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/debugfs.h>
9 #include <linux/delay.h>
10 #include <linux/of.h>
11 #include <linux/refcount.h>
12 #include <linux/uaccess.h>
13
14 #include <drm/drm_atomic.h>
15 #include <drm/drm_debugfs.h>
16 #include <drm/drm_device.h>
17 #include <drm/drm_encoder.h>
18 #include <drm/drm_fourcc.h>
19 #include <drm/drm_gem_dma_helper.h>
20 #include <drm/drm_managed.h>
21 #include <drm/drm_mm.h>
22 #include <drm/drm_modeset_lock.h>
23
24 #include <kunit/test-bug.h>
25
26 #include "uapi/drm/vc4_drm.h"
27
28 struct drm_device;
29 struct drm_gem_object;
30
31 extern const struct drm_driver vc4_drm_driver;
32 extern const struct drm_driver vc5_drm_driver;
33
34 /* Don't forget to update vc4_bo.c: bo_type_names[] when adding to
35 * this.
36 */
37 enum vc4_kernel_bo_type {
38 /* Any kernel allocation (gem_create_object hook) before it
39 * gets another type set.
40 */
41 VC4_BO_TYPE_KERNEL,
42 VC4_BO_TYPE_V3D,
43 VC4_BO_TYPE_V3D_SHADER,
44 VC4_BO_TYPE_DUMB,
45 VC4_BO_TYPE_BIN,
46 VC4_BO_TYPE_RCL,
47 VC4_BO_TYPE_BCL,
48 VC4_BO_TYPE_KERNEL_CACHE,
49 VC4_BO_TYPE_COUNT
50 };
51
52 /* Performance monitor object. The perform lifetime is controlled by userspace
53 * using perfmon related ioctls. A perfmon can be attached to a submit_cl
54 * request, and when this is the case, HW perf counters will be activated just
55 * before the submit_cl is submitted to the GPU and disabled when the job is
56 * done. This way, only events related to a specific job will be counted.
57 */
58 struct vc4_perfmon {
59 struct vc4_dev *dev;
60
61 /* Tracks the number of users of the perfmon, when this counter reaches
62 * zero the perfmon is destroyed.
63 */
64 refcount_t refcnt;
65
66 /* Number of counters activated in this perfmon instance
67 * (should be less than DRM_VC4_MAX_PERF_COUNTERS).
68 */
69 u8 ncounters;
70
71 /* Events counted by the HW perf counters. */
72 u8 events[DRM_VC4_MAX_PERF_COUNTERS];
73
74 /* Storage for counter values. Counters are incremented by the HW
75 * perf counter values every time the perfmon is attached to a GPU job.
76 * This way, perfmon users don't have to retrieve the results after
77 * each job if they want to track events covering several submissions.
78 * Note that counter values can't be reset, but you can fake a reset by
79 * destroying the perfmon and creating a new one.
80 */
81 u64 counters[] __counted_by(ncounters);
82 };
83
84 enum vc4_gen {
85 VC4_GEN_4,
86 VC4_GEN_5,
87 VC4_GEN_6_C,
88 VC4_GEN_6_D,
89 };
90
91 struct vc4_dev {
92 struct drm_device base;
93 struct device *dev;
94
95 enum vc4_gen gen;
96
97 unsigned int irq;
98
99 struct vc4_hvs *hvs;
100 struct vc4_v3d *v3d;
101
102 struct vc4_hang_state *hang_state;
103
104 /* The kernel-space BO cache. Tracks buffers that have been
105 * unreferenced by all other users (refcounts of 0!) but not
106 * yet freed, so we can do cheap allocations.
107 */
108 struct vc4_bo_cache {
109 /* Array of list heads for entries in the BO cache,
110 * based on number of pages, so we can do O(1) lookups
111 * in the cache when allocating.
112 */
113 struct list_head *size_list;
114 uint32_t size_list_size;
115
116 /* List of all BOs in the cache, ordered by age, so we
117 * can do O(1) lookups when trying to free old
118 * buffers.
119 */
120 struct list_head time_list;
121 struct work_struct time_work;
122 struct timer_list time_timer;
123 } bo_cache;
124
125 u32 num_labels;
126 struct vc4_label {
127 const char *name;
128 u32 num_allocated;
129 u32 size_allocated;
130 } *bo_labels;
131
132 /* Protects bo_cache and bo_labels. */
133 struct mutex bo_lock;
134
135 /* Purgeable BO pool. All BOs in this pool can have their memory
136 * reclaimed if the driver is unable to allocate new BOs. We also
137 * keep stats related to the purge mechanism here.
138 */
139 struct {
140 struct list_head list;
141 unsigned int num;
142 size_t size;
143 unsigned int purged_num;
144 size_t purged_size;
145 struct mutex lock;
146 } purgeable;
147
148 uint64_t dma_fence_context;
149
150 /* Sequence number for the last job queued in bin_job_list.
151 * Starts at 0 (no jobs emitted).
152 */
153 uint64_t emit_seqno;
154
155 /* Sequence number for the last completed job on the GPU.
156 * Starts at 0 (no jobs completed).
157 */
158 uint64_t finished_seqno;
159
160 /* List of all struct vc4_exec_info for jobs to be executed in
161 * the binner. The first job in the list is the one currently
162 * programmed into ct0ca for execution.
163 */
164 struct list_head bin_job_list;
165
166 /* List of all struct vc4_exec_info for jobs that have
167 * completed binning and are ready for rendering. The first
168 * job in the list is the one currently programmed into ct1ca
169 * for execution.
170 */
171 struct list_head render_job_list;
172
173 /* List of the finished vc4_exec_infos waiting to be freed by
174 * job_done_work.
175 */
176 struct list_head job_done_list;
177 /* Spinlock used to synchronize the job_list and seqno
178 * accesses between the IRQ handler and GEM ioctls.
179 */
180 spinlock_t job_lock;
181 wait_queue_head_t job_wait_queue;
182 struct work_struct job_done_work;
183
184 /* Used to track the active perfmon if any. Access to this field is
185 * protected by job_lock.
186 */
187 struct vc4_perfmon *active_perfmon;
188
189 /* The memory used for storing binner tile alloc, tile state,
190 * and overflow memory allocations. This is freed when V3D
191 * powers down.
192 */
193 struct vc4_bo *bin_bo;
194
195 /* Size of blocks allocated within bin_bo. */
196 uint32_t bin_alloc_size;
197
198 /* Bitmask of the bin_alloc_size chunks in bin_bo that are
199 * used.
200 */
201 uint32_t bin_alloc_used;
202
203 /* Bitmask of the current bin_alloc used for overflow memory. */
204 uint32_t bin_alloc_overflow;
205
206 /* Incremented when an underrun error happened after an atomic commit.
207 * This is particularly useful to detect when a specific modeset is too
208 * demanding in term of memory or HVS bandwidth which is hard to guess
209 * at atomic check time.
210 */
211 atomic_t underrun;
212
213 struct work_struct overflow_mem_work;
214
215 int power_refcount;
216
217 /* Set to true when the load tracker is active. */
218 bool load_tracker_enabled;
219
220 /* Mutex controlling the power refcount. */
221 struct mutex power_lock;
222
223 struct {
224 struct timer_list timer;
225 struct work_struct reset_work;
226 } hangcheck;
227
228 struct drm_modeset_lock ctm_state_lock;
229 struct drm_private_obj ctm_manager;
230 struct drm_private_obj hvs_channels;
231 struct drm_private_obj load_tracker;
232
233 /* Mutex for binner bo allocation. */
234 struct mutex bin_bo_lock;
235 /* Reference count for our binner bo. */
236 struct kref bin_bo_kref;
237 };
238
239 #define to_vc4_dev(_dev) \
240 container_of_const(_dev, struct vc4_dev, base)
241
242 struct vc4_bo {
243 struct drm_gem_dma_object base;
244
245 bool t_format;
246
247 /* List entry for the BO's position in either
248 * vc4_exec_info->unref_list or vc4_dev->bo_cache.time_list
249 */
250 struct list_head unref_head;
251
252 /* Time in jiffies when the BO was put in vc4->bo_cache. */
253 unsigned long free_time;
254
255 /* List entry for the BO's position in vc4_dev->bo_cache.size_list */
256 struct list_head size_head;
257
258 /* Struct for shader validation state, if created by
259 * DRM_IOCTL_VC4_CREATE_SHADER_BO.
260 */
261 struct vc4_validated_shader_info *validated_shader;
262
263 /* One of enum vc4_kernel_bo_type, or VC4_BO_TYPE_COUNT + i
264 * for user-allocated labels.
265 */
266 int label;
267
268 /* Count the number of active users. This is needed to determine
269 * whether we can move the BO to the purgeable list or not (when the BO
270 * is used by the GPU or the display engine we can't purge it).
271 */
272 refcount_t usecnt;
273
274 /* Store purgeable/purged state here */
275 u32 madv;
276 struct mutex madv_lock;
277 };
278
279 #define to_vc4_bo(_bo) \
280 container_of_const(to_drm_gem_dma_obj(_bo), struct vc4_bo, base)
281
282 struct vc4_fence {
283 struct dma_fence base;
284 struct drm_device *dev;
285 /* vc4 seqno for signaled() test */
286 uint64_t seqno;
287 };
288
289 #define to_vc4_fence(_fence) \
290 container_of_const(_fence, struct vc4_fence, base)
291
292 struct vc4_v3d {
293 struct vc4_dev *vc4;
294 struct platform_device *pdev;
295 void __iomem *regs;
296 struct clk *clk;
297 struct debugfs_regset32 regset;
298 };
299
300 #define VC4_NUM_UPM_HANDLES 32
301 struct vc4_upm_refcounts {
302 refcount_t refcount;
303
304 /* Allocation size */
305 size_t size;
306 /* Our allocation in UPM for prefetching. */
307 struct drm_mm_node upm;
308
309 /* Pointer back to the HVS structure */
310 struct vc4_hvs *hvs;
311 };
312
313 #define HVS_NUM_CHANNELS 3
314
315 struct vc4_hvs {
316 struct vc4_dev *vc4;
317 struct platform_device *pdev;
318 void __iomem *regs;
319 u32 __iomem *dlist;
320 unsigned int dlist_mem_size;
321
322 struct clk *core_clk;
323 struct clk *disp_clk;
324
325 unsigned long max_core_rate;
326
327 /* Memory manager for CRTCs to allocate space in the display
328 * list. Units are dwords.
329 */
330 struct drm_mm dlist_mm;
331
332 /* Memory manager for the LBM memory used by HVS scaling. */
333 struct drm_mm lbm_mm;
334
335 /* Memory manager for the UPM memory used for prefetching. */
336 struct drm_mm upm_mm;
337 struct ida upm_handles;
338 struct vc4_upm_refcounts upm_refcounts[VC4_NUM_UPM_HANDLES + 1];
339
340 spinlock_t mm_lock;
341
342 struct drm_mm_node mitchell_netravali_filter;
343
344 struct debugfs_regset32 regset;
345
346 /*
347 * Even if HDMI0 on the RPi4 can output modes requiring a pixel
348 * rate higher than 297MHz, it needs some adjustments in the
349 * config.txt file to be able to do so and thus won't always be
350 * available.
351 */
352 bool vc5_hdmi_enable_hdmi_20;
353
354 /*
355 * 4096x2160@60 requires a core overclock to work, so register
356 * whether that is sufficient.
357 */
358 bool vc5_hdmi_enable_4096by2160;
359 };
360
361 #define HVS_NUM_CHANNELS 3
362 #define HVS_UBM_WORD_SIZE 256
363
364 struct vc4_hvs_state {
365 struct drm_private_state base;
366 unsigned long core_clock_rate;
367
368 struct {
369 unsigned in_use: 1;
370 unsigned long fifo_load;
371 struct drm_crtc_commit *pending_commit;
372 } fifo_state[HVS_NUM_CHANNELS];
373 };
374
375 #define to_vc4_hvs_state(_state) \
376 container_of_const(_state, struct vc4_hvs_state, base)
377
378 struct vc4_hvs_state *vc4_hvs_get_global_state(struct drm_atomic_state *state);
379 struct vc4_hvs_state *vc4_hvs_get_old_global_state(const struct drm_atomic_state *state);
380 struct vc4_hvs_state *vc4_hvs_get_new_global_state(const struct drm_atomic_state *state);
381
382 struct vc4_plane {
383 struct drm_plane base;
384 };
385
386 #define to_vc4_plane(_plane) \
387 container_of_const(_plane, struct vc4_plane, base)
388
389 enum vc4_scaling_mode {
390 VC4_SCALING_NONE,
391 VC4_SCALING_TPZ,
392 VC4_SCALING_PPF,
393 };
394
395 struct vc4_plane_state {
396 struct drm_plane_state base;
397 /* System memory copy of the display list for this element, computed
398 * at atomic_check time.
399 */
400 u32 *dlist;
401 u32 dlist_size; /* Number of dwords allocated for the display list */
402 u32 dlist_count; /* Number of used dwords in the display list. */
403
404 /* Offset in the dlist to various words, for pageflip or
405 * cursor updates.
406 */
407 u32 pos0_offset;
408 u32 pos2_offset;
409 u32 ptr0_offset[DRM_FORMAT_MAX_PLANES];
410 u32 lbm_offset;
411
412 /* Offset where the plane's dlist was last stored in the
413 * hardware at vc4_crtc_atomic_flush() time.
414 */
415 u32 __iomem *hw_dlist;
416
417 /* Clipped coordinates of the plane on the display. */
418 int crtc_x, crtc_y, crtc_w, crtc_h;
419 /* Clipped area being scanned from in the FB in u16.16 format */
420 u32 src_x, src_y;
421
422 u32 src_w[2], src_h[2];
423
424 /* Scaling selection for the RGB/Y plane and the Cb/Cr planes. */
425 enum vc4_scaling_mode x_scaling[2], y_scaling[2];
426 bool is_unity;
427 bool is_yuv;
428
429 /* Our allocation in LBM for temporary storage during scaling. */
430 struct drm_mm_node lbm;
431
432 /* The Unified Pre-Fetcher Handle */
433 unsigned int upm_handle[DRM_FORMAT_MAX_PLANES];
434
435 /* Number of lines to pre-fetch */
436 unsigned int upm_buffer_lines;
437
438 /* Set when the plane has per-pixel alpha content or does not cover
439 * the entire screen. This is a hint to the CRTC that it might need
440 * to enable background color fill.
441 */
442 bool needs_bg_fill;
443
444 /* Mark the dlist as initialized. Useful to avoid initializing it twice
445 * when async update is not possible.
446 */
447 bool dlist_initialized;
448
449 /* Load of this plane on the HVS block. The load is expressed in HVS
450 * cycles/sec.
451 */
452 u64 hvs_load;
453
454 /* Memory bandwidth needed for this plane. This is expressed in
455 * bytes/sec.
456 */
457 u64 membus_load;
458 };
459
460 #define to_vc4_plane_state(_state) \
461 container_of_const(_state, struct vc4_plane_state, base)
462
463 enum vc4_encoder_type {
464 VC4_ENCODER_TYPE_NONE,
465 VC4_ENCODER_TYPE_HDMI0,
466 VC4_ENCODER_TYPE_HDMI1,
467 VC4_ENCODER_TYPE_VEC,
468 VC4_ENCODER_TYPE_DSI0,
469 VC4_ENCODER_TYPE_DSI1,
470 VC4_ENCODER_TYPE_SMI,
471 VC4_ENCODER_TYPE_DPI,
472 VC4_ENCODER_TYPE_TXP0,
473 VC4_ENCODER_TYPE_TXP1,
474 };
475
476 struct vc4_encoder {
477 struct drm_encoder base;
478 enum vc4_encoder_type type;
479 u32 clock_select;
480
481 void (*pre_crtc_configure)(struct drm_encoder *encoder, struct drm_atomic_state *state);
482 void (*pre_crtc_enable)(struct drm_encoder *encoder, struct drm_atomic_state *state);
483 void (*post_crtc_enable)(struct drm_encoder *encoder, struct drm_atomic_state *state);
484
485 void (*post_crtc_disable)(struct drm_encoder *encoder, struct drm_atomic_state *state);
486 void (*post_crtc_powerdown)(struct drm_encoder *encoder, struct drm_atomic_state *state);
487 };
488
489 #define to_vc4_encoder(_encoder) \
490 container_of_const(_encoder, struct vc4_encoder, base)
491
492 static inline
vc4_find_encoder_by_type(struct drm_device * drm,enum vc4_encoder_type type)493 struct drm_encoder *vc4_find_encoder_by_type(struct drm_device *drm,
494 enum vc4_encoder_type type)
495 {
496 struct drm_encoder *encoder;
497
498 drm_for_each_encoder(encoder, drm) {
499 struct vc4_encoder *vc4_encoder = to_vc4_encoder(encoder);
500
501 if (vc4_encoder->type == type)
502 return encoder;
503 }
504
505 return NULL;
506 }
507
508 struct vc4_crtc_data {
509 const char *name;
510
511 const char *debugfs_name;
512
513 /* Bitmask of channels (FIFOs) of the HVS that the output can source from */
514 unsigned int hvs_available_channels;
515
516 /* Which output of the HVS this pixelvalve sources from. */
517 int hvs_output;
518 };
519
520 struct vc4_txp_data {
521 struct vc4_crtc_data base;
522 enum vc4_encoder_type encoder_type;
523 unsigned int high_addr_ptr_reg;
524 unsigned int has_byte_enable:1;
525 unsigned int size_minus_one:1;
526 unsigned int supports_40bit_addresses:1;
527 };
528
529 extern const struct vc4_txp_data bcm2835_txp_data;
530
531 struct vc4_pv_data {
532 struct vc4_crtc_data base;
533
534 /* Depth of the PixelValve FIFO in bytes */
535 unsigned int fifo_depth;
536
537 /* Number of pixels output per clock period */
538 u8 pixels_per_clock;
539
540 enum vc4_encoder_type encoder_types[4];
541 };
542
543 extern const struct vc4_pv_data bcm2835_pv0_data;
544 extern const struct vc4_pv_data bcm2835_pv1_data;
545 extern const struct vc4_pv_data bcm2835_pv2_data;
546 extern const struct vc4_pv_data bcm2711_pv0_data;
547 extern const struct vc4_pv_data bcm2711_pv1_data;
548 extern const struct vc4_pv_data bcm2711_pv2_data;
549 extern const struct vc4_pv_data bcm2711_pv3_data;
550 extern const struct vc4_pv_data bcm2711_pv4_data;
551 extern const struct vc4_pv_data bcm2712_pv0_data;
552 extern const struct vc4_pv_data bcm2712_pv1_data;
553
554 struct vc4_crtc {
555 struct drm_crtc base;
556 struct platform_device *pdev;
557 const struct vc4_crtc_data *data;
558 void __iomem *regs;
559
560 /* Timestamp at start of vblank irq - unaffected by lock delays. */
561 ktime_t t_vblank;
562
563 u8 lut_r[256];
564 u8 lut_g[256];
565 u8 lut_b[256];
566
567 struct drm_pending_vblank_event *event;
568
569 struct debugfs_regset32 regset;
570
571 /**
572 * @feeds_txp: True if the CRTC feeds our writeback controller.
573 */
574 bool feeds_txp;
575
576 /**
577 * @irq_lock: Spinlock protecting the resources shared between
578 * the atomic code and our vblank handler.
579 */
580 spinlock_t irq_lock;
581
582 /**
583 * @current_dlist: Start offset of the display list currently
584 * set in the HVS for that CRTC. Protected by @irq_lock, and
585 * copied in vc4_hvs_update_dlist() for the CRTC interrupt
586 * handler to have access to that value.
587 */
588 unsigned int current_dlist;
589
590 /**
591 * @current_hvs_channel: HVS channel currently assigned to the
592 * CRTC. Protected by @irq_lock, and copied in
593 * vc4_hvs_atomic_begin() for the CRTC interrupt handler to have
594 * access to that value.
595 */
596 unsigned int current_hvs_channel;
597 };
598
599 #define to_vc4_crtc(_crtc) \
600 container_of_const(_crtc, struct vc4_crtc, base)
601
602 static inline const struct vc4_crtc_data *
vc4_crtc_to_vc4_crtc_data(const struct vc4_crtc * crtc)603 vc4_crtc_to_vc4_crtc_data(const struct vc4_crtc *crtc)
604 {
605 return crtc->data;
606 }
607
608 static inline const struct vc4_pv_data *
vc4_crtc_to_vc4_pv_data(const struct vc4_crtc * crtc)609 vc4_crtc_to_vc4_pv_data(const struct vc4_crtc *crtc)
610 {
611 const struct vc4_crtc_data *data = vc4_crtc_to_vc4_crtc_data(crtc);
612
613 return container_of_const(data, struct vc4_pv_data, base);
614 }
615
616 struct drm_encoder *vc4_get_crtc_encoder(struct drm_crtc *crtc,
617 struct drm_crtc_state *state);
618
619 struct vc4_crtc_state {
620 struct drm_crtc_state base;
621 /* Dlist area for this CRTC configuration. */
622 struct drm_mm_node mm;
623 bool txp_armed;
624 unsigned int assigned_channel;
625
626 struct drm_connector_tv_margins margins;
627
628 unsigned long hvs_load;
629
630 /* Transitional state below, only valid during atomic commits */
631 bool update_muxing;
632 };
633
634 #define VC4_HVS_CHANNEL_DISABLED ((unsigned int)-1)
635
636 #define to_vc4_crtc_state(_state) \
637 container_of_const(_state, struct vc4_crtc_state, base)
638
639 #define V3D_READ(offset) \
640 ({ \
641 kunit_fail_current_test("Accessing a register in a unit test!\n"); \
642 readl(vc4->v3d->regs + (offset)); \
643 })
644
645 #define V3D_WRITE(offset, val) \
646 do { \
647 kunit_fail_current_test("Accessing a register in a unit test!\n"); \
648 writel(val, vc4->v3d->regs + (offset)); \
649 } while (0)
650
651 #define HVS_READ(offset) \
652 ({ \
653 kunit_fail_current_test("Accessing a register in a unit test!\n"); \
654 readl(hvs->regs + (offset)); \
655 })
656
657 #define HVS_WRITE(offset, val) \
658 do { \
659 kunit_fail_current_test("Accessing a register in a unit test!\n"); \
660 writel(val, hvs->regs + (offset)); \
661 } while (0)
662
663 #define HVS_READ6(offset) \
664 HVS_READ(hvs->vc4->gen == VC4_GEN_6_C ? SCALER6_ ## offset : SCALER6D_ ## offset)
665
666 #define HVS_WRITE6(offset, val) \
667 HVS_WRITE(hvs->vc4->gen == VC4_GEN_6_C ? SCALER6_ ## offset : SCALER6D_ ## offset, val)
668
669 #define VC4_REG32(reg) { .name = #reg, .offset = reg }
670
671 struct vc4_exec_info {
672 struct vc4_dev *dev;
673
674 /* Sequence number for this bin/render job. */
675 uint64_t seqno;
676
677 struct dma_fence *fence;
678
679 /* Last current addresses the hardware was processing when the
680 * hangcheck timer checked on us.
681 */
682 uint32_t last_ct0ca, last_ct1ca;
683
684 /* Kernel-space copy of the ioctl arguments */
685 struct drm_vc4_submit_cl *args;
686
687 /* This is the array of BOs that were looked up at the start of exec.
688 * Command validation will use indices into this array.
689 */
690 struct drm_gem_object **bo;
691 uint32_t bo_count;
692
693 /* List of BOs that are being written by the RCL. Other than
694 * the binner temporary storage, this is all the BOs written
695 * by the job.
696 */
697 struct drm_gem_dma_object *rcl_write_bo[4];
698 uint32_t rcl_write_bo_count;
699
700 /* Pointers for our position in vc4->job_list */
701 struct list_head head;
702
703 /* List of other BOs used in the job that need to be released
704 * once the job is complete.
705 */
706 struct list_head unref_list;
707
708 /* Current unvalidated indices into @bo loaded by the non-hardware
709 * VC4_PACKET_GEM_HANDLES.
710 */
711 uint32_t bo_index[2];
712
713 /* This is the BO where we store the validated command lists, shader
714 * records, and uniforms.
715 */
716 struct drm_gem_dma_object *exec_bo;
717
718 /**
719 * This tracks the per-shader-record state (packet 64) that
720 * determines the length of the shader record and the offset
721 * it's expected to be found at. It gets read in from the
722 * command lists.
723 */
724 struct vc4_shader_state {
725 uint32_t addr;
726 /* Maximum vertex index referenced by any primitive using this
727 * shader state.
728 */
729 uint32_t max_index;
730 } *shader_state;
731
732 /** How many shader states the user declared they were using. */
733 uint32_t shader_state_size;
734 /** How many shader state records the validator has seen. */
735 uint32_t shader_state_count;
736
737 bool found_tile_binning_mode_config_packet;
738 bool found_start_tile_binning_packet;
739 bool found_increment_semaphore_packet;
740 bool found_flush;
741 uint8_t bin_tiles_x, bin_tiles_y;
742 /* Physical address of the start of the tile alloc array
743 * (where each tile's binned CL will start)
744 */
745 uint32_t tile_alloc_offset;
746 /* Bitmask of which binner slots are freed when this job completes. */
747 uint32_t bin_slots;
748
749 /**
750 * Computed addresses pointing into exec_bo where we start the
751 * bin thread (ct0) and render thread (ct1).
752 */
753 uint32_t ct0ca, ct0ea;
754 uint32_t ct1ca, ct1ea;
755
756 /* Pointer to the unvalidated bin CL (if present). */
757 void *bin_u;
758
759 /* Pointers to the shader recs. These paddr gets incremented as CL
760 * packets are relocated in validate_gl_shader_state, and the vaddrs
761 * (u and v) get incremented and size decremented as the shader recs
762 * themselves are validated.
763 */
764 void *shader_rec_u;
765 void *shader_rec_v;
766 uint32_t shader_rec_p;
767 uint32_t shader_rec_size;
768
769 /* Pointers to the uniform data. These pointers are incremented, and
770 * size decremented, as each batch of uniforms is uploaded.
771 */
772 void *uniforms_u;
773 void *uniforms_v;
774 uint32_t uniforms_p;
775 uint32_t uniforms_size;
776
777 /* Pointer to a performance monitor object if the user requested it,
778 * NULL otherwise.
779 */
780 struct vc4_perfmon *perfmon;
781
782 /* Whether the exec has taken a reference to the binner BO, which should
783 * happen with a VC4_PACKET_TILE_BINNING_MODE_CONFIG packet.
784 */
785 bool bin_bo_used;
786 };
787
788 /* Per-open file private data. Any driver-specific resource that has to be
789 * released when the DRM file is closed should be placed here.
790 */
791 struct vc4_file {
792 struct vc4_dev *dev;
793
794 struct {
795 struct idr idr;
796 struct mutex lock;
797 } perfmon;
798
799 bool bin_bo_used;
800 };
801
802 static inline struct vc4_exec_info *
vc4_first_bin_job(struct vc4_dev * vc4)803 vc4_first_bin_job(struct vc4_dev *vc4)
804 {
805 return list_first_entry_or_null(&vc4->bin_job_list,
806 struct vc4_exec_info, head);
807 }
808
809 static inline struct vc4_exec_info *
vc4_first_render_job(struct vc4_dev * vc4)810 vc4_first_render_job(struct vc4_dev *vc4)
811 {
812 return list_first_entry_or_null(&vc4->render_job_list,
813 struct vc4_exec_info, head);
814 }
815
816 static inline struct vc4_exec_info *
vc4_last_render_job(struct vc4_dev * vc4)817 vc4_last_render_job(struct vc4_dev *vc4)
818 {
819 if (list_empty(&vc4->render_job_list))
820 return NULL;
821 return list_last_entry(&vc4->render_job_list,
822 struct vc4_exec_info, head);
823 }
824
825 /**
826 * struct vc4_texture_sample_info - saves the offsets into the UBO for texture
827 * setup parameters.
828 *
829 * This will be used at draw time to relocate the reference to the texture
830 * contents in p0, and validate that the offset combined with
831 * width/height/stride/etc. from p1 and p2/p3 doesn't sample outside the BO.
832 * Note that the hardware treats unprovided config parameters as 0, so not all
833 * of them need to be set up for every texure sample, and we'll store ~0 as
834 * the offset to mark the unused ones.
835 *
836 * See the VC4 3D architecture guide page 41 ("Texture and Memory Lookup Unit
837 * Setup") for definitions of the texture parameters.
838 */
839 struct vc4_texture_sample_info {
840 bool is_direct;
841 uint32_t p_offset[4];
842 };
843
844 /**
845 * struct vc4_validated_shader_info - information about validated shaders that
846 * needs to be used from command list validation.
847 *
848 * For a given shader, each time a shader state record references it, we need
849 * to verify that the shader doesn't read more uniforms than the shader state
850 * record's uniform BO pointer can provide, and we need to apply relocations
851 * and validate the shader state record's uniforms that define the texture
852 * samples.
853 */
854 struct vc4_validated_shader_info {
855 uint32_t uniforms_size;
856 uint32_t uniforms_src_size;
857 uint32_t num_texture_samples;
858 struct vc4_texture_sample_info *texture_samples;
859
860 uint32_t num_uniform_addr_offsets;
861 uint32_t *uniform_addr_offsets;
862
863 bool is_threaded;
864 };
865
866 /**
867 * __wait_for - magic wait macro
868 *
869 * Macro to help avoid open coding check/wait/timeout patterns. Note that it's
870 * important that we check the condition again after having timed out, since the
871 * timeout could be due to preemption or similar and we've never had a chance to
872 * check the condition before the timeout.
873 */
874 #define __wait_for(OP, COND, US, Wmin, Wmax) ({ \
875 const ktime_t end__ = ktime_add_ns(ktime_get_raw(), 1000ll * (US)); \
876 long wait__ = (Wmin); /* recommended min for usleep is 10 us */ \
877 int ret__; \
878 might_sleep(); \
879 for (;;) { \
880 const bool expired__ = ktime_after(ktime_get_raw(), end__); \
881 OP; \
882 /* Guarantee COND check prior to timeout */ \
883 barrier(); \
884 if (COND) { \
885 ret__ = 0; \
886 break; \
887 } \
888 if (expired__) { \
889 ret__ = -ETIMEDOUT; \
890 break; \
891 } \
892 usleep_range(wait__, wait__ * 2); \
893 if (wait__ < (Wmax)) \
894 wait__ <<= 1; \
895 } \
896 ret__; \
897 })
898
899 #define _wait_for(COND, US, Wmin, Wmax) __wait_for(, (COND), (US), (Wmin), \
900 (Wmax))
901 #define wait_for(COND, MS) _wait_for((COND), (MS) * 1000, 10, 1000)
902
903 /* vc4_bo.c */
904 struct drm_gem_object *vc4_create_object(struct drm_device *dev, size_t size);
905 struct vc4_bo *vc4_bo_create(struct drm_device *dev, size_t size,
906 bool from_cache, enum vc4_kernel_bo_type type);
907 int vc4_bo_dumb_create(struct drm_file *file_priv,
908 struct drm_device *dev,
909 struct drm_mode_create_dumb *args);
910 int vc4_create_bo_ioctl(struct drm_device *dev, void *data,
911 struct drm_file *file_priv);
912 int vc4_create_shader_bo_ioctl(struct drm_device *dev, void *data,
913 struct drm_file *file_priv);
914 int vc4_mmap_bo_ioctl(struct drm_device *dev, void *data,
915 struct drm_file *file_priv);
916 int vc4_set_tiling_ioctl(struct drm_device *dev, void *data,
917 struct drm_file *file_priv);
918 int vc4_get_tiling_ioctl(struct drm_device *dev, void *data,
919 struct drm_file *file_priv);
920 int vc4_get_hang_state_ioctl(struct drm_device *dev, void *data,
921 struct drm_file *file_priv);
922 int vc4_label_bo_ioctl(struct drm_device *dev, void *data,
923 struct drm_file *file_priv);
924 int vc4_bo_cache_init(struct drm_device *dev);
925 int vc4_bo_inc_usecnt(struct vc4_bo *bo);
926 void vc4_bo_dec_usecnt(struct vc4_bo *bo);
927 void vc4_bo_add_to_purgeable_pool(struct vc4_bo *bo);
928 void vc4_bo_remove_from_purgeable_pool(struct vc4_bo *bo);
929 int vc4_bo_debugfs_init(struct drm_minor *minor);
930
931 /* vc4_crtc.c */
932 extern struct platform_driver vc4_crtc_driver;
933 int vc4_crtc_disable_at_boot(struct drm_crtc *crtc);
934 int __vc4_crtc_init(struct drm_device *drm, struct platform_device *pdev,
935 struct vc4_crtc *vc4_crtc, const struct vc4_crtc_data *data,
936 struct drm_plane *primary_plane,
937 const struct drm_crtc_funcs *crtc_funcs,
938 const struct drm_crtc_helper_funcs *crtc_helper_funcs,
939 bool feeds_txp);
940 int vc4_crtc_init(struct drm_device *drm, struct platform_device *pdev,
941 struct vc4_crtc *vc4_crtc, const struct vc4_crtc_data *data,
942 const struct drm_crtc_funcs *crtc_funcs,
943 const struct drm_crtc_helper_funcs *crtc_helper_funcs,
944 bool feeds_txp);
945 int vc4_page_flip(struct drm_crtc *crtc,
946 struct drm_framebuffer *fb,
947 struct drm_pending_vblank_event *event,
948 uint32_t flags,
949 struct drm_modeset_acquire_ctx *ctx);
950 int vc4_crtc_atomic_check(struct drm_crtc *crtc,
951 struct drm_atomic_state *state);
952 struct drm_crtc_state *vc4_crtc_duplicate_state(struct drm_crtc *crtc);
953 void vc4_crtc_destroy_state(struct drm_crtc *crtc,
954 struct drm_crtc_state *state);
955 void vc4_crtc_reset(struct drm_crtc *crtc);
956 void vc4_crtc_handle_vblank(struct vc4_crtc *crtc);
957 void vc4_crtc_send_vblank(struct drm_crtc *crtc);
958 int vc4_crtc_late_register(struct drm_crtc *crtc);
959 void vc4_crtc_get_margins(struct drm_crtc_state *state,
960 unsigned int *left, unsigned int *right,
961 unsigned int *top, unsigned int *bottom);
962
963 /* vc4_debugfs.c */
964 void vc4_debugfs_init(struct drm_minor *minor);
965 #ifdef CONFIG_DEBUG_FS
966 void vc4_debugfs_add_regset32(struct drm_device *drm,
967 const char *filename,
968 struct debugfs_regset32 *regset);
969 #else
970
vc4_debugfs_add_regset32(struct drm_device * drm,const char * filename,struct debugfs_regset32 * regset)971 static inline void vc4_debugfs_add_regset32(struct drm_device *drm,
972 const char *filename,
973 struct debugfs_regset32 *regset)
974 {}
975 #endif
976
977 /* vc4_drv.c */
978 void __iomem *vc4_ioremap_regs(struct platform_device *dev, int index);
979 int vc4_dumb_fixup_args(struct drm_mode_create_dumb *args);
980
981 /* vc4_dpi.c */
982 extern struct platform_driver vc4_dpi_driver;
983
984 /* vc4_dsi.c */
985 extern struct platform_driver vc4_dsi_driver;
986
987 /* vc4_fence.c */
988 extern const struct dma_fence_ops vc4_fence_ops;
989
990 /* vc4_gem.c */
991 int vc4_gem_init(struct drm_device *dev);
992 int vc4_submit_cl_ioctl(struct drm_device *dev, void *data,
993 struct drm_file *file_priv);
994 int vc4_wait_seqno_ioctl(struct drm_device *dev, void *data,
995 struct drm_file *file_priv);
996 int vc4_wait_bo_ioctl(struct drm_device *dev, void *data,
997 struct drm_file *file_priv);
998 void vc4_submit_next_bin_job(struct drm_device *dev);
999 void vc4_submit_next_render_job(struct drm_device *dev);
1000 void vc4_move_job_to_render(struct drm_device *dev, struct vc4_exec_info *exec);
1001 int vc4_wait_for_seqno(struct drm_device *dev, uint64_t seqno,
1002 uint64_t timeout_ns, bool interruptible);
1003 void vc4_job_handle_completed(struct vc4_dev *vc4);
1004 int vc4_gem_madvise_ioctl(struct drm_device *dev, void *data,
1005 struct drm_file *file_priv);
1006
1007 /* vc4_hdmi.c */
1008 extern struct platform_driver vc4_hdmi_driver;
1009
1010 /* vc4_vec.c */
1011 extern struct platform_driver vc4_vec_driver;
1012
1013 /* vc4_txp.c */
1014 extern struct platform_driver vc4_txp_driver;
1015
1016 /* vc4_irq.c */
1017 void vc4_irq_enable(struct drm_device *dev);
1018 void vc4_irq_disable(struct drm_device *dev);
1019 int vc4_irq_install(struct drm_device *dev, int irq);
1020 void vc4_irq_uninstall(struct drm_device *dev);
1021 void vc4_irq_reset(struct drm_device *dev);
1022
1023 /* vc4_hvs.c */
1024 extern struct platform_driver vc4_hvs_driver;
1025 struct vc4_hvs *__vc4_hvs_alloc(struct vc4_dev *vc4,
1026 void __iomem *regs,
1027 struct platform_device *pdev);
1028 void vc4_hvs_stop_channel(struct vc4_hvs *hvs, unsigned int output);
1029 int vc4_hvs_get_fifo_from_output(struct vc4_hvs *hvs, unsigned int output);
1030 u8 vc4_hvs_get_fifo_frame_count(struct vc4_hvs *hvs, unsigned int fifo);
1031 int vc4_hvs_atomic_check(struct drm_crtc *crtc, struct drm_atomic_state *state);
1032 void vc4_hvs_atomic_begin(struct drm_crtc *crtc, struct drm_atomic_state *state);
1033 void vc4_hvs_atomic_enable(struct drm_crtc *crtc, struct drm_atomic_state *state);
1034 void vc4_hvs_atomic_disable(struct drm_crtc *crtc, struct drm_atomic_state *state);
1035 void vc4_hvs_atomic_flush(struct drm_crtc *crtc, struct drm_atomic_state *state);
1036 void vc4_hvs_dump_state(struct vc4_hvs *hvs);
1037 void vc4_hvs_unmask_underrun(struct vc4_hvs *hvs, int channel);
1038 void vc4_hvs_mask_underrun(struct vc4_hvs *hvs, int channel);
1039 int vc4_hvs_debugfs_init(struct drm_minor *minor);
1040
1041 /* vc4_kms.c */
1042 int vc4_kms_load(struct drm_device *dev);
1043
1044 /* vc4_plane.c */
1045 struct drm_plane *vc4_plane_init(struct drm_device *dev,
1046 enum drm_plane_type type,
1047 uint32_t possible_crtcs);
1048 int vc4_plane_create_additional_planes(struct drm_device *dev);
1049 u32 vc4_plane_write_dlist(struct drm_plane *plane, u32 __iomem *dlist);
1050 u32 vc4_plane_dlist_size(const struct drm_plane_state *state);
1051 void vc4_plane_async_set_fb(struct drm_plane *plane,
1052 struct drm_framebuffer *fb);
1053
1054 /* vc4_v3d.c */
1055 extern struct platform_driver vc4_v3d_driver;
1056 extern const struct of_device_id vc4_v3d_dt_match[];
1057 int vc4_v3d_get_bin_slot(struct vc4_dev *vc4);
1058 int vc4_v3d_bin_bo_get(struct vc4_dev *vc4, bool *used);
1059 void vc4_v3d_bin_bo_put(struct vc4_dev *vc4);
1060 int vc4_v3d_pm_get(struct vc4_dev *vc4);
1061 void vc4_v3d_pm_put(struct vc4_dev *vc4);
1062 int vc4_v3d_debugfs_init(struct drm_minor *minor);
1063
1064 /* vc4_validate.c */
1065 int
1066 vc4_validate_bin_cl(struct drm_device *dev,
1067 void *validated,
1068 void *unvalidated,
1069 struct vc4_exec_info *exec);
1070
1071 int
1072 vc4_validate_shader_recs(struct drm_device *dev, struct vc4_exec_info *exec);
1073
1074 struct drm_gem_dma_object *vc4_use_bo(struct vc4_exec_info *exec,
1075 uint32_t hindex);
1076
1077 int vc4_get_rcl(struct drm_device *dev, struct vc4_exec_info *exec);
1078
1079 bool vc4_check_tex_size(struct vc4_exec_info *exec,
1080 struct drm_gem_dma_object *fbo,
1081 uint32_t offset, uint8_t tiling_format,
1082 uint32_t width, uint32_t height, uint8_t cpp);
1083
1084 /* vc4_validate_shader.c */
1085 struct vc4_validated_shader_info *
1086 vc4_validate_shader(struct drm_gem_dma_object *shader_obj);
1087
1088 /* vc4_perfmon.c */
1089 void vc4_perfmon_get(struct vc4_perfmon *perfmon);
1090 void vc4_perfmon_put(struct vc4_perfmon *perfmon);
1091 void vc4_perfmon_start(struct vc4_dev *vc4, struct vc4_perfmon *perfmon);
1092 void vc4_perfmon_stop(struct vc4_dev *vc4, struct vc4_perfmon *perfmon,
1093 bool capture);
1094 struct vc4_perfmon *vc4_perfmon_find(struct vc4_file *vc4file, int id);
1095 void vc4_perfmon_open_file(struct vc4_file *vc4file);
1096 void vc4_perfmon_close_file(struct vc4_file *vc4file);
1097 int vc4_perfmon_create_ioctl(struct drm_device *dev, void *data,
1098 struct drm_file *file_priv);
1099 int vc4_perfmon_destroy_ioctl(struct drm_device *dev, void *data,
1100 struct drm_file *file_priv);
1101 int vc4_perfmon_get_values_ioctl(struct drm_device *dev, void *data,
1102 struct drm_file *file_priv);
1103
1104 #endif /* _VC4_DRV_H_ */
1105