Loading...
Note: File does not exist in v3.1.
1// SPDX-License-Identifier: GPL-2.0 OR MIT
2/*
3 * Copyright 2014-2022 Advanced Micro Devices, Inc.
4 *
5 * Permission is hereby granted, free of charge, to any person obtaining a
6 * copy of this software and associated documentation files (the "Software"),
7 * to deal in the Software without restriction, including without limitation
8 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
9 * and/or sell copies of the Software, and to permit persons to whom the
10 * Software is furnished to do so, subject to the following conditions:
11 *
12 * The above copyright notice and this permission notice shall be included in
13 * all copies or substantial portions of the 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 COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
19 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
20 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
21 * OTHER DEALINGS IN THE SOFTWARE.
22 *
23 */
24
25#include <linux/printk.h>
26#include <linux/slab.h>
27#include <linux/mm_types.h>
28
29#include "kfd_priv.h"
30#include "kfd_mqd_manager.h"
31#include "vi_structs.h"
32#include "gca/gfx_8_0_sh_mask.h"
33#include "gca/gfx_8_0_enum.h"
34#include "oss/oss_3_0_sh_mask.h"
35
36#define CP_MQD_CONTROL__PRIV_STATE__SHIFT 0x8
37
38static inline struct vi_mqd *get_mqd(void *mqd)
39{
40 return (struct vi_mqd *)mqd;
41}
42
43static inline struct vi_sdma_mqd *get_sdma_mqd(void *mqd)
44{
45 return (struct vi_sdma_mqd *)mqd;
46}
47
48static void update_cu_mask(struct mqd_manager *mm, void *mqd,
49 struct mqd_update_info *minfo)
50{
51 struct vi_mqd *m;
52 uint32_t se_mask[4] = {0}; /* 4 is the max # of SEs */
53
54 if (!minfo || (minfo->update_flag != UPDATE_FLAG_CU_MASK) ||
55 !minfo->cu_mask.ptr)
56 return;
57
58 mqd_symmetrically_map_cu_mask(mm,
59 minfo->cu_mask.ptr, minfo->cu_mask.count, se_mask);
60
61 m = get_mqd(mqd);
62 m->compute_static_thread_mgmt_se0 = se_mask[0];
63 m->compute_static_thread_mgmt_se1 = se_mask[1];
64 m->compute_static_thread_mgmt_se2 = se_mask[2];
65 m->compute_static_thread_mgmt_se3 = se_mask[3];
66
67 pr_debug("Update cu mask to %#x %#x %#x %#x\n",
68 m->compute_static_thread_mgmt_se0,
69 m->compute_static_thread_mgmt_se1,
70 m->compute_static_thread_mgmt_se2,
71 m->compute_static_thread_mgmt_se3);
72}
73
74static void set_priority(struct vi_mqd *m, struct queue_properties *q)
75{
76 m->cp_hqd_pipe_priority = pipe_priority_map[q->priority];
77 m->cp_hqd_queue_priority = q->priority;
78}
79
80static struct kfd_mem_obj *allocate_mqd(struct kfd_dev *kfd,
81 struct queue_properties *q)
82{
83 struct kfd_mem_obj *mqd_mem_obj;
84
85 if (kfd_gtt_sa_allocate(kfd, sizeof(struct vi_mqd),
86 &mqd_mem_obj))
87 return NULL;
88
89 return mqd_mem_obj;
90}
91
92static void init_mqd(struct mqd_manager *mm, void **mqd,
93 struct kfd_mem_obj *mqd_mem_obj, uint64_t *gart_addr,
94 struct queue_properties *q)
95{
96 uint64_t addr;
97 struct vi_mqd *m;
98
99 m = (struct vi_mqd *) mqd_mem_obj->cpu_ptr;
100 addr = mqd_mem_obj->gpu_addr;
101
102 memset(m, 0, sizeof(struct vi_mqd));
103
104 m->header = 0xC0310800;
105 m->compute_pipelinestat_enable = 1;
106 m->compute_static_thread_mgmt_se0 = 0xFFFFFFFF;
107 m->compute_static_thread_mgmt_se1 = 0xFFFFFFFF;
108 m->compute_static_thread_mgmt_se2 = 0xFFFFFFFF;
109 m->compute_static_thread_mgmt_se3 = 0xFFFFFFFF;
110
111 m->cp_hqd_persistent_state = CP_HQD_PERSISTENT_STATE__PRELOAD_REQ_MASK |
112 0x53 << CP_HQD_PERSISTENT_STATE__PRELOAD_SIZE__SHIFT;
113
114 m->cp_mqd_control = 1 << CP_MQD_CONTROL__PRIV_STATE__SHIFT |
115 MTYPE_UC << CP_MQD_CONTROL__MTYPE__SHIFT;
116
117 m->cp_mqd_base_addr_lo = lower_32_bits(addr);
118 m->cp_mqd_base_addr_hi = upper_32_bits(addr);
119
120 m->cp_hqd_quantum = 1 << CP_HQD_QUANTUM__QUANTUM_EN__SHIFT |
121 1 << CP_HQD_QUANTUM__QUANTUM_SCALE__SHIFT |
122 1 << CP_HQD_QUANTUM__QUANTUM_DURATION__SHIFT;
123
124 set_priority(m, q);
125 m->cp_hqd_eop_rptr = 1 << CP_HQD_EOP_RPTR__INIT_FETCHER__SHIFT;
126
127 if (q->format == KFD_QUEUE_FORMAT_AQL)
128 m->cp_hqd_iq_rptr = 1;
129
130 if (q->tba_addr) {
131 m->compute_tba_lo = lower_32_bits(q->tba_addr >> 8);
132 m->compute_tba_hi = upper_32_bits(q->tba_addr >> 8);
133 m->compute_tma_lo = lower_32_bits(q->tma_addr >> 8);
134 m->compute_tma_hi = upper_32_bits(q->tma_addr >> 8);
135 m->compute_pgm_rsrc2 |=
136 (1 << COMPUTE_PGM_RSRC2__TRAP_PRESENT__SHIFT);
137 }
138
139 if (mm->dev->cwsr_enabled && q->ctx_save_restore_area_address) {
140 m->cp_hqd_persistent_state |=
141 (1 << CP_HQD_PERSISTENT_STATE__QSWITCH_MODE__SHIFT);
142 m->cp_hqd_ctx_save_base_addr_lo =
143 lower_32_bits(q->ctx_save_restore_area_address);
144 m->cp_hqd_ctx_save_base_addr_hi =
145 upper_32_bits(q->ctx_save_restore_area_address);
146 m->cp_hqd_ctx_save_size = q->ctx_save_restore_area_size;
147 m->cp_hqd_cntl_stack_size = q->ctl_stack_size;
148 m->cp_hqd_cntl_stack_offset = q->ctl_stack_size;
149 m->cp_hqd_wg_state_offset = q->ctl_stack_size;
150 }
151
152 *mqd = m;
153 if (gart_addr)
154 *gart_addr = addr;
155 mm->update_mqd(mm, m, q, NULL);
156}
157
158static int load_mqd(struct mqd_manager *mm, void *mqd,
159 uint32_t pipe_id, uint32_t queue_id,
160 struct queue_properties *p, struct mm_struct *mms)
161{
162 /* AQL write pointer counts in 64B packets, PM4/CP counts in dwords. */
163 uint32_t wptr_shift = (p->format == KFD_QUEUE_FORMAT_AQL ? 4 : 0);
164 uint32_t wptr_mask = (uint32_t)((p->queue_size / 4) - 1);
165
166 return mm->dev->kfd2kgd->hqd_load(mm->dev->adev, mqd, pipe_id, queue_id,
167 (uint32_t __user *)p->write_ptr,
168 wptr_shift, wptr_mask, mms);
169}
170
171static void __update_mqd(struct mqd_manager *mm, void *mqd,
172 struct queue_properties *q, struct mqd_update_info *minfo,
173 unsigned int mtype, unsigned int atc_bit)
174{
175 struct vi_mqd *m;
176
177 m = get_mqd(mqd);
178
179 m->cp_hqd_pq_control = 5 << CP_HQD_PQ_CONTROL__RPTR_BLOCK_SIZE__SHIFT |
180 atc_bit << CP_HQD_PQ_CONTROL__PQ_ATC__SHIFT |
181 mtype << CP_HQD_PQ_CONTROL__MTYPE__SHIFT;
182 m->cp_hqd_pq_control |= order_base_2(q->queue_size / 4) - 1;
183 pr_debug("cp_hqd_pq_control 0x%x\n", m->cp_hqd_pq_control);
184
185 m->cp_hqd_pq_base_lo = lower_32_bits((uint64_t)q->queue_address >> 8);
186 m->cp_hqd_pq_base_hi = upper_32_bits((uint64_t)q->queue_address >> 8);
187
188 m->cp_hqd_pq_rptr_report_addr_lo = lower_32_bits((uint64_t)q->read_ptr);
189 m->cp_hqd_pq_rptr_report_addr_hi = upper_32_bits((uint64_t)q->read_ptr);
190 m->cp_hqd_pq_wptr_poll_addr_lo = lower_32_bits((uint64_t)q->write_ptr);
191 m->cp_hqd_pq_wptr_poll_addr_hi = upper_32_bits((uint64_t)q->write_ptr);
192
193 m->cp_hqd_pq_doorbell_control =
194 q->doorbell_off <<
195 CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_OFFSET__SHIFT;
196 pr_debug("cp_hqd_pq_doorbell_control 0x%x\n",
197 m->cp_hqd_pq_doorbell_control);
198
199 m->cp_hqd_eop_control = atc_bit << CP_HQD_EOP_CONTROL__EOP_ATC__SHIFT |
200 mtype << CP_HQD_EOP_CONTROL__MTYPE__SHIFT;
201
202 m->cp_hqd_ib_control = atc_bit << CP_HQD_IB_CONTROL__IB_ATC__SHIFT |
203 3 << CP_HQD_IB_CONTROL__MIN_IB_AVAIL_SIZE__SHIFT |
204 mtype << CP_HQD_IB_CONTROL__MTYPE__SHIFT;
205
206 /*
207 * HW does not clamp this field correctly. Maximum EOP queue size
208 * is constrained by per-SE EOP done signal count, which is 8-bit.
209 * Limit is 0xFF EOP entries (= 0x7F8 dwords). CP will not submit
210 * more than (EOP entry count - 1) so a queue size of 0x800 dwords
211 * is safe, giving a maximum field value of 0xA.
212 */
213 m->cp_hqd_eop_control |= min(0xA,
214 order_base_2(q->eop_ring_buffer_size / 4) - 1);
215 m->cp_hqd_eop_base_addr_lo =
216 lower_32_bits(q->eop_ring_buffer_address >> 8);
217 m->cp_hqd_eop_base_addr_hi =
218 upper_32_bits(q->eop_ring_buffer_address >> 8);
219
220 m->cp_hqd_iq_timer = atc_bit << CP_HQD_IQ_TIMER__IQ_ATC__SHIFT |
221 mtype << CP_HQD_IQ_TIMER__MTYPE__SHIFT;
222
223 m->cp_hqd_vmid = q->vmid;
224
225 if (q->format == KFD_QUEUE_FORMAT_AQL) {
226 m->cp_hqd_pq_control |= CP_HQD_PQ_CONTROL__NO_UPDATE_RPTR_MASK |
227 2 << CP_HQD_PQ_CONTROL__SLOT_BASED_WPTR__SHIFT;
228 }
229
230 if (mm->dev->cwsr_enabled && q->ctx_save_restore_area_address)
231 m->cp_hqd_ctx_save_control =
232 atc_bit << CP_HQD_CTX_SAVE_CONTROL__ATC__SHIFT |
233 mtype << CP_HQD_CTX_SAVE_CONTROL__MTYPE__SHIFT;
234
235 update_cu_mask(mm, mqd, minfo);
236 set_priority(m, q);
237
238 q->is_active = QUEUE_IS_ACTIVE(*q);
239}
240
241
242static void update_mqd(struct mqd_manager *mm, void *mqd,
243 struct queue_properties *q,
244 struct mqd_update_info *minfo)
245{
246 __update_mqd(mm, mqd, q, minfo, MTYPE_CC, 1);
247}
248
249static uint32_t read_doorbell_id(void *mqd)
250{
251 struct vi_mqd *m = (struct vi_mqd *)mqd;
252
253 return m->queue_doorbell_id0;
254}
255
256static void update_mqd_tonga(struct mqd_manager *mm, void *mqd,
257 struct queue_properties *q,
258 struct mqd_update_info *minfo)
259{
260 __update_mqd(mm, mqd, q, minfo, MTYPE_UC, 0);
261}
262
263static int get_wave_state(struct mqd_manager *mm, void *mqd,
264 void __user *ctl_stack,
265 u32 *ctl_stack_used_size,
266 u32 *save_area_used_size)
267{
268 struct vi_mqd *m;
269
270 m = get_mqd(mqd);
271
272 *ctl_stack_used_size = m->cp_hqd_cntl_stack_size -
273 m->cp_hqd_cntl_stack_offset;
274 *save_area_used_size = m->cp_hqd_wg_state_offset -
275 m->cp_hqd_cntl_stack_size;
276
277 /* Control stack is not copied to user mode for GFXv8 because
278 * it's part of the context save area that is already
279 * accessible to user mode
280 */
281
282 return 0;
283}
284
285static void get_checkpoint_info(struct mqd_manager *mm, void *mqd, u32 *ctl_stack_size)
286{
287 /* Control stack is stored in user mode */
288 *ctl_stack_size = 0;
289}
290
291static void checkpoint_mqd(struct mqd_manager *mm, void *mqd, void *mqd_dst, void *ctl_stack_dst)
292{
293 struct vi_mqd *m;
294
295 m = get_mqd(mqd);
296
297 memcpy(mqd_dst, m, sizeof(struct vi_mqd));
298}
299
300static void restore_mqd(struct mqd_manager *mm, void **mqd,
301 struct kfd_mem_obj *mqd_mem_obj, uint64_t *gart_addr,
302 struct queue_properties *qp,
303 const void *mqd_src,
304 const void *ctl_stack_src, const u32 ctl_stack_size)
305{
306 uint64_t addr;
307 struct vi_mqd *m;
308
309 m = (struct vi_mqd *) mqd_mem_obj->cpu_ptr;
310 addr = mqd_mem_obj->gpu_addr;
311
312 memcpy(m, mqd_src, sizeof(*m));
313
314 *mqd = m;
315 if (gart_addr)
316 *gart_addr = addr;
317
318 m->cp_hqd_pq_doorbell_control =
319 qp->doorbell_off <<
320 CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_OFFSET__SHIFT;
321 pr_debug("cp_hqd_pq_doorbell_control 0x%x\n",
322 m->cp_hqd_pq_doorbell_control);
323
324 qp->is_active = 0;
325}
326
327static void init_mqd_hiq(struct mqd_manager *mm, void **mqd,
328 struct kfd_mem_obj *mqd_mem_obj, uint64_t *gart_addr,
329 struct queue_properties *q)
330{
331 struct vi_mqd *m;
332
333 init_mqd(mm, mqd, mqd_mem_obj, gart_addr, q);
334
335 m = get_mqd(*mqd);
336
337 m->cp_hqd_pq_control |= 1 << CP_HQD_PQ_CONTROL__PRIV_STATE__SHIFT |
338 1 << CP_HQD_PQ_CONTROL__KMD_QUEUE__SHIFT;
339}
340
341static void update_mqd_hiq(struct mqd_manager *mm, void *mqd,
342 struct queue_properties *q,
343 struct mqd_update_info *minfo)
344{
345 __update_mqd(mm, mqd, q, minfo, MTYPE_UC, 0);
346}
347
348static void init_mqd_sdma(struct mqd_manager *mm, void **mqd,
349 struct kfd_mem_obj *mqd_mem_obj, uint64_t *gart_addr,
350 struct queue_properties *q)
351{
352 struct vi_sdma_mqd *m;
353
354 m = (struct vi_sdma_mqd *) mqd_mem_obj->cpu_ptr;
355
356 memset(m, 0, sizeof(struct vi_sdma_mqd));
357
358 *mqd = m;
359 if (gart_addr)
360 *gart_addr = mqd_mem_obj->gpu_addr;
361
362 mm->update_mqd(mm, m, q, NULL);
363}
364
365static void update_mqd_sdma(struct mqd_manager *mm, void *mqd,
366 struct queue_properties *q,
367 struct mqd_update_info *minfo)
368{
369 struct vi_sdma_mqd *m;
370
371 m = get_sdma_mqd(mqd);
372 m->sdmax_rlcx_rb_cntl = order_base_2(q->queue_size / 4)
373 << SDMA0_RLC0_RB_CNTL__RB_SIZE__SHIFT |
374 q->vmid << SDMA0_RLC0_RB_CNTL__RB_VMID__SHIFT |
375 1 << SDMA0_RLC0_RB_CNTL__RPTR_WRITEBACK_ENABLE__SHIFT |
376 6 << SDMA0_RLC0_RB_CNTL__RPTR_WRITEBACK_TIMER__SHIFT;
377
378 m->sdmax_rlcx_rb_base = lower_32_bits(q->queue_address >> 8);
379 m->sdmax_rlcx_rb_base_hi = upper_32_bits(q->queue_address >> 8);
380 m->sdmax_rlcx_rb_rptr_addr_lo = lower_32_bits((uint64_t)q->read_ptr);
381 m->sdmax_rlcx_rb_rptr_addr_hi = upper_32_bits((uint64_t)q->read_ptr);
382 m->sdmax_rlcx_doorbell =
383 q->doorbell_off << SDMA0_RLC0_DOORBELL__OFFSET__SHIFT;
384
385 m->sdmax_rlcx_virtual_addr = q->sdma_vm_addr;
386
387 m->sdma_engine_id = q->sdma_engine_id;
388 m->sdma_queue_id = q->sdma_queue_id;
389
390 q->is_active = QUEUE_IS_ACTIVE(*q);
391}
392
393static void checkpoint_mqd_sdma(struct mqd_manager *mm,
394 void *mqd,
395 void *mqd_dst,
396 void *ctl_stack_dst)
397{
398 struct vi_sdma_mqd *m;
399
400 m = get_sdma_mqd(mqd);
401
402 memcpy(mqd_dst, m, sizeof(struct vi_sdma_mqd));
403}
404
405static void restore_mqd_sdma(struct mqd_manager *mm, void **mqd,
406 struct kfd_mem_obj *mqd_mem_obj, uint64_t *gart_addr,
407 struct queue_properties *qp,
408 const void *mqd_src,
409 const void *ctl_stack_src, const u32 ctl_stack_size)
410{
411 uint64_t addr;
412 struct vi_sdma_mqd *m;
413
414 m = (struct vi_sdma_mqd *) mqd_mem_obj->cpu_ptr;
415 addr = mqd_mem_obj->gpu_addr;
416
417 memcpy(m, mqd_src, sizeof(*m));
418
419 m->sdmax_rlcx_doorbell =
420 qp->doorbell_off << SDMA0_RLC0_DOORBELL__OFFSET__SHIFT;
421
422 *mqd = m;
423 if (gart_addr)
424 *gart_addr = addr;
425
426 qp->is_active = 0;
427}
428
429#if defined(CONFIG_DEBUG_FS)
430
431
432static int debugfs_show_mqd(struct seq_file *m, void *data)
433{
434 seq_hex_dump(m, " ", DUMP_PREFIX_OFFSET, 32, 4,
435 data, sizeof(struct vi_mqd), false);
436 return 0;
437}
438
439static int debugfs_show_mqd_sdma(struct seq_file *m, void *data)
440{
441 seq_hex_dump(m, " ", DUMP_PREFIX_OFFSET, 32, 4,
442 data, sizeof(struct vi_sdma_mqd), false);
443 return 0;
444}
445
446#endif
447
448struct mqd_manager *mqd_manager_init_vi(enum KFD_MQD_TYPE type,
449 struct kfd_dev *dev)
450{
451 struct mqd_manager *mqd;
452
453 if (WARN_ON(type >= KFD_MQD_TYPE_MAX))
454 return NULL;
455
456 mqd = kzalloc(sizeof(*mqd), GFP_KERNEL);
457 if (!mqd)
458 return NULL;
459
460 mqd->dev = dev;
461
462 switch (type) {
463 case KFD_MQD_TYPE_CP:
464 mqd->allocate_mqd = allocate_mqd;
465 mqd->init_mqd = init_mqd;
466 mqd->free_mqd = kfd_free_mqd_cp;
467 mqd->load_mqd = load_mqd;
468 mqd->update_mqd = update_mqd;
469 mqd->destroy_mqd = kfd_destroy_mqd_cp;
470 mqd->is_occupied = kfd_is_occupied_cp;
471 mqd->get_wave_state = get_wave_state;
472 mqd->get_checkpoint_info = get_checkpoint_info;
473 mqd->checkpoint_mqd = checkpoint_mqd;
474 mqd->restore_mqd = restore_mqd;
475 mqd->mqd_size = sizeof(struct vi_mqd);
476#if defined(CONFIG_DEBUG_FS)
477 mqd->debugfs_show_mqd = debugfs_show_mqd;
478#endif
479 break;
480 case KFD_MQD_TYPE_HIQ:
481 mqd->allocate_mqd = allocate_hiq_mqd;
482 mqd->init_mqd = init_mqd_hiq;
483 mqd->free_mqd = free_mqd_hiq_sdma;
484 mqd->load_mqd = load_mqd;
485 mqd->update_mqd = update_mqd_hiq;
486 mqd->destroy_mqd = kfd_destroy_mqd_cp;
487 mqd->is_occupied = kfd_is_occupied_cp;
488 mqd->mqd_size = sizeof(struct vi_mqd);
489#if defined(CONFIG_DEBUG_FS)
490 mqd->debugfs_show_mqd = debugfs_show_mqd;
491#endif
492 mqd->read_doorbell_id = read_doorbell_id;
493 break;
494 case KFD_MQD_TYPE_DIQ:
495 mqd->allocate_mqd = allocate_mqd;
496 mqd->init_mqd = init_mqd_hiq;
497 mqd->free_mqd = kfd_free_mqd_cp;
498 mqd->load_mqd = load_mqd;
499 mqd->update_mqd = update_mqd_hiq;
500 mqd->destroy_mqd = kfd_destroy_mqd_cp;
501 mqd->is_occupied = kfd_is_occupied_cp;
502 mqd->mqd_size = sizeof(struct vi_mqd);
503#if defined(CONFIG_DEBUG_FS)
504 mqd->debugfs_show_mqd = debugfs_show_mqd;
505#endif
506 break;
507 case KFD_MQD_TYPE_SDMA:
508 mqd->allocate_mqd = allocate_sdma_mqd;
509 mqd->init_mqd = init_mqd_sdma;
510 mqd->free_mqd = free_mqd_hiq_sdma;
511 mqd->load_mqd = kfd_load_mqd_sdma;
512 mqd->update_mqd = update_mqd_sdma;
513 mqd->destroy_mqd = kfd_destroy_mqd_sdma;
514 mqd->is_occupied = kfd_is_occupied_sdma;
515 mqd->checkpoint_mqd = checkpoint_mqd_sdma;
516 mqd->restore_mqd = restore_mqd_sdma;
517 mqd->mqd_size = sizeof(struct vi_sdma_mqd);
518#if defined(CONFIG_DEBUG_FS)
519 mqd->debugfs_show_mqd = debugfs_show_mqd_sdma;
520#endif
521 break;
522 default:
523 kfree(mqd);
524 return NULL;
525 }
526
527 return mqd;
528}
529
530struct mqd_manager *mqd_manager_init_vi_tonga(enum KFD_MQD_TYPE type,
531 struct kfd_dev *dev)
532{
533 struct mqd_manager *mqd;
534
535 mqd = mqd_manager_init_vi(type, dev);
536 if (!mqd)
537 return NULL;
538 if (type == KFD_MQD_TYPE_CP)
539 mqd->update_mqd = update_mqd_tonga;
540 return mqd;
541}