treewide: remove redundant IS_ERR() before error code check
[linux/fpc-iii.git] / drivers / gpu / drm / amd / amdgpu / vcn_v2_0.c
blobf4db8af6536bf9e9eb45a66789d42686f4944256
1 /*
2 * Copyright 2018 Advanced Micro Devices, Inc.
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:
11 * The above copyright notice and this permission notice shall be included in
12 * all copies or substantial portions of the Software.
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.
24 #include <linux/firmware.h>
26 #include "amdgpu.h"
27 #include "amdgpu_vcn.h"
28 #include "soc15.h"
29 #include "soc15d.h"
30 #include "amdgpu_pm.h"
31 #include "amdgpu_psp.h"
33 #include "vcn/vcn_2_0_0_offset.h"
34 #include "vcn/vcn_2_0_0_sh_mask.h"
35 #include "ivsrcid/vcn/irqsrcs_vcn_2_0.h"
37 #define mmUVD_CONTEXT_ID_INTERNAL_OFFSET 0x1fd
38 #define mmUVD_GPCOM_VCPU_CMD_INTERNAL_OFFSET 0x503
39 #define mmUVD_GPCOM_VCPU_DATA0_INTERNAL_OFFSET 0x504
40 #define mmUVD_GPCOM_VCPU_DATA1_INTERNAL_OFFSET 0x505
41 #define mmUVD_NO_OP_INTERNAL_OFFSET 0x53f
42 #define mmUVD_GP_SCRATCH8_INTERNAL_OFFSET 0x54a
43 #define mmUVD_SCRATCH9_INTERNAL_OFFSET 0xc01d
45 #define mmUVD_LMI_RBC_IB_VMID_INTERNAL_OFFSET 0x1e1
46 #define mmUVD_LMI_RBC_IB_64BIT_BAR_HIGH_INTERNAL_OFFSET 0x5a6
47 #define mmUVD_LMI_RBC_IB_64BIT_BAR_LOW_INTERNAL_OFFSET 0x5a7
48 #define mmUVD_RBC_IB_SIZE_INTERNAL_OFFSET 0x1e2
50 static void vcn_v2_0_set_dec_ring_funcs(struct amdgpu_device *adev);
51 static void vcn_v2_0_set_enc_ring_funcs(struct amdgpu_device *adev);
52 static void vcn_v2_0_set_irq_funcs(struct amdgpu_device *adev);
53 static int vcn_v2_0_set_powergating_state(void *handle,
54 enum amd_powergating_state state);
55 static int vcn_v2_0_pause_dpg_mode(struct amdgpu_device *adev,
56 int inst_idx, struct dpg_pause_state *new_state);
58 /**
59 * vcn_v2_0_early_init - set function pointers
61 * @handle: amdgpu_device pointer
63 * Set ring and irq function pointers
65 static int vcn_v2_0_early_init(void *handle)
67 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
69 adev->vcn.num_vcn_inst = 1;
70 adev->vcn.num_enc_rings = 2;
72 vcn_v2_0_set_dec_ring_funcs(adev);
73 vcn_v2_0_set_enc_ring_funcs(adev);
74 vcn_v2_0_set_irq_funcs(adev);
76 return 0;
79 /**
80 * vcn_v2_0_sw_init - sw init for VCN block
82 * @handle: amdgpu_device pointer
84 * Load firmware and sw initialization
86 static int vcn_v2_0_sw_init(void *handle)
88 struct amdgpu_ring *ring;
89 int i, r;
90 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
92 /* VCN DEC TRAP */
93 r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_VCN,
94 VCN_2_0__SRCID__UVD_SYSTEM_MESSAGE_INTERRUPT,
95 &adev->vcn.inst->irq);
96 if (r)
97 return r;
99 /* VCN ENC TRAP */
100 for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
101 r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_VCN,
102 i + VCN_2_0__SRCID__UVD_ENC_GENERAL_PURPOSE,
103 &adev->vcn.inst->irq);
104 if (r)
105 return r;
108 r = amdgpu_vcn_sw_init(adev);
109 if (r)
110 return r;
112 if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
113 const struct common_firmware_header *hdr;
114 hdr = (const struct common_firmware_header *)adev->vcn.fw->data;
115 adev->firmware.ucode[AMDGPU_UCODE_ID_VCN].ucode_id = AMDGPU_UCODE_ID_VCN;
116 adev->firmware.ucode[AMDGPU_UCODE_ID_VCN].fw = adev->vcn.fw;
117 adev->firmware.fw_size +=
118 ALIGN(le32_to_cpu(hdr->ucode_size_bytes), PAGE_SIZE);
119 DRM_INFO("PSP loading VCN firmware\n");
122 r = amdgpu_vcn_resume(adev);
123 if (r)
124 return r;
126 ring = &adev->vcn.inst->ring_dec;
128 ring->use_doorbell = true;
129 ring->doorbell_index = adev->doorbell_index.vcn.vcn_ring0_1 << 1;
131 sprintf(ring->name, "vcn_dec");
132 r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst->irq, 0);
133 if (r)
134 return r;
136 adev->vcn.internal.context_id = mmUVD_CONTEXT_ID_INTERNAL_OFFSET;
137 adev->vcn.internal.ib_vmid = mmUVD_LMI_RBC_IB_VMID_INTERNAL_OFFSET;
138 adev->vcn.internal.ib_bar_low = mmUVD_LMI_RBC_IB_64BIT_BAR_LOW_INTERNAL_OFFSET;
139 adev->vcn.internal.ib_bar_high = mmUVD_LMI_RBC_IB_64BIT_BAR_HIGH_INTERNAL_OFFSET;
140 adev->vcn.internal.ib_size = mmUVD_RBC_IB_SIZE_INTERNAL_OFFSET;
141 adev->vcn.internal.gp_scratch8 = mmUVD_GP_SCRATCH8_INTERNAL_OFFSET;
143 adev->vcn.internal.scratch9 = mmUVD_SCRATCH9_INTERNAL_OFFSET;
144 adev->vcn.inst->external.scratch9 = SOC15_REG_OFFSET(UVD, 0, mmUVD_SCRATCH9);
145 adev->vcn.internal.data0 = mmUVD_GPCOM_VCPU_DATA0_INTERNAL_OFFSET;
146 adev->vcn.inst->external.data0 = SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_DATA0);
147 adev->vcn.internal.data1 = mmUVD_GPCOM_VCPU_DATA1_INTERNAL_OFFSET;
148 adev->vcn.inst->external.data1 = SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_DATA1);
149 adev->vcn.internal.cmd = mmUVD_GPCOM_VCPU_CMD_INTERNAL_OFFSET;
150 adev->vcn.inst->external.cmd = SOC15_REG_OFFSET(UVD, 0, mmUVD_GPCOM_VCPU_CMD);
151 adev->vcn.internal.nop = mmUVD_NO_OP_INTERNAL_OFFSET;
152 adev->vcn.inst->external.nop = SOC15_REG_OFFSET(UVD, 0, mmUVD_NO_OP);
154 for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
155 ring = &adev->vcn.inst->ring_enc[i];
156 ring->use_doorbell = true;
157 ring->doorbell_index = (adev->doorbell_index.vcn.vcn_ring0_1 << 1) + 2 + i;
158 sprintf(ring->name, "vcn_enc%d", i);
159 r = amdgpu_ring_init(adev, ring, 512, &adev->vcn.inst->irq, 0);
160 if (r)
161 return r;
164 adev->vcn.pause_dpg_mode = vcn_v2_0_pause_dpg_mode;
166 return 0;
170 * vcn_v2_0_sw_fini - sw fini for VCN block
172 * @handle: amdgpu_device pointer
174 * VCN suspend and free up sw allocation
176 static int vcn_v2_0_sw_fini(void *handle)
178 int r;
179 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
181 r = amdgpu_vcn_suspend(adev);
182 if (r)
183 return r;
185 r = amdgpu_vcn_sw_fini(adev);
187 return r;
191 * vcn_v2_0_hw_init - start and test VCN block
193 * @handle: amdgpu_device pointer
195 * Initialize the hardware, boot up the VCPU and do some testing
197 static int vcn_v2_0_hw_init(void *handle)
199 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
200 struct amdgpu_ring *ring = &adev->vcn.inst->ring_dec;
201 int i, r;
203 adev->nbio.funcs->vcn_doorbell_range(adev, ring->use_doorbell,
204 ring->doorbell_index, 0);
206 r = amdgpu_ring_test_helper(ring);
207 if (r)
208 goto done;
210 for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
211 ring = &adev->vcn.inst->ring_enc[i];
212 r = amdgpu_ring_test_helper(ring);
213 if (r)
214 goto done;
217 done:
218 if (!r)
219 DRM_INFO("VCN decode and encode initialized successfully(under %s).\n",
220 (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)?"DPG Mode":"SPG Mode");
222 return r;
226 * vcn_v2_0_hw_fini - stop the hardware block
228 * @handle: amdgpu_device pointer
230 * Stop the VCN block, mark ring as not ready any more
232 static int vcn_v2_0_hw_fini(void *handle)
234 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
235 struct amdgpu_ring *ring = &adev->vcn.inst->ring_dec;
236 int i;
238 if ((adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) ||
239 (adev->vcn.cur_state != AMD_PG_STATE_GATE &&
240 RREG32_SOC15(VCN, 0, mmUVD_STATUS)))
241 vcn_v2_0_set_powergating_state(adev, AMD_PG_STATE_GATE);
243 ring->sched.ready = false;
245 for (i = 0; i < adev->vcn.num_enc_rings; ++i) {
246 ring = &adev->vcn.inst->ring_enc[i];
247 ring->sched.ready = false;
250 return 0;
254 * vcn_v2_0_suspend - suspend VCN block
256 * @handle: amdgpu_device pointer
258 * HW fini and suspend VCN block
260 static int vcn_v2_0_suspend(void *handle)
262 int r;
263 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
265 r = vcn_v2_0_hw_fini(adev);
266 if (r)
267 return r;
269 r = amdgpu_vcn_suspend(adev);
271 return r;
275 * vcn_v2_0_resume - resume VCN block
277 * @handle: amdgpu_device pointer
279 * Resume firmware and hw init VCN block
281 static int vcn_v2_0_resume(void *handle)
283 int r;
284 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
286 r = amdgpu_vcn_resume(adev);
287 if (r)
288 return r;
290 r = vcn_v2_0_hw_init(adev);
292 return r;
296 * vcn_v2_0_mc_resume - memory controller programming
298 * @adev: amdgpu_device pointer
300 * Let the VCN memory controller know it's offsets
302 static void vcn_v2_0_mc_resume(struct amdgpu_device *adev)
304 uint32_t size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4);
305 uint32_t offset;
307 /* cache window 0: fw */
308 if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
309 WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
310 (adev->firmware.ucode[AMDGPU_UCODE_ID_VCN].tmr_mc_addr_lo));
311 WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
312 (adev->firmware.ucode[AMDGPU_UCODE_ID_VCN].tmr_mc_addr_hi));
313 WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_OFFSET0, 0);
314 offset = 0;
315 } else {
316 WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW,
317 lower_32_bits(adev->vcn.inst->gpu_addr));
318 WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH,
319 upper_32_bits(adev->vcn.inst->gpu_addr));
320 offset = size;
321 WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_OFFSET0,
322 AMDGPU_UVD_FIRMWARE_OFFSET >> 3);
325 WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_SIZE0, size);
327 /* cache window 1: stack */
328 WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW,
329 lower_32_bits(adev->vcn.inst->gpu_addr + offset));
330 WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH,
331 upper_32_bits(adev->vcn.inst->gpu_addr + offset));
332 WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_OFFSET1, 0);
333 WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_SIZE1, AMDGPU_VCN_STACK_SIZE);
335 /* cache window 2: context */
336 WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW,
337 lower_32_bits(adev->vcn.inst->gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
338 WREG32_SOC15(UVD, 0, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH,
339 upper_32_bits(adev->vcn.inst->gpu_addr + offset + AMDGPU_VCN_STACK_SIZE));
340 WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_OFFSET2, 0);
341 WREG32_SOC15(UVD, 0, mmUVD_VCPU_CACHE_SIZE2, AMDGPU_VCN_CONTEXT_SIZE);
343 WREG32_SOC15(UVD, 0, mmUVD_GFX10_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
346 static void vcn_v2_0_mc_resume_dpg_mode(struct amdgpu_device *adev, bool indirect)
348 uint32_t size = AMDGPU_GPU_PAGE_ALIGN(adev->vcn.fw->size + 4);
349 uint32_t offset;
351 /* cache window 0: fw */
352 if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
353 if (!indirect) {
354 WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
355 UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
356 (adev->firmware.ucode[AMDGPU_UCODE_ID_VCN].tmr_mc_addr_lo), 0, indirect);
357 WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
358 UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
359 (adev->firmware.ucode[AMDGPU_UCODE_ID_VCN].tmr_mc_addr_hi), 0, indirect);
360 WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
361 UVD, 0, mmUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
362 } else {
363 WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
364 UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW), 0, 0, indirect);
365 WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
366 UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH), 0, 0, indirect);
367 WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
368 UVD, 0, mmUVD_VCPU_CACHE_OFFSET0), 0, 0, indirect);
370 offset = 0;
371 } else {
372 WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
373 UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_LOW),
374 lower_32_bits(adev->vcn.inst->gpu_addr), 0, indirect);
375 WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
376 UVD, 0, mmUVD_LMI_VCPU_CACHE_64BIT_BAR_HIGH),
377 upper_32_bits(adev->vcn.inst->gpu_addr), 0, indirect);
378 offset = size;
379 WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
380 UVD, 0, mmUVD_VCPU_CACHE_OFFSET0),
381 AMDGPU_UVD_FIRMWARE_OFFSET >> 3, 0, indirect);
384 if (!indirect)
385 WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
386 UVD, 0, mmUVD_VCPU_CACHE_SIZE0), size, 0, indirect);
387 else
388 WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
389 UVD, 0, mmUVD_VCPU_CACHE_SIZE0), 0, 0, indirect);
391 /* cache window 1: stack */
392 if (!indirect) {
393 WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
394 UVD, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW),
395 lower_32_bits(adev->vcn.inst->gpu_addr + offset), 0, indirect);
396 WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
397 UVD, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH),
398 upper_32_bits(adev->vcn.inst->gpu_addr + offset), 0, indirect);
399 WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
400 UVD, 0, mmUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
401 } else {
402 WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
403 UVD, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_LOW), 0, 0, indirect);
404 WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
405 UVD, 0, mmUVD_LMI_VCPU_CACHE1_64BIT_BAR_HIGH), 0, 0, indirect);
406 WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
407 UVD, 0, mmUVD_VCPU_CACHE_OFFSET1), 0, 0, indirect);
409 WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
410 UVD, 0, mmUVD_VCPU_CACHE_SIZE1), AMDGPU_VCN_STACK_SIZE, 0, indirect);
412 /* cache window 2: context */
413 WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
414 UVD, 0, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_LOW),
415 lower_32_bits(adev->vcn.inst->gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
416 WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
417 UVD, 0, mmUVD_LMI_VCPU_CACHE2_64BIT_BAR_HIGH),
418 upper_32_bits(adev->vcn.inst->gpu_addr + offset + AMDGPU_VCN_STACK_SIZE), 0, indirect);
419 WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
420 UVD, 0, mmUVD_VCPU_CACHE_OFFSET2), 0, 0, indirect);
421 WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
422 UVD, 0, mmUVD_VCPU_CACHE_SIZE2), AMDGPU_VCN_CONTEXT_SIZE, 0, indirect);
424 /* non-cache window */
425 WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
426 UVD, 0, mmUVD_LMI_VCPU_NC0_64BIT_BAR_LOW), 0, 0, indirect);
427 WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
428 UVD, 0, mmUVD_LMI_VCPU_NC0_64BIT_BAR_HIGH), 0, 0, indirect);
429 WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
430 UVD, 0, mmUVD_VCPU_NONCACHE_OFFSET0), 0, 0, indirect);
431 WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
432 UVD, 0, mmUVD_VCPU_NONCACHE_SIZE0), 0, 0, indirect);
434 /* VCN global tiling registers */
435 WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
436 UVD, 0, mmUVD_GFX10_ADDR_CONFIG), adev->gfx.config.gb_addr_config, 0, indirect);
440 * vcn_v2_0_disable_clock_gating - disable VCN clock gating
442 * @adev: amdgpu_device pointer
443 * @sw: enable SW clock gating
445 * Disable clock gating for VCN block
447 static void vcn_v2_0_disable_clock_gating(struct amdgpu_device *adev)
449 uint32_t data;
451 /* UVD disable CGC */
452 data = RREG32_SOC15(VCN, 0, mmUVD_CGC_CTRL);
453 if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
454 data |= 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
455 else
456 data &= ~UVD_CGC_CTRL__DYN_CLOCK_MODE_MASK;
457 data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
458 data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
459 WREG32_SOC15(VCN, 0, mmUVD_CGC_CTRL, data);
461 data = RREG32_SOC15(VCN, 0, mmUVD_CGC_GATE);
462 data &= ~(UVD_CGC_GATE__SYS_MASK
463 | UVD_CGC_GATE__UDEC_MASK
464 | UVD_CGC_GATE__MPEG2_MASK
465 | UVD_CGC_GATE__REGS_MASK
466 | UVD_CGC_GATE__RBC_MASK
467 | UVD_CGC_GATE__LMI_MC_MASK
468 | UVD_CGC_GATE__LMI_UMC_MASK
469 | UVD_CGC_GATE__IDCT_MASK
470 | UVD_CGC_GATE__MPRD_MASK
471 | UVD_CGC_GATE__MPC_MASK
472 | UVD_CGC_GATE__LBSI_MASK
473 | UVD_CGC_GATE__LRBBM_MASK
474 | UVD_CGC_GATE__UDEC_RE_MASK
475 | UVD_CGC_GATE__UDEC_CM_MASK
476 | UVD_CGC_GATE__UDEC_IT_MASK
477 | UVD_CGC_GATE__UDEC_DB_MASK
478 | UVD_CGC_GATE__UDEC_MP_MASK
479 | UVD_CGC_GATE__WCB_MASK
480 | UVD_CGC_GATE__VCPU_MASK
481 | UVD_CGC_GATE__SCPU_MASK);
482 WREG32_SOC15(VCN, 0, mmUVD_CGC_GATE, data);
484 data = RREG32_SOC15(VCN, 0, mmUVD_CGC_CTRL);
485 data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK
486 | UVD_CGC_CTRL__UDEC_CM_MODE_MASK
487 | UVD_CGC_CTRL__UDEC_IT_MODE_MASK
488 | UVD_CGC_CTRL__UDEC_DB_MODE_MASK
489 | UVD_CGC_CTRL__UDEC_MP_MODE_MASK
490 | UVD_CGC_CTRL__SYS_MODE_MASK
491 | UVD_CGC_CTRL__UDEC_MODE_MASK
492 | UVD_CGC_CTRL__MPEG2_MODE_MASK
493 | UVD_CGC_CTRL__REGS_MODE_MASK
494 | UVD_CGC_CTRL__RBC_MODE_MASK
495 | UVD_CGC_CTRL__LMI_MC_MODE_MASK
496 | UVD_CGC_CTRL__LMI_UMC_MODE_MASK
497 | UVD_CGC_CTRL__IDCT_MODE_MASK
498 | UVD_CGC_CTRL__MPRD_MODE_MASK
499 | UVD_CGC_CTRL__MPC_MODE_MASK
500 | UVD_CGC_CTRL__LBSI_MODE_MASK
501 | UVD_CGC_CTRL__LRBBM_MODE_MASK
502 | UVD_CGC_CTRL__WCB_MODE_MASK
503 | UVD_CGC_CTRL__VCPU_MODE_MASK
504 | UVD_CGC_CTRL__SCPU_MODE_MASK);
505 WREG32_SOC15(VCN, 0, mmUVD_CGC_CTRL, data);
507 /* turn on */
508 data = RREG32_SOC15(VCN, 0, mmUVD_SUVD_CGC_GATE);
509 data |= (UVD_SUVD_CGC_GATE__SRE_MASK
510 | UVD_SUVD_CGC_GATE__SIT_MASK
511 | UVD_SUVD_CGC_GATE__SMP_MASK
512 | UVD_SUVD_CGC_GATE__SCM_MASK
513 | UVD_SUVD_CGC_GATE__SDB_MASK
514 | UVD_SUVD_CGC_GATE__SRE_H264_MASK
515 | UVD_SUVD_CGC_GATE__SRE_HEVC_MASK
516 | UVD_SUVD_CGC_GATE__SIT_H264_MASK
517 | UVD_SUVD_CGC_GATE__SIT_HEVC_MASK
518 | UVD_SUVD_CGC_GATE__SCM_H264_MASK
519 | UVD_SUVD_CGC_GATE__SCM_HEVC_MASK
520 | UVD_SUVD_CGC_GATE__SDB_H264_MASK
521 | UVD_SUVD_CGC_GATE__SDB_HEVC_MASK
522 | UVD_SUVD_CGC_GATE__SCLR_MASK
523 | UVD_SUVD_CGC_GATE__UVD_SC_MASK
524 | UVD_SUVD_CGC_GATE__ENT_MASK
525 | UVD_SUVD_CGC_GATE__SIT_HEVC_DEC_MASK
526 | UVD_SUVD_CGC_GATE__SIT_HEVC_ENC_MASK
527 | UVD_SUVD_CGC_GATE__SITE_MASK
528 | UVD_SUVD_CGC_GATE__SRE_VP9_MASK
529 | UVD_SUVD_CGC_GATE__SCM_VP9_MASK
530 | UVD_SUVD_CGC_GATE__SIT_VP9_DEC_MASK
531 | UVD_SUVD_CGC_GATE__SDB_VP9_MASK
532 | UVD_SUVD_CGC_GATE__IME_HEVC_MASK);
533 WREG32_SOC15(VCN, 0, mmUVD_SUVD_CGC_GATE, data);
535 data = RREG32_SOC15(VCN, 0, mmUVD_SUVD_CGC_CTRL);
536 data &= ~(UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
537 | UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
538 | UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
539 | UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
540 | UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
541 | UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
542 | UVD_SUVD_CGC_CTRL__UVD_SC_MODE_MASK
543 | UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
544 | UVD_SUVD_CGC_CTRL__IME_MODE_MASK
545 | UVD_SUVD_CGC_CTRL__SITE_MODE_MASK);
546 WREG32_SOC15(VCN, 0, mmUVD_SUVD_CGC_CTRL, data);
549 static void vcn_v2_0_clock_gating_dpg_mode(struct amdgpu_device *adev,
550 uint8_t sram_sel, uint8_t indirect)
552 uint32_t reg_data = 0;
554 /* enable sw clock gating control */
555 if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
556 reg_data = 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
557 else
558 reg_data = 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
559 reg_data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
560 reg_data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
561 reg_data &= ~(UVD_CGC_CTRL__UDEC_RE_MODE_MASK |
562 UVD_CGC_CTRL__UDEC_CM_MODE_MASK |
563 UVD_CGC_CTRL__UDEC_IT_MODE_MASK |
564 UVD_CGC_CTRL__UDEC_DB_MODE_MASK |
565 UVD_CGC_CTRL__UDEC_MP_MODE_MASK |
566 UVD_CGC_CTRL__SYS_MODE_MASK |
567 UVD_CGC_CTRL__UDEC_MODE_MASK |
568 UVD_CGC_CTRL__MPEG2_MODE_MASK |
569 UVD_CGC_CTRL__REGS_MODE_MASK |
570 UVD_CGC_CTRL__RBC_MODE_MASK |
571 UVD_CGC_CTRL__LMI_MC_MODE_MASK |
572 UVD_CGC_CTRL__LMI_UMC_MODE_MASK |
573 UVD_CGC_CTRL__IDCT_MODE_MASK |
574 UVD_CGC_CTRL__MPRD_MODE_MASK |
575 UVD_CGC_CTRL__MPC_MODE_MASK |
576 UVD_CGC_CTRL__LBSI_MODE_MASK |
577 UVD_CGC_CTRL__LRBBM_MODE_MASK |
578 UVD_CGC_CTRL__WCB_MODE_MASK |
579 UVD_CGC_CTRL__VCPU_MODE_MASK |
580 UVD_CGC_CTRL__SCPU_MODE_MASK);
581 WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
582 UVD, 0, mmUVD_CGC_CTRL), reg_data, sram_sel, indirect);
584 /* turn off clock gating */
585 WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
586 UVD, 0, mmUVD_CGC_GATE), 0, sram_sel, indirect);
588 /* turn on SUVD clock gating */
589 WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
590 UVD, 0, mmUVD_SUVD_CGC_GATE), 1, sram_sel, indirect);
592 /* turn on sw mode in UVD_SUVD_CGC_CTRL */
593 WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
594 UVD, 0, mmUVD_SUVD_CGC_CTRL), 0, sram_sel, indirect);
598 * vcn_v2_0_enable_clock_gating - enable VCN clock gating
600 * @adev: amdgpu_device pointer
601 * @sw: enable SW clock gating
603 * Enable clock gating for VCN block
605 static void vcn_v2_0_enable_clock_gating(struct amdgpu_device *adev)
607 uint32_t data = 0;
609 /* enable UVD CGC */
610 data = RREG32_SOC15(VCN, 0, mmUVD_CGC_CTRL);
611 if (adev->cg_flags & AMD_CG_SUPPORT_VCN_MGCG)
612 data |= 1 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
613 else
614 data |= 0 << UVD_CGC_CTRL__DYN_CLOCK_MODE__SHIFT;
615 data |= 1 << UVD_CGC_CTRL__CLK_GATE_DLY_TIMER__SHIFT;
616 data |= 4 << UVD_CGC_CTRL__CLK_OFF_DELAY__SHIFT;
617 WREG32_SOC15(VCN, 0, mmUVD_CGC_CTRL, data);
619 data = RREG32_SOC15(VCN, 0, mmUVD_CGC_CTRL);
620 data |= (UVD_CGC_CTRL__UDEC_RE_MODE_MASK
621 | UVD_CGC_CTRL__UDEC_CM_MODE_MASK
622 | UVD_CGC_CTRL__UDEC_IT_MODE_MASK
623 | UVD_CGC_CTRL__UDEC_DB_MODE_MASK
624 | UVD_CGC_CTRL__UDEC_MP_MODE_MASK
625 | UVD_CGC_CTRL__SYS_MODE_MASK
626 | UVD_CGC_CTRL__UDEC_MODE_MASK
627 | UVD_CGC_CTRL__MPEG2_MODE_MASK
628 | UVD_CGC_CTRL__REGS_MODE_MASK
629 | UVD_CGC_CTRL__RBC_MODE_MASK
630 | UVD_CGC_CTRL__LMI_MC_MODE_MASK
631 | UVD_CGC_CTRL__LMI_UMC_MODE_MASK
632 | UVD_CGC_CTRL__IDCT_MODE_MASK
633 | UVD_CGC_CTRL__MPRD_MODE_MASK
634 | UVD_CGC_CTRL__MPC_MODE_MASK
635 | UVD_CGC_CTRL__LBSI_MODE_MASK
636 | UVD_CGC_CTRL__LRBBM_MODE_MASK
637 | UVD_CGC_CTRL__WCB_MODE_MASK
638 | UVD_CGC_CTRL__VCPU_MODE_MASK
639 | UVD_CGC_CTRL__SCPU_MODE_MASK);
640 WREG32_SOC15(VCN, 0, mmUVD_CGC_CTRL, data);
642 data = RREG32_SOC15(VCN, 0, mmUVD_SUVD_CGC_CTRL);
643 data |= (UVD_SUVD_CGC_CTRL__SRE_MODE_MASK
644 | UVD_SUVD_CGC_CTRL__SIT_MODE_MASK
645 | UVD_SUVD_CGC_CTRL__SMP_MODE_MASK
646 | UVD_SUVD_CGC_CTRL__SCM_MODE_MASK
647 | UVD_SUVD_CGC_CTRL__SDB_MODE_MASK
648 | UVD_SUVD_CGC_CTRL__SCLR_MODE_MASK
649 | UVD_SUVD_CGC_CTRL__UVD_SC_MODE_MASK
650 | UVD_SUVD_CGC_CTRL__ENT_MODE_MASK
651 | UVD_SUVD_CGC_CTRL__IME_MODE_MASK
652 | UVD_SUVD_CGC_CTRL__SITE_MODE_MASK);
653 WREG32_SOC15(VCN, 0, mmUVD_SUVD_CGC_CTRL, data);
656 static void vcn_v2_0_disable_static_power_gating(struct amdgpu_device *adev)
658 uint32_t data = 0;
659 int ret;
661 if (adev->pg_flags & AMD_PG_SUPPORT_VCN) {
662 data = (1 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT
663 | 1 << UVD_PGFSM_CONFIG__UVDU_PWR_CONFIG__SHIFT
664 | 2 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT
665 | 2 << UVD_PGFSM_CONFIG__UVDC_PWR_CONFIG__SHIFT
666 | 2 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT
667 | 2 << UVD_PGFSM_CONFIG__UVDIL_PWR_CONFIG__SHIFT
668 | 2 << UVD_PGFSM_CONFIG__UVDIR_PWR_CONFIG__SHIFT
669 | 2 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT
670 | 2 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT
671 | 2 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT);
673 WREG32_SOC15(VCN, 0, mmUVD_PGFSM_CONFIG, data);
674 SOC15_WAIT_ON_RREG(VCN, 0, mmUVD_PGFSM_STATUS,
675 UVD_PGFSM_STATUS__UVDM_UVDU_PWR_ON_2_0, 0xFFFFF, ret);
676 } else {
677 data = (1 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT
678 | 1 << UVD_PGFSM_CONFIG__UVDU_PWR_CONFIG__SHIFT
679 | 1 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT
680 | 1 << UVD_PGFSM_CONFIG__UVDC_PWR_CONFIG__SHIFT
681 | 1 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT
682 | 1 << UVD_PGFSM_CONFIG__UVDIL_PWR_CONFIG__SHIFT
683 | 1 << UVD_PGFSM_CONFIG__UVDIR_PWR_CONFIG__SHIFT
684 | 1 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT
685 | 1 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT
686 | 1 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT);
687 WREG32_SOC15(VCN, 0, mmUVD_PGFSM_CONFIG, data);
688 SOC15_WAIT_ON_RREG(VCN, 0, mmUVD_PGFSM_STATUS, 0, 0xFFFFF, ret);
691 /* polling UVD_PGFSM_STATUS to confirm UVDM_PWR_STATUS,
692 * UVDU_PWR_STATUS are 0 (power on) */
694 data = RREG32_SOC15(VCN, 0, mmUVD_POWER_STATUS);
695 data &= ~0x103;
696 if (adev->pg_flags & AMD_PG_SUPPORT_VCN)
697 data |= UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON |
698 UVD_POWER_STATUS__UVD_PG_EN_MASK;
700 WREG32_SOC15(VCN, 0, mmUVD_POWER_STATUS, data);
703 static void vcn_v2_0_enable_static_power_gating(struct amdgpu_device *adev)
705 uint32_t data = 0;
706 int ret;
708 if (adev->pg_flags & AMD_PG_SUPPORT_VCN) {
709 /* Before power off, this indicator has to be turned on */
710 data = RREG32_SOC15(VCN, 0, mmUVD_POWER_STATUS);
711 data &= ~UVD_POWER_STATUS__UVD_POWER_STATUS_MASK;
712 data |= UVD_POWER_STATUS__UVD_POWER_STATUS_TILES_OFF;
713 WREG32_SOC15(VCN, 0, mmUVD_POWER_STATUS, data);
716 data = (2 << UVD_PGFSM_CONFIG__UVDM_PWR_CONFIG__SHIFT
717 | 2 << UVD_PGFSM_CONFIG__UVDU_PWR_CONFIG__SHIFT
718 | 2 << UVD_PGFSM_CONFIG__UVDF_PWR_CONFIG__SHIFT
719 | 2 << UVD_PGFSM_CONFIG__UVDC_PWR_CONFIG__SHIFT
720 | 2 << UVD_PGFSM_CONFIG__UVDB_PWR_CONFIG__SHIFT
721 | 2 << UVD_PGFSM_CONFIG__UVDIL_PWR_CONFIG__SHIFT
722 | 2 << UVD_PGFSM_CONFIG__UVDIR_PWR_CONFIG__SHIFT
723 | 2 << UVD_PGFSM_CONFIG__UVDTD_PWR_CONFIG__SHIFT
724 | 2 << UVD_PGFSM_CONFIG__UVDTE_PWR_CONFIG__SHIFT
725 | 2 << UVD_PGFSM_CONFIG__UVDE_PWR_CONFIG__SHIFT);
727 WREG32_SOC15(VCN, 0, mmUVD_PGFSM_CONFIG, data);
729 data = (2 << UVD_PGFSM_STATUS__UVDM_PWR_STATUS__SHIFT
730 | 2 << UVD_PGFSM_STATUS__UVDU_PWR_STATUS__SHIFT
731 | 2 << UVD_PGFSM_STATUS__UVDF_PWR_STATUS__SHIFT
732 | 2 << UVD_PGFSM_STATUS__UVDC_PWR_STATUS__SHIFT
733 | 2 << UVD_PGFSM_STATUS__UVDB_PWR_STATUS__SHIFT
734 | 2 << UVD_PGFSM_STATUS__UVDIL_PWR_STATUS__SHIFT
735 | 2 << UVD_PGFSM_STATUS__UVDIR_PWR_STATUS__SHIFT
736 | 2 << UVD_PGFSM_STATUS__UVDTD_PWR_STATUS__SHIFT
737 | 2 << UVD_PGFSM_STATUS__UVDTE_PWR_STATUS__SHIFT
738 | 2 << UVD_PGFSM_STATUS__UVDE_PWR_STATUS__SHIFT);
739 SOC15_WAIT_ON_RREG(VCN, 0, mmUVD_PGFSM_STATUS, data, 0xFFFFF, ret);
743 static int vcn_v2_0_start_dpg_mode(struct amdgpu_device *adev, bool indirect)
745 struct amdgpu_ring *ring = &adev->vcn.inst->ring_dec;
746 uint32_t rb_bufsz, tmp;
748 vcn_v2_0_enable_static_power_gating(adev);
750 /* enable dynamic power gating mode */
751 tmp = RREG32_SOC15(UVD, 0, mmUVD_POWER_STATUS);
752 tmp |= UVD_POWER_STATUS__UVD_PG_MODE_MASK;
753 tmp |= UVD_POWER_STATUS__UVD_PG_EN_MASK;
754 WREG32_SOC15(UVD, 0, mmUVD_POWER_STATUS, tmp);
756 if (indirect)
757 adev->vcn.inst->dpg_sram_curr_addr = (uint32_t*)adev->vcn.inst->dpg_sram_cpu_addr;
759 /* enable clock gating */
760 vcn_v2_0_clock_gating_dpg_mode(adev, 0, indirect);
762 /* enable VCPU clock */
763 tmp = (0xFF << UVD_VCPU_CNTL__PRB_TIMEOUT_VAL__SHIFT);
764 tmp |= UVD_VCPU_CNTL__CLK_EN_MASK;
765 tmp |= UVD_VCPU_CNTL__MIF_WR_LOW_THRESHOLD_BP_MASK;
766 WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
767 UVD, 0, mmUVD_VCPU_CNTL), tmp, 0, indirect);
769 /* disable master interupt */
770 WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
771 UVD, 0, mmUVD_MASTINT_EN), 0, 0, indirect);
773 /* setup mmUVD_LMI_CTRL */
774 tmp = (UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK |
775 UVD_LMI_CTRL__REQ_MODE_MASK |
776 UVD_LMI_CTRL__CRC_RESET_MASK |
777 UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
778 UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
779 UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK |
780 (8 << UVD_LMI_CTRL__WRITE_CLEAN_TIMER__SHIFT) |
781 0x00100000L);
782 WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
783 UVD, 0, mmUVD_LMI_CTRL), tmp, 0, indirect);
785 WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
786 UVD, 0, mmUVD_MPC_CNTL),
787 0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT, 0, indirect);
789 WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
790 UVD, 0, mmUVD_MPC_SET_MUXA0),
791 ((0x1 << UVD_MPC_SET_MUXA0__VARA_1__SHIFT) |
792 (0x2 << UVD_MPC_SET_MUXA0__VARA_2__SHIFT) |
793 (0x3 << UVD_MPC_SET_MUXA0__VARA_3__SHIFT) |
794 (0x4 << UVD_MPC_SET_MUXA0__VARA_4__SHIFT)), 0, indirect);
796 WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
797 UVD, 0, mmUVD_MPC_SET_MUXB0),
798 ((0x1 << UVD_MPC_SET_MUXB0__VARB_1__SHIFT) |
799 (0x2 << UVD_MPC_SET_MUXB0__VARB_2__SHIFT) |
800 (0x3 << UVD_MPC_SET_MUXB0__VARB_3__SHIFT) |
801 (0x4 << UVD_MPC_SET_MUXB0__VARB_4__SHIFT)), 0, indirect);
803 WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
804 UVD, 0, mmUVD_MPC_SET_MUX),
805 ((0x0 << UVD_MPC_SET_MUX__SET_0__SHIFT) |
806 (0x1 << UVD_MPC_SET_MUX__SET_1__SHIFT) |
807 (0x2 << UVD_MPC_SET_MUX__SET_2__SHIFT)), 0, indirect);
809 vcn_v2_0_mc_resume_dpg_mode(adev, indirect);
811 WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
812 UVD, 0, mmUVD_REG_XX_MASK), 0x10, 0, indirect);
813 WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
814 UVD, 0, mmUVD_RBC_XX_IB_REG_CHECK), 0x3, 0, indirect);
816 /* release VCPU reset to boot */
817 WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
818 UVD, 0, mmUVD_SOFT_RESET), 0, 0, indirect);
820 /* enable LMI MC and UMC channels */
821 WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
822 UVD, 0, mmUVD_LMI_CTRL2),
823 0x1F << UVD_LMI_CTRL2__RE_OFLD_MIF_WR_REQ_NUM__SHIFT, 0, indirect);
825 /* enable master interrupt */
826 WREG32_SOC15_DPG_MODE_2_0(0, SOC15_DPG_MODE_OFFSET_2_0(
827 UVD, 0, mmUVD_MASTINT_EN),
828 UVD_MASTINT_EN__VCPU_EN_MASK, 0, indirect);
830 if (indirect)
831 psp_update_vcn_sram(adev, 0, adev->vcn.inst->dpg_sram_gpu_addr,
832 (uint32_t)((uintptr_t)adev->vcn.inst->dpg_sram_curr_addr -
833 (uintptr_t)adev->vcn.inst->dpg_sram_cpu_addr));
835 /* force RBC into idle state */
836 rb_bufsz = order_base_2(ring->ring_size);
837 tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz);
838 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
839 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
840 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
841 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
842 WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_CNTL, tmp);
844 /* set the write pointer delay */
845 WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_WPTR_CNTL, 0);
847 /* set the wb address */
848 WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_RPTR_ADDR,
849 (upper_32_bits(ring->gpu_addr) >> 2));
851 /* programm the RB_BASE for ring buffer */
852 WREG32_SOC15(UVD, 0, mmUVD_LMI_RBC_RB_64BIT_BAR_LOW,
853 lower_32_bits(ring->gpu_addr));
854 WREG32_SOC15(UVD, 0, mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH,
855 upper_32_bits(ring->gpu_addr));
857 /* Initialize the ring buffer's read and write pointers */
858 WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_RPTR, 0);
860 WREG32_SOC15(UVD, 0, mmUVD_SCRATCH2, 0);
862 ring->wptr = RREG32_SOC15(UVD, 0, mmUVD_RBC_RB_RPTR);
863 WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_WPTR,
864 lower_32_bits(ring->wptr));
866 return 0;
869 static int vcn_v2_0_start(struct amdgpu_device *adev)
871 struct amdgpu_ring *ring = &adev->vcn.inst->ring_dec;
872 uint32_t rb_bufsz, tmp;
873 uint32_t lmi_swap_cntl;
874 int i, j, r;
876 if (adev->pm.dpm_enabled)
877 amdgpu_dpm_enable_uvd(adev, true);
879 if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)
880 return vcn_v2_0_start_dpg_mode(adev, adev->vcn.indirect_sram);
882 vcn_v2_0_disable_static_power_gating(adev);
884 /* set uvd status busy */
885 tmp = RREG32_SOC15(UVD, 0, mmUVD_STATUS) | UVD_STATUS__UVD_BUSY;
886 WREG32_SOC15(UVD, 0, mmUVD_STATUS, tmp);
888 /*SW clock gating */
889 vcn_v2_0_disable_clock_gating(adev);
891 /* enable VCPU clock */
892 WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_VCPU_CNTL),
893 UVD_VCPU_CNTL__CLK_EN_MASK, ~UVD_VCPU_CNTL__CLK_EN_MASK);
895 /* disable master interrupt */
896 WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_MASTINT_EN), 0,
897 ~UVD_MASTINT_EN__VCPU_EN_MASK);
899 /* setup mmUVD_LMI_CTRL */
900 tmp = RREG32_SOC15(UVD, 0, mmUVD_LMI_CTRL);
901 WREG32_SOC15(UVD, 0, mmUVD_LMI_CTRL, tmp |
902 UVD_LMI_CTRL__WRITE_CLEAN_TIMER_EN_MASK |
903 UVD_LMI_CTRL__MASK_MC_URGENT_MASK |
904 UVD_LMI_CTRL__DATA_COHERENCY_EN_MASK |
905 UVD_LMI_CTRL__VCPU_DATA_COHERENCY_EN_MASK);
907 /* setup mmUVD_MPC_CNTL */
908 tmp = RREG32_SOC15(UVD, 0, mmUVD_MPC_CNTL);
909 tmp &= ~UVD_MPC_CNTL__REPLACEMENT_MODE_MASK;
910 tmp |= 0x2 << UVD_MPC_CNTL__REPLACEMENT_MODE__SHIFT;
911 WREG32_SOC15(VCN, 0, mmUVD_MPC_CNTL, tmp);
913 /* setup UVD_MPC_SET_MUXA0 */
914 WREG32_SOC15(UVD, 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)));
920 /* setup UVD_MPC_SET_MUXB0 */
921 WREG32_SOC15(UVD, 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)));
927 /* setup mmUVD_MPC_SET_MUX */
928 WREG32_SOC15(UVD, 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)));
933 vcn_v2_0_mc_resume(adev);
935 /* release VCPU reset to boot */
936 WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_SOFT_RESET), 0,
937 ~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
939 /* enable LMI MC and UMC channels */
940 WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_LMI_CTRL2), 0,
941 ~UVD_LMI_CTRL2__STALL_ARB_UMC_MASK);
943 tmp = RREG32_SOC15(VCN, 0, mmUVD_SOFT_RESET);
944 tmp &= ~UVD_SOFT_RESET__LMI_SOFT_RESET_MASK;
945 tmp &= ~UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK;
946 WREG32_SOC15(VCN, 0, mmUVD_SOFT_RESET, tmp);
948 /* disable byte swapping */
949 lmi_swap_cntl = 0;
950 #ifdef __BIG_ENDIAN
951 /* swap (8 in 32) RB and IB */
952 lmi_swap_cntl = 0xa;
953 #endif
954 WREG32_SOC15(UVD, 0, mmUVD_LMI_SWAP_CNTL, lmi_swap_cntl);
956 for (i = 0; i < 10; ++i) {
957 uint32_t status;
959 for (j = 0; j < 100; ++j) {
960 status = RREG32_SOC15(UVD, 0, mmUVD_STATUS);
961 if (status & 2)
962 break;
963 mdelay(10);
965 r = 0;
966 if (status & 2)
967 break;
969 DRM_ERROR("VCN decode not responding, trying to reset the VCPU!!!\n");
970 WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_SOFT_RESET),
971 UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK,
972 ~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
973 mdelay(10);
974 WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_SOFT_RESET), 0,
975 ~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
976 mdelay(10);
977 r = -1;
980 if (r) {
981 DRM_ERROR("VCN decode not responding, giving up!!!\n");
982 return r;
985 /* enable master interrupt */
986 WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_MASTINT_EN),
987 UVD_MASTINT_EN__VCPU_EN_MASK,
988 ~UVD_MASTINT_EN__VCPU_EN_MASK);
990 /* clear the busy bit of VCN_STATUS */
991 WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_STATUS), 0,
992 ~(2 << UVD_STATUS__VCPU_REPORT__SHIFT));
994 WREG32_SOC15(UVD, 0, mmUVD_LMI_RBC_RB_VMID, 0);
996 /* force RBC into idle state */
997 rb_bufsz = order_base_2(ring->ring_size);
998 tmp = REG_SET_FIELD(0, UVD_RBC_RB_CNTL, RB_BUFSZ, rb_bufsz);
999 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_BLKSZ, 1);
1000 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_FETCH, 1);
1001 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_NO_UPDATE, 1);
1002 tmp = REG_SET_FIELD(tmp, UVD_RBC_RB_CNTL, RB_RPTR_WR_EN, 1);
1003 WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_CNTL, tmp);
1005 /* programm the RB_BASE for ring buffer */
1006 WREG32_SOC15(UVD, 0, mmUVD_LMI_RBC_RB_64BIT_BAR_LOW,
1007 lower_32_bits(ring->gpu_addr));
1008 WREG32_SOC15(UVD, 0, mmUVD_LMI_RBC_RB_64BIT_BAR_HIGH,
1009 upper_32_bits(ring->gpu_addr));
1011 /* Initialize the ring buffer's read and write pointers */
1012 WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_RPTR, 0);
1014 ring->wptr = RREG32_SOC15(UVD, 0, mmUVD_RBC_RB_RPTR);
1015 WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_WPTR,
1016 lower_32_bits(ring->wptr));
1018 ring = &adev->vcn.inst->ring_enc[0];
1019 WREG32_SOC15(UVD, 0, mmUVD_RB_RPTR, lower_32_bits(ring->wptr));
1020 WREG32_SOC15(UVD, 0, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
1021 WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_LO, ring->gpu_addr);
1022 WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
1023 WREG32_SOC15(UVD, 0, mmUVD_RB_SIZE, ring->ring_size / 4);
1025 ring = &adev->vcn.inst->ring_enc[1];
1026 WREG32_SOC15(UVD, 0, mmUVD_RB_RPTR2, lower_32_bits(ring->wptr));
1027 WREG32_SOC15(UVD, 0, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
1028 WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_LO2, ring->gpu_addr);
1029 WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_HI2, upper_32_bits(ring->gpu_addr));
1030 WREG32_SOC15(UVD, 0, mmUVD_RB_SIZE2, ring->ring_size / 4);
1032 return 0;
1035 static int vcn_v2_0_stop_dpg_mode(struct amdgpu_device *adev)
1037 int ret_code = 0;
1038 uint32_t tmp;
1040 /* Wait for power status to be 1 */
1041 SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_POWER_STATUS, 1,
1042 UVD_POWER_STATUS__UVD_POWER_STATUS_MASK, ret_code);
1044 /* wait for read ptr to be equal to write ptr */
1045 tmp = RREG32_SOC15(UVD, 0, mmUVD_RB_WPTR);
1046 SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_RB_RPTR, tmp, 0xFFFFFFFF, ret_code);
1048 tmp = RREG32_SOC15(UVD, 0, mmUVD_RB_WPTR2);
1049 SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_RB_RPTR2, tmp, 0xFFFFFFFF, ret_code);
1051 tmp = RREG32_SOC15(UVD, 0, mmUVD_RBC_RB_WPTR) & 0x7FFFFFFF;
1052 SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_RBC_RB_RPTR, tmp, 0xFFFFFFFF, ret_code);
1054 SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_POWER_STATUS, 1,
1055 UVD_POWER_STATUS__UVD_POWER_STATUS_MASK, ret_code);
1057 /* disable dynamic power gating mode */
1058 WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_POWER_STATUS), 0,
1059 ~UVD_POWER_STATUS__UVD_PG_MODE_MASK);
1061 return 0;
1064 static int vcn_v2_0_stop(struct amdgpu_device *adev)
1066 uint32_t tmp;
1067 int r;
1069 if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG) {
1070 r = vcn_v2_0_stop_dpg_mode(adev);
1071 if (r)
1072 return r;
1073 goto power_off;
1076 /* wait for uvd idle */
1077 SOC15_WAIT_ON_RREG(VCN, 0, mmUVD_STATUS, UVD_STATUS__IDLE, 0x7, r);
1078 if (r)
1079 return r;
1081 tmp = UVD_LMI_STATUS__VCPU_LMI_WRITE_CLEAN_MASK |
1082 UVD_LMI_STATUS__READ_CLEAN_MASK |
1083 UVD_LMI_STATUS__WRITE_CLEAN_MASK |
1084 UVD_LMI_STATUS__WRITE_CLEAN_RAW_MASK;
1085 SOC15_WAIT_ON_RREG(VCN, 0, mmUVD_LMI_STATUS, tmp, tmp, r);
1086 if (r)
1087 return r;
1089 /* stall UMC channel */
1090 tmp = RREG32_SOC15(VCN, 0, mmUVD_LMI_CTRL2);
1091 tmp |= UVD_LMI_CTRL2__STALL_ARB_UMC_MASK;
1092 WREG32_SOC15(VCN, 0, mmUVD_LMI_CTRL2, tmp);
1094 tmp = UVD_LMI_STATUS__UMC_READ_CLEAN_RAW_MASK|
1095 UVD_LMI_STATUS__UMC_WRITE_CLEAN_RAW_MASK;
1096 SOC15_WAIT_ON_RREG(VCN, 0, mmUVD_LMI_STATUS, tmp, tmp, r);
1097 if (r)
1098 return r;
1100 /* disable VCPU clock */
1101 WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_VCPU_CNTL), 0,
1102 ~(UVD_VCPU_CNTL__CLK_EN_MASK));
1104 /* reset LMI UMC */
1105 WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_SOFT_RESET),
1106 UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK,
1107 ~UVD_SOFT_RESET__LMI_UMC_SOFT_RESET_MASK);
1109 /* reset LMI */
1110 WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_SOFT_RESET),
1111 UVD_SOFT_RESET__LMI_SOFT_RESET_MASK,
1112 ~UVD_SOFT_RESET__LMI_SOFT_RESET_MASK);
1114 /* reset VCPU */
1115 WREG32_P(SOC15_REG_OFFSET(UVD, 0, mmUVD_SOFT_RESET),
1116 UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK,
1117 ~UVD_SOFT_RESET__VCPU_SOFT_RESET_MASK);
1119 /* clear status */
1120 WREG32_SOC15(VCN, 0, mmUVD_STATUS, 0);
1122 vcn_v2_0_enable_clock_gating(adev);
1123 vcn_v2_0_enable_static_power_gating(adev);
1125 power_off:
1126 if (adev->pm.dpm_enabled)
1127 amdgpu_dpm_enable_uvd(adev, false);
1129 return 0;
1132 static int vcn_v2_0_pause_dpg_mode(struct amdgpu_device *adev,
1133 int inst_idx, struct dpg_pause_state *new_state)
1135 struct amdgpu_ring *ring;
1136 uint32_t reg_data = 0;
1137 int ret_code;
1139 /* pause/unpause if state is changed */
1140 if (adev->vcn.pause_state.fw_based != new_state->fw_based) {
1141 DRM_DEBUG("dpg pause state changed %d -> %d",
1142 adev->vcn.pause_state.fw_based, new_state->fw_based);
1143 reg_data = RREG32_SOC15(UVD, 0, mmUVD_DPG_PAUSE) &
1144 (~UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK);
1146 if (new_state->fw_based == VCN_DPG_STATE__PAUSE) {
1147 ret_code = 0;
1148 SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_POWER_STATUS, 0x1,
1149 UVD_POWER_STATUS__UVD_POWER_STATUS_MASK, ret_code);
1151 if (!ret_code) {
1152 /* pause DPG */
1153 reg_data |= UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1154 WREG32_SOC15(UVD, 0, mmUVD_DPG_PAUSE, reg_data);
1156 /* wait for ACK */
1157 SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_DPG_PAUSE,
1158 UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK,
1159 UVD_DPG_PAUSE__NJ_PAUSE_DPG_ACK_MASK, ret_code);
1161 /* Restore */
1162 ring = &adev->vcn.inst->ring_enc[0];
1163 WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_LO, ring->gpu_addr);
1164 WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_HI, upper_32_bits(ring->gpu_addr));
1165 WREG32_SOC15(UVD, 0, mmUVD_RB_SIZE, ring->ring_size / 4);
1166 WREG32_SOC15(UVD, 0, mmUVD_RB_RPTR, lower_32_bits(ring->wptr));
1167 WREG32_SOC15(UVD, 0, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
1169 ring = &adev->vcn.inst->ring_enc[1];
1170 WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_LO2, ring->gpu_addr);
1171 WREG32_SOC15(UVD, 0, mmUVD_RB_BASE_HI2, upper_32_bits(ring->gpu_addr));
1172 WREG32_SOC15(UVD, 0, mmUVD_RB_SIZE2, ring->ring_size / 4);
1173 WREG32_SOC15(UVD, 0, mmUVD_RB_RPTR2, lower_32_bits(ring->wptr));
1174 WREG32_SOC15(UVD, 0, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
1176 WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_WPTR,
1177 RREG32_SOC15(UVD, 0, mmUVD_SCRATCH2) & 0x7FFFFFFF);
1179 SOC15_WAIT_ON_RREG(UVD, 0, mmUVD_POWER_STATUS,
1180 UVD_PGFSM_CONFIG__UVDM_UVDU_PWR_ON,
1181 UVD_POWER_STATUS__UVD_POWER_STATUS_MASK, ret_code);
1183 } else {
1184 /* unpause dpg, no need to wait */
1185 reg_data &= ~UVD_DPG_PAUSE__NJ_PAUSE_DPG_REQ_MASK;
1186 WREG32_SOC15(UVD, 0, mmUVD_DPG_PAUSE, reg_data);
1188 adev->vcn.pause_state.fw_based = new_state->fw_based;
1191 return 0;
1194 static bool vcn_v2_0_is_idle(void *handle)
1196 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1198 return (RREG32_SOC15(VCN, 0, mmUVD_STATUS) == UVD_STATUS__IDLE);
1201 static int vcn_v2_0_wait_for_idle(void *handle)
1203 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1204 int ret = 0;
1206 SOC15_WAIT_ON_RREG(VCN, 0, mmUVD_STATUS, UVD_STATUS__IDLE,
1207 UVD_STATUS__IDLE, ret);
1209 return ret;
1212 static int vcn_v2_0_set_clockgating_state(void *handle,
1213 enum amd_clockgating_state state)
1215 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1216 bool enable = (state == AMD_CG_STATE_GATE) ? true : false;
1218 if (enable) {
1219 /* wait for STATUS to clear */
1220 if (vcn_v2_0_is_idle(handle))
1221 return -EBUSY;
1222 vcn_v2_0_enable_clock_gating(adev);
1223 } else {
1224 /* disable HW gating and enable Sw gating */
1225 vcn_v2_0_disable_clock_gating(adev);
1227 return 0;
1231 * vcn_v2_0_dec_ring_get_rptr - get read pointer
1233 * @ring: amdgpu_ring pointer
1235 * Returns the current hardware read pointer
1237 static uint64_t vcn_v2_0_dec_ring_get_rptr(struct amdgpu_ring *ring)
1239 struct amdgpu_device *adev = ring->adev;
1241 return RREG32_SOC15(UVD, 0, mmUVD_RBC_RB_RPTR);
1245 * vcn_v2_0_dec_ring_get_wptr - get write pointer
1247 * @ring: amdgpu_ring pointer
1249 * Returns the current hardware write pointer
1251 static uint64_t vcn_v2_0_dec_ring_get_wptr(struct amdgpu_ring *ring)
1253 struct amdgpu_device *adev = ring->adev;
1255 if (ring->use_doorbell)
1256 return adev->wb.wb[ring->wptr_offs];
1257 else
1258 return RREG32_SOC15(UVD, 0, mmUVD_RBC_RB_WPTR);
1262 * vcn_v2_0_dec_ring_set_wptr - set write pointer
1264 * @ring: amdgpu_ring pointer
1266 * Commits the write pointer to the hardware
1268 static void vcn_v2_0_dec_ring_set_wptr(struct amdgpu_ring *ring)
1270 struct amdgpu_device *adev = ring->adev;
1272 if (adev->pg_flags & AMD_PG_SUPPORT_VCN_DPG)
1273 WREG32_SOC15(UVD, 0, mmUVD_SCRATCH2,
1274 lower_32_bits(ring->wptr) | 0x80000000);
1276 if (ring->use_doorbell) {
1277 adev->wb.wb[ring->wptr_offs] = lower_32_bits(ring->wptr);
1278 WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1279 } else {
1280 WREG32_SOC15(UVD, 0, mmUVD_RBC_RB_WPTR, lower_32_bits(ring->wptr));
1285 * vcn_v2_0_dec_ring_insert_start - insert a start command
1287 * @ring: amdgpu_ring pointer
1289 * Write a start command to the ring.
1291 void vcn_v2_0_dec_ring_insert_start(struct amdgpu_ring *ring)
1293 struct amdgpu_device *adev = ring->adev;
1295 amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.data0, 0));
1296 amdgpu_ring_write(ring, 0);
1297 amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.cmd, 0));
1298 amdgpu_ring_write(ring, VCN_DEC_KMD_CMD | (VCN_DEC_CMD_PACKET_START << 1));
1302 * vcn_v2_0_dec_ring_insert_end - insert a end command
1304 * @ring: amdgpu_ring pointer
1306 * Write a end command to the ring.
1308 void vcn_v2_0_dec_ring_insert_end(struct amdgpu_ring *ring)
1310 struct amdgpu_device *adev = ring->adev;
1312 amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.cmd, 0));
1313 amdgpu_ring_write(ring, VCN_DEC_KMD_CMD | (VCN_DEC_CMD_PACKET_END << 1));
1317 * vcn_v2_0_dec_ring_insert_nop - insert a nop command
1319 * @ring: amdgpu_ring pointer
1321 * Write a nop command to the ring.
1323 void vcn_v2_0_dec_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count)
1325 struct amdgpu_device *adev = ring->adev;
1326 int i;
1328 WARN_ON(ring->wptr % 2 || count % 2);
1330 for (i = 0; i < count / 2; i++) {
1331 amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.nop, 0));
1332 amdgpu_ring_write(ring, 0);
1337 * vcn_v2_0_dec_ring_emit_fence - emit an fence & trap command
1339 * @ring: amdgpu_ring pointer
1340 * @fence: fence to emit
1342 * Write a fence and a trap command to the ring.
1344 void vcn_v2_0_dec_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, u64 seq,
1345 unsigned flags)
1347 struct amdgpu_device *adev = ring->adev;
1349 WARN_ON(flags & AMDGPU_FENCE_FLAG_64BIT);
1350 amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.context_id, 0));
1351 amdgpu_ring_write(ring, seq);
1353 amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.data0, 0));
1354 amdgpu_ring_write(ring, addr & 0xffffffff);
1356 amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.data1, 0));
1357 amdgpu_ring_write(ring, upper_32_bits(addr) & 0xff);
1359 amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.cmd, 0));
1360 amdgpu_ring_write(ring, VCN_DEC_KMD_CMD | (VCN_DEC_CMD_FENCE << 1));
1362 amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.data0, 0));
1363 amdgpu_ring_write(ring, 0);
1365 amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.data1, 0));
1366 amdgpu_ring_write(ring, 0);
1368 amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.cmd, 0));
1370 amdgpu_ring_write(ring, VCN_DEC_KMD_CMD | (VCN_DEC_CMD_TRAP << 1));
1374 * vcn_v2_0_dec_ring_emit_ib - execute indirect buffer
1376 * @ring: amdgpu_ring pointer
1377 * @ib: indirect buffer to execute
1379 * Write ring commands to execute the indirect buffer
1381 void vcn_v2_0_dec_ring_emit_ib(struct amdgpu_ring *ring,
1382 struct amdgpu_job *job,
1383 struct amdgpu_ib *ib,
1384 uint32_t flags)
1386 struct amdgpu_device *adev = ring->adev;
1387 unsigned vmid = AMDGPU_JOB_GET_VMID(job);
1389 amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.ib_vmid, 0));
1390 amdgpu_ring_write(ring, vmid);
1392 amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.ib_bar_low, 0));
1393 amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr));
1394 amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.ib_bar_high, 0));
1395 amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr));
1396 amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.ib_size, 0));
1397 amdgpu_ring_write(ring, ib->length_dw);
1400 void vcn_v2_0_dec_ring_emit_reg_wait(struct amdgpu_ring *ring, uint32_t reg,
1401 uint32_t val, uint32_t mask)
1403 struct amdgpu_device *adev = ring->adev;
1405 amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.data0, 0));
1406 amdgpu_ring_write(ring, reg << 2);
1408 amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.data1, 0));
1409 amdgpu_ring_write(ring, val);
1411 amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.gp_scratch8, 0));
1412 amdgpu_ring_write(ring, mask);
1414 amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.cmd, 0));
1416 amdgpu_ring_write(ring, VCN_DEC_KMD_CMD | (VCN_DEC_CMD_REG_READ_COND_WAIT << 1));
1419 void vcn_v2_0_dec_ring_emit_vm_flush(struct amdgpu_ring *ring,
1420 unsigned vmid, uint64_t pd_addr)
1422 struct amdgpu_vmhub *hub = &ring->adev->vmhub[ring->funcs->vmhub];
1423 uint32_t data0, data1, mask;
1425 pd_addr = amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr);
1427 /* wait for register write */
1428 data0 = hub->ctx0_ptb_addr_lo32 + vmid * 2;
1429 data1 = lower_32_bits(pd_addr);
1430 mask = 0xffffffff;
1431 vcn_v2_0_dec_ring_emit_reg_wait(ring, data0, data1, mask);
1434 void vcn_v2_0_dec_ring_emit_wreg(struct amdgpu_ring *ring,
1435 uint32_t reg, uint32_t val)
1437 struct amdgpu_device *adev = ring->adev;
1439 amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.data0, 0));
1440 amdgpu_ring_write(ring, reg << 2);
1442 amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.data1, 0));
1443 amdgpu_ring_write(ring, val);
1445 amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.cmd, 0));
1447 amdgpu_ring_write(ring, VCN_DEC_KMD_CMD | (VCN_DEC_CMD_WRITE_REG << 1));
1451 * vcn_v2_0_enc_ring_get_rptr - get enc read pointer
1453 * @ring: amdgpu_ring pointer
1455 * Returns the current hardware enc read pointer
1457 static uint64_t vcn_v2_0_enc_ring_get_rptr(struct amdgpu_ring *ring)
1459 struct amdgpu_device *adev = ring->adev;
1461 if (ring == &adev->vcn.inst->ring_enc[0])
1462 return RREG32_SOC15(UVD, 0, mmUVD_RB_RPTR);
1463 else
1464 return RREG32_SOC15(UVD, 0, mmUVD_RB_RPTR2);
1468 * vcn_v2_0_enc_ring_get_wptr - get enc write pointer
1470 * @ring: amdgpu_ring pointer
1472 * Returns the current hardware enc write pointer
1474 static uint64_t vcn_v2_0_enc_ring_get_wptr(struct amdgpu_ring *ring)
1476 struct amdgpu_device *adev = ring->adev;
1478 if (ring == &adev->vcn.inst->ring_enc[0]) {
1479 if (ring->use_doorbell)
1480 return adev->wb.wb[ring->wptr_offs];
1481 else
1482 return RREG32_SOC15(UVD, 0, mmUVD_RB_WPTR);
1483 } else {
1484 if (ring->use_doorbell)
1485 return adev->wb.wb[ring->wptr_offs];
1486 else
1487 return RREG32_SOC15(UVD, 0, mmUVD_RB_WPTR2);
1492 * vcn_v2_0_enc_ring_set_wptr - set enc write pointer
1494 * @ring: amdgpu_ring pointer
1496 * Commits the enc write pointer to the hardware
1498 static void vcn_v2_0_enc_ring_set_wptr(struct amdgpu_ring *ring)
1500 struct amdgpu_device *adev = ring->adev;
1502 if (ring == &adev->vcn.inst->ring_enc[0]) {
1503 if (ring->use_doorbell) {
1504 adev->wb.wb[ring->wptr_offs] = lower_32_bits(ring->wptr);
1505 WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1506 } else {
1507 WREG32_SOC15(UVD, 0, mmUVD_RB_WPTR, lower_32_bits(ring->wptr));
1509 } else {
1510 if (ring->use_doorbell) {
1511 adev->wb.wb[ring->wptr_offs] = lower_32_bits(ring->wptr);
1512 WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
1513 } else {
1514 WREG32_SOC15(UVD, 0, mmUVD_RB_WPTR2, lower_32_bits(ring->wptr));
1520 * vcn_v2_0_enc_ring_emit_fence - emit an enc fence & trap command
1522 * @ring: amdgpu_ring pointer
1523 * @fence: fence to emit
1525 * Write enc a fence and a trap command to the ring.
1527 void vcn_v2_0_enc_ring_emit_fence(struct amdgpu_ring *ring, u64 addr,
1528 u64 seq, unsigned flags)
1530 WARN_ON(flags & AMDGPU_FENCE_FLAG_64BIT);
1532 amdgpu_ring_write(ring, VCN_ENC_CMD_FENCE);
1533 amdgpu_ring_write(ring, addr);
1534 amdgpu_ring_write(ring, upper_32_bits(addr));
1535 amdgpu_ring_write(ring, seq);
1536 amdgpu_ring_write(ring, VCN_ENC_CMD_TRAP);
1539 void vcn_v2_0_enc_ring_insert_end(struct amdgpu_ring *ring)
1541 amdgpu_ring_write(ring, VCN_ENC_CMD_END);
1545 * vcn_v2_0_enc_ring_emit_ib - enc execute indirect buffer
1547 * @ring: amdgpu_ring pointer
1548 * @ib: indirect buffer to execute
1550 * Write enc ring commands to execute the indirect buffer
1552 void vcn_v2_0_enc_ring_emit_ib(struct amdgpu_ring *ring,
1553 struct amdgpu_job *job,
1554 struct amdgpu_ib *ib,
1555 uint32_t flags)
1557 unsigned vmid = AMDGPU_JOB_GET_VMID(job);
1559 amdgpu_ring_write(ring, VCN_ENC_CMD_IB);
1560 amdgpu_ring_write(ring, vmid);
1561 amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr));
1562 amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr));
1563 amdgpu_ring_write(ring, ib->length_dw);
1566 void vcn_v2_0_enc_ring_emit_reg_wait(struct amdgpu_ring *ring, uint32_t reg,
1567 uint32_t val, uint32_t mask)
1569 amdgpu_ring_write(ring, VCN_ENC_CMD_REG_WAIT);
1570 amdgpu_ring_write(ring, reg << 2);
1571 amdgpu_ring_write(ring, mask);
1572 amdgpu_ring_write(ring, val);
1575 void vcn_v2_0_enc_ring_emit_vm_flush(struct amdgpu_ring *ring,
1576 unsigned int vmid, uint64_t pd_addr)
1578 struct amdgpu_vmhub *hub = &ring->adev->vmhub[ring->funcs->vmhub];
1580 pd_addr = amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr);
1582 /* wait for reg writes */
1583 vcn_v2_0_enc_ring_emit_reg_wait(ring, hub->ctx0_ptb_addr_lo32 + vmid * 2,
1584 lower_32_bits(pd_addr), 0xffffffff);
1587 void vcn_v2_0_enc_ring_emit_wreg(struct amdgpu_ring *ring, uint32_t reg, uint32_t val)
1589 amdgpu_ring_write(ring, VCN_ENC_CMD_REG_WRITE);
1590 amdgpu_ring_write(ring, reg << 2);
1591 amdgpu_ring_write(ring, val);
1594 static int vcn_v2_0_set_interrupt_state(struct amdgpu_device *adev,
1595 struct amdgpu_irq_src *source,
1596 unsigned type,
1597 enum amdgpu_interrupt_state state)
1599 return 0;
1602 static int vcn_v2_0_process_interrupt(struct amdgpu_device *adev,
1603 struct amdgpu_irq_src *source,
1604 struct amdgpu_iv_entry *entry)
1606 DRM_DEBUG("IH: VCN TRAP\n");
1608 switch (entry->src_id) {
1609 case VCN_2_0__SRCID__UVD_SYSTEM_MESSAGE_INTERRUPT:
1610 amdgpu_fence_process(&adev->vcn.inst->ring_dec);
1611 break;
1612 case VCN_2_0__SRCID__UVD_ENC_GENERAL_PURPOSE:
1613 amdgpu_fence_process(&adev->vcn.inst->ring_enc[0]);
1614 break;
1615 case VCN_2_0__SRCID__UVD_ENC_LOW_LATENCY:
1616 amdgpu_fence_process(&adev->vcn.inst->ring_enc[1]);
1617 break;
1618 default:
1619 DRM_ERROR("Unhandled interrupt: %d %d\n",
1620 entry->src_id, entry->src_data[0]);
1621 break;
1624 return 0;
1627 static int vcn_v2_0_dec_ring_test_ring(struct amdgpu_ring *ring)
1629 struct amdgpu_device *adev = ring->adev;
1630 uint32_t tmp = 0;
1631 unsigned i;
1632 int r;
1634 WREG32(adev->vcn.inst[ring->me].external.scratch9, 0xCAFEDEAD);
1635 r = amdgpu_ring_alloc(ring, 4);
1636 if (r)
1637 return r;
1638 amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.cmd, 0));
1639 amdgpu_ring_write(ring, VCN_DEC_KMD_CMD | (VCN_DEC_CMD_PACKET_START << 1));
1640 amdgpu_ring_write(ring, PACKET0(adev->vcn.internal.scratch9, 0));
1641 amdgpu_ring_write(ring, 0xDEADBEEF);
1642 amdgpu_ring_commit(ring);
1643 for (i = 0; i < adev->usec_timeout; i++) {
1644 tmp = RREG32(adev->vcn.inst[ring->me].external.scratch9);
1645 if (tmp == 0xDEADBEEF)
1646 break;
1647 udelay(1);
1650 if (i >= adev->usec_timeout)
1651 r = -ETIMEDOUT;
1653 return r;
1657 static int vcn_v2_0_set_powergating_state(void *handle,
1658 enum amd_powergating_state state)
1660 /* This doesn't actually powergate the VCN block.
1661 * That's done in the dpm code via the SMC. This
1662 * just re-inits the block as necessary. The actual
1663 * gating still happens in the dpm code. We should
1664 * revisit this when there is a cleaner line between
1665 * the smc and the hw blocks
1667 int ret;
1668 struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1670 if (state == adev->vcn.cur_state)
1671 return 0;
1673 if (state == AMD_PG_STATE_GATE)
1674 ret = vcn_v2_0_stop(adev);
1675 else
1676 ret = vcn_v2_0_start(adev);
1678 if (!ret)
1679 adev->vcn.cur_state = state;
1680 return ret;
1683 static const struct amd_ip_funcs vcn_v2_0_ip_funcs = {
1684 .name = "vcn_v2_0",
1685 .early_init = vcn_v2_0_early_init,
1686 .late_init = NULL,
1687 .sw_init = vcn_v2_0_sw_init,
1688 .sw_fini = vcn_v2_0_sw_fini,
1689 .hw_init = vcn_v2_0_hw_init,
1690 .hw_fini = vcn_v2_0_hw_fini,
1691 .suspend = vcn_v2_0_suspend,
1692 .resume = vcn_v2_0_resume,
1693 .is_idle = vcn_v2_0_is_idle,
1694 .wait_for_idle = vcn_v2_0_wait_for_idle,
1695 .check_soft_reset = NULL,
1696 .pre_soft_reset = NULL,
1697 .soft_reset = NULL,
1698 .post_soft_reset = NULL,
1699 .set_clockgating_state = vcn_v2_0_set_clockgating_state,
1700 .set_powergating_state = vcn_v2_0_set_powergating_state,
1703 static const struct amdgpu_ring_funcs vcn_v2_0_dec_ring_vm_funcs = {
1704 .type = AMDGPU_RING_TYPE_VCN_DEC,
1705 .align_mask = 0xf,
1706 .vmhub = AMDGPU_MMHUB_0,
1707 .get_rptr = vcn_v2_0_dec_ring_get_rptr,
1708 .get_wptr = vcn_v2_0_dec_ring_get_wptr,
1709 .set_wptr = vcn_v2_0_dec_ring_set_wptr,
1710 .emit_frame_size =
1711 SOC15_FLUSH_GPU_TLB_NUM_WREG * 6 +
1712 SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 8 +
1713 8 + /* vcn_v2_0_dec_ring_emit_vm_flush */
1714 14 + 14 + /* vcn_v2_0_dec_ring_emit_fence x2 vm fence */
1716 .emit_ib_size = 8, /* vcn_v2_0_dec_ring_emit_ib */
1717 .emit_ib = vcn_v2_0_dec_ring_emit_ib,
1718 .emit_fence = vcn_v2_0_dec_ring_emit_fence,
1719 .emit_vm_flush = vcn_v2_0_dec_ring_emit_vm_flush,
1720 .test_ring = vcn_v2_0_dec_ring_test_ring,
1721 .test_ib = amdgpu_vcn_dec_ring_test_ib,
1722 .insert_nop = vcn_v2_0_dec_ring_insert_nop,
1723 .insert_start = vcn_v2_0_dec_ring_insert_start,
1724 .insert_end = vcn_v2_0_dec_ring_insert_end,
1725 .pad_ib = amdgpu_ring_generic_pad_ib,
1726 .begin_use = amdgpu_vcn_ring_begin_use,
1727 .end_use = amdgpu_vcn_ring_end_use,
1728 .emit_wreg = vcn_v2_0_dec_ring_emit_wreg,
1729 .emit_reg_wait = vcn_v2_0_dec_ring_emit_reg_wait,
1730 .emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
1733 static const struct amdgpu_ring_funcs vcn_v2_0_enc_ring_vm_funcs = {
1734 .type = AMDGPU_RING_TYPE_VCN_ENC,
1735 .align_mask = 0x3f,
1736 .nop = VCN_ENC_CMD_NO_OP,
1737 .vmhub = AMDGPU_MMHUB_0,
1738 .get_rptr = vcn_v2_0_enc_ring_get_rptr,
1739 .get_wptr = vcn_v2_0_enc_ring_get_wptr,
1740 .set_wptr = vcn_v2_0_enc_ring_set_wptr,
1741 .emit_frame_size =
1742 SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
1743 SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 4 +
1744 4 + /* vcn_v2_0_enc_ring_emit_vm_flush */
1745 5 + 5 + /* vcn_v2_0_enc_ring_emit_fence x2 vm fence */
1746 1, /* vcn_v2_0_enc_ring_insert_end */
1747 .emit_ib_size = 5, /* vcn_v2_0_enc_ring_emit_ib */
1748 .emit_ib = vcn_v2_0_enc_ring_emit_ib,
1749 .emit_fence = vcn_v2_0_enc_ring_emit_fence,
1750 .emit_vm_flush = vcn_v2_0_enc_ring_emit_vm_flush,
1751 .test_ring = amdgpu_vcn_enc_ring_test_ring,
1752 .test_ib = amdgpu_vcn_enc_ring_test_ib,
1753 .insert_nop = amdgpu_ring_insert_nop,
1754 .insert_end = vcn_v2_0_enc_ring_insert_end,
1755 .pad_ib = amdgpu_ring_generic_pad_ib,
1756 .begin_use = amdgpu_vcn_ring_begin_use,
1757 .end_use = amdgpu_vcn_ring_end_use,
1758 .emit_wreg = vcn_v2_0_enc_ring_emit_wreg,
1759 .emit_reg_wait = vcn_v2_0_enc_ring_emit_reg_wait,
1760 .emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
1763 static void vcn_v2_0_set_dec_ring_funcs(struct amdgpu_device *adev)
1765 adev->vcn.inst->ring_dec.funcs = &vcn_v2_0_dec_ring_vm_funcs;
1766 DRM_INFO("VCN decode is enabled in VM mode\n");
1769 static void vcn_v2_0_set_enc_ring_funcs(struct amdgpu_device *adev)
1771 int i;
1773 for (i = 0; i < adev->vcn.num_enc_rings; ++i)
1774 adev->vcn.inst->ring_enc[i].funcs = &vcn_v2_0_enc_ring_vm_funcs;
1776 DRM_INFO("VCN encode is enabled in VM mode\n");
1779 static const struct amdgpu_irq_src_funcs vcn_v2_0_irq_funcs = {
1780 .set = vcn_v2_0_set_interrupt_state,
1781 .process = vcn_v2_0_process_interrupt,
1784 static void vcn_v2_0_set_irq_funcs(struct amdgpu_device *adev)
1786 adev->vcn.inst->irq.num_types = adev->vcn.num_enc_rings + 1;
1787 adev->vcn.inst->irq.funcs = &vcn_v2_0_irq_funcs;
1790 const struct amdgpu_ip_block_version vcn_v2_0_ip_block =
1792 .type = AMD_IP_BLOCK_TYPE_VCN,
1793 .major = 2,
1794 .minor = 0,
1795 .rev = 0,
1796 .funcs = &vcn_v2_0_ip_funcs,