1 /*
2  * Copyright 2021 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/delay.h>
24 #include <linux/kernel.h>
25 #include <linux/firmware.h>
26 #include <linux/module.h>
27 #include <linux/pci.h>
28 #include "amdgpu.h"
29 #include "amdgpu_gfx.h"
30 #include "amdgpu_psp.h"
31 #include "amdgpu_smu.h"
32 #include "amdgpu_atomfirmware.h"
33 #include "imu_v11_0.h"
34 #include "soc21.h"
35 #include "nvd.h"
36 
37 #include "gc/gc_11_0_0_offset.h"
38 #include "gc/gc_11_0_0_sh_mask.h"
39 #include "smuio/smuio_13_0_6_offset.h"
40 #include "smuio/smuio_13_0_6_sh_mask.h"
41 #include "navi10_enum.h"
42 #include "ivsrcid/gfx/irqsrcs_gfx_11_0_0.h"
43 
44 #include "soc15.h"
45 #include "soc15d.h"
46 #include "clearstate_gfx11.h"
47 #include "v11_structs.h"
48 #include "gfx_v11_0.h"
49 #include "nbio_v4_3.h"
50 #include "mes_v11_0.h"
51 
52 #define GFX11_NUM_GFX_RINGS		1
53 #define GFX11_MEC_HPD_SIZE	2048
54 
55 #define RLCG_UCODE_LOADING_START_ADDRESS	0x00002000L
56 
57 #define regCGTT_WD_CLK_CTRL		0x5086
58 #define regCGTT_WD_CLK_CTRL_BASE_IDX	1
59 
60 MODULE_FIRMWARE("amdgpu/gc_11_0_0_pfp.bin");
61 MODULE_FIRMWARE("amdgpu/gc_11_0_0_me.bin");
62 MODULE_FIRMWARE("amdgpu/gc_11_0_0_mec.bin");
63 MODULE_FIRMWARE("amdgpu/gc_11_0_0_rlc.bin");
64 MODULE_FIRMWARE("amdgpu/gc_11_0_0_toc.bin");
65 MODULE_FIRMWARE("amdgpu/gc_11_0_1_pfp.bin");
66 MODULE_FIRMWARE("amdgpu/gc_11_0_1_me.bin");
67 MODULE_FIRMWARE("amdgpu/gc_11_0_1_mec.bin");
68 MODULE_FIRMWARE("amdgpu/gc_11_0_1_rlc.bin");
69 MODULE_FIRMWARE("amdgpu/gc_11_0_2_pfp.bin");
70 MODULE_FIRMWARE("amdgpu/gc_11_0_2_me.bin");
71 MODULE_FIRMWARE("amdgpu/gc_11_0_2_mec.bin");
72 MODULE_FIRMWARE("amdgpu/gc_11_0_2_rlc.bin");
73 
74 static const struct soc15_reg_golden golden_settings_gc_11_0[] =
75 {
76 	/* Pending on emulation bring up */
77 };
78 
79 static const struct soc15_reg_golden golden_settings_gc_11_0_0[] =
80 {
81 	/* Pending on emulation bring up */
82 };
83 
84 static const struct soc15_reg_golden golden_settings_gc_rlc_spm_11_0[] =
85 {
86 	/* Pending on emulation bring up */
87 };
88 
89 static const struct soc15_reg_golden golden_settings_gc_11_0_1[] =
90 {
91 	SOC15_REG_GOLDEN_VALUE(GC, 0, regCGTT_GS_NGG_CLK_CTRL, 0x9fff8fff, 0x00000010),
92 	SOC15_REG_GOLDEN_VALUE(GC, 0, regCGTT_WD_CLK_CTRL, 0xffff8fff, 0x00000010),
93 	SOC15_REG_GOLDEN_VALUE(GC, 0, regCPF_GCR_CNTL, 0x0007ffff, 0x0000c200),
94 	SOC15_REG_GOLDEN_VALUE(GC, 0, regGL2C_CTRL3, 0xffff001b, 0x00f01988),
95 	SOC15_REG_GOLDEN_VALUE(GC, 0, regPA_CL_ENHANCE, 0xf0ffffff, 0x00880007),
96 	SOC15_REG_GOLDEN_VALUE(GC, 0, regPA_SC_ENHANCE_3, 0xfffffffd, 0x00000008),
97 	SOC15_REG_GOLDEN_VALUE(GC, 0, regPA_SC_VRS_SURFACE_CNTL_1, 0xfff891ff, 0x55480100),
98 	SOC15_REG_GOLDEN_VALUE(GC, 0, regTA_CNTL_AUX, 0xf7f7ffff, 0x01030000),
99 	SOC15_REG_GOLDEN_VALUE(GC, 0, regTCP_CNTL2, 0xfcffffff, 0x0000000a)
100 };
101 
102 #define DEFAULT_SH_MEM_CONFIG \
103 	((SH_MEM_ADDRESS_MODE_64 << SH_MEM_CONFIG__ADDRESS_MODE__SHIFT) | \
104 	 (SH_MEM_ALIGNMENT_MODE_UNALIGNED << SH_MEM_CONFIG__ALIGNMENT_MODE__SHIFT) | \
105 	 (3 << SH_MEM_CONFIG__INITIAL_INST_PREFETCH__SHIFT))
106 
107 static void gfx_v11_0_disable_gpa_mode(struct amdgpu_device *adev);
108 static void gfx_v11_0_set_ring_funcs(struct amdgpu_device *adev);
109 static void gfx_v11_0_set_irq_funcs(struct amdgpu_device *adev);
110 static void gfx_v11_0_set_gds_init(struct amdgpu_device *adev);
111 static void gfx_v11_0_set_rlc_funcs(struct amdgpu_device *adev);
112 static void gfx_v11_0_set_mqd_funcs(struct amdgpu_device *adev);
113 static void gfx_v11_0_set_imu_funcs(struct amdgpu_device *adev);
114 static int gfx_v11_0_get_cu_info(struct amdgpu_device *adev,
115                                  struct amdgpu_cu_info *cu_info);
116 static uint64_t gfx_v11_0_get_gpu_clock_counter(struct amdgpu_device *adev);
117 static void gfx_v11_0_select_se_sh(struct amdgpu_device *adev, u32 se_num,
118 				   u32 sh_num, u32 instance);
119 static u32 gfx_v11_0_get_wgp_active_bitmap_per_sh(struct amdgpu_device *adev);
120 
121 static void gfx_v11_0_ring_emit_de_meta(struct amdgpu_ring *ring, bool resume);
122 static void gfx_v11_0_ring_emit_frame_cntl(struct amdgpu_ring *ring, bool start, bool secure);
123 static void gfx_v11_0_ring_emit_wreg(struct amdgpu_ring *ring, uint32_t reg,
124 				     uint32_t val);
125 static int gfx_v11_0_wait_for_rlc_autoload_complete(struct amdgpu_device *adev);
126 static void gfx_v11_0_ring_invalidate_tlbs(struct amdgpu_ring *ring,
127 					   uint16_t pasid, uint32_t flush_type,
128 					   bool all_hub, uint8_t dst_sel);
129 
130 static void gfx11_kiq_set_resources(struct amdgpu_ring *kiq_ring, uint64_t queue_mask)
131 {
132 	amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_SET_RESOURCES, 6));
133 	amdgpu_ring_write(kiq_ring, PACKET3_SET_RESOURCES_VMID_MASK(0) |
134 			  PACKET3_SET_RESOURCES_QUEUE_TYPE(0));	/* vmid_mask:0 queue_type:0 (KIQ) */
135 	amdgpu_ring_write(kiq_ring, lower_32_bits(queue_mask));	/* queue mask lo */
136 	amdgpu_ring_write(kiq_ring, upper_32_bits(queue_mask));	/* queue mask hi */
137 	amdgpu_ring_write(kiq_ring, 0);	/* gws mask lo */
138 	amdgpu_ring_write(kiq_ring, 0);	/* gws mask hi */
139 	amdgpu_ring_write(kiq_ring, 0);	/* oac mask */
140 	amdgpu_ring_write(kiq_ring, 0);	/* gds heap base:0, gds heap size:0 */
141 }
142 
143 static void gfx11_kiq_map_queues(struct amdgpu_ring *kiq_ring,
144 				 struct amdgpu_ring *ring)
145 {
146 	uint64_t mqd_addr = amdgpu_bo_gpu_offset(ring->mqd_obj);
147 	uint64_t wptr_addr = ring->wptr_gpu_addr;
148 	uint32_t me = 0, eng_sel = 0;
149 
150 	switch (ring->funcs->type) {
151 	case AMDGPU_RING_TYPE_COMPUTE:
152 		me = 1;
153 		eng_sel = 0;
154 		break;
155 	case AMDGPU_RING_TYPE_GFX:
156 		me = 0;
157 		eng_sel = 4;
158 		break;
159 	case AMDGPU_RING_TYPE_MES:
160 		me = 2;
161 		eng_sel = 5;
162 		break;
163 	default:
164 		WARN_ON(1);
165 	}
166 
167 	amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_MAP_QUEUES, 5));
168 	/* Q_sel:0, vmid:0, vidmem: 1, engine:0, num_Q:1*/
169 	amdgpu_ring_write(kiq_ring, /* Q_sel: 0, vmid: 0, engine: 0, num_Q: 1 */
170 			  PACKET3_MAP_QUEUES_QUEUE_SEL(0) | /* Queue_Sel */
171 			  PACKET3_MAP_QUEUES_VMID(0) | /* VMID */
172 			  PACKET3_MAP_QUEUES_QUEUE(ring->queue) |
173 			  PACKET3_MAP_QUEUES_PIPE(ring->pipe) |
174 			  PACKET3_MAP_QUEUES_ME((me)) |
175 			  PACKET3_MAP_QUEUES_QUEUE_TYPE(0) | /*queue_type: normal compute queue */
176 			  PACKET3_MAP_QUEUES_ALLOC_FORMAT(0) | /* alloc format: all_on_one_pipe */
177 			  PACKET3_MAP_QUEUES_ENGINE_SEL(eng_sel) |
178 			  PACKET3_MAP_QUEUES_NUM_QUEUES(1)); /* num_queues: must be 1 */
179 	amdgpu_ring_write(kiq_ring, PACKET3_MAP_QUEUES_DOORBELL_OFFSET(ring->doorbell_index));
180 	amdgpu_ring_write(kiq_ring, lower_32_bits(mqd_addr));
181 	amdgpu_ring_write(kiq_ring, upper_32_bits(mqd_addr));
182 	amdgpu_ring_write(kiq_ring, lower_32_bits(wptr_addr));
183 	amdgpu_ring_write(kiq_ring, upper_32_bits(wptr_addr));
184 }
185 
186 static void gfx11_kiq_unmap_queues(struct amdgpu_ring *kiq_ring,
187 				   struct amdgpu_ring *ring,
188 				   enum amdgpu_unmap_queues_action action,
189 				   u64 gpu_addr, u64 seq)
190 {
191 	struct amdgpu_device *adev = kiq_ring->adev;
192 	uint32_t eng_sel = ring->funcs->type == AMDGPU_RING_TYPE_GFX ? 4 : 0;
193 
194 	if (adev->enable_mes && !adev->gfx.kiq.ring.sched.ready) {
195 		amdgpu_mes_unmap_legacy_queue(adev, ring, action, gpu_addr, seq);
196 		return;
197 	}
198 
199 	amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_UNMAP_QUEUES, 4));
200 	amdgpu_ring_write(kiq_ring, /* Q_sel: 0, vmid: 0, engine: 0, num_Q: 1 */
201 			  PACKET3_UNMAP_QUEUES_ACTION(action) |
202 			  PACKET3_UNMAP_QUEUES_QUEUE_SEL(0) |
203 			  PACKET3_UNMAP_QUEUES_ENGINE_SEL(eng_sel) |
204 			  PACKET3_UNMAP_QUEUES_NUM_QUEUES(1));
205 	amdgpu_ring_write(kiq_ring,
206 		  PACKET3_UNMAP_QUEUES_DOORBELL_OFFSET0(ring->doorbell_index));
207 
208 	if (action == PREEMPT_QUEUES_NO_UNMAP) {
209 		amdgpu_ring_write(kiq_ring, lower_32_bits(gpu_addr));
210 		amdgpu_ring_write(kiq_ring, upper_32_bits(gpu_addr));
211 		amdgpu_ring_write(kiq_ring, seq);
212 	} else {
213 		amdgpu_ring_write(kiq_ring, 0);
214 		amdgpu_ring_write(kiq_ring, 0);
215 		amdgpu_ring_write(kiq_ring, 0);
216 	}
217 }
218 
219 static void gfx11_kiq_query_status(struct amdgpu_ring *kiq_ring,
220 				   struct amdgpu_ring *ring,
221 				   u64 addr,
222 				   u64 seq)
223 {
224 	uint32_t eng_sel = ring->funcs->type == AMDGPU_RING_TYPE_GFX ? 4 : 0;
225 
226 	amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_QUERY_STATUS, 5));
227 	amdgpu_ring_write(kiq_ring,
228 			  PACKET3_QUERY_STATUS_CONTEXT_ID(0) |
229 			  PACKET3_QUERY_STATUS_INTERRUPT_SEL(0) |
230 			  PACKET3_QUERY_STATUS_COMMAND(2));
231 	amdgpu_ring_write(kiq_ring, /* Q_sel: 0, vmid: 0, engine: 0, num_Q: 1 */
232 			  PACKET3_QUERY_STATUS_DOORBELL_OFFSET(ring->doorbell_index) |
233 			  PACKET3_QUERY_STATUS_ENG_SEL(eng_sel));
234 	amdgpu_ring_write(kiq_ring, lower_32_bits(addr));
235 	amdgpu_ring_write(kiq_ring, upper_32_bits(addr));
236 	amdgpu_ring_write(kiq_ring, lower_32_bits(seq));
237 	amdgpu_ring_write(kiq_ring, upper_32_bits(seq));
238 }
239 
240 static void gfx11_kiq_invalidate_tlbs(struct amdgpu_ring *kiq_ring,
241 				uint16_t pasid, uint32_t flush_type,
242 				bool all_hub)
243 {
244 	gfx_v11_0_ring_invalidate_tlbs(kiq_ring, pasid, flush_type, all_hub, 1);
245 }
246 
247 static const struct kiq_pm4_funcs gfx_v11_0_kiq_pm4_funcs = {
248 	.kiq_set_resources = gfx11_kiq_set_resources,
249 	.kiq_map_queues = gfx11_kiq_map_queues,
250 	.kiq_unmap_queues = gfx11_kiq_unmap_queues,
251 	.kiq_query_status = gfx11_kiq_query_status,
252 	.kiq_invalidate_tlbs = gfx11_kiq_invalidate_tlbs,
253 	.set_resources_size = 8,
254 	.map_queues_size = 7,
255 	.unmap_queues_size = 6,
256 	.query_status_size = 7,
257 	.invalidate_tlbs_size = 2,
258 };
259 
260 static void gfx_v11_0_set_kiq_pm4_funcs(struct amdgpu_device *adev)
261 {
262 	adev->gfx.kiq.pmf = &gfx_v11_0_kiq_pm4_funcs;
263 }
264 
265 static void gfx_v11_0_init_spm_golden_registers(struct amdgpu_device *adev)
266 {
267 	switch (adev->ip_versions[GC_HWIP][0]) {
268 	case IP_VERSION(11, 0, 0):
269 		soc15_program_register_sequence(adev,
270 						golden_settings_gc_rlc_spm_11_0,
271 						(const u32)ARRAY_SIZE(golden_settings_gc_rlc_spm_11_0));
272 		break;
273 	default:
274 		break;
275 	}
276 }
277 
278 static void gfx_v11_0_init_golden_registers(struct amdgpu_device *adev)
279 {
280 	switch (adev->ip_versions[GC_HWIP][0]) {
281 	case IP_VERSION(11, 0, 0):
282 		soc15_program_register_sequence(adev,
283 						golden_settings_gc_11_0,
284 						(const u32)ARRAY_SIZE(golden_settings_gc_11_0));
285 		soc15_program_register_sequence(adev,
286 						golden_settings_gc_11_0_0,
287 						(const u32)ARRAY_SIZE(golden_settings_gc_11_0_0));
288 		break;
289 	case IP_VERSION(11, 0, 1):
290 		soc15_program_register_sequence(adev,
291 						golden_settings_gc_11_0,
292 						(const u32)ARRAY_SIZE(golden_settings_gc_11_0));
293 		soc15_program_register_sequence(adev,
294 						golden_settings_gc_11_0_1,
295 						(const u32)ARRAY_SIZE(golden_settings_gc_11_0_1));
296 		break;
297 	default:
298 		break;
299 	}
300 	gfx_v11_0_init_spm_golden_registers(adev);
301 }
302 
303 static void gfx_v11_0_write_data_to_reg(struct amdgpu_ring *ring, int eng_sel,
304 				       bool wc, uint32_t reg, uint32_t val)
305 {
306 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
307 	amdgpu_ring_write(ring, WRITE_DATA_ENGINE_SEL(eng_sel) |
308 			  WRITE_DATA_DST_SEL(0) | (wc ? WR_CONFIRM : 0));
309 	amdgpu_ring_write(ring, reg);
310 	amdgpu_ring_write(ring, 0);
311 	amdgpu_ring_write(ring, val);
312 }
313 
314 static void gfx_v11_0_wait_reg_mem(struct amdgpu_ring *ring, int eng_sel,
315 				  int mem_space, int opt, uint32_t addr0,
316 				  uint32_t addr1, uint32_t ref, uint32_t mask,
317 				  uint32_t inv)
318 {
319 	amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
320 	amdgpu_ring_write(ring,
321 			  /* memory (1) or register (0) */
322 			  (WAIT_REG_MEM_MEM_SPACE(mem_space) |
323 			   WAIT_REG_MEM_OPERATION(opt) | /* wait */
324 			   WAIT_REG_MEM_FUNCTION(3) |  /* equal */
325 			   WAIT_REG_MEM_ENGINE(eng_sel)));
326 
327 	if (mem_space)
328 		BUG_ON(addr0 & 0x3); /* Dword align */
329 	amdgpu_ring_write(ring, addr0);
330 	amdgpu_ring_write(ring, addr1);
331 	amdgpu_ring_write(ring, ref);
332 	amdgpu_ring_write(ring, mask);
333 	amdgpu_ring_write(ring, inv); /* poll interval */
334 }
335 
336 static int gfx_v11_0_ring_test_ring(struct amdgpu_ring *ring)
337 {
338 	struct amdgpu_device *adev = ring->adev;
339 	uint32_t scratch = SOC15_REG_OFFSET(GC, 0, regSCRATCH_REG0);
340 	uint32_t tmp = 0;
341 	unsigned i;
342 	int r;
343 
344 	WREG32(scratch, 0xCAFEDEAD);
345 	r = amdgpu_ring_alloc(ring, 5);
346 	if (r) {
347 		DRM_ERROR("amdgpu: cp failed to lock ring %d (%d).\n",
348 			  ring->idx, r);
349 		return r;
350 	}
351 
352 	if (ring->funcs->type == AMDGPU_RING_TYPE_KIQ) {
353 		gfx_v11_0_ring_emit_wreg(ring, scratch, 0xDEADBEEF);
354 	} else {
355 		amdgpu_ring_write(ring, PACKET3(PACKET3_SET_UCONFIG_REG, 1));
356 		amdgpu_ring_write(ring, scratch -
357 				  PACKET3_SET_UCONFIG_REG_START);
358 		amdgpu_ring_write(ring, 0xDEADBEEF);
359 	}
360 	amdgpu_ring_commit(ring);
361 
362 	for (i = 0; i < adev->usec_timeout; i++) {
363 		tmp = RREG32(scratch);
364 		if (tmp == 0xDEADBEEF)
365 			break;
366 		if (amdgpu_emu_mode == 1)
367 			msleep(1);
368 		else
369 			udelay(1);
370 	}
371 
372 	if (i >= adev->usec_timeout)
373 		r = -ETIMEDOUT;
374 	return r;
375 }
376 
377 static int gfx_v11_0_ring_test_ib(struct amdgpu_ring *ring, long timeout)
378 {
379 	struct amdgpu_device *adev = ring->adev;
380 	struct amdgpu_ib ib;
381 	struct dma_fence *f = NULL;
382 	unsigned index;
383 	uint64_t gpu_addr;
384 	volatile uint32_t *cpu_ptr;
385 	long r;
386 
387 	/* MES KIQ fw hasn't indirect buffer support for now */
388 	if (adev->enable_mes_kiq &&
389 	    ring->funcs->type == AMDGPU_RING_TYPE_KIQ)
390 		return 0;
391 
392 	memset(&ib, 0, sizeof(ib));
393 
394 	if (ring->is_mes_queue) {
395 		uint32_t padding, offset;
396 
397 		offset = amdgpu_mes_ctx_get_offs(ring, AMDGPU_MES_CTX_IB_OFFS);
398 		padding = amdgpu_mes_ctx_get_offs(ring,
399 						  AMDGPU_MES_CTX_PADDING_OFFS);
400 
401 		ib.gpu_addr = amdgpu_mes_ctx_get_offs_gpu_addr(ring, offset);
402 		ib.ptr = amdgpu_mes_ctx_get_offs_cpu_addr(ring, offset);
403 
404 		gpu_addr = amdgpu_mes_ctx_get_offs_gpu_addr(ring, padding);
405 		cpu_ptr = amdgpu_mes_ctx_get_offs_cpu_addr(ring, padding);
406 		*cpu_ptr = cpu_to_le32(0xCAFEDEAD);
407 	} else {
408 		r = amdgpu_device_wb_get(adev, &index);
409 		if (r)
410 			return r;
411 
412 		gpu_addr = adev->wb.gpu_addr + (index * 4);
413 		adev->wb.wb[index] = cpu_to_le32(0xCAFEDEAD);
414 		cpu_ptr = &adev->wb.wb[index];
415 
416 		r = amdgpu_ib_get(adev, NULL, 16, AMDGPU_IB_POOL_DIRECT, &ib);
417 		if (r) {
418 			DRM_ERROR("amdgpu: failed to get ib (%ld).\n", r);
419 			goto err1;
420 		}
421 	}
422 
423 	ib.ptr[0] = PACKET3(PACKET3_WRITE_DATA, 3);
424 	ib.ptr[1] = WRITE_DATA_DST_SEL(5) | WR_CONFIRM;
425 	ib.ptr[2] = lower_32_bits(gpu_addr);
426 	ib.ptr[3] = upper_32_bits(gpu_addr);
427 	ib.ptr[4] = 0xDEADBEEF;
428 	ib.length_dw = 5;
429 
430 	r = amdgpu_ib_schedule(ring, 1, &ib, NULL, &f);
431 	if (r)
432 		goto err2;
433 
434 	r = dma_fence_wait_timeout(f, false, timeout);
435 	if (r == 0) {
436 		r = -ETIMEDOUT;
437 		goto err2;
438 	} else if (r < 0) {
439 		goto err2;
440 	}
441 
442 	if (le32_to_cpu(*cpu_ptr) == 0xDEADBEEF)
443 		r = 0;
444 	else
445 		r = -EINVAL;
446 err2:
447 	if (!ring->is_mes_queue)
448 		amdgpu_ib_free(adev, &ib, NULL);
449 	dma_fence_put(f);
450 err1:
451 	if (!ring->is_mes_queue)
452 		amdgpu_device_wb_free(adev, index);
453 	return r;
454 }
455 
456 static void gfx_v11_0_free_microcode(struct amdgpu_device *adev)
457 {
458 	release_firmware(adev->gfx.pfp_fw);
459 	adev->gfx.pfp_fw = NULL;
460 	release_firmware(adev->gfx.me_fw);
461 	adev->gfx.me_fw = NULL;
462 	release_firmware(adev->gfx.rlc_fw);
463 	adev->gfx.rlc_fw = NULL;
464 	release_firmware(adev->gfx.mec_fw);
465 	adev->gfx.mec_fw = NULL;
466 
467 	kfree(adev->gfx.rlc.register_list_format);
468 }
469 
470 static void gfx_v11_0_init_rlc_ext_microcode(struct amdgpu_device *adev)
471 {
472 	const struct rlc_firmware_header_v2_1 *rlc_hdr;
473 
474 	rlc_hdr = (const struct rlc_firmware_header_v2_1 *)adev->gfx.rlc_fw->data;
475 	adev->gfx.rlc_srlc_fw_version = le32_to_cpu(rlc_hdr->save_restore_list_cntl_ucode_ver);
476 	adev->gfx.rlc_srlc_feature_version = le32_to_cpu(rlc_hdr->save_restore_list_cntl_feature_ver);
477 	adev->gfx.rlc.save_restore_list_cntl_size_bytes = le32_to_cpu(rlc_hdr->save_restore_list_cntl_size_bytes);
478 	adev->gfx.rlc.save_restore_list_cntl = (u8 *)rlc_hdr + le32_to_cpu(rlc_hdr->save_restore_list_cntl_offset_bytes);
479 	adev->gfx.rlc_srlg_fw_version = le32_to_cpu(rlc_hdr->save_restore_list_gpm_ucode_ver);
480 	adev->gfx.rlc_srlg_feature_version = le32_to_cpu(rlc_hdr->save_restore_list_gpm_feature_ver);
481 	adev->gfx.rlc.save_restore_list_gpm_size_bytes = le32_to_cpu(rlc_hdr->save_restore_list_gpm_size_bytes);
482 	adev->gfx.rlc.save_restore_list_gpm = (u8 *)rlc_hdr + le32_to_cpu(rlc_hdr->save_restore_list_gpm_offset_bytes);
483 	adev->gfx.rlc_srls_fw_version = le32_to_cpu(rlc_hdr->save_restore_list_srm_ucode_ver);
484 	adev->gfx.rlc_srls_feature_version = le32_to_cpu(rlc_hdr->save_restore_list_srm_feature_ver);
485 	adev->gfx.rlc.save_restore_list_srm_size_bytes = le32_to_cpu(rlc_hdr->save_restore_list_srm_size_bytes);
486 	adev->gfx.rlc.save_restore_list_srm = (u8 *)rlc_hdr + le32_to_cpu(rlc_hdr->save_restore_list_srm_offset_bytes);
487 	adev->gfx.rlc.reg_list_format_direct_reg_list_length =
488 			le32_to_cpu(rlc_hdr->reg_list_format_direct_reg_list_length);
489 }
490 
491 static void gfx_v11_0_init_rlc_iram_dram_microcode(struct amdgpu_device *adev)
492 {
493 	const struct rlc_firmware_header_v2_2 *rlc_hdr;
494 
495 	rlc_hdr = (const struct rlc_firmware_header_v2_2 *)adev->gfx.rlc_fw->data;
496 	adev->gfx.rlc.rlc_iram_ucode_size_bytes = le32_to_cpu(rlc_hdr->rlc_iram_ucode_size_bytes);
497 	adev->gfx.rlc.rlc_iram_ucode = (u8 *)rlc_hdr + le32_to_cpu(rlc_hdr->rlc_iram_ucode_offset_bytes);
498 	adev->gfx.rlc.rlc_dram_ucode_size_bytes = le32_to_cpu(rlc_hdr->rlc_dram_ucode_size_bytes);
499 	adev->gfx.rlc.rlc_dram_ucode = (u8 *)rlc_hdr + le32_to_cpu(rlc_hdr->rlc_dram_ucode_offset_bytes);
500 }
501 
502 static void gfx_v11_0_init_rlcp_rlcv_microcode(struct amdgpu_device *adev)
503 {
504 	const struct rlc_firmware_header_v2_3 *rlc_hdr;
505 
506 	rlc_hdr = (const struct rlc_firmware_header_v2_3 *)adev->gfx.rlc_fw->data;
507 	adev->gfx.rlc.rlcp_ucode_size_bytes = le32_to_cpu(rlc_hdr->rlcp_ucode_size_bytes);
508 	adev->gfx.rlc.rlcp_ucode = (u8 *)rlc_hdr + le32_to_cpu(rlc_hdr->rlcp_ucode_offset_bytes);
509 	adev->gfx.rlc.rlcv_ucode_size_bytes = le32_to_cpu(rlc_hdr->rlcv_ucode_size_bytes);
510 	adev->gfx.rlc.rlcv_ucode = (u8 *)rlc_hdr + le32_to_cpu(rlc_hdr->rlcv_ucode_offset_bytes);
511 }
512 
513 static int gfx_v11_0_init_microcode(struct amdgpu_device *adev)
514 {
515 	char fw_name[40];
516 	char ucode_prefix[30];
517 	int err;
518 	struct amdgpu_firmware_info *info = NULL;
519 	const struct common_firmware_header *header = NULL;
520 	const struct gfx_firmware_header_v1_0 *cp_hdr;
521 	const struct gfx_firmware_header_v2_0 *cp_hdr_v2_0;
522 	const struct rlc_firmware_header_v2_0 *rlc_hdr;
523 	unsigned int *tmp = NULL;
524 	unsigned int i = 0;
525 	uint16_t version_major;
526 	uint16_t version_minor;
527 
528 	DRM_DEBUG("\n");
529 
530 	amdgpu_ucode_ip_version_decode(adev, GC_HWIP, ucode_prefix, sizeof(ucode_prefix));
531 
532 	snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_pfp.bin", ucode_prefix);
533 	err = request_firmware(&adev->gfx.pfp_fw, fw_name, adev->dev);
534 	if (err)
535 		goto out;
536 	err = amdgpu_ucode_validate(adev->gfx.pfp_fw);
537 	if (err)
538 		goto out;
539 	/* check pfp fw hdr version to decide if enable rs64 for gfx11.*/
540 	adev->gfx.rs64_enable = amdgpu_ucode_hdr_version(
541 				(union amdgpu_firmware_header *)
542 				adev->gfx.pfp_fw->data, 2, 0);
543 	if (adev->gfx.rs64_enable) {
544 		dev_info(adev->dev, "CP RS64 enable\n");
545 		cp_hdr_v2_0 = (const struct gfx_firmware_header_v2_0 *)adev->gfx.pfp_fw->data;
546 		adev->gfx.pfp_fw_version = le32_to_cpu(cp_hdr_v2_0->header.ucode_version);
547 		adev->gfx.pfp_feature_version = le32_to_cpu(cp_hdr_v2_0->ucode_feature_version);
548 
549 	} else {
550 		cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.pfp_fw->data;
551 		adev->gfx.pfp_fw_version = le32_to_cpu(cp_hdr->header.ucode_version);
552 		adev->gfx.pfp_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version);
553 	}
554 
555 	snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_me.bin", ucode_prefix);
556 	err = request_firmware(&adev->gfx.me_fw, fw_name, adev->dev);
557 	if (err)
558 		goto out;
559 	err = amdgpu_ucode_validate(adev->gfx.me_fw);
560 	if (err)
561 		goto out;
562 	if (adev->gfx.rs64_enable) {
563 		cp_hdr_v2_0 = (const struct gfx_firmware_header_v2_0 *)adev->gfx.me_fw->data;
564 		adev->gfx.me_fw_version = le32_to_cpu(cp_hdr_v2_0->header.ucode_version);
565 		adev->gfx.me_feature_version = le32_to_cpu(cp_hdr_v2_0->ucode_feature_version);
566 
567 	} else {
568 		cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.me_fw->data;
569 		adev->gfx.me_fw_version = le32_to_cpu(cp_hdr->header.ucode_version);
570 		adev->gfx.me_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version);
571 	}
572 
573 	if (!amdgpu_sriov_vf(adev)) {
574 		snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_rlc.bin", ucode_prefix);
575 		err = request_firmware(&adev->gfx.rlc_fw, fw_name, adev->dev);
576 		if (err)
577 			goto out;
578 		err = amdgpu_ucode_validate(adev->gfx.rlc_fw);
579 		rlc_hdr = (const struct rlc_firmware_header_v2_0 *)adev->gfx.rlc_fw->data;
580 		version_major = le16_to_cpu(rlc_hdr->header.header_version_major);
581 		version_minor = le16_to_cpu(rlc_hdr->header.header_version_minor);
582 
583 		adev->gfx.rlc_fw_version = le32_to_cpu(rlc_hdr->header.ucode_version);
584 		adev->gfx.rlc_feature_version = le32_to_cpu(rlc_hdr->ucode_feature_version);
585 		adev->gfx.rlc.save_and_restore_offset =
586 			le32_to_cpu(rlc_hdr->save_and_restore_offset);
587 		adev->gfx.rlc.clear_state_descriptor_offset =
588 			le32_to_cpu(rlc_hdr->clear_state_descriptor_offset);
589 		adev->gfx.rlc.avail_scratch_ram_locations =
590 			le32_to_cpu(rlc_hdr->avail_scratch_ram_locations);
591 		adev->gfx.rlc.reg_restore_list_size =
592 			le32_to_cpu(rlc_hdr->reg_restore_list_size);
593 		adev->gfx.rlc.reg_list_format_start =
594 			le32_to_cpu(rlc_hdr->reg_list_format_start);
595 		adev->gfx.rlc.reg_list_format_separate_start =
596 			le32_to_cpu(rlc_hdr->reg_list_format_separate_start);
597 		adev->gfx.rlc.starting_offsets_start =
598 			le32_to_cpu(rlc_hdr->starting_offsets_start);
599 		adev->gfx.rlc.reg_list_format_size_bytes =
600 			le32_to_cpu(rlc_hdr->reg_list_format_size_bytes);
601 		adev->gfx.rlc.reg_list_size_bytes =
602 			le32_to_cpu(rlc_hdr->reg_list_size_bytes);
603 		adev->gfx.rlc.register_list_format =
604 			kmalloc(adev->gfx.rlc.reg_list_format_size_bytes +
605 					adev->gfx.rlc.reg_list_size_bytes, GFP_KERNEL);
606 		if (!adev->gfx.rlc.register_list_format) {
607 			err = -ENOMEM;
608 			goto out;
609 		}
610 
611 		tmp = (unsigned int *)((uintptr_t)rlc_hdr +
612 							   le32_to_cpu(rlc_hdr->reg_list_format_array_offset_bytes));
613 		for (i = 0 ; i < (rlc_hdr->reg_list_format_size_bytes >> 2); i++)
614 			adev->gfx.rlc.register_list_format[i] =	le32_to_cpu(tmp[i]);
615 
616 		adev->gfx.rlc.register_restore = adev->gfx.rlc.register_list_format + i;
617 
618 		tmp = (unsigned int *)((uintptr_t)rlc_hdr +
619 							   le32_to_cpu(rlc_hdr->reg_list_array_offset_bytes));
620 		for (i = 0 ; i < (rlc_hdr->reg_list_size_bytes >> 2); i++)
621 			adev->gfx.rlc.register_restore[i] = le32_to_cpu(tmp[i]);
622 
623 		if (version_major == 2) {
624 			if (version_minor >= 1)
625 				gfx_v11_0_init_rlc_ext_microcode(adev);
626 			if (version_minor >= 2)
627 				gfx_v11_0_init_rlc_iram_dram_microcode(adev);
628 			if (version_minor == 3)
629 				gfx_v11_0_init_rlcp_rlcv_microcode(adev);
630 		}
631 	}
632 
633 	snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec.bin", ucode_prefix);
634 	err = request_firmware(&adev->gfx.mec_fw, fw_name, adev->dev);
635 	if (err)
636 		goto out;
637 	err = amdgpu_ucode_validate(adev->gfx.mec_fw);
638 	if (err)
639 		goto out;
640 	if (adev->gfx.rs64_enable) {
641 		cp_hdr_v2_0 = (const struct gfx_firmware_header_v2_0 *)adev->gfx.mec_fw->data;
642 		adev->gfx.mec_fw_version = le32_to_cpu(cp_hdr_v2_0->header.ucode_version);
643 		adev->gfx.mec_feature_version = le32_to_cpu(cp_hdr_v2_0->ucode_feature_version);
644 
645 	} else {
646 		cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data;
647 		adev->gfx.mec_fw_version = le32_to_cpu(cp_hdr->header.ucode_version);
648 		adev->gfx.mec_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version);
649 	}
650 
651 	/* only one MEC for gfx 11.0.0. */
652 	adev->gfx.mec2_fw = NULL;
653 
654 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
655 		if (adev->gfx.rs64_enable) {
656 			cp_hdr_v2_0 = (const struct gfx_firmware_header_v2_0 *)adev->gfx.pfp_fw->data;
657 			info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_RS64_PFP];
658 			info->ucode_id = AMDGPU_UCODE_ID_CP_RS64_PFP;
659 			info->fw = adev->gfx.pfp_fw;
660 			header = (const struct common_firmware_header *)info->fw->data;
661 			adev->firmware.fw_size +=
662 				ALIGN(le32_to_cpu(cp_hdr_v2_0->ucode_size_bytes), PAGE_SIZE);
663 
664 			info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_RS64_PFP_P0_STACK];
665 			info->ucode_id = AMDGPU_UCODE_ID_CP_RS64_PFP_P0_STACK;
666 			info->fw = adev->gfx.pfp_fw;
667 			header = (const struct common_firmware_header *)info->fw->data;
668 			adev->firmware.fw_size +=
669 				ALIGN(le32_to_cpu(cp_hdr_v2_0->data_size_bytes), PAGE_SIZE);
670 
671 			info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_RS64_PFP_P1_STACK];
672 			info->ucode_id = AMDGPU_UCODE_ID_CP_RS64_PFP_P1_STACK;
673 			info->fw = adev->gfx.pfp_fw;
674 			header = (const struct common_firmware_header *)info->fw->data;
675 			adev->firmware.fw_size +=
676 				ALIGN(le32_to_cpu(cp_hdr_v2_0->data_size_bytes), PAGE_SIZE);
677 
678 			cp_hdr_v2_0 = (const struct gfx_firmware_header_v2_0 *)adev->gfx.me_fw->data;
679 			info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_RS64_ME];
680 			info->ucode_id = AMDGPU_UCODE_ID_CP_RS64_ME;
681 			info->fw = adev->gfx.me_fw;
682 			header = (const struct common_firmware_header *)info->fw->data;
683 			adev->firmware.fw_size +=
684 				ALIGN(le32_to_cpu(cp_hdr_v2_0->ucode_size_bytes), PAGE_SIZE);
685 
686 			info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_RS64_ME_P0_STACK];
687 			info->ucode_id = AMDGPU_UCODE_ID_CP_RS64_ME_P0_STACK;
688 			info->fw = adev->gfx.me_fw;
689 			header = (const struct common_firmware_header *)info->fw->data;
690 			adev->firmware.fw_size +=
691 				ALIGN(le32_to_cpu(cp_hdr_v2_0->data_size_bytes), PAGE_SIZE);
692 
693 			info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_RS64_ME_P1_STACK];
694 			info->ucode_id = AMDGPU_UCODE_ID_CP_RS64_ME_P1_STACK;
695 			info->fw = adev->gfx.me_fw;
696 			header = (const struct common_firmware_header *)info->fw->data;
697 			adev->firmware.fw_size +=
698 				ALIGN(le32_to_cpu(cp_hdr_v2_0->data_size_bytes), PAGE_SIZE);
699 
700 			cp_hdr_v2_0 = (const struct gfx_firmware_header_v2_0 *)adev->gfx.mec_fw->data;
701 			info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_RS64_MEC];
702 			info->ucode_id = AMDGPU_UCODE_ID_CP_RS64_MEC;
703 			info->fw = adev->gfx.mec_fw;
704 			header = (const struct common_firmware_header *)info->fw->data;
705 			adev->firmware.fw_size +=
706 				ALIGN(le32_to_cpu(cp_hdr_v2_0->ucode_size_bytes), PAGE_SIZE);
707 
708 			info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_RS64_MEC_P0_STACK];
709 			info->ucode_id = AMDGPU_UCODE_ID_CP_RS64_MEC_P0_STACK;
710 			info->fw = adev->gfx.mec_fw;
711 			header = (const struct common_firmware_header *)info->fw->data;
712 			adev->firmware.fw_size +=
713 				ALIGN(le32_to_cpu(cp_hdr_v2_0->data_size_bytes), PAGE_SIZE);
714 
715 			info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_RS64_MEC_P1_STACK];
716 			info->ucode_id = AMDGPU_UCODE_ID_CP_RS64_MEC_P1_STACK;
717 			info->fw = adev->gfx.mec_fw;
718 			header = (const struct common_firmware_header *)info->fw->data;
719 			adev->firmware.fw_size +=
720 				ALIGN(le32_to_cpu(cp_hdr_v2_0->data_size_bytes), PAGE_SIZE);
721 
722 			info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_RS64_MEC_P2_STACK];
723 			info->ucode_id = AMDGPU_UCODE_ID_CP_RS64_MEC_P2_STACK;
724 			info->fw = adev->gfx.mec_fw;
725 			header = (const struct common_firmware_header *)info->fw->data;
726 			adev->firmware.fw_size +=
727 				ALIGN(le32_to_cpu(cp_hdr_v2_0->data_size_bytes), PAGE_SIZE);
728 
729 			info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_RS64_MEC_P3_STACK];
730 			info->ucode_id = AMDGPU_UCODE_ID_CP_RS64_MEC_P3_STACK;
731 			info->fw = adev->gfx.mec_fw;
732 			header = (const struct common_firmware_header *)info->fw->data;
733 			adev->firmware.fw_size +=
734 				ALIGN(le32_to_cpu(cp_hdr_v2_0->data_size_bytes), PAGE_SIZE);
735 		} else {
736 			info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_PFP];
737 			info->ucode_id = AMDGPU_UCODE_ID_CP_PFP;
738 			info->fw = adev->gfx.pfp_fw;
739 			header = (const struct common_firmware_header *)info->fw->data;
740 			adev->firmware.fw_size +=
741 				ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
742 
743 			info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_ME];
744 			info->ucode_id = AMDGPU_UCODE_ID_CP_ME;
745 			info->fw = adev->gfx.me_fw;
746 			header = (const struct common_firmware_header *)info->fw->data;
747 			adev->firmware.fw_size +=
748 				ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
749 
750 			info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_MEC1];
751 			info->ucode_id = AMDGPU_UCODE_ID_CP_MEC1;
752 			info->fw = adev->gfx.mec_fw;
753 			header = (const struct common_firmware_header *)info->fw->data;
754 			cp_hdr = (const struct gfx_firmware_header_v1_0 *)info->fw->data;
755 			adev->firmware.fw_size +=
756 				ALIGN(le32_to_cpu(header->ucode_size_bytes) -
757 				      le32_to_cpu(cp_hdr->jt_size) * 4, PAGE_SIZE);
758 
759 			info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_MEC1_JT];
760 			info->ucode_id = AMDGPU_UCODE_ID_CP_MEC1_JT;
761 			info->fw = adev->gfx.mec_fw;
762 			adev->firmware.fw_size +=
763 				ALIGN(le32_to_cpu(cp_hdr->jt_size) * 4, PAGE_SIZE);
764 		}
765 
766 		info = &adev->firmware.ucode[AMDGPU_UCODE_ID_RLC_G];
767 		info->ucode_id = AMDGPU_UCODE_ID_RLC_G;
768 		info->fw = adev->gfx.rlc_fw;
769 		if (info->fw) {
770 			header = (const struct common_firmware_header *)info->fw->data;
771 			adev->firmware.fw_size +=
772 				ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
773 		}
774 		if (adev->gfx.rlc.save_restore_list_gpm_size_bytes &&
775 		    adev->gfx.rlc.save_restore_list_srm_size_bytes) {
776 			info = &adev->firmware.ucode[AMDGPU_UCODE_ID_RLC_RESTORE_LIST_GPM_MEM];
777 			info->ucode_id = AMDGPU_UCODE_ID_RLC_RESTORE_LIST_GPM_MEM;
778 			info->fw = adev->gfx.rlc_fw;
779 			adev->firmware.fw_size +=
780 				ALIGN(adev->gfx.rlc.save_restore_list_gpm_size_bytes, PAGE_SIZE);
781 
782 			info = &adev->firmware.ucode[AMDGPU_UCODE_ID_RLC_RESTORE_LIST_SRM_MEM];
783 			info->ucode_id = AMDGPU_UCODE_ID_RLC_RESTORE_LIST_SRM_MEM;
784 			info->fw = adev->gfx.rlc_fw;
785 			adev->firmware.fw_size +=
786 				ALIGN(adev->gfx.rlc.save_restore_list_srm_size_bytes, PAGE_SIZE);
787 		}
788 
789 		if (adev->gfx.rlc.rlc_iram_ucode_size_bytes &&
790 		    adev->gfx.rlc.rlc_dram_ucode_size_bytes) {
791 			info = &adev->firmware.ucode[AMDGPU_UCODE_ID_RLC_IRAM];
792 			info->ucode_id = AMDGPU_UCODE_ID_RLC_IRAM;
793 			info->fw = adev->gfx.rlc_fw;
794 			adev->firmware.fw_size +=
795 				ALIGN(adev->gfx.rlc.rlc_iram_ucode_size_bytes, PAGE_SIZE);
796 
797 			info = &adev->firmware.ucode[AMDGPU_UCODE_ID_RLC_DRAM];
798 			info->ucode_id = AMDGPU_UCODE_ID_RLC_DRAM;
799 			info->fw = adev->gfx.rlc_fw;
800 			adev->firmware.fw_size +=
801 				ALIGN(adev->gfx.rlc.rlc_dram_ucode_size_bytes, PAGE_SIZE);
802 		}
803 
804 		if (adev->gfx.rlc.rlcp_ucode_size_bytes) {
805 			info = &adev->firmware.ucode[AMDGPU_UCODE_ID_RLC_P];
806 			info->ucode_id = AMDGPU_UCODE_ID_RLC_P;
807 			info->fw = adev->gfx.rlc_fw;
808 			adev->firmware.fw_size +=
809 				ALIGN(adev->gfx.rlc.rlcp_ucode_size_bytes, PAGE_SIZE);
810 		}
811 
812 		if (adev->gfx.rlc.rlcv_ucode_size_bytes) {
813 			info = &adev->firmware.ucode[AMDGPU_UCODE_ID_RLC_V];
814 			info->ucode_id = AMDGPU_UCODE_ID_RLC_V;
815 			info->fw = adev->gfx.rlc_fw;
816 			adev->firmware.fw_size +=
817 				ALIGN(adev->gfx.rlc.rlcv_ucode_size_bytes, PAGE_SIZE);
818 		}
819 	}
820 
821 out:
822 	if (err) {
823 		dev_err(adev->dev,
824 			"gfx11: Failed to load firmware \"%s\"\n",
825 			fw_name);
826 		release_firmware(adev->gfx.pfp_fw);
827 		adev->gfx.pfp_fw = NULL;
828 		release_firmware(adev->gfx.me_fw);
829 		adev->gfx.me_fw = NULL;
830 		release_firmware(adev->gfx.rlc_fw);
831 		adev->gfx.rlc_fw = NULL;
832 		release_firmware(adev->gfx.mec_fw);
833 		adev->gfx.mec_fw = NULL;
834 	}
835 
836 	return err;
837 }
838 
839 static int gfx_v11_0_init_toc_microcode(struct amdgpu_device *adev)
840 {
841 	const struct psp_firmware_header_v1_0 *toc_hdr;
842 	int err = 0;
843 	char fw_name[40];
844 	char ucode_prefix[30];
845 
846 	amdgpu_ucode_ip_version_decode(adev, GC_HWIP, ucode_prefix, sizeof(ucode_prefix));
847 
848 	snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_toc.bin", ucode_prefix);
849 	err = request_firmware(&adev->psp.toc_fw, fw_name, adev->dev);
850 	if (err)
851 		goto out;
852 
853 	err = amdgpu_ucode_validate(adev->psp.toc_fw);
854 	if (err)
855 		goto out;
856 
857 	toc_hdr = (const struct psp_firmware_header_v1_0 *)adev->psp.toc_fw->data;
858 	adev->psp.toc.fw_version = le32_to_cpu(toc_hdr->header.ucode_version);
859 	adev->psp.toc.feature_version = le32_to_cpu(toc_hdr->sos.fw_version);
860 	adev->psp.toc.size_bytes = le32_to_cpu(toc_hdr->header.ucode_size_bytes);
861 	adev->psp.toc.start_addr = (uint8_t *)toc_hdr +
862 				le32_to_cpu(toc_hdr->header.ucode_array_offset_bytes);
863 	return 0;
864 out:
865 	dev_err(adev->dev, "Failed to load TOC microcode\n");
866 	release_firmware(adev->psp.toc_fw);
867 	adev->psp.toc_fw = NULL;
868 	return err;
869 }
870 
871 static u32 gfx_v11_0_get_csb_size(struct amdgpu_device *adev)
872 {
873 	u32 count = 0;
874 	const struct cs_section_def *sect = NULL;
875 	const struct cs_extent_def *ext = NULL;
876 
877 	/* begin clear state */
878 	count += 2;
879 	/* context control state */
880 	count += 3;
881 
882 	for (sect = gfx11_cs_data; sect->section != NULL; ++sect) {
883 		for (ext = sect->section; ext->extent != NULL; ++ext) {
884 			if (sect->id == SECT_CONTEXT)
885 				count += 2 + ext->reg_count;
886 			else
887 				return 0;
888 		}
889 	}
890 
891 	/* set PA_SC_TILE_STEERING_OVERRIDE */
892 	count += 3;
893 	/* end clear state */
894 	count += 2;
895 	/* clear state */
896 	count += 2;
897 
898 	return count;
899 }
900 
901 static void gfx_v11_0_get_csb_buffer(struct amdgpu_device *adev,
902 				    volatile u32 *buffer)
903 {
904 	u32 count = 0, i;
905 	const struct cs_section_def *sect = NULL;
906 	const struct cs_extent_def *ext = NULL;
907 	int ctx_reg_offset;
908 
909 	if (adev->gfx.rlc.cs_data == NULL)
910 		return;
911 	if (buffer == NULL)
912 		return;
913 
914 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0));
915 	buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
916 
917 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CONTEXT_CONTROL, 1));
918 	buffer[count++] = cpu_to_le32(0x80000000);
919 	buffer[count++] = cpu_to_le32(0x80000000);
920 
921 	for (sect = adev->gfx.rlc.cs_data; sect->section != NULL; ++sect) {
922 		for (ext = sect->section; ext->extent != NULL; ++ext) {
923 			if (sect->id == SECT_CONTEXT) {
924 				buffer[count++] =
925 					cpu_to_le32(PACKET3(PACKET3_SET_CONTEXT_REG, ext->reg_count));
926 				buffer[count++] = cpu_to_le32(ext->reg_index -
927 						PACKET3_SET_CONTEXT_REG_START);
928 				for (i = 0; i < ext->reg_count; i++)
929 					buffer[count++] = cpu_to_le32(ext->extent[i]);
930 			} else {
931 				return;
932 			}
933 		}
934 	}
935 
936 	ctx_reg_offset =
937 		SOC15_REG_OFFSET(GC, 0, regPA_SC_TILE_STEERING_OVERRIDE) - PACKET3_SET_CONTEXT_REG_START;
938 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_SET_CONTEXT_REG, 1));
939 	buffer[count++] = cpu_to_le32(ctx_reg_offset);
940 	buffer[count++] = cpu_to_le32(adev->gfx.config.pa_sc_tile_steering_override);
941 
942 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0));
943 	buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_END_CLEAR_STATE);
944 
945 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CLEAR_STATE, 0));
946 	buffer[count++] = cpu_to_le32(0);
947 }
948 
949 static void gfx_v11_0_rlc_fini(struct amdgpu_device *adev)
950 {
951 	/* clear state block */
952 	amdgpu_bo_free_kernel(&adev->gfx.rlc.clear_state_obj,
953 			&adev->gfx.rlc.clear_state_gpu_addr,
954 			(void **)&adev->gfx.rlc.cs_ptr);
955 
956 	/* jump table block */
957 	amdgpu_bo_free_kernel(&adev->gfx.rlc.cp_table_obj,
958 			&adev->gfx.rlc.cp_table_gpu_addr,
959 			(void **)&adev->gfx.rlc.cp_table_ptr);
960 }
961 
962 static void gfx_v11_0_init_rlcg_reg_access_ctrl(struct amdgpu_device *adev)
963 {
964 	struct amdgpu_rlcg_reg_access_ctrl *reg_access_ctrl;
965 
966 	reg_access_ctrl = &adev->gfx.rlc.reg_access_ctrl;
967 	reg_access_ctrl->scratch_reg0 = SOC15_REG_OFFSET(GC, 0, regSCRATCH_REG0);
968 	reg_access_ctrl->scratch_reg1 = SOC15_REG_OFFSET(GC, 0, regSCRATCH_REG1);
969 	reg_access_ctrl->scratch_reg2 = SOC15_REG_OFFSET(GC, 0, regSCRATCH_REG2);
970 	reg_access_ctrl->scratch_reg3 = SOC15_REG_OFFSET(GC, 0, regSCRATCH_REG3);
971 	reg_access_ctrl->grbm_cntl = SOC15_REG_OFFSET(GC, 0, regGRBM_GFX_CNTL);
972 	reg_access_ctrl->grbm_idx = SOC15_REG_OFFSET(GC, 0, regGRBM_GFX_INDEX);
973 	reg_access_ctrl->spare_int = SOC15_REG_OFFSET(GC, 0, regRLC_SPARE_INT_0);
974 	adev->gfx.rlc.rlcg_reg_access_supported = true;
975 }
976 
977 static int gfx_v11_0_rlc_init(struct amdgpu_device *adev)
978 {
979 	const struct cs_section_def *cs_data;
980 	int r;
981 
982 	adev->gfx.rlc.cs_data = gfx11_cs_data;
983 
984 	cs_data = adev->gfx.rlc.cs_data;
985 
986 	if (cs_data) {
987 		/* init clear state block */
988 		r = amdgpu_gfx_rlc_init_csb(adev);
989 		if (r)
990 			return r;
991 	}
992 
993 	/* init spm vmid with 0xf */
994 	if (adev->gfx.rlc.funcs->update_spm_vmid)
995 		adev->gfx.rlc.funcs->update_spm_vmid(adev, 0xf);
996 
997 	return 0;
998 }
999 
1000 static void gfx_v11_0_mec_fini(struct amdgpu_device *adev)
1001 {
1002 	amdgpu_bo_free_kernel(&adev->gfx.mec.hpd_eop_obj, NULL, NULL);
1003 	amdgpu_bo_free_kernel(&adev->gfx.mec.mec_fw_obj, NULL, NULL);
1004 	amdgpu_bo_free_kernel(&adev->gfx.mec.mec_fw_data_obj, NULL, NULL);
1005 }
1006 
1007 static int gfx_v11_0_me_init(struct amdgpu_device *adev)
1008 {
1009 	int r;
1010 
1011 	bitmap_zero(adev->gfx.me.queue_bitmap, AMDGPU_MAX_GFX_QUEUES);
1012 
1013 	amdgpu_gfx_graphics_queue_acquire(adev);
1014 
1015 	r = gfx_v11_0_init_microcode(adev);
1016 	if (r)
1017 		DRM_ERROR("Failed to load gfx firmware!\n");
1018 
1019 	return r;
1020 }
1021 
1022 static int gfx_v11_0_mec_init(struct amdgpu_device *adev)
1023 {
1024 	int r;
1025 	u32 *hpd;
1026 	size_t mec_hpd_size;
1027 
1028 	bitmap_zero(adev->gfx.mec.queue_bitmap, AMDGPU_MAX_COMPUTE_QUEUES);
1029 
1030 	/* take ownership of the relevant compute queues */
1031 	amdgpu_gfx_compute_queue_acquire(adev);
1032 	mec_hpd_size = adev->gfx.num_compute_rings * GFX11_MEC_HPD_SIZE;
1033 
1034 	if (mec_hpd_size) {
1035 		r = amdgpu_bo_create_reserved(adev, mec_hpd_size, PAGE_SIZE,
1036 					      AMDGPU_GEM_DOMAIN_GTT,
1037 					      &adev->gfx.mec.hpd_eop_obj,
1038 					      &adev->gfx.mec.hpd_eop_gpu_addr,
1039 					      (void **)&hpd);
1040 		if (r) {
1041 			dev_warn(adev->dev, "(%d) create HDP EOP bo failed\n", r);
1042 			gfx_v11_0_mec_fini(adev);
1043 			return r;
1044 		}
1045 
1046 		memset(hpd, 0, mec_hpd_size);
1047 
1048 		amdgpu_bo_kunmap(adev->gfx.mec.hpd_eop_obj);
1049 		amdgpu_bo_unreserve(adev->gfx.mec.hpd_eop_obj);
1050 	}
1051 
1052 	return 0;
1053 }
1054 
1055 static uint32_t wave_read_ind(struct amdgpu_device *adev, uint32_t wave, uint32_t address)
1056 {
1057 	WREG32_SOC15(GC, 0, regSQ_IND_INDEX,
1058 		(wave << SQ_IND_INDEX__WAVE_ID__SHIFT) |
1059 		(address << SQ_IND_INDEX__INDEX__SHIFT));
1060 	return RREG32_SOC15(GC, 0, regSQ_IND_DATA);
1061 }
1062 
1063 static void wave_read_regs(struct amdgpu_device *adev, uint32_t wave,
1064 			   uint32_t thread, uint32_t regno,
1065 			   uint32_t num, uint32_t *out)
1066 {
1067 	WREG32_SOC15(GC, 0, regSQ_IND_INDEX,
1068 		(wave << SQ_IND_INDEX__WAVE_ID__SHIFT) |
1069 		(regno << SQ_IND_INDEX__INDEX__SHIFT) |
1070 		(thread << SQ_IND_INDEX__WORKITEM_ID__SHIFT) |
1071 		(SQ_IND_INDEX__AUTO_INCR_MASK));
1072 	while (num--)
1073 		*(out++) = RREG32_SOC15(GC, 0, regSQ_IND_DATA);
1074 }
1075 
1076 static void gfx_v11_0_read_wave_data(struct amdgpu_device *adev, uint32_t simd, uint32_t wave, uint32_t *dst, int *no_fields)
1077 {
1078 	/* in gfx11 the SIMD_ID is specified as part of the INSTANCE
1079 	 * field when performing a select_se_sh so it should be
1080 	 * zero here */
1081 	WARN_ON(simd != 0);
1082 
1083 	/* type 2 wave data */
1084 	dst[(*no_fields)++] = 2;
1085 	dst[(*no_fields)++] = wave_read_ind(adev, wave, ixSQ_WAVE_STATUS);
1086 	dst[(*no_fields)++] = wave_read_ind(adev, wave, ixSQ_WAVE_PC_LO);
1087 	dst[(*no_fields)++] = wave_read_ind(adev, wave, ixSQ_WAVE_PC_HI);
1088 	dst[(*no_fields)++] = wave_read_ind(adev, wave, ixSQ_WAVE_EXEC_LO);
1089 	dst[(*no_fields)++] = wave_read_ind(adev, wave, ixSQ_WAVE_EXEC_HI);
1090 	dst[(*no_fields)++] = wave_read_ind(adev, wave, ixSQ_WAVE_HW_ID1);
1091 	dst[(*no_fields)++] = wave_read_ind(adev, wave, ixSQ_WAVE_HW_ID2);
1092 	dst[(*no_fields)++] = wave_read_ind(adev, wave, ixSQ_WAVE_GPR_ALLOC);
1093 	dst[(*no_fields)++] = wave_read_ind(adev, wave, ixSQ_WAVE_LDS_ALLOC);
1094 	dst[(*no_fields)++] = wave_read_ind(adev, wave, ixSQ_WAVE_TRAPSTS);
1095 	dst[(*no_fields)++] = wave_read_ind(adev, wave, ixSQ_WAVE_IB_STS);
1096 	dst[(*no_fields)++] = wave_read_ind(adev, wave, ixSQ_WAVE_IB_STS2);
1097 	dst[(*no_fields)++] = wave_read_ind(adev, wave, ixSQ_WAVE_IB_DBG1);
1098 	dst[(*no_fields)++] = wave_read_ind(adev, wave, ixSQ_WAVE_M0);
1099 	dst[(*no_fields)++] = wave_read_ind(adev, wave, ixSQ_WAVE_MODE);
1100 }
1101 
1102 static void gfx_v11_0_read_wave_sgprs(struct amdgpu_device *adev, uint32_t simd,
1103 				     uint32_t wave, uint32_t start,
1104 				     uint32_t size, uint32_t *dst)
1105 {
1106 	WARN_ON(simd != 0);
1107 
1108 	wave_read_regs(
1109 		adev, wave, 0, start + SQIND_WAVE_SGPRS_OFFSET, size,
1110 		dst);
1111 }
1112 
1113 static void gfx_v11_0_read_wave_vgprs(struct amdgpu_device *adev, uint32_t simd,
1114 				      uint32_t wave, uint32_t thread,
1115 				      uint32_t start, uint32_t size,
1116 				      uint32_t *dst)
1117 {
1118 	wave_read_regs(
1119 		adev, wave, thread,
1120 		start + SQIND_WAVE_VGPRS_OFFSET, size, dst);
1121 }
1122 
1123 static void gfx_v11_0_select_me_pipe_q(struct amdgpu_device *adev,
1124 									  u32 me, u32 pipe, u32 q, u32 vm)
1125 {
1126 	soc21_grbm_select(adev, me, pipe, q, vm);
1127 }
1128 
1129 static const struct amdgpu_gfx_funcs gfx_v11_0_gfx_funcs = {
1130 	.get_gpu_clock_counter = &gfx_v11_0_get_gpu_clock_counter,
1131 	.select_se_sh = &gfx_v11_0_select_se_sh,
1132 	.read_wave_data = &gfx_v11_0_read_wave_data,
1133 	.read_wave_sgprs = &gfx_v11_0_read_wave_sgprs,
1134 	.read_wave_vgprs = &gfx_v11_0_read_wave_vgprs,
1135 	.select_me_pipe_q = &gfx_v11_0_select_me_pipe_q,
1136 	.init_spm_golden = &gfx_v11_0_init_spm_golden_registers,
1137 };
1138 
1139 static int gfx_v11_0_gpu_early_init(struct amdgpu_device *adev)
1140 {
1141 	adev->gfx.funcs = &gfx_v11_0_gfx_funcs;
1142 
1143 	switch (adev->ip_versions[GC_HWIP][0]) {
1144 	case IP_VERSION(11, 0, 0):
1145 	case IP_VERSION(11, 0, 2):
1146 		adev->gfx.config.max_hw_contexts = 8;
1147 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1148 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1149 		adev->gfx.config.sc_hiz_tile_fifo_size = 0;
1150 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x4C0;
1151 		break;
1152 	case IP_VERSION(11, 0, 1):
1153 		adev->gfx.config.max_hw_contexts = 8;
1154 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1155 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1156 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x80;
1157 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x300;
1158 		break;
1159 	default:
1160 		BUG();
1161 		break;
1162 	}
1163 
1164 	return 0;
1165 }
1166 
1167 static int gfx_v11_0_gfx_ring_init(struct amdgpu_device *adev, int ring_id,
1168 				   int me, int pipe, int queue)
1169 {
1170 	int r;
1171 	struct amdgpu_ring *ring;
1172 	unsigned int irq_type;
1173 
1174 	ring = &adev->gfx.gfx_ring[ring_id];
1175 
1176 	ring->me = me;
1177 	ring->pipe = pipe;
1178 	ring->queue = queue;
1179 
1180 	ring->ring_obj = NULL;
1181 	ring->use_doorbell = true;
1182 
1183 	if (!ring_id)
1184 		ring->doorbell_index = adev->doorbell_index.gfx_ring0 << 1;
1185 	else
1186 		ring->doorbell_index = adev->doorbell_index.gfx_ring1 << 1;
1187 	sprintf(ring->name, "gfx_%d.%d.%d", ring->me, ring->pipe, ring->queue);
1188 
1189 	irq_type = AMDGPU_CP_IRQ_GFX_ME0_PIPE0_EOP + ring->pipe;
1190 	r = amdgpu_ring_init(adev, ring, 1024, &adev->gfx.eop_irq, irq_type,
1191 			     AMDGPU_RING_PRIO_DEFAULT, NULL);
1192 	if (r)
1193 		return r;
1194 	return 0;
1195 }
1196 
1197 static int gfx_v11_0_compute_ring_init(struct amdgpu_device *adev, int ring_id,
1198 				       int mec, int pipe, int queue)
1199 {
1200 	int r;
1201 	unsigned irq_type;
1202 	struct amdgpu_ring *ring;
1203 	unsigned int hw_prio;
1204 
1205 	ring = &adev->gfx.compute_ring[ring_id];
1206 
1207 	/* mec0 is me1 */
1208 	ring->me = mec + 1;
1209 	ring->pipe = pipe;
1210 	ring->queue = queue;
1211 
1212 	ring->ring_obj = NULL;
1213 	ring->use_doorbell = true;
1214 	ring->doorbell_index = (adev->doorbell_index.mec_ring0 + ring_id) << 1;
1215 	ring->eop_gpu_addr = adev->gfx.mec.hpd_eop_gpu_addr
1216 				+ (ring_id * GFX11_MEC_HPD_SIZE);
1217 	sprintf(ring->name, "comp_%d.%d.%d", ring->me, ring->pipe, ring->queue);
1218 
1219 	irq_type = AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP
1220 		+ ((ring->me - 1) * adev->gfx.mec.num_pipe_per_mec)
1221 		+ ring->pipe;
1222 	hw_prio = amdgpu_gfx_is_high_priority_compute_queue(adev, ring) ?
1223 			AMDGPU_GFX_PIPE_PRIO_HIGH : AMDGPU_GFX_PIPE_PRIO_NORMAL;
1224 	/* type-2 packets are deprecated on MEC, use type-3 instead */
1225 	r = amdgpu_ring_init(adev, ring, 1024, &adev->gfx.eop_irq, irq_type,
1226 			     hw_prio, NULL);
1227 	if (r)
1228 		return r;
1229 
1230 	return 0;
1231 }
1232 
1233 static struct {
1234 	SOC21_FIRMWARE_ID	id;
1235 	unsigned int		offset;
1236 	unsigned int		size;
1237 } rlc_autoload_info[SOC21_FIRMWARE_ID_MAX];
1238 
1239 static void gfx_v11_0_parse_rlc_toc(struct amdgpu_device *adev, void *rlc_toc)
1240 {
1241 	RLC_TABLE_OF_CONTENT *ucode = rlc_toc;
1242 
1243 	while (ucode && (ucode->id > SOC21_FIRMWARE_ID_INVALID) &&
1244 			(ucode->id < SOC21_FIRMWARE_ID_MAX)) {
1245 		rlc_autoload_info[ucode->id].id = ucode->id;
1246 		rlc_autoload_info[ucode->id].offset = ucode->offset * 4;
1247 		rlc_autoload_info[ucode->id].size = ucode->size * 4;
1248 
1249 		ucode++;
1250 	}
1251 }
1252 
1253 static uint32_t gfx_v11_0_calc_toc_total_size(struct amdgpu_device *adev)
1254 {
1255 	uint32_t total_size = 0;
1256 	SOC21_FIRMWARE_ID id;
1257 
1258 	gfx_v11_0_parse_rlc_toc(adev, adev->psp.toc.start_addr);
1259 
1260 	for (id = SOC21_FIRMWARE_ID_RLC_G_UCODE; id < SOC21_FIRMWARE_ID_MAX; id++)
1261 		total_size += rlc_autoload_info[id].size;
1262 
1263 	/* In case the offset in rlc toc ucode is aligned */
1264 	if (total_size < rlc_autoload_info[SOC21_FIRMWARE_ID_MAX-1].offset)
1265 		total_size = rlc_autoload_info[SOC21_FIRMWARE_ID_MAX-1].offset +
1266 			rlc_autoload_info[SOC21_FIRMWARE_ID_MAX-1].size;
1267 
1268 	return total_size;
1269 }
1270 
1271 static int gfx_v11_0_rlc_autoload_buffer_init(struct amdgpu_device *adev)
1272 {
1273 	int r;
1274 	uint32_t total_size;
1275 
1276 	total_size = gfx_v11_0_calc_toc_total_size(adev);
1277 
1278 	r = amdgpu_bo_create_reserved(adev, total_size, 64 * 1024,
1279 			AMDGPU_GEM_DOMAIN_VRAM,
1280 			&adev->gfx.rlc.rlc_autoload_bo,
1281 			&adev->gfx.rlc.rlc_autoload_gpu_addr,
1282 			(void **)&adev->gfx.rlc.rlc_autoload_ptr);
1283 
1284 	if (r) {
1285 		dev_err(adev->dev, "(%d) failed to create fw autoload bo\n", r);
1286 		return r;
1287 	}
1288 
1289 	return 0;
1290 }
1291 
1292 static void gfx_v11_0_rlc_backdoor_autoload_copy_ucode(struct amdgpu_device *adev,
1293 					      SOC21_FIRMWARE_ID id,
1294 			    		      const void *fw_data,
1295 					      uint32_t fw_size,
1296 					      uint32_t *fw_autoload_mask)
1297 {
1298 	uint32_t toc_offset;
1299 	uint32_t toc_fw_size;
1300 	char *ptr = adev->gfx.rlc.rlc_autoload_ptr;
1301 
1302 	if (id <= SOC21_FIRMWARE_ID_INVALID || id >= SOC21_FIRMWARE_ID_MAX)
1303 		return;
1304 
1305 	toc_offset = rlc_autoload_info[id].offset;
1306 	toc_fw_size = rlc_autoload_info[id].size;
1307 
1308 	if (fw_size == 0)
1309 		fw_size = toc_fw_size;
1310 
1311 	if (fw_size > toc_fw_size)
1312 		fw_size = toc_fw_size;
1313 
1314 	memcpy(ptr + toc_offset, fw_data, fw_size);
1315 
1316 	if (fw_size < toc_fw_size)
1317 		memset(ptr + toc_offset + fw_size, 0, toc_fw_size - fw_size);
1318 
1319 	if ((id != SOC21_FIRMWARE_ID_RS64_PFP) && (id != SOC21_FIRMWARE_ID_RS64_ME))
1320 		*(uint64_t *)fw_autoload_mask |= 1ULL << id;
1321 }
1322 
1323 static void gfx_v11_0_rlc_backdoor_autoload_copy_toc_ucode(struct amdgpu_device *adev,
1324 							uint32_t *fw_autoload_mask)
1325 {
1326 	void *data;
1327 	uint32_t size;
1328 	uint64_t *toc_ptr;
1329 
1330 	*(uint64_t *)fw_autoload_mask |= 0x1;
1331 
1332 	DRM_DEBUG("rlc autoload enabled fw: 0x%llx\n", *(uint64_t *)fw_autoload_mask);
1333 
1334 	data = adev->psp.toc.start_addr;
1335 	size = rlc_autoload_info[SOC21_FIRMWARE_ID_RLC_TOC].size;
1336 
1337 	toc_ptr = (uint64_t *)data + size / 8 - 1;
1338 	*toc_ptr = *(uint64_t *)fw_autoload_mask;
1339 
1340 	gfx_v11_0_rlc_backdoor_autoload_copy_ucode(adev, SOC21_FIRMWARE_ID_RLC_TOC,
1341 					data, size, fw_autoload_mask);
1342 }
1343 
1344 static void gfx_v11_0_rlc_backdoor_autoload_copy_gfx_ucode(struct amdgpu_device *adev,
1345 							uint32_t *fw_autoload_mask)
1346 {
1347 	const __le32 *fw_data;
1348 	uint32_t fw_size;
1349 	const struct gfx_firmware_header_v1_0 *cp_hdr;
1350 	const struct gfx_firmware_header_v2_0 *cpv2_hdr;
1351 	const struct rlc_firmware_header_v2_0 *rlc_hdr;
1352 	const struct rlc_firmware_header_v2_2 *rlcv22_hdr;
1353 	uint16_t version_major, version_minor;
1354 
1355 	if (adev->gfx.rs64_enable) {
1356 		/* pfp ucode */
1357 		cpv2_hdr = (const struct gfx_firmware_header_v2_0 *)
1358 			adev->gfx.pfp_fw->data;
1359 		/* instruction */
1360 		fw_data = (const __le32 *)(adev->gfx.pfp_fw->data +
1361 			le32_to_cpu(cpv2_hdr->ucode_offset_bytes));
1362 		fw_size = le32_to_cpu(cpv2_hdr->ucode_size_bytes);
1363 		gfx_v11_0_rlc_backdoor_autoload_copy_ucode(adev, SOC21_FIRMWARE_ID_RS64_PFP,
1364 						fw_data, fw_size, fw_autoload_mask);
1365 		/* data */
1366 		fw_data = (const __le32 *)(adev->gfx.pfp_fw->data +
1367 			le32_to_cpu(cpv2_hdr->data_offset_bytes));
1368 		fw_size = le32_to_cpu(cpv2_hdr->data_size_bytes);
1369 		gfx_v11_0_rlc_backdoor_autoload_copy_ucode(adev, SOC21_FIRMWARE_ID_RS64_PFP_P0_STACK,
1370 						fw_data, fw_size, fw_autoload_mask);
1371 		gfx_v11_0_rlc_backdoor_autoload_copy_ucode(adev, SOC21_FIRMWARE_ID_RS64_PFP_P1_STACK,
1372 						fw_data, fw_size, fw_autoload_mask);
1373 		/* me ucode */
1374 		cpv2_hdr = (const struct gfx_firmware_header_v2_0 *)
1375 			adev->gfx.me_fw->data;
1376 		/* instruction */
1377 		fw_data = (const __le32 *)(adev->gfx.me_fw->data +
1378 			le32_to_cpu(cpv2_hdr->ucode_offset_bytes));
1379 		fw_size = le32_to_cpu(cpv2_hdr->ucode_size_bytes);
1380 		gfx_v11_0_rlc_backdoor_autoload_copy_ucode(adev, SOC21_FIRMWARE_ID_RS64_ME,
1381 						fw_data, fw_size, fw_autoload_mask);
1382 		/* data */
1383 		fw_data = (const __le32 *)(adev->gfx.me_fw->data +
1384 			le32_to_cpu(cpv2_hdr->data_offset_bytes));
1385 		fw_size = le32_to_cpu(cpv2_hdr->data_size_bytes);
1386 		gfx_v11_0_rlc_backdoor_autoload_copy_ucode(adev, SOC21_FIRMWARE_ID_RS64_ME_P0_STACK,
1387 						fw_data, fw_size, fw_autoload_mask);
1388 		gfx_v11_0_rlc_backdoor_autoload_copy_ucode(adev, SOC21_FIRMWARE_ID_RS64_ME_P1_STACK,
1389 						fw_data, fw_size, fw_autoload_mask);
1390 		/* mec ucode */
1391 		cpv2_hdr = (const struct gfx_firmware_header_v2_0 *)
1392 			adev->gfx.mec_fw->data;
1393 		/* instruction */
1394 		fw_data = (const __le32 *) (adev->gfx.mec_fw->data +
1395 			le32_to_cpu(cpv2_hdr->ucode_offset_bytes));
1396 		fw_size = le32_to_cpu(cpv2_hdr->ucode_size_bytes);
1397 		gfx_v11_0_rlc_backdoor_autoload_copy_ucode(adev, SOC21_FIRMWARE_ID_RS64_MEC,
1398 						fw_data, fw_size, fw_autoload_mask);
1399 		/* data */
1400 		fw_data = (const __le32 *) (adev->gfx.mec_fw->data +
1401 			le32_to_cpu(cpv2_hdr->data_offset_bytes));
1402 		fw_size = le32_to_cpu(cpv2_hdr->data_size_bytes);
1403 		gfx_v11_0_rlc_backdoor_autoload_copy_ucode(adev, SOC21_FIRMWARE_ID_RS64_MEC_P0_STACK,
1404 						fw_data, fw_size, fw_autoload_mask);
1405 		gfx_v11_0_rlc_backdoor_autoload_copy_ucode(adev, SOC21_FIRMWARE_ID_RS64_MEC_P1_STACK,
1406 						fw_data, fw_size, fw_autoload_mask);
1407 		gfx_v11_0_rlc_backdoor_autoload_copy_ucode(adev, SOC21_FIRMWARE_ID_RS64_MEC_P2_STACK,
1408 						fw_data, fw_size, fw_autoload_mask);
1409 		gfx_v11_0_rlc_backdoor_autoload_copy_ucode(adev, SOC21_FIRMWARE_ID_RS64_MEC_P3_STACK,
1410 						fw_data, fw_size, fw_autoload_mask);
1411 	} else {
1412 		/* pfp ucode */
1413 		cp_hdr = (const struct gfx_firmware_header_v1_0 *)
1414 			adev->gfx.pfp_fw->data;
1415 		fw_data = (const __le32 *)(adev->gfx.pfp_fw->data +
1416 				le32_to_cpu(cp_hdr->header.ucode_array_offset_bytes));
1417 		fw_size = le32_to_cpu(cp_hdr->header.ucode_size_bytes);
1418 		gfx_v11_0_rlc_backdoor_autoload_copy_ucode(adev, SOC21_FIRMWARE_ID_CP_PFP,
1419 						fw_data, fw_size, fw_autoload_mask);
1420 
1421 		/* me ucode */
1422 		cp_hdr = (const struct gfx_firmware_header_v1_0 *)
1423 			adev->gfx.me_fw->data;
1424 		fw_data = (const __le32 *)(adev->gfx.me_fw->data +
1425 				le32_to_cpu(cp_hdr->header.ucode_array_offset_bytes));
1426 		fw_size = le32_to_cpu(cp_hdr->header.ucode_size_bytes);
1427 		gfx_v11_0_rlc_backdoor_autoload_copy_ucode(adev, SOC21_FIRMWARE_ID_CP_ME,
1428 						fw_data, fw_size, fw_autoload_mask);
1429 
1430 		/* mec ucode */
1431 		cp_hdr = (const struct gfx_firmware_header_v1_0 *)
1432 			adev->gfx.mec_fw->data;
1433 		fw_data = (const __le32 *) (adev->gfx.mec_fw->data +
1434 				le32_to_cpu(cp_hdr->header.ucode_array_offset_bytes));
1435 		fw_size = le32_to_cpu(cp_hdr->header.ucode_size_bytes) -
1436 			cp_hdr->jt_size * 4;
1437 		gfx_v11_0_rlc_backdoor_autoload_copy_ucode(adev, SOC21_FIRMWARE_ID_CP_MEC,
1438 						fw_data, fw_size, fw_autoload_mask);
1439 	}
1440 
1441 	/* rlc ucode */
1442 	rlc_hdr = (const struct rlc_firmware_header_v2_0 *)
1443 		adev->gfx.rlc_fw->data;
1444 	fw_data = (const __le32 *)(adev->gfx.rlc_fw->data +
1445 			le32_to_cpu(rlc_hdr->header.ucode_array_offset_bytes));
1446 	fw_size = le32_to_cpu(rlc_hdr->header.ucode_size_bytes);
1447 	gfx_v11_0_rlc_backdoor_autoload_copy_ucode(adev, SOC21_FIRMWARE_ID_RLC_G_UCODE,
1448 					fw_data, fw_size, fw_autoload_mask);
1449 
1450 	version_major = le16_to_cpu(rlc_hdr->header.header_version_major);
1451 	version_minor = le16_to_cpu(rlc_hdr->header.header_version_minor);
1452 	if (version_major == 2) {
1453 		if (version_minor >= 2) {
1454 			rlcv22_hdr = (const struct rlc_firmware_header_v2_2 *)adev->gfx.rlc_fw->data;
1455 
1456 			fw_data = (const __le32 *)(adev->gfx.rlc_fw->data +
1457 					le32_to_cpu(rlcv22_hdr->rlc_iram_ucode_offset_bytes));
1458 			fw_size = le32_to_cpu(rlcv22_hdr->rlc_iram_ucode_size_bytes);
1459 			gfx_v11_0_rlc_backdoor_autoload_copy_ucode(adev, SOC21_FIRMWARE_ID_RLX6_UCODE,
1460 					fw_data, fw_size, fw_autoload_mask);
1461 
1462 			fw_data = (const __le32 *)(adev->gfx.rlc_fw->data +
1463 					le32_to_cpu(rlcv22_hdr->rlc_dram_ucode_offset_bytes));
1464 			fw_size = le32_to_cpu(rlcv22_hdr->rlc_dram_ucode_size_bytes);
1465 			gfx_v11_0_rlc_backdoor_autoload_copy_ucode(adev, SOC21_FIRMWARE_ID_RLX6_DRAM_BOOT,
1466 					fw_data, fw_size, fw_autoload_mask);
1467 		}
1468 	}
1469 }
1470 
1471 static void gfx_v11_0_rlc_backdoor_autoload_copy_sdma_ucode(struct amdgpu_device *adev,
1472 							uint32_t *fw_autoload_mask)
1473 {
1474 	const __le32 *fw_data;
1475 	uint32_t fw_size;
1476 	const struct sdma_firmware_header_v2_0 *sdma_hdr;
1477 
1478 	sdma_hdr = (const struct sdma_firmware_header_v2_0 *)
1479 		adev->sdma.instance[0].fw->data;
1480 	fw_data = (const __le32 *) (adev->sdma.instance[0].fw->data +
1481 			le32_to_cpu(sdma_hdr->header.ucode_array_offset_bytes));
1482 	fw_size = le32_to_cpu(sdma_hdr->ctx_ucode_size_bytes);
1483 
1484 	gfx_v11_0_rlc_backdoor_autoload_copy_ucode(adev,
1485 			SOC21_FIRMWARE_ID_SDMA_UCODE_TH0, fw_data, fw_size, fw_autoload_mask);
1486 
1487 	fw_data = (const __le32 *) (adev->sdma.instance[0].fw->data +
1488 			le32_to_cpu(sdma_hdr->ctl_ucode_offset));
1489 	fw_size = le32_to_cpu(sdma_hdr->ctl_ucode_size_bytes);
1490 
1491 	gfx_v11_0_rlc_backdoor_autoload_copy_ucode(adev,
1492 			SOC21_FIRMWARE_ID_SDMA_UCODE_TH1, fw_data, fw_size, fw_autoload_mask);
1493 }
1494 
1495 static void gfx_v11_0_rlc_backdoor_autoload_copy_mes_ucode(struct amdgpu_device *adev,
1496 							uint32_t *fw_autoload_mask)
1497 {
1498 	const __le32 *fw_data;
1499 	unsigned fw_size;
1500 	const struct mes_firmware_header_v1_0 *mes_hdr;
1501 	int pipe, ucode_id, data_id;
1502 
1503 	for (pipe = 0; pipe < 2; pipe++) {
1504 		if (pipe==0) {
1505 			ucode_id = SOC21_FIRMWARE_ID_RS64_MES_P0;
1506 			data_id  = SOC21_FIRMWARE_ID_RS64_MES_P0_STACK;
1507 		} else {
1508 			ucode_id = SOC21_FIRMWARE_ID_RS64_MES_P1;
1509 			data_id  = SOC21_FIRMWARE_ID_RS64_MES_P1_STACK;
1510 		}
1511 
1512 		mes_hdr = (const struct mes_firmware_header_v1_0 *)
1513 			adev->mes.fw[pipe]->data;
1514 
1515 		fw_data = (const __le32 *)(adev->mes.fw[pipe]->data +
1516 				le32_to_cpu(mes_hdr->mes_ucode_offset_bytes));
1517 		fw_size = le32_to_cpu(mes_hdr->mes_ucode_size_bytes);
1518 
1519 		gfx_v11_0_rlc_backdoor_autoload_copy_ucode(adev,
1520 				ucode_id, fw_data, fw_size, fw_autoload_mask);
1521 
1522 		fw_data = (const __le32 *)(adev->mes.fw[pipe]->data +
1523 				le32_to_cpu(mes_hdr->mes_ucode_data_offset_bytes));
1524 		fw_size = le32_to_cpu(mes_hdr->mes_ucode_data_size_bytes);
1525 
1526 		gfx_v11_0_rlc_backdoor_autoload_copy_ucode(adev,
1527 				data_id, fw_data, fw_size, fw_autoload_mask);
1528 	}
1529 }
1530 
1531 static int gfx_v11_0_rlc_backdoor_autoload_enable(struct amdgpu_device *adev)
1532 {
1533 	uint32_t rlc_g_offset, rlc_g_size;
1534 	uint64_t gpu_addr;
1535 	uint32_t autoload_fw_id[2];
1536 
1537 	memset(autoload_fw_id, 0, sizeof(uint32_t) * 2);
1538 
1539 	/* RLC autoload sequence 2: copy ucode */
1540 	gfx_v11_0_rlc_backdoor_autoload_copy_sdma_ucode(adev, autoload_fw_id);
1541 	gfx_v11_0_rlc_backdoor_autoload_copy_gfx_ucode(adev, autoload_fw_id);
1542 	gfx_v11_0_rlc_backdoor_autoload_copy_mes_ucode(adev, autoload_fw_id);
1543 	gfx_v11_0_rlc_backdoor_autoload_copy_toc_ucode(adev, autoload_fw_id);
1544 
1545 	rlc_g_offset = rlc_autoload_info[SOC21_FIRMWARE_ID_RLC_G_UCODE].offset;
1546 	rlc_g_size = rlc_autoload_info[SOC21_FIRMWARE_ID_RLC_G_UCODE].size;
1547 	gpu_addr = adev->gfx.rlc.rlc_autoload_gpu_addr + rlc_g_offset;
1548 
1549 	WREG32_SOC15(GC, 0, regGFX_IMU_RLC_BOOTLOADER_ADDR_HI, upper_32_bits(gpu_addr));
1550 	WREG32_SOC15(GC, 0, regGFX_IMU_RLC_BOOTLOADER_ADDR_LO, lower_32_bits(gpu_addr));
1551 
1552 	WREG32_SOC15(GC, 0, regGFX_IMU_RLC_BOOTLOADER_SIZE, rlc_g_size);
1553 
1554 	/* RLC autoload sequence 3: load IMU fw */
1555 	if (adev->gfx.imu.funcs->load_microcode)
1556 		adev->gfx.imu.funcs->load_microcode(adev);
1557 	/* RLC autoload sequence 4 init IMU fw */
1558 	if (adev->gfx.imu.funcs->setup_imu)
1559 		adev->gfx.imu.funcs->setup_imu(adev);
1560 	if (adev->gfx.imu.funcs->start_imu)
1561 		adev->gfx.imu.funcs->start_imu(adev);
1562 
1563 	/* RLC autoload sequence 5 disable gpa mode */
1564 	gfx_v11_0_disable_gpa_mode(adev);
1565 
1566 	return 0;
1567 }
1568 
1569 static int gfx_v11_0_sw_init(void *handle)
1570 {
1571 	int i, j, k, r, ring_id = 0;
1572 	struct amdgpu_kiq *kiq;
1573 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1574 
1575 	adev->gfxhub.funcs->init(adev);
1576 
1577 	switch (adev->ip_versions[GC_HWIP][0]) {
1578 	case IP_VERSION(11, 0, 0):
1579 	case IP_VERSION(11, 0, 1):
1580 	case IP_VERSION(11, 0, 2):
1581 		adev->gfx.me.num_me = 1;
1582 		adev->gfx.me.num_pipe_per_me = 1;
1583 		adev->gfx.me.num_queue_per_pipe = 1;
1584 		adev->gfx.mec.num_mec = 2;
1585 		adev->gfx.mec.num_pipe_per_mec = 4;
1586 		adev->gfx.mec.num_queue_per_pipe = 4;
1587 		break;
1588 	default:
1589 		adev->gfx.me.num_me = 1;
1590 		adev->gfx.me.num_pipe_per_me = 1;
1591 		adev->gfx.me.num_queue_per_pipe = 1;
1592 		adev->gfx.mec.num_mec = 1;
1593 		adev->gfx.mec.num_pipe_per_mec = 4;
1594 		adev->gfx.mec.num_queue_per_pipe = 8;
1595 		break;
1596 	}
1597 
1598 	/* EOP Event */
1599 	r = amdgpu_irq_add_id(adev, SOC21_IH_CLIENTID_GRBM_CP,
1600 			      GFX_11_0_0__SRCID__CP_EOP_INTERRUPT,
1601 			      &adev->gfx.eop_irq);
1602 	if (r)
1603 		return r;
1604 
1605 	/* Privileged reg */
1606 	r = amdgpu_irq_add_id(adev, SOC21_IH_CLIENTID_GRBM_CP,
1607 			      GFX_11_0_0__SRCID__CP_PRIV_REG_FAULT,
1608 			      &adev->gfx.priv_reg_irq);
1609 	if (r)
1610 		return r;
1611 
1612 	/* Privileged inst */
1613 	r = amdgpu_irq_add_id(adev, SOC21_IH_CLIENTID_GRBM_CP,
1614 			      GFX_11_0_0__SRCID__CP_PRIV_INSTR_FAULT,
1615 			      &adev->gfx.priv_inst_irq);
1616 	if (r)
1617 		return r;
1618 
1619 	adev->gfx.gfx_current_status = AMDGPU_GFX_NORMAL_MODE;
1620 
1621 	if (adev->gfx.imu.funcs) {
1622 		if (adev->gfx.imu.funcs->init_microcode) {
1623 			r = adev->gfx.imu.funcs->init_microcode(adev);
1624 			if (r)
1625 				DRM_ERROR("Failed to load imu firmware!\n");
1626 		}
1627 	}
1628 
1629 	r = gfx_v11_0_me_init(adev);
1630 	if (r)
1631 		return r;
1632 
1633 	r = gfx_v11_0_rlc_init(adev);
1634 	if (r) {
1635 		DRM_ERROR("Failed to init rlc BOs!\n");
1636 		return r;
1637 	}
1638 
1639 	r = gfx_v11_0_mec_init(adev);
1640 	if (r) {
1641 		DRM_ERROR("Failed to init MEC BOs!\n");
1642 		return r;
1643 	}
1644 
1645 	/* set up the gfx ring */
1646 	for (i = 0; i < adev->gfx.me.num_me; i++) {
1647 		for (j = 0; j < adev->gfx.me.num_queue_per_pipe; j++) {
1648 			for (k = 0; k < adev->gfx.me.num_pipe_per_me; k++) {
1649 				if (!amdgpu_gfx_is_me_queue_enabled(adev, i, k, j))
1650 					continue;
1651 
1652 				r = gfx_v11_0_gfx_ring_init(adev, ring_id,
1653 							    i, k, j);
1654 				if (r)
1655 					return r;
1656 				ring_id++;
1657 			}
1658 		}
1659 	}
1660 
1661 	ring_id = 0;
1662 	/* set up the compute queues - allocate horizontally across pipes */
1663 	for (i = 0; i < adev->gfx.mec.num_mec; ++i) {
1664 		for (j = 0; j < adev->gfx.mec.num_queue_per_pipe; j++) {
1665 			for (k = 0; k < adev->gfx.mec.num_pipe_per_mec; k++) {
1666 				if (!amdgpu_gfx_is_mec_queue_enabled(adev, i, k,
1667 								     j))
1668 					continue;
1669 
1670 				r = gfx_v11_0_compute_ring_init(adev, ring_id,
1671 								i, k, j);
1672 				if (r)
1673 					return r;
1674 
1675 				ring_id++;
1676 			}
1677 		}
1678 	}
1679 
1680 	if (!adev->enable_mes_kiq) {
1681 		r = amdgpu_gfx_kiq_init(adev, GFX11_MEC_HPD_SIZE);
1682 		if (r) {
1683 			DRM_ERROR("Failed to init KIQ BOs!\n");
1684 			return r;
1685 		}
1686 
1687 		kiq = &adev->gfx.kiq;
1688 		r = amdgpu_gfx_kiq_init_ring(adev, &kiq->ring, &kiq->irq);
1689 		if (r)
1690 			return r;
1691 	}
1692 
1693 	r = amdgpu_gfx_mqd_sw_init(adev, sizeof(struct v11_compute_mqd));
1694 	if (r)
1695 		return r;
1696 
1697 	/* allocate visible FB for rlc auto-loading fw */
1698 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_RLC_BACKDOOR_AUTO) {
1699 		r = gfx_v11_0_init_toc_microcode(adev);
1700 		if (r)
1701 			dev_err(adev->dev, "Failed to load toc firmware!\n");
1702 		r = gfx_v11_0_rlc_autoload_buffer_init(adev);
1703 		if (r)
1704 			return r;
1705 	}
1706 
1707 	r = gfx_v11_0_gpu_early_init(adev);
1708 	if (r)
1709 		return r;
1710 
1711 	return 0;
1712 }
1713 
1714 static void gfx_v11_0_pfp_fini(struct amdgpu_device *adev)
1715 {
1716 	amdgpu_bo_free_kernel(&adev->gfx.pfp.pfp_fw_obj,
1717 			      &adev->gfx.pfp.pfp_fw_gpu_addr,
1718 			      (void **)&adev->gfx.pfp.pfp_fw_ptr);
1719 
1720 	amdgpu_bo_free_kernel(&adev->gfx.pfp.pfp_fw_data_obj,
1721 			      &adev->gfx.pfp.pfp_fw_data_gpu_addr,
1722 			      (void **)&adev->gfx.pfp.pfp_fw_data_ptr);
1723 }
1724 
1725 static void gfx_v11_0_me_fini(struct amdgpu_device *adev)
1726 {
1727 	amdgpu_bo_free_kernel(&adev->gfx.me.me_fw_obj,
1728 			      &adev->gfx.me.me_fw_gpu_addr,
1729 			      (void **)&adev->gfx.me.me_fw_ptr);
1730 
1731 	amdgpu_bo_free_kernel(&adev->gfx.me.me_fw_data_obj,
1732 			       &adev->gfx.me.me_fw_data_gpu_addr,
1733 			       (void **)&adev->gfx.me.me_fw_data_ptr);
1734 }
1735 
1736 static void gfx_v11_0_rlc_autoload_buffer_fini(struct amdgpu_device *adev)
1737 {
1738 	amdgpu_bo_free_kernel(&adev->gfx.rlc.rlc_autoload_bo,
1739 			&adev->gfx.rlc.rlc_autoload_gpu_addr,
1740 			(void **)&adev->gfx.rlc.rlc_autoload_ptr);
1741 }
1742 
1743 static int gfx_v11_0_sw_fini(void *handle)
1744 {
1745 	int i;
1746 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1747 
1748 	for (i = 0; i < adev->gfx.num_gfx_rings; i++)
1749 		amdgpu_ring_fini(&adev->gfx.gfx_ring[i]);
1750 	for (i = 0; i < adev->gfx.num_compute_rings; i++)
1751 		amdgpu_ring_fini(&adev->gfx.compute_ring[i]);
1752 
1753 	amdgpu_gfx_mqd_sw_fini(adev);
1754 
1755 	if (!adev->enable_mes_kiq) {
1756 		amdgpu_gfx_kiq_free_ring(&adev->gfx.kiq.ring);
1757 		amdgpu_gfx_kiq_fini(adev);
1758 	}
1759 
1760 	gfx_v11_0_pfp_fini(adev);
1761 	gfx_v11_0_me_fini(adev);
1762 	gfx_v11_0_rlc_fini(adev);
1763 	gfx_v11_0_mec_fini(adev);
1764 
1765 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_RLC_BACKDOOR_AUTO)
1766 		gfx_v11_0_rlc_autoload_buffer_fini(adev);
1767 
1768 	gfx_v11_0_free_microcode(adev);
1769 
1770 	return 0;
1771 }
1772 
1773 static void gfx_v11_0_select_se_sh(struct amdgpu_device *adev, u32 se_num,
1774 				   u32 sh_num, u32 instance)
1775 {
1776 	u32 data;
1777 
1778 	if (instance == 0xffffffff)
1779 		data = REG_SET_FIELD(0, GRBM_GFX_INDEX,
1780 				     INSTANCE_BROADCAST_WRITES, 1);
1781 	else
1782 		data = REG_SET_FIELD(0, GRBM_GFX_INDEX, INSTANCE_INDEX,
1783 				     instance);
1784 
1785 	if (se_num == 0xffffffff)
1786 		data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SE_BROADCAST_WRITES,
1787 				     1);
1788 	else
1789 		data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SE_INDEX, se_num);
1790 
1791 	if (sh_num == 0xffffffff)
1792 		data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SA_BROADCAST_WRITES,
1793 				     1);
1794 	else
1795 		data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SA_INDEX, sh_num);
1796 
1797 	WREG32_SOC15(GC, 0, regGRBM_GFX_INDEX, data);
1798 }
1799 
1800 static u32 gfx_v11_0_get_rb_active_bitmap(struct amdgpu_device *adev)
1801 {
1802 	u32 data, mask;
1803 
1804 	data = RREG32_SOC15(GC, 0, regCC_RB_BACKEND_DISABLE);
1805 	data |= RREG32_SOC15(GC, 0, regGC_USER_RB_BACKEND_DISABLE);
1806 
1807 	data &= CC_RB_BACKEND_DISABLE__BACKEND_DISABLE_MASK;
1808 	data >>= GC_USER_RB_BACKEND_DISABLE__BACKEND_DISABLE__SHIFT;
1809 
1810 	mask = amdgpu_gfx_create_bitmask(adev->gfx.config.max_backends_per_se /
1811 					 adev->gfx.config.max_sh_per_se);
1812 
1813 	return (~data) & mask;
1814 }
1815 
1816 static void gfx_v11_0_setup_rb(struct amdgpu_device *adev)
1817 {
1818 	int i, j;
1819 	u32 data;
1820 	u32 active_rbs = 0;
1821 	u32 rb_bitmap_width_per_sh = adev->gfx.config.max_backends_per_se /
1822 					adev->gfx.config.max_sh_per_se;
1823 
1824 	mutex_lock(&adev->grbm_idx_mutex);
1825 	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
1826 		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
1827 			gfx_v11_0_select_se_sh(adev, i, j, 0xffffffff);
1828 			data = gfx_v11_0_get_rb_active_bitmap(adev);
1829 			active_rbs |= data << ((i * adev->gfx.config.max_sh_per_se + j) *
1830 					       rb_bitmap_width_per_sh);
1831 		}
1832 	}
1833 	gfx_v11_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
1834 	mutex_unlock(&adev->grbm_idx_mutex);
1835 
1836 	adev->gfx.config.backend_enable_mask = active_rbs;
1837 	adev->gfx.config.num_rbs = hweight32(active_rbs);
1838 }
1839 
1840 #define DEFAULT_SH_MEM_BASES	(0x6000)
1841 #define LDS_APP_BASE           0x1
1842 #define SCRATCH_APP_BASE       0x2
1843 
1844 static void gfx_v11_0_init_compute_vmid(struct amdgpu_device *adev)
1845 {
1846 	int i;
1847 	uint32_t sh_mem_bases;
1848 	uint32_t data;
1849 
1850 	/*
1851 	 * Configure apertures:
1852 	 * LDS:         0x60000000'00000000 - 0x60000001'00000000 (4GB)
1853 	 * Scratch:     0x60000001'00000000 - 0x60000002'00000000 (4GB)
1854 	 * GPUVM:       0x60010000'00000000 - 0x60020000'00000000 (1TB)
1855 	 */
1856 	sh_mem_bases = (LDS_APP_BASE << SH_MEM_BASES__SHARED_BASE__SHIFT) |
1857 			SCRATCH_APP_BASE;
1858 
1859 	mutex_lock(&adev->srbm_mutex);
1860 	for (i = adev->vm_manager.first_kfd_vmid; i < AMDGPU_NUM_VMID; i++) {
1861 		soc21_grbm_select(adev, 0, 0, 0, i);
1862 		/* CP and shaders */
1863 		WREG32_SOC15(GC, 0, regSH_MEM_CONFIG, DEFAULT_SH_MEM_CONFIG);
1864 		WREG32_SOC15(GC, 0, regSH_MEM_BASES, sh_mem_bases);
1865 
1866 		/* Enable trap for each kfd vmid. */
1867 		data = RREG32(SOC15_REG_OFFSET(GC, 0, regSPI_GDBG_PER_VMID_CNTL));
1868 		data = REG_SET_FIELD(data, SPI_GDBG_PER_VMID_CNTL, TRAP_EN, 1);
1869 	}
1870 	soc21_grbm_select(adev, 0, 0, 0, 0);
1871 	mutex_unlock(&adev->srbm_mutex);
1872 
1873 	/* Initialize all compute VMIDs to have no GDS, GWS, or OA
1874 	   acccess. These should be enabled by FW for target VMIDs. */
1875 	for (i = adev->vm_manager.first_kfd_vmid; i < AMDGPU_NUM_VMID; i++) {
1876 		WREG32_SOC15_OFFSET(GC, 0, regGDS_VMID0_BASE, 2 * i, 0);
1877 		WREG32_SOC15_OFFSET(GC, 0, regGDS_VMID0_SIZE, 2 * i, 0);
1878 		WREG32_SOC15_OFFSET(GC, 0, regGDS_GWS_VMID0, i, 0);
1879 		WREG32_SOC15_OFFSET(GC, 0, regGDS_OA_VMID0, i, 0);
1880 	}
1881 }
1882 
1883 static void gfx_v11_0_init_gds_vmid(struct amdgpu_device *adev)
1884 {
1885 	int vmid;
1886 
1887 	/*
1888 	 * Initialize all compute and user-gfx VMIDs to have no GDS, GWS, or OA
1889 	 * access. Compute VMIDs should be enabled by FW for target VMIDs,
1890 	 * the driver can enable them for graphics. VMID0 should maintain
1891 	 * access so that HWS firmware can save/restore entries.
1892 	 */
1893 	for (vmid = 1; vmid < 16; vmid++) {
1894 		WREG32_SOC15_OFFSET(GC, 0, regGDS_VMID0_BASE, 2 * vmid, 0);
1895 		WREG32_SOC15_OFFSET(GC, 0, regGDS_VMID0_SIZE, 2 * vmid, 0);
1896 		WREG32_SOC15_OFFSET(GC, 0, regGDS_GWS_VMID0, vmid, 0);
1897 		WREG32_SOC15_OFFSET(GC, 0, regGDS_OA_VMID0, vmid, 0);
1898 	}
1899 }
1900 
1901 static void gfx_v11_0_tcp_harvest(struct amdgpu_device *adev)
1902 {
1903 	/* TODO: harvest feature to be added later. */
1904 }
1905 
1906 static void gfx_v11_0_get_tcc_info(struct amdgpu_device *adev)
1907 {
1908 	/* TCCs are global (not instanced). */
1909 	uint32_t tcc_disable = RREG32_SOC15(GC, 0, regCGTS_TCC_DISABLE) |
1910 			       RREG32_SOC15(GC, 0, regCGTS_USER_TCC_DISABLE);
1911 
1912 	adev->gfx.config.tcc_disabled_mask =
1913 		REG_GET_FIELD(tcc_disable, CGTS_TCC_DISABLE, TCC_DISABLE) |
1914 		(REG_GET_FIELD(tcc_disable, CGTS_TCC_DISABLE, HI_TCC_DISABLE) << 16);
1915 }
1916 
1917 static void gfx_v11_0_constants_init(struct amdgpu_device *adev)
1918 {
1919 	u32 tmp;
1920 	int i;
1921 
1922 	WREG32_FIELD15_PREREG(GC, 0, GRBM_CNTL, READ_TIMEOUT, 0xff);
1923 
1924 	gfx_v11_0_setup_rb(adev);
1925 	gfx_v11_0_get_cu_info(adev, &adev->gfx.cu_info);
1926 	gfx_v11_0_get_tcc_info(adev);
1927 	adev->gfx.config.pa_sc_tile_steering_override = 0;
1928 
1929 	/* XXX SH_MEM regs */
1930 	/* where to put LDS, scratch, GPUVM in FSA64 space */
1931 	mutex_lock(&adev->srbm_mutex);
1932 	for (i = 0; i < adev->vm_manager.id_mgr[AMDGPU_GFXHUB_0].num_ids; i++) {
1933 		soc21_grbm_select(adev, 0, 0, 0, i);
1934 		/* CP and shaders */
1935 		WREG32_SOC15(GC, 0, regSH_MEM_CONFIG, DEFAULT_SH_MEM_CONFIG);
1936 		if (i != 0) {
1937 			tmp = REG_SET_FIELD(0, SH_MEM_BASES, PRIVATE_BASE,
1938 				(adev->gmc.private_aperture_start >> 48));
1939 			tmp = REG_SET_FIELD(tmp, SH_MEM_BASES, SHARED_BASE,
1940 				(adev->gmc.shared_aperture_start >> 48));
1941 			WREG32_SOC15(GC, 0, regSH_MEM_BASES, tmp);
1942 		}
1943 	}
1944 	soc21_grbm_select(adev, 0, 0, 0, 0);
1945 
1946 	mutex_unlock(&adev->srbm_mutex);
1947 
1948 	gfx_v11_0_init_compute_vmid(adev);
1949 	gfx_v11_0_init_gds_vmid(adev);
1950 }
1951 
1952 static void gfx_v11_0_enable_gui_idle_interrupt(struct amdgpu_device *adev,
1953 					       bool enable)
1954 {
1955 	u32 tmp;
1956 
1957 	if (amdgpu_sriov_vf(adev))
1958 		return;
1959 
1960 	tmp = RREG32_SOC15(GC, 0, regCP_INT_CNTL_RING0);
1961 
1962 	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CNTX_BUSY_INT_ENABLE,
1963 			    enable ? 1 : 0);
1964 	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CNTX_EMPTY_INT_ENABLE,
1965 			    enable ? 1 : 0);
1966 	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CMP_BUSY_INT_ENABLE,
1967 			    enable ? 1 : 0);
1968 	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, GFX_IDLE_INT_ENABLE,
1969 			    enable ? 1 : 0);
1970 
1971 	WREG32_SOC15(GC, 0, regCP_INT_CNTL_RING0, tmp);
1972 }
1973 
1974 static int gfx_v11_0_init_csb(struct amdgpu_device *adev)
1975 {
1976 	adev->gfx.rlc.funcs->get_csb_buffer(adev, adev->gfx.rlc.cs_ptr);
1977 
1978 	WREG32_SOC15(GC, 0, regRLC_CSIB_ADDR_HI,
1979 			adev->gfx.rlc.clear_state_gpu_addr >> 32);
1980 	WREG32_SOC15(GC, 0, regRLC_CSIB_ADDR_LO,
1981 			adev->gfx.rlc.clear_state_gpu_addr & 0xfffffffc);
1982 	WREG32_SOC15(GC, 0, regRLC_CSIB_LENGTH, adev->gfx.rlc.clear_state_size);
1983 
1984 	return 0;
1985 }
1986 
1987 static void gfx_v11_0_rlc_stop(struct amdgpu_device *adev)
1988 {
1989 	u32 tmp = RREG32_SOC15(GC, 0, regRLC_CNTL);
1990 
1991 	tmp = REG_SET_FIELD(tmp, RLC_CNTL, RLC_ENABLE_F32, 0);
1992 	WREG32_SOC15(GC, 0, regRLC_CNTL, tmp);
1993 }
1994 
1995 static void gfx_v11_0_rlc_reset(struct amdgpu_device *adev)
1996 {
1997 	WREG32_FIELD15_PREREG(GC, 0, GRBM_SOFT_RESET, SOFT_RESET_RLC, 1);
1998 	udelay(50);
1999 	WREG32_FIELD15_PREREG(GC, 0, GRBM_SOFT_RESET, SOFT_RESET_RLC, 0);
2000 	udelay(50);
2001 }
2002 
2003 static void gfx_v11_0_rlc_smu_handshake_cntl(struct amdgpu_device *adev,
2004 					     bool enable)
2005 {
2006 	uint32_t rlc_pg_cntl;
2007 
2008 	rlc_pg_cntl = RREG32_SOC15(GC, 0, regRLC_PG_CNTL);
2009 
2010 	if (!enable) {
2011 		/* RLC_PG_CNTL[23] = 0 (default)
2012 		 * RLC will wait for handshake acks with SMU
2013 		 * GFXOFF will be enabled
2014 		 * RLC_PG_CNTL[23] = 1
2015 		 * RLC will not issue any message to SMU
2016 		 * hence no handshake between SMU & RLC
2017 		 * GFXOFF will be disabled
2018 		 */
2019 		rlc_pg_cntl |= RLC_PG_CNTL__SMU_HANDSHAKE_DISABLE_MASK;
2020 	} else
2021 		rlc_pg_cntl &= ~RLC_PG_CNTL__SMU_HANDSHAKE_DISABLE_MASK;
2022 	WREG32_SOC15(GC, 0, regRLC_PG_CNTL, rlc_pg_cntl);
2023 }
2024 
2025 static void gfx_v11_0_rlc_start(struct amdgpu_device *adev)
2026 {
2027 	/* TODO: enable rlc & smu handshake until smu
2028 	 * and gfxoff feature works as expected */
2029 	if (!(amdgpu_pp_feature_mask & PP_GFXOFF_MASK))
2030 		gfx_v11_0_rlc_smu_handshake_cntl(adev, false);
2031 
2032 	WREG32_FIELD15_PREREG(GC, 0, RLC_CNTL, RLC_ENABLE_F32, 1);
2033 	udelay(50);
2034 }
2035 
2036 static void gfx_v11_0_rlc_enable_srm(struct amdgpu_device *adev)
2037 {
2038 	uint32_t tmp;
2039 
2040 	/* enable Save Restore Machine */
2041 	tmp = RREG32(SOC15_REG_OFFSET(GC, 0, regRLC_SRM_CNTL));
2042 	tmp |= RLC_SRM_CNTL__AUTO_INCR_ADDR_MASK;
2043 	tmp |= RLC_SRM_CNTL__SRM_ENABLE_MASK;
2044 	WREG32(SOC15_REG_OFFSET(GC, 0, regRLC_SRM_CNTL), tmp);
2045 }
2046 
2047 static void gfx_v11_0_load_rlcg_microcode(struct amdgpu_device *adev)
2048 {
2049 	const struct rlc_firmware_header_v2_0 *hdr;
2050 	const __le32 *fw_data;
2051 	unsigned i, fw_size;
2052 
2053 	hdr = (const struct rlc_firmware_header_v2_0 *)adev->gfx.rlc_fw->data;
2054 	fw_data = (const __le32 *)(adev->gfx.rlc_fw->data +
2055 			   le32_to_cpu(hdr->header.ucode_array_offset_bytes));
2056 	fw_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4;
2057 
2058 	WREG32_SOC15(GC, 0, regRLC_GPM_UCODE_ADDR,
2059 		     RLCG_UCODE_LOADING_START_ADDRESS);
2060 
2061 	for (i = 0; i < fw_size; i++)
2062 		WREG32_SOC15(GC, 0, regRLC_GPM_UCODE_DATA,
2063 			     le32_to_cpup(fw_data++));
2064 
2065 	WREG32_SOC15(GC, 0, regRLC_GPM_UCODE_ADDR, adev->gfx.rlc_fw_version);
2066 }
2067 
2068 static void gfx_v11_0_load_rlc_iram_dram_microcode(struct amdgpu_device *adev)
2069 {
2070 	const struct rlc_firmware_header_v2_2 *hdr;
2071 	const __le32 *fw_data;
2072 	unsigned i, fw_size;
2073 	u32 tmp;
2074 
2075 	hdr = (const struct rlc_firmware_header_v2_2 *)adev->gfx.rlc_fw->data;
2076 
2077 	fw_data = (const __le32 *)(adev->gfx.rlc_fw->data +
2078 			le32_to_cpu(hdr->rlc_iram_ucode_offset_bytes));
2079 	fw_size = le32_to_cpu(hdr->rlc_iram_ucode_size_bytes) / 4;
2080 
2081 	WREG32_SOC15(GC, 0, regRLC_LX6_IRAM_ADDR, 0);
2082 
2083 	for (i = 0; i < fw_size; i++) {
2084 		if ((amdgpu_emu_mode == 1) && (i % 100 == 99))
2085 			msleep(1);
2086 		WREG32_SOC15(GC, 0, regRLC_LX6_IRAM_DATA,
2087 				le32_to_cpup(fw_data++));
2088 	}
2089 
2090 	WREG32_SOC15(GC, 0, regRLC_LX6_IRAM_ADDR, adev->gfx.rlc_fw_version);
2091 
2092 	fw_data = (const __le32 *)(adev->gfx.rlc_fw->data +
2093 			le32_to_cpu(hdr->rlc_dram_ucode_offset_bytes));
2094 	fw_size = le32_to_cpu(hdr->rlc_dram_ucode_size_bytes) / 4;
2095 
2096 	WREG32_SOC15(GC, 0, regRLC_LX6_DRAM_ADDR, 0);
2097 	for (i = 0; i < fw_size; i++) {
2098 		if ((amdgpu_emu_mode == 1) && (i % 100 == 99))
2099 			msleep(1);
2100 		WREG32_SOC15(GC, 0, regRLC_LX6_DRAM_DATA,
2101 				le32_to_cpup(fw_data++));
2102 	}
2103 
2104 	WREG32_SOC15(GC, 0, regRLC_LX6_IRAM_ADDR, adev->gfx.rlc_fw_version);
2105 
2106 	tmp = RREG32_SOC15(GC, 0, regRLC_LX6_CNTL);
2107 	tmp = REG_SET_FIELD(tmp, RLC_LX6_CNTL, PDEBUG_ENABLE, 1);
2108 	tmp = REG_SET_FIELD(tmp, RLC_LX6_CNTL, BRESET, 0);
2109 	WREG32_SOC15(GC, 0, regRLC_LX6_CNTL, tmp);
2110 }
2111 
2112 static void gfx_v11_0_load_rlcp_rlcv_microcode(struct amdgpu_device *adev)
2113 {
2114 	const struct rlc_firmware_header_v2_3 *hdr;
2115 	const __le32 *fw_data;
2116 	unsigned i, fw_size;
2117 	u32 tmp;
2118 
2119 	hdr = (const struct rlc_firmware_header_v2_3 *)adev->gfx.rlc_fw->data;
2120 
2121 	fw_data = (const __le32 *)(adev->gfx.rlc_fw->data +
2122 			le32_to_cpu(hdr->rlcp_ucode_offset_bytes));
2123 	fw_size = le32_to_cpu(hdr->rlcp_ucode_size_bytes) / 4;
2124 
2125 	WREG32_SOC15(GC, 0, regRLC_PACE_UCODE_ADDR, 0);
2126 
2127 	for (i = 0; i < fw_size; i++) {
2128 		if ((amdgpu_emu_mode == 1) && (i % 100 == 99))
2129 			msleep(1);
2130 		WREG32_SOC15(GC, 0, regRLC_PACE_UCODE_DATA,
2131 				le32_to_cpup(fw_data++));
2132 	}
2133 
2134 	WREG32_SOC15(GC, 0, regRLC_PACE_UCODE_ADDR, adev->gfx.rlc_fw_version);
2135 
2136 	tmp = RREG32_SOC15(GC, 0, regRLC_GPM_THREAD_ENABLE);
2137 	tmp = REG_SET_FIELD(tmp, RLC_GPM_THREAD_ENABLE, THREAD1_ENABLE, 1);
2138 	WREG32_SOC15(GC, 0, regRLC_GPM_THREAD_ENABLE, tmp);
2139 
2140 	fw_data = (const __le32 *)(adev->gfx.rlc_fw->data +
2141 			le32_to_cpu(hdr->rlcv_ucode_offset_bytes));
2142 	fw_size = le32_to_cpu(hdr->rlcv_ucode_size_bytes) / 4;
2143 
2144 	WREG32_SOC15(GC, 0, regRLC_GPU_IOV_UCODE_ADDR, 0);
2145 
2146 	for (i = 0; i < fw_size; i++) {
2147 		if ((amdgpu_emu_mode == 1) && (i % 100 == 99))
2148 			msleep(1);
2149 		WREG32_SOC15(GC, 0, regRLC_GPU_IOV_UCODE_DATA,
2150 				le32_to_cpup(fw_data++));
2151 	}
2152 
2153 	WREG32_SOC15(GC, 0, regRLC_GPU_IOV_UCODE_ADDR, adev->gfx.rlc_fw_version);
2154 
2155 	tmp = RREG32_SOC15(GC, 0, regRLC_GPU_IOV_F32_CNTL);
2156 	tmp = REG_SET_FIELD(tmp, RLC_GPU_IOV_F32_CNTL, ENABLE, 1);
2157 	WREG32_SOC15(GC, 0, regRLC_GPU_IOV_F32_CNTL, tmp);
2158 }
2159 
2160 static int gfx_v11_0_rlc_load_microcode(struct amdgpu_device *adev)
2161 {
2162 	const struct rlc_firmware_header_v2_0 *hdr;
2163 	uint16_t version_major;
2164 	uint16_t version_minor;
2165 
2166 	if (!adev->gfx.rlc_fw)
2167 		return -EINVAL;
2168 
2169 	hdr = (const struct rlc_firmware_header_v2_0 *)adev->gfx.rlc_fw->data;
2170 	amdgpu_ucode_print_rlc_hdr(&hdr->header);
2171 
2172 	version_major = le16_to_cpu(hdr->header.header_version_major);
2173 	version_minor = le16_to_cpu(hdr->header.header_version_minor);
2174 
2175 	if (version_major == 2) {
2176 		gfx_v11_0_load_rlcg_microcode(adev);
2177 		if (amdgpu_dpm == 1) {
2178 			if (version_minor >= 2)
2179 				gfx_v11_0_load_rlc_iram_dram_microcode(adev);
2180 			if (version_minor == 3)
2181 				gfx_v11_0_load_rlcp_rlcv_microcode(adev);
2182 		}
2183 
2184 		return 0;
2185 	}
2186 
2187 	return -EINVAL;
2188 }
2189 
2190 static int gfx_v11_0_rlc_resume(struct amdgpu_device *adev)
2191 {
2192 	int r;
2193 
2194 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
2195 		gfx_v11_0_init_csb(adev);
2196 
2197 		if (!amdgpu_sriov_vf(adev)) /* enable RLC SRM */
2198 			gfx_v11_0_rlc_enable_srm(adev);
2199 	} else {
2200 		if (amdgpu_sriov_vf(adev)) {
2201 			gfx_v11_0_init_csb(adev);
2202 			return 0;
2203 		}
2204 
2205 		adev->gfx.rlc.funcs->stop(adev);
2206 
2207 		/* disable CG */
2208 		WREG32_SOC15(GC, 0, regRLC_CGCG_CGLS_CTRL, 0);
2209 
2210 		/* disable PG */
2211 		WREG32_SOC15(GC, 0, regRLC_PG_CNTL, 0);
2212 
2213 		if (adev->firmware.load_type == AMDGPU_FW_LOAD_DIRECT) {
2214 			/* legacy rlc firmware loading */
2215 			r = gfx_v11_0_rlc_load_microcode(adev);
2216 			if (r)
2217 				return r;
2218 		}
2219 
2220 		gfx_v11_0_init_csb(adev);
2221 
2222 		adev->gfx.rlc.funcs->start(adev);
2223 	}
2224 	return 0;
2225 }
2226 
2227 static int gfx_v11_0_config_me_cache(struct amdgpu_device *adev, uint64_t addr)
2228 {
2229 	uint32_t usec_timeout = 50000;  /* wait for 50ms */
2230 	uint32_t tmp;
2231 	int i;
2232 
2233 	/* Trigger an invalidation of the L1 instruction caches */
2234 	tmp = RREG32_SOC15(GC, 0, regCP_ME_IC_OP_CNTL);
2235 	tmp = REG_SET_FIELD(tmp, CP_ME_IC_OP_CNTL, INVALIDATE_CACHE, 1);
2236 	WREG32_SOC15(GC, 0, regCP_ME_IC_OP_CNTL, tmp);
2237 
2238 	/* Wait for invalidation complete */
2239 	for (i = 0; i < usec_timeout; i++) {
2240 		tmp = RREG32_SOC15(GC, 0, regCP_ME_IC_OP_CNTL);
2241 		if (1 == REG_GET_FIELD(tmp, CP_ME_IC_OP_CNTL,
2242 					INVALIDATE_CACHE_COMPLETE))
2243 			break;
2244 		udelay(1);
2245 	}
2246 
2247 	if (i >= usec_timeout) {
2248 		dev_err(adev->dev, "failed to invalidate instruction cache\n");
2249 		return -EINVAL;
2250 	}
2251 
2252 	if (amdgpu_emu_mode == 1)
2253 		adev->hdp.funcs->flush_hdp(adev, NULL);
2254 
2255 	tmp = RREG32_SOC15(GC, 0, regCP_ME_IC_BASE_CNTL);
2256 	tmp = REG_SET_FIELD(tmp, CP_ME_IC_BASE_CNTL, VMID, 0);
2257 	tmp = REG_SET_FIELD(tmp, CP_ME_IC_BASE_CNTL, CACHE_POLICY, 0);
2258 	tmp = REG_SET_FIELD(tmp, CP_ME_IC_BASE_CNTL, EXE_DISABLE, 0);
2259 	tmp = REG_SET_FIELD(tmp, CP_ME_IC_BASE_CNTL, ADDRESS_CLAMP, 1);
2260 	WREG32_SOC15(GC, 0, regCP_ME_IC_BASE_CNTL, tmp);
2261 
2262 	/* Program me ucode address into intruction cache address register */
2263 	WREG32_SOC15(GC, 0, regCP_ME_IC_BASE_LO,
2264 			lower_32_bits(addr) & 0xFFFFF000);
2265 	WREG32_SOC15(GC, 0, regCP_ME_IC_BASE_HI,
2266 			upper_32_bits(addr));
2267 
2268 	return 0;
2269 }
2270 
2271 static int gfx_v11_0_config_pfp_cache(struct amdgpu_device *adev, uint64_t addr)
2272 {
2273 	uint32_t usec_timeout = 50000;  /* wait for 50ms */
2274 	uint32_t tmp;
2275 	int i;
2276 
2277 	/* Trigger an invalidation of the L1 instruction caches */
2278 	tmp = RREG32_SOC15(GC, 0, regCP_PFP_IC_OP_CNTL);
2279 	tmp = REG_SET_FIELD(tmp, CP_PFP_IC_OP_CNTL, INVALIDATE_CACHE, 1);
2280 	WREG32_SOC15(GC, 0, regCP_PFP_IC_OP_CNTL, tmp);
2281 
2282 	/* Wait for invalidation complete */
2283 	for (i = 0; i < usec_timeout; i++) {
2284 		tmp = RREG32_SOC15(GC, 0, regCP_PFP_IC_OP_CNTL);
2285 		if (1 == REG_GET_FIELD(tmp, CP_PFP_IC_OP_CNTL,
2286 					INVALIDATE_CACHE_COMPLETE))
2287 			break;
2288 		udelay(1);
2289 	}
2290 
2291 	if (i >= usec_timeout) {
2292 		dev_err(adev->dev, "failed to invalidate instruction cache\n");
2293 		return -EINVAL;
2294 	}
2295 
2296 	if (amdgpu_emu_mode == 1)
2297 		adev->hdp.funcs->flush_hdp(adev, NULL);
2298 
2299 	tmp = RREG32_SOC15(GC, 0, regCP_PFP_IC_BASE_CNTL);
2300 	tmp = REG_SET_FIELD(tmp, CP_PFP_IC_BASE_CNTL, VMID, 0);
2301 	tmp = REG_SET_FIELD(tmp, CP_PFP_IC_BASE_CNTL, CACHE_POLICY, 0);
2302 	tmp = REG_SET_FIELD(tmp, CP_PFP_IC_BASE_CNTL, EXE_DISABLE, 0);
2303 	tmp = REG_SET_FIELD(tmp, CP_PFP_IC_BASE_CNTL, ADDRESS_CLAMP, 1);
2304 	WREG32_SOC15(GC, 0, regCP_PFP_IC_BASE_CNTL, tmp);
2305 
2306 	/* Program pfp ucode address into intruction cache address register */
2307 	WREG32_SOC15(GC, 0, regCP_PFP_IC_BASE_LO,
2308 			lower_32_bits(addr) & 0xFFFFF000);
2309 	WREG32_SOC15(GC, 0, regCP_PFP_IC_BASE_HI,
2310 			upper_32_bits(addr));
2311 
2312 	return 0;
2313 }
2314 
2315 static int gfx_v11_0_config_mec_cache(struct amdgpu_device *adev, uint64_t addr)
2316 {
2317 	uint32_t usec_timeout = 50000;  /* wait for 50ms */
2318 	uint32_t tmp;
2319 	int i;
2320 
2321 	/* Trigger an invalidation of the L1 instruction caches */
2322 	tmp = RREG32_SOC15(GC, 0, regCP_CPC_IC_OP_CNTL);
2323 	tmp = REG_SET_FIELD(tmp, CP_CPC_IC_OP_CNTL, INVALIDATE_CACHE, 1);
2324 
2325 	WREG32_SOC15(GC, 0, regCP_CPC_IC_OP_CNTL, tmp);
2326 
2327 	/* Wait for invalidation complete */
2328 	for (i = 0; i < usec_timeout; i++) {
2329 		tmp = RREG32_SOC15(GC, 0, regCP_CPC_IC_OP_CNTL);
2330 		if (1 == REG_GET_FIELD(tmp, CP_CPC_IC_OP_CNTL,
2331 					INVALIDATE_CACHE_COMPLETE))
2332 			break;
2333 		udelay(1);
2334 	}
2335 
2336 	if (i >= usec_timeout) {
2337 		dev_err(adev->dev, "failed to invalidate instruction cache\n");
2338 		return -EINVAL;
2339 	}
2340 
2341 	if (amdgpu_emu_mode == 1)
2342 		adev->hdp.funcs->flush_hdp(adev, NULL);
2343 
2344 	tmp = RREG32_SOC15(GC, 0, regCP_CPC_IC_BASE_CNTL);
2345 	tmp = REG_SET_FIELD(tmp, CP_CPC_IC_BASE_CNTL, CACHE_POLICY, 0);
2346 	tmp = REG_SET_FIELD(tmp, CP_CPC_IC_BASE_CNTL, EXE_DISABLE, 0);
2347 	tmp = REG_SET_FIELD(tmp, CP_CPC_IC_BASE_CNTL, ADDRESS_CLAMP, 1);
2348 	WREG32_SOC15(GC, 0, regCP_CPC_IC_BASE_CNTL, tmp);
2349 
2350 	/* Program mec1 ucode address into intruction cache address register */
2351 	WREG32_SOC15(GC, 0, regCP_CPC_IC_BASE_LO,
2352 			lower_32_bits(addr) & 0xFFFFF000);
2353 	WREG32_SOC15(GC, 0, regCP_CPC_IC_BASE_HI,
2354 			upper_32_bits(addr));
2355 
2356 	return 0;
2357 }
2358 
2359 static int gfx_v11_0_config_pfp_cache_rs64(struct amdgpu_device *adev, uint64_t addr, uint64_t addr2)
2360 {
2361 	uint32_t usec_timeout = 50000;  /* wait for 50ms */
2362 	uint32_t tmp;
2363 	unsigned i, pipe_id;
2364 	const struct gfx_firmware_header_v2_0 *pfp_hdr;
2365 
2366 	pfp_hdr = (const struct gfx_firmware_header_v2_0 *)
2367 		adev->gfx.pfp_fw->data;
2368 
2369 	WREG32_SOC15(GC, 0, regCP_PFP_IC_BASE_LO,
2370 		lower_32_bits(addr));
2371 	WREG32_SOC15(GC, 0, regCP_PFP_IC_BASE_HI,
2372 		upper_32_bits(addr));
2373 
2374 	tmp = RREG32_SOC15(GC, 0, regCP_PFP_IC_BASE_CNTL);
2375 	tmp = REG_SET_FIELD(tmp, CP_PFP_IC_BASE_CNTL, VMID, 0);
2376 	tmp = REG_SET_FIELD(tmp, CP_PFP_IC_BASE_CNTL, CACHE_POLICY, 0);
2377 	tmp = REG_SET_FIELD(tmp, CP_PFP_IC_BASE_CNTL, EXE_DISABLE, 0);
2378 	WREG32_SOC15(GC, 0, regCP_PFP_IC_BASE_CNTL, tmp);
2379 
2380 	/*
2381 	 * Programming any of the CP_PFP_IC_BASE registers
2382 	 * forces invalidation of the ME L1 I$. Wait for the
2383 	 * invalidation complete
2384 	 */
2385 	for (i = 0; i < usec_timeout; i++) {
2386 		tmp = RREG32_SOC15(GC, 0, regCP_PFP_IC_OP_CNTL);
2387 		if (1 == REG_GET_FIELD(tmp, CP_PFP_IC_OP_CNTL,
2388 			INVALIDATE_CACHE_COMPLETE))
2389 			break;
2390 		udelay(1);
2391 	}
2392 
2393 	if (i >= usec_timeout) {
2394 		dev_err(adev->dev, "failed to invalidate instruction cache\n");
2395 		return -EINVAL;
2396 	}
2397 
2398 	/* Prime the L1 instruction caches */
2399 	tmp = RREG32_SOC15(GC, 0, regCP_PFP_IC_OP_CNTL);
2400 	tmp = REG_SET_FIELD(tmp, CP_PFP_IC_OP_CNTL, PRIME_ICACHE, 1);
2401 	WREG32_SOC15(GC, 0, regCP_PFP_IC_OP_CNTL, tmp);
2402 	/* Waiting for cache primed*/
2403 	for (i = 0; i < usec_timeout; i++) {
2404 		tmp = RREG32_SOC15(GC, 0, regCP_PFP_IC_OP_CNTL);
2405 		if (1 == REG_GET_FIELD(tmp, CP_PFP_IC_OP_CNTL,
2406 			ICACHE_PRIMED))
2407 			break;
2408 		udelay(1);
2409 	}
2410 
2411 	if (i >= usec_timeout) {
2412 		dev_err(adev->dev, "failed to prime instruction cache\n");
2413 		return -EINVAL;
2414 	}
2415 
2416 	mutex_lock(&adev->srbm_mutex);
2417 	for (pipe_id = 0; pipe_id < adev->gfx.me.num_pipe_per_me; pipe_id++) {
2418 		soc21_grbm_select(adev, 0, pipe_id, 0, 0);
2419 		WREG32_SOC15(GC, 0, regCP_PFP_PRGRM_CNTR_START,
2420 			(pfp_hdr->ucode_start_addr_hi << 30) |
2421 			(pfp_hdr->ucode_start_addr_lo >> 2));
2422 		WREG32_SOC15(GC, 0, regCP_PFP_PRGRM_CNTR_START_HI,
2423 			pfp_hdr->ucode_start_addr_hi >> 2);
2424 
2425 		/*
2426 		 * Program CP_ME_CNTL to reset given PIPE to take
2427 		 * effect of CP_PFP_PRGRM_CNTR_START.
2428 		 */
2429 		tmp = RREG32_SOC15(GC, 0, regCP_ME_CNTL);
2430 		if (pipe_id == 0)
2431 			tmp = REG_SET_FIELD(tmp, CP_ME_CNTL,
2432 					PFP_PIPE0_RESET, 1);
2433 		else
2434 			tmp = REG_SET_FIELD(tmp, CP_ME_CNTL,
2435 					PFP_PIPE1_RESET, 1);
2436 		WREG32_SOC15(GC, 0, regCP_ME_CNTL, tmp);
2437 
2438 		/* Clear pfp pipe0 reset bit. */
2439 		if (pipe_id == 0)
2440 			tmp = REG_SET_FIELD(tmp, CP_ME_CNTL,
2441 					PFP_PIPE0_RESET, 0);
2442 		else
2443 			tmp = REG_SET_FIELD(tmp, CP_ME_CNTL,
2444 					PFP_PIPE1_RESET, 0);
2445 		WREG32_SOC15(GC, 0, regCP_ME_CNTL, tmp);
2446 
2447 		WREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_BASE0_LO,
2448 			lower_32_bits(addr2));
2449 		WREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_BASE0_HI,
2450 			upper_32_bits(addr2));
2451 	}
2452 	soc21_grbm_select(adev, 0, 0, 0, 0);
2453 	mutex_unlock(&adev->srbm_mutex);
2454 
2455 	tmp = RREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_BASE_CNTL);
2456 	tmp = REG_SET_FIELD(tmp, CP_GFX_RS64_DC_BASE_CNTL, VMID, 0);
2457 	tmp = REG_SET_FIELD(tmp, CP_GFX_RS64_DC_BASE_CNTL, CACHE_POLICY, 0);
2458 	WREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_BASE_CNTL, tmp);
2459 
2460 	/* Invalidate the data caches */
2461 	tmp = RREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_OP_CNTL);
2462 	tmp = REG_SET_FIELD(tmp, CP_GFX_RS64_DC_OP_CNTL, INVALIDATE_DCACHE, 1);
2463 	WREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_OP_CNTL, tmp);
2464 
2465 	for (i = 0; i < usec_timeout; i++) {
2466 		tmp = RREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_OP_CNTL);
2467 		if (1 == REG_GET_FIELD(tmp, CP_GFX_RS64_DC_OP_CNTL,
2468 			INVALIDATE_DCACHE_COMPLETE))
2469 			break;
2470 		udelay(1);
2471 	}
2472 
2473 	if (i >= usec_timeout) {
2474 		dev_err(adev->dev, "failed to invalidate RS64 data cache\n");
2475 		return -EINVAL;
2476 	}
2477 
2478 	return 0;
2479 }
2480 
2481 static int gfx_v11_0_config_me_cache_rs64(struct amdgpu_device *adev, uint64_t addr, uint64_t addr2)
2482 {
2483 	uint32_t usec_timeout = 50000;  /* wait for 50ms */
2484 	uint32_t tmp;
2485 	unsigned i, pipe_id;
2486 	const struct gfx_firmware_header_v2_0 *me_hdr;
2487 
2488 	me_hdr = (const struct gfx_firmware_header_v2_0 *)
2489 		adev->gfx.me_fw->data;
2490 
2491 	WREG32_SOC15(GC, 0, regCP_ME_IC_BASE_LO,
2492 		lower_32_bits(addr));
2493 	WREG32_SOC15(GC, 0, regCP_ME_IC_BASE_HI,
2494 		upper_32_bits(addr));
2495 
2496 	tmp = RREG32_SOC15(GC, 0, regCP_ME_IC_BASE_CNTL);
2497 	tmp = REG_SET_FIELD(tmp, CP_ME_IC_BASE_CNTL, VMID, 0);
2498 	tmp = REG_SET_FIELD(tmp, CP_ME_IC_BASE_CNTL, CACHE_POLICY, 0);
2499 	tmp = REG_SET_FIELD(tmp, CP_ME_IC_BASE_CNTL, EXE_DISABLE, 0);
2500 	WREG32_SOC15(GC, 0, regCP_ME_IC_BASE_CNTL, tmp);
2501 
2502 	/*
2503 	 * Programming any of the CP_ME_IC_BASE registers
2504 	 * forces invalidation of the ME L1 I$. Wait for the
2505 	 * invalidation complete
2506 	 */
2507 	for (i = 0; i < usec_timeout; i++) {
2508 		tmp = RREG32_SOC15(GC, 0, regCP_ME_IC_OP_CNTL);
2509 		if (1 == REG_GET_FIELD(tmp, CP_ME_IC_OP_CNTL,
2510 			INVALIDATE_CACHE_COMPLETE))
2511 			break;
2512 		udelay(1);
2513 	}
2514 
2515 	if (i >= usec_timeout) {
2516 		dev_err(adev->dev, "failed to invalidate instruction cache\n");
2517 		return -EINVAL;
2518 	}
2519 
2520 	/* Prime the instruction caches */
2521 	tmp = RREG32_SOC15(GC, 0, regCP_ME_IC_OP_CNTL);
2522 	tmp = REG_SET_FIELD(tmp, CP_ME_IC_OP_CNTL, PRIME_ICACHE, 1);
2523 	WREG32_SOC15(GC, 0, regCP_ME_IC_OP_CNTL, tmp);
2524 
2525 	/* Waiting for instruction cache primed*/
2526 	for (i = 0; i < usec_timeout; i++) {
2527 		tmp = RREG32_SOC15(GC, 0, regCP_ME_IC_OP_CNTL);
2528 		if (1 == REG_GET_FIELD(tmp, CP_ME_IC_OP_CNTL,
2529 			ICACHE_PRIMED))
2530 			break;
2531 		udelay(1);
2532 	}
2533 
2534 	if (i >= usec_timeout) {
2535 		dev_err(adev->dev, "failed to prime instruction cache\n");
2536 		return -EINVAL;
2537 	}
2538 
2539 	mutex_lock(&adev->srbm_mutex);
2540 	for (pipe_id = 0; pipe_id < adev->gfx.me.num_pipe_per_me; pipe_id++) {
2541 		soc21_grbm_select(adev, 0, pipe_id, 0, 0);
2542 		WREG32_SOC15(GC, 0, regCP_ME_PRGRM_CNTR_START,
2543 			(me_hdr->ucode_start_addr_hi << 30) |
2544 			(me_hdr->ucode_start_addr_lo >> 2) );
2545 		WREG32_SOC15(GC, 0, regCP_ME_PRGRM_CNTR_START_HI,
2546 			me_hdr->ucode_start_addr_hi>>2);
2547 
2548 		/*
2549 		 * Program CP_ME_CNTL to reset given PIPE to take
2550 		 * effect of CP_PFP_PRGRM_CNTR_START.
2551 		 */
2552 		tmp = RREG32_SOC15(GC, 0, regCP_ME_CNTL);
2553 		if (pipe_id == 0)
2554 			tmp = REG_SET_FIELD(tmp, CP_ME_CNTL,
2555 					ME_PIPE0_RESET, 1);
2556 		else
2557 			tmp = REG_SET_FIELD(tmp, CP_ME_CNTL,
2558 					ME_PIPE1_RESET, 1);
2559 		WREG32_SOC15(GC, 0, regCP_ME_CNTL, tmp);
2560 
2561 		/* Clear pfp pipe0 reset bit. */
2562 		if (pipe_id == 0)
2563 			tmp = REG_SET_FIELD(tmp, CP_ME_CNTL,
2564 					ME_PIPE0_RESET, 0);
2565 		else
2566 			tmp = REG_SET_FIELD(tmp, CP_ME_CNTL,
2567 					ME_PIPE1_RESET, 0);
2568 		WREG32_SOC15(GC, 0, regCP_ME_CNTL, tmp);
2569 
2570 		WREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_BASE1_LO,
2571 			lower_32_bits(addr2));
2572 		WREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_BASE1_HI,
2573 			upper_32_bits(addr2));
2574 	}
2575 	soc21_grbm_select(adev, 0, 0, 0, 0);
2576 	mutex_unlock(&adev->srbm_mutex);
2577 
2578 	tmp = RREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_BASE_CNTL);
2579 	tmp = REG_SET_FIELD(tmp, CP_GFX_RS64_DC_BASE_CNTL, VMID, 0);
2580 	tmp = REG_SET_FIELD(tmp, CP_GFX_RS64_DC_BASE_CNTL, CACHE_POLICY, 0);
2581 	WREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_BASE_CNTL, tmp);
2582 
2583 	/* Invalidate the data caches */
2584 	tmp = RREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_OP_CNTL);
2585 	tmp = REG_SET_FIELD(tmp, CP_GFX_RS64_DC_OP_CNTL, INVALIDATE_DCACHE, 1);
2586 	WREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_OP_CNTL, tmp);
2587 
2588 	for (i = 0; i < usec_timeout; i++) {
2589 		tmp = RREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_OP_CNTL);
2590 		if (1 == REG_GET_FIELD(tmp, CP_GFX_RS64_DC_OP_CNTL,
2591 			INVALIDATE_DCACHE_COMPLETE))
2592 			break;
2593 		udelay(1);
2594 	}
2595 
2596 	if (i >= usec_timeout) {
2597 		dev_err(adev->dev, "failed to invalidate RS64 data cache\n");
2598 		return -EINVAL;
2599 	}
2600 
2601 	return 0;
2602 }
2603 
2604 static int gfx_v11_0_config_mec_cache_rs64(struct amdgpu_device *adev, uint64_t addr, uint64_t addr2)
2605 {
2606 	uint32_t usec_timeout = 50000;  /* wait for 50ms */
2607 	uint32_t tmp;
2608 	unsigned i;
2609 	const struct gfx_firmware_header_v2_0 *mec_hdr;
2610 
2611 	mec_hdr = (const struct gfx_firmware_header_v2_0 *)
2612 		adev->gfx.mec_fw->data;
2613 
2614 	tmp = RREG32_SOC15(GC, 0, regCP_CPC_IC_BASE_CNTL);
2615 	tmp = REG_SET_FIELD(tmp, CP_CPC_IC_BASE_CNTL, VMID, 0);
2616 	tmp = REG_SET_FIELD(tmp, CP_CPC_IC_BASE_CNTL, EXE_DISABLE, 0);
2617 	tmp = REG_SET_FIELD(tmp, CP_CPC_IC_BASE_CNTL, CACHE_POLICY, 0);
2618 	WREG32_SOC15(GC, 0, regCP_CPC_IC_BASE_CNTL, tmp);
2619 
2620 	tmp = RREG32_SOC15(GC, 0, regCP_MEC_DC_BASE_CNTL);
2621 	tmp = REG_SET_FIELD(tmp, CP_MEC_DC_BASE_CNTL, VMID, 0);
2622 	tmp = REG_SET_FIELD(tmp, CP_MEC_DC_BASE_CNTL, CACHE_POLICY, 0);
2623 	WREG32_SOC15(GC, 0, regCP_MEC_DC_BASE_CNTL, tmp);
2624 
2625 	mutex_lock(&adev->srbm_mutex);
2626 	for (i = 0; i < adev->gfx.mec.num_pipe_per_mec; i++) {
2627 		soc21_grbm_select(adev, 1, i, 0, 0);
2628 
2629 		WREG32_SOC15(GC, 0, regCP_MEC_MDBASE_LO, addr2);
2630 		WREG32_SOC15(GC, 0, regCP_MEC_MDBASE_HI,
2631 		     upper_32_bits(addr2));
2632 
2633 		WREG32_SOC15(GC, 0, regCP_MEC_RS64_PRGRM_CNTR_START,
2634 					mec_hdr->ucode_start_addr_lo >> 2 |
2635 					mec_hdr->ucode_start_addr_hi << 30);
2636 		WREG32_SOC15(GC, 0, regCP_MEC_RS64_PRGRM_CNTR_START_HI,
2637 					mec_hdr->ucode_start_addr_hi >> 2);
2638 
2639 		WREG32_SOC15(GC, 0, regCP_CPC_IC_BASE_LO, addr);
2640 		WREG32_SOC15(GC, 0, regCP_CPC_IC_BASE_HI,
2641 		     upper_32_bits(addr));
2642 	}
2643 	mutex_unlock(&adev->srbm_mutex);
2644 	soc21_grbm_select(adev, 0, 0, 0, 0);
2645 
2646 	/* Trigger an invalidation of the L1 instruction caches */
2647 	tmp = RREG32_SOC15(GC, 0, regCP_MEC_DC_OP_CNTL);
2648 	tmp = REG_SET_FIELD(tmp, CP_MEC_DC_OP_CNTL, INVALIDATE_DCACHE, 1);
2649 	WREG32_SOC15(GC, 0, regCP_MEC_DC_OP_CNTL, tmp);
2650 
2651 	/* Wait for invalidation complete */
2652 	for (i = 0; i < usec_timeout; i++) {
2653 		tmp = RREG32_SOC15(GC, 0, regCP_MEC_DC_OP_CNTL);
2654 		if (1 == REG_GET_FIELD(tmp, CP_MEC_DC_OP_CNTL,
2655 				       INVALIDATE_DCACHE_COMPLETE))
2656 			break;
2657 		udelay(1);
2658 	}
2659 
2660 	if (i >= usec_timeout) {
2661 		dev_err(adev->dev, "failed to invalidate instruction cache\n");
2662 		return -EINVAL;
2663 	}
2664 
2665 	/* Trigger an invalidation of the L1 instruction caches */
2666 	tmp = RREG32_SOC15(GC, 0, regCP_CPC_IC_OP_CNTL);
2667 	tmp = REG_SET_FIELD(tmp, CP_CPC_IC_OP_CNTL, INVALIDATE_CACHE, 1);
2668 	WREG32_SOC15(GC, 0, regCP_CPC_IC_OP_CNTL, tmp);
2669 
2670 	/* Wait for invalidation complete */
2671 	for (i = 0; i < usec_timeout; i++) {
2672 		tmp = RREG32_SOC15(GC, 0, regCP_CPC_IC_OP_CNTL);
2673 		if (1 == REG_GET_FIELD(tmp, CP_CPC_IC_OP_CNTL,
2674 				       INVALIDATE_CACHE_COMPLETE))
2675 			break;
2676 		udelay(1);
2677 	}
2678 
2679 	if (i >= usec_timeout) {
2680 		dev_err(adev->dev, "failed to invalidate instruction cache\n");
2681 		return -EINVAL;
2682 	}
2683 
2684 	return 0;
2685 }
2686 
2687 static void gfx_v11_0_config_gfx_rs64(struct amdgpu_device *adev)
2688 {
2689 	const struct gfx_firmware_header_v2_0 *pfp_hdr;
2690 	const struct gfx_firmware_header_v2_0 *me_hdr;
2691 	const struct gfx_firmware_header_v2_0 *mec_hdr;
2692 	uint32_t pipe_id, tmp;
2693 
2694 	mec_hdr = (const struct gfx_firmware_header_v2_0 *)
2695 		adev->gfx.mec_fw->data;
2696 	me_hdr = (const struct gfx_firmware_header_v2_0 *)
2697 		adev->gfx.me_fw->data;
2698 	pfp_hdr = (const struct gfx_firmware_header_v2_0 *)
2699 		adev->gfx.pfp_fw->data;
2700 
2701 	/* config pfp program start addr */
2702 	for (pipe_id = 0; pipe_id < 2; pipe_id++) {
2703 		soc21_grbm_select(adev, 0, pipe_id, 0, 0);
2704 		WREG32_SOC15(GC, 0, regCP_PFP_PRGRM_CNTR_START,
2705 			(pfp_hdr->ucode_start_addr_hi << 30) |
2706 			(pfp_hdr->ucode_start_addr_lo >> 2));
2707 		WREG32_SOC15(GC, 0, regCP_PFP_PRGRM_CNTR_START_HI,
2708 			pfp_hdr->ucode_start_addr_hi >> 2);
2709 	}
2710 	soc21_grbm_select(adev, 0, 0, 0, 0);
2711 
2712 	/* reset pfp pipe */
2713 	tmp = RREG32_SOC15(GC, 0, regCP_ME_CNTL);
2714 	tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, PFP_PIPE0_RESET, 1);
2715 	tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, PFP_PIPE1_RESET, 1);
2716 	WREG32_SOC15(GC, 0, regCP_ME_CNTL, tmp);
2717 
2718 	/* clear pfp pipe reset */
2719 	tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, PFP_PIPE0_RESET, 0);
2720 	tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, PFP_PIPE1_RESET, 0);
2721 	WREG32_SOC15(GC, 0, regCP_ME_CNTL, tmp);
2722 
2723 	/* config me program start addr */
2724 	for (pipe_id = 0; pipe_id < 2; pipe_id++) {
2725 		soc21_grbm_select(adev, 0, pipe_id, 0, 0);
2726 		WREG32_SOC15(GC, 0, regCP_ME_PRGRM_CNTR_START,
2727 			(me_hdr->ucode_start_addr_hi << 30) |
2728 			(me_hdr->ucode_start_addr_lo >> 2) );
2729 		WREG32_SOC15(GC, 0, regCP_ME_PRGRM_CNTR_START_HI,
2730 			me_hdr->ucode_start_addr_hi>>2);
2731 	}
2732 	soc21_grbm_select(adev, 0, 0, 0, 0);
2733 
2734 	/* reset me pipe */
2735 	tmp = RREG32_SOC15(GC, 0, regCP_ME_CNTL);
2736 	tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, ME_PIPE0_RESET, 1);
2737 	tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, ME_PIPE1_RESET, 1);
2738 	WREG32_SOC15(GC, 0, regCP_ME_CNTL, tmp);
2739 
2740 	/* clear me pipe reset */
2741 	tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, ME_PIPE0_RESET, 0);
2742 	tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, ME_PIPE1_RESET, 0);
2743 	WREG32_SOC15(GC, 0, regCP_ME_CNTL, tmp);
2744 
2745 	/* config mec program start addr */
2746 	for (pipe_id = 0; pipe_id < 4; pipe_id++) {
2747 		soc21_grbm_select(adev, 1, pipe_id, 0, 0);
2748 		WREG32_SOC15(GC, 0, regCP_MEC_RS64_PRGRM_CNTR_START,
2749 					mec_hdr->ucode_start_addr_lo >> 2 |
2750 					mec_hdr->ucode_start_addr_hi << 30);
2751 		WREG32_SOC15(GC, 0, regCP_MEC_RS64_PRGRM_CNTR_START_HI,
2752 					mec_hdr->ucode_start_addr_hi >> 2);
2753 	}
2754 	soc21_grbm_select(adev, 0, 0, 0, 0);
2755 }
2756 
2757 static int gfx_v11_0_wait_for_rlc_autoload_complete(struct amdgpu_device *adev)
2758 {
2759 	uint32_t cp_status;
2760 	uint32_t bootload_status;
2761 	int i, r;
2762 	uint64_t addr, addr2;
2763 
2764 	for (i = 0; i < adev->usec_timeout; i++) {
2765 		cp_status = RREG32_SOC15(GC, 0, regCP_STAT);
2766 		bootload_status = RREG32_SOC15(GC, 0, regRLC_RLCS_BOOTLOAD_STATUS);
2767 		if ((cp_status == 0) &&
2768 		    (REG_GET_FIELD(bootload_status,
2769 			RLC_RLCS_BOOTLOAD_STATUS, BOOTLOAD_COMPLETE) == 1)) {
2770 			break;
2771 		}
2772 		udelay(1);
2773 	}
2774 
2775 	if (i >= adev->usec_timeout) {
2776 		dev_err(adev->dev, "rlc autoload: gc ucode autoload timeout\n");
2777 		return -ETIMEDOUT;
2778 	}
2779 
2780 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_RLC_BACKDOOR_AUTO) {
2781 		if (adev->gfx.rs64_enable) {
2782 			addr = adev->gfx.rlc.rlc_autoload_gpu_addr +
2783 				rlc_autoload_info[SOC21_FIRMWARE_ID_RS64_ME].offset;
2784 			addr2 = adev->gfx.rlc.rlc_autoload_gpu_addr +
2785 				rlc_autoload_info[SOC21_FIRMWARE_ID_RS64_ME_P0_STACK].offset;
2786 			r = gfx_v11_0_config_me_cache_rs64(adev, addr, addr2);
2787 			if (r)
2788 				return r;
2789 			addr = adev->gfx.rlc.rlc_autoload_gpu_addr +
2790 				rlc_autoload_info[SOC21_FIRMWARE_ID_RS64_PFP].offset;
2791 			addr2 = adev->gfx.rlc.rlc_autoload_gpu_addr +
2792 				rlc_autoload_info[SOC21_FIRMWARE_ID_RS64_PFP_P0_STACK].offset;
2793 			r = gfx_v11_0_config_pfp_cache_rs64(adev, addr, addr2);
2794 			if (r)
2795 				return r;
2796 			addr = adev->gfx.rlc.rlc_autoload_gpu_addr +
2797 				rlc_autoload_info[SOC21_FIRMWARE_ID_RS64_MEC].offset;
2798 			addr2 = adev->gfx.rlc.rlc_autoload_gpu_addr +
2799 				rlc_autoload_info[SOC21_FIRMWARE_ID_RS64_MEC_P0_STACK].offset;
2800 			r = gfx_v11_0_config_mec_cache_rs64(adev, addr, addr2);
2801 			if (r)
2802 				return r;
2803 		} else {
2804 			addr = adev->gfx.rlc.rlc_autoload_gpu_addr +
2805 				rlc_autoload_info[SOC21_FIRMWARE_ID_CP_ME].offset;
2806 			r = gfx_v11_0_config_me_cache(adev, addr);
2807 			if (r)
2808 				return r;
2809 			addr = adev->gfx.rlc.rlc_autoload_gpu_addr +
2810 				rlc_autoload_info[SOC21_FIRMWARE_ID_CP_PFP].offset;
2811 			r = gfx_v11_0_config_pfp_cache(adev, addr);
2812 			if (r)
2813 				return r;
2814 			addr = adev->gfx.rlc.rlc_autoload_gpu_addr +
2815 				rlc_autoload_info[SOC21_FIRMWARE_ID_CP_MEC].offset;
2816 			r = gfx_v11_0_config_mec_cache(adev, addr);
2817 			if (r)
2818 				return r;
2819 		}
2820 	}
2821 
2822 	return 0;
2823 }
2824 
2825 static int gfx_v11_0_cp_gfx_enable(struct amdgpu_device *adev, bool enable)
2826 {
2827 	int i;
2828 	u32 tmp = RREG32_SOC15(GC, 0, regCP_ME_CNTL);
2829 
2830 	tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, ME_HALT, enable ? 0 : 1);
2831 	tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, PFP_HALT, enable ? 0 : 1);
2832 	WREG32_SOC15(GC, 0, regCP_ME_CNTL, tmp);
2833 
2834 	for (i = 0; i < adev->usec_timeout; i++) {
2835 		if (RREG32_SOC15(GC, 0, regCP_STAT) == 0)
2836 			break;
2837 		udelay(1);
2838 	}
2839 
2840 	if (i >= adev->usec_timeout)
2841 		DRM_ERROR("failed to %s cp gfx\n", enable ? "unhalt" : "halt");
2842 
2843 	return 0;
2844 }
2845 
2846 static int gfx_v11_0_cp_gfx_load_pfp_microcode(struct amdgpu_device *adev)
2847 {
2848 	int r;
2849 	const struct gfx_firmware_header_v1_0 *pfp_hdr;
2850 	const __le32 *fw_data;
2851 	unsigned i, fw_size;
2852 
2853 	pfp_hdr = (const struct gfx_firmware_header_v1_0 *)
2854 		adev->gfx.pfp_fw->data;
2855 
2856 	amdgpu_ucode_print_gfx_hdr(&pfp_hdr->header);
2857 
2858 	fw_data = (const __le32 *)(adev->gfx.pfp_fw->data +
2859 		le32_to_cpu(pfp_hdr->header.ucode_array_offset_bytes));
2860 	fw_size = le32_to_cpu(pfp_hdr->header.ucode_size_bytes);
2861 
2862 	r = amdgpu_bo_create_reserved(adev, pfp_hdr->header.ucode_size_bytes,
2863 				      PAGE_SIZE, AMDGPU_GEM_DOMAIN_GTT,
2864 				      &adev->gfx.pfp.pfp_fw_obj,
2865 				      &adev->gfx.pfp.pfp_fw_gpu_addr,
2866 				      (void **)&adev->gfx.pfp.pfp_fw_ptr);
2867 	if (r) {
2868 		dev_err(adev->dev, "(%d) failed to create pfp fw bo\n", r);
2869 		gfx_v11_0_pfp_fini(adev);
2870 		return r;
2871 	}
2872 
2873 	memcpy(adev->gfx.pfp.pfp_fw_ptr, fw_data, fw_size);
2874 
2875 	amdgpu_bo_kunmap(adev->gfx.pfp.pfp_fw_obj);
2876 	amdgpu_bo_unreserve(adev->gfx.pfp.pfp_fw_obj);
2877 
2878 	gfx_v11_0_config_pfp_cache(adev, adev->gfx.pfp.pfp_fw_gpu_addr);
2879 
2880 	WREG32_SOC15(GC, 0, regCP_HYP_PFP_UCODE_ADDR, 0);
2881 
2882 	for (i = 0; i < pfp_hdr->jt_size; i++)
2883 		WREG32_SOC15(GC, 0, regCP_HYP_PFP_UCODE_DATA,
2884 			     le32_to_cpup(fw_data + pfp_hdr->jt_offset + i));
2885 
2886 	WREG32_SOC15(GC, 0, regCP_HYP_PFP_UCODE_ADDR, adev->gfx.pfp_fw_version);
2887 
2888 	return 0;
2889 }
2890 
2891 static int gfx_v11_0_cp_gfx_load_pfp_microcode_rs64(struct amdgpu_device *adev)
2892 {
2893 	int r;
2894 	const struct gfx_firmware_header_v2_0 *pfp_hdr;
2895 	const __le32 *fw_ucode, *fw_data;
2896 	unsigned i, pipe_id, fw_ucode_size, fw_data_size;
2897 	uint32_t tmp;
2898 	uint32_t usec_timeout = 50000;  /* wait for 50ms */
2899 
2900 	pfp_hdr = (const struct gfx_firmware_header_v2_0 *)
2901 		adev->gfx.pfp_fw->data;
2902 
2903 	amdgpu_ucode_print_gfx_hdr(&pfp_hdr->header);
2904 
2905 	/* instruction */
2906 	fw_ucode = (const __le32 *)(adev->gfx.pfp_fw->data +
2907 		le32_to_cpu(pfp_hdr->ucode_offset_bytes));
2908 	fw_ucode_size = le32_to_cpu(pfp_hdr->ucode_size_bytes);
2909 	/* data */
2910 	fw_data = (const __le32 *)(adev->gfx.pfp_fw->data +
2911 		le32_to_cpu(pfp_hdr->data_offset_bytes));
2912 	fw_data_size = le32_to_cpu(pfp_hdr->data_size_bytes);
2913 
2914 	/* 64kb align */
2915 	r = amdgpu_bo_create_reserved(adev, fw_ucode_size,
2916 				      64 * 1024, AMDGPU_GEM_DOMAIN_VRAM,
2917 				      &adev->gfx.pfp.pfp_fw_obj,
2918 				      &adev->gfx.pfp.pfp_fw_gpu_addr,
2919 				      (void **)&adev->gfx.pfp.pfp_fw_ptr);
2920 	if (r) {
2921 		dev_err(adev->dev, "(%d) failed to create pfp ucode fw bo\n", r);
2922 		gfx_v11_0_pfp_fini(adev);
2923 		return r;
2924 	}
2925 
2926 	r = amdgpu_bo_create_reserved(adev, fw_data_size,
2927 				      64 * 1024, AMDGPU_GEM_DOMAIN_VRAM,
2928 				      &adev->gfx.pfp.pfp_fw_data_obj,
2929 				      &adev->gfx.pfp.pfp_fw_data_gpu_addr,
2930 				      (void **)&adev->gfx.pfp.pfp_fw_data_ptr);
2931 	if (r) {
2932 		dev_err(adev->dev, "(%d) failed to create pfp data fw bo\n", r);
2933 		gfx_v11_0_pfp_fini(adev);
2934 		return r;
2935 	}
2936 
2937 	memcpy(adev->gfx.pfp.pfp_fw_ptr, fw_ucode, fw_ucode_size);
2938 	memcpy(adev->gfx.pfp.pfp_fw_data_ptr, fw_data, fw_data_size);
2939 
2940 	amdgpu_bo_kunmap(adev->gfx.pfp.pfp_fw_obj);
2941 	amdgpu_bo_kunmap(adev->gfx.pfp.pfp_fw_data_obj);
2942 	amdgpu_bo_unreserve(adev->gfx.pfp.pfp_fw_obj);
2943 	amdgpu_bo_unreserve(adev->gfx.pfp.pfp_fw_data_obj);
2944 
2945 	if (amdgpu_emu_mode == 1)
2946 		adev->hdp.funcs->flush_hdp(adev, NULL);
2947 
2948 	WREG32_SOC15(GC, 0, regCP_PFP_IC_BASE_LO,
2949 		lower_32_bits(adev->gfx.pfp.pfp_fw_gpu_addr));
2950 	WREG32_SOC15(GC, 0, regCP_PFP_IC_BASE_HI,
2951 		upper_32_bits(adev->gfx.pfp.pfp_fw_gpu_addr));
2952 
2953 	tmp = RREG32_SOC15(GC, 0, regCP_PFP_IC_BASE_CNTL);
2954 	tmp = REG_SET_FIELD(tmp, CP_PFP_IC_BASE_CNTL, VMID, 0);
2955 	tmp = REG_SET_FIELD(tmp, CP_PFP_IC_BASE_CNTL, CACHE_POLICY, 0);
2956 	tmp = REG_SET_FIELD(tmp, CP_PFP_IC_BASE_CNTL, EXE_DISABLE, 0);
2957 	WREG32_SOC15(GC, 0, regCP_PFP_IC_BASE_CNTL, tmp);
2958 
2959 	/*
2960 	 * Programming any of the CP_PFP_IC_BASE registers
2961 	 * forces invalidation of the ME L1 I$. Wait for the
2962 	 * invalidation complete
2963 	 */
2964 	for (i = 0; i < usec_timeout; i++) {
2965 		tmp = RREG32_SOC15(GC, 0, regCP_PFP_IC_OP_CNTL);
2966 		if (1 == REG_GET_FIELD(tmp, CP_PFP_IC_OP_CNTL,
2967 			INVALIDATE_CACHE_COMPLETE))
2968 			break;
2969 		udelay(1);
2970 	}
2971 
2972 	if (i >= usec_timeout) {
2973 		dev_err(adev->dev, "failed to invalidate instruction cache\n");
2974 		return -EINVAL;
2975 	}
2976 
2977 	/* Prime the L1 instruction caches */
2978 	tmp = RREG32_SOC15(GC, 0, regCP_PFP_IC_OP_CNTL);
2979 	tmp = REG_SET_FIELD(tmp, CP_PFP_IC_OP_CNTL, PRIME_ICACHE, 1);
2980 	WREG32_SOC15(GC, 0, regCP_PFP_IC_OP_CNTL, tmp);
2981 	/* Waiting for cache primed*/
2982 	for (i = 0; i < usec_timeout; i++) {
2983 		tmp = RREG32_SOC15(GC, 0, regCP_PFP_IC_OP_CNTL);
2984 		if (1 == REG_GET_FIELD(tmp, CP_PFP_IC_OP_CNTL,
2985 			ICACHE_PRIMED))
2986 			break;
2987 		udelay(1);
2988 	}
2989 
2990 	if (i >= usec_timeout) {
2991 		dev_err(adev->dev, "failed to prime instruction cache\n");
2992 		return -EINVAL;
2993 	}
2994 
2995 	mutex_lock(&adev->srbm_mutex);
2996 	for (pipe_id = 0; pipe_id < adev->gfx.me.num_pipe_per_me; pipe_id++) {
2997 		soc21_grbm_select(adev, 0, pipe_id, 0, 0);
2998 		WREG32_SOC15(GC, 0, regCP_PFP_PRGRM_CNTR_START,
2999 			(pfp_hdr->ucode_start_addr_hi << 30) |
3000 			(pfp_hdr->ucode_start_addr_lo >> 2) );
3001 		WREG32_SOC15(GC, 0, regCP_PFP_PRGRM_CNTR_START_HI,
3002 			pfp_hdr->ucode_start_addr_hi>>2);
3003 
3004 		/*
3005 		 * Program CP_ME_CNTL to reset given PIPE to take
3006 		 * effect of CP_PFP_PRGRM_CNTR_START.
3007 		 */
3008 		tmp = RREG32_SOC15(GC, 0, regCP_ME_CNTL);
3009 		if (pipe_id == 0)
3010 			tmp = REG_SET_FIELD(tmp, CP_ME_CNTL,
3011 					PFP_PIPE0_RESET, 1);
3012 		else
3013 			tmp = REG_SET_FIELD(tmp, CP_ME_CNTL,
3014 					PFP_PIPE1_RESET, 1);
3015 		WREG32_SOC15(GC, 0, regCP_ME_CNTL, tmp);
3016 
3017 		/* Clear pfp pipe0 reset bit. */
3018 		if (pipe_id == 0)
3019 			tmp = REG_SET_FIELD(tmp, CP_ME_CNTL,
3020 					PFP_PIPE0_RESET, 0);
3021 		else
3022 			tmp = REG_SET_FIELD(tmp, CP_ME_CNTL,
3023 					PFP_PIPE1_RESET, 0);
3024 		WREG32_SOC15(GC, 0, regCP_ME_CNTL, tmp);
3025 
3026 		WREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_BASE0_LO,
3027 			lower_32_bits(adev->gfx.pfp.pfp_fw_data_gpu_addr));
3028 		WREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_BASE0_HI,
3029 			upper_32_bits(adev->gfx.pfp.pfp_fw_data_gpu_addr));
3030 	}
3031 	soc21_grbm_select(adev, 0, 0, 0, 0);
3032 	mutex_unlock(&adev->srbm_mutex);
3033 
3034 	tmp = RREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_BASE_CNTL);
3035 	tmp = REG_SET_FIELD(tmp, CP_GFX_RS64_DC_BASE_CNTL, VMID, 0);
3036 	tmp = REG_SET_FIELD(tmp, CP_GFX_RS64_DC_BASE_CNTL, CACHE_POLICY, 0);
3037 	WREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_BASE_CNTL, tmp);
3038 
3039 	/* Invalidate the data caches */
3040 	tmp = RREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_OP_CNTL);
3041 	tmp = REG_SET_FIELD(tmp, CP_GFX_RS64_DC_OP_CNTL, INVALIDATE_DCACHE, 1);
3042 	WREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_OP_CNTL, tmp);
3043 
3044 	for (i = 0; i < usec_timeout; i++) {
3045 		tmp = RREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_OP_CNTL);
3046 		if (1 == REG_GET_FIELD(tmp, CP_GFX_RS64_DC_OP_CNTL,
3047 			INVALIDATE_DCACHE_COMPLETE))
3048 			break;
3049 		udelay(1);
3050 	}
3051 
3052 	if (i >= usec_timeout) {
3053 		dev_err(adev->dev, "failed to invalidate RS64 data cache\n");
3054 		return -EINVAL;
3055 	}
3056 
3057 	return 0;
3058 }
3059 
3060 static int gfx_v11_0_cp_gfx_load_me_microcode(struct amdgpu_device *adev)
3061 {
3062 	int r;
3063 	const struct gfx_firmware_header_v1_0 *me_hdr;
3064 	const __le32 *fw_data;
3065 	unsigned i, fw_size;
3066 
3067 	me_hdr = (const struct gfx_firmware_header_v1_0 *)
3068 		adev->gfx.me_fw->data;
3069 
3070 	amdgpu_ucode_print_gfx_hdr(&me_hdr->header);
3071 
3072 	fw_data = (const __le32 *)(adev->gfx.me_fw->data +
3073 		le32_to_cpu(me_hdr->header.ucode_array_offset_bytes));
3074 	fw_size = le32_to_cpu(me_hdr->header.ucode_size_bytes);
3075 
3076 	r = amdgpu_bo_create_reserved(adev, me_hdr->header.ucode_size_bytes,
3077 				      PAGE_SIZE, AMDGPU_GEM_DOMAIN_GTT,
3078 				      &adev->gfx.me.me_fw_obj,
3079 				      &adev->gfx.me.me_fw_gpu_addr,
3080 				      (void **)&adev->gfx.me.me_fw_ptr);
3081 	if (r) {
3082 		dev_err(adev->dev, "(%d) failed to create me fw bo\n", r);
3083 		gfx_v11_0_me_fini(adev);
3084 		return r;
3085 	}
3086 
3087 	memcpy(adev->gfx.me.me_fw_ptr, fw_data, fw_size);
3088 
3089 	amdgpu_bo_kunmap(adev->gfx.me.me_fw_obj);
3090 	amdgpu_bo_unreserve(adev->gfx.me.me_fw_obj);
3091 
3092 	gfx_v11_0_config_me_cache(adev, adev->gfx.me.me_fw_gpu_addr);
3093 
3094 	WREG32_SOC15(GC, 0, regCP_HYP_ME_UCODE_ADDR, 0);
3095 
3096 	for (i = 0; i < me_hdr->jt_size; i++)
3097 		WREG32_SOC15(GC, 0, regCP_HYP_ME_UCODE_DATA,
3098 			     le32_to_cpup(fw_data + me_hdr->jt_offset + i));
3099 
3100 	WREG32_SOC15(GC, 0, regCP_HYP_ME_UCODE_ADDR, adev->gfx.me_fw_version);
3101 
3102 	return 0;
3103 }
3104 
3105 static int gfx_v11_0_cp_gfx_load_me_microcode_rs64(struct amdgpu_device *adev)
3106 {
3107 	int r;
3108 	const struct gfx_firmware_header_v2_0 *me_hdr;
3109 	const __le32 *fw_ucode, *fw_data;
3110 	unsigned i, pipe_id, fw_ucode_size, fw_data_size;
3111 	uint32_t tmp;
3112 	uint32_t usec_timeout = 50000;  /* wait for 50ms */
3113 
3114 	me_hdr = (const struct gfx_firmware_header_v2_0 *)
3115 		adev->gfx.me_fw->data;
3116 
3117 	amdgpu_ucode_print_gfx_hdr(&me_hdr->header);
3118 
3119 	/* instruction */
3120 	fw_ucode = (const __le32 *)(adev->gfx.me_fw->data +
3121 		le32_to_cpu(me_hdr->ucode_offset_bytes));
3122 	fw_ucode_size = le32_to_cpu(me_hdr->ucode_size_bytes);
3123 	/* data */
3124 	fw_data = (const __le32 *)(adev->gfx.me_fw->data +
3125 		le32_to_cpu(me_hdr->data_offset_bytes));
3126 	fw_data_size = le32_to_cpu(me_hdr->data_size_bytes);
3127 
3128 	/* 64kb align*/
3129 	r = amdgpu_bo_create_reserved(adev, fw_ucode_size,
3130 				      64 * 1024, AMDGPU_GEM_DOMAIN_VRAM,
3131 				      &adev->gfx.me.me_fw_obj,
3132 				      &adev->gfx.me.me_fw_gpu_addr,
3133 				      (void **)&adev->gfx.me.me_fw_ptr);
3134 	if (r) {
3135 		dev_err(adev->dev, "(%d) failed to create me ucode bo\n", r);
3136 		gfx_v11_0_me_fini(adev);
3137 		return r;
3138 	}
3139 
3140 	r = amdgpu_bo_create_reserved(adev, fw_data_size,
3141 				      64 * 1024, AMDGPU_GEM_DOMAIN_VRAM,
3142 				      &adev->gfx.me.me_fw_data_obj,
3143 				      &adev->gfx.me.me_fw_data_gpu_addr,
3144 				      (void **)&adev->gfx.me.me_fw_data_ptr);
3145 	if (r) {
3146 		dev_err(adev->dev, "(%d) failed to create me data bo\n", r);
3147 		gfx_v11_0_pfp_fini(adev);
3148 		return r;
3149 	}
3150 
3151 	memcpy(adev->gfx.me.me_fw_ptr, fw_ucode, fw_ucode_size);
3152 	memcpy(adev->gfx.me.me_fw_data_ptr, fw_data, fw_data_size);
3153 
3154 	amdgpu_bo_kunmap(adev->gfx.me.me_fw_obj);
3155 	amdgpu_bo_kunmap(adev->gfx.me.me_fw_data_obj);
3156 	amdgpu_bo_unreserve(adev->gfx.me.me_fw_obj);
3157 	amdgpu_bo_unreserve(adev->gfx.me.me_fw_data_obj);
3158 
3159 	if (amdgpu_emu_mode == 1)
3160 		adev->hdp.funcs->flush_hdp(adev, NULL);
3161 
3162 	WREG32_SOC15(GC, 0, regCP_ME_IC_BASE_LO,
3163 		lower_32_bits(adev->gfx.me.me_fw_gpu_addr));
3164 	WREG32_SOC15(GC, 0, regCP_ME_IC_BASE_HI,
3165 		upper_32_bits(adev->gfx.me.me_fw_gpu_addr));
3166 
3167 	tmp = RREG32_SOC15(GC, 0, regCP_ME_IC_BASE_CNTL);
3168 	tmp = REG_SET_FIELD(tmp, CP_ME_IC_BASE_CNTL, VMID, 0);
3169 	tmp = REG_SET_FIELD(tmp, CP_ME_IC_BASE_CNTL, CACHE_POLICY, 0);
3170 	tmp = REG_SET_FIELD(tmp, CP_ME_IC_BASE_CNTL, EXE_DISABLE, 0);
3171 	WREG32_SOC15(GC, 0, regCP_ME_IC_BASE_CNTL, tmp);
3172 
3173 	/*
3174 	 * Programming any of the CP_ME_IC_BASE registers
3175 	 * forces invalidation of the ME L1 I$. Wait for the
3176 	 * invalidation complete
3177 	 */
3178 	for (i = 0; i < usec_timeout; i++) {
3179 		tmp = RREG32_SOC15(GC, 0, regCP_ME_IC_OP_CNTL);
3180 		if (1 == REG_GET_FIELD(tmp, CP_ME_IC_OP_CNTL,
3181 			INVALIDATE_CACHE_COMPLETE))
3182 			break;
3183 		udelay(1);
3184 	}
3185 
3186 	if (i >= usec_timeout) {
3187 		dev_err(adev->dev, "failed to invalidate instruction cache\n");
3188 		return -EINVAL;
3189 	}
3190 
3191 	/* Prime the instruction caches */
3192 	tmp = RREG32_SOC15(GC, 0, regCP_ME_IC_OP_CNTL);
3193 	tmp = REG_SET_FIELD(tmp, CP_ME_IC_OP_CNTL, PRIME_ICACHE, 1);
3194 	WREG32_SOC15(GC, 0, regCP_ME_IC_OP_CNTL, tmp);
3195 
3196 	/* Waiting for instruction cache primed*/
3197 	for (i = 0; i < usec_timeout; i++) {
3198 		tmp = RREG32_SOC15(GC, 0, regCP_ME_IC_OP_CNTL);
3199 		if (1 == REG_GET_FIELD(tmp, CP_ME_IC_OP_CNTL,
3200 			ICACHE_PRIMED))
3201 			break;
3202 		udelay(1);
3203 	}
3204 
3205 	if (i >= usec_timeout) {
3206 		dev_err(adev->dev, "failed to prime instruction cache\n");
3207 		return -EINVAL;
3208 	}
3209 
3210 	mutex_lock(&adev->srbm_mutex);
3211 	for (pipe_id = 0; pipe_id < adev->gfx.me.num_pipe_per_me; pipe_id++) {
3212 		soc21_grbm_select(adev, 0, pipe_id, 0, 0);
3213 		WREG32_SOC15(GC, 0, regCP_ME_PRGRM_CNTR_START,
3214 			(me_hdr->ucode_start_addr_hi << 30) |
3215 			(me_hdr->ucode_start_addr_lo >> 2) );
3216 		WREG32_SOC15(GC, 0, regCP_ME_PRGRM_CNTR_START_HI,
3217 			me_hdr->ucode_start_addr_hi>>2);
3218 
3219 		/*
3220 		 * Program CP_ME_CNTL to reset given PIPE to take
3221 		 * effect of CP_PFP_PRGRM_CNTR_START.
3222 		 */
3223 		tmp = RREG32_SOC15(GC, 0, regCP_ME_CNTL);
3224 		if (pipe_id == 0)
3225 			tmp = REG_SET_FIELD(tmp, CP_ME_CNTL,
3226 					ME_PIPE0_RESET, 1);
3227 		else
3228 			tmp = REG_SET_FIELD(tmp, CP_ME_CNTL,
3229 					ME_PIPE1_RESET, 1);
3230 		WREG32_SOC15(GC, 0, regCP_ME_CNTL, tmp);
3231 
3232 		/* Clear pfp pipe0 reset bit. */
3233 		if (pipe_id == 0)
3234 			tmp = REG_SET_FIELD(tmp, CP_ME_CNTL,
3235 					ME_PIPE0_RESET, 0);
3236 		else
3237 			tmp = REG_SET_FIELD(tmp, CP_ME_CNTL,
3238 					ME_PIPE1_RESET, 0);
3239 		WREG32_SOC15(GC, 0, regCP_ME_CNTL, tmp);
3240 
3241 		WREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_BASE1_LO,
3242 			lower_32_bits(adev->gfx.me.me_fw_data_gpu_addr));
3243 		WREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_BASE1_HI,
3244 			upper_32_bits(adev->gfx.me.me_fw_data_gpu_addr));
3245 	}
3246 	soc21_grbm_select(adev, 0, 0, 0, 0);
3247 	mutex_unlock(&adev->srbm_mutex);
3248 
3249 	tmp = RREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_BASE_CNTL);
3250 	tmp = REG_SET_FIELD(tmp, CP_GFX_RS64_DC_BASE_CNTL, VMID, 0);
3251 	tmp = REG_SET_FIELD(tmp, CP_GFX_RS64_DC_BASE_CNTL, CACHE_POLICY, 0);
3252 	WREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_BASE_CNTL, tmp);
3253 
3254 	/* Invalidate the data caches */
3255 	tmp = RREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_OP_CNTL);
3256 	tmp = REG_SET_FIELD(tmp, CP_GFX_RS64_DC_OP_CNTL, INVALIDATE_DCACHE, 1);
3257 	WREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_OP_CNTL, tmp);
3258 
3259 	for (i = 0; i < usec_timeout; i++) {
3260 		tmp = RREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_OP_CNTL);
3261 		if (1 == REG_GET_FIELD(tmp, CP_GFX_RS64_DC_OP_CNTL,
3262 			INVALIDATE_DCACHE_COMPLETE))
3263 			break;
3264 		udelay(1);
3265 	}
3266 
3267 	if (i >= usec_timeout) {
3268 		dev_err(adev->dev, "failed to invalidate RS64 data cache\n");
3269 		return -EINVAL;
3270 	}
3271 
3272 	return 0;
3273 }
3274 
3275 static int gfx_v11_0_cp_gfx_load_microcode(struct amdgpu_device *adev)
3276 {
3277 	int r;
3278 
3279 	if (!adev->gfx.me_fw || !adev->gfx.pfp_fw)
3280 		return -EINVAL;
3281 
3282 	gfx_v11_0_cp_gfx_enable(adev, false);
3283 
3284 	if (adev->gfx.rs64_enable)
3285 		r = gfx_v11_0_cp_gfx_load_pfp_microcode_rs64(adev);
3286 	else
3287 		r = gfx_v11_0_cp_gfx_load_pfp_microcode(adev);
3288 	if (r) {
3289 		dev_err(adev->dev, "(%d) failed to load pfp fw\n", r);
3290 		return r;
3291 	}
3292 
3293 	if (adev->gfx.rs64_enable)
3294 		r = gfx_v11_0_cp_gfx_load_me_microcode_rs64(adev);
3295 	else
3296 		r = gfx_v11_0_cp_gfx_load_me_microcode(adev);
3297 	if (r) {
3298 		dev_err(adev->dev, "(%d) failed to load me fw\n", r);
3299 		return r;
3300 	}
3301 
3302 	return 0;
3303 }
3304 
3305 static int gfx_v11_0_cp_gfx_start(struct amdgpu_device *adev)
3306 {
3307 	struct amdgpu_ring *ring;
3308 	const struct cs_section_def *sect = NULL;
3309 	const struct cs_extent_def *ext = NULL;
3310 	int r, i;
3311 	int ctx_reg_offset;
3312 
3313 	/* init the CP */
3314 	WREG32_SOC15(GC, 0, regCP_MAX_CONTEXT,
3315 		     adev->gfx.config.max_hw_contexts - 1);
3316 	WREG32_SOC15(GC, 0, regCP_DEVICE_ID, 1);
3317 
3318 	if (!amdgpu_async_gfx_ring)
3319 		gfx_v11_0_cp_gfx_enable(adev, true);
3320 
3321 	ring = &adev->gfx.gfx_ring[0];
3322 	r = amdgpu_ring_alloc(ring, gfx_v11_0_get_csb_size(adev));
3323 	if (r) {
3324 		DRM_ERROR("amdgpu: cp failed to lock ring (%d).\n", r);
3325 		return r;
3326 	}
3327 
3328 	amdgpu_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
3329 	amdgpu_ring_write(ring, PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
3330 
3331 	amdgpu_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1));
3332 	amdgpu_ring_write(ring, 0x80000000);
3333 	amdgpu_ring_write(ring, 0x80000000);
3334 
3335 	for (sect = gfx11_cs_data; sect->section != NULL; ++sect) {
3336 		for (ext = sect->section; ext->extent != NULL; ++ext) {
3337 			if (sect->id == SECT_CONTEXT) {
3338 				amdgpu_ring_write(ring,
3339 						  PACKET3(PACKET3_SET_CONTEXT_REG,
3340 							  ext->reg_count));
3341 				amdgpu_ring_write(ring, ext->reg_index -
3342 						  PACKET3_SET_CONTEXT_REG_START);
3343 				for (i = 0; i < ext->reg_count; i++)
3344 					amdgpu_ring_write(ring, ext->extent[i]);
3345 			}
3346 		}
3347 	}
3348 
3349 	ctx_reg_offset =
3350 		SOC15_REG_OFFSET(GC, 0, regPA_SC_TILE_STEERING_OVERRIDE) - PACKET3_SET_CONTEXT_REG_START;
3351 	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONTEXT_REG, 1));
3352 	amdgpu_ring_write(ring, ctx_reg_offset);
3353 	amdgpu_ring_write(ring, adev->gfx.config.pa_sc_tile_steering_override);
3354 
3355 	amdgpu_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
3356 	amdgpu_ring_write(ring, PACKET3_PREAMBLE_END_CLEAR_STATE);
3357 
3358 	amdgpu_ring_write(ring, PACKET3(PACKET3_CLEAR_STATE, 0));
3359 	amdgpu_ring_write(ring, 0);
3360 
3361 	amdgpu_ring_commit(ring);
3362 
3363 	/* submit cs packet to copy state 0 to next available state */
3364 	if (adev->gfx.num_gfx_rings > 1) {
3365 		/* maximum supported gfx ring is 2 */
3366 		ring = &adev->gfx.gfx_ring[1];
3367 		r = amdgpu_ring_alloc(ring, 2);
3368 		if (r) {
3369 			DRM_ERROR("amdgpu: cp failed to lock ring (%d).\n", r);
3370 			return r;
3371 		}
3372 
3373 		amdgpu_ring_write(ring, PACKET3(PACKET3_CLEAR_STATE, 0));
3374 		amdgpu_ring_write(ring, 0);
3375 
3376 		amdgpu_ring_commit(ring);
3377 	}
3378 	return 0;
3379 }
3380 
3381 static void gfx_v11_0_cp_gfx_switch_pipe(struct amdgpu_device *adev,
3382 					 CP_PIPE_ID pipe)
3383 {
3384 	u32 tmp;
3385 
3386 	tmp = RREG32_SOC15(GC, 0, regGRBM_GFX_CNTL);
3387 	tmp = REG_SET_FIELD(tmp, GRBM_GFX_CNTL, PIPEID, pipe);
3388 
3389 	WREG32_SOC15(GC, 0, regGRBM_GFX_CNTL, tmp);
3390 }
3391 
3392 static void gfx_v11_0_cp_gfx_set_doorbell(struct amdgpu_device *adev,
3393 					  struct amdgpu_ring *ring)
3394 {
3395 	u32 tmp;
3396 
3397 	tmp = RREG32_SOC15(GC, 0, regCP_RB_DOORBELL_CONTROL);
3398 	if (ring->use_doorbell) {
3399 		tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL,
3400 				    DOORBELL_OFFSET, ring->doorbell_index);
3401 		tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL,
3402 				    DOORBELL_EN, 1);
3403 	} else {
3404 		tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL,
3405 				    DOORBELL_EN, 0);
3406 	}
3407 	WREG32_SOC15(GC, 0, regCP_RB_DOORBELL_CONTROL, tmp);
3408 
3409 	tmp = REG_SET_FIELD(0, CP_RB_DOORBELL_RANGE_LOWER,
3410 			    DOORBELL_RANGE_LOWER, ring->doorbell_index);
3411 	WREG32_SOC15(GC, 0, regCP_RB_DOORBELL_RANGE_LOWER, tmp);
3412 
3413 	WREG32_SOC15(GC, 0, regCP_RB_DOORBELL_RANGE_UPPER,
3414 		     CP_RB_DOORBELL_RANGE_UPPER__DOORBELL_RANGE_UPPER_MASK);
3415 }
3416 
3417 static int gfx_v11_0_cp_gfx_resume(struct amdgpu_device *adev)
3418 {
3419 	struct amdgpu_ring *ring;
3420 	u32 tmp;
3421 	u32 rb_bufsz;
3422 	u64 rb_addr, rptr_addr, wptr_gpu_addr;
3423 	u32 i;
3424 
3425 	/* Set the write pointer delay */
3426 	WREG32_SOC15(GC, 0, regCP_RB_WPTR_DELAY, 0);
3427 
3428 	/* set the RB to use vmid 0 */
3429 	WREG32_SOC15(GC, 0, regCP_RB_VMID, 0);
3430 
3431 	/* Init gfx ring 0 for pipe 0 */
3432 	mutex_lock(&adev->srbm_mutex);
3433 	gfx_v11_0_cp_gfx_switch_pipe(adev, PIPE_ID0);
3434 
3435 	/* Set ring buffer size */
3436 	ring = &adev->gfx.gfx_ring[0];
3437 	rb_bufsz = order_base_2(ring->ring_size / 8);
3438 	tmp = REG_SET_FIELD(0, CP_RB0_CNTL, RB_BUFSZ, rb_bufsz);
3439 	tmp = REG_SET_FIELD(tmp, CP_RB0_CNTL, RB_BLKSZ, rb_bufsz - 2);
3440 	WREG32_SOC15(GC, 0, regCP_RB0_CNTL, tmp);
3441 
3442 	/* Initialize the ring buffer's write pointers */
3443 	ring->wptr = 0;
3444 	WREG32_SOC15(GC, 0, regCP_RB0_WPTR, lower_32_bits(ring->wptr));
3445 	WREG32_SOC15(GC, 0, regCP_RB0_WPTR_HI, upper_32_bits(ring->wptr));
3446 
3447 	/* set the wb address wether it's enabled or not */
3448 	rptr_addr = ring->rptr_gpu_addr;
3449 	WREG32_SOC15(GC, 0, regCP_RB0_RPTR_ADDR, lower_32_bits(rptr_addr));
3450 	WREG32_SOC15(GC, 0, regCP_RB0_RPTR_ADDR_HI, upper_32_bits(rptr_addr) &
3451 		     CP_RB_RPTR_ADDR_HI__RB_RPTR_ADDR_HI_MASK);
3452 
3453 	wptr_gpu_addr = ring->wptr_gpu_addr;
3454 	WREG32_SOC15(GC, 0, regCP_RB_WPTR_POLL_ADDR_LO,
3455 		     lower_32_bits(wptr_gpu_addr));
3456 	WREG32_SOC15(GC, 0, regCP_RB_WPTR_POLL_ADDR_HI,
3457 		     upper_32_bits(wptr_gpu_addr));
3458 
3459 	mdelay(1);
3460 	WREG32_SOC15(GC, 0, regCP_RB0_CNTL, tmp);
3461 
3462 	rb_addr = ring->gpu_addr >> 8;
3463 	WREG32_SOC15(GC, 0, regCP_RB0_BASE, rb_addr);
3464 	WREG32_SOC15(GC, 0, regCP_RB0_BASE_HI, upper_32_bits(rb_addr));
3465 
3466 	WREG32_SOC15(GC, 0, regCP_RB_ACTIVE, 1);
3467 
3468 	gfx_v11_0_cp_gfx_set_doorbell(adev, ring);
3469 	mutex_unlock(&adev->srbm_mutex);
3470 
3471 	/* Init gfx ring 1 for pipe 1 */
3472 	if (adev->gfx.num_gfx_rings > 1) {
3473 		mutex_lock(&adev->srbm_mutex);
3474 		gfx_v11_0_cp_gfx_switch_pipe(adev, PIPE_ID1);
3475 		/* maximum supported gfx ring is 2 */
3476 		ring = &adev->gfx.gfx_ring[1];
3477 		rb_bufsz = order_base_2(ring->ring_size / 8);
3478 		tmp = REG_SET_FIELD(0, CP_RB1_CNTL, RB_BUFSZ, rb_bufsz);
3479 		tmp = REG_SET_FIELD(tmp, CP_RB1_CNTL, RB_BLKSZ, rb_bufsz - 2);
3480 		WREG32_SOC15(GC, 0, regCP_RB1_CNTL, tmp);
3481 		/* Initialize the ring buffer's write pointers */
3482 		ring->wptr = 0;
3483 		WREG32_SOC15(GC, 0, regCP_RB1_WPTR, lower_32_bits(ring->wptr));
3484 		WREG32_SOC15(GC, 0, regCP_RB1_WPTR_HI, upper_32_bits(ring->wptr));
3485 		/* Set the wb address wether it's enabled or not */
3486 		rptr_addr = ring->rptr_gpu_addr;
3487 		WREG32_SOC15(GC, 0, regCP_RB1_RPTR_ADDR, lower_32_bits(rptr_addr));
3488 		WREG32_SOC15(GC, 0, regCP_RB1_RPTR_ADDR_HI, upper_32_bits(rptr_addr) &
3489 			     CP_RB1_RPTR_ADDR_HI__RB_RPTR_ADDR_HI_MASK);
3490 		wptr_gpu_addr = ring->wptr_gpu_addr;
3491 		WREG32_SOC15(GC, 0, regCP_RB_WPTR_POLL_ADDR_LO,
3492 			     lower_32_bits(wptr_gpu_addr));
3493 		WREG32_SOC15(GC, 0, regCP_RB_WPTR_POLL_ADDR_HI,
3494 			     upper_32_bits(wptr_gpu_addr));
3495 
3496 		mdelay(1);
3497 		WREG32_SOC15(GC, 0, regCP_RB1_CNTL, tmp);
3498 
3499 		rb_addr = ring->gpu_addr >> 8;
3500 		WREG32_SOC15(GC, 0, regCP_RB1_BASE, rb_addr);
3501 		WREG32_SOC15(GC, 0, regCP_RB1_BASE_HI, upper_32_bits(rb_addr));
3502 		WREG32_SOC15(GC, 0, regCP_RB1_ACTIVE, 1);
3503 
3504 		gfx_v11_0_cp_gfx_set_doorbell(adev, ring);
3505 		mutex_unlock(&adev->srbm_mutex);
3506 	}
3507 	/* Switch to pipe 0 */
3508 	mutex_lock(&adev->srbm_mutex);
3509 	gfx_v11_0_cp_gfx_switch_pipe(adev, PIPE_ID0);
3510 	mutex_unlock(&adev->srbm_mutex);
3511 
3512 	/* start the ring */
3513 	gfx_v11_0_cp_gfx_start(adev);
3514 
3515 	for (i = 0; i < adev->gfx.num_gfx_rings; i++) {
3516 		ring = &adev->gfx.gfx_ring[i];
3517 		ring->sched.ready = true;
3518 	}
3519 
3520 	return 0;
3521 }
3522 
3523 static void gfx_v11_0_cp_compute_enable(struct amdgpu_device *adev, bool enable)
3524 {
3525 	u32 data;
3526 
3527 	if (adev->gfx.rs64_enable) {
3528 		data = RREG32_SOC15(GC, 0, regCP_MEC_RS64_CNTL);
3529 		data = REG_SET_FIELD(data, CP_MEC_RS64_CNTL, MEC_INVALIDATE_ICACHE,
3530 							 enable ? 0 : 1);
3531 		data = REG_SET_FIELD(data, CP_MEC_RS64_CNTL, MEC_PIPE0_RESET,
3532 							 enable ? 0 : 1);
3533 		data = REG_SET_FIELD(data, CP_MEC_RS64_CNTL, MEC_PIPE1_RESET,
3534 							 enable ? 0 : 1);
3535 		data = REG_SET_FIELD(data, CP_MEC_RS64_CNTL, MEC_PIPE2_RESET,
3536 							 enable ? 0 : 1);
3537 		data = REG_SET_FIELD(data, CP_MEC_RS64_CNTL, MEC_PIPE3_RESET,
3538 							 enable ? 0 : 1);
3539 		data = REG_SET_FIELD(data, CP_MEC_RS64_CNTL, MEC_PIPE0_ACTIVE,
3540 							 enable ? 1 : 0);
3541 		data = REG_SET_FIELD(data, CP_MEC_RS64_CNTL, MEC_PIPE1_ACTIVE,
3542 				                         enable ? 1 : 0);
3543 		data = REG_SET_FIELD(data, CP_MEC_RS64_CNTL, MEC_PIPE2_ACTIVE,
3544 							 enable ? 1 : 0);
3545 		data = REG_SET_FIELD(data, CP_MEC_RS64_CNTL, MEC_PIPE3_ACTIVE,
3546 							 enable ? 1 : 0);
3547 		data = REG_SET_FIELD(data, CP_MEC_RS64_CNTL, MEC_HALT,
3548 							 enable ? 0 : 1);
3549 		WREG32_SOC15(GC, 0, regCP_MEC_RS64_CNTL, data);
3550 	} else {
3551 		data = RREG32_SOC15(GC, 0, regCP_MEC_CNTL);
3552 
3553 		if (enable) {
3554 			data = REG_SET_FIELD(data, CP_MEC_CNTL, MEC_ME1_HALT, 0);
3555 			if (!adev->enable_mes_kiq)
3556 				data = REG_SET_FIELD(data, CP_MEC_CNTL,
3557 						     MEC_ME2_HALT, 0);
3558 		} else {
3559 			data = REG_SET_FIELD(data, CP_MEC_CNTL, MEC_ME1_HALT, 1);
3560 			data = REG_SET_FIELD(data, CP_MEC_CNTL, MEC_ME2_HALT, 1);
3561 		}
3562 		WREG32_SOC15(GC, 0, regCP_MEC_CNTL, data);
3563 	}
3564 
3565 	adev->gfx.kiq.ring.sched.ready = enable;
3566 
3567 	udelay(50);
3568 }
3569 
3570 static int gfx_v11_0_cp_compute_load_microcode(struct amdgpu_device *adev)
3571 {
3572 	const struct gfx_firmware_header_v1_0 *mec_hdr;
3573 	const __le32 *fw_data;
3574 	unsigned i, fw_size;
3575 	u32 *fw = NULL;
3576 	int r;
3577 
3578 	if (!adev->gfx.mec_fw)
3579 		return -EINVAL;
3580 
3581 	gfx_v11_0_cp_compute_enable(adev, false);
3582 
3583 	mec_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data;
3584 	amdgpu_ucode_print_gfx_hdr(&mec_hdr->header);
3585 
3586 	fw_data = (const __le32 *)
3587 		(adev->gfx.mec_fw->data +
3588 		 le32_to_cpu(mec_hdr->header.ucode_array_offset_bytes));
3589 	fw_size = le32_to_cpu(mec_hdr->header.ucode_size_bytes);
3590 
3591 	r = amdgpu_bo_create_reserved(adev, mec_hdr->header.ucode_size_bytes,
3592 					  PAGE_SIZE, AMDGPU_GEM_DOMAIN_GTT,
3593 					  &adev->gfx.mec.mec_fw_obj,
3594 					  &adev->gfx.mec.mec_fw_gpu_addr,
3595 					  (void **)&fw);
3596 	if (r) {
3597 		dev_err(adev->dev, "(%d) failed to create mec fw bo\n", r);
3598 		gfx_v11_0_mec_fini(adev);
3599 		return r;
3600 	}
3601 
3602 	memcpy(fw, fw_data, fw_size);
3603 
3604 	amdgpu_bo_kunmap(adev->gfx.mec.mec_fw_obj);
3605 	amdgpu_bo_unreserve(adev->gfx.mec.mec_fw_obj);
3606 
3607 	gfx_v11_0_config_mec_cache(adev, adev->gfx.mec.mec_fw_gpu_addr);
3608 
3609 	/* MEC1 */
3610 	WREG32_SOC15(GC, 0, regCP_MEC_ME1_UCODE_ADDR, 0);
3611 
3612 	for (i = 0; i < mec_hdr->jt_size; i++)
3613 		WREG32_SOC15(GC, 0, regCP_MEC_ME1_UCODE_DATA,
3614 			     le32_to_cpup(fw_data + mec_hdr->jt_offset + i));
3615 
3616 	WREG32_SOC15(GC, 0, regCP_MEC_ME1_UCODE_ADDR, adev->gfx.mec_fw_version);
3617 
3618 	return 0;
3619 }
3620 
3621 static int gfx_v11_0_cp_compute_load_microcode_rs64(struct amdgpu_device *adev)
3622 {
3623 	const struct gfx_firmware_header_v2_0 *mec_hdr;
3624 	const __le32 *fw_ucode, *fw_data;
3625 	u32 tmp, fw_ucode_size, fw_data_size;
3626 	u32 i, usec_timeout = 50000; /* Wait for 50 ms */
3627 	u32 *fw_ucode_ptr, *fw_data_ptr;
3628 	int r;
3629 
3630 	if (!adev->gfx.mec_fw)
3631 		return -EINVAL;
3632 
3633 	gfx_v11_0_cp_compute_enable(adev, false);
3634 
3635 	mec_hdr = (const struct gfx_firmware_header_v2_0 *)adev->gfx.mec_fw->data;
3636 	amdgpu_ucode_print_gfx_hdr(&mec_hdr->header);
3637 
3638 	fw_ucode = (const __le32 *) (adev->gfx.mec_fw->data +
3639 				le32_to_cpu(mec_hdr->ucode_offset_bytes));
3640 	fw_ucode_size = le32_to_cpu(mec_hdr->ucode_size_bytes);
3641 
3642 	fw_data = (const __le32 *) (adev->gfx.mec_fw->data +
3643 				le32_to_cpu(mec_hdr->data_offset_bytes));
3644 	fw_data_size = le32_to_cpu(mec_hdr->data_size_bytes);
3645 
3646 	r = amdgpu_bo_create_reserved(adev, fw_ucode_size,
3647 				      64 * 1024, AMDGPU_GEM_DOMAIN_VRAM,
3648 				      &adev->gfx.mec.mec_fw_obj,
3649 				      &adev->gfx.mec.mec_fw_gpu_addr,
3650 				      (void **)&fw_ucode_ptr);
3651 	if (r) {
3652 		dev_err(adev->dev, "(%d) failed to create mec fw ucode bo\n", r);
3653 		gfx_v11_0_mec_fini(adev);
3654 		return r;
3655 	}
3656 
3657 	r = amdgpu_bo_create_reserved(adev, fw_data_size,
3658 				      64 * 1024, AMDGPU_GEM_DOMAIN_VRAM,
3659 				      &adev->gfx.mec.mec_fw_data_obj,
3660 				      &adev->gfx.mec.mec_fw_data_gpu_addr,
3661 				      (void **)&fw_data_ptr);
3662 	if (r) {
3663 		dev_err(adev->dev, "(%d) failed to create mec fw ucode bo\n", r);
3664 		gfx_v11_0_mec_fini(adev);
3665 		return r;
3666 	}
3667 
3668 	memcpy(fw_ucode_ptr, fw_ucode, fw_ucode_size);
3669 	memcpy(fw_data_ptr, fw_data, fw_data_size);
3670 
3671 	amdgpu_bo_kunmap(adev->gfx.mec.mec_fw_obj);
3672 	amdgpu_bo_kunmap(adev->gfx.mec.mec_fw_data_obj);
3673 	amdgpu_bo_unreserve(adev->gfx.mec.mec_fw_obj);
3674 	amdgpu_bo_unreserve(adev->gfx.mec.mec_fw_data_obj);
3675 
3676 	tmp = RREG32_SOC15(GC, 0, regCP_CPC_IC_BASE_CNTL);
3677 	tmp = REG_SET_FIELD(tmp, CP_CPC_IC_BASE_CNTL, VMID, 0);
3678 	tmp = REG_SET_FIELD(tmp, CP_CPC_IC_BASE_CNTL, EXE_DISABLE, 0);
3679 	tmp = REG_SET_FIELD(tmp, CP_CPC_IC_BASE_CNTL, CACHE_POLICY, 0);
3680 	WREG32_SOC15(GC, 0, regCP_CPC_IC_BASE_CNTL, tmp);
3681 
3682 	tmp = RREG32_SOC15(GC, 0, regCP_MEC_DC_BASE_CNTL);
3683 	tmp = REG_SET_FIELD(tmp, CP_MEC_DC_BASE_CNTL, VMID, 0);
3684 	tmp = REG_SET_FIELD(tmp, CP_MEC_DC_BASE_CNTL, CACHE_POLICY, 0);
3685 	WREG32_SOC15(GC, 0, regCP_MEC_DC_BASE_CNTL, tmp);
3686 
3687 	mutex_lock(&adev->srbm_mutex);
3688 	for (i = 0; i < adev->gfx.mec.num_pipe_per_mec; i++) {
3689 		soc21_grbm_select(adev, 1, i, 0, 0);
3690 
3691 		WREG32_SOC15(GC, 0, regCP_MEC_MDBASE_LO, adev->gfx.mec.mec_fw_data_gpu_addr);
3692 		WREG32_SOC15(GC, 0, regCP_MEC_MDBASE_HI,
3693 		     upper_32_bits(adev->gfx.mec.mec_fw_data_gpu_addr));
3694 
3695 		WREG32_SOC15(GC, 0, regCP_MEC_RS64_PRGRM_CNTR_START,
3696 					mec_hdr->ucode_start_addr_lo >> 2 |
3697 					mec_hdr->ucode_start_addr_hi << 30);
3698 		WREG32_SOC15(GC, 0, regCP_MEC_RS64_PRGRM_CNTR_START_HI,
3699 					mec_hdr->ucode_start_addr_hi >> 2);
3700 
3701 		WREG32_SOC15(GC, 0, regCP_CPC_IC_BASE_LO, adev->gfx.mec.mec_fw_gpu_addr);
3702 		WREG32_SOC15(GC, 0, regCP_CPC_IC_BASE_HI,
3703 		     upper_32_bits(adev->gfx.mec.mec_fw_gpu_addr));
3704 	}
3705 	mutex_unlock(&adev->srbm_mutex);
3706 	soc21_grbm_select(adev, 0, 0, 0, 0);
3707 
3708 	/* Trigger an invalidation of the L1 instruction caches */
3709 	tmp = RREG32_SOC15(GC, 0, regCP_MEC_DC_OP_CNTL);
3710 	tmp = REG_SET_FIELD(tmp, CP_MEC_DC_OP_CNTL, INVALIDATE_DCACHE, 1);
3711 	WREG32_SOC15(GC, 0, regCP_MEC_DC_OP_CNTL, tmp);
3712 
3713 	/* Wait for invalidation complete */
3714 	for (i = 0; i < usec_timeout; i++) {
3715 		tmp = RREG32_SOC15(GC, 0, regCP_MEC_DC_OP_CNTL);
3716 		if (1 == REG_GET_FIELD(tmp, CP_MEC_DC_OP_CNTL,
3717 				       INVALIDATE_DCACHE_COMPLETE))
3718 			break;
3719 		udelay(1);
3720 	}
3721 
3722 	if (i >= usec_timeout) {
3723 		dev_err(adev->dev, "failed to invalidate instruction cache\n");
3724 		return -EINVAL;
3725 	}
3726 
3727 	/* Trigger an invalidation of the L1 instruction caches */
3728 	tmp = RREG32_SOC15(GC, 0, regCP_CPC_IC_OP_CNTL);
3729 	tmp = REG_SET_FIELD(tmp, CP_CPC_IC_OP_CNTL, INVALIDATE_CACHE, 1);
3730 	WREG32_SOC15(GC, 0, regCP_CPC_IC_OP_CNTL, tmp);
3731 
3732 	/* Wait for invalidation complete */
3733 	for (i = 0; i < usec_timeout; i++) {
3734 		tmp = RREG32_SOC15(GC, 0, regCP_CPC_IC_OP_CNTL);
3735 		if (1 == REG_GET_FIELD(tmp, CP_CPC_IC_OP_CNTL,
3736 				       INVALIDATE_CACHE_COMPLETE))
3737 			break;
3738 		udelay(1);
3739 	}
3740 
3741 	if (i >= usec_timeout) {
3742 		dev_err(adev->dev, "failed to invalidate instruction cache\n");
3743 		return -EINVAL;
3744 	}
3745 
3746 	return 0;
3747 }
3748 
3749 static void gfx_v11_0_kiq_setting(struct amdgpu_ring *ring)
3750 {
3751 	uint32_t tmp;
3752 	struct amdgpu_device *adev = ring->adev;
3753 
3754 	/* tell RLC which is KIQ queue */
3755 	tmp = RREG32_SOC15(GC, 0, regRLC_CP_SCHEDULERS);
3756 	tmp &= 0xffffff00;
3757 	tmp |= (ring->me << 5) | (ring->pipe << 3) | (ring->queue);
3758 	WREG32_SOC15(GC, 0, regRLC_CP_SCHEDULERS, tmp);
3759 	tmp |= 0x80;
3760 	WREG32_SOC15(GC, 0, regRLC_CP_SCHEDULERS, tmp);
3761 }
3762 
3763 static void gfx_v11_0_cp_set_doorbell_range(struct amdgpu_device *adev)
3764 {
3765 	/* set graphics engine doorbell range */
3766 	WREG32_SOC15(GC, 0, regCP_RB_DOORBELL_RANGE_LOWER,
3767 		     (adev->doorbell_index.gfx_ring0 * 2) << 2);
3768 	WREG32_SOC15(GC, 0, regCP_RB_DOORBELL_RANGE_UPPER,
3769 		     (adev->doorbell_index.gfx_userqueue_end * 2) << 2);
3770 
3771 	/* set compute engine doorbell range */
3772 	WREG32_SOC15(GC, 0, regCP_MEC_DOORBELL_RANGE_LOWER,
3773 		     (adev->doorbell_index.kiq * 2) << 2);
3774 	WREG32_SOC15(GC, 0, regCP_MEC_DOORBELL_RANGE_UPPER,
3775 		     (adev->doorbell_index.userqueue_end * 2) << 2);
3776 }
3777 
3778 static int gfx_v11_0_gfx_mqd_init(struct amdgpu_device *adev, void *m,
3779 				  struct amdgpu_mqd_prop *prop)
3780 {
3781 	struct v11_gfx_mqd *mqd = m;
3782 	uint64_t hqd_gpu_addr, wb_gpu_addr;
3783 	uint32_t tmp;
3784 	uint32_t rb_bufsz;
3785 
3786 	/* set up gfx hqd wptr */
3787 	mqd->cp_gfx_hqd_wptr = 0;
3788 	mqd->cp_gfx_hqd_wptr_hi = 0;
3789 
3790 	/* set the pointer to the MQD */
3791 	mqd->cp_mqd_base_addr = prop->mqd_gpu_addr & 0xfffffffc;
3792 	mqd->cp_mqd_base_addr_hi = upper_32_bits(prop->mqd_gpu_addr);
3793 
3794 	/* set up mqd control */
3795 	tmp = RREG32_SOC15(GC, 0, regCP_GFX_MQD_CONTROL);
3796 	tmp = REG_SET_FIELD(tmp, CP_GFX_MQD_CONTROL, VMID, 0);
3797 	tmp = REG_SET_FIELD(tmp, CP_GFX_MQD_CONTROL, PRIV_STATE, 1);
3798 	tmp = REG_SET_FIELD(tmp, CP_GFX_MQD_CONTROL, CACHE_POLICY, 0);
3799 	mqd->cp_gfx_mqd_control = tmp;
3800 
3801 	/* set up gfx_hqd_vimd with 0x0 to indicate the ring buffer's vmid */
3802 	tmp = RREG32_SOC15(GC, 0, regCP_GFX_HQD_VMID);
3803 	tmp = REG_SET_FIELD(tmp, CP_GFX_HQD_VMID, VMID, 0);
3804 	mqd->cp_gfx_hqd_vmid = 0;
3805 
3806 	/* set up default queue priority level
3807 	 * 0x0 = low priority, 0x1 = high priority */
3808 	tmp = RREG32_SOC15(GC, 0, regCP_GFX_HQD_QUEUE_PRIORITY);
3809 	tmp = REG_SET_FIELD(tmp, CP_GFX_HQD_QUEUE_PRIORITY, PRIORITY_LEVEL, 0);
3810 	mqd->cp_gfx_hqd_queue_priority = tmp;
3811 
3812 	/* set up time quantum */
3813 	tmp = RREG32_SOC15(GC, 0, regCP_GFX_HQD_QUANTUM);
3814 	tmp = REG_SET_FIELD(tmp, CP_GFX_HQD_QUANTUM, QUANTUM_EN, 1);
3815 	mqd->cp_gfx_hqd_quantum = tmp;
3816 
3817 	/* set up gfx hqd base. this is similar as CP_RB_BASE */
3818 	hqd_gpu_addr = prop->hqd_base_gpu_addr >> 8;
3819 	mqd->cp_gfx_hqd_base = hqd_gpu_addr;
3820 	mqd->cp_gfx_hqd_base_hi = upper_32_bits(hqd_gpu_addr);
3821 
3822 	/* set up hqd_rptr_addr/_hi, similar as CP_RB_RPTR */
3823 	wb_gpu_addr = prop->rptr_gpu_addr;
3824 	mqd->cp_gfx_hqd_rptr_addr = wb_gpu_addr & 0xfffffffc;
3825 	mqd->cp_gfx_hqd_rptr_addr_hi =
3826 		upper_32_bits(wb_gpu_addr) & 0xffff;
3827 
3828 	/* set up rb_wptr_poll addr */
3829 	wb_gpu_addr = prop->wptr_gpu_addr;
3830 	mqd->cp_rb_wptr_poll_addr_lo = wb_gpu_addr & 0xfffffffc;
3831 	mqd->cp_rb_wptr_poll_addr_hi = upper_32_bits(wb_gpu_addr) & 0xffff;
3832 
3833 	/* set up the gfx_hqd_control, similar as CP_RB0_CNTL */
3834 	rb_bufsz = order_base_2(prop->queue_size / 4) - 1;
3835 	tmp = RREG32_SOC15(GC, 0, regCP_GFX_HQD_CNTL);
3836 	tmp = REG_SET_FIELD(tmp, CP_GFX_HQD_CNTL, RB_BUFSZ, rb_bufsz);
3837 	tmp = REG_SET_FIELD(tmp, CP_GFX_HQD_CNTL, RB_BLKSZ, rb_bufsz - 2);
3838 #ifdef __BIG_ENDIAN
3839 	tmp = REG_SET_FIELD(tmp, CP_GFX_HQD_CNTL, BUF_SWAP, 1);
3840 #endif
3841 	mqd->cp_gfx_hqd_cntl = tmp;
3842 
3843 	/* set up cp_doorbell_control */
3844 	tmp = RREG32_SOC15(GC, 0, regCP_RB_DOORBELL_CONTROL);
3845 	if (prop->use_doorbell) {
3846 		tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL,
3847 				    DOORBELL_OFFSET, prop->doorbell_index);
3848 		tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL,
3849 				    DOORBELL_EN, 1);
3850 	} else
3851 		tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL,
3852 				    DOORBELL_EN, 0);
3853 	mqd->cp_rb_doorbell_control = tmp;
3854 
3855 	/* reset read and write pointers, similar to CP_RB0_WPTR/_RPTR */
3856 	mqd->cp_gfx_hqd_rptr = RREG32_SOC15(GC, 0, regCP_GFX_HQD_RPTR);
3857 
3858 	/* active the queue */
3859 	mqd->cp_gfx_hqd_active = 1;
3860 
3861 	return 0;
3862 }
3863 
3864 #ifdef BRING_UP_DEBUG
3865 static int gfx_v11_0_gfx_queue_init_register(struct amdgpu_ring *ring)
3866 {
3867 	struct amdgpu_device *adev = ring->adev;
3868 	struct v11_gfx_mqd *mqd = ring->mqd_ptr;
3869 
3870 	/* set mmCP_GFX_HQD_WPTR/_HI to 0 */
3871 	WREG32_SOC15(GC, 0, regCP_GFX_HQD_WPTR, mqd->cp_gfx_hqd_wptr);
3872 	WREG32_SOC15(GC, 0, regCP_GFX_HQD_WPTR_HI, mqd->cp_gfx_hqd_wptr_hi);
3873 
3874 	/* set GFX_MQD_BASE */
3875 	WREG32_SOC15(GC, 0, regCP_MQD_BASE_ADDR, mqd->cp_mqd_base_addr);
3876 	WREG32_SOC15(GC, 0, regCP_MQD_BASE_ADDR_HI, mqd->cp_mqd_base_addr_hi);
3877 
3878 	/* set GFX_MQD_CONTROL */
3879 	WREG32_SOC15(GC, 0, regCP_GFX_MQD_CONTROL, mqd->cp_gfx_mqd_control);
3880 
3881 	/* set GFX_HQD_VMID to 0 */
3882 	WREG32_SOC15(GC, 0, regCP_GFX_HQD_VMID, mqd->cp_gfx_hqd_vmid);
3883 
3884 	WREG32_SOC15(GC, 0, regCP_GFX_HQD_QUEUE_PRIORITY,
3885 			mqd->cp_gfx_hqd_queue_priority);
3886 	WREG32_SOC15(GC, 0, regCP_GFX_HQD_QUANTUM, mqd->cp_gfx_hqd_quantum);
3887 
3888 	/* set GFX_HQD_BASE, similar as CP_RB_BASE */
3889 	WREG32_SOC15(GC, 0, regCP_GFX_HQD_BASE, mqd->cp_gfx_hqd_base);
3890 	WREG32_SOC15(GC, 0, regCP_GFX_HQD_BASE_HI, mqd->cp_gfx_hqd_base_hi);
3891 
3892 	/* set GFX_HQD_RPTR_ADDR, similar as CP_RB_RPTR */
3893 	WREG32_SOC15(GC, 0, regCP_GFX_HQD_RPTR_ADDR, mqd->cp_gfx_hqd_rptr_addr);
3894 	WREG32_SOC15(GC, 0, regCP_GFX_HQD_RPTR_ADDR_HI, mqd->cp_gfx_hqd_rptr_addr_hi);
3895 
3896 	/* set GFX_HQD_CNTL, similar as CP_RB_CNTL */
3897 	WREG32_SOC15(GC, 0, regCP_GFX_HQD_CNTL, mqd->cp_gfx_hqd_cntl);
3898 
3899 	/* set RB_WPTR_POLL_ADDR */
3900 	WREG32_SOC15(GC, 0, regCP_RB_WPTR_POLL_ADDR_LO, mqd->cp_rb_wptr_poll_addr_lo);
3901 	WREG32_SOC15(GC, 0, regCP_RB_WPTR_POLL_ADDR_HI, mqd->cp_rb_wptr_poll_addr_hi);
3902 
3903 	/* set RB_DOORBELL_CONTROL */
3904 	WREG32_SOC15(GC, 0, regCP_RB_DOORBELL_CONTROL, mqd->cp_rb_doorbell_control);
3905 
3906 	/* active the queue */
3907 	WREG32_SOC15(GC, 0, regCP_GFX_HQD_ACTIVE, mqd->cp_gfx_hqd_active);
3908 
3909 	return 0;
3910 }
3911 #endif
3912 
3913 static int gfx_v11_0_gfx_init_queue(struct amdgpu_ring *ring)
3914 {
3915 	struct amdgpu_device *adev = ring->adev;
3916 	struct v11_gfx_mqd *mqd = ring->mqd_ptr;
3917 	int mqd_idx = ring - &adev->gfx.gfx_ring[0];
3918 
3919 	if (!amdgpu_in_reset(adev) && !adev->in_suspend) {
3920 		memset((void *)mqd, 0, sizeof(*mqd));
3921 		mutex_lock(&adev->srbm_mutex);
3922 		soc21_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
3923 		amdgpu_ring_init_mqd(ring);
3924 #ifdef BRING_UP_DEBUG
3925 		gfx_v11_0_gfx_queue_init_register(ring);
3926 #endif
3927 		soc21_grbm_select(adev, 0, 0, 0, 0);
3928 		mutex_unlock(&adev->srbm_mutex);
3929 		if (adev->gfx.me.mqd_backup[mqd_idx])
3930 			memcpy(adev->gfx.me.mqd_backup[mqd_idx], mqd, sizeof(*mqd));
3931 	} else if (amdgpu_in_reset(adev)) {
3932 		/* reset mqd with the backup copy */
3933 		if (adev->gfx.me.mqd_backup[mqd_idx])
3934 			memcpy(mqd, adev->gfx.me.mqd_backup[mqd_idx], sizeof(*mqd));
3935 		/* reset the ring */
3936 		ring->wptr = 0;
3937 		*ring->wptr_cpu_addr = 0;
3938 		amdgpu_ring_clear_ring(ring);
3939 #ifdef BRING_UP_DEBUG
3940 		mutex_lock(&adev->srbm_mutex);
3941 		soc21_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
3942 		gfx_v11_0_gfx_queue_init_register(ring);
3943 		soc21_grbm_select(adev, 0, 0, 0, 0);
3944 		mutex_unlock(&adev->srbm_mutex);
3945 #endif
3946 	} else {
3947 		amdgpu_ring_clear_ring(ring);
3948 	}
3949 
3950 	return 0;
3951 }
3952 
3953 #ifndef BRING_UP_DEBUG
3954 static int gfx_v11_0_kiq_enable_kgq(struct amdgpu_device *adev)
3955 {
3956 	struct amdgpu_kiq *kiq = &adev->gfx.kiq;
3957 	struct amdgpu_ring *kiq_ring = &adev->gfx.kiq.ring;
3958 	int r, i;
3959 
3960 	if (!kiq->pmf || !kiq->pmf->kiq_map_queues)
3961 		return -EINVAL;
3962 
3963 	r = amdgpu_ring_alloc(kiq_ring, kiq->pmf->map_queues_size *
3964 					adev->gfx.num_gfx_rings);
3965 	if (r) {
3966 		DRM_ERROR("Failed to lock KIQ (%d).\n", r);
3967 		return r;
3968 	}
3969 
3970 	for (i = 0; i < adev->gfx.num_gfx_rings; i++)
3971 		kiq->pmf->kiq_map_queues(kiq_ring, &adev->gfx.gfx_ring[i]);
3972 
3973 	return amdgpu_ring_test_helper(kiq_ring);
3974 }
3975 #endif
3976 
3977 static int gfx_v11_0_cp_async_gfx_ring_resume(struct amdgpu_device *adev)
3978 {
3979 	int r, i;
3980 	struct amdgpu_ring *ring;
3981 
3982 	for (i = 0; i < adev->gfx.num_gfx_rings; i++) {
3983 		ring = &adev->gfx.gfx_ring[i];
3984 
3985 		r = amdgpu_bo_reserve(ring->mqd_obj, false);
3986 		if (unlikely(r != 0))
3987 			goto done;
3988 
3989 		r = amdgpu_bo_kmap(ring->mqd_obj, (void **)&ring->mqd_ptr);
3990 		if (!r) {
3991 			r = gfx_v11_0_gfx_init_queue(ring);
3992 			amdgpu_bo_kunmap(ring->mqd_obj);
3993 			ring->mqd_ptr = NULL;
3994 		}
3995 		amdgpu_bo_unreserve(ring->mqd_obj);
3996 		if (r)
3997 			goto done;
3998 	}
3999 #ifndef BRING_UP_DEBUG
4000 	r = gfx_v11_0_kiq_enable_kgq(adev);
4001 	if (r)
4002 		goto done;
4003 #endif
4004 	r = gfx_v11_0_cp_gfx_start(adev);
4005 	if (r)
4006 		goto done;
4007 
4008 	for (i = 0; i < adev->gfx.num_gfx_rings; i++) {
4009 		ring = &adev->gfx.gfx_ring[i];
4010 		ring->sched.ready = true;
4011 	}
4012 done:
4013 	return r;
4014 }
4015 
4016 static int gfx_v11_0_compute_mqd_init(struct amdgpu_device *adev, void *m,
4017 				      struct amdgpu_mqd_prop *prop)
4018 {
4019 	struct v11_compute_mqd *mqd = m;
4020 	uint64_t hqd_gpu_addr, wb_gpu_addr, eop_base_addr;
4021 	uint32_t tmp;
4022 
4023 	mqd->header = 0xC0310800;
4024 	mqd->compute_pipelinestat_enable = 0x00000001;
4025 	mqd->compute_static_thread_mgmt_se0 = 0xffffffff;
4026 	mqd->compute_static_thread_mgmt_se1 = 0xffffffff;
4027 	mqd->compute_static_thread_mgmt_se2 = 0xffffffff;
4028 	mqd->compute_static_thread_mgmt_se3 = 0xffffffff;
4029 	mqd->compute_misc_reserved = 0x00000007;
4030 
4031 	eop_base_addr = prop->eop_gpu_addr >> 8;
4032 	mqd->cp_hqd_eop_base_addr_lo = eop_base_addr;
4033 	mqd->cp_hqd_eop_base_addr_hi = upper_32_bits(eop_base_addr);
4034 
4035 	/* set the EOP size, register value is 2^(EOP_SIZE+1) dwords */
4036 	tmp = RREG32_SOC15(GC, 0, regCP_HQD_EOP_CONTROL);
4037 	tmp = REG_SET_FIELD(tmp, CP_HQD_EOP_CONTROL, EOP_SIZE,
4038 			(order_base_2(GFX11_MEC_HPD_SIZE / 4) - 1));
4039 
4040 	mqd->cp_hqd_eop_control = tmp;
4041 
4042 	/* enable doorbell? */
4043 	tmp = RREG32_SOC15(GC, 0, regCP_HQD_PQ_DOORBELL_CONTROL);
4044 
4045 	if (prop->use_doorbell) {
4046 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
4047 				    DOORBELL_OFFSET, prop->doorbell_index);
4048 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
4049 				    DOORBELL_EN, 1);
4050 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
4051 				    DOORBELL_SOURCE, 0);
4052 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
4053 				    DOORBELL_HIT, 0);
4054 	} else {
4055 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
4056 				    DOORBELL_EN, 0);
4057 	}
4058 
4059 	mqd->cp_hqd_pq_doorbell_control = tmp;
4060 
4061 	/* disable the queue if it's active */
4062 	mqd->cp_hqd_dequeue_request = 0;
4063 	mqd->cp_hqd_pq_rptr = 0;
4064 	mqd->cp_hqd_pq_wptr_lo = 0;
4065 	mqd->cp_hqd_pq_wptr_hi = 0;
4066 
4067 	/* set the pointer to the MQD */
4068 	mqd->cp_mqd_base_addr_lo = prop->mqd_gpu_addr & 0xfffffffc;
4069 	mqd->cp_mqd_base_addr_hi = upper_32_bits(prop->mqd_gpu_addr);
4070 
4071 	/* set MQD vmid to 0 */
4072 	tmp = RREG32_SOC15(GC, 0, regCP_MQD_CONTROL);
4073 	tmp = REG_SET_FIELD(tmp, CP_MQD_CONTROL, VMID, 0);
4074 	mqd->cp_mqd_control = tmp;
4075 
4076 	/* set the pointer to the HQD, this is similar CP_RB0_BASE/_HI */
4077 	hqd_gpu_addr = prop->hqd_base_gpu_addr >> 8;
4078 	mqd->cp_hqd_pq_base_lo = hqd_gpu_addr;
4079 	mqd->cp_hqd_pq_base_hi = upper_32_bits(hqd_gpu_addr);
4080 
4081 	/* set up the HQD, this is similar to CP_RB0_CNTL */
4082 	tmp = RREG32_SOC15(GC, 0, regCP_HQD_PQ_CONTROL);
4083 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, QUEUE_SIZE,
4084 			    (order_base_2(prop->queue_size / 4) - 1));
4085 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, RPTR_BLOCK_SIZE,
4086 			    (order_base_2(AMDGPU_GPU_PAGE_SIZE / 4) - 1));
4087 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, UNORD_DISPATCH, 0);
4088 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, TUNNEL_DISPATCH, 0);
4089 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, PRIV_STATE, 1);
4090 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, KMD_QUEUE, 1);
4091 	mqd->cp_hqd_pq_control = tmp;
4092 
4093 	/* set the wb address whether it's enabled or not */
4094 	wb_gpu_addr = prop->rptr_gpu_addr;
4095 	mqd->cp_hqd_pq_rptr_report_addr_lo = wb_gpu_addr & 0xfffffffc;
4096 	mqd->cp_hqd_pq_rptr_report_addr_hi =
4097 		upper_32_bits(wb_gpu_addr) & 0xffff;
4098 
4099 	/* only used if CP_PQ_WPTR_POLL_CNTL.CP_PQ_WPTR_POLL_CNTL__EN_MASK=1 */
4100 	wb_gpu_addr = prop->wptr_gpu_addr;
4101 	mqd->cp_hqd_pq_wptr_poll_addr_lo = wb_gpu_addr & 0xfffffffc;
4102 	mqd->cp_hqd_pq_wptr_poll_addr_hi = upper_32_bits(wb_gpu_addr) & 0xffff;
4103 
4104 	tmp = 0;
4105 	/* enable the doorbell if requested */
4106 	if (prop->use_doorbell) {
4107 		tmp = RREG32_SOC15(GC, 0, regCP_HQD_PQ_DOORBELL_CONTROL);
4108 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
4109 				DOORBELL_OFFSET, prop->doorbell_index);
4110 
4111 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
4112 				    DOORBELL_EN, 1);
4113 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
4114 				    DOORBELL_SOURCE, 0);
4115 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
4116 				    DOORBELL_HIT, 0);
4117 	}
4118 
4119 	mqd->cp_hqd_pq_doorbell_control = tmp;
4120 
4121 	/* reset read and write pointers, similar to CP_RB0_WPTR/_RPTR */
4122 	mqd->cp_hqd_pq_rptr = RREG32_SOC15(GC, 0, regCP_HQD_PQ_RPTR);
4123 
4124 	/* set the vmid for the queue */
4125 	mqd->cp_hqd_vmid = 0;
4126 
4127 	tmp = RREG32_SOC15(GC, 0, regCP_HQD_PERSISTENT_STATE);
4128 	tmp = REG_SET_FIELD(tmp, CP_HQD_PERSISTENT_STATE, PRELOAD_SIZE, 0x55);
4129 	mqd->cp_hqd_persistent_state = tmp;
4130 
4131 	/* set MIN_IB_AVAIL_SIZE */
4132 	tmp = RREG32_SOC15(GC, 0, regCP_HQD_IB_CONTROL);
4133 	tmp = REG_SET_FIELD(tmp, CP_HQD_IB_CONTROL, MIN_IB_AVAIL_SIZE, 3);
4134 	mqd->cp_hqd_ib_control = tmp;
4135 
4136 	/* set static priority for a compute queue/ring */
4137 	mqd->cp_hqd_pipe_priority = prop->hqd_pipe_priority;
4138 	mqd->cp_hqd_queue_priority = prop->hqd_queue_priority;
4139 
4140 	mqd->cp_hqd_active = prop->hqd_active;
4141 
4142 	return 0;
4143 }
4144 
4145 static int gfx_v11_0_kiq_init_register(struct amdgpu_ring *ring)
4146 {
4147 	struct amdgpu_device *adev = ring->adev;
4148 	struct v11_compute_mqd *mqd = ring->mqd_ptr;
4149 	int j;
4150 
4151 	/* inactivate the queue */
4152 	if (amdgpu_sriov_vf(adev))
4153 		WREG32_SOC15(GC, 0, regCP_HQD_ACTIVE, 0);
4154 
4155 	/* disable wptr polling */
4156 	WREG32_FIELD15_PREREG(GC, 0, CP_PQ_WPTR_POLL_CNTL, EN, 0);
4157 
4158 	/* write the EOP addr */
4159 	WREG32_SOC15(GC, 0, regCP_HQD_EOP_BASE_ADDR,
4160 	       mqd->cp_hqd_eop_base_addr_lo);
4161 	WREG32_SOC15(GC, 0, regCP_HQD_EOP_BASE_ADDR_HI,
4162 	       mqd->cp_hqd_eop_base_addr_hi);
4163 
4164 	/* set the EOP size, register value is 2^(EOP_SIZE+1) dwords */
4165 	WREG32_SOC15(GC, 0, regCP_HQD_EOP_CONTROL,
4166 	       mqd->cp_hqd_eop_control);
4167 
4168 	/* enable doorbell? */
4169 	WREG32_SOC15(GC, 0, regCP_HQD_PQ_DOORBELL_CONTROL,
4170 	       mqd->cp_hqd_pq_doorbell_control);
4171 
4172 	/* disable the queue if it's active */
4173 	if (RREG32_SOC15(GC, 0, regCP_HQD_ACTIVE) & 1) {
4174 		WREG32_SOC15(GC, 0, regCP_HQD_DEQUEUE_REQUEST, 1);
4175 		for (j = 0; j < adev->usec_timeout; j++) {
4176 			if (!(RREG32_SOC15(GC, 0, regCP_HQD_ACTIVE) & 1))
4177 				break;
4178 			udelay(1);
4179 		}
4180 		WREG32_SOC15(GC, 0, regCP_HQD_DEQUEUE_REQUEST,
4181 		       mqd->cp_hqd_dequeue_request);
4182 		WREG32_SOC15(GC, 0, regCP_HQD_PQ_RPTR,
4183 		       mqd->cp_hqd_pq_rptr);
4184 		WREG32_SOC15(GC, 0, regCP_HQD_PQ_WPTR_LO,
4185 		       mqd->cp_hqd_pq_wptr_lo);
4186 		WREG32_SOC15(GC, 0, regCP_HQD_PQ_WPTR_HI,
4187 		       mqd->cp_hqd_pq_wptr_hi);
4188 	}
4189 
4190 	/* set the pointer to the MQD */
4191 	WREG32_SOC15(GC, 0, regCP_MQD_BASE_ADDR,
4192 	       mqd->cp_mqd_base_addr_lo);
4193 	WREG32_SOC15(GC, 0, regCP_MQD_BASE_ADDR_HI,
4194 	       mqd->cp_mqd_base_addr_hi);
4195 
4196 	/* set MQD vmid to 0 */
4197 	WREG32_SOC15(GC, 0, regCP_MQD_CONTROL,
4198 	       mqd->cp_mqd_control);
4199 
4200 	/* set the pointer to the HQD, this is similar CP_RB0_BASE/_HI */
4201 	WREG32_SOC15(GC, 0, regCP_HQD_PQ_BASE,
4202 	       mqd->cp_hqd_pq_base_lo);
4203 	WREG32_SOC15(GC, 0, regCP_HQD_PQ_BASE_HI,
4204 	       mqd->cp_hqd_pq_base_hi);
4205 
4206 	/* set up the HQD, this is similar to CP_RB0_CNTL */
4207 	WREG32_SOC15(GC, 0, regCP_HQD_PQ_CONTROL,
4208 	       mqd->cp_hqd_pq_control);
4209 
4210 	/* set the wb address whether it's enabled or not */
4211 	WREG32_SOC15(GC, 0, regCP_HQD_PQ_RPTR_REPORT_ADDR,
4212 		mqd->cp_hqd_pq_rptr_report_addr_lo);
4213 	WREG32_SOC15(GC, 0, regCP_HQD_PQ_RPTR_REPORT_ADDR_HI,
4214 		mqd->cp_hqd_pq_rptr_report_addr_hi);
4215 
4216 	/* only used if CP_PQ_WPTR_POLL_CNTL.CP_PQ_WPTR_POLL_CNTL__EN_MASK=1 */
4217 	WREG32_SOC15(GC, 0, regCP_HQD_PQ_WPTR_POLL_ADDR,
4218 	       mqd->cp_hqd_pq_wptr_poll_addr_lo);
4219 	WREG32_SOC15(GC, 0, regCP_HQD_PQ_WPTR_POLL_ADDR_HI,
4220 	       mqd->cp_hqd_pq_wptr_poll_addr_hi);
4221 
4222 	/* enable the doorbell if requested */
4223 	if (ring->use_doorbell) {
4224 		WREG32_SOC15(GC, 0, regCP_MEC_DOORBELL_RANGE_LOWER,
4225 			(adev->doorbell_index.kiq * 2) << 2);
4226 		WREG32_SOC15(GC, 0, regCP_MEC_DOORBELL_RANGE_UPPER,
4227 			(adev->doorbell_index.userqueue_end * 2) << 2);
4228 	}
4229 
4230 	WREG32_SOC15(GC, 0, regCP_HQD_PQ_DOORBELL_CONTROL,
4231 	       mqd->cp_hqd_pq_doorbell_control);
4232 
4233 	/* reset read and write pointers, similar to CP_RB0_WPTR/_RPTR */
4234 	WREG32_SOC15(GC, 0, regCP_HQD_PQ_WPTR_LO,
4235 	       mqd->cp_hqd_pq_wptr_lo);
4236 	WREG32_SOC15(GC, 0, regCP_HQD_PQ_WPTR_HI,
4237 	       mqd->cp_hqd_pq_wptr_hi);
4238 
4239 	/* set the vmid for the queue */
4240 	WREG32_SOC15(GC, 0, regCP_HQD_VMID, mqd->cp_hqd_vmid);
4241 
4242 	WREG32_SOC15(GC, 0, regCP_HQD_PERSISTENT_STATE,
4243 	       mqd->cp_hqd_persistent_state);
4244 
4245 	/* activate the queue */
4246 	WREG32_SOC15(GC, 0, regCP_HQD_ACTIVE,
4247 	       mqd->cp_hqd_active);
4248 
4249 	if (ring->use_doorbell)
4250 		WREG32_FIELD15_PREREG(GC, 0, CP_PQ_STATUS, DOORBELL_ENABLE, 1);
4251 
4252 	return 0;
4253 }
4254 
4255 static int gfx_v11_0_kiq_init_queue(struct amdgpu_ring *ring)
4256 {
4257 	struct amdgpu_device *adev = ring->adev;
4258 	struct v11_compute_mqd *mqd = ring->mqd_ptr;
4259 	int mqd_idx = AMDGPU_MAX_COMPUTE_RINGS;
4260 
4261 	gfx_v11_0_kiq_setting(ring);
4262 
4263 	if (amdgpu_in_reset(adev)) { /* for GPU_RESET case */
4264 		/* reset MQD to a clean status */
4265 		if (adev->gfx.mec.mqd_backup[mqd_idx])
4266 			memcpy(mqd, adev->gfx.mec.mqd_backup[mqd_idx], sizeof(*mqd));
4267 
4268 		/* reset ring buffer */
4269 		ring->wptr = 0;
4270 		amdgpu_ring_clear_ring(ring);
4271 
4272 		mutex_lock(&adev->srbm_mutex);
4273 		soc21_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
4274 		gfx_v11_0_kiq_init_register(ring);
4275 		soc21_grbm_select(adev, 0, 0, 0, 0);
4276 		mutex_unlock(&adev->srbm_mutex);
4277 	} else {
4278 		memset((void *)mqd, 0, sizeof(*mqd));
4279 		mutex_lock(&adev->srbm_mutex);
4280 		soc21_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
4281 		amdgpu_ring_init_mqd(ring);
4282 		gfx_v11_0_kiq_init_register(ring);
4283 		soc21_grbm_select(adev, 0, 0, 0, 0);
4284 		mutex_unlock(&adev->srbm_mutex);
4285 
4286 		if (adev->gfx.mec.mqd_backup[mqd_idx])
4287 			memcpy(adev->gfx.mec.mqd_backup[mqd_idx], mqd, sizeof(*mqd));
4288 	}
4289 
4290 	return 0;
4291 }
4292 
4293 static int gfx_v11_0_kcq_init_queue(struct amdgpu_ring *ring)
4294 {
4295 	struct amdgpu_device *adev = ring->adev;
4296 	struct v11_compute_mqd *mqd = ring->mqd_ptr;
4297 	int mqd_idx = ring - &adev->gfx.compute_ring[0];
4298 
4299 	if (!amdgpu_in_reset(adev) && !adev->in_suspend) {
4300 		memset((void *)mqd, 0, sizeof(*mqd));
4301 		mutex_lock(&adev->srbm_mutex);
4302 		soc21_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
4303 		amdgpu_ring_init_mqd(ring);
4304 		soc21_grbm_select(adev, 0, 0, 0, 0);
4305 		mutex_unlock(&adev->srbm_mutex);
4306 
4307 		if (adev->gfx.mec.mqd_backup[mqd_idx])
4308 			memcpy(adev->gfx.mec.mqd_backup[mqd_idx], mqd, sizeof(*mqd));
4309 	} else if (amdgpu_in_reset(adev)) { /* for GPU_RESET case */
4310 		/* reset MQD to a clean status */
4311 		if (adev->gfx.mec.mqd_backup[mqd_idx])
4312 			memcpy(mqd, adev->gfx.mec.mqd_backup[mqd_idx], sizeof(*mqd));
4313 
4314 		/* reset ring buffer */
4315 		ring->wptr = 0;
4316 		atomic64_set((atomic64_t *)ring->wptr_cpu_addr, 0);
4317 		amdgpu_ring_clear_ring(ring);
4318 	} else {
4319 		amdgpu_ring_clear_ring(ring);
4320 	}
4321 
4322 	return 0;
4323 }
4324 
4325 static int gfx_v11_0_kiq_resume(struct amdgpu_device *adev)
4326 {
4327 	struct amdgpu_ring *ring;
4328 	int r;
4329 
4330 	ring = &adev->gfx.kiq.ring;
4331 
4332 	r = amdgpu_bo_reserve(ring->mqd_obj, false);
4333 	if (unlikely(r != 0))
4334 		return r;
4335 
4336 	r = amdgpu_bo_kmap(ring->mqd_obj, (void **)&ring->mqd_ptr);
4337 	if (unlikely(r != 0)) {
4338 		amdgpu_bo_unreserve(ring->mqd_obj);
4339 		return r;
4340 	}
4341 
4342 	gfx_v11_0_kiq_init_queue(ring);
4343 	amdgpu_bo_kunmap(ring->mqd_obj);
4344 	ring->mqd_ptr = NULL;
4345 	amdgpu_bo_unreserve(ring->mqd_obj);
4346 	ring->sched.ready = true;
4347 	return 0;
4348 }
4349 
4350 static int gfx_v11_0_kcq_resume(struct amdgpu_device *adev)
4351 {
4352 	struct amdgpu_ring *ring = NULL;
4353 	int r = 0, i;
4354 
4355 	if (!amdgpu_async_gfx_ring)
4356 		gfx_v11_0_cp_compute_enable(adev, true);
4357 
4358 	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
4359 		ring = &adev->gfx.compute_ring[i];
4360 
4361 		r = amdgpu_bo_reserve(ring->mqd_obj, false);
4362 		if (unlikely(r != 0))
4363 			goto done;
4364 		r = amdgpu_bo_kmap(ring->mqd_obj, (void **)&ring->mqd_ptr);
4365 		if (!r) {
4366 			r = gfx_v11_0_kcq_init_queue(ring);
4367 			amdgpu_bo_kunmap(ring->mqd_obj);
4368 			ring->mqd_ptr = NULL;
4369 		}
4370 		amdgpu_bo_unreserve(ring->mqd_obj);
4371 		if (r)
4372 			goto done;
4373 	}
4374 
4375 	r = amdgpu_gfx_enable_kcq(adev);
4376 done:
4377 	return r;
4378 }
4379 
4380 static int gfx_v11_0_cp_resume(struct amdgpu_device *adev)
4381 {
4382 	int r, i;
4383 	struct amdgpu_ring *ring;
4384 
4385 	if (!(adev->flags & AMD_IS_APU))
4386 		gfx_v11_0_enable_gui_idle_interrupt(adev, false);
4387 
4388 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_DIRECT) {
4389 		/* legacy firmware loading */
4390 		r = gfx_v11_0_cp_gfx_load_microcode(adev);
4391 		if (r)
4392 			return r;
4393 
4394 		if (adev->gfx.rs64_enable)
4395 			r = gfx_v11_0_cp_compute_load_microcode_rs64(adev);
4396 		else
4397 			r = gfx_v11_0_cp_compute_load_microcode(adev);
4398 		if (r)
4399 			return r;
4400 	}
4401 
4402 	gfx_v11_0_cp_set_doorbell_range(adev);
4403 
4404 	if (amdgpu_async_gfx_ring) {
4405 		gfx_v11_0_cp_compute_enable(adev, true);
4406 		gfx_v11_0_cp_gfx_enable(adev, true);
4407 	}
4408 
4409 	if (adev->enable_mes_kiq && adev->mes.kiq_hw_init)
4410 		r = amdgpu_mes_kiq_hw_init(adev);
4411 	else
4412 		r = gfx_v11_0_kiq_resume(adev);
4413 	if (r)
4414 		return r;
4415 
4416 	r = gfx_v11_0_kcq_resume(adev);
4417 	if (r)
4418 		return r;
4419 
4420 	if (!amdgpu_async_gfx_ring) {
4421 		r = gfx_v11_0_cp_gfx_resume(adev);
4422 		if (r)
4423 			return r;
4424 	} else {
4425 		r = gfx_v11_0_cp_async_gfx_ring_resume(adev);
4426 		if (r)
4427 			return r;
4428 	}
4429 
4430 	for (i = 0; i < adev->gfx.num_gfx_rings; i++) {
4431 		ring = &adev->gfx.gfx_ring[i];
4432 		r = amdgpu_ring_test_helper(ring);
4433 		if (r)
4434 			return r;
4435 	}
4436 
4437 	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
4438 		ring = &adev->gfx.compute_ring[i];
4439 		r = amdgpu_ring_test_helper(ring);
4440 		if (r)
4441 			return r;
4442 	}
4443 
4444 	return 0;
4445 }
4446 
4447 static void gfx_v11_0_cp_enable(struct amdgpu_device *adev, bool enable)
4448 {
4449 	gfx_v11_0_cp_gfx_enable(adev, enable);
4450 	gfx_v11_0_cp_compute_enable(adev, enable);
4451 }
4452 
4453 static int gfx_v11_0_gfxhub_enable(struct amdgpu_device *adev)
4454 {
4455 	int r;
4456 	bool value;
4457 
4458 	r = adev->gfxhub.funcs->gart_enable(adev);
4459 	if (r)
4460 		return r;
4461 
4462 	adev->hdp.funcs->flush_hdp(adev, NULL);
4463 
4464 	value = (amdgpu_vm_fault_stop == AMDGPU_VM_FAULT_STOP_ALWAYS) ?
4465 		false : true;
4466 
4467 	adev->gfxhub.funcs->set_fault_enable_default(adev, value);
4468 	amdgpu_gmc_flush_gpu_tlb(adev, 0, AMDGPU_GFXHUB_0, 0);
4469 
4470 	return 0;
4471 }
4472 
4473 static void gfx_v11_0_select_cp_fw_arch(struct amdgpu_device *adev)
4474 {
4475 	u32 tmp;
4476 
4477 	/* select RS64 */
4478 	if (adev->gfx.rs64_enable) {
4479 		tmp = RREG32_SOC15(GC, 0, regCP_GFX_CNTL);
4480 		tmp = REG_SET_FIELD(tmp, CP_GFX_CNTL, ENGINE_SEL, 1);
4481 		WREG32_SOC15(GC, 0, regCP_GFX_CNTL, tmp);
4482 
4483 		tmp = RREG32_SOC15(GC, 0, regCP_MEC_ISA_CNTL);
4484 		tmp = REG_SET_FIELD(tmp, CP_MEC_ISA_CNTL, ISA_MODE, 1);
4485 		WREG32_SOC15(GC, 0, regCP_MEC_ISA_CNTL, tmp);
4486 	}
4487 
4488 	if (amdgpu_emu_mode == 1)
4489 		msleep(100);
4490 }
4491 
4492 static int get_gb_addr_config(struct amdgpu_device * adev)
4493 {
4494 	u32 gb_addr_config;
4495 
4496 	gb_addr_config = RREG32_SOC15(GC, 0, regGB_ADDR_CONFIG);
4497 	if (gb_addr_config == 0)
4498 		return -EINVAL;
4499 
4500 	adev->gfx.config.gb_addr_config_fields.num_pkrs =
4501 		1 << REG_GET_FIELD(gb_addr_config, GB_ADDR_CONFIG, NUM_PKRS);
4502 
4503 	adev->gfx.config.gb_addr_config = gb_addr_config;
4504 
4505 	adev->gfx.config.gb_addr_config_fields.num_pipes = 1 <<
4506 			REG_GET_FIELD(adev->gfx.config.gb_addr_config,
4507 				      GB_ADDR_CONFIG, NUM_PIPES);
4508 
4509 	adev->gfx.config.max_tile_pipes =
4510 		adev->gfx.config.gb_addr_config_fields.num_pipes;
4511 
4512 	adev->gfx.config.gb_addr_config_fields.max_compress_frags = 1 <<
4513 			REG_GET_FIELD(adev->gfx.config.gb_addr_config,
4514 				      GB_ADDR_CONFIG, MAX_COMPRESSED_FRAGS);
4515 	adev->gfx.config.gb_addr_config_fields.num_rb_per_se = 1 <<
4516 			REG_GET_FIELD(adev->gfx.config.gb_addr_config,
4517 				      GB_ADDR_CONFIG, NUM_RB_PER_SE);
4518 	adev->gfx.config.gb_addr_config_fields.num_se = 1 <<
4519 			REG_GET_FIELD(adev->gfx.config.gb_addr_config,
4520 				      GB_ADDR_CONFIG, NUM_SHADER_ENGINES);
4521 	adev->gfx.config.gb_addr_config_fields.pipe_interleave_size = 1 << (8 +
4522 			REG_GET_FIELD(adev->gfx.config.gb_addr_config,
4523 				      GB_ADDR_CONFIG, PIPE_INTERLEAVE_SIZE));
4524 
4525 	return 0;
4526 }
4527 
4528 static void gfx_v11_0_disable_gpa_mode(struct amdgpu_device *adev)
4529 {
4530 	uint32_t data;
4531 
4532 	data = RREG32_SOC15(GC, 0, regCPC_PSP_DEBUG);
4533 	data |= CPC_PSP_DEBUG__GPA_OVERRIDE_MASK;
4534 	WREG32_SOC15(GC, 0, regCPC_PSP_DEBUG, data);
4535 
4536 	data = RREG32_SOC15(GC, 0, regCPG_PSP_DEBUG);
4537 	data |= CPG_PSP_DEBUG__GPA_OVERRIDE_MASK;
4538 	WREG32_SOC15(GC, 0, regCPG_PSP_DEBUG, data);
4539 }
4540 
4541 static int gfx_v11_0_hw_init(void *handle)
4542 {
4543 	int r;
4544 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4545 
4546 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_RLC_BACKDOOR_AUTO) {
4547 		if (adev->gfx.imu.funcs) {
4548 			/* RLC autoload sequence 1: Program rlc ram */
4549 			if (adev->gfx.imu.funcs->program_rlc_ram)
4550 				adev->gfx.imu.funcs->program_rlc_ram(adev);
4551 		}
4552 		/* rlc autoload firmware */
4553 		r = gfx_v11_0_rlc_backdoor_autoload_enable(adev);
4554 		if (r)
4555 			return r;
4556 	} else {
4557 		if (adev->firmware.load_type == AMDGPU_FW_LOAD_DIRECT) {
4558 			if (adev->gfx.imu.funcs && (amdgpu_dpm > 0)) {
4559 				if (adev->gfx.imu.funcs->load_microcode)
4560 					adev->gfx.imu.funcs->load_microcode(adev);
4561 				if (adev->gfx.imu.funcs->setup_imu)
4562 					adev->gfx.imu.funcs->setup_imu(adev);
4563 				if (adev->gfx.imu.funcs->start_imu)
4564 					adev->gfx.imu.funcs->start_imu(adev);
4565 			}
4566 		}
4567 	}
4568 
4569 	if ((adev->firmware.load_type == AMDGPU_FW_LOAD_RLC_BACKDOOR_AUTO) ||
4570 	    (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP)) {
4571 		r = gfx_v11_0_wait_for_rlc_autoload_complete(adev);
4572 		if (r) {
4573 			dev_err(adev->dev, "(%d) failed to wait rlc autoload complete\n", r);
4574 			return r;
4575 		}
4576 	}
4577 
4578 	adev->gfx.is_poweron = true;
4579 
4580 	if(get_gb_addr_config(adev))
4581 		DRM_WARN("Invalid gb_addr_config !\n");
4582 
4583 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP &&
4584 	    adev->gfx.rs64_enable)
4585 		gfx_v11_0_config_gfx_rs64(adev);
4586 
4587 	r = gfx_v11_0_gfxhub_enable(adev);
4588 	if (r)
4589 		return r;
4590 
4591 	if (!amdgpu_emu_mode)
4592 		gfx_v11_0_init_golden_registers(adev);
4593 
4594 	if ((adev->firmware.load_type == AMDGPU_FW_LOAD_DIRECT) ||
4595 	    (adev->firmware.load_type == AMDGPU_FW_LOAD_RLC_BACKDOOR_AUTO && amdgpu_dpm == 1)) {
4596 		/**
4597 		 * For gfx 11, rlc firmware loading relies on smu firmware is
4598 		 * loaded firstly, so in direct type, it has to load smc ucode
4599 		 * here before rlc.
4600 		 */
4601 		if (!(adev->flags & AMD_IS_APU)) {
4602 			r = amdgpu_pm_load_smu_firmware(adev, NULL);
4603 			if (r)
4604 				return r;
4605 		}
4606 	}
4607 
4608 	gfx_v11_0_constants_init(adev);
4609 
4610 	if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP)
4611 		gfx_v11_0_select_cp_fw_arch(adev);
4612 
4613 	if (adev->nbio.funcs->gc_doorbell_init)
4614 		adev->nbio.funcs->gc_doorbell_init(adev);
4615 
4616 	r = gfx_v11_0_rlc_resume(adev);
4617 	if (r)
4618 		return r;
4619 
4620 	/*
4621 	 * init golden registers and rlc resume may override some registers,
4622 	 * reconfig them here
4623 	 */
4624 	gfx_v11_0_tcp_harvest(adev);
4625 
4626 	r = gfx_v11_0_cp_resume(adev);
4627 	if (r)
4628 		return r;
4629 
4630 	return r;
4631 }
4632 
4633 #ifndef BRING_UP_DEBUG
4634 static int gfx_v11_0_kiq_disable_kgq(struct amdgpu_device *adev)
4635 {
4636 	struct amdgpu_kiq *kiq = &adev->gfx.kiq;
4637 	struct amdgpu_ring *kiq_ring = &kiq->ring;
4638 	int i, r = 0;
4639 
4640 	if (!kiq->pmf || !kiq->pmf->kiq_unmap_queues)
4641 		return -EINVAL;
4642 
4643 	if (amdgpu_ring_alloc(kiq_ring, kiq->pmf->unmap_queues_size *
4644 					adev->gfx.num_gfx_rings))
4645 		return -ENOMEM;
4646 
4647 	for (i = 0; i < adev->gfx.num_gfx_rings; i++)
4648 		kiq->pmf->kiq_unmap_queues(kiq_ring, &adev->gfx.gfx_ring[i],
4649 					   PREEMPT_QUEUES, 0, 0);
4650 
4651 	if (adev->gfx.kiq.ring.sched.ready)
4652 		r = amdgpu_ring_test_helper(kiq_ring);
4653 
4654 	return r;
4655 }
4656 #endif
4657 
4658 static int gfx_v11_0_hw_fini(void *handle)
4659 {
4660 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4661 	int r;
4662 	uint32_t tmp;
4663 
4664 	amdgpu_irq_put(adev, &adev->gfx.priv_reg_irq, 0);
4665 	amdgpu_irq_put(adev, &adev->gfx.priv_inst_irq, 0);
4666 
4667 	if (!adev->no_hw_access) {
4668 #ifndef BRING_UP_DEBUG
4669 		if (amdgpu_async_gfx_ring) {
4670 			r = gfx_v11_0_kiq_disable_kgq(adev);
4671 			if (r)
4672 				DRM_ERROR("KGQ disable failed\n");
4673 		}
4674 #endif
4675 		if (amdgpu_gfx_disable_kcq(adev))
4676 			DRM_ERROR("KCQ disable failed\n");
4677 
4678 		amdgpu_mes_kiq_hw_fini(adev);
4679 	}
4680 
4681 	if (amdgpu_sriov_vf(adev)) {
4682 		gfx_v11_0_cp_gfx_enable(adev, false);
4683 		/* Program KIQ position of RLC_CP_SCHEDULERS during destroy */
4684 		tmp = RREG32_SOC15(GC, 0, regRLC_CP_SCHEDULERS);
4685 		tmp &= 0xffffff00;
4686 		WREG32_SOC15(GC, 0, regRLC_CP_SCHEDULERS, tmp);
4687 
4688 		return 0;
4689 	}
4690 	gfx_v11_0_cp_enable(adev, false);
4691 	gfx_v11_0_enable_gui_idle_interrupt(adev, false);
4692 
4693 	adev->gfxhub.funcs->gart_disable(adev);
4694 
4695 	adev->gfx.is_poweron = false;
4696 
4697 	return 0;
4698 }
4699 
4700 static int gfx_v11_0_suspend(void *handle)
4701 {
4702 	return gfx_v11_0_hw_fini(handle);
4703 }
4704 
4705 static int gfx_v11_0_resume(void *handle)
4706 {
4707 	return gfx_v11_0_hw_init(handle);
4708 }
4709 
4710 static bool gfx_v11_0_is_idle(void *handle)
4711 {
4712 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4713 
4714 	if (REG_GET_FIELD(RREG32_SOC15(GC, 0, regGRBM_STATUS),
4715 				GRBM_STATUS, GUI_ACTIVE))
4716 		return false;
4717 	else
4718 		return true;
4719 }
4720 
4721 static int gfx_v11_0_wait_for_idle(void *handle)
4722 {
4723 	unsigned i;
4724 	u32 tmp;
4725 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4726 
4727 	for (i = 0; i < adev->usec_timeout; i++) {
4728 		/* read MC_STATUS */
4729 		tmp = RREG32_SOC15(GC, 0, regGRBM_STATUS) &
4730 			GRBM_STATUS__GUI_ACTIVE_MASK;
4731 
4732 		if (!REG_GET_FIELD(tmp, GRBM_STATUS, GUI_ACTIVE))
4733 			return 0;
4734 		udelay(1);
4735 	}
4736 	return -ETIMEDOUT;
4737 }
4738 
4739 static int gfx_v11_0_soft_reset(void *handle)
4740 {
4741 	u32 grbm_soft_reset = 0;
4742 	u32 tmp;
4743 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4744 
4745 	/* GRBM_STATUS */
4746 	tmp = RREG32_SOC15(GC, 0, regGRBM_STATUS);
4747 	if (tmp & (GRBM_STATUS__PA_BUSY_MASK | GRBM_STATUS__SC_BUSY_MASK |
4748 		   GRBM_STATUS__BCI_BUSY_MASK | GRBM_STATUS__SX_BUSY_MASK |
4749 		   GRBM_STATUS__TA_BUSY_MASK | GRBM_STATUS__DB_BUSY_MASK |
4750 		   GRBM_STATUS__CB_BUSY_MASK | GRBM_STATUS__GDS_BUSY_MASK |
4751 		   GRBM_STATUS__SPI_BUSY_MASK | GRBM_STATUS__GE_BUSY_NO_DMA_MASK)) {
4752 		grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset,
4753 						GRBM_SOFT_RESET, SOFT_RESET_CP,
4754 						1);
4755 		grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset,
4756 						GRBM_SOFT_RESET, SOFT_RESET_GFX,
4757 						1);
4758 	}
4759 
4760 	if (tmp & (GRBM_STATUS__CP_BUSY_MASK | GRBM_STATUS__CP_COHERENCY_BUSY_MASK)) {
4761 		grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset,
4762 						GRBM_SOFT_RESET, SOFT_RESET_CP,
4763 						1);
4764 	}
4765 
4766 	/* GRBM_STATUS2 */
4767 	tmp = RREG32_SOC15(GC, 0, regGRBM_STATUS2);
4768 	if (REG_GET_FIELD(tmp, GRBM_STATUS2, RLC_BUSY))
4769 		grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset,
4770 						GRBM_SOFT_RESET,
4771 						SOFT_RESET_RLC,
4772 						1);
4773 
4774 	if (grbm_soft_reset) {
4775 		/* stop the rlc */
4776 		gfx_v11_0_rlc_stop(adev);
4777 
4778 		/* Disable GFX parsing/prefetching */
4779 		gfx_v11_0_cp_gfx_enable(adev, false);
4780 
4781 		/* Disable MEC parsing/prefetching */
4782 		gfx_v11_0_cp_compute_enable(adev, false);
4783 
4784 		if (grbm_soft_reset) {
4785 			tmp = RREG32_SOC15(GC, 0, regGRBM_SOFT_RESET);
4786 			tmp |= grbm_soft_reset;
4787 			dev_info(adev->dev, "GRBM_SOFT_RESET=0x%08X\n", tmp);
4788 			WREG32_SOC15(GC, 0, regGRBM_SOFT_RESET, tmp);
4789 			tmp = RREG32_SOC15(GC, 0, regGRBM_SOFT_RESET);
4790 
4791 			udelay(50);
4792 
4793 			tmp &= ~grbm_soft_reset;
4794 			WREG32_SOC15(GC, 0, regGRBM_SOFT_RESET, tmp);
4795 			tmp = RREG32_SOC15(GC, 0, regGRBM_SOFT_RESET);
4796 		}
4797 
4798 		/* Wait a little for things to settle down */
4799 		udelay(50);
4800 	}
4801 	return 0;
4802 }
4803 
4804 static uint64_t gfx_v11_0_get_gpu_clock_counter(struct amdgpu_device *adev)
4805 {
4806 	uint64_t clock;
4807 
4808 	amdgpu_gfx_off_ctrl(adev, false);
4809 	mutex_lock(&adev->gfx.gpu_clock_mutex);
4810 	clock = (uint64_t)RREG32_SOC15(SMUIO, 0, regGOLDEN_TSC_COUNT_LOWER) |
4811 		((uint64_t)RREG32_SOC15(SMUIO, 0, regGOLDEN_TSC_COUNT_UPPER) << 32ULL);
4812 	mutex_unlock(&adev->gfx.gpu_clock_mutex);
4813 	amdgpu_gfx_off_ctrl(adev, true);
4814 	return clock;
4815 }
4816 
4817 static void gfx_v11_0_ring_emit_gds_switch(struct amdgpu_ring *ring,
4818 					   uint32_t vmid,
4819 					   uint32_t gds_base, uint32_t gds_size,
4820 					   uint32_t gws_base, uint32_t gws_size,
4821 					   uint32_t oa_base, uint32_t oa_size)
4822 {
4823 	struct amdgpu_device *adev = ring->adev;
4824 
4825 	/* GDS Base */
4826 	gfx_v11_0_write_data_to_reg(ring, 0, false,
4827 				    SOC15_REG_OFFSET(GC, 0, regGDS_VMID0_BASE) + 2 * vmid,
4828 				    gds_base);
4829 
4830 	/* GDS Size */
4831 	gfx_v11_0_write_data_to_reg(ring, 0, false,
4832 				    SOC15_REG_OFFSET(GC, 0, regGDS_VMID0_SIZE) + 2 * vmid,
4833 				    gds_size);
4834 
4835 	/* GWS */
4836 	gfx_v11_0_write_data_to_reg(ring, 0, false,
4837 				    SOC15_REG_OFFSET(GC, 0, regGDS_GWS_VMID0) + vmid,
4838 				    gws_size << GDS_GWS_VMID0__SIZE__SHIFT | gws_base);
4839 
4840 	/* OA */
4841 	gfx_v11_0_write_data_to_reg(ring, 0, false,
4842 				    SOC15_REG_OFFSET(GC, 0, regGDS_OA_VMID0) + vmid,
4843 				    (1 << (oa_size + oa_base)) - (1 << oa_base));
4844 }
4845 
4846 static int gfx_v11_0_early_init(void *handle)
4847 {
4848 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4849 
4850 	adev->gfx.num_gfx_rings = GFX11_NUM_GFX_RINGS;
4851 	adev->gfx.num_compute_rings = min(amdgpu_gfx_get_num_kcq(adev),
4852 					  AMDGPU_MAX_COMPUTE_RINGS);
4853 
4854 	gfx_v11_0_set_kiq_pm4_funcs(adev);
4855 	gfx_v11_0_set_ring_funcs(adev);
4856 	gfx_v11_0_set_irq_funcs(adev);
4857 	gfx_v11_0_set_gds_init(adev);
4858 	gfx_v11_0_set_rlc_funcs(adev);
4859 	gfx_v11_0_set_mqd_funcs(adev);
4860 	gfx_v11_0_set_imu_funcs(adev);
4861 
4862 	gfx_v11_0_init_rlcg_reg_access_ctrl(adev);
4863 
4864 	return 0;
4865 }
4866 
4867 static int gfx_v11_0_late_init(void *handle)
4868 {
4869 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4870 	int r;
4871 
4872 	r = amdgpu_irq_get(adev, &adev->gfx.priv_reg_irq, 0);
4873 	if (r)
4874 		return r;
4875 
4876 	r = amdgpu_irq_get(adev, &adev->gfx.priv_inst_irq, 0);
4877 	if (r)
4878 		return r;
4879 
4880 	return 0;
4881 }
4882 
4883 static bool gfx_v11_0_is_rlc_enabled(struct amdgpu_device *adev)
4884 {
4885 	uint32_t rlc_cntl;
4886 
4887 	/* if RLC is not enabled, do nothing */
4888 	rlc_cntl = RREG32_SOC15(GC, 0, regRLC_CNTL);
4889 	return (REG_GET_FIELD(rlc_cntl, RLC_CNTL, RLC_ENABLE_F32)) ? true : false;
4890 }
4891 
4892 static void gfx_v11_0_set_safe_mode(struct amdgpu_device *adev)
4893 {
4894 	uint32_t data;
4895 	unsigned i;
4896 
4897 	data = RLC_SAFE_MODE__CMD_MASK;
4898 	data |= (1 << RLC_SAFE_MODE__MESSAGE__SHIFT);
4899 
4900 	WREG32_SOC15(GC, 0, regRLC_SAFE_MODE, data);
4901 
4902 	/* wait for RLC_SAFE_MODE */
4903 	for (i = 0; i < adev->usec_timeout; i++) {
4904 		if (!REG_GET_FIELD(RREG32_SOC15(GC, 0, regRLC_SAFE_MODE),
4905 				   RLC_SAFE_MODE, CMD))
4906 			break;
4907 		udelay(1);
4908 	}
4909 }
4910 
4911 static void gfx_v11_0_unset_safe_mode(struct amdgpu_device *adev)
4912 {
4913 	WREG32_SOC15(GC, 0, regRLC_SAFE_MODE, RLC_SAFE_MODE__CMD_MASK);
4914 }
4915 
4916 static void gfx_v11_0_update_perf_clk(struct amdgpu_device *adev,
4917 				      bool enable)
4918 {
4919 	uint32_t def, data;
4920 
4921 	if (!(adev->cg_flags & AMD_CG_SUPPORT_GFX_PERF_CLK))
4922 		return;
4923 
4924 	def = data = RREG32_SOC15(GC, 0, regRLC_CGTT_MGCG_OVERRIDE);
4925 
4926 	if (enable)
4927 		data &= ~RLC_CGTT_MGCG_OVERRIDE__PERFMON_CLOCK_STATE_MASK;
4928 	else
4929 		data |= RLC_CGTT_MGCG_OVERRIDE__PERFMON_CLOCK_STATE_MASK;
4930 
4931 	if (def != data)
4932 		WREG32_SOC15(GC, 0, regRLC_CGTT_MGCG_OVERRIDE, data);
4933 }
4934 
4935 static void gfx_v11_0_update_sram_fgcg(struct amdgpu_device *adev,
4936 				       bool enable)
4937 {
4938 	uint32_t def, data;
4939 
4940 	if (!(adev->cg_flags & AMD_CG_SUPPORT_GFX_FGCG))
4941 		return;
4942 
4943 	def = data = RREG32_SOC15(GC, 0, regRLC_CGTT_MGCG_OVERRIDE);
4944 
4945 	if (enable)
4946 		data &= ~RLC_CGTT_MGCG_OVERRIDE__GFXIP_FGCG_OVERRIDE_MASK;
4947 	else
4948 		data |= RLC_CGTT_MGCG_OVERRIDE__GFXIP_FGCG_OVERRIDE_MASK;
4949 
4950 	if (def != data)
4951 		WREG32_SOC15(GC, 0, regRLC_CGTT_MGCG_OVERRIDE, data);
4952 }
4953 
4954 static void gfx_v11_0_update_repeater_fgcg(struct amdgpu_device *adev,
4955 					   bool enable)
4956 {
4957 	uint32_t def, data;
4958 
4959 	if (!(adev->cg_flags & AMD_CG_SUPPORT_REPEATER_FGCG))
4960 		return;
4961 
4962 	def = data = RREG32_SOC15(GC, 0, regRLC_CGTT_MGCG_OVERRIDE);
4963 
4964 	if (enable)
4965 		data &= ~RLC_CGTT_MGCG_OVERRIDE__GFXIP_REPEATER_FGCG_OVERRIDE_MASK;
4966 	else
4967 		data |= RLC_CGTT_MGCG_OVERRIDE__GFXIP_REPEATER_FGCG_OVERRIDE_MASK;
4968 
4969 	if (def != data)
4970 		WREG32_SOC15(GC, 0, regRLC_CGTT_MGCG_OVERRIDE, data);
4971 }
4972 
4973 static void gfx_v11_0_update_medium_grain_clock_gating(struct amdgpu_device *adev,
4974 						       bool enable)
4975 {
4976 	uint32_t data, def;
4977 
4978 	if (!(adev->cg_flags & (AMD_CG_SUPPORT_GFX_MGCG | AMD_CG_SUPPORT_GFX_MGLS)))
4979 		return;
4980 
4981 	/* It is disabled by HW by default */
4982 	if (enable) {
4983 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGCG) {
4984 			/* 1 - RLC_CGTT_MGCG_OVERRIDE */
4985 			def = data = RREG32_SOC15(GC, 0, regRLC_CGTT_MGCG_OVERRIDE);
4986 
4987 			data &= ~(RLC_CGTT_MGCG_OVERRIDE__GRBM_CGTT_SCLK_OVERRIDE_MASK |
4988 				  RLC_CGTT_MGCG_OVERRIDE__RLC_CGTT_SCLK_OVERRIDE_MASK |
4989 				  RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGCG_OVERRIDE_MASK);
4990 
4991 			if (def != data)
4992 				WREG32_SOC15(GC, 0, regRLC_CGTT_MGCG_OVERRIDE, data);
4993 		}
4994 	} else {
4995 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGCG) {
4996 			def = data = RREG32_SOC15(GC, 0, regRLC_CGTT_MGCG_OVERRIDE);
4997 
4998 			data |= (RLC_CGTT_MGCG_OVERRIDE__RLC_CGTT_SCLK_OVERRIDE_MASK |
4999 				 RLC_CGTT_MGCG_OVERRIDE__GRBM_CGTT_SCLK_OVERRIDE_MASK |
5000 				 RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGCG_OVERRIDE_MASK);
5001 
5002 			if (def != data)
5003 				WREG32_SOC15(GC, 0, regRLC_CGTT_MGCG_OVERRIDE, data);
5004 		}
5005 	}
5006 }
5007 
5008 static void gfx_v11_0_update_coarse_grain_clock_gating(struct amdgpu_device *adev,
5009 						       bool enable)
5010 {
5011 	uint32_t def, data;
5012 
5013 	if (!(adev->cg_flags &
5014 	      (AMD_CG_SUPPORT_GFX_CGCG |
5015 	      AMD_CG_SUPPORT_GFX_CGLS |
5016 	      AMD_CG_SUPPORT_GFX_3D_CGCG |
5017 	      AMD_CG_SUPPORT_GFX_3D_CGLS)))
5018 		return;
5019 
5020 	if (enable) {
5021 		def = data = RREG32_SOC15(GC, 0, regRLC_CGTT_MGCG_OVERRIDE);
5022 
5023 		/* unset CGCG override */
5024 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGCG)
5025 			data &= ~RLC_CGTT_MGCG_OVERRIDE__GFXIP_CGCG_OVERRIDE_MASK;
5026 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGLS)
5027 			data &= ~RLC_CGTT_MGCG_OVERRIDE__GFXIP_CGLS_OVERRIDE_MASK;
5028 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_3D_CGCG ||
5029 		    adev->cg_flags & AMD_CG_SUPPORT_GFX_3D_CGLS)
5030 			data &= ~RLC_CGTT_MGCG_OVERRIDE__GFXIP_GFX3D_CG_OVERRIDE_MASK;
5031 
5032 		/* update CGCG override bits */
5033 		if (def != data)
5034 			WREG32_SOC15(GC, 0, regRLC_CGTT_MGCG_OVERRIDE, data);
5035 
5036 		/* enable cgcg FSM(0x0000363F) */
5037 		def = data = RREG32_SOC15(GC, 0, regRLC_CGCG_CGLS_CTRL);
5038 
5039 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGCG) {
5040 			data &= ~RLC_CGCG_CGLS_CTRL__CGCG_GFX_IDLE_THRESHOLD_MASK;
5041 			data |= (0x36 << RLC_CGCG_CGLS_CTRL__CGCG_GFX_IDLE_THRESHOLD__SHIFT) |
5042 				 RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK;
5043 		}
5044 
5045 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGLS) {
5046 			data &= ~RLC_CGCG_CGLS_CTRL__CGLS_REP_COMPANSAT_DELAY_MASK;
5047 			data |= (0x000F << RLC_CGCG_CGLS_CTRL__CGLS_REP_COMPANSAT_DELAY__SHIFT) |
5048 				 RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK;
5049 		}
5050 
5051 		if (def != data)
5052 			WREG32_SOC15(GC, 0, regRLC_CGCG_CGLS_CTRL, data);
5053 
5054 		/* Program RLC_CGCG_CGLS_CTRL_3D */
5055 		def = data = RREG32_SOC15(GC, 0, regRLC_CGCG_CGLS_CTRL_3D);
5056 
5057 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_3D_CGCG) {
5058 			data &= ~RLC_CGCG_CGLS_CTRL_3D__CGCG_GFX_IDLE_THRESHOLD_MASK;
5059 			data |= (0x36 << RLC_CGCG_CGLS_CTRL_3D__CGCG_GFX_IDLE_THRESHOLD__SHIFT) |
5060 				 RLC_CGCG_CGLS_CTRL_3D__CGCG_EN_MASK;
5061 		}
5062 
5063 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_3D_CGLS) {
5064 			data &= ~RLC_CGCG_CGLS_CTRL_3D__CGLS_REP_COMPANSAT_DELAY_MASK;
5065 			data |= (0xf << RLC_CGCG_CGLS_CTRL_3D__CGLS_REP_COMPANSAT_DELAY__SHIFT) |
5066 				 RLC_CGCG_CGLS_CTRL_3D__CGLS_EN_MASK;
5067 		}
5068 
5069 		if (def != data)
5070 			WREG32_SOC15(GC, 0, regRLC_CGCG_CGLS_CTRL_3D, data);
5071 
5072 		/* set IDLE_POLL_COUNT(0x00900100) */
5073 		def = data = RREG32_SOC15(GC, 0, regCP_RB_WPTR_POLL_CNTL);
5074 
5075 		data &= ~(CP_RB_WPTR_POLL_CNTL__POLL_FREQUENCY_MASK | CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT_MASK);
5076 		data |= (0x0100 << CP_RB_WPTR_POLL_CNTL__POLL_FREQUENCY__SHIFT) |
5077 			(0x0090 << CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT__SHIFT);
5078 
5079 		if (def != data)
5080 			WREG32_SOC15(GC, 0, regCP_RB_WPTR_POLL_CNTL, data);
5081 
5082 		data = RREG32_SOC15(GC, 0, regCP_INT_CNTL);
5083 		data = REG_SET_FIELD(data, CP_INT_CNTL, CNTX_BUSY_INT_ENABLE, 1);
5084 		data = REG_SET_FIELD(data, CP_INT_CNTL, CNTX_EMPTY_INT_ENABLE, 1);
5085 		data = REG_SET_FIELD(data, CP_INT_CNTL, CMP_BUSY_INT_ENABLE, 1);
5086 		data = REG_SET_FIELD(data, CP_INT_CNTL, GFX_IDLE_INT_ENABLE, 1);
5087 		WREG32_SOC15(GC, 0, regCP_INT_CNTL, data);
5088 
5089 		data = RREG32_SOC15(GC, 0, regSDMA0_RLC_CGCG_CTRL);
5090 		data = REG_SET_FIELD(data, SDMA0_RLC_CGCG_CTRL, CGCG_INT_ENABLE, 1);
5091 		WREG32_SOC15(GC, 0, regSDMA0_RLC_CGCG_CTRL, data);
5092 
5093 		data = RREG32_SOC15(GC, 0, regSDMA1_RLC_CGCG_CTRL);
5094 		data = REG_SET_FIELD(data, SDMA1_RLC_CGCG_CTRL, CGCG_INT_ENABLE, 1);
5095 		WREG32_SOC15(GC, 0, regSDMA1_RLC_CGCG_CTRL, data);
5096 	} else {
5097 		/* Program RLC_CGCG_CGLS_CTRL */
5098 		def = data = RREG32_SOC15(GC, 0, regRLC_CGCG_CGLS_CTRL);
5099 
5100 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGCG)
5101 			data &= ~RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK;
5102 
5103 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGLS)
5104 			data &= ~RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK;
5105 
5106 		if (def != data)
5107 			WREG32_SOC15(GC, 0, regRLC_CGCG_CGLS_CTRL, data);
5108 
5109 		/* Program RLC_CGCG_CGLS_CTRL_3D */
5110 		def = data = RREG32_SOC15(GC, 0, regRLC_CGCG_CGLS_CTRL_3D);
5111 
5112 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_3D_CGCG)
5113 			data &= ~RLC_CGCG_CGLS_CTRL_3D__CGCG_EN_MASK;
5114 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_3D_CGLS)
5115 			data &= ~RLC_CGCG_CGLS_CTRL_3D__CGLS_EN_MASK;
5116 
5117 		if (def != data)
5118 			WREG32_SOC15(GC, 0, regRLC_CGCG_CGLS_CTRL_3D, data);
5119 
5120 		data = RREG32_SOC15(GC, 0, regSDMA0_RLC_CGCG_CTRL);
5121 		data &= ~SDMA0_RLC_CGCG_CTRL__CGCG_INT_ENABLE_MASK;
5122 		WREG32_SOC15(GC, 0, regSDMA0_RLC_CGCG_CTRL, data);
5123 
5124 		data = RREG32_SOC15(GC, 0, regSDMA1_RLC_CGCG_CTRL);
5125 		data &= ~SDMA1_RLC_CGCG_CTRL__CGCG_INT_ENABLE_MASK;
5126 		WREG32_SOC15(GC, 0, regSDMA1_RLC_CGCG_CTRL, data);
5127 	}
5128 }
5129 
5130 static int gfx_v11_0_update_gfx_clock_gating(struct amdgpu_device *adev,
5131 					    bool enable)
5132 {
5133 	amdgpu_gfx_rlc_enter_safe_mode(adev);
5134 
5135 	gfx_v11_0_update_coarse_grain_clock_gating(adev, enable);
5136 
5137 	gfx_v11_0_update_medium_grain_clock_gating(adev, enable);
5138 
5139 	gfx_v11_0_update_repeater_fgcg(adev, enable);
5140 
5141 	gfx_v11_0_update_sram_fgcg(adev, enable);
5142 
5143 	gfx_v11_0_update_perf_clk(adev, enable);
5144 
5145 	if (adev->cg_flags &
5146 	    (AMD_CG_SUPPORT_GFX_MGCG |
5147 	     AMD_CG_SUPPORT_GFX_CGLS |
5148 	     AMD_CG_SUPPORT_GFX_CGCG |
5149 	     AMD_CG_SUPPORT_GFX_3D_CGCG |
5150 	     AMD_CG_SUPPORT_GFX_3D_CGLS))
5151 	        gfx_v11_0_enable_gui_idle_interrupt(adev, enable);
5152 
5153 	amdgpu_gfx_rlc_exit_safe_mode(adev);
5154 
5155 	return 0;
5156 }
5157 
5158 static void gfx_v11_0_update_spm_vmid(struct amdgpu_device *adev, unsigned vmid)
5159 {
5160 	u32 reg, data;
5161 
5162 	reg = SOC15_REG_OFFSET(GC, 0, regRLC_SPM_MC_CNTL);
5163 	if (amdgpu_sriov_is_pp_one_vf(adev))
5164 		data = RREG32_NO_KIQ(reg);
5165 	else
5166 		data = RREG32(reg);
5167 
5168 	data &= ~RLC_SPM_MC_CNTL__RLC_SPM_VMID_MASK;
5169 	data |= (vmid & RLC_SPM_MC_CNTL__RLC_SPM_VMID_MASK) << RLC_SPM_MC_CNTL__RLC_SPM_VMID__SHIFT;
5170 
5171 	if (amdgpu_sriov_is_pp_one_vf(adev))
5172 		WREG32_SOC15_NO_KIQ(GC, 0, regRLC_SPM_MC_CNTL, data);
5173 	else
5174 		WREG32_SOC15(GC, 0, regRLC_SPM_MC_CNTL, data);
5175 }
5176 
5177 static const struct amdgpu_rlc_funcs gfx_v11_0_rlc_funcs = {
5178 	.is_rlc_enabled = gfx_v11_0_is_rlc_enabled,
5179 	.set_safe_mode = gfx_v11_0_set_safe_mode,
5180 	.unset_safe_mode = gfx_v11_0_unset_safe_mode,
5181 	.init = gfx_v11_0_rlc_init,
5182 	.get_csb_size = gfx_v11_0_get_csb_size,
5183 	.get_csb_buffer = gfx_v11_0_get_csb_buffer,
5184 	.resume = gfx_v11_0_rlc_resume,
5185 	.stop = gfx_v11_0_rlc_stop,
5186 	.reset = gfx_v11_0_rlc_reset,
5187 	.start = gfx_v11_0_rlc_start,
5188 	.update_spm_vmid = gfx_v11_0_update_spm_vmid,
5189 };
5190 
5191 static int gfx_v11_0_set_powergating_state(void *handle,
5192 					   enum amd_powergating_state state)
5193 {
5194 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5195 	bool enable = (state == AMD_PG_STATE_GATE);
5196 
5197 	if (amdgpu_sriov_vf(adev))
5198 		return 0;
5199 
5200 	switch (adev->ip_versions[GC_HWIP][0]) {
5201 	case IP_VERSION(11, 0, 0):
5202 	case IP_VERSION(11, 0, 2):
5203 		amdgpu_gfx_off_ctrl(adev, enable);
5204 		break;
5205 	default:
5206 		break;
5207 	}
5208 
5209 	return 0;
5210 }
5211 
5212 static int gfx_v11_0_set_clockgating_state(void *handle,
5213 					  enum amd_clockgating_state state)
5214 {
5215 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5216 
5217 	if (amdgpu_sriov_vf(adev))
5218 	        return 0;
5219 
5220 	switch (adev->ip_versions[GC_HWIP][0]) {
5221 	case IP_VERSION(11, 0, 0):
5222 	case IP_VERSION(11, 0, 2):
5223 	        gfx_v11_0_update_gfx_clock_gating(adev,
5224 	                        state ==  AMD_CG_STATE_GATE);
5225 	        break;
5226 	default:
5227 	        break;
5228 	}
5229 
5230 	return 0;
5231 }
5232 
5233 static void gfx_v11_0_get_clockgating_state(void *handle, u64 *flags)
5234 {
5235 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5236 	int data;
5237 
5238 	/* AMD_CG_SUPPORT_GFX_MGCG */
5239 	data = RREG32_SOC15(GC, 0, regRLC_CGTT_MGCG_OVERRIDE);
5240 	if (!(data & RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGCG_OVERRIDE_MASK))
5241 		*flags |= AMD_CG_SUPPORT_GFX_MGCG;
5242 
5243 	/* AMD_CG_SUPPORT_REPEATER_FGCG */
5244 	if (!(data & RLC_CGTT_MGCG_OVERRIDE__GFXIP_REPEATER_FGCG_OVERRIDE_MASK))
5245 		*flags |= AMD_CG_SUPPORT_REPEATER_FGCG;
5246 
5247 	/* AMD_CG_SUPPORT_GFX_FGCG */
5248 	if (!(data & RLC_CGTT_MGCG_OVERRIDE__GFXIP_FGCG_OVERRIDE_MASK))
5249 		*flags |= AMD_CG_SUPPORT_GFX_FGCG;
5250 
5251 	/* AMD_CG_SUPPORT_GFX_PERF_CLK */
5252 	if (!(data & RLC_CGTT_MGCG_OVERRIDE__PERFMON_CLOCK_STATE_MASK))
5253 		*flags |= AMD_CG_SUPPORT_GFX_PERF_CLK;
5254 
5255 	/* AMD_CG_SUPPORT_GFX_CGCG */
5256 	data = RREG32_SOC15(GC, 0, regRLC_CGCG_CGLS_CTRL);
5257 	if (data & RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK)
5258 		*flags |= AMD_CG_SUPPORT_GFX_CGCG;
5259 
5260 	/* AMD_CG_SUPPORT_GFX_CGLS */
5261 	if (data & RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK)
5262 		*flags |= AMD_CG_SUPPORT_GFX_CGLS;
5263 
5264 	/* AMD_CG_SUPPORT_GFX_3D_CGCG */
5265 	data = RREG32_SOC15(GC, 0, regRLC_CGCG_CGLS_CTRL_3D);
5266 	if (data & RLC_CGCG_CGLS_CTRL_3D__CGCG_EN_MASK)
5267 		*flags |= AMD_CG_SUPPORT_GFX_3D_CGCG;
5268 
5269 	/* AMD_CG_SUPPORT_GFX_3D_CGLS */
5270 	if (data & RLC_CGCG_CGLS_CTRL_3D__CGLS_EN_MASK)
5271 		*flags |= AMD_CG_SUPPORT_GFX_3D_CGLS;
5272 }
5273 
5274 static u64 gfx_v11_0_ring_get_rptr_gfx(struct amdgpu_ring *ring)
5275 {
5276 	/* gfx11 is 32bit rptr*/
5277 	return *(uint32_t *)ring->rptr_cpu_addr;
5278 }
5279 
5280 static u64 gfx_v11_0_ring_get_wptr_gfx(struct amdgpu_ring *ring)
5281 {
5282 	struct amdgpu_device *adev = ring->adev;
5283 	u64 wptr;
5284 
5285 	/* XXX check if swapping is necessary on BE */
5286 	if (ring->use_doorbell) {
5287 		wptr = atomic64_read((atomic64_t *)ring->wptr_cpu_addr);
5288 	} else {
5289 		wptr = RREG32_SOC15(GC, 0, regCP_RB0_WPTR);
5290 		wptr += (u64)RREG32_SOC15(GC, 0, regCP_RB0_WPTR_HI) << 32;
5291 	}
5292 
5293 	return wptr;
5294 }
5295 
5296 static void gfx_v11_0_ring_set_wptr_gfx(struct amdgpu_ring *ring)
5297 {
5298 	struct amdgpu_device *adev = ring->adev;
5299 
5300 	if (ring->use_doorbell) {
5301 		/* XXX check if swapping is necessary on BE */
5302 		atomic64_set((atomic64_t *)ring->wptr_cpu_addr, ring->wptr);
5303 		WDOORBELL64(ring->doorbell_index, ring->wptr);
5304 	} else {
5305 		WREG32_SOC15(GC, 0, regCP_RB0_WPTR, lower_32_bits(ring->wptr));
5306 		WREG32_SOC15(GC, 0, regCP_RB0_WPTR_HI, upper_32_bits(ring->wptr));
5307 	}
5308 }
5309 
5310 static u64 gfx_v11_0_ring_get_rptr_compute(struct amdgpu_ring *ring)
5311 {
5312 	/* gfx11 hardware is 32bit rptr */
5313 	return *(uint32_t *)ring->rptr_cpu_addr;
5314 }
5315 
5316 static u64 gfx_v11_0_ring_get_wptr_compute(struct amdgpu_ring *ring)
5317 {
5318 	u64 wptr;
5319 
5320 	/* XXX check if swapping is necessary on BE */
5321 	if (ring->use_doorbell)
5322 		wptr = atomic64_read((atomic64_t *)ring->wptr_cpu_addr);
5323 	else
5324 		BUG();
5325 	return wptr;
5326 }
5327 
5328 static void gfx_v11_0_ring_set_wptr_compute(struct amdgpu_ring *ring)
5329 {
5330 	struct amdgpu_device *adev = ring->adev;
5331 
5332 	/* XXX check if swapping is necessary on BE */
5333 	if (ring->use_doorbell) {
5334 		atomic64_set((atomic64_t *)ring->wptr_cpu_addr, ring->wptr);
5335 		WDOORBELL64(ring->doorbell_index, ring->wptr);
5336 	} else {
5337 		BUG(); /* only DOORBELL method supported on gfx11 now */
5338 	}
5339 }
5340 
5341 static void gfx_v11_0_ring_emit_hdp_flush(struct amdgpu_ring *ring)
5342 {
5343 	struct amdgpu_device *adev = ring->adev;
5344 	u32 ref_and_mask, reg_mem_engine;
5345 	const struct nbio_hdp_flush_reg *nbio_hf_reg = adev->nbio.hdp_flush_reg;
5346 
5347 	if (ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE) {
5348 		switch (ring->me) {
5349 		case 1:
5350 			ref_and_mask = nbio_hf_reg->ref_and_mask_cp2 << ring->pipe;
5351 			break;
5352 		case 2:
5353 			ref_and_mask = nbio_hf_reg->ref_and_mask_cp6 << ring->pipe;
5354 			break;
5355 		default:
5356 			return;
5357 		}
5358 		reg_mem_engine = 0;
5359 	} else {
5360 		ref_and_mask = nbio_hf_reg->ref_and_mask_cp0;
5361 		reg_mem_engine = 1; /* pfp */
5362 	}
5363 
5364 	gfx_v11_0_wait_reg_mem(ring, reg_mem_engine, 0, 1,
5365 			       adev->nbio.funcs->get_hdp_flush_req_offset(adev),
5366 			       adev->nbio.funcs->get_hdp_flush_done_offset(adev),
5367 			       ref_and_mask, ref_and_mask, 0x20);
5368 }
5369 
5370 static void gfx_v11_0_ring_emit_ib_gfx(struct amdgpu_ring *ring,
5371 				       struct amdgpu_job *job,
5372 				       struct amdgpu_ib *ib,
5373 				       uint32_t flags)
5374 {
5375 	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
5376 	u32 header, control = 0;
5377 
5378 	BUG_ON(ib->flags & AMDGPU_IB_FLAG_CE);
5379 
5380 	header = PACKET3(PACKET3_INDIRECT_BUFFER, 2);
5381 
5382 	control |= ib->length_dw | (vmid << 24);
5383 
5384 	if ((amdgpu_sriov_vf(ring->adev) || amdgpu_mcbp) && (ib->flags & AMDGPU_IB_FLAG_PREEMPT)) {
5385 		control |= INDIRECT_BUFFER_PRE_ENB(1);
5386 
5387 		if (flags & AMDGPU_IB_PREEMPTED)
5388 			control |= INDIRECT_BUFFER_PRE_RESUME(1);
5389 
5390 		if (vmid)
5391 			gfx_v11_0_ring_emit_de_meta(ring,
5392 				    (!amdgpu_sriov_vf(ring->adev) && flags & AMDGPU_IB_PREEMPTED) ? true : false);
5393 	}
5394 
5395 	if (ring->is_mes_queue)
5396 		/* inherit vmid from mqd */
5397 		control |= 0x400000;
5398 
5399 	amdgpu_ring_write(ring, header);
5400 	BUG_ON(ib->gpu_addr & 0x3); /* Dword align */
5401 	amdgpu_ring_write(ring,
5402 #ifdef __BIG_ENDIAN
5403 		(2 << 0) |
5404 #endif
5405 		lower_32_bits(ib->gpu_addr));
5406 	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr));
5407 	amdgpu_ring_write(ring, control);
5408 }
5409 
5410 static void gfx_v11_0_ring_emit_ib_compute(struct amdgpu_ring *ring,
5411 					   struct amdgpu_job *job,
5412 					   struct amdgpu_ib *ib,
5413 					   uint32_t flags)
5414 {
5415 	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
5416 	u32 control = INDIRECT_BUFFER_VALID | ib->length_dw | (vmid << 24);
5417 
5418 	if (ring->is_mes_queue)
5419 		/* inherit vmid from mqd */
5420 		control |= 0x40000000;
5421 
5422 	/* Currently, there is a high possibility to get wave ID mismatch
5423 	 * between ME and GDS, leading to a hw deadlock, because ME generates
5424 	 * different wave IDs than the GDS expects. This situation happens
5425 	 * randomly when at least 5 compute pipes use GDS ordered append.
5426 	 * The wave IDs generated by ME are also wrong after suspend/resume.
5427 	 * Those are probably bugs somewhere else in the kernel driver.
5428 	 *
5429 	 * Writing GDS_COMPUTE_MAX_WAVE_ID resets wave ID counters in ME and
5430 	 * GDS to 0 for this ring (me/pipe).
5431 	 */
5432 	if (ib->flags & AMDGPU_IB_FLAG_RESET_GDS_MAX_WAVE_ID) {
5433 		amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1));
5434 		amdgpu_ring_write(ring, regGDS_COMPUTE_MAX_WAVE_ID);
5435 		amdgpu_ring_write(ring, ring->adev->gds.gds_compute_max_wave_id);
5436 	}
5437 
5438 	amdgpu_ring_write(ring, PACKET3(PACKET3_INDIRECT_BUFFER, 2));
5439 	BUG_ON(ib->gpu_addr & 0x3); /* Dword align */
5440 	amdgpu_ring_write(ring,
5441 #ifdef __BIG_ENDIAN
5442 				(2 << 0) |
5443 #endif
5444 				lower_32_bits(ib->gpu_addr));
5445 	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr));
5446 	amdgpu_ring_write(ring, control);
5447 }
5448 
5449 static void gfx_v11_0_ring_emit_fence(struct amdgpu_ring *ring, u64 addr,
5450 				     u64 seq, unsigned flags)
5451 {
5452 	bool write64bit = flags & AMDGPU_FENCE_FLAG_64BIT;
5453 	bool int_sel = flags & AMDGPU_FENCE_FLAG_INT;
5454 
5455 	/* RELEASE_MEM - flush caches, send int */
5456 	amdgpu_ring_write(ring, PACKET3(PACKET3_RELEASE_MEM, 6));
5457 	amdgpu_ring_write(ring, (PACKET3_RELEASE_MEM_GCR_SEQ |
5458 				 PACKET3_RELEASE_MEM_GCR_GL2_WB |
5459 				 PACKET3_RELEASE_MEM_GCR_GL2_INV |
5460 				 PACKET3_RELEASE_MEM_GCR_GL2_US |
5461 				 PACKET3_RELEASE_MEM_GCR_GL1_INV |
5462 				 PACKET3_RELEASE_MEM_GCR_GLV_INV |
5463 				 PACKET3_RELEASE_MEM_GCR_GLM_INV |
5464 				 PACKET3_RELEASE_MEM_GCR_GLM_WB |
5465 				 PACKET3_RELEASE_MEM_CACHE_POLICY(3) |
5466 				 PACKET3_RELEASE_MEM_EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
5467 				 PACKET3_RELEASE_MEM_EVENT_INDEX(5)));
5468 	amdgpu_ring_write(ring, (PACKET3_RELEASE_MEM_DATA_SEL(write64bit ? 2 : 1) |
5469 				 PACKET3_RELEASE_MEM_INT_SEL(int_sel ? 2 : 0)));
5470 
5471 	/*
5472 	 * the address should be Qword aligned if 64bit write, Dword
5473 	 * aligned if only send 32bit data low (discard data high)
5474 	 */
5475 	if (write64bit)
5476 		BUG_ON(addr & 0x7);
5477 	else
5478 		BUG_ON(addr & 0x3);
5479 	amdgpu_ring_write(ring, lower_32_bits(addr));
5480 	amdgpu_ring_write(ring, upper_32_bits(addr));
5481 	amdgpu_ring_write(ring, lower_32_bits(seq));
5482 	amdgpu_ring_write(ring, upper_32_bits(seq));
5483 	amdgpu_ring_write(ring, ring->is_mes_queue ?
5484 			 (ring->hw_queue_id | AMDGPU_FENCE_MES_QUEUE_FLAG) : 0);
5485 }
5486 
5487 static void gfx_v11_0_ring_emit_pipeline_sync(struct amdgpu_ring *ring)
5488 {
5489 	int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX);
5490 	uint32_t seq = ring->fence_drv.sync_seq;
5491 	uint64_t addr = ring->fence_drv.gpu_addr;
5492 
5493 	gfx_v11_0_wait_reg_mem(ring, usepfp, 1, 0, lower_32_bits(addr),
5494 			       upper_32_bits(addr), seq, 0xffffffff, 4);
5495 }
5496 
5497 static void gfx_v11_0_ring_invalidate_tlbs(struct amdgpu_ring *ring,
5498 				   uint16_t pasid, uint32_t flush_type,
5499 				   bool all_hub, uint8_t dst_sel)
5500 {
5501 	amdgpu_ring_write(ring, PACKET3(PACKET3_INVALIDATE_TLBS, 0));
5502 	amdgpu_ring_write(ring,
5503 			  PACKET3_INVALIDATE_TLBS_DST_SEL(dst_sel) |
5504 			  PACKET3_INVALIDATE_TLBS_ALL_HUB(all_hub) |
5505 			  PACKET3_INVALIDATE_TLBS_PASID(pasid) |
5506 			  PACKET3_INVALIDATE_TLBS_FLUSH_TYPE(flush_type));
5507 }
5508 
5509 static void gfx_v11_0_ring_emit_vm_flush(struct amdgpu_ring *ring,
5510 					 unsigned vmid, uint64_t pd_addr)
5511 {
5512 	if (ring->is_mes_queue)
5513 		gfx_v11_0_ring_invalidate_tlbs(ring, 0, 0, false, 0);
5514 	else
5515 		amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr);
5516 
5517 	/* compute doesn't have PFP */
5518 	if (ring->funcs->type == AMDGPU_RING_TYPE_GFX) {
5519 		/* sync PFP to ME, otherwise we might get invalid PFP reads */
5520 		amdgpu_ring_write(ring, PACKET3(PACKET3_PFP_SYNC_ME, 0));
5521 		amdgpu_ring_write(ring, 0x0);
5522 	}
5523 }
5524 
5525 static void gfx_v11_0_ring_emit_fence_kiq(struct amdgpu_ring *ring, u64 addr,
5526 					  u64 seq, unsigned int flags)
5527 {
5528 	struct amdgpu_device *adev = ring->adev;
5529 
5530 	/* we only allocate 32bit for each seq wb address */
5531 	BUG_ON(flags & AMDGPU_FENCE_FLAG_64BIT);
5532 
5533 	/* write fence seq to the "addr" */
5534 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
5535 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
5536 				 WRITE_DATA_DST_SEL(5) | WR_CONFIRM));
5537 	amdgpu_ring_write(ring, lower_32_bits(addr));
5538 	amdgpu_ring_write(ring, upper_32_bits(addr));
5539 	amdgpu_ring_write(ring, lower_32_bits(seq));
5540 
5541 	if (flags & AMDGPU_FENCE_FLAG_INT) {
5542 		/* set register to trigger INT */
5543 		amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
5544 		amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
5545 					 WRITE_DATA_DST_SEL(0) | WR_CONFIRM));
5546 		amdgpu_ring_write(ring, SOC15_REG_OFFSET(GC, 0, regCPC_INT_STATUS));
5547 		amdgpu_ring_write(ring, 0);
5548 		amdgpu_ring_write(ring, 0x20000000); /* src_id is 178 */
5549 	}
5550 }
5551 
5552 static void gfx_v11_0_ring_emit_cntxcntl(struct amdgpu_ring *ring,
5553 					 uint32_t flags)
5554 {
5555 	uint32_t dw2 = 0;
5556 
5557 	dw2 |= 0x80000000; /* set load_enable otherwise this package is just NOPs */
5558 	if (flags & AMDGPU_HAVE_CTX_SWITCH) {
5559 		/* set load_global_config & load_global_uconfig */
5560 		dw2 |= 0x8001;
5561 		/* set load_cs_sh_regs */
5562 		dw2 |= 0x01000000;
5563 		/* set load_per_context_state & load_gfx_sh_regs for GFX */
5564 		dw2 |= 0x10002;
5565 	}
5566 
5567 	amdgpu_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1));
5568 	amdgpu_ring_write(ring, dw2);
5569 	amdgpu_ring_write(ring, 0);
5570 }
5571 
5572 static unsigned gfx_v11_0_ring_emit_init_cond_exec(struct amdgpu_ring *ring)
5573 {
5574 	unsigned ret;
5575 
5576 	amdgpu_ring_write(ring, PACKET3(PACKET3_COND_EXEC, 3));
5577 	amdgpu_ring_write(ring, lower_32_bits(ring->cond_exe_gpu_addr));
5578 	amdgpu_ring_write(ring, upper_32_bits(ring->cond_exe_gpu_addr));
5579 	amdgpu_ring_write(ring, 0); /* discard following DWs if *cond_exec_gpu_addr==0 */
5580 	ret = ring->wptr & ring->buf_mask;
5581 	amdgpu_ring_write(ring, 0x55aa55aa); /* patch dummy value later */
5582 
5583 	return ret;
5584 }
5585 
5586 static void gfx_v11_0_ring_emit_patch_cond_exec(struct amdgpu_ring *ring, unsigned offset)
5587 {
5588 	unsigned cur;
5589 	BUG_ON(offset > ring->buf_mask);
5590 	BUG_ON(ring->ring[offset] != 0x55aa55aa);
5591 
5592 	cur = (ring->wptr - 1) & ring->buf_mask;
5593 	if (likely(cur > offset))
5594 		ring->ring[offset] = cur - offset;
5595 	else
5596 		ring->ring[offset] = (ring->buf_mask + 1) - offset + cur;
5597 }
5598 
5599 static int gfx_v11_0_ring_preempt_ib(struct amdgpu_ring *ring)
5600 {
5601 	int i, r = 0;
5602 	struct amdgpu_device *adev = ring->adev;
5603 	struct amdgpu_kiq *kiq = &adev->gfx.kiq;
5604 	struct amdgpu_ring *kiq_ring = &kiq->ring;
5605 	unsigned long flags;
5606 
5607 	if (!kiq->pmf || !kiq->pmf->kiq_unmap_queues)
5608 		return -EINVAL;
5609 
5610 	spin_lock_irqsave(&kiq->ring_lock, flags);
5611 
5612 	if (amdgpu_ring_alloc(kiq_ring, kiq->pmf->unmap_queues_size)) {
5613 		spin_unlock_irqrestore(&kiq->ring_lock, flags);
5614 		return -ENOMEM;
5615 	}
5616 
5617 	/* assert preemption condition */
5618 	amdgpu_ring_set_preempt_cond_exec(ring, false);
5619 
5620 	/* assert IB preemption, emit the trailing fence */
5621 	kiq->pmf->kiq_unmap_queues(kiq_ring, ring, PREEMPT_QUEUES_NO_UNMAP,
5622 				   ring->trail_fence_gpu_addr,
5623 				   ++ring->trail_seq);
5624 	amdgpu_ring_commit(kiq_ring);
5625 
5626 	spin_unlock_irqrestore(&kiq->ring_lock, flags);
5627 
5628 	/* poll the trailing fence */
5629 	for (i = 0; i < adev->usec_timeout; i++) {
5630 		if (ring->trail_seq ==
5631 		    le32_to_cpu(*(ring->trail_fence_cpu_addr)))
5632 			break;
5633 		udelay(1);
5634 	}
5635 
5636 	if (i >= adev->usec_timeout) {
5637 		r = -EINVAL;
5638 		DRM_ERROR("ring %d failed to preempt ib\n", ring->idx);
5639 	}
5640 
5641 	/* deassert preemption condition */
5642 	amdgpu_ring_set_preempt_cond_exec(ring, true);
5643 	return r;
5644 }
5645 
5646 static void gfx_v11_0_ring_emit_de_meta(struct amdgpu_ring *ring, bool resume)
5647 {
5648 	struct amdgpu_device *adev = ring->adev;
5649 	struct v10_de_ib_state de_payload = {0};
5650 	uint64_t offset, gds_addr, de_payload_gpu_addr;
5651 	void *de_payload_cpu_addr;
5652 	int cnt;
5653 
5654 	if (ring->is_mes_queue) {
5655 		offset = offsetof(struct amdgpu_mes_ctx_meta_data,
5656 				  gfx[0].gfx_meta_data) +
5657 			offsetof(struct v10_gfx_meta_data, de_payload);
5658 		de_payload_gpu_addr =
5659 			amdgpu_mes_ctx_get_offs_gpu_addr(ring, offset);
5660 		de_payload_cpu_addr =
5661 			amdgpu_mes_ctx_get_offs_cpu_addr(ring, offset);
5662 
5663 		offset = offsetof(struct amdgpu_mes_ctx_meta_data,
5664 				  gfx[0].gds_backup) +
5665 			offsetof(struct v10_gfx_meta_data, de_payload);
5666 		gds_addr = amdgpu_mes_ctx_get_offs_gpu_addr(ring, offset);
5667 	} else {
5668 		offset = offsetof(struct v10_gfx_meta_data, de_payload);
5669 		de_payload_gpu_addr = amdgpu_csa_vaddr(ring->adev) + offset;
5670 		de_payload_cpu_addr = adev->virt.csa_cpu_addr + offset;
5671 
5672 		gds_addr = ALIGN(amdgpu_csa_vaddr(ring->adev) +
5673 				 AMDGPU_CSA_SIZE - adev->gds.gds_size,
5674 				 PAGE_SIZE);
5675 	}
5676 
5677 	de_payload.gds_backup_addrlo = lower_32_bits(gds_addr);
5678 	de_payload.gds_backup_addrhi = upper_32_bits(gds_addr);
5679 
5680 	cnt = (sizeof(de_payload) >> 2) + 4 - 2;
5681 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, cnt));
5682 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(1) |
5683 				 WRITE_DATA_DST_SEL(8) |
5684 				 WR_CONFIRM) |
5685 				 WRITE_DATA_CACHE_POLICY(0));
5686 	amdgpu_ring_write(ring, lower_32_bits(de_payload_gpu_addr));
5687 	amdgpu_ring_write(ring, upper_32_bits(de_payload_gpu_addr));
5688 
5689 	if (resume)
5690 		amdgpu_ring_write_multiple(ring, de_payload_cpu_addr,
5691 					   sizeof(de_payload) >> 2);
5692 	else
5693 		amdgpu_ring_write_multiple(ring, (void *)&de_payload,
5694 					   sizeof(de_payload) >> 2);
5695 }
5696 
5697 static void gfx_v11_0_ring_emit_frame_cntl(struct amdgpu_ring *ring, bool start,
5698 				    bool secure)
5699 {
5700 	uint32_t v = secure ? FRAME_TMZ : 0;
5701 
5702 	amdgpu_ring_write(ring, PACKET3(PACKET3_FRAME_CONTROL, 0));
5703 	amdgpu_ring_write(ring, v | FRAME_CMD(start ? 0 : 1));
5704 }
5705 
5706 static void gfx_v11_0_ring_emit_rreg(struct amdgpu_ring *ring, uint32_t reg,
5707 				     uint32_t reg_val_offs)
5708 {
5709 	struct amdgpu_device *adev = ring->adev;
5710 
5711 	amdgpu_ring_write(ring, PACKET3(PACKET3_COPY_DATA, 4));
5712 	amdgpu_ring_write(ring, 0 |	/* src: register*/
5713 				(5 << 8) |	/* dst: memory */
5714 				(1 << 20));	/* write confirm */
5715 	amdgpu_ring_write(ring, reg);
5716 	amdgpu_ring_write(ring, 0);
5717 	amdgpu_ring_write(ring, lower_32_bits(adev->wb.gpu_addr +
5718 				reg_val_offs * 4));
5719 	amdgpu_ring_write(ring, upper_32_bits(adev->wb.gpu_addr +
5720 				reg_val_offs * 4));
5721 }
5722 
5723 static void gfx_v11_0_ring_emit_wreg(struct amdgpu_ring *ring, uint32_t reg,
5724 				   uint32_t val)
5725 {
5726 	uint32_t cmd = 0;
5727 
5728 	switch (ring->funcs->type) {
5729 	case AMDGPU_RING_TYPE_GFX:
5730 		cmd = WRITE_DATA_ENGINE_SEL(1) | WR_CONFIRM;
5731 		break;
5732 	case AMDGPU_RING_TYPE_KIQ:
5733 		cmd = (1 << 16); /* no inc addr */
5734 		break;
5735 	default:
5736 		cmd = WR_CONFIRM;
5737 		break;
5738 	}
5739 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
5740 	amdgpu_ring_write(ring, cmd);
5741 	amdgpu_ring_write(ring, reg);
5742 	amdgpu_ring_write(ring, 0);
5743 	amdgpu_ring_write(ring, val);
5744 }
5745 
5746 static void gfx_v11_0_ring_emit_reg_wait(struct amdgpu_ring *ring, uint32_t reg,
5747 					uint32_t val, uint32_t mask)
5748 {
5749 	gfx_v11_0_wait_reg_mem(ring, 0, 0, 0, reg, 0, val, mask, 0x20);
5750 }
5751 
5752 static void gfx_v11_0_ring_emit_reg_write_reg_wait(struct amdgpu_ring *ring,
5753 						   uint32_t reg0, uint32_t reg1,
5754 						   uint32_t ref, uint32_t mask)
5755 {
5756 	int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX);
5757 
5758 	gfx_v11_0_wait_reg_mem(ring, usepfp, 0, 1, reg0, reg1,
5759 			       ref, mask, 0x20);
5760 }
5761 
5762 static void gfx_v11_0_ring_soft_recovery(struct amdgpu_ring *ring,
5763 					 unsigned vmid)
5764 {
5765 	struct amdgpu_device *adev = ring->adev;
5766 	uint32_t value = 0;
5767 
5768 	value = REG_SET_FIELD(value, SQ_CMD, CMD, 0x03);
5769 	value = REG_SET_FIELD(value, SQ_CMD, MODE, 0x01);
5770 	value = REG_SET_FIELD(value, SQ_CMD, CHECK_VMID, 1);
5771 	value = REG_SET_FIELD(value, SQ_CMD, VM_ID, vmid);
5772 	WREG32_SOC15(GC, 0, regSQ_CMD, value);
5773 }
5774 
5775 static void
5776 gfx_v11_0_set_gfx_eop_interrupt_state(struct amdgpu_device *adev,
5777 				      uint32_t me, uint32_t pipe,
5778 				      enum amdgpu_interrupt_state state)
5779 {
5780 	uint32_t cp_int_cntl, cp_int_cntl_reg;
5781 
5782 	if (!me) {
5783 		switch (pipe) {
5784 		case 0:
5785 			cp_int_cntl_reg = SOC15_REG_OFFSET(GC, 0, regCP_INT_CNTL_RING0);
5786 			break;
5787 		case 1:
5788 			cp_int_cntl_reg = SOC15_REG_OFFSET(GC, 0, regCP_INT_CNTL_RING1);
5789 			break;
5790 		default:
5791 			DRM_DEBUG("invalid pipe %d\n", pipe);
5792 			return;
5793 		}
5794 	} else {
5795 		DRM_DEBUG("invalid me %d\n", me);
5796 		return;
5797 	}
5798 
5799 	switch (state) {
5800 	case AMDGPU_IRQ_STATE_DISABLE:
5801 		cp_int_cntl = RREG32_SOC15_IP(GC, cp_int_cntl_reg);
5802 		cp_int_cntl = REG_SET_FIELD(cp_int_cntl, CP_INT_CNTL_RING0,
5803 					    TIME_STAMP_INT_ENABLE, 0);
5804 		cp_int_cntl = REG_SET_FIELD(cp_int_cntl, CP_INT_CNTL_RING0,
5805 					    GENERIC0_INT_ENABLE, 0);
5806 		WREG32_SOC15_IP(GC, cp_int_cntl_reg, cp_int_cntl);
5807 		break;
5808 	case AMDGPU_IRQ_STATE_ENABLE:
5809 		cp_int_cntl = RREG32_SOC15_IP(GC, cp_int_cntl_reg);
5810 		cp_int_cntl = REG_SET_FIELD(cp_int_cntl, CP_INT_CNTL_RING0,
5811 					    TIME_STAMP_INT_ENABLE, 1);
5812 		cp_int_cntl = REG_SET_FIELD(cp_int_cntl, CP_INT_CNTL_RING0,
5813 					    GENERIC0_INT_ENABLE, 1);
5814 		WREG32_SOC15_IP(GC, cp_int_cntl_reg, cp_int_cntl);
5815 		break;
5816 	default:
5817 		break;
5818 	}
5819 }
5820 
5821 static void gfx_v11_0_set_compute_eop_interrupt_state(struct amdgpu_device *adev,
5822 						     int me, int pipe,
5823 						     enum amdgpu_interrupt_state state)
5824 {
5825 	u32 mec_int_cntl, mec_int_cntl_reg;
5826 
5827 	/*
5828 	 * amdgpu controls only the first MEC. That's why this function only
5829 	 * handles the setting of interrupts for this specific MEC. All other
5830 	 * pipes' interrupts are set by amdkfd.
5831 	 */
5832 
5833 	if (me == 1) {
5834 		switch (pipe) {
5835 		case 0:
5836 			mec_int_cntl_reg = SOC15_REG_OFFSET(GC, 0, regCP_ME1_PIPE0_INT_CNTL);
5837 			break;
5838 		case 1:
5839 			mec_int_cntl_reg = SOC15_REG_OFFSET(GC, 0, regCP_ME1_PIPE1_INT_CNTL);
5840 			break;
5841 		case 2:
5842 			mec_int_cntl_reg = SOC15_REG_OFFSET(GC, 0, regCP_ME1_PIPE2_INT_CNTL);
5843 			break;
5844 		case 3:
5845 			mec_int_cntl_reg = SOC15_REG_OFFSET(GC, 0, regCP_ME1_PIPE3_INT_CNTL);
5846 			break;
5847 		default:
5848 			DRM_DEBUG("invalid pipe %d\n", pipe);
5849 			return;
5850 		}
5851 	} else {
5852 		DRM_DEBUG("invalid me %d\n", me);
5853 		return;
5854 	}
5855 
5856 	switch (state) {
5857 	case AMDGPU_IRQ_STATE_DISABLE:
5858 		mec_int_cntl = RREG32_SOC15_IP(GC, mec_int_cntl_reg);
5859 		mec_int_cntl = REG_SET_FIELD(mec_int_cntl, CP_ME1_PIPE0_INT_CNTL,
5860 					     TIME_STAMP_INT_ENABLE, 0);
5861 		mec_int_cntl = REG_SET_FIELD(mec_int_cntl, CP_ME1_PIPE0_INT_CNTL,
5862 					     GENERIC0_INT_ENABLE, 0);
5863 		WREG32_SOC15_IP(GC, mec_int_cntl_reg, mec_int_cntl);
5864 		break;
5865 	case AMDGPU_IRQ_STATE_ENABLE:
5866 		mec_int_cntl = RREG32_SOC15_IP(GC, mec_int_cntl_reg);
5867 		mec_int_cntl = REG_SET_FIELD(mec_int_cntl, CP_ME1_PIPE0_INT_CNTL,
5868 					     TIME_STAMP_INT_ENABLE, 1);
5869 		mec_int_cntl = REG_SET_FIELD(mec_int_cntl, CP_ME1_PIPE0_INT_CNTL,
5870 					     GENERIC0_INT_ENABLE, 1);
5871 		WREG32_SOC15_IP(GC, mec_int_cntl_reg, mec_int_cntl);
5872 		break;
5873 	default:
5874 		break;
5875 	}
5876 }
5877 
5878 static int gfx_v11_0_set_eop_interrupt_state(struct amdgpu_device *adev,
5879 					    struct amdgpu_irq_src *src,
5880 					    unsigned type,
5881 					    enum amdgpu_interrupt_state state)
5882 {
5883 	switch (type) {
5884 	case AMDGPU_CP_IRQ_GFX_ME0_PIPE0_EOP:
5885 		gfx_v11_0_set_gfx_eop_interrupt_state(adev, 0, 0, state);
5886 		break;
5887 	case AMDGPU_CP_IRQ_GFX_ME0_PIPE1_EOP:
5888 		gfx_v11_0_set_gfx_eop_interrupt_state(adev, 0, 1, state);
5889 		break;
5890 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP:
5891 		gfx_v11_0_set_compute_eop_interrupt_state(adev, 1, 0, state);
5892 		break;
5893 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE1_EOP:
5894 		gfx_v11_0_set_compute_eop_interrupt_state(adev, 1, 1, state);
5895 		break;
5896 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE2_EOP:
5897 		gfx_v11_0_set_compute_eop_interrupt_state(adev, 1, 2, state);
5898 		break;
5899 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE3_EOP:
5900 		gfx_v11_0_set_compute_eop_interrupt_state(adev, 1, 3, state);
5901 		break;
5902 	default:
5903 		break;
5904 	}
5905 	return 0;
5906 }
5907 
5908 static int gfx_v11_0_eop_irq(struct amdgpu_device *adev,
5909 			     struct amdgpu_irq_src *source,
5910 			     struct amdgpu_iv_entry *entry)
5911 {
5912 	int i;
5913 	u8 me_id, pipe_id, queue_id;
5914 	struct amdgpu_ring *ring;
5915 	uint32_t mes_queue_id = entry->src_data[0];
5916 
5917 	DRM_DEBUG("IH: CP EOP\n");
5918 
5919 	if (adev->enable_mes && (mes_queue_id & AMDGPU_FENCE_MES_QUEUE_FLAG)) {
5920 		struct amdgpu_mes_queue *queue;
5921 
5922 		mes_queue_id &= AMDGPU_FENCE_MES_QUEUE_ID_MASK;
5923 
5924 		spin_lock(&adev->mes.queue_id_lock);
5925 		queue = idr_find(&adev->mes.queue_id_idr, mes_queue_id);
5926 		if (queue) {
5927 			DRM_DEBUG("process mes queue id = %d\n", mes_queue_id);
5928 			amdgpu_fence_process(queue->ring);
5929 		}
5930 		spin_unlock(&adev->mes.queue_id_lock);
5931 	} else {
5932 		me_id = (entry->ring_id & 0x0c) >> 2;
5933 		pipe_id = (entry->ring_id & 0x03) >> 0;
5934 		queue_id = (entry->ring_id & 0x70) >> 4;
5935 
5936 		switch (me_id) {
5937 		case 0:
5938 			if (pipe_id == 0)
5939 				amdgpu_fence_process(&adev->gfx.gfx_ring[0]);
5940 			else
5941 				amdgpu_fence_process(&adev->gfx.gfx_ring[1]);
5942 			break;
5943 		case 1:
5944 		case 2:
5945 			for (i = 0; i < adev->gfx.num_compute_rings; i++) {
5946 				ring = &adev->gfx.compute_ring[i];
5947 				/* Per-queue interrupt is supported for MEC starting from VI.
5948 				 * The interrupt can only be enabled/disabled per pipe instead
5949 				 * of per queue.
5950 				 */
5951 				if ((ring->me == me_id) &&
5952 				    (ring->pipe == pipe_id) &&
5953 				    (ring->queue == queue_id))
5954 					amdgpu_fence_process(ring);
5955 			}
5956 			break;
5957 		}
5958 	}
5959 
5960 	return 0;
5961 }
5962 
5963 static int gfx_v11_0_set_priv_reg_fault_state(struct amdgpu_device *adev,
5964 					      struct amdgpu_irq_src *source,
5965 					      unsigned type,
5966 					      enum amdgpu_interrupt_state state)
5967 {
5968 	switch (state) {
5969 	case AMDGPU_IRQ_STATE_DISABLE:
5970 	case AMDGPU_IRQ_STATE_ENABLE:
5971 		WREG32_FIELD15_PREREG(GC, 0, CP_INT_CNTL_RING0,
5972 			       PRIV_REG_INT_ENABLE,
5973 			       state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0);
5974 		break;
5975 	default:
5976 		break;
5977 	}
5978 
5979 	return 0;
5980 }
5981 
5982 static int gfx_v11_0_set_priv_inst_fault_state(struct amdgpu_device *adev,
5983 					       struct amdgpu_irq_src *source,
5984 					       unsigned type,
5985 					       enum amdgpu_interrupt_state state)
5986 {
5987 	switch (state) {
5988 	case AMDGPU_IRQ_STATE_DISABLE:
5989 	case AMDGPU_IRQ_STATE_ENABLE:
5990 		WREG32_FIELD15_PREREG(GC, 0, CP_INT_CNTL_RING0,
5991 			       PRIV_INSTR_INT_ENABLE,
5992 			       state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0);
5993 		break;
5994 	default:
5995 		break;
5996 	}
5997 
5998 	return 0;
5999 }
6000 
6001 static void gfx_v11_0_handle_priv_fault(struct amdgpu_device *adev,
6002 					struct amdgpu_iv_entry *entry)
6003 {
6004 	u8 me_id, pipe_id, queue_id;
6005 	struct amdgpu_ring *ring;
6006 	int i;
6007 
6008 	me_id = (entry->ring_id & 0x0c) >> 2;
6009 	pipe_id = (entry->ring_id & 0x03) >> 0;
6010 	queue_id = (entry->ring_id & 0x70) >> 4;
6011 
6012 	switch (me_id) {
6013 	case 0:
6014 		for (i = 0; i < adev->gfx.num_gfx_rings; i++) {
6015 			ring = &adev->gfx.gfx_ring[i];
6016 			/* we only enabled 1 gfx queue per pipe for now */
6017 			if (ring->me == me_id && ring->pipe == pipe_id)
6018 				drm_sched_fault(&ring->sched);
6019 		}
6020 		break;
6021 	case 1:
6022 	case 2:
6023 		for (i = 0; i < adev->gfx.num_compute_rings; i++) {
6024 			ring = &adev->gfx.compute_ring[i];
6025 			if (ring->me == me_id && ring->pipe == pipe_id &&
6026 			    ring->queue == queue_id)
6027 				drm_sched_fault(&ring->sched);
6028 		}
6029 		break;
6030 	default:
6031 		BUG();
6032 		break;
6033 	}
6034 }
6035 
6036 static int gfx_v11_0_priv_reg_irq(struct amdgpu_device *adev,
6037 				  struct amdgpu_irq_src *source,
6038 				  struct amdgpu_iv_entry *entry)
6039 {
6040 	DRM_ERROR("Illegal register access in command stream\n");
6041 	gfx_v11_0_handle_priv_fault(adev, entry);
6042 	return 0;
6043 }
6044 
6045 static int gfx_v11_0_priv_inst_irq(struct amdgpu_device *adev,
6046 				   struct amdgpu_irq_src *source,
6047 				   struct amdgpu_iv_entry *entry)
6048 {
6049 	DRM_ERROR("Illegal instruction in command stream\n");
6050 	gfx_v11_0_handle_priv_fault(adev, entry);
6051 	return 0;
6052 }
6053 
6054 #if 0
6055 static int gfx_v11_0_kiq_set_interrupt_state(struct amdgpu_device *adev,
6056 					     struct amdgpu_irq_src *src,
6057 					     unsigned int type,
6058 					     enum amdgpu_interrupt_state state)
6059 {
6060 	uint32_t tmp, target;
6061 	struct amdgpu_ring *ring = &(adev->gfx.kiq.ring);
6062 
6063 	target = SOC15_REG_OFFSET(GC, 0, regCP_ME1_PIPE0_INT_CNTL);
6064 	target += ring->pipe;
6065 
6066 	switch (type) {
6067 	case AMDGPU_CP_KIQ_IRQ_DRIVER0:
6068 		if (state == AMDGPU_IRQ_STATE_DISABLE) {
6069 			tmp = RREG32_SOC15(GC, 0, regCPC_INT_CNTL);
6070 			tmp = REG_SET_FIELD(tmp, CPC_INT_CNTL,
6071 					    GENERIC2_INT_ENABLE, 0);
6072 			WREG32_SOC15(GC, 0, regCPC_INT_CNTL, tmp);
6073 
6074 			tmp = RREG32_SOC15_IP(GC, target);
6075 			tmp = REG_SET_FIELD(tmp, CP_ME1_PIPE0_INT_CNTL,
6076 					    GENERIC2_INT_ENABLE, 0);
6077 			WREG32_SOC15_IP(GC, target, tmp);
6078 		} else {
6079 			tmp = RREG32_SOC15(GC, 0, regCPC_INT_CNTL);
6080 			tmp = REG_SET_FIELD(tmp, CPC_INT_CNTL,
6081 					    GENERIC2_INT_ENABLE, 1);
6082 			WREG32_SOC15(GC, 0, regCPC_INT_CNTL, tmp);
6083 
6084 			tmp = RREG32_SOC15_IP(GC, target);
6085 			tmp = REG_SET_FIELD(tmp, CP_ME1_PIPE0_INT_CNTL,
6086 					    GENERIC2_INT_ENABLE, 1);
6087 			WREG32_SOC15_IP(GC, target, tmp);
6088 		}
6089 		break;
6090 	default:
6091 		BUG(); /* kiq only support GENERIC2_INT now */
6092 		break;
6093 	}
6094 	return 0;
6095 }
6096 #endif
6097 
6098 static void gfx_v11_0_emit_mem_sync(struct amdgpu_ring *ring)
6099 {
6100 	const unsigned int gcr_cntl =
6101 			PACKET3_ACQUIRE_MEM_GCR_CNTL_GL2_INV(1) |
6102 			PACKET3_ACQUIRE_MEM_GCR_CNTL_GL2_WB(1) |
6103 			PACKET3_ACQUIRE_MEM_GCR_CNTL_GLM_INV(1) |
6104 			PACKET3_ACQUIRE_MEM_GCR_CNTL_GLM_WB(1) |
6105 			PACKET3_ACQUIRE_MEM_GCR_CNTL_GL1_INV(1) |
6106 			PACKET3_ACQUIRE_MEM_GCR_CNTL_GLV_INV(1) |
6107 			PACKET3_ACQUIRE_MEM_GCR_CNTL_GLK_INV(1) |
6108 			PACKET3_ACQUIRE_MEM_GCR_CNTL_GLI_INV(1);
6109 
6110 	/* ACQUIRE_MEM - make one or more surfaces valid for use by the subsequent operations */
6111 	amdgpu_ring_write(ring, PACKET3(PACKET3_ACQUIRE_MEM, 6));
6112 	amdgpu_ring_write(ring, 0); /* CP_COHER_CNTL */
6113 	amdgpu_ring_write(ring, 0xffffffff);  /* CP_COHER_SIZE */
6114 	amdgpu_ring_write(ring, 0xffffff);  /* CP_COHER_SIZE_HI */
6115 	amdgpu_ring_write(ring, 0); /* CP_COHER_BASE */
6116 	amdgpu_ring_write(ring, 0);  /* CP_COHER_BASE_HI */
6117 	amdgpu_ring_write(ring, 0x0000000A); /* POLL_INTERVAL */
6118 	amdgpu_ring_write(ring, gcr_cntl); /* GCR_CNTL */
6119 }
6120 
6121 static const struct amd_ip_funcs gfx_v11_0_ip_funcs = {
6122 	.name = "gfx_v11_0",
6123 	.early_init = gfx_v11_0_early_init,
6124 	.late_init = gfx_v11_0_late_init,
6125 	.sw_init = gfx_v11_0_sw_init,
6126 	.sw_fini = gfx_v11_0_sw_fini,
6127 	.hw_init = gfx_v11_0_hw_init,
6128 	.hw_fini = gfx_v11_0_hw_fini,
6129 	.suspend = gfx_v11_0_suspend,
6130 	.resume = gfx_v11_0_resume,
6131 	.is_idle = gfx_v11_0_is_idle,
6132 	.wait_for_idle = gfx_v11_0_wait_for_idle,
6133 	.soft_reset = gfx_v11_0_soft_reset,
6134 	.set_clockgating_state = gfx_v11_0_set_clockgating_state,
6135 	.set_powergating_state = gfx_v11_0_set_powergating_state,
6136 	.get_clockgating_state = gfx_v11_0_get_clockgating_state,
6137 };
6138 
6139 static const struct amdgpu_ring_funcs gfx_v11_0_ring_funcs_gfx = {
6140 	.type = AMDGPU_RING_TYPE_GFX,
6141 	.align_mask = 0xff,
6142 	.nop = PACKET3(PACKET3_NOP, 0x3FFF),
6143 	.support_64bit_ptrs = true,
6144 	.vmhub = AMDGPU_GFXHUB_0,
6145 	.get_rptr = gfx_v11_0_ring_get_rptr_gfx,
6146 	.get_wptr = gfx_v11_0_ring_get_wptr_gfx,
6147 	.set_wptr = gfx_v11_0_ring_set_wptr_gfx,
6148 	.emit_frame_size = /* totally 242 maximum if 16 IBs */
6149 		5 + /* COND_EXEC */
6150 		7 + /* PIPELINE_SYNC */
6151 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 5 +
6152 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 7 +
6153 		2 + /* VM_FLUSH */
6154 		8 + /* FENCE for VM_FLUSH */
6155 		20 + /* GDS switch */
6156 		5 + /* COND_EXEC */
6157 		7 + /* HDP_flush */
6158 		4 + /* VGT_flush */
6159 		31 + /*	DE_META */
6160 		3 + /* CNTX_CTRL */
6161 		5 + /* HDP_INVL */
6162 		8 + 8 + /* FENCE x2 */
6163 		8, /* gfx_v11_0_emit_mem_sync */
6164 	.emit_ib_size =	4, /* gfx_v11_0_ring_emit_ib_gfx */
6165 	.emit_ib = gfx_v11_0_ring_emit_ib_gfx,
6166 	.emit_fence = gfx_v11_0_ring_emit_fence,
6167 	.emit_pipeline_sync = gfx_v11_0_ring_emit_pipeline_sync,
6168 	.emit_vm_flush = gfx_v11_0_ring_emit_vm_flush,
6169 	.emit_gds_switch = gfx_v11_0_ring_emit_gds_switch,
6170 	.emit_hdp_flush = gfx_v11_0_ring_emit_hdp_flush,
6171 	.test_ring = gfx_v11_0_ring_test_ring,
6172 	.test_ib = gfx_v11_0_ring_test_ib,
6173 	.insert_nop = amdgpu_ring_insert_nop,
6174 	.pad_ib = amdgpu_ring_generic_pad_ib,
6175 	.emit_cntxcntl = gfx_v11_0_ring_emit_cntxcntl,
6176 	.init_cond_exec = gfx_v11_0_ring_emit_init_cond_exec,
6177 	.patch_cond_exec = gfx_v11_0_ring_emit_patch_cond_exec,
6178 	.preempt_ib = gfx_v11_0_ring_preempt_ib,
6179 	.emit_frame_cntl = gfx_v11_0_ring_emit_frame_cntl,
6180 	.emit_wreg = gfx_v11_0_ring_emit_wreg,
6181 	.emit_reg_wait = gfx_v11_0_ring_emit_reg_wait,
6182 	.emit_reg_write_reg_wait = gfx_v11_0_ring_emit_reg_write_reg_wait,
6183 	.soft_recovery = gfx_v11_0_ring_soft_recovery,
6184 	.emit_mem_sync = gfx_v11_0_emit_mem_sync,
6185 };
6186 
6187 static const struct amdgpu_ring_funcs gfx_v11_0_ring_funcs_compute = {
6188 	.type = AMDGPU_RING_TYPE_COMPUTE,
6189 	.align_mask = 0xff,
6190 	.nop = PACKET3(PACKET3_NOP, 0x3FFF),
6191 	.support_64bit_ptrs = true,
6192 	.vmhub = AMDGPU_GFXHUB_0,
6193 	.get_rptr = gfx_v11_0_ring_get_rptr_compute,
6194 	.get_wptr = gfx_v11_0_ring_get_wptr_compute,
6195 	.set_wptr = gfx_v11_0_ring_set_wptr_compute,
6196 	.emit_frame_size =
6197 		20 + /* gfx_v11_0_ring_emit_gds_switch */
6198 		7 + /* gfx_v11_0_ring_emit_hdp_flush */
6199 		5 + /* hdp invalidate */
6200 		7 + /* gfx_v11_0_ring_emit_pipeline_sync */
6201 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 5 +
6202 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 7 +
6203 		2 + /* gfx_v11_0_ring_emit_vm_flush */
6204 		8 + 8 + 8 + /* gfx_v11_0_ring_emit_fence x3 for user fence, vm fence */
6205 		8, /* gfx_v11_0_emit_mem_sync */
6206 	.emit_ib_size =	7, /* gfx_v11_0_ring_emit_ib_compute */
6207 	.emit_ib = gfx_v11_0_ring_emit_ib_compute,
6208 	.emit_fence = gfx_v11_0_ring_emit_fence,
6209 	.emit_pipeline_sync = gfx_v11_0_ring_emit_pipeline_sync,
6210 	.emit_vm_flush = gfx_v11_0_ring_emit_vm_flush,
6211 	.emit_gds_switch = gfx_v11_0_ring_emit_gds_switch,
6212 	.emit_hdp_flush = gfx_v11_0_ring_emit_hdp_flush,
6213 	.test_ring = gfx_v11_0_ring_test_ring,
6214 	.test_ib = gfx_v11_0_ring_test_ib,
6215 	.insert_nop = amdgpu_ring_insert_nop,
6216 	.pad_ib = amdgpu_ring_generic_pad_ib,
6217 	.emit_wreg = gfx_v11_0_ring_emit_wreg,
6218 	.emit_reg_wait = gfx_v11_0_ring_emit_reg_wait,
6219 	.emit_reg_write_reg_wait = gfx_v11_0_ring_emit_reg_write_reg_wait,
6220 	.emit_mem_sync = gfx_v11_0_emit_mem_sync,
6221 };
6222 
6223 static const struct amdgpu_ring_funcs gfx_v11_0_ring_funcs_kiq = {
6224 	.type = AMDGPU_RING_TYPE_KIQ,
6225 	.align_mask = 0xff,
6226 	.nop = PACKET3(PACKET3_NOP, 0x3FFF),
6227 	.support_64bit_ptrs = true,
6228 	.vmhub = AMDGPU_GFXHUB_0,
6229 	.get_rptr = gfx_v11_0_ring_get_rptr_compute,
6230 	.get_wptr = gfx_v11_0_ring_get_wptr_compute,
6231 	.set_wptr = gfx_v11_0_ring_set_wptr_compute,
6232 	.emit_frame_size =
6233 		20 + /* gfx_v11_0_ring_emit_gds_switch */
6234 		7 + /* gfx_v11_0_ring_emit_hdp_flush */
6235 		5 + /*hdp invalidate */
6236 		7 + /* gfx_v11_0_ring_emit_pipeline_sync */
6237 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 5 +
6238 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 7 +
6239 		2 + /* gfx_v11_0_ring_emit_vm_flush */
6240 		8 + 8 + 8, /* gfx_v11_0_ring_emit_fence_kiq x3 for user fence, vm fence */
6241 	.emit_ib_size =	7, /* gfx_v11_0_ring_emit_ib_compute */
6242 	.emit_ib = gfx_v11_0_ring_emit_ib_compute,
6243 	.emit_fence = gfx_v11_0_ring_emit_fence_kiq,
6244 	.test_ring = gfx_v11_0_ring_test_ring,
6245 	.test_ib = gfx_v11_0_ring_test_ib,
6246 	.insert_nop = amdgpu_ring_insert_nop,
6247 	.pad_ib = amdgpu_ring_generic_pad_ib,
6248 	.emit_rreg = gfx_v11_0_ring_emit_rreg,
6249 	.emit_wreg = gfx_v11_0_ring_emit_wreg,
6250 	.emit_reg_wait = gfx_v11_0_ring_emit_reg_wait,
6251 	.emit_reg_write_reg_wait = gfx_v11_0_ring_emit_reg_write_reg_wait,
6252 };
6253 
6254 static void gfx_v11_0_set_ring_funcs(struct amdgpu_device *adev)
6255 {
6256 	int i;
6257 
6258 	adev->gfx.kiq.ring.funcs = &gfx_v11_0_ring_funcs_kiq;
6259 
6260 	for (i = 0; i < adev->gfx.num_gfx_rings; i++)
6261 		adev->gfx.gfx_ring[i].funcs = &gfx_v11_0_ring_funcs_gfx;
6262 
6263 	for (i = 0; i < adev->gfx.num_compute_rings; i++)
6264 		adev->gfx.compute_ring[i].funcs = &gfx_v11_0_ring_funcs_compute;
6265 }
6266 
6267 static const struct amdgpu_irq_src_funcs gfx_v11_0_eop_irq_funcs = {
6268 	.set = gfx_v11_0_set_eop_interrupt_state,
6269 	.process = gfx_v11_0_eop_irq,
6270 };
6271 
6272 static const struct amdgpu_irq_src_funcs gfx_v11_0_priv_reg_irq_funcs = {
6273 	.set = gfx_v11_0_set_priv_reg_fault_state,
6274 	.process = gfx_v11_0_priv_reg_irq,
6275 };
6276 
6277 static const struct amdgpu_irq_src_funcs gfx_v11_0_priv_inst_irq_funcs = {
6278 	.set = gfx_v11_0_set_priv_inst_fault_state,
6279 	.process = gfx_v11_0_priv_inst_irq,
6280 };
6281 
6282 static void gfx_v11_0_set_irq_funcs(struct amdgpu_device *adev)
6283 {
6284 	adev->gfx.eop_irq.num_types = AMDGPU_CP_IRQ_LAST;
6285 	adev->gfx.eop_irq.funcs = &gfx_v11_0_eop_irq_funcs;
6286 
6287 	adev->gfx.priv_reg_irq.num_types = 1;
6288 	adev->gfx.priv_reg_irq.funcs = &gfx_v11_0_priv_reg_irq_funcs;
6289 
6290 	adev->gfx.priv_inst_irq.num_types = 1;
6291 	adev->gfx.priv_inst_irq.funcs = &gfx_v11_0_priv_inst_irq_funcs;
6292 }
6293 
6294 static void gfx_v11_0_set_imu_funcs(struct amdgpu_device *adev)
6295 {
6296 	adev->gfx.imu.funcs = &gfx_v11_0_imu_funcs;
6297 }
6298 
6299 static void gfx_v11_0_set_rlc_funcs(struct amdgpu_device *adev)
6300 {
6301 	adev->gfx.rlc.funcs = &gfx_v11_0_rlc_funcs;
6302 }
6303 
6304 static void gfx_v11_0_set_gds_init(struct amdgpu_device *adev)
6305 {
6306 	unsigned total_cu = adev->gfx.config.max_cu_per_sh *
6307 			    adev->gfx.config.max_sh_per_se *
6308 			    adev->gfx.config.max_shader_engines;
6309 
6310 	adev->gds.gds_size = 0x1000;
6311 	adev->gds.gds_compute_max_wave_id = total_cu * 32 - 1;
6312 	adev->gds.gws_size = 64;
6313 	adev->gds.oa_size = 16;
6314 }
6315 
6316 static void gfx_v11_0_set_mqd_funcs(struct amdgpu_device *adev)
6317 {
6318 	/* set gfx eng mqd */
6319 	adev->mqds[AMDGPU_HW_IP_GFX].mqd_size =
6320 		sizeof(struct v11_gfx_mqd);
6321 	adev->mqds[AMDGPU_HW_IP_GFX].init_mqd =
6322 		gfx_v11_0_gfx_mqd_init;
6323 	/* set compute eng mqd */
6324 	adev->mqds[AMDGPU_HW_IP_COMPUTE].mqd_size =
6325 		sizeof(struct v11_compute_mqd);
6326 	adev->mqds[AMDGPU_HW_IP_COMPUTE].init_mqd =
6327 		gfx_v11_0_compute_mqd_init;
6328 }
6329 
6330 static void gfx_v11_0_set_user_wgp_inactive_bitmap_per_sh(struct amdgpu_device *adev,
6331 							  u32 bitmap)
6332 {
6333 	u32 data;
6334 
6335 	if (!bitmap)
6336 		return;
6337 
6338 	data = bitmap << GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_WGPS__SHIFT;
6339 	data &= GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_WGPS_MASK;
6340 
6341 	WREG32_SOC15(GC, 0, regGC_USER_SHADER_ARRAY_CONFIG, data);
6342 }
6343 
6344 static u32 gfx_v11_0_get_wgp_active_bitmap_per_sh(struct amdgpu_device *adev)
6345 {
6346 	u32 data, wgp_bitmask;
6347 	data = RREG32_SOC15(GC, 0, regCC_GC_SHADER_ARRAY_CONFIG);
6348 	data |= RREG32_SOC15(GC, 0, regGC_USER_SHADER_ARRAY_CONFIG);
6349 
6350 	data &= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_WGPS_MASK;
6351 	data >>= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_WGPS__SHIFT;
6352 
6353 	wgp_bitmask =
6354 		amdgpu_gfx_create_bitmask(adev->gfx.config.max_cu_per_sh >> 1);
6355 
6356 	return (~data) & wgp_bitmask;
6357 }
6358 
6359 static u32 gfx_v11_0_get_cu_active_bitmap_per_sh(struct amdgpu_device *adev)
6360 {
6361 	u32 wgp_idx, wgp_active_bitmap;
6362 	u32 cu_bitmap_per_wgp, cu_active_bitmap;
6363 
6364 	wgp_active_bitmap = gfx_v11_0_get_wgp_active_bitmap_per_sh(adev);
6365 	cu_active_bitmap = 0;
6366 
6367 	for (wgp_idx = 0; wgp_idx < 16; wgp_idx++) {
6368 		/* if there is one WGP enabled, it means 2 CUs will be enabled */
6369 		cu_bitmap_per_wgp = 3 << (2 * wgp_idx);
6370 		if (wgp_active_bitmap & (1 << wgp_idx))
6371 			cu_active_bitmap |= cu_bitmap_per_wgp;
6372 	}
6373 
6374 	return cu_active_bitmap;
6375 }
6376 
6377 static int gfx_v11_0_get_cu_info(struct amdgpu_device *adev,
6378 				 struct amdgpu_cu_info *cu_info)
6379 {
6380 	int i, j, k, counter, active_cu_number = 0;
6381 	u32 mask, bitmap;
6382 	unsigned disable_masks[8 * 2];
6383 
6384 	if (!adev || !cu_info)
6385 		return -EINVAL;
6386 
6387 	amdgpu_gfx_parse_disable_cu(disable_masks, 8, 2);
6388 
6389 	mutex_lock(&adev->grbm_idx_mutex);
6390 	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
6391 		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
6392 			mask = 1;
6393 			counter = 0;
6394 			gfx_v11_0_select_se_sh(adev, i, j, 0xffffffff);
6395 			if (i < 8 && j < 2)
6396 				gfx_v11_0_set_user_wgp_inactive_bitmap_per_sh(
6397 					adev, disable_masks[i * 2 + j]);
6398 			bitmap = gfx_v11_0_get_cu_active_bitmap_per_sh(adev);
6399 
6400 			/**
6401 			 * GFX11 could support more than 4 SEs, while the bitmap
6402 			 * in cu_info struct is 4x4 and ioctl interface struct
6403 			 * drm_amdgpu_info_device should keep stable.
6404 			 * So we use last two columns of bitmap to store cu mask for
6405 			 * SEs 4 to 7, the layout of the bitmap is as below:
6406 			 *    SE0: {SH0,SH1} --> {bitmap[0][0], bitmap[0][1]}
6407 			 *    SE1: {SH0,SH1} --> {bitmap[1][0], bitmap[1][1]}
6408 			 *    SE2: {SH0,SH1} --> {bitmap[2][0], bitmap[2][1]}
6409 			 *    SE3: {SH0,SH1} --> {bitmap[3][0], bitmap[3][1]}
6410 			 *    SE4: {SH0,SH1} --> {bitmap[0][2], bitmap[0][3]}
6411 			 *    SE5: {SH0,SH1} --> {bitmap[1][2], bitmap[1][3]}
6412 			 *    SE6: {SH0,SH1} --> {bitmap[2][2], bitmap[2][3]}
6413 			 *    SE7: {SH0,SH1} --> {bitmap[3][2], bitmap[3][3]}
6414 			 */
6415 			cu_info->bitmap[i % 4][j + (i / 4) * 2] = bitmap;
6416 
6417 			for (k = 0; k < adev->gfx.config.max_cu_per_sh; k++) {
6418 				if (bitmap & mask)
6419 					counter++;
6420 
6421 				mask <<= 1;
6422 			}
6423 			active_cu_number += counter;
6424 		}
6425 	}
6426 	gfx_v11_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
6427 	mutex_unlock(&adev->grbm_idx_mutex);
6428 
6429 	cu_info->number = active_cu_number;
6430 	cu_info->simd_per_cu = NUM_SIMD_PER_CU;
6431 
6432 	return 0;
6433 }
6434 
6435 const struct amdgpu_ip_block_version gfx_v11_0_ip_block =
6436 {
6437 	.type = AMD_IP_BLOCK_TYPE_GFX,
6438 	.major = 11,
6439 	.minor = 0,
6440 	.rev = 0,
6441 	.funcs = &gfx_v11_0_ip_funcs,
6442 };
6443