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 "gfx_v11_0_3.h"
50 #include "nbio_v4_3.h"
51 #include "mes_v11_0.h"
52 
53 #define GFX11_NUM_GFX_RINGS		1
54 #define GFX11_MEC_HPD_SIZE	2048
55 
56 #define RLCG_UCODE_LOADING_START_ADDRESS	0x00002000L
57 #define RLC_PG_DELAY_3_DEFAULT_GC_11_0_1	0x1388
58 
59 #define regCGTT_WD_CLK_CTRL		0x5086
60 #define regCGTT_WD_CLK_CTRL_BASE_IDX	1
61 #define regRLC_RLCS_BOOTLOAD_STATUS_gc_11_0_1	0x4e7e
62 #define regRLC_RLCS_BOOTLOAD_STATUS_gc_11_0_1_BASE_IDX	1
63 
64 MODULE_FIRMWARE("amdgpu/gc_11_0_0_pfp.bin");
65 MODULE_FIRMWARE("amdgpu/gc_11_0_0_me.bin");
66 MODULE_FIRMWARE("amdgpu/gc_11_0_0_mec.bin");
67 MODULE_FIRMWARE("amdgpu/gc_11_0_0_rlc.bin");
68 MODULE_FIRMWARE("amdgpu/gc_11_0_0_toc.bin");
69 MODULE_FIRMWARE("amdgpu/gc_11_0_1_pfp.bin");
70 MODULE_FIRMWARE("amdgpu/gc_11_0_1_me.bin");
71 MODULE_FIRMWARE("amdgpu/gc_11_0_1_mec.bin");
72 MODULE_FIRMWARE("amdgpu/gc_11_0_1_rlc.bin");
73 MODULE_FIRMWARE("amdgpu/gc_11_0_2_pfp.bin");
74 MODULE_FIRMWARE("amdgpu/gc_11_0_2_me.bin");
75 MODULE_FIRMWARE("amdgpu/gc_11_0_2_mec.bin");
76 MODULE_FIRMWARE("amdgpu/gc_11_0_2_rlc.bin");
77 MODULE_FIRMWARE("amdgpu/gc_11_0_3_pfp.bin");
78 MODULE_FIRMWARE("amdgpu/gc_11_0_3_me.bin");
79 MODULE_FIRMWARE("amdgpu/gc_11_0_3_mec.bin");
80 MODULE_FIRMWARE("amdgpu/gc_11_0_3_rlc.bin");
81 MODULE_FIRMWARE("amdgpu/gc_11_0_4_pfp.bin");
82 MODULE_FIRMWARE("amdgpu/gc_11_0_4_me.bin");
83 MODULE_FIRMWARE("amdgpu/gc_11_0_4_mec.bin");
84 MODULE_FIRMWARE("amdgpu/gc_11_0_4_rlc.bin");
85 
86 static const struct soc15_reg_golden golden_settings_gc_11_0[] = {
87 	SOC15_REG_GOLDEN_VALUE(GC, 0, regTCP_CNTL, 0x20000000, 0x20000000)
88 };
89 
90 static const struct soc15_reg_golden golden_settings_gc_11_0_1[] =
91 {
92 	SOC15_REG_GOLDEN_VALUE(GC, 0, regCGTT_GS_NGG_CLK_CTRL, 0x9fff8fff, 0x00000010),
93 	SOC15_REG_GOLDEN_VALUE(GC, 0, regCGTT_WD_CLK_CTRL, 0xffff8fff, 0x00000010),
94 	SOC15_REG_GOLDEN_VALUE(GC, 0, regCPF_GCR_CNTL, 0x0007ffff, 0x0000c200),
95 	SOC15_REG_GOLDEN_VALUE(GC, 0, regGL2C_CTRL3, 0xffff001b, 0x00f01988),
96 	SOC15_REG_GOLDEN_VALUE(GC, 0, regPA_CL_ENHANCE, 0xf0ffffff, 0x00880007),
97 	SOC15_REG_GOLDEN_VALUE(GC, 0, regPA_SC_ENHANCE_3, 0xfffffffd, 0x00000008),
98 	SOC15_REG_GOLDEN_VALUE(GC, 0, regPA_SC_VRS_SURFACE_CNTL_1, 0xfff891ff, 0x55480100),
99 	SOC15_REG_GOLDEN_VALUE(GC, 0, regTA_CNTL_AUX, 0xf7f7ffff, 0x01030000),
100 	SOC15_REG_GOLDEN_VALUE(GC, 0, regTCP_CNTL2, 0xfcffffff, 0x0000000a)
101 };
102 
103 #define DEFAULT_SH_MEM_CONFIG \
104 	((SH_MEM_ADDRESS_MODE_64 << SH_MEM_CONFIG__ADDRESS_MODE__SHIFT) | \
105 	 (SH_MEM_ALIGNMENT_MODE_UNALIGNED << SH_MEM_CONFIG__ALIGNMENT_MODE__SHIFT) | \
106 	 (3 << SH_MEM_CONFIG__INITIAL_INST_PREFETCH__SHIFT))
107 
108 static void gfx_v11_0_disable_gpa_mode(struct amdgpu_device *adev);
109 static void gfx_v11_0_set_ring_funcs(struct amdgpu_device *adev);
110 static void gfx_v11_0_set_irq_funcs(struct amdgpu_device *adev);
111 static void gfx_v11_0_set_gds_init(struct amdgpu_device *adev);
112 static void gfx_v11_0_set_rlc_funcs(struct amdgpu_device *adev);
113 static void gfx_v11_0_set_mqd_funcs(struct amdgpu_device *adev);
114 static void gfx_v11_0_set_imu_funcs(struct amdgpu_device *adev);
115 static int gfx_v11_0_get_cu_info(struct amdgpu_device *adev,
116                                  struct amdgpu_cu_info *cu_info);
117 static uint64_t gfx_v11_0_get_gpu_clock_counter(struct amdgpu_device *adev);
118 static void gfx_v11_0_select_se_sh(struct amdgpu_device *adev, u32 se_num,
119 				   u32 sh_num, u32 instance, int xcc_id);
120 static u32 gfx_v11_0_get_wgp_active_bitmap_per_sh(struct amdgpu_device *adev);
121 
122 static void gfx_v11_0_ring_emit_de_meta(struct amdgpu_ring *ring, bool resume);
123 static void gfx_v11_0_ring_emit_frame_cntl(struct amdgpu_ring *ring, bool start, bool secure);
124 static void gfx_v11_0_ring_emit_wreg(struct amdgpu_ring *ring, uint32_t reg,
125 				     uint32_t val);
126 static int gfx_v11_0_wait_for_rlc_autoload_complete(struct amdgpu_device *adev);
127 static void gfx_v11_0_ring_invalidate_tlbs(struct amdgpu_ring *ring,
128 					   uint16_t pasid, uint32_t flush_type,
129 					   bool all_hub, uint8_t dst_sel);
130 static void gfx_v11_0_set_safe_mode(struct amdgpu_device *adev, int xcc_id);
131 static void gfx_v11_0_unset_safe_mode(struct amdgpu_device *adev, int xcc_id);
132 static void gfx_v11_0_update_perf_clk(struct amdgpu_device *adev,
133 				      bool enable);
134 
135 static void gfx11_kiq_set_resources(struct amdgpu_ring *kiq_ring, uint64_t queue_mask)
136 {
137 	amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_SET_RESOURCES, 6));
138 	amdgpu_ring_write(kiq_ring, PACKET3_SET_RESOURCES_VMID_MASK(0) |
139 			  PACKET3_SET_RESOURCES_QUEUE_TYPE(0));	/* vmid_mask:0 queue_type:0 (KIQ) */
140 	amdgpu_ring_write(kiq_ring, lower_32_bits(queue_mask));	/* queue mask lo */
141 	amdgpu_ring_write(kiq_ring, upper_32_bits(queue_mask));	/* queue mask hi */
142 	amdgpu_ring_write(kiq_ring, 0);	/* gws mask lo */
143 	amdgpu_ring_write(kiq_ring, 0);	/* gws mask hi */
144 	amdgpu_ring_write(kiq_ring, 0);	/* oac mask */
145 	amdgpu_ring_write(kiq_ring, 0);	/* gds heap base:0, gds heap size:0 */
146 }
147 
148 static void gfx11_kiq_map_queues(struct amdgpu_ring *kiq_ring,
149 				 struct amdgpu_ring *ring)
150 {
151 	uint64_t mqd_addr = amdgpu_bo_gpu_offset(ring->mqd_obj);
152 	uint64_t wptr_addr = ring->wptr_gpu_addr;
153 	uint32_t me = 0, eng_sel = 0;
154 
155 	switch (ring->funcs->type) {
156 	case AMDGPU_RING_TYPE_COMPUTE:
157 		me = 1;
158 		eng_sel = 0;
159 		break;
160 	case AMDGPU_RING_TYPE_GFX:
161 		me = 0;
162 		eng_sel = 4;
163 		break;
164 	case AMDGPU_RING_TYPE_MES:
165 		me = 2;
166 		eng_sel = 5;
167 		break;
168 	default:
169 		WARN_ON(1);
170 	}
171 
172 	amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_MAP_QUEUES, 5));
173 	/* Q_sel:0, vmid:0, vidmem: 1, engine:0, num_Q:1*/
174 	amdgpu_ring_write(kiq_ring, /* Q_sel: 0, vmid: 0, engine: 0, num_Q: 1 */
175 			  PACKET3_MAP_QUEUES_QUEUE_SEL(0) | /* Queue_Sel */
176 			  PACKET3_MAP_QUEUES_VMID(0) | /* VMID */
177 			  PACKET3_MAP_QUEUES_QUEUE(ring->queue) |
178 			  PACKET3_MAP_QUEUES_PIPE(ring->pipe) |
179 			  PACKET3_MAP_QUEUES_ME((me)) |
180 			  PACKET3_MAP_QUEUES_QUEUE_TYPE(0) | /*queue_type: normal compute queue */
181 			  PACKET3_MAP_QUEUES_ALLOC_FORMAT(0) | /* alloc format: all_on_one_pipe */
182 			  PACKET3_MAP_QUEUES_ENGINE_SEL(eng_sel) |
183 			  PACKET3_MAP_QUEUES_NUM_QUEUES(1)); /* num_queues: must be 1 */
184 	amdgpu_ring_write(kiq_ring, PACKET3_MAP_QUEUES_DOORBELL_OFFSET(ring->doorbell_index));
185 	amdgpu_ring_write(kiq_ring, lower_32_bits(mqd_addr));
186 	amdgpu_ring_write(kiq_ring, upper_32_bits(mqd_addr));
187 	amdgpu_ring_write(kiq_ring, lower_32_bits(wptr_addr));
188 	amdgpu_ring_write(kiq_ring, upper_32_bits(wptr_addr));
189 }
190 
191 static void gfx11_kiq_unmap_queues(struct amdgpu_ring *kiq_ring,
192 				   struct amdgpu_ring *ring,
193 				   enum amdgpu_unmap_queues_action action,
194 				   u64 gpu_addr, u64 seq)
195 {
196 	struct amdgpu_device *adev = kiq_ring->adev;
197 	uint32_t eng_sel = ring->funcs->type == AMDGPU_RING_TYPE_GFX ? 4 : 0;
198 
199 	if (adev->enable_mes && !adev->gfx.kiq[0].ring.sched.ready) {
200 		amdgpu_mes_unmap_legacy_queue(adev, ring, action, gpu_addr, seq);
201 		return;
202 	}
203 
204 	amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_UNMAP_QUEUES, 4));
205 	amdgpu_ring_write(kiq_ring, /* Q_sel: 0, vmid: 0, engine: 0, num_Q: 1 */
206 			  PACKET3_UNMAP_QUEUES_ACTION(action) |
207 			  PACKET3_UNMAP_QUEUES_QUEUE_SEL(0) |
208 			  PACKET3_UNMAP_QUEUES_ENGINE_SEL(eng_sel) |
209 			  PACKET3_UNMAP_QUEUES_NUM_QUEUES(1));
210 	amdgpu_ring_write(kiq_ring,
211 		  PACKET3_UNMAP_QUEUES_DOORBELL_OFFSET0(ring->doorbell_index));
212 
213 	if (action == PREEMPT_QUEUES_NO_UNMAP) {
214 		amdgpu_ring_write(kiq_ring, lower_32_bits(gpu_addr));
215 		amdgpu_ring_write(kiq_ring, upper_32_bits(gpu_addr));
216 		amdgpu_ring_write(kiq_ring, seq);
217 	} else {
218 		amdgpu_ring_write(kiq_ring, 0);
219 		amdgpu_ring_write(kiq_ring, 0);
220 		amdgpu_ring_write(kiq_ring, 0);
221 	}
222 }
223 
224 static void gfx11_kiq_query_status(struct amdgpu_ring *kiq_ring,
225 				   struct amdgpu_ring *ring,
226 				   u64 addr,
227 				   u64 seq)
228 {
229 	uint32_t eng_sel = ring->funcs->type == AMDGPU_RING_TYPE_GFX ? 4 : 0;
230 
231 	amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_QUERY_STATUS, 5));
232 	amdgpu_ring_write(kiq_ring,
233 			  PACKET3_QUERY_STATUS_CONTEXT_ID(0) |
234 			  PACKET3_QUERY_STATUS_INTERRUPT_SEL(0) |
235 			  PACKET3_QUERY_STATUS_COMMAND(2));
236 	amdgpu_ring_write(kiq_ring, /* Q_sel: 0, vmid: 0, engine: 0, num_Q: 1 */
237 			  PACKET3_QUERY_STATUS_DOORBELL_OFFSET(ring->doorbell_index) |
238 			  PACKET3_QUERY_STATUS_ENG_SEL(eng_sel));
239 	amdgpu_ring_write(kiq_ring, lower_32_bits(addr));
240 	amdgpu_ring_write(kiq_ring, upper_32_bits(addr));
241 	amdgpu_ring_write(kiq_ring, lower_32_bits(seq));
242 	amdgpu_ring_write(kiq_ring, upper_32_bits(seq));
243 }
244 
245 static void gfx11_kiq_invalidate_tlbs(struct amdgpu_ring *kiq_ring,
246 				uint16_t pasid, uint32_t flush_type,
247 				bool all_hub)
248 {
249 	gfx_v11_0_ring_invalidate_tlbs(kiq_ring, pasid, flush_type, all_hub, 1);
250 }
251 
252 static const struct kiq_pm4_funcs gfx_v11_0_kiq_pm4_funcs = {
253 	.kiq_set_resources = gfx11_kiq_set_resources,
254 	.kiq_map_queues = gfx11_kiq_map_queues,
255 	.kiq_unmap_queues = gfx11_kiq_unmap_queues,
256 	.kiq_query_status = gfx11_kiq_query_status,
257 	.kiq_invalidate_tlbs = gfx11_kiq_invalidate_tlbs,
258 	.set_resources_size = 8,
259 	.map_queues_size = 7,
260 	.unmap_queues_size = 6,
261 	.query_status_size = 7,
262 	.invalidate_tlbs_size = 2,
263 };
264 
265 static void gfx_v11_0_set_kiq_pm4_funcs(struct amdgpu_device *adev)
266 {
267 	adev->gfx.kiq[0].pmf = &gfx_v11_0_kiq_pm4_funcs;
268 }
269 
270 static void gfx_v11_0_init_golden_registers(struct amdgpu_device *adev)
271 {
272 	switch (adev->ip_versions[GC_HWIP][0]) {
273 	case IP_VERSION(11, 0, 1):
274 	case IP_VERSION(11, 0, 4):
275 		soc15_program_register_sequence(adev,
276 						golden_settings_gc_11_0_1,
277 						(const u32)ARRAY_SIZE(golden_settings_gc_11_0_1));
278 		break;
279 	default:
280 		break;
281 	}
282 	soc15_program_register_sequence(adev,
283 					golden_settings_gc_11_0,
284 					(const u32)ARRAY_SIZE(golden_settings_gc_11_0));
285 
286 }
287 
288 static void gfx_v11_0_write_data_to_reg(struct amdgpu_ring *ring, int eng_sel,
289 				       bool wc, uint32_t reg, uint32_t val)
290 {
291 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
292 	amdgpu_ring_write(ring, WRITE_DATA_ENGINE_SEL(eng_sel) |
293 			  WRITE_DATA_DST_SEL(0) | (wc ? WR_CONFIRM : 0));
294 	amdgpu_ring_write(ring, reg);
295 	amdgpu_ring_write(ring, 0);
296 	amdgpu_ring_write(ring, val);
297 }
298 
299 static void gfx_v11_0_wait_reg_mem(struct amdgpu_ring *ring, int eng_sel,
300 				  int mem_space, int opt, uint32_t addr0,
301 				  uint32_t addr1, uint32_t ref, uint32_t mask,
302 				  uint32_t inv)
303 {
304 	amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
305 	amdgpu_ring_write(ring,
306 			  /* memory (1) or register (0) */
307 			  (WAIT_REG_MEM_MEM_SPACE(mem_space) |
308 			   WAIT_REG_MEM_OPERATION(opt) | /* wait */
309 			   WAIT_REG_MEM_FUNCTION(3) |  /* equal */
310 			   WAIT_REG_MEM_ENGINE(eng_sel)));
311 
312 	if (mem_space)
313 		BUG_ON(addr0 & 0x3); /* Dword align */
314 	amdgpu_ring_write(ring, addr0);
315 	amdgpu_ring_write(ring, addr1);
316 	amdgpu_ring_write(ring, ref);
317 	amdgpu_ring_write(ring, mask);
318 	amdgpu_ring_write(ring, inv); /* poll interval */
319 }
320 
321 static int gfx_v11_0_ring_test_ring(struct amdgpu_ring *ring)
322 {
323 	struct amdgpu_device *adev = ring->adev;
324 	uint32_t scratch = SOC15_REG_OFFSET(GC, 0, regSCRATCH_REG0);
325 	uint32_t tmp = 0;
326 	unsigned i;
327 	int r;
328 
329 	WREG32(scratch, 0xCAFEDEAD);
330 	r = amdgpu_ring_alloc(ring, 5);
331 	if (r) {
332 		DRM_ERROR("amdgpu: cp failed to lock ring %d (%d).\n",
333 			  ring->idx, r);
334 		return r;
335 	}
336 
337 	if (ring->funcs->type == AMDGPU_RING_TYPE_KIQ) {
338 		gfx_v11_0_ring_emit_wreg(ring, scratch, 0xDEADBEEF);
339 	} else {
340 		amdgpu_ring_write(ring, PACKET3(PACKET3_SET_UCONFIG_REG, 1));
341 		amdgpu_ring_write(ring, scratch -
342 				  PACKET3_SET_UCONFIG_REG_START);
343 		amdgpu_ring_write(ring, 0xDEADBEEF);
344 	}
345 	amdgpu_ring_commit(ring);
346 
347 	for (i = 0; i < adev->usec_timeout; i++) {
348 		tmp = RREG32(scratch);
349 		if (tmp == 0xDEADBEEF)
350 			break;
351 		if (amdgpu_emu_mode == 1)
352 			msleep(1);
353 		else
354 			udelay(1);
355 	}
356 
357 	if (i >= adev->usec_timeout)
358 		r = -ETIMEDOUT;
359 	return r;
360 }
361 
362 static int gfx_v11_0_ring_test_ib(struct amdgpu_ring *ring, long timeout)
363 {
364 	struct amdgpu_device *adev = ring->adev;
365 	struct amdgpu_ib ib;
366 	struct dma_fence *f = NULL;
367 	unsigned index;
368 	uint64_t gpu_addr;
369 	volatile uint32_t *cpu_ptr;
370 	long r;
371 
372 	/* MES KIQ fw hasn't indirect buffer support for now */
373 	if (adev->enable_mes_kiq &&
374 	    ring->funcs->type == AMDGPU_RING_TYPE_KIQ)
375 		return 0;
376 
377 	memset(&ib, 0, sizeof(ib));
378 
379 	if (ring->is_mes_queue) {
380 		uint32_t padding, offset;
381 
382 		offset = amdgpu_mes_ctx_get_offs(ring, AMDGPU_MES_CTX_IB_OFFS);
383 		padding = amdgpu_mes_ctx_get_offs(ring,
384 						  AMDGPU_MES_CTX_PADDING_OFFS);
385 
386 		ib.gpu_addr = amdgpu_mes_ctx_get_offs_gpu_addr(ring, offset);
387 		ib.ptr = amdgpu_mes_ctx_get_offs_cpu_addr(ring, offset);
388 
389 		gpu_addr = amdgpu_mes_ctx_get_offs_gpu_addr(ring, padding);
390 		cpu_ptr = amdgpu_mes_ctx_get_offs_cpu_addr(ring, padding);
391 		*cpu_ptr = cpu_to_le32(0xCAFEDEAD);
392 	} else {
393 		r = amdgpu_device_wb_get(adev, &index);
394 		if (r)
395 			return r;
396 
397 		gpu_addr = adev->wb.gpu_addr + (index * 4);
398 		adev->wb.wb[index] = cpu_to_le32(0xCAFEDEAD);
399 		cpu_ptr = &adev->wb.wb[index];
400 
401 		r = amdgpu_ib_get(adev, NULL, 20, AMDGPU_IB_POOL_DIRECT, &ib);
402 		if (r) {
403 			DRM_ERROR("amdgpu: failed to get ib (%ld).\n", r);
404 			goto err1;
405 		}
406 	}
407 
408 	ib.ptr[0] = PACKET3(PACKET3_WRITE_DATA, 3);
409 	ib.ptr[1] = WRITE_DATA_DST_SEL(5) | WR_CONFIRM;
410 	ib.ptr[2] = lower_32_bits(gpu_addr);
411 	ib.ptr[3] = upper_32_bits(gpu_addr);
412 	ib.ptr[4] = 0xDEADBEEF;
413 	ib.length_dw = 5;
414 
415 	r = amdgpu_ib_schedule(ring, 1, &ib, NULL, &f);
416 	if (r)
417 		goto err2;
418 
419 	r = dma_fence_wait_timeout(f, false, timeout);
420 	if (r == 0) {
421 		r = -ETIMEDOUT;
422 		goto err2;
423 	} else if (r < 0) {
424 		goto err2;
425 	}
426 
427 	if (le32_to_cpu(*cpu_ptr) == 0xDEADBEEF)
428 		r = 0;
429 	else
430 		r = -EINVAL;
431 err2:
432 	if (!ring->is_mes_queue)
433 		amdgpu_ib_free(adev, &ib, NULL);
434 	dma_fence_put(f);
435 err1:
436 	if (!ring->is_mes_queue)
437 		amdgpu_device_wb_free(adev, index);
438 	return r;
439 }
440 
441 static void gfx_v11_0_free_microcode(struct amdgpu_device *adev)
442 {
443 	amdgpu_ucode_release(&adev->gfx.pfp_fw);
444 	amdgpu_ucode_release(&adev->gfx.me_fw);
445 	amdgpu_ucode_release(&adev->gfx.rlc_fw);
446 	amdgpu_ucode_release(&adev->gfx.mec_fw);
447 
448 	kfree(adev->gfx.rlc.register_list_format);
449 }
450 
451 static int gfx_v11_0_init_toc_microcode(struct amdgpu_device *adev, const char *ucode_prefix)
452 {
453 	const struct psp_firmware_header_v1_0 *toc_hdr;
454 	int err = 0;
455 	char fw_name[40];
456 
457 	snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_toc.bin", ucode_prefix);
458 	err = amdgpu_ucode_request(adev, &adev->psp.toc_fw, fw_name);
459 	if (err)
460 		goto out;
461 
462 	toc_hdr = (const struct psp_firmware_header_v1_0 *)adev->psp.toc_fw->data;
463 	adev->psp.toc.fw_version = le32_to_cpu(toc_hdr->header.ucode_version);
464 	adev->psp.toc.feature_version = le32_to_cpu(toc_hdr->sos.fw_version);
465 	adev->psp.toc.size_bytes = le32_to_cpu(toc_hdr->header.ucode_size_bytes);
466 	adev->psp.toc.start_addr = (uint8_t *)toc_hdr +
467 				le32_to_cpu(toc_hdr->header.ucode_array_offset_bytes);
468 	return 0;
469 out:
470 	amdgpu_ucode_release(&adev->psp.toc_fw);
471 	return err;
472 }
473 
474 static void gfx_v11_0_check_fw_cp_gfx_shadow(struct amdgpu_device *adev)
475 {
476 	switch (adev->ip_versions[GC_HWIP][0]) {
477 	case IP_VERSION(11, 0, 0):
478 	case IP_VERSION(11, 0, 2):
479 	case IP_VERSION(11, 0, 3):
480 		if ((adev->gfx.me_fw_version >= 1505) &&
481 		    (adev->gfx.pfp_fw_version >= 1600) &&
482 		    (adev->gfx.mec_fw_version >= 512)) {
483 			if (amdgpu_sriov_vf(adev))
484 				adev->gfx.cp_gfx_shadow = true;
485 			else
486 				adev->gfx.cp_gfx_shadow = false;
487 		}
488 		break;
489 	default:
490 		adev->gfx.cp_gfx_shadow = false;
491 		break;
492 	}
493 }
494 
495 static int gfx_v11_0_init_microcode(struct amdgpu_device *adev)
496 {
497 	char fw_name[40];
498 	char ucode_prefix[30];
499 	int err;
500 	const struct rlc_firmware_header_v2_0 *rlc_hdr;
501 	uint16_t version_major;
502 	uint16_t version_minor;
503 
504 	DRM_DEBUG("\n");
505 
506 	amdgpu_ucode_ip_version_decode(adev, GC_HWIP, ucode_prefix, sizeof(ucode_prefix));
507 
508 	snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_pfp.bin", ucode_prefix);
509 	err = amdgpu_ucode_request(adev, &adev->gfx.pfp_fw, fw_name);
510 	if (err)
511 		goto out;
512 	/* check pfp fw hdr version to decide if enable rs64 for gfx11.*/
513 	adev->gfx.rs64_enable = amdgpu_ucode_hdr_version(
514 				(union amdgpu_firmware_header *)
515 				adev->gfx.pfp_fw->data, 2, 0);
516 	if (adev->gfx.rs64_enable) {
517 		dev_info(adev->dev, "CP RS64 enable\n");
518 		amdgpu_gfx_cp_init_microcode(adev, AMDGPU_UCODE_ID_CP_RS64_PFP);
519 		amdgpu_gfx_cp_init_microcode(adev, AMDGPU_UCODE_ID_CP_RS64_PFP_P0_STACK);
520 		amdgpu_gfx_cp_init_microcode(adev, AMDGPU_UCODE_ID_CP_RS64_PFP_P1_STACK);
521 	} else {
522 		amdgpu_gfx_cp_init_microcode(adev, AMDGPU_UCODE_ID_CP_PFP);
523 	}
524 
525 	snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_me.bin", ucode_prefix);
526 	err = amdgpu_ucode_request(adev, &adev->gfx.me_fw, fw_name);
527 	if (err)
528 		goto out;
529 	if (adev->gfx.rs64_enable) {
530 		amdgpu_gfx_cp_init_microcode(adev, AMDGPU_UCODE_ID_CP_RS64_ME);
531 		amdgpu_gfx_cp_init_microcode(adev, AMDGPU_UCODE_ID_CP_RS64_ME_P0_STACK);
532 		amdgpu_gfx_cp_init_microcode(adev, AMDGPU_UCODE_ID_CP_RS64_ME_P1_STACK);
533 	} else {
534 		amdgpu_gfx_cp_init_microcode(adev, AMDGPU_UCODE_ID_CP_ME);
535 	}
536 
537 	if (!amdgpu_sriov_vf(adev)) {
538 		snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_rlc.bin", ucode_prefix);
539 		err = amdgpu_ucode_request(adev, &adev->gfx.rlc_fw, fw_name);
540 		if (err)
541 			goto out;
542 		rlc_hdr = (const struct rlc_firmware_header_v2_0 *)adev->gfx.rlc_fw->data;
543 		version_major = le16_to_cpu(rlc_hdr->header.header_version_major);
544 		version_minor = le16_to_cpu(rlc_hdr->header.header_version_minor);
545 		err = amdgpu_gfx_rlc_init_microcode(adev, version_major, version_minor);
546 		if (err)
547 			goto out;
548 	}
549 
550 	snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec.bin", ucode_prefix);
551 	err = amdgpu_ucode_request(adev, &adev->gfx.mec_fw, fw_name);
552 	if (err)
553 		goto out;
554 	if (adev->gfx.rs64_enable) {
555 		amdgpu_gfx_cp_init_microcode(adev, AMDGPU_UCODE_ID_CP_RS64_MEC);
556 		amdgpu_gfx_cp_init_microcode(adev, AMDGPU_UCODE_ID_CP_RS64_MEC_P0_STACK);
557 		amdgpu_gfx_cp_init_microcode(adev, AMDGPU_UCODE_ID_CP_RS64_MEC_P1_STACK);
558 		amdgpu_gfx_cp_init_microcode(adev, AMDGPU_UCODE_ID_CP_RS64_MEC_P2_STACK);
559 		amdgpu_gfx_cp_init_microcode(adev, AMDGPU_UCODE_ID_CP_RS64_MEC_P3_STACK);
560 	} else {
561 		amdgpu_gfx_cp_init_microcode(adev, AMDGPU_UCODE_ID_CP_MEC1);
562 		amdgpu_gfx_cp_init_microcode(adev, AMDGPU_UCODE_ID_CP_MEC1_JT);
563 	}
564 
565 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_RLC_BACKDOOR_AUTO)
566 		err = gfx_v11_0_init_toc_microcode(adev, ucode_prefix);
567 
568 	/* only one MEC for gfx 11.0.0. */
569 	adev->gfx.mec2_fw = NULL;
570 
571 	gfx_v11_0_check_fw_cp_gfx_shadow(adev);
572 out:
573 	if (err) {
574 		amdgpu_ucode_release(&adev->gfx.pfp_fw);
575 		amdgpu_ucode_release(&adev->gfx.me_fw);
576 		amdgpu_ucode_release(&adev->gfx.rlc_fw);
577 		amdgpu_ucode_release(&adev->gfx.mec_fw);
578 	}
579 
580 	return err;
581 }
582 
583 static u32 gfx_v11_0_get_csb_size(struct amdgpu_device *adev)
584 {
585 	u32 count = 0;
586 	const struct cs_section_def *sect = NULL;
587 	const struct cs_extent_def *ext = NULL;
588 
589 	/* begin clear state */
590 	count += 2;
591 	/* context control state */
592 	count += 3;
593 
594 	for (sect = gfx11_cs_data; sect->section != NULL; ++sect) {
595 		for (ext = sect->section; ext->extent != NULL; ++ext) {
596 			if (sect->id == SECT_CONTEXT)
597 				count += 2 + ext->reg_count;
598 			else
599 				return 0;
600 		}
601 	}
602 
603 	/* set PA_SC_TILE_STEERING_OVERRIDE */
604 	count += 3;
605 	/* end clear state */
606 	count += 2;
607 	/* clear state */
608 	count += 2;
609 
610 	return count;
611 }
612 
613 static void gfx_v11_0_get_csb_buffer(struct amdgpu_device *adev,
614 				    volatile u32 *buffer)
615 {
616 	u32 count = 0, i;
617 	const struct cs_section_def *sect = NULL;
618 	const struct cs_extent_def *ext = NULL;
619 	int ctx_reg_offset;
620 
621 	if (adev->gfx.rlc.cs_data == NULL)
622 		return;
623 	if (buffer == NULL)
624 		return;
625 
626 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0));
627 	buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
628 
629 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CONTEXT_CONTROL, 1));
630 	buffer[count++] = cpu_to_le32(0x80000000);
631 	buffer[count++] = cpu_to_le32(0x80000000);
632 
633 	for (sect = adev->gfx.rlc.cs_data; sect->section != NULL; ++sect) {
634 		for (ext = sect->section; ext->extent != NULL; ++ext) {
635 			if (sect->id == SECT_CONTEXT) {
636 				buffer[count++] =
637 					cpu_to_le32(PACKET3(PACKET3_SET_CONTEXT_REG, ext->reg_count));
638 				buffer[count++] = cpu_to_le32(ext->reg_index -
639 						PACKET3_SET_CONTEXT_REG_START);
640 				for (i = 0; i < ext->reg_count; i++)
641 					buffer[count++] = cpu_to_le32(ext->extent[i]);
642 			} else {
643 				return;
644 			}
645 		}
646 	}
647 
648 	ctx_reg_offset =
649 		SOC15_REG_OFFSET(GC, 0, regPA_SC_TILE_STEERING_OVERRIDE) - PACKET3_SET_CONTEXT_REG_START;
650 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_SET_CONTEXT_REG, 1));
651 	buffer[count++] = cpu_to_le32(ctx_reg_offset);
652 	buffer[count++] = cpu_to_le32(adev->gfx.config.pa_sc_tile_steering_override);
653 
654 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0));
655 	buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_END_CLEAR_STATE);
656 
657 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CLEAR_STATE, 0));
658 	buffer[count++] = cpu_to_le32(0);
659 }
660 
661 static void gfx_v11_0_rlc_fini(struct amdgpu_device *adev)
662 {
663 	/* clear state block */
664 	amdgpu_bo_free_kernel(&adev->gfx.rlc.clear_state_obj,
665 			&adev->gfx.rlc.clear_state_gpu_addr,
666 			(void **)&adev->gfx.rlc.cs_ptr);
667 
668 	/* jump table block */
669 	amdgpu_bo_free_kernel(&adev->gfx.rlc.cp_table_obj,
670 			&adev->gfx.rlc.cp_table_gpu_addr,
671 			(void **)&adev->gfx.rlc.cp_table_ptr);
672 }
673 
674 static void gfx_v11_0_init_rlcg_reg_access_ctrl(struct amdgpu_device *adev)
675 {
676 	struct amdgpu_rlcg_reg_access_ctrl *reg_access_ctrl;
677 
678 	reg_access_ctrl = &adev->gfx.rlc.reg_access_ctrl[0];
679 	reg_access_ctrl->scratch_reg0 = SOC15_REG_OFFSET(GC, 0, regSCRATCH_REG0);
680 	reg_access_ctrl->scratch_reg1 = SOC15_REG_OFFSET(GC, 0, regSCRATCH_REG1);
681 	reg_access_ctrl->scratch_reg2 = SOC15_REG_OFFSET(GC, 0, regSCRATCH_REG2);
682 	reg_access_ctrl->scratch_reg3 = SOC15_REG_OFFSET(GC, 0, regSCRATCH_REG3);
683 	reg_access_ctrl->grbm_cntl = SOC15_REG_OFFSET(GC, 0, regGRBM_GFX_CNTL);
684 	reg_access_ctrl->grbm_idx = SOC15_REG_OFFSET(GC, 0, regGRBM_GFX_INDEX);
685 	reg_access_ctrl->spare_int = SOC15_REG_OFFSET(GC, 0, regRLC_SPARE_INT_0);
686 	adev->gfx.rlc.rlcg_reg_access_supported = true;
687 }
688 
689 static int gfx_v11_0_rlc_init(struct amdgpu_device *adev)
690 {
691 	const struct cs_section_def *cs_data;
692 	int r;
693 
694 	adev->gfx.rlc.cs_data = gfx11_cs_data;
695 
696 	cs_data = adev->gfx.rlc.cs_data;
697 
698 	if (cs_data) {
699 		/* init clear state block */
700 		r = amdgpu_gfx_rlc_init_csb(adev);
701 		if (r)
702 			return r;
703 	}
704 
705 	/* init spm vmid with 0xf */
706 	if (adev->gfx.rlc.funcs->update_spm_vmid)
707 		adev->gfx.rlc.funcs->update_spm_vmid(adev, 0xf);
708 
709 	return 0;
710 }
711 
712 static void gfx_v11_0_mec_fini(struct amdgpu_device *adev)
713 {
714 	amdgpu_bo_free_kernel(&adev->gfx.mec.hpd_eop_obj, NULL, NULL);
715 	amdgpu_bo_free_kernel(&adev->gfx.mec.mec_fw_obj, NULL, NULL);
716 	amdgpu_bo_free_kernel(&adev->gfx.mec.mec_fw_data_obj, NULL, NULL);
717 }
718 
719 static void gfx_v11_0_me_init(struct amdgpu_device *adev)
720 {
721 	bitmap_zero(adev->gfx.me.queue_bitmap, AMDGPU_MAX_GFX_QUEUES);
722 
723 	amdgpu_gfx_graphics_queue_acquire(adev);
724 }
725 
726 static int gfx_v11_0_mec_init(struct amdgpu_device *adev)
727 {
728 	int r;
729 	u32 *hpd;
730 	size_t mec_hpd_size;
731 
732 	bitmap_zero(adev->gfx.mec_bitmap[0].queue_bitmap, AMDGPU_MAX_COMPUTE_QUEUES);
733 
734 	/* take ownership of the relevant compute queues */
735 	amdgpu_gfx_compute_queue_acquire(adev);
736 	mec_hpd_size = adev->gfx.num_compute_rings * GFX11_MEC_HPD_SIZE;
737 
738 	if (mec_hpd_size) {
739 		r = amdgpu_bo_create_reserved(adev, mec_hpd_size, PAGE_SIZE,
740 					      AMDGPU_GEM_DOMAIN_GTT,
741 					      &adev->gfx.mec.hpd_eop_obj,
742 					      &adev->gfx.mec.hpd_eop_gpu_addr,
743 					      (void **)&hpd);
744 		if (r) {
745 			dev_warn(adev->dev, "(%d) create HDP EOP bo failed\n", r);
746 			gfx_v11_0_mec_fini(adev);
747 			return r;
748 		}
749 
750 		memset(hpd, 0, mec_hpd_size);
751 
752 		amdgpu_bo_kunmap(adev->gfx.mec.hpd_eop_obj);
753 		amdgpu_bo_unreserve(adev->gfx.mec.hpd_eop_obj);
754 	}
755 
756 	return 0;
757 }
758 
759 static uint32_t wave_read_ind(struct amdgpu_device *adev, uint32_t wave, uint32_t address)
760 {
761 	WREG32_SOC15(GC, 0, regSQ_IND_INDEX,
762 		(wave << SQ_IND_INDEX__WAVE_ID__SHIFT) |
763 		(address << SQ_IND_INDEX__INDEX__SHIFT));
764 	return RREG32_SOC15(GC, 0, regSQ_IND_DATA);
765 }
766 
767 static void wave_read_regs(struct amdgpu_device *adev, uint32_t wave,
768 			   uint32_t thread, uint32_t regno,
769 			   uint32_t num, uint32_t *out)
770 {
771 	WREG32_SOC15(GC, 0, regSQ_IND_INDEX,
772 		(wave << SQ_IND_INDEX__WAVE_ID__SHIFT) |
773 		(regno << SQ_IND_INDEX__INDEX__SHIFT) |
774 		(thread << SQ_IND_INDEX__WORKITEM_ID__SHIFT) |
775 		(SQ_IND_INDEX__AUTO_INCR_MASK));
776 	while (num--)
777 		*(out++) = RREG32_SOC15(GC, 0, regSQ_IND_DATA);
778 }
779 
780 static void gfx_v11_0_read_wave_data(struct amdgpu_device *adev, uint32_t xcc_id, uint32_t simd, uint32_t wave, uint32_t *dst, int *no_fields)
781 {
782 	/* in gfx11 the SIMD_ID is specified as part of the INSTANCE
783 	 * field when performing a select_se_sh so it should be
784 	 * zero here */
785 	WARN_ON(simd != 0);
786 
787 	/* type 3 wave data */
788 	dst[(*no_fields)++] = 3;
789 	dst[(*no_fields)++] = wave_read_ind(adev, wave, ixSQ_WAVE_STATUS);
790 	dst[(*no_fields)++] = wave_read_ind(adev, wave, ixSQ_WAVE_PC_LO);
791 	dst[(*no_fields)++] = wave_read_ind(adev, wave, ixSQ_WAVE_PC_HI);
792 	dst[(*no_fields)++] = wave_read_ind(adev, wave, ixSQ_WAVE_EXEC_LO);
793 	dst[(*no_fields)++] = wave_read_ind(adev, wave, ixSQ_WAVE_EXEC_HI);
794 	dst[(*no_fields)++] = wave_read_ind(adev, wave, ixSQ_WAVE_HW_ID1);
795 	dst[(*no_fields)++] = wave_read_ind(adev, wave, ixSQ_WAVE_HW_ID2);
796 	dst[(*no_fields)++] = wave_read_ind(adev, wave, ixSQ_WAVE_GPR_ALLOC);
797 	dst[(*no_fields)++] = wave_read_ind(adev, wave, ixSQ_WAVE_LDS_ALLOC);
798 	dst[(*no_fields)++] = wave_read_ind(adev, wave, ixSQ_WAVE_TRAPSTS);
799 	dst[(*no_fields)++] = wave_read_ind(adev, wave, ixSQ_WAVE_IB_STS);
800 	dst[(*no_fields)++] = wave_read_ind(adev, wave, ixSQ_WAVE_IB_STS2);
801 	dst[(*no_fields)++] = wave_read_ind(adev, wave, ixSQ_WAVE_IB_DBG1);
802 	dst[(*no_fields)++] = wave_read_ind(adev, wave, ixSQ_WAVE_M0);
803 	dst[(*no_fields)++] = wave_read_ind(adev, wave, ixSQ_WAVE_MODE);
804 }
805 
806 static void gfx_v11_0_read_wave_sgprs(struct amdgpu_device *adev, uint32_t xcc_id, uint32_t simd,
807 				     uint32_t wave, uint32_t start,
808 				     uint32_t size, uint32_t *dst)
809 {
810 	WARN_ON(simd != 0);
811 
812 	wave_read_regs(
813 		adev, wave, 0, start + SQIND_WAVE_SGPRS_OFFSET, size,
814 		dst);
815 }
816 
817 static void gfx_v11_0_read_wave_vgprs(struct amdgpu_device *adev, uint32_t xcc_id, uint32_t simd,
818 				      uint32_t wave, uint32_t thread,
819 				      uint32_t start, uint32_t size,
820 				      uint32_t *dst)
821 {
822 	wave_read_regs(
823 		adev, wave, thread,
824 		start + SQIND_WAVE_VGPRS_OFFSET, size, dst);
825 }
826 
827 static void gfx_v11_0_select_me_pipe_q(struct amdgpu_device *adev,
828 					u32 me, u32 pipe, u32 q, u32 vm, u32 xcc_id)
829 {
830 	soc21_grbm_select(adev, me, pipe, q, vm);
831 }
832 
833 /* all sizes are in bytes */
834 #define MQD_SHADOW_BASE_SIZE      73728
835 #define MQD_SHADOW_BASE_ALIGNMENT 256
836 #define MQD_FWWORKAREA_SIZE       484
837 #define MQD_FWWORKAREA_ALIGNMENT  256
838 
839 static int gfx_v11_0_get_gfx_shadow_info(struct amdgpu_device *adev,
840 					 struct amdgpu_gfx_shadow_info *shadow_info)
841 {
842 	if (adev->gfx.cp_gfx_shadow) {
843 		shadow_info->shadow_size = MQD_SHADOW_BASE_SIZE;
844 		shadow_info->shadow_alignment = MQD_SHADOW_BASE_ALIGNMENT;
845 		shadow_info->csa_size = MQD_FWWORKAREA_SIZE;
846 		shadow_info->csa_alignment = MQD_FWWORKAREA_ALIGNMENT;
847 		return 0;
848 	} else {
849 		memset(shadow_info, 0, sizeof(struct amdgpu_gfx_shadow_info));
850 		return -ENOTSUPP;
851 	}
852 }
853 
854 static const struct amdgpu_gfx_funcs gfx_v11_0_gfx_funcs = {
855 	.get_gpu_clock_counter = &gfx_v11_0_get_gpu_clock_counter,
856 	.select_se_sh = &gfx_v11_0_select_se_sh,
857 	.read_wave_data = &gfx_v11_0_read_wave_data,
858 	.read_wave_sgprs = &gfx_v11_0_read_wave_sgprs,
859 	.read_wave_vgprs = &gfx_v11_0_read_wave_vgprs,
860 	.select_me_pipe_q = &gfx_v11_0_select_me_pipe_q,
861 	.update_perfmon_mgcg = &gfx_v11_0_update_perf_clk,
862 	.get_gfx_shadow_info = &gfx_v11_0_get_gfx_shadow_info,
863 };
864 
865 static int gfx_v11_0_gpu_early_init(struct amdgpu_device *adev)
866 {
867 
868 	switch (adev->ip_versions[GC_HWIP][0]) {
869 	case IP_VERSION(11, 0, 0):
870 	case IP_VERSION(11, 0, 2):
871 		adev->gfx.config.max_hw_contexts = 8;
872 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
873 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
874 		adev->gfx.config.sc_hiz_tile_fifo_size = 0;
875 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x4C0;
876 		break;
877 	case IP_VERSION(11, 0, 3):
878 		adev->gfx.ras = &gfx_v11_0_3_ras;
879 		adev->gfx.config.max_hw_contexts = 8;
880 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
881 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
882 		adev->gfx.config.sc_hiz_tile_fifo_size = 0;
883 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x4C0;
884 		break;
885 	case IP_VERSION(11, 0, 1):
886 	case IP_VERSION(11, 0, 4):
887 		adev->gfx.config.max_hw_contexts = 8;
888 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
889 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
890 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x80;
891 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x300;
892 		break;
893 	default:
894 		BUG();
895 		break;
896 	}
897 
898 	return 0;
899 }
900 
901 static int gfx_v11_0_gfx_ring_init(struct amdgpu_device *adev, int ring_id,
902 				   int me, int pipe, int queue)
903 {
904 	int r;
905 	struct amdgpu_ring *ring;
906 	unsigned int irq_type;
907 
908 	ring = &adev->gfx.gfx_ring[ring_id];
909 
910 	ring->me = me;
911 	ring->pipe = pipe;
912 	ring->queue = queue;
913 
914 	ring->ring_obj = NULL;
915 	ring->use_doorbell = true;
916 
917 	if (!ring_id)
918 		ring->doorbell_index = adev->doorbell_index.gfx_ring0 << 1;
919 	else
920 		ring->doorbell_index = adev->doorbell_index.gfx_ring1 << 1;
921 	ring->vm_hub = AMDGPU_GFXHUB(0);
922 	sprintf(ring->name, "gfx_%d.%d.%d", ring->me, ring->pipe, ring->queue);
923 
924 	irq_type = AMDGPU_CP_IRQ_GFX_ME0_PIPE0_EOP + ring->pipe;
925 	r = amdgpu_ring_init(adev, ring, 1024, &adev->gfx.eop_irq, irq_type,
926 			     AMDGPU_RING_PRIO_DEFAULT, NULL);
927 	if (r)
928 		return r;
929 	return 0;
930 }
931 
932 static int gfx_v11_0_compute_ring_init(struct amdgpu_device *adev, int ring_id,
933 				       int mec, int pipe, int queue)
934 {
935 	int r;
936 	unsigned irq_type;
937 	struct amdgpu_ring *ring;
938 	unsigned int hw_prio;
939 
940 	ring = &adev->gfx.compute_ring[ring_id];
941 
942 	/* mec0 is me1 */
943 	ring->me = mec + 1;
944 	ring->pipe = pipe;
945 	ring->queue = queue;
946 
947 	ring->ring_obj = NULL;
948 	ring->use_doorbell = true;
949 	ring->doorbell_index = (adev->doorbell_index.mec_ring0 + ring_id) << 1;
950 	ring->eop_gpu_addr = adev->gfx.mec.hpd_eop_gpu_addr
951 				+ (ring_id * GFX11_MEC_HPD_SIZE);
952 	ring->vm_hub = AMDGPU_GFXHUB(0);
953 	sprintf(ring->name, "comp_%d.%d.%d", ring->me, ring->pipe, ring->queue);
954 
955 	irq_type = AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP
956 		+ ((ring->me - 1) * adev->gfx.mec.num_pipe_per_mec)
957 		+ ring->pipe;
958 	hw_prio = amdgpu_gfx_is_high_priority_compute_queue(adev, ring) ?
959 			AMDGPU_GFX_PIPE_PRIO_HIGH : AMDGPU_GFX_PIPE_PRIO_NORMAL;
960 	/* type-2 packets are deprecated on MEC, use type-3 instead */
961 	r = amdgpu_ring_init(adev, ring, 1024, &adev->gfx.eop_irq, irq_type,
962 			     hw_prio, NULL);
963 	if (r)
964 		return r;
965 
966 	return 0;
967 }
968 
969 static struct {
970 	SOC21_FIRMWARE_ID	id;
971 	unsigned int		offset;
972 	unsigned int		size;
973 } rlc_autoload_info[SOC21_FIRMWARE_ID_MAX];
974 
975 static void gfx_v11_0_parse_rlc_toc(struct amdgpu_device *adev, void *rlc_toc)
976 {
977 	RLC_TABLE_OF_CONTENT *ucode = rlc_toc;
978 
979 	while (ucode && (ucode->id > SOC21_FIRMWARE_ID_INVALID) &&
980 			(ucode->id < SOC21_FIRMWARE_ID_MAX)) {
981 		rlc_autoload_info[ucode->id].id = ucode->id;
982 		rlc_autoload_info[ucode->id].offset = ucode->offset * 4;
983 		rlc_autoload_info[ucode->id].size = ucode->size * 4;
984 
985 		ucode++;
986 	}
987 }
988 
989 static uint32_t gfx_v11_0_calc_toc_total_size(struct amdgpu_device *adev)
990 {
991 	uint32_t total_size = 0;
992 	SOC21_FIRMWARE_ID id;
993 
994 	gfx_v11_0_parse_rlc_toc(adev, adev->psp.toc.start_addr);
995 
996 	for (id = SOC21_FIRMWARE_ID_RLC_G_UCODE; id < SOC21_FIRMWARE_ID_MAX; id++)
997 		total_size += rlc_autoload_info[id].size;
998 
999 	/* In case the offset in rlc toc ucode is aligned */
1000 	if (total_size < rlc_autoload_info[SOC21_FIRMWARE_ID_MAX-1].offset)
1001 		total_size = rlc_autoload_info[SOC21_FIRMWARE_ID_MAX-1].offset +
1002 			rlc_autoload_info[SOC21_FIRMWARE_ID_MAX-1].size;
1003 
1004 	return total_size;
1005 }
1006 
1007 static int gfx_v11_0_rlc_autoload_buffer_init(struct amdgpu_device *adev)
1008 {
1009 	int r;
1010 	uint32_t total_size;
1011 
1012 	total_size = gfx_v11_0_calc_toc_total_size(adev);
1013 
1014 	r = amdgpu_bo_create_reserved(adev, total_size, 64 * 1024,
1015 				      AMDGPU_GEM_DOMAIN_VRAM |
1016 				      AMDGPU_GEM_DOMAIN_GTT,
1017 				      &adev->gfx.rlc.rlc_autoload_bo,
1018 				      &adev->gfx.rlc.rlc_autoload_gpu_addr,
1019 				      (void **)&adev->gfx.rlc.rlc_autoload_ptr);
1020 
1021 	if (r) {
1022 		dev_err(adev->dev, "(%d) failed to create fw autoload bo\n", r);
1023 		return r;
1024 	}
1025 
1026 	return 0;
1027 }
1028 
1029 static void gfx_v11_0_rlc_backdoor_autoload_copy_ucode(struct amdgpu_device *adev,
1030 					      SOC21_FIRMWARE_ID id,
1031 			    		      const void *fw_data,
1032 					      uint32_t fw_size,
1033 					      uint32_t *fw_autoload_mask)
1034 {
1035 	uint32_t toc_offset;
1036 	uint32_t toc_fw_size;
1037 	char *ptr = adev->gfx.rlc.rlc_autoload_ptr;
1038 
1039 	if (id <= SOC21_FIRMWARE_ID_INVALID || id >= SOC21_FIRMWARE_ID_MAX)
1040 		return;
1041 
1042 	toc_offset = rlc_autoload_info[id].offset;
1043 	toc_fw_size = rlc_autoload_info[id].size;
1044 
1045 	if (fw_size == 0)
1046 		fw_size = toc_fw_size;
1047 
1048 	if (fw_size > toc_fw_size)
1049 		fw_size = toc_fw_size;
1050 
1051 	memcpy(ptr + toc_offset, fw_data, fw_size);
1052 
1053 	if (fw_size < toc_fw_size)
1054 		memset(ptr + toc_offset + fw_size, 0, toc_fw_size - fw_size);
1055 
1056 	if ((id != SOC21_FIRMWARE_ID_RS64_PFP) && (id != SOC21_FIRMWARE_ID_RS64_ME))
1057 		*(uint64_t *)fw_autoload_mask |= 1ULL << id;
1058 }
1059 
1060 static void gfx_v11_0_rlc_backdoor_autoload_copy_toc_ucode(struct amdgpu_device *adev,
1061 							uint32_t *fw_autoload_mask)
1062 {
1063 	void *data;
1064 	uint32_t size;
1065 	uint64_t *toc_ptr;
1066 
1067 	*(uint64_t *)fw_autoload_mask |= 0x1;
1068 
1069 	DRM_DEBUG("rlc autoload enabled fw: 0x%llx\n", *(uint64_t *)fw_autoload_mask);
1070 
1071 	data = adev->psp.toc.start_addr;
1072 	size = rlc_autoload_info[SOC21_FIRMWARE_ID_RLC_TOC].size;
1073 
1074 	toc_ptr = (uint64_t *)data + size / 8 - 1;
1075 	*toc_ptr = *(uint64_t *)fw_autoload_mask;
1076 
1077 	gfx_v11_0_rlc_backdoor_autoload_copy_ucode(adev, SOC21_FIRMWARE_ID_RLC_TOC,
1078 					data, size, fw_autoload_mask);
1079 }
1080 
1081 static void gfx_v11_0_rlc_backdoor_autoload_copy_gfx_ucode(struct amdgpu_device *adev,
1082 							uint32_t *fw_autoload_mask)
1083 {
1084 	const __le32 *fw_data;
1085 	uint32_t fw_size;
1086 	const struct gfx_firmware_header_v1_0 *cp_hdr;
1087 	const struct gfx_firmware_header_v2_0 *cpv2_hdr;
1088 	const struct rlc_firmware_header_v2_0 *rlc_hdr;
1089 	const struct rlc_firmware_header_v2_2 *rlcv22_hdr;
1090 	uint16_t version_major, version_minor;
1091 
1092 	if (adev->gfx.rs64_enable) {
1093 		/* pfp ucode */
1094 		cpv2_hdr = (const struct gfx_firmware_header_v2_0 *)
1095 			adev->gfx.pfp_fw->data;
1096 		/* instruction */
1097 		fw_data = (const __le32 *)(adev->gfx.pfp_fw->data +
1098 			le32_to_cpu(cpv2_hdr->ucode_offset_bytes));
1099 		fw_size = le32_to_cpu(cpv2_hdr->ucode_size_bytes);
1100 		gfx_v11_0_rlc_backdoor_autoload_copy_ucode(adev, SOC21_FIRMWARE_ID_RS64_PFP,
1101 						fw_data, fw_size, fw_autoload_mask);
1102 		/* data */
1103 		fw_data = (const __le32 *)(adev->gfx.pfp_fw->data +
1104 			le32_to_cpu(cpv2_hdr->data_offset_bytes));
1105 		fw_size = le32_to_cpu(cpv2_hdr->data_size_bytes);
1106 		gfx_v11_0_rlc_backdoor_autoload_copy_ucode(adev, SOC21_FIRMWARE_ID_RS64_PFP_P0_STACK,
1107 						fw_data, fw_size, fw_autoload_mask);
1108 		gfx_v11_0_rlc_backdoor_autoload_copy_ucode(adev, SOC21_FIRMWARE_ID_RS64_PFP_P1_STACK,
1109 						fw_data, fw_size, fw_autoload_mask);
1110 		/* me ucode */
1111 		cpv2_hdr = (const struct gfx_firmware_header_v2_0 *)
1112 			adev->gfx.me_fw->data;
1113 		/* instruction */
1114 		fw_data = (const __le32 *)(adev->gfx.me_fw->data +
1115 			le32_to_cpu(cpv2_hdr->ucode_offset_bytes));
1116 		fw_size = le32_to_cpu(cpv2_hdr->ucode_size_bytes);
1117 		gfx_v11_0_rlc_backdoor_autoload_copy_ucode(adev, SOC21_FIRMWARE_ID_RS64_ME,
1118 						fw_data, fw_size, fw_autoload_mask);
1119 		/* data */
1120 		fw_data = (const __le32 *)(adev->gfx.me_fw->data +
1121 			le32_to_cpu(cpv2_hdr->data_offset_bytes));
1122 		fw_size = le32_to_cpu(cpv2_hdr->data_size_bytes);
1123 		gfx_v11_0_rlc_backdoor_autoload_copy_ucode(adev, SOC21_FIRMWARE_ID_RS64_ME_P0_STACK,
1124 						fw_data, fw_size, fw_autoload_mask);
1125 		gfx_v11_0_rlc_backdoor_autoload_copy_ucode(adev, SOC21_FIRMWARE_ID_RS64_ME_P1_STACK,
1126 						fw_data, fw_size, fw_autoload_mask);
1127 		/* mec ucode */
1128 		cpv2_hdr = (const struct gfx_firmware_header_v2_0 *)
1129 			adev->gfx.mec_fw->data;
1130 		/* instruction */
1131 		fw_data = (const __le32 *) (adev->gfx.mec_fw->data +
1132 			le32_to_cpu(cpv2_hdr->ucode_offset_bytes));
1133 		fw_size = le32_to_cpu(cpv2_hdr->ucode_size_bytes);
1134 		gfx_v11_0_rlc_backdoor_autoload_copy_ucode(adev, SOC21_FIRMWARE_ID_RS64_MEC,
1135 						fw_data, fw_size, fw_autoload_mask);
1136 		/* data */
1137 		fw_data = (const __le32 *) (adev->gfx.mec_fw->data +
1138 			le32_to_cpu(cpv2_hdr->data_offset_bytes));
1139 		fw_size = le32_to_cpu(cpv2_hdr->data_size_bytes);
1140 		gfx_v11_0_rlc_backdoor_autoload_copy_ucode(adev, SOC21_FIRMWARE_ID_RS64_MEC_P0_STACK,
1141 						fw_data, fw_size, fw_autoload_mask);
1142 		gfx_v11_0_rlc_backdoor_autoload_copy_ucode(adev, SOC21_FIRMWARE_ID_RS64_MEC_P1_STACK,
1143 						fw_data, fw_size, fw_autoload_mask);
1144 		gfx_v11_0_rlc_backdoor_autoload_copy_ucode(adev, SOC21_FIRMWARE_ID_RS64_MEC_P2_STACK,
1145 						fw_data, fw_size, fw_autoload_mask);
1146 		gfx_v11_0_rlc_backdoor_autoload_copy_ucode(adev, SOC21_FIRMWARE_ID_RS64_MEC_P3_STACK,
1147 						fw_data, fw_size, fw_autoload_mask);
1148 	} else {
1149 		/* pfp ucode */
1150 		cp_hdr = (const struct gfx_firmware_header_v1_0 *)
1151 			adev->gfx.pfp_fw->data;
1152 		fw_data = (const __le32 *)(adev->gfx.pfp_fw->data +
1153 				le32_to_cpu(cp_hdr->header.ucode_array_offset_bytes));
1154 		fw_size = le32_to_cpu(cp_hdr->header.ucode_size_bytes);
1155 		gfx_v11_0_rlc_backdoor_autoload_copy_ucode(adev, SOC21_FIRMWARE_ID_CP_PFP,
1156 						fw_data, fw_size, fw_autoload_mask);
1157 
1158 		/* me ucode */
1159 		cp_hdr = (const struct gfx_firmware_header_v1_0 *)
1160 			adev->gfx.me_fw->data;
1161 		fw_data = (const __le32 *)(adev->gfx.me_fw->data +
1162 				le32_to_cpu(cp_hdr->header.ucode_array_offset_bytes));
1163 		fw_size = le32_to_cpu(cp_hdr->header.ucode_size_bytes);
1164 		gfx_v11_0_rlc_backdoor_autoload_copy_ucode(adev, SOC21_FIRMWARE_ID_CP_ME,
1165 						fw_data, fw_size, fw_autoload_mask);
1166 
1167 		/* mec ucode */
1168 		cp_hdr = (const struct gfx_firmware_header_v1_0 *)
1169 			adev->gfx.mec_fw->data;
1170 		fw_data = (const __le32 *) (adev->gfx.mec_fw->data +
1171 				le32_to_cpu(cp_hdr->header.ucode_array_offset_bytes));
1172 		fw_size = le32_to_cpu(cp_hdr->header.ucode_size_bytes) -
1173 			cp_hdr->jt_size * 4;
1174 		gfx_v11_0_rlc_backdoor_autoload_copy_ucode(adev, SOC21_FIRMWARE_ID_CP_MEC,
1175 						fw_data, fw_size, fw_autoload_mask);
1176 	}
1177 
1178 	/* rlc ucode */
1179 	rlc_hdr = (const struct rlc_firmware_header_v2_0 *)
1180 		adev->gfx.rlc_fw->data;
1181 	fw_data = (const __le32 *)(adev->gfx.rlc_fw->data +
1182 			le32_to_cpu(rlc_hdr->header.ucode_array_offset_bytes));
1183 	fw_size = le32_to_cpu(rlc_hdr->header.ucode_size_bytes);
1184 	gfx_v11_0_rlc_backdoor_autoload_copy_ucode(adev, SOC21_FIRMWARE_ID_RLC_G_UCODE,
1185 					fw_data, fw_size, fw_autoload_mask);
1186 
1187 	version_major = le16_to_cpu(rlc_hdr->header.header_version_major);
1188 	version_minor = le16_to_cpu(rlc_hdr->header.header_version_minor);
1189 	if (version_major == 2) {
1190 		if (version_minor >= 2) {
1191 			rlcv22_hdr = (const struct rlc_firmware_header_v2_2 *)adev->gfx.rlc_fw->data;
1192 
1193 			fw_data = (const __le32 *)(adev->gfx.rlc_fw->data +
1194 					le32_to_cpu(rlcv22_hdr->rlc_iram_ucode_offset_bytes));
1195 			fw_size = le32_to_cpu(rlcv22_hdr->rlc_iram_ucode_size_bytes);
1196 			gfx_v11_0_rlc_backdoor_autoload_copy_ucode(adev, SOC21_FIRMWARE_ID_RLX6_UCODE,
1197 					fw_data, fw_size, fw_autoload_mask);
1198 
1199 			fw_data = (const __le32 *)(adev->gfx.rlc_fw->data +
1200 					le32_to_cpu(rlcv22_hdr->rlc_dram_ucode_offset_bytes));
1201 			fw_size = le32_to_cpu(rlcv22_hdr->rlc_dram_ucode_size_bytes);
1202 			gfx_v11_0_rlc_backdoor_autoload_copy_ucode(adev, SOC21_FIRMWARE_ID_RLX6_DRAM_BOOT,
1203 					fw_data, fw_size, fw_autoload_mask);
1204 		}
1205 	}
1206 }
1207 
1208 static void gfx_v11_0_rlc_backdoor_autoload_copy_sdma_ucode(struct amdgpu_device *adev,
1209 							uint32_t *fw_autoload_mask)
1210 {
1211 	const __le32 *fw_data;
1212 	uint32_t fw_size;
1213 	const struct sdma_firmware_header_v2_0 *sdma_hdr;
1214 
1215 	sdma_hdr = (const struct sdma_firmware_header_v2_0 *)
1216 		adev->sdma.instance[0].fw->data;
1217 	fw_data = (const __le32 *) (adev->sdma.instance[0].fw->data +
1218 			le32_to_cpu(sdma_hdr->header.ucode_array_offset_bytes));
1219 	fw_size = le32_to_cpu(sdma_hdr->ctx_ucode_size_bytes);
1220 
1221 	gfx_v11_0_rlc_backdoor_autoload_copy_ucode(adev,
1222 			SOC21_FIRMWARE_ID_SDMA_UCODE_TH0, fw_data, fw_size, fw_autoload_mask);
1223 
1224 	fw_data = (const __le32 *) (adev->sdma.instance[0].fw->data +
1225 			le32_to_cpu(sdma_hdr->ctl_ucode_offset));
1226 	fw_size = le32_to_cpu(sdma_hdr->ctl_ucode_size_bytes);
1227 
1228 	gfx_v11_0_rlc_backdoor_autoload_copy_ucode(adev,
1229 			SOC21_FIRMWARE_ID_SDMA_UCODE_TH1, fw_data, fw_size, fw_autoload_mask);
1230 }
1231 
1232 static void gfx_v11_0_rlc_backdoor_autoload_copy_mes_ucode(struct amdgpu_device *adev,
1233 							uint32_t *fw_autoload_mask)
1234 {
1235 	const __le32 *fw_data;
1236 	unsigned fw_size;
1237 	const struct mes_firmware_header_v1_0 *mes_hdr;
1238 	int pipe, ucode_id, data_id;
1239 
1240 	for (pipe = 0; pipe < 2; pipe++) {
1241 		if (pipe==0) {
1242 			ucode_id = SOC21_FIRMWARE_ID_RS64_MES_P0;
1243 			data_id  = SOC21_FIRMWARE_ID_RS64_MES_P0_STACK;
1244 		} else {
1245 			ucode_id = SOC21_FIRMWARE_ID_RS64_MES_P1;
1246 			data_id  = SOC21_FIRMWARE_ID_RS64_MES_P1_STACK;
1247 		}
1248 
1249 		mes_hdr = (const struct mes_firmware_header_v1_0 *)
1250 			adev->mes.fw[pipe]->data;
1251 
1252 		fw_data = (const __le32 *)(adev->mes.fw[pipe]->data +
1253 				le32_to_cpu(mes_hdr->mes_ucode_offset_bytes));
1254 		fw_size = le32_to_cpu(mes_hdr->mes_ucode_size_bytes);
1255 
1256 		gfx_v11_0_rlc_backdoor_autoload_copy_ucode(adev,
1257 				ucode_id, fw_data, fw_size, fw_autoload_mask);
1258 
1259 		fw_data = (const __le32 *)(adev->mes.fw[pipe]->data +
1260 				le32_to_cpu(mes_hdr->mes_ucode_data_offset_bytes));
1261 		fw_size = le32_to_cpu(mes_hdr->mes_ucode_data_size_bytes);
1262 
1263 		gfx_v11_0_rlc_backdoor_autoload_copy_ucode(adev,
1264 				data_id, fw_data, fw_size, fw_autoload_mask);
1265 	}
1266 }
1267 
1268 static int gfx_v11_0_rlc_backdoor_autoload_enable(struct amdgpu_device *adev)
1269 {
1270 	uint32_t rlc_g_offset, rlc_g_size;
1271 	uint64_t gpu_addr;
1272 	uint32_t autoload_fw_id[2];
1273 
1274 	memset(autoload_fw_id, 0, sizeof(uint32_t) * 2);
1275 
1276 	/* RLC autoload sequence 2: copy ucode */
1277 	gfx_v11_0_rlc_backdoor_autoload_copy_sdma_ucode(adev, autoload_fw_id);
1278 	gfx_v11_0_rlc_backdoor_autoload_copy_gfx_ucode(adev, autoload_fw_id);
1279 	gfx_v11_0_rlc_backdoor_autoload_copy_mes_ucode(adev, autoload_fw_id);
1280 	gfx_v11_0_rlc_backdoor_autoload_copy_toc_ucode(adev, autoload_fw_id);
1281 
1282 	rlc_g_offset = rlc_autoload_info[SOC21_FIRMWARE_ID_RLC_G_UCODE].offset;
1283 	rlc_g_size = rlc_autoload_info[SOC21_FIRMWARE_ID_RLC_G_UCODE].size;
1284 	gpu_addr = adev->gfx.rlc.rlc_autoload_gpu_addr + rlc_g_offset;
1285 
1286 	WREG32_SOC15(GC, 0, regGFX_IMU_RLC_BOOTLOADER_ADDR_HI, upper_32_bits(gpu_addr));
1287 	WREG32_SOC15(GC, 0, regGFX_IMU_RLC_BOOTLOADER_ADDR_LO, lower_32_bits(gpu_addr));
1288 
1289 	WREG32_SOC15(GC, 0, regGFX_IMU_RLC_BOOTLOADER_SIZE, rlc_g_size);
1290 
1291 	/* RLC autoload sequence 3: load IMU fw */
1292 	if (adev->gfx.imu.funcs->load_microcode)
1293 		adev->gfx.imu.funcs->load_microcode(adev);
1294 	/* RLC autoload sequence 4 init IMU fw */
1295 	if (adev->gfx.imu.funcs->setup_imu)
1296 		adev->gfx.imu.funcs->setup_imu(adev);
1297 	if (adev->gfx.imu.funcs->start_imu)
1298 		adev->gfx.imu.funcs->start_imu(adev);
1299 
1300 	/* RLC autoload sequence 5 disable gpa mode */
1301 	gfx_v11_0_disable_gpa_mode(adev);
1302 
1303 	return 0;
1304 }
1305 
1306 static int gfx_v11_0_sw_init(void *handle)
1307 {
1308 	int i, j, k, r, ring_id = 0;
1309 	struct amdgpu_kiq *kiq;
1310 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1311 
1312 	adev->gfxhub.funcs->init(adev);
1313 
1314 	switch (adev->ip_versions[GC_HWIP][0]) {
1315 	case IP_VERSION(11, 0, 0):
1316 	case IP_VERSION(11, 0, 2):
1317 	case IP_VERSION(11, 0, 3):
1318 		adev->gfx.me.num_me = 1;
1319 		adev->gfx.me.num_pipe_per_me = 1;
1320 		adev->gfx.me.num_queue_per_pipe = 1;
1321 		adev->gfx.mec.num_mec = 2;
1322 		adev->gfx.mec.num_pipe_per_mec = 4;
1323 		adev->gfx.mec.num_queue_per_pipe = 4;
1324 		break;
1325 	case IP_VERSION(11, 0, 1):
1326 	case IP_VERSION(11, 0, 4):
1327 		adev->gfx.me.num_me = 1;
1328 		adev->gfx.me.num_pipe_per_me = 1;
1329 		adev->gfx.me.num_queue_per_pipe = 1;
1330 		adev->gfx.mec.num_mec = 1;
1331 		adev->gfx.mec.num_pipe_per_mec = 4;
1332 		adev->gfx.mec.num_queue_per_pipe = 4;
1333 		break;
1334 	default:
1335 		adev->gfx.me.num_me = 1;
1336 		adev->gfx.me.num_pipe_per_me = 1;
1337 		adev->gfx.me.num_queue_per_pipe = 1;
1338 		adev->gfx.mec.num_mec = 1;
1339 		adev->gfx.mec.num_pipe_per_mec = 4;
1340 		adev->gfx.mec.num_queue_per_pipe = 8;
1341 		break;
1342 	}
1343 
1344 	/* Enable CG flag in one VF mode for enabling RLC safe mode enter/exit */
1345 	if (adev->ip_versions[GC_HWIP][0] == IP_VERSION(11, 0, 3) &&
1346 		amdgpu_sriov_is_pp_one_vf(adev))
1347 		adev->cg_flags = AMD_CG_SUPPORT_GFX_CGCG;
1348 
1349 	/* EOP Event */
1350 	r = amdgpu_irq_add_id(adev, SOC21_IH_CLIENTID_GRBM_CP,
1351 			      GFX_11_0_0__SRCID__CP_EOP_INTERRUPT,
1352 			      &adev->gfx.eop_irq);
1353 	if (r)
1354 		return r;
1355 
1356 	/* Privileged reg */
1357 	r = amdgpu_irq_add_id(adev, SOC21_IH_CLIENTID_GRBM_CP,
1358 			      GFX_11_0_0__SRCID__CP_PRIV_REG_FAULT,
1359 			      &adev->gfx.priv_reg_irq);
1360 	if (r)
1361 		return r;
1362 
1363 	/* Privileged inst */
1364 	r = amdgpu_irq_add_id(adev, SOC21_IH_CLIENTID_GRBM_CP,
1365 			      GFX_11_0_0__SRCID__CP_PRIV_INSTR_FAULT,
1366 			      &adev->gfx.priv_inst_irq);
1367 	if (r)
1368 		return r;
1369 
1370 	/* FED error */
1371 	r = amdgpu_irq_add_id(adev, SOC21_IH_CLIENTID_GFX,
1372 				  GFX_11_0_0__SRCID__RLC_GC_FED_INTERRUPT,
1373 				  &adev->gfx.rlc_gc_fed_irq);
1374 	if (r)
1375 		return r;
1376 
1377 	adev->gfx.gfx_current_status = AMDGPU_GFX_NORMAL_MODE;
1378 
1379 	if (adev->gfx.imu.funcs) {
1380 		if (adev->gfx.imu.funcs->init_microcode) {
1381 			r = adev->gfx.imu.funcs->init_microcode(adev);
1382 			if (r)
1383 				DRM_ERROR("Failed to load imu firmware!\n");
1384 		}
1385 	}
1386 
1387 	gfx_v11_0_me_init(adev);
1388 
1389 	r = gfx_v11_0_rlc_init(adev);
1390 	if (r) {
1391 		DRM_ERROR("Failed to init rlc BOs!\n");
1392 		return r;
1393 	}
1394 
1395 	r = gfx_v11_0_mec_init(adev);
1396 	if (r) {
1397 		DRM_ERROR("Failed to init MEC BOs!\n");
1398 		return r;
1399 	}
1400 
1401 	/* set up the gfx ring */
1402 	for (i = 0; i < adev->gfx.me.num_me; i++) {
1403 		for (j = 0; j < adev->gfx.me.num_queue_per_pipe; j++) {
1404 			for (k = 0; k < adev->gfx.me.num_pipe_per_me; k++) {
1405 				if (!amdgpu_gfx_is_me_queue_enabled(adev, i, k, j))
1406 					continue;
1407 
1408 				r = gfx_v11_0_gfx_ring_init(adev, ring_id,
1409 							    i, k, j);
1410 				if (r)
1411 					return r;
1412 				ring_id++;
1413 			}
1414 		}
1415 	}
1416 
1417 	ring_id = 0;
1418 	/* set up the compute queues - allocate horizontally across pipes */
1419 	for (i = 0; i < adev->gfx.mec.num_mec; ++i) {
1420 		for (j = 0; j < adev->gfx.mec.num_queue_per_pipe; j++) {
1421 			for (k = 0; k < adev->gfx.mec.num_pipe_per_mec; k++) {
1422 				if (!amdgpu_gfx_is_mec_queue_enabled(adev, 0, i,
1423 								     k, j))
1424 					continue;
1425 
1426 				r = gfx_v11_0_compute_ring_init(adev, ring_id,
1427 								i, k, j);
1428 				if (r)
1429 					return r;
1430 
1431 				ring_id++;
1432 			}
1433 		}
1434 	}
1435 
1436 	if (!adev->enable_mes_kiq) {
1437 		r = amdgpu_gfx_kiq_init(adev, GFX11_MEC_HPD_SIZE, 0);
1438 		if (r) {
1439 			DRM_ERROR("Failed to init KIQ BOs!\n");
1440 			return r;
1441 		}
1442 
1443 		kiq = &adev->gfx.kiq[0];
1444 		r = amdgpu_gfx_kiq_init_ring(adev, &kiq->ring, &kiq->irq, 0);
1445 		if (r)
1446 			return r;
1447 	}
1448 
1449 	r = amdgpu_gfx_mqd_sw_init(adev, sizeof(struct v11_compute_mqd), 0);
1450 	if (r)
1451 		return r;
1452 
1453 	/* allocate visible FB for rlc auto-loading fw */
1454 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_RLC_BACKDOOR_AUTO) {
1455 		r = gfx_v11_0_rlc_autoload_buffer_init(adev);
1456 		if (r)
1457 			return r;
1458 	}
1459 
1460 	r = gfx_v11_0_gpu_early_init(adev);
1461 	if (r)
1462 		return r;
1463 
1464 	if (amdgpu_gfx_ras_sw_init(adev)) {
1465 		dev_err(adev->dev, "Failed to initialize gfx ras block!\n");
1466 		return -EINVAL;
1467 	}
1468 
1469 	return 0;
1470 }
1471 
1472 static void gfx_v11_0_pfp_fini(struct amdgpu_device *adev)
1473 {
1474 	amdgpu_bo_free_kernel(&adev->gfx.pfp.pfp_fw_obj,
1475 			      &adev->gfx.pfp.pfp_fw_gpu_addr,
1476 			      (void **)&adev->gfx.pfp.pfp_fw_ptr);
1477 
1478 	amdgpu_bo_free_kernel(&adev->gfx.pfp.pfp_fw_data_obj,
1479 			      &adev->gfx.pfp.pfp_fw_data_gpu_addr,
1480 			      (void **)&adev->gfx.pfp.pfp_fw_data_ptr);
1481 }
1482 
1483 static void gfx_v11_0_me_fini(struct amdgpu_device *adev)
1484 {
1485 	amdgpu_bo_free_kernel(&adev->gfx.me.me_fw_obj,
1486 			      &adev->gfx.me.me_fw_gpu_addr,
1487 			      (void **)&adev->gfx.me.me_fw_ptr);
1488 
1489 	amdgpu_bo_free_kernel(&adev->gfx.me.me_fw_data_obj,
1490 			       &adev->gfx.me.me_fw_data_gpu_addr,
1491 			       (void **)&adev->gfx.me.me_fw_data_ptr);
1492 }
1493 
1494 static void gfx_v11_0_rlc_autoload_buffer_fini(struct amdgpu_device *adev)
1495 {
1496 	amdgpu_bo_free_kernel(&adev->gfx.rlc.rlc_autoload_bo,
1497 			&adev->gfx.rlc.rlc_autoload_gpu_addr,
1498 			(void **)&adev->gfx.rlc.rlc_autoload_ptr);
1499 }
1500 
1501 static int gfx_v11_0_sw_fini(void *handle)
1502 {
1503 	int i;
1504 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1505 
1506 	for (i = 0; i < adev->gfx.num_gfx_rings; i++)
1507 		amdgpu_ring_fini(&adev->gfx.gfx_ring[i]);
1508 	for (i = 0; i < adev->gfx.num_compute_rings; i++)
1509 		amdgpu_ring_fini(&adev->gfx.compute_ring[i]);
1510 
1511 	amdgpu_gfx_mqd_sw_fini(adev, 0);
1512 
1513 	if (!adev->enable_mes_kiq) {
1514 		amdgpu_gfx_kiq_free_ring(&adev->gfx.kiq[0].ring);
1515 		amdgpu_gfx_kiq_fini(adev, 0);
1516 	}
1517 
1518 	gfx_v11_0_pfp_fini(adev);
1519 	gfx_v11_0_me_fini(adev);
1520 	gfx_v11_0_rlc_fini(adev);
1521 	gfx_v11_0_mec_fini(adev);
1522 
1523 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_RLC_BACKDOOR_AUTO)
1524 		gfx_v11_0_rlc_autoload_buffer_fini(adev);
1525 
1526 	gfx_v11_0_free_microcode(adev);
1527 
1528 	return 0;
1529 }
1530 
1531 static void gfx_v11_0_select_se_sh(struct amdgpu_device *adev, u32 se_num,
1532 				   u32 sh_num, u32 instance, int xcc_id)
1533 {
1534 	u32 data;
1535 
1536 	if (instance == 0xffffffff)
1537 		data = REG_SET_FIELD(0, GRBM_GFX_INDEX,
1538 				     INSTANCE_BROADCAST_WRITES, 1);
1539 	else
1540 		data = REG_SET_FIELD(0, GRBM_GFX_INDEX, INSTANCE_INDEX,
1541 				     instance);
1542 
1543 	if (se_num == 0xffffffff)
1544 		data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SE_BROADCAST_WRITES,
1545 				     1);
1546 	else
1547 		data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SE_INDEX, se_num);
1548 
1549 	if (sh_num == 0xffffffff)
1550 		data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SA_BROADCAST_WRITES,
1551 				     1);
1552 	else
1553 		data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SA_INDEX, sh_num);
1554 
1555 	WREG32_SOC15(GC, 0, regGRBM_GFX_INDEX, data);
1556 }
1557 
1558 static u32 gfx_v11_0_get_sa_active_bitmap(struct amdgpu_device *adev)
1559 {
1560 	u32 gc_disabled_sa_mask, gc_user_disabled_sa_mask, sa_mask;
1561 
1562 	gc_disabled_sa_mask = RREG32_SOC15(GC, 0, regCC_GC_SA_UNIT_DISABLE);
1563 	gc_disabled_sa_mask = REG_GET_FIELD(gc_disabled_sa_mask,
1564 					   CC_GC_SA_UNIT_DISABLE,
1565 					   SA_DISABLE);
1566 	gc_user_disabled_sa_mask = RREG32_SOC15(GC, 0, regGC_USER_SA_UNIT_DISABLE);
1567 	gc_user_disabled_sa_mask = REG_GET_FIELD(gc_user_disabled_sa_mask,
1568 						 GC_USER_SA_UNIT_DISABLE,
1569 						 SA_DISABLE);
1570 	sa_mask = amdgpu_gfx_create_bitmask(adev->gfx.config.max_sh_per_se *
1571 					    adev->gfx.config.max_shader_engines);
1572 
1573 	return sa_mask & (~(gc_disabled_sa_mask | gc_user_disabled_sa_mask));
1574 }
1575 
1576 static u32 gfx_v11_0_get_rb_active_bitmap(struct amdgpu_device *adev)
1577 {
1578 	u32 gc_disabled_rb_mask, gc_user_disabled_rb_mask;
1579 	u32 rb_mask;
1580 
1581 	gc_disabled_rb_mask = RREG32_SOC15(GC, 0, regCC_RB_BACKEND_DISABLE);
1582 	gc_disabled_rb_mask = REG_GET_FIELD(gc_disabled_rb_mask,
1583 					    CC_RB_BACKEND_DISABLE,
1584 					    BACKEND_DISABLE);
1585 	gc_user_disabled_rb_mask = RREG32_SOC15(GC, 0, regGC_USER_RB_BACKEND_DISABLE);
1586 	gc_user_disabled_rb_mask = REG_GET_FIELD(gc_user_disabled_rb_mask,
1587 						 GC_USER_RB_BACKEND_DISABLE,
1588 						 BACKEND_DISABLE);
1589 	rb_mask = amdgpu_gfx_create_bitmask(adev->gfx.config.max_backends_per_se *
1590 					    adev->gfx.config.max_shader_engines);
1591 
1592 	return rb_mask & (~(gc_disabled_rb_mask | gc_user_disabled_rb_mask));
1593 }
1594 
1595 static void gfx_v11_0_setup_rb(struct amdgpu_device *adev)
1596 {
1597 	u32 rb_bitmap_width_per_sa;
1598 	u32 max_sa;
1599 	u32 active_sa_bitmap;
1600 	u32 global_active_rb_bitmap;
1601 	u32 active_rb_bitmap = 0;
1602 	u32 i;
1603 
1604 	/* query sa bitmap from SA_UNIT_DISABLE registers */
1605 	active_sa_bitmap = gfx_v11_0_get_sa_active_bitmap(adev);
1606 	/* query rb bitmap from RB_BACKEND_DISABLE registers */
1607 	global_active_rb_bitmap = gfx_v11_0_get_rb_active_bitmap(adev);
1608 
1609 	/* generate active rb bitmap according to active sa bitmap */
1610 	max_sa = adev->gfx.config.max_shader_engines *
1611 		 adev->gfx.config.max_sh_per_se;
1612 	rb_bitmap_width_per_sa = adev->gfx.config.max_backends_per_se /
1613 				 adev->gfx.config.max_sh_per_se;
1614 	for (i = 0; i < max_sa; i++) {
1615 		if (active_sa_bitmap & (1 << i))
1616 			active_rb_bitmap |= (0x3 << (i * rb_bitmap_width_per_sa));
1617 	}
1618 
1619 	active_rb_bitmap |= global_active_rb_bitmap;
1620 	adev->gfx.config.backend_enable_mask = active_rb_bitmap;
1621 	adev->gfx.config.num_rbs = hweight32(active_rb_bitmap);
1622 }
1623 
1624 #define DEFAULT_SH_MEM_BASES	(0x6000)
1625 #define LDS_APP_BASE           0x1
1626 #define SCRATCH_APP_BASE       0x2
1627 
1628 static void gfx_v11_0_init_compute_vmid(struct amdgpu_device *adev)
1629 {
1630 	int i;
1631 	uint32_t sh_mem_bases;
1632 	uint32_t data;
1633 
1634 	/*
1635 	 * Configure apertures:
1636 	 * LDS:         0x60000000'00000000 - 0x60000001'00000000 (4GB)
1637 	 * Scratch:     0x60000001'00000000 - 0x60000002'00000000 (4GB)
1638 	 * GPUVM:       0x60010000'00000000 - 0x60020000'00000000 (1TB)
1639 	 */
1640 	sh_mem_bases = (LDS_APP_BASE << SH_MEM_BASES__SHARED_BASE__SHIFT) |
1641 			SCRATCH_APP_BASE;
1642 
1643 	mutex_lock(&adev->srbm_mutex);
1644 	for (i = adev->vm_manager.first_kfd_vmid; i < AMDGPU_NUM_VMID; i++) {
1645 		soc21_grbm_select(adev, 0, 0, 0, i);
1646 		/* CP and shaders */
1647 		WREG32_SOC15(GC, 0, regSH_MEM_CONFIG, DEFAULT_SH_MEM_CONFIG);
1648 		WREG32_SOC15(GC, 0, regSH_MEM_BASES, sh_mem_bases);
1649 
1650 		/* Enable trap for each kfd vmid. */
1651 		data = RREG32_SOC15(GC, 0, regSPI_GDBG_PER_VMID_CNTL);
1652 		data = REG_SET_FIELD(data, SPI_GDBG_PER_VMID_CNTL, TRAP_EN, 1);
1653 		WREG32_SOC15(GC, 0, regSPI_GDBG_PER_VMID_CNTL, data);
1654 	}
1655 	soc21_grbm_select(adev, 0, 0, 0, 0);
1656 	mutex_unlock(&adev->srbm_mutex);
1657 
1658 	/* Initialize all compute VMIDs to have no GDS, GWS, or OA
1659 	   acccess. These should be enabled by FW for target VMIDs. */
1660 	for (i = adev->vm_manager.first_kfd_vmid; i < AMDGPU_NUM_VMID; i++) {
1661 		WREG32_SOC15_OFFSET(GC, 0, regGDS_VMID0_BASE, 2 * i, 0);
1662 		WREG32_SOC15_OFFSET(GC, 0, regGDS_VMID0_SIZE, 2 * i, 0);
1663 		WREG32_SOC15_OFFSET(GC, 0, regGDS_GWS_VMID0, i, 0);
1664 		WREG32_SOC15_OFFSET(GC, 0, regGDS_OA_VMID0, i, 0);
1665 	}
1666 }
1667 
1668 static void gfx_v11_0_init_gds_vmid(struct amdgpu_device *adev)
1669 {
1670 	int vmid;
1671 
1672 	/*
1673 	 * Initialize all compute and user-gfx VMIDs to have no GDS, GWS, or OA
1674 	 * access. Compute VMIDs should be enabled by FW for target VMIDs,
1675 	 * the driver can enable them for graphics. VMID0 should maintain
1676 	 * access so that HWS firmware can save/restore entries.
1677 	 */
1678 	for (vmid = 1; vmid < 16; vmid++) {
1679 		WREG32_SOC15_OFFSET(GC, 0, regGDS_VMID0_BASE, 2 * vmid, 0);
1680 		WREG32_SOC15_OFFSET(GC, 0, regGDS_VMID0_SIZE, 2 * vmid, 0);
1681 		WREG32_SOC15_OFFSET(GC, 0, regGDS_GWS_VMID0, vmid, 0);
1682 		WREG32_SOC15_OFFSET(GC, 0, regGDS_OA_VMID0, vmid, 0);
1683 	}
1684 }
1685 
1686 static void gfx_v11_0_tcp_harvest(struct amdgpu_device *adev)
1687 {
1688 	/* TODO: harvest feature to be added later. */
1689 }
1690 
1691 static void gfx_v11_0_get_tcc_info(struct amdgpu_device *adev)
1692 {
1693 	/* TCCs are global (not instanced). */
1694 	uint32_t tcc_disable = RREG32_SOC15(GC, 0, regCGTS_TCC_DISABLE) |
1695 			       RREG32_SOC15(GC, 0, regCGTS_USER_TCC_DISABLE);
1696 
1697 	adev->gfx.config.tcc_disabled_mask =
1698 		REG_GET_FIELD(tcc_disable, CGTS_TCC_DISABLE, TCC_DISABLE) |
1699 		(REG_GET_FIELD(tcc_disable, CGTS_TCC_DISABLE, HI_TCC_DISABLE) << 16);
1700 }
1701 
1702 static void gfx_v11_0_constants_init(struct amdgpu_device *adev)
1703 {
1704 	u32 tmp;
1705 	int i;
1706 
1707 	if (!amdgpu_sriov_vf(adev))
1708 		WREG32_FIELD15_PREREG(GC, 0, GRBM_CNTL, READ_TIMEOUT, 0xff);
1709 
1710 	gfx_v11_0_setup_rb(adev);
1711 	gfx_v11_0_get_cu_info(adev, &adev->gfx.cu_info);
1712 	gfx_v11_0_get_tcc_info(adev);
1713 	adev->gfx.config.pa_sc_tile_steering_override = 0;
1714 
1715 	/* Set whether texture coordinate truncation is conformant. */
1716 	tmp = RREG32_SOC15(GC, 0, regTA_CNTL2);
1717 	adev->gfx.config.ta_cntl2_truncate_coord_mode =
1718 		REG_GET_FIELD(tmp, TA_CNTL2, TRUNCATE_COORD_MODE);
1719 
1720 	/* XXX SH_MEM regs */
1721 	/* where to put LDS, scratch, GPUVM in FSA64 space */
1722 	mutex_lock(&adev->srbm_mutex);
1723 	for (i = 0; i < adev->vm_manager.id_mgr[AMDGPU_GFXHUB(0)].num_ids; i++) {
1724 		soc21_grbm_select(adev, 0, 0, 0, i);
1725 		/* CP and shaders */
1726 		WREG32_SOC15(GC, 0, regSH_MEM_CONFIG, DEFAULT_SH_MEM_CONFIG);
1727 		if (i != 0) {
1728 			tmp = REG_SET_FIELD(0, SH_MEM_BASES, PRIVATE_BASE,
1729 				(adev->gmc.private_aperture_start >> 48));
1730 			tmp = REG_SET_FIELD(tmp, SH_MEM_BASES, SHARED_BASE,
1731 				(adev->gmc.shared_aperture_start >> 48));
1732 			WREG32_SOC15(GC, 0, regSH_MEM_BASES, tmp);
1733 		}
1734 	}
1735 	soc21_grbm_select(adev, 0, 0, 0, 0);
1736 
1737 	mutex_unlock(&adev->srbm_mutex);
1738 
1739 	gfx_v11_0_init_compute_vmid(adev);
1740 	gfx_v11_0_init_gds_vmid(adev);
1741 }
1742 
1743 static void gfx_v11_0_enable_gui_idle_interrupt(struct amdgpu_device *adev,
1744 					       bool enable)
1745 {
1746 	u32 tmp;
1747 
1748 	if (amdgpu_sriov_vf(adev))
1749 		return;
1750 
1751 	tmp = RREG32_SOC15(GC, 0, regCP_INT_CNTL_RING0);
1752 
1753 	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CNTX_BUSY_INT_ENABLE,
1754 			    enable ? 1 : 0);
1755 	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CNTX_EMPTY_INT_ENABLE,
1756 			    enable ? 1 : 0);
1757 	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CMP_BUSY_INT_ENABLE,
1758 			    enable ? 1 : 0);
1759 	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, GFX_IDLE_INT_ENABLE,
1760 			    enable ? 1 : 0);
1761 
1762 	WREG32_SOC15(GC, 0, regCP_INT_CNTL_RING0, tmp);
1763 }
1764 
1765 static int gfx_v11_0_init_csb(struct amdgpu_device *adev)
1766 {
1767 	adev->gfx.rlc.funcs->get_csb_buffer(adev, adev->gfx.rlc.cs_ptr);
1768 
1769 	WREG32_SOC15(GC, 0, regRLC_CSIB_ADDR_HI,
1770 			adev->gfx.rlc.clear_state_gpu_addr >> 32);
1771 	WREG32_SOC15(GC, 0, regRLC_CSIB_ADDR_LO,
1772 			adev->gfx.rlc.clear_state_gpu_addr & 0xfffffffc);
1773 	WREG32_SOC15(GC, 0, regRLC_CSIB_LENGTH, adev->gfx.rlc.clear_state_size);
1774 
1775 	return 0;
1776 }
1777 
1778 static void gfx_v11_0_rlc_stop(struct amdgpu_device *adev)
1779 {
1780 	u32 tmp = RREG32_SOC15(GC, 0, regRLC_CNTL);
1781 
1782 	tmp = REG_SET_FIELD(tmp, RLC_CNTL, RLC_ENABLE_F32, 0);
1783 	WREG32_SOC15(GC, 0, regRLC_CNTL, tmp);
1784 }
1785 
1786 static void gfx_v11_0_rlc_reset(struct amdgpu_device *adev)
1787 {
1788 	WREG32_FIELD15_PREREG(GC, 0, GRBM_SOFT_RESET, SOFT_RESET_RLC, 1);
1789 	udelay(50);
1790 	WREG32_FIELD15_PREREG(GC, 0, GRBM_SOFT_RESET, SOFT_RESET_RLC, 0);
1791 	udelay(50);
1792 }
1793 
1794 static void gfx_v11_0_rlc_smu_handshake_cntl(struct amdgpu_device *adev,
1795 					     bool enable)
1796 {
1797 	uint32_t rlc_pg_cntl;
1798 
1799 	rlc_pg_cntl = RREG32_SOC15(GC, 0, regRLC_PG_CNTL);
1800 
1801 	if (!enable) {
1802 		/* RLC_PG_CNTL[23] = 0 (default)
1803 		 * RLC will wait for handshake acks with SMU
1804 		 * GFXOFF will be enabled
1805 		 * RLC_PG_CNTL[23] = 1
1806 		 * RLC will not issue any message to SMU
1807 		 * hence no handshake between SMU & RLC
1808 		 * GFXOFF will be disabled
1809 		 */
1810 		rlc_pg_cntl |= RLC_PG_CNTL__SMU_HANDSHAKE_DISABLE_MASK;
1811 	} else
1812 		rlc_pg_cntl &= ~RLC_PG_CNTL__SMU_HANDSHAKE_DISABLE_MASK;
1813 	WREG32_SOC15(GC, 0, regRLC_PG_CNTL, rlc_pg_cntl);
1814 }
1815 
1816 static void gfx_v11_0_rlc_start(struct amdgpu_device *adev)
1817 {
1818 	/* TODO: enable rlc & smu handshake until smu
1819 	 * and gfxoff feature works as expected */
1820 	if (!(amdgpu_pp_feature_mask & PP_GFXOFF_MASK))
1821 		gfx_v11_0_rlc_smu_handshake_cntl(adev, false);
1822 
1823 	WREG32_FIELD15_PREREG(GC, 0, RLC_CNTL, RLC_ENABLE_F32, 1);
1824 	udelay(50);
1825 }
1826 
1827 static void gfx_v11_0_rlc_enable_srm(struct amdgpu_device *adev)
1828 {
1829 	uint32_t tmp;
1830 
1831 	/* enable Save Restore Machine */
1832 	tmp = RREG32(SOC15_REG_OFFSET(GC, 0, regRLC_SRM_CNTL));
1833 	tmp |= RLC_SRM_CNTL__AUTO_INCR_ADDR_MASK;
1834 	tmp |= RLC_SRM_CNTL__SRM_ENABLE_MASK;
1835 	WREG32(SOC15_REG_OFFSET(GC, 0, regRLC_SRM_CNTL), tmp);
1836 }
1837 
1838 static void gfx_v11_0_load_rlcg_microcode(struct amdgpu_device *adev)
1839 {
1840 	const struct rlc_firmware_header_v2_0 *hdr;
1841 	const __le32 *fw_data;
1842 	unsigned i, fw_size;
1843 
1844 	hdr = (const struct rlc_firmware_header_v2_0 *)adev->gfx.rlc_fw->data;
1845 	fw_data = (const __le32 *)(adev->gfx.rlc_fw->data +
1846 			   le32_to_cpu(hdr->header.ucode_array_offset_bytes));
1847 	fw_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4;
1848 
1849 	WREG32_SOC15(GC, 0, regRLC_GPM_UCODE_ADDR,
1850 		     RLCG_UCODE_LOADING_START_ADDRESS);
1851 
1852 	for (i = 0; i < fw_size; i++)
1853 		WREG32_SOC15(GC, 0, regRLC_GPM_UCODE_DATA,
1854 			     le32_to_cpup(fw_data++));
1855 
1856 	WREG32_SOC15(GC, 0, regRLC_GPM_UCODE_ADDR, adev->gfx.rlc_fw_version);
1857 }
1858 
1859 static void gfx_v11_0_load_rlc_iram_dram_microcode(struct amdgpu_device *adev)
1860 {
1861 	const struct rlc_firmware_header_v2_2 *hdr;
1862 	const __le32 *fw_data;
1863 	unsigned i, fw_size;
1864 	u32 tmp;
1865 
1866 	hdr = (const struct rlc_firmware_header_v2_2 *)adev->gfx.rlc_fw->data;
1867 
1868 	fw_data = (const __le32 *)(adev->gfx.rlc_fw->data +
1869 			le32_to_cpu(hdr->rlc_iram_ucode_offset_bytes));
1870 	fw_size = le32_to_cpu(hdr->rlc_iram_ucode_size_bytes) / 4;
1871 
1872 	WREG32_SOC15(GC, 0, regRLC_LX6_IRAM_ADDR, 0);
1873 
1874 	for (i = 0; i < fw_size; i++) {
1875 		if ((amdgpu_emu_mode == 1) && (i % 100 == 99))
1876 			msleep(1);
1877 		WREG32_SOC15(GC, 0, regRLC_LX6_IRAM_DATA,
1878 				le32_to_cpup(fw_data++));
1879 	}
1880 
1881 	WREG32_SOC15(GC, 0, regRLC_LX6_IRAM_ADDR, adev->gfx.rlc_fw_version);
1882 
1883 	fw_data = (const __le32 *)(adev->gfx.rlc_fw->data +
1884 			le32_to_cpu(hdr->rlc_dram_ucode_offset_bytes));
1885 	fw_size = le32_to_cpu(hdr->rlc_dram_ucode_size_bytes) / 4;
1886 
1887 	WREG32_SOC15(GC, 0, regRLC_LX6_DRAM_ADDR, 0);
1888 	for (i = 0; i < fw_size; i++) {
1889 		if ((amdgpu_emu_mode == 1) && (i % 100 == 99))
1890 			msleep(1);
1891 		WREG32_SOC15(GC, 0, regRLC_LX6_DRAM_DATA,
1892 				le32_to_cpup(fw_data++));
1893 	}
1894 
1895 	WREG32_SOC15(GC, 0, regRLC_LX6_IRAM_ADDR, adev->gfx.rlc_fw_version);
1896 
1897 	tmp = RREG32_SOC15(GC, 0, regRLC_LX6_CNTL);
1898 	tmp = REG_SET_FIELD(tmp, RLC_LX6_CNTL, PDEBUG_ENABLE, 1);
1899 	tmp = REG_SET_FIELD(tmp, RLC_LX6_CNTL, BRESET, 0);
1900 	WREG32_SOC15(GC, 0, regRLC_LX6_CNTL, tmp);
1901 }
1902 
1903 static void gfx_v11_0_load_rlcp_rlcv_microcode(struct amdgpu_device *adev)
1904 {
1905 	const struct rlc_firmware_header_v2_3 *hdr;
1906 	const __le32 *fw_data;
1907 	unsigned i, fw_size;
1908 	u32 tmp;
1909 
1910 	hdr = (const struct rlc_firmware_header_v2_3 *)adev->gfx.rlc_fw->data;
1911 
1912 	fw_data = (const __le32 *)(adev->gfx.rlc_fw->data +
1913 			le32_to_cpu(hdr->rlcp_ucode_offset_bytes));
1914 	fw_size = le32_to_cpu(hdr->rlcp_ucode_size_bytes) / 4;
1915 
1916 	WREG32_SOC15(GC, 0, regRLC_PACE_UCODE_ADDR, 0);
1917 
1918 	for (i = 0; i < fw_size; i++) {
1919 		if ((amdgpu_emu_mode == 1) && (i % 100 == 99))
1920 			msleep(1);
1921 		WREG32_SOC15(GC, 0, regRLC_PACE_UCODE_DATA,
1922 				le32_to_cpup(fw_data++));
1923 	}
1924 
1925 	WREG32_SOC15(GC, 0, regRLC_PACE_UCODE_ADDR, adev->gfx.rlc_fw_version);
1926 
1927 	tmp = RREG32_SOC15(GC, 0, regRLC_GPM_THREAD_ENABLE);
1928 	tmp = REG_SET_FIELD(tmp, RLC_GPM_THREAD_ENABLE, THREAD1_ENABLE, 1);
1929 	WREG32_SOC15(GC, 0, regRLC_GPM_THREAD_ENABLE, tmp);
1930 
1931 	fw_data = (const __le32 *)(adev->gfx.rlc_fw->data +
1932 			le32_to_cpu(hdr->rlcv_ucode_offset_bytes));
1933 	fw_size = le32_to_cpu(hdr->rlcv_ucode_size_bytes) / 4;
1934 
1935 	WREG32_SOC15(GC, 0, regRLC_GPU_IOV_UCODE_ADDR, 0);
1936 
1937 	for (i = 0; i < fw_size; i++) {
1938 		if ((amdgpu_emu_mode == 1) && (i % 100 == 99))
1939 			msleep(1);
1940 		WREG32_SOC15(GC, 0, regRLC_GPU_IOV_UCODE_DATA,
1941 				le32_to_cpup(fw_data++));
1942 	}
1943 
1944 	WREG32_SOC15(GC, 0, regRLC_GPU_IOV_UCODE_ADDR, adev->gfx.rlc_fw_version);
1945 
1946 	tmp = RREG32_SOC15(GC, 0, regRLC_GPU_IOV_F32_CNTL);
1947 	tmp = REG_SET_FIELD(tmp, RLC_GPU_IOV_F32_CNTL, ENABLE, 1);
1948 	WREG32_SOC15(GC, 0, regRLC_GPU_IOV_F32_CNTL, tmp);
1949 }
1950 
1951 static int gfx_v11_0_rlc_load_microcode(struct amdgpu_device *adev)
1952 {
1953 	const struct rlc_firmware_header_v2_0 *hdr;
1954 	uint16_t version_major;
1955 	uint16_t version_minor;
1956 
1957 	if (!adev->gfx.rlc_fw)
1958 		return -EINVAL;
1959 
1960 	hdr = (const struct rlc_firmware_header_v2_0 *)adev->gfx.rlc_fw->data;
1961 	amdgpu_ucode_print_rlc_hdr(&hdr->header);
1962 
1963 	version_major = le16_to_cpu(hdr->header.header_version_major);
1964 	version_minor = le16_to_cpu(hdr->header.header_version_minor);
1965 
1966 	if (version_major == 2) {
1967 		gfx_v11_0_load_rlcg_microcode(adev);
1968 		if (amdgpu_dpm == 1) {
1969 			if (version_minor >= 2)
1970 				gfx_v11_0_load_rlc_iram_dram_microcode(adev);
1971 			if (version_minor == 3)
1972 				gfx_v11_0_load_rlcp_rlcv_microcode(adev);
1973 		}
1974 
1975 		return 0;
1976 	}
1977 
1978 	return -EINVAL;
1979 }
1980 
1981 static int gfx_v11_0_rlc_resume(struct amdgpu_device *adev)
1982 {
1983 	int r;
1984 
1985 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
1986 		gfx_v11_0_init_csb(adev);
1987 
1988 		if (!amdgpu_sriov_vf(adev)) /* enable RLC SRM */
1989 			gfx_v11_0_rlc_enable_srm(adev);
1990 	} else {
1991 		if (amdgpu_sriov_vf(adev)) {
1992 			gfx_v11_0_init_csb(adev);
1993 			return 0;
1994 		}
1995 
1996 		adev->gfx.rlc.funcs->stop(adev);
1997 
1998 		/* disable CG */
1999 		WREG32_SOC15(GC, 0, regRLC_CGCG_CGLS_CTRL, 0);
2000 
2001 		/* disable PG */
2002 		WREG32_SOC15(GC, 0, regRLC_PG_CNTL, 0);
2003 
2004 		if (adev->firmware.load_type == AMDGPU_FW_LOAD_DIRECT) {
2005 			/* legacy rlc firmware loading */
2006 			r = gfx_v11_0_rlc_load_microcode(adev);
2007 			if (r)
2008 				return r;
2009 		}
2010 
2011 		gfx_v11_0_init_csb(adev);
2012 
2013 		adev->gfx.rlc.funcs->start(adev);
2014 	}
2015 	return 0;
2016 }
2017 
2018 static int gfx_v11_0_config_me_cache(struct amdgpu_device *adev, uint64_t addr)
2019 {
2020 	uint32_t usec_timeout = 50000;  /* wait for 50ms */
2021 	uint32_t tmp;
2022 	int i;
2023 
2024 	/* Trigger an invalidation of the L1 instruction caches */
2025 	tmp = RREG32_SOC15(GC, 0, regCP_ME_IC_OP_CNTL);
2026 	tmp = REG_SET_FIELD(tmp, CP_ME_IC_OP_CNTL, INVALIDATE_CACHE, 1);
2027 	WREG32_SOC15(GC, 0, regCP_ME_IC_OP_CNTL, tmp);
2028 
2029 	/* Wait for invalidation complete */
2030 	for (i = 0; i < usec_timeout; i++) {
2031 		tmp = RREG32_SOC15(GC, 0, regCP_ME_IC_OP_CNTL);
2032 		if (1 == REG_GET_FIELD(tmp, CP_ME_IC_OP_CNTL,
2033 					INVALIDATE_CACHE_COMPLETE))
2034 			break;
2035 		udelay(1);
2036 	}
2037 
2038 	if (i >= usec_timeout) {
2039 		dev_err(adev->dev, "failed to invalidate instruction cache\n");
2040 		return -EINVAL;
2041 	}
2042 
2043 	if (amdgpu_emu_mode == 1)
2044 		adev->hdp.funcs->flush_hdp(adev, NULL);
2045 
2046 	tmp = RREG32_SOC15(GC, 0, regCP_ME_IC_BASE_CNTL);
2047 	tmp = REG_SET_FIELD(tmp, CP_ME_IC_BASE_CNTL, VMID, 0);
2048 	tmp = REG_SET_FIELD(tmp, CP_ME_IC_BASE_CNTL, CACHE_POLICY, 0);
2049 	tmp = REG_SET_FIELD(tmp, CP_ME_IC_BASE_CNTL, EXE_DISABLE, 0);
2050 	tmp = REG_SET_FIELD(tmp, CP_ME_IC_BASE_CNTL, ADDRESS_CLAMP, 1);
2051 	WREG32_SOC15(GC, 0, regCP_ME_IC_BASE_CNTL, tmp);
2052 
2053 	/* Program me ucode address into intruction cache address register */
2054 	WREG32_SOC15(GC, 0, regCP_ME_IC_BASE_LO,
2055 			lower_32_bits(addr) & 0xFFFFF000);
2056 	WREG32_SOC15(GC, 0, regCP_ME_IC_BASE_HI,
2057 			upper_32_bits(addr));
2058 
2059 	return 0;
2060 }
2061 
2062 static int gfx_v11_0_config_pfp_cache(struct amdgpu_device *adev, uint64_t addr)
2063 {
2064 	uint32_t usec_timeout = 50000;  /* wait for 50ms */
2065 	uint32_t tmp;
2066 	int i;
2067 
2068 	/* Trigger an invalidation of the L1 instruction caches */
2069 	tmp = RREG32_SOC15(GC, 0, regCP_PFP_IC_OP_CNTL);
2070 	tmp = REG_SET_FIELD(tmp, CP_PFP_IC_OP_CNTL, INVALIDATE_CACHE, 1);
2071 	WREG32_SOC15(GC, 0, regCP_PFP_IC_OP_CNTL, tmp);
2072 
2073 	/* Wait for invalidation complete */
2074 	for (i = 0; i < usec_timeout; i++) {
2075 		tmp = RREG32_SOC15(GC, 0, regCP_PFP_IC_OP_CNTL);
2076 		if (1 == REG_GET_FIELD(tmp, CP_PFP_IC_OP_CNTL,
2077 					INVALIDATE_CACHE_COMPLETE))
2078 			break;
2079 		udelay(1);
2080 	}
2081 
2082 	if (i >= usec_timeout) {
2083 		dev_err(adev->dev, "failed to invalidate instruction cache\n");
2084 		return -EINVAL;
2085 	}
2086 
2087 	if (amdgpu_emu_mode == 1)
2088 		adev->hdp.funcs->flush_hdp(adev, NULL);
2089 
2090 	tmp = RREG32_SOC15(GC, 0, regCP_PFP_IC_BASE_CNTL);
2091 	tmp = REG_SET_FIELD(tmp, CP_PFP_IC_BASE_CNTL, VMID, 0);
2092 	tmp = REG_SET_FIELD(tmp, CP_PFP_IC_BASE_CNTL, CACHE_POLICY, 0);
2093 	tmp = REG_SET_FIELD(tmp, CP_PFP_IC_BASE_CNTL, EXE_DISABLE, 0);
2094 	tmp = REG_SET_FIELD(tmp, CP_PFP_IC_BASE_CNTL, ADDRESS_CLAMP, 1);
2095 	WREG32_SOC15(GC, 0, regCP_PFP_IC_BASE_CNTL, tmp);
2096 
2097 	/* Program pfp ucode address into intruction cache address register */
2098 	WREG32_SOC15(GC, 0, regCP_PFP_IC_BASE_LO,
2099 			lower_32_bits(addr) & 0xFFFFF000);
2100 	WREG32_SOC15(GC, 0, regCP_PFP_IC_BASE_HI,
2101 			upper_32_bits(addr));
2102 
2103 	return 0;
2104 }
2105 
2106 static int gfx_v11_0_config_mec_cache(struct amdgpu_device *adev, uint64_t addr)
2107 {
2108 	uint32_t usec_timeout = 50000;  /* wait for 50ms */
2109 	uint32_t tmp;
2110 	int i;
2111 
2112 	/* Trigger an invalidation of the L1 instruction caches */
2113 	tmp = RREG32_SOC15(GC, 0, regCP_CPC_IC_OP_CNTL);
2114 	tmp = REG_SET_FIELD(tmp, CP_CPC_IC_OP_CNTL, INVALIDATE_CACHE, 1);
2115 
2116 	WREG32_SOC15(GC, 0, regCP_CPC_IC_OP_CNTL, tmp);
2117 
2118 	/* Wait for invalidation complete */
2119 	for (i = 0; i < usec_timeout; i++) {
2120 		tmp = RREG32_SOC15(GC, 0, regCP_CPC_IC_OP_CNTL);
2121 		if (1 == REG_GET_FIELD(tmp, CP_CPC_IC_OP_CNTL,
2122 					INVALIDATE_CACHE_COMPLETE))
2123 			break;
2124 		udelay(1);
2125 	}
2126 
2127 	if (i >= usec_timeout) {
2128 		dev_err(adev->dev, "failed to invalidate instruction cache\n");
2129 		return -EINVAL;
2130 	}
2131 
2132 	if (amdgpu_emu_mode == 1)
2133 		adev->hdp.funcs->flush_hdp(adev, NULL);
2134 
2135 	tmp = RREG32_SOC15(GC, 0, regCP_CPC_IC_BASE_CNTL);
2136 	tmp = REG_SET_FIELD(tmp, CP_CPC_IC_BASE_CNTL, CACHE_POLICY, 0);
2137 	tmp = REG_SET_FIELD(tmp, CP_CPC_IC_BASE_CNTL, EXE_DISABLE, 0);
2138 	tmp = REG_SET_FIELD(tmp, CP_CPC_IC_BASE_CNTL, ADDRESS_CLAMP, 1);
2139 	WREG32_SOC15(GC, 0, regCP_CPC_IC_BASE_CNTL, tmp);
2140 
2141 	/* Program mec1 ucode address into intruction cache address register */
2142 	WREG32_SOC15(GC, 0, regCP_CPC_IC_BASE_LO,
2143 			lower_32_bits(addr) & 0xFFFFF000);
2144 	WREG32_SOC15(GC, 0, regCP_CPC_IC_BASE_HI,
2145 			upper_32_bits(addr));
2146 
2147 	return 0;
2148 }
2149 
2150 static int gfx_v11_0_config_pfp_cache_rs64(struct amdgpu_device *adev, uint64_t addr, uint64_t addr2)
2151 {
2152 	uint32_t usec_timeout = 50000;  /* wait for 50ms */
2153 	uint32_t tmp;
2154 	unsigned i, pipe_id;
2155 	const struct gfx_firmware_header_v2_0 *pfp_hdr;
2156 
2157 	pfp_hdr = (const struct gfx_firmware_header_v2_0 *)
2158 		adev->gfx.pfp_fw->data;
2159 
2160 	WREG32_SOC15(GC, 0, regCP_PFP_IC_BASE_LO,
2161 		lower_32_bits(addr));
2162 	WREG32_SOC15(GC, 0, regCP_PFP_IC_BASE_HI,
2163 		upper_32_bits(addr));
2164 
2165 	tmp = RREG32_SOC15(GC, 0, regCP_PFP_IC_BASE_CNTL);
2166 	tmp = REG_SET_FIELD(tmp, CP_PFP_IC_BASE_CNTL, VMID, 0);
2167 	tmp = REG_SET_FIELD(tmp, CP_PFP_IC_BASE_CNTL, CACHE_POLICY, 0);
2168 	tmp = REG_SET_FIELD(tmp, CP_PFP_IC_BASE_CNTL, EXE_DISABLE, 0);
2169 	WREG32_SOC15(GC, 0, regCP_PFP_IC_BASE_CNTL, tmp);
2170 
2171 	/*
2172 	 * Programming any of the CP_PFP_IC_BASE registers
2173 	 * forces invalidation of the ME L1 I$. Wait for the
2174 	 * invalidation complete
2175 	 */
2176 	for (i = 0; i < usec_timeout; i++) {
2177 		tmp = RREG32_SOC15(GC, 0, regCP_PFP_IC_OP_CNTL);
2178 		if (1 == REG_GET_FIELD(tmp, CP_PFP_IC_OP_CNTL,
2179 			INVALIDATE_CACHE_COMPLETE))
2180 			break;
2181 		udelay(1);
2182 	}
2183 
2184 	if (i >= usec_timeout) {
2185 		dev_err(adev->dev, "failed to invalidate instruction cache\n");
2186 		return -EINVAL;
2187 	}
2188 
2189 	/* Prime the L1 instruction caches */
2190 	tmp = RREG32_SOC15(GC, 0, regCP_PFP_IC_OP_CNTL);
2191 	tmp = REG_SET_FIELD(tmp, CP_PFP_IC_OP_CNTL, PRIME_ICACHE, 1);
2192 	WREG32_SOC15(GC, 0, regCP_PFP_IC_OP_CNTL, tmp);
2193 	/* Waiting for cache primed*/
2194 	for (i = 0; i < usec_timeout; i++) {
2195 		tmp = RREG32_SOC15(GC, 0, regCP_PFP_IC_OP_CNTL);
2196 		if (1 == REG_GET_FIELD(tmp, CP_PFP_IC_OP_CNTL,
2197 			ICACHE_PRIMED))
2198 			break;
2199 		udelay(1);
2200 	}
2201 
2202 	if (i >= usec_timeout) {
2203 		dev_err(adev->dev, "failed to prime instruction cache\n");
2204 		return -EINVAL;
2205 	}
2206 
2207 	mutex_lock(&adev->srbm_mutex);
2208 	for (pipe_id = 0; pipe_id < adev->gfx.me.num_pipe_per_me; pipe_id++) {
2209 		soc21_grbm_select(adev, 0, pipe_id, 0, 0);
2210 		WREG32_SOC15(GC, 0, regCP_PFP_PRGRM_CNTR_START,
2211 			(pfp_hdr->ucode_start_addr_hi << 30) |
2212 			(pfp_hdr->ucode_start_addr_lo >> 2));
2213 		WREG32_SOC15(GC, 0, regCP_PFP_PRGRM_CNTR_START_HI,
2214 			pfp_hdr->ucode_start_addr_hi >> 2);
2215 
2216 		/*
2217 		 * Program CP_ME_CNTL to reset given PIPE to take
2218 		 * effect of CP_PFP_PRGRM_CNTR_START.
2219 		 */
2220 		tmp = RREG32_SOC15(GC, 0, regCP_ME_CNTL);
2221 		if (pipe_id == 0)
2222 			tmp = REG_SET_FIELD(tmp, CP_ME_CNTL,
2223 					PFP_PIPE0_RESET, 1);
2224 		else
2225 			tmp = REG_SET_FIELD(tmp, CP_ME_CNTL,
2226 					PFP_PIPE1_RESET, 1);
2227 		WREG32_SOC15(GC, 0, regCP_ME_CNTL, tmp);
2228 
2229 		/* Clear pfp pipe0 reset bit. */
2230 		if (pipe_id == 0)
2231 			tmp = REG_SET_FIELD(tmp, CP_ME_CNTL,
2232 					PFP_PIPE0_RESET, 0);
2233 		else
2234 			tmp = REG_SET_FIELD(tmp, CP_ME_CNTL,
2235 					PFP_PIPE1_RESET, 0);
2236 		WREG32_SOC15(GC, 0, regCP_ME_CNTL, tmp);
2237 
2238 		WREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_BASE0_LO,
2239 			lower_32_bits(addr2));
2240 		WREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_BASE0_HI,
2241 			upper_32_bits(addr2));
2242 	}
2243 	soc21_grbm_select(adev, 0, 0, 0, 0);
2244 	mutex_unlock(&adev->srbm_mutex);
2245 
2246 	tmp = RREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_BASE_CNTL);
2247 	tmp = REG_SET_FIELD(tmp, CP_GFX_RS64_DC_BASE_CNTL, VMID, 0);
2248 	tmp = REG_SET_FIELD(tmp, CP_GFX_RS64_DC_BASE_CNTL, CACHE_POLICY, 0);
2249 	WREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_BASE_CNTL, tmp);
2250 
2251 	/* Invalidate the data caches */
2252 	tmp = RREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_OP_CNTL);
2253 	tmp = REG_SET_FIELD(tmp, CP_GFX_RS64_DC_OP_CNTL, INVALIDATE_DCACHE, 1);
2254 	WREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_OP_CNTL, tmp);
2255 
2256 	for (i = 0; i < usec_timeout; i++) {
2257 		tmp = RREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_OP_CNTL);
2258 		if (1 == REG_GET_FIELD(tmp, CP_GFX_RS64_DC_OP_CNTL,
2259 			INVALIDATE_DCACHE_COMPLETE))
2260 			break;
2261 		udelay(1);
2262 	}
2263 
2264 	if (i >= usec_timeout) {
2265 		dev_err(adev->dev, "failed to invalidate RS64 data cache\n");
2266 		return -EINVAL;
2267 	}
2268 
2269 	return 0;
2270 }
2271 
2272 static int gfx_v11_0_config_me_cache_rs64(struct amdgpu_device *adev, uint64_t addr, uint64_t addr2)
2273 {
2274 	uint32_t usec_timeout = 50000;  /* wait for 50ms */
2275 	uint32_t tmp;
2276 	unsigned i, pipe_id;
2277 	const struct gfx_firmware_header_v2_0 *me_hdr;
2278 
2279 	me_hdr = (const struct gfx_firmware_header_v2_0 *)
2280 		adev->gfx.me_fw->data;
2281 
2282 	WREG32_SOC15(GC, 0, regCP_ME_IC_BASE_LO,
2283 		lower_32_bits(addr));
2284 	WREG32_SOC15(GC, 0, regCP_ME_IC_BASE_HI,
2285 		upper_32_bits(addr));
2286 
2287 	tmp = RREG32_SOC15(GC, 0, regCP_ME_IC_BASE_CNTL);
2288 	tmp = REG_SET_FIELD(tmp, CP_ME_IC_BASE_CNTL, VMID, 0);
2289 	tmp = REG_SET_FIELD(tmp, CP_ME_IC_BASE_CNTL, CACHE_POLICY, 0);
2290 	tmp = REG_SET_FIELD(tmp, CP_ME_IC_BASE_CNTL, EXE_DISABLE, 0);
2291 	WREG32_SOC15(GC, 0, regCP_ME_IC_BASE_CNTL, tmp);
2292 
2293 	/*
2294 	 * Programming any of the CP_ME_IC_BASE registers
2295 	 * forces invalidation of the ME L1 I$. Wait for the
2296 	 * invalidation complete
2297 	 */
2298 	for (i = 0; i < usec_timeout; i++) {
2299 		tmp = RREG32_SOC15(GC, 0, regCP_ME_IC_OP_CNTL);
2300 		if (1 == REG_GET_FIELD(tmp, CP_ME_IC_OP_CNTL,
2301 			INVALIDATE_CACHE_COMPLETE))
2302 			break;
2303 		udelay(1);
2304 	}
2305 
2306 	if (i >= usec_timeout) {
2307 		dev_err(adev->dev, "failed to invalidate instruction cache\n");
2308 		return -EINVAL;
2309 	}
2310 
2311 	/* Prime the instruction caches */
2312 	tmp = RREG32_SOC15(GC, 0, regCP_ME_IC_OP_CNTL);
2313 	tmp = REG_SET_FIELD(tmp, CP_ME_IC_OP_CNTL, PRIME_ICACHE, 1);
2314 	WREG32_SOC15(GC, 0, regCP_ME_IC_OP_CNTL, tmp);
2315 
2316 	/* Waiting for instruction cache primed*/
2317 	for (i = 0; i < usec_timeout; i++) {
2318 		tmp = RREG32_SOC15(GC, 0, regCP_ME_IC_OP_CNTL);
2319 		if (1 == REG_GET_FIELD(tmp, CP_ME_IC_OP_CNTL,
2320 			ICACHE_PRIMED))
2321 			break;
2322 		udelay(1);
2323 	}
2324 
2325 	if (i >= usec_timeout) {
2326 		dev_err(adev->dev, "failed to prime instruction cache\n");
2327 		return -EINVAL;
2328 	}
2329 
2330 	mutex_lock(&adev->srbm_mutex);
2331 	for (pipe_id = 0; pipe_id < adev->gfx.me.num_pipe_per_me; pipe_id++) {
2332 		soc21_grbm_select(adev, 0, pipe_id, 0, 0);
2333 		WREG32_SOC15(GC, 0, regCP_ME_PRGRM_CNTR_START,
2334 			(me_hdr->ucode_start_addr_hi << 30) |
2335 			(me_hdr->ucode_start_addr_lo >> 2) );
2336 		WREG32_SOC15(GC, 0, regCP_ME_PRGRM_CNTR_START_HI,
2337 			me_hdr->ucode_start_addr_hi>>2);
2338 
2339 		/*
2340 		 * Program CP_ME_CNTL to reset given PIPE to take
2341 		 * effect of CP_PFP_PRGRM_CNTR_START.
2342 		 */
2343 		tmp = RREG32_SOC15(GC, 0, regCP_ME_CNTL);
2344 		if (pipe_id == 0)
2345 			tmp = REG_SET_FIELD(tmp, CP_ME_CNTL,
2346 					ME_PIPE0_RESET, 1);
2347 		else
2348 			tmp = REG_SET_FIELD(tmp, CP_ME_CNTL,
2349 					ME_PIPE1_RESET, 1);
2350 		WREG32_SOC15(GC, 0, regCP_ME_CNTL, tmp);
2351 
2352 		/* Clear pfp pipe0 reset bit. */
2353 		if (pipe_id == 0)
2354 			tmp = REG_SET_FIELD(tmp, CP_ME_CNTL,
2355 					ME_PIPE0_RESET, 0);
2356 		else
2357 			tmp = REG_SET_FIELD(tmp, CP_ME_CNTL,
2358 					ME_PIPE1_RESET, 0);
2359 		WREG32_SOC15(GC, 0, regCP_ME_CNTL, tmp);
2360 
2361 		WREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_BASE1_LO,
2362 			lower_32_bits(addr2));
2363 		WREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_BASE1_HI,
2364 			upper_32_bits(addr2));
2365 	}
2366 	soc21_grbm_select(adev, 0, 0, 0, 0);
2367 	mutex_unlock(&adev->srbm_mutex);
2368 
2369 	tmp = RREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_BASE_CNTL);
2370 	tmp = REG_SET_FIELD(tmp, CP_GFX_RS64_DC_BASE_CNTL, VMID, 0);
2371 	tmp = REG_SET_FIELD(tmp, CP_GFX_RS64_DC_BASE_CNTL, CACHE_POLICY, 0);
2372 	WREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_BASE_CNTL, tmp);
2373 
2374 	/* Invalidate the data caches */
2375 	tmp = RREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_OP_CNTL);
2376 	tmp = REG_SET_FIELD(tmp, CP_GFX_RS64_DC_OP_CNTL, INVALIDATE_DCACHE, 1);
2377 	WREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_OP_CNTL, tmp);
2378 
2379 	for (i = 0; i < usec_timeout; i++) {
2380 		tmp = RREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_OP_CNTL);
2381 		if (1 == REG_GET_FIELD(tmp, CP_GFX_RS64_DC_OP_CNTL,
2382 			INVALIDATE_DCACHE_COMPLETE))
2383 			break;
2384 		udelay(1);
2385 	}
2386 
2387 	if (i >= usec_timeout) {
2388 		dev_err(adev->dev, "failed to invalidate RS64 data cache\n");
2389 		return -EINVAL;
2390 	}
2391 
2392 	return 0;
2393 }
2394 
2395 static int gfx_v11_0_config_mec_cache_rs64(struct amdgpu_device *adev, uint64_t addr, uint64_t addr2)
2396 {
2397 	uint32_t usec_timeout = 50000;  /* wait for 50ms */
2398 	uint32_t tmp;
2399 	unsigned i;
2400 	const struct gfx_firmware_header_v2_0 *mec_hdr;
2401 
2402 	mec_hdr = (const struct gfx_firmware_header_v2_0 *)
2403 		adev->gfx.mec_fw->data;
2404 
2405 	tmp = RREG32_SOC15(GC, 0, regCP_CPC_IC_BASE_CNTL);
2406 	tmp = REG_SET_FIELD(tmp, CP_CPC_IC_BASE_CNTL, VMID, 0);
2407 	tmp = REG_SET_FIELD(tmp, CP_CPC_IC_BASE_CNTL, EXE_DISABLE, 0);
2408 	tmp = REG_SET_FIELD(tmp, CP_CPC_IC_BASE_CNTL, CACHE_POLICY, 0);
2409 	WREG32_SOC15(GC, 0, regCP_CPC_IC_BASE_CNTL, tmp);
2410 
2411 	tmp = RREG32_SOC15(GC, 0, regCP_MEC_DC_BASE_CNTL);
2412 	tmp = REG_SET_FIELD(tmp, CP_MEC_DC_BASE_CNTL, VMID, 0);
2413 	tmp = REG_SET_FIELD(tmp, CP_MEC_DC_BASE_CNTL, CACHE_POLICY, 0);
2414 	WREG32_SOC15(GC, 0, regCP_MEC_DC_BASE_CNTL, tmp);
2415 
2416 	mutex_lock(&adev->srbm_mutex);
2417 	for (i = 0; i < adev->gfx.mec.num_pipe_per_mec; i++) {
2418 		soc21_grbm_select(adev, 1, i, 0, 0);
2419 
2420 		WREG32_SOC15(GC, 0, regCP_MEC_MDBASE_LO, addr2);
2421 		WREG32_SOC15(GC, 0, regCP_MEC_MDBASE_HI,
2422 		     upper_32_bits(addr2));
2423 
2424 		WREG32_SOC15(GC, 0, regCP_MEC_RS64_PRGRM_CNTR_START,
2425 					mec_hdr->ucode_start_addr_lo >> 2 |
2426 					mec_hdr->ucode_start_addr_hi << 30);
2427 		WREG32_SOC15(GC, 0, regCP_MEC_RS64_PRGRM_CNTR_START_HI,
2428 					mec_hdr->ucode_start_addr_hi >> 2);
2429 
2430 		WREG32_SOC15(GC, 0, regCP_CPC_IC_BASE_LO, addr);
2431 		WREG32_SOC15(GC, 0, regCP_CPC_IC_BASE_HI,
2432 		     upper_32_bits(addr));
2433 	}
2434 	mutex_unlock(&adev->srbm_mutex);
2435 	soc21_grbm_select(adev, 0, 0, 0, 0);
2436 
2437 	/* Trigger an invalidation of the L1 instruction caches */
2438 	tmp = RREG32_SOC15(GC, 0, regCP_MEC_DC_OP_CNTL);
2439 	tmp = REG_SET_FIELD(tmp, CP_MEC_DC_OP_CNTL, INVALIDATE_DCACHE, 1);
2440 	WREG32_SOC15(GC, 0, regCP_MEC_DC_OP_CNTL, tmp);
2441 
2442 	/* Wait for invalidation complete */
2443 	for (i = 0; i < usec_timeout; i++) {
2444 		tmp = RREG32_SOC15(GC, 0, regCP_MEC_DC_OP_CNTL);
2445 		if (1 == REG_GET_FIELD(tmp, CP_MEC_DC_OP_CNTL,
2446 				       INVALIDATE_DCACHE_COMPLETE))
2447 			break;
2448 		udelay(1);
2449 	}
2450 
2451 	if (i >= usec_timeout) {
2452 		dev_err(adev->dev, "failed to invalidate instruction cache\n");
2453 		return -EINVAL;
2454 	}
2455 
2456 	/* Trigger an invalidation of the L1 instruction caches */
2457 	tmp = RREG32_SOC15(GC, 0, regCP_CPC_IC_OP_CNTL);
2458 	tmp = REG_SET_FIELD(tmp, CP_CPC_IC_OP_CNTL, INVALIDATE_CACHE, 1);
2459 	WREG32_SOC15(GC, 0, regCP_CPC_IC_OP_CNTL, tmp);
2460 
2461 	/* Wait for invalidation complete */
2462 	for (i = 0; i < usec_timeout; i++) {
2463 		tmp = RREG32_SOC15(GC, 0, regCP_CPC_IC_OP_CNTL);
2464 		if (1 == REG_GET_FIELD(tmp, CP_CPC_IC_OP_CNTL,
2465 				       INVALIDATE_CACHE_COMPLETE))
2466 			break;
2467 		udelay(1);
2468 	}
2469 
2470 	if (i >= usec_timeout) {
2471 		dev_err(adev->dev, "failed to invalidate instruction cache\n");
2472 		return -EINVAL;
2473 	}
2474 
2475 	return 0;
2476 }
2477 
2478 static void gfx_v11_0_config_gfx_rs64(struct amdgpu_device *adev)
2479 {
2480 	const struct gfx_firmware_header_v2_0 *pfp_hdr;
2481 	const struct gfx_firmware_header_v2_0 *me_hdr;
2482 	const struct gfx_firmware_header_v2_0 *mec_hdr;
2483 	uint32_t pipe_id, tmp;
2484 
2485 	mec_hdr = (const struct gfx_firmware_header_v2_0 *)
2486 		adev->gfx.mec_fw->data;
2487 	me_hdr = (const struct gfx_firmware_header_v2_0 *)
2488 		adev->gfx.me_fw->data;
2489 	pfp_hdr = (const struct gfx_firmware_header_v2_0 *)
2490 		adev->gfx.pfp_fw->data;
2491 
2492 	/* config pfp program start addr */
2493 	for (pipe_id = 0; pipe_id < 2; pipe_id++) {
2494 		soc21_grbm_select(adev, 0, pipe_id, 0, 0);
2495 		WREG32_SOC15(GC, 0, regCP_PFP_PRGRM_CNTR_START,
2496 			(pfp_hdr->ucode_start_addr_hi << 30) |
2497 			(pfp_hdr->ucode_start_addr_lo >> 2));
2498 		WREG32_SOC15(GC, 0, regCP_PFP_PRGRM_CNTR_START_HI,
2499 			pfp_hdr->ucode_start_addr_hi >> 2);
2500 	}
2501 	soc21_grbm_select(adev, 0, 0, 0, 0);
2502 
2503 	/* reset pfp pipe */
2504 	tmp = RREG32_SOC15(GC, 0, regCP_ME_CNTL);
2505 	tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, PFP_PIPE0_RESET, 1);
2506 	tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, PFP_PIPE1_RESET, 1);
2507 	WREG32_SOC15(GC, 0, regCP_ME_CNTL, tmp);
2508 
2509 	/* clear pfp pipe reset */
2510 	tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, PFP_PIPE0_RESET, 0);
2511 	tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, PFP_PIPE1_RESET, 0);
2512 	WREG32_SOC15(GC, 0, regCP_ME_CNTL, tmp);
2513 
2514 	/* config me program start addr */
2515 	for (pipe_id = 0; pipe_id < 2; pipe_id++) {
2516 		soc21_grbm_select(adev, 0, pipe_id, 0, 0);
2517 		WREG32_SOC15(GC, 0, regCP_ME_PRGRM_CNTR_START,
2518 			(me_hdr->ucode_start_addr_hi << 30) |
2519 			(me_hdr->ucode_start_addr_lo >> 2) );
2520 		WREG32_SOC15(GC, 0, regCP_ME_PRGRM_CNTR_START_HI,
2521 			me_hdr->ucode_start_addr_hi>>2);
2522 	}
2523 	soc21_grbm_select(adev, 0, 0, 0, 0);
2524 
2525 	/* reset me pipe */
2526 	tmp = RREG32_SOC15(GC, 0, regCP_ME_CNTL);
2527 	tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, ME_PIPE0_RESET, 1);
2528 	tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, ME_PIPE1_RESET, 1);
2529 	WREG32_SOC15(GC, 0, regCP_ME_CNTL, tmp);
2530 
2531 	/* clear me pipe reset */
2532 	tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, ME_PIPE0_RESET, 0);
2533 	tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, ME_PIPE1_RESET, 0);
2534 	WREG32_SOC15(GC, 0, regCP_ME_CNTL, tmp);
2535 
2536 	/* config mec program start addr */
2537 	for (pipe_id = 0; pipe_id < 4; pipe_id++) {
2538 		soc21_grbm_select(adev, 1, pipe_id, 0, 0);
2539 		WREG32_SOC15(GC, 0, regCP_MEC_RS64_PRGRM_CNTR_START,
2540 					mec_hdr->ucode_start_addr_lo >> 2 |
2541 					mec_hdr->ucode_start_addr_hi << 30);
2542 		WREG32_SOC15(GC, 0, regCP_MEC_RS64_PRGRM_CNTR_START_HI,
2543 					mec_hdr->ucode_start_addr_hi >> 2);
2544 	}
2545 	soc21_grbm_select(adev, 0, 0, 0, 0);
2546 
2547 	/* reset mec pipe */
2548 	tmp = RREG32_SOC15(GC, 0, regCP_MEC_RS64_CNTL);
2549 	tmp = REG_SET_FIELD(tmp, CP_MEC_RS64_CNTL, MEC_PIPE0_RESET, 1);
2550 	tmp = REG_SET_FIELD(tmp, CP_MEC_RS64_CNTL, MEC_PIPE1_RESET, 1);
2551 	tmp = REG_SET_FIELD(tmp, CP_MEC_RS64_CNTL, MEC_PIPE2_RESET, 1);
2552 	tmp = REG_SET_FIELD(tmp, CP_MEC_RS64_CNTL, MEC_PIPE3_RESET, 1);
2553 	WREG32_SOC15(GC, 0, regCP_MEC_RS64_CNTL, tmp);
2554 
2555 	/* clear mec pipe reset */
2556 	tmp = REG_SET_FIELD(tmp, CP_MEC_RS64_CNTL, MEC_PIPE0_RESET, 0);
2557 	tmp = REG_SET_FIELD(tmp, CP_MEC_RS64_CNTL, MEC_PIPE1_RESET, 0);
2558 	tmp = REG_SET_FIELD(tmp, CP_MEC_RS64_CNTL, MEC_PIPE2_RESET, 0);
2559 	tmp = REG_SET_FIELD(tmp, CP_MEC_RS64_CNTL, MEC_PIPE3_RESET, 0);
2560 	WREG32_SOC15(GC, 0, regCP_MEC_RS64_CNTL, tmp);
2561 }
2562 
2563 static int gfx_v11_0_wait_for_rlc_autoload_complete(struct amdgpu_device *adev)
2564 {
2565 	uint32_t cp_status;
2566 	uint32_t bootload_status;
2567 	int i, r;
2568 	uint64_t addr, addr2;
2569 
2570 	for (i = 0; i < adev->usec_timeout; i++) {
2571 		cp_status = RREG32_SOC15(GC, 0, regCP_STAT);
2572 
2573 		if (adev->ip_versions[GC_HWIP][0] == IP_VERSION(11, 0, 1) ||
2574 				adev->ip_versions[GC_HWIP][0] == IP_VERSION(11, 0, 4))
2575 			bootload_status = RREG32_SOC15(GC, 0,
2576 					regRLC_RLCS_BOOTLOAD_STATUS_gc_11_0_1);
2577 		else
2578 			bootload_status = RREG32_SOC15(GC, 0, regRLC_RLCS_BOOTLOAD_STATUS);
2579 
2580 		if ((cp_status == 0) &&
2581 		    (REG_GET_FIELD(bootload_status,
2582 			RLC_RLCS_BOOTLOAD_STATUS, BOOTLOAD_COMPLETE) == 1)) {
2583 			break;
2584 		}
2585 		udelay(1);
2586 	}
2587 
2588 	if (i >= adev->usec_timeout) {
2589 		dev_err(adev->dev, "rlc autoload: gc ucode autoload timeout\n");
2590 		return -ETIMEDOUT;
2591 	}
2592 
2593 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_RLC_BACKDOOR_AUTO) {
2594 		if (adev->gfx.rs64_enable) {
2595 			addr = adev->gfx.rlc.rlc_autoload_gpu_addr +
2596 				rlc_autoload_info[SOC21_FIRMWARE_ID_RS64_ME].offset;
2597 			addr2 = adev->gfx.rlc.rlc_autoload_gpu_addr +
2598 				rlc_autoload_info[SOC21_FIRMWARE_ID_RS64_ME_P0_STACK].offset;
2599 			r = gfx_v11_0_config_me_cache_rs64(adev, addr, addr2);
2600 			if (r)
2601 				return r;
2602 			addr = adev->gfx.rlc.rlc_autoload_gpu_addr +
2603 				rlc_autoload_info[SOC21_FIRMWARE_ID_RS64_PFP].offset;
2604 			addr2 = adev->gfx.rlc.rlc_autoload_gpu_addr +
2605 				rlc_autoload_info[SOC21_FIRMWARE_ID_RS64_PFP_P0_STACK].offset;
2606 			r = gfx_v11_0_config_pfp_cache_rs64(adev, addr, addr2);
2607 			if (r)
2608 				return r;
2609 			addr = adev->gfx.rlc.rlc_autoload_gpu_addr +
2610 				rlc_autoload_info[SOC21_FIRMWARE_ID_RS64_MEC].offset;
2611 			addr2 = adev->gfx.rlc.rlc_autoload_gpu_addr +
2612 				rlc_autoload_info[SOC21_FIRMWARE_ID_RS64_MEC_P0_STACK].offset;
2613 			r = gfx_v11_0_config_mec_cache_rs64(adev, addr, addr2);
2614 			if (r)
2615 				return r;
2616 		} else {
2617 			addr = adev->gfx.rlc.rlc_autoload_gpu_addr +
2618 				rlc_autoload_info[SOC21_FIRMWARE_ID_CP_ME].offset;
2619 			r = gfx_v11_0_config_me_cache(adev, addr);
2620 			if (r)
2621 				return r;
2622 			addr = adev->gfx.rlc.rlc_autoload_gpu_addr +
2623 				rlc_autoload_info[SOC21_FIRMWARE_ID_CP_PFP].offset;
2624 			r = gfx_v11_0_config_pfp_cache(adev, addr);
2625 			if (r)
2626 				return r;
2627 			addr = adev->gfx.rlc.rlc_autoload_gpu_addr +
2628 				rlc_autoload_info[SOC21_FIRMWARE_ID_CP_MEC].offset;
2629 			r = gfx_v11_0_config_mec_cache(adev, addr);
2630 			if (r)
2631 				return r;
2632 		}
2633 	}
2634 
2635 	return 0;
2636 }
2637 
2638 static int gfx_v11_0_cp_gfx_enable(struct amdgpu_device *adev, bool enable)
2639 {
2640 	int i;
2641 	u32 tmp = RREG32_SOC15(GC, 0, regCP_ME_CNTL);
2642 
2643 	tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, ME_HALT, enable ? 0 : 1);
2644 	tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, PFP_HALT, enable ? 0 : 1);
2645 	WREG32_SOC15(GC, 0, regCP_ME_CNTL, tmp);
2646 
2647 	for (i = 0; i < adev->usec_timeout; i++) {
2648 		if (RREG32_SOC15(GC, 0, regCP_STAT) == 0)
2649 			break;
2650 		udelay(1);
2651 	}
2652 
2653 	if (i >= adev->usec_timeout)
2654 		DRM_ERROR("failed to %s cp gfx\n", enable ? "unhalt" : "halt");
2655 
2656 	return 0;
2657 }
2658 
2659 static int gfx_v11_0_cp_gfx_load_pfp_microcode(struct amdgpu_device *adev)
2660 {
2661 	int r;
2662 	const struct gfx_firmware_header_v1_0 *pfp_hdr;
2663 	const __le32 *fw_data;
2664 	unsigned i, fw_size;
2665 
2666 	pfp_hdr = (const struct gfx_firmware_header_v1_0 *)
2667 		adev->gfx.pfp_fw->data;
2668 
2669 	amdgpu_ucode_print_gfx_hdr(&pfp_hdr->header);
2670 
2671 	fw_data = (const __le32 *)(adev->gfx.pfp_fw->data +
2672 		le32_to_cpu(pfp_hdr->header.ucode_array_offset_bytes));
2673 	fw_size = le32_to_cpu(pfp_hdr->header.ucode_size_bytes);
2674 
2675 	r = amdgpu_bo_create_reserved(adev, pfp_hdr->header.ucode_size_bytes,
2676 				      PAGE_SIZE, AMDGPU_GEM_DOMAIN_GTT,
2677 				      &adev->gfx.pfp.pfp_fw_obj,
2678 				      &adev->gfx.pfp.pfp_fw_gpu_addr,
2679 				      (void **)&adev->gfx.pfp.pfp_fw_ptr);
2680 	if (r) {
2681 		dev_err(adev->dev, "(%d) failed to create pfp fw bo\n", r);
2682 		gfx_v11_0_pfp_fini(adev);
2683 		return r;
2684 	}
2685 
2686 	memcpy(adev->gfx.pfp.pfp_fw_ptr, fw_data, fw_size);
2687 
2688 	amdgpu_bo_kunmap(adev->gfx.pfp.pfp_fw_obj);
2689 	amdgpu_bo_unreserve(adev->gfx.pfp.pfp_fw_obj);
2690 
2691 	gfx_v11_0_config_pfp_cache(adev, adev->gfx.pfp.pfp_fw_gpu_addr);
2692 
2693 	WREG32_SOC15(GC, 0, regCP_HYP_PFP_UCODE_ADDR, 0);
2694 
2695 	for (i = 0; i < pfp_hdr->jt_size; i++)
2696 		WREG32_SOC15(GC, 0, regCP_HYP_PFP_UCODE_DATA,
2697 			     le32_to_cpup(fw_data + pfp_hdr->jt_offset + i));
2698 
2699 	WREG32_SOC15(GC, 0, regCP_HYP_PFP_UCODE_ADDR, adev->gfx.pfp_fw_version);
2700 
2701 	return 0;
2702 }
2703 
2704 static int gfx_v11_0_cp_gfx_load_pfp_microcode_rs64(struct amdgpu_device *adev)
2705 {
2706 	int r;
2707 	const struct gfx_firmware_header_v2_0 *pfp_hdr;
2708 	const __le32 *fw_ucode, *fw_data;
2709 	unsigned i, pipe_id, fw_ucode_size, fw_data_size;
2710 	uint32_t tmp;
2711 	uint32_t usec_timeout = 50000;  /* wait for 50ms */
2712 
2713 	pfp_hdr = (const struct gfx_firmware_header_v2_0 *)
2714 		adev->gfx.pfp_fw->data;
2715 
2716 	amdgpu_ucode_print_gfx_hdr(&pfp_hdr->header);
2717 
2718 	/* instruction */
2719 	fw_ucode = (const __le32 *)(adev->gfx.pfp_fw->data +
2720 		le32_to_cpu(pfp_hdr->ucode_offset_bytes));
2721 	fw_ucode_size = le32_to_cpu(pfp_hdr->ucode_size_bytes);
2722 	/* data */
2723 	fw_data = (const __le32 *)(adev->gfx.pfp_fw->data +
2724 		le32_to_cpu(pfp_hdr->data_offset_bytes));
2725 	fw_data_size = le32_to_cpu(pfp_hdr->data_size_bytes);
2726 
2727 	/* 64kb align */
2728 	r = amdgpu_bo_create_reserved(adev, fw_ucode_size,
2729 				      64 * 1024,
2730 				      AMDGPU_GEM_DOMAIN_VRAM |
2731 				      AMDGPU_GEM_DOMAIN_GTT,
2732 				      &adev->gfx.pfp.pfp_fw_obj,
2733 				      &adev->gfx.pfp.pfp_fw_gpu_addr,
2734 				      (void **)&adev->gfx.pfp.pfp_fw_ptr);
2735 	if (r) {
2736 		dev_err(adev->dev, "(%d) failed to create pfp ucode fw bo\n", r);
2737 		gfx_v11_0_pfp_fini(adev);
2738 		return r;
2739 	}
2740 
2741 	r = amdgpu_bo_create_reserved(adev, fw_data_size,
2742 				      64 * 1024,
2743 				      AMDGPU_GEM_DOMAIN_VRAM |
2744 				      AMDGPU_GEM_DOMAIN_GTT,
2745 				      &adev->gfx.pfp.pfp_fw_data_obj,
2746 				      &adev->gfx.pfp.pfp_fw_data_gpu_addr,
2747 				      (void **)&adev->gfx.pfp.pfp_fw_data_ptr);
2748 	if (r) {
2749 		dev_err(adev->dev, "(%d) failed to create pfp data fw bo\n", r);
2750 		gfx_v11_0_pfp_fini(adev);
2751 		return r;
2752 	}
2753 
2754 	memcpy(adev->gfx.pfp.pfp_fw_ptr, fw_ucode, fw_ucode_size);
2755 	memcpy(adev->gfx.pfp.pfp_fw_data_ptr, fw_data, fw_data_size);
2756 
2757 	amdgpu_bo_kunmap(adev->gfx.pfp.pfp_fw_obj);
2758 	amdgpu_bo_kunmap(adev->gfx.pfp.pfp_fw_data_obj);
2759 	amdgpu_bo_unreserve(adev->gfx.pfp.pfp_fw_obj);
2760 	amdgpu_bo_unreserve(adev->gfx.pfp.pfp_fw_data_obj);
2761 
2762 	if (amdgpu_emu_mode == 1)
2763 		adev->hdp.funcs->flush_hdp(adev, NULL);
2764 
2765 	WREG32_SOC15(GC, 0, regCP_PFP_IC_BASE_LO,
2766 		lower_32_bits(adev->gfx.pfp.pfp_fw_gpu_addr));
2767 	WREG32_SOC15(GC, 0, regCP_PFP_IC_BASE_HI,
2768 		upper_32_bits(adev->gfx.pfp.pfp_fw_gpu_addr));
2769 
2770 	tmp = RREG32_SOC15(GC, 0, regCP_PFP_IC_BASE_CNTL);
2771 	tmp = REG_SET_FIELD(tmp, CP_PFP_IC_BASE_CNTL, VMID, 0);
2772 	tmp = REG_SET_FIELD(tmp, CP_PFP_IC_BASE_CNTL, CACHE_POLICY, 0);
2773 	tmp = REG_SET_FIELD(tmp, CP_PFP_IC_BASE_CNTL, EXE_DISABLE, 0);
2774 	WREG32_SOC15(GC, 0, regCP_PFP_IC_BASE_CNTL, tmp);
2775 
2776 	/*
2777 	 * Programming any of the CP_PFP_IC_BASE registers
2778 	 * forces invalidation of the ME L1 I$. Wait for the
2779 	 * invalidation complete
2780 	 */
2781 	for (i = 0; i < usec_timeout; i++) {
2782 		tmp = RREG32_SOC15(GC, 0, regCP_PFP_IC_OP_CNTL);
2783 		if (1 == REG_GET_FIELD(tmp, CP_PFP_IC_OP_CNTL,
2784 			INVALIDATE_CACHE_COMPLETE))
2785 			break;
2786 		udelay(1);
2787 	}
2788 
2789 	if (i >= usec_timeout) {
2790 		dev_err(adev->dev, "failed to invalidate instruction cache\n");
2791 		return -EINVAL;
2792 	}
2793 
2794 	/* Prime the L1 instruction caches */
2795 	tmp = RREG32_SOC15(GC, 0, regCP_PFP_IC_OP_CNTL);
2796 	tmp = REG_SET_FIELD(tmp, CP_PFP_IC_OP_CNTL, PRIME_ICACHE, 1);
2797 	WREG32_SOC15(GC, 0, regCP_PFP_IC_OP_CNTL, tmp);
2798 	/* Waiting for cache primed*/
2799 	for (i = 0; i < usec_timeout; i++) {
2800 		tmp = RREG32_SOC15(GC, 0, regCP_PFP_IC_OP_CNTL);
2801 		if (1 == REG_GET_FIELD(tmp, CP_PFP_IC_OP_CNTL,
2802 			ICACHE_PRIMED))
2803 			break;
2804 		udelay(1);
2805 	}
2806 
2807 	if (i >= usec_timeout) {
2808 		dev_err(adev->dev, "failed to prime instruction cache\n");
2809 		return -EINVAL;
2810 	}
2811 
2812 	mutex_lock(&adev->srbm_mutex);
2813 	for (pipe_id = 0; pipe_id < adev->gfx.me.num_pipe_per_me; pipe_id++) {
2814 		soc21_grbm_select(adev, 0, pipe_id, 0, 0);
2815 		WREG32_SOC15(GC, 0, regCP_PFP_PRGRM_CNTR_START,
2816 			(pfp_hdr->ucode_start_addr_hi << 30) |
2817 			(pfp_hdr->ucode_start_addr_lo >> 2) );
2818 		WREG32_SOC15(GC, 0, regCP_PFP_PRGRM_CNTR_START_HI,
2819 			pfp_hdr->ucode_start_addr_hi>>2);
2820 
2821 		/*
2822 		 * Program CP_ME_CNTL to reset given PIPE to take
2823 		 * effect of CP_PFP_PRGRM_CNTR_START.
2824 		 */
2825 		tmp = RREG32_SOC15(GC, 0, regCP_ME_CNTL);
2826 		if (pipe_id == 0)
2827 			tmp = REG_SET_FIELD(tmp, CP_ME_CNTL,
2828 					PFP_PIPE0_RESET, 1);
2829 		else
2830 			tmp = REG_SET_FIELD(tmp, CP_ME_CNTL,
2831 					PFP_PIPE1_RESET, 1);
2832 		WREG32_SOC15(GC, 0, regCP_ME_CNTL, tmp);
2833 
2834 		/* Clear pfp pipe0 reset bit. */
2835 		if (pipe_id == 0)
2836 			tmp = REG_SET_FIELD(tmp, CP_ME_CNTL,
2837 					PFP_PIPE0_RESET, 0);
2838 		else
2839 			tmp = REG_SET_FIELD(tmp, CP_ME_CNTL,
2840 					PFP_PIPE1_RESET, 0);
2841 		WREG32_SOC15(GC, 0, regCP_ME_CNTL, tmp);
2842 
2843 		WREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_BASE0_LO,
2844 			lower_32_bits(adev->gfx.pfp.pfp_fw_data_gpu_addr));
2845 		WREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_BASE0_HI,
2846 			upper_32_bits(adev->gfx.pfp.pfp_fw_data_gpu_addr));
2847 	}
2848 	soc21_grbm_select(adev, 0, 0, 0, 0);
2849 	mutex_unlock(&adev->srbm_mutex);
2850 
2851 	tmp = RREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_BASE_CNTL);
2852 	tmp = REG_SET_FIELD(tmp, CP_GFX_RS64_DC_BASE_CNTL, VMID, 0);
2853 	tmp = REG_SET_FIELD(tmp, CP_GFX_RS64_DC_BASE_CNTL, CACHE_POLICY, 0);
2854 	WREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_BASE_CNTL, tmp);
2855 
2856 	/* Invalidate the data caches */
2857 	tmp = RREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_OP_CNTL);
2858 	tmp = REG_SET_FIELD(tmp, CP_GFX_RS64_DC_OP_CNTL, INVALIDATE_DCACHE, 1);
2859 	WREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_OP_CNTL, tmp);
2860 
2861 	for (i = 0; i < usec_timeout; i++) {
2862 		tmp = RREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_OP_CNTL);
2863 		if (1 == REG_GET_FIELD(tmp, CP_GFX_RS64_DC_OP_CNTL,
2864 			INVALIDATE_DCACHE_COMPLETE))
2865 			break;
2866 		udelay(1);
2867 	}
2868 
2869 	if (i >= usec_timeout) {
2870 		dev_err(adev->dev, "failed to invalidate RS64 data cache\n");
2871 		return -EINVAL;
2872 	}
2873 
2874 	return 0;
2875 }
2876 
2877 static int gfx_v11_0_cp_gfx_load_me_microcode(struct amdgpu_device *adev)
2878 {
2879 	int r;
2880 	const struct gfx_firmware_header_v1_0 *me_hdr;
2881 	const __le32 *fw_data;
2882 	unsigned i, fw_size;
2883 
2884 	me_hdr = (const struct gfx_firmware_header_v1_0 *)
2885 		adev->gfx.me_fw->data;
2886 
2887 	amdgpu_ucode_print_gfx_hdr(&me_hdr->header);
2888 
2889 	fw_data = (const __le32 *)(adev->gfx.me_fw->data +
2890 		le32_to_cpu(me_hdr->header.ucode_array_offset_bytes));
2891 	fw_size = le32_to_cpu(me_hdr->header.ucode_size_bytes);
2892 
2893 	r = amdgpu_bo_create_reserved(adev, me_hdr->header.ucode_size_bytes,
2894 				      PAGE_SIZE, AMDGPU_GEM_DOMAIN_GTT,
2895 				      &adev->gfx.me.me_fw_obj,
2896 				      &adev->gfx.me.me_fw_gpu_addr,
2897 				      (void **)&adev->gfx.me.me_fw_ptr);
2898 	if (r) {
2899 		dev_err(adev->dev, "(%d) failed to create me fw bo\n", r);
2900 		gfx_v11_0_me_fini(adev);
2901 		return r;
2902 	}
2903 
2904 	memcpy(adev->gfx.me.me_fw_ptr, fw_data, fw_size);
2905 
2906 	amdgpu_bo_kunmap(adev->gfx.me.me_fw_obj);
2907 	amdgpu_bo_unreserve(adev->gfx.me.me_fw_obj);
2908 
2909 	gfx_v11_0_config_me_cache(adev, adev->gfx.me.me_fw_gpu_addr);
2910 
2911 	WREG32_SOC15(GC, 0, regCP_HYP_ME_UCODE_ADDR, 0);
2912 
2913 	for (i = 0; i < me_hdr->jt_size; i++)
2914 		WREG32_SOC15(GC, 0, regCP_HYP_ME_UCODE_DATA,
2915 			     le32_to_cpup(fw_data + me_hdr->jt_offset + i));
2916 
2917 	WREG32_SOC15(GC, 0, regCP_HYP_ME_UCODE_ADDR, adev->gfx.me_fw_version);
2918 
2919 	return 0;
2920 }
2921 
2922 static int gfx_v11_0_cp_gfx_load_me_microcode_rs64(struct amdgpu_device *adev)
2923 {
2924 	int r;
2925 	const struct gfx_firmware_header_v2_0 *me_hdr;
2926 	const __le32 *fw_ucode, *fw_data;
2927 	unsigned i, pipe_id, fw_ucode_size, fw_data_size;
2928 	uint32_t tmp;
2929 	uint32_t usec_timeout = 50000;  /* wait for 50ms */
2930 
2931 	me_hdr = (const struct gfx_firmware_header_v2_0 *)
2932 		adev->gfx.me_fw->data;
2933 
2934 	amdgpu_ucode_print_gfx_hdr(&me_hdr->header);
2935 
2936 	/* instruction */
2937 	fw_ucode = (const __le32 *)(adev->gfx.me_fw->data +
2938 		le32_to_cpu(me_hdr->ucode_offset_bytes));
2939 	fw_ucode_size = le32_to_cpu(me_hdr->ucode_size_bytes);
2940 	/* data */
2941 	fw_data = (const __le32 *)(adev->gfx.me_fw->data +
2942 		le32_to_cpu(me_hdr->data_offset_bytes));
2943 	fw_data_size = le32_to_cpu(me_hdr->data_size_bytes);
2944 
2945 	/* 64kb align*/
2946 	r = amdgpu_bo_create_reserved(adev, fw_ucode_size,
2947 				      64 * 1024,
2948 				      AMDGPU_GEM_DOMAIN_VRAM |
2949 				      AMDGPU_GEM_DOMAIN_GTT,
2950 				      &adev->gfx.me.me_fw_obj,
2951 				      &adev->gfx.me.me_fw_gpu_addr,
2952 				      (void **)&adev->gfx.me.me_fw_ptr);
2953 	if (r) {
2954 		dev_err(adev->dev, "(%d) failed to create me ucode bo\n", r);
2955 		gfx_v11_0_me_fini(adev);
2956 		return r;
2957 	}
2958 
2959 	r = amdgpu_bo_create_reserved(adev, fw_data_size,
2960 				      64 * 1024,
2961 				      AMDGPU_GEM_DOMAIN_VRAM |
2962 				      AMDGPU_GEM_DOMAIN_GTT,
2963 				      &adev->gfx.me.me_fw_data_obj,
2964 				      &adev->gfx.me.me_fw_data_gpu_addr,
2965 				      (void **)&adev->gfx.me.me_fw_data_ptr);
2966 	if (r) {
2967 		dev_err(adev->dev, "(%d) failed to create me data bo\n", r);
2968 		gfx_v11_0_pfp_fini(adev);
2969 		return r;
2970 	}
2971 
2972 	memcpy(adev->gfx.me.me_fw_ptr, fw_ucode, fw_ucode_size);
2973 	memcpy(adev->gfx.me.me_fw_data_ptr, fw_data, fw_data_size);
2974 
2975 	amdgpu_bo_kunmap(adev->gfx.me.me_fw_obj);
2976 	amdgpu_bo_kunmap(adev->gfx.me.me_fw_data_obj);
2977 	amdgpu_bo_unreserve(adev->gfx.me.me_fw_obj);
2978 	amdgpu_bo_unreserve(adev->gfx.me.me_fw_data_obj);
2979 
2980 	if (amdgpu_emu_mode == 1)
2981 		adev->hdp.funcs->flush_hdp(adev, NULL);
2982 
2983 	WREG32_SOC15(GC, 0, regCP_ME_IC_BASE_LO,
2984 		lower_32_bits(adev->gfx.me.me_fw_gpu_addr));
2985 	WREG32_SOC15(GC, 0, regCP_ME_IC_BASE_HI,
2986 		upper_32_bits(adev->gfx.me.me_fw_gpu_addr));
2987 
2988 	tmp = RREG32_SOC15(GC, 0, regCP_ME_IC_BASE_CNTL);
2989 	tmp = REG_SET_FIELD(tmp, CP_ME_IC_BASE_CNTL, VMID, 0);
2990 	tmp = REG_SET_FIELD(tmp, CP_ME_IC_BASE_CNTL, CACHE_POLICY, 0);
2991 	tmp = REG_SET_FIELD(tmp, CP_ME_IC_BASE_CNTL, EXE_DISABLE, 0);
2992 	WREG32_SOC15(GC, 0, regCP_ME_IC_BASE_CNTL, tmp);
2993 
2994 	/*
2995 	 * Programming any of the CP_ME_IC_BASE registers
2996 	 * forces invalidation of the ME L1 I$. Wait for the
2997 	 * invalidation complete
2998 	 */
2999 	for (i = 0; i < usec_timeout; i++) {
3000 		tmp = RREG32_SOC15(GC, 0, regCP_ME_IC_OP_CNTL);
3001 		if (1 == REG_GET_FIELD(tmp, CP_ME_IC_OP_CNTL,
3002 			INVALIDATE_CACHE_COMPLETE))
3003 			break;
3004 		udelay(1);
3005 	}
3006 
3007 	if (i >= usec_timeout) {
3008 		dev_err(adev->dev, "failed to invalidate instruction cache\n");
3009 		return -EINVAL;
3010 	}
3011 
3012 	/* Prime the instruction caches */
3013 	tmp = RREG32_SOC15(GC, 0, regCP_ME_IC_OP_CNTL);
3014 	tmp = REG_SET_FIELD(tmp, CP_ME_IC_OP_CNTL, PRIME_ICACHE, 1);
3015 	WREG32_SOC15(GC, 0, regCP_ME_IC_OP_CNTL, tmp);
3016 
3017 	/* Waiting for instruction cache primed*/
3018 	for (i = 0; i < usec_timeout; i++) {
3019 		tmp = RREG32_SOC15(GC, 0, regCP_ME_IC_OP_CNTL);
3020 		if (1 == REG_GET_FIELD(tmp, CP_ME_IC_OP_CNTL,
3021 			ICACHE_PRIMED))
3022 			break;
3023 		udelay(1);
3024 	}
3025 
3026 	if (i >= usec_timeout) {
3027 		dev_err(adev->dev, "failed to prime instruction cache\n");
3028 		return -EINVAL;
3029 	}
3030 
3031 	mutex_lock(&adev->srbm_mutex);
3032 	for (pipe_id = 0; pipe_id < adev->gfx.me.num_pipe_per_me; pipe_id++) {
3033 		soc21_grbm_select(adev, 0, pipe_id, 0, 0);
3034 		WREG32_SOC15(GC, 0, regCP_ME_PRGRM_CNTR_START,
3035 			(me_hdr->ucode_start_addr_hi << 30) |
3036 			(me_hdr->ucode_start_addr_lo >> 2) );
3037 		WREG32_SOC15(GC, 0, regCP_ME_PRGRM_CNTR_START_HI,
3038 			me_hdr->ucode_start_addr_hi>>2);
3039 
3040 		/*
3041 		 * Program CP_ME_CNTL to reset given PIPE to take
3042 		 * effect of CP_PFP_PRGRM_CNTR_START.
3043 		 */
3044 		tmp = RREG32_SOC15(GC, 0, regCP_ME_CNTL);
3045 		if (pipe_id == 0)
3046 			tmp = REG_SET_FIELD(tmp, CP_ME_CNTL,
3047 					ME_PIPE0_RESET, 1);
3048 		else
3049 			tmp = REG_SET_FIELD(tmp, CP_ME_CNTL,
3050 					ME_PIPE1_RESET, 1);
3051 		WREG32_SOC15(GC, 0, regCP_ME_CNTL, tmp);
3052 
3053 		/* Clear pfp pipe0 reset bit. */
3054 		if (pipe_id == 0)
3055 			tmp = REG_SET_FIELD(tmp, CP_ME_CNTL,
3056 					ME_PIPE0_RESET, 0);
3057 		else
3058 			tmp = REG_SET_FIELD(tmp, CP_ME_CNTL,
3059 					ME_PIPE1_RESET, 0);
3060 		WREG32_SOC15(GC, 0, regCP_ME_CNTL, tmp);
3061 
3062 		WREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_BASE1_LO,
3063 			lower_32_bits(adev->gfx.me.me_fw_data_gpu_addr));
3064 		WREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_BASE1_HI,
3065 			upper_32_bits(adev->gfx.me.me_fw_data_gpu_addr));
3066 	}
3067 	soc21_grbm_select(adev, 0, 0, 0, 0);
3068 	mutex_unlock(&adev->srbm_mutex);
3069 
3070 	tmp = RREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_BASE_CNTL);
3071 	tmp = REG_SET_FIELD(tmp, CP_GFX_RS64_DC_BASE_CNTL, VMID, 0);
3072 	tmp = REG_SET_FIELD(tmp, CP_GFX_RS64_DC_BASE_CNTL, CACHE_POLICY, 0);
3073 	WREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_BASE_CNTL, tmp);
3074 
3075 	/* Invalidate the data caches */
3076 	tmp = RREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_OP_CNTL);
3077 	tmp = REG_SET_FIELD(tmp, CP_GFX_RS64_DC_OP_CNTL, INVALIDATE_DCACHE, 1);
3078 	WREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_OP_CNTL, tmp);
3079 
3080 	for (i = 0; i < usec_timeout; i++) {
3081 		tmp = RREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_OP_CNTL);
3082 		if (1 == REG_GET_FIELD(tmp, CP_GFX_RS64_DC_OP_CNTL,
3083 			INVALIDATE_DCACHE_COMPLETE))
3084 			break;
3085 		udelay(1);
3086 	}
3087 
3088 	if (i >= usec_timeout) {
3089 		dev_err(adev->dev, "failed to invalidate RS64 data cache\n");
3090 		return -EINVAL;
3091 	}
3092 
3093 	return 0;
3094 }
3095 
3096 static int gfx_v11_0_cp_gfx_load_microcode(struct amdgpu_device *adev)
3097 {
3098 	int r;
3099 
3100 	if (!adev->gfx.me_fw || !adev->gfx.pfp_fw)
3101 		return -EINVAL;
3102 
3103 	gfx_v11_0_cp_gfx_enable(adev, false);
3104 
3105 	if (adev->gfx.rs64_enable)
3106 		r = gfx_v11_0_cp_gfx_load_pfp_microcode_rs64(adev);
3107 	else
3108 		r = gfx_v11_0_cp_gfx_load_pfp_microcode(adev);
3109 	if (r) {
3110 		dev_err(adev->dev, "(%d) failed to load pfp fw\n", r);
3111 		return r;
3112 	}
3113 
3114 	if (adev->gfx.rs64_enable)
3115 		r = gfx_v11_0_cp_gfx_load_me_microcode_rs64(adev);
3116 	else
3117 		r = gfx_v11_0_cp_gfx_load_me_microcode(adev);
3118 	if (r) {
3119 		dev_err(adev->dev, "(%d) failed to load me fw\n", r);
3120 		return r;
3121 	}
3122 
3123 	return 0;
3124 }
3125 
3126 static int gfx_v11_0_cp_gfx_start(struct amdgpu_device *adev)
3127 {
3128 	struct amdgpu_ring *ring;
3129 	const struct cs_section_def *sect = NULL;
3130 	const struct cs_extent_def *ext = NULL;
3131 	int r, i;
3132 	int ctx_reg_offset;
3133 
3134 	/* init the CP */
3135 	WREG32_SOC15(GC, 0, regCP_MAX_CONTEXT,
3136 		     adev->gfx.config.max_hw_contexts - 1);
3137 	WREG32_SOC15(GC, 0, regCP_DEVICE_ID, 1);
3138 
3139 	if (!amdgpu_async_gfx_ring)
3140 		gfx_v11_0_cp_gfx_enable(adev, true);
3141 
3142 	ring = &adev->gfx.gfx_ring[0];
3143 	r = amdgpu_ring_alloc(ring, gfx_v11_0_get_csb_size(adev));
3144 	if (r) {
3145 		DRM_ERROR("amdgpu: cp failed to lock ring (%d).\n", r);
3146 		return r;
3147 	}
3148 
3149 	amdgpu_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
3150 	amdgpu_ring_write(ring, PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
3151 
3152 	amdgpu_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1));
3153 	amdgpu_ring_write(ring, 0x80000000);
3154 	amdgpu_ring_write(ring, 0x80000000);
3155 
3156 	for (sect = gfx11_cs_data; sect->section != NULL; ++sect) {
3157 		for (ext = sect->section; ext->extent != NULL; ++ext) {
3158 			if (sect->id == SECT_CONTEXT) {
3159 				amdgpu_ring_write(ring,
3160 						  PACKET3(PACKET3_SET_CONTEXT_REG,
3161 							  ext->reg_count));
3162 				amdgpu_ring_write(ring, ext->reg_index -
3163 						  PACKET3_SET_CONTEXT_REG_START);
3164 				for (i = 0; i < ext->reg_count; i++)
3165 					amdgpu_ring_write(ring, ext->extent[i]);
3166 			}
3167 		}
3168 	}
3169 
3170 	ctx_reg_offset =
3171 		SOC15_REG_OFFSET(GC, 0, regPA_SC_TILE_STEERING_OVERRIDE) - PACKET3_SET_CONTEXT_REG_START;
3172 	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONTEXT_REG, 1));
3173 	amdgpu_ring_write(ring, ctx_reg_offset);
3174 	amdgpu_ring_write(ring, adev->gfx.config.pa_sc_tile_steering_override);
3175 
3176 	amdgpu_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
3177 	amdgpu_ring_write(ring, PACKET3_PREAMBLE_END_CLEAR_STATE);
3178 
3179 	amdgpu_ring_write(ring, PACKET3(PACKET3_CLEAR_STATE, 0));
3180 	amdgpu_ring_write(ring, 0);
3181 
3182 	amdgpu_ring_commit(ring);
3183 
3184 	/* submit cs packet to copy state 0 to next available state */
3185 	if (adev->gfx.num_gfx_rings > 1) {
3186 		/* maximum supported gfx ring is 2 */
3187 		ring = &adev->gfx.gfx_ring[1];
3188 		r = amdgpu_ring_alloc(ring, 2);
3189 		if (r) {
3190 			DRM_ERROR("amdgpu: cp failed to lock ring (%d).\n", r);
3191 			return r;
3192 		}
3193 
3194 		amdgpu_ring_write(ring, PACKET3(PACKET3_CLEAR_STATE, 0));
3195 		amdgpu_ring_write(ring, 0);
3196 
3197 		amdgpu_ring_commit(ring);
3198 	}
3199 	return 0;
3200 }
3201 
3202 static void gfx_v11_0_cp_gfx_switch_pipe(struct amdgpu_device *adev,
3203 					 CP_PIPE_ID pipe)
3204 {
3205 	u32 tmp;
3206 
3207 	tmp = RREG32_SOC15(GC, 0, regGRBM_GFX_CNTL);
3208 	tmp = REG_SET_FIELD(tmp, GRBM_GFX_CNTL, PIPEID, pipe);
3209 
3210 	WREG32_SOC15(GC, 0, regGRBM_GFX_CNTL, tmp);
3211 }
3212 
3213 static void gfx_v11_0_cp_gfx_set_doorbell(struct amdgpu_device *adev,
3214 					  struct amdgpu_ring *ring)
3215 {
3216 	u32 tmp;
3217 
3218 	tmp = RREG32_SOC15(GC, 0, regCP_RB_DOORBELL_CONTROL);
3219 	if (ring->use_doorbell) {
3220 		tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL,
3221 				    DOORBELL_OFFSET, ring->doorbell_index);
3222 		tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL,
3223 				    DOORBELL_EN, 1);
3224 	} else {
3225 		tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL,
3226 				    DOORBELL_EN, 0);
3227 	}
3228 	WREG32_SOC15(GC, 0, regCP_RB_DOORBELL_CONTROL, tmp);
3229 
3230 	tmp = REG_SET_FIELD(0, CP_RB_DOORBELL_RANGE_LOWER,
3231 			    DOORBELL_RANGE_LOWER, ring->doorbell_index);
3232 	WREG32_SOC15(GC, 0, regCP_RB_DOORBELL_RANGE_LOWER, tmp);
3233 
3234 	WREG32_SOC15(GC, 0, regCP_RB_DOORBELL_RANGE_UPPER,
3235 		     CP_RB_DOORBELL_RANGE_UPPER__DOORBELL_RANGE_UPPER_MASK);
3236 }
3237 
3238 static int gfx_v11_0_cp_gfx_resume(struct amdgpu_device *adev)
3239 {
3240 	struct amdgpu_ring *ring;
3241 	u32 tmp;
3242 	u32 rb_bufsz;
3243 	u64 rb_addr, rptr_addr, wptr_gpu_addr;
3244 
3245 	/* Set the write pointer delay */
3246 	WREG32_SOC15(GC, 0, regCP_RB_WPTR_DELAY, 0);
3247 
3248 	/* set the RB to use vmid 0 */
3249 	WREG32_SOC15(GC, 0, regCP_RB_VMID, 0);
3250 
3251 	/* Init gfx ring 0 for pipe 0 */
3252 	mutex_lock(&adev->srbm_mutex);
3253 	gfx_v11_0_cp_gfx_switch_pipe(adev, PIPE_ID0);
3254 
3255 	/* Set ring buffer size */
3256 	ring = &adev->gfx.gfx_ring[0];
3257 	rb_bufsz = order_base_2(ring->ring_size / 8);
3258 	tmp = REG_SET_FIELD(0, CP_RB0_CNTL, RB_BUFSZ, rb_bufsz);
3259 	tmp = REG_SET_FIELD(tmp, CP_RB0_CNTL, RB_BLKSZ, rb_bufsz - 2);
3260 	WREG32_SOC15(GC, 0, regCP_RB0_CNTL, tmp);
3261 
3262 	/* Initialize the ring buffer's write pointers */
3263 	ring->wptr = 0;
3264 	WREG32_SOC15(GC, 0, regCP_RB0_WPTR, lower_32_bits(ring->wptr));
3265 	WREG32_SOC15(GC, 0, regCP_RB0_WPTR_HI, upper_32_bits(ring->wptr));
3266 
3267 	/* set the wb address wether it's enabled or not */
3268 	rptr_addr = ring->rptr_gpu_addr;
3269 	WREG32_SOC15(GC, 0, regCP_RB0_RPTR_ADDR, lower_32_bits(rptr_addr));
3270 	WREG32_SOC15(GC, 0, regCP_RB0_RPTR_ADDR_HI, upper_32_bits(rptr_addr) &
3271 		     CP_RB_RPTR_ADDR_HI__RB_RPTR_ADDR_HI_MASK);
3272 
3273 	wptr_gpu_addr = ring->wptr_gpu_addr;
3274 	WREG32_SOC15(GC, 0, regCP_RB_WPTR_POLL_ADDR_LO,
3275 		     lower_32_bits(wptr_gpu_addr));
3276 	WREG32_SOC15(GC, 0, regCP_RB_WPTR_POLL_ADDR_HI,
3277 		     upper_32_bits(wptr_gpu_addr));
3278 
3279 	mdelay(1);
3280 	WREG32_SOC15(GC, 0, regCP_RB0_CNTL, tmp);
3281 
3282 	rb_addr = ring->gpu_addr >> 8;
3283 	WREG32_SOC15(GC, 0, regCP_RB0_BASE, rb_addr);
3284 	WREG32_SOC15(GC, 0, regCP_RB0_BASE_HI, upper_32_bits(rb_addr));
3285 
3286 	WREG32_SOC15(GC, 0, regCP_RB_ACTIVE, 1);
3287 
3288 	gfx_v11_0_cp_gfx_set_doorbell(adev, ring);
3289 	mutex_unlock(&adev->srbm_mutex);
3290 
3291 	/* Init gfx ring 1 for pipe 1 */
3292 	if (adev->gfx.num_gfx_rings > 1) {
3293 		mutex_lock(&adev->srbm_mutex);
3294 		gfx_v11_0_cp_gfx_switch_pipe(adev, PIPE_ID1);
3295 		/* maximum supported gfx ring is 2 */
3296 		ring = &adev->gfx.gfx_ring[1];
3297 		rb_bufsz = order_base_2(ring->ring_size / 8);
3298 		tmp = REG_SET_FIELD(0, CP_RB1_CNTL, RB_BUFSZ, rb_bufsz);
3299 		tmp = REG_SET_FIELD(tmp, CP_RB1_CNTL, RB_BLKSZ, rb_bufsz - 2);
3300 		WREG32_SOC15(GC, 0, regCP_RB1_CNTL, tmp);
3301 		/* Initialize the ring buffer's write pointers */
3302 		ring->wptr = 0;
3303 		WREG32_SOC15(GC, 0, regCP_RB1_WPTR, lower_32_bits(ring->wptr));
3304 		WREG32_SOC15(GC, 0, regCP_RB1_WPTR_HI, upper_32_bits(ring->wptr));
3305 		/* Set the wb address wether it's enabled or not */
3306 		rptr_addr = ring->rptr_gpu_addr;
3307 		WREG32_SOC15(GC, 0, regCP_RB1_RPTR_ADDR, lower_32_bits(rptr_addr));
3308 		WREG32_SOC15(GC, 0, regCP_RB1_RPTR_ADDR_HI, upper_32_bits(rptr_addr) &
3309 			     CP_RB1_RPTR_ADDR_HI__RB_RPTR_ADDR_HI_MASK);
3310 		wptr_gpu_addr = ring->wptr_gpu_addr;
3311 		WREG32_SOC15(GC, 0, regCP_RB_WPTR_POLL_ADDR_LO,
3312 			     lower_32_bits(wptr_gpu_addr));
3313 		WREG32_SOC15(GC, 0, regCP_RB_WPTR_POLL_ADDR_HI,
3314 			     upper_32_bits(wptr_gpu_addr));
3315 
3316 		mdelay(1);
3317 		WREG32_SOC15(GC, 0, regCP_RB1_CNTL, tmp);
3318 
3319 		rb_addr = ring->gpu_addr >> 8;
3320 		WREG32_SOC15(GC, 0, regCP_RB1_BASE, rb_addr);
3321 		WREG32_SOC15(GC, 0, regCP_RB1_BASE_HI, upper_32_bits(rb_addr));
3322 		WREG32_SOC15(GC, 0, regCP_RB1_ACTIVE, 1);
3323 
3324 		gfx_v11_0_cp_gfx_set_doorbell(adev, ring);
3325 		mutex_unlock(&adev->srbm_mutex);
3326 	}
3327 	/* Switch to pipe 0 */
3328 	mutex_lock(&adev->srbm_mutex);
3329 	gfx_v11_0_cp_gfx_switch_pipe(adev, PIPE_ID0);
3330 	mutex_unlock(&adev->srbm_mutex);
3331 
3332 	/* start the ring */
3333 	gfx_v11_0_cp_gfx_start(adev);
3334 
3335 	return 0;
3336 }
3337 
3338 static void gfx_v11_0_cp_compute_enable(struct amdgpu_device *adev, bool enable)
3339 {
3340 	u32 data;
3341 
3342 	if (adev->gfx.rs64_enable) {
3343 		data = RREG32_SOC15(GC, 0, regCP_MEC_RS64_CNTL);
3344 		data = REG_SET_FIELD(data, CP_MEC_RS64_CNTL, MEC_INVALIDATE_ICACHE,
3345 							 enable ? 0 : 1);
3346 		data = REG_SET_FIELD(data, CP_MEC_RS64_CNTL, MEC_PIPE0_RESET,
3347 							 enable ? 0 : 1);
3348 		data = REG_SET_FIELD(data, CP_MEC_RS64_CNTL, MEC_PIPE1_RESET,
3349 							 enable ? 0 : 1);
3350 		data = REG_SET_FIELD(data, CP_MEC_RS64_CNTL, MEC_PIPE2_RESET,
3351 							 enable ? 0 : 1);
3352 		data = REG_SET_FIELD(data, CP_MEC_RS64_CNTL, MEC_PIPE3_RESET,
3353 							 enable ? 0 : 1);
3354 		data = REG_SET_FIELD(data, CP_MEC_RS64_CNTL, MEC_PIPE0_ACTIVE,
3355 							 enable ? 1 : 0);
3356 		data = REG_SET_FIELD(data, CP_MEC_RS64_CNTL, MEC_PIPE1_ACTIVE,
3357 				                         enable ? 1 : 0);
3358 		data = REG_SET_FIELD(data, CP_MEC_RS64_CNTL, MEC_PIPE2_ACTIVE,
3359 							 enable ? 1 : 0);
3360 		data = REG_SET_FIELD(data, CP_MEC_RS64_CNTL, MEC_PIPE3_ACTIVE,
3361 							 enable ? 1 : 0);
3362 		data = REG_SET_FIELD(data, CP_MEC_RS64_CNTL, MEC_HALT,
3363 							 enable ? 0 : 1);
3364 		WREG32_SOC15(GC, 0, regCP_MEC_RS64_CNTL, data);
3365 	} else {
3366 		data = RREG32_SOC15(GC, 0, regCP_MEC_CNTL);
3367 
3368 		if (enable) {
3369 			data = REG_SET_FIELD(data, CP_MEC_CNTL, MEC_ME1_HALT, 0);
3370 			if (!adev->enable_mes_kiq)
3371 				data = REG_SET_FIELD(data, CP_MEC_CNTL,
3372 						     MEC_ME2_HALT, 0);
3373 		} else {
3374 			data = REG_SET_FIELD(data, CP_MEC_CNTL, MEC_ME1_HALT, 1);
3375 			data = REG_SET_FIELD(data, CP_MEC_CNTL, MEC_ME2_HALT, 1);
3376 		}
3377 		WREG32_SOC15(GC, 0, regCP_MEC_CNTL, data);
3378 	}
3379 
3380 	udelay(50);
3381 }
3382 
3383 static int gfx_v11_0_cp_compute_load_microcode(struct amdgpu_device *adev)
3384 {
3385 	const struct gfx_firmware_header_v1_0 *mec_hdr;
3386 	const __le32 *fw_data;
3387 	unsigned i, fw_size;
3388 	u32 *fw = NULL;
3389 	int r;
3390 
3391 	if (!adev->gfx.mec_fw)
3392 		return -EINVAL;
3393 
3394 	gfx_v11_0_cp_compute_enable(adev, false);
3395 
3396 	mec_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data;
3397 	amdgpu_ucode_print_gfx_hdr(&mec_hdr->header);
3398 
3399 	fw_data = (const __le32 *)
3400 		(adev->gfx.mec_fw->data +
3401 		 le32_to_cpu(mec_hdr->header.ucode_array_offset_bytes));
3402 	fw_size = le32_to_cpu(mec_hdr->header.ucode_size_bytes);
3403 
3404 	r = amdgpu_bo_create_reserved(adev, mec_hdr->header.ucode_size_bytes,
3405 					  PAGE_SIZE, AMDGPU_GEM_DOMAIN_GTT,
3406 					  &adev->gfx.mec.mec_fw_obj,
3407 					  &adev->gfx.mec.mec_fw_gpu_addr,
3408 					  (void **)&fw);
3409 	if (r) {
3410 		dev_err(adev->dev, "(%d) failed to create mec fw bo\n", r);
3411 		gfx_v11_0_mec_fini(adev);
3412 		return r;
3413 	}
3414 
3415 	memcpy(fw, fw_data, fw_size);
3416 
3417 	amdgpu_bo_kunmap(adev->gfx.mec.mec_fw_obj);
3418 	amdgpu_bo_unreserve(adev->gfx.mec.mec_fw_obj);
3419 
3420 	gfx_v11_0_config_mec_cache(adev, adev->gfx.mec.mec_fw_gpu_addr);
3421 
3422 	/* MEC1 */
3423 	WREG32_SOC15(GC, 0, regCP_MEC_ME1_UCODE_ADDR, 0);
3424 
3425 	for (i = 0; i < mec_hdr->jt_size; i++)
3426 		WREG32_SOC15(GC, 0, regCP_MEC_ME1_UCODE_DATA,
3427 			     le32_to_cpup(fw_data + mec_hdr->jt_offset + i));
3428 
3429 	WREG32_SOC15(GC, 0, regCP_MEC_ME1_UCODE_ADDR, adev->gfx.mec_fw_version);
3430 
3431 	return 0;
3432 }
3433 
3434 static int gfx_v11_0_cp_compute_load_microcode_rs64(struct amdgpu_device *adev)
3435 {
3436 	const struct gfx_firmware_header_v2_0 *mec_hdr;
3437 	const __le32 *fw_ucode, *fw_data;
3438 	u32 tmp, fw_ucode_size, fw_data_size;
3439 	u32 i, usec_timeout = 50000; /* Wait for 50 ms */
3440 	u32 *fw_ucode_ptr, *fw_data_ptr;
3441 	int r;
3442 
3443 	if (!adev->gfx.mec_fw)
3444 		return -EINVAL;
3445 
3446 	gfx_v11_0_cp_compute_enable(adev, false);
3447 
3448 	mec_hdr = (const struct gfx_firmware_header_v2_0 *)adev->gfx.mec_fw->data;
3449 	amdgpu_ucode_print_gfx_hdr(&mec_hdr->header);
3450 
3451 	fw_ucode = (const __le32 *) (adev->gfx.mec_fw->data +
3452 				le32_to_cpu(mec_hdr->ucode_offset_bytes));
3453 	fw_ucode_size = le32_to_cpu(mec_hdr->ucode_size_bytes);
3454 
3455 	fw_data = (const __le32 *) (adev->gfx.mec_fw->data +
3456 				le32_to_cpu(mec_hdr->data_offset_bytes));
3457 	fw_data_size = le32_to_cpu(mec_hdr->data_size_bytes);
3458 
3459 	r = amdgpu_bo_create_reserved(adev, fw_ucode_size,
3460 				      64 * 1024,
3461 				      AMDGPU_GEM_DOMAIN_VRAM |
3462 				      AMDGPU_GEM_DOMAIN_GTT,
3463 				      &adev->gfx.mec.mec_fw_obj,
3464 				      &adev->gfx.mec.mec_fw_gpu_addr,
3465 				      (void **)&fw_ucode_ptr);
3466 	if (r) {
3467 		dev_err(adev->dev, "(%d) failed to create mec fw ucode bo\n", r);
3468 		gfx_v11_0_mec_fini(adev);
3469 		return r;
3470 	}
3471 
3472 	r = amdgpu_bo_create_reserved(adev, fw_data_size,
3473 				      64 * 1024,
3474 				      AMDGPU_GEM_DOMAIN_VRAM |
3475 				      AMDGPU_GEM_DOMAIN_GTT,
3476 				      &adev->gfx.mec.mec_fw_data_obj,
3477 				      &adev->gfx.mec.mec_fw_data_gpu_addr,
3478 				      (void **)&fw_data_ptr);
3479 	if (r) {
3480 		dev_err(adev->dev, "(%d) failed to create mec fw ucode bo\n", r);
3481 		gfx_v11_0_mec_fini(adev);
3482 		return r;
3483 	}
3484 
3485 	memcpy(fw_ucode_ptr, fw_ucode, fw_ucode_size);
3486 	memcpy(fw_data_ptr, fw_data, fw_data_size);
3487 
3488 	amdgpu_bo_kunmap(adev->gfx.mec.mec_fw_obj);
3489 	amdgpu_bo_kunmap(adev->gfx.mec.mec_fw_data_obj);
3490 	amdgpu_bo_unreserve(adev->gfx.mec.mec_fw_obj);
3491 	amdgpu_bo_unreserve(adev->gfx.mec.mec_fw_data_obj);
3492 
3493 	tmp = RREG32_SOC15(GC, 0, regCP_CPC_IC_BASE_CNTL);
3494 	tmp = REG_SET_FIELD(tmp, CP_CPC_IC_BASE_CNTL, VMID, 0);
3495 	tmp = REG_SET_FIELD(tmp, CP_CPC_IC_BASE_CNTL, EXE_DISABLE, 0);
3496 	tmp = REG_SET_FIELD(tmp, CP_CPC_IC_BASE_CNTL, CACHE_POLICY, 0);
3497 	WREG32_SOC15(GC, 0, regCP_CPC_IC_BASE_CNTL, tmp);
3498 
3499 	tmp = RREG32_SOC15(GC, 0, regCP_MEC_DC_BASE_CNTL);
3500 	tmp = REG_SET_FIELD(tmp, CP_MEC_DC_BASE_CNTL, VMID, 0);
3501 	tmp = REG_SET_FIELD(tmp, CP_MEC_DC_BASE_CNTL, CACHE_POLICY, 0);
3502 	WREG32_SOC15(GC, 0, regCP_MEC_DC_BASE_CNTL, tmp);
3503 
3504 	mutex_lock(&adev->srbm_mutex);
3505 	for (i = 0; i < adev->gfx.mec.num_pipe_per_mec; i++) {
3506 		soc21_grbm_select(adev, 1, i, 0, 0);
3507 
3508 		WREG32_SOC15(GC, 0, regCP_MEC_MDBASE_LO, adev->gfx.mec.mec_fw_data_gpu_addr);
3509 		WREG32_SOC15(GC, 0, regCP_MEC_MDBASE_HI,
3510 		     upper_32_bits(adev->gfx.mec.mec_fw_data_gpu_addr));
3511 
3512 		WREG32_SOC15(GC, 0, regCP_MEC_RS64_PRGRM_CNTR_START,
3513 					mec_hdr->ucode_start_addr_lo >> 2 |
3514 					mec_hdr->ucode_start_addr_hi << 30);
3515 		WREG32_SOC15(GC, 0, regCP_MEC_RS64_PRGRM_CNTR_START_HI,
3516 					mec_hdr->ucode_start_addr_hi >> 2);
3517 
3518 		WREG32_SOC15(GC, 0, regCP_CPC_IC_BASE_LO, adev->gfx.mec.mec_fw_gpu_addr);
3519 		WREG32_SOC15(GC, 0, regCP_CPC_IC_BASE_HI,
3520 		     upper_32_bits(adev->gfx.mec.mec_fw_gpu_addr));
3521 	}
3522 	mutex_unlock(&adev->srbm_mutex);
3523 	soc21_grbm_select(adev, 0, 0, 0, 0);
3524 
3525 	/* Trigger an invalidation of the L1 instruction caches */
3526 	tmp = RREG32_SOC15(GC, 0, regCP_MEC_DC_OP_CNTL);
3527 	tmp = REG_SET_FIELD(tmp, CP_MEC_DC_OP_CNTL, INVALIDATE_DCACHE, 1);
3528 	WREG32_SOC15(GC, 0, regCP_MEC_DC_OP_CNTL, tmp);
3529 
3530 	/* Wait for invalidation complete */
3531 	for (i = 0; i < usec_timeout; i++) {
3532 		tmp = RREG32_SOC15(GC, 0, regCP_MEC_DC_OP_CNTL);
3533 		if (1 == REG_GET_FIELD(tmp, CP_MEC_DC_OP_CNTL,
3534 				       INVALIDATE_DCACHE_COMPLETE))
3535 			break;
3536 		udelay(1);
3537 	}
3538 
3539 	if (i >= usec_timeout) {
3540 		dev_err(adev->dev, "failed to invalidate instruction cache\n");
3541 		return -EINVAL;
3542 	}
3543 
3544 	/* Trigger an invalidation of the L1 instruction caches */
3545 	tmp = RREG32_SOC15(GC, 0, regCP_CPC_IC_OP_CNTL);
3546 	tmp = REG_SET_FIELD(tmp, CP_CPC_IC_OP_CNTL, INVALIDATE_CACHE, 1);
3547 	WREG32_SOC15(GC, 0, regCP_CPC_IC_OP_CNTL, tmp);
3548 
3549 	/* Wait for invalidation complete */
3550 	for (i = 0; i < usec_timeout; i++) {
3551 		tmp = RREG32_SOC15(GC, 0, regCP_CPC_IC_OP_CNTL);
3552 		if (1 == REG_GET_FIELD(tmp, CP_CPC_IC_OP_CNTL,
3553 				       INVALIDATE_CACHE_COMPLETE))
3554 			break;
3555 		udelay(1);
3556 	}
3557 
3558 	if (i >= usec_timeout) {
3559 		dev_err(adev->dev, "failed to invalidate instruction cache\n");
3560 		return -EINVAL;
3561 	}
3562 
3563 	return 0;
3564 }
3565 
3566 static void gfx_v11_0_kiq_setting(struct amdgpu_ring *ring)
3567 {
3568 	uint32_t tmp;
3569 	struct amdgpu_device *adev = ring->adev;
3570 
3571 	/* tell RLC which is KIQ queue */
3572 	tmp = RREG32_SOC15(GC, 0, regRLC_CP_SCHEDULERS);
3573 	tmp &= 0xffffff00;
3574 	tmp |= (ring->me << 5) | (ring->pipe << 3) | (ring->queue);
3575 	WREG32_SOC15(GC, 0, regRLC_CP_SCHEDULERS, tmp);
3576 	tmp |= 0x80;
3577 	WREG32_SOC15(GC, 0, regRLC_CP_SCHEDULERS, tmp);
3578 }
3579 
3580 static void gfx_v11_0_cp_set_doorbell_range(struct amdgpu_device *adev)
3581 {
3582 	/* set graphics engine doorbell range */
3583 	WREG32_SOC15(GC, 0, regCP_RB_DOORBELL_RANGE_LOWER,
3584 		     (adev->doorbell_index.gfx_ring0 * 2) << 2);
3585 	WREG32_SOC15(GC, 0, regCP_RB_DOORBELL_RANGE_UPPER,
3586 		     (adev->doorbell_index.gfx_userqueue_end * 2) << 2);
3587 
3588 	/* set compute engine doorbell range */
3589 	WREG32_SOC15(GC, 0, regCP_MEC_DOORBELL_RANGE_LOWER,
3590 		     (adev->doorbell_index.kiq * 2) << 2);
3591 	WREG32_SOC15(GC, 0, regCP_MEC_DOORBELL_RANGE_UPPER,
3592 		     (adev->doorbell_index.userqueue_end * 2) << 2);
3593 }
3594 
3595 static int gfx_v11_0_gfx_mqd_init(struct amdgpu_device *adev, void *m,
3596 				  struct amdgpu_mqd_prop *prop)
3597 {
3598 	struct v11_gfx_mqd *mqd = m;
3599 	uint64_t hqd_gpu_addr, wb_gpu_addr;
3600 	uint32_t tmp;
3601 	uint32_t rb_bufsz;
3602 
3603 	/* set up gfx hqd wptr */
3604 	mqd->cp_gfx_hqd_wptr = 0;
3605 	mqd->cp_gfx_hqd_wptr_hi = 0;
3606 
3607 	/* set the pointer to the MQD */
3608 	mqd->cp_mqd_base_addr = prop->mqd_gpu_addr & 0xfffffffc;
3609 	mqd->cp_mqd_base_addr_hi = upper_32_bits(prop->mqd_gpu_addr);
3610 
3611 	/* set up mqd control */
3612 	tmp = RREG32_SOC15(GC, 0, regCP_GFX_MQD_CONTROL);
3613 	tmp = REG_SET_FIELD(tmp, CP_GFX_MQD_CONTROL, VMID, 0);
3614 	tmp = REG_SET_FIELD(tmp, CP_GFX_MQD_CONTROL, PRIV_STATE, 1);
3615 	tmp = REG_SET_FIELD(tmp, CP_GFX_MQD_CONTROL, CACHE_POLICY, 0);
3616 	mqd->cp_gfx_mqd_control = tmp;
3617 
3618 	/* set up gfx_hqd_vimd with 0x0 to indicate the ring buffer's vmid */
3619 	tmp = RREG32_SOC15(GC, 0, regCP_GFX_HQD_VMID);
3620 	tmp = REG_SET_FIELD(tmp, CP_GFX_HQD_VMID, VMID, 0);
3621 	mqd->cp_gfx_hqd_vmid = 0;
3622 
3623 	/* set up default queue priority level
3624 	 * 0x0 = low priority, 0x1 = high priority */
3625 	tmp = RREG32_SOC15(GC, 0, regCP_GFX_HQD_QUEUE_PRIORITY);
3626 	tmp = REG_SET_FIELD(tmp, CP_GFX_HQD_QUEUE_PRIORITY, PRIORITY_LEVEL, 0);
3627 	mqd->cp_gfx_hqd_queue_priority = tmp;
3628 
3629 	/* set up time quantum */
3630 	tmp = RREG32_SOC15(GC, 0, regCP_GFX_HQD_QUANTUM);
3631 	tmp = REG_SET_FIELD(tmp, CP_GFX_HQD_QUANTUM, QUANTUM_EN, 1);
3632 	mqd->cp_gfx_hqd_quantum = tmp;
3633 
3634 	/* set up gfx hqd base. this is similar as CP_RB_BASE */
3635 	hqd_gpu_addr = prop->hqd_base_gpu_addr >> 8;
3636 	mqd->cp_gfx_hqd_base = hqd_gpu_addr;
3637 	mqd->cp_gfx_hqd_base_hi = upper_32_bits(hqd_gpu_addr);
3638 
3639 	/* set up hqd_rptr_addr/_hi, similar as CP_RB_RPTR */
3640 	wb_gpu_addr = prop->rptr_gpu_addr;
3641 	mqd->cp_gfx_hqd_rptr_addr = wb_gpu_addr & 0xfffffffc;
3642 	mqd->cp_gfx_hqd_rptr_addr_hi =
3643 		upper_32_bits(wb_gpu_addr) & 0xffff;
3644 
3645 	/* set up rb_wptr_poll addr */
3646 	wb_gpu_addr = prop->wptr_gpu_addr;
3647 	mqd->cp_rb_wptr_poll_addr_lo = wb_gpu_addr & 0xfffffffc;
3648 	mqd->cp_rb_wptr_poll_addr_hi = upper_32_bits(wb_gpu_addr) & 0xffff;
3649 
3650 	/* set up the gfx_hqd_control, similar as CP_RB0_CNTL */
3651 	rb_bufsz = order_base_2(prop->queue_size / 4) - 1;
3652 	tmp = RREG32_SOC15(GC, 0, regCP_GFX_HQD_CNTL);
3653 	tmp = REG_SET_FIELD(tmp, CP_GFX_HQD_CNTL, RB_BUFSZ, rb_bufsz);
3654 	tmp = REG_SET_FIELD(tmp, CP_GFX_HQD_CNTL, RB_BLKSZ, rb_bufsz - 2);
3655 #ifdef __BIG_ENDIAN
3656 	tmp = REG_SET_FIELD(tmp, CP_GFX_HQD_CNTL, BUF_SWAP, 1);
3657 #endif
3658 	mqd->cp_gfx_hqd_cntl = tmp;
3659 
3660 	/* set up cp_doorbell_control */
3661 	tmp = RREG32_SOC15(GC, 0, regCP_RB_DOORBELL_CONTROL);
3662 	if (prop->use_doorbell) {
3663 		tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL,
3664 				    DOORBELL_OFFSET, prop->doorbell_index);
3665 		tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL,
3666 				    DOORBELL_EN, 1);
3667 	} else
3668 		tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL,
3669 				    DOORBELL_EN, 0);
3670 	mqd->cp_rb_doorbell_control = tmp;
3671 
3672 	/* reset read and write pointers, similar to CP_RB0_WPTR/_RPTR */
3673 	mqd->cp_gfx_hqd_rptr = RREG32_SOC15(GC, 0, regCP_GFX_HQD_RPTR);
3674 
3675 	/* active the queue */
3676 	mqd->cp_gfx_hqd_active = 1;
3677 
3678 	return 0;
3679 }
3680 
3681 static int gfx_v11_0_gfx_init_queue(struct amdgpu_ring *ring)
3682 {
3683 	struct amdgpu_device *adev = ring->adev;
3684 	struct v11_gfx_mqd *mqd = ring->mqd_ptr;
3685 	int mqd_idx = ring - &adev->gfx.gfx_ring[0];
3686 
3687 	if (!amdgpu_in_reset(adev) && !adev->in_suspend) {
3688 		memset((void *)mqd, 0, sizeof(*mqd));
3689 		mutex_lock(&adev->srbm_mutex);
3690 		soc21_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
3691 		amdgpu_ring_init_mqd(ring);
3692 		soc21_grbm_select(adev, 0, 0, 0, 0);
3693 		mutex_unlock(&adev->srbm_mutex);
3694 		if (adev->gfx.me.mqd_backup[mqd_idx])
3695 			memcpy_fromio(adev->gfx.me.mqd_backup[mqd_idx], mqd, sizeof(*mqd));
3696 	} else {
3697 		/* restore mqd with the backup copy */
3698 		if (adev->gfx.me.mqd_backup[mqd_idx])
3699 			memcpy_toio(mqd, adev->gfx.me.mqd_backup[mqd_idx], sizeof(*mqd));
3700 		/* reset the ring */
3701 		ring->wptr = 0;
3702 		*ring->wptr_cpu_addr = 0;
3703 		amdgpu_ring_clear_ring(ring);
3704 	}
3705 
3706 	return 0;
3707 }
3708 
3709 static int gfx_v11_0_cp_async_gfx_ring_resume(struct amdgpu_device *adev)
3710 {
3711 	int r, i;
3712 	struct amdgpu_ring *ring;
3713 
3714 	for (i = 0; i < adev->gfx.num_gfx_rings; i++) {
3715 		ring = &adev->gfx.gfx_ring[i];
3716 
3717 		r = amdgpu_bo_reserve(ring->mqd_obj, false);
3718 		if (unlikely(r != 0))
3719 			return r;
3720 
3721 		r = amdgpu_bo_kmap(ring->mqd_obj, (void **)&ring->mqd_ptr);
3722 		if (!r) {
3723 			r = gfx_v11_0_gfx_init_queue(ring);
3724 			amdgpu_bo_kunmap(ring->mqd_obj);
3725 			ring->mqd_ptr = NULL;
3726 		}
3727 		amdgpu_bo_unreserve(ring->mqd_obj);
3728 		if (r)
3729 			return r;
3730 	}
3731 
3732 	r = amdgpu_gfx_enable_kgq(adev, 0);
3733 	if (r)
3734 		return r;
3735 
3736 	return gfx_v11_0_cp_gfx_start(adev);
3737 }
3738 
3739 static int gfx_v11_0_compute_mqd_init(struct amdgpu_device *adev, void *m,
3740 				      struct amdgpu_mqd_prop *prop)
3741 {
3742 	struct v11_compute_mqd *mqd = m;
3743 	uint64_t hqd_gpu_addr, wb_gpu_addr, eop_base_addr;
3744 	uint32_t tmp;
3745 
3746 	mqd->header = 0xC0310800;
3747 	mqd->compute_pipelinestat_enable = 0x00000001;
3748 	mqd->compute_static_thread_mgmt_se0 = 0xffffffff;
3749 	mqd->compute_static_thread_mgmt_se1 = 0xffffffff;
3750 	mqd->compute_static_thread_mgmt_se2 = 0xffffffff;
3751 	mqd->compute_static_thread_mgmt_se3 = 0xffffffff;
3752 	mqd->compute_misc_reserved = 0x00000007;
3753 
3754 	eop_base_addr = prop->eop_gpu_addr >> 8;
3755 	mqd->cp_hqd_eop_base_addr_lo = eop_base_addr;
3756 	mqd->cp_hqd_eop_base_addr_hi = upper_32_bits(eop_base_addr);
3757 
3758 	/* set the EOP size, register value is 2^(EOP_SIZE+1) dwords */
3759 	tmp = RREG32_SOC15(GC, 0, regCP_HQD_EOP_CONTROL);
3760 	tmp = REG_SET_FIELD(tmp, CP_HQD_EOP_CONTROL, EOP_SIZE,
3761 			(order_base_2(GFX11_MEC_HPD_SIZE / 4) - 1));
3762 
3763 	mqd->cp_hqd_eop_control = tmp;
3764 
3765 	/* enable doorbell? */
3766 	tmp = RREG32_SOC15(GC, 0, regCP_HQD_PQ_DOORBELL_CONTROL);
3767 
3768 	if (prop->use_doorbell) {
3769 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3770 				    DOORBELL_OFFSET, prop->doorbell_index);
3771 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3772 				    DOORBELL_EN, 1);
3773 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3774 				    DOORBELL_SOURCE, 0);
3775 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3776 				    DOORBELL_HIT, 0);
3777 	} else {
3778 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3779 				    DOORBELL_EN, 0);
3780 	}
3781 
3782 	mqd->cp_hqd_pq_doorbell_control = tmp;
3783 
3784 	/* disable the queue if it's active */
3785 	mqd->cp_hqd_dequeue_request = 0;
3786 	mqd->cp_hqd_pq_rptr = 0;
3787 	mqd->cp_hqd_pq_wptr_lo = 0;
3788 	mqd->cp_hqd_pq_wptr_hi = 0;
3789 
3790 	/* set the pointer to the MQD */
3791 	mqd->cp_mqd_base_addr_lo = prop->mqd_gpu_addr & 0xfffffffc;
3792 	mqd->cp_mqd_base_addr_hi = upper_32_bits(prop->mqd_gpu_addr);
3793 
3794 	/* set MQD vmid to 0 */
3795 	tmp = RREG32_SOC15(GC, 0, regCP_MQD_CONTROL);
3796 	tmp = REG_SET_FIELD(tmp, CP_MQD_CONTROL, VMID, 0);
3797 	mqd->cp_mqd_control = tmp;
3798 
3799 	/* set the pointer to the HQD, this is similar CP_RB0_BASE/_HI */
3800 	hqd_gpu_addr = prop->hqd_base_gpu_addr >> 8;
3801 	mqd->cp_hqd_pq_base_lo = hqd_gpu_addr;
3802 	mqd->cp_hqd_pq_base_hi = upper_32_bits(hqd_gpu_addr);
3803 
3804 	/* set up the HQD, this is similar to CP_RB0_CNTL */
3805 	tmp = RREG32_SOC15(GC, 0, regCP_HQD_PQ_CONTROL);
3806 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, QUEUE_SIZE,
3807 			    (order_base_2(prop->queue_size / 4) - 1));
3808 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, RPTR_BLOCK_SIZE,
3809 			    (order_base_2(AMDGPU_GPU_PAGE_SIZE / 4) - 1));
3810 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, UNORD_DISPATCH, 1);
3811 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, TUNNEL_DISPATCH, 0);
3812 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, PRIV_STATE, 1);
3813 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, KMD_QUEUE, 1);
3814 	mqd->cp_hqd_pq_control = tmp;
3815 
3816 	/* set the wb address whether it's enabled or not */
3817 	wb_gpu_addr = prop->rptr_gpu_addr;
3818 	mqd->cp_hqd_pq_rptr_report_addr_lo = wb_gpu_addr & 0xfffffffc;
3819 	mqd->cp_hqd_pq_rptr_report_addr_hi =
3820 		upper_32_bits(wb_gpu_addr) & 0xffff;
3821 
3822 	/* only used if CP_PQ_WPTR_POLL_CNTL.CP_PQ_WPTR_POLL_CNTL__EN_MASK=1 */
3823 	wb_gpu_addr = prop->wptr_gpu_addr;
3824 	mqd->cp_hqd_pq_wptr_poll_addr_lo = wb_gpu_addr & 0xfffffffc;
3825 	mqd->cp_hqd_pq_wptr_poll_addr_hi = upper_32_bits(wb_gpu_addr) & 0xffff;
3826 
3827 	tmp = 0;
3828 	/* enable the doorbell if requested */
3829 	if (prop->use_doorbell) {
3830 		tmp = RREG32_SOC15(GC, 0, regCP_HQD_PQ_DOORBELL_CONTROL);
3831 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3832 				DOORBELL_OFFSET, prop->doorbell_index);
3833 
3834 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3835 				    DOORBELL_EN, 1);
3836 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3837 				    DOORBELL_SOURCE, 0);
3838 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3839 				    DOORBELL_HIT, 0);
3840 	}
3841 
3842 	mqd->cp_hqd_pq_doorbell_control = tmp;
3843 
3844 	/* reset read and write pointers, similar to CP_RB0_WPTR/_RPTR */
3845 	mqd->cp_hqd_pq_rptr = RREG32_SOC15(GC, 0, regCP_HQD_PQ_RPTR);
3846 
3847 	/* set the vmid for the queue */
3848 	mqd->cp_hqd_vmid = 0;
3849 
3850 	tmp = RREG32_SOC15(GC, 0, regCP_HQD_PERSISTENT_STATE);
3851 	tmp = REG_SET_FIELD(tmp, CP_HQD_PERSISTENT_STATE, PRELOAD_SIZE, 0x55);
3852 	mqd->cp_hqd_persistent_state = tmp;
3853 
3854 	/* set MIN_IB_AVAIL_SIZE */
3855 	tmp = RREG32_SOC15(GC, 0, regCP_HQD_IB_CONTROL);
3856 	tmp = REG_SET_FIELD(tmp, CP_HQD_IB_CONTROL, MIN_IB_AVAIL_SIZE, 3);
3857 	mqd->cp_hqd_ib_control = tmp;
3858 
3859 	/* set static priority for a compute queue/ring */
3860 	mqd->cp_hqd_pipe_priority = prop->hqd_pipe_priority;
3861 	mqd->cp_hqd_queue_priority = prop->hqd_queue_priority;
3862 
3863 	mqd->cp_hqd_active = prop->hqd_active;
3864 
3865 	return 0;
3866 }
3867 
3868 static int gfx_v11_0_kiq_init_register(struct amdgpu_ring *ring)
3869 {
3870 	struct amdgpu_device *adev = ring->adev;
3871 	struct v11_compute_mqd *mqd = ring->mqd_ptr;
3872 	int j;
3873 
3874 	/* inactivate the queue */
3875 	if (amdgpu_sriov_vf(adev))
3876 		WREG32_SOC15(GC, 0, regCP_HQD_ACTIVE, 0);
3877 
3878 	/* disable wptr polling */
3879 	WREG32_FIELD15_PREREG(GC, 0, CP_PQ_WPTR_POLL_CNTL, EN, 0);
3880 
3881 	/* write the EOP addr */
3882 	WREG32_SOC15(GC, 0, regCP_HQD_EOP_BASE_ADDR,
3883 	       mqd->cp_hqd_eop_base_addr_lo);
3884 	WREG32_SOC15(GC, 0, regCP_HQD_EOP_BASE_ADDR_HI,
3885 	       mqd->cp_hqd_eop_base_addr_hi);
3886 
3887 	/* set the EOP size, register value is 2^(EOP_SIZE+1) dwords */
3888 	WREG32_SOC15(GC, 0, regCP_HQD_EOP_CONTROL,
3889 	       mqd->cp_hqd_eop_control);
3890 
3891 	/* enable doorbell? */
3892 	WREG32_SOC15(GC, 0, regCP_HQD_PQ_DOORBELL_CONTROL,
3893 	       mqd->cp_hqd_pq_doorbell_control);
3894 
3895 	/* disable the queue if it's active */
3896 	if (RREG32_SOC15(GC, 0, regCP_HQD_ACTIVE) & 1) {
3897 		WREG32_SOC15(GC, 0, regCP_HQD_DEQUEUE_REQUEST, 1);
3898 		for (j = 0; j < adev->usec_timeout; j++) {
3899 			if (!(RREG32_SOC15(GC, 0, regCP_HQD_ACTIVE) & 1))
3900 				break;
3901 			udelay(1);
3902 		}
3903 		WREG32_SOC15(GC, 0, regCP_HQD_DEQUEUE_REQUEST,
3904 		       mqd->cp_hqd_dequeue_request);
3905 		WREG32_SOC15(GC, 0, regCP_HQD_PQ_RPTR,
3906 		       mqd->cp_hqd_pq_rptr);
3907 		WREG32_SOC15(GC, 0, regCP_HQD_PQ_WPTR_LO,
3908 		       mqd->cp_hqd_pq_wptr_lo);
3909 		WREG32_SOC15(GC, 0, regCP_HQD_PQ_WPTR_HI,
3910 		       mqd->cp_hqd_pq_wptr_hi);
3911 	}
3912 
3913 	/* set the pointer to the MQD */
3914 	WREG32_SOC15(GC, 0, regCP_MQD_BASE_ADDR,
3915 	       mqd->cp_mqd_base_addr_lo);
3916 	WREG32_SOC15(GC, 0, regCP_MQD_BASE_ADDR_HI,
3917 	       mqd->cp_mqd_base_addr_hi);
3918 
3919 	/* set MQD vmid to 0 */
3920 	WREG32_SOC15(GC, 0, regCP_MQD_CONTROL,
3921 	       mqd->cp_mqd_control);
3922 
3923 	/* set the pointer to the HQD, this is similar CP_RB0_BASE/_HI */
3924 	WREG32_SOC15(GC, 0, regCP_HQD_PQ_BASE,
3925 	       mqd->cp_hqd_pq_base_lo);
3926 	WREG32_SOC15(GC, 0, regCP_HQD_PQ_BASE_HI,
3927 	       mqd->cp_hqd_pq_base_hi);
3928 
3929 	/* set up the HQD, this is similar to CP_RB0_CNTL */
3930 	WREG32_SOC15(GC, 0, regCP_HQD_PQ_CONTROL,
3931 	       mqd->cp_hqd_pq_control);
3932 
3933 	/* set the wb address whether it's enabled or not */
3934 	WREG32_SOC15(GC, 0, regCP_HQD_PQ_RPTR_REPORT_ADDR,
3935 		mqd->cp_hqd_pq_rptr_report_addr_lo);
3936 	WREG32_SOC15(GC, 0, regCP_HQD_PQ_RPTR_REPORT_ADDR_HI,
3937 		mqd->cp_hqd_pq_rptr_report_addr_hi);
3938 
3939 	/* only used if CP_PQ_WPTR_POLL_CNTL.CP_PQ_WPTR_POLL_CNTL__EN_MASK=1 */
3940 	WREG32_SOC15(GC, 0, regCP_HQD_PQ_WPTR_POLL_ADDR,
3941 	       mqd->cp_hqd_pq_wptr_poll_addr_lo);
3942 	WREG32_SOC15(GC, 0, regCP_HQD_PQ_WPTR_POLL_ADDR_HI,
3943 	       mqd->cp_hqd_pq_wptr_poll_addr_hi);
3944 
3945 	/* enable the doorbell if requested */
3946 	if (ring->use_doorbell) {
3947 		WREG32_SOC15(GC, 0, regCP_MEC_DOORBELL_RANGE_LOWER,
3948 			(adev->doorbell_index.kiq * 2) << 2);
3949 		WREG32_SOC15(GC, 0, regCP_MEC_DOORBELL_RANGE_UPPER,
3950 			(adev->doorbell_index.userqueue_end * 2) << 2);
3951 	}
3952 
3953 	WREG32_SOC15(GC, 0, regCP_HQD_PQ_DOORBELL_CONTROL,
3954 	       mqd->cp_hqd_pq_doorbell_control);
3955 
3956 	/* reset read and write pointers, similar to CP_RB0_WPTR/_RPTR */
3957 	WREG32_SOC15(GC, 0, regCP_HQD_PQ_WPTR_LO,
3958 	       mqd->cp_hqd_pq_wptr_lo);
3959 	WREG32_SOC15(GC, 0, regCP_HQD_PQ_WPTR_HI,
3960 	       mqd->cp_hqd_pq_wptr_hi);
3961 
3962 	/* set the vmid for the queue */
3963 	WREG32_SOC15(GC, 0, regCP_HQD_VMID, mqd->cp_hqd_vmid);
3964 
3965 	WREG32_SOC15(GC, 0, regCP_HQD_PERSISTENT_STATE,
3966 	       mqd->cp_hqd_persistent_state);
3967 
3968 	/* activate the queue */
3969 	WREG32_SOC15(GC, 0, regCP_HQD_ACTIVE,
3970 	       mqd->cp_hqd_active);
3971 
3972 	if (ring->use_doorbell)
3973 		WREG32_FIELD15_PREREG(GC, 0, CP_PQ_STATUS, DOORBELL_ENABLE, 1);
3974 
3975 	return 0;
3976 }
3977 
3978 static int gfx_v11_0_kiq_init_queue(struct amdgpu_ring *ring)
3979 {
3980 	struct amdgpu_device *adev = ring->adev;
3981 	struct v11_compute_mqd *mqd = ring->mqd_ptr;
3982 
3983 	gfx_v11_0_kiq_setting(ring);
3984 
3985 	if (amdgpu_in_reset(adev)) { /* for GPU_RESET case */
3986 		/* reset MQD to a clean status */
3987 		if (adev->gfx.kiq[0].mqd_backup)
3988 			memcpy_toio(mqd, adev->gfx.kiq[0].mqd_backup, sizeof(*mqd));
3989 
3990 		/* reset ring buffer */
3991 		ring->wptr = 0;
3992 		amdgpu_ring_clear_ring(ring);
3993 
3994 		mutex_lock(&adev->srbm_mutex);
3995 		soc21_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
3996 		gfx_v11_0_kiq_init_register(ring);
3997 		soc21_grbm_select(adev, 0, 0, 0, 0);
3998 		mutex_unlock(&adev->srbm_mutex);
3999 	} else {
4000 		memset((void *)mqd, 0, sizeof(*mqd));
4001 		if (amdgpu_sriov_vf(adev) && adev->in_suspend)
4002 			amdgpu_ring_clear_ring(ring);
4003 		mutex_lock(&adev->srbm_mutex);
4004 		soc21_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
4005 		amdgpu_ring_init_mqd(ring);
4006 		gfx_v11_0_kiq_init_register(ring);
4007 		soc21_grbm_select(adev, 0, 0, 0, 0);
4008 		mutex_unlock(&adev->srbm_mutex);
4009 
4010 		if (adev->gfx.kiq[0].mqd_backup)
4011 			memcpy_fromio(adev->gfx.kiq[0].mqd_backup, mqd, sizeof(*mqd));
4012 	}
4013 
4014 	return 0;
4015 }
4016 
4017 static int gfx_v11_0_kcq_init_queue(struct amdgpu_ring *ring)
4018 {
4019 	struct amdgpu_device *adev = ring->adev;
4020 	struct v11_compute_mqd *mqd = ring->mqd_ptr;
4021 	int mqd_idx = ring - &adev->gfx.compute_ring[0];
4022 
4023 	if (!amdgpu_in_reset(adev) && !adev->in_suspend) {
4024 		memset((void *)mqd, 0, sizeof(*mqd));
4025 		mutex_lock(&adev->srbm_mutex);
4026 		soc21_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
4027 		amdgpu_ring_init_mqd(ring);
4028 		soc21_grbm_select(adev, 0, 0, 0, 0);
4029 		mutex_unlock(&adev->srbm_mutex);
4030 
4031 		if (adev->gfx.mec.mqd_backup[mqd_idx])
4032 			memcpy_fromio(adev->gfx.mec.mqd_backup[mqd_idx], mqd, sizeof(*mqd));
4033 	} else {
4034 		/* restore MQD to a clean status */
4035 		if (adev->gfx.mec.mqd_backup[mqd_idx])
4036 			memcpy_toio(mqd, adev->gfx.mec.mqd_backup[mqd_idx], sizeof(*mqd));
4037 		/* reset ring buffer */
4038 		ring->wptr = 0;
4039 		atomic64_set((atomic64_t *)ring->wptr_cpu_addr, 0);
4040 		amdgpu_ring_clear_ring(ring);
4041 	}
4042 
4043 	return 0;
4044 }
4045 
4046 static int gfx_v11_0_kiq_resume(struct amdgpu_device *adev)
4047 {
4048 	struct amdgpu_ring *ring;
4049 	int r;
4050 
4051 	ring = &adev->gfx.kiq[0].ring;
4052 
4053 	r = amdgpu_bo_reserve(ring->mqd_obj, false);
4054 	if (unlikely(r != 0))
4055 		return r;
4056 
4057 	r = amdgpu_bo_kmap(ring->mqd_obj, (void **)&ring->mqd_ptr);
4058 	if (unlikely(r != 0)) {
4059 		amdgpu_bo_unreserve(ring->mqd_obj);
4060 		return r;
4061 	}
4062 
4063 	gfx_v11_0_kiq_init_queue(ring);
4064 	amdgpu_bo_kunmap(ring->mqd_obj);
4065 	ring->mqd_ptr = NULL;
4066 	amdgpu_bo_unreserve(ring->mqd_obj);
4067 	ring->sched.ready = true;
4068 	return 0;
4069 }
4070 
4071 static int gfx_v11_0_kcq_resume(struct amdgpu_device *adev)
4072 {
4073 	struct amdgpu_ring *ring = NULL;
4074 	int r = 0, i;
4075 
4076 	if (!amdgpu_async_gfx_ring)
4077 		gfx_v11_0_cp_compute_enable(adev, true);
4078 
4079 	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
4080 		ring = &adev->gfx.compute_ring[i];
4081 
4082 		r = amdgpu_bo_reserve(ring->mqd_obj, false);
4083 		if (unlikely(r != 0))
4084 			goto done;
4085 		r = amdgpu_bo_kmap(ring->mqd_obj, (void **)&ring->mqd_ptr);
4086 		if (!r) {
4087 			r = gfx_v11_0_kcq_init_queue(ring);
4088 			amdgpu_bo_kunmap(ring->mqd_obj);
4089 			ring->mqd_ptr = NULL;
4090 		}
4091 		amdgpu_bo_unreserve(ring->mqd_obj);
4092 		if (r)
4093 			goto done;
4094 	}
4095 
4096 	r = amdgpu_gfx_enable_kcq(adev, 0);
4097 done:
4098 	return r;
4099 }
4100 
4101 static int gfx_v11_0_cp_resume(struct amdgpu_device *adev)
4102 {
4103 	int r, i;
4104 	struct amdgpu_ring *ring;
4105 
4106 	if (!(adev->flags & AMD_IS_APU))
4107 		gfx_v11_0_enable_gui_idle_interrupt(adev, false);
4108 
4109 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_DIRECT) {
4110 		/* legacy firmware loading */
4111 		r = gfx_v11_0_cp_gfx_load_microcode(adev);
4112 		if (r)
4113 			return r;
4114 
4115 		if (adev->gfx.rs64_enable)
4116 			r = gfx_v11_0_cp_compute_load_microcode_rs64(adev);
4117 		else
4118 			r = gfx_v11_0_cp_compute_load_microcode(adev);
4119 		if (r)
4120 			return r;
4121 	}
4122 
4123 	gfx_v11_0_cp_set_doorbell_range(adev);
4124 
4125 	if (amdgpu_async_gfx_ring) {
4126 		gfx_v11_0_cp_compute_enable(adev, true);
4127 		gfx_v11_0_cp_gfx_enable(adev, true);
4128 	}
4129 
4130 	if (adev->enable_mes_kiq && adev->mes.kiq_hw_init)
4131 		r = amdgpu_mes_kiq_hw_init(adev);
4132 	else
4133 		r = gfx_v11_0_kiq_resume(adev);
4134 	if (r)
4135 		return r;
4136 
4137 	r = gfx_v11_0_kcq_resume(adev);
4138 	if (r)
4139 		return r;
4140 
4141 	if (!amdgpu_async_gfx_ring) {
4142 		r = gfx_v11_0_cp_gfx_resume(adev);
4143 		if (r)
4144 			return r;
4145 	} else {
4146 		r = gfx_v11_0_cp_async_gfx_ring_resume(adev);
4147 		if (r)
4148 			return r;
4149 	}
4150 
4151 	for (i = 0; i < adev->gfx.num_gfx_rings; i++) {
4152 		ring = &adev->gfx.gfx_ring[i];
4153 		r = amdgpu_ring_test_helper(ring);
4154 		if (r)
4155 			return r;
4156 	}
4157 
4158 	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
4159 		ring = &adev->gfx.compute_ring[i];
4160 		r = amdgpu_ring_test_helper(ring);
4161 		if (r)
4162 			return r;
4163 	}
4164 
4165 	return 0;
4166 }
4167 
4168 static void gfx_v11_0_cp_enable(struct amdgpu_device *adev, bool enable)
4169 {
4170 	gfx_v11_0_cp_gfx_enable(adev, enable);
4171 	gfx_v11_0_cp_compute_enable(adev, enable);
4172 }
4173 
4174 static int gfx_v11_0_gfxhub_enable(struct amdgpu_device *adev)
4175 {
4176 	int r;
4177 	bool value;
4178 
4179 	r = adev->gfxhub.funcs->gart_enable(adev);
4180 	if (r)
4181 		return r;
4182 
4183 	adev->hdp.funcs->flush_hdp(adev, NULL);
4184 
4185 	value = (amdgpu_vm_fault_stop == AMDGPU_VM_FAULT_STOP_ALWAYS) ?
4186 		false : true;
4187 
4188 	adev->gfxhub.funcs->set_fault_enable_default(adev, value);
4189 	amdgpu_gmc_flush_gpu_tlb(adev, 0, AMDGPU_GFXHUB(0), 0);
4190 
4191 	return 0;
4192 }
4193 
4194 static void gfx_v11_0_select_cp_fw_arch(struct amdgpu_device *adev)
4195 {
4196 	u32 tmp;
4197 
4198 	/* select RS64 */
4199 	if (adev->gfx.rs64_enable) {
4200 		tmp = RREG32_SOC15(GC, 0, regCP_GFX_CNTL);
4201 		tmp = REG_SET_FIELD(tmp, CP_GFX_CNTL, ENGINE_SEL, 1);
4202 		WREG32_SOC15(GC, 0, regCP_GFX_CNTL, tmp);
4203 
4204 		tmp = RREG32_SOC15(GC, 0, regCP_MEC_ISA_CNTL);
4205 		tmp = REG_SET_FIELD(tmp, CP_MEC_ISA_CNTL, ISA_MODE, 1);
4206 		WREG32_SOC15(GC, 0, regCP_MEC_ISA_CNTL, tmp);
4207 	}
4208 
4209 	if (amdgpu_emu_mode == 1)
4210 		msleep(100);
4211 }
4212 
4213 static int get_gb_addr_config(struct amdgpu_device * adev)
4214 {
4215 	u32 gb_addr_config;
4216 
4217 	gb_addr_config = RREG32_SOC15(GC, 0, regGB_ADDR_CONFIG);
4218 	if (gb_addr_config == 0)
4219 		return -EINVAL;
4220 
4221 	adev->gfx.config.gb_addr_config_fields.num_pkrs =
4222 		1 << REG_GET_FIELD(gb_addr_config, GB_ADDR_CONFIG, NUM_PKRS);
4223 
4224 	adev->gfx.config.gb_addr_config = gb_addr_config;
4225 
4226 	adev->gfx.config.gb_addr_config_fields.num_pipes = 1 <<
4227 			REG_GET_FIELD(adev->gfx.config.gb_addr_config,
4228 				      GB_ADDR_CONFIG, NUM_PIPES);
4229 
4230 	adev->gfx.config.max_tile_pipes =
4231 		adev->gfx.config.gb_addr_config_fields.num_pipes;
4232 
4233 	adev->gfx.config.gb_addr_config_fields.max_compress_frags = 1 <<
4234 			REG_GET_FIELD(adev->gfx.config.gb_addr_config,
4235 				      GB_ADDR_CONFIG, MAX_COMPRESSED_FRAGS);
4236 	adev->gfx.config.gb_addr_config_fields.num_rb_per_se = 1 <<
4237 			REG_GET_FIELD(adev->gfx.config.gb_addr_config,
4238 				      GB_ADDR_CONFIG, NUM_RB_PER_SE);
4239 	adev->gfx.config.gb_addr_config_fields.num_se = 1 <<
4240 			REG_GET_FIELD(adev->gfx.config.gb_addr_config,
4241 				      GB_ADDR_CONFIG, NUM_SHADER_ENGINES);
4242 	adev->gfx.config.gb_addr_config_fields.pipe_interleave_size = 1 << (8 +
4243 			REG_GET_FIELD(adev->gfx.config.gb_addr_config,
4244 				      GB_ADDR_CONFIG, PIPE_INTERLEAVE_SIZE));
4245 
4246 	return 0;
4247 }
4248 
4249 static void gfx_v11_0_disable_gpa_mode(struct amdgpu_device *adev)
4250 {
4251 	uint32_t data;
4252 
4253 	data = RREG32_SOC15(GC, 0, regCPC_PSP_DEBUG);
4254 	data |= CPC_PSP_DEBUG__GPA_OVERRIDE_MASK;
4255 	WREG32_SOC15(GC, 0, regCPC_PSP_DEBUG, data);
4256 
4257 	data = RREG32_SOC15(GC, 0, regCPG_PSP_DEBUG);
4258 	data |= CPG_PSP_DEBUG__GPA_OVERRIDE_MASK;
4259 	WREG32_SOC15(GC, 0, regCPG_PSP_DEBUG, data);
4260 }
4261 
4262 static int gfx_v11_0_hw_init(void *handle)
4263 {
4264 	int r;
4265 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4266 
4267 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_RLC_BACKDOOR_AUTO) {
4268 		if (adev->gfx.imu.funcs) {
4269 			/* RLC autoload sequence 1: Program rlc ram */
4270 			if (adev->gfx.imu.funcs->program_rlc_ram)
4271 				adev->gfx.imu.funcs->program_rlc_ram(adev);
4272 		}
4273 		/* rlc autoload firmware */
4274 		r = gfx_v11_0_rlc_backdoor_autoload_enable(adev);
4275 		if (r)
4276 			return r;
4277 	} else {
4278 		if (adev->firmware.load_type == AMDGPU_FW_LOAD_DIRECT) {
4279 			if (adev->gfx.imu.funcs && (amdgpu_dpm > 0)) {
4280 				if (adev->gfx.imu.funcs->load_microcode)
4281 					adev->gfx.imu.funcs->load_microcode(adev);
4282 				if (adev->gfx.imu.funcs->setup_imu)
4283 					adev->gfx.imu.funcs->setup_imu(adev);
4284 				if (adev->gfx.imu.funcs->start_imu)
4285 					adev->gfx.imu.funcs->start_imu(adev);
4286 			}
4287 
4288 			/* disable gpa mode in backdoor loading */
4289 			gfx_v11_0_disable_gpa_mode(adev);
4290 		}
4291 	}
4292 
4293 	if ((adev->firmware.load_type == AMDGPU_FW_LOAD_RLC_BACKDOOR_AUTO) ||
4294 	    (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP)) {
4295 		r = gfx_v11_0_wait_for_rlc_autoload_complete(adev);
4296 		if (r) {
4297 			dev_err(adev->dev, "(%d) failed to wait rlc autoload complete\n", r);
4298 			return r;
4299 		}
4300 	}
4301 
4302 	adev->gfx.is_poweron = true;
4303 
4304 	if(get_gb_addr_config(adev))
4305 		DRM_WARN("Invalid gb_addr_config !\n");
4306 
4307 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP &&
4308 	    adev->gfx.rs64_enable)
4309 		gfx_v11_0_config_gfx_rs64(adev);
4310 
4311 	r = gfx_v11_0_gfxhub_enable(adev);
4312 	if (r)
4313 		return r;
4314 
4315 	if (!amdgpu_emu_mode)
4316 		gfx_v11_0_init_golden_registers(adev);
4317 
4318 	if ((adev->firmware.load_type == AMDGPU_FW_LOAD_DIRECT) ||
4319 	    (adev->firmware.load_type == AMDGPU_FW_LOAD_RLC_BACKDOOR_AUTO && amdgpu_dpm == 1)) {
4320 		/**
4321 		 * For gfx 11, rlc firmware loading relies on smu firmware is
4322 		 * loaded firstly, so in direct type, it has to load smc ucode
4323 		 * here before rlc.
4324 		 */
4325 		if (!(adev->flags & AMD_IS_APU)) {
4326 			r = amdgpu_pm_load_smu_firmware(adev, NULL);
4327 			if (r)
4328 				return r;
4329 		}
4330 	}
4331 
4332 	gfx_v11_0_constants_init(adev);
4333 
4334 	if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP)
4335 		gfx_v11_0_select_cp_fw_arch(adev);
4336 
4337 	if (adev->nbio.funcs->gc_doorbell_init)
4338 		adev->nbio.funcs->gc_doorbell_init(adev);
4339 
4340 	r = gfx_v11_0_rlc_resume(adev);
4341 	if (r)
4342 		return r;
4343 
4344 	/*
4345 	 * init golden registers and rlc resume may override some registers,
4346 	 * reconfig them here
4347 	 */
4348 	gfx_v11_0_tcp_harvest(adev);
4349 
4350 	r = gfx_v11_0_cp_resume(adev);
4351 	if (r)
4352 		return r;
4353 
4354 	return r;
4355 }
4356 
4357 static int gfx_v11_0_hw_fini(void *handle)
4358 {
4359 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4360 
4361 	amdgpu_irq_put(adev, &adev->gfx.priv_reg_irq, 0);
4362 	amdgpu_irq_put(adev, &adev->gfx.priv_inst_irq, 0);
4363 
4364 	if (!adev->no_hw_access) {
4365 		if (amdgpu_async_gfx_ring) {
4366 			if (amdgpu_gfx_disable_kgq(adev, 0))
4367 				DRM_ERROR("KGQ disable failed\n");
4368 		}
4369 
4370 		if (amdgpu_gfx_disable_kcq(adev, 0))
4371 			DRM_ERROR("KCQ disable failed\n");
4372 
4373 		amdgpu_mes_kiq_hw_fini(adev);
4374 	}
4375 
4376 	if (amdgpu_sriov_vf(adev))
4377 		/* Remove the steps disabling CPG and clearing KIQ position,
4378 		 * so that CP could perform IDLE-SAVE during switch. Those
4379 		 * steps are necessary to avoid a DMAR error in gfx9 but it is
4380 		 * not reproduced on gfx11.
4381 		 */
4382 		return 0;
4383 
4384 	gfx_v11_0_cp_enable(adev, false);
4385 	gfx_v11_0_enable_gui_idle_interrupt(adev, false);
4386 
4387 	adev->gfxhub.funcs->gart_disable(adev);
4388 
4389 	adev->gfx.is_poweron = false;
4390 
4391 	return 0;
4392 }
4393 
4394 static int gfx_v11_0_suspend(void *handle)
4395 {
4396 	return gfx_v11_0_hw_fini(handle);
4397 }
4398 
4399 static int gfx_v11_0_resume(void *handle)
4400 {
4401 	return gfx_v11_0_hw_init(handle);
4402 }
4403 
4404 static bool gfx_v11_0_is_idle(void *handle)
4405 {
4406 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4407 
4408 	if (REG_GET_FIELD(RREG32_SOC15(GC, 0, regGRBM_STATUS),
4409 				GRBM_STATUS, GUI_ACTIVE))
4410 		return false;
4411 	else
4412 		return true;
4413 }
4414 
4415 static int gfx_v11_0_wait_for_idle(void *handle)
4416 {
4417 	unsigned i;
4418 	u32 tmp;
4419 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4420 
4421 	for (i = 0; i < adev->usec_timeout; i++) {
4422 		/* read MC_STATUS */
4423 		tmp = RREG32_SOC15(GC, 0, regGRBM_STATUS) &
4424 			GRBM_STATUS__GUI_ACTIVE_MASK;
4425 
4426 		if (!REG_GET_FIELD(tmp, GRBM_STATUS, GUI_ACTIVE))
4427 			return 0;
4428 		udelay(1);
4429 	}
4430 	return -ETIMEDOUT;
4431 }
4432 
4433 static int gfx_v11_0_soft_reset(void *handle)
4434 {
4435 	u32 grbm_soft_reset = 0;
4436 	u32 tmp;
4437 	int i, j, k;
4438 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4439 
4440 	tmp = RREG32_SOC15(GC, 0, regCP_INT_CNTL);
4441 	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL, CMP_BUSY_INT_ENABLE, 0);
4442 	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL, CNTX_BUSY_INT_ENABLE, 0);
4443 	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL, CNTX_EMPTY_INT_ENABLE, 0);
4444 	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL, GFX_IDLE_INT_ENABLE, 0);
4445 	WREG32_SOC15(GC, 0, regCP_INT_CNTL, tmp);
4446 
4447 	gfx_v11_0_set_safe_mode(adev, 0);
4448 
4449 	for (i = 0; i < adev->gfx.mec.num_mec; ++i) {
4450 		for (j = 0; j < adev->gfx.mec.num_queue_per_pipe; j++) {
4451 			for (k = 0; k < adev->gfx.mec.num_pipe_per_mec; k++) {
4452 				tmp = RREG32_SOC15(GC, 0, regGRBM_GFX_CNTL);
4453 				tmp = REG_SET_FIELD(tmp, GRBM_GFX_CNTL, MEID, i);
4454 				tmp = REG_SET_FIELD(tmp, GRBM_GFX_CNTL, QUEUEID, j);
4455 				tmp = REG_SET_FIELD(tmp, GRBM_GFX_CNTL, PIPEID, k);
4456 				WREG32_SOC15(GC, 0, regGRBM_GFX_CNTL, tmp);
4457 
4458 				WREG32_SOC15(GC, 0, regCP_HQD_DEQUEUE_REQUEST, 0x2);
4459 				WREG32_SOC15(GC, 0, regSPI_COMPUTE_QUEUE_RESET, 0x1);
4460 			}
4461 		}
4462 	}
4463 	for (i = 0; i < adev->gfx.me.num_me; ++i) {
4464 		for (j = 0; j < adev->gfx.me.num_queue_per_pipe; j++) {
4465 			for (k = 0; k < adev->gfx.me.num_pipe_per_me; k++) {
4466 				tmp = RREG32_SOC15(GC, 0, regGRBM_GFX_CNTL);
4467 				tmp = REG_SET_FIELD(tmp, GRBM_GFX_CNTL, MEID, i);
4468 				tmp = REG_SET_FIELD(tmp, GRBM_GFX_CNTL, QUEUEID, j);
4469 				tmp = REG_SET_FIELD(tmp, GRBM_GFX_CNTL, PIPEID, k);
4470 				WREG32_SOC15(GC, 0, regGRBM_GFX_CNTL, tmp);
4471 
4472 				WREG32_SOC15(GC, 0, regCP_GFX_HQD_DEQUEUE_REQUEST, 0x1);
4473 			}
4474 		}
4475 	}
4476 
4477 	WREG32_SOC15(GC, 0, regCP_VMID_RESET, 0xfffffffe);
4478 
4479 	// Read CP_VMID_RESET register three times.
4480 	// to get sufficient time for GFX_HQD_ACTIVE reach 0
4481 	RREG32_SOC15(GC, 0, regCP_VMID_RESET);
4482 	RREG32_SOC15(GC, 0, regCP_VMID_RESET);
4483 	RREG32_SOC15(GC, 0, regCP_VMID_RESET);
4484 
4485 	for (i = 0; i < adev->usec_timeout; i++) {
4486 		if (!RREG32_SOC15(GC, 0, regCP_HQD_ACTIVE) &&
4487 		    !RREG32_SOC15(GC, 0, regCP_GFX_HQD_ACTIVE))
4488 			break;
4489 		udelay(1);
4490 	}
4491 	if (i >= adev->usec_timeout) {
4492 		printk("Failed to wait all pipes clean\n");
4493 		return -EINVAL;
4494 	}
4495 
4496 	/**********  trigger soft reset  ***********/
4497 	grbm_soft_reset = RREG32_SOC15(GC, 0, regGRBM_SOFT_RESET);
4498 	grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET,
4499 					SOFT_RESET_CP, 1);
4500 	grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET,
4501 					SOFT_RESET_GFX, 1);
4502 	grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET,
4503 					SOFT_RESET_CPF, 1);
4504 	grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET,
4505 					SOFT_RESET_CPC, 1);
4506 	grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET,
4507 					SOFT_RESET_CPG, 1);
4508 	WREG32_SOC15(GC, 0, regGRBM_SOFT_RESET, grbm_soft_reset);
4509 	/**********  exit soft reset  ***********/
4510 	grbm_soft_reset = RREG32_SOC15(GC, 0, regGRBM_SOFT_RESET);
4511 	grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET,
4512 					SOFT_RESET_CP, 0);
4513 	grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET,
4514 					SOFT_RESET_GFX, 0);
4515 	grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET,
4516 					SOFT_RESET_CPF, 0);
4517 	grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET,
4518 					SOFT_RESET_CPC, 0);
4519 	grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET,
4520 					SOFT_RESET_CPG, 0);
4521 	WREG32_SOC15(GC, 0, regGRBM_SOFT_RESET, grbm_soft_reset);
4522 
4523 	tmp = RREG32_SOC15(GC, 0, regCP_SOFT_RESET_CNTL);
4524 	tmp = REG_SET_FIELD(tmp, CP_SOFT_RESET_CNTL, CMP_HQD_REG_RESET, 0x1);
4525 	WREG32_SOC15(GC, 0, regCP_SOFT_RESET_CNTL, tmp);
4526 
4527 	WREG32_SOC15(GC, 0, regCP_ME_CNTL, 0x0);
4528 	WREG32_SOC15(GC, 0, regCP_MEC_RS64_CNTL, 0x0);
4529 
4530 	for (i = 0; i < adev->usec_timeout; i++) {
4531 		if (!RREG32_SOC15(GC, 0, regCP_VMID_RESET))
4532 			break;
4533 		udelay(1);
4534 	}
4535 	if (i >= adev->usec_timeout) {
4536 		printk("Failed to wait CP_VMID_RESET to 0\n");
4537 		return -EINVAL;
4538 	}
4539 
4540 	tmp = RREG32_SOC15(GC, 0, regCP_INT_CNTL);
4541 	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL, CMP_BUSY_INT_ENABLE, 1);
4542 	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL, CNTX_BUSY_INT_ENABLE, 1);
4543 	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL, CNTX_EMPTY_INT_ENABLE, 1);
4544 	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL, GFX_IDLE_INT_ENABLE, 1);
4545 	WREG32_SOC15(GC, 0, regCP_INT_CNTL, tmp);
4546 
4547 	gfx_v11_0_unset_safe_mode(adev, 0);
4548 
4549 	return gfx_v11_0_cp_resume(adev);
4550 }
4551 
4552 static bool gfx_v11_0_check_soft_reset(void *handle)
4553 {
4554 	int i, r;
4555 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4556 	struct amdgpu_ring *ring;
4557 	long tmo = msecs_to_jiffies(1000);
4558 
4559 	for (i = 0; i < adev->gfx.num_gfx_rings; i++) {
4560 		ring = &adev->gfx.gfx_ring[i];
4561 		r = amdgpu_ring_test_ib(ring, tmo);
4562 		if (r)
4563 			return true;
4564 	}
4565 
4566 	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
4567 		ring = &adev->gfx.compute_ring[i];
4568 		r = amdgpu_ring_test_ib(ring, tmo);
4569 		if (r)
4570 			return true;
4571 	}
4572 
4573 	return false;
4574 }
4575 
4576 static int gfx_v11_0_post_soft_reset(void *handle)
4577 {
4578 	/**
4579 	 * GFX soft reset will impact MES, need resume MES when do GFX soft reset
4580 	 */
4581 	return amdgpu_mes_resume((struct amdgpu_device *)handle);
4582 }
4583 
4584 static uint64_t gfx_v11_0_get_gpu_clock_counter(struct amdgpu_device *adev)
4585 {
4586 	uint64_t clock;
4587 	uint64_t clock_counter_lo, clock_counter_hi_pre, clock_counter_hi_after;
4588 
4589 	if (amdgpu_sriov_vf(adev)) {
4590 		amdgpu_gfx_off_ctrl(adev, false);
4591 		mutex_lock(&adev->gfx.gpu_clock_mutex);
4592 		clock_counter_hi_pre = (uint64_t)RREG32_SOC15(GC, 0, regCP_MES_MTIME_HI);
4593 		clock_counter_lo = (uint64_t)RREG32_SOC15(GC, 0, regCP_MES_MTIME_LO);
4594 		clock_counter_hi_after = (uint64_t)RREG32_SOC15(GC, 0, regCP_MES_MTIME_HI);
4595 		if (clock_counter_hi_pre != clock_counter_hi_after)
4596 			clock_counter_lo = (uint64_t)RREG32_SOC15(GC, 0, regCP_MES_MTIME_LO);
4597 		mutex_unlock(&adev->gfx.gpu_clock_mutex);
4598 		amdgpu_gfx_off_ctrl(adev, true);
4599 	} else {
4600 		preempt_disable();
4601 		clock_counter_hi_pre = (uint64_t)RREG32_SOC15(SMUIO, 0, regGOLDEN_TSC_COUNT_UPPER);
4602 		clock_counter_lo = (uint64_t)RREG32_SOC15(SMUIO, 0, regGOLDEN_TSC_COUNT_LOWER);
4603 		clock_counter_hi_after = (uint64_t)RREG32_SOC15(SMUIO, 0, regGOLDEN_TSC_COUNT_UPPER);
4604 		if (clock_counter_hi_pre != clock_counter_hi_after)
4605 			clock_counter_lo = (uint64_t)RREG32_SOC15(SMUIO, 0, regGOLDEN_TSC_COUNT_LOWER);
4606 		preempt_enable();
4607 	}
4608 	clock = clock_counter_lo | (clock_counter_hi_after << 32ULL);
4609 
4610 	return clock;
4611 }
4612 
4613 static void gfx_v11_0_ring_emit_gds_switch(struct amdgpu_ring *ring,
4614 					   uint32_t vmid,
4615 					   uint32_t gds_base, uint32_t gds_size,
4616 					   uint32_t gws_base, uint32_t gws_size,
4617 					   uint32_t oa_base, uint32_t oa_size)
4618 {
4619 	struct amdgpu_device *adev = ring->adev;
4620 
4621 	/* GDS Base */
4622 	gfx_v11_0_write_data_to_reg(ring, 0, false,
4623 				    SOC15_REG_OFFSET(GC, 0, regGDS_VMID0_BASE) + 2 * vmid,
4624 				    gds_base);
4625 
4626 	/* GDS Size */
4627 	gfx_v11_0_write_data_to_reg(ring, 0, false,
4628 				    SOC15_REG_OFFSET(GC, 0, regGDS_VMID0_SIZE) + 2 * vmid,
4629 				    gds_size);
4630 
4631 	/* GWS */
4632 	gfx_v11_0_write_data_to_reg(ring, 0, false,
4633 				    SOC15_REG_OFFSET(GC, 0, regGDS_GWS_VMID0) + vmid,
4634 				    gws_size << GDS_GWS_VMID0__SIZE__SHIFT | gws_base);
4635 
4636 	/* OA */
4637 	gfx_v11_0_write_data_to_reg(ring, 0, false,
4638 				    SOC15_REG_OFFSET(GC, 0, regGDS_OA_VMID0) + vmid,
4639 				    (1 << (oa_size + oa_base)) - (1 << oa_base));
4640 }
4641 
4642 static int gfx_v11_0_early_init(void *handle)
4643 {
4644 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4645 
4646 	adev->gfx.funcs = &gfx_v11_0_gfx_funcs;
4647 
4648 	adev->gfx.num_gfx_rings = GFX11_NUM_GFX_RINGS;
4649 	adev->gfx.num_compute_rings = min(amdgpu_gfx_get_num_kcq(adev),
4650 					  AMDGPU_MAX_COMPUTE_RINGS);
4651 
4652 	gfx_v11_0_set_kiq_pm4_funcs(adev);
4653 	gfx_v11_0_set_ring_funcs(adev);
4654 	gfx_v11_0_set_irq_funcs(adev);
4655 	gfx_v11_0_set_gds_init(adev);
4656 	gfx_v11_0_set_rlc_funcs(adev);
4657 	gfx_v11_0_set_mqd_funcs(adev);
4658 	gfx_v11_0_set_imu_funcs(adev);
4659 
4660 	gfx_v11_0_init_rlcg_reg_access_ctrl(adev);
4661 
4662 	return gfx_v11_0_init_microcode(adev);
4663 }
4664 
4665 static int gfx_v11_0_late_init(void *handle)
4666 {
4667 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4668 	int r;
4669 
4670 	r = amdgpu_irq_get(adev, &adev->gfx.priv_reg_irq, 0);
4671 	if (r)
4672 		return r;
4673 
4674 	r = amdgpu_irq_get(adev, &adev->gfx.priv_inst_irq, 0);
4675 	if (r)
4676 		return r;
4677 
4678 	return 0;
4679 }
4680 
4681 static bool gfx_v11_0_is_rlc_enabled(struct amdgpu_device *adev)
4682 {
4683 	uint32_t rlc_cntl;
4684 
4685 	/* if RLC is not enabled, do nothing */
4686 	rlc_cntl = RREG32_SOC15(GC, 0, regRLC_CNTL);
4687 	return (REG_GET_FIELD(rlc_cntl, RLC_CNTL, RLC_ENABLE_F32)) ? true : false;
4688 }
4689 
4690 static void gfx_v11_0_set_safe_mode(struct amdgpu_device *adev, int xcc_id)
4691 {
4692 	uint32_t data;
4693 	unsigned i;
4694 
4695 	data = RLC_SAFE_MODE__CMD_MASK;
4696 	data |= (1 << RLC_SAFE_MODE__MESSAGE__SHIFT);
4697 
4698 	WREG32_SOC15(GC, 0, regRLC_SAFE_MODE, data);
4699 
4700 	/* wait for RLC_SAFE_MODE */
4701 	for (i = 0; i < adev->usec_timeout; i++) {
4702 		if (!REG_GET_FIELD(RREG32_SOC15(GC, 0, regRLC_SAFE_MODE),
4703 				   RLC_SAFE_MODE, CMD))
4704 			break;
4705 		udelay(1);
4706 	}
4707 }
4708 
4709 static void gfx_v11_0_unset_safe_mode(struct amdgpu_device *adev, int xcc_id)
4710 {
4711 	WREG32_SOC15(GC, 0, regRLC_SAFE_MODE, RLC_SAFE_MODE__CMD_MASK);
4712 }
4713 
4714 static void gfx_v11_0_update_perf_clk(struct amdgpu_device *adev,
4715 				      bool enable)
4716 {
4717 	uint32_t def, data;
4718 
4719 	if (!(adev->cg_flags & AMD_CG_SUPPORT_GFX_PERF_CLK))
4720 		return;
4721 
4722 	def = data = RREG32_SOC15(GC, 0, regRLC_CGTT_MGCG_OVERRIDE);
4723 
4724 	if (enable)
4725 		data &= ~RLC_CGTT_MGCG_OVERRIDE__PERFMON_CLOCK_STATE_MASK;
4726 	else
4727 		data |= RLC_CGTT_MGCG_OVERRIDE__PERFMON_CLOCK_STATE_MASK;
4728 
4729 	if (def != data)
4730 		WREG32_SOC15(GC, 0, regRLC_CGTT_MGCG_OVERRIDE, data);
4731 }
4732 
4733 static void gfx_v11_0_update_sram_fgcg(struct amdgpu_device *adev,
4734 				       bool enable)
4735 {
4736 	uint32_t def, data;
4737 
4738 	if (!(adev->cg_flags & AMD_CG_SUPPORT_GFX_FGCG))
4739 		return;
4740 
4741 	def = data = RREG32_SOC15(GC, 0, regRLC_CGTT_MGCG_OVERRIDE);
4742 
4743 	if (enable)
4744 		data &= ~RLC_CGTT_MGCG_OVERRIDE__GFXIP_FGCG_OVERRIDE_MASK;
4745 	else
4746 		data |= RLC_CGTT_MGCG_OVERRIDE__GFXIP_FGCG_OVERRIDE_MASK;
4747 
4748 	if (def != data)
4749 		WREG32_SOC15(GC, 0, regRLC_CGTT_MGCG_OVERRIDE, data);
4750 }
4751 
4752 static void gfx_v11_0_update_repeater_fgcg(struct amdgpu_device *adev,
4753 					   bool enable)
4754 {
4755 	uint32_t def, data;
4756 
4757 	if (!(adev->cg_flags & AMD_CG_SUPPORT_REPEATER_FGCG))
4758 		return;
4759 
4760 	def = data = RREG32_SOC15(GC, 0, regRLC_CGTT_MGCG_OVERRIDE);
4761 
4762 	if (enable)
4763 		data &= ~RLC_CGTT_MGCG_OVERRIDE__GFXIP_REPEATER_FGCG_OVERRIDE_MASK;
4764 	else
4765 		data |= RLC_CGTT_MGCG_OVERRIDE__GFXIP_REPEATER_FGCG_OVERRIDE_MASK;
4766 
4767 	if (def != data)
4768 		WREG32_SOC15(GC, 0, regRLC_CGTT_MGCG_OVERRIDE, data);
4769 }
4770 
4771 static void gfx_v11_0_update_medium_grain_clock_gating(struct amdgpu_device *adev,
4772 						       bool enable)
4773 {
4774 	uint32_t data, def;
4775 
4776 	if (!(adev->cg_flags & (AMD_CG_SUPPORT_GFX_MGCG | AMD_CG_SUPPORT_GFX_MGLS)))
4777 		return;
4778 
4779 	/* It is disabled by HW by default */
4780 	if (enable) {
4781 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGCG) {
4782 			/* 1 - RLC_CGTT_MGCG_OVERRIDE */
4783 			def = data = RREG32_SOC15(GC, 0, regRLC_CGTT_MGCG_OVERRIDE);
4784 
4785 			data &= ~(RLC_CGTT_MGCG_OVERRIDE__GRBM_CGTT_SCLK_OVERRIDE_MASK |
4786 				  RLC_CGTT_MGCG_OVERRIDE__RLC_CGTT_SCLK_OVERRIDE_MASK |
4787 				  RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGCG_OVERRIDE_MASK);
4788 
4789 			if (def != data)
4790 				WREG32_SOC15(GC, 0, regRLC_CGTT_MGCG_OVERRIDE, data);
4791 		}
4792 	} else {
4793 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGCG) {
4794 			def = data = RREG32_SOC15(GC, 0, regRLC_CGTT_MGCG_OVERRIDE);
4795 
4796 			data |= (RLC_CGTT_MGCG_OVERRIDE__RLC_CGTT_SCLK_OVERRIDE_MASK |
4797 				 RLC_CGTT_MGCG_OVERRIDE__GRBM_CGTT_SCLK_OVERRIDE_MASK |
4798 				 RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGCG_OVERRIDE_MASK);
4799 
4800 			if (def != data)
4801 				WREG32_SOC15(GC, 0, regRLC_CGTT_MGCG_OVERRIDE, data);
4802 		}
4803 	}
4804 }
4805 
4806 static void gfx_v11_0_update_coarse_grain_clock_gating(struct amdgpu_device *adev,
4807 						       bool enable)
4808 {
4809 	uint32_t def, data;
4810 
4811 	if (!(adev->cg_flags &
4812 	      (AMD_CG_SUPPORT_GFX_CGCG |
4813 	      AMD_CG_SUPPORT_GFX_CGLS |
4814 	      AMD_CG_SUPPORT_GFX_3D_CGCG |
4815 	      AMD_CG_SUPPORT_GFX_3D_CGLS)))
4816 		return;
4817 
4818 	if (enable) {
4819 		def = data = RREG32_SOC15(GC, 0, regRLC_CGTT_MGCG_OVERRIDE);
4820 
4821 		/* unset CGCG override */
4822 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGCG)
4823 			data &= ~RLC_CGTT_MGCG_OVERRIDE__GFXIP_CGCG_OVERRIDE_MASK;
4824 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGLS)
4825 			data &= ~RLC_CGTT_MGCG_OVERRIDE__GFXIP_CGLS_OVERRIDE_MASK;
4826 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_3D_CGCG ||
4827 		    adev->cg_flags & AMD_CG_SUPPORT_GFX_3D_CGLS)
4828 			data &= ~RLC_CGTT_MGCG_OVERRIDE__GFXIP_GFX3D_CG_OVERRIDE_MASK;
4829 
4830 		/* update CGCG override bits */
4831 		if (def != data)
4832 			WREG32_SOC15(GC, 0, regRLC_CGTT_MGCG_OVERRIDE, data);
4833 
4834 		/* enable cgcg FSM(0x0000363F) */
4835 		def = data = RREG32_SOC15(GC, 0, regRLC_CGCG_CGLS_CTRL);
4836 
4837 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGCG) {
4838 			data &= ~RLC_CGCG_CGLS_CTRL__CGCG_GFX_IDLE_THRESHOLD_MASK;
4839 			data |= (0x36 << RLC_CGCG_CGLS_CTRL__CGCG_GFX_IDLE_THRESHOLD__SHIFT) |
4840 				 RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK;
4841 		}
4842 
4843 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGLS) {
4844 			data &= ~RLC_CGCG_CGLS_CTRL__CGLS_REP_COMPANSAT_DELAY_MASK;
4845 			data |= (0x000F << RLC_CGCG_CGLS_CTRL__CGLS_REP_COMPANSAT_DELAY__SHIFT) |
4846 				 RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK;
4847 		}
4848 
4849 		if (def != data)
4850 			WREG32_SOC15(GC, 0, regRLC_CGCG_CGLS_CTRL, data);
4851 
4852 		/* Program RLC_CGCG_CGLS_CTRL_3D */
4853 		def = data = RREG32_SOC15(GC, 0, regRLC_CGCG_CGLS_CTRL_3D);
4854 
4855 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_3D_CGCG) {
4856 			data &= ~RLC_CGCG_CGLS_CTRL_3D__CGCG_GFX_IDLE_THRESHOLD_MASK;
4857 			data |= (0x36 << RLC_CGCG_CGLS_CTRL_3D__CGCG_GFX_IDLE_THRESHOLD__SHIFT) |
4858 				 RLC_CGCG_CGLS_CTRL_3D__CGCG_EN_MASK;
4859 		}
4860 
4861 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_3D_CGLS) {
4862 			data &= ~RLC_CGCG_CGLS_CTRL_3D__CGLS_REP_COMPANSAT_DELAY_MASK;
4863 			data |= (0xf << RLC_CGCG_CGLS_CTRL_3D__CGLS_REP_COMPANSAT_DELAY__SHIFT) |
4864 				 RLC_CGCG_CGLS_CTRL_3D__CGLS_EN_MASK;
4865 		}
4866 
4867 		if (def != data)
4868 			WREG32_SOC15(GC, 0, regRLC_CGCG_CGLS_CTRL_3D, data);
4869 
4870 		/* set IDLE_POLL_COUNT(0x00900100) */
4871 		def = data = RREG32_SOC15(GC, 0, regCP_RB_WPTR_POLL_CNTL);
4872 
4873 		data &= ~(CP_RB_WPTR_POLL_CNTL__POLL_FREQUENCY_MASK | CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT_MASK);
4874 		data |= (0x0100 << CP_RB_WPTR_POLL_CNTL__POLL_FREQUENCY__SHIFT) |
4875 			(0x0090 << CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT__SHIFT);
4876 
4877 		if (def != data)
4878 			WREG32_SOC15(GC, 0, regCP_RB_WPTR_POLL_CNTL, data);
4879 
4880 		data = RREG32_SOC15(GC, 0, regCP_INT_CNTL);
4881 		data = REG_SET_FIELD(data, CP_INT_CNTL, CNTX_BUSY_INT_ENABLE, 1);
4882 		data = REG_SET_FIELD(data, CP_INT_CNTL, CNTX_EMPTY_INT_ENABLE, 1);
4883 		data = REG_SET_FIELD(data, CP_INT_CNTL, CMP_BUSY_INT_ENABLE, 1);
4884 		data = REG_SET_FIELD(data, CP_INT_CNTL, GFX_IDLE_INT_ENABLE, 1);
4885 		WREG32_SOC15(GC, 0, regCP_INT_CNTL, data);
4886 
4887 		data = RREG32_SOC15(GC, 0, regSDMA0_RLC_CGCG_CTRL);
4888 		data = REG_SET_FIELD(data, SDMA0_RLC_CGCG_CTRL, CGCG_INT_ENABLE, 1);
4889 		WREG32_SOC15(GC, 0, regSDMA0_RLC_CGCG_CTRL, data);
4890 
4891 		/* Some ASICs only have one SDMA instance, not need to configure SDMA1 */
4892 		if (adev->sdma.num_instances > 1) {
4893 			data = RREG32_SOC15(GC, 0, regSDMA1_RLC_CGCG_CTRL);
4894 			data = REG_SET_FIELD(data, SDMA1_RLC_CGCG_CTRL, CGCG_INT_ENABLE, 1);
4895 			WREG32_SOC15(GC, 0, regSDMA1_RLC_CGCG_CTRL, data);
4896 		}
4897 	} else {
4898 		/* Program RLC_CGCG_CGLS_CTRL */
4899 		def = data = RREG32_SOC15(GC, 0, regRLC_CGCG_CGLS_CTRL);
4900 
4901 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGCG)
4902 			data &= ~RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK;
4903 
4904 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGLS)
4905 			data &= ~RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK;
4906 
4907 		if (def != data)
4908 			WREG32_SOC15(GC, 0, regRLC_CGCG_CGLS_CTRL, data);
4909 
4910 		/* Program RLC_CGCG_CGLS_CTRL_3D */
4911 		def = data = RREG32_SOC15(GC, 0, regRLC_CGCG_CGLS_CTRL_3D);
4912 
4913 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_3D_CGCG)
4914 			data &= ~RLC_CGCG_CGLS_CTRL_3D__CGCG_EN_MASK;
4915 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_3D_CGLS)
4916 			data &= ~RLC_CGCG_CGLS_CTRL_3D__CGLS_EN_MASK;
4917 
4918 		if (def != data)
4919 			WREG32_SOC15(GC, 0, regRLC_CGCG_CGLS_CTRL_3D, data);
4920 
4921 		data = RREG32_SOC15(GC, 0, regSDMA0_RLC_CGCG_CTRL);
4922 		data &= ~SDMA0_RLC_CGCG_CTRL__CGCG_INT_ENABLE_MASK;
4923 		WREG32_SOC15(GC, 0, regSDMA0_RLC_CGCG_CTRL, data);
4924 
4925 		/* Some ASICs only have one SDMA instance, not need to configure SDMA1 */
4926 		if (adev->sdma.num_instances > 1) {
4927 			data = RREG32_SOC15(GC, 0, regSDMA1_RLC_CGCG_CTRL);
4928 			data &= ~SDMA1_RLC_CGCG_CTRL__CGCG_INT_ENABLE_MASK;
4929 			WREG32_SOC15(GC, 0, regSDMA1_RLC_CGCG_CTRL, data);
4930 		}
4931 	}
4932 }
4933 
4934 static int gfx_v11_0_update_gfx_clock_gating(struct amdgpu_device *adev,
4935 					    bool enable)
4936 {
4937 	amdgpu_gfx_rlc_enter_safe_mode(adev, 0);
4938 
4939 	gfx_v11_0_update_coarse_grain_clock_gating(adev, enable);
4940 
4941 	gfx_v11_0_update_medium_grain_clock_gating(adev, enable);
4942 
4943 	gfx_v11_0_update_repeater_fgcg(adev, enable);
4944 
4945 	gfx_v11_0_update_sram_fgcg(adev, enable);
4946 
4947 	gfx_v11_0_update_perf_clk(adev, enable);
4948 
4949 	if (adev->cg_flags &
4950 	    (AMD_CG_SUPPORT_GFX_MGCG |
4951 	     AMD_CG_SUPPORT_GFX_CGLS |
4952 	     AMD_CG_SUPPORT_GFX_CGCG |
4953 	     AMD_CG_SUPPORT_GFX_3D_CGCG |
4954 	     AMD_CG_SUPPORT_GFX_3D_CGLS))
4955 	        gfx_v11_0_enable_gui_idle_interrupt(adev, enable);
4956 
4957 	amdgpu_gfx_rlc_exit_safe_mode(adev, 0);
4958 
4959 	return 0;
4960 }
4961 
4962 static void gfx_v11_0_update_spm_vmid(struct amdgpu_device *adev, unsigned vmid)
4963 {
4964 	u32 reg, data;
4965 
4966 	amdgpu_gfx_off_ctrl(adev, false);
4967 
4968 	reg = SOC15_REG_OFFSET(GC, 0, regRLC_SPM_MC_CNTL);
4969 	if (amdgpu_sriov_is_pp_one_vf(adev))
4970 		data = RREG32_NO_KIQ(reg);
4971 	else
4972 		data = RREG32(reg);
4973 
4974 	data &= ~RLC_SPM_MC_CNTL__RLC_SPM_VMID_MASK;
4975 	data |= (vmid & RLC_SPM_MC_CNTL__RLC_SPM_VMID_MASK) << RLC_SPM_MC_CNTL__RLC_SPM_VMID__SHIFT;
4976 
4977 	if (amdgpu_sriov_is_pp_one_vf(adev))
4978 		WREG32_SOC15_NO_KIQ(GC, 0, regRLC_SPM_MC_CNTL, data);
4979 	else
4980 		WREG32_SOC15(GC, 0, regRLC_SPM_MC_CNTL, data);
4981 
4982 	amdgpu_gfx_off_ctrl(adev, true);
4983 }
4984 
4985 static const struct amdgpu_rlc_funcs gfx_v11_0_rlc_funcs = {
4986 	.is_rlc_enabled = gfx_v11_0_is_rlc_enabled,
4987 	.set_safe_mode = gfx_v11_0_set_safe_mode,
4988 	.unset_safe_mode = gfx_v11_0_unset_safe_mode,
4989 	.init = gfx_v11_0_rlc_init,
4990 	.get_csb_size = gfx_v11_0_get_csb_size,
4991 	.get_csb_buffer = gfx_v11_0_get_csb_buffer,
4992 	.resume = gfx_v11_0_rlc_resume,
4993 	.stop = gfx_v11_0_rlc_stop,
4994 	.reset = gfx_v11_0_rlc_reset,
4995 	.start = gfx_v11_0_rlc_start,
4996 	.update_spm_vmid = gfx_v11_0_update_spm_vmid,
4997 };
4998 
4999 static void gfx_v11_cntl_power_gating(struct amdgpu_device *adev, bool enable)
5000 {
5001 	u32 data = RREG32_SOC15(GC, 0, regRLC_PG_CNTL);
5002 
5003 	if (enable && (adev->pg_flags & AMD_PG_SUPPORT_GFX_PG))
5004 		data |= RLC_PG_CNTL__GFX_POWER_GATING_ENABLE_MASK;
5005 	else
5006 		data &= ~RLC_PG_CNTL__GFX_POWER_GATING_ENABLE_MASK;
5007 
5008 	WREG32_SOC15(GC, 0, regRLC_PG_CNTL, data);
5009 
5010 	// Program RLC_PG_DELAY3 for CGPG hysteresis
5011 	if (enable && (adev->pg_flags & AMD_PG_SUPPORT_GFX_PG)) {
5012 		switch (adev->ip_versions[GC_HWIP][0]) {
5013 		case IP_VERSION(11, 0, 1):
5014 		case IP_VERSION(11, 0, 4):
5015 			WREG32_SOC15(GC, 0, regRLC_PG_DELAY_3, RLC_PG_DELAY_3_DEFAULT_GC_11_0_1);
5016 			break;
5017 		default:
5018 			break;
5019 		}
5020 	}
5021 }
5022 
5023 static void gfx_v11_cntl_pg(struct amdgpu_device *adev, bool enable)
5024 {
5025 	amdgpu_gfx_rlc_enter_safe_mode(adev, 0);
5026 
5027 	gfx_v11_cntl_power_gating(adev, enable);
5028 
5029 	amdgpu_gfx_rlc_exit_safe_mode(adev, 0);
5030 }
5031 
5032 static int gfx_v11_0_set_powergating_state(void *handle,
5033 					   enum amd_powergating_state state)
5034 {
5035 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5036 	bool enable = (state == AMD_PG_STATE_GATE);
5037 
5038 	if (amdgpu_sriov_vf(adev))
5039 		return 0;
5040 
5041 	switch (adev->ip_versions[GC_HWIP][0]) {
5042 	case IP_VERSION(11, 0, 0):
5043 	case IP_VERSION(11, 0, 2):
5044 	case IP_VERSION(11, 0, 3):
5045 		amdgpu_gfx_off_ctrl(adev, enable);
5046 		break;
5047 	case IP_VERSION(11, 0, 1):
5048 	case IP_VERSION(11, 0, 4):
5049 		if (!enable)
5050 			amdgpu_gfx_off_ctrl(adev, false);
5051 
5052 		gfx_v11_cntl_pg(adev, enable);
5053 
5054 		if (enable)
5055 			amdgpu_gfx_off_ctrl(adev, true);
5056 
5057 		break;
5058 	default:
5059 		break;
5060 	}
5061 
5062 	return 0;
5063 }
5064 
5065 static int gfx_v11_0_set_clockgating_state(void *handle,
5066 					  enum amd_clockgating_state state)
5067 {
5068 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5069 
5070 	if (amdgpu_sriov_vf(adev))
5071 	        return 0;
5072 
5073 	switch (adev->ip_versions[GC_HWIP][0]) {
5074 	case IP_VERSION(11, 0, 0):
5075 	case IP_VERSION(11, 0, 1):
5076 	case IP_VERSION(11, 0, 2):
5077 	case IP_VERSION(11, 0, 3):
5078 	case IP_VERSION(11, 0, 4):
5079 	        gfx_v11_0_update_gfx_clock_gating(adev,
5080 	                        state ==  AMD_CG_STATE_GATE);
5081 	        break;
5082 	default:
5083 	        break;
5084 	}
5085 
5086 	return 0;
5087 }
5088 
5089 static void gfx_v11_0_get_clockgating_state(void *handle, u64 *flags)
5090 {
5091 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5092 	int data;
5093 
5094 	/* AMD_CG_SUPPORT_GFX_MGCG */
5095 	data = RREG32_SOC15(GC, 0, regRLC_CGTT_MGCG_OVERRIDE);
5096 	if (!(data & RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGCG_OVERRIDE_MASK))
5097 		*flags |= AMD_CG_SUPPORT_GFX_MGCG;
5098 
5099 	/* AMD_CG_SUPPORT_REPEATER_FGCG */
5100 	if (!(data & RLC_CGTT_MGCG_OVERRIDE__GFXIP_REPEATER_FGCG_OVERRIDE_MASK))
5101 		*flags |= AMD_CG_SUPPORT_REPEATER_FGCG;
5102 
5103 	/* AMD_CG_SUPPORT_GFX_FGCG */
5104 	if (!(data & RLC_CGTT_MGCG_OVERRIDE__GFXIP_FGCG_OVERRIDE_MASK))
5105 		*flags |= AMD_CG_SUPPORT_GFX_FGCG;
5106 
5107 	/* AMD_CG_SUPPORT_GFX_PERF_CLK */
5108 	if (!(data & RLC_CGTT_MGCG_OVERRIDE__PERFMON_CLOCK_STATE_MASK))
5109 		*flags |= AMD_CG_SUPPORT_GFX_PERF_CLK;
5110 
5111 	/* AMD_CG_SUPPORT_GFX_CGCG */
5112 	data = RREG32_SOC15(GC, 0, regRLC_CGCG_CGLS_CTRL);
5113 	if (data & RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK)
5114 		*flags |= AMD_CG_SUPPORT_GFX_CGCG;
5115 
5116 	/* AMD_CG_SUPPORT_GFX_CGLS */
5117 	if (data & RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK)
5118 		*flags |= AMD_CG_SUPPORT_GFX_CGLS;
5119 
5120 	/* AMD_CG_SUPPORT_GFX_3D_CGCG */
5121 	data = RREG32_SOC15(GC, 0, regRLC_CGCG_CGLS_CTRL_3D);
5122 	if (data & RLC_CGCG_CGLS_CTRL_3D__CGCG_EN_MASK)
5123 		*flags |= AMD_CG_SUPPORT_GFX_3D_CGCG;
5124 
5125 	/* AMD_CG_SUPPORT_GFX_3D_CGLS */
5126 	if (data & RLC_CGCG_CGLS_CTRL_3D__CGLS_EN_MASK)
5127 		*flags |= AMD_CG_SUPPORT_GFX_3D_CGLS;
5128 }
5129 
5130 static u64 gfx_v11_0_ring_get_rptr_gfx(struct amdgpu_ring *ring)
5131 {
5132 	/* gfx11 is 32bit rptr*/
5133 	return *(uint32_t *)ring->rptr_cpu_addr;
5134 }
5135 
5136 static u64 gfx_v11_0_ring_get_wptr_gfx(struct amdgpu_ring *ring)
5137 {
5138 	struct amdgpu_device *adev = ring->adev;
5139 	u64 wptr;
5140 
5141 	/* XXX check if swapping is necessary on BE */
5142 	if (ring->use_doorbell) {
5143 		wptr = atomic64_read((atomic64_t *)ring->wptr_cpu_addr);
5144 	} else {
5145 		wptr = RREG32_SOC15(GC, 0, regCP_RB0_WPTR);
5146 		wptr += (u64)RREG32_SOC15(GC, 0, regCP_RB0_WPTR_HI) << 32;
5147 	}
5148 
5149 	return wptr;
5150 }
5151 
5152 static void gfx_v11_0_ring_set_wptr_gfx(struct amdgpu_ring *ring)
5153 {
5154 	struct amdgpu_device *adev = ring->adev;
5155 	uint32_t *wptr_saved;
5156 	uint32_t *is_queue_unmap;
5157 	uint64_t aggregated_db_index;
5158 	uint32_t mqd_size = adev->mqds[AMDGPU_HW_IP_GFX].mqd_size;
5159 	uint64_t wptr_tmp;
5160 
5161 	if (ring->is_mes_queue) {
5162 		wptr_saved = (uint32_t *)(ring->mqd_ptr + mqd_size);
5163 		is_queue_unmap = (uint32_t *)(ring->mqd_ptr + mqd_size +
5164 					      sizeof(uint32_t));
5165 		aggregated_db_index =
5166 			amdgpu_mes_get_aggregated_doorbell_index(adev,
5167 								 ring->hw_prio);
5168 
5169 		wptr_tmp = ring->wptr & ring->buf_mask;
5170 		atomic64_set((atomic64_t *)ring->wptr_cpu_addr, wptr_tmp);
5171 		*wptr_saved = wptr_tmp;
5172 		/* assume doorbell always being used by mes mapped queue */
5173 		if (*is_queue_unmap) {
5174 			WDOORBELL64(aggregated_db_index, wptr_tmp);
5175 			WDOORBELL64(ring->doorbell_index, wptr_tmp);
5176 		} else {
5177 			WDOORBELL64(ring->doorbell_index, wptr_tmp);
5178 
5179 			if (*is_queue_unmap)
5180 				WDOORBELL64(aggregated_db_index, wptr_tmp);
5181 		}
5182 	} else {
5183 		if (ring->use_doorbell) {
5184 			/* XXX check if swapping is necessary on BE */
5185 			atomic64_set((atomic64_t *)ring->wptr_cpu_addr,
5186 				     ring->wptr);
5187 			WDOORBELL64(ring->doorbell_index, ring->wptr);
5188 		} else {
5189 			WREG32_SOC15(GC, 0, regCP_RB0_WPTR,
5190 				     lower_32_bits(ring->wptr));
5191 			WREG32_SOC15(GC, 0, regCP_RB0_WPTR_HI,
5192 				     upper_32_bits(ring->wptr));
5193 		}
5194 	}
5195 }
5196 
5197 static u64 gfx_v11_0_ring_get_rptr_compute(struct amdgpu_ring *ring)
5198 {
5199 	/* gfx11 hardware is 32bit rptr */
5200 	return *(uint32_t *)ring->rptr_cpu_addr;
5201 }
5202 
5203 static u64 gfx_v11_0_ring_get_wptr_compute(struct amdgpu_ring *ring)
5204 {
5205 	u64 wptr;
5206 
5207 	/* XXX check if swapping is necessary on BE */
5208 	if (ring->use_doorbell)
5209 		wptr = atomic64_read((atomic64_t *)ring->wptr_cpu_addr);
5210 	else
5211 		BUG();
5212 	return wptr;
5213 }
5214 
5215 static void gfx_v11_0_ring_set_wptr_compute(struct amdgpu_ring *ring)
5216 {
5217 	struct amdgpu_device *adev = ring->adev;
5218 	uint32_t *wptr_saved;
5219 	uint32_t *is_queue_unmap;
5220 	uint64_t aggregated_db_index;
5221 	uint32_t mqd_size = adev->mqds[AMDGPU_HW_IP_COMPUTE].mqd_size;
5222 	uint64_t wptr_tmp;
5223 
5224 	if (ring->is_mes_queue) {
5225 		wptr_saved = (uint32_t *)(ring->mqd_ptr + mqd_size);
5226 		is_queue_unmap = (uint32_t *)(ring->mqd_ptr + mqd_size +
5227 					      sizeof(uint32_t));
5228 		aggregated_db_index =
5229 			amdgpu_mes_get_aggregated_doorbell_index(adev,
5230 								 ring->hw_prio);
5231 
5232 		wptr_tmp = ring->wptr & ring->buf_mask;
5233 		atomic64_set((atomic64_t *)ring->wptr_cpu_addr, wptr_tmp);
5234 		*wptr_saved = wptr_tmp;
5235 		/* assume doorbell always used by mes mapped queue */
5236 		if (*is_queue_unmap) {
5237 			WDOORBELL64(aggregated_db_index, wptr_tmp);
5238 			WDOORBELL64(ring->doorbell_index, wptr_tmp);
5239 		} else {
5240 			WDOORBELL64(ring->doorbell_index, wptr_tmp);
5241 
5242 			if (*is_queue_unmap)
5243 				WDOORBELL64(aggregated_db_index, wptr_tmp);
5244 		}
5245 	} else {
5246 		/* XXX check if swapping is necessary on BE */
5247 		if (ring->use_doorbell) {
5248 			atomic64_set((atomic64_t *)ring->wptr_cpu_addr,
5249 				     ring->wptr);
5250 			WDOORBELL64(ring->doorbell_index, ring->wptr);
5251 		} else {
5252 			BUG(); /* only DOORBELL method supported on gfx11 now */
5253 		}
5254 	}
5255 }
5256 
5257 static void gfx_v11_0_ring_emit_hdp_flush(struct amdgpu_ring *ring)
5258 {
5259 	struct amdgpu_device *adev = ring->adev;
5260 	u32 ref_and_mask, reg_mem_engine;
5261 	const struct nbio_hdp_flush_reg *nbio_hf_reg = adev->nbio.hdp_flush_reg;
5262 
5263 	if (ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE) {
5264 		switch (ring->me) {
5265 		case 1:
5266 			ref_and_mask = nbio_hf_reg->ref_and_mask_cp2 << ring->pipe;
5267 			break;
5268 		case 2:
5269 			ref_and_mask = nbio_hf_reg->ref_and_mask_cp6 << ring->pipe;
5270 			break;
5271 		default:
5272 			return;
5273 		}
5274 		reg_mem_engine = 0;
5275 	} else {
5276 		ref_and_mask = nbio_hf_reg->ref_and_mask_cp0;
5277 		reg_mem_engine = 1; /* pfp */
5278 	}
5279 
5280 	gfx_v11_0_wait_reg_mem(ring, reg_mem_engine, 0, 1,
5281 			       adev->nbio.funcs->get_hdp_flush_req_offset(adev),
5282 			       adev->nbio.funcs->get_hdp_flush_done_offset(adev),
5283 			       ref_and_mask, ref_and_mask, 0x20);
5284 }
5285 
5286 static void gfx_v11_0_ring_emit_ib_gfx(struct amdgpu_ring *ring,
5287 				       struct amdgpu_job *job,
5288 				       struct amdgpu_ib *ib,
5289 				       uint32_t flags)
5290 {
5291 	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
5292 	u32 header, control = 0;
5293 
5294 	BUG_ON(ib->flags & AMDGPU_IB_FLAG_CE);
5295 
5296 	header = PACKET3(PACKET3_INDIRECT_BUFFER, 2);
5297 
5298 	control |= ib->length_dw | (vmid << 24);
5299 
5300 	if (ring->adev->gfx.mcbp && (ib->flags & AMDGPU_IB_FLAG_PREEMPT)) {
5301 		control |= INDIRECT_BUFFER_PRE_ENB(1);
5302 
5303 		if (flags & AMDGPU_IB_PREEMPTED)
5304 			control |= INDIRECT_BUFFER_PRE_RESUME(1);
5305 
5306 		if (vmid)
5307 			gfx_v11_0_ring_emit_de_meta(ring,
5308 				    (!amdgpu_sriov_vf(ring->adev) && flags & AMDGPU_IB_PREEMPTED) ? true : false);
5309 	}
5310 
5311 	if (ring->is_mes_queue)
5312 		/* inherit vmid from mqd */
5313 		control |= 0x400000;
5314 
5315 	amdgpu_ring_write(ring, header);
5316 	BUG_ON(ib->gpu_addr & 0x3); /* Dword align */
5317 	amdgpu_ring_write(ring,
5318 #ifdef __BIG_ENDIAN
5319 		(2 << 0) |
5320 #endif
5321 		lower_32_bits(ib->gpu_addr));
5322 	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr));
5323 	amdgpu_ring_write(ring, control);
5324 }
5325 
5326 static void gfx_v11_0_ring_emit_ib_compute(struct amdgpu_ring *ring,
5327 					   struct amdgpu_job *job,
5328 					   struct amdgpu_ib *ib,
5329 					   uint32_t flags)
5330 {
5331 	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
5332 	u32 control = INDIRECT_BUFFER_VALID | ib->length_dw | (vmid << 24);
5333 
5334 	if (ring->is_mes_queue)
5335 		/* inherit vmid from mqd */
5336 		control |= 0x40000000;
5337 
5338 	/* Currently, there is a high possibility to get wave ID mismatch
5339 	 * between ME and GDS, leading to a hw deadlock, because ME generates
5340 	 * different wave IDs than the GDS expects. This situation happens
5341 	 * randomly when at least 5 compute pipes use GDS ordered append.
5342 	 * The wave IDs generated by ME are also wrong after suspend/resume.
5343 	 * Those are probably bugs somewhere else in the kernel driver.
5344 	 *
5345 	 * Writing GDS_COMPUTE_MAX_WAVE_ID resets wave ID counters in ME and
5346 	 * GDS to 0 for this ring (me/pipe).
5347 	 */
5348 	if (ib->flags & AMDGPU_IB_FLAG_RESET_GDS_MAX_WAVE_ID) {
5349 		amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1));
5350 		amdgpu_ring_write(ring, regGDS_COMPUTE_MAX_WAVE_ID);
5351 		amdgpu_ring_write(ring, ring->adev->gds.gds_compute_max_wave_id);
5352 	}
5353 
5354 	amdgpu_ring_write(ring, PACKET3(PACKET3_INDIRECT_BUFFER, 2));
5355 	BUG_ON(ib->gpu_addr & 0x3); /* Dword align */
5356 	amdgpu_ring_write(ring,
5357 #ifdef __BIG_ENDIAN
5358 				(2 << 0) |
5359 #endif
5360 				lower_32_bits(ib->gpu_addr));
5361 	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr));
5362 	amdgpu_ring_write(ring, control);
5363 }
5364 
5365 static void gfx_v11_0_ring_emit_fence(struct amdgpu_ring *ring, u64 addr,
5366 				     u64 seq, unsigned flags)
5367 {
5368 	bool write64bit = flags & AMDGPU_FENCE_FLAG_64BIT;
5369 	bool int_sel = flags & AMDGPU_FENCE_FLAG_INT;
5370 
5371 	/* RELEASE_MEM - flush caches, send int */
5372 	amdgpu_ring_write(ring, PACKET3(PACKET3_RELEASE_MEM, 6));
5373 	amdgpu_ring_write(ring, (PACKET3_RELEASE_MEM_GCR_SEQ |
5374 				 PACKET3_RELEASE_MEM_GCR_GL2_WB |
5375 				 PACKET3_RELEASE_MEM_GCR_GL2_INV |
5376 				 PACKET3_RELEASE_MEM_GCR_GL2_US |
5377 				 PACKET3_RELEASE_MEM_GCR_GL1_INV |
5378 				 PACKET3_RELEASE_MEM_GCR_GLV_INV |
5379 				 PACKET3_RELEASE_MEM_GCR_GLM_INV |
5380 				 PACKET3_RELEASE_MEM_GCR_GLM_WB |
5381 				 PACKET3_RELEASE_MEM_CACHE_POLICY(3) |
5382 				 PACKET3_RELEASE_MEM_EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
5383 				 PACKET3_RELEASE_MEM_EVENT_INDEX(5)));
5384 	amdgpu_ring_write(ring, (PACKET3_RELEASE_MEM_DATA_SEL(write64bit ? 2 : 1) |
5385 				 PACKET3_RELEASE_MEM_INT_SEL(int_sel ? 2 : 0)));
5386 
5387 	/*
5388 	 * the address should be Qword aligned if 64bit write, Dword
5389 	 * aligned if only send 32bit data low (discard data high)
5390 	 */
5391 	if (write64bit)
5392 		BUG_ON(addr & 0x7);
5393 	else
5394 		BUG_ON(addr & 0x3);
5395 	amdgpu_ring_write(ring, lower_32_bits(addr));
5396 	amdgpu_ring_write(ring, upper_32_bits(addr));
5397 	amdgpu_ring_write(ring, lower_32_bits(seq));
5398 	amdgpu_ring_write(ring, upper_32_bits(seq));
5399 	amdgpu_ring_write(ring, ring->is_mes_queue ?
5400 			 (ring->hw_queue_id | AMDGPU_FENCE_MES_QUEUE_FLAG) : 0);
5401 }
5402 
5403 static void gfx_v11_0_ring_emit_pipeline_sync(struct amdgpu_ring *ring)
5404 {
5405 	int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX);
5406 	uint32_t seq = ring->fence_drv.sync_seq;
5407 	uint64_t addr = ring->fence_drv.gpu_addr;
5408 
5409 	gfx_v11_0_wait_reg_mem(ring, usepfp, 1, 0, lower_32_bits(addr),
5410 			       upper_32_bits(addr), seq, 0xffffffff, 4);
5411 }
5412 
5413 static void gfx_v11_0_ring_invalidate_tlbs(struct amdgpu_ring *ring,
5414 				   uint16_t pasid, uint32_t flush_type,
5415 				   bool all_hub, uint8_t dst_sel)
5416 {
5417 	amdgpu_ring_write(ring, PACKET3(PACKET3_INVALIDATE_TLBS, 0));
5418 	amdgpu_ring_write(ring,
5419 			  PACKET3_INVALIDATE_TLBS_DST_SEL(dst_sel) |
5420 			  PACKET3_INVALIDATE_TLBS_ALL_HUB(all_hub) |
5421 			  PACKET3_INVALIDATE_TLBS_PASID(pasid) |
5422 			  PACKET3_INVALIDATE_TLBS_FLUSH_TYPE(flush_type));
5423 }
5424 
5425 static void gfx_v11_0_ring_emit_vm_flush(struct amdgpu_ring *ring,
5426 					 unsigned vmid, uint64_t pd_addr)
5427 {
5428 	if (ring->is_mes_queue)
5429 		gfx_v11_0_ring_invalidate_tlbs(ring, 0, 0, false, 0);
5430 	else
5431 		amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr);
5432 
5433 	/* compute doesn't have PFP */
5434 	if (ring->funcs->type == AMDGPU_RING_TYPE_GFX) {
5435 		/* sync PFP to ME, otherwise we might get invalid PFP reads */
5436 		amdgpu_ring_write(ring, PACKET3(PACKET3_PFP_SYNC_ME, 0));
5437 		amdgpu_ring_write(ring, 0x0);
5438 	}
5439 }
5440 
5441 static void gfx_v11_0_ring_emit_fence_kiq(struct amdgpu_ring *ring, u64 addr,
5442 					  u64 seq, unsigned int flags)
5443 {
5444 	struct amdgpu_device *adev = ring->adev;
5445 
5446 	/* we only allocate 32bit for each seq wb address */
5447 	BUG_ON(flags & AMDGPU_FENCE_FLAG_64BIT);
5448 
5449 	/* write fence seq to the "addr" */
5450 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
5451 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
5452 				 WRITE_DATA_DST_SEL(5) | WR_CONFIRM));
5453 	amdgpu_ring_write(ring, lower_32_bits(addr));
5454 	amdgpu_ring_write(ring, upper_32_bits(addr));
5455 	amdgpu_ring_write(ring, lower_32_bits(seq));
5456 
5457 	if (flags & AMDGPU_FENCE_FLAG_INT) {
5458 		/* set register to trigger INT */
5459 		amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
5460 		amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
5461 					 WRITE_DATA_DST_SEL(0) | WR_CONFIRM));
5462 		amdgpu_ring_write(ring, SOC15_REG_OFFSET(GC, 0, regCPC_INT_STATUS));
5463 		amdgpu_ring_write(ring, 0);
5464 		amdgpu_ring_write(ring, 0x20000000); /* src_id is 178 */
5465 	}
5466 }
5467 
5468 static void gfx_v11_0_ring_emit_cntxcntl(struct amdgpu_ring *ring,
5469 					 uint32_t flags)
5470 {
5471 	uint32_t dw2 = 0;
5472 
5473 	dw2 |= 0x80000000; /* set load_enable otherwise this package is just NOPs */
5474 	if (flags & AMDGPU_HAVE_CTX_SWITCH) {
5475 		/* set load_global_config & load_global_uconfig */
5476 		dw2 |= 0x8001;
5477 		/* set load_cs_sh_regs */
5478 		dw2 |= 0x01000000;
5479 		/* set load_per_context_state & load_gfx_sh_regs for GFX */
5480 		dw2 |= 0x10002;
5481 	}
5482 
5483 	amdgpu_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1));
5484 	amdgpu_ring_write(ring, dw2);
5485 	amdgpu_ring_write(ring, 0);
5486 }
5487 
5488 static void gfx_v11_0_ring_emit_gfx_shadow(struct amdgpu_ring *ring,
5489 					   u64 shadow_va, u64 csa_va,
5490 					   u64 gds_va, bool init_shadow,
5491 					   int vmid)
5492 {
5493 	struct amdgpu_device *adev = ring->adev;
5494 
5495 	if (!adev->gfx.cp_gfx_shadow)
5496 		return;
5497 
5498 	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_Q_PREEMPTION_MODE, 7));
5499 	amdgpu_ring_write(ring, lower_32_bits(shadow_va));
5500 	amdgpu_ring_write(ring, upper_32_bits(shadow_va));
5501 	amdgpu_ring_write(ring, lower_32_bits(gds_va));
5502 	amdgpu_ring_write(ring, upper_32_bits(gds_va));
5503 	amdgpu_ring_write(ring, lower_32_bits(csa_va));
5504 	amdgpu_ring_write(ring, upper_32_bits(csa_va));
5505 	amdgpu_ring_write(ring, shadow_va ?
5506 			  PACKET3_SET_Q_PREEMPTION_MODE_IB_VMID(vmid) : 0);
5507 	amdgpu_ring_write(ring, init_shadow ?
5508 			  PACKET3_SET_Q_PREEMPTION_MODE_INIT_SHADOW_MEM : 0);
5509 }
5510 
5511 static unsigned gfx_v11_0_ring_emit_init_cond_exec(struct amdgpu_ring *ring)
5512 {
5513 	unsigned ret;
5514 
5515 	amdgpu_ring_write(ring, PACKET3(PACKET3_COND_EXEC, 3));
5516 	amdgpu_ring_write(ring, lower_32_bits(ring->cond_exe_gpu_addr));
5517 	amdgpu_ring_write(ring, upper_32_bits(ring->cond_exe_gpu_addr));
5518 	amdgpu_ring_write(ring, 0); /* discard following DWs if *cond_exec_gpu_addr==0 */
5519 	ret = ring->wptr & ring->buf_mask;
5520 	amdgpu_ring_write(ring, 0x55aa55aa); /* patch dummy value later */
5521 
5522 	return ret;
5523 }
5524 
5525 static void gfx_v11_0_ring_emit_patch_cond_exec(struct amdgpu_ring *ring, unsigned offset)
5526 {
5527 	unsigned cur;
5528 	BUG_ON(offset > ring->buf_mask);
5529 	BUG_ON(ring->ring[offset] != 0x55aa55aa);
5530 
5531 	cur = (ring->wptr - 1) & ring->buf_mask;
5532 	if (likely(cur > offset))
5533 		ring->ring[offset] = cur - offset;
5534 	else
5535 		ring->ring[offset] = (ring->buf_mask + 1) - offset + cur;
5536 }
5537 
5538 static int gfx_v11_0_ring_preempt_ib(struct amdgpu_ring *ring)
5539 {
5540 	int i, r = 0;
5541 	struct amdgpu_device *adev = ring->adev;
5542 	struct amdgpu_kiq *kiq = &adev->gfx.kiq[0];
5543 	struct amdgpu_ring *kiq_ring = &kiq->ring;
5544 	unsigned long flags;
5545 
5546 	if (!kiq->pmf || !kiq->pmf->kiq_unmap_queues)
5547 		return -EINVAL;
5548 
5549 	spin_lock_irqsave(&kiq->ring_lock, flags);
5550 
5551 	if (amdgpu_ring_alloc(kiq_ring, kiq->pmf->unmap_queues_size)) {
5552 		spin_unlock_irqrestore(&kiq->ring_lock, flags);
5553 		return -ENOMEM;
5554 	}
5555 
5556 	/* assert preemption condition */
5557 	amdgpu_ring_set_preempt_cond_exec(ring, false);
5558 
5559 	/* assert IB preemption, emit the trailing fence */
5560 	kiq->pmf->kiq_unmap_queues(kiq_ring, ring, PREEMPT_QUEUES_NO_UNMAP,
5561 				   ring->trail_fence_gpu_addr,
5562 				   ++ring->trail_seq);
5563 	amdgpu_ring_commit(kiq_ring);
5564 
5565 	spin_unlock_irqrestore(&kiq->ring_lock, flags);
5566 
5567 	/* poll the trailing fence */
5568 	for (i = 0; i < adev->usec_timeout; i++) {
5569 		if (ring->trail_seq ==
5570 		    le32_to_cpu(*(ring->trail_fence_cpu_addr)))
5571 			break;
5572 		udelay(1);
5573 	}
5574 
5575 	if (i >= adev->usec_timeout) {
5576 		r = -EINVAL;
5577 		DRM_ERROR("ring %d failed to preempt ib\n", ring->idx);
5578 	}
5579 
5580 	/* deassert preemption condition */
5581 	amdgpu_ring_set_preempt_cond_exec(ring, true);
5582 	return r;
5583 }
5584 
5585 static void gfx_v11_0_ring_emit_de_meta(struct amdgpu_ring *ring, bool resume)
5586 {
5587 	struct amdgpu_device *adev = ring->adev;
5588 	struct v10_de_ib_state de_payload = {0};
5589 	uint64_t offset, gds_addr, de_payload_gpu_addr;
5590 	void *de_payload_cpu_addr;
5591 	int cnt;
5592 
5593 	if (ring->is_mes_queue) {
5594 		offset = offsetof(struct amdgpu_mes_ctx_meta_data,
5595 				  gfx[0].gfx_meta_data) +
5596 			offsetof(struct v10_gfx_meta_data, de_payload);
5597 		de_payload_gpu_addr =
5598 			amdgpu_mes_ctx_get_offs_gpu_addr(ring, offset);
5599 		de_payload_cpu_addr =
5600 			amdgpu_mes_ctx_get_offs_cpu_addr(ring, offset);
5601 
5602 		offset = offsetof(struct amdgpu_mes_ctx_meta_data,
5603 				  gfx[0].gds_backup) +
5604 			offsetof(struct v10_gfx_meta_data, de_payload);
5605 		gds_addr = amdgpu_mes_ctx_get_offs_gpu_addr(ring, offset);
5606 	} else {
5607 		offset = offsetof(struct v10_gfx_meta_data, de_payload);
5608 		de_payload_gpu_addr = amdgpu_csa_vaddr(ring->adev) + offset;
5609 		de_payload_cpu_addr = adev->virt.csa_cpu_addr + offset;
5610 
5611 		gds_addr = ALIGN(amdgpu_csa_vaddr(ring->adev) +
5612 				 AMDGPU_CSA_SIZE - adev->gds.gds_size,
5613 				 PAGE_SIZE);
5614 	}
5615 
5616 	de_payload.gds_backup_addrlo = lower_32_bits(gds_addr);
5617 	de_payload.gds_backup_addrhi = upper_32_bits(gds_addr);
5618 
5619 	cnt = (sizeof(de_payload) >> 2) + 4 - 2;
5620 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, cnt));
5621 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(1) |
5622 				 WRITE_DATA_DST_SEL(8) |
5623 				 WR_CONFIRM) |
5624 				 WRITE_DATA_CACHE_POLICY(0));
5625 	amdgpu_ring_write(ring, lower_32_bits(de_payload_gpu_addr));
5626 	amdgpu_ring_write(ring, upper_32_bits(de_payload_gpu_addr));
5627 
5628 	if (resume)
5629 		amdgpu_ring_write_multiple(ring, de_payload_cpu_addr,
5630 					   sizeof(de_payload) >> 2);
5631 	else
5632 		amdgpu_ring_write_multiple(ring, (void *)&de_payload,
5633 					   sizeof(de_payload) >> 2);
5634 }
5635 
5636 static void gfx_v11_0_ring_emit_frame_cntl(struct amdgpu_ring *ring, bool start,
5637 				    bool secure)
5638 {
5639 	uint32_t v = secure ? FRAME_TMZ : 0;
5640 
5641 	amdgpu_ring_write(ring, PACKET3(PACKET3_FRAME_CONTROL, 0));
5642 	amdgpu_ring_write(ring, v | FRAME_CMD(start ? 0 : 1));
5643 }
5644 
5645 static void gfx_v11_0_ring_emit_rreg(struct amdgpu_ring *ring, uint32_t reg,
5646 				     uint32_t reg_val_offs)
5647 {
5648 	struct amdgpu_device *adev = ring->adev;
5649 
5650 	amdgpu_ring_write(ring, PACKET3(PACKET3_COPY_DATA, 4));
5651 	amdgpu_ring_write(ring, 0 |	/* src: register*/
5652 				(5 << 8) |	/* dst: memory */
5653 				(1 << 20));	/* write confirm */
5654 	amdgpu_ring_write(ring, reg);
5655 	amdgpu_ring_write(ring, 0);
5656 	amdgpu_ring_write(ring, lower_32_bits(adev->wb.gpu_addr +
5657 				reg_val_offs * 4));
5658 	amdgpu_ring_write(ring, upper_32_bits(adev->wb.gpu_addr +
5659 				reg_val_offs * 4));
5660 }
5661 
5662 static void gfx_v11_0_ring_emit_wreg(struct amdgpu_ring *ring, uint32_t reg,
5663 				   uint32_t val)
5664 {
5665 	uint32_t cmd = 0;
5666 
5667 	switch (ring->funcs->type) {
5668 	case AMDGPU_RING_TYPE_GFX:
5669 		cmd = WRITE_DATA_ENGINE_SEL(1) | WR_CONFIRM;
5670 		break;
5671 	case AMDGPU_RING_TYPE_KIQ:
5672 		cmd = (1 << 16); /* no inc addr */
5673 		break;
5674 	default:
5675 		cmd = WR_CONFIRM;
5676 		break;
5677 	}
5678 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
5679 	amdgpu_ring_write(ring, cmd);
5680 	amdgpu_ring_write(ring, reg);
5681 	amdgpu_ring_write(ring, 0);
5682 	amdgpu_ring_write(ring, val);
5683 }
5684 
5685 static void gfx_v11_0_ring_emit_reg_wait(struct amdgpu_ring *ring, uint32_t reg,
5686 					uint32_t val, uint32_t mask)
5687 {
5688 	gfx_v11_0_wait_reg_mem(ring, 0, 0, 0, reg, 0, val, mask, 0x20);
5689 }
5690 
5691 static void gfx_v11_0_ring_emit_reg_write_reg_wait(struct amdgpu_ring *ring,
5692 						   uint32_t reg0, uint32_t reg1,
5693 						   uint32_t ref, uint32_t mask)
5694 {
5695 	int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX);
5696 
5697 	gfx_v11_0_wait_reg_mem(ring, usepfp, 0, 1, reg0, reg1,
5698 			       ref, mask, 0x20);
5699 }
5700 
5701 static void gfx_v11_0_ring_soft_recovery(struct amdgpu_ring *ring,
5702 					 unsigned vmid)
5703 {
5704 	struct amdgpu_device *adev = ring->adev;
5705 	uint32_t value = 0;
5706 
5707 	value = REG_SET_FIELD(value, SQ_CMD, CMD, 0x03);
5708 	value = REG_SET_FIELD(value, SQ_CMD, MODE, 0x01);
5709 	value = REG_SET_FIELD(value, SQ_CMD, CHECK_VMID, 1);
5710 	value = REG_SET_FIELD(value, SQ_CMD, VM_ID, vmid);
5711 	WREG32_SOC15(GC, 0, regSQ_CMD, value);
5712 }
5713 
5714 static void
5715 gfx_v11_0_set_gfx_eop_interrupt_state(struct amdgpu_device *adev,
5716 				      uint32_t me, uint32_t pipe,
5717 				      enum amdgpu_interrupt_state state)
5718 {
5719 	uint32_t cp_int_cntl, cp_int_cntl_reg;
5720 
5721 	if (!me) {
5722 		switch (pipe) {
5723 		case 0:
5724 			cp_int_cntl_reg = SOC15_REG_OFFSET(GC, 0, regCP_INT_CNTL_RING0);
5725 			break;
5726 		case 1:
5727 			cp_int_cntl_reg = SOC15_REG_OFFSET(GC, 0, regCP_INT_CNTL_RING1);
5728 			break;
5729 		default:
5730 			DRM_DEBUG("invalid pipe %d\n", pipe);
5731 			return;
5732 		}
5733 	} else {
5734 		DRM_DEBUG("invalid me %d\n", me);
5735 		return;
5736 	}
5737 
5738 	switch (state) {
5739 	case AMDGPU_IRQ_STATE_DISABLE:
5740 		cp_int_cntl = RREG32_SOC15_IP(GC, cp_int_cntl_reg);
5741 		cp_int_cntl = REG_SET_FIELD(cp_int_cntl, CP_INT_CNTL_RING0,
5742 					    TIME_STAMP_INT_ENABLE, 0);
5743 		cp_int_cntl = REG_SET_FIELD(cp_int_cntl, CP_INT_CNTL_RING0,
5744 					    GENERIC0_INT_ENABLE, 0);
5745 		WREG32_SOC15_IP(GC, cp_int_cntl_reg, cp_int_cntl);
5746 		break;
5747 	case AMDGPU_IRQ_STATE_ENABLE:
5748 		cp_int_cntl = RREG32_SOC15_IP(GC, cp_int_cntl_reg);
5749 		cp_int_cntl = REG_SET_FIELD(cp_int_cntl, CP_INT_CNTL_RING0,
5750 					    TIME_STAMP_INT_ENABLE, 1);
5751 		cp_int_cntl = REG_SET_FIELD(cp_int_cntl, CP_INT_CNTL_RING0,
5752 					    GENERIC0_INT_ENABLE, 1);
5753 		WREG32_SOC15_IP(GC, cp_int_cntl_reg, cp_int_cntl);
5754 		break;
5755 	default:
5756 		break;
5757 	}
5758 }
5759 
5760 static void gfx_v11_0_set_compute_eop_interrupt_state(struct amdgpu_device *adev,
5761 						     int me, int pipe,
5762 						     enum amdgpu_interrupt_state state)
5763 {
5764 	u32 mec_int_cntl, mec_int_cntl_reg;
5765 
5766 	/*
5767 	 * amdgpu controls only the first MEC. That's why this function only
5768 	 * handles the setting of interrupts for this specific MEC. All other
5769 	 * pipes' interrupts are set by amdkfd.
5770 	 */
5771 
5772 	if (me == 1) {
5773 		switch (pipe) {
5774 		case 0:
5775 			mec_int_cntl_reg = SOC15_REG_OFFSET(GC, 0, regCP_ME1_PIPE0_INT_CNTL);
5776 			break;
5777 		case 1:
5778 			mec_int_cntl_reg = SOC15_REG_OFFSET(GC, 0, regCP_ME1_PIPE1_INT_CNTL);
5779 			break;
5780 		case 2:
5781 			mec_int_cntl_reg = SOC15_REG_OFFSET(GC, 0, regCP_ME1_PIPE2_INT_CNTL);
5782 			break;
5783 		case 3:
5784 			mec_int_cntl_reg = SOC15_REG_OFFSET(GC, 0, regCP_ME1_PIPE3_INT_CNTL);
5785 			break;
5786 		default:
5787 			DRM_DEBUG("invalid pipe %d\n", pipe);
5788 			return;
5789 		}
5790 	} else {
5791 		DRM_DEBUG("invalid me %d\n", me);
5792 		return;
5793 	}
5794 
5795 	switch (state) {
5796 	case AMDGPU_IRQ_STATE_DISABLE:
5797 		mec_int_cntl = RREG32_SOC15_IP(GC, mec_int_cntl_reg);
5798 		mec_int_cntl = REG_SET_FIELD(mec_int_cntl, CP_ME1_PIPE0_INT_CNTL,
5799 					     TIME_STAMP_INT_ENABLE, 0);
5800 		mec_int_cntl = REG_SET_FIELD(mec_int_cntl, CP_ME1_PIPE0_INT_CNTL,
5801 					     GENERIC0_INT_ENABLE, 0);
5802 		WREG32_SOC15_IP(GC, mec_int_cntl_reg, mec_int_cntl);
5803 		break;
5804 	case AMDGPU_IRQ_STATE_ENABLE:
5805 		mec_int_cntl = RREG32_SOC15_IP(GC, mec_int_cntl_reg);
5806 		mec_int_cntl = REG_SET_FIELD(mec_int_cntl, CP_ME1_PIPE0_INT_CNTL,
5807 					     TIME_STAMP_INT_ENABLE, 1);
5808 		mec_int_cntl = REG_SET_FIELD(mec_int_cntl, CP_ME1_PIPE0_INT_CNTL,
5809 					     GENERIC0_INT_ENABLE, 1);
5810 		WREG32_SOC15_IP(GC, mec_int_cntl_reg, mec_int_cntl);
5811 		break;
5812 	default:
5813 		break;
5814 	}
5815 }
5816 
5817 static int gfx_v11_0_set_eop_interrupt_state(struct amdgpu_device *adev,
5818 					    struct amdgpu_irq_src *src,
5819 					    unsigned type,
5820 					    enum amdgpu_interrupt_state state)
5821 {
5822 	switch (type) {
5823 	case AMDGPU_CP_IRQ_GFX_ME0_PIPE0_EOP:
5824 		gfx_v11_0_set_gfx_eop_interrupt_state(adev, 0, 0, state);
5825 		break;
5826 	case AMDGPU_CP_IRQ_GFX_ME0_PIPE1_EOP:
5827 		gfx_v11_0_set_gfx_eop_interrupt_state(adev, 0, 1, state);
5828 		break;
5829 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP:
5830 		gfx_v11_0_set_compute_eop_interrupt_state(adev, 1, 0, state);
5831 		break;
5832 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE1_EOP:
5833 		gfx_v11_0_set_compute_eop_interrupt_state(adev, 1, 1, state);
5834 		break;
5835 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE2_EOP:
5836 		gfx_v11_0_set_compute_eop_interrupt_state(adev, 1, 2, state);
5837 		break;
5838 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE3_EOP:
5839 		gfx_v11_0_set_compute_eop_interrupt_state(adev, 1, 3, state);
5840 		break;
5841 	default:
5842 		break;
5843 	}
5844 	return 0;
5845 }
5846 
5847 static int gfx_v11_0_eop_irq(struct amdgpu_device *adev,
5848 			     struct amdgpu_irq_src *source,
5849 			     struct amdgpu_iv_entry *entry)
5850 {
5851 	int i;
5852 	u8 me_id, pipe_id, queue_id;
5853 	struct amdgpu_ring *ring;
5854 	uint32_t mes_queue_id = entry->src_data[0];
5855 
5856 	DRM_DEBUG("IH: CP EOP\n");
5857 
5858 	if (adev->enable_mes && (mes_queue_id & AMDGPU_FENCE_MES_QUEUE_FLAG)) {
5859 		struct amdgpu_mes_queue *queue;
5860 
5861 		mes_queue_id &= AMDGPU_FENCE_MES_QUEUE_ID_MASK;
5862 
5863 		spin_lock(&adev->mes.queue_id_lock);
5864 		queue = idr_find(&adev->mes.queue_id_idr, mes_queue_id);
5865 		if (queue) {
5866 			DRM_DEBUG("process mes queue id = %d\n", mes_queue_id);
5867 			amdgpu_fence_process(queue->ring);
5868 		}
5869 		spin_unlock(&adev->mes.queue_id_lock);
5870 	} else {
5871 		me_id = (entry->ring_id & 0x0c) >> 2;
5872 		pipe_id = (entry->ring_id & 0x03) >> 0;
5873 		queue_id = (entry->ring_id & 0x70) >> 4;
5874 
5875 		switch (me_id) {
5876 		case 0:
5877 			if (pipe_id == 0)
5878 				amdgpu_fence_process(&adev->gfx.gfx_ring[0]);
5879 			else
5880 				amdgpu_fence_process(&adev->gfx.gfx_ring[1]);
5881 			break;
5882 		case 1:
5883 		case 2:
5884 			for (i = 0; i < adev->gfx.num_compute_rings; i++) {
5885 				ring = &adev->gfx.compute_ring[i];
5886 				/* Per-queue interrupt is supported for MEC starting from VI.
5887 				 * The interrupt can only be enabled/disabled per pipe instead
5888 				 * of per queue.
5889 				 */
5890 				if ((ring->me == me_id) &&
5891 				    (ring->pipe == pipe_id) &&
5892 				    (ring->queue == queue_id))
5893 					amdgpu_fence_process(ring);
5894 			}
5895 			break;
5896 		}
5897 	}
5898 
5899 	return 0;
5900 }
5901 
5902 static int gfx_v11_0_set_priv_reg_fault_state(struct amdgpu_device *adev,
5903 					      struct amdgpu_irq_src *source,
5904 					      unsigned type,
5905 					      enum amdgpu_interrupt_state state)
5906 {
5907 	switch (state) {
5908 	case AMDGPU_IRQ_STATE_DISABLE:
5909 	case AMDGPU_IRQ_STATE_ENABLE:
5910 		WREG32_FIELD15_PREREG(GC, 0, CP_INT_CNTL_RING0,
5911 			       PRIV_REG_INT_ENABLE,
5912 			       state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0);
5913 		break;
5914 	default:
5915 		break;
5916 	}
5917 
5918 	return 0;
5919 }
5920 
5921 static int gfx_v11_0_set_priv_inst_fault_state(struct amdgpu_device *adev,
5922 					       struct amdgpu_irq_src *source,
5923 					       unsigned type,
5924 					       enum amdgpu_interrupt_state state)
5925 {
5926 	switch (state) {
5927 	case AMDGPU_IRQ_STATE_DISABLE:
5928 	case AMDGPU_IRQ_STATE_ENABLE:
5929 		WREG32_FIELD15_PREREG(GC, 0, CP_INT_CNTL_RING0,
5930 			       PRIV_INSTR_INT_ENABLE,
5931 			       state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0);
5932 		break;
5933 	default:
5934 		break;
5935 	}
5936 
5937 	return 0;
5938 }
5939 
5940 static void gfx_v11_0_handle_priv_fault(struct amdgpu_device *adev,
5941 					struct amdgpu_iv_entry *entry)
5942 {
5943 	u8 me_id, pipe_id, queue_id;
5944 	struct amdgpu_ring *ring;
5945 	int i;
5946 
5947 	me_id = (entry->ring_id & 0x0c) >> 2;
5948 	pipe_id = (entry->ring_id & 0x03) >> 0;
5949 	queue_id = (entry->ring_id & 0x70) >> 4;
5950 
5951 	switch (me_id) {
5952 	case 0:
5953 		for (i = 0; i < adev->gfx.num_gfx_rings; i++) {
5954 			ring = &adev->gfx.gfx_ring[i];
5955 			/* we only enabled 1 gfx queue per pipe for now */
5956 			if (ring->me == me_id && ring->pipe == pipe_id)
5957 				drm_sched_fault(&ring->sched);
5958 		}
5959 		break;
5960 	case 1:
5961 	case 2:
5962 		for (i = 0; i < adev->gfx.num_compute_rings; i++) {
5963 			ring = &adev->gfx.compute_ring[i];
5964 			if (ring->me == me_id && ring->pipe == pipe_id &&
5965 			    ring->queue == queue_id)
5966 				drm_sched_fault(&ring->sched);
5967 		}
5968 		break;
5969 	default:
5970 		BUG();
5971 		break;
5972 	}
5973 }
5974 
5975 static int gfx_v11_0_priv_reg_irq(struct amdgpu_device *adev,
5976 				  struct amdgpu_irq_src *source,
5977 				  struct amdgpu_iv_entry *entry)
5978 {
5979 	DRM_ERROR("Illegal register access in command stream\n");
5980 	gfx_v11_0_handle_priv_fault(adev, entry);
5981 	return 0;
5982 }
5983 
5984 static int gfx_v11_0_priv_inst_irq(struct amdgpu_device *adev,
5985 				   struct amdgpu_irq_src *source,
5986 				   struct amdgpu_iv_entry *entry)
5987 {
5988 	DRM_ERROR("Illegal instruction in command stream\n");
5989 	gfx_v11_0_handle_priv_fault(adev, entry);
5990 	return 0;
5991 }
5992 
5993 static int gfx_v11_0_rlc_gc_fed_irq(struct amdgpu_device *adev,
5994 				  struct amdgpu_irq_src *source,
5995 				  struct amdgpu_iv_entry *entry)
5996 {
5997 	if (adev->gfx.ras && adev->gfx.ras->rlc_gc_fed_irq)
5998 		return adev->gfx.ras->rlc_gc_fed_irq(adev, source, entry);
5999 
6000 	return 0;
6001 }
6002 
6003 #if 0
6004 static int gfx_v11_0_kiq_set_interrupt_state(struct amdgpu_device *adev,
6005 					     struct amdgpu_irq_src *src,
6006 					     unsigned int type,
6007 					     enum amdgpu_interrupt_state state)
6008 {
6009 	uint32_t tmp, target;
6010 	struct amdgpu_ring *ring = &(adev->gfx.kiq[0].ring);
6011 
6012 	target = SOC15_REG_OFFSET(GC, 0, regCP_ME1_PIPE0_INT_CNTL);
6013 	target += ring->pipe;
6014 
6015 	switch (type) {
6016 	case AMDGPU_CP_KIQ_IRQ_DRIVER0:
6017 		if (state == AMDGPU_IRQ_STATE_DISABLE) {
6018 			tmp = RREG32_SOC15(GC, 0, regCPC_INT_CNTL);
6019 			tmp = REG_SET_FIELD(tmp, CPC_INT_CNTL,
6020 					    GENERIC2_INT_ENABLE, 0);
6021 			WREG32_SOC15(GC, 0, regCPC_INT_CNTL, tmp);
6022 
6023 			tmp = RREG32_SOC15_IP(GC, target);
6024 			tmp = REG_SET_FIELD(tmp, CP_ME1_PIPE0_INT_CNTL,
6025 					    GENERIC2_INT_ENABLE, 0);
6026 			WREG32_SOC15_IP(GC, target, tmp);
6027 		} else {
6028 			tmp = RREG32_SOC15(GC, 0, regCPC_INT_CNTL);
6029 			tmp = REG_SET_FIELD(tmp, CPC_INT_CNTL,
6030 					    GENERIC2_INT_ENABLE, 1);
6031 			WREG32_SOC15(GC, 0, regCPC_INT_CNTL, tmp);
6032 
6033 			tmp = RREG32_SOC15_IP(GC, target);
6034 			tmp = REG_SET_FIELD(tmp, CP_ME1_PIPE0_INT_CNTL,
6035 					    GENERIC2_INT_ENABLE, 1);
6036 			WREG32_SOC15_IP(GC, target, tmp);
6037 		}
6038 		break;
6039 	default:
6040 		BUG(); /* kiq only support GENERIC2_INT now */
6041 		break;
6042 	}
6043 	return 0;
6044 }
6045 #endif
6046 
6047 static void gfx_v11_0_emit_mem_sync(struct amdgpu_ring *ring)
6048 {
6049 	const unsigned int gcr_cntl =
6050 			PACKET3_ACQUIRE_MEM_GCR_CNTL_GL2_INV(1) |
6051 			PACKET3_ACQUIRE_MEM_GCR_CNTL_GL2_WB(1) |
6052 			PACKET3_ACQUIRE_MEM_GCR_CNTL_GLM_INV(1) |
6053 			PACKET3_ACQUIRE_MEM_GCR_CNTL_GLM_WB(1) |
6054 			PACKET3_ACQUIRE_MEM_GCR_CNTL_GL1_INV(1) |
6055 			PACKET3_ACQUIRE_MEM_GCR_CNTL_GLV_INV(1) |
6056 			PACKET3_ACQUIRE_MEM_GCR_CNTL_GLK_INV(1) |
6057 			PACKET3_ACQUIRE_MEM_GCR_CNTL_GLI_INV(1);
6058 
6059 	/* ACQUIRE_MEM - make one or more surfaces valid for use by the subsequent operations */
6060 	amdgpu_ring_write(ring, PACKET3(PACKET3_ACQUIRE_MEM, 6));
6061 	amdgpu_ring_write(ring, 0); /* CP_COHER_CNTL */
6062 	amdgpu_ring_write(ring, 0xffffffff);  /* CP_COHER_SIZE */
6063 	amdgpu_ring_write(ring, 0xffffff);  /* CP_COHER_SIZE_HI */
6064 	amdgpu_ring_write(ring, 0); /* CP_COHER_BASE */
6065 	amdgpu_ring_write(ring, 0);  /* CP_COHER_BASE_HI */
6066 	amdgpu_ring_write(ring, 0x0000000A); /* POLL_INTERVAL */
6067 	amdgpu_ring_write(ring, gcr_cntl); /* GCR_CNTL */
6068 }
6069 
6070 static const struct amd_ip_funcs gfx_v11_0_ip_funcs = {
6071 	.name = "gfx_v11_0",
6072 	.early_init = gfx_v11_0_early_init,
6073 	.late_init = gfx_v11_0_late_init,
6074 	.sw_init = gfx_v11_0_sw_init,
6075 	.sw_fini = gfx_v11_0_sw_fini,
6076 	.hw_init = gfx_v11_0_hw_init,
6077 	.hw_fini = gfx_v11_0_hw_fini,
6078 	.suspend = gfx_v11_0_suspend,
6079 	.resume = gfx_v11_0_resume,
6080 	.is_idle = gfx_v11_0_is_idle,
6081 	.wait_for_idle = gfx_v11_0_wait_for_idle,
6082 	.soft_reset = gfx_v11_0_soft_reset,
6083 	.check_soft_reset = gfx_v11_0_check_soft_reset,
6084 	.post_soft_reset = gfx_v11_0_post_soft_reset,
6085 	.set_clockgating_state = gfx_v11_0_set_clockgating_state,
6086 	.set_powergating_state = gfx_v11_0_set_powergating_state,
6087 	.get_clockgating_state = gfx_v11_0_get_clockgating_state,
6088 };
6089 
6090 static const struct amdgpu_ring_funcs gfx_v11_0_ring_funcs_gfx = {
6091 	.type = AMDGPU_RING_TYPE_GFX,
6092 	.align_mask = 0xff,
6093 	.nop = PACKET3(PACKET3_NOP, 0x3FFF),
6094 	.support_64bit_ptrs = true,
6095 	.secure_submission_supported = true,
6096 	.get_rptr = gfx_v11_0_ring_get_rptr_gfx,
6097 	.get_wptr = gfx_v11_0_ring_get_wptr_gfx,
6098 	.set_wptr = gfx_v11_0_ring_set_wptr_gfx,
6099 	.emit_frame_size = /* totally 242 maximum if 16 IBs */
6100 		5 + /* COND_EXEC */
6101 		9 + /* SET_Q_PREEMPTION_MODE */
6102 		7 + /* PIPELINE_SYNC */
6103 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 5 +
6104 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 7 +
6105 		2 + /* VM_FLUSH */
6106 		8 + /* FENCE for VM_FLUSH */
6107 		20 + /* GDS switch */
6108 		5 + /* COND_EXEC */
6109 		7 + /* HDP_flush */
6110 		4 + /* VGT_flush */
6111 		31 + /*	DE_META */
6112 		3 + /* CNTX_CTRL */
6113 		5 + /* HDP_INVL */
6114 		8 + 8 + /* FENCE x2 */
6115 		8, /* gfx_v11_0_emit_mem_sync */
6116 	.emit_ib_size =	4, /* gfx_v11_0_ring_emit_ib_gfx */
6117 	.emit_ib = gfx_v11_0_ring_emit_ib_gfx,
6118 	.emit_fence = gfx_v11_0_ring_emit_fence,
6119 	.emit_pipeline_sync = gfx_v11_0_ring_emit_pipeline_sync,
6120 	.emit_vm_flush = gfx_v11_0_ring_emit_vm_flush,
6121 	.emit_gds_switch = gfx_v11_0_ring_emit_gds_switch,
6122 	.emit_hdp_flush = gfx_v11_0_ring_emit_hdp_flush,
6123 	.test_ring = gfx_v11_0_ring_test_ring,
6124 	.test_ib = gfx_v11_0_ring_test_ib,
6125 	.insert_nop = amdgpu_ring_insert_nop,
6126 	.pad_ib = amdgpu_ring_generic_pad_ib,
6127 	.emit_cntxcntl = gfx_v11_0_ring_emit_cntxcntl,
6128 	.emit_gfx_shadow = gfx_v11_0_ring_emit_gfx_shadow,
6129 	.init_cond_exec = gfx_v11_0_ring_emit_init_cond_exec,
6130 	.patch_cond_exec = gfx_v11_0_ring_emit_patch_cond_exec,
6131 	.preempt_ib = gfx_v11_0_ring_preempt_ib,
6132 	.emit_frame_cntl = gfx_v11_0_ring_emit_frame_cntl,
6133 	.emit_wreg = gfx_v11_0_ring_emit_wreg,
6134 	.emit_reg_wait = gfx_v11_0_ring_emit_reg_wait,
6135 	.emit_reg_write_reg_wait = gfx_v11_0_ring_emit_reg_write_reg_wait,
6136 	.soft_recovery = gfx_v11_0_ring_soft_recovery,
6137 	.emit_mem_sync = gfx_v11_0_emit_mem_sync,
6138 };
6139 
6140 static const struct amdgpu_ring_funcs gfx_v11_0_ring_funcs_compute = {
6141 	.type = AMDGPU_RING_TYPE_COMPUTE,
6142 	.align_mask = 0xff,
6143 	.nop = PACKET3(PACKET3_NOP, 0x3FFF),
6144 	.support_64bit_ptrs = true,
6145 	.get_rptr = gfx_v11_0_ring_get_rptr_compute,
6146 	.get_wptr = gfx_v11_0_ring_get_wptr_compute,
6147 	.set_wptr = gfx_v11_0_ring_set_wptr_compute,
6148 	.emit_frame_size =
6149 		20 + /* gfx_v11_0_ring_emit_gds_switch */
6150 		7 + /* gfx_v11_0_ring_emit_hdp_flush */
6151 		5 + /* hdp invalidate */
6152 		7 + /* gfx_v11_0_ring_emit_pipeline_sync */
6153 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 5 +
6154 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 7 +
6155 		2 + /* gfx_v11_0_ring_emit_vm_flush */
6156 		8 + 8 + 8 + /* gfx_v11_0_ring_emit_fence x3 for user fence, vm fence */
6157 		8, /* gfx_v11_0_emit_mem_sync */
6158 	.emit_ib_size =	7, /* gfx_v11_0_ring_emit_ib_compute */
6159 	.emit_ib = gfx_v11_0_ring_emit_ib_compute,
6160 	.emit_fence = gfx_v11_0_ring_emit_fence,
6161 	.emit_pipeline_sync = gfx_v11_0_ring_emit_pipeline_sync,
6162 	.emit_vm_flush = gfx_v11_0_ring_emit_vm_flush,
6163 	.emit_gds_switch = gfx_v11_0_ring_emit_gds_switch,
6164 	.emit_hdp_flush = gfx_v11_0_ring_emit_hdp_flush,
6165 	.test_ring = gfx_v11_0_ring_test_ring,
6166 	.test_ib = gfx_v11_0_ring_test_ib,
6167 	.insert_nop = amdgpu_ring_insert_nop,
6168 	.pad_ib = amdgpu_ring_generic_pad_ib,
6169 	.emit_wreg = gfx_v11_0_ring_emit_wreg,
6170 	.emit_reg_wait = gfx_v11_0_ring_emit_reg_wait,
6171 	.emit_reg_write_reg_wait = gfx_v11_0_ring_emit_reg_write_reg_wait,
6172 	.emit_mem_sync = gfx_v11_0_emit_mem_sync,
6173 };
6174 
6175 static const struct amdgpu_ring_funcs gfx_v11_0_ring_funcs_kiq = {
6176 	.type = AMDGPU_RING_TYPE_KIQ,
6177 	.align_mask = 0xff,
6178 	.nop = PACKET3(PACKET3_NOP, 0x3FFF),
6179 	.support_64bit_ptrs = true,
6180 	.get_rptr = gfx_v11_0_ring_get_rptr_compute,
6181 	.get_wptr = gfx_v11_0_ring_get_wptr_compute,
6182 	.set_wptr = gfx_v11_0_ring_set_wptr_compute,
6183 	.emit_frame_size =
6184 		20 + /* gfx_v11_0_ring_emit_gds_switch */
6185 		7 + /* gfx_v11_0_ring_emit_hdp_flush */
6186 		5 + /*hdp invalidate */
6187 		7 + /* gfx_v11_0_ring_emit_pipeline_sync */
6188 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 5 +
6189 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 7 +
6190 		2 + /* gfx_v11_0_ring_emit_vm_flush */
6191 		8 + 8 + 8, /* gfx_v11_0_ring_emit_fence_kiq x3 for user fence, vm fence */
6192 	.emit_ib_size =	7, /* gfx_v11_0_ring_emit_ib_compute */
6193 	.emit_ib = gfx_v11_0_ring_emit_ib_compute,
6194 	.emit_fence = gfx_v11_0_ring_emit_fence_kiq,
6195 	.test_ring = gfx_v11_0_ring_test_ring,
6196 	.test_ib = gfx_v11_0_ring_test_ib,
6197 	.insert_nop = amdgpu_ring_insert_nop,
6198 	.pad_ib = amdgpu_ring_generic_pad_ib,
6199 	.emit_rreg = gfx_v11_0_ring_emit_rreg,
6200 	.emit_wreg = gfx_v11_0_ring_emit_wreg,
6201 	.emit_reg_wait = gfx_v11_0_ring_emit_reg_wait,
6202 	.emit_reg_write_reg_wait = gfx_v11_0_ring_emit_reg_write_reg_wait,
6203 };
6204 
6205 static void gfx_v11_0_set_ring_funcs(struct amdgpu_device *adev)
6206 {
6207 	int i;
6208 
6209 	adev->gfx.kiq[0].ring.funcs = &gfx_v11_0_ring_funcs_kiq;
6210 
6211 	for (i = 0; i < adev->gfx.num_gfx_rings; i++)
6212 		adev->gfx.gfx_ring[i].funcs = &gfx_v11_0_ring_funcs_gfx;
6213 
6214 	for (i = 0; i < adev->gfx.num_compute_rings; i++)
6215 		adev->gfx.compute_ring[i].funcs = &gfx_v11_0_ring_funcs_compute;
6216 }
6217 
6218 static const struct amdgpu_irq_src_funcs gfx_v11_0_eop_irq_funcs = {
6219 	.set = gfx_v11_0_set_eop_interrupt_state,
6220 	.process = gfx_v11_0_eop_irq,
6221 };
6222 
6223 static const struct amdgpu_irq_src_funcs gfx_v11_0_priv_reg_irq_funcs = {
6224 	.set = gfx_v11_0_set_priv_reg_fault_state,
6225 	.process = gfx_v11_0_priv_reg_irq,
6226 };
6227 
6228 static const struct amdgpu_irq_src_funcs gfx_v11_0_priv_inst_irq_funcs = {
6229 	.set = gfx_v11_0_set_priv_inst_fault_state,
6230 	.process = gfx_v11_0_priv_inst_irq,
6231 };
6232 
6233 static const struct amdgpu_irq_src_funcs gfx_v11_0_rlc_gc_fed_irq_funcs = {
6234 	.process = gfx_v11_0_rlc_gc_fed_irq,
6235 };
6236 
6237 static void gfx_v11_0_set_irq_funcs(struct amdgpu_device *adev)
6238 {
6239 	adev->gfx.eop_irq.num_types = AMDGPU_CP_IRQ_LAST;
6240 	adev->gfx.eop_irq.funcs = &gfx_v11_0_eop_irq_funcs;
6241 
6242 	adev->gfx.priv_reg_irq.num_types = 1;
6243 	adev->gfx.priv_reg_irq.funcs = &gfx_v11_0_priv_reg_irq_funcs;
6244 
6245 	adev->gfx.priv_inst_irq.num_types = 1;
6246 	adev->gfx.priv_inst_irq.funcs = &gfx_v11_0_priv_inst_irq_funcs;
6247 
6248 	adev->gfx.rlc_gc_fed_irq.num_types = 1; /* 0x80 FED error */
6249 	adev->gfx.rlc_gc_fed_irq.funcs = &gfx_v11_0_rlc_gc_fed_irq_funcs;
6250 
6251 }
6252 
6253 static void gfx_v11_0_set_imu_funcs(struct amdgpu_device *adev)
6254 {
6255 	if (adev->flags & AMD_IS_APU)
6256 		adev->gfx.imu.mode = MISSION_MODE;
6257 	else
6258 		adev->gfx.imu.mode = DEBUG_MODE;
6259 
6260 	adev->gfx.imu.funcs = &gfx_v11_0_imu_funcs;
6261 }
6262 
6263 static void gfx_v11_0_set_rlc_funcs(struct amdgpu_device *adev)
6264 {
6265 	adev->gfx.rlc.funcs = &gfx_v11_0_rlc_funcs;
6266 }
6267 
6268 static void gfx_v11_0_set_gds_init(struct amdgpu_device *adev)
6269 {
6270 	unsigned total_cu = adev->gfx.config.max_cu_per_sh *
6271 			    adev->gfx.config.max_sh_per_se *
6272 			    adev->gfx.config.max_shader_engines;
6273 
6274 	adev->gds.gds_size = 0x1000;
6275 	adev->gds.gds_compute_max_wave_id = total_cu * 32 - 1;
6276 	adev->gds.gws_size = 64;
6277 	adev->gds.oa_size = 16;
6278 }
6279 
6280 static void gfx_v11_0_set_mqd_funcs(struct amdgpu_device *adev)
6281 {
6282 	/* set gfx eng mqd */
6283 	adev->mqds[AMDGPU_HW_IP_GFX].mqd_size =
6284 		sizeof(struct v11_gfx_mqd);
6285 	adev->mqds[AMDGPU_HW_IP_GFX].init_mqd =
6286 		gfx_v11_0_gfx_mqd_init;
6287 	/* set compute eng mqd */
6288 	adev->mqds[AMDGPU_HW_IP_COMPUTE].mqd_size =
6289 		sizeof(struct v11_compute_mqd);
6290 	adev->mqds[AMDGPU_HW_IP_COMPUTE].init_mqd =
6291 		gfx_v11_0_compute_mqd_init;
6292 }
6293 
6294 static void gfx_v11_0_set_user_wgp_inactive_bitmap_per_sh(struct amdgpu_device *adev,
6295 							  u32 bitmap)
6296 {
6297 	u32 data;
6298 
6299 	if (!bitmap)
6300 		return;
6301 
6302 	data = bitmap << GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_WGPS__SHIFT;
6303 	data &= GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_WGPS_MASK;
6304 
6305 	WREG32_SOC15(GC, 0, regGC_USER_SHADER_ARRAY_CONFIG, data);
6306 }
6307 
6308 static u32 gfx_v11_0_get_wgp_active_bitmap_per_sh(struct amdgpu_device *adev)
6309 {
6310 	u32 data, wgp_bitmask;
6311 	data = RREG32_SOC15(GC, 0, regCC_GC_SHADER_ARRAY_CONFIG);
6312 	data |= RREG32_SOC15(GC, 0, regGC_USER_SHADER_ARRAY_CONFIG);
6313 
6314 	data &= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_WGPS_MASK;
6315 	data >>= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_WGPS__SHIFT;
6316 
6317 	wgp_bitmask =
6318 		amdgpu_gfx_create_bitmask(adev->gfx.config.max_cu_per_sh >> 1);
6319 
6320 	return (~data) & wgp_bitmask;
6321 }
6322 
6323 static u32 gfx_v11_0_get_cu_active_bitmap_per_sh(struct amdgpu_device *adev)
6324 {
6325 	u32 wgp_idx, wgp_active_bitmap;
6326 	u32 cu_bitmap_per_wgp, cu_active_bitmap;
6327 
6328 	wgp_active_bitmap = gfx_v11_0_get_wgp_active_bitmap_per_sh(adev);
6329 	cu_active_bitmap = 0;
6330 
6331 	for (wgp_idx = 0; wgp_idx < 16; wgp_idx++) {
6332 		/* if there is one WGP enabled, it means 2 CUs will be enabled */
6333 		cu_bitmap_per_wgp = 3 << (2 * wgp_idx);
6334 		if (wgp_active_bitmap & (1 << wgp_idx))
6335 			cu_active_bitmap |= cu_bitmap_per_wgp;
6336 	}
6337 
6338 	return cu_active_bitmap;
6339 }
6340 
6341 static int gfx_v11_0_get_cu_info(struct amdgpu_device *adev,
6342 				 struct amdgpu_cu_info *cu_info)
6343 {
6344 	int i, j, k, counter, active_cu_number = 0;
6345 	u32 mask, bitmap;
6346 	unsigned disable_masks[8 * 2];
6347 
6348 	if (!adev || !cu_info)
6349 		return -EINVAL;
6350 
6351 	amdgpu_gfx_parse_disable_cu(disable_masks, 8, 2);
6352 
6353 	mutex_lock(&adev->grbm_idx_mutex);
6354 	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
6355 		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
6356 			bitmap = i * adev->gfx.config.max_sh_per_se + j;
6357 			if (!((gfx_v11_0_get_sa_active_bitmap(adev) >> bitmap) & 1))
6358 				continue;
6359 			mask = 1;
6360 			counter = 0;
6361 			gfx_v11_0_select_se_sh(adev, i, j, 0xffffffff, 0);
6362 			if (i < 8 && j < 2)
6363 				gfx_v11_0_set_user_wgp_inactive_bitmap_per_sh(
6364 					adev, disable_masks[i * 2 + j]);
6365 			bitmap = gfx_v11_0_get_cu_active_bitmap_per_sh(adev);
6366 
6367 			/**
6368 			 * GFX11 could support more than 4 SEs, while the bitmap
6369 			 * in cu_info struct is 4x4 and ioctl interface struct
6370 			 * drm_amdgpu_info_device should keep stable.
6371 			 * So we use last two columns of bitmap to store cu mask for
6372 			 * SEs 4 to 7, the layout of the bitmap is as below:
6373 			 *    SE0: {SH0,SH1} --> {bitmap[0][0], bitmap[0][1]}
6374 			 *    SE1: {SH0,SH1} --> {bitmap[1][0], bitmap[1][1]}
6375 			 *    SE2: {SH0,SH1} --> {bitmap[2][0], bitmap[2][1]}
6376 			 *    SE3: {SH0,SH1} --> {bitmap[3][0], bitmap[3][1]}
6377 			 *    SE4: {SH0,SH1} --> {bitmap[0][2], bitmap[0][3]}
6378 			 *    SE5: {SH0,SH1} --> {bitmap[1][2], bitmap[1][3]}
6379 			 *    SE6: {SH0,SH1} --> {bitmap[2][2], bitmap[2][3]}
6380 			 *    SE7: {SH0,SH1} --> {bitmap[3][2], bitmap[3][3]}
6381 			 */
6382 			cu_info->bitmap[0][i % 4][j + (i / 4) * 2] = bitmap;
6383 
6384 			for (k = 0; k < adev->gfx.config.max_cu_per_sh; k++) {
6385 				if (bitmap & mask)
6386 					counter++;
6387 
6388 				mask <<= 1;
6389 			}
6390 			active_cu_number += counter;
6391 		}
6392 	}
6393 	gfx_v11_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 0);
6394 	mutex_unlock(&adev->grbm_idx_mutex);
6395 
6396 	cu_info->number = active_cu_number;
6397 	cu_info->simd_per_cu = NUM_SIMD_PER_CU;
6398 
6399 	return 0;
6400 }
6401 
6402 const struct amdgpu_ip_block_version gfx_v11_0_ip_block =
6403 {
6404 	.type = AMD_IP_BLOCK_TYPE_GFX,
6405 	.major = 11,
6406 	.minor = 0,
6407 	.rev = 0,
6408 	.funcs = &gfx_v11_0_ip_funcs,
6409 };
6410