax)	__wait_for(, (COND), (US), (Wmin), \
782						   (Wmax))
783#define wait_for(COND, MS)		_wait_for((COND), (MS) * 1000, 10, 1000)
784
785/* vc4_bo.c */
786struct drm_gem_object *vc4_create_object(struct drm_device *dev, size_t size);
787void vc4_free_object(struct drm_gem_object *gem_obj);
788struct vc4_bo *vc4_bo_create(struct drm_device *dev, size_t size,
789			     bool from_cache, enum vc4_kernel_bo_type type);
790int vc4_dumb_create(struct drm_file *file_priv,
791		    struct drm_device *dev,
792		    struct drm_mode_create_dumb *args);
793struct dma_buf *vc4_prime_export(struct drm_gem_object *obj, int flags);
794int vc4_create_bo_ioctl(struct drm_device *dev, void *data,
795			struct drm_file *file_priv);
796int vc4_create_shader_bo_ioctl(struct drm_device *dev, void *data,
797			       struct drm_file *file_priv);
798int vc4_mmap_bo_ioctl(struct drm_device *dev, void *data,
799		      struct drm_file *file_priv);
800int vc4_set_tiling_ioctl(struct drm_device *dev, void *data,
801			 struct drm_file *file_priv);
802int vc4_get_tiling_ioctl(struct drm_device *dev, void *data,
803			 struct drm_file *file_priv);
804int vc4_get_hang_state_ioctl(struct drm_device *dev, void *data,
805			     struct drm_file *file_priv);
806int vc4_label_bo_ioctl(struct drm_device *dev, void *data,
807		       struct drm_file *file_priv);
808vm_fault_t vc4_fault(struct vm_fault *vmf);
809int vc4_mmap(struct file *filp, struct vm_area_struct *vma);
810int vc4_prime_mmap(struct drm_gem_object *obj, struct vm_area_struct *vma);
811struct drm_gem_object *vc4_prime_import_sg_table(struct drm_device *dev,
812						 struct dma_buf_attachment *attach,
813						 struct sg_table *sgt);
814void *vc4_prime_vmap(struct drm_gem_object *obj);
815int vc4_bo_cache_init(struct drm_device *dev);
816int vc4_bo_inc_usecnt(struct vc4_bo *bo);
817void vc4_bo_dec_usecnt(struct vc4_bo *bo);
818void vc4_bo_add_to_purgeable_pool(struct vc4_bo *bo);
819void vc4_bo_remove_from_purgeable_pool(struct vc4_bo *bo);
820
821/* vc4_crtc.c */
822extern struct platform_driver vc4_crtc_driver;
823int vc4_crtc_disable_at_boot(struct drm_crtc *crtc);
824int 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);
827void vc4_crtc_destroy(struct drm_crtc *crtc);
828int 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);
833struct drm_crtc_state *vc4_crtc_duplicate_state(struct drm_crtc *crtc);
834void vc4_crtc_destroy_state(struct drm_crtc *crtc,
835			    struct drm_crtc_state *state);
836void vc4_crtc_reset(struct drm_crtc *crtc);
837void vc4_crtc_handle_vblank(struct vc4_crtc *crtc);
838void vc4_crtc_get_margins(struct drm_crtc_state *state,
839			  unsigned int *left, unsigned int *right,
840			  unsigned int *top, unsigned int *bottom);
841
842/* vc4_debugfs.c */
843void vc4_debugfs_init(struct drm_minor *minor);
844#ifdef CONFIG_DEBUG_FS
845void vc4_debugfs_add_file(struct drm_device *drm,
846			  const char *filename,
847			  int (*show)(struct seq_file*, void*),
848			  void *data);
849void vc4_debugfs_add_regset32(struct drm_device *drm,
850			      const char *filename,
851			      struct debugfs_regset32 *regset);
852#else
853static 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
860static 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 */
868void __iomem *vc4_ioremap_regs(struct platform_device *dev, int index);
869
870/* vc4_dpi.c */
871extern struct platform_driver vc4_dpi_driver;
872
873/* vc4_dsi.c */
874extern struct platform_driver vc4_dsi_driver;
875
876/* vc4_fence.c */
877extern const struct dma_fence_ops vc4_fence_ops;
878
879/* vc4_gem.c */
880int vc4_gem_init(struct drm_device *dev);
881int vc4_submit_cl_ioctl(struct drm_device *dev, void *data,
882			struct drm_file *file_priv);
883int vc4_wait_seqno_ioctl(struct drm_device *dev, void *data,
884			 struct drm_file *file_priv);
885int vc4_wait_bo_ioctl(struct drm_device *dev, void *data,
886		      struct drm_file *file_priv);
887void vc4_submit_next_bin_job(struct drm_device *dev);
888void vc4_submit_next_render_job(struct drm_device *dev);
889void vc4_move_job_to_render(struct drm_device *dev, struct vc4_exec_info *exec);
890int vc4_wait_for_seqno(struct drm_device *dev, uint64_t seqno,
891		       uint64_t timeout_ns, bool interruptible);
892void vc4_job_handle_completed(struct vc4_dev *vc4);
893int 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));
896int vc4_gem_madvise_ioctl(struct drm_device *dev, void *data,
897			  struct drm_file *file_priv);
898
899/* vc4_hdmi.c */
900extern struct platform_driver vc4_hdmi_driver;
901
902/* vc4_vec.c */
903extern struct platform_driver vc4_vec_driver;
904
905/* vc4_txp.c */
906extern struct platform_driver vc4_txp_driver;
907
908/* vc4_irq.c */
909irqreturn_t vc4_irq(int irq, void *arg);
910void vc4_irq_preinstall(struct drm_device *dev);
911int vc4_irq_postinstall(struct drm_device *dev);
912void vc4_irq_uninstall(struct drm_device *dev);
913void vc4_irq_reset(struct drm_device *dev);
914
915/* vc4_hvs.c */
916extern struct platform_driver vc4_hvs_driver;
917void vc4_hvs_stop_channel(struct drm_device *dev, unsigned int output);
918int vc4_hvs_get_fifo_from_output(struct drm_device *dev, unsigned int output);
919int vc4_hvs_atomic_check(struct drm_crtc *crtc, struct drm_crtc_state *state);
920void vc4_hvs_atomic_enable(struct drm_crtc *crtc, struct drm_crtc_state *old_state);
921void vc4_hvs_atomic_disable(struct drm_crtc *crtc, struct drm_crtc_state *old_state);
922void vc4_hvs_atomic_flush(struct drm_crtc *crtc, struct drm_crtc_state *state);
923void vc4_hvs_dump_state(struct drm_device *dev);
924void vc4_hvs_unmask_underrun(struct drm_device *dev, int channel);
925void vc4_hvs_mask_underrun(struct drm_device *dev, int channel);
926
927/* vc4_kms.c */
928int vc4_kms_load(struct drm_device *dev);
929
930/* vc4_plane.c */
931struct drm_plane *vc4_plane_init(struct drm_device *dev,
932				 enum drm_plane_type type);
933int vc4_plane_create_additional_planes(struct drm_device *dev);
934u32 vc4_plane_write_dlist(struct drm_plane *plane, u32 __iomem *dlist);
935u32 vc4_plane_dlist_size(const struct drm_plane_state *state);
936void vc4_plane_async_set_fb(struct drm_plane *plane,
937			    struct drm_framebuffer *fb);
938
939/* vc4_v3d.c */
940extern struct platform_driver vc4_v3d_driver;
941extern const struct of_device_id vc4_v3d_dt_match[];
942int vc4_v3d_get_bin_slot(struct vc4_dev *vc4);
943int vc4_v3d_bin_bo_get(struct vc4_dev *vc4, bool *used);
944void vc4_v3d_bin_bo_put(struct vc4_dev *vc4);
945int vc4_v3d_pm_get(struct vc4_dev *vc4);
946void vc4_v3d_pm_put(struct vc4_dev *vc4);
947
948/* vc4_validate.c */
949int
950vc4_validate_bin_cl(struct drm_device *dev,
951		    void *validated,
952		    void *unvalidated,
953		    struct vc4_exec_info *exec);
954
955int
956vc4_validate_shader_recs(struct drm_device *dev, struct vc4_exec_info *exec);
957
958struct drm_gem_cma_object *vc4_use_bo(struct vc4_exec_info *exec,
959				      uint32_t hindex);
960
961int vc4_get_rcl(struct drm_device *dev, struct vc4_exec_info *exec);
962
963bool vc4_check_tex_size(struct vc4_exec_info *exec,
964			struct drm_gem_cma_object *fbo,
965			uint32_t offset, uint8_t tiling_format,
966			uint32_t width, uint32_t height, uint8_t cpp);
967
968/* vc4_validate_shader.c */
969struct vc4_validated_shader_info *
970vc4_validate_shader(struct drm_gem_cma_object *shader_obj);
971
972/* vc4_perfmon.c */
973void vc4_perfmon_get(struct vc4_perfmon *perfmon);
974void vc4_perfmon_put(struct vc4_perfmon *perfmon);
975void vc4_perfmon_start(struct vc4_dev *vc4, struct vc4_perfmon *perfmon);
976void vc4_perfmon_stop(struct vc4_dev *vc4, struct vc4_perfmon *perfmon,
977		      bool capture);
978struct vc4_perfmon *vc4_perfmon_find(struct vc4_file *vc4file, int id);
979void vc4_perfmon_open_file(struct vc4_file *vc4file);
980void vc4_perfmon_close_file(struct vc4_file *vc4file);
981int vc4_perfmon_create_ioctl(struct drm_device *dev, void *data,
982			     struct drm_file *file_priv);
983int vc4_perfmon_destroy_ioctl(struct drm_device *dev, void *data,
984			      struct drm_file *file_priv);
985int vc4_perfmon_get_values_ioctl(struct drm_device *dev, void *data,
986				 struct drm_file *file_priv);
987
988#endif /* _VC4_DRV_H_ */
989