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 <linux/firmware.h>
25 #include <drm/drm_drv.h>
26
27 #include "amdgpu.h"
28 #include "amdgpu_vcn.h"
29 #include "amdgpu_pm.h"
30 #include "soc15.h"
31 #include "soc15d.h"
32 #include "vcn_v2_0.h"
33 #include "mmsch_v1_0.h"
34 #include "vcn_v2_5.h"
35
36 #include "vcn/vcn_2_5_offset.h"
37 #include "vcn/vcn_2_5_sh_mask.h"
38 #include "ivsrcid/vcn/irqsrcs_vcn_2_0.h"
39
40 #define VCN_VID_SOC_ADDRESS_2_0 0x1fa00
41 #define VCN1_VID_SOC_ADDRESS_3_0 0x48200
42
43 #define mmUVD_CONTEXT_ID_INTERNAL_OFFSET 0x27
44 #define mmUVD_GPCOM_VCPU_CMD_INTERNAL_OFFSET 0x0f
45 #define mmUVD_GPCOM_VCPU_DATA0_INTERNAL_OFFSET 0x10
46 #define mmUVD_GPCOM_VCPU_DATA1_INTERNAL_OFFSET 0x11
47 #define mmUVD_NO_OP_INTERNAL_OFFSET 0x29
48 #define mmUVD_GP_SCRATCH8_INTERNAL_OFFSET 0x66
49 #define mmUVD_SCRATCH9_INTERNAL_OFFSET 0xc01d
50
51 #define mmUVD_LMI_RBC_IB_VMID_INTERNAL_OFFSET 0x431
52 #define mmUVD_LMI_RBC_IB_64BIT_BAR_LOW_INTERNAL_OFFSET 0x3b4
53 #define mmUVD_LMI_RBC_IB_64BIT_BAR_HIGH_INTERNAL_OFFSET 0x3b5
54 #define mmUVD_RBC_IB_SIZE_INTERNAL_OFFSET 0x25c
55
56 #define VCN25_MAX_HW_INSTANCES_ARCTURUS 2
57
58 static const struct amdgpu_hwip_reg_entry vcn_reg_list_2_5[] = {
59 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_POWER_STATUS),
60 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_POWER_STATUS),
61 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_STATUS),
62 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_CONTEXT_ID),
63 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_CONTEXT_ID2),
64 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_GPCOM_VCPU_DATA0),
65 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_GPCOM_VCPU_DATA1),
66 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_GPCOM_VCPU_CMD),
67 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_BASE_HI),
68 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_BASE_LO),
69 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_BASE_HI2),
70 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_BASE_LO2),
71 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_BASE_HI3),
72 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_BASE_LO3),
73 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_BASE_HI4),
74 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_BASE_LO4),
75 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_RPTR),
76 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_WPTR),
77 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_RPTR2),
78 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_WPTR2),
79 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_RPTR3),
80 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_WPTR3),
81 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_RPTR4),
82 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_WPTR4),
83 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_SIZE),
84 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_SIZE2),
85 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_SIZE3),
86 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_RB_SIZE4),
87 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_PGFSM_CONFIG),
88 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_PGFSM_STATUS),
89 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_DPG_LMA_CTL),
90 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_DPG_LMA_DATA),
91 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_DPG_LMA_MASK),
92 SOC15_REG_ENTRY_STR(VCN, 0, mmUVD_DPG_PAUSE)
93 };
94
95 static void vcn_v2_5_set_dec_ring_funcs(struct amdgpu_device *adev);
96 static void vcn_v2_5_set_enc_ring_funcs(struct amdgpu_device *adev);
97 static void vcn_v2_5_set_irq_funcs(struct amdgpu_device *adev);
98 static int vcn_v2_5_set_powergating_state(void *handle,
99 enum amd_powergating_state state);
100 static int vcn_v2_5_pause_dpg_mode(struct amdgpu_device *adev,
101 int inst_idx, struct dpg_pause_state *new_state);
102 static int vcn_v2_5_sriov_start(struct amdgpu_device *adev);
103 static void vcn_v2_5_set_ras_funcs(struct amdgpu_device *adev);
104
105 static int amdgpu_ih_clientid_vcns[] = {
106 SOC15_IH_CLIENTID_VCN,
107 SOC15_IH_CLIENTID_VCN1
108 };
109
110 /**
111 * vcn_v2_5_early_init - set function pointers and load microcode
112 *
113 * @ip_block: Pointer to the amdgpu_ip_block for this hw instance.
114 *
115 * Set ring and irq function pointers
116 * Load microcode from filesystem
117 */
vcn_v2_5_early_init(struct amdgpu_ip_block * ip_block)118 static int vcn_v2_5_early_init(struct amdgpu_ip_block *ip_block)
119 {
120 struct amdgpu_device *adev = ip_block->adev;
121
122 if (amdgpu_sriov_vf(adev)) {
123 adev->vcn.num_vcn_inst = 2;
124 adev->vcn.harvest_config = 0;
125 adev->vcn.num_enc_rings = 1;
126 } else {
127 u32 harvest;
128 int i;
129
130 for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
131 harvest = RREG32_SOC15(VCN, i, mmCC_UVD_HARVESTING);
132 if (harvest & CC_UVD_HARVESTING__UVD_DISABLE_MASK)
133 adev->vcn.harvest_config |= 1 << i;
134 }
135 if (adev->vcn.harvest_config == (AMDGPU_VCN_HARVEST_VCN0 |
136 AMDGPU_VCN_HARVEST_VCN1))
137 /* both instances are harvested, disable the block */
138 return -ENOENT;
139
140 adev->vcn.num_enc_rings = 2;
141 }
142
143 vcn_v2_5_set_dec_ring_funcs(adev);
144 vcn_v2_5_set_enc_ring_funcs(adev);
145 vcn_v2_5_set_irq_funcs(adev);
146 vcn_v2_5_set_ras_funcs(adev);
147
148 return amdgpu_vcn_early_init(adev);
149 }
150
151 /**
152 * vcn_v2_5_sw_init - sw init for VCN block
153 *
154 * @ip_block: Pointer to the amdgpu_ip_block for this hw instance.
155 *
156 * Load firmware and sw initialization
157 */
vcn_v2_5_sw_init(struct amdgpu_ip_block * ip_block)158 static int vcn_v2_5_sw_init(struct amdgpu_ip_block *ip_block)
159 {
160 struct amdgpu_ring *ring;
161 int i, j, r;
162 uint32_t reg_count = ARRAY_SIZE(vcn_reg_list_2_5);
163 uint32_t *ptr;
164 struct amdgpu_device *adev = ip_block->adev;
165
166 for (j = 0; j < adev->vcn.num_vcn_inst; j++) {
167 if (adev->vcn.harvest_config & (1 << j))
168 continue;
169 /* VCN DEC TRAP */
170 r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[j],
171 VCN_2_0__SRCID__UVD_SYSTEM_MESSAGE_INTERRUPT, &adev->vcn.inst[j].irq);
172 if (r)
173 return r;
174
175 /* VCN ENC TRAP */
176 for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
177 r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[j],
178 i + VCN_2_0__SRCID__UVD_ENC_GENERAL_PURPOSE, &adev->vcn.inst[j].irq);
179 if (r)
180 return r;
181 }
182
183 /* VCN POISON TRAP */
184 r = amdgpu_irq_add_id(adev, amdgpu_ih_clientid_vcns[j],
185 VCN_2_6__SRCID_UVD_POISON, &adev->vcn.inst[j].ras_poison_irq);
186 if (r)
187 return r;
188 }
189
190 r = amdgpu_vcn_sw_init(adev);
191 if (r)
192 return r;
193
194 amdgpu_vcn_setup_ucode(adev);
195
196 r = amdgpu_vcn_resume(adev);
197 if (r)
198 return r;
199
200 for (j = 0; j < adev->vcn.num_vcn_inst; j++) {
201 volatile struct amdgpu_fw_shared *fw_shared;
202
203 if (adev->vcn.harvest_config & (1 << j))
204 continue;
205 adev->vcn.internal.context_id = mmUVD_CONTEXT_ID_INTERNAL_OFFSET;
206 adev->vcn.internal.ib_vmid = mmUVD_LMI_RBC_IB_VMID_INTERNAL_OFFSET;
207 adev->vcn.internal.ib_bar_low = mmUVD_LMI_RBC_IB_64BIT_BAR_LOW_INTERNAL_OFFSET;
208 adev->vcn.internal.ib_bar_high = mmUVD_LMI_RBC_IB_64BIT_BAR_HIGH_INTERNAL_OFFSET;
209 adev->vcn.internal.ib_size = mmUVD_RBC_IB_SIZE_INTERNAL_OFFSET;
210 adev->vcn.internal.gp_scratch8 = mmUVD_GP_SCRATCH8_INTERNAL_OFFSET;
211
212 adev->vcn.internal.scratch9 = mmUVD_SCRATCH9_INTERNAL_OFFSET;
213 adev->vcn.inst[j].external.scratch9 = SOC15_REG_OFFSET(VCN, j, mmUVD_SCRATCH9);
214 adev->vcn.internal.data0 = mmUVD_GPCOM_VCPU_DATA0_INTERNAL_OFFSET;
215 adev->vcn.inst[j].external.data0 = SOC15_REG_OFFSET(VCN, j, mmUVD_GPCOM_VCPU_DATA0);
216 adev->vcn.internal.data1 = mmUVD_GPCOM_VCPU_DATA1_INTERNAL_OFFSET;
217 adev->vcn.inst[j].external.data1 = SOC15_REG_OFFSET(VCN, j, mmUVD_GPCOM_VCPU_DATA1);
218 adev->vcn.internal.cmd = mmUVD_GPCOM_VCPU_CMD_INTERNAL_OFFSET;
219 adev->vcn.inst[j].external.cmd = SOC15_REG_OFFSET(VCN, j, mmUVD_GPCOM_VCPU_CMD);
220 adev->vcn.internal.nop = mmUVD_NO_OP_INTERNAL_OFFSET;
221 adev->vcn.inst[j].external.nop = SOC15_REG_OFFSET(VCN, j, mmUVD_NO_OP);
222
223 ring = &adev->vcn.inst[j].ring_dec;
224 ring->use_doorbell = true;
225
226 ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) +
227 (amdgpu_sriov_vf(adev) ? 2*j : 8*j);
228
229 if (amdgpu_ip_version(adev, UVD_HWIP, 0) == IP_VERSION(2, 5, 0))
230 ring->vm_hub = AMDGPU_MMHUB1(0);
231 else
232 ring->vm_hub = AMDGPU_MMHUB0(0);
233
234 sprintf(ring->name, "vcn_dec_%d", j);
235 r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst[j].irq,
236 0, AMDGPU_RING_PRIO_DEFAULT, NULL);
237 if (r)
238 return r;
239
240 for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
241 enum amdgpu_ring_priority_level hw_prio = amdgpu_vcn_get_enc_ring_prio(i);
242
243 ring = &adev->vcn.inst[j].ring_enc[i];
244 ring->use_doorbell = true;
245
246 ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) +
247 (amdgpu_sriov_vf(adev) ? (1 + i + 2*j) : (2 + i + 8*j));
248
249 if (amdgpu_ip_version(adev, UVD_HWIP, 0) ==
250 IP_VERSION(2, 5, 0))
251 ring->vm_hub = AMDGPU_MMHUB1(0);
252 else
253 ring->vm_hub = AMDGPU_MMHUB0(0);
254
255 sprintf(ring->name, "vcn_enc_%d.%d", j, i);
256 r = amdgpu_ring_init(adev, ring, 512,
257 &adev->vcn.inst[j].irq, 0,
258 hw_prio, NULL);
259 if (r)
260 return r;
261 }
262
263 fw_shared = adev->vcn.inst[j].fw_shared.cpu_addr;
264 fw_shared->present_flag_0 = cpu_to_le32(AMDGPU_VCN_MULTI_QUEUE_FLAG);
265
266 if (amdgpu_vcnfw_log)
267 amdgpu_vcn_fwlog_init(&adev->vcn.inst[i]);
268 }
269
270 if (amdgpu_sriov_vf(adev)) {
271 r = amdgpu_virt_alloc_mm_table(adev);
272 if (r)
273 return r;
274 }
275
276 if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)
277 adev->vcn.pause_dpg_mode = vcn_v2_5_pause_dpg_mode;
278
279 r = amdgpu_vcn_ras_sw_init(adev);
280 if (r)
281 return r;
282
283 /* Allocate memory for VCN IP Dump buffer */
284 ptr = kcalloc(adev->vcn.num_vcn_inst * reg_count, sizeof(uint32_t), GFP_KERNEL);
285 if (!ptr) {
286 DRM_ERROR("Failed to allocate memory for VCN IP Dump\n");
287 adev->vcn.ip_dump = NULL;
288 } else {
289 adev->vcn.ip_dump = ptr;
290 }
291
292 return 0;
293 }
294
295 /**
296 * vcn_v2_5_sw_fini - sw fini for VCN block
297 *
298 * @ip_block: Pointer to the amdgpu_ip_block for this hw instance.
299 *
300 * VCN suspend and free up sw allocation
301 */
vcn_v2_5_sw_fini(struct amdgpu_ip_block * ip_block)302 static int vcn_v2_5_sw_fini(struct amdgpu_ip_block *ip_block)
303 {
304 int i, r, idx;
305 struct amdgpu_device *adev = ip_block->adev;
306 volatile struct amdgpu_fw_shared *fw_shared;
307
308 if (drm_dev_enter(adev_to_drm(adev), &idx)) {
309 for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
310 if (adev->vcn.harvest_config & (1 << i))
311 continue;
312 fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
313 fw_shared->present_flag_0 = 0;
314 }
315 drm_dev_exit(idx);
316 }
317
318
319 if (amdgpu_sriov_vf(adev))
320 amdgpu_virt_free_mm_table(adev);
321
322 r = amdgpu_vcn_suspend(adev);
323 if (r)
324 return r;
325
326 r = amdgpu_vcn_sw_fini(adev);
327
328 kfree(adev->vcn.ip_dump);
329
330 return r;
331 }
332
333 /**
334 * vcn_v2_5_hw_init - start and test VCN block
335 *
336 * @ip_block: Pointer to the amdgpu_ip_block for this hw instance.
337 *
338 * Initialize the hardware, boot up the VCPU and do some testing
339 */
vcn_v2_5_hw_init(struct amdgpu_ip_block * ip_block)340 static int vcn_v2_5_hw_init(struct amdgpu_ip_block *ip_block)
341 {
342 struct amdgpu_device *adev = ip_block->adev;
343 struct amdgpu_ring *ring;
344 int i, j, r = 0;
345
346 if (amdgpu_sriov_vf(adev))
347 r = vcn_v2_5_sriov_start(adev);
348
349 for (j = 0; j < adev->vcn.num_vcn_inst; ++j) {
350 if (adev->vcn.harvest_config & (1 << j))
351 continue;
352
353 if (amdgpu_sriov_vf(adev)) {
354 adev->vcn.inst[j].ring_enc[0].sched.ready = true;
355 adev->vcn.inst[j].ring_enc[1].sched.ready = false;
356 adev->vcn.inst[j].ring_enc[2].sched.ready = false;
357 adev->vcn.inst[j].ring_dec.sched.ready = true;
358 } else {
359
360 ring = &adev->vcn.inst[j].ring_dec;
361
362 adev->nbio.funcs->vcn_doorbell_range(adev, ring->use_doorbell,
363 ring->doorbell_index, j);
364
365 r = amdgpu_ring_test_helper(ring);
366 if (r)
367 return r;
368
369 for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
370 ring = &adev->vcn.inst[j].ring_enc[i];
371 r = amdgpu_ring_test_helper(ring);
372 if (r)
373 return r;
374 }
375 }
376 }
377
378 return r;
379 }
380
381 /**
382 * vcn_v2_5_hw_fini - stop the hardware block
383 *
384 * @ip_block: Pointer to the amdgpu_ip_block for this hw instance.
385 *
386 * Stop the VCN block, mark ring as not ready any more
387 */
vcn_v2_5_hw_fini(struct amdgpu_ip_block * ip_block)388 static int vcn_v2_5_hw_fini(struct amdgpu_ip_block *ip_block)
389 {
390 struct amdgpu_device *adev = ip_block->adev;
391 int i;
392
393 cancel_delayed_work_sync(&adev->vcn.idle_work);
394
395 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
396 if (adev->vcn.harvest_config & (1 << i))
397 continue;
398
399 if ((adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) ||
400 (adev->vcn.cur_state != AMD_PG_STATE_GATE &&
401 RREG32_SOC15(VCN, i, mmUVD_STATUS)))
402 vcn_v2_5_set_powergating_state(adev, AMD_PG_STATE_GATE);
403
404 if (amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__VCN))
405 amdgpu_irq_put(adev, &adev->vcn.inst[i].ras_poison_irq, 0);
406 }
407
408 return 0;
409 }
410
411 /**
412 * vcn_v2_5_suspend - suspend VCN block
413 *
414 * @ip_block: Pointer to the amdgpu_ip_block for this hw instance.
415 *
416 * HW fini and suspend VCN block
417 */
vcn_v2_5_suspend(struct amdgpu_ip_block * ip_block)418 static int vcn_v2_5_suspend(struct amdgpu_ip_block *ip_block)
419 {
420 int r;
421
422 r = vcn_v2_5_hw_fini(ip_block);
423 if (r)
424 return r;
425
426 r = amdgpu_vcn_suspend(ip_block->adev);
427
428 return r;
429 }
430
431 /**
432 * vcn_v2_5_resume - resume VCN block
433 *
434 * @ip_block: Pointer to the amdgpu_ip_block for this hw instance.
435 *
436 * Resume firmware and hw init VCN block
437 */
vcn_v2_5_resume(struct amdgpu_ip_block * ip_block)438 static int vcn_v2_5_resume(struct amdgpu_ip_block *ip_block)
439 {
440 int r;
441
442 r = amdgpu_vcn_resume(ip_block->adev);
443 if (r)
444 return r;
445
446 r = vcn_v2_5_hw_init(ip_block);
447
448 return r;
449 }
450
451 /**
452 * vcn_v2_5_mc_resume - memory controller programming
453 *
454 * @adev: amdgpu_device pointer
455 *
456 * Let the VCN memory controller know it's offsets
457 */
vcn_v2_5_mc_resume(struct amdgpu_device * adev)458 static void vcn_v2_5_mc_resume(struct amdgpu_device *adev)
459 {
460 uint32_t size;
461 uint32_t offset;
462 int i;
463
464 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
465 if (adev->vcn.harvest_config & (1 << i))
466 continue;
467
468 size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw[i]->size + 4);
469 /* cache window 0: fw */
470 if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
471 WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
472 (adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_lo));
473 WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
474 (adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_hi));
475 WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_OFFSET0, 0);
476 offset = 0;
477 } else {
478 WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
479 lower_32_bits(adev->vcn.inst[i].gpu_addr));
480 WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
481 upper_32_bits(adev->vcn.inst[i].gpu_addr));
482 offset = size;
483 WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_OFFSET0,
484 AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
485 }
486 WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_SIZE0, size);
487
488 /* cache window 1: stack */
489 WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW,
490 lower_32_bits(adev->vcn.inst[i].gpu_addr + offset));
491 WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH,
492 upper_32_bits(adev->vcn.inst[i].gpu_addr + offset));
493 WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_OFFSET1, 0);
494 WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_SIZE1, AMDGPU_VCN_STACK_SIZE);
495
496 /* cache window 2: context */
497 WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW,
498 lower_32_bits(adev->vcn.inst[i].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
499 WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH,
500 upper_32_bits(adev->vcn.inst[i].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
501 WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_OFFSET2, 0);
502 WREG32_SOC15(VCN, i, mmUVD_VCPU_CACHE_SIZE2, AMDGPU_VCN_CONTEXT_SIZE);
503
504 /* non-cache window */
505 WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_NC0_64BIT_BAR_LOW,
506 lower_32_bits(adev->vcn.inst[i].fw_shared.gpu_addr));
507 WREG32_SOC15(VCN, i, mmUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH,
508 upper_32_bits(adev->vcn.inst[i].fw_shared.gpu_addr));
509 WREG32_SOC15(VCN, i, mmUVD_VCPU_NONCACHE_OFFSET0, 0);
510 WREG32_SOC15(VCN, i, mmUVD_VCPU_NONCACHE_SIZE0,
511 AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared)));
512 }
513 }
514
vcn_v2_5_mc_resume_dpg_mode(struct amdgpu_device * adev,int inst_idx,bool indirect)515 static void vcn_v2_5_mc_resume_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
516 {
517 uint32_t size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw[inst_idx]->size + 4);
518 uint32_t offset;
519
520 /* cache window 0: fw */
521 if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
522 if (!indirect) {
523 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
524 VCN, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
525 (adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_lo), 0, indirect);
526 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
527 VCN, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
528 (adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + inst_idx].tmr_mc_addr_hi), 0, indirect);
529 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
530 VCN, 0, mmUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
531 } else {
532 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
533 VCN, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW), 0, 0, indirect);
534 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
535 VCN, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH), 0, 0, indirect);
536 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
537 VCN, 0, mmUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
538 }
539 offset = 0;
540 } else {
541 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
542 VCN, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
543 lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect);
544 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
545 VCN, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
546 upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr), 0, indirect);
547 offset = size;
548 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
549 VCN, 0, mmUVD_VCPU_CACHE_OFFSET0),
550 AMDGPU_UVD_FIRMWARE_OFFSET >> 3, 0, indirect);
551 }
552
553 if (!indirect)
554 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
555 VCN, 0, mmUVD_VCPU_CACHE_SIZE0), size, 0, indirect);
556 else
557 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
558 VCN, 0, mmUVD_VCPU_CACHE_SIZE0), 0, 0, indirect);
559
560 /* cache window 1: stack */
561 if (!indirect) {
562 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
563 VCN, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
564 lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect);
565 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
566 VCN, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
567 upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset), 0, indirect);
568 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
569 VCN, 0, mmUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
570 } else {
571 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
572 VCN, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW), 0, 0, indirect);
573 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
574 VCN, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH), 0, 0, indirect);
575 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
576 VCN, 0, mmUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
577 }
578 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
579 VCN, 0, mmUVD_VCPU_CACHE_SIZE1), AMDGPU_VCN_STACK_SIZE, 0, indirect);
580
581 /* cache window 2: context */
582 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
583 VCN, 0, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
584 lower_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
585 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
586 VCN, 0, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
587 upper_32_bits(adev->vcn.inst[inst_idx].gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
588 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
589 VCN, 0, mmUVD_VCPU_CACHE_OFFSET2), 0, 0, indirect);
590 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
591 VCN, 0, mmUVD_VCPU_CACHE_SIZE2), AMDGPU_VCN_CONTEXT_SIZE, 0, indirect);
592
593 /* non-cache window */
594 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
595 VCN, 0, mmUVD_LMI_VCPU_NC0_64BIT_BAR_LOW),
596 lower_32_bits(adev->vcn.inst[inst_idx].fw_shared.gpu_addr), 0, indirect);
597 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
598 VCN, 0, mmUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH),
599 upper_32_bits(adev->vcn.inst[inst_idx].fw_shared.gpu_addr), 0, indirect);
600 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
601 VCN, 0, mmUVD_VCPU_NONCACHE_OFFSET0), 0, 0, indirect);
602 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
603 VCN, 0, mmUVD_VCPU_NONCACHE_SIZE0),
604 AMDGPU_GPU_PAGE_ALIGN(sizeof(struct amdgpu_fw_shared)), 0, indirect);
605
606 /* VCN global tiling registers */
607 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
608 VCN, 0, mmUVD_GFX8_ADDR_CONFIG), adev->gfx.config.gb_addr_config, 0, indirect);
609 }
610
611 /**
612 * vcn_v2_5_disable_clock_gating - disable VCN clock gating
613 *
614 * @adev: amdgpu_device pointer
615 *
616 * Disable clock gating for VCN block
617 */
vcn_v2_5_disable_clock_gating(struct amdgpu_device * adev)618 static void vcn_v2_5_disable_clock_gating(struct amdgpu_device *adev)
619 {
620 uint32_t data;
621 int i;
622
623 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
624 if (adev->vcn.harvest_config & (1 << i))
625 continue;
626 /* UVD disable CGC */
627 data = RREG32_SOC15(VCN, i, mmUVD_CGC_CTRL);
628 if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
629 data |= 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
630 else
631 data &= ~UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK;
632 data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
633 data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
634 WREG32_SOC15(VCN, i, mmUVD_CGC_CTRL, data);
635
636 data = RREG32_SOC15(VCN, i, mmUVD_CGC_GATE);
637 data &= ~(UVD_CGC_GATE__SYS_MASK
638 | UVD_CGC_GATE__UDEC_MASK
639 | UVD_CGC_GATE__MPEG2_MASK
640 | UVD_CGC_GATE__REGS_MASK
641 | UVD_CGC_GATE__RBC_MASK
642 | UVD_CGC_GATE__LMI_MC_MASK
643 | UVD_CGC_GATE__LMI_UMC_MASK
644 | UVD_CGC_GATE__IDCT_MASK
645 | UVD_CGC_GATE__MPRD_MASK
646 | UVD_CGC_GATE__MPC_MASK
647 | UVD_CGC_GATE__LBSI_MASK
648 | UVD_CGC_GATE__LRBBM_MASK
649 | UVD_CGC_GATE__UDEC_RE_MASK
650 | UVD_CGC_GATE__UDEC_CM_MASK
651 | UVD_CGC_GATE__UDEC_IT_MASK
652 | UVD_CGC_GATE__UDEC_DB_MASK
653 | UVD_CGC_GATE__UDEC_MP_MASK
654 | UVD_CGC_GATE__WCB_MASK
655 | UVD_CGC_GATE__VCPU_MASK
656 | UVD_CGC_GATE__MMSCH_MASK);
657
658 WREG32_SOC15(VCN, i, mmUVD_CGC_GATE, data);
659
660 SOC15_WAIT_ON_RREG(VCN, i, mmUVD_CGC_GATE, 0, 0xFFFFFFFF);
661
662 data = RREG32_SOC15(VCN, i, mmUVD_CGC_CTRL);
663 data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK
664 | UVD_CGC_CTRL__UDEC_CM_MODE_MASK
665 | UVD_CGC_CTRL__UDEC_IT_MODE_MASK
666 | UVD_CGC_CTRL__UDEC_DB_MODE_MASK
667 | UVD_CGC_CTRL__UDEC_MP_MODE_MASK
668 | UVD_CGC_CTRL__SYS_MODE_MASK
669 | UVD_CGC_CTRL__UDEC_MODE_MASK
670 | UVD_CGC_CTRL__MPEG2_MODE_MASK
671 | UVD_CGC_CTRL__REGS_MODE_MASK
672 | UVD_CGC_CTRL__RBC_MODE_MASK
673 | UVD_CGC_CTRL__LMI_MC_MODE_MASK
674 | UVD_CGC_CTRL__LMI_UMC_MODE_MASK
675 | UVD_CGC_CTRL__IDCT_MODE_MASK
676 | UVD_CGC_CTRL__MPRD_MODE_MASK
677 | UVD_CGC_CTRL__MPC_MODE_MASK
678 | UVD_CGC_CTRL__LBSI_MODE_MASK
679 | UVD_CGC_CTRL__LRBBM_MODE_MASK
680 | UVD_CGC_CTRL__WCB_MODE_MASK
681 | UVD_CGC_CTRL__VCPU_MODE_MASK
682 | UVD_CGC_CTRL__MMSCH_MODE_MASK);
683 WREG32_SOC15(VCN, i, mmUVD_CGC_CTRL, data);
684
685 /* turn on */
686 data = RREG32_SOC15(VCN, i, mmUVD_SUVD_CGC_GATE);
687 data |= (UVD_SUVD_CGC_GATE__SRE_MASK
688 | UVD_SUVD_CGC_GATE__SIT_MASK
689 | UVD_SUVD_CGC_GATE__SMP_MASK
690 | UVD_SUVD_CGC_GATE__SCM_MASK
691 | UVD_SUVD_CGC_GATE__SDB_MASK
692 | UVD_SUVD_CGC_GATE__SRE_H264_MASK
693 | UVD_SUVD_CGC_GATE__SRE_HEVC_MASK
694 | UVD_SUVD_CGC_GATE__SIT_H264_MASK
695 | UVD_SUVD_CGC_GATE__SIT_HEVC_MASK
696 | UVD_SUVD_CGC_GATE__SCM_H264_MASK
697 | UVD_SUVD_CGC_GATE__SCM_HEVC_MASK
698 | UVD_SUVD_CGC_GATE__SDB_H264_MASK
699 | UVD_SUVD_CGC_GATE__SDB_HEVC_MASK
700 | UVD_SUVD_CGC_GATE__SCLR_MASK
701 | UVD_SUVD_CGC_GATE__UVD_SC_MASK
702 | UVD_SUVD_CGC_GATE__ENT_MASK
703 | UVD_SUVD_CGC_GATE__SIT_HEVC_DEC_MASK
704 | UVD_SUVD_CGC_GATE__SIT_HEVC_ENC_MASK
705 | UVD_SUVD_CGC_GATE__SITE_MASK
706 | UVD_SUVD_CGC_GATE__SRE_VP9_MASK
707 | UVD_SUVD_CGC_GATE__SCM_VP9_MASK
708 | UVD_SUVD_CGC_GATE__SIT_VP9_DEC_MASK
709 | UVD_SUVD_CGC_GATE__SDB_VP9_MASK
710 | UVD_SUVD_CGC_GATE__IME_HEVC_MASK);
711 WREG32_SOC15(VCN, i, mmUVD_SUVD_CGC_GATE, data);
712
713 data = RREG32_SOC15(VCN, i, mmUVD_SUVD_CGC_CTRL);
714 data &= ~(UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
715 | UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
716 | UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
717 | UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
718 | UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
719 | UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
720 | UVD_SUVD_CGC_CTRL__UVD_SC_MODE_MASK
721 | UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
722 | UVD_SUVD_CGC_CTRL__IME_MODE_MASK
723 | UVD_SUVD_CGC_CTRL__SITE_MODE_MASK);
724 WREG32_SOC15(VCN, i, mmUVD_SUVD_CGC_CTRL, data);
725 }
726 }
727
vcn_v2_5_clock_gating_dpg_mode(struct amdgpu_device * adev,uint8_t sram_sel,int inst_idx,uint8_t indirect)728 static void vcn_v2_5_clock_gating_dpg_mode(struct amdgpu_device *adev,
729 uint8_t sram_sel, int inst_idx, uint8_t indirect)
730 {
731 uint32_t reg_data = 0;
732
733 /* enable sw clock gating control */
734 if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
735 reg_data = 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
736 else
737 reg_data = 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
738 reg_data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
739 reg_data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
740 reg_data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK |
741 UVD_CGC_CTRL__UDEC_CM_MODE_MASK |
742 UVD_CGC_CTRL__UDEC_IT_MODE_MASK |
743 UVD_CGC_CTRL__UDEC_DB_MODE_MASK |
744 UVD_CGC_CTRL__UDEC_MP_MODE_MASK |
745 UVD_CGC_CTRL__SYS_MODE_MASK |
746 UVD_CGC_CTRL__UDEC_MODE_MASK |
747 UVD_CGC_CTRL__MPEG2_MODE_MASK |
748 UVD_CGC_CTRL__REGS_MODE_MASK |
749 UVD_CGC_CTRL__RBC_MODE_MASK |
750 UVD_CGC_CTRL__LMI_MC_MODE_MASK |
751 UVD_CGC_CTRL__LMI_UMC_MODE_MASK |
752 UVD_CGC_CTRL__IDCT_MODE_MASK |
753 UVD_CGC_CTRL__MPRD_MODE_MASK |
754 UVD_CGC_CTRL__MPC_MODE_MASK |
755 UVD_CGC_CTRL__LBSI_MODE_MASK |
756 UVD_CGC_CTRL__LRBBM_MODE_MASK |
757 UVD_CGC_CTRL__WCB_MODE_MASK |
758 UVD_CGC_CTRL__VCPU_MODE_MASK |
759 UVD_CGC_CTRL__MMSCH_MODE_MASK);
760 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
761 VCN, 0, mmUVD_CGC_CTRL), reg_data, sram_sel, indirect);
762
763 /* turn off clock gating */
764 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
765 VCN, 0, mmUVD_CGC_GATE), 0, sram_sel, indirect);
766
767 /* turn on SUVD clock gating */
768 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
769 VCN, 0, mmUVD_SUVD_CGC_GATE), 1, sram_sel, indirect);
770
771 /* turn on sw mode in UVD_SUVD_CGC_CTRL */
772 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
773 VCN, 0, mmUVD_SUVD_CGC_CTRL), 0, sram_sel, indirect);
774 }
775
776 /**
777 * vcn_v2_5_enable_clock_gating - enable VCN clock gating
778 *
779 * @adev: amdgpu_device pointer
780 *
781 * Enable clock gating for VCN block
782 */
vcn_v2_5_enable_clock_gating(struct amdgpu_device * adev)783 static void vcn_v2_5_enable_clock_gating(struct amdgpu_device *adev)
784 {
785 uint32_t data = 0;
786 int i;
787
788 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
789 if (adev->vcn.harvest_config & (1 << i))
790 continue;
791 /* enable UVD CGC */
792 data = RREG32_SOC15(VCN, i, mmUVD_CGC_CTRL);
793 if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
794 data |= 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
795 else
796 data |= 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
797 data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
798 data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
799 WREG32_SOC15(VCN, i, mmUVD_CGC_CTRL, data);
800
801 data = RREG32_SOC15(VCN, i, mmUVD_CGC_CTRL);
802 data |= (UVD_CGC_CTRL__UDEC_RE_MODE_MASK
803 | UVD_CGC_CTRL__UDEC_CM_MODE_MASK
804 | UVD_CGC_CTRL__UDEC_IT_MODE_MASK
805 | UVD_CGC_CTRL__UDEC_DB_MODE_MASK
806 | UVD_CGC_CTRL__UDEC_MP_MODE_MASK
807 | UVD_CGC_CTRL__SYS_MODE_MASK
808 | UVD_CGC_CTRL__UDEC_MODE_MASK
809 | UVD_CGC_CTRL__MPEG2_MODE_MASK
810 | UVD_CGC_CTRL__REGS_MODE_MASK
811 | UVD_CGC_CTRL__RBC_MODE_MASK
812 | UVD_CGC_CTRL__LMI_MC_MODE_MASK
813 | UVD_CGC_CTRL__LMI_UMC_MODE_MASK
814 | UVD_CGC_CTRL__IDCT_MODE_MASK
815 | UVD_CGC_CTRL__MPRD_MODE_MASK
816 | UVD_CGC_CTRL__MPC_MODE_MASK
817 | UVD_CGC_CTRL__LBSI_MODE_MASK
818 | UVD_CGC_CTRL__LRBBM_MODE_MASK
819 | UVD_CGC_CTRL__WCB_MODE_MASK
820 | UVD_CGC_CTRL__VCPU_MODE_MASK);
821 WREG32_SOC15(VCN, i, mmUVD_CGC_CTRL, data);
822
823 data = RREG32_SOC15(VCN, i, mmUVD_SUVD_CGC_CTRL);
824 data |= (UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
825 | UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
826 | UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
827 | UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
828 | UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
829 | UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
830 | UVD_SUVD_CGC_CTRL__UVD_SC_MODE_MASK
831 | UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
832 | UVD_SUVD_CGC_CTRL__IME_MODE_MASK
833 | UVD_SUVD_CGC_CTRL__SITE_MODE_MASK);
834 WREG32_SOC15(VCN, i, mmUVD_SUVD_CGC_CTRL, data);
835 }
836 }
837
vcn_v2_6_enable_ras(struct amdgpu_device * adev,int inst_idx,bool indirect)838 static void vcn_v2_6_enable_ras(struct amdgpu_device *adev, int inst_idx,
839 bool indirect)
840 {
841 uint32_t tmp;
842
843 if (amdgpu_ip_version(adev, UVD_HWIP, 0) != IP_VERSION(2, 6, 0))
844 return;
845
846 tmp = VCN_RAS_CNTL__VCPU_VCODEC_REARM_MASK |
847 VCN_RAS_CNTL__VCPU_VCODEC_IH_EN_MASK |
848 VCN_RAS_CNTL__VCPU_VCODEC_PMI_EN_MASK |
849 VCN_RAS_CNTL__VCPU_VCODEC_STALL_EN_MASK;
850 WREG32_SOC15_DPG_MODE(inst_idx,
851 SOC15_DPG_MODE_OFFSET(VCN, 0, mmVCN_RAS_CNTL),
852 tmp, 0, indirect);
853
854 tmp = UVD_VCPU_INT_EN__RASCNTL_VCPU_VCODEC_EN_MASK;
855 WREG32_SOC15_DPG_MODE(inst_idx,
856 SOC15_DPG_MODE_OFFSET(VCN, 0, mmUVD_VCPU_INT_EN),
857 tmp, 0, indirect);
858
859 tmp = UVD_SYS_INT_EN__RASCNTL_VCPU_VCODEC_EN_MASK;
860 WREG32_SOC15_DPG_MODE(inst_idx,
861 SOC15_DPG_MODE_OFFSET(VCN, 0, mmUVD_SYS_INT_EN),
862 tmp, 0, indirect);
863 }
864
vcn_v2_5_start_dpg_mode(struct amdgpu_device * adev,int inst_idx,bool indirect)865 static int vcn_v2_5_start_dpg_mode(struct amdgpu_device *adev, int inst_idx, bool indirect)
866 {
867 volatile struct amdgpu_fw_shared *fw_shared = adev->vcn.inst[inst_idx].fw_shared.cpu_addr;
868 struct amdgpu_ring *ring;
869 uint32_t rb_bufsz, tmp;
870
871 /* disable register anti-hang mechanism */
872 WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS), 1,
873 ~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
874 /* enable dynamic power gating mode */
875 tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_POWER_STATUS);
876 tmp |= UVD_POWER_STATUS__UVD_PG_MODE_MASK;
877 tmp |= UVD_POWER_STATUS__UVD_PG_EN_MASK;
878 WREG32_SOC15(VCN, inst_idx, mmUVD_POWER_STATUS, tmp);
879
880 if (indirect)
881 adev->vcn.inst[inst_idx].dpg_sram_curr_addr = (uint32_t *)adev->vcn.inst[inst_idx].dpg_sram_cpu_addr;
882
883 /* enable clock gating */
884 vcn_v2_5_clock_gating_dpg_mode(adev, 0, inst_idx, indirect);
885
886 /* enable VCPU clock */
887 tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
888 tmp |= UVD_VCPU_CNTL__CLK_EN_MASK;
889 tmp |= UVD_VCPU_CNTL__BLK_RST_MASK;
890 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
891 VCN, 0, mmUVD_VCPU_CNTL), tmp, 0, indirect);
892
893 /* disable master interupt */
894 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
895 VCN, 0, mmUVD_MASTINT_EN), 0, 0, indirect);
896
897 /* setup mmUVD_LMI_CTRL */
898 tmp = (0x8 | UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK |
899 UVD_LMI_CTRL__REQ_MODE_MASK |
900 UVD_LMI_CTRL__CRC_RESET_MASK |
901 UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
902 UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
903 UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK |
904 (8 << UVD_LMI_CTRL__WRITE_CLEAN_TIMER__SHIFT) |
905 0x00100000L);
906 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
907 VCN, 0, mmUVD_LMI_CTRL), tmp, 0, indirect);
908
909 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
910 VCN, 0, mmUVD_MPC_CNTL),
911 0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT, 0, indirect);
912
913 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
914 VCN, 0, mmUVD_MPC_SET_MUXA0),
915 ((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
916 (0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
917 (0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
918 (0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)), 0, indirect);
919
920 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
921 VCN, 0, mmUVD_MPC_SET_MUXB0),
922 ((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
923 (0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
924 (0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
925 (0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)), 0, indirect);
926
927 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
928 VCN, 0, mmUVD_MPC_SET_MUX),
929 ((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
930 (0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
931 (0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)), 0, indirect);
932
933 vcn_v2_5_mc_resume_dpg_mode(adev, inst_idx, indirect);
934
935 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
936 VCN, 0, mmUVD_REG_XX_MASK), 0x10, 0, indirect);
937 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
938 VCN, 0, mmUVD_RBC_XX_IB_REG_CHECK), 0x3, 0, indirect);
939
940 /* enable LMI MC and UMC channels */
941 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
942 VCN, 0, mmUVD_LMI_CTRL2), 0, 0, indirect);
943
944 vcn_v2_6_enable_ras(adev, inst_idx, indirect);
945
946 /* unblock VCPU register access */
947 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
948 VCN, 0, mmUVD_RB_ARB_CTRL), 0, 0, indirect);
949
950 tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
951 tmp |= UVD_VCPU_CNTL__CLK_EN_MASK;
952 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
953 VCN, 0, mmUVD_VCPU_CNTL), tmp, 0, indirect);
954
955 /* enable master interrupt */
956 WREG32_SOC15_DPG_MODE(inst_idx, SOC15_DPG_MODE_OFFSET(
957 VCN, 0, mmUVD_MASTINT_EN),
958 UVD_MASTINT_EN__VCPU_EN_MASK, 0, indirect);
959
960 if (indirect)
961 amdgpu_vcn_psp_update_sram(adev, inst_idx, 0);
962
963 ring = &adev->vcn.inst[inst_idx].ring_dec;
964 /* force RBC into idle state */
965 rb_bufsz = order_base_2(ring->ring_size);
966 tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz);
967 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
968 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
969 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
970 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
971 WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_CNTL, tmp);
972
973 /* Stall DPG before WPTR/RPTR reset */
974 WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
975 UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK,
976 ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
977 fw_shared->multi_queue.decode_queue_mode |= FW_QUEUE_RING_RESET;
978
979 /* set the write pointer delay */
980 WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_WPTR_CNTL, 0);
981
982 /* set the wb address */
983 WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_RPTR_ADDR,
984 (upper_32_bits(ring->gpu_addr) >> 2));
985
986 /* program the RB_BASE for ring buffer */
987 WREG32_SOC15(VCN, inst_idx, mmUVD_LMI_RBC_RB_64BIT_BAR_LOW,
988 lower_32_bits(ring->gpu_addr));
989 WREG32_SOC15(VCN, inst_idx, mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH,
990 upper_32_bits(ring->gpu_addr));
991
992 /* Initialize the ring buffer's read and write pointers */
993 WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_RPTR, 0);
994
995 WREG32_SOC15(VCN, inst_idx, mmUVD_SCRATCH2, 0);
996
997 ring->wptr = RREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_RPTR);
998 WREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_WPTR,
999 lower_32_bits(ring->wptr));
1000
1001 fw_shared->multi_queue.decode_queue_mode &= ~FW_QUEUE_RING_RESET;
1002 /* Unstall DPG */
1003 WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
1004 0, ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
1005
1006 return 0;
1007 }
1008
vcn_v2_5_start(struct amdgpu_device * adev)1009 static int vcn_v2_5_start(struct amdgpu_device *adev)
1010 {
1011 struct amdgpu_ring *ring;
1012 uint32_t rb_bufsz, tmp;
1013 int i, j, k, r;
1014
1015 if (adev->pm.dpm_enabled)
1016 amdgpu_dpm_enable_uvd(adev, true);
1017
1018 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1019 if (adev->vcn.harvest_config & (1 << i))
1020 continue;
1021 if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
1022 r = vcn_v2_5_start_dpg_mode(adev, i, adev->vcn.indirect_sram);
1023 continue;
1024 }
1025
1026 /* disable register anti-hang mechanism */
1027 WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_POWER_STATUS), 0,
1028 ~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1029
1030 /* set uvd status busy */
1031 tmp = RREG32_SOC15(VCN, i, mmUVD_STATUS) | UVD_STATUS__UVD_BUSY;
1032 WREG32_SOC15(VCN, i, mmUVD_STATUS, tmp);
1033 }
1034
1035 if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)
1036 return 0;
1037
1038 /*SW clock gating */
1039 vcn_v2_5_disable_clock_gating(adev);
1040
1041 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1042 if (adev->vcn.harvest_config & (1 << i))
1043 continue;
1044 /* enable VCPU clock */
1045 WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL),
1046 UVD_VCPU_CNTL__CLK_EN_MASK, ~UVD_VCPU_CNTL__CLK_EN_MASK);
1047
1048 /* disable master interrupt */
1049 WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_MASTINT_EN), 0,
1050 ~UVD_MASTINT_EN__VCPU_EN_MASK);
1051
1052 /* setup mmUVD_LMI_CTRL */
1053 tmp = RREG32_SOC15(VCN, i, mmUVD_LMI_CTRL);
1054 tmp &= ~0xff;
1055 WREG32_SOC15(VCN, i, mmUVD_LMI_CTRL, tmp | 0x8|
1056 UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK |
1057 UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
1058 UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
1059 UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK);
1060
1061 /* setup mmUVD_MPC_CNTL */
1062 tmp = RREG32_SOC15(VCN, i, mmUVD_MPC_CNTL);
1063 tmp &= ~UVD_MPC_CNTL__REPLACEMENT_MODE_MASK;
1064 tmp |= 0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT;
1065 WREG32_SOC15(VCN, i, mmUVD_MPC_CNTL, tmp);
1066
1067 /* setup UVD_MPC_SET_MUXA0 */
1068 WREG32_SOC15(VCN, i, mmUVD_MPC_SET_MUXA0,
1069 ((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
1070 (0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
1071 (0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
1072 (0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)));
1073
1074 /* setup UVD_MPC_SET_MUXB0 */
1075 WREG32_SOC15(VCN, i, mmUVD_MPC_SET_MUXB0,
1076 ((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
1077 (0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
1078 (0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
1079 (0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)));
1080
1081 /* setup mmUVD_MPC_SET_MUX */
1082 WREG32_SOC15(VCN, i, mmUVD_MPC_SET_MUX,
1083 ((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
1084 (0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
1085 (0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)));
1086 }
1087
1088 vcn_v2_5_mc_resume(adev);
1089
1090 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1091 volatile struct amdgpu_fw_shared *fw_shared = adev->vcn.inst[i].fw_shared.cpu_addr;
1092 if (adev->vcn.harvest_config & (1 << i))
1093 continue;
1094 /* VCN global tiling registers */
1095 WREG32_SOC15(VCN, i, mmUVD_GFX8_ADDR_CONFIG,
1096 adev->gfx.config.gb_addr_config);
1097 WREG32_SOC15(VCN, i, mmUVD_GFX8_ADDR_CONFIG,
1098 adev->gfx.config.gb_addr_config);
1099
1100 /* enable LMI MC and UMC channels */
1101 WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_LMI_CTRL2), 0,
1102 ~UVD_LMI_CTRL2__STALL_ARB_UMC_MASK);
1103
1104 /* unblock VCPU register access */
1105 WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_RB_ARB_CTRL), 0,
1106 ~UVD_RB_ARB_CTRL__VCPU_DIS_MASK);
1107
1108 WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL), 0,
1109 ~UVD_VCPU_CNTL__BLK_RST_MASK);
1110
1111 for (k = 0; k < 10; ++k) {
1112 uint32_t status;
1113
1114 for (j = 0; j < 100; ++j) {
1115 status = RREG32_SOC15(VCN, i, mmUVD_STATUS);
1116 if (status & 2)
1117 break;
1118 if (amdgpu_emu_mode == 1)
1119 msleep(500);
1120 else
1121 mdelay(10);
1122 }
1123 r = 0;
1124 if (status & 2)
1125 break;
1126
1127 DRM_ERROR("VCN decode not responding, trying to reset the VCPU!!!\n");
1128 WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL),
1129 UVD_VCPU_CNTL__BLK_RST_MASK,
1130 ~UVD_VCPU_CNTL__BLK_RST_MASK);
1131 mdelay(10);
1132 WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL), 0,
1133 ~UVD_VCPU_CNTL__BLK_RST_MASK);
1134
1135 mdelay(10);
1136 r = -1;
1137 }
1138
1139 if (r) {
1140 DRM_ERROR("VCN decode not responding, giving up!!!\n");
1141 return r;
1142 }
1143
1144 /* enable master interrupt */
1145 WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_MASTINT_EN),
1146 UVD_MASTINT_EN__VCPU_EN_MASK,
1147 ~UVD_MASTINT_EN__VCPU_EN_MASK);
1148
1149 /* clear the busy bit of VCN_STATUS */
1150 WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_STATUS), 0,
1151 ~(2 << UVD_STATUS__VCPU_REPORT__SHIFT));
1152
1153 WREG32_SOC15(VCN, i, mmUVD_LMI_RBC_RB_VMID, 0);
1154
1155 ring = &adev->vcn.inst[i].ring_dec;
1156 /* force RBC into idle state */
1157 rb_bufsz = order_base_2(ring->ring_size);
1158 tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz);
1159 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
1160 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
1161 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
1162 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
1163 WREG32_SOC15(VCN, i, mmUVD_RBC_RB_CNTL, tmp);
1164
1165 fw_shared->multi_queue.decode_queue_mode |= FW_QUEUE_RING_RESET;
1166 /* program the RB_BASE for ring buffer */
1167 WREG32_SOC15(VCN, i, mmUVD_LMI_RBC_RB_64BIT_BAR_LOW,
1168 lower_32_bits(ring->gpu_addr));
1169 WREG32_SOC15(VCN, i, mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH,
1170 upper_32_bits(ring->gpu_addr));
1171
1172 /* Initialize the ring buffer's read and write pointers */
1173 WREG32_SOC15(VCN, i, mmUVD_RBC_RB_RPTR, 0);
1174
1175 ring->wptr = RREG32_SOC15(VCN, i, mmUVD_RBC_RB_RPTR);
1176 WREG32_SOC15(VCN, i, mmUVD_RBC_RB_WPTR,
1177 lower_32_bits(ring->wptr));
1178 fw_shared->multi_queue.decode_queue_mode &= ~FW_QUEUE_RING_RESET;
1179
1180 fw_shared->multi_queue.encode_generalpurpose_queue_mode |= FW_QUEUE_RING_RESET;
1181 ring = &adev->vcn.inst[i].ring_enc[0];
1182 WREG32_SOC15(VCN, i, mmUVD_RB_RPTR, lower_32_bits(ring->wptr));
1183 WREG32_SOC15(VCN, i, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
1184 WREG32_SOC15(VCN, i, mmUVD_RB_BASE_LO, ring->gpu_addr);
1185 WREG32_SOC15(VCN, i, mmUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
1186 WREG32_SOC15(VCN, i, mmUVD_RB_SIZE, ring->ring_size / 4);
1187 fw_shared->multi_queue.encode_generalpurpose_queue_mode &= ~FW_QUEUE_RING_RESET;
1188
1189 fw_shared->multi_queue.encode_lowlatency_queue_mode |= FW_QUEUE_RING_RESET;
1190 ring = &adev->vcn.inst[i].ring_enc[1];
1191 WREG32_SOC15(VCN, i, mmUVD_RB_RPTR2, lower_32_bits(ring->wptr));
1192 WREG32_SOC15(VCN, i, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
1193 WREG32_SOC15(VCN, i, mmUVD_RB_BASE_LO2, ring->gpu_addr);
1194 WREG32_SOC15(VCN, i, mmUVD_RB_BASE_HI2, upper_32_bits(ring->gpu_addr));
1195 WREG32_SOC15(VCN, i, mmUVD_RB_SIZE2, ring->ring_size / 4);
1196 fw_shared->multi_queue.encode_lowlatency_queue_mode &= ~FW_QUEUE_RING_RESET;
1197 }
1198
1199 return 0;
1200 }
1201
vcn_v2_5_mmsch_start(struct amdgpu_device * adev,struct amdgpu_mm_table * table)1202 static int vcn_v2_5_mmsch_start(struct amdgpu_device *adev,
1203 struct amdgpu_mm_table *table)
1204 {
1205 uint32_t data = 0, loop = 0, size = 0;
1206 uint64_t addr = table->gpu_addr;
1207 struct mmsch_v1_1_init_header *header = NULL;
1208
1209 header = (struct mmsch_v1_1_init_header *)table->cpu_addr;
1210 size = header->total_size;
1211
1212 /*
1213 * 1, write to vce_mmsch_vf_ctx_addr_lo/hi register with GPU mc addr of
1214 * memory descriptor location
1215 */
1216 WREG32_SOC15(VCN, 0, mmMMSCH_VF_CTX_ADDR_LO, lower_32_bits(addr));
1217 WREG32_SOC15(VCN, 0, mmMMSCH_VF_CTX_ADDR_HI, upper_32_bits(addr));
1218
1219 /* 2, update vmid of descriptor */
1220 data = RREG32_SOC15(VCN, 0, mmMMSCH_VF_VMID);
1221 data &= ~MMSCH_VF_VMID__VF_CTX_VMID_MASK;
1222 /* use domain0 for MM scheduler */
1223 data |= (0 << MMSCH_VF_VMID__VF_CTX_VMID__SHIFT);
1224 WREG32_SOC15(VCN, 0, mmMMSCH_VF_VMID, data);
1225
1226 /* 3, notify mmsch about the size of this descriptor */
1227 WREG32_SOC15(VCN, 0, mmMMSCH_VF_CTX_SIZE, size);
1228
1229 /* 4, set resp to zero */
1230 WREG32_SOC15(VCN, 0, mmMMSCH_VF_MAILBOX_RESP, 0);
1231
1232 /*
1233 * 5, kick off the initialization and wait until
1234 * VCE_MMSCH_VF_MAILBOX_RESP becomes non-zero
1235 */
1236 WREG32_SOC15(VCN, 0, mmMMSCH_VF_MAILBOX_HOST, 0x10000001);
1237
1238 data = RREG32_SOC15(VCN, 0, mmMMSCH_VF_MAILBOX_RESP);
1239 loop = 10;
1240 while ((data & 0x10000002) != 0x10000002) {
1241 udelay(100);
1242 data = RREG32_SOC15(VCN, 0, mmMMSCH_VF_MAILBOX_RESP);
1243 loop--;
1244 if (!loop)
1245 break;
1246 }
1247
1248 if (!loop) {
1249 dev_err(adev->dev,
1250 "failed to init MMSCH, mmMMSCH_VF_MAILBOX_RESP = %x\n",
1251 data);
1252 return -EBUSY;
1253 }
1254
1255 return 0;
1256 }
1257
vcn_v2_5_sriov_start(struct amdgpu_device * adev)1258 static int vcn_v2_5_sriov_start(struct amdgpu_device *adev)
1259 {
1260 struct amdgpu_ring *ring;
1261 uint32_t offset, size, tmp, i, rb_bufsz;
1262 uint32_t table_size = 0;
1263 struct mmsch_v1_0_cmd_direct_write direct_wt = { { 0 } };
1264 struct mmsch_v1_0_cmd_direct_read_modify_write direct_rd_mod_wt = { { 0 } };
1265 struct mmsch_v1_0_cmd_end end = { { 0 } };
1266 uint32_t *init_table = adev->virt.mm_table.cpu_addr;
1267 struct mmsch_v1_1_init_header *header = (struct mmsch_v1_1_init_header *)init_table;
1268
1269 direct_wt.cmd_header.command_type = MMSCH_COMMAND__DIRECT_REG_WRITE;
1270 direct_rd_mod_wt.cmd_header.command_type = MMSCH_COMMAND__DIRECT_REG_READ_MODIFY_WRITE;
1271 end.cmd_header.command_type = MMSCH_COMMAND__END;
1272
1273 header->version = MMSCH_VERSION;
1274 header->total_size = sizeof(struct mmsch_v1_1_init_header) >> 2;
1275 init_table += header->total_size;
1276
1277 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1278 header->eng[i].table_offset = header->total_size;
1279 header->eng[i].init_status = 0;
1280 header->eng[i].table_size = 0;
1281
1282 table_size = 0;
1283
1284 MMSCH_V1_0_INSERT_DIRECT_RD_MOD_WT(
1285 SOC15_REG_OFFSET(VCN, i, mmUVD_STATUS),
1286 ~UVD_STATUS__UVD_BUSY, UVD_STATUS__UVD_BUSY);
1287
1288 size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw[i]->size + 4);
1289 /* mc resume*/
1290 if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
1291 MMSCH_V1_0_INSERT_DIRECT_WT(
1292 SOC15_REG_OFFSET(VCN, i,
1293 mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
1294 adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_lo);
1295 MMSCH_V1_0_INSERT_DIRECT_WT(
1296 SOC15_REG_OFFSET(VCN, i,
1297 mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
1298 adev->firmware.ucode[AMDGPU_UCODE_ID_VCN + i].tmr_mc_addr_hi);
1299 offset = 0;
1300 MMSCH_V1_0_INSERT_DIRECT_WT(
1301 SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_OFFSET0), 0);
1302 } else {
1303 MMSCH_V1_0_INSERT_DIRECT_WT(
1304 SOC15_REG_OFFSET(VCN, i,
1305 mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
1306 lower_32_bits(adev->vcn.inst[i].gpu_addr));
1307 MMSCH_V1_0_INSERT_DIRECT_WT(
1308 SOC15_REG_OFFSET(VCN, i,
1309 mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
1310 upper_32_bits(adev->vcn.inst[i].gpu_addr));
1311 offset = size;
1312 MMSCH_V1_0_INSERT_DIRECT_WT(
1313 SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_OFFSET0),
1314 AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
1315 }
1316
1317 MMSCH_V1_0_INSERT_DIRECT_WT(
1318 SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_SIZE0),
1319 size);
1320 MMSCH_V1_0_INSERT_DIRECT_WT(
1321 SOC15_REG_OFFSET(VCN, i,
1322 mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
1323 lower_32_bits(adev->vcn.inst[i].gpu_addr + offset));
1324 MMSCH_V1_0_INSERT_DIRECT_WT(
1325 SOC15_REG_OFFSET(VCN, i,
1326 mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
1327 upper_32_bits(adev->vcn.inst[i].gpu_addr + offset));
1328 MMSCH_V1_0_INSERT_DIRECT_WT(
1329 SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_OFFSET1),
1330 0);
1331 MMSCH_V1_0_INSERT_DIRECT_WT(
1332 SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_SIZE1),
1333 AMDGPU_VCN_STACK_SIZE);
1334 MMSCH_V1_0_INSERT_DIRECT_WT(
1335 SOC15_REG_OFFSET(VCN, i,
1336 mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
1337 lower_32_bits(adev->vcn.inst[i].gpu_addr + offset +
1338 AMDGPU_VCN_STACK_SIZE));
1339 MMSCH_V1_0_INSERT_DIRECT_WT(
1340 SOC15_REG_OFFSET(VCN, i,
1341 mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
1342 upper_32_bits(adev->vcn.inst[i].gpu_addr + offset +
1343 AMDGPU_VCN_STACK_SIZE));
1344 MMSCH_V1_0_INSERT_DIRECT_WT(
1345 SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_OFFSET2),
1346 0);
1347 MMSCH_V1_0_INSERT_DIRECT_WT(
1348 SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CACHE_SIZE2),
1349 AMDGPU_VCN_CONTEXT_SIZE);
1350
1351 ring = &adev->vcn.inst[i].ring_enc[0];
1352 ring->wptr = 0;
1353
1354 MMSCH_V1_0_INSERT_DIRECT_WT(
1355 SOC15_REG_OFFSET(VCN, i, mmUVD_RB_BASE_LO),
1356 lower_32_bits(ring->gpu_addr));
1357 MMSCH_V1_0_INSERT_DIRECT_WT(
1358 SOC15_REG_OFFSET(VCN, i, mmUVD_RB_BASE_HI),
1359 upper_32_bits(ring->gpu_addr));
1360 MMSCH_V1_0_INSERT_DIRECT_WT(
1361 SOC15_REG_OFFSET(VCN, i, mmUVD_RB_SIZE),
1362 ring->ring_size / 4);
1363
1364 ring = &adev->vcn.inst[i].ring_dec;
1365 ring->wptr = 0;
1366 MMSCH_V1_0_INSERT_DIRECT_WT(
1367 SOC15_REG_OFFSET(VCN, i,
1368 mmUVD_LMI_RBC_RB_64BIT_BAR_LOW),
1369 lower_32_bits(ring->gpu_addr));
1370 MMSCH_V1_0_INSERT_DIRECT_WT(
1371 SOC15_REG_OFFSET(VCN, i,
1372 mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH),
1373 upper_32_bits(ring->gpu_addr));
1374
1375 /* force RBC into idle state */
1376 rb_bufsz = order_base_2(ring->ring_size);
1377 tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz);
1378 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
1379 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
1380 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
1381 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
1382 MMSCH_V1_0_INSERT_DIRECT_WT(
1383 SOC15_REG_OFFSET(VCN, i, mmUVD_RBC_RB_CNTL), tmp);
1384
1385 /* add end packet */
1386 memcpy((void *)init_table, &end, sizeof(struct mmsch_v1_0_cmd_end));
1387 table_size += sizeof(struct mmsch_v1_0_cmd_end) / 4;
1388 init_table += sizeof(struct mmsch_v1_0_cmd_end) / 4;
1389
1390 /* refine header */
1391 header->eng[i].table_size = table_size;
1392 header->total_size += table_size;
1393 }
1394
1395 return vcn_v2_5_mmsch_start(adev, &adev->virt.mm_table);
1396 }
1397
vcn_v2_5_stop_dpg_mode(struct amdgpu_device * adev,int inst_idx)1398 static int vcn_v2_5_stop_dpg_mode(struct amdgpu_device *adev, int inst_idx)
1399 {
1400 uint32_t tmp;
1401
1402 /* Wait for power status to be 1 */
1403 SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS, 1,
1404 UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1405
1406 /* wait for read ptr to be equal to write ptr */
1407 tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR);
1408 SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_RB_RPTR, tmp, 0xFFFFFFFF);
1409
1410 tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR2);
1411 SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_RB_RPTR2, tmp, 0xFFFFFFFF);
1412
1413 tmp = RREG32_SOC15(VCN, inst_idx, mmUVD_RBC_RB_WPTR) & 0x7FFFFFFF;
1414 SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_RBC_RB_RPTR, tmp, 0xFFFFFFFF);
1415
1416 SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS, 1,
1417 UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1418
1419 /* disable dynamic power gating mode */
1420 WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS), 0,
1421 ~UVD_POWER_STATUS__UVD_PG_MODE_MASK);
1422
1423 return 0;
1424 }
1425
vcn_v2_5_stop(struct amdgpu_device * adev)1426 static int vcn_v2_5_stop(struct amdgpu_device *adev)
1427 {
1428 uint32_t tmp;
1429 int i, r = 0;
1430
1431 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1432 if (adev->vcn.harvest_config & (1 << i))
1433 continue;
1434 if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
1435 r = vcn_v2_5_stop_dpg_mode(adev, i);
1436 continue;
1437 }
1438
1439 /* wait for vcn idle */
1440 r = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_STATUS, UVD_STATUS__IDLE, 0x7);
1441 if (r)
1442 return r;
1443
1444 tmp = UVD_LMI_STATUS__VCPU_LMI_WRITE_CLEAN_MASK |
1445 UVD_LMI_STATUS__READ_CLEAN_MASK |
1446 UVD_LMI_STATUS__WRITE_CLEAN_MASK |
1447 UVD_LMI_STATUS__WRITE_CLEAN_RAW_MASK;
1448 r = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_LMI_STATUS, tmp, tmp);
1449 if (r)
1450 return r;
1451
1452 /* block LMI UMC channel */
1453 tmp = RREG32_SOC15(VCN, i, mmUVD_LMI_CTRL2);
1454 tmp |= UVD_LMI_CTRL2__STALL_ARB_UMC_MASK;
1455 WREG32_SOC15(VCN, i, mmUVD_LMI_CTRL2, tmp);
1456
1457 tmp = UVD_LMI_STATUS__UMC_READ_CLEAN_RAW_MASK|
1458 UVD_LMI_STATUS__UMC_WRITE_CLEAN_RAW_MASK;
1459 r = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_LMI_STATUS, tmp, tmp);
1460 if (r)
1461 return r;
1462
1463 /* block VCPU register access */
1464 WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_RB_ARB_CTRL),
1465 UVD_RB_ARB_CTRL__VCPU_DIS_MASK,
1466 ~UVD_RB_ARB_CTRL__VCPU_DIS_MASK);
1467
1468 /* reset VCPU */
1469 WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL),
1470 UVD_VCPU_CNTL__BLK_RST_MASK,
1471 ~UVD_VCPU_CNTL__BLK_RST_MASK);
1472
1473 /* disable VCPU clock */
1474 WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_VCPU_CNTL), 0,
1475 ~(UVD_VCPU_CNTL__CLK_EN_MASK));
1476
1477 /* clear status */
1478 WREG32_SOC15(VCN, i, mmUVD_STATUS, 0);
1479
1480 vcn_v2_5_enable_clock_gating(adev);
1481
1482 /* enable register anti-hang mechanism */
1483 WREG32_P(SOC15_REG_OFFSET(VCN, i, mmUVD_POWER_STATUS),
1484 UVD_POWER_STATUS__UVD_POWER_STATUS_MASK,
1485 ~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1486 }
1487
1488 if (adev->pm.dpm_enabled)
1489 amdgpu_dpm_enable_uvd(adev, false);
1490
1491 return 0;
1492 }
1493
vcn_v2_5_pause_dpg_mode(struct amdgpu_device * adev,int inst_idx,struct dpg_pause_state * new_state)1494 static int vcn_v2_5_pause_dpg_mode(struct amdgpu_device *adev,
1495 int inst_idx, struct dpg_pause_state *new_state)
1496 {
1497 struct amdgpu_ring *ring;
1498 uint32_t reg_data = 0;
1499 int ret_code = 0;
1500
1501 /* pause/unpause if state is changed */
1502 if (adev->vcn.inst[inst_idx].pause_state.fw_based != new_state->fw_based) {
1503 DRM_DEBUG("dpg pause state changed %d -> %d",
1504 adev->vcn.inst[inst_idx].pause_state.fw_based, new_state->fw_based);
1505 reg_data = RREG32_SOC15(VCN, inst_idx, mmUVD_DPG_PAUSE) &
1506 (~UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
1507
1508 if (new_state->fw_based == VCN_DPG_STATE__PAUSE) {
1509 ret_code = SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS, 0x1,
1510 UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1511
1512 if (!ret_code) {
1513 volatile struct amdgpu_fw_shared *fw_shared = adev->vcn.inst[inst_idx].fw_shared.cpu_addr;
1514
1515 /* pause DPG */
1516 reg_data |= UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1517 WREG32_SOC15(VCN, inst_idx, mmUVD_DPG_PAUSE, reg_data);
1518
1519 /* wait for ACK */
1520 SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_DPG_PAUSE,
1521 UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK,
1522 UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
1523
1524 /* Stall DPG before WPTR/RPTR reset */
1525 WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
1526 UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK,
1527 ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
1528
1529 /* Restore */
1530 fw_shared->multi_queue.encode_generalpurpose_queue_mode |= FW_QUEUE_RING_RESET;
1531 ring = &adev->vcn.inst[inst_idx].ring_enc[0];
1532 ring->wptr = 0;
1533 WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_LO, ring->gpu_addr);
1534 WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
1535 WREG32_SOC15(VCN, inst_idx, mmUVD_RB_SIZE, ring->ring_size / 4);
1536 WREG32_SOC15(VCN, inst_idx, mmUVD_RB_RPTR, lower_32_bits(ring->wptr));
1537 WREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
1538 fw_shared->multi_queue.encode_generalpurpose_queue_mode &= ~FW_QUEUE_RING_RESET;
1539
1540 fw_shared->multi_queue.encode_lowlatency_queue_mode |= FW_QUEUE_RING_RESET;
1541 ring = &adev->vcn.inst[inst_idx].ring_enc[1];
1542 ring->wptr = 0;
1543 WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_LO2, ring->gpu_addr);
1544 WREG32_SOC15(VCN, inst_idx, mmUVD_RB_BASE_HI2, upper_32_bits(ring->gpu_addr));
1545 WREG32_SOC15(VCN, inst_idx, mmUVD_RB_SIZE2, ring->ring_size / 4);
1546 WREG32_SOC15(VCN, inst_idx, mmUVD_RB_RPTR2, lower_32_bits(ring->wptr));
1547 WREG32_SOC15(VCN, inst_idx, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
1548 fw_shared->multi_queue.encode_lowlatency_queue_mode &= ~FW_QUEUE_RING_RESET;
1549
1550 /* Unstall DPG */
1551 WREG32_P(SOC15_REG_OFFSET(VCN, inst_idx, mmUVD_POWER_STATUS),
1552 0, ~UVD_POWER_STATUS__STALL_DPG_POWER_UP_MASK);
1553
1554 SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS,
1555 UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON, UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1556 }
1557 } else {
1558 reg_data &= ~UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1559 WREG32_SOC15(VCN, inst_idx, mmUVD_DPG_PAUSE, reg_data);
1560 SOC15_WAIT_ON_RREG(VCN, inst_idx, mmUVD_POWER_STATUS, 0x1,
1561 UVD_POWER_STATUS__UVD_POWER_STATUS_MASK);
1562 }
1563 adev->vcn.inst[inst_idx].pause_state.fw_based = new_state->fw_based;
1564 }
1565
1566 return 0;
1567 }
1568
1569 /**
1570 * vcn_v2_5_dec_ring_get_rptr - get read pointer
1571 *
1572 * @ring: amdgpu_ring pointer
1573 *
1574 * Returns the current hardware read pointer
1575 */
vcn_v2_5_dec_ring_get_rptr(struct amdgpu_ring * ring)1576 static uint64_t vcn_v2_5_dec_ring_get_rptr(struct amdgpu_ring *ring)
1577 {
1578 struct amdgpu_device *adev = ring->adev;
1579
1580 return RREG32_SOC15(VCN, ring->me, mmUVD_RBC_RB_RPTR);
1581 }
1582
1583 /**
1584 * vcn_v2_5_dec_ring_get_wptr - get write pointer
1585 *
1586 * @ring: amdgpu_ring pointer
1587 *
1588 * Returns the current hardware write pointer
1589 */
vcn_v2_5_dec_ring_get_wptr(struct amdgpu_ring * ring)1590 static uint64_t vcn_v2_5_dec_ring_get_wptr(struct amdgpu_ring *ring)
1591 {
1592 struct amdgpu_device *adev = ring->adev;
1593
1594 if (ring->use_doorbell)
1595 return *ring->wptr_cpu_addr;
1596 else
1597 return RREG32_SOC15(VCN, ring->me, mmUVD_RBC_RB_WPTR);
1598 }
1599
1600 /**
1601 * vcn_v2_5_dec_ring_set_wptr - set write pointer
1602 *
1603 * @ring: amdgpu_ring pointer
1604 *
1605 * Commits the write pointer to the hardware
1606 */
vcn_v2_5_dec_ring_set_wptr(struct amdgpu_ring * ring)1607 static void vcn_v2_5_dec_ring_set_wptr(struct amdgpu_ring *ring)
1608 {
1609 struct amdgpu_device *adev = ring->adev;
1610
1611 if (ring->use_doorbell) {
1612 *ring->wptr_cpu_addr = lower_32_bits(ring->wptr);
1613 WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1614 } else {
1615 WREG32_SOC15(VCN, ring->me, mmUVD_RBC_RB_WPTR, lower_32_bits(ring->wptr));
1616 }
1617 }
1618
1619 static const struct amdgpu_ring_funcs vcn_v2_5_dec_ring_vm_funcs = {
1620 .type = AMDGPU_RING_TYPE_VCN_DEC,
1621 .align_mask = 0xf,
1622 .secure_submission_supported = true,
1623 .get_rptr = vcn_v2_5_dec_ring_get_rptr,
1624 .get_wptr = vcn_v2_5_dec_ring_get_wptr,
1625 .set_wptr = vcn_v2_5_dec_ring_set_wptr,
1626 .emit_frame_size =
1627 SOC15_FLUSH_GPU_TLB_NUM_WREG * 6 +
1628 SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 8 +
1629 8 + /* vcn_v2_0_dec_ring_emit_vm_flush */
1630 14 + 14 + /* vcn_v2_0_dec_ring_emit_fence x2 vm fence */
1631 6,
1632 .emit_ib_size = 8, /* vcn_v2_0_dec_ring_emit_ib */
1633 .emit_ib = vcn_v2_0_dec_ring_emit_ib,
1634 .emit_fence = vcn_v2_0_dec_ring_emit_fence,
1635 .emit_vm_flush = vcn_v2_0_dec_ring_emit_vm_flush,
1636 .test_ring = vcn_v2_0_dec_ring_test_ring,
1637 .test_ib = amdgpu_vcn_dec_ring_test_ib,
1638 .insert_nop = vcn_v2_0_dec_ring_insert_nop,
1639 .insert_start = vcn_v2_0_dec_ring_insert_start,
1640 .insert_end = vcn_v2_0_dec_ring_insert_end,
1641 .pad_ib = amdgpu_ring_generic_pad_ib,
1642 .begin_use = amdgpu_vcn_ring_begin_use,
1643 .end_use = amdgpu_vcn_ring_end_use,
1644 .emit_wreg = vcn_v2_0_dec_ring_emit_wreg,
1645 .emit_reg_wait = vcn_v2_0_dec_ring_emit_reg_wait,
1646 .emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
1647 };
1648
1649 /**
1650 * vcn_v2_5_enc_ring_get_rptr - get enc read pointer
1651 *
1652 * @ring: amdgpu_ring pointer
1653 *
1654 * Returns the current hardware enc read pointer
1655 */
vcn_v2_5_enc_ring_get_rptr(struct amdgpu_ring * ring)1656 static uint64_t vcn_v2_5_enc_ring_get_rptr(struct amdgpu_ring *ring)
1657 {
1658 struct amdgpu_device *adev = ring->adev;
1659
1660 if (ring == &adev->vcn.inst[ring->me].ring_enc[0])
1661 return RREG32_SOC15(VCN, ring->me, mmUVD_RB_RPTR);
1662 else
1663 return RREG32_SOC15(VCN, ring->me, mmUVD_RB_RPTR2);
1664 }
1665
1666 /**
1667 * vcn_v2_5_enc_ring_get_wptr - get enc write pointer
1668 *
1669 * @ring: amdgpu_ring pointer
1670 *
1671 * Returns the current hardware enc write pointer
1672 */
vcn_v2_5_enc_ring_get_wptr(struct amdgpu_ring * ring)1673 static uint64_t vcn_v2_5_enc_ring_get_wptr(struct amdgpu_ring *ring)
1674 {
1675 struct amdgpu_device *adev = ring->adev;
1676
1677 if (ring == &adev->vcn.inst[ring->me].ring_enc[0]) {
1678 if (ring->use_doorbell)
1679 return *ring->wptr_cpu_addr;
1680 else
1681 return RREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR);
1682 } else {
1683 if (ring->use_doorbell)
1684 return *ring->wptr_cpu_addr;
1685 else
1686 return RREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR2);
1687 }
1688 }
1689
1690 /**
1691 * vcn_v2_5_enc_ring_set_wptr - set enc write pointer
1692 *
1693 * @ring: amdgpu_ring pointer
1694 *
1695 * Commits the enc write pointer to the hardware
1696 */
vcn_v2_5_enc_ring_set_wptr(struct amdgpu_ring * ring)1697 static void vcn_v2_5_enc_ring_set_wptr(struct amdgpu_ring *ring)
1698 {
1699 struct amdgpu_device *adev = ring->adev;
1700
1701 if (ring == &adev->vcn.inst[ring->me].ring_enc[0]) {
1702 if (ring->use_doorbell) {
1703 *ring->wptr_cpu_addr = lower_32_bits(ring->wptr);
1704 WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1705 } else {
1706 WREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
1707 }
1708 } else {
1709 if (ring->use_doorbell) {
1710 *ring->wptr_cpu_addr = lower_32_bits(ring->wptr);
1711 WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1712 } else {
1713 WREG32_SOC15(VCN, ring->me, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
1714 }
1715 }
1716 }
1717
1718 static const struct amdgpu_ring_funcs vcn_v2_5_enc_ring_vm_funcs = {
1719 .type = AMDGPU_RING_TYPE_VCN_ENC,
1720 .align_mask = 0x3f,
1721 .nop = VCN_ENC_CMD_NO_OP,
1722 .get_rptr = vcn_v2_5_enc_ring_get_rptr,
1723 .get_wptr = vcn_v2_5_enc_ring_get_wptr,
1724 .set_wptr = vcn_v2_5_enc_ring_set_wptr,
1725 .emit_frame_size =
1726 SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
1727 SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 4 +
1728 4 + /* vcn_v2_0_enc_ring_emit_vm_flush */
1729 5 + 5 + /* vcn_v2_0_enc_ring_emit_fence x2 vm fence */
1730 1, /* vcn_v2_0_enc_ring_insert_end */
1731 .emit_ib_size = 5, /* vcn_v2_0_enc_ring_emit_ib */
1732 .emit_ib = vcn_v2_0_enc_ring_emit_ib,
1733 .emit_fence = vcn_v2_0_enc_ring_emit_fence,
1734 .emit_vm_flush = vcn_v2_0_enc_ring_emit_vm_flush,
1735 .test_ring = amdgpu_vcn_enc_ring_test_ring,
1736 .test_ib = amdgpu_vcn_enc_ring_test_ib,
1737 .insert_nop = amdgpu_ring_insert_nop,
1738 .insert_end = vcn_v2_0_enc_ring_insert_end,
1739 .pad_ib = amdgpu_ring_generic_pad_ib,
1740 .begin_use = amdgpu_vcn_ring_begin_use,
1741 .end_use = amdgpu_vcn_ring_end_use,
1742 .emit_wreg = vcn_v2_0_enc_ring_emit_wreg,
1743 .emit_reg_wait = vcn_v2_0_enc_ring_emit_reg_wait,
1744 .emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
1745 };
1746
vcn_v2_5_set_dec_ring_funcs(struct amdgpu_device * adev)1747 static void vcn_v2_5_set_dec_ring_funcs(struct amdgpu_device *adev)
1748 {
1749 int i;
1750
1751 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1752 if (adev->vcn.harvest_config & (1 << i))
1753 continue;
1754 adev->vcn.inst[i].ring_dec.funcs = &vcn_v2_5_dec_ring_vm_funcs;
1755 adev->vcn.inst[i].ring_dec.me = i;
1756 }
1757 }
1758
vcn_v2_5_set_enc_ring_funcs(struct amdgpu_device * adev)1759 static void vcn_v2_5_set_enc_ring_funcs(struct amdgpu_device *adev)
1760 {
1761 int i, j;
1762
1763 for (j = 0; j < adev->vcn.num_vcn_inst; ++j) {
1764 if (adev->vcn.harvest_config & (1 << j))
1765 continue;
1766 for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
1767 adev->vcn.inst[j].ring_enc[i].funcs = &vcn_v2_5_enc_ring_vm_funcs;
1768 adev->vcn.inst[j].ring_enc[i].me = j;
1769 }
1770 }
1771 }
1772
vcn_v2_5_is_idle(void * handle)1773 static bool vcn_v2_5_is_idle(void *handle)
1774 {
1775 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1776 int i, ret = 1;
1777
1778 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1779 if (adev->vcn.harvest_config & (1 << i))
1780 continue;
1781 ret &= (RREG32_SOC15(VCN, i, mmUVD_STATUS) == UVD_STATUS__IDLE);
1782 }
1783
1784 return ret;
1785 }
1786
vcn_v2_5_wait_for_idle(struct amdgpu_ip_block * ip_block)1787 static int vcn_v2_5_wait_for_idle(struct amdgpu_ip_block *ip_block)
1788 {
1789 struct amdgpu_device *adev = ip_block->adev;
1790 int i, ret = 0;
1791
1792 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1793 if (adev->vcn.harvest_config & (1 << i))
1794 continue;
1795 ret = SOC15_WAIT_ON_RREG(VCN, i, mmUVD_STATUS, UVD_STATUS__IDLE,
1796 UVD_STATUS__IDLE);
1797 if (ret)
1798 return ret;
1799 }
1800
1801 return ret;
1802 }
1803
vcn_v2_5_set_clockgating_state(void * handle,enum amd_clockgating_state state)1804 static int vcn_v2_5_set_clockgating_state(void *handle,
1805 enum amd_clockgating_state state)
1806 {
1807 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1808 bool enable = (state == AMD_CG_STATE_GATE);
1809
1810 if (amdgpu_sriov_vf(adev))
1811 return 0;
1812
1813 if (enable) {
1814 if (!vcn_v2_5_is_idle(handle))
1815 return -EBUSY;
1816 vcn_v2_5_enable_clock_gating(adev);
1817 } else {
1818 vcn_v2_5_disable_clock_gating(adev);
1819 }
1820
1821 return 0;
1822 }
1823
vcn_v2_5_set_powergating_state(void * handle,enum amd_powergating_state state)1824 static int vcn_v2_5_set_powergating_state(void *handle,
1825 enum amd_powergating_state state)
1826 {
1827 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1828 int ret;
1829
1830 if (amdgpu_sriov_vf(adev))
1831 return 0;
1832
1833 if(state == adev->vcn.cur_state)
1834 return 0;
1835
1836 if (state == AMD_PG_STATE_GATE)
1837 ret = vcn_v2_5_stop(adev);
1838 else
1839 ret = vcn_v2_5_start(adev);
1840
1841 if(!ret)
1842 adev->vcn.cur_state = state;
1843
1844 return ret;
1845 }
1846
vcn_v2_5_set_interrupt_state(struct amdgpu_device * adev,struct amdgpu_irq_src * source,unsigned type,enum amdgpu_interrupt_state state)1847 static int vcn_v2_5_set_interrupt_state(struct amdgpu_device *adev,
1848 struct amdgpu_irq_src *source,
1849 unsigned type,
1850 enum amdgpu_interrupt_state state)
1851 {
1852 return 0;
1853 }
1854
vcn_v2_6_set_ras_interrupt_state(struct amdgpu_device * adev,struct amdgpu_irq_src * source,unsigned int type,enum amdgpu_interrupt_state state)1855 static int vcn_v2_6_set_ras_interrupt_state(struct amdgpu_device *adev,
1856 struct amdgpu_irq_src *source,
1857 unsigned int type,
1858 enum amdgpu_interrupt_state state)
1859 {
1860 return 0;
1861 }
1862
vcn_v2_5_process_interrupt(struct amdgpu_device * adev,struct amdgpu_irq_src * source,struct amdgpu_iv_entry * entry)1863 static int vcn_v2_5_process_interrupt(struct amdgpu_device *adev,
1864 struct amdgpu_irq_src *source,
1865 struct amdgpu_iv_entry *entry)
1866 {
1867 uint32_t ip_instance;
1868
1869 switch (entry->client_id) {
1870 case SOC15_IH_CLIENTID_VCN:
1871 ip_instance = 0;
1872 break;
1873 case SOC15_IH_CLIENTID_VCN1:
1874 ip_instance = 1;
1875 break;
1876 default:
1877 DRM_ERROR("Unhandled client id: %d\n", entry->client_id);
1878 return 0;
1879 }
1880
1881 DRM_DEBUG("IH: VCN TRAP\n");
1882
1883 switch (entry->src_id) {
1884 case VCN_2_0__SRCID__UVD_SYSTEM_MESSAGE_INTERRUPT:
1885 amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_dec);
1886 break;
1887 case VCN_2_0__SRCID__UVD_ENC_GENERAL_PURPOSE:
1888 amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_enc[0]);
1889 break;
1890 case VCN_2_0__SRCID__UVD_ENC_LOW_LATENCY:
1891 amdgpu_fence_process(&adev->vcn.inst[ip_instance].ring_enc[1]);
1892 break;
1893 default:
1894 DRM_ERROR("Unhandled interrupt: %d %d\n",
1895 entry->src_id, entry->src_data[0]);
1896 break;
1897 }
1898
1899 return 0;
1900 }
1901
1902 static const struct amdgpu_irq_src_funcs vcn_v2_5_irq_funcs = {
1903 .set = vcn_v2_5_set_interrupt_state,
1904 .process = vcn_v2_5_process_interrupt,
1905 };
1906
1907 static const struct amdgpu_irq_src_funcs vcn_v2_6_ras_irq_funcs = {
1908 .set = vcn_v2_6_set_ras_interrupt_state,
1909 .process = amdgpu_vcn_process_poison_irq,
1910 };
1911
vcn_v2_5_set_irq_funcs(struct amdgpu_device * adev)1912 static void vcn_v2_5_set_irq_funcs(struct amdgpu_device *adev)
1913 {
1914 int i;
1915
1916 for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
1917 if (adev->vcn.harvest_config & (1 << i))
1918 continue;
1919 adev->vcn.inst[i].irq.num_types = adev->vcn.num_enc_rings + 1;
1920 adev->vcn.inst[i].irq.funcs = &vcn_v2_5_irq_funcs;
1921
1922 adev->vcn.inst[i].ras_poison_irq.num_types = adev->vcn.num_enc_rings + 1;
1923 adev->vcn.inst[i].ras_poison_irq.funcs = &vcn_v2_6_ras_irq_funcs;
1924 }
1925 }
1926
vcn_v2_5_print_ip_state(struct amdgpu_ip_block * ip_block,struct drm_printer * p)1927 static void vcn_v2_5_print_ip_state(struct amdgpu_ip_block *ip_block, struct drm_printer *p)
1928 {
1929 struct amdgpu_device *adev = ip_block->adev;
1930 int i, j;
1931 uint32_t reg_count = ARRAY_SIZE(vcn_reg_list_2_5);
1932 uint32_t inst_off, is_powered;
1933
1934 if (!adev->vcn.ip_dump)
1935 return;
1936
1937 drm_printf(p, "num_instances:%d\n", adev->vcn.num_vcn_inst);
1938 for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
1939 if (adev->vcn.harvest_config & (1 << i)) {
1940 drm_printf(p, "\nHarvested Instance:VCN%d Skipping dump\n", i);
1941 continue;
1942 }
1943
1944 inst_off = i * reg_count;
1945 is_powered = (adev->vcn.ip_dump[inst_off] &
1946 UVD_POWER_STATUS__UVD_POWER_STATUS_MASK) != 1;
1947
1948 if (is_powered) {
1949 drm_printf(p, "\nActive Instance:VCN%d\n", i);
1950 for (j = 0; j < reg_count; j++)
1951 drm_printf(p, "%-50s \t 0x%08x\n", vcn_reg_list_2_5[j].reg_name,
1952 adev->vcn.ip_dump[inst_off + j]);
1953 } else {
1954 drm_printf(p, "\nInactive Instance:VCN%d\n", i);
1955 }
1956 }
1957 }
1958
vcn_v2_5_dump_ip_state(struct amdgpu_ip_block * ip_block)1959 static void vcn_v2_5_dump_ip_state(struct amdgpu_ip_block *ip_block)
1960 {
1961 struct amdgpu_device *adev = ip_block->adev;
1962 int i, j;
1963 bool is_powered;
1964 uint32_t inst_off;
1965 uint32_t reg_count = ARRAY_SIZE(vcn_reg_list_2_5);
1966
1967 if (!adev->vcn.ip_dump)
1968 return;
1969
1970 for (i = 0; i < adev->vcn.num_vcn_inst; i++) {
1971 if (adev->vcn.harvest_config & (1 << i))
1972 continue;
1973
1974 inst_off = i * reg_count;
1975 /* mmUVD_POWER_STATUS is always readable and is first element of the array */
1976 adev->vcn.ip_dump[inst_off] = RREG32_SOC15(VCN, i, mmUVD_POWER_STATUS);
1977 is_powered = (adev->vcn.ip_dump[inst_off] &
1978 UVD_POWER_STATUS__UVD_POWER_STATUS_MASK) != 1;
1979
1980 if (is_powered)
1981 for (j = 1; j < reg_count; j++)
1982 adev->vcn.ip_dump[inst_off + j] =
1983 RREG32(SOC15_REG_ENTRY_OFFSET_INST(vcn_reg_list_2_5[j], i));
1984 }
1985 }
1986
1987 static const struct amd_ip_funcs vcn_v2_5_ip_funcs = {
1988 .name = "vcn_v2_5",
1989 .early_init = vcn_v2_5_early_init,
1990 .sw_init = vcn_v2_5_sw_init,
1991 .sw_fini = vcn_v2_5_sw_fini,
1992 .hw_init = vcn_v2_5_hw_init,
1993 .hw_fini = vcn_v2_5_hw_fini,
1994 .suspend = vcn_v2_5_suspend,
1995 .resume = vcn_v2_5_resume,
1996 .is_idle = vcn_v2_5_is_idle,
1997 .wait_for_idle = vcn_v2_5_wait_for_idle,
1998 .set_clockgating_state = vcn_v2_5_set_clockgating_state,
1999 .set_powergating_state = vcn_v2_5_set_powergating_state,
2000 .dump_ip_state = vcn_v2_5_dump_ip_state,
2001 .print_ip_state = vcn_v2_5_print_ip_state,
2002 };
2003
2004 static const struct amd_ip_funcs vcn_v2_6_ip_funcs = {
2005 .name = "vcn_v2_6",
2006 .early_init = vcn_v2_5_early_init,
2007 .sw_init = vcn_v2_5_sw_init,
2008 .sw_fini = vcn_v2_5_sw_fini,
2009 .hw_init = vcn_v2_5_hw_init,
2010 .hw_fini = vcn_v2_5_hw_fini,
2011 .suspend = vcn_v2_5_suspend,
2012 .resume = vcn_v2_5_resume,
2013 .is_idle = vcn_v2_5_is_idle,
2014 .wait_for_idle = vcn_v2_5_wait_for_idle,
2015 .set_clockgating_state = vcn_v2_5_set_clockgating_state,
2016 .set_powergating_state = vcn_v2_5_set_powergating_state,
2017 .dump_ip_state = vcn_v2_5_dump_ip_state,
2018 .print_ip_state = vcn_v2_5_print_ip_state,
2019 };
2020
2021 const struct amdgpu_ip_block_version vcn_v2_5_ip_block =
2022 {
2023 .type = AMD_IP_BLOCK_TYPE_VCN,
2024 .major = 2,
2025 .minor = 5,
2026 .rev = 0,
2027 .funcs = &vcn_v2_5_ip_funcs,
2028 };
2029
2030 const struct amdgpu_ip_block_version vcn_v2_6_ip_block =
2031 {
2032 .type = AMD_IP_BLOCK_TYPE_VCN,
2033 .major = 2,
2034 .minor = 6,
2035 .rev = 0,
2036 .funcs = &vcn_v2_6_ip_funcs,
2037 };
2038
vcn_v2_6_query_poison_by_instance(struct amdgpu_device * adev,uint32_t instance,uint32_t sub_block)2039 static uint32_t vcn_v2_6_query_poison_by_instance(struct amdgpu_device *adev,
2040 uint32_t instance, uint32_t sub_block)
2041 {
2042 uint32_t poison_stat = 0, reg_value = 0;
2043
2044 switch (sub_block) {
2045 case AMDGPU_VCN_V2_6_VCPU_VCODEC:
2046 reg_value = RREG32_SOC15(VCN, instance, mmUVD_RAS_VCPU_VCODEC_STATUS);
2047 poison_stat = REG_GET_FIELD(reg_value, UVD_RAS_VCPU_VCODEC_STATUS, POISONED_PF);
2048 break;
2049 default:
2050 break;
2051 }
2052
2053 if (poison_stat)
2054 dev_info(adev->dev, "Poison detected in VCN%d, sub_block%d\n",
2055 instance, sub_block);
2056
2057 return poison_stat;
2058 }
2059
vcn_v2_6_query_poison_status(struct amdgpu_device * adev)2060 static bool vcn_v2_6_query_poison_status(struct amdgpu_device *adev)
2061 {
2062 uint32_t inst, sub;
2063 uint32_t poison_stat = 0;
2064
2065 for (inst = 0; inst < adev->vcn.num_vcn_inst; inst++)
2066 for (sub = 0; sub < AMDGPU_VCN_V2_6_MAX_SUB_BLOCK; sub++)
2067 poison_stat +=
2068 vcn_v2_6_query_poison_by_instance(adev, inst, sub);
2069
2070 return !!poison_stat;
2071 }
2072
2073 const struct amdgpu_ras_block_hw_ops vcn_v2_6_ras_hw_ops = {
2074 .query_poison_status = vcn_v2_6_query_poison_status,
2075 };
2076
2077 static struct amdgpu_vcn_ras vcn_v2_6_ras = {
2078 .ras_block = {
2079 .hw_ops = &vcn_v2_6_ras_hw_ops,
2080 .ras_late_init = amdgpu_vcn_ras_late_init,
2081 },
2082 };
2083
vcn_v2_5_set_ras_funcs(struct amdgpu_device * adev)2084 static void vcn_v2_5_set_ras_funcs(struct amdgpu_device *adev)
2085 {
2086 switch (amdgpu_ip_version(adev, VCN_HWIP, 0)) {
2087 case IP_VERSION(2, 6, 0):
2088 adev->vcn.ras = &vcn_v2_6_ras;
2089 break;
2090 default:
2091 break;
2092 }
2093 }
2094