1 // SPDX-License-Identifier: MIT
2 /*
3 * Copyright © 2019 Intel Corporation
4 */
5
6 #include "i915_selftest.h"
7
8 #include "gt/intel_context.h"
9 #include "gt/intel_engine_regs.h"
10 #include "gt/intel_engine_user.h"
11 #include "gt/intel_gpu_commands.h"
12 #include "gt/intel_gt.h"
13 #include "gt/intel_gt_regs.h"
14 #include "gem/i915_gem_lmem.h"
15
16 #include "gem/selftests/igt_gem_utils.h"
17 #include "selftests/igt_flush_test.h"
18 #include "selftests/mock_drm.h"
19 #include "selftests/i915_random.h"
20 #include "huge_gem_object.h"
21 #include "mock_context.h"
22
23 #define OW_SIZE 16 /* in bytes */
24 #define F_SUBTILE_SIZE 64 /* in bytes */
25 #define F_TILE_WIDTH 128 /* in bytes */
26 #define F_TILE_HEIGHT 32 /* in pixels */
27 #define F_SUBTILE_WIDTH OW_SIZE /* in bytes */
28 #define F_SUBTILE_HEIGHT 4 /* in pixels */
29
linear_x_y_to_ftiled_pos(int x,int y,u32 stride,int bpp)30 static int linear_x_y_to_ftiled_pos(int x, int y, u32 stride, int bpp)
31 {
32 int tile_base;
33 int tile_x, tile_y;
34 int swizzle, subtile;
35 int pixel_size = bpp / 8;
36 int pos;
37
38 /*
39 * Subtile remapping for F tile. Note that map[a]==b implies map[b]==a
40 * so we can use the same table to tile and until.
41 */
42 static const u8 f_subtile_map[] = {
43 0, 1, 2, 3, 8, 9, 10, 11,
44 4, 5, 6, 7, 12, 13, 14, 15,
45 16, 17, 18, 19, 24, 25, 26, 27,
46 20, 21, 22, 23, 28, 29, 30, 31,
47 32, 33, 34, 35, 40, 41, 42, 43,
48 36, 37, 38, 39, 44, 45, 46, 47,
49 48, 49, 50, 51, 56, 57, 58, 59,
50 52, 53, 54, 55, 60, 61, 62, 63
51 };
52
53 x *= pixel_size;
54 /*
55 * Where does the 4k tile start (in bytes)? This is the same for Y and
56 * F so we can use the Y-tile algorithm to get to that point.
57 */
58 tile_base =
59 y / F_TILE_HEIGHT * stride * F_TILE_HEIGHT +
60 x / F_TILE_WIDTH * 4096;
61
62 /* Find pixel within tile */
63 tile_x = x % F_TILE_WIDTH;
64 tile_y = y % F_TILE_HEIGHT;
65
66 /* And figure out the subtile within the 4k tile */
67 subtile = tile_y / F_SUBTILE_HEIGHT * 8 + tile_x / F_SUBTILE_WIDTH;
68
69 /* Swizzle the subtile number according to the bspec diagram */
70 swizzle = f_subtile_map[subtile];
71
72 /* Calculate new position */
73 pos = tile_base +
74 swizzle * F_SUBTILE_SIZE +
75 tile_y % F_SUBTILE_HEIGHT * OW_SIZE +
76 tile_x % F_SUBTILE_WIDTH;
77
78 GEM_BUG_ON(!IS_ALIGNED(pos, pixel_size));
79
80 return pos / pixel_size * 4;
81 }
82
83 enum client_tiling {
84 CLIENT_TILING_LINEAR,
85 CLIENT_TILING_X,
86 CLIENT_TILING_Y,
87 CLIENT_TILING_4,
88 CLIENT_NUM_TILING_TYPES
89 };
90
91 #define WIDTH 512
92 #define HEIGHT 32
93
94 struct blit_buffer {
95 struct i915_vma *vma;
96 u32 start_val;
97 enum client_tiling tiling;
98 };
99
100 struct tiled_blits {
101 struct intel_context *ce;
102 struct blit_buffer buffers[3];
103 struct blit_buffer scratch;
104 struct i915_vma *batch;
105 u64 hole;
106 u64 align;
107 u32 width;
108 u32 height;
109 };
110
supports_x_tiling(const struct drm_i915_private * i915)111 static bool supports_x_tiling(const struct drm_i915_private *i915)
112 {
113 int gen = GRAPHICS_VER(i915);
114
115 if (gen < 12)
116 return true;
117
118 if (!HAS_LMEM(i915) || IS_DG1(i915))
119 return false;
120
121 return true;
122 }
123
fast_blit_ok(const struct blit_buffer * buf)124 static bool fast_blit_ok(const struct blit_buffer *buf)
125 {
126 int gen = GRAPHICS_VER(buf->vma->vm->i915);
127
128 if (gen < 9)
129 return false;
130
131 if (gen < 12)
132 return true;
133
134 /* filter out platforms with unsupported X-tile support in fastblit */
135 if (buf->tiling == CLIENT_TILING_X && !supports_x_tiling(buf->vma->vm->i915))
136 return false;
137
138 return true;
139 }
140
prepare_blit(const struct tiled_blits * t,struct blit_buffer * dst,struct blit_buffer * src,struct drm_i915_gem_object * batch)141 static int prepare_blit(const struct tiled_blits *t,
142 struct blit_buffer *dst,
143 struct blit_buffer *src,
144 struct drm_i915_gem_object *batch)
145 {
146 const int ver = GRAPHICS_VER(to_i915(batch->base.dev));
147 bool use_64b_reloc = ver >= 8;
148 u32 src_pitch, dst_pitch;
149 u32 cmd, *cs;
150
151 cs = i915_gem_object_pin_map_unlocked(batch, I915_MAP_WC);
152 if (IS_ERR(cs))
153 return PTR_ERR(cs);
154
155 if (fast_blit_ok(dst) && fast_blit_ok(src)) {
156 struct intel_gt *gt = t->ce->engine->gt;
157 u32 src_tiles = 0, dst_tiles = 0;
158 u32 src_4t = 0, dst_4t = 0;
159
160 /* Need to program BLIT_CCTL if it is not done previously
161 * before using XY_FAST_COPY_BLT
162 */
163 *cs++ = MI_LOAD_REGISTER_IMM(1);
164 *cs++ = i915_mmio_reg_offset(BLIT_CCTL(t->ce->engine->mmio_base));
165 *cs++ = (BLIT_CCTL_SRC_MOCS(gt->mocs.uc_index) |
166 BLIT_CCTL_DST_MOCS(gt->mocs.uc_index));
167
168 src_pitch = t->width; /* in dwords */
169 if (src->tiling == CLIENT_TILING_4) {
170 src_tiles = XY_FAST_COPY_BLT_D0_SRC_TILE_MODE(YMAJOR);
171 src_4t = XY_FAST_COPY_BLT_D1_SRC_TILE4;
172 } else if (src->tiling == CLIENT_TILING_Y) {
173 src_tiles = XY_FAST_COPY_BLT_D0_SRC_TILE_MODE(YMAJOR);
174 } else if (src->tiling == CLIENT_TILING_X) {
175 src_tiles = XY_FAST_COPY_BLT_D0_SRC_TILE_MODE(TILE_X);
176 } else {
177 src_pitch *= 4; /* in bytes */
178 }
179
180 dst_pitch = t->width; /* in dwords */
181 if (dst->tiling == CLIENT_TILING_4) {
182 dst_tiles = XY_FAST_COPY_BLT_D0_DST_TILE_MODE(YMAJOR);
183 dst_4t = XY_FAST_COPY_BLT_D1_DST_TILE4;
184 } else if (dst->tiling == CLIENT_TILING_Y) {
185 dst_tiles = XY_FAST_COPY_BLT_D0_DST_TILE_MODE(YMAJOR);
186 } else if (dst->tiling == CLIENT_TILING_X) {
187 dst_tiles = XY_FAST_COPY_BLT_D0_DST_TILE_MODE(TILE_X);
188 } else {
189 dst_pitch *= 4; /* in bytes */
190 }
191
192 *cs++ = GEN9_XY_FAST_COPY_BLT_CMD | (10 - 2) |
193 src_tiles | dst_tiles;
194 *cs++ = src_4t | dst_4t | BLT_DEPTH_32 | dst_pitch;
195 *cs++ = 0;
196 *cs++ = t->height << 16 | t->width;
197 *cs++ = lower_32_bits(i915_vma_offset(dst->vma));
198 *cs++ = upper_32_bits(i915_vma_offset(dst->vma));
199 *cs++ = 0;
200 *cs++ = src_pitch;
201 *cs++ = lower_32_bits(i915_vma_offset(src->vma));
202 *cs++ = upper_32_bits(i915_vma_offset(src->vma));
203 } else {
204 if (ver >= 6) {
205 *cs++ = MI_LOAD_REGISTER_IMM(1);
206 *cs++ = i915_mmio_reg_offset(BCS_SWCTRL);
207 cmd = (BCS_SRC_Y | BCS_DST_Y) << 16;
208 if (src->tiling == CLIENT_TILING_Y)
209 cmd |= BCS_SRC_Y;
210 if (dst->tiling == CLIENT_TILING_Y)
211 cmd |= BCS_DST_Y;
212 *cs++ = cmd;
213
214 cmd = MI_FLUSH_DW;
215 if (ver >= 8)
216 cmd++;
217 *cs++ = cmd;
218 *cs++ = 0;
219 *cs++ = 0;
220 *cs++ = 0;
221 }
222
223 cmd = XY_SRC_COPY_BLT_CMD | BLT_WRITE_RGBA | (8 - 2);
224 if (ver >= 8)
225 cmd += 2;
226
227 src_pitch = t->width * 4;
228 if (src->tiling) {
229 cmd |= XY_SRC_COPY_BLT_SRC_TILED;
230 src_pitch /= 4;
231 }
232
233 dst_pitch = t->width * 4;
234 if (dst->tiling) {
235 cmd |= XY_SRC_COPY_BLT_DST_TILED;
236 dst_pitch /= 4;
237 }
238
239 *cs++ = cmd;
240 *cs++ = BLT_DEPTH_32 | BLT_ROP_SRC_COPY | dst_pitch;
241 *cs++ = 0;
242 *cs++ = t->height << 16 | t->width;
243 *cs++ = lower_32_bits(i915_vma_offset(dst->vma));
244 if (use_64b_reloc)
245 *cs++ = upper_32_bits(i915_vma_offset(dst->vma));
246 *cs++ = 0;
247 *cs++ = src_pitch;
248 *cs++ = lower_32_bits(i915_vma_offset(src->vma));
249 if (use_64b_reloc)
250 *cs++ = upper_32_bits(i915_vma_offset(src->vma));
251 }
252
253 *cs++ = MI_BATCH_BUFFER_END;
254
255 i915_gem_object_flush_map(batch);
256 i915_gem_object_unpin_map(batch);
257
258 return 0;
259 }
260
tiled_blits_destroy_buffers(struct tiled_blits * t)261 static void tiled_blits_destroy_buffers(struct tiled_blits *t)
262 {
263 int i;
264
265 for (i = 0; i < ARRAY_SIZE(t->buffers); i++)
266 i915_vma_put(t->buffers[i].vma);
267
268 i915_vma_put(t->scratch.vma);
269 i915_vma_put(t->batch);
270 }
271
272 static struct i915_vma *
__create_vma(struct tiled_blits * t,size_t size,bool lmem)273 __create_vma(struct tiled_blits *t, size_t size, bool lmem)
274 {
275 struct drm_i915_private *i915 = t->ce->vm->i915;
276 struct drm_i915_gem_object *obj;
277 struct i915_vma *vma;
278
279 if (lmem)
280 obj = i915_gem_object_create_lmem(i915, size, 0);
281 else
282 obj = i915_gem_object_create_shmem(i915, size);
283 if (IS_ERR(obj))
284 return ERR_CAST(obj);
285
286 vma = i915_vma_instance(obj, t->ce->vm, NULL);
287 if (IS_ERR(vma))
288 i915_gem_object_put(obj);
289
290 return vma;
291 }
292
create_vma(struct tiled_blits * t,bool lmem)293 static struct i915_vma *create_vma(struct tiled_blits *t, bool lmem)
294 {
295 return __create_vma(t, PAGE_ALIGN(t->width * t->height * 4), lmem);
296 }
297
tiled_blits_create_buffers(struct tiled_blits * t,int width,int height,struct rnd_state * prng)298 static int tiled_blits_create_buffers(struct tiled_blits *t,
299 int width, int height,
300 struct rnd_state *prng)
301 {
302 struct drm_i915_private *i915 = t->ce->engine->i915;
303 int i;
304
305 t->width = width;
306 t->height = height;
307
308 t->batch = __create_vma(t, PAGE_SIZE, false);
309 if (IS_ERR(t->batch))
310 return PTR_ERR(t->batch);
311
312 t->scratch.vma = create_vma(t, false);
313 if (IS_ERR(t->scratch.vma)) {
314 i915_vma_put(t->batch);
315 return PTR_ERR(t->scratch.vma);
316 }
317
318 for (i = 0; i < ARRAY_SIZE(t->buffers); i++) {
319 struct i915_vma *vma;
320
321 vma = create_vma(t, HAS_LMEM(i915) && i % 2);
322 if (IS_ERR(vma)) {
323 tiled_blits_destroy_buffers(t);
324 return PTR_ERR(vma);
325 }
326
327 t->buffers[i].vma = vma;
328 t->buffers[i].tiling =
329 i915_prandom_u32_max_state(CLIENT_NUM_TILING_TYPES, prng);
330
331 /* Platforms support either TileY or Tile4, not both */
332 if (HAS_4TILE(i915) && t->buffers[i].tiling == CLIENT_TILING_Y)
333 t->buffers[i].tiling = CLIENT_TILING_4;
334 else if (!HAS_4TILE(i915) && t->buffers[i].tiling == CLIENT_TILING_4)
335 t->buffers[i].tiling = CLIENT_TILING_Y;
336 }
337
338 return 0;
339 }
340
fill_scratch(struct tiled_blits * t,u32 * vaddr,u32 val)341 static void fill_scratch(struct tiled_blits *t, u32 *vaddr, u32 val)
342 {
343 int i;
344
345 t->scratch.start_val = val;
346 for (i = 0; i < t->width * t->height; i++)
347 vaddr[i] = val++;
348
349 i915_gem_object_flush_map(t->scratch.vma->obj);
350 }
351
swizzle_bit(unsigned int bit,u64 offset)352 static u64 swizzle_bit(unsigned int bit, u64 offset)
353 {
354 return (offset & BIT_ULL(bit)) >> (bit - 6);
355 }
356
tiled_offset(const struct intel_gt * gt,u64 v,unsigned int stride,enum client_tiling tiling,int x_pos,int y_pos)357 static u64 tiled_offset(const struct intel_gt *gt,
358 u64 v,
359 unsigned int stride,
360 enum client_tiling tiling,
361 int x_pos, int y_pos)
362 {
363 unsigned int swizzle;
364 u64 x, y;
365
366 if (tiling == CLIENT_TILING_LINEAR)
367 return v;
368
369 y = div64_u64_rem(v, stride, &x);
370
371 if (tiling == CLIENT_TILING_4) {
372 v = linear_x_y_to_ftiled_pos(x_pos, y_pos, stride, 32);
373
374 /* no swizzling for f-tiling */
375 swizzle = I915_BIT_6_SWIZZLE_NONE;
376 } else if (tiling == CLIENT_TILING_X) {
377 v = div64_u64_rem(y, 8, &y) * stride * 8;
378 v += y * 512;
379 v += div64_u64_rem(x, 512, &x) << 12;
380 v += x;
381
382 swizzle = gt->ggtt->bit_6_swizzle_x;
383 } else {
384 const unsigned int ytile_span = 16;
385 const unsigned int ytile_height = 512;
386
387 v = div64_u64_rem(y, 32, &y) * stride * 32;
388 v += y * ytile_span;
389 v += div64_u64_rem(x, ytile_span, &x) * ytile_height;
390 v += x;
391
392 swizzle = gt->ggtt->bit_6_swizzle_y;
393 }
394
395 switch (swizzle) {
396 case I915_BIT_6_SWIZZLE_9:
397 v ^= swizzle_bit(9, v);
398 break;
399 case I915_BIT_6_SWIZZLE_9_10:
400 v ^= swizzle_bit(9, v) ^ swizzle_bit(10, v);
401 break;
402 case I915_BIT_6_SWIZZLE_9_11:
403 v ^= swizzle_bit(9, v) ^ swizzle_bit(11, v);
404 break;
405 case I915_BIT_6_SWIZZLE_9_10_11:
406 v ^= swizzle_bit(9, v) ^ swizzle_bit(10, v) ^ swizzle_bit(11, v);
407 break;
408 }
409
410 return v;
411 }
412
repr_tiling(enum client_tiling tiling)413 static const char *repr_tiling(enum client_tiling tiling)
414 {
415 switch (tiling) {
416 case CLIENT_TILING_LINEAR: return "linear";
417 case CLIENT_TILING_X: return "X";
418 case CLIENT_TILING_Y: return "Y";
419 case CLIENT_TILING_4: return "F";
420 default: return "unknown";
421 }
422 }
423
verify_buffer(const struct tiled_blits * t,struct blit_buffer * buf,struct rnd_state * prng)424 static int verify_buffer(const struct tiled_blits *t,
425 struct blit_buffer *buf,
426 struct rnd_state *prng)
427 {
428 const u32 *vaddr;
429 int ret = 0;
430 int x, y, p;
431
432 x = i915_prandom_u32_max_state(t->width, prng);
433 y = i915_prandom_u32_max_state(t->height, prng);
434 p = y * t->width + x;
435
436 vaddr = i915_gem_object_pin_map_unlocked(buf->vma->obj, I915_MAP_WC);
437 if (IS_ERR(vaddr))
438 return PTR_ERR(vaddr);
439
440 if (vaddr[0] != buf->start_val) {
441 ret = -EINVAL;
442 } else {
443 u64 v = tiled_offset(buf->vma->vm->gt,
444 p * 4, t->width * 4,
445 buf->tiling, x, y);
446
447 if (vaddr[v / sizeof(*vaddr)] != buf->start_val + p)
448 ret = -EINVAL;
449 }
450 if (ret) {
451 pr_err("Invalid %s tiling detected at (%d, %d), start_val %x\n",
452 repr_tiling(buf->tiling),
453 x, y, buf->start_val);
454 igt_hexdump(vaddr, 4096);
455 }
456
457 i915_gem_object_unpin_map(buf->vma->obj);
458 return ret;
459 }
460
pin_buffer(struct i915_vma * vma,u64 addr)461 static int pin_buffer(struct i915_vma *vma, u64 addr)
462 {
463 int err;
464
465 if (drm_mm_node_allocated(&vma->node) && i915_vma_offset(vma) != addr) {
466 err = i915_vma_unbind_unlocked(vma);
467 if (err)
468 return err;
469 }
470
471 err = i915_vma_pin(vma, 0, 0, PIN_USER | PIN_OFFSET_FIXED | addr);
472 if (err)
473 return err;
474
475 GEM_BUG_ON(i915_vma_offset(vma) != addr);
476 return 0;
477 }
478
479 static int
tiled_blit(struct tiled_blits * t,struct blit_buffer * dst,u64 dst_addr,struct blit_buffer * src,u64 src_addr)480 tiled_blit(struct tiled_blits *t,
481 struct blit_buffer *dst, u64 dst_addr,
482 struct blit_buffer *src, u64 src_addr)
483 {
484 struct i915_request *rq;
485 int err;
486
487 err = pin_buffer(src->vma, src_addr);
488 if (err) {
489 pr_err("Cannot pin src @ %llx\n", src_addr);
490 return err;
491 }
492
493 err = pin_buffer(dst->vma, dst_addr);
494 if (err) {
495 pr_err("Cannot pin dst @ %llx\n", dst_addr);
496 goto err_src;
497 }
498
499 err = i915_vma_pin(t->batch, 0, 0, PIN_USER | PIN_HIGH);
500 if (err) {
501 pr_err("cannot pin batch\n");
502 goto err_dst;
503 }
504
505 err = prepare_blit(t, dst, src, t->batch->obj);
506 if (err)
507 goto err_bb;
508
509 rq = intel_context_create_request(t->ce);
510 if (IS_ERR(rq)) {
511 err = PTR_ERR(rq);
512 goto err_bb;
513 }
514
515 err = igt_vma_move_to_active_unlocked(t->batch, rq, 0);
516 if (!err)
517 err = igt_vma_move_to_active_unlocked(src->vma, rq, 0);
518 if (!err)
519 err = igt_vma_move_to_active_unlocked(dst->vma, rq, 0);
520 if (!err)
521 err = rq->engine->emit_bb_start(rq,
522 i915_vma_offset(t->batch),
523 i915_vma_size(t->batch),
524 0);
525 i915_request_get(rq);
526 i915_request_add(rq);
527 if (i915_request_wait(rq, 0, HZ / 2) < 0)
528 err = -ETIME;
529 i915_request_put(rq);
530
531 dst->start_val = src->start_val;
532 err_bb:
533 i915_vma_unpin(t->batch);
534 err_dst:
535 i915_vma_unpin(dst->vma);
536 err_src:
537 i915_vma_unpin(src->vma);
538 return err;
539 }
540
541 static struct tiled_blits *
tiled_blits_create(struct intel_engine_cs * engine,struct rnd_state * prng)542 tiled_blits_create(struct intel_engine_cs *engine, struct rnd_state *prng)
543 {
544 struct drm_mm_node hole;
545 struct tiled_blits *t;
546 u64 hole_size;
547 int err;
548
549 t = kzalloc(sizeof(*t), GFP_KERNEL);
550 if (!t)
551 return ERR_PTR(-ENOMEM);
552
553 t->ce = intel_context_create(engine);
554 if (IS_ERR(t->ce)) {
555 err = PTR_ERR(t->ce);
556 goto err_free;
557 }
558
559 t->align = i915_vm_min_alignment(t->ce->vm, INTEL_MEMORY_LOCAL);
560 t->align = max(t->align,
561 i915_vm_min_alignment(t->ce->vm, INTEL_MEMORY_SYSTEM));
562
563 hole_size = 2 * round_up(WIDTH * HEIGHT * 4, t->align);
564 hole_size *= 2; /* room to maneuver */
565 hole_size += 2 * t->align; /* padding on either side */
566
567 mutex_lock(&t->ce->vm->mutex);
568 memset(&hole, 0, sizeof(hole));
569 err = drm_mm_insert_node_in_range(&t->ce->vm->mm, &hole,
570 hole_size, t->align,
571 I915_COLOR_UNEVICTABLE,
572 0, U64_MAX,
573 DRM_MM_INSERT_BEST);
574 if (!err)
575 drm_mm_remove_node(&hole);
576 mutex_unlock(&t->ce->vm->mutex);
577 if (err) {
578 err = -ENODEV;
579 goto err_put;
580 }
581
582 t->hole = hole.start + t->align;
583 pr_info("Using hole at %llx\n", t->hole);
584
585 err = tiled_blits_create_buffers(t, WIDTH, HEIGHT, prng);
586 if (err)
587 goto err_put;
588
589 return t;
590
591 err_put:
592 intel_context_put(t->ce);
593 err_free:
594 kfree(t);
595 return ERR_PTR(err);
596 }
597
tiled_blits_destroy(struct tiled_blits * t)598 static void tiled_blits_destroy(struct tiled_blits *t)
599 {
600 tiled_blits_destroy_buffers(t);
601
602 intel_context_put(t->ce);
603 kfree(t);
604 }
605
tiled_blits_prepare(struct tiled_blits * t,struct rnd_state * prng)606 static int tiled_blits_prepare(struct tiled_blits *t,
607 struct rnd_state *prng)
608 {
609 u64 offset = round_up(t->width * t->height * 4, t->align);
610 u32 *map;
611 int err;
612 int i;
613
614 map = i915_gem_object_pin_map_unlocked(t->scratch.vma->obj, I915_MAP_WC);
615 if (IS_ERR(map))
616 return PTR_ERR(map);
617
618 /* Use scratch to fill objects */
619 for (i = 0; i < ARRAY_SIZE(t->buffers); i++) {
620 fill_scratch(t, map, prandom_u32_state(prng));
621 GEM_BUG_ON(verify_buffer(t, &t->scratch, prng));
622
623 err = tiled_blit(t,
624 &t->buffers[i], t->hole + offset,
625 &t->scratch, t->hole);
626 if (err == 0)
627 err = verify_buffer(t, &t->buffers[i], prng);
628 if (err) {
629 pr_err("Failed to create buffer %d\n", i);
630 break;
631 }
632 }
633
634 i915_gem_object_unpin_map(t->scratch.vma->obj);
635 return err;
636 }
637
tiled_blits_bounce(struct tiled_blits * t,struct rnd_state * prng)638 static int tiled_blits_bounce(struct tiled_blits *t, struct rnd_state *prng)
639 {
640 u64 offset = round_up(t->width * t->height * 4, 2 * t->align);
641 int err;
642
643 /* We want to check position invariant tiling across GTT eviction */
644
645 err = tiled_blit(t,
646 &t->buffers[1], t->hole + offset / 2,
647 &t->buffers[0], t->hole + 2 * offset);
648 if (err)
649 return err;
650
651 /* Simulating GTT eviction of the same buffer / layout */
652 t->buffers[2].tiling = t->buffers[0].tiling;
653
654 /* Reposition so that we overlap the old addresses, and slightly off */
655 err = tiled_blit(t,
656 &t->buffers[2], t->hole + t->align,
657 &t->buffers[1], t->hole + 3 * offset / 2);
658 if (err)
659 return err;
660
661 err = verify_buffer(t, &t->buffers[2], prng);
662 if (err)
663 return err;
664
665 return 0;
666 }
667
__igt_client_tiled_blits(struct intel_engine_cs * engine,struct rnd_state * prng)668 static int __igt_client_tiled_blits(struct intel_engine_cs *engine,
669 struct rnd_state *prng)
670 {
671 struct tiled_blits *t;
672 int err;
673
674 t = tiled_blits_create(engine, prng);
675 if (IS_ERR(t))
676 return PTR_ERR(t);
677
678 err = tiled_blits_prepare(t, prng);
679 if (err)
680 goto out;
681
682 err = tiled_blits_bounce(t, prng);
683 if (err)
684 goto out;
685
686 out:
687 tiled_blits_destroy(t);
688 return err;
689 }
690
has_bit17_swizzle(int sw)691 static bool has_bit17_swizzle(int sw)
692 {
693 return (sw == I915_BIT_6_SWIZZLE_9_10_17 ||
694 sw == I915_BIT_6_SWIZZLE_9_17);
695 }
696
bad_swizzling(struct drm_i915_private * i915)697 static bool bad_swizzling(struct drm_i915_private *i915)
698 {
699 struct i915_ggtt *ggtt = to_gt(i915)->ggtt;
700
701 if (i915->gem_quirks & GEM_QUIRK_PIN_SWIZZLED_PAGES)
702 return true;
703
704 if (has_bit17_swizzle(ggtt->bit_6_swizzle_x) ||
705 has_bit17_swizzle(ggtt->bit_6_swizzle_y))
706 return true;
707
708 return false;
709 }
710
igt_client_tiled_blits(void * arg)711 static int igt_client_tiled_blits(void *arg)
712 {
713 struct drm_i915_private *i915 = arg;
714 I915_RND_STATE(prng);
715 int inst = 0;
716
717 /* Test requires explicit BLT tiling controls */
718 if (GRAPHICS_VER(i915) < 4)
719 return 0;
720
721 if (bad_swizzling(i915)) /* Requires sane (sub-page) swizzling */
722 return 0;
723
724 do {
725 struct intel_engine_cs *engine;
726 int err;
727
728 engine = intel_engine_lookup_user(i915,
729 I915_ENGINE_CLASS_COPY,
730 inst++);
731 if (!engine)
732 return 0;
733
734 err = __igt_client_tiled_blits(engine, &prng);
735 if (err == -ENODEV)
736 err = 0;
737 if (err)
738 return err;
739 } while (1);
740 }
741
i915_gem_client_blt_live_selftests(struct drm_i915_private * i915)742 int i915_gem_client_blt_live_selftests(struct drm_i915_private *i915)
743 {
744 static const struct i915_subtest tests[] = {
745 SUBTEST(igt_client_tiled_blits),
746 };
747
748 if (intel_gt_is_wedged(to_gt(i915)))
749 return 0;
750
751 return i915_live_subtests(tests, i915);
752 }
753