1 /*
2  * Copyright 2022 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 
25 #include "amdgpu.h"
26 #include "amdgpu_gfx.h"
27 #include "soc15.h"
28 #include "soc15d.h"
29 #include "soc15_common.h"
30 #include "vega10_enum.h"
31 
32 #include "v9_structs.h"
33 
34 #include "ivsrcid/gfx/irqsrcs_gfx_9_0.h"
35 
36 #include "gc/gc_9_4_3_offset.h"
37 #include "gc/gc_9_4_3_sh_mask.h"
38 
39 #include "gfx_v9_4_3.h"
40 #include "amdgpu_xcp.h"
41 
42 MODULE_FIRMWARE("amdgpu/gc_9_4_3_mec.bin");
43 MODULE_FIRMWARE("amdgpu/gc_9_4_3_rlc.bin");
44 
45 #define GFX9_MEC_HPD_SIZE 4096
46 #define RLCG_UCODE_LOADING_START_ADDRESS 0x00002000L
47 
48 #define GOLDEN_GB_ADDR_CONFIG 0x2a114042
49 #define CP_HQD_PERSISTENT_STATE_DEFAULT 0xbe05301
50 
51 struct amdgpu_gfx_ras gfx_v9_4_3_ras;
52 
53 static void gfx_v9_4_3_set_ring_funcs(struct amdgpu_device *adev);
54 static void gfx_v9_4_3_set_irq_funcs(struct amdgpu_device *adev);
55 static void gfx_v9_4_3_set_gds_init(struct amdgpu_device *adev);
56 static void gfx_v9_4_3_set_rlc_funcs(struct amdgpu_device *adev);
57 static int gfx_v9_4_3_get_cu_info(struct amdgpu_device *adev,
58 				struct amdgpu_cu_info *cu_info);
59 
gfx_v9_4_3_kiq_set_resources(struct amdgpu_ring * kiq_ring,uint64_t queue_mask)60 static void gfx_v9_4_3_kiq_set_resources(struct amdgpu_ring *kiq_ring,
61 				uint64_t queue_mask)
62 {
63 	amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_SET_RESOURCES, 6));
64 	amdgpu_ring_write(kiq_ring,
65 		PACKET3_SET_RESOURCES_VMID_MASK(0) |
66 		/* vmid_mask:0* queue_type:0 (KIQ) */
67 		PACKET3_SET_RESOURCES_QUEUE_TYPE(0));
68 	amdgpu_ring_write(kiq_ring,
69 			lower_32_bits(queue_mask));	/* queue mask lo */
70 	amdgpu_ring_write(kiq_ring,
71 			upper_32_bits(queue_mask));	/* queue mask hi */
72 	amdgpu_ring_write(kiq_ring, 0);	/* gws mask lo */
73 	amdgpu_ring_write(kiq_ring, 0);	/* gws mask hi */
74 	amdgpu_ring_write(kiq_ring, 0);	/* oac mask */
75 	amdgpu_ring_write(kiq_ring, 0);	/* gds heap base:0, gds heap size:0 */
76 }
77 
gfx_v9_4_3_kiq_map_queues(struct amdgpu_ring * kiq_ring,struct amdgpu_ring * ring)78 static void gfx_v9_4_3_kiq_map_queues(struct amdgpu_ring *kiq_ring,
79 				 struct amdgpu_ring *ring)
80 {
81 	struct amdgpu_device *adev = kiq_ring->adev;
82 	uint64_t mqd_addr = amdgpu_bo_gpu_offset(ring->mqd_obj);
83 	uint64_t wptr_addr = adev->wb.gpu_addr + (ring->wptr_offs * 4);
84 	uint32_t eng_sel = ring->funcs->type == AMDGPU_RING_TYPE_GFX ? 4 : 0;
85 
86 	amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_MAP_QUEUES, 5));
87 	/* Q_sel:0, vmid:0, vidmem: 1, engine:0, num_Q:1*/
88 	amdgpu_ring_write(kiq_ring, /* Q_sel: 0, vmid: 0, engine: 0, num_Q: 1 */
89 			 PACKET3_MAP_QUEUES_QUEUE_SEL(0) | /* Queue_Sel */
90 			 PACKET3_MAP_QUEUES_VMID(0) | /* VMID */
91 			 PACKET3_MAP_QUEUES_QUEUE(ring->queue) |
92 			 PACKET3_MAP_QUEUES_PIPE(ring->pipe) |
93 			 PACKET3_MAP_QUEUES_ME((ring->me == 1 ? 0 : 1)) |
94 			 /*queue_type: normal compute queue */
95 			 PACKET3_MAP_QUEUES_QUEUE_TYPE(0) |
96 			 /* alloc format: all_on_one_pipe */
97 			 PACKET3_MAP_QUEUES_ALLOC_FORMAT(0) |
98 			 PACKET3_MAP_QUEUES_ENGINE_SEL(eng_sel) |
99 			 /* num_queues: must be 1 */
100 			 PACKET3_MAP_QUEUES_NUM_QUEUES(1));
101 	amdgpu_ring_write(kiq_ring,
102 			PACKET3_MAP_QUEUES_DOORBELL_OFFSET(ring->doorbell_index));
103 	amdgpu_ring_write(kiq_ring, lower_32_bits(mqd_addr));
104 	amdgpu_ring_write(kiq_ring, upper_32_bits(mqd_addr));
105 	amdgpu_ring_write(kiq_ring, lower_32_bits(wptr_addr));
106 	amdgpu_ring_write(kiq_ring, upper_32_bits(wptr_addr));
107 }
108 
gfx_v9_4_3_kiq_unmap_queues(struct amdgpu_ring * kiq_ring,struct amdgpu_ring * ring,enum amdgpu_unmap_queues_action action,u64 gpu_addr,u64 seq)109 static void gfx_v9_4_3_kiq_unmap_queues(struct amdgpu_ring *kiq_ring,
110 				   struct amdgpu_ring *ring,
111 				   enum amdgpu_unmap_queues_action action,
112 				   u64 gpu_addr, u64 seq)
113 {
114 	uint32_t eng_sel = ring->funcs->type == AMDGPU_RING_TYPE_GFX ? 4 : 0;
115 
116 	amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_UNMAP_QUEUES, 4));
117 	amdgpu_ring_write(kiq_ring, /* Q_sel: 0, vmid: 0, engine: 0, num_Q: 1 */
118 			  PACKET3_UNMAP_QUEUES_ACTION(action) |
119 			  PACKET3_UNMAP_QUEUES_QUEUE_SEL(0) |
120 			  PACKET3_UNMAP_QUEUES_ENGINE_SEL(eng_sel) |
121 			  PACKET3_UNMAP_QUEUES_NUM_QUEUES(1));
122 	amdgpu_ring_write(kiq_ring,
123 			PACKET3_UNMAP_QUEUES_DOORBELL_OFFSET0(ring->doorbell_index));
124 
125 	if (action == PREEMPT_QUEUES_NO_UNMAP) {
126 		amdgpu_ring_write(kiq_ring, lower_32_bits(gpu_addr));
127 		amdgpu_ring_write(kiq_ring, upper_32_bits(gpu_addr));
128 		amdgpu_ring_write(kiq_ring, seq);
129 	} else {
130 		amdgpu_ring_write(kiq_ring, 0);
131 		amdgpu_ring_write(kiq_ring, 0);
132 		amdgpu_ring_write(kiq_ring, 0);
133 	}
134 }
135 
gfx_v9_4_3_kiq_query_status(struct amdgpu_ring * kiq_ring,struct amdgpu_ring * ring,u64 addr,u64 seq)136 static void gfx_v9_4_3_kiq_query_status(struct amdgpu_ring *kiq_ring,
137 				   struct amdgpu_ring *ring,
138 				   u64 addr,
139 				   u64 seq)
140 {
141 	uint32_t eng_sel = ring->funcs->type == AMDGPU_RING_TYPE_GFX ? 4 : 0;
142 
143 	amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_QUERY_STATUS, 5));
144 	amdgpu_ring_write(kiq_ring,
145 			  PACKET3_QUERY_STATUS_CONTEXT_ID(0) |
146 			  PACKET3_QUERY_STATUS_INTERRUPT_SEL(0) |
147 			  PACKET3_QUERY_STATUS_COMMAND(2));
148 	/* Q_sel: 0, vmid: 0, engine: 0, num_Q: 1 */
149 	amdgpu_ring_write(kiq_ring,
150 			PACKET3_QUERY_STATUS_DOORBELL_OFFSET(ring->doorbell_index) |
151 			PACKET3_QUERY_STATUS_ENG_SEL(eng_sel));
152 	amdgpu_ring_write(kiq_ring, lower_32_bits(addr));
153 	amdgpu_ring_write(kiq_ring, upper_32_bits(addr));
154 	amdgpu_ring_write(kiq_ring, lower_32_bits(seq));
155 	amdgpu_ring_write(kiq_ring, upper_32_bits(seq));
156 }
157 
gfx_v9_4_3_kiq_invalidate_tlbs(struct amdgpu_ring * kiq_ring,uint16_t pasid,uint32_t flush_type,bool all_hub)158 static void gfx_v9_4_3_kiq_invalidate_tlbs(struct amdgpu_ring *kiq_ring,
159 				uint16_t pasid, uint32_t flush_type,
160 				bool all_hub)
161 {
162 	amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_INVALIDATE_TLBS, 0));
163 	amdgpu_ring_write(kiq_ring,
164 			PACKET3_INVALIDATE_TLBS_DST_SEL(1) |
165 			PACKET3_INVALIDATE_TLBS_ALL_HUB(all_hub) |
166 			PACKET3_INVALIDATE_TLBS_PASID(pasid) |
167 			PACKET3_INVALIDATE_TLBS_FLUSH_TYPE(flush_type));
168 }
169 
170 static const struct kiq_pm4_funcs gfx_v9_4_3_kiq_pm4_funcs = {
171 	.kiq_set_resources = gfx_v9_4_3_kiq_set_resources,
172 	.kiq_map_queues = gfx_v9_4_3_kiq_map_queues,
173 	.kiq_unmap_queues = gfx_v9_4_3_kiq_unmap_queues,
174 	.kiq_query_status = gfx_v9_4_3_kiq_query_status,
175 	.kiq_invalidate_tlbs = gfx_v9_4_3_kiq_invalidate_tlbs,
176 	.set_resources_size = 8,
177 	.map_queues_size = 7,
178 	.unmap_queues_size = 6,
179 	.query_status_size = 7,
180 	.invalidate_tlbs_size = 2,
181 };
182 
gfx_v9_4_3_set_kiq_pm4_funcs(struct amdgpu_device * adev)183 static void gfx_v9_4_3_set_kiq_pm4_funcs(struct amdgpu_device *adev)
184 {
185 	int i, num_xcc;
186 
187 	num_xcc = NUM_XCC(adev->gfx.xcc_mask);
188 	for (i = 0; i < num_xcc; i++)
189 		adev->gfx.kiq[i].pmf = &gfx_v9_4_3_kiq_pm4_funcs;
190 }
191 
gfx_v9_4_3_init_golden_registers(struct amdgpu_device * adev)192 static void gfx_v9_4_3_init_golden_registers(struct amdgpu_device *adev)
193 {
194 	int i, num_xcc, dev_inst;
195 
196 	num_xcc = NUM_XCC(adev->gfx.xcc_mask);
197 	for (i = 0; i < num_xcc; i++) {
198 		dev_inst = GET_INST(GC, i);
199 
200 		WREG32_SOC15(GC, dev_inst, regGB_ADDR_CONFIG,
201 			     GOLDEN_GB_ADDR_CONFIG);
202 		/* Golden settings applied by driver for ASIC with rev_id 0 */
203 		if (adev->rev_id == 0) {
204 			WREG32_FIELD15_PREREG(GC, dev_inst, TCP_UTCL1_CNTL1,
205 					      REDUCE_FIFO_DEPTH_BY_2, 2);
206 		} else {
207 			WREG32_FIELD15_PREREG(GC, dev_inst, TCP_UTCL1_CNTL2,
208 						SPARE, 0x1);
209 		}
210 	}
211 }
212 
gfx_v9_4_3_write_data_to_reg(struct amdgpu_ring * ring,int eng_sel,bool wc,uint32_t reg,uint32_t val)213 static void gfx_v9_4_3_write_data_to_reg(struct amdgpu_ring *ring, int eng_sel,
214 				       bool wc, uint32_t reg, uint32_t val)
215 {
216 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
217 	amdgpu_ring_write(ring, WRITE_DATA_ENGINE_SEL(eng_sel) |
218 				WRITE_DATA_DST_SEL(0) |
219 				(wc ? WR_CONFIRM : 0));
220 	amdgpu_ring_write(ring, reg);
221 	amdgpu_ring_write(ring, 0);
222 	amdgpu_ring_write(ring, val);
223 }
224 
gfx_v9_4_3_wait_reg_mem(struct amdgpu_ring * ring,int eng_sel,int mem_space,int opt,uint32_t addr0,uint32_t addr1,uint32_t ref,uint32_t mask,uint32_t inv)225 static void gfx_v9_4_3_wait_reg_mem(struct amdgpu_ring *ring, int eng_sel,
226 				  int mem_space, int opt, uint32_t addr0,
227 				  uint32_t addr1, uint32_t ref, uint32_t mask,
228 				  uint32_t inv)
229 {
230 	amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
231 	amdgpu_ring_write(ring,
232 				 /* memory (1) or register (0) */
233 				 (WAIT_REG_MEM_MEM_SPACE(mem_space) |
234 				 WAIT_REG_MEM_OPERATION(opt) | /* wait */
235 				 WAIT_REG_MEM_FUNCTION(3) |  /* equal */
236 				 WAIT_REG_MEM_ENGINE(eng_sel)));
237 
238 	if (mem_space)
239 		BUG_ON(addr0 & 0x3); /* Dword align */
240 	amdgpu_ring_write(ring, addr0);
241 	amdgpu_ring_write(ring, addr1);
242 	amdgpu_ring_write(ring, ref);
243 	amdgpu_ring_write(ring, mask);
244 	amdgpu_ring_write(ring, inv); /* poll interval */
245 }
246 
gfx_v9_4_3_ring_test_ring(struct amdgpu_ring * ring)247 static int gfx_v9_4_3_ring_test_ring(struct amdgpu_ring *ring)
248 {
249 	uint32_t scratch_reg0_offset, xcc_offset;
250 	struct amdgpu_device *adev = ring->adev;
251 	uint32_t tmp = 0;
252 	unsigned i;
253 	int r;
254 
255 	/* Use register offset which is local to XCC in the packet */
256 	xcc_offset = SOC15_REG_OFFSET(GC, 0, regSCRATCH_REG0);
257 	scratch_reg0_offset = SOC15_REG_OFFSET(GC, GET_INST(GC, ring->xcc_id), regSCRATCH_REG0);
258 	WREG32(scratch_reg0_offset, 0xCAFEDEAD);
259 
260 	r = amdgpu_ring_alloc(ring, 3);
261 	if (r)
262 		return r;
263 
264 	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_UCONFIG_REG, 1));
265 	amdgpu_ring_write(ring, xcc_offset - PACKET3_SET_UCONFIG_REG_START);
266 	amdgpu_ring_write(ring, 0xDEADBEEF);
267 	amdgpu_ring_commit(ring);
268 
269 	for (i = 0; i < adev->usec_timeout; i++) {
270 		tmp = RREG32(scratch_reg0_offset);
271 		if (tmp == 0xDEADBEEF)
272 			break;
273 		udelay(1);
274 	}
275 
276 	if (i >= adev->usec_timeout)
277 		r = -ETIMEDOUT;
278 	return r;
279 }
280 
gfx_v9_4_3_ring_test_ib(struct amdgpu_ring * ring,long timeout)281 static int gfx_v9_4_3_ring_test_ib(struct amdgpu_ring *ring, long timeout)
282 {
283 	struct amdgpu_device *adev = ring->adev;
284 	struct amdgpu_ib ib;
285 	struct dma_fence *f = NULL;
286 
287 	unsigned index;
288 	uint64_t gpu_addr;
289 	uint32_t tmp;
290 	long r;
291 
292 	r = amdgpu_device_wb_get(adev, &index);
293 	if (r)
294 		return r;
295 
296 	gpu_addr = adev->wb.gpu_addr + (index * 4);
297 	adev->wb.wb[index] = cpu_to_le32(0xCAFEDEAD);
298 	memset(&ib, 0, sizeof(ib));
299 
300 	r = amdgpu_ib_get(adev, NULL, 20, AMDGPU_IB_POOL_DIRECT, &ib);
301 	if (r)
302 		goto err1;
303 
304 	ib.ptr[0] = PACKET3(PACKET3_WRITE_DATA, 3);
305 	ib.ptr[1] = WRITE_DATA_DST_SEL(5) | WR_CONFIRM;
306 	ib.ptr[2] = lower_32_bits(gpu_addr);
307 	ib.ptr[3] = upper_32_bits(gpu_addr);
308 	ib.ptr[4] = 0xDEADBEEF;
309 	ib.length_dw = 5;
310 
311 	r = amdgpu_ib_schedule(ring, 1, &ib, NULL, &f);
312 	if (r)
313 		goto err2;
314 
315 	r = dma_fence_wait_timeout(f, false, timeout);
316 	if (r == 0) {
317 		r = -ETIMEDOUT;
318 		goto err2;
319 	} else if (r < 0) {
320 		goto err2;
321 	}
322 
323 	tmp = adev->wb.wb[index];
324 	if (tmp == 0xDEADBEEF)
325 		r = 0;
326 	else
327 		r = -EINVAL;
328 
329 err2:
330 	amdgpu_ib_free(adev, &ib, NULL);
331 	dma_fence_put(f);
332 err1:
333 	amdgpu_device_wb_free(adev, index);
334 	return r;
335 }
336 
337 
338 /* This value might differs per partition */
gfx_v9_4_3_get_gpu_clock_counter(struct amdgpu_device * adev)339 static uint64_t gfx_v9_4_3_get_gpu_clock_counter(struct amdgpu_device *adev)
340 {
341 	uint64_t clock;
342 
343 	mutex_lock(&adev->gfx.gpu_clock_mutex);
344 	WREG32_SOC15(GC, GET_INST(GC, 0), regRLC_CAPTURE_GPU_CLOCK_COUNT, 1);
345 	clock = (uint64_t)RREG32_SOC15(GC, GET_INST(GC, 0), regRLC_GPU_CLOCK_COUNT_LSB) |
346 		((uint64_t)RREG32_SOC15(GC, GET_INST(GC, 0), regRLC_GPU_CLOCK_COUNT_MSB) << 32ULL);
347 	mutex_unlock(&adev->gfx.gpu_clock_mutex);
348 
349 	return clock;
350 }
351 
gfx_v9_4_3_free_microcode(struct amdgpu_device * adev)352 static void gfx_v9_4_3_free_microcode(struct amdgpu_device *adev)
353 {
354 	amdgpu_ucode_release(&adev->gfx.pfp_fw);
355 	amdgpu_ucode_release(&adev->gfx.me_fw);
356 	amdgpu_ucode_release(&adev->gfx.ce_fw);
357 	amdgpu_ucode_release(&adev->gfx.rlc_fw);
358 	amdgpu_ucode_release(&adev->gfx.mec_fw);
359 	amdgpu_ucode_release(&adev->gfx.mec2_fw);
360 
361 	kfree(adev->gfx.rlc.register_list_format);
362 }
363 
gfx_v9_4_3_init_rlc_microcode(struct amdgpu_device * adev,const char * chip_name)364 static int gfx_v9_4_3_init_rlc_microcode(struct amdgpu_device *adev,
365 					  const char *chip_name)
366 {
367 	char fw_name[30];
368 	int err;
369 	const struct rlc_firmware_header_v2_0 *rlc_hdr;
370 	uint16_t version_major;
371 	uint16_t version_minor;
372 
373 	snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_rlc.bin", chip_name);
374 
375 	err = amdgpu_ucode_request(adev, &adev->gfx.rlc_fw, fw_name);
376 	if (err)
377 		goto out;
378 	rlc_hdr = (const struct rlc_firmware_header_v2_0 *)adev->gfx.rlc_fw->data;
379 
380 	version_major = le16_to_cpu(rlc_hdr->header.header_version_major);
381 	version_minor = le16_to_cpu(rlc_hdr->header.header_version_minor);
382 	err = amdgpu_gfx_rlc_init_microcode(adev, version_major, version_minor);
383 out:
384 	if (err)
385 		amdgpu_ucode_release(&adev->gfx.rlc_fw);
386 
387 	return err;
388 }
389 
gfx_v9_4_3_should_disable_gfxoff(struct pci_dev * pdev)390 static bool gfx_v9_4_3_should_disable_gfxoff(struct pci_dev *pdev)
391 {
392 	return true;
393 }
394 
gfx_v9_4_3_check_if_need_gfxoff(struct amdgpu_device * adev)395 static void gfx_v9_4_3_check_if_need_gfxoff(struct amdgpu_device *adev)
396 {
397 	if (gfx_v9_4_3_should_disable_gfxoff(adev->pdev))
398 		adev->pm.pp_feature &= ~PP_GFXOFF_MASK;
399 }
400 
gfx_v9_4_3_init_cp_compute_microcode(struct amdgpu_device * adev,const char * chip_name)401 static int gfx_v9_4_3_init_cp_compute_microcode(struct amdgpu_device *adev,
402 					  const char *chip_name)
403 {
404 	char fw_name[30];
405 	int err;
406 
407 	snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec.bin", chip_name);
408 
409 	err = amdgpu_ucode_request(adev, &adev->gfx.mec_fw, fw_name);
410 	if (err)
411 		goto out;
412 	amdgpu_gfx_cp_init_microcode(adev, AMDGPU_UCODE_ID_CP_MEC1);
413 	amdgpu_gfx_cp_init_microcode(adev, AMDGPU_UCODE_ID_CP_MEC1_JT);
414 
415 	adev->gfx.mec2_fw_version = adev->gfx.mec_fw_version;
416 	adev->gfx.mec2_feature_version = adev->gfx.mec_feature_version;
417 
418 	gfx_v9_4_3_check_if_need_gfxoff(adev);
419 
420 out:
421 	if (err)
422 		amdgpu_ucode_release(&adev->gfx.mec_fw);
423 	return err;
424 }
425 
gfx_v9_4_3_init_microcode(struct amdgpu_device * adev)426 static int gfx_v9_4_3_init_microcode(struct amdgpu_device *adev)
427 {
428 	char ucode_prefix[15];
429 	int r;
430 
431 	amdgpu_ucode_ip_version_decode(adev, GC_HWIP, ucode_prefix, sizeof(ucode_prefix));
432 
433 	r = gfx_v9_4_3_init_rlc_microcode(adev, ucode_prefix);
434 	if (r)
435 		return r;
436 
437 	r = gfx_v9_4_3_init_cp_compute_microcode(adev, ucode_prefix);
438 	if (r)
439 		return r;
440 
441 	return r;
442 }
443 
gfx_v9_4_3_mec_fini(struct amdgpu_device * adev)444 static void gfx_v9_4_3_mec_fini(struct amdgpu_device *adev)
445 {
446 	amdgpu_bo_free_kernel(&adev->gfx.mec.hpd_eop_obj, NULL, NULL);
447 	amdgpu_bo_free_kernel(&adev->gfx.mec.mec_fw_obj, NULL, NULL);
448 }
449 
gfx_v9_4_3_mec_init(struct amdgpu_device * adev)450 static int gfx_v9_4_3_mec_init(struct amdgpu_device *adev)
451 {
452 	int r, i, num_xcc;
453 	u32 *hpd;
454 	const __le32 *fw_data;
455 	unsigned fw_size;
456 	u32 *fw;
457 	size_t mec_hpd_size;
458 
459 	const struct gfx_firmware_header_v1_0 *mec_hdr;
460 
461 	num_xcc = NUM_XCC(adev->gfx.xcc_mask);
462 	for (i = 0; i < num_xcc; i++)
463 		bitmap_zero(adev->gfx.mec_bitmap[i].queue_bitmap,
464 			AMDGPU_MAX_COMPUTE_QUEUES);
465 
466 	/* take ownership of the relevant compute queues */
467 	amdgpu_gfx_compute_queue_acquire(adev);
468 	mec_hpd_size =
469 		adev->gfx.num_compute_rings * num_xcc * GFX9_MEC_HPD_SIZE;
470 	if (mec_hpd_size) {
471 		r = amdgpu_bo_create_reserved(adev, mec_hpd_size, PAGE_SIZE,
472 					      AMDGPU_GEM_DOMAIN_VRAM |
473 					      AMDGPU_GEM_DOMAIN_GTT,
474 					      &adev->gfx.mec.hpd_eop_obj,
475 					      &adev->gfx.mec.hpd_eop_gpu_addr,
476 					      (void **)&hpd);
477 		if (r) {
478 			dev_warn(adev->dev, "(%d) create HDP EOP bo failed\n", r);
479 			gfx_v9_4_3_mec_fini(adev);
480 			return r;
481 		}
482 
483 		if (amdgpu_emu_mode == 1) {
484 			for (i = 0; i < mec_hpd_size / 4; i++) {
485 				memset((void *)(hpd + i), 0, 4);
486 				if (i % 50 == 0)
487 					msleep(1);
488 			}
489 		} else {
490 			memset(hpd, 0, mec_hpd_size);
491 		}
492 
493 		amdgpu_bo_kunmap(adev->gfx.mec.hpd_eop_obj);
494 		amdgpu_bo_unreserve(adev->gfx.mec.hpd_eop_obj);
495 	}
496 
497 	mec_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data;
498 
499 	fw_data = (const __le32 *)
500 		(adev->gfx.mec_fw->data +
501 		 le32_to_cpu(mec_hdr->header.ucode_array_offset_bytes));
502 	fw_size = le32_to_cpu(mec_hdr->header.ucode_size_bytes);
503 
504 	r = amdgpu_bo_create_reserved(adev, mec_hdr->header.ucode_size_bytes,
505 				      PAGE_SIZE, AMDGPU_GEM_DOMAIN_GTT,
506 				      &adev->gfx.mec.mec_fw_obj,
507 				      &adev->gfx.mec.mec_fw_gpu_addr,
508 				      (void **)&fw);
509 	if (r) {
510 		dev_warn(adev->dev, "(%d) create mec firmware bo failed\n", r);
511 		gfx_v9_4_3_mec_fini(adev);
512 		return r;
513 	}
514 
515 	memcpy(fw, fw_data, fw_size);
516 
517 	amdgpu_bo_kunmap(adev->gfx.mec.mec_fw_obj);
518 	amdgpu_bo_unreserve(adev->gfx.mec.mec_fw_obj);
519 
520 	return 0;
521 }
522 
gfx_v9_4_3_xcc_select_se_sh(struct amdgpu_device * adev,u32 se_num,u32 sh_num,u32 instance,int xcc_id)523 static void gfx_v9_4_3_xcc_select_se_sh(struct amdgpu_device *adev, u32 se_num,
524 					u32 sh_num, u32 instance, int xcc_id)
525 {
526 	u32 data;
527 
528 	if (instance == 0xffffffff)
529 		data = REG_SET_FIELD(0, GRBM_GFX_INDEX,
530 				     INSTANCE_BROADCAST_WRITES, 1);
531 	else
532 		data = REG_SET_FIELD(0, GRBM_GFX_INDEX,
533 				     INSTANCE_INDEX, instance);
534 
535 	if (se_num == 0xffffffff)
536 		data = REG_SET_FIELD(data, GRBM_GFX_INDEX,
537 				     SE_BROADCAST_WRITES, 1);
538 	else
539 		data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SE_INDEX, se_num);
540 
541 	if (sh_num == 0xffffffff)
542 		data = REG_SET_FIELD(data, GRBM_GFX_INDEX,
543 				     SH_BROADCAST_WRITES, 1);
544 	else
545 		data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SH_INDEX, sh_num);
546 
547 	WREG32_SOC15_RLC_SHADOW_EX(reg, GC, GET_INST(GC, xcc_id), regGRBM_GFX_INDEX, data);
548 }
549 
wave_read_ind(struct amdgpu_device * adev,uint32_t xcc_id,uint32_t simd,uint32_t wave,uint32_t address)550 static uint32_t wave_read_ind(struct amdgpu_device *adev, uint32_t xcc_id, uint32_t simd, uint32_t wave, uint32_t address)
551 {
552 	WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regSQ_IND_INDEX,
553 		(wave << SQ_IND_INDEX__WAVE_ID__SHIFT) |
554 		(simd << SQ_IND_INDEX__SIMD_ID__SHIFT) |
555 		(address << SQ_IND_INDEX__INDEX__SHIFT) |
556 		(SQ_IND_INDEX__FORCE_READ_MASK));
557 	return RREG32_SOC15(GC, GET_INST(GC, xcc_id), regSQ_IND_DATA);
558 }
559 
wave_read_regs(struct amdgpu_device * adev,uint32_t xcc_id,uint32_t simd,uint32_t wave,uint32_t thread,uint32_t regno,uint32_t num,uint32_t * out)560 static void wave_read_regs(struct amdgpu_device *adev, uint32_t xcc_id, uint32_t simd,
561 			   uint32_t wave, uint32_t thread,
562 			   uint32_t regno, uint32_t num, uint32_t *out)
563 {
564 	WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regSQ_IND_INDEX,
565 		(wave << SQ_IND_INDEX__WAVE_ID__SHIFT) |
566 		(simd << SQ_IND_INDEX__SIMD_ID__SHIFT) |
567 		(regno << SQ_IND_INDEX__INDEX__SHIFT) |
568 		(thread << SQ_IND_INDEX__THREAD_ID__SHIFT) |
569 		(SQ_IND_INDEX__FORCE_READ_MASK) |
570 		(SQ_IND_INDEX__AUTO_INCR_MASK));
571 	while (num--)
572 		*(out++) = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regSQ_IND_DATA);
573 }
574 
gfx_v9_4_3_read_wave_data(struct amdgpu_device * adev,uint32_t xcc_id,uint32_t simd,uint32_t wave,uint32_t * dst,int * no_fields)575 static void gfx_v9_4_3_read_wave_data(struct amdgpu_device *adev,
576 				      uint32_t xcc_id, uint32_t simd, uint32_t wave,
577 				      uint32_t *dst, int *no_fields)
578 {
579 	/* type 1 wave data */
580 	dst[(*no_fields)++] = 1;
581 	dst[(*no_fields)++] = wave_read_ind(adev, xcc_id, simd, wave, ixSQ_WAVE_STATUS);
582 	dst[(*no_fields)++] = wave_read_ind(adev, xcc_id, simd, wave, ixSQ_WAVE_PC_LO);
583 	dst[(*no_fields)++] = wave_read_ind(adev, xcc_id, simd, wave, ixSQ_WAVE_PC_HI);
584 	dst[(*no_fields)++] = wave_read_ind(adev, xcc_id, simd, wave, ixSQ_WAVE_EXEC_LO);
585 	dst[(*no_fields)++] = wave_read_ind(adev, xcc_id, simd, wave, ixSQ_WAVE_EXEC_HI);
586 	dst[(*no_fields)++] = wave_read_ind(adev, xcc_id, simd, wave, ixSQ_WAVE_HW_ID);
587 	dst[(*no_fields)++] = wave_read_ind(adev, xcc_id, simd, wave, ixSQ_WAVE_INST_DW0);
588 	dst[(*no_fields)++] = wave_read_ind(adev, xcc_id, simd, wave, ixSQ_WAVE_INST_DW1);
589 	dst[(*no_fields)++] = wave_read_ind(adev, xcc_id, simd, wave, ixSQ_WAVE_GPR_ALLOC);
590 	dst[(*no_fields)++] = wave_read_ind(adev, xcc_id, simd, wave, ixSQ_WAVE_LDS_ALLOC);
591 	dst[(*no_fields)++] = wave_read_ind(adev, xcc_id, simd, wave, ixSQ_WAVE_TRAPSTS);
592 	dst[(*no_fields)++] = wave_read_ind(adev, xcc_id, simd, wave, ixSQ_WAVE_IB_STS);
593 	dst[(*no_fields)++] = wave_read_ind(adev, xcc_id, simd, wave, ixSQ_WAVE_IB_DBG0);
594 	dst[(*no_fields)++] = wave_read_ind(adev, xcc_id, simd, wave, ixSQ_WAVE_M0);
595 	dst[(*no_fields)++] = wave_read_ind(adev, xcc_id, simd, wave, ixSQ_WAVE_MODE);
596 }
597 
gfx_v9_4_3_read_wave_sgprs(struct amdgpu_device * adev,uint32_t xcc_id,uint32_t simd,uint32_t wave,uint32_t start,uint32_t size,uint32_t * dst)598 static void gfx_v9_4_3_read_wave_sgprs(struct amdgpu_device *adev, uint32_t xcc_id, uint32_t simd,
599 				       uint32_t wave, uint32_t start,
600 				       uint32_t size, uint32_t *dst)
601 {
602 	wave_read_regs(adev, xcc_id, simd, wave, 0,
603 		       start + SQIND_WAVE_SGPRS_OFFSET, size, dst);
604 }
605 
gfx_v9_4_3_read_wave_vgprs(struct amdgpu_device * adev,uint32_t xcc_id,uint32_t simd,uint32_t wave,uint32_t thread,uint32_t start,uint32_t size,uint32_t * dst)606 static void gfx_v9_4_3_read_wave_vgprs(struct amdgpu_device *adev, uint32_t xcc_id, uint32_t simd,
607 				       uint32_t wave, uint32_t thread,
608 				       uint32_t start, uint32_t size,
609 				       uint32_t *dst)
610 {
611 	wave_read_regs(adev, xcc_id, simd, wave, thread,
612 		       start + SQIND_WAVE_VGPRS_OFFSET, size, dst);
613 }
614 
gfx_v9_4_3_select_me_pipe_q(struct amdgpu_device * adev,u32 me,u32 pipe,u32 q,u32 vm,u32 xcc_id)615 static void gfx_v9_4_3_select_me_pipe_q(struct amdgpu_device *adev,
616 					u32 me, u32 pipe, u32 q, u32 vm, u32 xcc_id)
617 {
618 	soc15_grbm_select(adev, me, pipe, q, vm, GET_INST(GC, xcc_id));
619 }
620 
621 
gfx_v9_4_3_switch_compute_partition(struct amdgpu_device * adev,int num_xccs_per_xcp)622 static int gfx_v9_4_3_switch_compute_partition(struct amdgpu_device *adev,
623 						int num_xccs_per_xcp)
624 {
625 	int ret, i, num_xcc;
626 	u32 tmp = 0, regval;
627 
628 	if (adev->psp.funcs) {
629 		ret = psp_spatial_partition(&adev->psp,
630 					    NUM_XCC(adev->gfx.xcc_mask) /
631 						    num_xccs_per_xcp);
632 		if (ret)
633 			return ret;
634 	}
635 
636 	num_xcc = NUM_XCC(adev->gfx.xcc_mask);
637 
638 	for (i = 0; i < num_xcc; i++) {
639 		tmp = REG_SET_FIELD(tmp, CP_HYP_XCP_CTL, NUM_XCC_IN_XCP,
640 				    num_xccs_per_xcp);
641 		tmp = REG_SET_FIELD(tmp, CP_HYP_XCP_CTL, VIRTUAL_XCC_ID,
642 				    i % num_xccs_per_xcp);
643 		regval = RREG32_SOC15(GC, GET_INST(GC, i), regCP_HYP_XCP_CTL);
644 		if (regval != tmp)
645 			WREG32_SOC15(GC, GET_INST(GC, i), regCP_HYP_XCP_CTL,
646 				     tmp);
647 	}
648 
649 	adev->gfx.num_xcc_per_xcp = num_xccs_per_xcp;
650 
651 	return 0;
652 }
653 
gfx_v9_4_3_ih_to_xcc_inst(struct amdgpu_device * adev,int ih_node)654 static int gfx_v9_4_3_ih_to_xcc_inst(struct amdgpu_device *adev, int ih_node)
655 {
656 	int xcc;
657 
658 	xcc = hweight8(adev->gfx.xcc_mask & GENMASK(ih_node / 2, 0));
659 	if (!xcc) {
660 		dev_err(adev->dev, "Couldn't find xcc mapping from IH node");
661 		return -EINVAL;
662 	}
663 
664 	return xcc - 1;
665 }
666 
667 static const struct amdgpu_gfx_funcs gfx_v9_4_3_gfx_funcs = {
668 	.get_gpu_clock_counter = &gfx_v9_4_3_get_gpu_clock_counter,
669 	.select_se_sh = &gfx_v9_4_3_xcc_select_se_sh,
670 	.read_wave_data = &gfx_v9_4_3_read_wave_data,
671 	.read_wave_sgprs = &gfx_v9_4_3_read_wave_sgprs,
672 	.read_wave_vgprs = &gfx_v9_4_3_read_wave_vgprs,
673 	.select_me_pipe_q = &gfx_v9_4_3_select_me_pipe_q,
674 	.switch_partition_mode = &gfx_v9_4_3_switch_compute_partition,
675 	.ih_node_to_logical_xcc = &gfx_v9_4_3_ih_to_xcc_inst,
676 };
677 
gfx_v9_4_3_gpu_early_init(struct amdgpu_device * adev)678 static int gfx_v9_4_3_gpu_early_init(struct amdgpu_device *adev)
679 {
680 	u32 gb_addr_config;
681 
682 	adev->gfx.funcs = &gfx_v9_4_3_gfx_funcs;
683 	adev->gfx.ras = &gfx_v9_4_3_ras;
684 
685 	switch (adev->ip_versions[GC_HWIP][0]) {
686 	case IP_VERSION(9, 4, 3):
687 		adev->gfx.config.max_hw_contexts = 8;
688 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
689 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
690 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
691 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x4C0;
692 		gb_addr_config = RREG32_SOC15(GC, GET_INST(GC, 0), regGB_ADDR_CONFIG);
693 		break;
694 	default:
695 		BUG();
696 		break;
697 	}
698 
699 	adev->gfx.config.gb_addr_config = gb_addr_config;
700 
701 	adev->gfx.config.gb_addr_config_fields.num_pipes = 1 <<
702 			REG_GET_FIELD(
703 					adev->gfx.config.gb_addr_config,
704 					GB_ADDR_CONFIG,
705 					NUM_PIPES);
706 
707 	adev->gfx.config.max_tile_pipes =
708 		adev->gfx.config.gb_addr_config_fields.num_pipes;
709 
710 	adev->gfx.config.gb_addr_config_fields.num_banks = 1 <<
711 			REG_GET_FIELD(
712 					adev->gfx.config.gb_addr_config,
713 					GB_ADDR_CONFIG,
714 					NUM_BANKS);
715 	adev->gfx.config.gb_addr_config_fields.max_compress_frags = 1 <<
716 			REG_GET_FIELD(
717 					adev->gfx.config.gb_addr_config,
718 					GB_ADDR_CONFIG,
719 					MAX_COMPRESSED_FRAGS);
720 	adev->gfx.config.gb_addr_config_fields.num_rb_per_se = 1 <<
721 			REG_GET_FIELD(
722 					adev->gfx.config.gb_addr_config,
723 					GB_ADDR_CONFIG,
724 					NUM_RB_PER_SE);
725 	adev->gfx.config.gb_addr_config_fields.num_se = 1 <<
726 			REG_GET_FIELD(
727 					adev->gfx.config.gb_addr_config,
728 					GB_ADDR_CONFIG,
729 					NUM_SHADER_ENGINES);
730 	adev->gfx.config.gb_addr_config_fields.pipe_interleave_size = 1 << (8 +
731 			REG_GET_FIELD(
732 					adev->gfx.config.gb_addr_config,
733 					GB_ADDR_CONFIG,
734 					PIPE_INTERLEAVE_SIZE));
735 
736 	return 0;
737 }
738 
gfx_v9_4_3_compute_ring_init(struct amdgpu_device * adev,int ring_id,int xcc_id,int mec,int pipe,int queue)739 static int gfx_v9_4_3_compute_ring_init(struct amdgpu_device *adev, int ring_id,
740 				        int xcc_id, int mec, int pipe, int queue)
741 {
742 	unsigned irq_type;
743 	struct amdgpu_ring *ring = &adev->gfx.compute_ring[ring_id];
744 	unsigned int hw_prio;
745 	uint32_t xcc_doorbell_start;
746 
747 	ring = &adev->gfx.compute_ring[xcc_id * adev->gfx.num_compute_rings +
748 				       ring_id];
749 
750 	/* mec0 is me1 */
751 	ring->xcc_id = xcc_id;
752 	ring->me = mec + 1;
753 	ring->pipe = pipe;
754 	ring->queue = queue;
755 
756 	ring->ring_obj = NULL;
757 	ring->use_doorbell = true;
758 	xcc_doorbell_start = adev->doorbell_index.mec_ring0 +
759 			     xcc_id * adev->doorbell_index.xcc_doorbell_range;
760 	ring->doorbell_index = (xcc_doorbell_start + ring_id) << 1;
761 	ring->eop_gpu_addr = adev->gfx.mec.hpd_eop_gpu_addr +
762 			     (ring_id + xcc_id * adev->gfx.num_compute_rings) *
763 				     GFX9_MEC_HPD_SIZE;
764 	ring->vm_hub = AMDGPU_GFXHUB(xcc_id);
765 	sprintf(ring->name, "comp_%d.%d.%d.%d",
766 			ring->xcc_id, ring->me, ring->pipe, ring->queue);
767 
768 	irq_type = AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP
769 		+ ((ring->me - 1) * adev->gfx.mec.num_pipe_per_mec)
770 		+ ring->pipe;
771 	hw_prio = amdgpu_gfx_is_high_priority_compute_queue(adev, ring) ?
772 			AMDGPU_GFX_PIPE_PRIO_HIGH : AMDGPU_GFX_PIPE_PRIO_NORMAL;
773 	/* type-2 packets are deprecated on MEC, use type-3 instead */
774 	return amdgpu_ring_init(adev, ring, 1024, &adev->gfx.eop_irq, irq_type,
775 				hw_prio, NULL);
776 }
777 
gfx_v9_4_3_sw_init(void * handle)778 static int gfx_v9_4_3_sw_init(void *handle)
779 {
780 	int i, j, k, r, ring_id, xcc_id, num_xcc;
781 	struct amdgpu_kiq *kiq;
782 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
783 
784 	adev->gfx.mec.num_mec = 2;
785 	adev->gfx.mec.num_pipe_per_mec = 4;
786 	adev->gfx.mec.num_queue_per_pipe = 8;
787 
788 	num_xcc = NUM_XCC(adev->gfx.xcc_mask);
789 
790 	/* EOP Event */
791 	r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_GRBM_CP, GFX_9_0__SRCID__CP_EOP_INTERRUPT, &adev->gfx.eop_irq);
792 	if (r)
793 		return r;
794 
795 	/* Privileged reg */
796 	r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_GRBM_CP, GFX_9_0__SRCID__CP_PRIV_REG_FAULT,
797 			      &adev->gfx.priv_reg_irq);
798 	if (r)
799 		return r;
800 
801 	/* Privileged inst */
802 	r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_GRBM_CP, GFX_9_0__SRCID__CP_PRIV_INSTR_FAULT,
803 			      &adev->gfx.priv_inst_irq);
804 	if (r)
805 		return r;
806 
807 	adev->gfx.gfx_current_status = AMDGPU_GFX_NORMAL_MODE;
808 
809 	r = adev->gfx.rlc.funcs->init(adev);
810 	if (r) {
811 		DRM_ERROR("Failed to init rlc BOs!\n");
812 		return r;
813 	}
814 
815 	r = gfx_v9_4_3_mec_init(adev);
816 	if (r) {
817 		DRM_ERROR("Failed to init MEC BOs!\n");
818 		return r;
819 	}
820 
821 	/* set up the compute queues - allocate horizontally across pipes */
822 	for (xcc_id = 0; xcc_id < num_xcc; xcc_id++) {
823 		ring_id = 0;
824 		for (i = 0; i < adev->gfx.mec.num_mec; ++i) {
825 			for (j = 0; j < adev->gfx.mec.num_queue_per_pipe; j++) {
826 				for (k = 0; k < adev->gfx.mec.num_pipe_per_mec;
827 				     k++) {
828 					if (!amdgpu_gfx_is_mec_queue_enabled(
829 							adev, xcc_id, i, k, j))
830 						continue;
831 
832 					r = gfx_v9_4_3_compute_ring_init(adev,
833 								       ring_id,
834 								       xcc_id,
835 								       i, k, j);
836 					if (r)
837 						return r;
838 
839 					ring_id++;
840 				}
841 			}
842 		}
843 
844 		r = amdgpu_gfx_kiq_init(adev, GFX9_MEC_HPD_SIZE, xcc_id);
845 		if (r) {
846 			DRM_ERROR("Failed to init KIQ BOs!\n");
847 			return r;
848 		}
849 
850 		kiq = &adev->gfx.kiq[xcc_id];
851 		r = amdgpu_gfx_kiq_init_ring(adev, &kiq->ring, &kiq->irq, xcc_id);
852 		if (r)
853 			return r;
854 
855 		/* create MQD for all compute queues as wel as KIQ for SRIOV case */
856 		r = amdgpu_gfx_mqd_sw_init(adev,
857 				sizeof(struct v9_mqd_allocation), xcc_id);
858 		if (r)
859 			return r;
860 	}
861 
862 	r = gfx_v9_4_3_gpu_early_init(adev);
863 	if (r)
864 		return r;
865 
866 	r = amdgpu_gfx_ras_sw_init(adev);
867 	if (r)
868 		return r;
869 
870 
871 	if (!amdgpu_sriov_vf(adev))
872 		r = amdgpu_gfx_sysfs_init(adev);
873 
874 	return r;
875 }
876 
gfx_v9_4_3_sw_fini(void * handle)877 static int gfx_v9_4_3_sw_fini(void *handle)
878 {
879 	int i, num_xcc;
880 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
881 
882 	num_xcc = NUM_XCC(adev->gfx.xcc_mask);
883 	for (i = 0; i < adev->gfx.num_compute_rings * num_xcc; i++)
884 		amdgpu_ring_fini(&adev->gfx.compute_ring[i]);
885 
886 	for (i = 0; i < num_xcc; i++) {
887 		amdgpu_gfx_mqd_sw_fini(adev, i);
888 		amdgpu_gfx_kiq_free_ring(&adev->gfx.kiq[i].ring);
889 		amdgpu_gfx_kiq_fini(adev, i);
890 	}
891 
892 	gfx_v9_4_3_mec_fini(adev);
893 	amdgpu_bo_unref(&adev->gfx.rlc.clear_state_obj);
894 	gfx_v9_4_3_free_microcode(adev);
895 	if (!amdgpu_sriov_vf(adev))
896 		amdgpu_gfx_sysfs_fini(adev);
897 
898 	return 0;
899 }
900 
901 #define DEFAULT_SH_MEM_BASES	(0x6000)
gfx_v9_4_3_xcc_init_compute_vmid(struct amdgpu_device * adev,int xcc_id)902 static void gfx_v9_4_3_xcc_init_compute_vmid(struct amdgpu_device *adev,
903 					     int xcc_id)
904 {
905 	int i;
906 	uint32_t sh_mem_config;
907 	uint32_t sh_mem_bases;
908 	uint32_t data;
909 
910 	/*
911 	 * Configure apertures:
912 	 * LDS:         0x60000000'00000000 - 0x60000001'00000000 (4GB)
913 	 * Scratch:     0x60000001'00000000 - 0x60000002'00000000 (4GB)
914 	 * GPUVM:       0x60010000'00000000 - 0x60020000'00000000 (1TB)
915 	 */
916 	sh_mem_bases = DEFAULT_SH_MEM_BASES | (DEFAULT_SH_MEM_BASES << 16);
917 
918 	sh_mem_config = SH_MEM_ADDRESS_MODE_64 |
919 			SH_MEM_ALIGNMENT_MODE_UNALIGNED <<
920 			SH_MEM_CONFIG__ALIGNMENT_MODE__SHIFT;
921 
922 	mutex_lock(&adev->srbm_mutex);
923 	for (i = adev->vm_manager.first_kfd_vmid; i < AMDGPU_NUM_VMID; i++) {
924 		soc15_grbm_select(adev, 0, 0, 0, i, GET_INST(GC, xcc_id));
925 		/* CP and shaders */
926 		WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regSH_MEM_CONFIG, sh_mem_config);
927 		WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regSH_MEM_BASES, sh_mem_bases);
928 
929 		/* Enable trap for each kfd vmid. */
930 		data = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regSPI_GDBG_PER_VMID_CNTL);
931 		data = REG_SET_FIELD(data, SPI_GDBG_PER_VMID_CNTL, TRAP_EN, 1);
932 		WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regSPI_GDBG_PER_VMID_CNTL, data);
933 	}
934 	soc15_grbm_select(adev, 0, 0, 0, 0, GET_INST(GC, xcc_id));
935 	mutex_unlock(&adev->srbm_mutex);
936 
937 	/* Initialize all compute VMIDs to have no GDS, GWS, or OA
938 	   acccess. These should be enabled by FW for target VMIDs. */
939 	for (i = adev->vm_manager.first_kfd_vmid; i < AMDGPU_NUM_VMID; i++) {
940 		WREG32_SOC15_OFFSET(GC, GET_INST(GC, xcc_id), regGDS_VMID0_BASE, 2 * i, 0);
941 		WREG32_SOC15_OFFSET(GC, GET_INST(GC, xcc_id), regGDS_VMID0_SIZE, 2 * i, 0);
942 		WREG32_SOC15_OFFSET(GC, GET_INST(GC, xcc_id), regGDS_GWS_VMID0, i, 0);
943 		WREG32_SOC15_OFFSET(GC, GET_INST(GC, xcc_id), regGDS_OA_VMID0, i, 0);
944 	}
945 }
946 
gfx_v9_4_3_xcc_init_gds_vmid(struct amdgpu_device * adev,int xcc_id)947 static void gfx_v9_4_3_xcc_init_gds_vmid(struct amdgpu_device *adev, int xcc_id)
948 {
949 	int vmid;
950 
951 	/*
952 	 * Initialize all compute and user-gfx VMIDs to have no GDS, GWS, or OA
953 	 * access. Compute VMIDs should be enabled by FW for target VMIDs,
954 	 * the driver can enable them for graphics. VMID0 should maintain
955 	 * access so that HWS firmware can save/restore entries.
956 	 */
957 	for (vmid = 1; vmid < AMDGPU_NUM_VMID; vmid++) {
958 		WREG32_SOC15_OFFSET(GC, GET_INST(GC, xcc_id), regGDS_VMID0_BASE, 2 * vmid, 0);
959 		WREG32_SOC15_OFFSET(GC, GET_INST(GC, xcc_id), regGDS_VMID0_SIZE, 2 * vmid, 0);
960 		WREG32_SOC15_OFFSET(GC, GET_INST(GC, xcc_id), regGDS_GWS_VMID0, vmid, 0);
961 		WREG32_SOC15_OFFSET(GC, GET_INST(GC, xcc_id), regGDS_OA_VMID0, vmid, 0);
962 	}
963 }
964 
gfx_v9_4_3_xcc_constants_init(struct amdgpu_device * adev,int xcc_id)965 static void gfx_v9_4_3_xcc_constants_init(struct amdgpu_device *adev,
966 					  int xcc_id)
967 {
968 	u32 tmp;
969 	int i;
970 
971 	/* XXX SH_MEM regs */
972 	/* where to put LDS, scratch, GPUVM in FSA64 space */
973 	mutex_lock(&adev->srbm_mutex);
974 	for (i = 0; i < adev->vm_manager.id_mgr[AMDGPU_GFXHUB(0)].num_ids; i++) {
975 		soc15_grbm_select(adev, 0, 0, 0, i, GET_INST(GC, xcc_id));
976 		/* CP and shaders */
977 		if (i == 0) {
978 			tmp = REG_SET_FIELD(0, SH_MEM_CONFIG, ALIGNMENT_MODE,
979 					    SH_MEM_ALIGNMENT_MODE_UNALIGNED);
980 			tmp = REG_SET_FIELD(tmp, SH_MEM_CONFIG, RETRY_DISABLE,
981 					    !!adev->gmc.noretry);
982 			WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id),
983 					 regSH_MEM_CONFIG, tmp);
984 			WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id),
985 					 regSH_MEM_BASES, 0);
986 		} else {
987 			tmp = REG_SET_FIELD(0, SH_MEM_CONFIG, ALIGNMENT_MODE,
988 					    SH_MEM_ALIGNMENT_MODE_UNALIGNED);
989 			tmp = REG_SET_FIELD(tmp, SH_MEM_CONFIG, RETRY_DISABLE,
990 					    !!adev->gmc.noretry);
991 			WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id),
992 					 regSH_MEM_CONFIG, tmp);
993 			tmp = REG_SET_FIELD(0, SH_MEM_BASES, PRIVATE_BASE,
994 					    (adev->gmc.private_aperture_start >>
995 					     48));
996 			tmp = REG_SET_FIELD(tmp, SH_MEM_BASES, SHARED_BASE,
997 					    (adev->gmc.shared_aperture_start >>
998 					     48));
999 			WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id),
1000 					 regSH_MEM_BASES, tmp);
1001 		}
1002 	}
1003 	soc15_grbm_select(adev, 0, 0, 0, 0, GET_INST(GC, 0));
1004 
1005 	mutex_unlock(&adev->srbm_mutex);
1006 
1007 	gfx_v9_4_3_xcc_init_compute_vmid(adev, xcc_id);
1008 	gfx_v9_4_3_xcc_init_gds_vmid(adev, xcc_id);
1009 }
1010 
gfx_v9_4_3_constants_init(struct amdgpu_device * adev)1011 static void gfx_v9_4_3_constants_init(struct amdgpu_device *adev)
1012 {
1013 	int i, num_xcc;
1014 
1015 	num_xcc = NUM_XCC(adev->gfx.xcc_mask);
1016 
1017 	gfx_v9_4_3_get_cu_info(adev, &adev->gfx.cu_info);
1018 	adev->gfx.config.db_debug2 =
1019 		RREG32_SOC15(GC, GET_INST(GC, 0), regDB_DEBUG2);
1020 
1021 	for (i = 0; i < num_xcc; i++)
1022 		gfx_v9_4_3_xcc_constants_init(adev, i);
1023 }
1024 
1025 static void
gfx_v9_4_3_xcc_enable_save_restore_machine(struct amdgpu_device * adev,int xcc_id)1026 gfx_v9_4_3_xcc_enable_save_restore_machine(struct amdgpu_device *adev,
1027 					   int xcc_id)
1028 {
1029 	WREG32_FIELD15_PREREG(GC, GET_INST(GC, xcc_id), RLC_SRM_CNTL, SRM_ENABLE, 1);
1030 }
1031 
gfx_v9_4_3_xcc_init_pg(struct amdgpu_device * adev,int xcc_id)1032 static void gfx_v9_4_3_xcc_init_pg(struct amdgpu_device *adev, int xcc_id)
1033 {
1034 	/*
1035 	 * Rlc save restore list is workable since v2_1.
1036 	 * And it's needed by gfxoff feature.
1037 	 */
1038 	if (adev->gfx.rlc.is_rlc_v2_1)
1039 		gfx_v9_4_3_xcc_enable_save_restore_machine(adev, xcc_id);
1040 }
1041 
gfx_v9_4_3_xcc_disable_gpa_mode(struct amdgpu_device * adev,int xcc_id)1042 static void gfx_v9_4_3_xcc_disable_gpa_mode(struct amdgpu_device *adev, int xcc_id)
1043 {
1044 	uint32_t data;
1045 
1046 	data = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regCPC_PSP_DEBUG);
1047 	data |= CPC_PSP_DEBUG__UTCL2IUGPAOVERRIDE_MASK;
1048 	WREG32_SOC15(GC, GET_INST(GC, xcc_id), regCPC_PSP_DEBUG, data);
1049 }
1050 
gfx_v9_4_3_is_rlc_enabled(struct amdgpu_device * adev)1051 static bool gfx_v9_4_3_is_rlc_enabled(struct amdgpu_device *adev)
1052 {
1053 	uint32_t rlc_setting;
1054 
1055 	/* if RLC is not enabled, do nothing */
1056 	rlc_setting = RREG32_SOC15(GC, GET_INST(GC, 0), regRLC_CNTL);
1057 	if (!(rlc_setting & RLC_CNTL__RLC_ENABLE_F32_MASK))
1058 		return false;
1059 
1060 	return true;
1061 }
1062 
gfx_v9_4_3_xcc_set_safe_mode(struct amdgpu_device * adev,int xcc_id)1063 static void gfx_v9_4_3_xcc_set_safe_mode(struct amdgpu_device *adev, int xcc_id)
1064 {
1065 	uint32_t data;
1066 	unsigned i;
1067 
1068 	data = RLC_SAFE_MODE__CMD_MASK;
1069 	data |= (1 << RLC_SAFE_MODE__MESSAGE__SHIFT);
1070 	WREG32_SOC15(GC, GET_INST(GC, xcc_id), regRLC_SAFE_MODE, data);
1071 
1072 	/* wait for RLC_SAFE_MODE */
1073 	for (i = 0; i < adev->usec_timeout; i++) {
1074 		if (!REG_GET_FIELD(RREG32_SOC15(GC, GET_INST(GC, xcc_id), regRLC_SAFE_MODE), RLC_SAFE_MODE, CMD))
1075 			break;
1076 		udelay(1);
1077 	}
1078 }
1079 
gfx_v9_4_3_xcc_unset_safe_mode(struct amdgpu_device * adev,int xcc_id)1080 static void gfx_v9_4_3_xcc_unset_safe_mode(struct amdgpu_device *adev,
1081 					   int xcc_id)
1082 {
1083 	uint32_t data;
1084 
1085 	data = RLC_SAFE_MODE__CMD_MASK;
1086 	WREG32_SOC15(GC, GET_INST(GC, xcc_id), regRLC_SAFE_MODE, data);
1087 }
1088 
gfx_v9_4_3_init_rlcg_reg_access_ctrl(struct amdgpu_device * adev)1089 static void gfx_v9_4_3_init_rlcg_reg_access_ctrl(struct amdgpu_device *adev)
1090 {
1091 	int xcc_id, num_xcc;
1092 	struct amdgpu_rlcg_reg_access_ctrl *reg_access_ctrl;
1093 
1094 	num_xcc = NUM_XCC(adev->gfx.xcc_mask);
1095 	for (xcc_id = 0; xcc_id < num_xcc; xcc_id++) {
1096 		reg_access_ctrl = &adev->gfx.rlc.reg_access_ctrl[GET_INST(GC, xcc_id)];
1097 		reg_access_ctrl->scratch_reg0 = SOC15_REG_OFFSET(GC, GET_INST(GC, xcc_id), regSCRATCH_REG0);
1098 		reg_access_ctrl->scratch_reg1 = SOC15_REG_OFFSET(GC, GET_INST(GC, xcc_id), regSCRATCH_REG1);
1099 		reg_access_ctrl->scratch_reg2 = SOC15_REG_OFFSET(GC, GET_INST(GC, xcc_id), regSCRATCH_REG2);
1100 		reg_access_ctrl->scratch_reg3 = SOC15_REG_OFFSET(GC, GET_INST(GC, xcc_id), regSCRATCH_REG3);
1101 		reg_access_ctrl->grbm_cntl = SOC15_REG_OFFSET(GC, GET_INST(GC, xcc_id), regGRBM_GFX_CNTL);
1102 		reg_access_ctrl->grbm_idx = SOC15_REG_OFFSET(GC, GET_INST(GC, xcc_id), regGRBM_GFX_INDEX);
1103 		reg_access_ctrl->spare_int = SOC15_REG_OFFSET(GC, GET_INST(GC, xcc_id), regRLC_SPARE_INT);
1104 	}
1105 }
1106 
gfx_v9_4_3_rlc_init(struct amdgpu_device * adev)1107 static int gfx_v9_4_3_rlc_init(struct amdgpu_device *adev)
1108 {
1109 	/* init spm vmid with 0xf */
1110 	if (adev->gfx.rlc.funcs->update_spm_vmid)
1111 		adev->gfx.rlc.funcs->update_spm_vmid(adev, 0xf);
1112 
1113 	return 0;
1114 }
1115 
gfx_v9_4_3_xcc_wait_for_rlc_serdes(struct amdgpu_device * adev,int xcc_id)1116 static void gfx_v9_4_3_xcc_wait_for_rlc_serdes(struct amdgpu_device *adev,
1117 					       int xcc_id)
1118 {
1119 	u32 i, j, k;
1120 	u32 mask;
1121 
1122 	mutex_lock(&adev->grbm_idx_mutex);
1123 	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
1124 		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
1125 			gfx_v9_4_3_xcc_select_se_sh(adev, i, j, 0xffffffff,
1126 						    xcc_id);
1127 			for (k = 0; k < adev->usec_timeout; k++) {
1128 				if (RREG32_SOC15(GC, GET_INST(GC, xcc_id), regRLC_SERDES_CU_MASTER_BUSY) == 0)
1129 					break;
1130 				udelay(1);
1131 			}
1132 			if (k == adev->usec_timeout) {
1133 				gfx_v9_4_3_xcc_select_se_sh(adev, 0xffffffff,
1134 							    0xffffffff,
1135 							    0xffffffff, xcc_id);
1136 				mutex_unlock(&adev->grbm_idx_mutex);
1137 				DRM_INFO("Timeout wait for RLC serdes %u,%u\n",
1138 					 i, j);
1139 				return;
1140 			}
1141 		}
1142 	}
1143 	gfx_v9_4_3_xcc_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff,
1144 				    xcc_id);
1145 	mutex_unlock(&adev->grbm_idx_mutex);
1146 
1147 	mask = RLC_SERDES_NONCU_MASTER_BUSY__SE_MASTER_BUSY_MASK |
1148 		RLC_SERDES_NONCU_MASTER_BUSY__GC_MASTER_BUSY_MASK |
1149 		RLC_SERDES_NONCU_MASTER_BUSY__TC0_MASTER_BUSY_MASK |
1150 		RLC_SERDES_NONCU_MASTER_BUSY__TC1_MASTER_BUSY_MASK;
1151 	for (k = 0; k < adev->usec_timeout; k++) {
1152 		if ((RREG32_SOC15(GC, GET_INST(GC, xcc_id), regRLC_SERDES_NONCU_MASTER_BUSY) & mask) == 0)
1153 			break;
1154 		udelay(1);
1155 	}
1156 }
1157 
gfx_v9_4_3_xcc_enable_gui_idle_interrupt(struct amdgpu_device * adev,bool enable,int xcc_id)1158 static void gfx_v9_4_3_xcc_enable_gui_idle_interrupt(struct amdgpu_device *adev,
1159 						     bool enable, int xcc_id)
1160 {
1161 	u32 tmp;
1162 
1163 	/* These interrupts should be enabled to drive DS clock */
1164 
1165 	tmp = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_INT_CNTL_RING0);
1166 
1167 	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CNTX_BUSY_INT_ENABLE, enable ? 1 : 0);
1168 	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CNTX_EMPTY_INT_ENABLE, enable ? 1 : 0);
1169 	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CMP_BUSY_INT_ENABLE, enable ? 1 : 0);
1170 
1171 	WREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_INT_CNTL_RING0, tmp);
1172 }
1173 
gfx_v9_4_3_xcc_rlc_stop(struct amdgpu_device * adev,int xcc_id)1174 static void gfx_v9_4_3_xcc_rlc_stop(struct amdgpu_device *adev, int xcc_id)
1175 {
1176 	WREG32_FIELD15_PREREG(GC, GET_INST(GC, xcc_id), RLC_CNTL,
1177 			      RLC_ENABLE_F32, 0);
1178 	gfx_v9_4_3_xcc_enable_gui_idle_interrupt(adev, false, xcc_id);
1179 	gfx_v9_4_3_xcc_wait_for_rlc_serdes(adev, xcc_id);
1180 }
1181 
gfx_v9_4_3_rlc_stop(struct amdgpu_device * adev)1182 static void gfx_v9_4_3_rlc_stop(struct amdgpu_device *adev)
1183 {
1184 	int i, num_xcc;
1185 
1186 	num_xcc = NUM_XCC(adev->gfx.xcc_mask);
1187 	for (i = 0; i < num_xcc; i++)
1188 		gfx_v9_4_3_xcc_rlc_stop(adev, i);
1189 }
1190 
gfx_v9_4_3_xcc_rlc_reset(struct amdgpu_device * adev,int xcc_id)1191 static void gfx_v9_4_3_xcc_rlc_reset(struct amdgpu_device *adev, int xcc_id)
1192 {
1193 	WREG32_FIELD15_PREREG(GC, GET_INST(GC, xcc_id), GRBM_SOFT_RESET,
1194 			      SOFT_RESET_RLC, 1);
1195 	udelay(50);
1196 	WREG32_FIELD15_PREREG(GC, GET_INST(GC, xcc_id), GRBM_SOFT_RESET,
1197 			      SOFT_RESET_RLC, 0);
1198 	udelay(50);
1199 }
1200 
gfx_v9_4_3_rlc_reset(struct amdgpu_device * adev)1201 static void gfx_v9_4_3_rlc_reset(struct amdgpu_device *adev)
1202 {
1203 	int i, num_xcc;
1204 
1205 	num_xcc = NUM_XCC(adev->gfx.xcc_mask);
1206 	for (i = 0; i < num_xcc; i++)
1207 		gfx_v9_4_3_xcc_rlc_reset(adev, i);
1208 }
1209 
gfx_v9_4_3_xcc_rlc_start(struct amdgpu_device * adev,int xcc_id)1210 static void gfx_v9_4_3_xcc_rlc_start(struct amdgpu_device *adev, int xcc_id)
1211 {
1212 	WREG32_FIELD15_PREREG(GC, GET_INST(GC, xcc_id), RLC_CNTL,
1213 			      RLC_ENABLE_F32, 1);
1214 	udelay(50);
1215 
1216 	/* carrizo do enable cp interrupt after cp inited */
1217 	if (!(adev->flags & AMD_IS_APU)) {
1218 		gfx_v9_4_3_xcc_enable_gui_idle_interrupt(adev, true, xcc_id);
1219 		udelay(50);
1220 	}
1221 }
1222 
gfx_v9_4_3_rlc_start(struct amdgpu_device * adev)1223 static void gfx_v9_4_3_rlc_start(struct amdgpu_device *adev)
1224 {
1225 #ifdef AMDGPU_RLC_DEBUG_RETRY
1226 	u32 rlc_ucode_ver;
1227 #endif
1228 	int i, num_xcc;
1229 
1230 	num_xcc = NUM_XCC(adev->gfx.xcc_mask);
1231 	for (i = 0; i < num_xcc; i++) {
1232 		gfx_v9_4_3_xcc_rlc_start(adev, i);
1233 #ifdef AMDGPU_RLC_DEBUG_RETRY
1234 		/* RLC_GPM_GENERAL_6 : RLC Ucode version */
1235 		rlc_ucode_ver = RREG32_SOC15(GC, GET_INST(GC, i), regRLC_GPM_GENERAL_6);
1236 		if (rlc_ucode_ver == 0x108) {
1237 			dev_info(adev->dev,
1238 				 "Using rlc debug ucode. regRLC_GPM_GENERAL_6 ==0x08%x / fw_ver == %i \n",
1239 				 rlc_ucode_ver, adev->gfx.rlc_fw_version);
1240 			/* RLC_GPM_TIMER_INT_3 : Timer interval in RefCLK cycles,
1241 			 * default is 0x9C4 to create a 100us interval */
1242 			WREG32_SOC15(GC, GET_INST(GC, i), regRLC_GPM_TIMER_INT_3, 0x9C4);
1243 			/* RLC_GPM_GENERAL_12 : Minimum gap between wptr and rptr
1244 			 * to disable the page fault retry interrupts, default is
1245 			 * 0x100 (256) */
1246 			WREG32_SOC15(GC, GET_INST(GC, i), regRLC_GPM_GENERAL_12, 0x100);
1247 		}
1248 #endif
1249 	}
1250 }
1251 
gfx_v9_4_3_xcc_rlc_load_microcode(struct amdgpu_device * adev,int xcc_id)1252 static int gfx_v9_4_3_xcc_rlc_load_microcode(struct amdgpu_device *adev,
1253 					     int xcc_id)
1254 {
1255 	const struct rlc_firmware_header_v2_0 *hdr;
1256 	const __le32 *fw_data;
1257 	unsigned i, fw_size;
1258 
1259 	if (!adev->gfx.rlc_fw)
1260 		return -EINVAL;
1261 
1262 	hdr = (const struct rlc_firmware_header_v2_0 *)adev->gfx.rlc_fw->data;
1263 	amdgpu_ucode_print_rlc_hdr(&hdr->header);
1264 
1265 	fw_data = (const __le32 *)(adev->gfx.rlc_fw->data +
1266 			   le32_to_cpu(hdr->header.ucode_array_offset_bytes));
1267 	fw_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4;
1268 
1269 	WREG32_SOC15(GC, GET_INST(GC, xcc_id), regRLC_GPM_UCODE_ADDR,
1270 			RLCG_UCODE_LOADING_START_ADDRESS);
1271 	for (i = 0; i < fw_size; i++) {
1272 		if (amdgpu_emu_mode == 1 && i % 100 == 0) {
1273 			dev_info(adev->dev, "Write RLC ucode data %u DWs\n", i);
1274 			msleep(1);
1275 		}
1276 		WREG32_SOC15(GC, GET_INST(GC, xcc_id), regRLC_GPM_UCODE_DATA, le32_to_cpup(fw_data++));
1277 	}
1278 	WREG32_SOC15(GC, GET_INST(GC, xcc_id), regRLC_GPM_UCODE_ADDR, adev->gfx.rlc_fw_version);
1279 
1280 	return 0;
1281 }
1282 
gfx_v9_4_3_xcc_rlc_resume(struct amdgpu_device * adev,int xcc_id)1283 static int gfx_v9_4_3_xcc_rlc_resume(struct amdgpu_device *adev, int xcc_id)
1284 {
1285 	int r;
1286 
1287 	if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP) {
1288 		gfx_v9_4_3_xcc_rlc_stop(adev, xcc_id);
1289 		/* legacy rlc firmware loading */
1290 		r = gfx_v9_4_3_xcc_rlc_load_microcode(adev, xcc_id);
1291 		if (r)
1292 			return r;
1293 		gfx_v9_4_3_xcc_rlc_start(adev, xcc_id);
1294 	}
1295 
1296 	amdgpu_gfx_rlc_enter_safe_mode(adev, xcc_id);
1297 	/* disable CG */
1298 	WREG32_SOC15(GC, GET_INST(GC, xcc_id), regRLC_CGCG_CGLS_CTRL, 0);
1299 	gfx_v9_4_3_xcc_init_pg(adev, xcc_id);
1300 	amdgpu_gfx_rlc_exit_safe_mode(adev, xcc_id);
1301 
1302 	return 0;
1303 }
1304 
gfx_v9_4_3_rlc_resume(struct amdgpu_device * adev)1305 static int gfx_v9_4_3_rlc_resume(struct amdgpu_device *adev)
1306 {
1307 	int r, i, num_xcc;
1308 
1309 	if (amdgpu_sriov_vf(adev))
1310 		return 0;
1311 
1312 	num_xcc = NUM_XCC(adev->gfx.xcc_mask);
1313 	for (i = 0; i < num_xcc; i++) {
1314 		r = gfx_v9_4_3_xcc_rlc_resume(adev, i);
1315 		if (r)
1316 			return r;
1317 	}
1318 
1319 	return 0;
1320 }
1321 
gfx_v9_4_3_update_spm_vmid(struct amdgpu_device * adev,unsigned vmid)1322 static void gfx_v9_4_3_update_spm_vmid(struct amdgpu_device *adev,
1323 				       unsigned vmid)
1324 {
1325 	u32 reg, data;
1326 
1327 	reg = SOC15_REG_OFFSET(GC, GET_INST(GC, 0), regRLC_SPM_MC_CNTL);
1328 	if (amdgpu_sriov_is_pp_one_vf(adev))
1329 		data = RREG32_NO_KIQ(reg);
1330 	else
1331 		data = RREG32(reg);
1332 
1333 	data &= ~RLC_SPM_MC_CNTL__RLC_SPM_VMID_MASK;
1334 	data |= (vmid & RLC_SPM_MC_CNTL__RLC_SPM_VMID_MASK) << RLC_SPM_MC_CNTL__RLC_SPM_VMID__SHIFT;
1335 
1336 	if (amdgpu_sriov_is_pp_one_vf(adev))
1337 		WREG32_SOC15_NO_KIQ(GC, GET_INST(GC, 0), regRLC_SPM_MC_CNTL, data);
1338 	else
1339 		WREG32_SOC15(GC, GET_INST(GC, 0), regRLC_SPM_MC_CNTL, data);
1340 }
1341 
1342 static const struct soc15_reg_rlcg rlcg_access_gc_9_4_3[] = {
1343 	{SOC15_REG_ENTRY(GC, 0, regGRBM_GFX_INDEX)},
1344 	{SOC15_REG_ENTRY(GC, 0, regSQ_IND_INDEX)},
1345 };
1346 
gfx_v9_4_3_check_rlcg_range(struct amdgpu_device * adev,uint32_t offset,struct soc15_reg_rlcg * entries,int arr_size)1347 static bool gfx_v9_4_3_check_rlcg_range(struct amdgpu_device *adev,
1348 					uint32_t offset,
1349 					struct soc15_reg_rlcg *entries, int arr_size)
1350 {
1351 	int i, inst;
1352 	uint32_t reg;
1353 
1354 	if (!entries)
1355 		return false;
1356 
1357 	for (i = 0; i < arr_size; i++) {
1358 		const struct soc15_reg_rlcg *entry;
1359 
1360 		entry = &entries[i];
1361 		inst = adev->ip_map.logical_to_dev_inst ?
1362 			       adev->ip_map.logical_to_dev_inst(
1363 				       adev, entry->hwip, entry->instance) :
1364 			       entry->instance;
1365 		reg = adev->reg_offset[entry->hwip][inst][entry->segment] +
1366 		      entry->reg;
1367 		if (offset == reg)
1368 			return true;
1369 	}
1370 
1371 	return false;
1372 }
1373 
gfx_v9_4_3_is_rlcg_access_range(struct amdgpu_device * adev,u32 offset)1374 static bool gfx_v9_4_3_is_rlcg_access_range(struct amdgpu_device *adev, u32 offset)
1375 {
1376 	return gfx_v9_4_3_check_rlcg_range(adev, offset,
1377 					(void *)rlcg_access_gc_9_4_3,
1378 					ARRAY_SIZE(rlcg_access_gc_9_4_3));
1379 }
1380 
gfx_v9_4_3_xcc_cp_compute_enable(struct amdgpu_device * adev,bool enable,int xcc_id)1381 static void gfx_v9_4_3_xcc_cp_compute_enable(struct amdgpu_device *adev,
1382 					     bool enable, int xcc_id)
1383 {
1384 	if (enable) {
1385 		WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_MEC_CNTL, 0);
1386 	} else {
1387 		WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_MEC_CNTL,
1388 			(CP_MEC_CNTL__MEC_ME1_HALT_MASK | CP_MEC_CNTL__MEC_ME2_HALT_MASK));
1389 		adev->gfx.kiq[xcc_id].ring.sched.ready = false;
1390 	}
1391 	udelay(50);
1392 }
1393 
gfx_v9_4_3_xcc_cp_compute_load_microcode(struct amdgpu_device * adev,int xcc_id)1394 static int gfx_v9_4_3_xcc_cp_compute_load_microcode(struct amdgpu_device *adev,
1395 						    int xcc_id)
1396 {
1397 	const struct gfx_firmware_header_v1_0 *mec_hdr;
1398 	const __le32 *fw_data;
1399 	unsigned i;
1400 	u32 tmp;
1401 	u32 mec_ucode_addr_offset;
1402 	u32 mec_ucode_data_offset;
1403 
1404 	if (!adev->gfx.mec_fw)
1405 		return -EINVAL;
1406 
1407 	gfx_v9_4_3_xcc_cp_compute_enable(adev, false, xcc_id);
1408 
1409 	mec_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data;
1410 	amdgpu_ucode_print_gfx_hdr(&mec_hdr->header);
1411 
1412 	fw_data = (const __le32 *)
1413 		(adev->gfx.mec_fw->data +
1414 		 le32_to_cpu(mec_hdr->header.ucode_array_offset_bytes));
1415 	tmp = 0;
1416 	tmp = REG_SET_FIELD(tmp, CP_CPC_IC_BASE_CNTL, VMID, 0);
1417 	tmp = REG_SET_FIELD(tmp, CP_CPC_IC_BASE_CNTL, CACHE_POLICY, 0);
1418 	WREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_CPC_IC_BASE_CNTL, tmp);
1419 
1420 	WREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_CPC_IC_BASE_LO,
1421 		adev->gfx.mec.mec_fw_gpu_addr & 0xFFFFF000);
1422 	WREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_CPC_IC_BASE_HI,
1423 		upper_32_bits(adev->gfx.mec.mec_fw_gpu_addr));
1424 
1425 	mec_ucode_addr_offset =
1426 		SOC15_REG_OFFSET(GC, GET_INST(GC, xcc_id), regCP_MEC_ME1_UCODE_ADDR);
1427 	mec_ucode_data_offset =
1428 		SOC15_REG_OFFSET(GC, GET_INST(GC, xcc_id), regCP_MEC_ME1_UCODE_DATA);
1429 
1430 	/* MEC1 */
1431 	WREG32(mec_ucode_addr_offset, mec_hdr->jt_offset);
1432 	for (i = 0; i < mec_hdr->jt_size; i++)
1433 		WREG32(mec_ucode_data_offset,
1434 		       le32_to_cpup(fw_data + mec_hdr->jt_offset + i));
1435 
1436 	WREG32(mec_ucode_addr_offset, adev->gfx.mec_fw_version);
1437 	/* Todo : Loading MEC2 firmware is only necessary if MEC2 should run different microcode than MEC1. */
1438 
1439 	return 0;
1440 }
1441 
1442 /* KIQ functions */
gfx_v9_4_3_xcc_kiq_setting(struct amdgpu_ring * ring,int xcc_id)1443 static void gfx_v9_4_3_xcc_kiq_setting(struct amdgpu_ring *ring, int xcc_id)
1444 {
1445 	uint32_t tmp;
1446 	struct amdgpu_device *adev = ring->adev;
1447 
1448 	/* tell RLC which is KIQ queue */
1449 	tmp = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regRLC_CP_SCHEDULERS);
1450 	tmp &= 0xffffff00;
1451 	tmp |= (ring->me << 5) | (ring->pipe << 3) | (ring->queue);
1452 	WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regRLC_CP_SCHEDULERS, tmp);
1453 	tmp |= 0x80;
1454 	WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regRLC_CP_SCHEDULERS, tmp);
1455 }
1456 
gfx_v9_4_3_mqd_set_priority(struct amdgpu_ring * ring,struct v9_mqd * mqd)1457 static void gfx_v9_4_3_mqd_set_priority(struct amdgpu_ring *ring, struct v9_mqd *mqd)
1458 {
1459 	struct amdgpu_device *adev = ring->adev;
1460 
1461 	if (ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE) {
1462 		if (amdgpu_gfx_is_high_priority_compute_queue(adev, ring)) {
1463 			mqd->cp_hqd_pipe_priority = AMDGPU_GFX_PIPE_PRIO_HIGH;
1464 			mqd->cp_hqd_queue_priority =
1465 				AMDGPU_GFX_QUEUE_PRIORITY_MAXIMUM;
1466 		}
1467 	}
1468 }
1469 
gfx_v9_4_3_xcc_mqd_init(struct amdgpu_ring * ring,int xcc_id)1470 static int gfx_v9_4_3_xcc_mqd_init(struct amdgpu_ring *ring, int xcc_id)
1471 {
1472 	struct amdgpu_device *adev = ring->adev;
1473 	struct v9_mqd *mqd = ring->mqd_ptr;
1474 	uint64_t hqd_gpu_addr, wb_gpu_addr, eop_base_addr;
1475 	uint32_t tmp;
1476 
1477 	mqd->header = 0xC0310800;
1478 	mqd->compute_pipelinestat_enable = 0x00000001;
1479 	mqd->compute_static_thread_mgmt_se0 = 0xffffffff;
1480 	mqd->compute_static_thread_mgmt_se1 = 0xffffffff;
1481 	mqd->compute_static_thread_mgmt_se2 = 0xffffffff;
1482 	mqd->compute_static_thread_mgmt_se3 = 0xffffffff;
1483 	mqd->compute_misc_reserved = 0x00000003;
1484 
1485 	mqd->dynamic_cu_mask_addr_lo =
1486 		lower_32_bits(ring->mqd_gpu_addr
1487 			      + offsetof(struct v9_mqd_allocation, dynamic_cu_mask));
1488 	mqd->dynamic_cu_mask_addr_hi =
1489 		upper_32_bits(ring->mqd_gpu_addr
1490 			      + offsetof(struct v9_mqd_allocation, dynamic_cu_mask));
1491 
1492 	eop_base_addr = ring->eop_gpu_addr >> 8;
1493 	mqd->cp_hqd_eop_base_addr_lo = eop_base_addr;
1494 	mqd->cp_hqd_eop_base_addr_hi = upper_32_bits(eop_base_addr);
1495 
1496 	/* set the EOP size, register value is 2^(EOP_SIZE+1) dwords */
1497 	tmp = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_HQD_EOP_CONTROL);
1498 	tmp = REG_SET_FIELD(tmp, CP_HQD_EOP_CONTROL, EOP_SIZE,
1499 			(order_base_2(GFX9_MEC_HPD_SIZE / 4) - 1));
1500 
1501 	mqd->cp_hqd_eop_control = tmp;
1502 
1503 	/* enable doorbell? */
1504 	tmp = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_HQD_PQ_DOORBELL_CONTROL);
1505 
1506 	if (ring->use_doorbell) {
1507 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
1508 				    DOORBELL_OFFSET, ring->doorbell_index);
1509 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
1510 				    DOORBELL_EN, 1);
1511 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
1512 				    DOORBELL_SOURCE, 0);
1513 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
1514 				    DOORBELL_HIT, 0);
1515 	} else {
1516 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
1517 					 DOORBELL_EN, 0);
1518 	}
1519 
1520 	mqd->cp_hqd_pq_doorbell_control = tmp;
1521 
1522 	/* disable the queue if it's active */
1523 	ring->wptr = 0;
1524 	mqd->cp_hqd_dequeue_request = 0;
1525 	mqd->cp_hqd_pq_rptr = 0;
1526 	mqd->cp_hqd_pq_wptr_lo = 0;
1527 	mqd->cp_hqd_pq_wptr_hi = 0;
1528 
1529 	/* set the pointer to the MQD */
1530 	mqd->cp_mqd_base_addr_lo = ring->mqd_gpu_addr & 0xfffffffc;
1531 	mqd->cp_mqd_base_addr_hi = upper_32_bits(ring->mqd_gpu_addr);
1532 
1533 	/* set MQD vmid to 0 */
1534 	tmp = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_MQD_CONTROL);
1535 	tmp = REG_SET_FIELD(tmp, CP_MQD_CONTROL, VMID, 0);
1536 	mqd->cp_mqd_control = tmp;
1537 
1538 	/* set the pointer to the HQD, this is similar CP_RB0_BASE/_HI */
1539 	hqd_gpu_addr = ring->gpu_addr >> 8;
1540 	mqd->cp_hqd_pq_base_lo = hqd_gpu_addr;
1541 	mqd->cp_hqd_pq_base_hi = upper_32_bits(hqd_gpu_addr);
1542 
1543 	/* set up the HQD, this is similar to CP_RB0_CNTL */
1544 	tmp = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_HQD_PQ_CONTROL);
1545 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, QUEUE_SIZE,
1546 			    (order_base_2(ring->ring_size / 4) - 1));
1547 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, RPTR_BLOCK_SIZE,
1548 			((order_base_2(AMDGPU_GPU_PAGE_SIZE / 4) - 1) << 8));
1549 #ifdef __BIG_ENDIAN
1550 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, ENDIAN_SWAP, 1);
1551 #endif
1552 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, UNORD_DISPATCH, 0);
1553 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, ROQ_PQ_IB_FLIP, 0);
1554 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, PRIV_STATE, 1);
1555 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, KMD_QUEUE, 1);
1556 	mqd->cp_hqd_pq_control = tmp;
1557 
1558 	/* set the wb address whether it's enabled or not */
1559 	wb_gpu_addr = adev->wb.gpu_addr + (ring->rptr_offs * 4);
1560 	mqd->cp_hqd_pq_rptr_report_addr_lo = wb_gpu_addr & 0xfffffffc;
1561 	mqd->cp_hqd_pq_rptr_report_addr_hi =
1562 		upper_32_bits(wb_gpu_addr) & 0xffff;
1563 
1564 	/* only used if CP_PQ_WPTR_POLL_CNTL.CP_PQ_WPTR_POLL_CNTL__EN_MASK=1 */
1565 	wb_gpu_addr = adev->wb.gpu_addr + (ring->wptr_offs * 4);
1566 	mqd->cp_hqd_pq_wptr_poll_addr_lo = wb_gpu_addr & 0xfffffffc;
1567 	mqd->cp_hqd_pq_wptr_poll_addr_hi = upper_32_bits(wb_gpu_addr) & 0xffff;
1568 
1569 	/* reset read and write pointers, similar to CP_RB0_WPTR/_RPTR */
1570 	ring->wptr = 0;
1571 	mqd->cp_hqd_pq_rptr = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_HQD_PQ_RPTR);
1572 
1573 	/* set the vmid for the queue */
1574 	mqd->cp_hqd_vmid = 0;
1575 
1576 	tmp = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_HQD_PERSISTENT_STATE);
1577 	tmp = REG_SET_FIELD(tmp, CP_HQD_PERSISTENT_STATE, PRELOAD_SIZE, 0x53);
1578 	mqd->cp_hqd_persistent_state = tmp;
1579 
1580 	/* set MIN_IB_AVAIL_SIZE */
1581 	tmp = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_HQD_IB_CONTROL);
1582 	tmp = REG_SET_FIELD(tmp, CP_HQD_IB_CONTROL, MIN_IB_AVAIL_SIZE, 3);
1583 	mqd->cp_hqd_ib_control = tmp;
1584 
1585 	/* set static priority for a queue/ring */
1586 	gfx_v9_4_3_mqd_set_priority(ring, mqd);
1587 	mqd->cp_hqd_quantum = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_HQD_QUANTUM);
1588 
1589 	/* map_queues packet doesn't need activate the queue,
1590 	 * so only kiq need set this field.
1591 	 */
1592 	if (ring->funcs->type == AMDGPU_RING_TYPE_KIQ)
1593 		mqd->cp_hqd_active = 1;
1594 
1595 	return 0;
1596 }
1597 
gfx_v9_4_3_xcc_kiq_init_register(struct amdgpu_ring * ring,int xcc_id)1598 static int gfx_v9_4_3_xcc_kiq_init_register(struct amdgpu_ring *ring,
1599 					    int xcc_id)
1600 {
1601 	struct amdgpu_device *adev = ring->adev;
1602 	struct v9_mqd *mqd = ring->mqd_ptr;
1603 	int j;
1604 
1605 	/* disable wptr polling */
1606 	WREG32_FIELD15_PREREG(GC, GET_INST(GC, xcc_id), CP_PQ_WPTR_POLL_CNTL, EN, 0);
1607 
1608 	WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_HQD_EOP_BASE_ADDR,
1609 	       mqd->cp_hqd_eop_base_addr_lo);
1610 	WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_HQD_EOP_BASE_ADDR_HI,
1611 	       mqd->cp_hqd_eop_base_addr_hi);
1612 
1613 	/* set the EOP size, register value is 2^(EOP_SIZE+1) dwords */
1614 	WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_HQD_EOP_CONTROL,
1615 	       mqd->cp_hqd_eop_control);
1616 
1617 	/* enable doorbell? */
1618 	WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_HQD_PQ_DOORBELL_CONTROL,
1619 	       mqd->cp_hqd_pq_doorbell_control);
1620 
1621 	/* disable the queue if it's active */
1622 	if (RREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_HQD_ACTIVE) & 1) {
1623 		WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_HQD_DEQUEUE_REQUEST, 1);
1624 		for (j = 0; j < adev->usec_timeout; j++) {
1625 			if (!(RREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_HQD_ACTIVE) & 1))
1626 				break;
1627 			udelay(1);
1628 		}
1629 		WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_HQD_DEQUEUE_REQUEST,
1630 		       mqd->cp_hqd_dequeue_request);
1631 		WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_HQD_PQ_RPTR,
1632 		       mqd->cp_hqd_pq_rptr);
1633 		WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_HQD_PQ_WPTR_LO,
1634 		       mqd->cp_hqd_pq_wptr_lo);
1635 		WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_HQD_PQ_WPTR_HI,
1636 		       mqd->cp_hqd_pq_wptr_hi);
1637 	}
1638 
1639 	/* set the pointer to the MQD */
1640 	WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_MQD_BASE_ADDR,
1641 	       mqd->cp_mqd_base_addr_lo);
1642 	WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_MQD_BASE_ADDR_HI,
1643 	       mqd->cp_mqd_base_addr_hi);
1644 
1645 	/* set MQD vmid to 0 */
1646 	WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_MQD_CONTROL,
1647 	       mqd->cp_mqd_control);
1648 
1649 	/* set the pointer to the HQD, this is similar CP_RB0_BASE/_HI */
1650 	WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_HQD_PQ_BASE,
1651 	       mqd->cp_hqd_pq_base_lo);
1652 	WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_HQD_PQ_BASE_HI,
1653 	       mqd->cp_hqd_pq_base_hi);
1654 
1655 	/* set up the HQD, this is similar to CP_RB0_CNTL */
1656 	WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_HQD_PQ_CONTROL,
1657 	       mqd->cp_hqd_pq_control);
1658 
1659 	/* set the wb address whether it's enabled or not */
1660 	WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_HQD_PQ_RPTR_REPORT_ADDR,
1661 				mqd->cp_hqd_pq_rptr_report_addr_lo);
1662 	WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_HQD_PQ_RPTR_REPORT_ADDR_HI,
1663 				mqd->cp_hqd_pq_rptr_report_addr_hi);
1664 
1665 	/* only used if CP_PQ_WPTR_POLL_CNTL.CP_PQ_WPTR_POLL_CNTL__EN_MASK=1 */
1666 	WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_HQD_PQ_WPTR_POLL_ADDR,
1667 	       mqd->cp_hqd_pq_wptr_poll_addr_lo);
1668 	WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_HQD_PQ_WPTR_POLL_ADDR_HI,
1669 	       mqd->cp_hqd_pq_wptr_poll_addr_hi);
1670 
1671 	/* enable the doorbell if requested */
1672 	if (ring->use_doorbell) {
1673 		WREG32_SOC15(
1674 			GC, GET_INST(GC, xcc_id),
1675 			regCP_MEC_DOORBELL_RANGE_LOWER,
1676 			((adev->doorbell_index.kiq +
1677 			  xcc_id * adev->doorbell_index.xcc_doorbell_range) *
1678 			 2) << 2);
1679 		WREG32_SOC15(
1680 			GC, GET_INST(GC, xcc_id),
1681 			regCP_MEC_DOORBELL_RANGE_UPPER,
1682 			((adev->doorbell_index.userqueue_end +
1683 			  xcc_id * adev->doorbell_index.xcc_doorbell_range) *
1684 			 2) << 2);
1685 	}
1686 
1687 	WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_HQD_PQ_DOORBELL_CONTROL,
1688 	       mqd->cp_hqd_pq_doorbell_control);
1689 
1690 	/* reset read and write pointers, similar to CP_RB0_WPTR/_RPTR */
1691 	WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_HQD_PQ_WPTR_LO,
1692 	       mqd->cp_hqd_pq_wptr_lo);
1693 	WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_HQD_PQ_WPTR_HI,
1694 	       mqd->cp_hqd_pq_wptr_hi);
1695 
1696 	/* set the vmid for the queue */
1697 	WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_HQD_VMID, mqd->cp_hqd_vmid);
1698 
1699 	WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_HQD_PERSISTENT_STATE,
1700 	       mqd->cp_hqd_persistent_state);
1701 
1702 	/* activate the queue */
1703 	WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_HQD_ACTIVE,
1704 	       mqd->cp_hqd_active);
1705 
1706 	if (ring->use_doorbell)
1707 		WREG32_FIELD15_PREREG(GC, GET_INST(GC, xcc_id), CP_PQ_STATUS, DOORBELL_ENABLE, 1);
1708 
1709 	return 0;
1710 }
1711 
gfx_v9_4_3_xcc_q_fini_register(struct amdgpu_ring * ring,int xcc_id)1712 static int gfx_v9_4_3_xcc_q_fini_register(struct amdgpu_ring *ring,
1713 					    int xcc_id)
1714 {
1715 	struct amdgpu_device *adev = ring->adev;
1716 	int j;
1717 
1718 	/* disable the queue if it's active */
1719 	if (RREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_HQD_ACTIVE) & 1) {
1720 
1721 		WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_HQD_DEQUEUE_REQUEST, 1);
1722 
1723 		for (j = 0; j < adev->usec_timeout; j++) {
1724 			if (!(RREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_HQD_ACTIVE) & 1))
1725 				break;
1726 			udelay(1);
1727 		}
1728 
1729 		if (j == AMDGPU_MAX_USEC_TIMEOUT) {
1730 			DRM_DEBUG("%s dequeue request failed.\n", ring->name);
1731 
1732 			/* Manual disable if dequeue request times out */
1733 			WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_HQD_ACTIVE, 0);
1734 		}
1735 
1736 		WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_HQD_DEQUEUE_REQUEST,
1737 		      0);
1738 	}
1739 
1740 	WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_HQD_IQ_TIMER, 0);
1741 	WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_HQD_IB_CONTROL, 0);
1742 	WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_HQD_PERSISTENT_STATE, CP_HQD_PERSISTENT_STATE_DEFAULT);
1743 	WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_HQD_PQ_DOORBELL_CONTROL, 0x40000000);
1744 	WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_HQD_PQ_DOORBELL_CONTROL, 0);
1745 	WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_HQD_PQ_RPTR, 0);
1746 	WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_HQD_PQ_WPTR_HI, 0);
1747 	WREG32_SOC15_RLC(GC, GET_INST(GC, xcc_id), regCP_HQD_PQ_WPTR_LO, 0);
1748 
1749 	return 0;
1750 }
1751 
gfx_v9_4_3_xcc_kiq_init_queue(struct amdgpu_ring * ring,int xcc_id)1752 static int gfx_v9_4_3_xcc_kiq_init_queue(struct amdgpu_ring *ring, int xcc_id)
1753 {
1754 	struct amdgpu_device *adev = ring->adev;
1755 	struct v9_mqd *mqd = ring->mqd_ptr;
1756 	struct v9_mqd *tmp_mqd;
1757 
1758 	gfx_v9_4_3_xcc_kiq_setting(ring, xcc_id);
1759 
1760 	/* GPU could be in bad state during probe, driver trigger the reset
1761 	 * after load the SMU, in this case , the mqd is not be initialized.
1762 	 * driver need to re-init the mqd.
1763 	 * check mqd->cp_hqd_pq_control since this value should not be 0
1764 	 */
1765 	tmp_mqd = (struct v9_mqd *)adev->gfx.kiq[xcc_id].mqd_backup;
1766 	if (amdgpu_in_reset(adev) && tmp_mqd->cp_hqd_pq_control) {
1767 		/* for GPU_RESET case , reset MQD to a clean status */
1768 		if (adev->gfx.kiq[xcc_id].mqd_backup)
1769 			memcpy(mqd, adev->gfx.kiq[xcc_id].mqd_backup, sizeof(struct v9_mqd_allocation));
1770 
1771 		/* reset ring buffer */
1772 		ring->wptr = 0;
1773 		amdgpu_ring_clear_ring(ring);
1774 		mutex_lock(&adev->srbm_mutex);
1775 		soc15_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0, GET_INST(GC, xcc_id));
1776 		gfx_v9_4_3_xcc_kiq_init_register(ring, xcc_id);
1777 		soc15_grbm_select(adev, 0, 0, 0, 0, GET_INST(GC, xcc_id));
1778 		mutex_unlock(&adev->srbm_mutex);
1779 	} else {
1780 		memset((void *)mqd, 0, sizeof(struct v9_mqd_allocation));
1781 		((struct v9_mqd_allocation *)mqd)->dynamic_cu_mask = 0xFFFFFFFF;
1782 		((struct v9_mqd_allocation *)mqd)->dynamic_rb_mask = 0xFFFFFFFF;
1783 		mutex_lock(&adev->srbm_mutex);
1784 		if (amdgpu_sriov_vf(adev) && adev->in_suspend)
1785 			amdgpu_ring_clear_ring(ring);
1786 		soc15_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0, GET_INST(GC, xcc_id));
1787 		gfx_v9_4_3_xcc_mqd_init(ring, xcc_id);
1788 		gfx_v9_4_3_xcc_kiq_init_register(ring, xcc_id);
1789 		soc15_grbm_select(adev, 0, 0, 0, 0, GET_INST(GC, xcc_id));
1790 		mutex_unlock(&adev->srbm_mutex);
1791 
1792 		if (adev->gfx.kiq[xcc_id].mqd_backup)
1793 			memcpy(adev->gfx.kiq[xcc_id].mqd_backup, mqd, sizeof(struct v9_mqd_allocation));
1794 	}
1795 
1796 	return 0;
1797 }
1798 
gfx_v9_4_3_xcc_kcq_init_queue(struct amdgpu_ring * ring,int xcc_id)1799 static int gfx_v9_4_3_xcc_kcq_init_queue(struct amdgpu_ring *ring, int xcc_id)
1800 {
1801 	struct amdgpu_device *adev = ring->adev;
1802 	struct v9_mqd *mqd = ring->mqd_ptr;
1803 	int mqd_idx = ring - &adev->gfx.compute_ring[0];
1804 	struct v9_mqd *tmp_mqd;
1805 
1806 	/* Same as above kiq init, driver need to re-init the mqd if mqd->cp_hqd_pq_control
1807 	 * is not be initialized before
1808 	 */
1809 	tmp_mqd = (struct v9_mqd *)adev->gfx.mec.mqd_backup[mqd_idx];
1810 
1811 	if (!tmp_mqd->cp_hqd_pq_control ||
1812 	    (!amdgpu_in_reset(adev) && !adev->in_suspend)) {
1813 		memset((void *)mqd, 0, sizeof(struct v9_mqd_allocation));
1814 		((struct v9_mqd_allocation *)mqd)->dynamic_cu_mask = 0xFFFFFFFF;
1815 		((struct v9_mqd_allocation *)mqd)->dynamic_rb_mask = 0xFFFFFFFF;
1816 		mutex_lock(&adev->srbm_mutex);
1817 		soc15_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0, GET_INST(GC, xcc_id));
1818 		gfx_v9_4_3_xcc_mqd_init(ring, xcc_id);
1819 		soc15_grbm_select(adev, 0, 0, 0, 0, GET_INST(GC, xcc_id));
1820 		mutex_unlock(&adev->srbm_mutex);
1821 
1822 		if (adev->gfx.mec.mqd_backup[mqd_idx])
1823 			memcpy(adev->gfx.mec.mqd_backup[mqd_idx], mqd, sizeof(struct v9_mqd_allocation));
1824 	} else {
1825 		/* restore MQD to a clean status */
1826 		if (adev->gfx.mec.mqd_backup[mqd_idx])
1827 			memcpy(mqd, adev->gfx.mec.mqd_backup[mqd_idx], sizeof(struct v9_mqd_allocation));
1828 		/* reset ring buffer */
1829 		ring->wptr = 0;
1830 		atomic64_set((atomic64_t *)&adev->wb.wb[ring->wptr_offs], 0);
1831 		amdgpu_ring_clear_ring(ring);
1832 	}
1833 
1834 	return 0;
1835 }
1836 
gfx_v9_4_3_xcc_kcq_fini_register(struct amdgpu_device * adev,int xcc_id)1837 static int gfx_v9_4_3_xcc_kcq_fini_register(struct amdgpu_device *adev, int xcc_id)
1838 {
1839 	struct amdgpu_ring *ring;
1840 	int j;
1841 
1842 	for (j = 0; j < adev->gfx.num_compute_rings; j++) {
1843 		ring = &adev->gfx.compute_ring[j +  xcc_id * adev->gfx.num_compute_rings];
1844 		if (!amdgpu_in_reset(adev) && !adev->in_suspend) {
1845 			mutex_lock(&adev->srbm_mutex);
1846 			soc15_grbm_select(adev, ring->me,
1847 					ring->pipe,
1848 					ring->queue, 0, GET_INST(GC, xcc_id));
1849 			gfx_v9_4_3_xcc_q_fini_register(ring, xcc_id);
1850 			soc15_grbm_select(adev, 0, 0, 0, 0, GET_INST(GC, xcc_id));
1851 			mutex_unlock(&adev->srbm_mutex);
1852 		}
1853 	}
1854 
1855 	return 0;
1856 }
1857 
gfx_v9_4_3_xcc_kiq_resume(struct amdgpu_device * adev,int xcc_id)1858 static int gfx_v9_4_3_xcc_kiq_resume(struct amdgpu_device *adev, int xcc_id)
1859 {
1860 	struct amdgpu_ring *ring;
1861 	int r;
1862 
1863 	ring = &adev->gfx.kiq[xcc_id].ring;
1864 
1865 	r = amdgpu_bo_reserve(ring->mqd_obj, false);
1866 	if (unlikely(r != 0))
1867 		return r;
1868 
1869 	r = amdgpu_bo_kmap(ring->mqd_obj, (void **)&ring->mqd_ptr);
1870 	if (unlikely(r != 0)) {
1871 		amdgpu_bo_unreserve(ring->mqd_obj);
1872 		return r;
1873 	}
1874 
1875 	gfx_v9_4_3_xcc_kiq_init_queue(ring, xcc_id);
1876 	amdgpu_bo_kunmap(ring->mqd_obj);
1877 	ring->mqd_ptr = NULL;
1878 	amdgpu_bo_unreserve(ring->mqd_obj);
1879 	return 0;
1880 }
1881 
gfx_v9_4_3_xcc_kcq_resume(struct amdgpu_device * adev,int xcc_id)1882 static int gfx_v9_4_3_xcc_kcq_resume(struct amdgpu_device *adev, int xcc_id)
1883 {
1884 	struct amdgpu_ring *ring = NULL;
1885 	int r = 0, i;
1886 
1887 	gfx_v9_4_3_xcc_cp_compute_enable(adev, true, xcc_id);
1888 
1889 	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
1890 		ring = &adev->gfx.compute_ring[i + xcc_id * adev->gfx.num_compute_rings];
1891 
1892 		r = amdgpu_bo_reserve(ring->mqd_obj, false);
1893 		if (unlikely(r != 0))
1894 			goto done;
1895 		r = amdgpu_bo_kmap(ring->mqd_obj, (void **)&ring->mqd_ptr);
1896 		if (!r) {
1897 			r = gfx_v9_4_3_xcc_kcq_init_queue(ring, xcc_id);
1898 			amdgpu_bo_kunmap(ring->mqd_obj);
1899 			ring->mqd_ptr = NULL;
1900 		}
1901 		amdgpu_bo_unreserve(ring->mqd_obj);
1902 		if (r)
1903 			goto done;
1904 	}
1905 
1906 	r = amdgpu_gfx_enable_kcq(adev, xcc_id);
1907 done:
1908 	return r;
1909 }
1910 
gfx_v9_4_3_xcc_cp_resume(struct amdgpu_device * adev,int xcc_id)1911 static int gfx_v9_4_3_xcc_cp_resume(struct amdgpu_device *adev, int xcc_id)
1912 {
1913 	struct amdgpu_ring *ring;
1914 	int r, j;
1915 
1916 	gfx_v9_4_3_xcc_enable_gui_idle_interrupt(adev, false, xcc_id);
1917 
1918 	if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP) {
1919 		gfx_v9_4_3_xcc_disable_gpa_mode(adev, xcc_id);
1920 
1921 		r = gfx_v9_4_3_xcc_cp_compute_load_microcode(adev, xcc_id);
1922 		if (r)
1923 			return r;
1924 	}
1925 
1926 	r = gfx_v9_4_3_xcc_kiq_resume(adev, xcc_id);
1927 	if (r)
1928 		return r;
1929 
1930 	r = gfx_v9_4_3_xcc_kcq_resume(adev, xcc_id);
1931 	if (r)
1932 		return r;
1933 
1934 	for (j = 0; j < adev->gfx.num_compute_rings; j++) {
1935 		ring = &adev->gfx.compute_ring
1936 				[j + xcc_id * adev->gfx.num_compute_rings];
1937 		r = amdgpu_ring_test_helper(ring);
1938 		if (r)
1939 			return r;
1940 	}
1941 
1942 	gfx_v9_4_3_xcc_enable_gui_idle_interrupt(adev, true, xcc_id);
1943 
1944 	return 0;
1945 }
1946 
gfx_v9_4_3_cp_resume(struct amdgpu_device * adev)1947 static int gfx_v9_4_3_cp_resume(struct amdgpu_device *adev)
1948 {
1949 	int r = 0, i, num_xcc;
1950 
1951 	if (amdgpu_xcp_query_partition_mode(adev->xcp_mgr,
1952 					    AMDGPU_XCP_FL_NONE) ==
1953 	    AMDGPU_UNKNOWN_COMPUTE_PARTITION_MODE)
1954 		r = amdgpu_xcp_switch_partition_mode(adev->xcp_mgr,
1955 						     amdgpu_user_partt_mode);
1956 
1957 	if (r)
1958 		return r;
1959 
1960 	num_xcc = NUM_XCC(adev->gfx.xcc_mask);
1961 	for (i = 0; i < num_xcc; i++) {
1962 		r = gfx_v9_4_3_xcc_cp_resume(adev, i);
1963 		if (r)
1964 			return r;
1965 	}
1966 
1967 	return 0;
1968 }
1969 
gfx_v9_4_3_xcc_cp_enable(struct amdgpu_device * adev,bool enable,int xcc_id)1970 static void gfx_v9_4_3_xcc_cp_enable(struct amdgpu_device *adev, bool enable,
1971 				     int xcc_id)
1972 {
1973 	gfx_v9_4_3_xcc_cp_compute_enable(adev, enable, xcc_id);
1974 }
1975 
gfx_v9_4_3_xcc_fini(struct amdgpu_device * adev,int xcc_id)1976 static void gfx_v9_4_3_xcc_fini(struct amdgpu_device *adev, int xcc_id)
1977 {
1978 	if (amdgpu_gfx_disable_kcq(adev, xcc_id))
1979 		DRM_ERROR("XCD %d KCQ disable failed\n", xcc_id);
1980 
1981 	if (amdgpu_sriov_vf(adev)) {
1982 		/* must disable polling for SRIOV when hw finished, otherwise
1983 		 * CPC engine may still keep fetching WB address which is already
1984 		 * invalid after sw finished and trigger DMAR reading error in
1985 		 * hypervisor side.
1986 		 */
1987 		WREG32_FIELD15_PREREG(GC, GET_INST(GC, xcc_id), CP_PQ_WPTR_POLL_CNTL, EN, 0);
1988 		return;
1989 	}
1990 
1991 	/* Use deinitialize sequence from CAIL when unbinding device
1992 	 * from driver, otherwise KIQ is hanging when binding back
1993 	 */
1994 	if (!amdgpu_in_reset(adev) && !adev->in_suspend) {
1995 		mutex_lock(&adev->srbm_mutex);
1996 		soc15_grbm_select(adev, adev->gfx.kiq[xcc_id].ring.me,
1997 				  adev->gfx.kiq[xcc_id].ring.pipe,
1998 				  adev->gfx.kiq[xcc_id].ring.queue, 0,
1999 				  GET_INST(GC, xcc_id));
2000 		gfx_v9_4_3_xcc_q_fini_register(&adev->gfx.kiq[xcc_id].ring,
2001 						 xcc_id);
2002 		soc15_grbm_select(adev, 0, 0, 0, 0, GET_INST(GC, xcc_id));
2003 		mutex_unlock(&adev->srbm_mutex);
2004 	}
2005 
2006 	gfx_v9_4_3_xcc_kcq_fini_register(adev, xcc_id);
2007 	gfx_v9_4_3_xcc_cp_enable(adev, false, xcc_id);
2008 }
2009 
gfx_v9_4_3_hw_init(void * handle)2010 static int gfx_v9_4_3_hw_init(void *handle)
2011 {
2012 	int r;
2013 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2014 
2015 	if (!amdgpu_sriov_vf(adev))
2016 		gfx_v9_4_3_init_golden_registers(adev);
2017 
2018 	gfx_v9_4_3_constants_init(adev);
2019 
2020 	r = adev->gfx.rlc.funcs->resume(adev);
2021 	if (r)
2022 		return r;
2023 
2024 	r = gfx_v9_4_3_cp_resume(adev);
2025 	if (r)
2026 		return r;
2027 
2028 	return r;
2029 }
2030 
gfx_v9_4_3_hw_fini(void * handle)2031 static int gfx_v9_4_3_hw_fini(void *handle)
2032 {
2033 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2034 	int i, num_xcc;
2035 
2036 	amdgpu_irq_put(adev, &adev->gfx.priv_reg_irq, 0);
2037 	amdgpu_irq_put(adev, &adev->gfx.priv_inst_irq, 0);
2038 
2039 	num_xcc = NUM_XCC(adev->gfx.xcc_mask);
2040 	for (i = 0; i < num_xcc; i++) {
2041 		gfx_v9_4_3_xcc_fini(adev, i);
2042 	}
2043 
2044 	return 0;
2045 }
2046 
gfx_v9_4_3_suspend(void * handle)2047 static int gfx_v9_4_3_suspend(void *handle)
2048 {
2049 	return gfx_v9_4_3_hw_fini(handle);
2050 }
2051 
gfx_v9_4_3_resume(void * handle)2052 static int gfx_v9_4_3_resume(void *handle)
2053 {
2054 	return gfx_v9_4_3_hw_init(handle);
2055 }
2056 
gfx_v9_4_3_is_idle(void * handle)2057 static bool gfx_v9_4_3_is_idle(void *handle)
2058 {
2059 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2060 	int i, num_xcc;
2061 
2062 	num_xcc = NUM_XCC(adev->gfx.xcc_mask);
2063 	for (i = 0; i < num_xcc; i++) {
2064 		if (REG_GET_FIELD(RREG32_SOC15(GC, GET_INST(GC, i), regGRBM_STATUS),
2065 					GRBM_STATUS, GUI_ACTIVE))
2066 			return false;
2067 	}
2068 	return true;
2069 }
2070 
gfx_v9_4_3_wait_for_idle(void * handle)2071 static int gfx_v9_4_3_wait_for_idle(void *handle)
2072 {
2073 	unsigned i;
2074 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2075 
2076 	for (i = 0; i < adev->usec_timeout; i++) {
2077 		if (gfx_v9_4_3_is_idle(handle))
2078 			return 0;
2079 		udelay(1);
2080 	}
2081 	return -ETIMEDOUT;
2082 }
2083 
gfx_v9_4_3_soft_reset(void * handle)2084 static int gfx_v9_4_3_soft_reset(void *handle)
2085 {
2086 	u32 grbm_soft_reset = 0;
2087 	u32 tmp;
2088 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2089 
2090 	/* GRBM_STATUS */
2091 	tmp = RREG32_SOC15(GC, GET_INST(GC, 0), regGRBM_STATUS);
2092 	if (tmp & (GRBM_STATUS__PA_BUSY_MASK | GRBM_STATUS__SC_BUSY_MASK |
2093 		   GRBM_STATUS__BCI_BUSY_MASK | GRBM_STATUS__SX_BUSY_MASK |
2094 		   GRBM_STATUS__TA_BUSY_MASK | GRBM_STATUS__VGT_BUSY_MASK |
2095 		   GRBM_STATUS__DB_BUSY_MASK | GRBM_STATUS__CB_BUSY_MASK |
2096 		   GRBM_STATUS__GDS_BUSY_MASK | GRBM_STATUS__SPI_BUSY_MASK |
2097 		   GRBM_STATUS__IA_BUSY_MASK | GRBM_STATUS__IA_BUSY_NO_DMA_MASK)) {
2098 		grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset,
2099 						GRBM_SOFT_RESET, SOFT_RESET_CP, 1);
2100 		grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset,
2101 						GRBM_SOFT_RESET, SOFT_RESET_GFX, 1);
2102 	}
2103 
2104 	if (tmp & (GRBM_STATUS__CP_BUSY_MASK | GRBM_STATUS__CP_COHERENCY_BUSY_MASK)) {
2105 		grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset,
2106 						GRBM_SOFT_RESET, SOFT_RESET_CP, 1);
2107 	}
2108 
2109 	/* GRBM_STATUS2 */
2110 	tmp = RREG32_SOC15(GC, GET_INST(GC, 0), regGRBM_STATUS2);
2111 	if (REG_GET_FIELD(tmp, GRBM_STATUS2, RLC_BUSY))
2112 		grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset,
2113 						GRBM_SOFT_RESET, SOFT_RESET_RLC, 1);
2114 
2115 
2116 	if (grbm_soft_reset) {
2117 		/* stop the rlc */
2118 		adev->gfx.rlc.funcs->stop(adev);
2119 
2120 		/* Disable MEC parsing/prefetching */
2121 		gfx_v9_4_3_xcc_cp_compute_enable(adev, false, 0);
2122 
2123 		if (grbm_soft_reset) {
2124 			tmp = RREG32_SOC15(GC, GET_INST(GC, 0), regGRBM_SOFT_RESET);
2125 			tmp |= grbm_soft_reset;
2126 			dev_info(adev->dev, "GRBM_SOFT_RESET=0x%08X\n", tmp);
2127 			WREG32_SOC15(GC, GET_INST(GC, 0), regGRBM_SOFT_RESET, tmp);
2128 			tmp = RREG32_SOC15(GC, GET_INST(GC, 0), regGRBM_SOFT_RESET);
2129 
2130 			udelay(50);
2131 
2132 			tmp &= ~grbm_soft_reset;
2133 			WREG32_SOC15(GC, GET_INST(GC, 0), regGRBM_SOFT_RESET, tmp);
2134 			tmp = RREG32_SOC15(GC, GET_INST(GC, 0), regGRBM_SOFT_RESET);
2135 		}
2136 
2137 		/* Wait a little for things to settle down */
2138 		udelay(50);
2139 	}
2140 	return 0;
2141 }
2142 
gfx_v9_4_3_ring_emit_gds_switch(struct amdgpu_ring * ring,uint32_t vmid,uint32_t gds_base,uint32_t gds_size,uint32_t gws_base,uint32_t gws_size,uint32_t oa_base,uint32_t oa_size)2143 static void gfx_v9_4_3_ring_emit_gds_switch(struct amdgpu_ring *ring,
2144 					  uint32_t vmid,
2145 					  uint32_t gds_base, uint32_t gds_size,
2146 					  uint32_t gws_base, uint32_t gws_size,
2147 					  uint32_t oa_base, uint32_t oa_size)
2148 {
2149 	struct amdgpu_device *adev = ring->adev;
2150 
2151 	/* GDS Base */
2152 	gfx_v9_4_3_write_data_to_reg(ring, 0, false,
2153 				   SOC15_REG_OFFSET(GC, GET_INST(GC, 0), regGDS_VMID0_BASE) + 2 * vmid,
2154 				   gds_base);
2155 
2156 	/* GDS Size */
2157 	gfx_v9_4_3_write_data_to_reg(ring, 0, false,
2158 				   SOC15_REG_OFFSET(GC, GET_INST(GC, 0), regGDS_VMID0_SIZE) + 2 * vmid,
2159 				   gds_size);
2160 
2161 	/* GWS */
2162 	gfx_v9_4_3_write_data_to_reg(ring, 0, false,
2163 				   SOC15_REG_OFFSET(GC, GET_INST(GC, 0), regGDS_GWS_VMID0) + vmid,
2164 				   gws_size << GDS_GWS_VMID0__SIZE__SHIFT | gws_base);
2165 
2166 	/* OA */
2167 	gfx_v9_4_3_write_data_to_reg(ring, 0, false,
2168 				   SOC15_REG_OFFSET(GC, GET_INST(GC, 0), regGDS_OA_VMID0) + vmid,
2169 				   (1 << (oa_size + oa_base)) - (1 << oa_base));
2170 }
2171 
gfx_v9_4_3_early_init(void * handle)2172 static int gfx_v9_4_3_early_init(void *handle)
2173 {
2174 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2175 
2176 	adev->gfx.num_compute_rings = min(amdgpu_gfx_get_num_kcq(adev),
2177 					  AMDGPU_MAX_COMPUTE_RINGS);
2178 	gfx_v9_4_3_set_kiq_pm4_funcs(adev);
2179 	gfx_v9_4_3_set_ring_funcs(adev);
2180 	gfx_v9_4_3_set_irq_funcs(adev);
2181 	gfx_v9_4_3_set_gds_init(adev);
2182 	gfx_v9_4_3_set_rlc_funcs(adev);
2183 
2184 	/* init rlcg reg access ctrl */
2185 	gfx_v9_4_3_init_rlcg_reg_access_ctrl(adev);
2186 
2187 	return gfx_v9_4_3_init_microcode(adev);
2188 }
2189 
gfx_v9_4_3_late_init(void * handle)2190 static int gfx_v9_4_3_late_init(void *handle)
2191 {
2192 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2193 	int r;
2194 
2195 	r = amdgpu_irq_get(adev, &adev->gfx.priv_reg_irq, 0);
2196 	if (r)
2197 		return r;
2198 
2199 	r = amdgpu_irq_get(adev, &adev->gfx.priv_inst_irq, 0);
2200 	if (r)
2201 		return r;
2202 
2203 	if (adev->gfx.ras &&
2204 	    adev->gfx.ras->enable_watchdog_timer)
2205 		adev->gfx.ras->enable_watchdog_timer(adev);
2206 
2207 	return 0;
2208 }
2209 
gfx_v9_4_3_xcc_update_sram_fgcg(struct amdgpu_device * adev,bool enable,int xcc_id)2210 static void gfx_v9_4_3_xcc_update_sram_fgcg(struct amdgpu_device *adev,
2211 					    bool enable, int xcc_id)
2212 {
2213 	uint32_t def, data;
2214 
2215 	if (!(adev->cg_flags & AMD_CG_SUPPORT_GFX_FGCG))
2216 		return;
2217 
2218 	def = data = RREG32_SOC15(GC, GET_INST(GC, xcc_id),
2219 				  regRLC_CGTT_MGCG_OVERRIDE);
2220 
2221 	if (enable)
2222 		data &= ~RLC_CGTT_MGCG_OVERRIDE__GFXIP_FGCG_OVERRIDE_MASK;
2223 	else
2224 		data |= RLC_CGTT_MGCG_OVERRIDE__GFXIP_FGCG_OVERRIDE_MASK;
2225 
2226 	if (def != data)
2227 		WREG32_SOC15(GC, GET_INST(GC, xcc_id),
2228 			     regRLC_CGTT_MGCG_OVERRIDE, data);
2229 
2230 }
2231 
gfx_v9_4_3_xcc_update_repeater_fgcg(struct amdgpu_device * adev,bool enable,int xcc_id)2232 static void gfx_v9_4_3_xcc_update_repeater_fgcg(struct amdgpu_device *adev,
2233 						bool enable, int xcc_id)
2234 {
2235 	uint32_t def, data;
2236 
2237 	if (!(adev->cg_flags & AMD_CG_SUPPORT_REPEATER_FGCG))
2238 		return;
2239 
2240 	def = data = RREG32_SOC15(GC, GET_INST(GC, xcc_id),
2241 				  regRLC_CGTT_MGCG_OVERRIDE);
2242 
2243 	if (enable)
2244 		data &= ~RLC_CGTT_MGCG_OVERRIDE__GFXIP_REP_FGCG_OVERRIDE_MASK;
2245 	else
2246 		data |= RLC_CGTT_MGCG_OVERRIDE__GFXIP_REP_FGCG_OVERRIDE_MASK;
2247 
2248 	if (def != data)
2249 		WREG32_SOC15(GC, GET_INST(GC, xcc_id),
2250 			     regRLC_CGTT_MGCG_OVERRIDE, data);
2251 }
2252 
2253 static void
gfx_v9_4_3_xcc_update_medium_grain_clock_gating(struct amdgpu_device * adev,bool enable,int xcc_id)2254 gfx_v9_4_3_xcc_update_medium_grain_clock_gating(struct amdgpu_device *adev,
2255 						bool enable, int xcc_id)
2256 {
2257 	uint32_t data, def;
2258 
2259 	/* It is disabled by HW by default */
2260 	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGCG)) {
2261 		/* 1 - RLC_CGTT_MGCG_OVERRIDE */
2262 		def = data = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regRLC_CGTT_MGCG_OVERRIDE);
2263 
2264 		data &= ~(RLC_CGTT_MGCG_OVERRIDE__GRBM_CGTT_SCLK_OVERRIDE_MASK |
2265 			  RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGCG_OVERRIDE_MASK |
2266 			  RLC_CGTT_MGCG_OVERRIDE__RLC_CGTT_SCLK_OVERRIDE_MASK |
2267 			  RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGLS_OVERRIDE_MASK);
2268 
2269 		if (def != data)
2270 			WREG32_SOC15(GC, GET_INST(GC, xcc_id), regRLC_CGTT_MGCG_OVERRIDE, data);
2271 
2272 		/* MGLS is a global flag to control all MGLS in GFX */
2273 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS) {
2274 			/* 2 - RLC memory Light sleep */
2275 			if (adev->cg_flags & AMD_CG_SUPPORT_GFX_RLC_LS) {
2276 				def = data = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regRLC_MEM_SLP_CNTL);
2277 				data |= RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK;
2278 				if (def != data)
2279 					WREG32_SOC15(GC, GET_INST(GC, xcc_id), regRLC_MEM_SLP_CNTL, data);
2280 			}
2281 			/* 3 - CP memory Light sleep */
2282 			if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CP_LS) {
2283 				def = data = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_MEM_SLP_CNTL);
2284 				data |= CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK;
2285 				if (def != data)
2286 					WREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_MEM_SLP_CNTL, data);
2287 			}
2288 		}
2289 	} else {
2290 		/* 1 - MGCG_OVERRIDE */
2291 		def = data = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regRLC_CGTT_MGCG_OVERRIDE);
2292 
2293 		data |= (RLC_CGTT_MGCG_OVERRIDE__RLC_CGTT_SCLK_OVERRIDE_MASK |
2294 			 RLC_CGTT_MGCG_OVERRIDE__GRBM_CGTT_SCLK_OVERRIDE_MASK |
2295 			 RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGCG_OVERRIDE_MASK |
2296 			 RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGLS_OVERRIDE_MASK);
2297 
2298 		if (def != data)
2299 			WREG32_SOC15(GC, GET_INST(GC, xcc_id), regRLC_CGTT_MGCG_OVERRIDE, data);
2300 
2301 		/* 2 - disable MGLS in RLC */
2302 		data = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regRLC_MEM_SLP_CNTL);
2303 		if (data & RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK) {
2304 			data &= ~RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK;
2305 			WREG32_SOC15(GC, GET_INST(GC, xcc_id), regRLC_MEM_SLP_CNTL, data);
2306 		}
2307 
2308 		/* 3 - disable MGLS in CP */
2309 		data = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_MEM_SLP_CNTL);
2310 		if (data & CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK) {
2311 			data &= ~CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK;
2312 			WREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_MEM_SLP_CNTL, data);
2313 		}
2314 	}
2315 
2316 }
2317 
2318 static void
gfx_v9_4_3_xcc_update_coarse_grain_clock_gating(struct amdgpu_device * adev,bool enable,int xcc_id)2319 gfx_v9_4_3_xcc_update_coarse_grain_clock_gating(struct amdgpu_device *adev,
2320 						bool enable, int xcc_id)
2321 {
2322 	uint32_t def, data;
2323 
2324 	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGCG)) {
2325 
2326 		def = data = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regRLC_CGTT_MGCG_OVERRIDE);
2327 		/* unset CGCG override */
2328 		data &= ~RLC_CGTT_MGCG_OVERRIDE__GFXIP_CGCG_OVERRIDE_MASK;
2329 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGLS)
2330 			data &= ~RLC_CGTT_MGCG_OVERRIDE__GFXIP_CGLS_OVERRIDE_MASK;
2331 		else
2332 			data |= RLC_CGTT_MGCG_OVERRIDE__GFXIP_CGLS_OVERRIDE_MASK;
2333 		/* update CGCG and CGLS override bits */
2334 		if (def != data)
2335 			WREG32_SOC15(GC, GET_INST(GC, xcc_id), regRLC_CGTT_MGCG_OVERRIDE, data);
2336 
2337 		/* enable cgcg FSM(0x0000363F) */
2338 		def = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regRLC_CGCG_CGLS_CTRL);
2339 
2340 		data = (0x36
2341 			<< RLC_CGCG_CGLS_CTRL__CGCG_GFX_IDLE_THRESHOLD__SHIFT) |
2342 		       RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK;
2343 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGLS)
2344 			data |= (0x000F << RLC_CGCG_CGLS_CTRL__CGLS_REP_COMPANSAT_DELAY__SHIFT) |
2345 				RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK;
2346 		if (def != data)
2347 			WREG32_SOC15(GC, GET_INST(GC, xcc_id), regRLC_CGCG_CGLS_CTRL, data);
2348 
2349 		/* set IDLE_POLL_COUNT(0x00900100) */
2350 		def = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_RB_WPTR_POLL_CNTL);
2351 		data = (0x0100 << CP_RB_WPTR_POLL_CNTL__POLL_FREQUENCY__SHIFT) |
2352 			(0x0090 << CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT__SHIFT);
2353 		if (def != data)
2354 			WREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_RB_WPTR_POLL_CNTL, data);
2355 	} else {
2356 		def = data = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regRLC_CGCG_CGLS_CTRL);
2357 		/* reset CGCG/CGLS bits */
2358 		data &= ~(RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK | RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK);
2359 		/* disable cgcg and cgls in FSM */
2360 		if (def != data)
2361 			WREG32_SOC15(GC, GET_INST(GC, xcc_id), regRLC_CGCG_CGLS_CTRL, data);
2362 	}
2363 
2364 }
2365 
gfx_v9_4_3_xcc_update_gfx_clock_gating(struct amdgpu_device * adev,bool enable,int xcc_id)2366 static int gfx_v9_4_3_xcc_update_gfx_clock_gating(struct amdgpu_device *adev,
2367 						  bool enable, int xcc_id)
2368 {
2369 	amdgpu_gfx_rlc_enter_safe_mode(adev, xcc_id);
2370 
2371 	if (enable) {
2372 		/* FGCG */
2373 		gfx_v9_4_3_xcc_update_sram_fgcg(adev, enable, xcc_id);
2374 		gfx_v9_4_3_xcc_update_repeater_fgcg(adev, enable, xcc_id);
2375 
2376 		/* CGCG/CGLS should be enabled after MGCG/MGLS
2377 		 * ===  MGCG + MGLS ===
2378 		 */
2379 		gfx_v9_4_3_xcc_update_medium_grain_clock_gating(adev, enable,
2380 								xcc_id);
2381 		/* ===  CGCG + CGLS === */
2382 		gfx_v9_4_3_xcc_update_coarse_grain_clock_gating(adev, enable,
2383 								xcc_id);
2384 	} else {
2385 		/* CGCG/CGLS should be disabled before MGCG/MGLS
2386 		 * ===  CGCG + CGLS ===
2387 		 */
2388 		gfx_v9_4_3_xcc_update_coarse_grain_clock_gating(adev, enable,
2389 								xcc_id);
2390 		/* ===  MGCG + MGLS === */
2391 		gfx_v9_4_3_xcc_update_medium_grain_clock_gating(adev, enable,
2392 								xcc_id);
2393 
2394 		/* FGCG */
2395 		gfx_v9_4_3_xcc_update_sram_fgcg(adev, enable, xcc_id);
2396 		gfx_v9_4_3_xcc_update_repeater_fgcg(adev, enable, xcc_id);
2397 	}
2398 
2399 	amdgpu_gfx_rlc_exit_safe_mode(adev, xcc_id);
2400 
2401 	return 0;
2402 }
2403 
2404 static const struct amdgpu_rlc_funcs gfx_v9_4_3_rlc_funcs = {
2405 	.is_rlc_enabled = gfx_v9_4_3_is_rlc_enabled,
2406 	.set_safe_mode = gfx_v9_4_3_xcc_set_safe_mode,
2407 	.unset_safe_mode = gfx_v9_4_3_xcc_unset_safe_mode,
2408 	.init = gfx_v9_4_3_rlc_init,
2409 	.resume = gfx_v9_4_3_rlc_resume,
2410 	.stop = gfx_v9_4_3_rlc_stop,
2411 	.reset = gfx_v9_4_3_rlc_reset,
2412 	.start = gfx_v9_4_3_rlc_start,
2413 	.update_spm_vmid = gfx_v9_4_3_update_spm_vmid,
2414 	.is_rlcg_access_range = gfx_v9_4_3_is_rlcg_access_range,
2415 };
2416 
gfx_v9_4_3_set_powergating_state(void * handle,enum amd_powergating_state state)2417 static int gfx_v9_4_3_set_powergating_state(void *handle,
2418 					  enum amd_powergating_state state)
2419 {
2420 	return 0;
2421 }
2422 
gfx_v9_4_3_set_clockgating_state(void * handle,enum amd_clockgating_state state)2423 static int gfx_v9_4_3_set_clockgating_state(void *handle,
2424 					  enum amd_clockgating_state state)
2425 {
2426 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2427 	int i, num_xcc;
2428 
2429 	if (amdgpu_sriov_vf(adev))
2430 		return 0;
2431 
2432 	num_xcc = NUM_XCC(adev->gfx.xcc_mask);
2433 	switch (adev->ip_versions[GC_HWIP][0]) {
2434 	case IP_VERSION(9, 4, 3):
2435 		for (i = 0; i < num_xcc; i++)
2436 			gfx_v9_4_3_xcc_update_gfx_clock_gating(
2437 				adev, state == AMD_CG_STATE_GATE, i);
2438 		break;
2439 	default:
2440 		break;
2441 	}
2442 	return 0;
2443 }
2444 
gfx_v9_4_3_get_clockgating_state(void * handle,u64 * flags)2445 static void gfx_v9_4_3_get_clockgating_state(void *handle, u64 *flags)
2446 {
2447 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2448 	int data;
2449 
2450 	if (amdgpu_sriov_vf(adev))
2451 		*flags = 0;
2452 
2453 	/* AMD_CG_SUPPORT_GFX_MGCG */
2454 	data = RREG32_KIQ(SOC15_REG_OFFSET(GC, GET_INST(GC, 0), regRLC_CGTT_MGCG_OVERRIDE));
2455 	if (!(data & RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGCG_OVERRIDE_MASK))
2456 		*flags |= AMD_CG_SUPPORT_GFX_MGCG;
2457 
2458 	/* AMD_CG_SUPPORT_GFX_CGCG */
2459 	data = RREG32_KIQ(SOC15_REG_OFFSET(GC, GET_INST(GC, 0), regRLC_CGCG_CGLS_CTRL));
2460 	if (data & RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK)
2461 		*flags |= AMD_CG_SUPPORT_GFX_CGCG;
2462 
2463 	/* AMD_CG_SUPPORT_GFX_CGLS */
2464 	if (data & RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK)
2465 		*flags |= AMD_CG_SUPPORT_GFX_CGLS;
2466 
2467 	/* AMD_CG_SUPPORT_GFX_RLC_LS */
2468 	data = RREG32_KIQ(SOC15_REG_OFFSET(GC, GET_INST(GC, 0), regRLC_MEM_SLP_CNTL));
2469 	if (data & RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK)
2470 		*flags |= AMD_CG_SUPPORT_GFX_RLC_LS | AMD_CG_SUPPORT_GFX_MGLS;
2471 
2472 	/* AMD_CG_SUPPORT_GFX_CP_LS */
2473 	data = RREG32_KIQ(SOC15_REG_OFFSET(GC, GET_INST(GC, 0), regCP_MEM_SLP_CNTL));
2474 	if (data & CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK)
2475 		*flags |= AMD_CG_SUPPORT_GFX_CP_LS | AMD_CG_SUPPORT_GFX_MGLS;
2476 }
2477 
gfx_v9_4_3_ring_emit_hdp_flush(struct amdgpu_ring * ring)2478 static void gfx_v9_4_3_ring_emit_hdp_flush(struct amdgpu_ring *ring)
2479 {
2480 	struct amdgpu_device *adev = ring->adev;
2481 	u32 ref_and_mask, reg_mem_engine;
2482 	const struct nbio_hdp_flush_reg *nbio_hf_reg = adev->nbio.hdp_flush_reg;
2483 
2484 	if (ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE) {
2485 		switch (ring->me) {
2486 		case 1:
2487 			ref_and_mask = nbio_hf_reg->ref_and_mask_cp2 << ring->pipe;
2488 			break;
2489 		case 2:
2490 			ref_and_mask = nbio_hf_reg->ref_and_mask_cp6 << ring->pipe;
2491 			break;
2492 		default:
2493 			return;
2494 		}
2495 		reg_mem_engine = 0;
2496 	} else {
2497 		ref_and_mask = nbio_hf_reg->ref_and_mask_cp0;
2498 		reg_mem_engine = 1; /* pfp */
2499 	}
2500 
2501 	gfx_v9_4_3_wait_reg_mem(ring, reg_mem_engine, 0, 1,
2502 			      adev->nbio.funcs->get_hdp_flush_req_offset(adev),
2503 			      adev->nbio.funcs->get_hdp_flush_done_offset(adev),
2504 			      ref_and_mask, ref_and_mask, 0x20);
2505 }
2506 
gfx_v9_4_3_ring_emit_ib_compute(struct amdgpu_ring * ring,struct amdgpu_job * job,struct amdgpu_ib * ib,uint32_t flags)2507 static void gfx_v9_4_3_ring_emit_ib_compute(struct amdgpu_ring *ring,
2508 					  struct amdgpu_job *job,
2509 					  struct amdgpu_ib *ib,
2510 					  uint32_t flags)
2511 {
2512 	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
2513 	u32 control = INDIRECT_BUFFER_VALID | ib->length_dw | (vmid << 24);
2514 
2515 	/* Currently, there is a high possibility to get wave ID mismatch
2516 	 * between ME and GDS, leading to a hw deadlock, because ME generates
2517 	 * different wave IDs than the GDS expects. This situation happens
2518 	 * randomly when at least 5 compute pipes use GDS ordered append.
2519 	 * The wave IDs generated by ME are also wrong after suspend/resume.
2520 	 * Those are probably bugs somewhere else in the kernel driver.
2521 	 *
2522 	 * Writing GDS_COMPUTE_MAX_WAVE_ID resets wave ID counters in ME and
2523 	 * GDS to 0 for this ring (me/pipe).
2524 	 */
2525 	if (ib->flags & AMDGPU_IB_FLAG_RESET_GDS_MAX_WAVE_ID) {
2526 		amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1));
2527 		amdgpu_ring_write(ring, regGDS_COMPUTE_MAX_WAVE_ID);
2528 		amdgpu_ring_write(ring, ring->adev->gds.gds_compute_max_wave_id);
2529 	}
2530 
2531 	amdgpu_ring_write(ring, PACKET3(PACKET3_INDIRECT_BUFFER, 2));
2532 	BUG_ON(ib->gpu_addr & 0x3); /* Dword align */
2533 	amdgpu_ring_write(ring,
2534 #ifdef __BIG_ENDIAN
2535 				(2 << 0) |
2536 #endif
2537 				lower_32_bits(ib->gpu_addr));
2538 	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr));
2539 	amdgpu_ring_write(ring, control);
2540 }
2541 
gfx_v9_4_3_ring_emit_fence(struct amdgpu_ring * ring,u64 addr,u64 seq,unsigned flags)2542 static void gfx_v9_4_3_ring_emit_fence(struct amdgpu_ring *ring, u64 addr,
2543 				     u64 seq, unsigned flags)
2544 {
2545 	bool write64bit = flags & AMDGPU_FENCE_FLAG_64BIT;
2546 	bool int_sel = flags & AMDGPU_FENCE_FLAG_INT;
2547 	bool writeback = flags & AMDGPU_FENCE_FLAG_TC_WB_ONLY;
2548 
2549 	/* RELEASE_MEM - flush caches, send int */
2550 	amdgpu_ring_write(ring, PACKET3(PACKET3_RELEASE_MEM, 6));
2551 	amdgpu_ring_write(ring, ((writeback ? (EOP_TC_WB_ACTION_EN |
2552 					       EOP_TC_NC_ACTION_EN) :
2553 					      (EOP_TCL1_ACTION_EN |
2554 					       EOP_TC_ACTION_EN |
2555 					       EOP_TC_WB_ACTION_EN |
2556 					       EOP_TC_MD_ACTION_EN)) |
2557 				 EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
2558 				 EVENT_INDEX(5)));
2559 	amdgpu_ring_write(ring, DATA_SEL(write64bit ? 2 : 1) | INT_SEL(int_sel ? 2 : 0));
2560 
2561 	/*
2562 	 * the address should be Qword aligned if 64bit write, Dword
2563 	 * aligned if only send 32bit data low (discard data high)
2564 	 */
2565 	if (write64bit)
2566 		BUG_ON(addr & 0x7);
2567 	else
2568 		BUG_ON(addr & 0x3);
2569 	amdgpu_ring_write(ring, lower_32_bits(addr));
2570 	amdgpu_ring_write(ring, upper_32_bits(addr));
2571 	amdgpu_ring_write(ring, lower_32_bits(seq));
2572 	amdgpu_ring_write(ring, upper_32_bits(seq));
2573 	amdgpu_ring_write(ring, 0);
2574 }
2575 
gfx_v9_4_3_ring_emit_pipeline_sync(struct amdgpu_ring * ring)2576 static void gfx_v9_4_3_ring_emit_pipeline_sync(struct amdgpu_ring *ring)
2577 {
2578 	int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX);
2579 	uint32_t seq = ring->fence_drv.sync_seq;
2580 	uint64_t addr = ring->fence_drv.gpu_addr;
2581 
2582 	gfx_v9_4_3_wait_reg_mem(ring, usepfp, 1, 0,
2583 			      lower_32_bits(addr), upper_32_bits(addr),
2584 			      seq, 0xffffffff, 4);
2585 }
2586 
gfx_v9_4_3_ring_emit_vm_flush(struct amdgpu_ring * ring,unsigned vmid,uint64_t pd_addr)2587 static void gfx_v9_4_3_ring_emit_vm_flush(struct amdgpu_ring *ring,
2588 					unsigned vmid, uint64_t pd_addr)
2589 {
2590 	amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr);
2591 }
2592 
gfx_v9_4_3_ring_get_rptr_compute(struct amdgpu_ring * ring)2593 static u64 gfx_v9_4_3_ring_get_rptr_compute(struct amdgpu_ring *ring)
2594 {
2595 	return ring->adev->wb.wb[ring->rptr_offs]; /* gfx9 hardware is 32bit rptr */
2596 }
2597 
gfx_v9_4_3_ring_get_wptr_compute(struct amdgpu_ring * ring)2598 static u64 gfx_v9_4_3_ring_get_wptr_compute(struct amdgpu_ring *ring)
2599 {
2600 	u64 wptr;
2601 
2602 	/* XXX check if swapping is necessary on BE */
2603 	if (ring->use_doorbell)
2604 		wptr = atomic64_read((atomic64_t *)&ring->adev->wb.wb[ring->wptr_offs]);
2605 	else
2606 		BUG();
2607 	return wptr;
2608 }
2609 
gfx_v9_4_3_ring_set_wptr_compute(struct amdgpu_ring * ring)2610 static void gfx_v9_4_3_ring_set_wptr_compute(struct amdgpu_ring *ring)
2611 {
2612 	struct amdgpu_device *adev = ring->adev;
2613 
2614 	/* XXX check if swapping is necessary on BE */
2615 	if (ring->use_doorbell) {
2616 		atomic64_set((atomic64_t *)&adev->wb.wb[ring->wptr_offs], ring->wptr);
2617 		WDOORBELL64(ring->doorbell_index, ring->wptr);
2618 	} else {
2619 		BUG(); /* only DOORBELL method supported on gfx9 now */
2620 	}
2621 }
2622 
gfx_v9_4_3_ring_emit_fence_kiq(struct amdgpu_ring * ring,u64 addr,u64 seq,unsigned int flags)2623 static void gfx_v9_4_3_ring_emit_fence_kiq(struct amdgpu_ring *ring, u64 addr,
2624 					 u64 seq, unsigned int flags)
2625 {
2626 	struct amdgpu_device *adev = ring->adev;
2627 
2628 	/* we only allocate 32bit for each seq wb address */
2629 	BUG_ON(flags & AMDGPU_FENCE_FLAG_64BIT);
2630 
2631 	/* write fence seq to the "addr" */
2632 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
2633 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
2634 				 WRITE_DATA_DST_SEL(5) | WR_CONFIRM));
2635 	amdgpu_ring_write(ring, lower_32_bits(addr));
2636 	amdgpu_ring_write(ring, upper_32_bits(addr));
2637 	amdgpu_ring_write(ring, lower_32_bits(seq));
2638 
2639 	if (flags & AMDGPU_FENCE_FLAG_INT) {
2640 		/* set register to trigger INT */
2641 		amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
2642 		amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
2643 					 WRITE_DATA_DST_SEL(0) | WR_CONFIRM));
2644 		amdgpu_ring_write(ring, SOC15_REG_OFFSET(GC, GET_INST(GC, 0), regCPC_INT_STATUS));
2645 		amdgpu_ring_write(ring, 0);
2646 		amdgpu_ring_write(ring, 0x20000000); /* src_id is 178 */
2647 	}
2648 }
2649 
gfx_v9_4_3_ring_emit_rreg(struct amdgpu_ring * ring,uint32_t reg,uint32_t reg_val_offs)2650 static void gfx_v9_4_3_ring_emit_rreg(struct amdgpu_ring *ring, uint32_t reg,
2651 				    uint32_t reg_val_offs)
2652 {
2653 	struct amdgpu_device *adev = ring->adev;
2654 
2655 	amdgpu_ring_write(ring, PACKET3(PACKET3_COPY_DATA, 4));
2656 	amdgpu_ring_write(ring, 0 |	/* src: register*/
2657 				(5 << 8) |	/* dst: memory */
2658 				(1 << 20));	/* write confirm */
2659 	amdgpu_ring_write(ring, reg);
2660 	amdgpu_ring_write(ring, 0);
2661 	amdgpu_ring_write(ring, lower_32_bits(adev->wb.gpu_addr +
2662 				reg_val_offs * 4));
2663 	amdgpu_ring_write(ring, upper_32_bits(adev->wb.gpu_addr +
2664 				reg_val_offs * 4));
2665 }
2666 
gfx_v9_4_3_ring_emit_wreg(struct amdgpu_ring * ring,uint32_t reg,uint32_t val)2667 static void gfx_v9_4_3_ring_emit_wreg(struct amdgpu_ring *ring, uint32_t reg,
2668 				    uint32_t val)
2669 {
2670 	uint32_t cmd = 0;
2671 
2672 	switch (ring->funcs->type) {
2673 	case AMDGPU_RING_TYPE_GFX:
2674 		cmd = WRITE_DATA_ENGINE_SEL(1) | WR_CONFIRM;
2675 		break;
2676 	case AMDGPU_RING_TYPE_KIQ:
2677 		cmd = (1 << 16); /* no inc addr */
2678 		break;
2679 	default:
2680 		cmd = WR_CONFIRM;
2681 		break;
2682 	}
2683 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
2684 	amdgpu_ring_write(ring, cmd);
2685 	amdgpu_ring_write(ring, reg);
2686 	amdgpu_ring_write(ring, 0);
2687 	amdgpu_ring_write(ring, val);
2688 }
2689 
gfx_v9_4_3_ring_emit_reg_wait(struct amdgpu_ring * ring,uint32_t reg,uint32_t val,uint32_t mask)2690 static void gfx_v9_4_3_ring_emit_reg_wait(struct amdgpu_ring *ring, uint32_t reg,
2691 					uint32_t val, uint32_t mask)
2692 {
2693 	gfx_v9_4_3_wait_reg_mem(ring, 0, 0, 0, reg, 0, val, mask, 0x20);
2694 }
2695 
gfx_v9_4_3_ring_emit_reg_write_reg_wait(struct amdgpu_ring * ring,uint32_t reg0,uint32_t reg1,uint32_t ref,uint32_t mask)2696 static void gfx_v9_4_3_ring_emit_reg_write_reg_wait(struct amdgpu_ring *ring,
2697 						  uint32_t reg0, uint32_t reg1,
2698 						  uint32_t ref, uint32_t mask)
2699 {
2700 	amdgpu_ring_emit_reg_write_reg_wait_helper(ring, reg0, reg1,
2701 						   ref, mask);
2702 }
2703 
gfx_v9_4_3_xcc_set_compute_eop_interrupt_state(struct amdgpu_device * adev,int me,int pipe,enum amdgpu_interrupt_state state,int xcc_id)2704 static void gfx_v9_4_3_xcc_set_compute_eop_interrupt_state(
2705 	struct amdgpu_device *adev, int me, int pipe,
2706 	enum amdgpu_interrupt_state state, int xcc_id)
2707 {
2708 	u32 mec_int_cntl, mec_int_cntl_reg;
2709 
2710 	/*
2711 	 * amdgpu controls only the first MEC. That's why this function only
2712 	 * handles the setting of interrupts for this specific MEC. All other
2713 	 * pipes' interrupts are set by amdkfd.
2714 	 */
2715 
2716 	if (me == 1) {
2717 		switch (pipe) {
2718 		case 0:
2719 			mec_int_cntl_reg = SOC15_REG_OFFSET(GC, GET_INST(GC, xcc_id), regCP_ME1_PIPE0_INT_CNTL);
2720 			break;
2721 		case 1:
2722 			mec_int_cntl_reg = SOC15_REG_OFFSET(GC, GET_INST(GC, xcc_id), regCP_ME1_PIPE1_INT_CNTL);
2723 			break;
2724 		case 2:
2725 			mec_int_cntl_reg = SOC15_REG_OFFSET(GC, GET_INST(GC, xcc_id), regCP_ME1_PIPE2_INT_CNTL);
2726 			break;
2727 		case 3:
2728 			mec_int_cntl_reg = SOC15_REG_OFFSET(GC, GET_INST(GC, xcc_id), regCP_ME1_PIPE3_INT_CNTL);
2729 			break;
2730 		default:
2731 			DRM_DEBUG("invalid pipe %d\n", pipe);
2732 			return;
2733 		}
2734 	} else {
2735 		DRM_DEBUG("invalid me %d\n", me);
2736 		return;
2737 	}
2738 
2739 	switch (state) {
2740 	case AMDGPU_IRQ_STATE_DISABLE:
2741 		mec_int_cntl = RREG32(mec_int_cntl_reg);
2742 		mec_int_cntl = REG_SET_FIELD(mec_int_cntl, CP_ME1_PIPE0_INT_CNTL,
2743 					     TIME_STAMP_INT_ENABLE, 0);
2744 		WREG32(mec_int_cntl_reg, mec_int_cntl);
2745 		break;
2746 	case AMDGPU_IRQ_STATE_ENABLE:
2747 		mec_int_cntl = RREG32(mec_int_cntl_reg);
2748 		mec_int_cntl = REG_SET_FIELD(mec_int_cntl, CP_ME1_PIPE0_INT_CNTL,
2749 					     TIME_STAMP_INT_ENABLE, 1);
2750 		WREG32(mec_int_cntl_reg, mec_int_cntl);
2751 		break;
2752 	default:
2753 		break;
2754 	}
2755 }
2756 
gfx_v9_4_3_set_priv_reg_fault_state(struct amdgpu_device * adev,struct amdgpu_irq_src * source,unsigned type,enum amdgpu_interrupt_state state)2757 static int gfx_v9_4_3_set_priv_reg_fault_state(struct amdgpu_device *adev,
2758 					     struct amdgpu_irq_src *source,
2759 					     unsigned type,
2760 					     enum amdgpu_interrupt_state state)
2761 {
2762 	int i, num_xcc;
2763 
2764 	num_xcc = NUM_XCC(adev->gfx.xcc_mask);
2765 	switch (state) {
2766 	case AMDGPU_IRQ_STATE_DISABLE:
2767 	case AMDGPU_IRQ_STATE_ENABLE:
2768 		for (i = 0; i < num_xcc; i++)
2769 			WREG32_FIELD15_PREREG(GC, GET_INST(GC, i), CP_INT_CNTL_RING0,
2770 				PRIV_REG_INT_ENABLE,
2771 				state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0);
2772 		break;
2773 	default:
2774 		break;
2775 	}
2776 
2777 	return 0;
2778 }
2779 
gfx_v9_4_3_set_priv_inst_fault_state(struct amdgpu_device * adev,struct amdgpu_irq_src * source,unsigned type,enum amdgpu_interrupt_state state)2780 static int gfx_v9_4_3_set_priv_inst_fault_state(struct amdgpu_device *adev,
2781 					      struct amdgpu_irq_src *source,
2782 					      unsigned type,
2783 					      enum amdgpu_interrupt_state state)
2784 {
2785 	int i, num_xcc;
2786 
2787 	num_xcc = NUM_XCC(adev->gfx.xcc_mask);
2788 	switch (state) {
2789 	case AMDGPU_IRQ_STATE_DISABLE:
2790 	case AMDGPU_IRQ_STATE_ENABLE:
2791 		for (i = 0; i < num_xcc; i++)
2792 			WREG32_FIELD15_PREREG(GC, GET_INST(GC, i), CP_INT_CNTL_RING0,
2793 				PRIV_INSTR_INT_ENABLE,
2794 				state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0);
2795 		break;
2796 	default:
2797 		break;
2798 	}
2799 
2800 	return 0;
2801 }
2802 
gfx_v9_4_3_set_eop_interrupt_state(struct amdgpu_device * adev,struct amdgpu_irq_src * src,unsigned type,enum amdgpu_interrupt_state state)2803 static int gfx_v9_4_3_set_eop_interrupt_state(struct amdgpu_device *adev,
2804 					    struct amdgpu_irq_src *src,
2805 					    unsigned type,
2806 					    enum amdgpu_interrupt_state state)
2807 {
2808 	int i, num_xcc;
2809 
2810 	num_xcc = NUM_XCC(adev->gfx.xcc_mask);
2811 	for (i = 0; i < num_xcc; i++) {
2812 		switch (type) {
2813 		case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP:
2814 			gfx_v9_4_3_xcc_set_compute_eop_interrupt_state(
2815 				adev, 1, 0, state, i);
2816 			break;
2817 		case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE1_EOP:
2818 			gfx_v9_4_3_xcc_set_compute_eop_interrupt_state(
2819 				adev, 1, 1, state, i);
2820 			break;
2821 		case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE2_EOP:
2822 			gfx_v9_4_3_xcc_set_compute_eop_interrupt_state(
2823 				adev, 1, 2, state, i);
2824 			break;
2825 		case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE3_EOP:
2826 			gfx_v9_4_3_xcc_set_compute_eop_interrupt_state(
2827 				adev, 1, 3, state, i);
2828 			break;
2829 		case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE0_EOP:
2830 			gfx_v9_4_3_xcc_set_compute_eop_interrupt_state(
2831 				adev, 2, 0, state, i);
2832 			break;
2833 		case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE1_EOP:
2834 			gfx_v9_4_3_xcc_set_compute_eop_interrupt_state(
2835 				adev, 2, 1, state, i);
2836 			break;
2837 		case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE2_EOP:
2838 			gfx_v9_4_3_xcc_set_compute_eop_interrupt_state(
2839 				adev, 2, 2, state, i);
2840 			break;
2841 		case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE3_EOP:
2842 			gfx_v9_4_3_xcc_set_compute_eop_interrupt_state(
2843 				adev, 2, 3, state, i);
2844 			break;
2845 		default:
2846 			break;
2847 		}
2848 	}
2849 
2850 	return 0;
2851 }
2852 
gfx_v9_4_3_eop_irq(struct amdgpu_device * adev,struct amdgpu_irq_src * source,struct amdgpu_iv_entry * entry)2853 static int gfx_v9_4_3_eop_irq(struct amdgpu_device *adev,
2854 			    struct amdgpu_irq_src *source,
2855 			    struct amdgpu_iv_entry *entry)
2856 {
2857 	int i, xcc_id;
2858 	u8 me_id, pipe_id, queue_id;
2859 	struct amdgpu_ring *ring;
2860 
2861 	DRM_DEBUG("IH: CP EOP\n");
2862 	me_id = (entry->ring_id & 0x0c) >> 2;
2863 	pipe_id = (entry->ring_id & 0x03) >> 0;
2864 	queue_id = (entry->ring_id & 0x70) >> 4;
2865 
2866 	xcc_id = gfx_v9_4_3_ih_to_xcc_inst(adev, entry->node_id);
2867 
2868 	if (xcc_id == -EINVAL)
2869 		return -EINVAL;
2870 
2871 	switch (me_id) {
2872 	case 0:
2873 	case 1:
2874 	case 2:
2875 		for (i = 0; i < adev->gfx.num_compute_rings; i++) {
2876 			ring = &adev->gfx.compute_ring
2877 					[i +
2878 					 xcc_id * adev->gfx.num_compute_rings];
2879 			/* Per-queue interrupt is supported for MEC starting from VI.
2880 			  * The interrupt can only be enabled/disabled per pipe instead of per queue.
2881 			  */
2882 
2883 			if ((ring->me == me_id) && (ring->pipe == pipe_id) && (ring->queue == queue_id))
2884 				amdgpu_fence_process(ring);
2885 		}
2886 		break;
2887 	}
2888 	return 0;
2889 }
2890 
gfx_v9_4_3_fault(struct amdgpu_device * adev,struct amdgpu_iv_entry * entry)2891 static void gfx_v9_4_3_fault(struct amdgpu_device *adev,
2892 			   struct amdgpu_iv_entry *entry)
2893 {
2894 	u8 me_id, pipe_id, queue_id;
2895 	struct amdgpu_ring *ring;
2896 	int i, xcc_id;
2897 
2898 	me_id = (entry->ring_id & 0x0c) >> 2;
2899 	pipe_id = (entry->ring_id & 0x03) >> 0;
2900 	queue_id = (entry->ring_id & 0x70) >> 4;
2901 
2902 	xcc_id = gfx_v9_4_3_ih_to_xcc_inst(adev, entry->node_id);
2903 
2904 	if (xcc_id == -EINVAL)
2905 		return;
2906 
2907 	switch (me_id) {
2908 	case 0:
2909 	case 1:
2910 	case 2:
2911 		for (i = 0; i < adev->gfx.num_compute_rings; i++) {
2912 			ring = &adev->gfx.compute_ring
2913 					[i +
2914 					 xcc_id * adev->gfx.num_compute_rings];
2915 			if (ring->me == me_id && ring->pipe == pipe_id &&
2916 			    ring->queue == queue_id)
2917 				drm_sched_fault(&ring->sched);
2918 		}
2919 		break;
2920 	}
2921 }
2922 
gfx_v9_4_3_priv_reg_irq(struct amdgpu_device * adev,struct amdgpu_irq_src * source,struct amdgpu_iv_entry * entry)2923 static int gfx_v9_4_3_priv_reg_irq(struct amdgpu_device *adev,
2924 				 struct amdgpu_irq_src *source,
2925 				 struct amdgpu_iv_entry *entry)
2926 {
2927 	DRM_ERROR("Illegal register access in command stream\n");
2928 	gfx_v9_4_3_fault(adev, entry);
2929 	return 0;
2930 }
2931 
gfx_v9_4_3_priv_inst_irq(struct amdgpu_device * adev,struct amdgpu_irq_src * source,struct amdgpu_iv_entry * entry)2932 static int gfx_v9_4_3_priv_inst_irq(struct amdgpu_device *adev,
2933 				  struct amdgpu_irq_src *source,
2934 				  struct amdgpu_iv_entry *entry)
2935 {
2936 	DRM_ERROR("Illegal instruction in command stream\n");
2937 	gfx_v9_4_3_fault(adev, entry);
2938 	return 0;
2939 }
2940 
gfx_v9_4_3_emit_mem_sync(struct amdgpu_ring * ring)2941 static void gfx_v9_4_3_emit_mem_sync(struct amdgpu_ring *ring)
2942 {
2943 	const unsigned int cp_coher_cntl =
2944 			PACKET3_ACQUIRE_MEM_CP_COHER_CNTL_SH_ICACHE_ACTION_ENA(1) |
2945 			PACKET3_ACQUIRE_MEM_CP_COHER_CNTL_SH_KCACHE_ACTION_ENA(1) |
2946 			PACKET3_ACQUIRE_MEM_CP_COHER_CNTL_TC_ACTION_ENA(1) |
2947 			PACKET3_ACQUIRE_MEM_CP_COHER_CNTL_TCL1_ACTION_ENA(1) |
2948 			PACKET3_ACQUIRE_MEM_CP_COHER_CNTL_TC_WB_ACTION_ENA(1);
2949 
2950 	/* ACQUIRE_MEM -make one or more surfaces valid for use by the subsequent operations */
2951 	amdgpu_ring_write(ring, PACKET3(PACKET3_ACQUIRE_MEM, 5));
2952 	amdgpu_ring_write(ring, cp_coher_cntl); /* CP_COHER_CNTL */
2953 	amdgpu_ring_write(ring, 0xffffffff);  /* CP_COHER_SIZE */
2954 	amdgpu_ring_write(ring, 0xffffff);  /* CP_COHER_SIZE_HI */
2955 	amdgpu_ring_write(ring, 0); /* CP_COHER_BASE */
2956 	amdgpu_ring_write(ring, 0);  /* CP_COHER_BASE_HI */
2957 	amdgpu_ring_write(ring, 0x0000000A); /* POLL_INTERVAL */
2958 }
2959 
gfx_v9_4_3_emit_wave_limit_cs(struct amdgpu_ring * ring,uint32_t pipe,bool enable)2960 static void gfx_v9_4_3_emit_wave_limit_cs(struct amdgpu_ring *ring,
2961 					uint32_t pipe, bool enable)
2962 {
2963 	struct amdgpu_device *adev = ring->adev;
2964 	uint32_t val;
2965 	uint32_t wcl_cs_reg;
2966 
2967 	/* regSPI_WCL_PIPE_PERCENT_CS[0-7]_DEFAULT values are same */
2968 	val = enable ? 0x1 : 0x7f;
2969 
2970 	switch (pipe) {
2971 	case 0:
2972 		wcl_cs_reg = SOC15_REG_OFFSET(GC, GET_INST(GC, 0), regSPI_WCL_PIPE_PERCENT_CS0);
2973 		break;
2974 	case 1:
2975 		wcl_cs_reg = SOC15_REG_OFFSET(GC, GET_INST(GC, 0), regSPI_WCL_PIPE_PERCENT_CS1);
2976 		break;
2977 	case 2:
2978 		wcl_cs_reg = SOC15_REG_OFFSET(GC, GET_INST(GC, 0), regSPI_WCL_PIPE_PERCENT_CS2);
2979 		break;
2980 	case 3:
2981 		wcl_cs_reg = SOC15_REG_OFFSET(GC, GET_INST(GC, 0), regSPI_WCL_PIPE_PERCENT_CS3);
2982 		break;
2983 	default:
2984 		DRM_DEBUG("invalid pipe %d\n", pipe);
2985 		return;
2986 	}
2987 
2988 	amdgpu_ring_emit_wreg(ring, wcl_cs_reg, val);
2989 
2990 }
gfx_v9_4_3_emit_wave_limit(struct amdgpu_ring * ring,bool enable)2991 static void gfx_v9_4_3_emit_wave_limit(struct amdgpu_ring *ring, bool enable)
2992 {
2993 	struct amdgpu_device *adev = ring->adev;
2994 	uint32_t val;
2995 	int i;
2996 
2997 	/* regSPI_WCL_PIPE_PERCENT_GFX is 7 bit multiplier register to limit
2998 	 * number of gfx waves. Setting 5 bit will make sure gfx only gets
2999 	 * around 25% of gpu resources.
3000 	 */
3001 	val = enable ? 0x1f : 0x07ffffff;
3002 	amdgpu_ring_emit_wreg(ring,
3003 			      SOC15_REG_OFFSET(GC, GET_INST(GC, 0), regSPI_WCL_PIPE_PERCENT_GFX),
3004 			      val);
3005 
3006 	/* Restrict waves for normal/low priority compute queues as well
3007 	 * to get best QoS for high priority compute jobs.
3008 	 *
3009 	 * amdgpu controls only 1st ME(0-3 CS pipes).
3010 	 */
3011 	for (i = 0; i < adev->gfx.mec.num_pipe_per_mec; i++) {
3012 		if (i != ring->pipe)
3013 			gfx_v9_4_3_emit_wave_limit_cs(ring, i, enable);
3014 
3015 	}
3016 }
3017 
3018 enum amdgpu_gfx_cp_ras_mem_id {
3019 	AMDGPU_GFX_CP_MEM1 = 1,
3020 	AMDGPU_GFX_CP_MEM2,
3021 	AMDGPU_GFX_CP_MEM3,
3022 	AMDGPU_GFX_CP_MEM4,
3023 	AMDGPU_GFX_CP_MEM5,
3024 };
3025 
3026 enum amdgpu_gfx_gcea_ras_mem_id {
3027 	AMDGPU_GFX_GCEA_IOWR_CMDMEM = 4,
3028 	AMDGPU_GFX_GCEA_IORD_CMDMEM,
3029 	AMDGPU_GFX_GCEA_GMIWR_CMDMEM,
3030 	AMDGPU_GFX_GCEA_GMIRD_CMDMEM,
3031 	AMDGPU_GFX_GCEA_DRAMWR_CMDMEM,
3032 	AMDGPU_GFX_GCEA_DRAMRD_CMDMEM,
3033 	AMDGPU_GFX_GCEA_MAM_DMEM0,
3034 	AMDGPU_GFX_GCEA_MAM_DMEM1,
3035 	AMDGPU_GFX_GCEA_MAM_DMEM2,
3036 	AMDGPU_GFX_GCEA_MAM_DMEM3,
3037 	AMDGPU_GFX_GCEA_MAM_AMEM0,
3038 	AMDGPU_GFX_GCEA_MAM_AMEM1,
3039 	AMDGPU_GFX_GCEA_MAM_AMEM2,
3040 	AMDGPU_GFX_GCEA_MAM_AMEM3,
3041 	AMDGPU_GFX_GCEA_MAM_AFLUSH_BUFFER,
3042 	AMDGPU_GFX_GCEA_WRET_TAGMEM,
3043 	AMDGPU_GFX_GCEA_RRET_TAGMEM,
3044 	AMDGPU_GFX_GCEA_IOWR_DATAMEM,
3045 	AMDGPU_GFX_GCEA_GMIWR_DATAMEM,
3046 	AMDGPU_GFX_GCEA_DRAM_DATAMEM,
3047 };
3048 
3049 enum amdgpu_gfx_gc_cane_ras_mem_id {
3050 	AMDGPU_GFX_GC_CANE_MEM0 = 0,
3051 };
3052 
3053 enum amdgpu_gfx_gcutcl2_ras_mem_id {
3054 	AMDGPU_GFX_GCUTCL2_MEM2P512X95 = 160,
3055 };
3056 
3057 enum amdgpu_gfx_gds_ras_mem_id {
3058 	AMDGPU_GFX_GDS_MEM0 = 0,
3059 };
3060 
3061 enum amdgpu_gfx_lds_ras_mem_id {
3062 	AMDGPU_GFX_LDS_BANK0 = 0,
3063 	AMDGPU_GFX_LDS_BANK1,
3064 	AMDGPU_GFX_LDS_BANK2,
3065 	AMDGPU_GFX_LDS_BANK3,
3066 	AMDGPU_GFX_LDS_BANK4,
3067 	AMDGPU_GFX_LDS_BANK5,
3068 	AMDGPU_GFX_LDS_BANK6,
3069 	AMDGPU_GFX_LDS_BANK7,
3070 	AMDGPU_GFX_LDS_BANK8,
3071 	AMDGPU_GFX_LDS_BANK9,
3072 	AMDGPU_GFX_LDS_BANK10,
3073 	AMDGPU_GFX_LDS_BANK11,
3074 	AMDGPU_GFX_LDS_BANK12,
3075 	AMDGPU_GFX_LDS_BANK13,
3076 	AMDGPU_GFX_LDS_BANK14,
3077 	AMDGPU_GFX_LDS_BANK15,
3078 	AMDGPU_GFX_LDS_BANK16,
3079 	AMDGPU_GFX_LDS_BANK17,
3080 	AMDGPU_GFX_LDS_BANK18,
3081 	AMDGPU_GFX_LDS_BANK19,
3082 	AMDGPU_GFX_LDS_BANK20,
3083 	AMDGPU_GFX_LDS_BANK21,
3084 	AMDGPU_GFX_LDS_BANK22,
3085 	AMDGPU_GFX_LDS_BANK23,
3086 	AMDGPU_GFX_LDS_BANK24,
3087 	AMDGPU_GFX_LDS_BANK25,
3088 	AMDGPU_GFX_LDS_BANK26,
3089 	AMDGPU_GFX_LDS_BANK27,
3090 	AMDGPU_GFX_LDS_BANK28,
3091 	AMDGPU_GFX_LDS_BANK29,
3092 	AMDGPU_GFX_LDS_BANK30,
3093 	AMDGPU_GFX_LDS_BANK31,
3094 	AMDGPU_GFX_LDS_SP_BUFFER_A,
3095 	AMDGPU_GFX_LDS_SP_BUFFER_B,
3096 };
3097 
3098 enum amdgpu_gfx_rlc_ras_mem_id {
3099 	AMDGPU_GFX_RLC_GPMF32 = 1,
3100 	AMDGPU_GFX_RLC_RLCVF32,
3101 	AMDGPU_GFX_RLC_SCRATCH,
3102 	AMDGPU_GFX_RLC_SRM_ARAM,
3103 	AMDGPU_GFX_RLC_SRM_DRAM,
3104 	AMDGPU_GFX_RLC_TCTAG,
3105 	AMDGPU_GFX_RLC_SPM_SE,
3106 	AMDGPU_GFX_RLC_SPM_GRBMT,
3107 };
3108 
3109 enum amdgpu_gfx_sp_ras_mem_id {
3110 	AMDGPU_GFX_SP_SIMDID0 = 0,
3111 };
3112 
3113 enum amdgpu_gfx_spi_ras_mem_id {
3114 	AMDGPU_GFX_SPI_MEM0 = 0,
3115 	AMDGPU_GFX_SPI_MEM1,
3116 	AMDGPU_GFX_SPI_MEM2,
3117 	AMDGPU_GFX_SPI_MEM3,
3118 };
3119 
3120 enum amdgpu_gfx_sqc_ras_mem_id {
3121 	AMDGPU_GFX_SQC_INST_CACHE_A = 100,
3122 	AMDGPU_GFX_SQC_INST_CACHE_B = 101,
3123 	AMDGPU_GFX_SQC_INST_CACHE_TAG_A = 102,
3124 	AMDGPU_GFX_SQC_INST_CACHE_TAG_B = 103,
3125 	AMDGPU_GFX_SQC_INST_CACHE_MISS_FIFO_A = 104,
3126 	AMDGPU_GFX_SQC_INST_CACHE_MISS_FIFO_B = 105,
3127 	AMDGPU_GFX_SQC_INST_CACHE_GATCL1_MISS_FIFO_A = 106,
3128 	AMDGPU_GFX_SQC_INST_CACHE_GATCL1_MISS_FIFO_B = 107,
3129 	AMDGPU_GFX_SQC_DATA_CACHE_A = 200,
3130 	AMDGPU_GFX_SQC_DATA_CACHE_B = 201,
3131 	AMDGPU_GFX_SQC_DATA_CACHE_TAG_A = 202,
3132 	AMDGPU_GFX_SQC_DATA_CACHE_TAG_B = 203,
3133 	AMDGPU_GFX_SQC_DATA_CACHE_MISS_FIFO_A = 204,
3134 	AMDGPU_GFX_SQC_DATA_CACHE_MISS_FIFO_B = 205,
3135 	AMDGPU_GFX_SQC_DATA_CACHE_HIT_FIFO_A = 206,
3136 	AMDGPU_GFX_SQC_DATA_CACHE_HIT_FIFO_B = 207,
3137 	AMDGPU_GFX_SQC_DIRTY_BIT_A = 208,
3138 	AMDGPU_GFX_SQC_DIRTY_BIT_B = 209,
3139 	AMDGPU_GFX_SQC_WRITE_DATA_BUFFER_CU0 = 210,
3140 	AMDGPU_GFX_SQC_WRITE_DATA_BUFFER_CU1 = 211,
3141 	AMDGPU_GFX_SQC_UTCL1_MISS_LFIFO_DATA_CACHE_A = 212,
3142 	AMDGPU_GFX_SQC_UTCL1_MISS_LFIFO_DATA_CACHE_B = 213,
3143 	AMDGPU_GFX_SQC_UTCL1_MISS_LFIFO_INST_CACHE = 108,
3144 };
3145 
3146 enum amdgpu_gfx_sq_ras_mem_id {
3147 	AMDGPU_GFX_SQ_SGPR_MEM0 = 0,
3148 	AMDGPU_GFX_SQ_SGPR_MEM1,
3149 	AMDGPU_GFX_SQ_SGPR_MEM2,
3150 	AMDGPU_GFX_SQ_SGPR_MEM3,
3151 };
3152 
3153 enum amdgpu_gfx_ta_ras_mem_id {
3154 	AMDGPU_GFX_TA_FS_AFIFO_RAM_LO = 1,
3155 	AMDGPU_GFX_TA_FS_AFIFO_RAM_HI,
3156 	AMDGPU_GFX_TA_FS_CFIFO_RAM,
3157 	AMDGPU_GFX_TA_FSX_LFIFO,
3158 	AMDGPU_GFX_TA_FS_DFIFO_RAM,
3159 };
3160 
3161 enum amdgpu_gfx_tcc_ras_mem_id {
3162 	AMDGPU_GFX_TCC_MEM1 = 1,
3163 };
3164 
3165 enum amdgpu_gfx_tca_ras_mem_id {
3166 	AMDGPU_GFX_TCA_MEM1 = 1,
3167 };
3168 
3169 enum amdgpu_gfx_tci_ras_mem_id {
3170 	AMDGPU_GFX_TCIW_MEM = 1,
3171 };
3172 
3173 enum amdgpu_gfx_tcp_ras_mem_id {
3174 	AMDGPU_GFX_TCP_LFIFO0 = 1,
3175 	AMDGPU_GFX_TCP_SET0BANK0_RAM,
3176 	AMDGPU_GFX_TCP_SET0BANK1_RAM,
3177 	AMDGPU_GFX_TCP_SET0BANK2_RAM,
3178 	AMDGPU_GFX_TCP_SET0BANK3_RAM,
3179 	AMDGPU_GFX_TCP_SET1BANK0_RAM,
3180 	AMDGPU_GFX_TCP_SET1BANK1_RAM,
3181 	AMDGPU_GFX_TCP_SET1BANK2_RAM,
3182 	AMDGPU_GFX_TCP_SET1BANK3_RAM,
3183 	AMDGPU_GFX_TCP_SET2BANK0_RAM,
3184 	AMDGPU_GFX_TCP_SET2BANK1_RAM,
3185 	AMDGPU_GFX_TCP_SET2BANK2_RAM,
3186 	AMDGPU_GFX_TCP_SET2BANK3_RAM,
3187 	AMDGPU_GFX_TCP_SET3BANK0_RAM,
3188 	AMDGPU_GFX_TCP_SET3BANK1_RAM,
3189 	AMDGPU_GFX_TCP_SET3BANK2_RAM,
3190 	AMDGPU_GFX_TCP_SET3BANK3_RAM,
3191 	AMDGPU_GFX_TCP_VM_FIFO,
3192 	AMDGPU_GFX_TCP_DB_TAGRAM0,
3193 	AMDGPU_GFX_TCP_DB_TAGRAM1,
3194 	AMDGPU_GFX_TCP_DB_TAGRAM2,
3195 	AMDGPU_GFX_TCP_DB_TAGRAM3,
3196 	AMDGPU_GFX_TCP_UTCL1_LFIFO_PROBE0,
3197 	AMDGPU_GFX_TCP_UTCL1_LFIFO_PROBE1,
3198 	AMDGPU_GFX_TCP_CMD_FIFO,
3199 };
3200 
3201 enum amdgpu_gfx_td_ras_mem_id {
3202 	AMDGPU_GFX_TD_UTD_CS_FIFO_MEM = 1,
3203 	AMDGPU_GFX_TD_UTD_SS_FIFO_LO_MEM,
3204 	AMDGPU_GFX_TD_UTD_SS_FIFO_HI_MEM,
3205 };
3206 
3207 enum amdgpu_gfx_tcx_ras_mem_id {
3208 	AMDGPU_GFX_TCX_FIFOD0 = 0,
3209 	AMDGPU_GFX_TCX_FIFOD1,
3210 	AMDGPU_GFX_TCX_FIFOD2,
3211 	AMDGPU_GFX_TCX_FIFOD3,
3212 	AMDGPU_GFX_TCX_FIFOD4,
3213 	AMDGPU_GFX_TCX_FIFOD5,
3214 	AMDGPU_GFX_TCX_FIFOD6,
3215 	AMDGPU_GFX_TCX_FIFOD7,
3216 	AMDGPU_GFX_TCX_FIFOB0,
3217 	AMDGPU_GFX_TCX_FIFOB1,
3218 	AMDGPU_GFX_TCX_FIFOB2,
3219 	AMDGPU_GFX_TCX_FIFOB3,
3220 	AMDGPU_GFX_TCX_FIFOB4,
3221 	AMDGPU_GFX_TCX_FIFOB5,
3222 	AMDGPU_GFX_TCX_FIFOB6,
3223 	AMDGPU_GFX_TCX_FIFOB7,
3224 	AMDGPU_GFX_TCX_FIFOA0,
3225 	AMDGPU_GFX_TCX_FIFOA1,
3226 	AMDGPU_GFX_TCX_FIFOA2,
3227 	AMDGPU_GFX_TCX_FIFOA3,
3228 	AMDGPU_GFX_TCX_FIFOA4,
3229 	AMDGPU_GFX_TCX_FIFOA5,
3230 	AMDGPU_GFX_TCX_FIFOA6,
3231 	AMDGPU_GFX_TCX_FIFOA7,
3232 	AMDGPU_GFX_TCX_CFIFO0,
3233 	AMDGPU_GFX_TCX_CFIFO1,
3234 	AMDGPU_GFX_TCX_CFIFO2,
3235 	AMDGPU_GFX_TCX_CFIFO3,
3236 	AMDGPU_GFX_TCX_CFIFO4,
3237 	AMDGPU_GFX_TCX_CFIFO5,
3238 	AMDGPU_GFX_TCX_CFIFO6,
3239 	AMDGPU_GFX_TCX_CFIFO7,
3240 	AMDGPU_GFX_TCX_FIFO_ACKB0,
3241 	AMDGPU_GFX_TCX_FIFO_ACKB1,
3242 	AMDGPU_GFX_TCX_FIFO_ACKB2,
3243 	AMDGPU_GFX_TCX_FIFO_ACKB3,
3244 	AMDGPU_GFX_TCX_FIFO_ACKB4,
3245 	AMDGPU_GFX_TCX_FIFO_ACKB5,
3246 	AMDGPU_GFX_TCX_FIFO_ACKB6,
3247 	AMDGPU_GFX_TCX_FIFO_ACKB7,
3248 	AMDGPU_GFX_TCX_FIFO_ACKD0,
3249 	AMDGPU_GFX_TCX_FIFO_ACKD1,
3250 	AMDGPU_GFX_TCX_FIFO_ACKD2,
3251 	AMDGPU_GFX_TCX_FIFO_ACKD3,
3252 	AMDGPU_GFX_TCX_FIFO_ACKD4,
3253 	AMDGPU_GFX_TCX_FIFO_ACKD5,
3254 	AMDGPU_GFX_TCX_FIFO_ACKD6,
3255 	AMDGPU_GFX_TCX_FIFO_ACKD7,
3256 	AMDGPU_GFX_TCX_DST_FIFOA0,
3257 	AMDGPU_GFX_TCX_DST_FIFOA1,
3258 	AMDGPU_GFX_TCX_DST_FIFOA2,
3259 	AMDGPU_GFX_TCX_DST_FIFOA3,
3260 	AMDGPU_GFX_TCX_DST_FIFOA4,
3261 	AMDGPU_GFX_TCX_DST_FIFOA5,
3262 	AMDGPU_GFX_TCX_DST_FIFOA6,
3263 	AMDGPU_GFX_TCX_DST_FIFOA7,
3264 	AMDGPU_GFX_TCX_DST_FIFOB0,
3265 	AMDGPU_GFX_TCX_DST_FIFOB1,
3266 	AMDGPU_GFX_TCX_DST_FIFOB2,
3267 	AMDGPU_GFX_TCX_DST_FIFOB3,
3268 	AMDGPU_GFX_TCX_DST_FIFOB4,
3269 	AMDGPU_GFX_TCX_DST_FIFOB5,
3270 	AMDGPU_GFX_TCX_DST_FIFOB6,
3271 	AMDGPU_GFX_TCX_DST_FIFOB7,
3272 	AMDGPU_GFX_TCX_DST_FIFOD0,
3273 	AMDGPU_GFX_TCX_DST_FIFOD1,
3274 	AMDGPU_GFX_TCX_DST_FIFOD2,
3275 	AMDGPU_GFX_TCX_DST_FIFOD3,
3276 	AMDGPU_GFX_TCX_DST_FIFOD4,
3277 	AMDGPU_GFX_TCX_DST_FIFOD5,
3278 	AMDGPU_GFX_TCX_DST_FIFOD6,
3279 	AMDGPU_GFX_TCX_DST_FIFOD7,
3280 	AMDGPU_GFX_TCX_DST_FIFO_ACKB0,
3281 	AMDGPU_GFX_TCX_DST_FIFO_ACKB1,
3282 	AMDGPU_GFX_TCX_DST_FIFO_ACKB2,
3283 	AMDGPU_GFX_TCX_DST_FIFO_ACKB3,
3284 	AMDGPU_GFX_TCX_DST_FIFO_ACKB4,
3285 	AMDGPU_GFX_TCX_DST_FIFO_ACKB5,
3286 	AMDGPU_GFX_TCX_DST_FIFO_ACKB6,
3287 	AMDGPU_GFX_TCX_DST_FIFO_ACKB7,
3288 	AMDGPU_GFX_TCX_DST_FIFO_ACKD0,
3289 	AMDGPU_GFX_TCX_DST_FIFO_ACKD1,
3290 	AMDGPU_GFX_TCX_DST_FIFO_ACKD2,
3291 	AMDGPU_GFX_TCX_DST_FIFO_ACKD3,
3292 	AMDGPU_GFX_TCX_DST_FIFO_ACKD4,
3293 	AMDGPU_GFX_TCX_DST_FIFO_ACKD5,
3294 	AMDGPU_GFX_TCX_DST_FIFO_ACKD6,
3295 	AMDGPU_GFX_TCX_DST_FIFO_ACKD7,
3296 };
3297 
3298 enum amdgpu_gfx_atc_l2_ras_mem_id {
3299 	AMDGPU_GFX_ATC_L2_MEM0 = 0,
3300 };
3301 
3302 enum amdgpu_gfx_utcl2_ras_mem_id {
3303 	AMDGPU_GFX_UTCL2_MEM0 = 0,
3304 };
3305 
3306 enum amdgpu_gfx_vml2_ras_mem_id {
3307 	AMDGPU_GFX_VML2_MEM0 = 0,
3308 };
3309 
3310 enum amdgpu_gfx_vml2_walker_ras_mem_id {
3311 	AMDGPU_GFX_VML2_WALKER_MEM0 = 0,
3312 };
3313 
3314 static const struct amdgpu_ras_memory_id_entry gfx_v9_4_3_ras_cp_mem_list[] = {
3315 	{AMDGPU_GFX_CP_MEM1, "CP_MEM1"},
3316 	{AMDGPU_GFX_CP_MEM2, "CP_MEM2"},
3317 	{AMDGPU_GFX_CP_MEM3, "CP_MEM3"},
3318 	{AMDGPU_GFX_CP_MEM4, "CP_MEM4"},
3319 	{AMDGPU_GFX_CP_MEM5, "CP_MEM5"},
3320 };
3321 
3322 static const struct amdgpu_ras_memory_id_entry gfx_v9_4_3_ras_gcea_mem_list[] = {
3323 	{AMDGPU_GFX_GCEA_IOWR_CMDMEM, "GCEA_IOWR_CMDMEM"},
3324 	{AMDGPU_GFX_GCEA_IORD_CMDMEM, "GCEA_IORD_CMDMEM"},
3325 	{AMDGPU_GFX_GCEA_GMIWR_CMDMEM, "GCEA_GMIWR_CMDMEM"},
3326 	{AMDGPU_GFX_GCEA_GMIRD_CMDMEM, "GCEA_GMIRD_CMDMEM"},
3327 	{AMDGPU_GFX_GCEA_DRAMWR_CMDMEM, "GCEA_DRAMWR_CMDMEM"},
3328 	{AMDGPU_GFX_GCEA_DRAMRD_CMDMEM, "GCEA_DRAMRD_CMDMEM"},
3329 	{AMDGPU_GFX_GCEA_MAM_DMEM0, "GCEA_MAM_DMEM0"},
3330 	{AMDGPU_GFX_GCEA_MAM_DMEM1, "GCEA_MAM_DMEM1"},
3331 	{AMDGPU_GFX_GCEA_MAM_DMEM2, "GCEA_MAM_DMEM2"},
3332 	{AMDGPU_GFX_GCEA_MAM_DMEM3, "GCEA_MAM_DMEM3"},
3333 	{AMDGPU_GFX_GCEA_MAM_AMEM0, "GCEA_MAM_AMEM0"},
3334 	{AMDGPU_GFX_GCEA_MAM_AMEM1, "GCEA_MAM_AMEM1"},
3335 	{AMDGPU_GFX_GCEA_MAM_AMEM2, "GCEA_MAM_AMEM2"},
3336 	{AMDGPU_GFX_GCEA_MAM_AMEM3, "GCEA_MAM_AMEM3"},
3337 	{AMDGPU_GFX_GCEA_MAM_AFLUSH_BUFFER, "GCEA_MAM_AFLUSH_BUFFER"},
3338 	{AMDGPU_GFX_GCEA_WRET_TAGMEM, "GCEA_WRET_TAGMEM"},
3339 	{AMDGPU_GFX_GCEA_RRET_TAGMEM, "GCEA_RRET_TAGMEM"},
3340 	{AMDGPU_GFX_GCEA_IOWR_DATAMEM, "GCEA_IOWR_DATAMEM"},
3341 	{AMDGPU_GFX_GCEA_GMIWR_DATAMEM, "GCEA_GMIWR_DATAMEM"},
3342 	{AMDGPU_GFX_GCEA_DRAM_DATAMEM, "GCEA_DRAM_DATAMEM"},
3343 };
3344 
3345 static const struct amdgpu_ras_memory_id_entry gfx_v9_4_3_ras_gc_cane_mem_list[] = {
3346 	{AMDGPU_GFX_GC_CANE_MEM0, "GC_CANE_MEM0"},
3347 };
3348 
3349 static const struct amdgpu_ras_memory_id_entry gfx_v9_4_3_ras_gcutcl2_mem_list[] = {
3350 	{AMDGPU_GFX_GCUTCL2_MEM2P512X95, "GCUTCL2_MEM2P512X95"},
3351 };
3352 
3353 static const struct amdgpu_ras_memory_id_entry gfx_v9_4_3_ras_gds_mem_list[] = {
3354 	{AMDGPU_GFX_GDS_MEM0, "GDS_MEM"},
3355 };
3356 
3357 static const struct amdgpu_ras_memory_id_entry gfx_v9_4_3_ras_lds_mem_list[] = {
3358 	{AMDGPU_GFX_LDS_BANK0, "LDS_BANK0"},
3359 	{AMDGPU_GFX_LDS_BANK1, "LDS_BANK1"},
3360 	{AMDGPU_GFX_LDS_BANK2, "LDS_BANK2"},
3361 	{AMDGPU_GFX_LDS_BANK3, "LDS_BANK3"},
3362 	{AMDGPU_GFX_LDS_BANK4, "LDS_BANK4"},
3363 	{AMDGPU_GFX_LDS_BANK5, "LDS_BANK5"},
3364 	{AMDGPU_GFX_LDS_BANK6, "LDS_BANK6"},
3365 	{AMDGPU_GFX_LDS_BANK7, "LDS_BANK7"},
3366 	{AMDGPU_GFX_LDS_BANK8, "LDS_BANK8"},
3367 	{AMDGPU_GFX_LDS_BANK9, "LDS_BANK9"},
3368 	{AMDGPU_GFX_LDS_BANK10, "LDS_BANK10"},
3369 	{AMDGPU_GFX_LDS_BANK11, "LDS_BANK11"},
3370 	{AMDGPU_GFX_LDS_BANK12, "LDS_BANK12"},
3371 	{AMDGPU_GFX_LDS_BANK13, "LDS_BANK13"},
3372 	{AMDGPU_GFX_LDS_BANK14, "LDS_BANK14"},
3373 	{AMDGPU_GFX_LDS_BANK15, "LDS_BANK15"},
3374 	{AMDGPU_GFX_LDS_BANK16, "LDS_BANK16"},
3375 	{AMDGPU_GFX_LDS_BANK17, "LDS_BANK17"},
3376 	{AMDGPU_GFX_LDS_BANK18, "LDS_BANK18"},
3377 	{AMDGPU_GFX_LDS_BANK19, "LDS_BANK19"},
3378 	{AMDGPU_GFX_LDS_BANK20, "LDS_BANK20"},
3379 	{AMDGPU_GFX_LDS_BANK21, "LDS_BANK21"},
3380 	{AMDGPU_GFX_LDS_BANK22, "LDS_BANK22"},
3381 	{AMDGPU_GFX_LDS_BANK23, "LDS_BANK23"},
3382 	{AMDGPU_GFX_LDS_BANK24, "LDS_BANK24"},
3383 	{AMDGPU_GFX_LDS_BANK25, "LDS_BANK25"},
3384 	{AMDGPU_GFX_LDS_BANK26, "LDS_BANK26"},
3385 	{AMDGPU_GFX_LDS_BANK27, "LDS_BANK27"},
3386 	{AMDGPU_GFX_LDS_BANK28, "LDS_BANK28"},
3387 	{AMDGPU_GFX_LDS_BANK29, "LDS_BANK29"},
3388 	{AMDGPU_GFX_LDS_BANK30, "LDS_BANK30"},
3389 	{AMDGPU_GFX_LDS_BANK31, "LDS_BANK31"},
3390 	{AMDGPU_GFX_LDS_SP_BUFFER_A, "LDS_SP_BUFFER_A"},
3391 	{AMDGPU_GFX_LDS_SP_BUFFER_B, "LDS_SP_BUFFER_B"},
3392 };
3393 
3394 static const struct amdgpu_ras_memory_id_entry gfx_v9_4_3_ras_rlc_mem_list[] = {
3395 	{AMDGPU_GFX_RLC_GPMF32, "RLC_GPMF32"},
3396 	{AMDGPU_GFX_RLC_RLCVF32, "RLC_RLCVF32"},
3397 	{AMDGPU_GFX_RLC_SCRATCH, "RLC_SCRATCH"},
3398 	{AMDGPU_GFX_RLC_SRM_ARAM, "RLC_SRM_ARAM"},
3399 	{AMDGPU_GFX_RLC_SRM_DRAM, "RLC_SRM_DRAM"},
3400 	{AMDGPU_GFX_RLC_TCTAG, "RLC_TCTAG"},
3401 	{AMDGPU_GFX_RLC_SPM_SE, "RLC_SPM_SE"},
3402 	{AMDGPU_GFX_RLC_SPM_GRBMT, "RLC_SPM_GRBMT"},
3403 };
3404 
3405 static const struct amdgpu_ras_memory_id_entry gfx_v9_4_3_ras_sp_mem_list[] = {
3406 	{AMDGPU_GFX_SP_SIMDID0, "SP_SIMDID0"},
3407 };
3408 
3409 static const struct amdgpu_ras_memory_id_entry gfx_v9_4_3_ras_spi_mem_list[] = {
3410 	{AMDGPU_GFX_SPI_MEM0, "SPI_MEM0"},
3411 	{AMDGPU_GFX_SPI_MEM1, "SPI_MEM1"},
3412 	{AMDGPU_GFX_SPI_MEM2, "SPI_MEM2"},
3413 	{AMDGPU_GFX_SPI_MEM3, "SPI_MEM3"},
3414 };
3415 
3416 static const struct amdgpu_ras_memory_id_entry gfx_v9_4_3_ras_sqc_mem_list[] = {
3417 	{AMDGPU_GFX_SQC_INST_CACHE_A, "SQC_INST_CACHE_A"},
3418 	{AMDGPU_GFX_SQC_INST_CACHE_B, "SQC_INST_CACHE_B"},
3419 	{AMDGPU_GFX_SQC_INST_CACHE_TAG_A, "SQC_INST_CACHE_TAG_A"},
3420 	{AMDGPU_GFX_SQC_INST_CACHE_TAG_B, "SQC_INST_CACHE_TAG_B"},
3421 	{AMDGPU_GFX_SQC_INST_CACHE_MISS_FIFO_A, "SQC_INST_CACHE_MISS_FIFO_A"},
3422 	{AMDGPU_GFX_SQC_INST_CACHE_MISS_FIFO_B, "SQC_INST_CACHE_MISS_FIFO_B"},
3423 	{AMDGPU_GFX_SQC_INST_CACHE_GATCL1_MISS_FIFO_A, "SQC_INST_CACHE_GATCL1_MISS_FIFO_A"},
3424 	{AMDGPU_GFX_SQC_INST_CACHE_GATCL1_MISS_FIFO_B, "SQC_INST_CACHE_GATCL1_MISS_FIFO_B"},
3425 	{AMDGPU_GFX_SQC_DATA_CACHE_A, "SQC_DATA_CACHE_A"},
3426 	{AMDGPU_GFX_SQC_DATA_CACHE_B, "SQC_DATA_CACHE_B"},
3427 	{AMDGPU_GFX_SQC_DATA_CACHE_TAG_A, "SQC_DATA_CACHE_TAG_A"},
3428 	{AMDGPU_GFX_SQC_DATA_CACHE_TAG_B, "SQC_DATA_CACHE_TAG_B"},
3429 	{AMDGPU_GFX_SQC_DATA_CACHE_MISS_FIFO_A, "SQC_DATA_CACHE_MISS_FIFO_A"},
3430 	{AMDGPU_GFX_SQC_DATA_CACHE_MISS_FIFO_B, "SQC_DATA_CACHE_MISS_FIFO_B"},
3431 	{AMDGPU_GFX_SQC_DATA_CACHE_HIT_FIFO_A, "SQC_DATA_CACHE_HIT_FIFO_A"},
3432 	{AMDGPU_GFX_SQC_DATA_CACHE_HIT_FIFO_B, "SQC_DATA_CACHE_HIT_FIFO_B"},
3433 	{AMDGPU_GFX_SQC_DIRTY_BIT_A, "SQC_DIRTY_BIT_A"},
3434 	{AMDGPU_GFX_SQC_DIRTY_BIT_B, "SQC_DIRTY_BIT_B"},
3435 	{AMDGPU_GFX_SQC_WRITE_DATA_BUFFER_CU0, "SQC_WRITE_DATA_BUFFER_CU0"},
3436 	{AMDGPU_GFX_SQC_WRITE_DATA_BUFFER_CU1, "SQC_WRITE_DATA_BUFFER_CU1"},
3437 	{AMDGPU_GFX_SQC_UTCL1_MISS_LFIFO_DATA_CACHE_A, "SQC_UTCL1_MISS_LFIFO_DATA_CACHE_A"},
3438 	{AMDGPU_GFX_SQC_UTCL1_MISS_LFIFO_DATA_CACHE_B, "SQC_UTCL1_MISS_LFIFO_DATA_CACHE_B"},
3439 	{AMDGPU_GFX_SQC_UTCL1_MISS_LFIFO_INST_CACHE, "SQC_UTCL1_MISS_LFIFO_INST_CACHE"},
3440 };
3441 
3442 static const struct amdgpu_ras_memory_id_entry gfx_v9_4_3_ras_sq_mem_list[] = {
3443 	{AMDGPU_GFX_SQ_SGPR_MEM0, "SQ_SGPR_MEM0"},
3444 	{AMDGPU_GFX_SQ_SGPR_MEM1, "SQ_SGPR_MEM1"},
3445 	{AMDGPU_GFX_SQ_SGPR_MEM2, "SQ_SGPR_MEM2"},
3446 	{AMDGPU_GFX_SQ_SGPR_MEM3, "SQ_SGPR_MEM3"},
3447 };
3448 
3449 static const struct amdgpu_ras_memory_id_entry gfx_v9_4_3_ras_ta_mem_list[] = {
3450 	{AMDGPU_GFX_TA_FS_AFIFO_RAM_LO, "TA_FS_AFIFO_RAM_LO"},
3451 	{AMDGPU_GFX_TA_FS_AFIFO_RAM_HI, "TA_FS_AFIFO_RAM_HI"},
3452 	{AMDGPU_GFX_TA_FS_CFIFO_RAM, "TA_FS_CFIFO_RAM"},
3453 	{AMDGPU_GFX_TA_FSX_LFIFO, "TA_FSX_LFIFO"},
3454 	{AMDGPU_GFX_TA_FS_DFIFO_RAM, "TA_FS_DFIFO_RAM"},
3455 };
3456 
3457 static const struct amdgpu_ras_memory_id_entry gfx_v9_4_3_ras_tcc_mem_list[] = {
3458 	{AMDGPU_GFX_TCC_MEM1, "TCC_MEM1"},
3459 };
3460 
3461 static const struct amdgpu_ras_memory_id_entry gfx_v9_4_3_ras_tca_mem_list[] = {
3462 	{AMDGPU_GFX_TCA_MEM1, "TCA_MEM1"},
3463 };
3464 
3465 static const struct amdgpu_ras_memory_id_entry gfx_v9_4_3_ras_tci_mem_list[] = {
3466 	{AMDGPU_GFX_TCIW_MEM, "TCIW_MEM"},
3467 };
3468 
3469 static const struct amdgpu_ras_memory_id_entry gfx_v9_4_3_ras_tcp_mem_list[] = {
3470 	{AMDGPU_GFX_TCP_LFIFO0, "TCP_LFIFO0"},
3471 	{AMDGPU_GFX_TCP_SET0BANK0_RAM, "TCP_SET0BANK0_RAM"},
3472 	{AMDGPU_GFX_TCP_SET0BANK1_RAM, "TCP_SET0BANK1_RAM"},
3473 	{AMDGPU_GFX_TCP_SET0BANK2_RAM, "TCP_SET0BANK2_RAM"},
3474 	{AMDGPU_GFX_TCP_SET0BANK3_RAM, "TCP_SET0BANK3_RAM"},
3475 	{AMDGPU_GFX_TCP_SET1BANK0_RAM, "TCP_SET1BANK0_RAM"},
3476 	{AMDGPU_GFX_TCP_SET1BANK1_RAM, "TCP_SET1BANK1_RAM"},
3477 	{AMDGPU_GFX_TCP_SET1BANK2_RAM, "TCP_SET1BANK2_RAM"},
3478 	{AMDGPU_GFX_TCP_SET1BANK3_RAM, "TCP_SET1BANK3_RAM"},
3479 	{AMDGPU_GFX_TCP_SET2BANK0_RAM, "TCP_SET2BANK0_RAM"},
3480 	{AMDGPU_GFX_TCP_SET2BANK1_RAM, "TCP_SET2BANK1_RAM"},
3481 	{AMDGPU_GFX_TCP_SET2BANK2_RAM, "TCP_SET2BANK2_RAM"},
3482 	{AMDGPU_GFX_TCP_SET2BANK3_RAM, "TCP_SET2BANK3_RAM"},
3483 	{AMDGPU_GFX_TCP_SET3BANK0_RAM, "TCP_SET3BANK0_RAM"},
3484 	{AMDGPU_GFX_TCP_SET3BANK1_RAM, "TCP_SET3BANK1_RAM"},
3485 	{AMDGPU_GFX_TCP_SET3BANK2_RAM, "TCP_SET3BANK2_RAM"},
3486 	{AMDGPU_GFX_TCP_SET3BANK3_RAM, "TCP_SET3BANK3_RAM"},
3487 	{AMDGPU_GFX_TCP_VM_FIFO, "TCP_VM_FIFO"},
3488 	{AMDGPU_GFX_TCP_DB_TAGRAM0, "TCP_DB_TAGRAM0"},
3489 	{AMDGPU_GFX_TCP_DB_TAGRAM1, "TCP_DB_TAGRAM1"},
3490 	{AMDGPU_GFX_TCP_DB_TAGRAM2, "TCP_DB_TAGRAM2"},
3491 	{AMDGPU_GFX_TCP_DB_TAGRAM3, "TCP_DB_TAGRAM3"},
3492 	{AMDGPU_GFX_TCP_UTCL1_LFIFO_PROBE0, "TCP_UTCL1_LFIFO_PROBE0"},
3493 	{AMDGPU_GFX_TCP_UTCL1_LFIFO_PROBE1, "TCP_UTCL1_LFIFO_PROBE1"},
3494 	{AMDGPU_GFX_TCP_CMD_FIFO, "TCP_CMD_FIFO"},
3495 };
3496 
3497 static const struct amdgpu_ras_memory_id_entry gfx_v9_4_3_ras_td_mem_list[] = {
3498 	{AMDGPU_GFX_TD_UTD_CS_FIFO_MEM, "TD_UTD_CS_FIFO_MEM"},
3499 	{AMDGPU_GFX_TD_UTD_SS_FIFO_LO_MEM, "TD_UTD_SS_FIFO_LO_MEM"},
3500 	{AMDGPU_GFX_TD_UTD_SS_FIFO_HI_MEM, "TD_UTD_SS_FIFO_HI_MEM"},
3501 };
3502 
3503 static const struct amdgpu_ras_memory_id_entry gfx_v9_4_3_ras_tcx_mem_list[] = {
3504 	{AMDGPU_GFX_TCX_FIFOD0, "TCX_FIFOD0"},
3505 	{AMDGPU_GFX_TCX_FIFOD1, "TCX_FIFOD1"},
3506 	{AMDGPU_GFX_TCX_FIFOD2, "TCX_FIFOD2"},
3507 	{AMDGPU_GFX_TCX_FIFOD3, "TCX_FIFOD3"},
3508 	{AMDGPU_GFX_TCX_FIFOD4, "TCX_FIFOD4"},
3509 	{AMDGPU_GFX_TCX_FIFOD5, "TCX_FIFOD5"},
3510 	{AMDGPU_GFX_TCX_FIFOD6, "TCX_FIFOD6"},
3511 	{AMDGPU_GFX_TCX_FIFOD7, "TCX_FIFOD7"},
3512 	{AMDGPU_GFX_TCX_FIFOB0, "TCX_FIFOB0"},
3513 	{AMDGPU_GFX_TCX_FIFOB1, "TCX_FIFOB1"},
3514 	{AMDGPU_GFX_TCX_FIFOB2, "TCX_FIFOB2"},
3515 	{AMDGPU_GFX_TCX_FIFOB3, "TCX_FIFOB3"},
3516 	{AMDGPU_GFX_TCX_FIFOB4, "TCX_FIFOB4"},
3517 	{AMDGPU_GFX_TCX_FIFOB5, "TCX_FIFOB5"},
3518 	{AMDGPU_GFX_TCX_FIFOB6, "TCX_FIFOB6"},
3519 	{AMDGPU_GFX_TCX_FIFOB7, "TCX_FIFOB7"},
3520 	{AMDGPU_GFX_TCX_FIFOA0, "TCX_FIFOA0"},
3521 	{AMDGPU_GFX_TCX_FIFOA1, "TCX_FIFOA1"},
3522 	{AMDGPU_GFX_TCX_FIFOA2, "TCX_FIFOA2"},
3523 	{AMDGPU_GFX_TCX_FIFOA3, "TCX_FIFOA3"},
3524 	{AMDGPU_GFX_TCX_FIFOA4, "TCX_FIFOA4"},
3525 	{AMDGPU_GFX_TCX_FIFOA5, "TCX_FIFOA5"},
3526 	{AMDGPU_GFX_TCX_FIFOA6, "TCX_FIFOA6"},
3527 	{AMDGPU_GFX_TCX_FIFOA7, "TCX_FIFOA7"},
3528 	{AMDGPU_GFX_TCX_CFIFO0, "TCX_CFIFO0"},
3529 	{AMDGPU_GFX_TCX_CFIFO1, "TCX_CFIFO1"},
3530 	{AMDGPU_GFX_TCX_CFIFO2, "TCX_CFIFO2"},
3531 	{AMDGPU_GFX_TCX_CFIFO3, "TCX_CFIFO3"},
3532 	{AMDGPU_GFX_TCX_CFIFO4, "TCX_CFIFO4"},
3533 	{AMDGPU_GFX_TCX_CFIFO5, "TCX_CFIFO5"},
3534 	{AMDGPU_GFX_TCX_CFIFO6, "TCX_CFIFO6"},
3535 	{AMDGPU_GFX_TCX_CFIFO7, "TCX_CFIFO7"},
3536 	{AMDGPU_GFX_TCX_FIFO_ACKB0, "TCX_FIFO_ACKB0"},
3537 	{AMDGPU_GFX_TCX_FIFO_ACKB1, "TCX_FIFO_ACKB1"},
3538 	{AMDGPU_GFX_TCX_FIFO_ACKB2, "TCX_FIFO_ACKB2"},
3539 	{AMDGPU_GFX_TCX_FIFO_ACKB3, "TCX_FIFO_ACKB3"},
3540 	{AMDGPU_GFX_TCX_FIFO_ACKB4, "TCX_FIFO_ACKB4"},
3541 	{AMDGPU_GFX_TCX_FIFO_ACKB5, "TCX_FIFO_ACKB5"},
3542 	{AMDGPU_GFX_TCX_FIFO_ACKB6, "TCX_FIFO_ACKB6"},
3543 	{AMDGPU_GFX_TCX_FIFO_ACKB7, "TCX_FIFO_ACKB7"},
3544 	{AMDGPU_GFX_TCX_FIFO_ACKD0, "TCX_FIFO_ACKD0"},
3545 	{AMDGPU_GFX_TCX_FIFO_ACKD1, "TCX_FIFO_ACKD1"},
3546 	{AMDGPU_GFX_TCX_FIFO_ACKD2, "TCX_FIFO_ACKD2"},
3547 	{AMDGPU_GFX_TCX_FIFO_ACKD3, "TCX_FIFO_ACKD3"},
3548 	{AMDGPU_GFX_TCX_FIFO_ACKD4, "TCX_FIFO_ACKD4"},
3549 	{AMDGPU_GFX_TCX_FIFO_ACKD5, "TCX_FIFO_ACKD5"},
3550 	{AMDGPU_GFX_TCX_FIFO_ACKD6, "TCX_FIFO_ACKD6"},
3551 	{AMDGPU_GFX_TCX_FIFO_ACKD7, "TCX_FIFO_ACKD7"},
3552 	{AMDGPU_GFX_TCX_DST_FIFOA0, "TCX_DST_FIFOA0"},
3553 	{AMDGPU_GFX_TCX_DST_FIFOA1, "TCX_DST_FIFOA1"},
3554 	{AMDGPU_GFX_TCX_DST_FIFOA2, "TCX_DST_FIFOA2"},
3555 	{AMDGPU_GFX_TCX_DST_FIFOA3, "TCX_DST_FIFOA3"},
3556 	{AMDGPU_GFX_TCX_DST_FIFOA4, "TCX_DST_FIFOA4"},
3557 	{AMDGPU_GFX_TCX_DST_FIFOA5, "TCX_DST_FIFOA5"},
3558 	{AMDGPU_GFX_TCX_DST_FIFOA6, "TCX_DST_FIFOA6"},
3559 	{AMDGPU_GFX_TCX_DST_FIFOA7, "TCX_DST_FIFOA7"},
3560 	{AMDGPU_GFX_TCX_DST_FIFOB0, "TCX_DST_FIFOB0"},
3561 	{AMDGPU_GFX_TCX_DST_FIFOB1, "TCX_DST_FIFOB1"},
3562 	{AMDGPU_GFX_TCX_DST_FIFOB2, "TCX_DST_FIFOB2"},
3563 	{AMDGPU_GFX_TCX_DST_FIFOB3, "TCX_DST_FIFOB3"},
3564 	{AMDGPU_GFX_TCX_DST_FIFOB4, "TCX_DST_FIFOB4"},
3565 	{AMDGPU_GFX_TCX_DST_FIFOB5, "TCX_DST_FIFOB5"},
3566 	{AMDGPU_GFX_TCX_DST_FIFOB6, "TCX_DST_FIFOB6"},
3567 	{AMDGPU_GFX_TCX_DST_FIFOB7, "TCX_DST_FIFOB7"},
3568 	{AMDGPU_GFX_TCX_DST_FIFOD0, "TCX_DST_FIFOD0"},
3569 	{AMDGPU_GFX_TCX_DST_FIFOD1, "TCX_DST_FIFOD1"},
3570 	{AMDGPU_GFX_TCX_DST_FIFOD2, "TCX_DST_FIFOD2"},
3571 	{AMDGPU_GFX_TCX_DST_FIFOD3, "TCX_DST_FIFOD3"},
3572 	{AMDGPU_GFX_TCX_DST_FIFOD4, "TCX_DST_FIFOD4"},
3573 	{AMDGPU_GFX_TCX_DST_FIFOD5, "TCX_DST_FIFOD5"},
3574 	{AMDGPU_GFX_TCX_DST_FIFOD6, "TCX_DST_FIFOD6"},
3575 	{AMDGPU_GFX_TCX_DST_FIFOD7, "TCX_DST_FIFOD7"},
3576 	{AMDGPU_GFX_TCX_DST_FIFO_ACKB0, "TCX_DST_FIFO_ACKB0"},
3577 	{AMDGPU_GFX_TCX_DST_FIFO_ACKB1, "TCX_DST_FIFO_ACKB1"},
3578 	{AMDGPU_GFX_TCX_DST_FIFO_ACKB2, "TCX_DST_FIFO_ACKB2"},
3579 	{AMDGPU_GFX_TCX_DST_FIFO_ACKB3, "TCX_DST_FIFO_ACKB3"},
3580 	{AMDGPU_GFX_TCX_DST_FIFO_ACKB4, "TCX_DST_FIFO_ACKB4"},
3581 	{AMDGPU_GFX_TCX_DST_FIFO_ACKB5, "TCX_DST_FIFO_ACKB5"},
3582 	{AMDGPU_GFX_TCX_DST_FIFO_ACKB6, "TCX_DST_FIFO_ACKB6"},
3583 	{AMDGPU_GFX_TCX_DST_FIFO_ACKB7, "TCX_DST_FIFO_ACKB7"},
3584 	{AMDGPU_GFX_TCX_DST_FIFO_ACKD0, "TCX_DST_FIFO_ACKD0"},
3585 	{AMDGPU_GFX_TCX_DST_FIFO_ACKD1, "TCX_DST_FIFO_ACKD1"},
3586 	{AMDGPU_GFX_TCX_DST_FIFO_ACKD2, "TCX_DST_FIFO_ACKD2"},
3587 	{AMDGPU_GFX_TCX_DST_FIFO_ACKD3, "TCX_DST_FIFO_ACKD3"},
3588 	{AMDGPU_GFX_TCX_DST_FIFO_ACKD4, "TCX_DST_FIFO_ACKD4"},
3589 	{AMDGPU_GFX_TCX_DST_FIFO_ACKD5, "TCX_DST_FIFO_ACKD5"},
3590 	{AMDGPU_GFX_TCX_DST_FIFO_ACKD6, "TCX_DST_FIFO_ACKD6"},
3591 	{AMDGPU_GFX_TCX_DST_FIFO_ACKD7, "TCX_DST_FIFO_ACKD7"},
3592 };
3593 
3594 static const struct amdgpu_ras_memory_id_entry gfx_v9_4_3_ras_atc_l2_mem_list[] = {
3595 	{AMDGPU_GFX_ATC_L2_MEM, "ATC_L2_MEM"},
3596 };
3597 
3598 static const struct amdgpu_ras_memory_id_entry gfx_v9_4_3_ras_utcl2_mem_list[] = {
3599 	{AMDGPU_GFX_UTCL2_MEM, "UTCL2_MEM"},
3600 };
3601 
3602 static const struct amdgpu_ras_memory_id_entry gfx_v9_4_3_ras_vml2_mem_list[] = {
3603 	{AMDGPU_GFX_VML2_MEM, "VML2_MEM"},
3604 };
3605 
3606 static const struct amdgpu_ras_memory_id_entry gfx_v9_4_3_ras_vml2_walker_mem_list[] = {
3607 	{AMDGPU_GFX_VML2_WALKER_MEM, "VML2_WALKER_MEM"},
3608 };
3609 
3610 static const struct amdgpu_gfx_ras_mem_id_entry gfx_v9_4_3_ras_mem_list_array[AMDGPU_GFX_MEM_TYPE_NUM] = {
3611 	AMDGPU_GFX_MEMID_ENT(gfx_v9_4_3_ras_cp_mem_list)
3612 	AMDGPU_GFX_MEMID_ENT(gfx_v9_4_3_ras_gcea_mem_list)
3613 	AMDGPU_GFX_MEMID_ENT(gfx_v9_4_3_ras_gc_cane_mem_list)
3614 	AMDGPU_GFX_MEMID_ENT(gfx_v9_4_3_ras_gcutcl2_mem_list)
3615 	AMDGPU_GFX_MEMID_ENT(gfx_v9_4_3_ras_gds_mem_list)
3616 	AMDGPU_GFX_MEMID_ENT(gfx_v9_4_3_ras_lds_mem_list)
3617 	AMDGPU_GFX_MEMID_ENT(gfx_v9_4_3_ras_rlc_mem_list)
3618 	AMDGPU_GFX_MEMID_ENT(gfx_v9_4_3_ras_sp_mem_list)
3619 	AMDGPU_GFX_MEMID_ENT(gfx_v9_4_3_ras_spi_mem_list)
3620 	AMDGPU_GFX_MEMID_ENT(gfx_v9_4_3_ras_sqc_mem_list)
3621 	AMDGPU_GFX_MEMID_ENT(gfx_v9_4_3_ras_sq_mem_list)
3622 	AMDGPU_GFX_MEMID_ENT(gfx_v9_4_3_ras_ta_mem_list)
3623 	AMDGPU_GFX_MEMID_ENT(gfx_v9_4_3_ras_tcc_mem_list)
3624 	AMDGPU_GFX_MEMID_ENT(gfx_v9_4_3_ras_tca_mem_list)
3625 	AMDGPU_GFX_MEMID_ENT(gfx_v9_4_3_ras_tci_mem_list)
3626 	AMDGPU_GFX_MEMID_ENT(gfx_v9_4_3_ras_tcp_mem_list)
3627 	AMDGPU_GFX_MEMID_ENT(gfx_v9_4_3_ras_td_mem_list)
3628 	AMDGPU_GFX_MEMID_ENT(gfx_v9_4_3_ras_tcx_mem_list)
3629 	AMDGPU_GFX_MEMID_ENT(gfx_v9_4_3_ras_atc_l2_mem_list)
3630 	AMDGPU_GFX_MEMID_ENT(gfx_v9_4_3_ras_utcl2_mem_list)
3631 	AMDGPU_GFX_MEMID_ENT(gfx_v9_4_3_ras_vml2_mem_list)
3632 	AMDGPU_GFX_MEMID_ENT(gfx_v9_4_3_ras_vml2_walker_mem_list)
3633 };
3634 
3635 static const struct amdgpu_gfx_ras_reg_entry gfx_v9_4_3_ce_reg_list[] = {
3636 	{{AMDGPU_RAS_REG_ENTRY(GC, 0, regRLC_CE_ERR_STATUS_LOW, regRLC_CE_ERR_STATUS_HIGH),
3637 	    1, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "RLC"},
3638 	    AMDGPU_GFX_RLC_MEM, 1},
3639 	{{AMDGPU_RAS_REG_ENTRY(GC, 0, regCPC_CE_ERR_STATUS_LO, regCPC_CE_ERR_STATUS_HI),
3640 	    1, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "CPC"},
3641 	    AMDGPU_GFX_CP_MEM, 1},
3642 	{{AMDGPU_RAS_REG_ENTRY(GC, 0, regCPF_CE_ERR_STATUS_LO, regCPF_CE_ERR_STATUS_HI),
3643 	    1, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "CPF"},
3644 	    AMDGPU_GFX_CP_MEM, 1},
3645 	{{AMDGPU_RAS_REG_ENTRY(GC, 0, regCPG_CE_ERR_STATUS_LO, regCPG_CE_ERR_STATUS_HI),
3646 	    1, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "CPG"},
3647 	    AMDGPU_GFX_CP_MEM, 1},
3648 	{{AMDGPU_RAS_REG_ENTRY(GC, 0, regGDS_CE_ERR_STATUS_LO, regGDS_CE_ERR_STATUS_HI),
3649 	    1, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "GDS"},
3650 	    AMDGPU_GFX_GDS_MEM, 1},
3651 	{{AMDGPU_RAS_REG_ENTRY(GC, 0, regGC_CANE_CE_ERR_STATUS_LO, regGC_CANE_CE_ERR_STATUS_HI),
3652 	    1, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "CANE"},
3653 	    AMDGPU_GFX_GC_CANE_MEM, 1},
3654 	{{AMDGPU_RAS_REG_ENTRY(GC, 0, regSPI_CE_ERR_STATUS_LO, regSPI_CE_ERR_STATUS_HI),
3655 	    1, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "SPI"},
3656 	    AMDGPU_GFX_SPI_MEM, 8},
3657 	{{AMDGPU_RAS_REG_ENTRY(GC, 0, regSP0_CE_ERR_STATUS_LO, regSP0_CE_ERR_STATUS_HI),
3658 	    10, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "SP0"},
3659 	    AMDGPU_GFX_SP_MEM, 1},
3660 	{{AMDGPU_RAS_REG_ENTRY(GC, 0, regSP1_CE_ERR_STATUS_LO, regSP1_CE_ERR_STATUS_HI),
3661 	    10, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "SP1"},
3662 	    AMDGPU_GFX_SP_MEM, 1},
3663 	{{AMDGPU_RAS_REG_ENTRY(GC, 0, regSQ_CE_ERR_STATUS_LO, regSQ_CE_ERR_STATUS_HI),
3664 	    10, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "SQ"},
3665 	    AMDGPU_GFX_SQ_MEM, 8},
3666 	{{AMDGPU_RAS_REG_ENTRY(GC, 0, regSQC_CE_EDC_LO, regSQC_CE_EDC_HI),
3667 	    5, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "SQC"},
3668 	    AMDGPU_GFX_SQC_MEM, 8},
3669 	{{AMDGPU_RAS_REG_ENTRY(GC, 0, regTCX_CE_ERR_STATUS_LO, regTCX_CE_ERR_STATUS_HI),
3670 	    2, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "TCX"},
3671 	    AMDGPU_GFX_TCX_MEM, 1},
3672 	{{AMDGPU_RAS_REG_ENTRY(GC, 0, regTCC_CE_ERR_STATUS_LO, regTCC_CE_ERR_STATUS_HI),
3673 	    16, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "TCC"},
3674 	    AMDGPU_GFX_TCC_MEM, 1},
3675 	{{AMDGPU_RAS_REG_ENTRY(GC, 0, regTA_CE_EDC_LO, regTA_CE_EDC_HI),
3676 	    10, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "TA"},
3677 	    AMDGPU_GFX_TA_MEM, 8},
3678 	{{AMDGPU_RAS_REG_ENTRY(GC, 0, regTCI_CE_EDC_LO_REG, regTCI_CE_EDC_HI_REG),
3679 	    31, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "TCI"},
3680 	    AMDGPU_GFX_TCI_MEM, 1},
3681 	{{AMDGPU_RAS_REG_ENTRY(GC, 0, regTCP_CE_EDC_LO_REG, regTCP_CE_EDC_HI_REG),
3682 	    10, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "TCP"},
3683 	    AMDGPU_GFX_TCP_MEM, 8},
3684 	{{AMDGPU_RAS_REG_ENTRY(GC, 0, regTD_CE_EDC_LO, regTD_CE_EDC_HI),
3685 	    10, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "TD"},
3686 	    AMDGPU_GFX_TD_MEM, 8},
3687 	{{AMDGPU_RAS_REG_ENTRY(GC, 0, regGCEA_CE_ERR_STATUS_LO, regGCEA_CE_ERR_STATUS_HI),
3688 	    16, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "GCEA"},
3689 	    AMDGPU_GFX_GCEA_MEM, 1},
3690 	{{AMDGPU_RAS_REG_ENTRY(GC, 0, regLDS_CE_ERR_STATUS_LO, regLDS_CE_ERR_STATUS_HI),
3691 	    10, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "LDS"},
3692 	    AMDGPU_GFX_LDS_MEM, 1},
3693 };
3694 
3695 static const struct amdgpu_gfx_ras_reg_entry gfx_v9_4_3_ue_reg_list[] = {
3696 	{{AMDGPU_RAS_REG_ENTRY(GC, 0, regRLC_UE_ERR_STATUS_LOW, regRLC_UE_ERR_STATUS_HIGH),
3697 	    1, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "RLC"},
3698 	    AMDGPU_GFX_RLC_MEM, 1},
3699 	{{AMDGPU_RAS_REG_ENTRY(GC, 0, regCPC_UE_ERR_STATUS_LO, regCPC_UE_ERR_STATUS_HI),
3700 	    1, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "CPC"},
3701 	    AMDGPU_GFX_CP_MEM, 1},
3702 	{{AMDGPU_RAS_REG_ENTRY(GC, 0, regCPF_UE_ERR_STATUS_LO, regCPF_UE_ERR_STATUS_HI),
3703 	    1, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "CPF"},
3704 	    AMDGPU_GFX_CP_MEM, 1},
3705 	{{AMDGPU_RAS_REG_ENTRY(GC, 0, regCPG_UE_ERR_STATUS_LO, regCPG_UE_ERR_STATUS_HI),
3706 	    1, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "CPG"},
3707 	    AMDGPU_GFX_CP_MEM, 1},
3708 	{{AMDGPU_RAS_REG_ENTRY(GC, 0, regGDS_UE_ERR_STATUS_LO, regGDS_UE_ERR_STATUS_HI),
3709 	    1, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "GDS"},
3710 	    AMDGPU_GFX_GDS_MEM, 1},
3711 	{{AMDGPU_RAS_REG_ENTRY(GC, 0, regGC_CANE_UE_ERR_STATUS_LO, regGC_CANE_UE_ERR_STATUS_HI),
3712 	    1, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "CANE"},
3713 	    AMDGPU_GFX_GC_CANE_MEM, 1},
3714 	{{AMDGPU_RAS_REG_ENTRY(GC, 0, regSPI_UE_ERR_STATUS_LO, regSPI_UE_ERR_STATUS_HI),
3715 	    1, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "SPI"},
3716 	    AMDGPU_GFX_SPI_MEM, 8},
3717 	{{AMDGPU_RAS_REG_ENTRY(GC, 0, regSP0_UE_ERR_STATUS_LO, regSP0_UE_ERR_STATUS_HI),
3718 	    10, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "SP0"},
3719 	    AMDGPU_GFX_SP_MEM, 1},
3720 	{{AMDGPU_RAS_REG_ENTRY(GC, 0, regSP1_UE_ERR_STATUS_LO, regSP1_UE_ERR_STATUS_HI),
3721 	    10, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "SP1"},
3722 	    AMDGPU_GFX_SP_MEM, 1},
3723 	{{AMDGPU_RAS_REG_ENTRY(GC, 0, regSQ_UE_ERR_STATUS_LO, regSQ_UE_ERR_STATUS_HI),
3724 	    10, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "SQ"},
3725 	    AMDGPU_GFX_SQ_MEM, 8},
3726 	{{AMDGPU_RAS_REG_ENTRY(GC, 0, regSQC_UE_EDC_LO, regSQC_UE_EDC_HI),
3727 	    5, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "SQC"},
3728 	    AMDGPU_GFX_SQC_MEM, 8},
3729 	{{AMDGPU_RAS_REG_ENTRY(GC, 0, regTCX_UE_ERR_STATUS_LO, regTCX_UE_ERR_STATUS_HI),
3730 	    2, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "TCX"},
3731 	    AMDGPU_GFX_TCX_MEM, 1},
3732 	{{AMDGPU_RAS_REG_ENTRY(GC, 0, regTCC_UE_ERR_STATUS_LO, regTCC_UE_ERR_STATUS_HI),
3733 	    16, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "TCC"},
3734 	    AMDGPU_GFX_TCC_MEM, 1},
3735 	{{AMDGPU_RAS_REG_ENTRY(GC, 0, regTA_UE_EDC_LO, regTA_UE_EDC_HI),
3736 	    10, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "TA"},
3737 	    AMDGPU_GFX_TA_MEM, 8},
3738 	{{AMDGPU_RAS_REG_ENTRY(GC, 0, regTCI_UE_EDC_LO_REG, regTCI_UE_EDC_HI_REG),
3739 	    31, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "TCI"},
3740 	    AMDGPU_GFX_TCI_MEM, 1},
3741 	{{AMDGPU_RAS_REG_ENTRY(GC, 0, regTCP_UE_EDC_LO_REG, regTCP_UE_EDC_HI_REG),
3742 	    10, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "TCP"},
3743 	    AMDGPU_GFX_TCP_MEM, 8},
3744 	{{AMDGPU_RAS_REG_ENTRY(GC, 0, regTD_UE_EDC_LO, regTD_UE_EDC_HI),
3745 	    10, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "TD"},
3746 	    AMDGPU_GFX_TD_MEM, 8},
3747 	{{AMDGPU_RAS_REG_ENTRY(GC, 0, regTCA_UE_ERR_STATUS_LO, regTCA_UE_ERR_STATUS_HI),
3748 	    2, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "TCA"},
3749 	    AMDGPU_GFX_TCA_MEM, 1},
3750 	{{AMDGPU_RAS_REG_ENTRY(GC, 0, regGCEA_UE_ERR_STATUS_LO, regGCEA_UE_ERR_STATUS_HI),
3751 	    16, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "GCEA"},
3752 	    AMDGPU_GFX_GCEA_MEM, 1},
3753 	{{AMDGPU_RAS_REG_ENTRY(GC, 0, regLDS_UE_ERR_STATUS_LO, regLDS_UE_ERR_STATUS_HI),
3754 	    10, (AMDGPU_RAS_ERR_INFO_VALID | AMDGPU_RAS_ERR_STATUS_VALID), "LDS"},
3755 	    AMDGPU_GFX_LDS_MEM, 1},
3756 };
3757 
3758 static const struct soc15_reg_entry gfx_v9_4_3_ea_err_status_regs = {
3759 	SOC15_REG_ENTRY(GC, 0, regGCEA_ERR_STATUS), 0, 1, 16
3760 };
3761 
gfx_v9_4_3_inst_query_ras_err_count(struct amdgpu_device * adev,void * ras_error_status,int xcc_id)3762 static void gfx_v9_4_3_inst_query_ras_err_count(struct amdgpu_device *adev,
3763 					void *ras_error_status, int xcc_id)
3764 {
3765 	struct ras_err_data *err_data = (struct ras_err_data *)ras_error_status;
3766 	unsigned long ce_count = 0, ue_count = 0;
3767 	uint32_t i, j, k;
3768 
3769 	mutex_lock(&adev->grbm_idx_mutex);
3770 
3771 	for (i = 0; i < ARRAY_SIZE(gfx_v9_4_3_ce_reg_list); i++) {
3772 		for (j = 0; j < gfx_v9_4_3_ce_reg_list[i].se_num; j++) {
3773 			for (k = 0; k < gfx_v9_4_3_ce_reg_list[i].reg_entry.reg_inst; k++) {
3774 				/* no need to select if instance number is 1 */
3775 				if (gfx_v9_4_3_ce_reg_list[i].se_num > 1 ||
3776 				    gfx_v9_4_3_ce_reg_list[i].reg_entry.reg_inst > 1)
3777 					gfx_v9_4_3_xcc_select_se_sh(adev, j, 0, k, xcc_id);
3778 
3779 				amdgpu_ras_inst_query_ras_error_count(adev,
3780 					&(gfx_v9_4_3_ce_reg_list[i].reg_entry),
3781 					1,
3782 					gfx_v9_4_3_ras_mem_list_array[gfx_v9_4_3_ce_reg_list[i].mem_id_type].mem_id_ent,
3783 					gfx_v9_4_3_ras_mem_list_array[gfx_v9_4_3_ce_reg_list[i].mem_id_type].size,
3784 					GET_INST(GC, xcc_id),
3785 					AMDGPU_RAS_ERROR__SINGLE_CORRECTABLE,
3786 					&ce_count);
3787 
3788 				amdgpu_ras_inst_query_ras_error_count(adev,
3789 					&(gfx_v9_4_3_ue_reg_list[i].reg_entry),
3790 					1,
3791 					gfx_v9_4_3_ras_mem_list_array[gfx_v9_4_3_ue_reg_list[i].mem_id_type].mem_id_ent,
3792 					gfx_v9_4_3_ras_mem_list_array[gfx_v9_4_3_ue_reg_list[i].mem_id_type].size,
3793 					GET_INST(GC, xcc_id),
3794 					AMDGPU_RAS_ERROR__MULTI_UNCORRECTABLE,
3795 					&ue_count);
3796 			}
3797 		}
3798 	}
3799 
3800 	gfx_v9_4_3_xcc_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff,
3801 			xcc_id);
3802 	mutex_unlock(&adev->grbm_idx_mutex);
3803 
3804 	/* the caller should make sure initialize value of
3805 	 * err_data->ue_count and err_data->ce_count
3806 	 */
3807 	err_data->ce_count += ce_count;
3808 	err_data->ue_count += ue_count;
3809 }
3810 
gfx_v9_4_3_inst_reset_ras_err_count(struct amdgpu_device * adev,void * ras_error_status,int xcc_id)3811 static void gfx_v9_4_3_inst_reset_ras_err_count(struct amdgpu_device *adev,
3812 					void *ras_error_status, int xcc_id)
3813 {
3814 	uint32_t i, j, k;
3815 
3816 	mutex_lock(&adev->grbm_idx_mutex);
3817 
3818 	for (i = 0; i < ARRAY_SIZE(gfx_v9_4_3_ce_reg_list); i++) {
3819 		for (j = 0; j < gfx_v9_4_3_ce_reg_list[i].se_num; j++) {
3820 			for (k = 0; k < gfx_v9_4_3_ce_reg_list[i].reg_entry.reg_inst; k++) {
3821 				/* no need to select if instance number is 1 */
3822 				if (gfx_v9_4_3_ce_reg_list[i].se_num > 1 ||
3823 				    gfx_v9_4_3_ce_reg_list[i].reg_entry.reg_inst > 1)
3824 					gfx_v9_4_3_xcc_select_se_sh(adev, j, 0, k, xcc_id);
3825 
3826 				amdgpu_ras_inst_reset_ras_error_count(adev,
3827 					&(gfx_v9_4_3_ce_reg_list[i].reg_entry),
3828 					1,
3829 					GET_INST(GC, xcc_id));
3830 
3831 				amdgpu_ras_inst_reset_ras_error_count(adev,
3832 					&(gfx_v9_4_3_ue_reg_list[i].reg_entry),
3833 					1,
3834 					GET_INST(GC, xcc_id));
3835 			}
3836 		}
3837 	}
3838 
3839 	gfx_v9_4_3_xcc_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff,
3840 			xcc_id);
3841 	mutex_unlock(&adev->grbm_idx_mutex);
3842 }
3843 
gfx_v9_4_3_inst_query_ea_err_status(struct amdgpu_device * adev,int xcc_id)3844 static void gfx_v9_4_3_inst_query_ea_err_status(struct amdgpu_device *adev,
3845 					int xcc_id)
3846 {
3847 	uint32_t i, j;
3848 	uint32_t reg_value;
3849 
3850 	mutex_lock(&adev->grbm_idx_mutex);
3851 
3852 	for (i = 0; i < gfx_v9_4_3_ea_err_status_regs.se_num; i++) {
3853 		for (j = 0; j < gfx_v9_4_3_ea_err_status_regs.instance; j++) {
3854 			gfx_v9_4_3_xcc_select_se_sh(adev, i, 0, j, xcc_id);
3855 			reg_value = RREG32_SOC15(GC, GET_INST(GC, xcc_id),
3856 					regGCEA_ERR_STATUS);
3857 			if (REG_GET_FIELD(reg_value, GCEA_ERR_STATUS, SDP_RDRSP_STATUS) ||
3858 			    REG_GET_FIELD(reg_value, GCEA_ERR_STATUS, SDP_WRRSP_STATUS) ||
3859 			    REG_GET_FIELD(reg_value, GCEA_ERR_STATUS, SDP_RDRSP_DATAPARITY_ERROR)) {
3860 				dev_warn(adev->dev,
3861 					"GCEA err detected at instance: %d, status: 0x%x!\n",
3862 					j, reg_value);
3863 			}
3864 			/* clear after read */
3865 			reg_value = REG_SET_FIELD(reg_value, GCEA_ERR_STATUS,
3866 						  CLEAR_ERROR_STATUS, 0x1);
3867 			WREG32_SOC15(GC, GET_INST(GC, xcc_id), regGCEA_ERR_STATUS,
3868 					reg_value);
3869 		}
3870 	}
3871 
3872 	gfx_v9_4_3_xcc_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff,
3873 			xcc_id);
3874 	mutex_unlock(&adev->grbm_idx_mutex);
3875 }
3876 
gfx_v9_4_3_inst_query_utc_err_status(struct amdgpu_device * adev,int xcc_id)3877 static void gfx_v9_4_3_inst_query_utc_err_status(struct amdgpu_device *adev,
3878 					int xcc_id)
3879 {
3880 	uint32_t data;
3881 
3882 	data = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regUTCL2_MEM_ECC_STATUS);
3883 	if (data) {
3884 		dev_warn(adev->dev, "GFX UTCL2 Mem Ecc Status: 0x%x!\n", data);
3885 		WREG32_SOC15(GC, GET_INST(GC, xcc_id), regUTCL2_MEM_ECC_STATUS, 0x3);
3886 	}
3887 
3888 	data = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regVML2_MEM_ECC_STATUS);
3889 	if (data) {
3890 		dev_warn(adev->dev, "GFX VML2 Mem Ecc Status: 0x%x!\n", data);
3891 		WREG32_SOC15(GC, GET_INST(GC, xcc_id), regVML2_MEM_ECC_STATUS, 0x3);
3892 	}
3893 
3894 	data = RREG32_SOC15(GC, GET_INST(GC, xcc_id),
3895 				regVML2_WALKER_MEM_ECC_STATUS);
3896 	if (data) {
3897 		dev_warn(adev->dev, "GFX VML2 Walker Mem Ecc Status: 0x%x!\n", data);
3898 		WREG32_SOC15(GC, GET_INST(GC, xcc_id), regVML2_WALKER_MEM_ECC_STATUS,
3899 				0x3);
3900 	}
3901 }
3902 
gfx_v9_4_3_log_cu_timeout_status(struct amdgpu_device * adev,uint32_t status,int xcc_id)3903 static void gfx_v9_4_3_log_cu_timeout_status(struct amdgpu_device *adev,
3904 					uint32_t status, int xcc_id)
3905 {
3906 	struct amdgpu_cu_info *cu_info = &adev->gfx.cu_info;
3907 	uint32_t i, simd, wave;
3908 	uint32_t wave_status;
3909 	uint32_t wave_pc_lo, wave_pc_hi;
3910 	uint32_t wave_exec_lo, wave_exec_hi;
3911 	uint32_t wave_inst_dw0, wave_inst_dw1;
3912 	uint32_t wave_ib_sts;
3913 
3914 	for (i = 0; i < 32; i++) {
3915 		if (!((i << 1) & status))
3916 			continue;
3917 
3918 		simd = i / cu_info->max_waves_per_simd;
3919 		wave = i % cu_info->max_waves_per_simd;
3920 
3921 		wave_status = wave_read_ind(adev, xcc_id, simd, wave, ixSQ_WAVE_STATUS);
3922 		wave_pc_lo = wave_read_ind(adev, xcc_id, simd, wave, ixSQ_WAVE_PC_LO);
3923 		wave_pc_hi = wave_read_ind(adev, xcc_id, simd, wave, ixSQ_WAVE_PC_HI);
3924 		wave_exec_lo =
3925 			wave_read_ind(adev, xcc_id, simd, wave, ixSQ_WAVE_EXEC_LO);
3926 		wave_exec_hi =
3927 			wave_read_ind(adev, xcc_id, simd, wave, ixSQ_WAVE_EXEC_HI);
3928 		wave_inst_dw0 =
3929 			wave_read_ind(adev, xcc_id, simd, wave, ixSQ_WAVE_INST_DW0);
3930 		wave_inst_dw1 =
3931 			wave_read_ind(adev, xcc_id, simd, wave, ixSQ_WAVE_INST_DW1);
3932 		wave_ib_sts = wave_read_ind(adev, xcc_id, simd, wave, ixSQ_WAVE_IB_STS);
3933 
3934 		dev_info(
3935 			adev->dev,
3936 			"\t SIMD %d, Wave %d: status 0x%x, pc 0x%llx, exec 0x%llx, inst 0x%llx, ib_sts 0x%x\n",
3937 			simd, wave, wave_status,
3938 			((uint64_t)wave_pc_hi << 32 | wave_pc_lo),
3939 			((uint64_t)wave_exec_hi << 32 | wave_exec_lo),
3940 			((uint64_t)wave_inst_dw1 << 32 | wave_inst_dw0),
3941 			wave_ib_sts);
3942 	}
3943 }
3944 
gfx_v9_4_3_inst_query_sq_timeout_status(struct amdgpu_device * adev,int xcc_id)3945 static void gfx_v9_4_3_inst_query_sq_timeout_status(struct amdgpu_device *adev,
3946 					int xcc_id)
3947 {
3948 	uint32_t se_idx, sh_idx, cu_idx;
3949 	uint32_t status;
3950 
3951 	mutex_lock(&adev->grbm_idx_mutex);
3952 	for (se_idx = 0; se_idx < adev->gfx.config.max_shader_engines; se_idx++) {
3953 		for (sh_idx = 0; sh_idx < adev->gfx.config.max_sh_per_se; sh_idx++) {
3954 			for (cu_idx = 0; cu_idx < adev->gfx.config.max_cu_per_sh; cu_idx++) {
3955 				gfx_v9_4_3_xcc_select_se_sh(adev, se_idx, sh_idx,
3956 							cu_idx, xcc_id);
3957 				status = RREG32_SOC15(GC, GET_INST(GC, xcc_id),
3958 						      regSQ_TIMEOUT_STATUS);
3959 				if (status != 0) {
3960 					dev_info(
3961 						adev->dev,
3962 						"GFX Watchdog Timeout: SE %d, SH %d, CU %d\n",
3963 						se_idx, sh_idx, cu_idx);
3964 					gfx_v9_4_3_log_cu_timeout_status(
3965 						adev, status, xcc_id);
3966 				}
3967 				/* clear old status */
3968 				WREG32_SOC15(GC, GET_INST(GC, xcc_id),
3969 						regSQ_TIMEOUT_STATUS, 0);
3970 			}
3971 		}
3972 	}
3973 	gfx_v9_4_3_xcc_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff,
3974 			xcc_id);
3975 	mutex_unlock(&adev->grbm_idx_mutex);
3976 }
3977 
gfx_v9_4_3_inst_query_ras_err_status(struct amdgpu_device * adev,void * ras_error_status,int xcc_id)3978 static void gfx_v9_4_3_inst_query_ras_err_status(struct amdgpu_device *adev,
3979 					void *ras_error_status, int xcc_id)
3980 {
3981 	gfx_v9_4_3_inst_query_ea_err_status(adev, xcc_id);
3982 	gfx_v9_4_3_inst_query_utc_err_status(adev, xcc_id);
3983 	gfx_v9_4_3_inst_query_sq_timeout_status(adev, xcc_id);
3984 }
3985 
gfx_v9_4_3_inst_reset_utc_err_status(struct amdgpu_device * adev,int xcc_id)3986 static void gfx_v9_4_3_inst_reset_utc_err_status(struct amdgpu_device *adev,
3987 					int xcc_id)
3988 {
3989 	WREG32_SOC15(GC, GET_INST(GC, xcc_id), regUTCL2_MEM_ECC_STATUS, 0x3);
3990 	WREG32_SOC15(GC, GET_INST(GC, xcc_id), regVML2_MEM_ECC_STATUS, 0x3);
3991 	WREG32_SOC15(GC, GET_INST(GC, xcc_id), regVML2_WALKER_MEM_ECC_STATUS, 0x3);
3992 }
3993 
gfx_v9_4_3_inst_reset_ea_err_status(struct amdgpu_device * adev,int xcc_id)3994 static void gfx_v9_4_3_inst_reset_ea_err_status(struct amdgpu_device *adev,
3995 					int xcc_id)
3996 {
3997 	uint32_t i, j;
3998 	uint32_t value;
3999 
4000 	mutex_lock(&adev->grbm_idx_mutex);
4001 	for (i = 0; i < gfx_v9_4_3_ea_err_status_regs.se_num; i++) {
4002 		for (j = 0; j < gfx_v9_4_3_ea_err_status_regs.instance; j++) {
4003 			gfx_v9_4_3_xcc_select_se_sh(adev, i, 0, j, xcc_id);
4004 			value = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regGCEA_ERR_STATUS);
4005 			value = REG_SET_FIELD(value, GCEA_ERR_STATUS,
4006 						CLEAR_ERROR_STATUS, 0x1);
4007 			WREG32_SOC15(GC, GET_INST(GC, xcc_id), regGCEA_ERR_STATUS, value);
4008 		}
4009 	}
4010 	gfx_v9_4_3_xcc_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff,
4011 			xcc_id);
4012 	mutex_unlock(&adev->grbm_idx_mutex);
4013 }
4014 
gfx_v9_4_3_inst_reset_sq_timeout_status(struct amdgpu_device * adev,int xcc_id)4015 static void gfx_v9_4_3_inst_reset_sq_timeout_status(struct amdgpu_device *adev,
4016 					int xcc_id)
4017 {
4018 	uint32_t se_idx, sh_idx, cu_idx;
4019 
4020 	mutex_lock(&adev->grbm_idx_mutex);
4021 	for (se_idx = 0; se_idx < adev->gfx.config.max_shader_engines; se_idx++) {
4022 		for (sh_idx = 0; sh_idx < adev->gfx.config.max_sh_per_se; sh_idx++) {
4023 			for (cu_idx = 0; cu_idx < adev->gfx.config.max_cu_per_sh; cu_idx++) {
4024 				gfx_v9_4_3_xcc_select_se_sh(adev, se_idx, sh_idx,
4025 							cu_idx, xcc_id);
4026 				WREG32_SOC15(GC, GET_INST(GC, xcc_id),
4027 						regSQ_TIMEOUT_STATUS, 0);
4028 			}
4029 		}
4030 	}
4031 	gfx_v9_4_3_xcc_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff,
4032 			xcc_id);
4033 	mutex_unlock(&adev->grbm_idx_mutex);
4034 }
4035 
gfx_v9_4_3_inst_reset_ras_err_status(struct amdgpu_device * adev,void * ras_error_status,int xcc_id)4036 static void gfx_v9_4_3_inst_reset_ras_err_status(struct amdgpu_device *adev,
4037 					void *ras_error_status, int xcc_id)
4038 {
4039 	gfx_v9_4_3_inst_reset_utc_err_status(adev, xcc_id);
4040 	gfx_v9_4_3_inst_reset_ea_err_status(adev, xcc_id);
4041 	gfx_v9_4_3_inst_reset_sq_timeout_status(adev, xcc_id);
4042 }
4043 
gfx_v9_4_3_inst_enable_watchdog_timer(struct amdgpu_device * adev,void * ras_error_status,int xcc_id)4044 static void gfx_v9_4_3_inst_enable_watchdog_timer(struct amdgpu_device *adev,
4045 					void *ras_error_status, int xcc_id)
4046 {
4047 	uint32_t i;
4048 	uint32_t data;
4049 
4050 	data = RREG32_SOC15(GC, GET_INST(GC, 0), regSQ_TIMEOUT_CONFIG);
4051 	data = REG_SET_FIELD(data, SQ_TIMEOUT_CONFIG, TIMEOUT_FATAL_DISABLE,
4052 			     amdgpu_watchdog_timer.timeout_fatal_disable ? 1 : 0);
4053 
4054 	if (amdgpu_watchdog_timer.timeout_fatal_disable &&
4055 	    (amdgpu_watchdog_timer.period < 1 ||
4056 	     amdgpu_watchdog_timer.period > 0x23)) {
4057 		dev_warn(adev->dev, "Watchdog period range is 1 to 0x23\n");
4058 		amdgpu_watchdog_timer.period = 0x23;
4059 	}
4060 	data = REG_SET_FIELD(data, SQ_TIMEOUT_CONFIG, PERIOD_SEL,
4061 			     amdgpu_watchdog_timer.period);
4062 
4063 	mutex_lock(&adev->grbm_idx_mutex);
4064 	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
4065 		gfx_v9_4_3_xcc_select_se_sh(adev, i, 0xffffffff, 0xffffffff, xcc_id);
4066 		WREG32_SOC15(GC, GET_INST(GC, xcc_id), regSQ_TIMEOUT_CONFIG, data);
4067 	}
4068 	gfx_v9_4_3_xcc_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff,
4069 			xcc_id);
4070 	mutex_unlock(&adev->grbm_idx_mutex);
4071 }
4072 
gfx_v9_4_3_query_ras_error_count(struct amdgpu_device * adev,void * ras_error_status)4073 static void gfx_v9_4_3_query_ras_error_count(struct amdgpu_device *adev,
4074 					void *ras_error_status)
4075 {
4076 	amdgpu_gfx_ras_error_func(adev, ras_error_status,
4077 			gfx_v9_4_3_inst_query_ras_err_count);
4078 }
4079 
gfx_v9_4_3_reset_ras_error_count(struct amdgpu_device * adev)4080 static void gfx_v9_4_3_reset_ras_error_count(struct amdgpu_device *adev)
4081 {
4082 	amdgpu_gfx_ras_error_func(adev, NULL, gfx_v9_4_3_inst_reset_ras_err_count);
4083 }
4084 
gfx_v9_4_3_query_ras_error_status(struct amdgpu_device * adev)4085 static void gfx_v9_4_3_query_ras_error_status(struct amdgpu_device *adev)
4086 {
4087 	amdgpu_gfx_ras_error_func(adev, NULL, gfx_v9_4_3_inst_query_ras_err_status);
4088 }
4089 
gfx_v9_4_3_reset_ras_error_status(struct amdgpu_device * adev)4090 static void gfx_v9_4_3_reset_ras_error_status(struct amdgpu_device *adev)
4091 {
4092 	amdgpu_gfx_ras_error_func(adev, NULL, gfx_v9_4_3_inst_reset_ras_err_status);
4093 }
4094 
gfx_v9_4_3_enable_watchdog_timer(struct amdgpu_device * adev)4095 static void gfx_v9_4_3_enable_watchdog_timer(struct amdgpu_device *adev)
4096 {
4097 	amdgpu_gfx_ras_error_func(adev, NULL, gfx_v9_4_3_inst_enable_watchdog_timer);
4098 }
4099 
4100 static const struct amd_ip_funcs gfx_v9_4_3_ip_funcs = {
4101 	.name = "gfx_v9_4_3",
4102 	.early_init = gfx_v9_4_3_early_init,
4103 	.late_init = gfx_v9_4_3_late_init,
4104 	.sw_init = gfx_v9_4_3_sw_init,
4105 	.sw_fini = gfx_v9_4_3_sw_fini,
4106 	.hw_init = gfx_v9_4_3_hw_init,
4107 	.hw_fini = gfx_v9_4_3_hw_fini,
4108 	.suspend = gfx_v9_4_3_suspend,
4109 	.resume = gfx_v9_4_3_resume,
4110 	.is_idle = gfx_v9_4_3_is_idle,
4111 	.wait_for_idle = gfx_v9_4_3_wait_for_idle,
4112 	.soft_reset = gfx_v9_4_3_soft_reset,
4113 	.set_clockgating_state = gfx_v9_4_3_set_clockgating_state,
4114 	.set_powergating_state = gfx_v9_4_3_set_powergating_state,
4115 	.get_clockgating_state = gfx_v9_4_3_get_clockgating_state,
4116 };
4117 
4118 static const struct amdgpu_ring_funcs gfx_v9_4_3_ring_funcs_compute = {
4119 	.type = AMDGPU_RING_TYPE_COMPUTE,
4120 	.align_mask = 0xff,
4121 	.nop = PACKET3(PACKET3_NOP, 0x3FFF),
4122 	.support_64bit_ptrs = true,
4123 	.get_rptr = gfx_v9_4_3_ring_get_rptr_compute,
4124 	.get_wptr = gfx_v9_4_3_ring_get_wptr_compute,
4125 	.set_wptr = gfx_v9_4_3_ring_set_wptr_compute,
4126 	.emit_frame_size =
4127 		20 + /* gfx_v9_4_3_ring_emit_gds_switch */
4128 		7 + /* gfx_v9_4_3_ring_emit_hdp_flush */
4129 		5 + /* hdp invalidate */
4130 		7 + /* gfx_v9_4_3_ring_emit_pipeline_sync */
4131 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 5 +
4132 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 7 +
4133 		2 + /* gfx_v9_4_3_ring_emit_vm_flush */
4134 		8 + 8 + 8 + /* gfx_v9_4_3_ring_emit_fence x3 for user fence, vm fence */
4135 		7 + /* gfx_v9_4_3_emit_mem_sync */
4136 		5 + /* gfx_v9_4_3_emit_wave_limit for updating regSPI_WCL_PIPE_PERCENT_GFX register */
4137 		15, /* for updating 3 regSPI_WCL_PIPE_PERCENT_CS registers */
4138 	.emit_ib_size =	7, /* gfx_v9_4_3_ring_emit_ib_compute */
4139 	.emit_ib = gfx_v9_4_3_ring_emit_ib_compute,
4140 	.emit_fence = gfx_v9_4_3_ring_emit_fence,
4141 	.emit_pipeline_sync = gfx_v9_4_3_ring_emit_pipeline_sync,
4142 	.emit_vm_flush = gfx_v9_4_3_ring_emit_vm_flush,
4143 	.emit_gds_switch = gfx_v9_4_3_ring_emit_gds_switch,
4144 	.emit_hdp_flush = gfx_v9_4_3_ring_emit_hdp_flush,
4145 	.test_ring = gfx_v9_4_3_ring_test_ring,
4146 	.test_ib = gfx_v9_4_3_ring_test_ib,
4147 	.insert_nop = amdgpu_ring_insert_nop,
4148 	.pad_ib = amdgpu_ring_generic_pad_ib,
4149 	.emit_wreg = gfx_v9_4_3_ring_emit_wreg,
4150 	.emit_reg_wait = gfx_v9_4_3_ring_emit_reg_wait,
4151 	.emit_reg_write_reg_wait = gfx_v9_4_3_ring_emit_reg_write_reg_wait,
4152 	.emit_mem_sync = gfx_v9_4_3_emit_mem_sync,
4153 	.emit_wave_limit = gfx_v9_4_3_emit_wave_limit,
4154 };
4155 
4156 static const struct amdgpu_ring_funcs gfx_v9_4_3_ring_funcs_kiq = {
4157 	.type = AMDGPU_RING_TYPE_KIQ,
4158 	.align_mask = 0xff,
4159 	.nop = PACKET3(PACKET3_NOP, 0x3FFF),
4160 	.support_64bit_ptrs = true,
4161 	.get_rptr = gfx_v9_4_3_ring_get_rptr_compute,
4162 	.get_wptr = gfx_v9_4_3_ring_get_wptr_compute,
4163 	.set_wptr = gfx_v9_4_3_ring_set_wptr_compute,
4164 	.emit_frame_size =
4165 		20 + /* gfx_v9_4_3_ring_emit_gds_switch */
4166 		7 + /* gfx_v9_4_3_ring_emit_hdp_flush */
4167 		5 + /* hdp invalidate */
4168 		7 + /* gfx_v9_4_3_ring_emit_pipeline_sync */
4169 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 5 +
4170 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 7 +
4171 		2 + /* gfx_v9_4_3_ring_emit_vm_flush */
4172 		8 + 8 + 8, /* gfx_v9_4_3_ring_emit_fence_kiq x3 for user fence, vm fence */
4173 	.emit_ib_size =	7, /* gfx_v9_4_3_ring_emit_ib_compute */
4174 	.emit_fence = gfx_v9_4_3_ring_emit_fence_kiq,
4175 	.test_ring = gfx_v9_4_3_ring_test_ring,
4176 	.insert_nop = amdgpu_ring_insert_nop,
4177 	.pad_ib = amdgpu_ring_generic_pad_ib,
4178 	.emit_rreg = gfx_v9_4_3_ring_emit_rreg,
4179 	.emit_wreg = gfx_v9_4_3_ring_emit_wreg,
4180 	.emit_reg_wait = gfx_v9_4_3_ring_emit_reg_wait,
4181 	.emit_reg_write_reg_wait = gfx_v9_4_3_ring_emit_reg_write_reg_wait,
4182 };
4183 
gfx_v9_4_3_set_ring_funcs(struct amdgpu_device * adev)4184 static void gfx_v9_4_3_set_ring_funcs(struct amdgpu_device *adev)
4185 {
4186 	int i, j, num_xcc;
4187 
4188 	num_xcc = NUM_XCC(adev->gfx.xcc_mask);
4189 	for (i = 0; i < num_xcc; i++) {
4190 		adev->gfx.kiq[i].ring.funcs = &gfx_v9_4_3_ring_funcs_kiq;
4191 
4192 		for (j = 0; j < adev->gfx.num_compute_rings; j++)
4193 			adev->gfx.compute_ring[j + i * adev->gfx.num_compute_rings].funcs
4194 					= &gfx_v9_4_3_ring_funcs_compute;
4195 	}
4196 }
4197 
4198 static const struct amdgpu_irq_src_funcs gfx_v9_4_3_eop_irq_funcs = {
4199 	.set = gfx_v9_4_3_set_eop_interrupt_state,
4200 	.process = gfx_v9_4_3_eop_irq,
4201 };
4202 
4203 static const struct amdgpu_irq_src_funcs gfx_v9_4_3_priv_reg_irq_funcs = {
4204 	.set = gfx_v9_4_3_set_priv_reg_fault_state,
4205 	.process = gfx_v9_4_3_priv_reg_irq,
4206 };
4207 
4208 static const struct amdgpu_irq_src_funcs gfx_v9_4_3_priv_inst_irq_funcs = {
4209 	.set = gfx_v9_4_3_set_priv_inst_fault_state,
4210 	.process = gfx_v9_4_3_priv_inst_irq,
4211 };
4212 
gfx_v9_4_3_set_irq_funcs(struct amdgpu_device * adev)4213 static void gfx_v9_4_3_set_irq_funcs(struct amdgpu_device *adev)
4214 {
4215 	adev->gfx.eop_irq.num_types = AMDGPU_CP_IRQ_LAST;
4216 	adev->gfx.eop_irq.funcs = &gfx_v9_4_3_eop_irq_funcs;
4217 
4218 	adev->gfx.priv_reg_irq.num_types = 1;
4219 	adev->gfx.priv_reg_irq.funcs = &gfx_v9_4_3_priv_reg_irq_funcs;
4220 
4221 	adev->gfx.priv_inst_irq.num_types = 1;
4222 	adev->gfx.priv_inst_irq.funcs = &gfx_v9_4_3_priv_inst_irq_funcs;
4223 }
4224 
gfx_v9_4_3_set_rlc_funcs(struct amdgpu_device * adev)4225 static void gfx_v9_4_3_set_rlc_funcs(struct amdgpu_device *adev)
4226 {
4227 	adev->gfx.rlc.funcs = &gfx_v9_4_3_rlc_funcs;
4228 }
4229 
4230 
gfx_v9_4_3_set_gds_init(struct amdgpu_device * adev)4231 static void gfx_v9_4_3_set_gds_init(struct amdgpu_device *adev)
4232 {
4233 	/* init asci gds info */
4234 	switch (adev->ip_versions[GC_HWIP][0]) {
4235 	case IP_VERSION(9, 4, 3):
4236 		/* 9.4.3 removed all the GDS internal memory,
4237 		 * only support GWS opcode in kernel, like barrier
4238 		 * semaphore.etc */
4239 		adev->gds.gds_size = 0;
4240 		break;
4241 	default:
4242 		adev->gds.gds_size = 0x10000;
4243 		break;
4244 	}
4245 
4246 	switch (adev->ip_versions[GC_HWIP][0]) {
4247 	case IP_VERSION(9, 4, 3):
4248 		/* deprecated for 9.4.3, no usage at all */
4249 		adev->gds.gds_compute_max_wave_id = 0;
4250 		break;
4251 	default:
4252 		/* this really depends on the chip */
4253 		adev->gds.gds_compute_max_wave_id = 0x7ff;
4254 		break;
4255 	}
4256 
4257 	adev->gds.gws_size = 64;
4258 	adev->gds.oa_size = 16;
4259 }
4260 
gfx_v9_4_3_set_user_cu_inactive_bitmap(struct amdgpu_device * adev,u32 bitmap,int xcc_id)4261 static void gfx_v9_4_3_set_user_cu_inactive_bitmap(struct amdgpu_device *adev,
4262 						 u32 bitmap, int xcc_id)
4263 {
4264 	u32 data;
4265 
4266 	if (!bitmap)
4267 		return;
4268 
4269 	data = bitmap << GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_CUS__SHIFT;
4270 	data &= GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_CUS_MASK;
4271 
4272 	WREG32_SOC15(GC, GET_INST(GC, xcc_id), regGC_USER_SHADER_ARRAY_CONFIG, data);
4273 }
4274 
gfx_v9_4_3_get_cu_active_bitmap(struct amdgpu_device * adev,int xcc_id)4275 static u32 gfx_v9_4_3_get_cu_active_bitmap(struct amdgpu_device *adev, int xcc_id)
4276 {
4277 	u32 data, mask;
4278 
4279 	data = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regCC_GC_SHADER_ARRAY_CONFIG);
4280 	data |= RREG32_SOC15(GC, GET_INST(GC, xcc_id), regGC_USER_SHADER_ARRAY_CONFIG);
4281 
4282 	data &= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_CUS_MASK;
4283 	data >>= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_CUS__SHIFT;
4284 
4285 	mask = amdgpu_gfx_create_bitmask(adev->gfx.config.max_cu_per_sh);
4286 
4287 	return (~data) & mask;
4288 }
4289 
gfx_v9_4_3_get_cu_info(struct amdgpu_device * adev,struct amdgpu_cu_info * cu_info)4290 static int gfx_v9_4_3_get_cu_info(struct amdgpu_device *adev,
4291 				 struct amdgpu_cu_info *cu_info)
4292 {
4293 	int i, j, k, prev_counter, counter, xcc_id, active_cu_number = 0;
4294 	u32 mask, bitmap, ao_bitmap, ao_cu_mask = 0, tmp;
4295 	unsigned disable_masks[4 * 4];
4296 	bool is_symmetric_cus;
4297 
4298 	if (!adev || !cu_info)
4299 		return -EINVAL;
4300 
4301 	/*
4302 	 * 16 comes from bitmap array size 4*4, and it can cover all gfx9 ASICs
4303 	 */
4304 	if (adev->gfx.config.max_shader_engines *
4305 		adev->gfx.config.max_sh_per_se > 16)
4306 		return -EINVAL;
4307 
4308 	amdgpu_gfx_parse_disable_cu(disable_masks,
4309 				    adev->gfx.config.max_shader_engines,
4310 				    adev->gfx.config.max_sh_per_se);
4311 
4312 	mutex_lock(&adev->grbm_idx_mutex);
4313 	for (xcc_id = 0; xcc_id < NUM_XCC(adev->gfx.xcc_mask); xcc_id++) {
4314 		is_symmetric_cus = true;
4315 		for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
4316 			for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
4317 				mask = 1;
4318 				ao_bitmap = 0;
4319 				counter = 0;
4320 				gfx_v9_4_3_xcc_select_se_sh(adev, i, j, 0xffffffff, xcc_id);
4321 				gfx_v9_4_3_set_user_cu_inactive_bitmap(
4322 					adev,
4323 					disable_masks[i * adev->gfx.config.max_sh_per_se + j],
4324 					xcc_id);
4325 				bitmap = gfx_v9_4_3_get_cu_active_bitmap(adev, xcc_id);
4326 
4327 				cu_info->bitmap[xcc_id][i][j] = bitmap;
4328 
4329 				for (k = 0; k < adev->gfx.config.max_cu_per_sh; k++) {
4330 					if (bitmap & mask) {
4331 						if (counter < adev->gfx.config.max_cu_per_sh)
4332 							ao_bitmap |= mask;
4333 						counter++;
4334 					}
4335 					mask <<= 1;
4336 				}
4337 				active_cu_number += counter;
4338 				if (i < 2 && j < 2)
4339 					ao_cu_mask |= (ao_bitmap << (i * 16 + j * 8));
4340 				cu_info->ao_cu_bitmap[i][j] = ao_bitmap;
4341 			}
4342 			if (i && is_symmetric_cus && prev_counter != counter)
4343 				is_symmetric_cus = false;
4344 			prev_counter = counter;
4345 		}
4346 		if (is_symmetric_cus) {
4347 			tmp = RREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_CPC_DEBUG);
4348 			tmp = REG_SET_FIELD(tmp, CP_CPC_DEBUG, CPC_HARVESTING_RELAUNCH_DISABLE, 1);
4349 			tmp = REG_SET_FIELD(tmp, CP_CPC_DEBUG, CPC_HARVESTING_DISPATCH_DISABLE, 1);
4350 			WREG32_SOC15(GC, GET_INST(GC, xcc_id), regCP_CPC_DEBUG, tmp);
4351 		}
4352 		gfx_v9_4_3_xcc_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff,
4353 					    xcc_id);
4354 	}
4355 	mutex_unlock(&adev->grbm_idx_mutex);
4356 
4357 	cu_info->number = active_cu_number;
4358 	cu_info->ao_cu_mask = ao_cu_mask;
4359 	cu_info->simd_per_cu = NUM_SIMD_PER_CU;
4360 
4361 	return 0;
4362 }
4363 
4364 const struct amdgpu_ip_block_version gfx_v9_4_3_ip_block = {
4365 	.type = AMD_IP_BLOCK_TYPE_GFX,
4366 	.major = 9,
4367 	.minor = 4,
4368 	.rev = 0,
4369 	.funcs = &gfx_v9_4_3_ip_funcs,
4370 };
4371 
gfx_v9_4_3_xcp_resume(void * handle,uint32_t inst_mask)4372 static int gfx_v9_4_3_xcp_resume(void *handle, uint32_t inst_mask)
4373 {
4374 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4375 	uint32_t tmp_mask;
4376 	int i, r;
4377 
4378 	/* TODO : Initialize golden regs */
4379 	/* gfx_v9_4_3_init_golden_registers(adev); */
4380 
4381 	tmp_mask = inst_mask;
4382 	for_each_inst(i, tmp_mask)
4383 		gfx_v9_4_3_xcc_constants_init(adev, i);
4384 
4385 	if (!amdgpu_sriov_vf(adev)) {
4386 		tmp_mask = inst_mask;
4387 		for_each_inst(i, tmp_mask) {
4388 			r = gfx_v9_4_3_xcc_rlc_resume(adev, i);
4389 			if (r)
4390 				return r;
4391 		}
4392 	}
4393 
4394 	tmp_mask = inst_mask;
4395 	for_each_inst(i, tmp_mask) {
4396 		r = gfx_v9_4_3_xcc_cp_resume(adev, i);
4397 		if (r)
4398 			return r;
4399 	}
4400 
4401 	return 0;
4402 }
4403 
gfx_v9_4_3_xcp_suspend(void * handle,uint32_t inst_mask)4404 static int gfx_v9_4_3_xcp_suspend(void *handle, uint32_t inst_mask)
4405 {
4406 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4407 	int i;
4408 
4409 	for_each_inst(i, inst_mask)
4410 		gfx_v9_4_3_xcc_fini(adev, i);
4411 
4412 	return 0;
4413 }
4414 
4415 struct amdgpu_xcp_ip_funcs gfx_v9_4_3_xcp_funcs = {
4416 	.suspend = &gfx_v9_4_3_xcp_suspend,
4417 	.resume = &gfx_v9_4_3_xcp_resume
4418 };
4419 
4420 struct amdgpu_ras_block_hw_ops  gfx_v9_4_3_ras_ops = {
4421 	.query_ras_error_count = &gfx_v9_4_3_query_ras_error_count,
4422 	.reset_ras_error_count = &gfx_v9_4_3_reset_ras_error_count,
4423 	.query_ras_error_status = &gfx_v9_4_3_query_ras_error_status,
4424 	.reset_ras_error_status = &gfx_v9_4_3_reset_ras_error_status,
4425 };
4426 
4427 struct amdgpu_gfx_ras gfx_v9_4_3_ras = {
4428 	.ras_block = {
4429 		.hw_ops = &gfx_v9_4_3_ras_ops,
4430 	},
4431 	.enable_watchdog_timer = &gfx_v9_4_3_enable_watchdog_timer,
4432 };
4433