projects/linux·blob master · refs

793 lines · 23.8 KiB

1/*
2 * Copyright © 2014-2018 Broadcom
3 *
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
10 *
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
13 * Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
21 * IN THE SOFTWARE.
22 */
23
24#ifndef _V3D_DRM_H_
25#define _V3D_DRM_H_
26
27#include "drm.h"
28
29#if defined(__cplusplus)
30extern "C" {
31#endif
32
33#define DRM_V3D_SUBMIT_CL 0x00
34#define DRM_V3D_WAIT_BO 0x01
35#define DRM_V3D_CREATE_BO 0x02
36#define DRM_V3D_MMAP_BO 0x03
37#define DRM_V3D_GET_PARAM 0x04
38#define DRM_V3D_GET_BO_OFFSET 0x05
39#define DRM_V3D_SUBMIT_TFU 0x06
40#define DRM_V3D_SUBMIT_CSD 0x07
41#define DRM_V3D_PERFMON_CREATE 0x08
42#define DRM_V3D_PERFMON_DESTROY 0x09
43#define DRM_V3D_PERFMON_GET_VALUES 0x0a
44#define DRM_V3D_SUBMIT_CPU 0x0b
45#define DRM_V3D_PERFMON_GET_COUNTER 0x0c
46#define DRM_V3D_PERFMON_SET_GLOBAL 0x0d
47
48#define DRM_IOCTL_V3D_SUBMIT_CL DRM_IOWR(DRM_COMMAND_BASE + DRM_V3D_SUBMIT_CL, struct drm_v3d_submit_cl)
49#define DRM_IOCTL_V3D_WAIT_BO DRM_IOWR(DRM_COMMAND_BASE + DRM_V3D_WAIT_BO, struct drm_v3d_wait_bo)
50#define DRM_IOCTL_V3D_CREATE_BO DRM_IOWR(DRM_COMMAND_BASE + DRM_V3D_CREATE_BO, struct drm_v3d_create_bo)
51#define DRM_IOCTL_V3D_MMAP_BO DRM_IOWR(DRM_COMMAND_BASE + DRM_V3D_MMAP_BO, struct drm_v3d_mmap_bo)
52#define DRM_IOCTL_V3D_GET_PARAM DRM_IOWR(DRM_COMMAND_BASE + DRM_V3D_GET_PARAM, struct drm_v3d_get_param)
53#define DRM_IOCTL_V3D_GET_BO_OFFSET DRM_IOWR(DRM_COMMAND_BASE + DRM_V3D_GET_BO_OFFSET, struct drm_v3d_get_bo_offset)
54#define DRM_IOCTL_V3D_SUBMIT_TFU DRM_IOW(DRM_COMMAND_BASE + DRM_V3D_SUBMIT_TFU, struct drm_v3d_submit_tfu)
55#define DRM_IOCTL_V3D_SUBMIT_CSD DRM_IOW(DRM_COMMAND_BASE + DRM_V3D_SUBMIT_CSD, struct drm_v3d_submit_csd)
56#define DRM_IOCTL_V3D_PERFMON_CREATE DRM_IOWR(DRM_COMMAND_BASE + DRM_V3D_PERFMON_CREATE, \
57 struct drm_v3d_perfmon_create)
58#define DRM_IOCTL_V3D_PERFMON_DESTROY DRM_IOWR(DRM_COMMAND_BASE + DRM_V3D_PERFMON_DESTROY, \
59 struct drm_v3d_perfmon_destroy)
60#define DRM_IOCTL_V3D_PERFMON_GET_VALUES DRM_IOWR(DRM_COMMAND_BASE + DRM_V3D_PERFMON_GET_VALUES, \
61 struct drm_v3d_perfmon_get_values)
62#define DRM_IOCTL_V3D_SUBMIT_CPU DRM_IOW(DRM_COMMAND_BASE + DRM_V3D_SUBMIT_CPU, struct drm_v3d_submit_cpu)
63#define DRM_IOCTL_V3D_PERFMON_GET_COUNTER DRM_IOWR(DRM_COMMAND_BASE + DRM_V3D_PERFMON_GET_COUNTER, \
64 struct drm_v3d_perfmon_get_counter)
65#define DRM_IOCTL_V3D_PERFMON_SET_GLOBAL DRM_IOW(DRM_COMMAND_BASE + DRM_V3D_PERFMON_SET_GLOBAL, \
66 struct drm_v3d_perfmon_set_global)
67
68#define DRM_V3D_SUBMIT_CL_FLUSH_CACHE 0x01
69#define DRM_V3D_SUBMIT_EXTENSION 0x02
70
71/* struct drm_v3d_extension - ioctl extensions
72 *
73 * Linked-list of generic extensions where the id identify which struct is
74 * pointed by ext_data. Therefore, DRM_V3D_EXT_ID_* is used on id to identify
75 * the extension type.
76 */
77struct drm_v3d_extension {
78 __u64 next;
79 __u32 id;
80#define DRM_V3D_EXT_ID_MULTI_SYNC 0x01
81#define DRM_V3D_EXT_ID_CPU_INDIRECT_CSD 0x02
82#define DRM_V3D_EXT_ID_CPU_TIMESTAMP_QUERY 0x03
83#define DRM_V3D_EXT_ID_CPU_RESET_TIMESTAMP_QUERY 0x04
84#define DRM_V3D_EXT_ID_CPU_COPY_TIMESTAMP_QUERY 0x05
85#define DRM_V3D_EXT_ID_CPU_RESET_PERFORMANCE_QUERY 0x06
86#define DRM_V3D_EXT_ID_CPU_COPY_PERFORMANCE_QUERY 0x07
87 __u32 flags; /* mbz */
88};
89
90/* struct drm_v3d_sem - wait/signal semaphore
91 *
92 * If binary semaphore, it only takes syncobj handle and ignores flags and
93 * point fields. Point is defined for timeline syncobj feature.
94 */
95struct drm_v3d_sem {
96 __u32 handle; /* syncobj */
97 /* rsv below, for future uses */
98 __u32 flags;
99 __u64 point; /* for timeline sem support */
100 __u64 mbz[2]; /* must be zero, rsv */
101};
102
103/* Enum for each of the V3D queues. */
104enum v3d_queue {
105 V3D_BIN,
106 V3D_RENDER,
107 V3D_TFU,
108 V3D_CSD,
109 V3D_CACHE_CLEAN,
110 V3D_CPU,
111};
112
113/**
114 * struct drm_v3d_multi_sync - ioctl extension to add support multiples
115 * syncobjs for commands submission.
116 *
117 * When an extension of DRM_V3D_EXT_ID_MULTI_SYNC id is defined, it points to
118 * this extension to define wait and signal dependencies, instead of single
119 * in/out sync entries on submitting commands. The field flags is used to
120 * determine the stage to set wait dependencies.
121 */
122struct drm_v3d_multi_sync {
123 struct drm_v3d_extension base;
124 /* Array of wait and signal semaphores */
125 __u64 in_syncs;
126 __u64 out_syncs;
127
128 /* Number of entries */
129 __u32 in_sync_count;
130 __u32 out_sync_count;
131
132 /* set the stage (v3d_queue) to sync */
133 __u32 wait_stage;
134
135 __u32 pad; /* mbz */
136};
137
138/**
139 * struct drm_v3d_submit_cl - ioctl argument for submitting commands to the 3D
140 * engine.
141 *
142 * This asks the kernel to have the GPU execute an optional binner
143 * command list, and a render command list.
144 *
145 * The L1T, slice, L2C, L2T, and GCA caches will be flushed before
146 * each CL executes. The VCD cache should be flushed (if necessary)
147 * by the submitted CLs. The TLB writes are guaranteed to have been
148 * flushed by the time the render done IRQ happens, which is the
149 * trigger for out_sync. Any dirtying of cachelines by the job (only
150 * possible using TMU writes) must be flushed by the caller using the
151 * DRM_V3D_SUBMIT_CL_FLUSH_CACHE_FLAG flag.
152 */
153struct drm_v3d_submit_cl {
154 /* Pointer to the binner command list.
155 *
156 * This is the first set of commands executed, which runs the
157 * coordinate shader to determine where primitives land on the screen,
158 * then writes out the state updates and draw calls necessary per tile
159 * to the tile allocation BO.
160 *
161 * This BCL will block on any previous BCL submitted on the
162 * same FD, but not on any RCL or BCLs submitted by other
163 * clients -- that is left up to the submitter to control
164 * using in_sync_bcl if necessary.
165 */
166 __u32 bcl_start;
167
168 /** End address of the BCL (first byte after the BCL) */
169 __u32 bcl_end;
170
171 /* Offset of the render command list.
172 *
173 * This is the second set of commands executed, which will either
174 * execute the tiles that have been set up by the BCL, or a fixed set
175 * of tiles (in the case of RCL-only blits).
176 *
177 * This RCL will block on this submit's BCL, and any previous
178 * RCL submitted on the same FD, but not on any RCL or BCLs
179 * submitted by other clients -- that is left up to the
180 * submitter to control using in_sync_rcl if necessary.
181 */
182 __u32 rcl_start;
183
184 /** End address of the RCL (first byte after the RCL) */
185 __u32 rcl_end;
186
187 /** An optional sync object to wait on before starting the BCL. */
188 __u32 in_sync_bcl;
189 /** An optional sync object to wait on before starting the RCL. */
190 __u32 in_sync_rcl;
191 /** An optional sync object to place the completion fence in. */
192 __u32 out_sync;
193
194 /* Offset of the tile alloc memory
195 *
196 * This is optional on V3D 3.3 (where the CL can set the value) but
197 * required on V3D 4.1.
198 */
199 __u32 qma;
200
201 /** Size of the tile alloc memory. */
202 __u32 qms;
203
204 /** Offset of the tile state data array. */
205 __u32 qts;
206
207 /* Pointer to a u32 array of the BOs that are referenced by the job.
208 */
209 __u64 bo_handles;
210
211 /* Number of BO handles passed in (size is that times 4). */
212 __u32 bo_handle_count;
213
214 /* DRM_V3D_SUBMIT_* properties */
215 __u32 flags;
216
217 /* ID of the perfmon to attach to this job. 0 means no perfmon. */
218 __u32 perfmon_id;
219
220 __u32 pad;
221
222 /* Pointer to an array of ioctl extensions*/
223 __u64 extensions;
224};
225
226/**
227 * struct drm_v3d_wait_bo - ioctl argument for waiting for
228 * completion of the last DRM_V3D_SUBMIT_CL on a BO.
229 *
230 * This is useful for cases where multiple processes might be
231 * rendering to a BO and you want to wait for all rendering to be
232 * completed.
233 */
234struct drm_v3d_wait_bo {
235 __u32 handle;
236 __u32 pad;
237 __u64 timeout_ns;
238};
239
240/**
241 * struct drm_v3d_create_bo - ioctl argument for creating V3D BOs.
242 *
243 * There are currently no values for the flags argument, but it may be
244 * used in a future extension.
245 */
246struct drm_v3d_create_bo {
247 __u32 size;
248 __u32 flags;
249 /** Returned GEM handle for the BO. */
250 __u32 handle;
251 /**
252 * Returned offset for the BO in the V3D address space. This offset
253 * is private to the DRM fd and is valid for the lifetime of the GEM
254 * handle.
255 *
256 * This offset value will always be nonzero, since various HW
257 * units treat 0 specially.
258 */
259 __u32 offset;
260};
261
262/**
263 * struct drm_v3d_mmap_bo - ioctl argument for mapping V3D BOs.
264 *
265 * This doesn't actually perform an mmap. Instead, it returns the
266 * offset you need to use in an mmap on the DRM device node. This
267 * means that tools like valgrind end up knowing about the mapped
268 * memory.
269 *
270 * There are currently no values for the flags argument, but it may be
271 * used in a future extension.
272 */
273struct drm_v3d_mmap_bo {
274 /** Handle for the object being mapped. */
275 __u32 handle;
276 __u32 flags;
277 /** offset into the drm node to use for subsequent mmap call. */
278 __u64 offset;
279};
280
281enum drm_v3d_param {
282 DRM_V3D_PARAM_V3D_UIFCFG,
283 DRM_V3D_PARAM_V3D_HUB_IDENT1,
284 DRM_V3D_PARAM_V3D_HUB_IDENT2,
285 DRM_V3D_PARAM_V3D_HUB_IDENT3,
286 DRM_V3D_PARAM_V3D_CORE0_IDENT0,
287 DRM_V3D_PARAM_V3D_CORE0_IDENT1,
288 DRM_V3D_PARAM_V3D_CORE0_IDENT2,
289 DRM_V3D_PARAM_SUPPORTS_TFU,
290 DRM_V3D_PARAM_SUPPORTS_CSD,
291 DRM_V3D_PARAM_SUPPORTS_CACHE_FLUSH,
292 DRM_V3D_PARAM_SUPPORTS_PERFMON,
293 DRM_V3D_PARAM_SUPPORTS_MULTISYNC_EXT,
294 DRM_V3D_PARAM_SUPPORTS_CPU_QUEUE,
295 DRM_V3D_PARAM_MAX_PERF_COUNTERS,
296 DRM_V3D_PARAM_SUPPORTS_SUPER_PAGES,
297 DRM_V3D_PARAM_GLOBAL_RESET_COUNTER,
298 DRM_V3D_PARAM_CONTEXT_RESET_COUNTER,
299};
300
301struct drm_v3d_get_param {
302 __u32 param;
303 __u32 pad;
304 __u64 value;
305};
306
307/**
308 * Returns the offset for the BO in the V3D address space for this DRM fd.
309 * This is the same value returned by drm_v3d_create_bo, if that was called
310 * from this DRM fd.
311 */
312struct drm_v3d_get_bo_offset {
313 __u32 handle;
314 __u32 offset;
315};
316
317struct drm_v3d_submit_tfu {
318 __u32 icfg;
319 __u32 iia;
320 __u32 iis;
321 __u32 ica;
322 __u32 iua;
323 __u32 ioa;
324 __u32 ios;
325 __u32 coef[4];
326 /* First handle is the output BO, following are other inputs.
327 * 0 for unused.
328 */
329 __u32 bo_handles[4];
330 /* sync object to block on before running the TFU job. Each TFU
331 * job will execute in the order submitted to its FD. Synchronization
332 * against rendering jobs requires using sync objects.
333 */
334 __u32 in_sync;
335 /* Sync object to signal when the TFU job is done. */
336 __u32 out_sync;
337
338 __u32 flags;
339
340 /* Pointer to an array of ioctl extensions*/
341 __u64 extensions;
342
343 struct {
344 __u32 ioc;
345 __u32 pad;
346 } v71;
347};
348
349/* Submits a compute shader for dispatch. This job will block on any
350 * previous compute shaders submitted on this fd, and any other
351 * synchronization must be performed with in_sync/out_sync.
352 */
353struct drm_v3d_submit_csd {
354 __u32 cfg[7];
355 __u32 coef[4];
356
357 /* Pointer to a u32 array of the BOs that are referenced by the job.
358 */
359 __u64 bo_handles;
360
361 /* Number of BO handles passed in (size is that times 4). */
362 __u32 bo_handle_count;
363
364 /* sync object to block on before running the CSD job. Each
365 * CSD job will execute in the order submitted to its FD.
366 * Synchronization against rendering/TFU jobs or CSD from
367 * other fds requires using sync objects.
368 */
369 __u32 in_sync;
370 /* Sync object to signal when the CSD job is done. */
371 __u32 out_sync;
372
373 /* ID of the perfmon to attach to this job. 0 means no perfmon. */
374 __u32 perfmon_id;
375
376 /* Pointer to an array of ioctl extensions*/
377 __u64 extensions;
378
379 __u32 flags;
380
381 __u32 pad;
382};
383
384/**
385 * struct drm_v3d_indirect_csd - ioctl extension for the CPU job to create an
386 * indirect CSD
387 *
388 * When an extension of DRM_V3D_EXT_ID_CPU_INDIRECT_CSD id is defined, it
389 * points to this extension to define a indirect CSD submission. It creates a
390 * CPU job linked to a CSD job. The CPU job waits for the indirect CSD
391 * dependencies and, once they are signaled, it updates the CSD job config
392 * before allowing the CSD job execution.
393 */
394struct drm_v3d_indirect_csd {
395 struct drm_v3d_extension base;
396
397 /* Indirect CSD */
398 struct drm_v3d_submit_csd submit;
399
400 /* Handle of the indirect BO, that should be also attached to the
401 * indirect CSD.
402 */
403 __u32 indirect;
404
405 /* Offset within the BO where the workgroup counts are stored */
406 __u32 offset;
407
408 /* Workgroups size */
409 __u32 wg_size;
410
411 /* Indices of the uniforms with the workgroup dispatch counts
412 * in the uniform stream. If the uniform rewrite is not needed,
413 * the offset must be 0xffffffff.
414 */
415 __u32 wg_uniform_offsets[3];
416};
417
418/**
419 * struct drm_v3d_timestamp_query - ioctl extension for the CPU job to calculate
420 * a timestamp query
421 *
422 * When an extension DRM_V3D_EXT_ID_TIMESTAMP_QUERY is defined, it points to
423 * this extension to define a timestamp query submission. This CPU job will
424 * calculate the timestamp query and update the query value within the
425 * timestamp BO. Moreover, it will signal the timestamp syncobj to indicate
426 * query availability.
427 */
428struct drm_v3d_timestamp_query {
429 struct drm_v3d_extension base;
430
431 /* Array of queries' offsets within the timestamp BO for their value */
432 __u64 offsets;
433
434 /* Array of timestamp's syncobjs to indicate its availability */
435 __u64 syncs;
436
437 /* Number of queries */
438 __u32 count;
439
440 /* mbz */
441 __u32 pad;
442};
443
444/**
445 * struct drm_v3d_reset_timestamp_query - ioctl extension for the CPU job to
446 * reset timestamp queries
447 *
448 * When an extension DRM_V3D_EXT_ID_CPU_RESET_TIMESTAMP_QUERY is defined, it
449 * points to this extension to define a reset timestamp submission. This CPU
450 * job will reset the timestamp queries based on value offset of the first
451 * query. Moreover, it will reset the timestamp syncobj to reset query
452 * availability.
453 */
454struct drm_v3d_reset_timestamp_query {
455 struct drm_v3d_extension base;
456
457 /* Array of timestamp's syncobjs to indicate its availability */
458 __u64 syncs;
459
460 /* Offset of the first query within the timestamp BO for its value */
461 __u32 offset;
462
463 /* Number of queries */
464 __u32 count;
465};
466
467/**
468 * struct drm_v3d_copy_timestamp_query - ioctl extension for the CPU job to copy
469 * query results to a buffer
470 *
471 * When an extension DRM_V3D_EXT_ID_CPU_COPY_TIMESTAMP_QUERY is defined, it
472 * points to this extension to define a copy timestamp query submission. This
473 * CPU job will copy the timestamp queries results to a BO with the offset
474 * and stride defined in the extension.
475 */
476struct drm_v3d_copy_timestamp_query {
477 struct drm_v3d_extension base;
478
479 /* Define if should write to buffer using 64 or 32 bits */
480 __u8 do_64bit;
481
482 /* Define if it can write to buffer even if the query is not available */
483 __u8 do_partial;
484
485 /* Define if it should write availability bit to buffer */
486 __u8 availability_bit;
487
488 /* mbz */
489 __u8 pad;
490
491 /* Offset of the buffer in the BO */
492 __u32 offset;
493
494 /* Stride of the buffer in the BO */
495 __u32 stride;
496
497 /* Number of queries */
498 __u32 count;
499
500 /* Array of queries' offsets within the timestamp BO for their value */
501 __u64 offsets;
502
503 /* Array of timestamp's syncobjs to indicate its availability */
504 __u64 syncs;
505};
506
507/**
508 * struct drm_v3d_reset_performance_query - ioctl extension for the CPU job to
509 * reset performance queries
510 *
511 * When an extension DRM_V3D_EXT_ID_CPU_RESET_PERFORMANCE_QUERY is defined, it
512 * points to this extension to define a reset performance submission. This CPU
513 * job will reset the performance queries by resetting the values of the
514 * performance monitors. Moreover, it will reset the syncobj to reset query
515 * availability.
516 */
517struct drm_v3d_reset_performance_query {
518 struct drm_v3d_extension base;
519
520 /* Array of performance queries's syncobjs to indicate its availability */
521 __u64 syncs;
522
523 /* Number of queries */
524 __u32 count;
525
526 /* Number of performance monitors */
527 __u32 nperfmons;
528
529 /* Array of u64 user-pointers that point to an array of kperfmon_ids */
530 __u64 kperfmon_ids;
531};
532
533/**
534 * struct drm_v3d_copy_performance_query - ioctl extension for the CPU job to copy
535 * performance query results to a buffer
536 *
537 * When an extension DRM_V3D_EXT_ID_CPU_COPY_PERFORMANCE_QUERY is defined, it
538 * points to this extension to define a copy performance query submission. This
539 * CPU job will copy the performance queries results to a BO with the offset
540 * and stride defined in the extension.
541 */
542struct drm_v3d_copy_performance_query {
543 struct drm_v3d_extension base;
544
545 /* Define if should write to buffer using 64 or 32 bits */
546 __u8 do_64bit;
547
548 /* Define if it can write to buffer even if the query is not available */
549 __u8 do_partial;
550
551 /* Define if it should write availability bit to buffer */
552 __u8 availability_bit;
553
554 /* mbz */
555 __u8 pad;
556
557 /* Offset of the buffer in the BO */
558 __u32 offset;
559
560 /* Stride of the buffer in the BO */
561 __u32 stride;
562
563 /* Number of performance monitors */
564 __u32 nperfmons;
565
566 /* Number of performance counters related to this query pool */
567 __u32 ncounters;
568
569 /* Number of queries */
570 __u32 count;
571
572 /* Array of performance queries's syncobjs to indicate its availability */
573 __u64 syncs;
574
575 /* Array of u64 user-pointers that point to an array of kperfmon_ids */
576 __u64 kperfmon_ids;
577};
578
579struct drm_v3d_submit_cpu {
580 /* Pointer to a u32 array of the BOs that are referenced by the job.
581 *
582 * For DRM_V3D_EXT_ID_CPU_INDIRECT_CSD, it must contain only one BO,
583 * that contains the workgroup counts.
584 *
585 * For DRM_V3D_EXT_ID_TIMESTAMP_QUERY, it must contain only one BO,
586 * that will contain the timestamp.
587 *
588 * For DRM_V3D_EXT_ID_CPU_RESET_TIMESTAMP_QUERY, it must contain only
589 * one BO, that contains the timestamp.
590 *
591 * For DRM_V3D_EXT_ID_CPU_COPY_TIMESTAMP_QUERY, it must contain two
592 * BOs. The first is the BO where the timestamp queries will be written
593 * to. The second is the BO that contains the timestamp.
594 *
595 * For DRM_V3D_EXT_ID_CPU_RESET_PERFORMANCE_QUERY, it must contain no
596 * BOs.
597 *
598 * For DRM_V3D_EXT_ID_CPU_COPY_PERFORMANCE_QUERY, it must contain one
599 * BO, where the performance queries will be written.
600 */
601 __u64 bo_handles;
602
603 /* Number of BO handles passed in (size is that times 4). */
604 __u32 bo_handle_count;
605
606 __u32 flags;
607
608 /* Pointer to an array of ioctl extensions*/
609 __u64 extensions;
610};
611
612/* The performance counters index represented by this enum are deprecated and
613 * must no longer be used. These counters are only valid for V3D 4.2.
614 *
615 * In order to check for performance counter information,
616 * use DRM_IOCTL_V3D_PERFMON_GET_COUNTER.
617 *
618 * Don't use V3D_PERFCNT_NUM to retrieve the maximum number of performance
619 * counters. You should use DRM_IOCTL_V3D_GET_PARAM with the following
620 * parameter: DRM_V3D_PARAM_MAX_PERF_COUNTERS.
621 */
622enum {
623 V3D_PERFCNT_FEP_VALID_PRIMTS_NO_PIXELS,
624 V3D_PERFCNT_FEP_VALID_PRIMS,
625 V3D_PERFCNT_FEP_EZ_NFCLIP_QUADS,
626 V3D_PERFCNT_FEP_VALID_QUADS,
627 V3D_PERFCNT_TLB_QUADS_STENCIL_FAIL,
628 V3D_PERFCNT_TLB_QUADS_STENCILZ_FAIL,
629 V3D_PERFCNT_TLB_QUADS_STENCILZ_PASS,
630 V3D_PERFCNT_TLB_QUADS_ZERO_COV,
631 V3D_PERFCNT_TLB_QUADS_NONZERO_COV,
632 V3D_PERFCNT_TLB_QUADS_WRITTEN,
633 V3D_PERFCNT_PTB_PRIM_VIEWPOINT_DISCARD,
634 V3D_PERFCNT_PTB_PRIM_CLIP,
635 V3D_PERFCNT_PTB_PRIM_REV,
636 V3D_PERFCNT_QPU_IDLE_CYCLES,
637 V3D_PERFCNT_QPU_ACTIVE_CYCLES_VERTEX_COORD_USER,
638 V3D_PERFCNT_QPU_ACTIVE_CYCLES_FRAG,
639 V3D_PERFCNT_QPU_CYCLES_VALID_INSTR,
640 V3D_PERFCNT_QPU_CYCLES_TMU_STALL,
641 V3D_PERFCNT_QPU_CYCLES_SCOREBOARD_STALL,
642 V3D_PERFCNT_QPU_CYCLES_VARYINGS_STALL,
643 V3D_PERFCNT_QPU_IC_HIT,
644 V3D_PERFCNT_QPU_IC_MISS,
645 V3D_PERFCNT_QPU_UC_HIT,
646 V3D_PERFCNT_QPU_UC_MISS,
647 V3D_PERFCNT_TMU_TCACHE_ACCESS,
648 V3D_PERFCNT_TMU_TCACHE_MISS,
649 V3D_PERFCNT_VPM_VDW_STALL,
650 V3D_PERFCNT_VPM_VCD_STALL,
651 V3D_PERFCNT_BIN_ACTIVE,
652 V3D_PERFCNT_RDR_ACTIVE,
653 V3D_PERFCNT_L2T_HITS,
654 V3D_PERFCNT_L2T_MISSES,
655 V3D_PERFCNT_CYCLE_COUNT,
656 V3D_PERFCNT_QPU_CYCLES_STALLED_VERTEX_COORD_USER,
657 V3D_PERFCNT_QPU_CYCLES_STALLED_FRAGMENT,
658 V3D_PERFCNT_PTB_PRIMS_BINNED,
659 V3D_PERFCNT_AXI_WRITES_WATCH_0,
660 V3D_PERFCNT_AXI_READS_WATCH_0,
661 V3D_PERFCNT_AXI_WRITE_STALLS_WATCH_0,
662 V3D_PERFCNT_AXI_READ_STALLS_WATCH_0,
663 V3D_PERFCNT_AXI_WRITE_BYTES_WATCH_0,
664 V3D_PERFCNT_AXI_READ_BYTES_WATCH_0,
665 V3D_PERFCNT_AXI_WRITES_WATCH_1,
666 V3D_PERFCNT_AXI_READS_WATCH_1,
667 V3D_PERFCNT_AXI_WRITE_STALLS_WATCH_1,
668 V3D_PERFCNT_AXI_READ_STALLS_WATCH_1,
669 V3D_PERFCNT_AXI_WRITE_BYTES_WATCH_1,
670 V3D_PERFCNT_AXI_READ_BYTES_WATCH_1,
671 V3D_PERFCNT_TLB_PARTIAL_QUADS,
672 V3D_PERFCNT_TMU_CONFIG_ACCESSES,
673 V3D_PERFCNT_L2T_NO_ID_STALL,
674 V3D_PERFCNT_L2T_COM_QUE_STALL,
675 V3D_PERFCNT_L2T_TMU_WRITES,
676 V3D_PERFCNT_TMU_ACTIVE_CYCLES,
677 V3D_PERFCNT_TMU_STALLED_CYCLES,
678 V3D_PERFCNT_CLE_ACTIVE,
679 V3D_PERFCNT_L2T_TMU_READS,
680 V3D_PERFCNT_L2T_CLE_READS,
681 V3D_PERFCNT_L2T_VCD_READS,
682 V3D_PERFCNT_L2T_TMUCFG_READS,
683 V3D_PERFCNT_L2T_SLC0_READS,
684 V3D_PERFCNT_L2T_SLC1_READS,
685 V3D_PERFCNT_L2T_SLC2_READS,
686 V3D_PERFCNT_L2T_TMU_W_MISSES,
687 V3D_PERFCNT_L2T_TMU_R_MISSES,
688 V3D_PERFCNT_L2T_CLE_MISSES,
689 V3D_PERFCNT_L2T_VCD_MISSES,
690 V3D_PERFCNT_L2T_TMUCFG_MISSES,
691 V3D_PERFCNT_L2T_SLC0_MISSES,
692 V3D_PERFCNT_L2T_SLC1_MISSES,
693 V3D_PERFCNT_L2T_SLC2_MISSES,
694 V3D_PERFCNT_CORE_MEM_WRITES,
695 V3D_PERFCNT_L2T_MEM_WRITES,
696 V3D_PERFCNT_PTB_MEM_WRITES,
697 V3D_PERFCNT_TLB_MEM_WRITES,
698 V3D_PERFCNT_CORE_MEM_READS,
699 V3D_PERFCNT_L2T_MEM_READS,
700 V3D_PERFCNT_PTB_MEM_READS,
701 V3D_PERFCNT_PSE_MEM_READS,
702 V3D_PERFCNT_TLB_MEM_READS,
703 V3D_PERFCNT_GMP_MEM_READS,
704 V3D_PERFCNT_PTB_W_MEM_WORDS,
705 V3D_PERFCNT_TLB_W_MEM_WORDS,
706 V3D_PERFCNT_PSE_R_MEM_WORDS,
707 V3D_PERFCNT_TLB_R_MEM_WORDS,
708 V3D_PERFCNT_TMU_MRU_HITS,
709 V3D_PERFCNT_COMPUTE_ACTIVE,
710 V3D_PERFCNT_NUM,
711};
712
713#define DRM_V3D_MAX_PERF_COUNTERS 32
714
715struct drm_v3d_perfmon_create {
716 __u32 id;
717 __u32 ncounters;
718 __u8 counters[DRM_V3D_MAX_PERF_COUNTERS];
719};
720
721struct drm_v3d_perfmon_destroy {
722 __u32 id;
723};
724
725/*
726 * Returns the values of the performance counters tracked by this
727 * perfmon (as an array of ncounters u64 values).
728 *
729 * No implicit synchronization is performed, so the user has to
730 * guarantee that any jobs using this perfmon have already been
731 * completed (probably by blocking on the seqno returned by the
732 * last exec that used the perfmon).
733 */
734struct drm_v3d_perfmon_get_values {
735 __u32 id;
736 __u32 pad;
737 __u64 values_ptr;
738};
739
740#define DRM_V3D_PERFCNT_MAX_NAME 64
741#define DRM_V3D_PERFCNT_MAX_CATEGORY 32
742#define DRM_V3D_PERFCNT_MAX_DESCRIPTION 256
743
744/**
745 * struct drm_v3d_perfmon_get_counter - ioctl to get the description of a
746 * performance counter
747 *
748 * As userspace needs to retrieve information about the performance counters
749 * available, this IOCTL allows users to get information about a performance
750 * counter (name, category and description).
751 */
752struct drm_v3d_perfmon_get_counter {
753 /*
754 * Counter ID
755 *
756 * Must be smaller than the maximum number of performance counters, which
757 * can be retrieve through DRM_V3D_PARAM_MAX_PERF_COUNTERS.
758 */
759 __u8 counter;
760
761 /* Name of the counter */
762 __u8 name[DRM_V3D_PERFCNT_MAX_NAME];
763
764 /* Category of the counter */
765 __u8 category[DRM_V3D_PERFCNT_MAX_CATEGORY];
766
767 /* Description of the counter */
768 __u8 description[DRM_V3D_PERFCNT_MAX_DESCRIPTION];
769
770 /* mbz */
771 __u8 reserved[7];
772};
773
774#define DRM_V3D_PERFMON_CLEAR_GLOBAL 0x0001
775
776/**
777 * struct drm_v3d_perfmon_set_global - ioctl to define a global performance
778 * monitor
779 *
780 * The global performance monitor will be used for all jobs. If a global
781 * performance monitor is defined, jobs with a self-defined performance
782 * monitor won't be allowed.
783 */
784struct drm_v3d_perfmon_set_global {
785 __u32 flags;
786 __u32 id;
787};
788
789#if defined(__cplusplus)
790}
791#endif
792
793#endif /* _V3D_DRM_H_ */