xref: /openbmc/linux/drivers/gpu/drm/amd/amdgpu/soc15.c (revision 55b37d9c)
1 /*
2  * Copyright 2016 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 #include <linux/firmware.h>
24 #include <linux/slab.h>
25 #include <linux/module.h>
26 #include <linux/pci.h>
27 
28 #include <drm/amdgpu_drm.h>
29 
30 #include "amdgpu.h"
31 #include "amdgpu_atombios.h"
32 #include "amdgpu_ih.h"
33 #include "amdgpu_uvd.h"
34 #include "amdgpu_vce.h"
35 #include "amdgpu_ucode.h"
36 #include "amdgpu_psp.h"
37 #include "atom.h"
38 #include "amd_pcie.h"
39 
40 #include "uvd/uvd_7_0_offset.h"
41 #include "gc/gc_9_0_offset.h"
42 #include "gc/gc_9_0_sh_mask.h"
43 #include "sdma0/sdma0_4_0_offset.h"
44 #include "sdma1/sdma1_4_0_offset.h"
45 #include "nbio/nbio_7_0_default.h"
46 #include "nbio/nbio_7_0_offset.h"
47 #include "nbio/nbio_7_0_sh_mask.h"
48 #include "nbio/nbio_7_0_smn.h"
49 #include "mp/mp_9_0_offset.h"
50 
51 #include "soc15.h"
52 #include "soc15_common.h"
53 #include "gfx_v9_0.h"
54 #include "gmc_v9_0.h"
55 #include "gfxhub_v1_0.h"
56 #include "mmhub_v1_0.h"
57 #include "df_v1_7.h"
58 #include "df_v3_6.h"
59 #include "nbio_v6_1.h"
60 #include "nbio_v7_0.h"
61 #include "nbio_v7_4.h"
62 #include "hdp_v4_0.h"
63 #include "vega10_ih.h"
64 #include "vega20_ih.h"
65 #include "navi10_ih.h"
66 #include "sdma_v4_0.h"
67 #include "uvd_v7_0.h"
68 #include "vce_v4_0.h"
69 #include "vcn_v1_0.h"
70 #include "vcn_v2_0.h"
71 #include "jpeg_v2_0.h"
72 #include "vcn_v2_5.h"
73 #include "jpeg_v2_5.h"
74 #include "smuio_v9_0.h"
75 #include "smuio_v11_0.h"
76 #include "smuio_v13_0.h"
77 #include "amdgpu_vkms.h"
78 #include "mxgpu_ai.h"
79 #include "amdgpu_ras.h"
80 #include "amdgpu_xgmi.h"
81 #include <uapi/linux/kfd_ioctl.h>
82 
83 #define mmMP0_MISC_CGTT_CTRL0                                                                   0x01b9
84 #define mmMP0_MISC_CGTT_CTRL0_BASE_IDX                                                          0
85 #define mmMP0_MISC_LIGHT_SLEEP_CTRL                                                             0x01ba
86 #define mmMP0_MISC_LIGHT_SLEEP_CTRL_BASE_IDX                                                    0
87 
88 static const struct amd_ip_funcs soc15_common_ip_funcs;
89 
90 /* Vega, Raven, Arcturus */
91 static const struct amdgpu_video_codec_info vega_video_codecs_encode_array[] =
92 {
93 	{codec_info_build(AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG4_AVC, 4096, 2304, 0)},
94 	{codec_info_build(AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_HEVC, 4096, 2304, 0)},
95 };
96 
97 static const struct amdgpu_video_codecs vega_video_codecs_encode =
98 {
99 	.codec_count = ARRAY_SIZE(vega_video_codecs_encode_array),
100 	.codec_array = vega_video_codecs_encode_array,
101 };
102 
103 /* Vega */
104 static const struct amdgpu_video_codec_info vega_video_codecs_decode_array[] =
105 {
106 	{codec_info_build(AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG2, 4096, 4096, 3)},
107 	{codec_info_build(AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG4, 4096, 4096, 5)},
108 	{codec_info_build(AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG4_AVC, 4096, 4096, 52)},
109 	{codec_info_build(AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_VC1, 4096, 4096, 4)},
110 	{codec_info_build(AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_HEVC, 4096, 4096, 186)},
111 	{codec_info_build(AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_JPEG, 4096, 4096, 0)},
112 };
113 
114 static const struct amdgpu_video_codecs vega_video_codecs_decode =
115 {
116 	.codec_count = ARRAY_SIZE(vega_video_codecs_decode_array),
117 	.codec_array = vega_video_codecs_decode_array,
118 };
119 
120 /* Raven */
121 static const struct amdgpu_video_codec_info rv_video_codecs_decode_array[] =
122 {
123 	{codec_info_build(AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG2, 4096, 4096, 3)},
124 	{codec_info_build(AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG4, 4096, 4096, 5)},
125 	{codec_info_build(AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG4_AVC, 4096, 4096, 52)},
126 	{codec_info_build(AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_VC1, 4096, 4096, 4)},
127 	{codec_info_build(AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_HEVC, 4096, 4096, 186)},
128 	{codec_info_build(AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_JPEG, 4096, 4096, 0)},
129 	{codec_info_build(AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_VP9, 4096, 4096, 0)},
130 };
131 
132 static const struct amdgpu_video_codecs rv_video_codecs_decode =
133 {
134 	.codec_count = ARRAY_SIZE(rv_video_codecs_decode_array),
135 	.codec_array = rv_video_codecs_decode_array,
136 };
137 
138 /* Renoir, Arcturus */
139 static const struct amdgpu_video_codec_info rn_video_codecs_decode_array[] =
140 {
141 	{codec_info_build(AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG2, 4096, 4096, 3)},
142 	{codec_info_build(AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG4, 4096, 4096, 5)},
143 	{codec_info_build(AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_MPEG4_AVC, 4096, 4096, 52)},
144 	{codec_info_build(AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_VC1, 4096, 4096, 4)},
145 	{codec_info_build(AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_HEVC, 8192, 4352, 186)},
146 	{codec_info_build(AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_JPEG, 4096, 4096, 0)},
147 	{codec_info_build(AMDGPU_INFO_VIDEO_CAPS_CODEC_IDX_VP9, 8192, 4352, 0)},
148 };
149 
150 static const struct amdgpu_video_codecs rn_video_codecs_decode =
151 {
152 	.codec_count = ARRAY_SIZE(rn_video_codecs_decode_array),
153 	.codec_array = rn_video_codecs_decode_array,
154 };
155 
156 static int soc15_query_video_codecs(struct amdgpu_device *adev, bool encode,
157 				    const struct amdgpu_video_codecs **codecs)
158 {
159 	if (adev->ip_versions[VCE_HWIP][0]) {
160 		switch (adev->ip_versions[VCE_HWIP][0]) {
161 		case IP_VERSION(4, 0, 0):
162 		case IP_VERSION(4, 1, 0):
163 			if (encode)
164 				*codecs = &vega_video_codecs_encode;
165 			else
166 				*codecs = &vega_video_codecs_decode;
167 			return 0;
168 		default:
169 			return -EINVAL;
170 		}
171 	} else {
172 		switch (adev->ip_versions[UVD_HWIP][0]) {
173 		case IP_VERSION(1, 0, 0):
174 		case IP_VERSION(1, 0, 1):
175 			if (encode)
176 				*codecs = &vega_video_codecs_encode;
177 			else
178 				*codecs = &rv_video_codecs_decode;
179 			return 0;
180 		case IP_VERSION(2, 5, 0):
181 		case IP_VERSION(2, 6, 0):
182 		case IP_VERSION(2, 2, 0):
183 			if (encode)
184 				*codecs = &vega_video_codecs_encode;
185 			else
186 				*codecs = &rn_video_codecs_decode;
187 			return 0;
188 		default:
189 			return -EINVAL;
190 		}
191 	}
192 }
193 
194 static u32 soc15_uvd_ctx_rreg(struct amdgpu_device *adev, u32 reg)
195 {
196 	unsigned long flags, address, data;
197 	u32 r;
198 
199 	address = SOC15_REG_OFFSET(UVD, 0, mmUVD_CTX_INDEX);
200 	data = SOC15_REG_OFFSET(UVD, 0, mmUVD_CTX_DATA);
201 
202 	spin_lock_irqsave(&adev->uvd_ctx_idx_lock, flags);
203 	WREG32(address, ((reg) & 0x1ff));
204 	r = RREG32(data);
205 	spin_unlock_irqrestore(&adev->uvd_ctx_idx_lock, flags);
206 	return r;
207 }
208 
209 static void soc15_uvd_ctx_wreg(struct amdgpu_device *adev, u32 reg, u32 v)
210 {
211 	unsigned long flags, address, data;
212 
213 	address = SOC15_REG_OFFSET(UVD, 0, mmUVD_CTX_INDEX);
214 	data = SOC15_REG_OFFSET(UVD, 0, mmUVD_CTX_DATA);
215 
216 	spin_lock_irqsave(&adev->uvd_ctx_idx_lock, flags);
217 	WREG32(address, ((reg) & 0x1ff));
218 	WREG32(data, (v));
219 	spin_unlock_irqrestore(&adev->uvd_ctx_idx_lock, flags);
220 }
221 
222 static u32 soc15_didt_rreg(struct amdgpu_device *adev, u32 reg)
223 {
224 	unsigned long flags, address, data;
225 	u32 r;
226 
227 	address = SOC15_REG_OFFSET(GC, 0, mmDIDT_IND_INDEX);
228 	data = SOC15_REG_OFFSET(GC, 0, mmDIDT_IND_DATA);
229 
230 	spin_lock_irqsave(&adev->didt_idx_lock, flags);
231 	WREG32(address, (reg));
232 	r = RREG32(data);
233 	spin_unlock_irqrestore(&adev->didt_idx_lock, flags);
234 	return r;
235 }
236 
237 static void soc15_didt_wreg(struct amdgpu_device *adev, u32 reg, u32 v)
238 {
239 	unsigned long flags, address, data;
240 
241 	address = SOC15_REG_OFFSET(GC, 0, mmDIDT_IND_INDEX);
242 	data = SOC15_REG_OFFSET(GC, 0, mmDIDT_IND_DATA);
243 
244 	spin_lock_irqsave(&adev->didt_idx_lock, flags);
245 	WREG32(address, (reg));
246 	WREG32(data, (v));
247 	spin_unlock_irqrestore(&adev->didt_idx_lock, flags);
248 }
249 
250 static u32 soc15_gc_cac_rreg(struct amdgpu_device *adev, u32 reg)
251 {
252 	unsigned long flags;
253 	u32 r;
254 
255 	spin_lock_irqsave(&adev->gc_cac_idx_lock, flags);
256 	WREG32_SOC15(GC, 0, mmGC_CAC_IND_INDEX, (reg));
257 	r = RREG32_SOC15(GC, 0, mmGC_CAC_IND_DATA);
258 	spin_unlock_irqrestore(&adev->gc_cac_idx_lock, flags);
259 	return r;
260 }
261 
262 static void soc15_gc_cac_wreg(struct amdgpu_device *adev, u32 reg, u32 v)
263 {
264 	unsigned long flags;
265 
266 	spin_lock_irqsave(&adev->gc_cac_idx_lock, flags);
267 	WREG32_SOC15(GC, 0, mmGC_CAC_IND_INDEX, (reg));
268 	WREG32_SOC15(GC, 0, mmGC_CAC_IND_DATA, (v));
269 	spin_unlock_irqrestore(&adev->gc_cac_idx_lock, flags);
270 }
271 
272 static u32 soc15_se_cac_rreg(struct amdgpu_device *adev, u32 reg)
273 {
274 	unsigned long flags;
275 	u32 r;
276 
277 	spin_lock_irqsave(&adev->se_cac_idx_lock, flags);
278 	WREG32_SOC15(GC, 0, mmSE_CAC_IND_INDEX, (reg));
279 	r = RREG32_SOC15(GC, 0, mmSE_CAC_IND_DATA);
280 	spin_unlock_irqrestore(&adev->se_cac_idx_lock, flags);
281 	return r;
282 }
283 
284 static void soc15_se_cac_wreg(struct amdgpu_device *adev, u32 reg, u32 v)
285 {
286 	unsigned long flags;
287 
288 	spin_lock_irqsave(&adev->se_cac_idx_lock, flags);
289 	WREG32_SOC15(GC, 0, mmSE_CAC_IND_INDEX, (reg));
290 	WREG32_SOC15(GC, 0, mmSE_CAC_IND_DATA, (v));
291 	spin_unlock_irqrestore(&adev->se_cac_idx_lock, flags);
292 }
293 
294 static u32 soc15_get_config_memsize(struct amdgpu_device *adev)
295 {
296 	return adev->nbio.funcs->get_memsize(adev);
297 }
298 
299 static u32 soc15_get_xclk(struct amdgpu_device *adev)
300 {
301 	u32 reference_clock = adev->clock.spll.reference_freq;
302 
303 	if (adev->ip_versions[MP1_HWIP][0] == IP_VERSION(12, 0, 0) ||
304 	    adev->ip_versions[MP1_HWIP][0] == IP_VERSION(12, 0, 1) ||
305 	    adev->ip_versions[MP1_HWIP][0] == IP_VERSION(10, 0, 0) ||
306 	    adev->ip_versions[MP1_HWIP][0] == IP_VERSION(10, 0, 1))
307 		return 10000;
308 
309 	return reference_clock;
310 }
311 
312 
313 void soc15_grbm_select(struct amdgpu_device *adev,
314 		     u32 me, u32 pipe, u32 queue, u32 vmid)
315 {
316 	u32 grbm_gfx_cntl = 0;
317 	grbm_gfx_cntl = REG_SET_FIELD(grbm_gfx_cntl, GRBM_GFX_CNTL, PIPEID, pipe);
318 	grbm_gfx_cntl = REG_SET_FIELD(grbm_gfx_cntl, GRBM_GFX_CNTL, MEID, me);
319 	grbm_gfx_cntl = REG_SET_FIELD(grbm_gfx_cntl, GRBM_GFX_CNTL, VMID, vmid);
320 	grbm_gfx_cntl = REG_SET_FIELD(grbm_gfx_cntl, GRBM_GFX_CNTL, QUEUEID, queue);
321 
322 	WREG32_SOC15_RLC_SHADOW(GC, 0, mmGRBM_GFX_CNTL, grbm_gfx_cntl);
323 }
324 
325 static void soc15_vga_set_state(struct amdgpu_device *adev, bool state)
326 {
327 	/* todo */
328 }
329 
330 static bool soc15_read_disabled_bios(struct amdgpu_device *adev)
331 {
332 	/* todo */
333 	return false;
334 }
335 
336 static struct soc15_allowed_register_entry soc15_allowed_read_registers[] = {
337 	{ SOC15_REG_ENTRY(GC, 0, mmGRBM_STATUS)},
338 	{ SOC15_REG_ENTRY(GC, 0, mmGRBM_STATUS2)},
339 	{ SOC15_REG_ENTRY(GC, 0, mmGRBM_STATUS_SE0)},
340 	{ SOC15_REG_ENTRY(GC, 0, mmGRBM_STATUS_SE1)},
341 	{ SOC15_REG_ENTRY(GC, 0, mmGRBM_STATUS_SE2)},
342 	{ SOC15_REG_ENTRY(GC, 0, mmGRBM_STATUS_SE3)},
343 	{ SOC15_REG_ENTRY(SDMA0, 0, mmSDMA0_STATUS_REG)},
344 	{ SOC15_REG_ENTRY(SDMA1, 0, mmSDMA1_STATUS_REG)},
345 	{ SOC15_REG_ENTRY(GC, 0, mmCP_STAT)},
346 	{ SOC15_REG_ENTRY(GC, 0, mmCP_STALLED_STAT1)},
347 	{ SOC15_REG_ENTRY(GC, 0, mmCP_STALLED_STAT2)},
348 	{ SOC15_REG_ENTRY(GC, 0, mmCP_STALLED_STAT3)},
349 	{ SOC15_REG_ENTRY(GC, 0, mmCP_CPF_BUSY_STAT)},
350 	{ SOC15_REG_ENTRY(GC, 0, mmCP_CPF_STALLED_STAT1)},
351 	{ SOC15_REG_ENTRY(GC, 0, mmCP_CPF_STATUS)},
352 	{ SOC15_REG_ENTRY(GC, 0, mmCP_CPC_BUSY_STAT)},
353 	{ SOC15_REG_ENTRY(GC, 0, mmCP_CPC_STALLED_STAT1)},
354 	{ SOC15_REG_ENTRY(GC, 0, mmCP_CPC_STATUS)},
355 	{ SOC15_REG_ENTRY(GC, 0, mmGB_ADDR_CONFIG)},
356 	{ SOC15_REG_ENTRY(GC, 0, mmDB_DEBUG2)},
357 };
358 
359 static uint32_t soc15_read_indexed_register(struct amdgpu_device *adev, u32 se_num,
360 					 u32 sh_num, u32 reg_offset)
361 {
362 	uint32_t val;
363 
364 	mutex_lock(&adev->grbm_idx_mutex);
365 	if (se_num != 0xffffffff || sh_num != 0xffffffff)
366 		amdgpu_gfx_select_se_sh(adev, se_num, sh_num, 0xffffffff);
367 
368 	val = RREG32(reg_offset);
369 
370 	if (se_num != 0xffffffff || sh_num != 0xffffffff)
371 		amdgpu_gfx_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
372 	mutex_unlock(&adev->grbm_idx_mutex);
373 	return val;
374 }
375 
376 static uint32_t soc15_get_register_value(struct amdgpu_device *adev,
377 					 bool indexed, u32 se_num,
378 					 u32 sh_num, u32 reg_offset)
379 {
380 	if (indexed) {
381 		return soc15_read_indexed_register(adev, se_num, sh_num, reg_offset);
382 	} else {
383 		if (reg_offset == SOC15_REG_OFFSET(GC, 0, mmGB_ADDR_CONFIG))
384 			return adev->gfx.config.gb_addr_config;
385 		else if (reg_offset == SOC15_REG_OFFSET(GC, 0, mmDB_DEBUG2))
386 			return adev->gfx.config.db_debug2;
387 		return RREG32(reg_offset);
388 	}
389 }
390 
391 static int soc15_read_register(struct amdgpu_device *adev, u32 se_num,
392 			    u32 sh_num, u32 reg_offset, u32 *value)
393 {
394 	uint32_t i;
395 	struct soc15_allowed_register_entry  *en;
396 
397 	*value = 0;
398 	for (i = 0; i < ARRAY_SIZE(soc15_allowed_read_registers); i++) {
399 		en = &soc15_allowed_read_registers[i];
400 		if (!adev->reg_offset[en->hwip][en->inst])
401 			continue;
402 		else if (reg_offset != (adev->reg_offset[en->hwip][en->inst][en->seg]
403 					+ en->reg_offset))
404 			continue;
405 
406 		*value = soc15_get_register_value(adev,
407 						  soc15_allowed_read_registers[i].grbm_indexed,
408 						  se_num, sh_num, reg_offset);
409 		return 0;
410 	}
411 	return -EINVAL;
412 }
413 
414 
415 /**
416  * soc15_program_register_sequence - program an array of registers.
417  *
418  * @adev: amdgpu_device pointer
419  * @regs: pointer to the register array
420  * @array_size: size of the register array
421  *
422  * Programs an array or registers with and and or masks.
423  * This is a helper for setting golden registers.
424  */
425 
426 void soc15_program_register_sequence(struct amdgpu_device *adev,
427 					     const struct soc15_reg_golden *regs,
428 					     const u32 array_size)
429 {
430 	const struct soc15_reg_golden *entry;
431 	u32 tmp, reg;
432 	int i;
433 
434 	for (i = 0; i < array_size; ++i) {
435 		entry = &regs[i];
436 		reg =  adev->reg_offset[entry->hwip][entry->instance][entry->segment] + entry->reg;
437 
438 		if (entry->and_mask == 0xffffffff) {
439 			tmp = entry->or_mask;
440 		} else {
441 			tmp = (entry->hwip == GC_HWIP) ?
442 				RREG32_SOC15_IP(GC, reg) : RREG32(reg);
443 
444 			tmp &= ~(entry->and_mask);
445 			tmp |= (entry->or_mask & entry->and_mask);
446 		}
447 
448 		if (reg == SOC15_REG_OFFSET(GC, 0, mmPA_SC_BINNER_EVENT_CNTL_3) ||
449 			reg == SOC15_REG_OFFSET(GC, 0, mmPA_SC_ENHANCE) ||
450 			reg == SOC15_REG_OFFSET(GC, 0, mmPA_SC_ENHANCE_1) ||
451 			reg == SOC15_REG_OFFSET(GC, 0, mmSH_MEM_CONFIG))
452 			WREG32_RLC(reg, tmp);
453 		else
454 			(entry->hwip == GC_HWIP) ?
455 				WREG32_SOC15_IP(GC, reg, tmp) : WREG32(reg, tmp);
456 
457 	}
458 
459 }
460 
461 static int soc15_asic_baco_reset(struct amdgpu_device *adev)
462 {
463 	struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
464 	int ret = 0;
465 
466 	/* avoid NBIF got stuck when do RAS recovery in BACO reset */
467 	if (ras && adev->ras_enabled)
468 		adev->nbio.funcs->enable_doorbell_interrupt(adev, false);
469 
470 	ret = amdgpu_dpm_baco_reset(adev);
471 	if (ret)
472 		return ret;
473 
474 	/* re-enable doorbell interrupt after BACO exit */
475 	if (ras && adev->ras_enabled)
476 		adev->nbio.funcs->enable_doorbell_interrupt(adev, true);
477 
478 	return 0;
479 }
480 
481 static enum amd_reset_method
482 soc15_asic_reset_method(struct amdgpu_device *adev)
483 {
484 	bool baco_reset = false;
485 	bool connected_to_cpu = false;
486 	struct amdgpu_ras *ras = amdgpu_ras_get_context(adev);
487 
488         if (adev->gmc.xgmi.supported && adev->gmc.xgmi.connected_to_cpu)
489                 connected_to_cpu = true;
490 
491 	if (amdgpu_reset_method == AMD_RESET_METHOD_MODE1 ||
492 	    amdgpu_reset_method == AMD_RESET_METHOD_MODE2 ||
493 	    amdgpu_reset_method == AMD_RESET_METHOD_BACO ||
494 	    amdgpu_reset_method == AMD_RESET_METHOD_PCI) {
495 		/* If connected to cpu, driver only support mode2 */
496                 if (connected_to_cpu)
497                         return AMD_RESET_METHOD_MODE2;
498                 return amdgpu_reset_method;
499         }
500 
501 	if (amdgpu_reset_method != -1)
502 		dev_warn(adev->dev, "Specified reset method:%d isn't supported, using AUTO instead.\n",
503 				  amdgpu_reset_method);
504 
505 	switch (adev->ip_versions[MP1_HWIP][0]) {
506 	case IP_VERSION(10, 0, 0):
507 	case IP_VERSION(10, 0, 1):
508 	case IP_VERSION(12, 0, 0):
509 	case IP_VERSION(12, 0, 1):
510 		return AMD_RESET_METHOD_MODE2;
511 	case IP_VERSION(9, 0, 0):
512 	case IP_VERSION(11, 0, 2):
513 		if (adev->asic_type == CHIP_VEGA20) {
514 			if (adev->psp.sos.fw_version >= 0x80067)
515 				baco_reset = amdgpu_dpm_is_baco_supported(adev);
516 			/*
517 			 * 1. PMFW version > 0x284300: all cases use baco
518 			 * 2. PMFW version <= 0x284300: only sGPU w/o RAS use baco
519 			 */
520 			if (ras && adev->ras_enabled &&
521 			    adev->pm.fw_version <= 0x283400)
522 				baco_reset = false;
523 		} else {
524 			baco_reset = amdgpu_dpm_is_baco_supported(adev);
525 		}
526 		break;
527 	case IP_VERSION(13, 0, 2):
528 		 /*
529 		 * 1.connected to cpu: driver issue mode2 reset
530 		 * 2.discret gpu: driver issue mode1 reset
531 		 */
532 		if (connected_to_cpu)
533 			return AMD_RESET_METHOD_MODE2;
534 		break;
535 	default:
536 		break;
537 	}
538 
539 	if (baco_reset)
540 		return AMD_RESET_METHOD_BACO;
541 	else
542 		return AMD_RESET_METHOD_MODE1;
543 }
544 
545 static int soc15_asic_reset(struct amdgpu_device *adev)
546 {
547 	/* original raven doesn't have full asic reset */
548 	if ((adev->apu_flags & AMD_APU_IS_RAVEN) ||
549 	    (adev->apu_flags & AMD_APU_IS_RAVEN2))
550 		return 0;
551 
552 	switch (soc15_asic_reset_method(adev)) {
553 	case AMD_RESET_METHOD_PCI:
554 		dev_info(adev->dev, "PCI reset\n");
555 		return amdgpu_device_pci_reset(adev);
556 	case AMD_RESET_METHOD_BACO:
557 		dev_info(adev->dev, "BACO reset\n");
558 		return soc15_asic_baco_reset(adev);
559 	case AMD_RESET_METHOD_MODE2:
560 		dev_info(adev->dev, "MODE2 reset\n");
561 		return amdgpu_dpm_mode2_reset(adev);
562 	default:
563 		dev_info(adev->dev, "MODE1 reset\n");
564 		return amdgpu_device_mode1_reset(adev);
565 	}
566 }
567 
568 static bool soc15_supports_baco(struct amdgpu_device *adev)
569 {
570 	switch (adev->ip_versions[MP1_HWIP][0]) {
571 	case IP_VERSION(9, 0, 0):
572 	case IP_VERSION(11, 0, 2):
573 		if (adev->asic_type == CHIP_VEGA20) {
574 			if (adev->psp.sos.fw_version >= 0x80067)
575 				return amdgpu_dpm_is_baco_supported(adev);
576 			return false;
577 		} else {
578 			return amdgpu_dpm_is_baco_supported(adev);
579 		}
580 		break;
581 	default:
582 		return false;
583 	}
584 }
585 
586 /*static int soc15_set_uvd_clock(struct amdgpu_device *adev, u32 clock,
587 			u32 cntl_reg, u32 status_reg)
588 {
589 	return 0;
590 }*/
591 
592 static int soc15_set_uvd_clocks(struct amdgpu_device *adev, u32 vclk, u32 dclk)
593 {
594 	/*int r;
595 
596 	r = soc15_set_uvd_clock(adev, vclk, ixCG_VCLK_CNTL, ixCG_VCLK_STATUS);
597 	if (r)
598 		return r;
599 
600 	r = soc15_set_uvd_clock(adev, dclk, ixCG_DCLK_CNTL, ixCG_DCLK_STATUS);
601 	*/
602 	return 0;
603 }
604 
605 static int soc15_set_vce_clocks(struct amdgpu_device *adev, u32 evclk, u32 ecclk)
606 {
607 	/* todo */
608 
609 	return 0;
610 }
611 
612 static void soc15_program_aspm(struct amdgpu_device *adev)
613 {
614 	if (!amdgpu_device_should_use_aspm(adev))
615 		return;
616 
617 	if (!(adev->flags & AMD_IS_APU) &&
618 	    (adev->nbio.funcs->program_aspm))
619 		adev->nbio.funcs->program_aspm(adev);
620 }
621 
622 const struct amdgpu_ip_block_version vega10_common_ip_block =
623 {
624 	.type = AMD_IP_BLOCK_TYPE_COMMON,
625 	.major = 2,
626 	.minor = 0,
627 	.rev = 0,
628 	.funcs = &soc15_common_ip_funcs,
629 };
630 
631 static void soc15_reg_base_init(struct amdgpu_device *adev)
632 {
633 	/* Set IP register base before any HW register access */
634 	switch (adev->asic_type) {
635 	case CHIP_VEGA10:
636 	case CHIP_VEGA12:
637 	case CHIP_RAVEN:
638 	case CHIP_RENOIR:
639 		vega10_reg_base_init(adev);
640 		break;
641 	case CHIP_VEGA20:
642 		vega20_reg_base_init(adev);
643 		break;
644 	case CHIP_ARCTURUS:
645 		arct_reg_base_init(adev);
646 		break;
647 	case CHIP_ALDEBARAN:
648 		aldebaran_reg_base_init(adev);
649 		break;
650 	default:
651 		DRM_ERROR("Unsupported asic type: %d!\n", adev->asic_type);
652 		break;
653 	}
654 }
655 
656 void soc15_set_virt_ops(struct amdgpu_device *adev)
657 {
658 	adev->virt.ops = &xgpu_ai_virt_ops;
659 
660 	/* init soc15 reg base early enough so we can
661 	 * request request full access for sriov before
662 	 * set_ip_blocks. */
663 	soc15_reg_base_init(adev);
664 }
665 
666 static bool soc15_need_full_reset(struct amdgpu_device *adev)
667 {
668 	/* change this when we implement soft reset */
669 	return true;
670 }
671 
672 static void soc15_get_pcie_usage(struct amdgpu_device *adev, uint64_t *count0,
673 				 uint64_t *count1)
674 {
675 	uint32_t perfctr = 0;
676 	uint64_t cnt0_of, cnt1_of;
677 	int tmp;
678 
679 	/* This reports 0 on APUs, so return to avoid writing/reading registers
680 	 * that may or may not be different from their GPU counterparts
681 	 */
682 	if (adev->flags & AMD_IS_APU)
683 		return;
684 
685 	/* Set the 2 events that we wish to watch, defined above */
686 	/* Reg 40 is # received msgs */
687 	/* Reg 104 is # of posted requests sent */
688 	perfctr = REG_SET_FIELD(perfctr, PCIE_PERF_CNTL_TXCLK, EVENT0_SEL, 40);
689 	perfctr = REG_SET_FIELD(perfctr, PCIE_PERF_CNTL_TXCLK, EVENT1_SEL, 104);
690 
691 	/* Write to enable desired perf counters */
692 	WREG32_PCIE(smnPCIE_PERF_CNTL_TXCLK, perfctr);
693 	/* Zero out and enable the perf counters
694 	 * Write 0x5:
695 	 * Bit 0 = Start all counters(1)
696 	 * Bit 2 = Global counter reset enable(1)
697 	 */
698 	WREG32_PCIE(smnPCIE_PERF_COUNT_CNTL, 0x00000005);
699 
700 	msleep(1000);
701 
702 	/* Load the shadow and disable the perf counters
703 	 * Write 0x2:
704 	 * Bit 0 = Stop counters(0)
705 	 * Bit 1 = Load the shadow counters(1)
706 	 */
707 	WREG32_PCIE(smnPCIE_PERF_COUNT_CNTL, 0x00000002);
708 
709 	/* Read register values to get any >32bit overflow */
710 	tmp = RREG32_PCIE(smnPCIE_PERF_CNTL_TXCLK);
711 	cnt0_of = REG_GET_FIELD(tmp, PCIE_PERF_CNTL_TXCLK, COUNTER0_UPPER);
712 	cnt1_of = REG_GET_FIELD(tmp, PCIE_PERF_CNTL_TXCLK, COUNTER1_UPPER);
713 
714 	/* Get the values and add the overflow */
715 	*count0 = RREG32_PCIE(smnPCIE_PERF_COUNT0_TXCLK) | (cnt0_of << 32);
716 	*count1 = RREG32_PCIE(smnPCIE_PERF_COUNT1_TXCLK) | (cnt1_of << 32);
717 }
718 
719 static void vega20_get_pcie_usage(struct amdgpu_device *adev, uint64_t *count0,
720 				 uint64_t *count1)
721 {
722 	uint32_t perfctr = 0;
723 	uint64_t cnt0_of, cnt1_of;
724 	int tmp;
725 
726 	/* This reports 0 on APUs, so return to avoid writing/reading registers
727 	 * that may or may not be different from their GPU counterparts
728 	 */
729 	if (adev->flags & AMD_IS_APU)
730 		return;
731 
732 	/* Set the 2 events that we wish to watch, defined above */
733 	/* Reg 40 is # received msgs */
734 	/* Reg 108 is # of posted requests sent on VG20 */
735 	perfctr = REG_SET_FIELD(perfctr, PCIE_PERF_CNTL_TXCLK3,
736 				EVENT0_SEL, 40);
737 	perfctr = REG_SET_FIELD(perfctr, PCIE_PERF_CNTL_TXCLK3,
738 				EVENT1_SEL, 108);
739 
740 	/* Write to enable desired perf counters */
741 	WREG32_PCIE(smnPCIE_PERF_CNTL_TXCLK3, perfctr);
742 	/* Zero out and enable the perf counters
743 	 * Write 0x5:
744 	 * Bit 0 = Start all counters(1)
745 	 * Bit 2 = Global counter reset enable(1)
746 	 */
747 	WREG32_PCIE(smnPCIE_PERF_COUNT_CNTL, 0x00000005);
748 
749 	msleep(1000);
750 
751 	/* Load the shadow and disable the perf counters
752 	 * Write 0x2:
753 	 * Bit 0 = Stop counters(0)
754 	 * Bit 1 = Load the shadow counters(1)
755 	 */
756 	WREG32_PCIE(smnPCIE_PERF_COUNT_CNTL, 0x00000002);
757 
758 	/* Read register values to get any >32bit overflow */
759 	tmp = RREG32_PCIE(smnPCIE_PERF_CNTL_TXCLK3);
760 	cnt0_of = REG_GET_FIELD(tmp, PCIE_PERF_CNTL_TXCLK3, COUNTER0_UPPER);
761 	cnt1_of = REG_GET_FIELD(tmp, PCIE_PERF_CNTL_TXCLK3, COUNTER1_UPPER);
762 
763 	/* Get the values and add the overflow */
764 	*count0 = RREG32_PCIE(smnPCIE_PERF_COUNT0_TXCLK3) | (cnt0_of << 32);
765 	*count1 = RREG32_PCIE(smnPCIE_PERF_COUNT1_TXCLK3) | (cnt1_of << 32);
766 }
767 
768 static bool soc15_need_reset_on_init(struct amdgpu_device *adev)
769 {
770 	u32 sol_reg;
771 
772 	/* CP hangs in IGT reloading test on RN, reset to WA */
773 	if (adev->asic_type == CHIP_RENOIR)
774 		return true;
775 
776 	/* Just return false for soc15 GPUs.  Reset does not seem to
777 	 * be necessary.
778 	 */
779 	if (!amdgpu_passthrough(adev))
780 		return false;
781 
782 	if (adev->flags & AMD_IS_APU)
783 		return false;
784 
785 	/* Check sOS sign of life register to confirm sys driver and sOS
786 	 * are already been loaded.
787 	 */
788 	sol_reg = RREG32_SOC15(MP0, 0, mmMP0_SMN_C2PMSG_81);
789 	if (sol_reg)
790 		return true;
791 
792 	return false;
793 }
794 
795 static uint64_t soc15_get_pcie_replay_count(struct amdgpu_device *adev)
796 {
797 	uint64_t nak_r, nak_g;
798 
799 	/* Get the number of NAKs received and generated */
800 	nak_r = RREG32_PCIE(smnPCIE_RX_NUM_NAK);
801 	nak_g = RREG32_PCIE(smnPCIE_RX_NUM_NAK_GENERATED);
802 
803 	/* Add the total number of NAKs, i.e the number of replays */
804 	return (nak_r + nak_g);
805 }
806 
807 static void soc15_pre_asic_init(struct amdgpu_device *adev)
808 {
809 	gmc_v9_0_restore_registers(adev);
810 }
811 
812 static const struct amdgpu_asic_funcs soc15_asic_funcs =
813 {
814 	.read_disabled_bios = &soc15_read_disabled_bios,
815 	.read_bios_from_rom = &amdgpu_soc15_read_bios_from_rom,
816 	.read_register = &soc15_read_register,
817 	.reset = &soc15_asic_reset,
818 	.reset_method = &soc15_asic_reset_method,
819 	.set_vga_state = &soc15_vga_set_state,
820 	.get_xclk = &soc15_get_xclk,
821 	.set_uvd_clocks = &soc15_set_uvd_clocks,
822 	.set_vce_clocks = &soc15_set_vce_clocks,
823 	.get_config_memsize = &soc15_get_config_memsize,
824 	.need_full_reset = &soc15_need_full_reset,
825 	.init_doorbell_index = &vega10_doorbell_index_init,
826 	.get_pcie_usage = &soc15_get_pcie_usage,
827 	.need_reset_on_init = &soc15_need_reset_on_init,
828 	.get_pcie_replay_count = &soc15_get_pcie_replay_count,
829 	.supports_baco = &soc15_supports_baco,
830 	.pre_asic_init = &soc15_pre_asic_init,
831 	.query_video_codecs = &soc15_query_video_codecs,
832 };
833 
834 static const struct amdgpu_asic_funcs vega20_asic_funcs =
835 {
836 	.read_disabled_bios = &soc15_read_disabled_bios,
837 	.read_bios_from_rom = &amdgpu_soc15_read_bios_from_rom,
838 	.read_register = &soc15_read_register,
839 	.reset = &soc15_asic_reset,
840 	.reset_method = &soc15_asic_reset_method,
841 	.set_vga_state = &soc15_vga_set_state,
842 	.get_xclk = &soc15_get_xclk,
843 	.set_uvd_clocks = &soc15_set_uvd_clocks,
844 	.set_vce_clocks = &soc15_set_vce_clocks,
845 	.get_config_memsize = &soc15_get_config_memsize,
846 	.need_full_reset = &soc15_need_full_reset,
847 	.init_doorbell_index = &vega20_doorbell_index_init,
848 	.get_pcie_usage = &vega20_get_pcie_usage,
849 	.need_reset_on_init = &soc15_need_reset_on_init,
850 	.get_pcie_replay_count = &soc15_get_pcie_replay_count,
851 	.supports_baco = &soc15_supports_baco,
852 	.pre_asic_init = &soc15_pre_asic_init,
853 	.query_video_codecs = &soc15_query_video_codecs,
854 };
855 
856 static int soc15_common_early_init(void *handle)
857 {
858 #define MMIO_REG_HOLE_OFFSET (0x80000 - PAGE_SIZE)
859 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
860 
861 	if (!amdgpu_sriov_vf(adev)) {
862 		adev->rmmio_remap.reg_offset = MMIO_REG_HOLE_OFFSET;
863 		adev->rmmio_remap.bus_addr = adev->rmmio_base + MMIO_REG_HOLE_OFFSET;
864 	}
865 	adev->smc_rreg = NULL;
866 	adev->smc_wreg = NULL;
867 	adev->pcie_rreg = &amdgpu_device_indirect_rreg;
868 	adev->pcie_wreg = &amdgpu_device_indirect_wreg;
869 	adev->pcie_rreg64 = &amdgpu_device_indirect_rreg64;
870 	adev->pcie_wreg64 = &amdgpu_device_indirect_wreg64;
871 	adev->uvd_ctx_rreg = &soc15_uvd_ctx_rreg;
872 	adev->uvd_ctx_wreg = &soc15_uvd_ctx_wreg;
873 	adev->didt_rreg = &soc15_didt_rreg;
874 	adev->didt_wreg = &soc15_didt_wreg;
875 	adev->gc_cac_rreg = &soc15_gc_cac_rreg;
876 	adev->gc_cac_wreg = &soc15_gc_cac_wreg;
877 	adev->se_cac_rreg = &soc15_se_cac_rreg;
878 	adev->se_cac_wreg = &soc15_se_cac_wreg;
879 
880 	adev->rev_id = amdgpu_device_get_rev_id(adev);
881 	adev->external_rev_id = 0xFF;
882 	/* TODO: split the GC and PG flags based on the relevant IP version for which
883 	 * they are relevant.
884 	 */
885 	switch (adev->ip_versions[GC_HWIP][0]) {
886 	case IP_VERSION(9, 0, 1):
887 		adev->asic_funcs = &soc15_asic_funcs;
888 		adev->cg_flags = AMD_CG_SUPPORT_GFX_MGCG |
889 			AMD_CG_SUPPORT_GFX_MGLS |
890 			AMD_CG_SUPPORT_GFX_RLC_LS |
891 			AMD_CG_SUPPORT_GFX_CP_LS |
892 			AMD_CG_SUPPORT_GFX_3D_CGCG |
893 			AMD_CG_SUPPORT_GFX_3D_CGLS |
894 			AMD_CG_SUPPORT_GFX_CGCG |
895 			AMD_CG_SUPPORT_GFX_CGLS |
896 			AMD_CG_SUPPORT_BIF_MGCG |
897 			AMD_CG_SUPPORT_BIF_LS |
898 			AMD_CG_SUPPORT_HDP_LS |
899 			AMD_CG_SUPPORT_DRM_MGCG |
900 			AMD_CG_SUPPORT_DRM_LS |
901 			AMD_CG_SUPPORT_ROM_MGCG |
902 			AMD_CG_SUPPORT_DF_MGCG |
903 			AMD_CG_SUPPORT_SDMA_MGCG |
904 			AMD_CG_SUPPORT_SDMA_LS |
905 			AMD_CG_SUPPORT_MC_MGCG |
906 			AMD_CG_SUPPORT_MC_LS;
907 		adev->pg_flags = 0;
908 		adev->external_rev_id = 0x1;
909 		break;
910 	case IP_VERSION(9, 2, 1):
911 		adev->asic_funcs = &soc15_asic_funcs;
912 		adev->cg_flags = AMD_CG_SUPPORT_GFX_MGCG |
913 			AMD_CG_SUPPORT_GFX_MGLS |
914 			AMD_CG_SUPPORT_GFX_CGCG |
915 			AMD_CG_SUPPORT_GFX_CGLS |
916 			AMD_CG_SUPPORT_GFX_3D_CGCG |
917 			AMD_CG_SUPPORT_GFX_3D_CGLS |
918 			AMD_CG_SUPPORT_GFX_CP_LS |
919 			AMD_CG_SUPPORT_MC_LS |
920 			AMD_CG_SUPPORT_MC_MGCG |
921 			AMD_CG_SUPPORT_SDMA_MGCG |
922 			AMD_CG_SUPPORT_SDMA_LS |
923 			AMD_CG_SUPPORT_BIF_MGCG |
924 			AMD_CG_SUPPORT_BIF_LS |
925 			AMD_CG_SUPPORT_HDP_MGCG |
926 			AMD_CG_SUPPORT_HDP_LS |
927 			AMD_CG_SUPPORT_ROM_MGCG |
928 			AMD_CG_SUPPORT_VCE_MGCG |
929 			AMD_CG_SUPPORT_UVD_MGCG;
930 		adev->pg_flags = 0;
931 		adev->external_rev_id = adev->rev_id + 0x14;
932 		break;
933 	case IP_VERSION(9, 4, 0):
934 		adev->asic_funcs = &vega20_asic_funcs;
935 		adev->cg_flags = AMD_CG_SUPPORT_GFX_MGCG |
936 			AMD_CG_SUPPORT_GFX_MGLS |
937 			AMD_CG_SUPPORT_GFX_CGCG |
938 			AMD_CG_SUPPORT_GFX_CGLS |
939 			AMD_CG_SUPPORT_GFX_3D_CGCG |
940 			AMD_CG_SUPPORT_GFX_3D_CGLS |
941 			AMD_CG_SUPPORT_GFX_CP_LS |
942 			AMD_CG_SUPPORT_MC_LS |
943 			AMD_CG_SUPPORT_MC_MGCG |
944 			AMD_CG_SUPPORT_SDMA_MGCG |
945 			AMD_CG_SUPPORT_SDMA_LS |
946 			AMD_CG_SUPPORT_BIF_MGCG |
947 			AMD_CG_SUPPORT_BIF_LS |
948 			AMD_CG_SUPPORT_HDP_MGCG |
949 			AMD_CG_SUPPORT_HDP_LS |
950 			AMD_CG_SUPPORT_ROM_MGCG |
951 			AMD_CG_SUPPORT_VCE_MGCG |
952 			AMD_CG_SUPPORT_UVD_MGCG;
953 		adev->pg_flags = 0;
954 		adev->external_rev_id = adev->rev_id + 0x28;
955 		break;
956 	case IP_VERSION(9, 1, 0):
957 	case IP_VERSION(9, 2, 2):
958 		adev->asic_funcs = &soc15_asic_funcs;
959 
960 		if (adev->rev_id >= 0x8)
961 			adev->apu_flags |= AMD_APU_IS_RAVEN2;
962 
963 		if (adev->apu_flags & AMD_APU_IS_RAVEN2)
964 			adev->external_rev_id = adev->rev_id + 0x79;
965 		else if (adev->apu_flags & AMD_APU_IS_PICASSO)
966 			adev->external_rev_id = adev->rev_id + 0x41;
967 		else if (adev->rev_id == 1)
968 			adev->external_rev_id = adev->rev_id + 0x20;
969 		else
970 			adev->external_rev_id = adev->rev_id + 0x01;
971 
972 		if (adev->apu_flags & AMD_APU_IS_RAVEN2) {
973 			adev->cg_flags = AMD_CG_SUPPORT_GFX_MGCG |
974 				AMD_CG_SUPPORT_GFX_MGLS |
975 				AMD_CG_SUPPORT_GFX_CP_LS |
976 				AMD_CG_SUPPORT_GFX_3D_CGCG |
977 				AMD_CG_SUPPORT_GFX_3D_CGLS |
978 				AMD_CG_SUPPORT_GFX_CGCG |
979 				AMD_CG_SUPPORT_GFX_CGLS |
980 				AMD_CG_SUPPORT_BIF_LS |
981 				AMD_CG_SUPPORT_HDP_LS |
982 				AMD_CG_SUPPORT_MC_MGCG |
983 				AMD_CG_SUPPORT_MC_LS |
984 				AMD_CG_SUPPORT_SDMA_MGCG |
985 				AMD_CG_SUPPORT_SDMA_LS |
986 				AMD_CG_SUPPORT_VCN_MGCG;
987 
988 			adev->pg_flags = AMD_PG_SUPPORT_SDMA | AMD_PG_SUPPORT_VCN;
989 		} else if (adev->apu_flags & AMD_APU_IS_PICASSO) {
990 			adev->cg_flags = AMD_CG_SUPPORT_GFX_MGCG |
991 				AMD_CG_SUPPORT_GFX_MGLS |
992 				AMD_CG_SUPPORT_GFX_CP_LS |
993 				AMD_CG_SUPPORT_GFX_3D_CGLS |
994 				AMD_CG_SUPPORT_GFX_CGCG |
995 				AMD_CG_SUPPORT_GFX_CGLS |
996 				AMD_CG_SUPPORT_BIF_LS |
997 				AMD_CG_SUPPORT_HDP_LS |
998 				AMD_CG_SUPPORT_MC_MGCG |
999 				AMD_CG_SUPPORT_MC_LS |
1000 				AMD_CG_SUPPORT_SDMA_MGCG |
1001 				AMD_CG_SUPPORT_SDMA_LS |
1002 				AMD_CG_SUPPORT_VCN_MGCG;
1003 
1004 			/*
1005 			 * MMHUB PG needs to be disabled for Picasso for
1006 			 * stability reasons.
1007 			 */
1008 			adev->pg_flags = AMD_PG_SUPPORT_SDMA |
1009 				AMD_PG_SUPPORT_VCN;
1010 		} else {
1011 			adev->cg_flags = AMD_CG_SUPPORT_GFX_MGCG |
1012 				AMD_CG_SUPPORT_GFX_MGLS |
1013 				AMD_CG_SUPPORT_GFX_RLC_LS |
1014 				AMD_CG_SUPPORT_GFX_CP_LS |
1015 				AMD_CG_SUPPORT_GFX_3D_CGLS |
1016 				AMD_CG_SUPPORT_GFX_CGCG |
1017 				AMD_CG_SUPPORT_GFX_CGLS |
1018 				AMD_CG_SUPPORT_BIF_MGCG |
1019 				AMD_CG_SUPPORT_BIF_LS |
1020 				AMD_CG_SUPPORT_HDP_MGCG |
1021 				AMD_CG_SUPPORT_HDP_LS |
1022 				AMD_CG_SUPPORT_DRM_MGCG |
1023 				AMD_CG_SUPPORT_DRM_LS |
1024 				AMD_CG_SUPPORT_MC_MGCG |
1025 				AMD_CG_SUPPORT_MC_LS |
1026 				AMD_CG_SUPPORT_SDMA_MGCG |
1027 				AMD_CG_SUPPORT_SDMA_LS |
1028 				AMD_CG_SUPPORT_VCN_MGCG;
1029 
1030 			adev->pg_flags = AMD_PG_SUPPORT_SDMA | AMD_PG_SUPPORT_VCN;
1031 		}
1032 		break;
1033 	case IP_VERSION(9, 4, 1):
1034 		adev->asic_funcs = &vega20_asic_funcs;
1035 		adev->cg_flags = AMD_CG_SUPPORT_GFX_MGCG |
1036 			AMD_CG_SUPPORT_GFX_MGLS |
1037 			AMD_CG_SUPPORT_GFX_CGCG |
1038 			AMD_CG_SUPPORT_GFX_CGLS |
1039 			AMD_CG_SUPPORT_GFX_CP_LS |
1040 			AMD_CG_SUPPORT_HDP_MGCG |
1041 			AMD_CG_SUPPORT_HDP_LS |
1042 			AMD_CG_SUPPORT_SDMA_MGCG |
1043 			AMD_CG_SUPPORT_SDMA_LS |
1044 			AMD_CG_SUPPORT_MC_MGCG |
1045 			AMD_CG_SUPPORT_MC_LS |
1046 			AMD_CG_SUPPORT_IH_CG |
1047 			AMD_CG_SUPPORT_VCN_MGCG |
1048 			AMD_CG_SUPPORT_JPEG_MGCG;
1049 		adev->pg_flags = AMD_PG_SUPPORT_VCN | AMD_PG_SUPPORT_VCN_DPG;
1050 		adev->external_rev_id = adev->rev_id + 0x32;
1051 		break;
1052 	case IP_VERSION(9, 3, 0):
1053 		adev->asic_funcs = &soc15_asic_funcs;
1054 
1055 		if (adev->apu_flags & AMD_APU_IS_RENOIR)
1056 			adev->external_rev_id = adev->rev_id + 0x91;
1057 		else
1058 			adev->external_rev_id = adev->rev_id + 0xa1;
1059 		adev->cg_flags = AMD_CG_SUPPORT_GFX_MGCG |
1060 				 AMD_CG_SUPPORT_GFX_MGLS |
1061 				 AMD_CG_SUPPORT_GFX_3D_CGCG |
1062 				 AMD_CG_SUPPORT_GFX_3D_CGLS |
1063 				 AMD_CG_SUPPORT_GFX_CGCG |
1064 				 AMD_CG_SUPPORT_GFX_CGLS |
1065 				 AMD_CG_SUPPORT_GFX_CP_LS |
1066 				 AMD_CG_SUPPORT_MC_MGCG |
1067 				 AMD_CG_SUPPORT_MC_LS |
1068 				 AMD_CG_SUPPORT_SDMA_MGCG |
1069 				 AMD_CG_SUPPORT_SDMA_LS |
1070 				 AMD_CG_SUPPORT_BIF_LS |
1071 				 AMD_CG_SUPPORT_HDP_LS |
1072 				 AMD_CG_SUPPORT_VCN_MGCG |
1073 				 AMD_CG_SUPPORT_JPEG_MGCG |
1074 				 AMD_CG_SUPPORT_IH_CG |
1075 				 AMD_CG_SUPPORT_ATHUB_LS |
1076 				 AMD_CG_SUPPORT_ATHUB_MGCG |
1077 				 AMD_CG_SUPPORT_DF_MGCG;
1078 		adev->pg_flags = AMD_PG_SUPPORT_SDMA |
1079 				 AMD_PG_SUPPORT_VCN |
1080 				 AMD_PG_SUPPORT_JPEG |
1081 				 AMD_PG_SUPPORT_VCN_DPG;
1082 		break;
1083 	case IP_VERSION(9, 4, 2):
1084 		adev->asic_funcs = &vega20_asic_funcs;
1085 		adev->cg_flags = AMD_CG_SUPPORT_GFX_MGCG |
1086 			AMD_CG_SUPPORT_GFX_MGLS |
1087 			AMD_CG_SUPPORT_GFX_CP_LS |
1088 			AMD_CG_SUPPORT_HDP_LS |
1089 			AMD_CG_SUPPORT_SDMA_MGCG |
1090 			AMD_CG_SUPPORT_SDMA_LS |
1091 			AMD_CG_SUPPORT_IH_CG |
1092 			AMD_CG_SUPPORT_VCN_MGCG | AMD_CG_SUPPORT_JPEG_MGCG;
1093 		adev->pg_flags = AMD_PG_SUPPORT_VCN_DPG;
1094 		adev->external_rev_id = adev->rev_id + 0x3c;
1095 		break;
1096 	case IP_VERSION(9, 4, 3):
1097 		adev->asic_funcs = &vega20_asic_funcs;
1098 		adev->cg_flags = 0;
1099 		adev->pg_flags = 0;
1100 		break;
1101 	default:
1102 		/* FIXME: not supported yet */
1103 		return -EINVAL;
1104 	}
1105 
1106 	if (amdgpu_sriov_vf(adev)) {
1107 		amdgpu_virt_init_setting(adev);
1108 		xgpu_ai_mailbox_set_irq_funcs(adev);
1109 	}
1110 
1111 	return 0;
1112 }
1113 
1114 static int soc15_common_late_init(void *handle)
1115 {
1116 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1117 
1118 	if (amdgpu_sriov_vf(adev))
1119 		xgpu_ai_mailbox_get_irq(adev);
1120 
1121 	/* Enable selfring doorbell aperture late because doorbell BAR
1122 	 * aperture will change if resize BAR successfully in gmc sw_init.
1123 	 */
1124 	adev->nbio.funcs->enable_doorbell_selfring_aperture(adev, true);
1125 
1126 	return 0;
1127 }
1128 
1129 static int soc15_common_sw_init(void *handle)
1130 {
1131 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1132 
1133 	if (amdgpu_sriov_vf(adev))
1134 		xgpu_ai_mailbox_add_irq_id(adev);
1135 
1136 	if (adev->df.funcs &&
1137 	    adev->df.funcs->sw_init)
1138 		adev->df.funcs->sw_init(adev);
1139 
1140 	return 0;
1141 }
1142 
1143 static int soc15_common_sw_fini(void *handle)
1144 {
1145 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1146 
1147 	if (adev->df.funcs &&
1148 	    adev->df.funcs->sw_fini)
1149 		adev->df.funcs->sw_fini(adev);
1150 	return 0;
1151 }
1152 
1153 static void soc15_sdma_doorbell_range_init(struct amdgpu_device *adev)
1154 {
1155 	int i;
1156 
1157 	/* sdma doorbell range is programed by hypervisor */
1158 	if (!amdgpu_sriov_vf(adev)) {
1159 		for (i = 0; i < adev->sdma.num_instances; i++) {
1160 			adev->nbio.funcs->sdma_doorbell_range(adev, i,
1161 				true, adev->doorbell_index.sdma_engine[i] << 1,
1162 				adev->doorbell_index.sdma_doorbell_range);
1163 		}
1164 	}
1165 }
1166 
1167 static int soc15_common_hw_init(void *handle)
1168 {
1169 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1170 
1171 	/* enable aspm */
1172 	soc15_program_aspm(adev);
1173 	/* setup nbio registers */
1174 	adev->nbio.funcs->init_registers(adev);
1175 	/* remap HDP registers to a hole in mmio space,
1176 	 * for the purpose of expose those registers
1177 	 * to process space
1178 	 */
1179 	if (adev->nbio.funcs->remap_hdp_registers && !amdgpu_sriov_vf(adev))
1180 		adev->nbio.funcs->remap_hdp_registers(adev);
1181 
1182 	/* enable the doorbell aperture */
1183 	adev->nbio.funcs->enable_doorbell_aperture(adev, true);
1184 
1185 	/* HW doorbell routing policy: doorbell writing not
1186 	 * in SDMA/IH/MM/ACV range will be routed to CP. So
1187 	 * we need to init SDMA doorbell range prior
1188 	 * to CP ip block init and ring test.  IH already
1189 	 * happens before CP.
1190 	 */
1191 	soc15_sdma_doorbell_range_init(adev);
1192 
1193 	return 0;
1194 }
1195 
1196 static int soc15_common_hw_fini(void *handle)
1197 {
1198 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1199 
1200 	/* Disable the doorbell aperture and selfring doorbell aperture
1201 	 * separately in hw_fini because soc15_enable_doorbell_aperture
1202 	 * has been removed and there is no need to delay disabling
1203 	 * selfring doorbell.
1204 	 */
1205 	adev->nbio.funcs->enable_doorbell_aperture(adev, false);
1206 	adev->nbio.funcs->enable_doorbell_selfring_aperture(adev, false);
1207 
1208 	if (amdgpu_sriov_vf(adev))
1209 		xgpu_ai_mailbox_put_irq(adev);
1210 
1211 	if (adev->nbio.ras_if &&
1212 	    amdgpu_ras_is_supported(adev, adev->nbio.ras_if->block)) {
1213 		if (adev->nbio.ras &&
1214 		    adev->nbio.ras->init_ras_controller_interrupt)
1215 			amdgpu_irq_put(adev, &adev->nbio.ras_controller_irq, 0);
1216 		if (adev->nbio.ras &&
1217 		    adev->nbio.ras->init_ras_err_event_athub_interrupt)
1218 			amdgpu_irq_put(adev, &adev->nbio.ras_err_event_athub_irq, 0);
1219 	}
1220 
1221 	return 0;
1222 }
1223 
1224 static int soc15_common_suspend(void *handle)
1225 {
1226 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1227 
1228 	return soc15_common_hw_fini(adev);
1229 }
1230 
1231 static int soc15_common_resume(void *handle)
1232 {
1233 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1234 
1235 	return soc15_common_hw_init(adev);
1236 }
1237 
1238 static bool soc15_common_is_idle(void *handle)
1239 {
1240 	return true;
1241 }
1242 
1243 static int soc15_common_wait_for_idle(void *handle)
1244 {
1245 	return 0;
1246 }
1247 
1248 static int soc15_common_soft_reset(void *handle)
1249 {
1250 	return 0;
1251 }
1252 
1253 static void soc15_update_drm_clock_gating(struct amdgpu_device *adev, bool enable)
1254 {
1255 	uint32_t def, data;
1256 
1257 	def = data = RREG32(SOC15_REG_OFFSET(MP0, 0, mmMP0_MISC_CGTT_CTRL0));
1258 
1259 	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_DRM_MGCG))
1260 		data &= ~(0x01000000 |
1261 			  0x02000000 |
1262 			  0x04000000 |
1263 			  0x08000000 |
1264 			  0x10000000 |
1265 			  0x20000000 |
1266 			  0x40000000 |
1267 			  0x80000000);
1268 	else
1269 		data |= (0x01000000 |
1270 			 0x02000000 |
1271 			 0x04000000 |
1272 			 0x08000000 |
1273 			 0x10000000 |
1274 			 0x20000000 |
1275 			 0x40000000 |
1276 			 0x80000000);
1277 
1278 	if (def != data)
1279 		WREG32(SOC15_REG_OFFSET(MP0, 0, mmMP0_MISC_CGTT_CTRL0), data);
1280 }
1281 
1282 static void soc15_update_drm_light_sleep(struct amdgpu_device *adev, bool enable)
1283 {
1284 	uint32_t def, data;
1285 
1286 	def = data = RREG32(SOC15_REG_OFFSET(MP0, 0, mmMP0_MISC_LIGHT_SLEEP_CTRL));
1287 
1288 	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_DRM_LS))
1289 		data |= 1;
1290 	else
1291 		data &= ~1;
1292 
1293 	if (def != data)
1294 		WREG32(SOC15_REG_OFFSET(MP0, 0, mmMP0_MISC_LIGHT_SLEEP_CTRL), data);
1295 }
1296 
1297 static int soc15_common_set_clockgating_state(void *handle,
1298 					    enum amd_clockgating_state state)
1299 {
1300 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1301 
1302 	if (amdgpu_sriov_vf(adev))
1303 		return 0;
1304 
1305 	switch (adev->ip_versions[NBIO_HWIP][0]) {
1306 	case IP_VERSION(6, 1, 0):
1307 	case IP_VERSION(6, 2, 0):
1308 	case IP_VERSION(7, 4, 0):
1309 		adev->nbio.funcs->update_medium_grain_clock_gating(adev,
1310 				state == AMD_CG_STATE_GATE);
1311 		adev->nbio.funcs->update_medium_grain_light_sleep(adev,
1312 				state == AMD_CG_STATE_GATE);
1313 		adev->hdp.funcs->update_clock_gating(adev,
1314 				state == AMD_CG_STATE_GATE);
1315 		soc15_update_drm_clock_gating(adev,
1316 				state == AMD_CG_STATE_GATE);
1317 		soc15_update_drm_light_sleep(adev,
1318 				state == AMD_CG_STATE_GATE);
1319 		adev->smuio.funcs->update_rom_clock_gating(adev,
1320 				state == AMD_CG_STATE_GATE);
1321 		adev->df.funcs->update_medium_grain_clock_gating(adev,
1322 				state == AMD_CG_STATE_GATE);
1323 		break;
1324 	case IP_VERSION(7, 0, 0):
1325 	case IP_VERSION(7, 0, 1):
1326 	case IP_VERSION(2, 5, 0):
1327 		adev->nbio.funcs->update_medium_grain_clock_gating(adev,
1328 				state == AMD_CG_STATE_GATE);
1329 		adev->nbio.funcs->update_medium_grain_light_sleep(adev,
1330 				state == AMD_CG_STATE_GATE);
1331 		adev->hdp.funcs->update_clock_gating(adev,
1332 				state == AMD_CG_STATE_GATE);
1333 		soc15_update_drm_clock_gating(adev,
1334 				state == AMD_CG_STATE_GATE);
1335 		soc15_update_drm_light_sleep(adev,
1336 				state == AMD_CG_STATE_GATE);
1337 		break;
1338 	case IP_VERSION(7, 4, 1):
1339 	case IP_VERSION(7, 4, 4):
1340 		adev->hdp.funcs->update_clock_gating(adev,
1341 				state == AMD_CG_STATE_GATE);
1342 		break;
1343 	default:
1344 		break;
1345 	}
1346 	return 0;
1347 }
1348 
1349 static void soc15_common_get_clockgating_state(void *handle, u64 *flags)
1350 {
1351 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1352 	int data;
1353 
1354 	if (amdgpu_sriov_vf(adev))
1355 		*flags = 0;
1356 
1357 	adev->nbio.funcs->get_clockgating_state(adev, flags);
1358 
1359 	adev->hdp.funcs->get_clock_gating_state(adev, flags);
1360 
1361 	if (adev->ip_versions[MP0_HWIP][0] != IP_VERSION(13, 0, 2)) {
1362 
1363 		/* AMD_CG_SUPPORT_DRM_MGCG */
1364 		data = RREG32(SOC15_REG_OFFSET(MP0, 0, mmMP0_MISC_CGTT_CTRL0));
1365 		if (!(data & 0x01000000))
1366 			*flags |= AMD_CG_SUPPORT_DRM_MGCG;
1367 
1368 		/* AMD_CG_SUPPORT_DRM_LS */
1369 		data = RREG32(SOC15_REG_OFFSET(MP0, 0, mmMP0_MISC_LIGHT_SLEEP_CTRL));
1370 		if (data & 0x1)
1371 			*flags |= AMD_CG_SUPPORT_DRM_LS;
1372 	}
1373 
1374 	/* AMD_CG_SUPPORT_ROM_MGCG */
1375 	adev->smuio.funcs->get_clock_gating_state(adev, flags);
1376 
1377 	adev->df.funcs->get_clockgating_state(adev, flags);
1378 }
1379 
1380 static int soc15_common_set_powergating_state(void *handle,
1381 					    enum amd_powergating_state state)
1382 {
1383 	/* todo */
1384 	return 0;
1385 }
1386 
1387 static const struct amd_ip_funcs soc15_common_ip_funcs = {
1388 	.name = "soc15_common",
1389 	.early_init = soc15_common_early_init,
1390 	.late_init = soc15_common_late_init,
1391 	.sw_init = soc15_common_sw_init,
1392 	.sw_fini = soc15_common_sw_fini,
1393 	.hw_init = soc15_common_hw_init,
1394 	.hw_fini = soc15_common_hw_fini,
1395 	.suspend = soc15_common_suspend,
1396 	.resume = soc15_common_resume,
1397 	.is_idle = soc15_common_is_idle,
1398 	.wait_for_idle = soc15_common_wait_for_idle,
1399 	.soft_reset = soc15_common_soft_reset,
1400 	.set_clockgating_state = soc15_common_set_clockgating_state,
1401 	.set_powergating_state = soc15_common_set_powergating_state,
1402 	.get_clockgating_state= soc15_common_get_clockgating_state,
1403 };
1404