1 /*
2 * Copyright 2019 Advanced Micro Devices, Inc.
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 shall be included in
12 * all copies or substantial portions of the Software.
13 *
14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
17 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20 * OTHER DEALINGS IN THE SOFTWARE.
21 *
22 */
23
24 #include "amdgpu.h"
25 #include "amdgpu_jpeg.h"
26 #include "amdgpu_pm.h"
27 #include "soc15.h"
28 #include "soc15d.h"
29 #include "jpeg_v2_0.h"
30
31 #include "vcn/vcn_2_0_0_offset.h"
32 #include "vcn/vcn_2_0_0_sh_mask.h"
33 #include "ivsrcid/vcn/irqsrcs_vcn_2_0.h"
34
35 static void jpeg_v2_0_set_dec_ring_funcs(struct amdgpu_device *adev);
36 static void jpeg_v2_0_set_irq_funcs(struct amdgpu_device *adev);
37 static int jpeg_v2_0_set_powergating_state(void *handle,
38 enum amd_powergating_state state);
39
40 /**
41 * jpeg_v2_0_early_init - set function pointers
42 *
43 * @handle: amdgpu_device pointer
44 *
45 * Set ring and irq function pointers
46 */
jpeg_v2_0_early_init(void * handle)47 static int jpeg_v2_0_early_init(void *handle)
48 {
49 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
50
51 adev->jpeg.num_jpeg_inst = 1;
52 adev->jpeg.num_jpeg_rings = 1;
53
54 jpeg_v2_0_set_dec_ring_funcs(adev);
55 jpeg_v2_0_set_irq_funcs(adev);
56
57 return 0;
58 }
59
60 /**
61 * jpeg_v2_0_sw_init - sw init for JPEG block
62 *
63 * @handle: amdgpu_device pointer
64 *
65 * Load firmware and sw initialization
66 */
jpeg_v2_0_sw_init(void * handle)67 static int jpeg_v2_0_sw_init(void *handle)
68 {
69 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
70 struct amdgpu_ring *ring;
71 int r;
72
73 /* JPEG TRAP */
74 r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_VCN,
75 VCN_2_0__SRCID__JPEG_DECODE, &adev->jpeg.inst->irq);
76 if (r)
77 return r;
78
79 r = amdgpu_jpeg_sw_init(adev);
80 if (r)
81 return r;
82
83 r = amdgpu_jpeg_resume(adev);
84 if (r)
85 return r;
86
87 ring = adev->jpeg.inst->ring_dec;
88 ring->use_doorbell = true;
89 ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) + 1;
90 ring->vm_hub = AMDGPU_MMHUB0(0);
91 sprintf(ring->name, "jpeg_dec");
92 r = amdgpu_ring_init(adev, ring, 512, &adev->jpeg.inst->irq,
93 0, AMDGPU_RING_PRIO_DEFAULT, NULL);
94 if (r)
95 return r;
96
97 adev->jpeg.internal.jpeg_pitch[0] = mmUVD_JPEG_PITCH_INTERNAL_OFFSET;
98 adev->jpeg.inst->external.jpeg_pitch[0] = SOC15_REG_OFFSET(JPEG, 0, mmUVD_JPEG_PITCH);
99
100 return 0;
101 }
102
103 /**
104 * jpeg_v2_0_sw_fini - sw fini for JPEG block
105 *
106 * @handle: amdgpu_device pointer
107 *
108 * JPEG suspend and free up sw allocation
109 */
jpeg_v2_0_sw_fini(void * handle)110 static int jpeg_v2_0_sw_fini(void *handle)
111 {
112 int r;
113 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
114
115 r = amdgpu_jpeg_suspend(adev);
116 if (r)
117 return r;
118
119 r = amdgpu_jpeg_sw_fini(adev);
120
121 return r;
122 }
123
124 /**
125 * jpeg_v2_0_hw_init - start and test JPEG block
126 *
127 * @handle: amdgpu_device pointer
128 *
129 */
jpeg_v2_0_hw_init(void * handle)130 static int jpeg_v2_0_hw_init(void *handle)
131 {
132 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
133 struct amdgpu_ring *ring = adev->jpeg.inst->ring_dec;
134 int r;
135
136 adev->nbio.funcs->vcn_doorbell_range(adev, ring->use_doorbell,
137 (adev->doorbell_index.vcn.vcn_ring0_1 << 1), 0);
138
139 r = amdgpu_ring_test_helper(ring);
140 if (!r)
141 DRM_INFO("JPEG decode initialized successfully.\n");
142
143 return r;
144 }
145
146 /**
147 * jpeg_v2_0_hw_fini - stop the hardware block
148 *
149 * @handle: amdgpu_device pointer
150 *
151 * Stop the JPEG block, mark ring as not ready any more
152 */
jpeg_v2_0_hw_fini(void * handle)153 static int jpeg_v2_0_hw_fini(void *handle)
154 {
155 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
156
157 cancel_delayed_work_sync(&adev->vcn.idle_work);
158
159 if (adev->jpeg.cur_state != AMD_PG_STATE_GATE &&
160 RREG32_SOC15(JPEG, 0, mmUVD_JRBC_STATUS))
161 jpeg_v2_0_set_powergating_state(adev, AMD_PG_STATE_GATE);
162
163 return 0;
164 }
165
166 /**
167 * jpeg_v2_0_suspend - suspend JPEG block
168 *
169 * @handle: amdgpu_device pointer
170 *
171 * HW fini and suspend JPEG block
172 */
jpeg_v2_0_suspend(void * handle)173 static int jpeg_v2_0_suspend(void *handle)
174 {
175 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
176 int r;
177
178 r = jpeg_v2_0_hw_fini(adev);
179 if (r)
180 return r;
181
182 r = amdgpu_jpeg_suspend(adev);
183
184 return r;
185 }
186
187 /**
188 * jpeg_v2_0_resume - resume JPEG block
189 *
190 * @handle: amdgpu_device pointer
191 *
192 * Resume firmware and hw init JPEG block
193 */
jpeg_v2_0_resume(void * handle)194 static int jpeg_v2_0_resume(void *handle)
195 {
196 int r;
197 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
198
199 r = amdgpu_jpeg_resume(adev);
200 if (r)
201 return r;
202
203 r = jpeg_v2_0_hw_init(adev);
204
205 return r;
206 }
207
jpeg_v2_0_disable_power_gating(struct amdgpu_device * adev)208 static int jpeg_v2_0_disable_power_gating(struct amdgpu_device *adev)
209 {
210 uint32_t data;
211 int r = 0;
212
213 if (adev->pg_flags & AMD_PG_SUPPORT_JPEG) {
214 data = 1 << UVD_PGFSM_CONFIG__UVDJ_PWR_CONFIG__SHIFT;
215 WREG32(SOC15_REG_OFFSET(JPEG, 0, mmUVD_PGFSM_CONFIG), data);
216
217 r = SOC15_WAIT_ON_RREG(JPEG, 0,
218 mmUVD_PGFSM_STATUS, UVD_PGFSM_STATUS_UVDJ_PWR_ON,
219 UVD_PGFSM_STATUS__UVDJ_PWR_STATUS_MASK);
220
221 if (r) {
222 DRM_ERROR("amdgpu: JPEG disable power gating failed\n");
223 return r;
224 }
225 }
226
227 /* Removing the anti hang mechanism to indicate the UVDJ tile is ON */
228 data = RREG32(SOC15_REG_OFFSET(JPEG, 0, mmUVD_JPEG_POWER_STATUS)) & ~0x1;
229 WREG32(SOC15_REG_OFFSET(JPEG, 0, mmUVD_JPEG_POWER_STATUS), data);
230
231 return 0;
232 }
233
jpeg_v2_0_enable_power_gating(struct amdgpu_device * adev)234 static int jpeg_v2_0_enable_power_gating(struct amdgpu_device *adev)
235 {
236 if (adev->pg_flags & AMD_PG_SUPPORT_JPEG) {
237 uint32_t data;
238 int r = 0;
239
240 data = RREG32(SOC15_REG_OFFSET(JPEG, 0, mmUVD_JPEG_POWER_STATUS));
241 data &= ~UVD_JPEG_POWER_STATUS__JPEG_POWER_STATUS_MASK;
242 data |= 0x1; //UVD_JPEG_POWER_STATUS__JPEG_POWER_STATUS_TILES_OFF;
243 WREG32(SOC15_REG_OFFSET(JPEG, 0, mmUVD_JPEG_POWER_STATUS), data);
244
245 data = 2 << UVD_PGFSM_CONFIG__UVDJ_PWR_CONFIG__SHIFT;
246 WREG32(SOC15_REG_OFFSET(JPEG, 0, mmUVD_PGFSM_CONFIG), data);
247
248 r = SOC15_WAIT_ON_RREG(JPEG, 0, mmUVD_PGFSM_STATUS,
249 (2 << UVD_PGFSM_STATUS__UVDJ_PWR_STATUS__SHIFT),
250 UVD_PGFSM_STATUS__UVDJ_PWR_STATUS_MASK);
251
252 if (r) {
253 DRM_ERROR("amdgpu: JPEG enable power gating failed\n");
254 return r;
255 }
256 }
257
258 return 0;
259 }
260
jpeg_v2_0_disable_clock_gating(struct amdgpu_device * adev)261 static void jpeg_v2_0_disable_clock_gating(struct amdgpu_device *adev)
262 {
263 uint32_t data;
264
265 data = RREG32_SOC15(JPEG, 0, mmJPEG_CGC_CTRL);
266 if (adev->cg_flags & AMD_CG_SUPPORT_JPEG_MGCG)
267 data |= 1 << JPEG_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
268 else
269 data &= ~JPEG_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
270
271 data |= 1 << JPEG_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
272 data |= 4 << JPEG_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
273 WREG32_SOC15(JPEG, 0, mmJPEG_CGC_CTRL, data);
274
275 data = RREG32_SOC15(JPEG, 0, mmJPEG_CGC_GATE);
276 data &= ~(JPEG_CGC_GATE__JPEG_DEC_MASK
277 | JPEG_CGC_GATE__JPEG2_DEC_MASK
278 | JPEG_CGC_GATE__JPEG_ENC_MASK
279 | JPEG_CGC_GATE__JMCIF_MASK
280 | JPEG_CGC_GATE__JRBBM_MASK);
281 WREG32_SOC15(JPEG, 0, mmJPEG_CGC_GATE, data);
282 }
283
jpeg_v2_0_enable_clock_gating(struct amdgpu_device * adev)284 static void jpeg_v2_0_enable_clock_gating(struct amdgpu_device *adev)
285 {
286 uint32_t data;
287
288 data = RREG32_SOC15(JPEG, 0, mmJPEG_CGC_CTRL);
289 if (adev->cg_flags & AMD_CG_SUPPORT_JPEG_MGCG)
290 data |= 1 << JPEG_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
291 else
292 data |= 0 << JPEG_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
293
294 data |= 1 << JPEG_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
295 data |= 4 << JPEG_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
296 WREG32_SOC15(JPEG, 0, mmJPEG_CGC_CTRL, data);
297
298 data = RREG32_SOC15(JPEG, 0, mmJPEG_CGC_GATE);
299 data |= (JPEG_CGC_GATE__JPEG_DEC_MASK
300 |JPEG_CGC_GATE__JPEG2_DEC_MASK
301 |JPEG_CGC_GATE__JPEG_ENC_MASK
302 |JPEG_CGC_GATE__JMCIF_MASK
303 |JPEG_CGC_GATE__JRBBM_MASK);
304 WREG32_SOC15(JPEG, 0, mmJPEG_CGC_GATE, data);
305 }
306
307 /**
308 * jpeg_v2_0_start - start JPEG block
309 *
310 * @adev: amdgpu_device pointer
311 *
312 * Setup and start the JPEG block
313 */
jpeg_v2_0_start(struct amdgpu_device * adev)314 static int jpeg_v2_0_start(struct amdgpu_device *adev)
315 {
316 struct amdgpu_ring *ring = adev->jpeg.inst->ring_dec;
317 int r;
318
319 if (adev->pm.dpm_enabled)
320 amdgpu_dpm_enable_jpeg(adev, true);
321
322 /* disable power gating */
323 r = jpeg_v2_0_disable_power_gating(adev);
324 if (r)
325 return r;
326
327 /* JPEG disable CGC */
328 jpeg_v2_0_disable_clock_gating(adev);
329
330 WREG32_SOC15(JPEG, 0, mmJPEG_DEC_GFX10_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
331
332 /* enable JMI channel */
333 WREG32_P(SOC15_REG_OFFSET(JPEG, 0, mmUVD_JMI_CNTL), 0,
334 ~UVD_JMI_CNTL__SOFT_RESET_MASK);
335
336 /* enable System Interrupt for JRBC */
337 WREG32_P(SOC15_REG_OFFSET(JPEG, 0, mmJPEG_SYS_INT_EN),
338 JPEG_SYS_INT_EN__DJRBC_MASK,
339 ~JPEG_SYS_INT_EN__DJRBC_MASK);
340
341 WREG32_SOC15(JPEG, 0, mmUVD_LMI_JRBC_RB_VMID, 0);
342 WREG32_SOC15(JPEG, 0, mmUVD_JRBC_RB_CNTL, (0x00000001L | 0x00000002L));
343 WREG32_SOC15(JPEG, 0, mmUVD_LMI_JRBC_RB_64BIT_BAR_LOW,
344 lower_32_bits(ring->gpu_addr));
345 WREG32_SOC15(JPEG, 0, mmUVD_LMI_JRBC_RB_64BIT_BAR_HIGH,
346 upper_32_bits(ring->gpu_addr));
347 WREG32_SOC15(JPEG, 0, mmUVD_JRBC_RB_RPTR, 0);
348 WREG32_SOC15(JPEG, 0, mmUVD_JRBC_RB_WPTR, 0);
349 WREG32_SOC15(JPEG, 0, mmUVD_JRBC_RB_CNTL, 0x00000002L);
350 WREG32_SOC15(JPEG, 0, mmUVD_JRBC_RB_SIZE, ring->ring_size / 4);
351 ring->wptr = RREG32_SOC15(JPEG, 0, mmUVD_JRBC_RB_WPTR);
352
353 return 0;
354 }
355
356 /**
357 * jpeg_v2_0_stop - stop JPEG block
358 *
359 * @adev: amdgpu_device pointer
360 *
361 * stop the JPEG block
362 */
jpeg_v2_0_stop(struct amdgpu_device * adev)363 static int jpeg_v2_0_stop(struct amdgpu_device *adev)
364 {
365 int r;
366
367 /* reset JMI */
368 WREG32_P(SOC15_REG_OFFSET(JPEG, 0, mmUVD_JMI_CNTL),
369 UVD_JMI_CNTL__SOFT_RESET_MASK,
370 ~UVD_JMI_CNTL__SOFT_RESET_MASK);
371
372 /* enable JPEG CGC */
373 jpeg_v2_0_enable_clock_gating(adev);
374
375 /* enable power gating */
376 r = jpeg_v2_0_enable_power_gating(adev);
377 if (r)
378 return r;
379
380 if (adev->pm.dpm_enabled)
381 amdgpu_dpm_enable_jpeg(adev, false);
382
383 return 0;
384 }
385
386 /**
387 * jpeg_v2_0_dec_ring_get_rptr - get read pointer
388 *
389 * @ring: amdgpu_ring pointer
390 *
391 * Returns the current hardware read pointer
392 */
jpeg_v2_0_dec_ring_get_rptr(struct amdgpu_ring * ring)393 static uint64_t jpeg_v2_0_dec_ring_get_rptr(struct amdgpu_ring *ring)
394 {
395 struct amdgpu_device *adev = ring->adev;
396
397 return RREG32_SOC15(JPEG, 0, mmUVD_JRBC_RB_RPTR);
398 }
399
400 /**
401 * jpeg_v2_0_dec_ring_get_wptr - get write pointer
402 *
403 * @ring: amdgpu_ring pointer
404 *
405 * Returns the current hardware write pointer
406 */
jpeg_v2_0_dec_ring_get_wptr(struct amdgpu_ring * ring)407 static uint64_t jpeg_v2_0_dec_ring_get_wptr(struct amdgpu_ring *ring)
408 {
409 struct amdgpu_device *adev = ring->adev;
410
411 if (ring->use_doorbell)
412 return *ring->wptr_cpu_addr;
413 else
414 return RREG32_SOC15(JPEG, 0, mmUVD_JRBC_RB_WPTR);
415 }
416
417 /**
418 * jpeg_v2_0_dec_ring_set_wptr - set write pointer
419 *
420 * @ring: amdgpu_ring pointer
421 *
422 * Commits the write pointer to the hardware
423 */
jpeg_v2_0_dec_ring_set_wptr(struct amdgpu_ring * ring)424 static void jpeg_v2_0_dec_ring_set_wptr(struct amdgpu_ring *ring)
425 {
426 struct amdgpu_device *adev = ring->adev;
427
428 if (ring->use_doorbell) {
429 *ring->wptr_cpu_addr = lower_32_bits(ring->wptr);
430 WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
431 } else {
432 WREG32_SOC15(JPEG, 0, mmUVD_JRBC_RB_WPTR, lower_32_bits(ring->wptr));
433 }
434 }
435
436 /**
437 * jpeg_v2_0_dec_ring_insert_start - insert a start command
438 *
439 * @ring: amdgpu_ring pointer
440 *
441 * Write a start command to the ring.
442 */
jpeg_v2_0_dec_ring_insert_start(struct amdgpu_ring * ring)443 void jpeg_v2_0_dec_ring_insert_start(struct amdgpu_ring *ring)
444 {
445 amdgpu_ring_write(ring, PACKETJ(mmUVD_JRBC_EXTERNAL_REG_INTERNAL_OFFSET,
446 0, 0, PACKETJ_TYPE0));
447 amdgpu_ring_write(ring, 0x68e04);
448
449 amdgpu_ring_write(ring, PACKETJ(JRBC_DEC_EXTERNAL_REG_WRITE_ADDR,
450 0, 0, PACKETJ_TYPE0));
451 amdgpu_ring_write(ring, 0x80010000);
452 }
453
454 /**
455 * jpeg_v2_0_dec_ring_insert_end - insert a end command
456 *
457 * @ring: amdgpu_ring pointer
458 *
459 * Write a end command to the ring.
460 */
jpeg_v2_0_dec_ring_insert_end(struct amdgpu_ring * ring)461 void jpeg_v2_0_dec_ring_insert_end(struct amdgpu_ring *ring)
462 {
463 amdgpu_ring_write(ring, PACKETJ(mmUVD_JRBC_EXTERNAL_REG_INTERNAL_OFFSET,
464 0, 0, PACKETJ_TYPE0));
465 amdgpu_ring_write(ring, 0x68e04);
466
467 amdgpu_ring_write(ring, PACKETJ(JRBC_DEC_EXTERNAL_REG_WRITE_ADDR,
468 0, 0, PACKETJ_TYPE0));
469 amdgpu_ring_write(ring, 0x00010000);
470 }
471
472 /**
473 * jpeg_v2_0_dec_ring_emit_fence - emit an fence & trap command
474 *
475 * @ring: amdgpu_ring pointer
476 * @addr: address
477 * @seq: sequence number
478 * @flags: fence related flags
479 *
480 * Write a fence and a trap command to the ring.
481 */
jpeg_v2_0_dec_ring_emit_fence(struct amdgpu_ring * ring,u64 addr,u64 seq,unsigned flags)482 void jpeg_v2_0_dec_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, u64 seq,
483 unsigned flags)
484 {
485 WARN_ON(flags & AMDGPU_FENCE_FLAG_64BIT);
486
487 amdgpu_ring_write(ring, PACKETJ(mmUVD_JPEG_GPCOM_DATA0_INTERNAL_OFFSET,
488 0, 0, PACKETJ_TYPE0));
489 amdgpu_ring_write(ring, seq);
490
491 amdgpu_ring_write(ring, PACKETJ(mmUVD_JPEG_GPCOM_DATA1_INTERNAL_OFFSET,
492 0, 0, PACKETJ_TYPE0));
493 amdgpu_ring_write(ring, seq);
494
495 amdgpu_ring_write(ring, PACKETJ(mmUVD_LMI_JRBC_RB_MEM_WR_64BIT_BAR_LOW_INTERNAL_OFFSET,
496 0, 0, PACKETJ_TYPE0));
497 amdgpu_ring_write(ring, lower_32_bits(addr));
498
499 amdgpu_ring_write(ring, PACKETJ(mmUVD_LMI_JRBC_RB_MEM_WR_64BIT_BAR_HIGH_INTERNAL_OFFSET,
500 0, 0, PACKETJ_TYPE0));
501 amdgpu_ring_write(ring, upper_32_bits(addr));
502
503 amdgpu_ring_write(ring, PACKETJ(mmUVD_JPEG_GPCOM_CMD_INTERNAL_OFFSET,
504 0, 0, PACKETJ_TYPE0));
505 amdgpu_ring_write(ring, 0x8);
506
507 amdgpu_ring_write(ring, PACKETJ(mmUVD_JPEG_GPCOM_CMD_INTERNAL_OFFSET,
508 0, PACKETJ_CONDITION_CHECK0, PACKETJ_TYPE4));
509 amdgpu_ring_write(ring, 0);
510
511 amdgpu_ring_write(ring, PACKETJ(mmUVD_JRBC_EXTERNAL_REG_INTERNAL_OFFSET,
512 0, 0, PACKETJ_TYPE0));
513 amdgpu_ring_write(ring, 0x3fbc);
514
515 amdgpu_ring_write(ring, PACKETJ(JRBC_DEC_EXTERNAL_REG_WRITE_ADDR,
516 0, 0, PACKETJ_TYPE0));
517 amdgpu_ring_write(ring, 0x1);
518
519 amdgpu_ring_write(ring, PACKETJ(0, 0, 0, PACKETJ_TYPE7));
520 amdgpu_ring_write(ring, 0);
521 }
522
523 /**
524 * jpeg_v2_0_dec_ring_emit_ib - execute indirect buffer
525 *
526 * @ring: amdgpu_ring pointer
527 * @job: job to retrieve vmid from
528 * @ib: indirect buffer to execute
529 * @flags: unused
530 *
531 * Write ring commands to execute the indirect buffer.
532 */
jpeg_v2_0_dec_ring_emit_ib(struct amdgpu_ring * ring,struct amdgpu_job * job,struct amdgpu_ib * ib,uint32_t flags)533 void jpeg_v2_0_dec_ring_emit_ib(struct amdgpu_ring *ring,
534 struct amdgpu_job *job,
535 struct amdgpu_ib *ib,
536 uint32_t flags)
537 {
538 unsigned vmid = AMDGPU_JOB_GET_VMID(job);
539
540 amdgpu_ring_write(ring, PACKETJ(mmUVD_JPEG_IH_CTRL_INTERNAL_OFFSET,
541 0, 0, PACKETJ_TYPE0));
542 amdgpu_ring_write(ring, (vmid << JPEG_IH_CTRL__IH_VMID__SHIFT));
543
544 amdgpu_ring_write(ring, PACKETJ(mmUVD_LMI_JRBC_IB_VMID_INTERNAL_OFFSET,
545 0, 0, PACKETJ_TYPE0));
546 amdgpu_ring_write(ring, (vmid | (vmid << 4) | (vmid << 8)));
547
548 amdgpu_ring_write(ring, PACKETJ(mmUVD_LMI_JPEG_VMID_INTERNAL_OFFSET,
549 0, 0, PACKETJ_TYPE0));
550 amdgpu_ring_write(ring, (vmid | (vmid << 4) | (vmid << 8)));
551
552 amdgpu_ring_write(ring, PACKETJ(mmUVD_LMI_JRBC_IB_64BIT_BAR_LOW_INTERNAL_OFFSET,
553 0, 0, PACKETJ_TYPE0));
554 amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr));
555
556 amdgpu_ring_write(ring, PACKETJ(mmUVD_LMI_JRBC_IB_64BIT_BAR_HIGH_INTERNAL_OFFSET,
557 0, 0, PACKETJ_TYPE0));
558 amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr));
559
560 amdgpu_ring_write(ring, PACKETJ(mmUVD_JRBC_IB_SIZE_INTERNAL_OFFSET,
561 0, 0, PACKETJ_TYPE0));
562 amdgpu_ring_write(ring, ib->length_dw);
563
564 amdgpu_ring_write(ring, PACKETJ(mmUVD_LMI_JRBC_RB_MEM_RD_64BIT_BAR_LOW_INTERNAL_OFFSET,
565 0, 0, PACKETJ_TYPE0));
566 amdgpu_ring_write(ring, lower_32_bits(ring->gpu_addr));
567
568 amdgpu_ring_write(ring, PACKETJ(mmUVD_LMI_JRBC_RB_MEM_RD_64BIT_BAR_HIGH_INTERNAL_OFFSET,
569 0, 0, PACKETJ_TYPE0));
570 amdgpu_ring_write(ring, upper_32_bits(ring->gpu_addr));
571
572 amdgpu_ring_write(ring, PACKETJ(0, 0, PACKETJ_CONDITION_CHECK0, PACKETJ_TYPE2));
573 amdgpu_ring_write(ring, 0);
574
575 amdgpu_ring_write(ring, PACKETJ(mmUVD_JRBC_RB_COND_RD_TIMER_INTERNAL_OFFSET,
576 0, 0, PACKETJ_TYPE0));
577 amdgpu_ring_write(ring, 0x01400200);
578
579 amdgpu_ring_write(ring, PACKETJ(mmUVD_JRBC_RB_REF_DATA_INTERNAL_OFFSET,
580 0, 0, PACKETJ_TYPE0));
581 amdgpu_ring_write(ring, 0x2);
582
583 amdgpu_ring_write(ring, PACKETJ(mmUVD_JRBC_STATUS_INTERNAL_OFFSET,
584 0, PACKETJ_CONDITION_CHECK3, PACKETJ_TYPE3));
585 amdgpu_ring_write(ring, 0x2);
586 }
587
jpeg_v2_0_dec_ring_emit_reg_wait(struct amdgpu_ring * ring,uint32_t reg,uint32_t val,uint32_t mask)588 void jpeg_v2_0_dec_ring_emit_reg_wait(struct amdgpu_ring *ring, uint32_t reg,
589 uint32_t val, uint32_t mask)
590 {
591 uint32_t reg_offset = (reg << 2);
592
593 amdgpu_ring_write(ring, PACKETJ(mmUVD_JRBC_RB_COND_RD_TIMER_INTERNAL_OFFSET,
594 0, 0, PACKETJ_TYPE0));
595 amdgpu_ring_write(ring, 0x01400200);
596
597 amdgpu_ring_write(ring, PACKETJ(mmUVD_JRBC_RB_REF_DATA_INTERNAL_OFFSET,
598 0, 0, PACKETJ_TYPE0));
599 amdgpu_ring_write(ring, val);
600
601 amdgpu_ring_write(ring, PACKETJ(mmUVD_JRBC_EXTERNAL_REG_INTERNAL_OFFSET,
602 0, 0, PACKETJ_TYPE0));
603 if (reg_offset >= 0x10000 && reg_offset <= 0x105ff) {
604 amdgpu_ring_write(ring, 0);
605 amdgpu_ring_write(ring,
606 PACKETJ((reg_offset >> 2), 0, 0, PACKETJ_TYPE3));
607 } else {
608 amdgpu_ring_write(ring, reg_offset);
609 amdgpu_ring_write(ring, PACKETJ(JRBC_DEC_EXTERNAL_REG_WRITE_ADDR,
610 0, 0, PACKETJ_TYPE3));
611 }
612 amdgpu_ring_write(ring, mask);
613 }
614
jpeg_v2_0_dec_ring_emit_vm_flush(struct amdgpu_ring * ring,unsigned vmid,uint64_t pd_addr)615 void jpeg_v2_0_dec_ring_emit_vm_flush(struct amdgpu_ring *ring,
616 unsigned vmid, uint64_t pd_addr)
617 {
618 struct amdgpu_vmhub *hub = &ring->adev->vmhub[ring->vm_hub];
619 uint32_t data0, data1, mask;
620
621 pd_addr = amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr);
622
623 /* wait for register write */
624 data0 = hub->ctx0_ptb_addr_lo32 + vmid * hub->ctx_addr_distance;
625 data1 = lower_32_bits(pd_addr);
626 mask = 0xffffffff;
627 jpeg_v2_0_dec_ring_emit_reg_wait(ring, data0, data1, mask);
628 }
629
jpeg_v2_0_dec_ring_emit_wreg(struct amdgpu_ring * ring,uint32_t reg,uint32_t val)630 void jpeg_v2_0_dec_ring_emit_wreg(struct amdgpu_ring *ring, uint32_t reg, uint32_t val)
631 {
632 uint32_t reg_offset = (reg << 2);
633
634 amdgpu_ring_write(ring, PACKETJ(mmUVD_JRBC_EXTERNAL_REG_INTERNAL_OFFSET,
635 0, 0, PACKETJ_TYPE0));
636 if (reg_offset >= 0x10000 && reg_offset <= 0x105ff) {
637 amdgpu_ring_write(ring, 0);
638 amdgpu_ring_write(ring,
639 PACKETJ((reg_offset >> 2), 0, 0, PACKETJ_TYPE0));
640 } else {
641 amdgpu_ring_write(ring, reg_offset);
642 amdgpu_ring_write(ring, PACKETJ(JRBC_DEC_EXTERNAL_REG_WRITE_ADDR,
643 0, 0, PACKETJ_TYPE0));
644 }
645 amdgpu_ring_write(ring, val);
646 }
647
jpeg_v2_0_dec_ring_nop(struct amdgpu_ring * ring,uint32_t count)648 void jpeg_v2_0_dec_ring_nop(struct amdgpu_ring *ring, uint32_t count)
649 {
650 int i;
651
652 WARN_ON(ring->wptr % 2 || count % 2);
653
654 for (i = 0; i < count / 2; i++) {
655 amdgpu_ring_write(ring, PACKETJ(0, 0, 0, PACKETJ_TYPE6));
656 amdgpu_ring_write(ring, 0);
657 }
658 }
659
jpeg_v2_0_is_idle(void * handle)660 static bool jpeg_v2_0_is_idle(void *handle)
661 {
662 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
663
664 return ((RREG32_SOC15(JPEG, 0, mmUVD_JRBC_STATUS) &
665 UVD_JRBC_STATUS__RB_JOB_DONE_MASK) ==
666 UVD_JRBC_STATUS__RB_JOB_DONE_MASK);
667 }
668
jpeg_v2_0_wait_for_idle(void * handle)669 static int jpeg_v2_0_wait_for_idle(void *handle)
670 {
671 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
672 int ret;
673
674 ret = SOC15_WAIT_ON_RREG(JPEG, 0, mmUVD_JRBC_STATUS, UVD_JRBC_STATUS__RB_JOB_DONE_MASK,
675 UVD_JRBC_STATUS__RB_JOB_DONE_MASK);
676
677 return ret;
678 }
679
jpeg_v2_0_set_clockgating_state(void * handle,enum amd_clockgating_state state)680 static int jpeg_v2_0_set_clockgating_state(void *handle,
681 enum amd_clockgating_state state)
682 {
683 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
684 bool enable = (state == AMD_CG_STATE_GATE);
685
686 if (enable) {
687 if (!jpeg_v2_0_is_idle(handle))
688 return -EBUSY;
689 jpeg_v2_0_enable_clock_gating(adev);
690 } else {
691 jpeg_v2_0_disable_clock_gating(adev);
692 }
693
694 return 0;
695 }
696
jpeg_v2_0_set_powergating_state(void * handle,enum amd_powergating_state state)697 static int jpeg_v2_0_set_powergating_state(void *handle,
698 enum amd_powergating_state state)
699 {
700 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
701 int ret;
702
703 if (state == adev->jpeg.cur_state)
704 return 0;
705
706 if (state == AMD_PG_STATE_GATE)
707 ret = jpeg_v2_0_stop(adev);
708 else
709 ret = jpeg_v2_0_start(adev);
710
711 if (!ret)
712 adev->jpeg.cur_state = state;
713
714 return ret;
715 }
716
jpeg_v2_0_set_interrupt_state(struct amdgpu_device * adev,struct amdgpu_irq_src * source,unsigned type,enum amdgpu_interrupt_state state)717 static int jpeg_v2_0_set_interrupt_state(struct amdgpu_device *adev,
718 struct amdgpu_irq_src *source,
719 unsigned type,
720 enum amdgpu_interrupt_state state)
721 {
722 return 0;
723 }
724
jpeg_v2_0_process_interrupt(struct amdgpu_device * adev,struct amdgpu_irq_src * source,struct amdgpu_iv_entry * entry)725 static int jpeg_v2_0_process_interrupt(struct amdgpu_device *adev,
726 struct amdgpu_irq_src *source,
727 struct amdgpu_iv_entry *entry)
728 {
729 DRM_DEBUG("IH: JPEG TRAP\n");
730
731 switch (entry->src_id) {
732 case VCN_2_0__SRCID__JPEG_DECODE:
733 amdgpu_fence_process(adev->jpeg.inst->ring_dec);
734 break;
735 default:
736 DRM_ERROR("Unhandled interrupt: %d %d\n",
737 entry->src_id, entry->src_data[0]);
738 break;
739 }
740
741 return 0;
742 }
743
744 static const struct amd_ip_funcs jpeg_v2_0_ip_funcs = {
745 .name = "jpeg_v2_0",
746 .early_init = jpeg_v2_0_early_init,
747 .late_init = NULL,
748 .sw_init = jpeg_v2_0_sw_init,
749 .sw_fini = jpeg_v2_0_sw_fini,
750 .hw_init = jpeg_v2_0_hw_init,
751 .hw_fini = jpeg_v2_0_hw_fini,
752 .suspend = jpeg_v2_0_suspend,
753 .resume = jpeg_v2_0_resume,
754 .is_idle = jpeg_v2_0_is_idle,
755 .wait_for_idle = jpeg_v2_0_wait_for_idle,
756 .check_soft_reset = NULL,
757 .pre_soft_reset = NULL,
758 .soft_reset = NULL,
759 .post_soft_reset = NULL,
760 .set_clockgating_state = jpeg_v2_0_set_clockgating_state,
761 .set_powergating_state = jpeg_v2_0_set_powergating_state,
762 };
763
764 static const struct amdgpu_ring_funcs jpeg_v2_0_dec_ring_vm_funcs = {
765 .type = AMDGPU_RING_TYPE_VCN_JPEG,
766 .align_mask = 0xf,
767 .get_rptr = jpeg_v2_0_dec_ring_get_rptr,
768 .get_wptr = jpeg_v2_0_dec_ring_get_wptr,
769 .set_wptr = jpeg_v2_0_dec_ring_set_wptr,
770 .emit_frame_size =
771 SOC15_FLUSH_GPU_TLB_NUM_WREG * 6 +
772 SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 8 +
773 8 + /* jpeg_v2_0_dec_ring_emit_vm_flush */
774 18 + 18 + /* jpeg_v2_0_dec_ring_emit_fence x2 vm fence */
775 8 + 16,
776 .emit_ib_size = 24, /* jpeg_v2_0_dec_ring_emit_ib */
777 .emit_ib = jpeg_v2_0_dec_ring_emit_ib,
778 .emit_fence = jpeg_v2_0_dec_ring_emit_fence,
779 .emit_vm_flush = jpeg_v2_0_dec_ring_emit_vm_flush,
780 .test_ring = amdgpu_jpeg_dec_ring_test_ring,
781 .test_ib = amdgpu_jpeg_dec_ring_test_ib,
782 .insert_nop = jpeg_v2_0_dec_ring_nop,
783 .insert_start = jpeg_v2_0_dec_ring_insert_start,
784 .insert_end = jpeg_v2_0_dec_ring_insert_end,
785 .pad_ib = amdgpu_ring_generic_pad_ib,
786 .begin_use = amdgpu_jpeg_ring_begin_use,
787 .end_use = amdgpu_jpeg_ring_end_use,
788 .emit_wreg = jpeg_v2_0_dec_ring_emit_wreg,
789 .emit_reg_wait = jpeg_v2_0_dec_ring_emit_reg_wait,
790 .emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
791 };
792
jpeg_v2_0_set_dec_ring_funcs(struct amdgpu_device * adev)793 static void jpeg_v2_0_set_dec_ring_funcs(struct amdgpu_device *adev)
794 {
795 adev->jpeg.inst->ring_dec->funcs = &jpeg_v2_0_dec_ring_vm_funcs;
796 DRM_INFO("JPEG decode is enabled in VM mode\n");
797 }
798
799 static const struct amdgpu_irq_src_funcs jpeg_v2_0_irq_funcs = {
800 .set = jpeg_v2_0_set_interrupt_state,
801 .process = jpeg_v2_0_process_interrupt,
802 };
803
jpeg_v2_0_set_irq_funcs(struct amdgpu_device * adev)804 static void jpeg_v2_0_set_irq_funcs(struct amdgpu_device *adev)
805 {
806 adev->jpeg.inst->irq.num_types = 1;
807 adev->jpeg.inst->irq.funcs = &jpeg_v2_0_irq_funcs;
808 }
809
810 const struct amdgpu_ip_block_version jpeg_v2_0_ip_block = {
811 .type = AMD_IP_BLOCK_TYPE_JPEG,
812 .major = 2,
813 .minor = 0,
814 .rev = 0,
815 .funcs = &jpeg_v2_0_ip_funcs,
816 };
817