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_1[] =
87 {
88 	SOC15_REG_GOLDEN_VALUE(GC, 0, regCGTT_GS_NGG_CLK_CTRL, 0x9fff8fff, 0x00000010),
89 	SOC15_REG_GOLDEN_VALUE(GC, 0, regCGTT_WD_CLK_CTRL, 0xffff8fff, 0x00000010),
90 	SOC15_REG_GOLDEN_VALUE(GC, 0, regCPF_GCR_CNTL, 0x0007ffff, 0x0000c200),
91 	SOC15_REG_GOLDEN_VALUE(GC, 0, regGL2C_CTRL3, 0xffff001b, 0x00f01988),
92 	SOC15_REG_GOLDEN_VALUE(GC, 0, regPA_CL_ENHANCE, 0xf0ffffff, 0x00880007),
93 	SOC15_REG_GOLDEN_VALUE(GC, 0, regPA_SC_ENHANCE_3, 0xfffffffd, 0x00000008),
94 	SOC15_REG_GOLDEN_VALUE(GC, 0, regPA_SC_VRS_SURFACE_CNTL_1, 0xfff891ff, 0x55480100),
95 	SOC15_REG_GOLDEN_VALUE(GC, 0, regTA_CNTL_AUX, 0xf7f7ffff, 0x01030000),
96 	SOC15_REG_GOLDEN_VALUE(GC, 0, regTCP_CNTL2, 0xfcffffff, 0x0000000a)
97 };
98 
99 #define DEFAULT_SH_MEM_CONFIG \
100 	((SH_MEM_ADDRESS_MODE_64 << SH_MEM_CONFIG__ADDRESS_MODE__SHIFT) | \
101 	 (SH_MEM_ALIGNMENT_MODE_UNALIGNED << SH_MEM_CONFIG__ALIGNMENT_MODE__SHIFT) | \
102 	 (3 << SH_MEM_CONFIG__INITIAL_INST_PREFETCH__SHIFT))
103 
104 static void gfx_v11_0_disable_gpa_mode(struct amdgpu_device *adev);
105 static void gfx_v11_0_set_ring_funcs(struct amdgpu_device *adev);
106 static void gfx_v11_0_set_irq_funcs(struct amdgpu_device *adev);
107 static void gfx_v11_0_set_gds_init(struct amdgpu_device *adev);
108 static void gfx_v11_0_set_rlc_funcs(struct amdgpu_device *adev);
109 static void gfx_v11_0_set_mqd_funcs(struct amdgpu_device *adev);
110 static void gfx_v11_0_set_imu_funcs(struct amdgpu_device *adev);
111 static int gfx_v11_0_get_cu_info(struct amdgpu_device *adev,
112                                  struct amdgpu_cu_info *cu_info);
113 static uint64_t gfx_v11_0_get_gpu_clock_counter(struct amdgpu_device *adev);
114 static void gfx_v11_0_select_se_sh(struct amdgpu_device *adev, u32 se_num,
115 				   u32 sh_num, u32 instance, int xcc_id);
116 static u32 gfx_v11_0_get_wgp_active_bitmap_per_sh(struct amdgpu_device *adev);
117 
118 static void gfx_v11_0_ring_emit_de_meta(struct amdgpu_ring *ring, bool resume);
119 static void gfx_v11_0_ring_emit_frame_cntl(struct amdgpu_ring *ring, bool start, bool secure);
120 static void gfx_v11_0_ring_emit_wreg(struct amdgpu_ring *ring, uint32_t reg,
121 				     uint32_t val);
122 static int gfx_v11_0_wait_for_rlc_autoload_complete(struct amdgpu_device *adev);
123 static void gfx_v11_0_ring_invalidate_tlbs(struct amdgpu_ring *ring,
124 					   uint16_t pasid, uint32_t flush_type,
125 					   bool all_hub, uint8_t dst_sel);
126 static void gfx_v11_0_set_safe_mode(struct amdgpu_device *adev, int xcc_id);
127 static void gfx_v11_0_unset_safe_mode(struct amdgpu_device *adev, int xcc_id);
128 static void gfx_v11_0_update_perf_clk(struct amdgpu_device *adev,
129 				      bool enable);
130 
131 static void gfx11_kiq_set_resources(struct amdgpu_ring *kiq_ring, uint64_t queue_mask)
132 {
133 	amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_SET_RESOURCES, 6));
134 	amdgpu_ring_write(kiq_ring, PACKET3_SET_RESOURCES_VMID_MASK(0) |
135 			  PACKET3_SET_RESOURCES_QUEUE_TYPE(0));	/* vmid_mask:0 queue_type:0 (KIQ) */
136 	amdgpu_ring_write(kiq_ring, lower_32_bits(queue_mask));	/* queue mask lo */
137 	amdgpu_ring_write(kiq_ring, upper_32_bits(queue_mask));	/* queue mask hi */
138 	amdgpu_ring_write(kiq_ring, 0);	/* gws mask lo */
139 	amdgpu_ring_write(kiq_ring, 0);	/* gws mask hi */
140 	amdgpu_ring_write(kiq_ring, 0);	/* oac mask */
141 	amdgpu_ring_write(kiq_ring, 0);	/* gds heap base:0, gds heap size:0 */
142 }
143 
144 static void gfx11_kiq_map_queues(struct amdgpu_ring *kiq_ring,
145 				 struct amdgpu_ring *ring)
146 {
147 	uint64_t mqd_addr = amdgpu_bo_gpu_offset(ring->mqd_obj);
148 	uint64_t wptr_addr = ring->wptr_gpu_addr;
149 	uint32_t me = 0, eng_sel = 0;
150 
151 	switch (ring->funcs->type) {
152 	case AMDGPU_RING_TYPE_COMPUTE:
153 		me = 1;
154 		eng_sel = 0;
155 		break;
156 	case AMDGPU_RING_TYPE_GFX:
157 		me = 0;
158 		eng_sel = 4;
159 		break;
160 	case AMDGPU_RING_TYPE_MES:
161 		me = 2;
162 		eng_sel = 5;
163 		break;
164 	default:
165 		WARN_ON(1);
166 	}
167 
168 	amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_MAP_QUEUES, 5));
169 	/* Q_sel:0, vmid:0, vidmem: 1, engine:0, num_Q:1*/
170 	amdgpu_ring_write(kiq_ring, /* Q_sel: 0, vmid: 0, engine: 0, num_Q: 1 */
171 			  PACKET3_MAP_QUEUES_QUEUE_SEL(0) | /* Queue_Sel */
172 			  PACKET3_MAP_QUEUES_VMID(0) | /* VMID */
173 			  PACKET3_MAP_QUEUES_QUEUE(ring->queue) |
174 			  PACKET3_MAP_QUEUES_PIPE(ring->pipe) |
175 			  PACKET3_MAP_QUEUES_ME((me)) |
176 			  PACKET3_MAP_QUEUES_QUEUE_TYPE(0) | /*queue_type: normal compute queue */
177 			  PACKET3_MAP_QUEUES_ALLOC_FORMAT(0) | /* alloc format: all_on_one_pipe */
178 			  PACKET3_MAP_QUEUES_ENGINE_SEL(eng_sel) |
179 			  PACKET3_MAP_QUEUES_NUM_QUEUES(1)); /* num_queues: must be 1 */
180 	amdgpu_ring_write(kiq_ring, PACKET3_MAP_QUEUES_DOORBELL_OFFSET(ring->doorbell_index));
181 	amdgpu_ring_write(kiq_ring, lower_32_bits(mqd_addr));
182 	amdgpu_ring_write(kiq_ring, upper_32_bits(mqd_addr));
183 	amdgpu_ring_write(kiq_ring, lower_32_bits(wptr_addr));
184 	amdgpu_ring_write(kiq_ring, upper_32_bits(wptr_addr));
185 }
186 
187 static void gfx11_kiq_unmap_queues(struct amdgpu_ring *kiq_ring,
188 				   struct amdgpu_ring *ring,
189 				   enum amdgpu_unmap_queues_action action,
190 				   u64 gpu_addr, u64 seq)
191 {
192 	struct amdgpu_device *adev = kiq_ring->adev;
193 	uint32_t eng_sel = ring->funcs->type == AMDGPU_RING_TYPE_GFX ? 4 : 0;
194 
195 	if (adev->enable_mes && !adev->gfx.kiq[0].ring.sched.ready) {
196 		amdgpu_mes_unmap_legacy_queue(adev, ring, action, gpu_addr, seq);
197 		return;
198 	}
199 
200 	amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_UNMAP_QUEUES, 4));
201 	amdgpu_ring_write(kiq_ring, /* Q_sel: 0, vmid: 0, engine: 0, num_Q: 1 */
202 			  PACKET3_UNMAP_QUEUES_ACTION(action) |
203 			  PACKET3_UNMAP_QUEUES_QUEUE_SEL(0) |
204 			  PACKET3_UNMAP_QUEUES_ENGINE_SEL(eng_sel) |
205 			  PACKET3_UNMAP_QUEUES_NUM_QUEUES(1));
206 	amdgpu_ring_write(kiq_ring,
207 		  PACKET3_UNMAP_QUEUES_DOORBELL_OFFSET0(ring->doorbell_index));
208 
209 	if (action == PREEMPT_QUEUES_NO_UNMAP) {
210 		amdgpu_ring_write(kiq_ring, lower_32_bits(gpu_addr));
211 		amdgpu_ring_write(kiq_ring, upper_32_bits(gpu_addr));
212 		amdgpu_ring_write(kiq_ring, seq);
213 	} else {
214 		amdgpu_ring_write(kiq_ring, 0);
215 		amdgpu_ring_write(kiq_ring, 0);
216 		amdgpu_ring_write(kiq_ring, 0);
217 	}
218 }
219 
220 static void gfx11_kiq_query_status(struct amdgpu_ring *kiq_ring,
221 				   struct amdgpu_ring *ring,
222 				   u64 addr,
223 				   u64 seq)
224 {
225 	uint32_t eng_sel = ring->funcs->type == AMDGPU_RING_TYPE_GFX ? 4 : 0;
226 
227 	amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_QUERY_STATUS, 5));
228 	amdgpu_ring_write(kiq_ring,
229 			  PACKET3_QUERY_STATUS_CONTEXT_ID(0) |
230 			  PACKET3_QUERY_STATUS_INTERRUPT_SEL(0) |
231 			  PACKET3_QUERY_STATUS_COMMAND(2));
232 	amdgpu_ring_write(kiq_ring, /* Q_sel: 0, vmid: 0, engine: 0, num_Q: 1 */
233 			  PACKET3_QUERY_STATUS_DOORBELL_OFFSET(ring->doorbell_index) |
234 			  PACKET3_QUERY_STATUS_ENG_SEL(eng_sel));
235 	amdgpu_ring_write(kiq_ring, lower_32_bits(addr));
236 	amdgpu_ring_write(kiq_ring, upper_32_bits(addr));
237 	amdgpu_ring_write(kiq_ring, lower_32_bits(seq));
238 	amdgpu_ring_write(kiq_ring, upper_32_bits(seq));
239 }
240 
241 static void gfx11_kiq_invalidate_tlbs(struct amdgpu_ring *kiq_ring,
242 				uint16_t pasid, uint32_t flush_type,
243 				bool all_hub)
244 {
245 	gfx_v11_0_ring_invalidate_tlbs(kiq_ring, pasid, flush_type, all_hub, 1);
246 }
247 
248 static const struct kiq_pm4_funcs gfx_v11_0_kiq_pm4_funcs = {
249 	.kiq_set_resources = gfx11_kiq_set_resources,
250 	.kiq_map_queues = gfx11_kiq_map_queues,
251 	.kiq_unmap_queues = gfx11_kiq_unmap_queues,
252 	.kiq_query_status = gfx11_kiq_query_status,
253 	.kiq_invalidate_tlbs = gfx11_kiq_invalidate_tlbs,
254 	.set_resources_size = 8,
255 	.map_queues_size = 7,
256 	.unmap_queues_size = 6,
257 	.query_status_size = 7,
258 	.invalidate_tlbs_size = 2,
259 };
260 
261 static void gfx_v11_0_set_kiq_pm4_funcs(struct amdgpu_device *adev)
262 {
263 	adev->gfx.kiq[0].pmf = &gfx_v11_0_kiq_pm4_funcs;
264 }
265 
266 static void gfx_v11_0_init_golden_registers(struct amdgpu_device *adev)
267 {
268 	switch (adev->ip_versions[GC_HWIP][0]) {
269 	case IP_VERSION(11, 0, 1):
270 	case IP_VERSION(11, 0, 4):
271 		soc15_program_register_sequence(adev,
272 						golden_settings_gc_11_0_1,
273 						(const u32)ARRAY_SIZE(golden_settings_gc_11_0_1));
274 		break;
275 	default:
276 		break;
277 	}
278 }
279 
280 static void gfx_v11_0_write_data_to_reg(struct amdgpu_ring *ring, int eng_sel,
281 				       bool wc, uint32_t reg, uint32_t val)
282 {
283 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
284 	amdgpu_ring_write(ring, WRITE_DATA_ENGINE_SEL(eng_sel) |
285 			  WRITE_DATA_DST_SEL(0) | (wc ? WR_CONFIRM : 0));
286 	amdgpu_ring_write(ring, reg);
287 	amdgpu_ring_write(ring, 0);
288 	amdgpu_ring_write(ring, val);
289 }
290 
291 static void gfx_v11_0_wait_reg_mem(struct amdgpu_ring *ring, int eng_sel,
292 				  int mem_space, int opt, uint32_t addr0,
293 				  uint32_t addr1, uint32_t ref, uint32_t mask,
294 				  uint32_t inv)
295 {
296 	amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
297 	amdgpu_ring_write(ring,
298 			  /* memory (1) or register (0) */
299 			  (WAIT_REG_MEM_MEM_SPACE(mem_space) |
300 			   WAIT_REG_MEM_OPERATION(opt) | /* wait */
301 			   WAIT_REG_MEM_FUNCTION(3) |  /* equal */
302 			   WAIT_REG_MEM_ENGINE(eng_sel)));
303 
304 	if (mem_space)
305 		BUG_ON(addr0 & 0x3); /* Dword align */
306 	amdgpu_ring_write(ring, addr0);
307 	amdgpu_ring_write(ring, addr1);
308 	amdgpu_ring_write(ring, ref);
309 	amdgpu_ring_write(ring, mask);
310 	amdgpu_ring_write(ring, inv); /* poll interval */
311 }
312 
313 static int gfx_v11_0_ring_test_ring(struct amdgpu_ring *ring)
314 {
315 	struct amdgpu_device *adev = ring->adev;
316 	uint32_t scratch = SOC15_REG_OFFSET(GC, 0, regSCRATCH_REG0);
317 	uint32_t tmp = 0;
318 	unsigned i;
319 	int r;
320 
321 	WREG32(scratch, 0xCAFEDEAD);
322 	r = amdgpu_ring_alloc(ring, 5);
323 	if (r) {
324 		DRM_ERROR("amdgpu: cp failed to lock ring %d (%d).\n",
325 			  ring->idx, r);
326 		return r;
327 	}
328 
329 	if (ring->funcs->type == AMDGPU_RING_TYPE_KIQ) {
330 		gfx_v11_0_ring_emit_wreg(ring, scratch, 0xDEADBEEF);
331 	} else {
332 		amdgpu_ring_write(ring, PACKET3(PACKET3_SET_UCONFIG_REG, 1));
333 		amdgpu_ring_write(ring, scratch -
334 				  PACKET3_SET_UCONFIG_REG_START);
335 		amdgpu_ring_write(ring, 0xDEADBEEF);
336 	}
337 	amdgpu_ring_commit(ring);
338 
339 	for (i = 0; i < adev->usec_timeout; i++) {
340 		tmp = RREG32(scratch);
341 		if (tmp == 0xDEADBEEF)
342 			break;
343 		if (amdgpu_emu_mode == 1)
344 			msleep(1);
345 		else
346 			udelay(1);
347 	}
348 
349 	if (i >= adev->usec_timeout)
350 		r = -ETIMEDOUT;
351 	return r;
352 }
353 
354 static int gfx_v11_0_ring_test_ib(struct amdgpu_ring *ring, long timeout)
355 {
356 	struct amdgpu_device *adev = ring->adev;
357 	struct amdgpu_ib ib;
358 	struct dma_fence *f = NULL;
359 	unsigned index;
360 	uint64_t gpu_addr;
361 	volatile uint32_t *cpu_ptr;
362 	long r;
363 
364 	/* MES KIQ fw hasn't indirect buffer support for now */
365 	if (adev->enable_mes_kiq &&
366 	    ring->funcs->type == AMDGPU_RING_TYPE_KIQ)
367 		return 0;
368 
369 	memset(&ib, 0, sizeof(ib));
370 
371 	if (ring->is_mes_queue) {
372 		uint32_t padding, offset;
373 
374 		offset = amdgpu_mes_ctx_get_offs(ring, AMDGPU_MES_CTX_IB_OFFS);
375 		padding = amdgpu_mes_ctx_get_offs(ring,
376 						  AMDGPU_MES_CTX_PADDING_OFFS);
377 
378 		ib.gpu_addr = amdgpu_mes_ctx_get_offs_gpu_addr(ring, offset);
379 		ib.ptr = amdgpu_mes_ctx_get_offs_cpu_addr(ring, offset);
380 
381 		gpu_addr = amdgpu_mes_ctx_get_offs_gpu_addr(ring, padding);
382 		cpu_ptr = amdgpu_mes_ctx_get_offs_cpu_addr(ring, padding);
383 		*cpu_ptr = cpu_to_le32(0xCAFEDEAD);
384 	} else {
385 		r = amdgpu_device_wb_get(adev, &index);
386 		if (r)
387 			return r;
388 
389 		gpu_addr = adev->wb.gpu_addr + (index * 4);
390 		adev->wb.wb[index] = cpu_to_le32(0xCAFEDEAD);
391 		cpu_ptr = &adev->wb.wb[index];
392 
393 		r = amdgpu_ib_get(adev, NULL, 16, AMDGPU_IB_POOL_DIRECT, &ib);
394 		if (r) {
395 			DRM_ERROR("amdgpu: failed to get ib (%ld).\n", r);
396 			goto err1;
397 		}
398 	}
399 
400 	ib.ptr[0] = PACKET3(PACKET3_WRITE_DATA, 3);
401 	ib.ptr[1] = WRITE_DATA_DST_SEL(5) | WR_CONFIRM;
402 	ib.ptr[2] = lower_32_bits(gpu_addr);
403 	ib.ptr[3] = upper_32_bits(gpu_addr);
404 	ib.ptr[4] = 0xDEADBEEF;
405 	ib.length_dw = 5;
406 
407 	r = amdgpu_ib_schedule(ring, 1, &ib, NULL, &f);
408 	if (r)
409 		goto err2;
410 
411 	r = dma_fence_wait_timeout(f, false, timeout);
412 	if (r == 0) {
413 		r = -ETIMEDOUT;
414 		goto err2;
415 	} else if (r < 0) {
416 		goto err2;
417 	}
418 
419 	if (le32_to_cpu(*cpu_ptr) == 0xDEADBEEF)
420 		r = 0;
421 	else
422 		r = -EINVAL;
423 err2:
424 	if (!ring->is_mes_queue)
425 		amdgpu_ib_free(adev, &ib, NULL);
426 	dma_fence_put(f);
427 err1:
428 	if (!ring->is_mes_queue)
429 		amdgpu_device_wb_free(adev, index);
430 	return r;
431 }
432 
433 static void gfx_v11_0_free_microcode(struct amdgpu_device *adev)
434 {
435 	amdgpu_ucode_release(&adev->gfx.pfp_fw);
436 	amdgpu_ucode_release(&adev->gfx.me_fw);
437 	amdgpu_ucode_release(&adev->gfx.rlc_fw);
438 	amdgpu_ucode_release(&adev->gfx.mec_fw);
439 
440 	kfree(adev->gfx.rlc.register_list_format);
441 }
442 
443 static int gfx_v11_0_init_toc_microcode(struct amdgpu_device *adev, const char *ucode_prefix)
444 {
445 	const struct psp_firmware_header_v1_0 *toc_hdr;
446 	int err = 0;
447 	char fw_name[40];
448 
449 	snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_toc.bin", ucode_prefix);
450 	err = amdgpu_ucode_request(adev, &adev->psp.toc_fw, fw_name);
451 	if (err)
452 		goto out;
453 
454 	toc_hdr = (const struct psp_firmware_header_v1_0 *)adev->psp.toc_fw->data;
455 	adev->psp.toc.fw_version = le32_to_cpu(toc_hdr->header.ucode_version);
456 	adev->psp.toc.feature_version = le32_to_cpu(toc_hdr->sos.fw_version);
457 	adev->psp.toc.size_bytes = le32_to_cpu(toc_hdr->header.ucode_size_bytes);
458 	adev->psp.toc.start_addr = (uint8_t *)toc_hdr +
459 				le32_to_cpu(toc_hdr->header.ucode_array_offset_bytes);
460 	return 0;
461 out:
462 	amdgpu_ucode_release(&adev->psp.toc_fw);
463 	return err;
464 }
465 
466 static void gfx_v11_0_check_fw_cp_gfx_shadow(struct amdgpu_device *adev)
467 {
468 	switch (adev->ip_versions[GC_HWIP][0]) {
469 	case IP_VERSION(11, 0, 0):
470 	case IP_VERSION(11, 0, 2):
471 	case IP_VERSION(11, 0, 3):
472 		if ((adev->gfx.me_fw_version >= 1505) &&
473 		    (adev->gfx.pfp_fw_version >= 1600) &&
474 		    (adev->gfx.mec_fw_version >= 512))
475 			adev->gfx.cp_gfx_shadow = true;
476 		break;
477 	default:
478 		adev->gfx.cp_gfx_shadow = false;
479 		break;
480 	}
481 }
482 
483 static int gfx_v11_0_init_microcode(struct amdgpu_device *adev)
484 {
485 	char fw_name[40];
486 	char ucode_prefix[30];
487 	int err;
488 	const struct rlc_firmware_header_v2_0 *rlc_hdr;
489 	uint16_t version_major;
490 	uint16_t version_minor;
491 
492 	DRM_DEBUG("\n");
493 
494 	amdgpu_ucode_ip_version_decode(adev, GC_HWIP, ucode_prefix, sizeof(ucode_prefix));
495 
496 	snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_pfp.bin", ucode_prefix);
497 	err = amdgpu_ucode_request(adev, &adev->gfx.pfp_fw, fw_name);
498 	if (err)
499 		goto out;
500 	/* check pfp fw hdr version to decide if enable rs64 for gfx11.*/
501 	adev->gfx.rs64_enable = amdgpu_ucode_hdr_version(
502 				(union amdgpu_firmware_header *)
503 				adev->gfx.pfp_fw->data, 2, 0);
504 	if (adev->gfx.rs64_enable) {
505 		dev_info(adev->dev, "CP RS64 enable\n");
506 		amdgpu_gfx_cp_init_microcode(adev, AMDGPU_UCODE_ID_CP_RS64_PFP);
507 		amdgpu_gfx_cp_init_microcode(adev, AMDGPU_UCODE_ID_CP_RS64_PFP_P0_STACK);
508 		amdgpu_gfx_cp_init_microcode(adev, AMDGPU_UCODE_ID_CP_RS64_PFP_P1_STACK);
509 	} else {
510 		amdgpu_gfx_cp_init_microcode(adev, AMDGPU_UCODE_ID_CP_PFP);
511 	}
512 
513 	snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_me.bin", ucode_prefix);
514 	err = amdgpu_ucode_request(adev, &adev->gfx.me_fw, fw_name);
515 	if (err)
516 		goto out;
517 	if (adev->gfx.rs64_enable) {
518 		amdgpu_gfx_cp_init_microcode(adev, AMDGPU_UCODE_ID_CP_RS64_ME);
519 		amdgpu_gfx_cp_init_microcode(adev, AMDGPU_UCODE_ID_CP_RS64_ME_P0_STACK);
520 		amdgpu_gfx_cp_init_microcode(adev, AMDGPU_UCODE_ID_CP_RS64_ME_P1_STACK);
521 	} else {
522 		amdgpu_gfx_cp_init_microcode(adev, AMDGPU_UCODE_ID_CP_ME);
523 	}
524 
525 	if (!amdgpu_sriov_vf(adev)) {
526 		snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_rlc.bin", ucode_prefix);
527 		err = amdgpu_ucode_request(adev, &adev->gfx.rlc_fw, fw_name);
528 		if (err)
529 			goto out;
530 		rlc_hdr = (const struct rlc_firmware_header_v2_0 *)adev->gfx.rlc_fw->data;
531 		version_major = le16_to_cpu(rlc_hdr->header.header_version_major);
532 		version_minor = le16_to_cpu(rlc_hdr->header.header_version_minor);
533 		err = amdgpu_gfx_rlc_init_microcode(adev, version_major, version_minor);
534 		if (err)
535 			goto out;
536 	}
537 
538 	snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec.bin", ucode_prefix);
539 	err = amdgpu_ucode_request(adev, &adev->gfx.mec_fw, fw_name);
540 	if (err)
541 		goto out;
542 	if (adev->gfx.rs64_enable) {
543 		amdgpu_gfx_cp_init_microcode(adev, AMDGPU_UCODE_ID_CP_RS64_MEC);
544 		amdgpu_gfx_cp_init_microcode(adev, AMDGPU_UCODE_ID_CP_RS64_MEC_P0_STACK);
545 		amdgpu_gfx_cp_init_microcode(adev, AMDGPU_UCODE_ID_CP_RS64_MEC_P1_STACK);
546 		amdgpu_gfx_cp_init_microcode(adev, AMDGPU_UCODE_ID_CP_RS64_MEC_P2_STACK);
547 		amdgpu_gfx_cp_init_microcode(adev, AMDGPU_UCODE_ID_CP_RS64_MEC_P3_STACK);
548 	} else {
549 		amdgpu_gfx_cp_init_microcode(adev, AMDGPU_UCODE_ID_CP_MEC1);
550 		amdgpu_gfx_cp_init_microcode(adev, AMDGPU_UCODE_ID_CP_MEC1_JT);
551 	}
552 
553 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_RLC_BACKDOOR_AUTO)
554 		err = gfx_v11_0_init_toc_microcode(adev, ucode_prefix);
555 
556 	/* only one MEC for gfx 11.0.0. */
557 	adev->gfx.mec2_fw = NULL;
558 
559 	gfx_v11_0_check_fw_cp_gfx_shadow(adev);
560 out:
561 	if (err) {
562 		amdgpu_ucode_release(&adev->gfx.pfp_fw);
563 		amdgpu_ucode_release(&adev->gfx.me_fw);
564 		amdgpu_ucode_release(&adev->gfx.rlc_fw);
565 		amdgpu_ucode_release(&adev->gfx.mec_fw);
566 	}
567 
568 	return err;
569 }
570 
571 static u32 gfx_v11_0_get_csb_size(struct amdgpu_device *adev)
572 {
573 	u32 count = 0;
574 	const struct cs_section_def *sect = NULL;
575 	const struct cs_extent_def *ext = NULL;
576 
577 	/* begin clear state */
578 	count += 2;
579 	/* context control state */
580 	count += 3;
581 
582 	for (sect = gfx11_cs_data; sect->section != NULL; ++sect) {
583 		for (ext = sect->section; ext->extent != NULL; ++ext) {
584 			if (sect->id == SECT_CONTEXT)
585 				count += 2 + ext->reg_count;
586 			else
587 				return 0;
588 		}
589 	}
590 
591 	/* set PA_SC_TILE_STEERING_OVERRIDE */
592 	count += 3;
593 	/* end clear state */
594 	count += 2;
595 	/* clear state */
596 	count += 2;
597 
598 	return count;
599 }
600 
601 static void gfx_v11_0_get_csb_buffer(struct amdgpu_device *adev,
602 				    volatile u32 *buffer)
603 {
604 	u32 count = 0, i;
605 	const struct cs_section_def *sect = NULL;
606 	const struct cs_extent_def *ext = NULL;
607 	int ctx_reg_offset;
608 
609 	if (adev->gfx.rlc.cs_data == NULL)
610 		return;
611 	if (buffer == NULL)
612 		return;
613 
614 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0));
615 	buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
616 
617 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CONTEXT_CONTROL, 1));
618 	buffer[count++] = cpu_to_le32(0x80000000);
619 	buffer[count++] = cpu_to_le32(0x80000000);
620 
621 	for (sect = adev->gfx.rlc.cs_data; sect->section != NULL; ++sect) {
622 		for (ext = sect->section; ext->extent != NULL; ++ext) {
623 			if (sect->id == SECT_CONTEXT) {
624 				buffer[count++] =
625 					cpu_to_le32(PACKET3(PACKET3_SET_CONTEXT_REG, ext->reg_count));
626 				buffer[count++] = cpu_to_le32(ext->reg_index -
627 						PACKET3_SET_CONTEXT_REG_START);
628 				for (i = 0; i < ext->reg_count; i++)
629 					buffer[count++] = cpu_to_le32(ext->extent[i]);
630 			} else {
631 				return;
632 			}
633 		}
634 	}
635 
636 	ctx_reg_offset =
637 		SOC15_REG_OFFSET(GC, 0, regPA_SC_TILE_STEERING_OVERRIDE) - PACKET3_SET_CONTEXT_REG_START;
638 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_SET_CONTEXT_REG, 1));
639 	buffer[count++] = cpu_to_le32(ctx_reg_offset);
640 	buffer[count++] = cpu_to_le32(adev->gfx.config.pa_sc_tile_steering_override);
641 
642 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0));
643 	buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_END_CLEAR_STATE);
644 
645 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CLEAR_STATE, 0));
646 	buffer[count++] = cpu_to_le32(0);
647 }
648 
649 static void gfx_v11_0_rlc_fini(struct amdgpu_device *adev)
650 {
651 	/* clear state block */
652 	amdgpu_bo_free_kernel(&adev->gfx.rlc.clear_state_obj,
653 			&adev->gfx.rlc.clear_state_gpu_addr,
654 			(void **)&adev->gfx.rlc.cs_ptr);
655 
656 	/* jump table block */
657 	amdgpu_bo_free_kernel(&adev->gfx.rlc.cp_table_obj,
658 			&adev->gfx.rlc.cp_table_gpu_addr,
659 			(void **)&adev->gfx.rlc.cp_table_ptr);
660 }
661 
662 static void gfx_v11_0_init_rlcg_reg_access_ctrl(struct amdgpu_device *adev)
663 {
664 	struct amdgpu_rlcg_reg_access_ctrl *reg_access_ctrl;
665 
666 	reg_access_ctrl = &adev->gfx.rlc.reg_access_ctrl[0];
667 	reg_access_ctrl->scratch_reg0 = SOC15_REG_OFFSET(GC, 0, regSCRATCH_REG0);
668 	reg_access_ctrl->scratch_reg1 = SOC15_REG_OFFSET(GC, 0, regSCRATCH_REG1);
669 	reg_access_ctrl->scratch_reg2 = SOC15_REG_OFFSET(GC, 0, regSCRATCH_REG2);
670 	reg_access_ctrl->scratch_reg3 = SOC15_REG_OFFSET(GC, 0, regSCRATCH_REG3);
671 	reg_access_ctrl->grbm_cntl = SOC15_REG_OFFSET(GC, 0, regGRBM_GFX_CNTL);
672 	reg_access_ctrl->grbm_idx = SOC15_REG_OFFSET(GC, 0, regGRBM_GFX_INDEX);
673 	reg_access_ctrl->spare_int = SOC15_REG_OFFSET(GC, 0, regRLC_SPARE_INT_0);
674 	adev->gfx.rlc.rlcg_reg_access_supported = true;
675 }
676 
677 static int gfx_v11_0_rlc_init(struct amdgpu_device *adev)
678 {
679 	const struct cs_section_def *cs_data;
680 	int r;
681 
682 	adev->gfx.rlc.cs_data = gfx11_cs_data;
683 
684 	cs_data = adev->gfx.rlc.cs_data;
685 
686 	if (cs_data) {
687 		/* init clear state block */
688 		r = amdgpu_gfx_rlc_init_csb(adev);
689 		if (r)
690 			return r;
691 	}
692 
693 	/* init spm vmid with 0xf */
694 	if (adev->gfx.rlc.funcs->update_spm_vmid)
695 		adev->gfx.rlc.funcs->update_spm_vmid(adev, 0xf);
696 
697 	return 0;
698 }
699 
700 static void gfx_v11_0_mec_fini(struct amdgpu_device *adev)
701 {
702 	amdgpu_bo_free_kernel(&adev->gfx.mec.hpd_eop_obj, NULL, NULL);
703 	amdgpu_bo_free_kernel(&adev->gfx.mec.mec_fw_obj, NULL, NULL);
704 	amdgpu_bo_free_kernel(&adev->gfx.mec.mec_fw_data_obj, NULL, NULL);
705 }
706 
707 static void gfx_v11_0_me_init(struct amdgpu_device *adev)
708 {
709 	bitmap_zero(adev->gfx.me.queue_bitmap, AMDGPU_MAX_GFX_QUEUES);
710 
711 	amdgpu_gfx_graphics_queue_acquire(adev);
712 }
713 
714 static int gfx_v11_0_mec_init(struct amdgpu_device *adev)
715 {
716 	int r;
717 	u32 *hpd;
718 	size_t mec_hpd_size;
719 
720 	bitmap_zero(adev->gfx.mec_bitmap[0].queue_bitmap, AMDGPU_MAX_COMPUTE_QUEUES);
721 
722 	/* take ownership of the relevant compute queues */
723 	amdgpu_gfx_compute_queue_acquire(adev);
724 	mec_hpd_size = adev->gfx.num_compute_rings * GFX11_MEC_HPD_SIZE;
725 
726 	if (mec_hpd_size) {
727 		r = amdgpu_bo_create_reserved(adev, mec_hpd_size, PAGE_SIZE,
728 					      AMDGPU_GEM_DOMAIN_GTT,
729 					      &adev->gfx.mec.hpd_eop_obj,
730 					      &adev->gfx.mec.hpd_eop_gpu_addr,
731 					      (void **)&hpd);
732 		if (r) {
733 			dev_warn(adev->dev, "(%d) create HDP EOP bo failed\n", r);
734 			gfx_v11_0_mec_fini(adev);
735 			return r;
736 		}
737 
738 		memset(hpd, 0, mec_hpd_size);
739 
740 		amdgpu_bo_kunmap(adev->gfx.mec.hpd_eop_obj);
741 		amdgpu_bo_unreserve(adev->gfx.mec.hpd_eop_obj);
742 	}
743 
744 	return 0;
745 }
746 
747 static uint32_t wave_read_ind(struct amdgpu_device *adev, uint32_t wave, uint32_t address)
748 {
749 	WREG32_SOC15(GC, 0, regSQ_IND_INDEX,
750 		(wave << SQ_IND_INDEX__WAVE_ID__SHIFT) |
751 		(address << SQ_IND_INDEX__INDEX__SHIFT));
752 	return RREG32_SOC15(GC, 0, regSQ_IND_DATA);
753 }
754 
755 static void wave_read_regs(struct amdgpu_device *adev, uint32_t wave,
756 			   uint32_t thread, uint32_t regno,
757 			   uint32_t num, uint32_t *out)
758 {
759 	WREG32_SOC15(GC, 0, regSQ_IND_INDEX,
760 		(wave << SQ_IND_INDEX__WAVE_ID__SHIFT) |
761 		(regno << SQ_IND_INDEX__INDEX__SHIFT) |
762 		(thread << SQ_IND_INDEX__WORKITEM_ID__SHIFT) |
763 		(SQ_IND_INDEX__AUTO_INCR_MASK));
764 	while (num--)
765 		*(out++) = RREG32_SOC15(GC, 0, regSQ_IND_DATA);
766 }
767 
768 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)
769 {
770 	/* in gfx11 the SIMD_ID is specified as part of the INSTANCE
771 	 * field when performing a select_se_sh so it should be
772 	 * zero here */
773 	WARN_ON(simd != 0);
774 
775 	/* type 3 wave data */
776 	dst[(*no_fields)++] = 3;
777 	dst[(*no_fields)++] = wave_read_ind(adev, wave, ixSQ_WAVE_STATUS);
778 	dst[(*no_fields)++] = wave_read_ind(adev, wave, ixSQ_WAVE_PC_LO);
779 	dst[(*no_fields)++] = wave_read_ind(adev, wave, ixSQ_WAVE_PC_HI);
780 	dst[(*no_fields)++] = wave_read_ind(adev, wave, ixSQ_WAVE_EXEC_LO);
781 	dst[(*no_fields)++] = wave_read_ind(adev, wave, ixSQ_WAVE_EXEC_HI);
782 	dst[(*no_fields)++] = wave_read_ind(adev, wave, ixSQ_WAVE_HW_ID1);
783 	dst[(*no_fields)++] = wave_read_ind(adev, wave, ixSQ_WAVE_HW_ID2);
784 	dst[(*no_fields)++] = wave_read_ind(adev, wave, ixSQ_WAVE_GPR_ALLOC);
785 	dst[(*no_fields)++] = wave_read_ind(adev, wave, ixSQ_WAVE_LDS_ALLOC);
786 	dst[(*no_fields)++] = wave_read_ind(adev, wave, ixSQ_WAVE_TRAPSTS);
787 	dst[(*no_fields)++] = wave_read_ind(adev, wave, ixSQ_WAVE_IB_STS);
788 	dst[(*no_fields)++] = wave_read_ind(adev, wave, ixSQ_WAVE_IB_STS2);
789 	dst[(*no_fields)++] = wave_read_ind(adev, wave, ixSQ_WAVE_IB_DBG1);
790 	dst[(*no_fields)++] = wave_read_ind(adev, wave, ixSQ_WAVE_M0);
791 	dst[(*no_fields)++] = wave_read_ind(adev, wave, ixSQ_WAVE_MODE);
792 }
793 
794 static void gfx_v11_0_read_wave_sgprs(struct amdgpu_device *adev, uint32_t xcc_id, uint32_t simd,
795 				     uint32_t wave, uint32_t start,
796 				     uint32_t size, uint32_t *dst)
797 {
798 	WARN_ON(simd != 0);
799 
800 	wave_read_regs(
801 		adev, wave, 0, start + SQIND_WAVE_SGPRS_OFFSET, size,
802 		dst);
803 }
804 
805 static void gfx_v11_0_read_wave_vgprs(struct amdgpu_device *adev, uint32_t xcc_id, uint32_t simd,
806 				      uint32_t wave, uint32_t thread,
807 				      uint32_t start, uint32_t size,
808 				      uint32_t *dst)
809 {
810 	wave_read_regs(
811 		adev, wave, thread,
812 		start + SQIND_WAVE_VGPRS_OFFSET, size, dst);
813 }
814 
815 static void gfx_v11_0_select_me_pipe_q(struct amdgpu_device *adev,
816 					u32 me, u32 pipe, u32 q, u32 vm, u32 xcc_id)
817 {
818 	soc21_grbm_select(adev, me, pipe, q, vm);
819 }
820 
821 /* all sizes are in bytes */
822 #define MQD_SHADOW_BASE_SIZE      73728
823 #define MQD_SHADOW_BASE_ALIGNMENT 256
824 #define MQD_FWWORKAREA_SIZE       484
825 #define MQD_FWWORKAREA_ALIGNMENT  256
826 
827 static int gfx_v11_0_get_gfx_shadow_info(struct amdgpu_device *adev,
828 					 struct amdgpu_gfx_shadow_info *shadow_info)
829 {
830 	if (adev->gfx.cp_gfx_shadow) {
831 		shadow_info->shadow_size = MQD_SHADOW_BASE_SIZE;
832 		shadow_info->shadow_alignment = MQD_SHADOW_BASE_ALIGNMENT;
833 		shadow_info->csa_size = MQD_FWWORKAREA_SIZE;
834 		shadow_info->csa_alignment = MQD_FWWORKAREA_ALIGNMENT;
835 		return 0;
836 	} else {
837 		memset(shadow_info, 0, sizeof(struct amdgpu_gfx_shadow_info));
838 		return -ENOTSUPP;
839 	}
840 }
841 
842 static const struct amdgpu_gfx_funcs gfx_v11_0_gfx_funcs = {
843 	.get_gpu_clock_counter = &gfx_v11_0_get_gpu_clock_counter,
844 	.select_se_sh = &gfx_v11_0_select_se_sh,
845 	.read_wave_data = &gfx_v11_0_read_wave_data,
846 	.read_wave_sgprs = &gfx_v11_0_read_wave_sgprs,
847 	.read_wave_vgprs = &gfx_v11_0_read_wave_vgprs,
848 	.select_me_pipe_q = &gfx_v11_0_select_me_pipe_q,
849 	.update_perfmon_mgcg = &gfx_v11_0_update_perf_clk,
850 	.get_gfx_shadow_info = &gfx_v11_0_get_gfx_shadow_info,
851 };
852 
853 static int gfx_v11_0_gpu_early_init(struct amdgpu_device *adev)
854 {
855 
856 	switch (adev->ip_versions[GC_HWIP][0]) {
857 	case IP_VERSION(11, 0, 0):
858 	case IP_VERSION(11, 0, 2):
859 		adev->gfx.config.max_hw_contexts = 8;
860 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
861 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
862 		adev->gfx.config.sc_hiz_tile_fifo_size = 0;
863 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x4C0;
864 		break;
865 	case IP_VERSION(11, 0, 3):
866 		adev->gfx.ras = &gfx_v11_0_3_ras;
867 		adev->gfx.config.max_hw_contexts = 8;
868 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
869 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
870 		adev->gfx.config.sc_hiz_tile_fifo_size = 0;
871 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x4C0;
872 		break;
873 	case IP_VERSION(11, 0, 1):
874 	case IP_VERSION(11, 0, 4):
875 		adev->gfx.config.max_hw_contexts = 8;
876 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
877 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
878 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x80;
879 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x300;
880 		break;
881 	default:
882 		BUG();
883 		break;
884 	}
885 
886 	return 0;
887 }
888 
889 static int gfx_v11_0_gfx_ring_init(struct amdgpu_device *adev, int ring_id,
890 				   int me, int pipe, int queue)
891 {
892 	int r;
893 	struct amdgpu_ring *ring;
894 	unsigned int irq_type;
895 
896 	ring = &adev->gfx.gfx_ring[ring_id];
897 
898 	ring->me = me;
899 	ring->pipe = pipe;
900 	ring->queue = queue;
901 
902 	ring->ring_obj = NULL;
903 	ring->use_doorbell = true;
904 
905 	if (!ring_id)
906 		ring->doorbell_index = adev->doorbell_index.gfx_ring0 << 1;
907 	else
908 		ring->doorbell_index = adev->doorbell_index.gfx_ring1 << 1;
909 	ring->vm_hub = AMDGPU_GFXHUB(0);
910 	sprintf(ring->name, "gfx_%d.%d.%d", ring->me, ring->pipe, ring->queue);
911 
912 	irq_type = AMDGPU_CP_IRQ_GFX_ME0_PIPE0_EOP + ring->pipe;
913 	r = amdgpu_ring_init(adev, ring, 1024, &adev->gfx.eop_irq, irq_type,
914 			     AMDGPU_RING_PRIO_DEFAULT, NULL);
915 	if (r)
916 		return r;
917 	return 0;
918 }
919 
920 static int gfx_v11_0_compute_ring_init(struct amdgpu_device *adev, int ring_id,
921 				       int mec, int pipe, int queue)
922 {
923 	int r;
924 	unsigned irq_type;
925 	struct amdgpu_ring *ring;
926 	unsigned int hw_prio;
927 
928 	ring = &adev->gfx.compute_ring[ring_id];
929 
930 	/* mec0 is me1 */
931 	ring->me = mec + 1;
932 	ring->pipe = pipe;
933 	ring->queue = queue;
934 
935 	ring->ring_obj = NULL;
936 	ring->use_doorbell = true;
937 	ring->doorbell_index = (adev->doorbell_index.mec_ring0 + ring_id) << 1;
938 	ring->eop_gpu_addr = adev->gfx.mec.hpd_eop_gpu_addr
939 				+ (ring_id * GFX11_MEC_HPD_SIZE);
940 	ring->vm_hub = AMDGPU_GFXHUB(0);
941 	sprintf(ring->name, "comp_%d.%d.%d", ring->me, ring->pipe, ring->queue);
942 
943 	irq_type = AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP
944 		+ ((ring->me - 1) * adev->gfx.mec.num_pipe_per_mec)
945 		+ ring->pipe;
946 	hw_prio = amdgpu_gfx_is_high_priority_compute_queue(adev, ring) ?
947 			AMDGPU_GFX_PIPE_PRIO_HIGH : AMDGPU_GFX_PIPE_PRIO_NORMAL;
948 	/* type-2 packets are deprecated on MEC, use type-3 instead */
949 	r = amdgpu_ring_init(adev, ring, 1024, &adev->gfx.eop_irq, irq_type,
950 			     hw_prio, NULL);
951 	if (r)
952 		return r;
953 
954 	return 0;
955 }
956 
957 static struct {
958 	SOC21_FIRMWARE_ID	id;
959 	unsigned int		offset;
960 	unsigned int		size;
961 } rlc_autoload_info[SOC21_FIRMWARE_ID_MAX];
962 
963 static void gfx_v11_0_parse_rlc_toc(struct amdgpu_device *adev, void *rlc_toc)
964 {
965 	RLC_TABLE_OF_CONTENT *ucode = rlc_toc;
966 
967 	while (ucode && (ucode->id > SOC21_FIRMWARE_ID_INVALID) &&
968 			(ucode->id < SOC21_FIRMWARE_ID_MAX)) {
969 		rlc_autoload_info[ucode->id].id = ucode->id;
970 		rlc_autoload_info[ucode->id].offset = ucode->offset * 4;
971 		rlc_autoload_info[ucode->id].size = ucode->size * 4;
972 
973 		ucode++;
974 	}
975 }
976 
977 static uint32_t gfx_v11_0_calc_toc_total_size(struct amdgpu_device *adev)
978 {
979 	uint32_t total_size = 0;
980 	SOC21_FIRMWARE_ID id;
981 
982 	gfx_v11_0_parse_rlc_toc(adev, adev->psp.toc.start_addr);
983 
984 	for (id = SOC21_FIRMWARE_ID_RLC_G_UCODE; id < SOC21_FIRMWARE_ID_MAX; id++)
985 		total_size += rlc_autoload_info[id].size;
986 
987 	/* In case the offset in rlc toc ucode is aligned */
988 	if (total_size < rlc_autoload_info[SOC21_FIRMWARE_ID_MAX-1].offset)
989 		total_size = rlc_autoload_info[SOC21_FIRMWARE_ID_MAX-1].offset +
990 			rlc_autoload_info[SOC21_FIRMWARE_ID_MAX-1].size;
991 
992 	return total_size;
993 }
994 
995 static int gfx_v11_0_rlc_autoload_buffer_init(struct amdgpu_device *adev)
996 {
997 	int r;
998 	uint32_t total_size;
999 
1000 	total_size = gfx_v11_0_calc_toc_total_size(adev);
1001 
1002 	r = amdgpu_bo_create_reserved(adev, total_size, 64 * 1024,
1003 				      AMDGPU_GEM_DOMAIN_VRAM |
1004 				      AMDGPU_GEM_DOMAIN_GTT,
1005 				      &adev->gfx.rlc.rlc_autoload_bo,
1006 				      &adev->gfx.rlc.rlc_autoload_gpu_addr,
1007 				      (void **)&adev->gfx.rlc.rlc_autoload_ptr);
1008 
1009 	if (r) {
1010 		dev_err(adev->dev, "(%d) failed to create fw autoload bo\n", r);
1011 		return r;
1012 	}
1013 
1014 	return 0;
1015 }
1016 
1017 static void gfx_v11_0_rlc_backdoor_autoload_copy_ucode(struct amdgpu_device *adev,
1018 					      SOC21_FIRMWARE_ID id,
1019 			    		      const void *fw_data,
1020 					      uint32_t fw_size,
1021 					      uint32_t *fw_autoload_mask)
1022 {
1023 	uint32_t toc_offset;
1024 	uint32_t toc_fw_size;
1025 	char *ptr = adev->gfx.rlc.rlc_autoload_ptr;
1026 
1027 	if (id <= SOC21_FIRMWARE_ID_INVALID || id >= SOC21_FIRMWARE_ID_MAX)
1028 		return;
1029 
1030 	toc_offset = rlc_autoload_info[id].offset;
1031 	toc_fw_size = rlc_autoload_info[id].size;
1032 
1033 	if (fw_size == 0)
1034 		fw_size = toc_fw_size;
1035 
1036 	if (fw_size > toc_fw_size)
1037 		fw_size = toc_fw_size;
1038 
1039 	memcpy(ptr + toc_offset, fw_data, fw_size);
1040 
1041 	if (fw_size < toc_fw_size)
1042 		memset(ptr + toc_offset + fw_size, 0, toc_fw_size - fw_size);
1043 
1044 	if ((id != SOC21_FIRMWARE_ID_RS64_PFP) && (id != SOC21_FIRMWARE_ID_RS64_ME))
1045 		*(uint64_t *)fw_autoload_mask |= 1ULL << id;
1046 }
1047 
1048 static void gfx_v11_0_rlc_backdoor_autoload_copy_toc_ucode(struct amdgpu_device *adev,
1049 							uint32_t *fw_autoload_mask)
1050 {
1051 	void *data;
1052 	uint32_t size;
1053 	uint64_t *toc_ptr;
1054 
1055 	*(uint64_t *)fw_autoload_mask |= 0x1;
1056 
1057 	DRM_DEBUG("rlc autoload enabled fw: 0x%llx\n", *(uint64_t *)fw_autoload_mask);
1058 
1059 	data = adev->psp.toc.start_addr;
1060 	size = rlc_autoload_info[SOC21_FIRMWARE_ID_RLC_TOC].size;
1061 
1062 	toc_ptr = (uint64_t *)data + size / 8 - 1;
1063 	*toc_ptr = *(uint64_t *)fw_autoload_mask;
1064 
1065 	gfx_v11_0_rlc_backdoor_autoload_copy_ucode(adev, SOC21_FIRMWARE_ID_RLC_TOC,
1066 					data, size, fw_autoload_mask);
1067 }
1068 
1069 static void gfx_v11_0_rlc_backdoor_autoload_copy_gfx_ucode(struct amdgpu_device *adev,
1070 							uint32_t *fw_autoload_mask)
1071 {
1072 	const __le32 *fw_data;
1073 	uint32_t fw_size;
1074 	const struct gfx_firmware_header_v1_0 *cp_hdr;
1075 	const struct gfx_firmware_header_v2_0 *cpv2_hdr;
1076 	const struct rlc_firmware_header_v2_0 *rlc_hdr;
1077 	const struct rlc_firmware_header_v2_2 *rlcv22_hdr;
1078 	uint16_t version_major, version_minor;
1079 
1080 	if (adev->gfx.rs64_enable) {
1081 		/* pfp ucode */
1082 		cpv2_hdr = (const struct gfx_firmware_header_v2_0 *)
1083 			adev->gfx.pfp_fw->data;
1084 		/* instruction */
1085 		fw_data = (const __le32 *)(adev->gfx.pfp_fw->data +
1086 			le32_to_cpu(cpv2_hdr->ucode_offset_bytes));
1087 		fw_size = le32_to_cpu(cpv2_hdr->ucode_size_bytes);
1088 		gfx_v11_0_rlc_backdoor_autoload_copy_ucode(adev, SOC21_FIRMWARE_ID_RS64_PFP,
1089 						fw_data, fw_size, fw_autoload_mask);
1090 		/* data */
1091 		fw_data = (const __le32 *)(adev->gfx.pfp_fw->data +
1092 			le32_to_cpu(cpv2_hdr->data_offset_bytes));
1093 		fw_size = le32_to_cpu(cpv2_hdr->data_size_bytes);
1094 		gfx_v11_0_rlc_backdoor_autoload_copy_ucode(adev, SOC21_FIRMWARE_ID_RS64_PFP_P0_STACK,
1095 						fw_data, fw_size, fw_autoload_mask);
1096 		gfx_v11_0_rlc_backdoor_autoload_copy_ucode(adev, SOC21_FIRMWARE_ID_RS64_PFP_P1_STACK,
1097 						fw_data, fw_size, fw_autoload_mask);
1098 		/* me ucode */
1099 		cpv2_hdr = (const struct gfx_firmware_header_v2_0 *)
1100 			adev->gfx.me_fw->data;
1101 		/* instruction */
1102 		fw_data = (const __le32 *)(adev->gfx.me_fw->data +
1103 			le32_to_cpu(cpv2_hdr->ucode_offset_bytes));
1104 		fw_size = le32_to_cpu(cpv2_hdr->ucode_size_bytes);
1105 		gfx_v11_0_rlc_backdoor_autoload_copy_ucode(adev, SOC21_FIRMWARE_ID_RS64_ME,
1106 						fw_data, fw_size, fw_autoload_mask);
1107 		/* data */
1108 		fw_data = (const __le32 *)(adev->gfx.me_fw->data +
1109 			le32_to_cpu(cpv2_hdr->data_offset_bytes));
1110 		fw_size = le32_to_cpu(cpv2_hdr->data_size_bytes);
1111 		gfx_v11_0_rlc_backdoor_autoload_copy_ucode(adev, SOC21_FIRMWARE_ID_RS64_ME_P0_STACK,
1112 						fw_data, fw_size, fw_autoload_mask);
1113 		gfx_v11_0_rlc_backdoor_autoload_copy_ucode(adev, SOC21_FIRMWARE_ID_RS64_ME_P1_STACK,
1114 						fw_data, fw_size, fw_autoload_mask);
1115 		/* mec ucode */
1116 		cpv2_hdr = (const struct gfx_firmware_header_v2_0 *)
1117 			adev->gfx.mec_fw->data;
1118 		/* instruction */
1119 		fw_data = (const __le32 *) (adev->gfx.mec_fw->data +
1120 			le32_to_cpu(cpv2_hdr->ucode_offset_bytes));
1121 		fw_size = le32_to_cpu(cpv2_hdr->ucode_size_bytes);
1122 		gfx_v11_0_rlc_backdoor_autoload_copy_ucode(adev, SOC21_FIRMWARE_ID_RS64_MEC,
1123 						fw_data, fw_size, fw_autoload_mask);
1124 		/* data */
1125 		fw_data = (const __le32 *) (adev->gfx.mec_fw->data +
1126 			le32_to_cpu(cpv2_hdr->data_offset_bytes));
1127 		fw_size = le32_to_cpu(cpv2_hdr->data_size_bytes);
1128 		gfx_v11_0_rlc_backdoor_autoload_copy_ucode(adev, SOC21_FIRMWARE_ID_RS64_MEC_P0_STACK,
1129 						fw_data, fw_size, fw_autoload_mask);
1130 		gfx_v11_0_rlc_backdoor_autoload_copy_ucode(adev, SOC21_FIRMWARE_ID_RS64_MEC_P1_STACK,
1131 						fw_data, fw_size, fw_autoload_mask);
1132 		gfx_v11_0_rlc_backdoor_autoload_copy_ucode(adev, SOC21_FIRMWARE_ID_RS64_MEC_P2_STACK,
1133 						fw_data, fw_size, fw_autoload_mask);
1134 		gfx_v11_0_rlc_backdoor_autoload_copy_ucode(adev, SOC21_FIRMWARE_ID_RS64_MEC_P3_STACK,
1135 						fw_data, fw_size, fw_autoload_mask);
1136 	} else {
1137 		/* pfp ucode */
1138 		cp_hdr = (const struct gfx_firmware_header_v1_0 *)
1139 			adev->gfx.pfp_fw->data;
1140 		fw_data = (const __le32 *)(adev->gfx.pfp_fw->data +
1141 				le32_to_cpu(cp_hdr->header.ucode_array_offset_bytes));
1142 		fw_size = le32_to_cpu(cp_hdr->header.ucode_size_bytes);
1143 		gfx_v11_0_rlc_backdoor_autoload_copy_ucode(adev, SOC21_FIRMWARE_ID_CP_PFP,
1144 						fw_data, fw_size, fw_autoload_mask);
1145 
1146 		/* me ucode */
1147 		cp_hdr = (const struct gfx_firmware_header_v1_0 *)
1148 			adev->gfx.me_fw->data;
1149 		fw_data = (const __le32 *)(adev->gfx.me_fw->data +
1150 				le32_to_cpu(cp_hdr->header.ucode_array_offset_bytes));
1151 		fw_size = le32_to_cpu(cp_hdr->header.ucode_size_bytes);
1152 		gfx_v11_0_rlc_backdoor_autoload_copy_ucode(adev, SOC21_FIRMWARE_ID_CP_ME,
1153 						fw_data, fw_size, fw_autoload_mask);
1154 
1155 		/* mec ucode */
1156 		cp_hdr = (const struct gfx_firmware_header_v1_0 *)
1157 			adev->gfx.mec_fw->data;
1158 		fw_data = (const __le32 *) (adev->gfx.mec_fw->data +
1159 				le32_to_cpu(cp_hdr->header.ucode_array_offset_bytes));
1160 		fw_size = le32_to_cpu(cp_hdr->header.ucode_size_bytes) -
1161 			cp_hdr->jt_size * 4;
1162 		gfx_v11_0_rlc_backdoor_autoload_copy_ucode(adev, SOC21_FIRMWARE_ID_CP_MEC,
1163 						fw_data, fw_size, fw_autoload_mask);
1164 	}
1165 
1166 	/* rlc ucode */
1167 	rlc_hdr = (const struct rlc_firmware_header_v2_0 *)
1168 		adev->gfx.rlc_fw->data;
1169 	fw_data = (const __le32 *)(adev->gfx.rlc_fw->data +
1170 			le32_to_cpu(rlc_hdr->header.ucode_array_offset_bytes));
1171 	fw_size = le32_to_cpu(rlc_hdr->header.ucode_size_bytes);
1172 	gfx_v11_0_rlc_backdoor_autoload_copy_ucode(adev, SOC21_FIRMWARE_ID_RLC_G_UCODE,
1173 					fw_data, fw_size, fw_autoload_mask);
1174 
1175 	version_major = le16_to_cpu(rlc_hdr->header.header_version_major);
1176 	version_minor = le16_to_cpu(rlc_hdr->header.header_version_minor);
1177 	if (version_major == 2) {
1178 		if (version_minor >= 2) {
1179 			rlcv22_hdr = (const struct rlc_firmware_header_v2_2 *)adev->gfx.rlc_fw->data;
1180 
1181 			fw_data = (const __le32 *)(adev->gfx.rlc_fw->data +
1182 					le32_to_cpu(rlcv22_hdr->rlc_iram_ucode_offset_bytes));
1183 			fw_size = le32_to_cpu(rlcv22_hdr->rlc_iram_ucode_size_bytes);
1184 			gfx_v11_0_rlc_backdoor_autoload_copy_ucode(adev, SOC21_FIRMWARE_ID_RLX6_UCODE,
1185 					fw_data, fw_size, fw_autoload_mask);
1186 
1187 			fw_data = (const __le32 *)(adev->gfx.rlc_fw->data +
1188 					le32_to_cpu(rlcv22_hdr->rlc_dram_ucode_offset_bytes));
1189 			fw_size = le32_to_cpu(rlcv22_hdr->rlc_dram_ucode_size_bytes);
1190 			gfx_v11_0_rlc_backdoor_autoload_copy_ucode(adev, SOC21_FIRMWARE_ID_RLX6_DRAM_BOOT,
1191 					fw_data, fw_size, fw_autoload_mask);
1192 		}
1193 	}
1194 }
1195 
1196 static void gfx_v11_0_rlc_backdoor_autoload_copy_sdma_ucode(struct amdgpu_device *adev,
1197 							uint32_t *fw_autoload_mask)
1198 {
1199 	const __le32 *fw_data;
1200 	uint32_t fw_size;
1201 	const struct sdma_firmware_header_v2_0 *sdma_hdr;
1202 
1203 	sdma_hdr = (const struct sdma_firmware_header_v2_0 *)
1204 		adev->sdma.instance[0].fw->data;
1205 	fw_data = (const __le32 *) (adev->sdma.instance[0].fw->data +
1206 			le32_to_cpu(sdma_hdr->header.ucode_array_offset_bytes));
1207 	fw_size = le32_to_cpu(sdma_hdr->ctx_ucode_size_bytes);
1208 
1209 	gfx_v11_0_rlc_backdoor_autoload_copy_ucode(adev,
1210 			SOC21_FIRMWARE_ID_SDMA_UCODE_TH0, fw_data, fw_size, fw_autoload_mask);
1211 
1212 	fw_data = (const __le32 *) (adev->sdma.instance[0].fw->data +
1213 			le32_to_cpu(sdma_hdr->ctl_ucode_offset));
1214 	fw_size = le32_to_cpu(sdma_hdr->ctl_ucode_size_bytes);
1215 
1216 	gfx_v11_0_rlc_backdoor_autoload_copy_ucode(adev,
1217 			SOC21_FIRMWARE_ID_SDMA_UCODE_TH1, fw_data, fw_size, fw_autoload_mask);
1218 }
1219 
1220 static void gfx_v11_0_rlc_backdoor_autoload_copy_mes_ucode(struct amdgpu_device *adev,
1221 							uint32_t *fw_autoload_mask)
1222 {
1223 	const __le32 *fw_data;
1224 	unsigned fw_size;
1225 	const struct mes_firmware_header_v1_0 *mes_hdr;
1226 	int pipe, ucode_id, data_id;
1227 
1228 	for (pipe = 0; pipe < 2; pipe++) {
1229 		if (pipe==0) {
1230 			ucode_id = SOC21_FIRMWARE_ID_RS64_MES_P0;
1231 			data_id  = SOC21_FIRMWARE_ID_RS64_MES_P0_STACK;
1232 		} else {
1233 			ucode_id = SOC21_FIRMWARE_ID_RS64_MES_P1;
1234 			data_id  = SOC21_FIRMWARE_ID_RS64_MES_P1_STACK;
1235 		}
1236 
1237 		mes_hdr = (const struct mes_firmware_header_v1_0 *)
1238 			adev->mes.fw[pipe]->data;
1239 
1240 		fw_data = (const __le32 *)(adev->mes.fw[pipe]->data +
1241 				le32_to_cpu(mes_hdr->mes_ucode_offset_bytes));
1242 		fw_size = le32_to_cpu(mes_hdr->mes_ucode_size_bytes);
1243 
1244 		gfx_v11_0_rlc_backdoor_autoload_copy_ucode(adev,
1245 				ucode_id, fw_data, fw_size, fw_autoload_mask);
1246 
1247 		fw_data = (const __le32 *)(adev->mes.fw[pipe]->data +
1248 				le32_to_cpu(mes_hdr->mes_ucode_data_offset_bytes));
1249 		fw_size = le32_to_cpu(mes_hdr->mes_ucode_data_size_bytes);
1250 
1251 		gfx_v11_0_rlc_backdoor_autoload_copy_ucode(adev,
1252 				data_id, fw_data, fw_size, fw_autoload_mask);
1253 	}
1254 }
1255 
1256 static int gfx_v11_0_rlc_backdoor_autoload_enable(struct amdgpu_device *adev)
1257 {
1258 	uint32_t rlc_g_offset, rlc_g_size;
1259 	uint64_t gpu_addr;
1260 	uint32_t autoload_fw_id[2];
1261 
1262 	memset(autoload_fw_id, 0, sizeof(uint32_t) * 2);
1263 
1264 	/* RLC autoload sequence 2: copy ucode */
1265 	gfx_v11_0_rlc_backdoor_autoload_copy_sdma_ucode(adev, autoload_fw_id);
1266 	gfx_v11_0_rlc_backdoor_autoload_copy_gfx_ucode(adev, autoload_fw_id);
1267 	gfx_v11_0_rlc_backdoor_autoload_copy_mes_ucode(adev, autoload_fw_id);
1268 	gfx_v11_0_rlc_backdoor_autoload_copy_toc_ucode(adev, autoload_fw_id);
1269 
1270 	rlc_g_offset = rlc_autoload_info[SOC21_FIRMWARE_ID_RLC_G_UCODE].offset;
1271 	rlc_g_size = rlc_autoload_info[SOC21_FIRMWARE_ID_RLC_G_UCODE].size;
1272 	gpu_addr = adev->gfx.rlc.rlc_autoload_gpu_addr + rlc_g_offset;
1273 
1274 	WREG32_SOC15(GC, 0, regGFX_IMU_RLC_BOOTLOADER_ADDR_HI, upper_32_bits(gpu_addr));
1275 	WREG32_SOC15(GC, 0, regGFX_IMU_RLC_BOOTLOADER_ADDR_LO, lower_32_bits(gpu_addr));
1276 
1277 	WREG32_SOC15(GC, 0, regGFX_IMU_RLC_BOOTLOADER_SIZE, rlc_g_size);
1278 
1279 	/* RLC autoload sequence 3: load IMU fw */
1280 	if (adev->gfx.imu.funcs->load_microcode)
1281 		adev->gfx.imu.funcs->load_microcode(adev);
1282 	/* RLC autoload sequence 4 init IMU fw */
1283 	if (adev->gfx.imu.funcs->setup_imu)
1284 		adev->gfx.imu.funcs->setup_imu(adev);
1285 	if (adev->gfx.imu.funcs->start_imu)
1286 		adev->gfx.imu.funcs->start_imu(adev);
1287 
1288 	/* RLC autoload sequence 5 disable gpa mode */
1289 	gfx_v11_0_disable_gpa_mode(adev);
1290 
1291 	return 0;
1292 }
1293 
1294 static int gfx_v11_0_sw_init(void *handle)
1295 {
1296 	int i, j, k, r, ring_id = 0;
1297 	struct amdgpu_kiq *kiq;
1298 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1299 
1300 	adev->gfxhub.funcs->init(adev);
1301 
1302 	switch (adev->ip_versions[GC_HWIP][0]) {
1303 	case IP_VERSION(11, 0, 0):
1304 	case IP_VERSION(11, 0, 2):
1305 	case IP_VERSION(11, 0, 3):
1306 		adev->gfx.me.num_me = 1;
1307 		adev->gfx.me.num_pipe_per_me = 1;
1308 		adev->gfx.me.num_queue_per_pipe = 1;
1309 		adev->gfx.mec.num_mec = 2;
1310 		adev->gfx.mec.num_pipe_per_mec = 4;
1311 		adev->gfx.mec.num_queue_per_pipe = 4;
1312 		break;
1313 	case IP_VERSION(11, 0, 1):
1314 	case IP_VERSION(11, 0, 4):
1315 		adev->gfx.me.num_me = 1;
1316 		adev->gfx.me.num_pipe_per_me = 1;
1317 		adev->gfx.me.num_queue_per_pipe = 1;
1318 		adev->gfx.mec.num_mec = 1;
1319 		adev->gfx.mec.num_pipe_per_mec = 4;
1320 		adev->gfx.mec.num_queue_per_pipe = 4;
1321 		break;
1322 	default:
1323 		adev->gfx.me.num_me = 1;
1324 		adev->gfx.me.num_pipe_per_me = 1;
1325 		adev->gfx.me.num_queue_per_pipe = 1;
1326 		adev->gfx.mec.num_mec = 1;
1327 		adev->gfx.mec.num_pipe_per_mec = 4;
1328 		adev->gfx.mec.num_queue_per_pipe = 8;
1329 		break;
1330 	}
1331 
1332 	/* Enable CG flag in one VF mode for enabling RLC safe mode enter/exit */
1333 	if (adev->ip_versions[GC_HWIP][0] == IP_VERSION(11, 0, 3) &&
1334 		amdgpu_sriov_is_pp_one_vf(adev))
1335 		adev->cg_flags = AMD_CG_SUPPORT_GFX_CGCG;
1336 
1337 	/* EOP Event */
1338 	r = amdgpu_irq_add_id(adev, SOC21_IH_CLIENTID_GRBM_CP,
1339 			      GFX_11_0_0__SRCID__CP_EOP_INTERRUPT,
1340 			      &adev->gfx.eop_irq);
1341 	if (r)
1342 		return r;
1343 
1344 	/* Privileged reg */
1345 	r = amdgpu_irq_add_id(adev, SOC21_IH_CLIENTID_GRBM_CP,
1346 			      GFX_11_0_0__SRCID__CP_PRIV_REG_FAULT,
1347 			      &adev->gfx.priv_reg_irq);
1348 	if (r)
1349 		return r;
1350 
1351 	/* Privileged inst */
1352 	r = amdgpu_irq_add_id(adev, SOC21_IH_CLIENTID_GRBM_CP,
1353 			      GFX_11_0_0__SRCID__CP_PRIV_INSTR_FAULT,
1354 			      &adev->gfx.priv_inst_irq);
1355 	if (r)
1356 		return r;
1357 
1358 	/* FED error */
1359 	r = amdgpu_irq_add_id(adev, SOC21_IH_CLIENTID_GFX,
1360 				  GFX_11_0_0__SRCID__RLC_GC_FED_INTERRUPT,
1361 				  &adev->gfx.rlc_gc_fed_irq);
1362 	if (r)
1363 		return r;
1364 
1365 	adev->gfx.gfx_current_status = AMDGPU_GFX_NORMAL_MODE;
1366 
1367 	if (adev->gfx.imu.funcs) {
1368 		if (adev->gfx.imu.funcs->init_microcode) {
1369 			r = adev->gfx.imu.funcs->init_microcode(adev);
1370 			if (r)
1371 				DRM_ERROR("Failed to load imu firmware!\n");
1372 		}
1373 	}
1374 
1375 	gfx_v11_0_me_init(adev);
1376 
1377 	r = gfx_v11_0_rlc_init(adev);
1378 	if (r) {
1379 		DRM_ERROR("Failed to init rlc BOs!\n");
1380 		return r;
1381 	}
1382 
1383 	r = gfx_v11_0_mec_init(adev);
1384 	if (r) {
1385 		DRM_ERROR("Failed to init MEC BOs!\n");
1386 		return r;
1387 	}
1388 
1389 	/* set up the gfx ring */
1390 	for (i = 0; i < adev->gfx.me.num_me; i++) {
1391 		for (j = 0; j < adev->gfx.me.num_queue_per_pipe; j++) {
1392 			for (k = 0; k < adev->gfx.me.num_pipe_per_me; k++) {
1393 				if (!amdgpu_gfx_is_me_queue_enabled(adev, i, k, j))
1394 					continue;
1395 
1396 				r = gfx_v11_0_gfx_ring_init(adev, ring_id,
1397 							    i, k, j);
1398 				if (r)
1399 					return r;
1400 				ring_id++;
1401 			}
1402 		}
1403 	}
1404 
1405 	ring_id = 0;
1406 	/* set up the compute queues - allocate horizontally across pipes */
1407 	for (i = 0; i < adev->gfx.mec.num_mec; ++i) {
1408 		for (j = 0; j < adev->gfx.mec.num_queue_per_pipe; j++) {
1409 			for (k = 0; k < adev->gfx.mec.num_pipe_per_mec; k++) {
1410 				if (!amdgpu_gfx_is_mec_queue_enabled(adev, 0, i,
1411 								     k, j))
1412 					continue;
1413 
1414 				r = gfx_v11_0_compute_ring_init(adev, ring_id,
1415 								i, k, j);
1416 				if (r)
1417 					return r;
1418 
1419 				ring_id++;
1420 			}
1421 		}
1422 	}
1423 
1424 	if (!adev->enable_mes_kiq) {
1425 		r = amdgpu_gfx_kiq_init(adev, GFX11_MEC_HPD_SIZE, 0);
1426 		if (r) {
1427 			DRM_ERROR("Failed to init KIQ BOs!\n");
1428 			return r;
1429 		}
1430 
1431 		kiq = &adev->gfx.kiq[0];
1432 		r = amdgpu_gfx_kiq_init_ring(adev, &kiq->ring, &kiq->irq, 0);
1433 		if (r)
1434 			return r;
1435 	}
1436 
1437 	r = amdgpu_gfx_mqd_sw_init(adev, sizeof(struct v11_compute_mqd), 0);
1438 	if (r)
1439 		return r;
1440 
1441 	/* allocate visible FB for rlc auto-loading fw */
1442 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_RLC_BACKDOOR_AUTO) {
1443 		r = gfx_v11_0_rlc_autoload_buffer_init(adev);
1444 		if (r)
1445 			return r;
1446 	}
1447 
1448 	r = gfx_v11_0_gpu_early_init(adev);
1449 	if (r)
1450 		return r;
1451 
1452 	if (amdgpu_gfx_ras_sw_init(adev)) {
1453 		dev_err(adev->dev, "Failed to initialize gfx ras block!\n");
1454 		return -EINVAL;
1455 	}
1456 
1457 	return 0;
1458 }
1459 
1460 static void gfx_v11_0_pfp_fini(struct amdgpu_device *adev)
1461 {
1462 	amdgpu_bo_free_kernel(&adev->gfx.pfp.pfp_fw_obj,
1463 			      &adev->gfx.pfp.pfp_fw_gpu_addr,
1464 			      (void **)&adev->gfx.pfp.pfp_fw_ptr);
1465 
1466 	amdgpu_bo_free_kernel(&adev->gfx.pfp.pfp_fw_data_obj,
1467 			      &adev->gfx.pfp.pfp_fw_data_gpu_addr,
1468 			      (void **)&adev->gfx.pfp.pfp_fw_data_ptr);
1469 }
1470 
1471 static void gfx_v11_0_me_fini(struct amdgpu_device *adev)
1472 {
1473 	amdgpu_bo_free_kernel(&adev->gfx.me.me_fw_obj,
1474 			      &adev->gfx.me.me_fw_gpu_addr,
1475 			      (void **)&adev->gfx.me.me_fw_ptr);
1476 
1477 	amdgpu_bo_free_kernel(&adev->gfx.me.me_fw_data_obj,
1478 			       &adev->gfx.me.me_fw_data_gpu_addr,
1479 			       (void **)&adev->gfx.me.me_fw_data_ptr);
1480 }
1481 
1482 static void gfx_v11_0_rlc_autoload_buffer_fini(struct amdgpu_device *adev)
1483 {
1484 	amdgpu_bo_free_kernel(&adev->gfx.rlc.rlc_autoload_bo,
1485 			&adev->gfx.rlc.rlc_autoload_gpu_addr,
1486 			(void **)&adev->gfx.rlc.rlc_autoload_ptr);
1487 }
1488 
1489 static int gfx_v11_0_sw_fini(void *handle)
1490 {
1491 	int i;
1492 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1493 
1494 	for (i = 0; i < adev->gfx.num_gfx_rings; i++)
1495 		amdgpu_ring_fini(&adev->gfx.gfx_ring[i]);
1496 	for (i = 0; i < adev->gfx.num_compute_rings; i++)
1497 		amdgpu_ring_fini(&adev->gfx.compute_ring[i]);
1498 
1499 	amdgpu_gfx_mqd_sw_fini(adev, 0);
1500 
1501 	if (!adev->enable_mes_kiq) {
1502 		amdgpu_gfx_kiq_free_ring(&adev->gfx.kiq[0].ring);
1503 		amdgpu_gfx_kiq_fini(adev, 0);
1504 	}
1505 
1506 	gfx_v11_0_pfp_fini(adev);
1507 	gfx_v11_0_me_fini(adev);
1508 	gfx_v11_0_rlc_fini(adev);
1509 	gfx_v11_0_mec_fini(adev);
1510 
1511 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_RLC_BACKDOOR_AUTO)
1512 		gfx_v11_0_rlc_autoload_buffer_fini(adev);
1513 
1514 	gfx_v11_0_free_microcode(adev);
1515 
1516 	return 0;
1517 }
1518 
1519 static void gfx_v11_0_select_se_sh(struct amdgpu_device *adev, u32 se_num,
1520 				   u32 sh_num, u32 instance, int xcc_id)
1521 {
1522 	u32 data;
1523 
1524 	if (instance == 0xffffffff)
1525 		data = REG_SET_FIELD(0, GRBM_GFX_INDEX,
1526 				     INSTANCE_BROADCAST_WRITES, 1);
1527 	else
1528 		data = REG_SET_FIELD(0, GRBM_GFX_INDEX, INSTANCE_INDEX,
1529 				     instance);
1530 
1531 	if (se_num == 0xffffffff)
1532 		data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SE_BROADCAST_WRITES,
1533 				     1);
1534 	else
1535 		data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SE_INDEX, se_num);
1536 
1537 	if (sh_num == 0xffffffff)
1538 		data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SA_BROADCAST_WRITES,
1539 				     1);
1540 	else
1541 		data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SA_INDEX, sh_num);
1542 
1543 	WREG32_SOC15(GC, 0, regGRBM_GFX_INDEX, data);
1544 }
1545 
1546 static u32 gfx_v11_0_get_sa_active_bitmap(struct amdgpu_device *adev)
1547 {
1548 	u32 gc_disabled_sa_mask, gc_user_disabled_sa_mask, sa_mask;
1549 
1550 	gc_disabled_sa_mask = RREG32_SOC15(GC, 0, regCC_GC_SA_UNIT_DISABLE);
1551 	gc_disabled_sa_mask = REG_GET_FIELD(gc_disabled_sa_mask,
1552 					   CC_GC_SA_UNIT_DISABLE,
1553 					   SA_DISABLE);
1554 	gc_user_disabled_sa_mask = RREG32_SOC15(GC, 0, regGC_USER_SA_UNIT_DISABLE);
1555 	gc_user_disabled_sa_mask = REG_GET_FIELD(gc_user_disabled_sa_mask,
1556 						 GC_USER_SA_UNIT_DISABLE,
1557 						 SA_DISABLE);
1558 	sa_mask = amdgpu_gfx_create_bitmask(adev->gfx.config.max_sh_per_se *
1559 					    adev->gfx.config.max_shader_engines);
1560 
1561 	return sa_mask & (~(gc_disabled_sa_mask | gc_user_disabled_sa_mask));
1562 }
1563 
1564 static u32 gfx_v11_0_get_rb_active_bitmap(struct amdgpu_device *adev)
1565 {
1566 	u32 gc_disabled_rb_mask, gc_user_disabled_rb_mask;
1567 	u32 rb_mask;
1568 
1569 	gc_disabled_rb_mask = RREG32_SOC15(GC, 0, regCC_RB_BACKEND_DISABLE);
1570 	gc_disabled_rb_mask = REG_GET_FIELD(gc_disabled_rb_mask,
1571 					    CC_RB_BACKEND_DISABLE,
1572 					    BACKEND_DISABLE);
1573 	gc_user_disabled_rb_mask = RREG32_SOC15(GC, 0, regGC_USER_RB_BACKEND_DISABLE);
1574 	gc_user_disabled_rb_mask = REG_GET_FIELD(gc_user_disabled_rb_mask,
1575 						 GC_USER_RB_BACKEND_DISABLE,
1576 						 BACKEND_DISABLE);
1577 	rb_mask = amdgpu_gfx_create_bitmask(adev->gfx.config.max_backends_per_se *
1578 					    adev->gfx.config.max_shader_engines);
1579 
1580 	return rb_mask & (~(gc_disabled_rb_mask | gc_user_disabled_rb_mask));
1581 }
1582 
1583 static void gfx_v11_0_setup_rb(struct amdgpu_device *adev)
1584 {
1585 	u32 rb_bitmap_width_per_sa;
1586 	u32 max_sa;
1587 	u32 active_sa_bitmap;
1588 	u32 global_active_rb_bitmap;
1589 	u32 active_rb_bitmap = 0;
1590 	u32 i;
1591 
1592 	/* query sa bitmap from SA_UNIT_DISABLE registers */
1593 	active_sa_bitmap = gfx_v11_0_get_sa_active_bitmap(adev);
1594 	/* query rb bitmap from RB_BACKEND_DISABLE registers */
1595 	global_active_rb_bitmap = gfx_v11_0_get_rb_active_bitmap(adev);
1596 
1597 	/* generate active rb bitmap according to active sa bitmap */
1598 	max_sa = adev->gfx.config.max_shader_engines *
1599 		 adev->gfx.config.max_sh_per_se;
1600 	rb_bitmap_width_per_sa = adev->gfx.config.max_backends_per_se /
1601 				 adev->gfx.config.max_sh_per_se;
1602 	for (i = 0; i < max_sa; i++) {
1603 		if (active_sa_bitmap & (1 << i))
1604 			active_rb_bitmap |= (0x3 << (i * rb_bitmap_width_per_sa));
1605 	}
1606 
1607 	active_rb_bitmap |= global_active_rb_bitmap;
1608 	adev->gfx.config.backend_enable_mask = active_rb_bitmap;
1609 	adev->gfx.config.num_rbs = hweight32(active_rb_bitmap);
1610 }
1611 
1612 #define DEFAULT_SH_MEM_BASES	(0x6000)
1613 #define LDS_APP_BASE           0x1
1614 #define SCRATCH_APP_BASE       0x2
1615 
1616 static void gfx_v11_0_init_compute_vmid(struct amdgpu_device *adev)
1617 {
1618 	int i;
1619 	uint32_t sh_mem_bases;
1620 	uint32_t data;
1621 
1622 	/*
1623 	 * Configure apertures:
1624 	 * LDS:         0x60000000'00000000 - 0x60000001'00000000 (4GB)
1625 	 * Scratch:     0x60000001'00000000 - 0x60000002'00000000 (4GB)
1626 	 * GPUVM:       0x60010000'00000000 - 0x60020000'00000000 (1TB)
1627 	 */
1628 	sh_mem_bases = (LDS_APP_BASE << SH_MEM_BASES__SHARED_BASE__SHIFT) |
1629 			SCRATCH_APP_BASE;
1630 
1631 	mutex_lock(&adev->srbm_mutex);
1632 	for (i = adev->vm_manager.first_kfd_vmid; i < AMDGPU_NUM_VMID; i++) {
1633 		soc21_grbm_select(adev, 0, 0, 0, i);
1634 		/* CP and shaders */
1635 		WREG32_SOC15(GC, 0, regSH_MEM_CONFIG, DEFAULT_SH_MEM_CONFIG);
1636 		WREG32_SOC15(GC, 0, regSH_MEM_BASES, sh_mem_bases);
1637 
1638 		/* Enable trap for each kfd vmid. */
1639 		data = RREG32_SOC15(GC, 0, regSPI_GDBG_PER_VMID_CNTL);
1640 		data = REG_SET_FIELD(data, SPI_GDBG_PER_VMID_CNTL, TRAP_EN, 1);
1641 		WREG32_SOC15(GC, 0, regSPI_GDBG_PER_VMID_CNTL, data);
1642 	}
1643 	soc21_grbm_select(adev, 0, 0, 0, 0);
1644 	mutex_unlock(&adev->srbm_mutex);
1645 
1646 	/* Initialize all compute VMIDs to have no GDS, GWS, or OA
1647 	   acccess. These should be enabled by FW for target VMIDs. */
1648 	for (i = adev->vm_manager.first_kfd_vmid; i < AMDGPU_NUM_VMID; i++) {
1649 		WREG32_SOC15_OFFSET(GC, 0, regGDS_VMID0_BASE, 2 * i, 0);
1650 		WREG32_SOC15_OFFSET(GC, 0, regGDS_VMID0_SIZE, 2 * i, 0);
1651 		WREG32_SOC15_OFFSET(GC, 0, regGDS_GWS_VMID0, i, 0);
1652 		WREG32_SOC15_OFFSET(GC, 0, regGDS_OA_VMID0, i, 0);
1653 	}
1654 }
1655 
1656 static void gfx_v11_0_init_gds_vmid(struct amdgpu_device *adev)
1657 {
1658 	int vmid;
1659 
1660 	/*
1661 	 * Initialize all compute and user-gfx VMIDs to have no GDS, GWS, or OA
1662 	 * access. Compute VMIDs should be enabled by FW for target VMIDs,
1663 	 * the driver can enable them for graphics. VMID0 should maintain
1664 	 * access so that HWS firmware can save/restore entries.
1665 	 */
1666 	for (vmid = 1; vmid < 16; vmid++) {
1667 		WREG32_SOC15_OFFSET(GC, 0, regGDS_VMID0_BASE, 2 * vmid, 0);
1668 		WREG32_SOC15_OFFSET(GC, 0, regGDS_VMID0_SIZE, 2 * vmid, 0);
1669 		WREG32_SOC15_OFFSET(GC, 0, regGDS_GWS_VMID0, vmid, 0);
1670 		WREG32_SOC15_OFFSET(GC, 0, regGDS_OA_VMID0, vmid, 0);
1671 	}
1672 }
1673 
1674 static void gfx_v11_0_tcp_harvest(struct amdgpu_device *adev)
1675 {
1676 	/* TODO: harvest feature to be added later. */
1677 }
1678 
1679 static void gfx_v11_0_get_tcc_info(struct amdgpu_device *adev)
1680 {
1681 	/* TCCs are global (not instanced). */
1682 	uint32_t tcc_disable = RREG32_SOC15(GC, 0, regCGTS_TCC_DISABLE) |
1683 			       RREG32_SOC15(GC, 0, regCGTS_USER_TCC_DISABLE);
1684 
1685 	adev->gfx.config.tcc_disabled_mask =
1686 		REG_GET_FIELD(tcc_disable, CGTS_TCC_DISABLE, TCC_DISABLE) |
1687 		(REG_GET_FIELD(tcc_disable, CGTS_TCC_DISABLE, HI_TCC_DISABLE) << 16);
1688 }
1689 
1690 static void gfx_v11_0_constants_init(struct amdgpu_device *adev)
1691 {
1692 	u32 tmp;
1693 	int i;
1694 
1695 	if (!amdgpu_sriov_vf(adev))
1696 		WREG32_FIELD15_PREREG(GC, 0, GRBM_CNTL, READ_TIMEOUT, 0xff);
1697 
1698 	gfx_v11_0_setup_rb(adev);
1699 	gfx_v11_0_get_cu_info(adev, &adev->gfx.cu_info);
1700 	gfx_v11_0_get_tcc_info(adev);
1701 	adev->gfx.config.pa_sc_tile_steering_override = 0;
1702 
1703 	/* Set whether texture coordinate truncation is conformant. */
1704 	tmp = RREG32_SOC15(GC, 0, regTA_CNTL2);
1705 	adev->gfx.config.ta_cntl2_truncate_coord_mode =
1706 		REG_GET_FIELD(tmp, TA_CNTL2, TRUNCATE_COORD_MODE);
1707 
1708 	/* XXX SH_MEM regs */
1709 	/* where to put LDS, scratch, GPUVM in FSA64 space */
1710 	mutex_lock(&adev->srbm_mutex);
1711 	for (i = 0; i < adev->vm_manager.id_mgr[AMDGPU_GFXHUB(0)].num_ids; i++) {
1712 		soc21_grbm_select(adev, 0, 0, 0, i);
1713 		/* CP and shaders */
1714 		WREG32_SOC15(GC, 0, regSH_MEM_CONFIG, DEFAULT_SH_MEM_CONFIG);
1715 		if (i != 0) {
1716 			tmp = REG_SET_FIELD(0, SH_MEM_BASES, PRIVATE_BASE,
1717 				(adev->gmc.private_aperture_start >> 48));
1718 			tmp = REG_SET_FIELD(tmp, SH_MEM_BASES, SHARED_BASE,
1719 				(adev->gmc.shared_aperture_start >> 48));
1720 			WREG32_SOC15(GC, 0, regSH_MEM_BASES, tmp);
1721 		}
1722 	}
1723 	soc21_grbm_select(adev, 0, 0, 0, 0);
1724 
1725 	mutex_unlock(&adev->srbm_mutex);
1726 
1727 	gfx_v11_0_init_compute_vmid(adev);
1728 	gfx_v11_0_init_gds_vmid(adev);
1729 }
1730 
1731 static void gfx_v11_0_enable_gui_idle_interrupt(struct amdgpu_device *adev,
1732 					       bool enable)
1733 {
1734 	u32 tmp;
1735 
1736 	if (amdgpu_sriov_vf(adev))
1737 		return;
1738 
1739 	tmp = RREG32_SOC15(GC, 0, regCP_INT_CNTL_RING0);
1740 
1741 	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CNTX_BUSY_INT_ENABLE,
1742 			    enable ? 1 : 0);
1743 	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CNTX_EMPTY_INT_ENABLE,
1744 			    enable ? 1 : 0);
1745 	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CMP_BUSY_INT_ENABLE,
1746 			    enable ? 1 : 0);
1747 	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, GFX_IDLE_INT_ENABLE,
1748 			    enable ? 1 : 0);
1749 
1750 	WREG32_SOC15(GC, 0, regCP_INT_CNTL_RING0, tmp);
1751 }
1752 
1753 static int gfx_v11_0_init_csb(struct amdgpu_device *adev)
1754 {
1755 	adev->gfx.rlc.funcs->get_csb_buffer(adev, adev->gfx.rlc.cs_ptr);
1756 
1757 	WREG32_SOC15(GC, 0, regRLC_CSIB_ADDR_HI,
1758 			adev->gfx.rlc.clear_state_gpu_addr >> 32);
1759 	WREG32_SOC15(GC, 0, regRLC_CSIB_ADDR_LO,
1760 			adev->gfx.rlc.clear_state_gpu_addr & 0xfffffffc);
1761 	WREG32_SOC15(GC, 0, regRLC_CSIB_LENGTH, adev->gfx.rlc.clear_state_size);
1762 
1763 	return 0;
1764 }
1765 
1766 static void gfx_v11_0_rlc_stop(struct amdgpu_device *adev)
1767 {
1768 	u32 tmp = RREG32_SOC15(GC, 0, regRLC_CNTL);
1769 
1770 	tmp = REG_SET_FIELD(tmp, RLC_CNTL, RLC_ENABLE_F32, 0);
1771 	WREG32_SOC15(GC, 0, regRLC_CNTL, tmp);
1772 }
1773 
1774 static void gfx_v11_0_rlc_reset(struct amdgpu_device *adev)
1775 {
1776 	WREG32_FIELD15_PREREG(GC, 0, GRBM_SOFT_RESET, SOFT_RESET_RLC, 1);
1777 	udelay(50);
1778 	WREG32_FIELD15_PREREG(GC, 0, GRBM_SOFT_RESET, SOFT_RESET_RLC, 0);
1779 	udelay(50);
1780 }
1781 
1782 static void gfx_v11_0_rlc_smu_handshake_cntl(struct amdgpu_device *adev,
1783 					     bool enable)
1784 {
1785 	uint32_t rlc_pg_cntl;
1786 
1787 	rlc_pg_cntl = RREG32_SOC15(GC, 0, regRLC_PG_CNTL);
1788 
1789 	if (!enable) {
1790 		/* RLC_PG_CNTL[23] = 0 (default)
1791 		 * RLC will wait for handshake acks with SMU
1792 		 * GFXOFF will be enabled
1793 		 * RLC_PG_CNTL[23] = 1
1794 		 * RLC will not issue any message to SMU
1795 		 * hence no handshake between SMU & RLC
1796 		 * GFXOFF will be disabled
1797 		 */
1798 		rlc_pg_cntl |= RLC_PG_CNTL__SMU_HANDSHAKE_DISABLE_MASK;
1799 	} else
1800 		rlc_pg_cntl &= ~RLC_PG_CNTL__SMU_HANDSHAKE_DISABLE_MASK;
1801 	WREG32_SOC15(GC, 0, regRLC_PG_CNTL, rlc_pg_cntl);
1802 }
1803 
1804 static void gfx_v11_0_rlc_start(struct amdgpu_device *adev)
1805 {
1806 	/* TODO: enable rlc & smu handshake until smu
1807 	 * and gfxoff feature works as expected */
1808 	if (!(amdgpu_pp_feature_mask & PP_GFXOFF_MASK))
1809 		gfx_v11_0_rlc_smu_handshake_cntl(adev, false);
1810 
1811 	WREG32_FIELD15_PREREG(GC, 0, RLC_CNTL, RLC_ENABLE_F32, 1);
1812 	udelay(50);
1813 }
1814 
1815 static void gfx_v11_0_rlc_enable_srm(struct amdgpu_device *adev)
1816 {
1817 	uint32_t tmp;
1818 
1819 	/* enable Save Restore Machine */
1820 	tmp = RREG32(SOC15_REG_OFFSET(GC, 0, regRLC_SRM_CNTL));
1821 	tmp |= RLC_SRM_CNTL__AUTO_INCR_ADDR_MASK;
1822 	tmp |= RLC_SRM_CNTL__SRM_ENABLE_MASK;
1823 	WREG32(SOC15_REG_OFFSET(GC, 0, regRLC_SRM_CNTL), tmp);
1824 }
1825 
1826 static void gfx_v11_0_load_rlcg_microcode(struct amdgpu_device *adev)
1827 {
1828 	const struct rlc_firmware_header_v2_0 *hdr;
1829 	const __le32 *fw_data;
1830 	unsigned i, fw_size;
1831 
1832 	hdr = (const struct rlc_firmware_header_v2_0 *)adev->gfx.rlc_fw->data;
1833 	fw_data = (const __le32 *)(adev->gfx.rlc_fw->data +
1834 			   le32_to_cpu(hdr->header.ucode_array_offset_bytes));
1835 	fw_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4;
1836 
1837 	WREG32_SOC15(GC, 0, regRLC_GPM_UCODE_ADDR,
1838 		     RLCG_UCODE_LOADING_START_ADDRESS);
1839 
1840 	for (i = 0; i < fw_size; i++)
1841 		WREG32_SOC15(GC, 0, regRLC_GPM_UCODE_DATA,
1842 			     le32_to_cpup(fw_data++));
1843 
1844 	WREG32_SOC15(GC, 0, regRLC_GPM_UCODE_ADDR, adev->gfx.rlc_fw_version);
1845 }
1846 
1847 static void gfx_v11_0_load_rlc_iram_dram_microcode(struct amdgpu_device *adev)
1848 {
1849 	const struct rlc_firmware_header_v2_2 *hdr;
1850 	const __le32 *fw_data;
1851 	unsigned i, fw_size;
1852 	u32 tmp;
1853 
1854 	hdr = (const struct rlc_firmware_header_v2_2 *)adev->gfx.rlc_fw->data;
1855 
1856 	fw_data = (const __le32 *)(adev->gfx.rlc_fw->data +
1857 			le32_to_cpu(hdr->rlc_iram_ucode_offset_bytes));
1858 	fw_size = le32_to_cpu(hdr->rlc_iram_ucode_size_bytes) / 4;
1859 
1860 	WREG32_SOC15(GC, 0, regRLC_LX6_IRAM_ADDR, 0);
1861 
1862 	for (i = 0; i < fw_size; i++) {
1863 		if ((amdgpu_emu_mode == 1) && (i % 100 == 99))
1864 			msleep(1);
1865 		WREG32_SOC15(GC, 0, regRLC_LX6_IRAM_DATA,
1866 				le32_to_cpup(fw_data++));
1867 	}
1868 
1869 	WREG32_SOC15(GC, 0, regRLC_LX6_IRAM_ADDR, adev->gfx.rlc_fw_version);
1870 
1871 	fw_data = (const __le32 *)(adev->gfx.rlc_fw->data +
1872 			le32_to_cpu(hdr->rlc_dram_ucode_offset_bytes));
1873 	fw_size = le32_to_cpu(hdr->rlc_dram_ucode_size_bytes) / 4;
1874 
1875 	WREG32_SOC15(GC, 0, regRLC_LX6_DRAM_ADDR, 0);
1876 	for (i = 0; i < fw_size; i++) {
1877 		if ((amdgpu_emu_mode == 1) && (i % 100 == 99))
1878 			msleep(1);
1879 		WREG32_SOC15(GC, 0, regRLC_LX6_DRAM_DATA,
1880 				le32_to_cpup(fw_data++));
1881 	}
1882 
1883 	WREG32_SOC15(GC, 0, regRLC_LX6_IRAM_ADDR, adev->gfx.rlc_fw_version);
1884 
1885 	tmp = RREG32_SOC15(GC, 0, regRLC_LX6_CNTL);
1886 	tmp = REG_SET_FIELD(tmp, RLC_LX6_CNTL, PDEBUG_ENABLE, 1);
1887 	tmp = REG_SET_FIELD(tmp, RLC_LX6_CNTL, BRESET, 0);
1888 	WREG32_SOC15(GC, 0, regRLC_LX6_CNTL, tmp);
1889 }
1890 
1891 static void gfx_v11_0_load_rlcp_rlcv_microcode(struct amdgpu_device *adev)
1892 {
1893 	const struct rlc_firmware_header_v2_3 *hdr;
1894 	const __le32 *fw_data;
1895 	unsigned i, fw_size;
1896 	u32 tmp;
1897 
1898 	hdr = (const struct rlc_firmware_header_v2_3 *)adev->gfx.rlc_fw->data;
1899 
1900 	fw_data = (const __le32 *)(adev->gfx.rlc_fw->data +
1901 			le32_to_cpu(hdr->rlcp_ucode_offset_bytes));
1902 	fw_size = le32_to_cpu(hdr->rlcp_ucode_size_bytes) / 4;
1903 
1904 	WREG32_SOC15(GC, 0, regRLC_PACE_UCODE_ADDR, 0);
1905 
1906 	for (i = 0; i < fw_size; i++) {
1907 		if ((amdgpu_emu_mode == 1) && (i % 100 == 99))
1908 			msleep(1);
1909 		WREG32_SOC15(GC, 0, regRLC_PACE_UCODE_DATA,
1910 				le32_to_cpup(fw_data++));
1911 	}
1912 
1913 	WREG32_SOC15(GC, 0, regRLC_PACE_UCODE_ADDR, adev->gfx.rlc_fw_version);
1914 
1915 	tmp = RREG32_SOC15(GC, 0, regRLC_GPM_THREAD_ENABLE);
1916 	tmp = REG_SET_FIELD(tmp, RLC_GPM_THREAD_ENABLE, THREAD1_ENABLE, 1);
1917 	WREG32_SOC15(GC, 0, regRLC_GPM_THREAD_ENABLE, tmp);
1918 
1919 	fw_data = (const __le32 *)(adev->gfx.rlc_fw->data +
1920 			le32_to_cpu(hdr->rlcv_ucode_offset_bytes));
1921 	fw_size = le32_to_cpu(hdr->rlcv_ucode_size_bytes) / 4;
1922 
1923 	WREG32_SOC15(GC, 0, regRLC_GPU_IOV_UCODE_ADDR, 0);
1924 
1925 	for (i = 0; i < fw_size; i++) {
1926 		if ((amdgpu_emu_mode == 1) && (i % 100 == 99))
1927 			msleep(1);
1928 		WREG32_SOC15(GC, 0, regRLC_GPU_IOV_UCODE_DATA,
1929 				le32_to_cpup(fw_data++));
1930 	}
1931 
1932 	WREG32_SOC15(GC, 0, regRLC_GPU_IOV_UCODE_ADDR, adev->gfx.rlc_fw_version);
1933 
1934 	tmp = RREG32_SOC15(GC, 0, regRLC_GPU_IOV_F32_CNTL);
1935 	tmp = REG_SET_FIELD(tmp, RLC_GPU_IOV_F32_CNTL, ENABLE, 1);
1936 	WREG32_SOC15(GC, 0, regRLC_GPU_IOV_F32_CNTL, tmp);
1937 }
1938 
1939 static int gfx_v11_0_rlc_load_microcode(struct amdgpu_device *adev)
1940 {
1941 	const struct rlc_firmware_header_v2_0 *hdr;
1942 	uint16_t version_major;
1943 	uint16_t version_minor;
1944 
1945 	if (!adev->gfx.rlc_fw)
1946 		return -EINVAL;
1947 
1948 	hdr = (const struct rlc_firmware_header_v2_0 *)adev->gfx.rlc_fw->data;
1949 	amdgpu_ucode_print_rlc_hdr(&hdr->header);
1950 
1951 	version_major = le16_to_cpu(hdr->header.header_version_major);
1952 	version_minor = le16_to_cpu(hdr->header.header_version_minor);
1953 
1954 	if (version_major == 2) {
1955 		gfx_v11_0_load_rlcg_microcode(adev);
1956 		if (amdgpu_dpm == 1) {
1957 			if (version_minor >= 2)
1958 				gfx_v11_0_load_rlc_iram_dram_microcode(adev);
1959 			if (version_minor == 3)
1960 				gfx_v11_0_load_rlcp_rlcv_microcode(adev);
1961 		}
1962 
1963 		return 0;
1964 	}
1965 
1966 	return -EINVAL;
1967 }
1968 
1969 static int gfx_v11_0_rlc_resume(struct amdgpu_device *adev)
1970 {
1971 	int r;
1972 
1973 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
1974 		gfx_v11_0_init_csb(adev);
1975 
1976 		if (!amdgpu_sriov_vf(adev)) /* enable RLC SRM */
1977 			gfx_v11_0_rlc_enable_srm(adev);
1978 	} else {
1979 		if (amdgpu_sriov_vf(adev)) {
1980 			gfx_v11_0_init_csb(adev);
1981 			return 0;
1982 		}
1983 
1984 		adev->gfx.rlc.funcs->stop(adev);
1985 
1986 		/* disable CG */
1987 		WREG32_SOC15(GC, 0, regRLC_CGCG_CGLS_CTRL, 0);
1988 
1989 		/* disable PG */
1990 		WREG32_SOC15(GC, 0, regRLC_PG_CNTL, 0);
1991 
1992 		if (adev->firmware.load_type == AMDGPU_FW_LOAD_DIRECT) {
1993 			/* legacy rlc firmware loading */
1994 			r = gfx_v11_0_rlc_load_microcode(adev);
1995 			if (r)
1996 				return r;
1997 		}
1998 
1999 		gfx_v11_0_init_csb(adev);
2000 
2001 		adev->gfx.rlc.funcs->start(adev);
2002 	}
2003 	return 0;
2004 }
2005 
2006 static int gfx_v11_0_config_me_cache(struct amdgpu_device *adev, uint64_t addr)
2007 {
2008 	uint32_t usec_timeout = 50000;  /* wait for 50ms */
2009 	uint32_t tmp;
2010 	int i;
2011 
2012 	/* Trigger an invalidation of the L1 instruction caches */
2013 	tmp = RREG32_SOC15(GC, 0, regCP_ME_IC_OP_CNTL);
2014 	tmp = REG_SET_FIELD(tmp, CP_ME_IC_OP_CNTL, INVALIDATE_CACHE, 1);
2015 	WREG32_SOC15(GC, 0, regCP_ME_IC_OP_CNTL, tmp);
2016 
2017 	/* Wait for invalidation complete */
2018 	for (i = 0; i < usec_timeout; i++) {
2019 		tmp = RREG32_SOC15(GC, 0, regCP_ME_IC_OP_CNTL);
2020 		if (1 == REG_GET_FIELD(tmp, CP_ME_IC_OP_CNTL,
2021 					INVALIDATE_CACHE_COMPLETE))
2022 			break;
2023 		udelay(1);
2024 	}
2025 
2026 	if (i >= usec_timeout) {
2027 		dev_err(adev->dev, "failed to invalidate instruction cache\n");
2028 		return -EINVAL;
2029 	}
2030 
2031 	if (amdgpu_emu_mode == 1)
2032 		adev->hdp.funcs->flush_hdp(adev, NULL);
2033 
2034 	tmp = RREG32_SOC15(GC, 0, regCP_ME_IC_BASE_CNTL);
2035 	tmp = REG_SET_FIELD(tmp, CP_ME_IC_BASE_CNTL, VMID, 0);
2036 	tmp = REG_SET_FIELD(tmp, CP_ME_IC_BASE_CNTL, CACHE_POLICY, 0);
2037 	tmp = REG_SET_FIELD(tmp, CP_ME_IC_BASE_CNTL, EXE_DISABLE, 0);
2038 	tmp = REG_SET_FIELD(tmp, CP_ME_IC_BASE_CNTL, ADDRESS_CLAMP, 1);
2039 	WREG32_SOC15(GC, 0, regCP_ME_IC_BASE_CNTL, tmp);
2040 
2041 	/* Program me ucode address into intruction cache address register */
2042 	WREG32_SOC15(GC, 0, regCP_ME_IC_BASE_LO,
2043 			lower_32_bits(addr) & 0xFFFFF000);
2044 	WREG32_SOC15(GC, 0, regCP_ME_IC_BASE_HI,
2045 			upper_32_bits(addr));
2046 
2047 	return 0;
2048 }
2049 
2050 static int gfx_v11_0_config_pfp_cache(struct amdgpu_device *adev, uint64_t addr)
2051 {
2052 	uint32_t usec_timeout = 50000;  /* wait for 50ms */
2053 	uint32_t tmp;
2054 	int i;
2055 
2056 	/* Trigger an invalidation of the L1 instruction caches */
2057 	tmp = RREG32_SOC15(GC, 0, regCP_PFP_IC_OP_CNTL);
2058 	tmp = REG_SET_FIELD(tmp, CP_PFP_IC_OP_CNTL, INVALIDATE_CACHE, 1);
2059 	WREG32_SOC15(GC, 0, regCP_PFP_IC_OP_CNTL, tmp);
2060 
2061 	/* Wait for invalidation complete */
2062 	for (i = 0; i < usec_timeout; i++) {
2063 		tmp = RREG32_SOC15(GC, 0, regCP_PFP_IC_OP_CNTL);
2064 		if (1 == REG_GET_FIELD(tmp, CP_PFP_IC_OP_CNTL,
2065 					INVALIDATE_CACHE_COMPLETE))
2066 			break;
2067 		udelay(1);
2068 	}
2069 
2070 	if (i >= usec_timeout) {
2071 		dev_err(adev->dev, "failed to invalidate instruction cache\n");
2072 		return -EINVAL;
2073 	}
2074 
2075 	if (amdgpu_emu_mode == 1)
2076 		adev->hdp.funcs->flush_hdp(adev, NULL);
2077 
2078 	tmp = RREG32_SOC15(GC, 0, regCP_PFP_IC_BASE_CNTL);
2079 	tmp = REG_SET_FIELD(tmp, CP_PFP_IC_BASE_CNTL, VMID, 0);
2080 	tmp = REG_SET_FIELD(tmp, CP_PFP_IC_BASE_CNTL, CACHE_POLICY, 0);
2081 	tmp = REG_SET_FIELD(tmp, CP_PFP_IC_BASE_CNTL, EXE_DISABLE, 0);
2082 	tmp = REG_SET_FIELD(tmp, CP_PFP_IC_BASE_CNTL, ADDRESS_CLAMP, 1);
2083 	WREG32_SOC15(GC, 0, regCP_PFP_IC_BASE_CNTL, tmp);
2084 
2085 	/* Program pfp ucode address into intruction cache address register */
2086 	WREG32_SOC15(GC, 0, regCP_PFP_IC_BASE_LO,
2087 			lower_32_bits(addr) & 0xFFFFF000);
2088 	WREG32_SOC15(GC, 0, regCP_PFP_IC_BASE_HI,
2089 			upper_32_bits(addr));
2090 
2091 	return 0;
2092 }
2093 
2094 static int gfx_v11_0_config_mec_cache(struct amdgpu_device *adev, uint64_t addr)
2095 {
2096 	uint32_t usec_timeout = 50000;  /* wait for 50ms */
2097 	uint32_t tmp;
2098 	int i;
2099 
2100 	/* Trigger an invalidation of the L1 instruction caches */
2101 	tmp = RREG32_SOC15(GC, 0, regCP_CPC_IC_OP_CNTL);
2102 	tmp = REG_SET_FIELD(tmp, CP_CPC_IC_OP_CNTL, INVALIDATE_CACHE, 1);
2103 
2104 	WREG32_SOC15(GC, 0, regCP_CPC_IC_OP_CNTL, tmp);
2105 
2106 	/* Wait for invalidation complete */
2107 	for (i = 0; i < usec_timeout; i++) {
2108 		tmp = RREG32_SOC15(GC, 0, regCP_CPC_IC_OP_CNTL);
2109 		if (1 == REG_GET_FIELD(tmp, CP_CPC_IC_OP_CNTL,
2110 					INVALIDATE_CACHE_COMPLETE))
2111 			break;
2112 		udelay(1);
2113 	}
2114 
2115 	if (i >= usec_timeout) {
2116 		dev_err(adev->dev, "failed to invalidate instruction cache\n");
2117 		return -EINVAL;
2118 	}
2119 
2120 	if (amdgpu_emu_mode == 1)
2121 		adev->hdp.funcs->flush_hdp(adev, NULL);
2122 
2123 	tmp = RREG32_SOC15(GC, 0, regCP_CPC_IC_BASE_CNTL);
2124 	tmp = REG_SET_FIELD(tmp, CP_CPC_IC_BASE_CNTL, CACHE_POLICY, 0);
2125 	tmp = REG_SET_FIELD(tmp, CP_CPC_IC_BASE_CNTL, EXE_DISABLE, 0);
2126 	tmp = REG_SET_FIELD(tmp, CP_CPC_IC_BASE_CNTL, ADDRESS_CLAMP, 1);
2127 	WREG32_SOC15(GC, 0, regCP_CPC_IC_BASE_CNTL, tmp);
2128 
2129 	/* Program mec1 ucode address into intruction cache address register */
2130 	WREG32_SOC15(GC, 0, regCP_CPC_IC_BASE_LO,
2131 			lower_32_bits(addr) & 0xFFFFF000);
2132 	WREG32_SOC15(GC, 0, regCP_CPC_IC_BASE_HI,
2133 			upper_32_bits(addr));
2134 
2135 	return 0;
2136 }
2137 
2138 static int gfx_v11_0_config_pfp_cache_rs64(struct amdgpu_device *adev, uint64_t addr, uint64_t addr2)
2139 {
2140 	uint32_t usec_timeout = 50000;  /* wait for 50ms */
2141 	uint32_t tmp;
2142 	unsigned i, pipe_id;
2143 	const struct gfx_firmware_header_v2_0 *pfp_hdr;
2144 
2145 	pfp_hdr = (const struct gfx_firmware_header_v2_0 *)
2146 		adev->gfx.pfp_fw->data;
2147 
2148 	WREG32_SOC15(GC, 0, regCP_PFP_IC_BASE_LO,
2149 		lower_32_bits(addr));
2150 	WREG32_SOC15(GC, 0, regCP_PFP_IC_BASE_HI,
2151 		upper_32_bits(addr));
2152 
2153 	tmp = RREG32_SOC15(GC, 0, regCP_PFP_IC_BASE_CNTL);
2154 	tmp = REG_SET_FIELD(tmp, CP_PFP_IC_BASE_CNTL, VMID, 0);
2155 	tmp = REG_SET_FIELD(tmp, CP_PFP_IC_BASE_CNTL, CACHE_POLICY, 0);
2156 	tmp = REG_SET_FIELD(tmp, CP_PFP_IC_BASE_CNTL, EXE_DISABLE, 0);
2157 	WREG32_SOC15(GC, 0, regCP_PFP_IC_BASE_CNTL, tmp);
2158 
2159 	/*
2160 	 * Programming any of the CP_PFP_IC_BASE registers
2161 	 * forces invalidation of the ME L1 I$. Wait for the
2162 	 * invalidation complete
2163 	 */
2164 	for (i = 0; i < usec_timeout; i++) {
2165 		tmp = RREG32_SOC15(GC, 0, regCP_PFP_IC_OP_CNTL);
2166 		if (1 == REG_GET_FIELD(tmp, CP_PFP_IC_OP_CNTL,
2167 			INVALIDATE_CACHE_COMPLETE))
2168 			break;
2169 		udelay(1);
2170 	}
2171 
2172 	if (i >= usec_timeout) {
2173 		dev_err(adev->dev, "failed to invalidate instruction cache\n");
2174 		return -EINVAL;
2175 	}
2176 
2177 	/* Prime the L1 instruction caches */
2178 	tmp = RREG32_SOC15(GC, 0, regCP_PFP_IC_OP_CNTL);
2179 	tmp = REG_SET_FIELD(tmp, CP_PFP_IC_OP_CNTL, PRIME_ICACHE, 1);
2180 	WREG32_SOC15(GC, 0, regCP_PFP_IC_OP_CNTL, tmp);
2181 	/* Waiting for cache primed*/
2182 	for (i = 0; i < usec_timeout; i++) {
2183 		tmp = RREG32_SOC15(GC, 0, regCP_PFP_IC_OP_CNTL);
2184 		if (1 == REG_GET_FIELD(tmp, CP_PFP_IC_OP_CNTL,
2185 			ICACHE_PRIMED))
2186 			break;
2187 		udelay(1);
2188 	}
2189 
2190 	if (i >= usec_timeout) {
2191 		dev_err(adev->dev, "failed to prime instruction cache\n");
2192 		return -EINVAL;
2193 	}
2194 
2195 	mutex_lock(&adev->srbm_mutex);
2196 	for (pipe_id = 0; pipe_id < adev->gfx.me.num_pipe_per_me; pipe_id++) {
2197 		soc21_grbm_select(adev, 0, pipe_id, 0, 0);
2198 		WREG32_SOC15(GC, 0, regCP_PFP_PRGRM_CNTR_START,
2199 			(pfp_hdr->ucode_start_addr_hi << 30) |
2200 			(pfp_hdr->ucode_start_addr_lo >> 2));
2201 		WREG32_SOC15(GC, 0, regCP_PFP_PRGRM_CNTR_START_HI,
2202 			pfp_hdr->ucode_start_addr_hi >> 2);
2203 
2204 		/*
2205 		 * Program CP_ME_CNTL to reset given PIPE to take
2206 		 * effect of CP_PFP_PRGRM_CNTR_START.
2207 		 */
2208 		tmp = RREG32_SOC15(GC, 0, regCP_ME_CNTL);
2209 		if (pipe_id == 0)
2210 			tmp = REG_SET_FIELD(tmp, CP_ME_CNTL,
2211 					PFP_PIPE0_RESET, 1);
2212 		else
2213 			tmp = REG_SET_FIELD(tmp, CP_ME_CNTL,
2214 					PFP_PIPE1_RESET, 1);
2215 		WREG32_SOC15(GC, 0, regCP_ME_CNTL, tmp);
2216 
2217 		/* Clear pfp pipe0 reset bit. */
2218 		if (pipe_id == 0)
2219 			tmp = REG_SET_FIELD(tmp, CP_ME_CNTL,
2220 					PFP_PIPE0_RESET, 0);
2221 		else
2222 			tmp = REG_SET_FIELD(tmp, CP_ME_CNTL,
2223 					PFP_PIPE1_RESET, 0);
2224 		WREG32_SOC15(GC, 0, regCP_ME_CNTL, tmp);
2225 
2226 		WREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_BASE0_LO,
2227 			lower_32_bits(addr2));
2228 		WREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_BASE0_HI,
2229 			upper_32_bits(addr2));
2230 	}
2231 	soc21_grbm_select(adev, 0, 0, 0, 0);
2232 	mutex_unlock(&adev->srbm_mutex);
2233 
2234 	tmp = RREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_BASE_CNTL);
2235 	tmp = REG_SET_FIELD(tmp, CP_GFX_RS64_DC_BASE_CNTL, VMID, 0);
2236 	tmp = REG_SET_FIELD(tmp, CP_GFX_RS64_DC_BASE_CNTL, CACHE_POLICY, 0);
2237 	WREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_BASE_CNTL, tmp);
2238 
2239 	/* Invalidate the data caches */
2240 	tmp = RREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_OP_CNTL);
2241 	tmp = REG_SET_FIELD(tmp, CP_GFX_RS64_DC_OP_CNTL, INVALIDATE_DCACHE, 1);
2242 	WREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_OP_CNTL, tmp);
2243 
2244 	for (i = 0; i < usec_timeout; i++) {
2245 		tmp = RREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_OP_CNTL);
2246 		if (1 == REG_GET_FIELD(tmp, CP_GFX_RS64_DC_OP_CNTL,
2247 			INVALIDATE_DCACHE_COMPLETE))
2248 			break;
2249 		udelay(1);
2250 	}
2251 
2252 	if (i >= usec_timeout) {
2253 		dev_err(adev->dev, "failed to invalidate RS64 data cache\n");
2254 		return -EINVAL;
2255 	}
2256 
2257 	return 0;
2258 }
2259 
2260 static int gfx_v11_0_config_me_cache_rs64(struct amdgpu_device *adev, uint64_t addr, uint64_t addr2)
2261 {
2262 	uint32_t usec_timeout = 50000;  /* wait for 50ms */
2263 	uint32_t tmp;
2264 	unsigned i, pipe_id;
2265 	const struct gfx_firmware_header_v2_0 *me_hdr;
2266 
2267 	me_hdr = (const struct gfx_firmware_header_v2_0 *)
2268 		adev->gfx.me_fw->data;
2269 
2270 	WREG32_SOC15(GC, 0, regCP_ME_IC_BASE_LO,
2271 		lower_32_bits(addr));
2272 	WREG32_SOC15(GC, 0, regCP_ME_IC_BASE_HI,
2273 		upper_32_bits(addr));
2274 
2275 	tmp = RREG32_SOC15(GC, 0, regCP_ME_IC_BASE_CNTL);
2276 	tmp = REG_SET_FIELD(tmp, CP_ME_IC_BASE_CNTL, VMID, 0);
2277 	tmp = REG_SET_FIELD(tmp, CP_ME_IC_BASE_CNTL, CACHE_POLICY, 0);
2278 	tmp = REG_SET_FIELD(tmp, CP_ME_IC_BASE_CNTL, EXE_DISABLE, 0);
2279 	WREG32_SOC15(GC, 0, regCP_ME_IC_BASE_CNTL, tmp);
2280 
2281 	/*
2282 	 * Programming any of the CP_ME_IC_BASE registers
2283 	 * forces invalidation of the ME L1 I$. Wait for the
2284 	 * invalidation complete
2285 	 */
2286 	for (i = 0; i < usec_timeout; i++) {
2287 		tmp = RREG32_SOC15(GC, 0, regCP_ME_IC_OP_CNTL);
2288 		if (1 == REG_GET_FIELD(tmp, CP_ME_IC_OP_CNTL,
2289 			INVALIDATE_CACHE_COMPLETE))
2290 			break;
2291 		udelay(1);
2292 	}
2293 
2294 	if (i >= usec_timeout) {
2295 		dev_err(adev->dev, "failed to invalidate instruction cache\n");
2296 		return -EINVAL;
2297 	}
2298 
2299 	/* Prime the instruction caches */
2300 	tmp = RREG32_SOC15(GC, 0, regCP_ME_IC_OP_CNTL);
2301 	tmp = REG_SET_FIELD(tmp, CP_ME_IC_OP_CNTL, PRIME_ICACHE, 1);
2302 	WREG32_SOC15(GC, 0, regCP_ME_IC_OP_CNTL, tmp);
2303 
2304 	/* Waiting for instruction cache primed*/
2305 	for (i = 0; i < usec_timeout; i++) {
2306 		tmp = RREG32_SOC15(GC, 0, regCP_ME_IC_OP_CNTL);
2307 		if (1 == REG_GET_FIELD(tmp, CP_ME_IC_OP_CNTL,
2308 			ICACHE_PRIMED))
2309 			break;
2310 		udelay(1);
2311 	}
2312 
2313 	if (i >= usec_timeout) {
2314 		dev_err(adev->dev, "failed to prime instruction cache\n");
2315 		return -EINVAL;
2316 	}
2317 
2318 	mutex_lock(&adev->srbm_mutex);
2319 	for (pipe_id = 0; pipe_id < adev->gfx.me.num_pipe_per_me; pipe_id++) {
2320 		soc21_grbm_select(adev, 0, pipe_id, 0, 0);
2321 		WREG32_SOC15(GC, 0, regCP_ME_PRGRM_CNTR_START,
2322 			(me_hdr->ucode_start_addr_hi << 30) |
2323 			(me_hdr->ucode_start_addr_lo >> 2) );
2324 		WREG32_SOC15(GC, 0, regCP_ME_PRGRM_CNTR_START_HI,
2325 			me_hdr->ucode_start_addr_hi>>2);
2326 
2327 		/*
2328 		 * Program CP_ME_CNTL to reset given PIPE to take
2329 		 * effect of CP_PFP_PRGRM_CNTR_START.
2330 		 */
2331 		tmp = RREG32_SOC15(GC, 0, regCP_ME_CNTL);
2332 		if (pipe_id == 0)
2333 			tmp = REG_SET_FIELD(tmp, CP_ME_CNTL,
2334 					ME_PIPE0_RESET, 1);
2335 		else
2336 			tmp = REG_SET_FIELD(tmp, CP_ME_CNTL,
2337 					ME_PIPE1_RESET, 1);
2338 		WREG32_SOC15(GC, 0, regCP_ME_CNTL, tmp);
2339 
2340 		/* Clear pfp pipe0 reset bit. */
2341 		if (pipe_id == 0)
2342 			tmp = REG_SET_FIELD(tmp, CP_ME_CNTL,
2343 					ME_PIPE0_RESET, 0);
2344 		else
2345 			tmp = REG_SET_FIELD(tmp, CP_ME_CNTL,
2346 					ME_PIPE1_RESET, 0);
2347 		WREG32_SOC15(GC, 0, regCP_ME_CNTL, tmp);
2348 
2349 		WREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_BASE1_LO,
2350 			lower_32_bits(addr2));
2351 		WREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_BASE1_HI,
2352 			upper_32_bits(addr2));
2353 	}
2354 	soc21_grbm_select(adev, 0, 0, 0, 0);
2355 	mutex_unlock(&adev->srbm_mutex);
2356 
2357 	tmp = RREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_BASE_CNTL);
2358 	tmp = REG_SET_FIELD(tmp, CP_GFX_RS64_DC_BASE_CNTL, VMID, 0);
2359 	tmp = REG_SET_FIELD(tmp, CP_GFX_RS64_DC_BASE_CNTL, CACHE_POLICY, 0);
2360 	WREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_BASE_CNTL, tmp);
2361 
2362 	/* Invalidate the data caches */
2363 	tmp = RREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_OP_CNTL);
2364 	tmp = REG_SET_FIELD(tmp, CP_GFX_RS64_DC_OP_CNTL, INVALIDATE_DCACHE, 1);
2365 	WREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_OP_CNTL, tmp);
2366 
2367 	for (i = 0; i < usec_timeout; i++) {
2368 		tmp = RREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_OP_CNTL);
2369 		if (1 == REG_GET_FIELD(tmp, CP_GFX_RS64_DC_OP_CNTL,
2370 			INVALIDATE_DCACHE_COMPLETE))
2371 			break;
2372 		udelay(1);
2373 	}
2374 
2375 	if (i >= usec_timeout) {
2376 		dev_err(adev->dev, "failed to invalidate RS64 data cache\n");
2377 		return -EINVAL;
2378 	}
2379 
2380 	return 0;
2381 }
2382 
2383 static int gfx_v11_0_config_mec_cache_rs64(struct amdgpu_device *adev, uint64_t addr, uint64_t addr2)
2384 {
2385 	uint32_t usec_timeout = 50000;  /* wait for 50ms */
2386 	uint32_t tmp;
2387 	unsigned i;
2388 	const struct gfx_firmware_header_v2_0 *mec_hdr;
2389 
2390 	mec_hdr = (const struct gfx_firmware_header_v2_0 *)
2391 		adev->gfx.mec_fw->data;
2392 
2393 	tmp = RREG32_SOC15(GC, 0, regCP_CPC_IC_BASE_CNTL);
2394 	tmp = REG_SET_FIELD(tmp, CP_CPC_IC_BASE_CNTL, VMID, 0);
2395 	tmp = REG_SET_FIELD(tmp, CP_CPC_IC_BASE_CNTL, EXE_DISABLE, 0);
2396 	tmp = REG_SET_FIELD(tmp, CP_CPC_IC_BASE_CNTL, CACHE_POLICY, 0);
2397 	WREG32_SOC15(GC, 0, regCP_CPC_IC_BASE_CNTL, tmp);
2398 
2399 	tmp = RREG32_SOC15(GC, 0, regCP_MEC_DC_BASE_CNTL);
2400 	tmp = REG_SET_FIELD(tmp, CP_MEC_DC_BASE_CNTL, VMID, 0);
2401 	tmp = REG_SET_FIELD(tmp, CP_MEC_DC_BASE_CNTL, CACHE_POLICY, 0);
2402 	WREG32_SOC15(GC, 0, regCP_MEC_DC_BASE_CNTL, tmp);
2403 
2404 	mutex_lock(&adev->srbm_mutex);
2405 	for (i = 0; i < adev->gfx.mec.num_pipe_per_mec; i++) {
2406 		soc21_grbm_select(adev, 1, i, 0, 0);
2407 
2408 		WREG32_SOC15(GC, 0, regCP_MEC_MDBASE_LO, addr2);
2409 		WREG32_SOC15(GC, 0, regCP_MEC_MDBASE_HI,
2410 		     upper_32_bits(addr2));
2411 
2412 		WREG32_SOC15(GC, 0, regCP_MEC_RS64_PRGRM_CNTR_START,
2413 					mec_hdr->ucode_start_addr_lo >> 2 |
2414 					mec_hdr->ucode_start_addr_hi << 30);
2415 		WREG32_SOC15(GC, 0, regCP_MEC_RS64_PRGRM_CNTR_START_HI,
2416 					mec_hdr->ucode_start_addr_hi >> 2);
2417 
2418 		WREG32_SOC15(GC, 0, regCP_CPC_IC_BASE_LO, addr);
2419 		WREG32_SOC15(GC, 0, regCP_CPC_IC_BASE_HI,
2420 		     upper_32_bits(addr));
2421 	}
2422 	mutex_unlock(&adev->srbm_mutex);
2423 	soc21_grbm_select(adev, 0, 0, 0, 0);
2424 
2425 	/* Trigger an invalidation of the L1 instruction caches */
2426 	tmp = RREG32_SOC15(GC, 0, regCP_MEC_DC_OP_CNTL);
2427 	tmp = REG_SET_FIELD(tmp, CP_MEC_DC_OP_CNTL, INVALIDATE_DCACHE, 1);
2428 	WREG32_SOC15(GC, 0, regCP_MEC_DC_OP_CNTL, tmp);
2429 
2430 	/* Wait for invalidation complete */
2431 	for (i = 0; i < usec_timeout; i++) {
2432 		tmp = RREG32_SOC15(GC, 0, regCP_MEC_DC_OP_CNTL);
2433 		if (1 == REG_GET_FIELD(tmp, CP_MEC_DC_OP_CNTL,
2434 				       INVALIDATE_DCACHE_COMPLETE))
2435 			break;
2436 		udelay(1);
2437 	}
2438 
2439 	if (i >= usec_timeout) {
2440 		dev_err(adev->dev, "failed to invalidate instruction cache\n");
2441 		return -EINVAL;
2442 	}
2443 
2444 	/* Trigger an invalidation of the L1 instruction caches */
2445 	tmp = RREG32_SOC15(GC, 0, regCP_CPC_IC_OP_CNTL);
2446 	tmp = REG_SET_FIELD(tmp, CP_CPC_IC_OP_CNTL, INVALIDATE_CACHE, 1);
2447 	WREG32_SOC15(GC, 0, regCP_CPC_IC_OP_CNTL, tmp);
2448 
2449 	/* Wait for invalidation complete */
2450 	for (i = 0; i < usec_timeout; i++) {
2451 		tmp = RREG32_SOC15(GC, 0, regCP_CPC_IC_OP_CNTL);
2452 		if (1 == REG_GET_FIELD(tmp, CP_CPC_IC_OP_CNTL,
2453 				       INVALIDATE_CACHE_COMPLETE))
2454 			break;
2455 		udelay(1);
2456 	}
2457 
2458 	if (i >= usec_timeout) {
2459 		dev_err(adev->dev, "failed to invalidate instruction cache\n");
2460 		return -EINVAL;
2461 	}
2462 
2463 	return 0;
2464 }
2465 
2466 static void gfx_v11_0_config_gfx_rs64(struct amdgpu_device *adev)
2467 {
2468 	const struct gfx_firmware_header_v2_0 *pfp_hdr;
2469 	const struct gfx_firmware_header_v2_0 *me_hdr;
2470 	const struct gfx_firmware_header_v2_0 *mec_hdr;
2471 	uint32_t pipe_id, tmp;
2472 
2473 	mec_hdr = (const struct gfx_firmware_header_v2_0 *)
2474 		adev->gfx.mec_fw->data;
2475 	me_hdr = (const struct gfx_firmware_header_v2_0 *)
2476 		adev->gfx.me_fw->data;
2477 	pfp_hdr = (const struct gfx_firmware_header_v2_0 *)
2478 		adev->gfx.pfp_fw->data;
2479 
2480 	/* config pfp program start addr */
2481 	for (pipe_id = 0; pipe_id < 2; pipe_id++) {
2482 		soc21_grbm_select(adev, 0, pipe_id, 0, 0);
2483 		WREG32_SOC15(GC, 0, regCP_PFP_PRGRM_CNTR_START,
2484 			(pfp_hdr->ucode_start_addr_hi << 30) |
2485 			(pfp_hdr->ucode_start_addr_lo >> 2));
2486 		WREG32_SOC15(GC, 0, regCP_PFP_PRGRM_CNTR_START_HI,
2487 			pfp_hdr->ucode_start_addr_hi >> 2);
2488 	}
2489 	soc21_grbm_select(adev, 0, 0, 0, 0);
2490 
2491 	/* reset pfp pipe */
2492 	tmp = RREG32_SOC15(GC, 0, regCP_ME_CNTL);
2493 	tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, PFP_PIPE0_RESET, 1);
2494 	tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, PFP_PIPE1_RESET, 1);
2495 	WREG32_SOC15(GC, 0, regCP_ME_CNTL, tmp);
2496 
2497 	/* clear pfp pipe reset */
2498 	tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, PFP_PIPE0_RESET, 0);
2499 	tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, PFP_PIPE1_RESET, 0);
2500 	WREG32_SOC15(GC, 0, regCP_ME_CNTL, tmp);
2501 
2502 	/* config me program start addr */
2503 	for (pipe_id = 0; pipe_id < 2; pipe_id++) {
2504 		soc21_grbm_select(adev, 0, pipe_id, 0, 0);
2505 		WREG32_SOC15(GC, 0, regCP_ME_PRGRM_CNTR_START,
2506 			(me_hdr->ucode_start_addr_hi << 30) |
2507 			(me_hdr->ucode_start_addr_lo >> 2) );
2508 		WREG32_SOC15(GC, 0, regCP_ME_PRGRM_CNTR_START_HI,
2509 			me_hdr->ucode_start_addr_hi>>2);
2510 	}
2511 	soc21_grbm_select(adev, 0, 0, 0, 0);
2512 
2513 	/* reset me pipe */
2514 	tmp = RREG32_SOC15(GC, 0, regCP_ME_CNTL);
2515 	tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, ME_PIPE0_RESET, 1);
2516 	tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, ME_PIPE1_RESET, 1);
2517 	WREG32_SOC15(GC, 0, regCP_ME_CNTL, tmp);
2518 
2519 	/* clear me pipe reset */
2520 	tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, ME_PIPE0_RESET, 0);
2521 	tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, ME_PIPE1_RESET, 0);
2522 	WREG32_SOC15(GC, 0, regCP_ME_CNTL, tmp);
2523 
2524 	/* config mec program start addr */
2525 	for (pipe_id = 0; pipe_id < 4; pipe_id++) {
2526 		soc21_grbm_select(adev, 1, pipe_id, 0, 0);
2527 		WREG32_SOC15(GC, 0, regCP_MEC_RS64_PRGRM_CNTR_START,
2528 					mec_hdr->ucode_start_addr_lo >> 2 |
2529 					mec_hdr->ucode_start_addr_hi << 30);
2530 		WREG32_SOC15(GC, 0, regCP_MEC_RS64_PRGRM_CNTR_START_HI,
2531 					mec_hdr->ucode_start_addr_hi >> 2);
2532 	}
2533 	soc21_grbm_select(adev, 0, 0, 0, 0);
2534 
2535 	/* reset mec pipe */
2536 	tmp = RREG32_SOC15(GC, 0, regCP_MEC_RS64_CNTL);
2537 	tmp = REG_SET_FIELD(tmp, CP_MEC_RS64_CNTL, MEC_PIPE0_RESET, 1);
2538 	tmp = REG_SET_FIELD(tmp, CP_MEC_RS64_CNTL, MEC_PIPE1_RESET, 1);
2539 	tmp = REG_SET_FIELD(tmp, CP_MEC_RS64_CNTL, MEC_PIPE2_RESET, 1);
2540 	tmp = REG_SET_FIELD(tmp, CP_MEC_RS64_CNTL, MEC_PIPE3_RESET, 1);
2541 	WREG32_SOC15(GC, 0, regCP_MEC_RS64_CNTL, tmp);
2542 
2543 	/* clear mec pipe reset */
2544 	tmp = REG_SET_FIELD(tmp, CP_MEC_RS64_CNTL, MEC_PIPE0_RESET, 0);
2545 	tmp = REG_SET_FIELD(tmp, CP_MEC_RS64_CNTL, MEC_PIPE1_RESET, 0);
2546 	tmp = REG_SET_FIELD(tmp, CP_MEC_RS64_CNTL, MEC_PIPE2_RESET, 0);
2547 	tmp = REG_SET_FIELD(tmp, CP_MEC_RS64_CNTL, MEC_PIPE3_RESET, 0);
2548 	WREG32_SOC15(GC, 0, regCP_MEC_RS64_CNTL, tmp);
2549 }
2550 
2551 static int gfx_v11_0_wait_for_rlc_autoload_complete(struct amdgpu_device *adev)
2552 {
2553 	uint32_t cp_status;
2554 	uint32_t bootload_status;
2555 	int i, r;
2556 	uint64_t addr, addr2;
2557 
2558 	for (i = 0; i < adev->usec_timeout; i++) {
2559 		cp_status = RREG32_SOC15(GC, 0, regCP_STAT);
2560 
2561 		if (adev->ip_versions[GC_HWIP][0] == IP_VERSION(11, 0, 1) ||
2562 				adev->ip_versions[GC_HWIP][0] == IP_VERSION(11, 0, 4))
2563 			bootload_status = RREG32_SOC15(GC, 0,
2564 					regRLC_RLCS_BOOTLOAD_STATUS_gc_11_0_1);
2565 		else
2566 			bootload_status = RREG32_SOC15(GC, 0, regRLC_RLCS_BOOTLOAD_STATUS);
2567 
2568 		if ((cp_status == 0) &&
2569 		    (REG_GET_FIELD(bootload_status,
2570 			RLC_RLCS_BOOTLOAD_STATUS, BOOTLOAD_COMPLETE) == 1)) {
2571 			break;
2572 		}
2573 		udelay(1);
2574 	}
2575 
2576 	if (i >= adev->usec_timeout) {
2577 		dev_err(adev->dev, "rlc autoload: gc ucode autoload timeout\n");
2578 		return -ETIMEDOUT;
2579 	}
2580 
2581 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_RLC_BACKDOOR_AUTO) {
2582 		if (adev->gfx.rs64_enable) {
2583 			addr = adev->gfx.rlc.rlc_autoload_gpu_addr +
2584 				rlc_autoload_info[SOC21_FIRMWARE_ID_RS64_ME].offset;
2585 			addr2 = adev->gfx.rlc.rlc_autoload_gpu_addr +
2586 				rlc_autoload_info[SOC21_FIRMWARE_ID_RS64_ME_P0_STACK].offset;
2587 			r = gfx_v11_0_config_me_cache_rs64(adev, addr, addr2);
2588 			if (r)
2589 				return r;
2590 			addr = adev->gfx.rlc.rlc_autoload_gpu_addr +
2591 				rlc_autoload_info[SOC21_FIRMWARE_ID_RS64_PFP].offset;
2592 			addr2 = adev->gfx.rlc.rlc_autoload_gpu_addr +
2593 				rlc_autoload_info[SOC21_FIRMWARE_ID_RS64_PFP_P0_STACK].offset;
2594 			r = gfx_v11_0_config_pfp_cache_rs64(adev, addr, addr2);
2595 			if (r)
2596 				return r;
2597 			addr = adev->gfx.rlc.rlc_autoload_gpu_addr +
2598 				rlc_autoload_info[SOC21_FIRMWARE_ID_RS64_MEC].offset;
2599 			addr2 = adev->gfx.rlc.rlc_autoload_gpu_addr +
2600 				rlc_autoload_info[SOC21_FIRMWARE_ID_RS64_MEC_P0_STACK].offset;
2601 			r = gfx_v11_0_config_mec_cache_rs64(adev, addr, addr2);
2602 			if (r)
2603 				return r;
2604 		} else {
2605 			addr = adev->gfx.rlc.rlc_autoload_gpu_addr +
2606 				rlc_autoload_info[SOC21_FIRMWARE_ID_CP_ME].offset;
2607 			r = gfx_v11_0_config_me_cache(adev, addr);
2608 			if (r)
2609 				return r;
2610 			addr = adev->gfx.rlc.rlc_autoload_gpu_addr +
2611 				rlc_autoload_info[SOC21_FIRMWARE_ID_CP_PFP].offset;
2612 			r = gfx_v11_0_config_pfp_cache(adev, addr);
2613 			if (r)
2614 				return r;
2615 			addr = adev->gfx.rlc.rlc_autoload_gpu_addr +
2616 				rlc_autoload_info[SOC21_FIRMWARE_ID_CP_MEC].offset;
2617 			r = gfx_v11_0_config_mec_cache(adev, addr);
2618 			if (r)
2619 				return r;
2620 		}
2621 	}
2622 
2623 	return 0;
2624 }
2625 
2626 static int gfx_v11_0_cp_gfx_enable(struct amdgpu_device *adev, bool enable)
2627 {
2628 	int i;
2629 	u32 tmp = RREG32_SOC15(GC, 0, regCP_ME_CNTL);
2630 
2631 	tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, ME_HALT, enable ? 0 : 1);
2632 	tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, PFP_HALT, enable ? 0 : 1);
2633 	WREG32_SOC15(GC, 0, regCP_ME_CNTL, tmp);
2634 
2635 	for (i = 0; i < adev->usec_timeout; i++) {
2636 		if (RREG32_SOC15(GC, 0, regCP_STAT) == 0)
2637 			break;
2638 		udelay(1);
2639 	}
2640 
2641 	if (i >= adev->usec_timeout)
2642 		DRM_ERROR("failed to %s cp gfx\n", enable ? "unhalt" : "halt");
2643 
2644 	return 0;
2645 }
2646 
2647 static int gfx_v11_0_cp_gfx_load_pfp_microcode(struct amdgpu_device *adev)
2648 {
2649 	int r;
2650 	const struct gfx_firmware_header_v1_0 *pfp_hdr;
2651 	const __le32 *fw_data;
2652 	unsigned i, fw_size;
2653 
2654 	pfp_hdr = (const struct gfx_firmware_header_v1_0 *)
2655 		adev->gfx.pfp_fw->data;
2656 
2657 	amdgpu_ucode_print_gfx_hdr(&pfp_hdr->header);
2658 
2659 	fw_data = (const __le32 *)(adev->gfx.pfp_fw->data +
2660 		le32_to_cpu(pfp_hdr->header.ucode_array_offset_bytes));
2661 	fw_size = le32_to_cpu(pfp_hdr->header.ucode_size_bytes);
2662 
2663 	r = amdgpu_bo_create_reserved(adev, pfp_hdr->header.ucode_size_bytes,
2664 				      PAGE_SIZE, AMDGPU_GEM_DOMAIN_GTT,
2665 				      &adev->gfx.pfp.pfp_fw_obj,
2666 				      &adev->gfx.pfp.pfp_fw_gpu_addr,
2667 				      (void **)&adev->gfx.pfp.pfp_fw_ptr);
2668 	if (r) {
2669 		dev_err(adev->dev, "(%d) failed to create pfp fw bo\n", r);
2670 		gfx_v11_0_pfp_fini(adev);
2671 		return r;
2672 	}
2673 
2674 	memcpy(adev->gfx.pfp.pfp_fw_ptr, fw_data, fw_size);
2675 
2676 	amdgpu_bo_kunmap(adev->gfx.pfp.pfp_fw_obj);
2677 	amdgpu_bo_unreserve(adev->gfx.pfp.pfp_fw_obj);
2678 
2679 	gfx_v11_0_config_pfp_cache(adev, adev->gfx.pfp.pfp_fw_gpu_addr);
2680 
2681 	WREG32_SOC15(GC, 0, regCP_HYP_PFP_UCODE_ADDR, 0);
2682 
2683 	for (i = 0; i < pfp_hdr->jt_size; i++)
2684 		WREG32_SOC15(GC, 0, regCP_HYP_PFP_UCODE_DATA,
2685 			     le32_to_cpup(fw_data + pfp_hdr->jt_offset + i));
2686 
2687 	WREG32_SOC15(GC, 0, regCP_HYP_PFP_UCODE_ADDR, adev->gfx.pfp_fw_version);
2688 
2689 	return 0;
2690 }
2691 
2692 static int gfx_v11_0_cp_gfx_load_pfp_microcode_rs64(struct amdgpu_device *adev)
2693 {
2694 	int r;
2695 	const struct gfx_firmware_header_v2_0 *pfp_hdr;
2696 	const __le32 *fw_ucode, *fw_data;
2697 	unsigned i, pipe_id, fw_ucode_size, fw_data_size;
2698 	uint32_t tmp;
2699 	uint32_t usec_timeout = 50000;  /* wait for 50ms */
2700 
2701 	pfp_hdr = (const struct gfx_firmware_header_v2_0 *)
2702 		adev->gfx.pfp_fw->data;
2703 
2704 	amdgpu_ucode_print_gfx_hdr(&pfp_hdr->header);
2705 
2706 	/* instruction */
2707 	fw_ucode = (const __le32 *)(adev->gfx.pfp_fw->data +
2708 		le32_to_cpu(pfp_hdr->ucode_offset_bytes));
2709 	fw_ucode_size = le32_to_cpu(pfp_hdr->ucode_size_bytes);
2710 	/* data */
2711 	fw_data = (const __le32 *)(adev->gfx.pfp_fw->data +
2712 		le32_to_cpu(pfp_hdr->data_offset_bytes));
2713 	fw_data_size = le32_to_cpu(pfp_hdr->data_size_bytes);
2714 
2715 	/* 64kb align */
2716 	r = amdgpu_bo_create_reserved(adev, fw_ucode_size,
2717 				      64 * 1024,
2718 				      AMDGPU_GEM_DOMAIN_VRAM |
2719 				      AMDGPU_GEM_DOMAIN_GTT,
2720 				      &adev->gfx.pfp.pfp_fw_obj,
2721 				      &adev->gfx.pfp.pfp_fw_gpu_addr,
2722 				      (void **)&adev->gfx.pfp.pfp_fw_ptr);
2723 	if (r) {
2724 		dev_err(adev->dev, "(%d) failed to create pfp ucode fw bo\n", r);
2725 		gfx_v11_0_pfp_fini(adev);
2726 		return r;
2727 	}
2728 
2729 	r = amdgpu_bo_create_reserved(adev, fw_data_size,
2730 				      64 * 1024,
2731 				      AMDGPU_GEM_DOMAIN_VRAM |
2732 				      AMDGPU_GEM_DOMAIN_GTT,
2733 				      &adev->gfx.pfp.pfp_fw_data_obj,
2734 				      &adev->gfx.pfp.pfp_fw_data_gpu_addr,
2735 				      (void **)&adev->gfx.pfp.pfp_fw_data_ptr);
2736 	if (r) {
2737 		dev_err(adev->dev, "(%d) failed to create pfp data fw bo\n", r);
2738 		gfx_v11_0_pfp_fini(adev);
2739 		return r;
2740 	}
2741 
2742 	memcpy(adev->gfx.pfp.pfp_fw_ptr, fw_ucode, fw_ucode_size);
2743 	memcpy(adev->gfx.pfp.pfp_fw_data_ptr, fw_data, fw_data_size);
2744 
2745 	amdgpu_bo_kunmap(adev->gfx.pfp.pfp_fw_obj);
2746 	amdgpu_bo_kunmap(adev->gfx.pfp.pfp_fw_data_obj);
2747 	amdgpu_bo_unreserve(adev->gfx.pfp.pfp_fw_obj);
2748 	amdgpu_bo_unreserve(adev->gfx.pfp.pfp_fw_data_obj);
2749 
2750 	if (amdgpu_emu_mode == 1)
2751 		adev->hdp.funcs->flush_hdp(adev, NULL);
2752 
2753 	WREG32_SOC15(GC, 0, regCP_PFP_IC_BASE_LO,
2754 		lower_32_bits(adev->gfx.pfp.pfp_fw_gpu_addr));
2755 	WREG32_SOC15(GC, 0, regCP_PFP_IC_BASE_HI,
2756 		upper_32_bits(adev->gfx.pfp.pfp_fw_gpu_addr));
2757 
2758 	tmp = RREG32_SOC15(GC, 0, regCP_PFP_IC_BASE_CNTL);
2759 	tmp = REG_SET_FIELD(tmp, CP_PFP_IC_BASE_CNTL, VMID, 0);
2760 	tmp = REG_SET_FIELD(tmp, CP_PFP_IC_BASE_CNTL, CACHE_POLICY, 0);
2761 	tmp = REG_SET_FIELD(tmp, CP_PFP_IC_BASE_CNTL, EXE_DISABLE, 0);
2762 	WREG32_SOC15(GC, 0, regCP_PFP_IC_BASE_CNTL, tmp);
2763 
2764 	/*
2765 	 * Programming any of the CP_PFP_IC_BASE registers
2766 	 * forces invalidation of the ME L1 I$. Wait for the
2767 	 * invalidation complete
2768 	 */
2769 	for (i = 0; i < usec_timeout; i++) {
2770 		tmp = RREG32_SOC15(GC, 0, regCP_PFP_IC_OP_CNTL);
2771 		if (1 == REG_GET_FIELD(tmp, CP_PFP_IC_OP_CNTL,
2772 			INVALIDATE_CACHE_COMPLETE))
2773 			break;
2774 		udelay(1);
2775 	}
2776 
2777 	if (i >= usec_timeout) {
2778 		dev_err(adev->dev, "failed to invalidate instruction cache\n");
2779 		return -EINVAL;
2780 	}
2781 
2782 	/* Prime the L1 instruction caches */
2783 	tmp = RREG32_SOC15(GC, 0, regCP_PFP_IC_OP_CNTL);
2784 	tmp = REG_SET_FIELD(tmp, CP_PFP_IC_OP_CNTL, PRIME_ICACHE, 1);
2785 	WREG32_SOC15(GC, 0, regCP_PFP_IC_OP_CNTL, tmp);
2786 	/* Waiting for cache primed*/
2787 	for (i = 0; i < usec_timeout; i++) {
2788 		tmp = RREG32_SOC15(GC, 0, regCP_PFP_IC_OP_CNTL);
2789 		if (1 == REG_GET_FIELD(tmp, CP_PFP_IC_OP_CNTL,
2790 			ICACHE_PRIMED))
2791 			break;
2792 		udelay(1);
2793 	}
2794 
2795 	if (i >= usec_timeout) {
2796 		dev_err(adev->dev, "failed to prime instruction cache\n");
2797 		return -EINVAL;
2798 	}
2799 
2800 	mutex_lock(&adev->srbm_mutex);
2801 	for (pipe_id = 0; pipe_id < adev->gfx.me.num_pipe_per_me; pipe_id++) {
2802 		soc21_grbm_select(adev, 0, pipe_id, 0, 0);
2803 		WREG32_SOC15(GC, 0, regCP_PFP_PRGRM_CNTR_START,
2804 			(pfp_hdr->ucode_start_addr_hi << 30) |
2805 			(pfp_hdr->ucode_start_addr_lo >> 2) );
2806 		WREG32_SOC15(GC, 0, regCP_PFP_PRGRM_CNTR_START_HI,
2807 			pfp_hdr->ucode_start_addr_hi>>2);
2808 
2809 		/*
2810 		 * Program CP_ME_CNTL to reset given PIPE to take
2811 		 * effect of CP_PFP_PRGRM_CNTR_START.
2812 		 */
2813 		tmp = RREG32_SOC15(GC, 0, regCP_ME_CNTL);
2814 		if (pipe_id == 0)
2815 			tmp = REG_SET_FIELD(tmp, CP_ME_CNTL,
2816 					PFP_PIPE0_RESET, 1);
2817 		else
2818 			tmp = REG_SET_FIELD(tmp, CP_ME_CNTL,
2819 					PFP_PIPE1_RESET, 1);
2820 		WREG32_SOC15(GC, 0, regCP_ME_CNTL, tmp);
2821 
2822 		/* Clear pfp pipe0 reset bit. */
2823 		if (pipe_id == 0)
2824 			tmp = REG_SET_FIELD(tmp, CP_ME_CNTL,
2825 					PFP_PIPE0_RESET, 0);
2826 		else
2827 			tmp = REG_SET_FIELD(tmp, CP_ME_CNTL,
2828 					PFP_PIPE1_RESET, 0);
2829 		WREG32_SOC15(GC, 0, regCP_ME_CNTL, tmp);
2830 
2831 		WREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_BASE0_LO,
2832 			lower_32_bits(adev->gfx.pfp.pfp_fw_data_gpu_addr));
2833 		WREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_BASE0_HI,
2834 			upper_32_bits(adev->gfx.pfp.pfp_fw_data_gpu_addr));
2835 	}
2836 	soc21_grbm_select(adev, 0, 0, 0, 0);
2837 	mutex_unlock(&adev->srbm_mutex);
2838 
2839 	tmp = RREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_BASE_CNTL);
2840 	tmp = REG_SET_FIELD(tmp, CP_GFX_RS64_DC_BASE_CNTL, VMID, 0);
2841 	tmp = REG_SET_FIELD(tmp, CP_GFX_RS64_DC_BASE_CNTL, CACHE_POLICY, 0);
2842 	WREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_BASE_CNTL, tmp);
2843 
2844 	/* Invalidate the data caches */
2845 	tmp = RREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_OP_CNTL);
2846 	tmp = REG_SET_FIELD(tmp, CP_GFX_RS64_DC_OP_CNTL, INVALIDATE_DCACHE, 1);
2847 	WREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_OP_CNTL, tmp);
2848 
2849 	for (i = 0; i < usec_timeout; i++) {
2850 		tmp = RREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_OP_CNTL);
2851 		if (1 == REG_GET_FIELD(tmp, CP_GFX_RS64_DC_OP_CNTL,
2852 			INVALIDATE_DCACHE_COMPLETE))
2853 			break;
2854 		udelay(1);
2855 	}
2856 
2857 	if (i >= usec_timeout) {
2858 		dev_err(adev->dev, "failed to invalidate RS64 data cache\n");
2859 		return -EINVAL;
2860 	}
2861 
2862 	return 0;
2863 }
2864 
2865 static int gfx_v11_0_cp_gfx_load_me_microcode(struct amdgpu_device *adev)
2866 {
2867 	int r;
2868 	const struct gfx_firmware_header_v1_0 *me_hdr;
2869 	const __le32 *fw_data;
2870 	unsigned i, fw_size;
2871 
2872 	me_hdr = (const struct gfx_firmware_header_v1_0 *)
2873 		adev->gfx.me_fw->data;
2874 
2875 	amdgpu_ucode_print_gfx_hdr(&me_hdr->header);
2876 
2877 	fw_data = (const __le32 *)(adev->gfx.me_fw->data +
2878 		le32_to_cpu(me_hdr->header.ucode_array_offset_bytes));
2879 	fw_size = le32_to_cpu(me_hdr->header.ucode_size_bytes);
2880 
2881 	r = amdgpu_bo_create_reserved(adev, me_hdr->header.ucode_size_bytes,
2882 				      PAGE_SIZE, AMDGPU_GEM_DOMAIN_GTT,
2883 				      &adev->gfx.me.me_fw_obj,
2884 				      &adev->gfx.me.me_fw_gpu_addr,
2885 				      (void **)&adev->gfx.me.me_fw_ptr);
2886 	if (r) {
2887 		dev_err(adev->dev, "(%d) failed to create me fw bo\n", r);
2888 		gfx_v11_0_me_fini(adev);
2889 		return r;
2890 	}
2891 
2892 	memcpy(adev->gfx.me.me_fw_ptr, fw_data, fw_size);
2893 
2894 	amdgpu_bo_kunmap(adev->gfx.me.me_fw_obj);
2895 	amdgpu_bo_unreserve(adev->gfx.me.me_fw_obj);
2896 
2897 	gfx_v11_0_config_me_cache(adev, adev->gfx.me.me_fw_gpu_addr);
2898 
2899 	WREG32_SOC15(GC, 0, regCP_HYP_ME_UCODE_ADDR, 0);
2900 
2901 	for (i = 0; i < me_hdr->jt_size; i++)
2902 		WREG32_SOC15(GC, 0, regCP_HYP_ME_UCODE_DATA,
2903 			     le32_to_cpup(fw_data + me_hdr->jt_offset + i));
2904 
2905 	WREG32_SOC15(GC, 0, regCP_HYP_ME_UCODE_ADDR, adev->gfx.me_fw_version);
2906 
2907 	return 0;
2908 }
2909 
2910 static int gfx_v11_0_cp_gfx_load_me_microcode_rs64(struct amdgpu_device *adev)
2911 {
2912 	int r;
2913 	const struct gfx_firmware_header_v2_0 *me_hdr;
2914 	const __le32 *fw_ucode, *fw_data;
2915 	unsigned i, pipe_id, fw_ucode_size, fw_data_size;
2916 	uint32_t tmp;
2917 	uint32_t usec_timeout = 50000;  /* wait for 50ms */
2918 
2919 	me_hdr = (const struct gfx_firmware_header_v2_0 *)
2920 		adev->gfx.me_fw->data;
2921 
2922 	amdgpu_ucode_print_gfx_hdr(&me_hdr->header);
2923 
2924 	/* instruction */
2925 	fw_ucode = (const __le32 *)(adev->gfx.me_fw->data +
2926 		le32_to_cpu(me_hdr->ucode_offset_bytes));
2927 	fw_ucode_size = le32_to_cpu(me_hdr->ucode_size_bytes);
2928 	/* data */
2929 	fw_data = (const __le32 *)(adev->gfx.me_fw->data +
2930 		le32_to_cpu(me_hdr->data_offset_bytes));
2931 	fw_data_size = le32_to_cpu(me_hdr->data_size_bytes);
2932 
2933 	/* 64kb align*/
2934 	r = amdgpu_bo_create_reserved(adev, fw_ucode_size,
2935 				      64 * 1024,
2936 				      AMDGPU_GEM_DOMAIN_VRAM |
2937 				      AMDGPU_GEM_DOMAIN_GTT,
2938 				      &adev->gfx.me.me_fw_obj,
2939 				      &adev->gfx.me.me_fw_gpu_addr,
2940 				      (void **)&adev->gfx.me.me_fw_ptr);
2941 	if (r) {
2942 		dev_err(adev->dev, "(%d) failed to create me ucode bo\n", r);
2943 		gfx_v11_0_me_fini(adev);
2944 		return r;
2945 	}
2946 
2947 	r = amdgpu_bo_create_reserved(adev, fw_data_size,
2948 				      64 * 1024,
2949 				      AMDGPU_GEM_DOMAIN_VRAM |
2950 				      AMDGPU_GEM_DOMAIN_GTT,
2951 				      &adev->gfx.me.me_fw_data_obj,
2952 				      &adev->gfx.me.me_fw_data_gpu_addr,
2953 				      (void **)&adev->gfx.me.me_fw_data_ptr);
2954 	if (r) {
2955 		dev_err(adev->dev, "(%d) failed to create me data bo\n", r);
2956 		gfx_v11_0_pfp_fini(adev);
2957 		return r;
2958 	}
2959 
2960 	memcpy(adev->gfx.me.me_fw_ptr, fw_ucode, fw_ucode_size);
2961 	memcpy(adev->gfx.me.me_fw_data_ptr, fw_data, fw_data_size);
2962 
2963 	amdgpu_bo_kunmap(adev->gfx.me.me_fw_obj);
2964 	amdgpu_bo_kunmap(adev->gfx.me.me_fw_data_obj);
2965 	amdgpu_bo_unreserve(adev->gfx.me.me_fw_obj);
2966 	amdgpu_bo_unreserve(adev->gfx.me.me_fw_data_obj);
2967 
2968 	if (amdgpu_emu_mode == 1)
2969 		adev->hdp.funcs->flush_hdp(adev, NULL);
2970 
2971 	WREG32_SOC15(GC, 0, regCP_ME_IC_BASE_LO,
2972 		lower_32_bits(adev->gfx.me.me_fw_gpu_addr));
2973 	WREG32_SOC15(GC, 0, regCP_ME_IC_BASE_HI,
2974 		upper_32_bits(adev->gfx.me.me_fw_gpu_addr));
2975 
2976 	tmp = RREG32_SOC15(GC, 0, regCP_ME_IC_BASE_CNTL);
2977 	tmp = REG_SET_FIELD(tmp, CP_ME_IC_BASE_CNTL, VMID, 0);
2978 	tmp = REG_SET_FIELD(tmp, CP_ME_IC_BASE_CNTL, CACHE_POLICY, 0);
2979 	tmp = REG_SET_FIELD(tmp, CP_ME_IC_BASE_CNTL, EXE_DISABLE, 0);
2980 	WREG32_SOC15(GC, 0, regCP_ME_IC_BASE_CNTL, tmp);
2981 
2982 	/*
2983 	 * Programming any of the CP_ME_IC_BASE registers
2984 	 * forces invalidation of the ME L1 I$. Wait for the
2985 	 * invalidation complete
2986 	 */
2987 	for (i = 0; i < usec_timeout; i++) {
2988 		tmp = RREG32_SOC15(GC, 0, regCP_ME_IC_OP_CNTL);
2989 		if (1 == REG_GET_FIELD(tmp, CP_ME_IC_OP_CNTL,
2990 			INVALIDATE_CACHE_COMPLETE))
2991 			break;
2992 		udelay(1);
2993 	}
2994 
2995 	if (i >= usec_timeout) {
2996 		dev_err(adev->dev, "failed to invalidate instruction cache\n");
2997 		return -EINVAL;
2998 	}
2999 
3000 	/* Prime the instruction caches */
3001 	tmp = RREG32_SOC15(GC, 0, regCP_ME_IC_OP_CNTL);
3002 	tmp = REG_SET_FIELD(tmp, CP_ME_IC_OP_CNTL, PRIME_ICACHE, 1);
3003 	WREG32_SOC15(GC, 0, regCP_ME_IC_OP_CNTL, tmp);
3004 
3005 	/* Waiting for instruction cache primed*/
3006 	for (i = 0; i < usec_timeout; i++) {
3007 		tmp = RREG32_SOC15(GC, 0, regCP_ME_IC_OP_CNTL);
3008 		if (1 == REG_GET_FIELD(tmp, CP_ME_IC_OP_CNTL,
3009 			ICACHE_PRIMED))
3010 			break;
3011 		udelay(1);
3012 	}
3013 
3014 	if (i >= usec_timeout) {
3015 		dev_err(adev->dev, "failed to prime instruction cache\n");
3016 		return -EINVAL;
3017 	}
3018 
3019 	mutex_lock(&adev->srbm_mutex);
3020 	for (pipe_id = 0; pipe_id < adev->gfx.me.num_pipe_per_me; pipe_id++) {
3021 		soc21_grbm_select(adev, 0, pipe_id, 0, 0);
3022 		WREG32_SOC15(GC, 0, regCP_ME_PRGRM_CNTR_START,
3023 			(me_hdr->ucode_start_addr_hi << 30) |
3024 			(me_hdr->ucode_start_addr_lo >> 2) );
3025 		WREG32_SOC15(GC, 0, regCP_ME_PRGRM_CNTR_START_HI,
3026 			me_hdr->ucode_start_addr_hi>>2);
3027 
3028 		/*
3029 		 * Program CP_ME_CNTL to reset given PIPE to take
3030 		 * effect of CP_PFP_PRGRM_CNTR_START.
3031 		 */
3032 		tmp = RREG32_SOC15(GC, 0, regCP_ME_CNTL);
3033 		if (pipe_id == 0)
3034 			tmp = REG_SET_FIELD(tmp, CP_ME_CNTL,
3035 					ME_PIPE0_RESET, 1);
3036 		else
3037 			tmp = REG_SET_FIELD(tmp, CP_ME_CNTL,
3038 					ME_PIPE1_RESET, 1);
3039 		WREG32_SOC15(GC, 0, regCP_ME_CNTL, tmp);
3040 
3041 		/* Clear pfp pipe0 reset bit. */
3042 		if (pipe_id == 0)
3043 			tmp = REG_SET_FIELD(tmp, CP_ME_CNTL,
3044 					ME_PIPE0_RESET, 0);
3045 		else
3046 			tmp = REG_SET_FIELD(tmp, CP_ME_CNTL,
3047 					ME_PIPE1_RESET, 0);
3048 		WREG32_SOC15(GC, 0, regCP_ME_CNTL, tmp);
3049 
3050 		WREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_BASE1_LO,
3051 			lower_32_bits(adev->gfx.me.me_fw_data_gpu_addr));
3052 		WREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_BASE1_HI,
3053 			upper_32_bits(adev->gfx.me.me_fw_data_gpu_addr));
3054 	}
3055 	soc21_grbm_select(adev, 0, 0, 0, 0);
3056 	mutex_unlock(&adev->srbm_mutex);
3057 
3058 	tmp = RREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_BASE_CNTL);
3059 	tmp = REG_SET_FIELD(tmp, CP_GFX_RS64_DC_BASE_CNTL, VMID, 0);
3060 	tmp = REG_SET_FIELD(tmp, CP_GFX_RS64_DC_BASE_CNTL, CACHE_POLICY, 0);
3061 	WREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_BASE_CNTL, tmp);
3062 
3063 	/* Invalidate the data caches */
3064 	tmp = RREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_OP_CNTL);
3065 	tmp = REG_SET_FIELD(tmp, CP_GFX_RS64_DC_OP_CNTL, INVALIDATE_DCACHE, 1);
3066 	WREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_OP_CNTL, tmp);
3067 
3068 	for (i = 0; i < usec_timeout; i++) {
3069 		tmp = RREG32_SOC15(GC, 0, regCP_GFX_RS64_DC_OP_CNTL);
3070 		if (1 == REG_GET_FIELD(tmp, CP_GFX_RS64_DC_OP_CNTL,
3071 			INVALIDATE_DCACHE_COMPLETE))
3072 			break;
3073 		udelay(1);
3074 	}
3075 
3076 	if (i >= usec_timeout) {
3077 		dev_err(adev->dev, "failed to invalidate RS64 data cache\n");
3078 		return -EINVAL;
3079 	}
3080 
3081 	return 0;
3082 }
3083 
3084 static int gfx_v11_0_cp_gfx_load_microcode(struct amdgpu_device *adev)
3085 {
3086 	int r;
3087 
3088 	if (!adev->gfx.me_fw || !adev->gfx.pfp_fw)
3089 		return -EINVAL;
3090 
3091 	gfx_v11_0_cp_gfx_enable(adev, false);
3092 
3093 	if (adev->gfx.rs64_enable)
3094 		r = gfx_v11_0_cp_gfx_load_pfp_microcode_rs64(adev);
3095 	else
3096 		r = gfx_v11_0_cp_gfx_load_pfp_microcode(adev);
3097 	if (r) {
3098 		dev_err(adev->dev, "(%d) failed to load pfp fw\n", r);
3099 		return r;
3100 	}
3101 
3102 	if (adev->gfx.rs64_enable)
3103 		r = gfx_v11_0_cp_gfx_load_me_microcode_rs64(adev);
3104 	else
3105 		r = gfx_v11_0_cp_gfx_load_me_microcode(adev);
3106 	if (r) {
3107 		dev_err(adev->dev, "(%d) failed to load me fw\n", r);
3108 		return r;
3109 	}
3110 
3111 	return 0;
3112 }
3113 
3114 static int gfx_v11_0_cp_gfx_start(struct amdgpu_device *adev)
3115 {
3116 	struct amdgpu_ring *ring;
3117 	const struct cs_section_def *sect = NULL;
3118 	const struct cs_extent_def *ext = NULL;
3119 	int r, i;
3120 	int ctx_reg_offset;
3121 
3122 	/* init the CP */
3123 	WREG32_SOC15(GC, 0, regCP_MAX_CONTEXT,
3124 		     adev->gfx.config.max_hw_contexts - 1);
3125 	WREG32_SOC15(GC, 0, regCP_DEVICE_ID, 1);
3126 
3127 	if (!amdgpu_async_gfx_ring)
3128 		gfx_v11_0_cp_gfx_enable(adev, true);
3129 
3130 	ring = &adev->gfx.gfx_ring[0];
3131 	r = amdgpu_ring_alloc(ring, gfx_v11_0_get_csb_size(adev));
3132 	if (r) {
3133 		DRM_ERROR("amdgpu: cp failed to lock ring (%d).\n", r);
3134 		return r;
3135 	}
3136 
3137 	amdgpu_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
3138 	amdgpu_ring_write(ring, PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
3139 
3140 	amdgpu_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1));
3141 	amdgpu_ring_write(ring, 0x80000000);
3142 	amdgpu_ring_write(ring, 0x80000000);
3143 
3144 	for (sect = gfx11_cs_data; sect->section != NULL; ++sect) {
3145 		for (ext = sect->section; ext->extent != NULL; ++ext) {
3146 			if (sect->id == SECT_CONTEXT) {
3147 				amdgpu_ring_write(ring,
3148 						  PACKET3(PACKET3_SET_CONTEXT_REG,
3149 							  ext->reg_count));
3150 				amdgpu_ring_write(ring, ext->reg_index -
3151 						  PACKET3_SET_CONTEXT_REG_START);
3152 				for (i = 0; i < ext->reg_count; i++)
3153 					amdgpu_ring_write(ring, ext->extent[i]);
3154 			}
3155 		}
3156 	}
3157 
3158 	ctx_reg_offset =
3159 		SOC15_REG_OFFSET(GC, 0, regPA_SC_TILE_STEERING_OVERRIDE) - PACKET3_SET_CONTEXT_REG_START;
3160 	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONTEXT_REG, 1));
3161 	amdgpu_ring_write(ring, ctx_reg_offset);
3162 	amdgpu_ring_write(ring, adev->gfx.config.pa_sc_tile_steering_override);
3163 
3164 	amdgpu_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
3165 	amdgpu_ring_write(ring, PACKET3_PREAMBLE_END_CLEAR_STATE);
3166 
3167 	amdgpu_ring_write(ring, PACKET3(PACKET3_CLEAR_STATE, 0));
3168 	amdgpu_ring_write(ring, 0);
3169 
3170 	amdgpu_ring_commit(ring);
3171 
3172 	/* submit cs packet to copy state 0 to next available state */
3173 	if (adev->gfx.num_gfx_rings > 1) {
3174 		/* maximum supported gfx ring is 2 */
3175 		ring = &adev->gfx.gfx_ring[1];
3176 		r = amdgpu_ring_alloc(ring, 2);
3177 		if (r) {
3178 			DRM_ERROR("amdgpu: cp failed to lock ring (%d).\n", r);
3179 			return r;
3180 		}
3181 
3182 		amdgpu_ring_write(ring, PACKET3(PACKET3_CLEAR_STATE, 0));
3183 		amdgpu_ring_write(ring, 0);
3184 
3185 		amdgpu_ring_commit(ring);
3186 	}
3187 	return 0;
3188 }
3189 
3190 static void gfx_v11_0_cp_gfx_switch_pipe(struct amdgpu_device *adev,
3191 					 CP_PIPE_ID pipe)
3192 {
3193 	u32 tmp;
3194 
3195 	tmp = RREG32_SOC15(GC, 0, regGRBM_GFX_CNTL);
3196 	tmp = REG_SET_FIELD(tmp, GRBM_GFX_CNTL, PIPEID, pipe);
3197 
3198 	WREG32_SOC15(GC, 0, regGRBM_GFX_CNTL, tmp);
3199 }
3200 
3201 static void gfx_v11_0_cp_gfx_set_doorbell(struct amdgpu_device *adev,
3202 					  struct amdgpu_ring *ring)
3203 {
3204 	u32 tmp;
3205 
3206 	tmp = RREG32_SOC15(GC, 0, regCP_RB_DOORBELL_CONTROL);
3207 	if (ring->use_doorbell) {
3208 		tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL,
3209 				    DOORBELL_OFFSET, ring->doorbell_index);
3210 		tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL,
3211 				    DOORBELL_EN, 1);
3212 	} else {
3213 		tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL,
3214 				    DOORBELL_EN, 0);
3215 	}
3216 	WREG32_SOC15(GC, 0, regCP_RB_DOORBELL_CONTROL, tmp);
3217 
3218 	tmp = REG_SET_FIELD(0, CP_RB_DOORBELL_RANGE_LOWER,
3219 			    DOORBELL_RANGE_LOWER, ring->doorbell_index);
3220 	WREG32_SOC15(GC, 0, regCP_RB_DOORBELL_RANGE_LOWER, tmp);
3221 
3222 	WREG32_SOC15(GC, 0, regCP_RB_DOORBELL_RANGE_UPPER,
3223 		     CP_RB_DOORBELL_RANGE_UPPER__DOORBELL_RANGE_UPPER_MASK);
3224 }
3225 
3226 static int gfx_v11_0_cp_gfx_resume(struct amdgpu_device *adev)
3227 {
3228 	struct amdgpu_ring *ring;
3229 	u32 tmp;
3230 	u32 rb_bufsz;
3231 	u64 rb_addr, rptr_addr, wptr_gpu_addr;
3232 
3233 	/* Set the write pointer delay */
3234 	WREG32_SOC15(GC, 0, regCP_RB_WPTR_DELAY, 0);
3235 
3236 	/* set the RB to use vmid 0 */
3237 	WREG32_SOC15(GC, 0, regCP_RB_VMID, 0);
3238 
3239 	/* Init gfx ring 0 for pipe 0 */
3240 	mutex_lock(&adev->srbm_mutex);
3241 	gfx_v11_0_cp_gfx_switch_pipe(adev, PIPE_ID0);
3242 
3243 	/* Set ring buffer size */
3244 	ring = &adev->gfx.gfx_ring[0];
3245 	rb_bufsz = order_base_2(ring->ring_size / 8);
3246 	tmp = REG_SET_FIELD(0, CP_RB0_CNTL, RB_BUFSZ, rb_bufsz);
3247 	tmp = REG_SET_FIELD(tmp, CP_RB0_CNTL, RB_BLKSZ, rb_bufsz - 2);
3248 	WREG32_SOC15(GC, 0, regCP_RB0_CNTL, tmp);
3249 
3250 	/* Initialize the ring buffer's write pointers */
3251 	ring->wptr = 0;
3252 	WREG32_SOC15(GC, 0, regCP_RB0_WPTR, lower_32_bits(ring->wptr));
3253 	WREG32_SOC15(GC, 0, regCP_RB0_WPTR_HI, upper_32_bits(ring->wptr));
3254 
3255 	/* set the wb address wether it's enabled or not */
3256 	rptr_addr = ring->rptr_gpu_addr;
3257 	WREG32_SOC15(GC, 0, regCP_RB0_RPTR_ADDR, lower_32_bits(rptr_addr));
3258 	WREG32_SOC15(GC, 0, regCP_RB0_RPTR_ADDR_HI, upper_32_bits(rptr_addr) &
3259 		     CP_RB_RPTR_ADDR_HI__RB_RPTR_ADDR_HI_MASK);
3260 
3261 	wptr_gpu_addr = ring->wptr_gpu_addr;
3262 	WREG32_SOC15(GC, 0, regCP_RB_WPTR_POLL_ADDR_LO,
3263 		     lower_32_bits(wptr_gpu_addr));
3264 	WREG32_SOC15(GC, 0, regCP_RB_WPTR_POLL_ADDR_HI,
3265 		     upper_32_bits(wptr_gpu_addr));
3266 
3267 	mdelay(1);
3268 	WREG32_SOC15(GC, 0, regCP_RB0_CNTL, tmp);
3269 
3270 	rb_addr = ring->gpu_addr >> 8;
3271 	WREG32_SOC15(GC, 0, regCP_RB0_BASE, rb_addr);
3272 	WREG32_SOC15(GC, 0, regCP_RB0_BASE_HI, upper_32_bits(rb_addr));
3273 
3274 	WREG32_SOC15(GC, 0, regCP_RB_ACTIVE, 1);
3275 
3276 	gfx_v11_0_cp_gfx_set_doorbell(adev, ring);
3277 	mutex_unlock(&adev->srbm_mutex);
3278 
3279 	/* Init gfx ring 1 for pipe 1 */
3280 	if (adev->gfx.num_gfx_rings > 1) {
3281 		mutex_lock(&adev->srbm_mutex);
3282 		gfx_v11_0_cp_gfx_switch_pipe(adev, PIPE_ID1);
3283 		/* maximum supported gfx ring is 2 */
3284 		ring = &adev->gfx.gfx_ring[1];
3285 		rb_bufsz = order_base_2(ring->ring_size / 8);
3286 		tmp = REG_SET_FIELD(0, CP_RB1_CNTL, RB_BUFSZ, rb_bufsz);
3287 		tmp = REG_SET_FIELD(tmp, CP_RB1_CNTL, RB_BLKSZ, rb_bufsz - 2);
3288 		WREG32_SOC15(GC, 0, regCP_RB1_CNTL, tmp);
3289 		/* Initialize the ring buffer's write pointers */
3290 		ring->wptr = 0;
3291 		WREG32_SOC15(GC, 0, regCP_RB1_WPTR, lower_32_bits(ring->wptr));
3292 		WREG32_SOC15(GC, 0, regCP_RB1_WPTR_HI, upper_32_bits(ring->wptr));
3293 		/* Set the wb address wether it's enabled or not */
3294 		rptr_addr = ring->rptr_gpu_addr;
3295 		WREG32_SOC15(GC, 0, regCP_RB1_RPTR_ADDR, lower_32_bits(rptr_addr));
3296 		WREG32_SOC15(GC, 0, regCP_RB1_RPTR_ADDR_HI, upper_32_bits(rptr_addr) &
3297 			     CP_RB1_RPTR_ADDR_HI__RB_RPTR_ADDR_HI_MASK);
3298 		wptr_gpu_addr = ring->wptr_gpu_addr;
3299 		WREG32_SOC15(GC, 0, regCP_RB_WPTR_POLL_ADDR_LO,
3300 			     lower_32_bits(wptr_gpu_addr));
3301 		WREG32_SOC15(GC, 0, regCP_RB_WPTR_POLL_ADDR_HI,
3302 			     upper_32_bits(wptr_gpu_addr));
3303 
3304 		mdelay(1);
3305 		WREG32_SOC15(GC, 0, regCP_RB1_CNTL, tmp);
3306 
3307 		rb_addr = ring->gpu_addr >> 8;
3308 		WREG32_SOC15(GC, 0, regCP_RB1_BASE, rb_addr);
3309 		WREG32_SOC15(GC, 0, regCP_RB1_BASE_HI, upper_32_bits(rb_addr));
3310 		WREG32_SOC15(GC, 0, regCP_RB1_ACTIVE, 1);
3311 
3312 		gfx_v11_0_cp_gfx_set_doorbell(adev, ring);
3313 		mutex_unlock(&adev->srbm_mutex);
3314 	}
3315 	/* Switch to pipe 0 */
3316 	mutex_lock(&adev->srbm_mutex);
3317 	gfx_v11_0_cp_gfx_switch_pipe(adev, PIPE_ID0);
3318 	mutex_unlock(&adev->srbm_mutex);
3319 
3320 	/* start the ring */
3321 	gfx_v11_0_cp_gfx_start(adev);
3322 
3323 	return 0;
3324 }
3325 
3326 static void gfx_v11_0_cp_compute_enable(struct amdgpu_device *adev, bool enable)
3327 {
3328 	u32 data;
3329 
3330 	if (adev->gfx.rs64_enable) {
3331 		data = RREG32_SOC15(GC, 0, regCP_MEC_RS64_CNTL);
3332 		data = REG_SET_FIELD(data, CP_MEC_RS64_CNTL, MEC_INVALIDATE_ICACHE,
3333 							 enable ? 0 : 1);
3334 		data = REG_SET_FIELD(data, CP_MEC_RS64_CNTL, MEC_PIPE0_RESET,
3335 							 enable ? 0 : 1);
3336 		data = REG_SET_FIELD(data, CP_MEC_RS64_CNTL, MEC_PIPE1_RESET,
3337 							 enable ? 0 : 1);
3338 		data = REG_SET_FIELD(data, CP_MEC_RS64_CNTL, MEC_PIPE2_RESET,
3339 							 enable ? 0 : 1);
3340 		data = REG_SET_FIELD(data, CP_MEC_RS64_CNTL, MEC_PIPE3_RESET,
3341 							 enable ? 0 : 1);
3342 		data = REG_SET_FIELD(data, CP_MEC_RS64_CNTL, MEC_PIPE0_ACTIVE,
3343 							 enable ? 1 : 0);
3344 		data = REG_SET_FIELD(data, CP_MEC_RS64_CNTL, MEC_PIPE1_ACTIVE,
3345 				                         enable ? 1 : 0);
3346 		data = REG_SET_FIELD(data, CP_MEC_RS64_CNTL, MEC_PIPE2_ACTIVE,
3347 							 enable ? 1 : 0);
3348 		data = REG_SET_FIELD(data, CP_MEC_RS64_CNTL, MEC_PIPE3_ACTIVE,
3349 							 enable ? 1 : 0);
3350 		data = REG_SET_FIELD(data, CP_MEC_RS64_CNTL, MEC_HALT,
3351 							 enable ? 0 : 1);
3352 		WREG32_SOC15(GC, 0, regCP_MEC_RS64_CNTL, data);
3353 	} else {
3354 		data = RREG32_SOC15(GC, 0, regCP_MEC_CNTL);
3355 
3356 		if (enable) {
3357 			data = REG_SET_FIELD(data, CP_MEC_CNTL, MEC_ME1_HALT, 0);
3358 			if (!adev->enable_mes_kiq)
3359 				data = REG_SET_FIELD(data, CP_MEC_CNTL,
3360 						     MEC_ME2_HALT, 0);
3361 		} else {
3362 			data = REG_SET_FIELD(data, CP_MEC_CNTL, MEC_ME1_HALT, 1);
3363 			data = REG_SET_FIELD(data, CP_MEC_CNTL, MEC_ME2_HALT, 1);
3364 		}
3365 		WREG32_SOC15(GC, 0, regCP_MEC_CNTL, data);
3366 	}
3367 
3368 	udelay(50);
3369 }
3370 
3371 static int gfx_v11_0_cp_compute_load_microcode(struct amdgpu_device *adev)
3372 {
3373 	const struct gfx_firmware_header_v1_0 *mec_hdr;
3374 	const __le32 *fw_data;
3375 	unsigned i, fw_size;
3376 	u32 *fw = NULL;
3377 	int r;
3378 
3379 	if (!adev->gfx.mec_fw)
3380 		return -EINVAL;
3381 
3382 	gfx_v11_0_cp_compute_enable(adev, false);
3383 
3384 	mec_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data;
3385 	amdgpu_ucode_print_gfx_hdr(&mec_hdr->header);
3386 
3387 	fw_data = (const __le32 *)
3388 		(adev->gfx.mec_fw->data +
3389 		 le32_to_cpu(mec_hdr->header.ucode_array_offset_bytes));
3390 	fw_size = le32_to_cpu(mec_hdr->header.ucode_size_bytes);
3391 
3392 	r = amdgpu_bo_create_reserved(adev, mec_hdr->header.ucode_size_bytes,
3393 					  PAGE_SIZE, AMDGPU_GEM_DOMAIN_GTT,
3394 					  &adev->gfx.mec.mec_fw_obj,
3395 					  &adev->gfx.mec.mec_fw_gpu_addr,
3396 					  (void **)&fw);
3397 	if (r) {
3398 		dev_err(adev->dev, "(%d) failed to create mec fw bo\n", r);
3399 		gfx_v11_0_mec_fini(adev);
3400 		return r;
3401 	}
3402 
3403 	memcpy(fw, fw_data, fw_size);
3404 
3405 	amdgpu_bo_kunmap(adev->gfx.mec.mec_fw_obj);
3406 	amdgpu_bo_unreserve(adev->gfx.mec.mec_fw_obj);
3407 
3408 	gfx_v11_0_config_mec_cache(adev, adev->gfx.mec.mec_fw_gpu_addr);
3409 
3410 	/* MEC1 */
3411 	WREG32_SOC15(GC, 0, regCP_MEC_ME1_UCODE_ADDR, 0);
3412 
3413 	for (i = 0; i < mec_hdr->jt_size; i++)
3414 		WREG32_SOC15(GC, 0, regCP_MEC_ME1_UCODE_DATA,
3415 			     le32_to_cpup(fw_data + mec_hdr->jt_offset + i));
3416 
3417 	WREG32_SOC15(GC, 0, regCP_MEC_ME1_UCODE_ADDR, adev->gfx.mec_fw_version);
3418 
3419 	return 0;
3420 }
3421 
3422 static int gfx_v11_0_cp_compute_load_microcode_rs64(struct amdgpu_device *adev)
3423 {
3424 	const struct gfx_firmware_header_v2_0 *mec_hdr;
3425 	const __le32 *fw_ucode, *fw_data;
3426 	u32 tmp, fw_ucode_size, fw_data_size;
3427 	u32 i, usec_timeout = 50000; /* Wait for 50 ms */
3428 	u32 *fw_ucode_ptr, *fw_data_ptr;
3429 	int r;
3430 
3431 	if (!adev->gfx.mec_fw)
3432 		return -EINVAL;
3433 
3434 	gfx_v11_0_cp_compute_enable(adev, false);
3435 
3436 	mec_hdr = (const struct gfx_firmware_header_v2_0 *)adev->gfx.mec_fw->data;
3437 	amdgpu_ucode_print_gfx_hdr(&mec_hdr->header);
3438 
3439 	fw_ucode = (const __le32 *) (adev->gfx.mec_fw->data +
3440 				le32_to_cpu(mec_hdr->ucode_offset_bytes));
3441 	fw_ucode_size = le32_to_cpu(mec_hdr->ucode_size_bytes);
3442 
3443 	fw_data = (const __le32 *) (adev->gfx.mec_fw->data +
3444 				le32_to_cpu(mec_hdr->data_offset_bytes));
3445 	fw_data_size = le32_to_cpu(mec_hdr->data_size_bytes);
3446 
3447 	r = amdgpu_bo_create_reserved(adev, fw_ucode_size,
3448 				      64 * 1024,
3449 				      AMDGPU_GEM_DOMAIN_VRAM |
3450 				      AMDGPU_GEM_DOMAIN_GTT,
3451 				      &adev->gfx.mec.mec_fw_obj,
3452 				      &adev->gfx.mec.mec_fw_gpu_addr,
3453 				      (void **)&fw_ucode_ptr);
3454 	if (r) {
3455 		dev_err(adev->dev, "(%d) failed to create mec fw ucode bo\n", r);
3456 		gfx_v11_0_mec_fini(adev);
3457 		return r;
3458 	}
3459 
3460 	r = amdgpu_bo_create_reserved(adev, fw_data_size,
3461 				      64 * 1024,
3462 				      AMDGPU_GEM_DOMAIN_VRAM |
3463 				      AMDGPU_GEM_DOMAIN_GTT,
3464 				      &adev->gfx.mec.mec_fw_data_obj,
3465 				      &adev->gfx.mec.mec_fw_data_gpu_addr,
3466 				      (void **)&fw_data_ptr);
3467 	if (r) {
3468 		dev_err(adev->dev, "(%d) failed to create mec fw ucode bo\n", r);
3469 		gfx_v11_0_mec_fini(adev);
3470 		return r;
3471 	}
3472 
3473 	memcpy(fw_ucode_ptr, fw_ucode, fw_ucode_size);
3474 	memcpy(fw_data_ptr, fw_data, fw_data_size);
3475 
3476 	amdgpu_bo_kunmap(adev->gfx.mec.mec_fw_obj);
3477 	amdgpu_bo_kunmap(adev->gfx.mec.mec_fw_data_obj);
3478 	amdgpu_bo_unreserve(adev->gfx.mec.mec_fw_obj);
3479 	amdgpu_bo_unreserve(adev->gfx.mec.mec_fw_data_obj);
3480 
3481 	tmp = RREG32_SOC15(GC, 0, regCP_CPC_IC_BASE_CNTL);
3482 	tmp = REG_SET_FIELD(tmp, CP_CPC_IC_BASE_CNTL, VMID, 0);
3483 	tmp = REG_SET_FIELD(tmp, CP_CPC_IC_BASE_CNTL, EXE_DISABLE, 0);
3484 	tmp = REG_SET_FIELD(tmp, CP_CPC_IC_BASE_CNTL, CACHE_POLICY, 0);
3485 	WREG32_SOC15(GC, 0, regCP_CPC_IC_BASE_CNTL, tmp);
3486 
3487 	tmp = RREG32_SOC15(GC, 0, regCP_MEC_DC_BASE_CNTL);
3488 	tmp = REG_SET_FIELD(tmp, CP_MEC_DC_BASE_CNTL, VMID, 0);
3489 	tmp = REG_SET_FIELD(tmp, CP_MEC_DC_BASE_CNTL, CACHE_POLICY, 0);
3490 	WREG32_SOC15(GC, 0, regCP_MEC_DC_BASE_CNTL, tmp);
3491 
3492 	mutex_lock(&adev->srbm_mutex);
3493 	for (i = 0; i < adev->gfx.mec.num_pipe_per_mec; i++) {
3494 		soc21_grbm_select(adev, 1, i, 0, 0);
3495 
3496 		WREG32_SOC15(GC, 0, regCP_MEC_MDBASE_LO, adev->gfx.mec.mec_fw_data_gpu_addr);
3497 		WREG32_SOC15(GC, 0, regCP_MEC_MDBASE_HI,
3498 		     upper_32_bits(adev->gfx.mec.mec_fw_data_gpu_addr));
3499 
3500 		WREG32_SOC15(GC, 0, regCP_MEC_RS64_PRGRM_CNTR_START,
3501 					mec_hdr->ucode_start_addr_lo >> 2 |
3502 					mec_hdr->ucode_start_addr_hi << 30);
3503 		WREG32_SOC15(GC, 0, regCP_MEC_RS64_PRGRM_CNTR_START_HI,
3504 					mec_hdr->ucode_start_addr_hi >> 2);
3505 
3506 		WREG32_SOC15(GC, 0, regCP_CPC_IC_BASE_LO, adev->gfx.mec.mec_fw_gpu_addr);
3507 		WREG32_SOC15(GC, 0, regCP_CPC_IC_BASE_HI,
3508 		     upper_32_bits(adev->gfx.mec.mec_fw_gpu_addr));
3509 	}
3510 	mutex_unlock(&adev->srbm_mutex);
3511 	soc21_grbm_select(adev, 0, 0, 0, 0);
3512 
3513 	/* Trigger an invalidation of the L1 instruction caches */
3514 	tmp = RREG32_SOC15(GC, 0, regCP_MEC_DC_OP_CNTL);
3515 	tmp = REG_SET_FIELD(tmp, CP_MEC_DC_OP_CNTL, INVALIDATE_DCACHE, 1);
3516 	WREG32_SOC15(GC, 0, regCP_MEC_DC_OP_CNTL, tmp);
3517 
3518 	/* Wait for invalidation complete */
3519 	for (i = 0; i < usec_timeout; i++) {
3520 		tmp = RREG32_SOC15(GC, 0, regCP_MEC_DC_OP_CNTL);
3521 		if (1 == REG_GET_FIELD(tmp, CP_MEC_DC_OP_CNTL,
3522 				       INVALIDATE_DCACHE_COMPLETE))
3523 			break;
3524 		udelay(1);
3525 	}
3526 
3527 	if (i >= usec_timeout) {
3528 		dev_err(adev->dev, "failed to invalidate instruction cache\n");
3529 		return -EINVAL;
3530 	}
3531 
3532 	/* Trigger an invalidation of the L1 instruction caches */
3533 	tmp = RREG32_SOC15(GC, 0, regCP_CPC_IC_OP_CNTL);
3534 	tmp = REG_SET_FIELD(tmp, CP_CPC_IC_OP_CNTL, INVALIDATE_CACHE, 1);
3535 	WREG32_SOC15(GC, 0, regCP_CPC_IC_OP_CNTL, tmp);
3536 
3537 	/* Wait for invalidation complete */
3538 	for (i = 0; i < usec_timeout; i++) {
3539 		tmp = RREG32_SOC15(GC, 0, regCP_CPC_IC_OP_CNTL);
3540 		if (1 == REG_GET_FIELD(tmp, CP_CPC_IC_OP_CNTL,
3541 				       INVALIDATE_CACHE_COMPLETE))
3542 			break;
3543 		udelay(1);
3544 	}
3545 
3546 	if (i >= usec_timeout) {
3547 		dev_err(adev->dev, "failed to invalidate instruction cache\n");
3548 		return -EINVAL;
3549 	}
3550 
3551 	return 0;
3552 }
3553 
3554 static void gfx_v11_0_kiq_setting(struct amdgpu_ring *ring)
3555 {
3556 	uint32_t tmp;
3557 	struct amdgpu_device *adev = ring->adev;
3558 
3559 	/* tell RLC which is KIQ queue */
3560 	tmp = RREG32_SOC15(GC, 0, regRLC_CP_SCHEDULERS);
3561 	tmp &= 0xffffff00;
3562 	tmp |= (ring->me << 5) | (ring->pipe << 3) | (ring->queue);
3563 	WREG32_SOC15(GC, 0, regRLC_CP_SCHEDULERS, tmp);
3564 	tmp |= 0x80;
3565 	WREG32_SOC15(GC, 0, regRLC_CP_SCHEDULERS, tmp);
3566 }
3567 
3568 static void gfx_v11_0_cp_set_doorbell_range(struct amdgpu_device *adev)
3569 {
3570 	/* set graphics engine doorbell range */
3571 	WREG32_SOC15(GC, 0, regCP_RB_DOORBELL_RANGE_LOWER,
3572 		     (adev->doorbell_index.gfx_ring0 * 2) << 2);
3573 	WREG32_SOC15(GC, 0, regCP_RB_DOORBELL_RANGE_UPPER,
3574 		     (adev->doorbell_index.gfx_userqueue_end * 2) << 2);
3575 
3576 	/* set compute engine doorbell range */
3577 	WREG32_SOC15(GC, 0, regCP_MEC_DOORBELL_RANGE_LOWER,
3578 		     (adev->doorbell_index.kiq * 2) << 2);
3579 	WREG32_SOC15(GC, 0, regCP_MEC_DOORBELL_RANGE_UPPER,
3580 		     (adev->doorbell_index.userqueue_end * 2) << 2);
3581 }
3582 
3583 static int gfx_v11_0_gfx_mqd_init(struct amdgpu_device *adev, void *m,
3584 				  struct amdgpu_mqd_prop *prop)
3585 {
3586 	struct v11_gfx_mqd *mqd = m;
3587 	uint64_t hqd_gpu_addr, wb_gpu_addr;
3588 	uint32_t tmp;
3589 	uint32_t rb_bufsz;
3590 
3591 	/* set up gfx hqd wptr */
3592 	mqd->cp_gfx_hqd_wptr = 0;
3593 	mqd->cp_gfx_hqd_wptr_hi = 0;
3594 
3595 	/* set the pointer to the MQD */
3596 	mqd->cp_mqd_base_addr = prop->mqd_gpu_addr & 0xfffffffc;
3597 	mqd->cp_mqd_base_addr_hi = upper_32_bits(prop->mqd_gpu_addr);
3598 
3599 	/* set up mqd control */
3600 	tmp = RREG32_SOC15(GC, 0, regCP_GFX_MQD_CONTROL);
3601 	tmp = REG_SET_FIELD(tmp, CP_GFX_MQD_CONTROL, VMID, 0);
3602 	tmp = REG_SET_FIELD(tmp, CP_GFX_MQD_CONTROL, PRIV_STATE, 1);
3603 	tmp = REG_SET_FIELD(tmp, CP_GFX_MQD_CONTROL, CACHE_POLICY, 0);
3604 	mqd->cp_gfx_mqd_control = tmp;
3605 
3606 	/* set up gfx_hqd_vimd with 0x0 to indicate the ring buffer's vmid */
3607 	tmp = RREG32_SOC15(GC, 0, regCP_GFX_HQD_VMID);
3608 	tmp = REG_SET_FIELD(tmp, CP_GFX_HQD_VMID, VMID, 0);
3609 	mqd->cp_gfx_hqd_vmid = 0;
3610 
3611 	/* set up default queue priority level
3612 	 * 0x0 = low priority, 0x1 = high priority */
3613 	tmp = RREG32_SOC15(GC, 0, regCP_GFX_HQD_QUEUE_PRIORITY);
3614 	tmp = REG_SET_FIELD(tmp, CP_GFX_HQD_QUEUE_PRIORITY, PRIORITY_LEVEL, 0);
3615 	mqd->cp_gfx_hqd_queue_priority = tmp;
3616 
3617 	/* set up time quantum */
3618 	tmp = RREG32_SOC15(GC, 0, regCP_GFX_HQD_QUANTUM);
3619 	tmp = REG_SET_FIELD(tmp, CP_GFX_HQD_QUANTUM, QUANTUM_EN, 1);
3620 	mqd->cp_gfx_hqd_quantum = tmp;
3621 
3622 	/* set up gfx hqd base. this is similar as CP_RB_BASE */
3623 	hqd_gpu_addr = prop->hqd_base_gpu_addr >> 8;
3624 	mqd->cp_gfx_hqd_base = hqd_gpu_addr;
3625 	mqd->cp_gfx_hqd_base_hi = upper_32_bits(hqd_gpu_addr);
3626 
3627 	/* set up hqd_rptr_addr/_hi, similar as CP_RB_RPTR */
3628 	wb_gpu_addr = prop->rptr_gpu_addr;
3629 	mqd->cp_gfx_hqd_rptr_addr = wb_gpu_addr & 0xfffffffc;
3630 	mqd->cp_gfx_hqd_rptr_addr_hi =
3631 		upper_32_bits(wb_gpu_addr) & 0xffff;
3632 
3633 	/* set up rb_wptr_poll addr */
3634 	wb_gpu_addr = prop->wptr_gpu_addr;
3635 	mqd->cp_rb_wptr_poll_addr_lo = wb_gpu_addr & 0xfffffffc;
3636 	mqd->cp_rb_wptr_poll_addr_hi = upper_32_bits(wb_gpu_addr) & 0xffff;
3637 
3638 	/* set up the gfx_hqd_control, similar as CP_RB0_CNTL */
3639 	rb_bufsz = order_base_2(prop->queue_size / 4) - 1;
3640 	tmp = RREG32_SOC15(GC, 0, regCP_GFX_HQD_CNTL);
3641 	tmp = REG_SET_FIELD(tmp, CP_GFX_HQD_CNTL, RB_BUFSZ, rb_bufsz);
3642 	tmp = REG_SET_FIELD(tmp, CP_GFX_HQD_CNTL, RB_BLKSZ, rb_bufsz - 2);
3643 #ifdef __BIG_ENDIAN
3644 	tmp = REG_SET_FIELD(tmp, CP_GFX_HQD_CNTL, BUF_SWAP, 1);
3645 #endif
3646 	mqd->cp_gfx_hqd_cntl = tmp;
3647 
3648 	/* set up cp_doorbell_control */
3649 	tmp = RREG32_SOC15(GC, 0, regCP_RB_DOORBELL_CONTROL);
3650 	if (prop->use_doorbell) {
3651 		tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL,
3652 				    DOORBELL_OFFSET, prop->doorbell_index);
3653 		tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL,
3654 				    DOORBELL_EN, 1);
3655 	} else
3656 		tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL,
3657 				    DOORBELL_EN, 0);
3658 	mqd->cp_rb_doorbell_control = tmp;
3659 
3660 	/* reset read and write pointers, similar to CP_RB0_WPTR/_RPTR */
3661 	mqd->cp_gfx_hqd_rptr = RREG32_SOC15(GC, 0, regCP_GFX_HQD_RPTR);
3662 
3663 	/* active the queue */
3664 	mqd->cp_gfx_hqd_active = 1;
3665 
3666 	return 0;
3667 }
3668 
3669 static int gfx_v11_0_gfx_init_queue(struct amdgpu_ring *ring)
3670 {
3671 	struct amdgpu_device *adev = ring->adev;
3672 	struct v11_gfx_mqd *mqd = ring->mqd_ptr;
3673 	int mqd_idx = ring - &adev->gfx.gfx_ring[0];
3674 
3675 	if (!amdgpu_in_reset(adev) && !adev->in_suspend) {
3676 		memset((void *)mqd, 0, sizeof(*mqd));
3677 		mutex_lock(&adev->srbm_mutex);
3678 		soc21_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
3679 		amdgpu_ring_init_mqd(ring);
3680 		soc21_grbm_select(adev, 0, 0, 0, 0);
3681 		mutex_unlock(&adev->srbm_mutex);
3682 		if (adev->gfx.me.mqd_backup[mqd_idx])
3683 			memcpy(adev->gfx.me.mqd_backup[mqd_idx], mqd, sizeof(*mqd));
3684 	} else {
3685 		/* restore mqd with the backup copy */
3686 		if (adev->gfx.me.mqd_backup[mqd_idx])
3687 			memcpy(mqd, adev->gfx.me.mqd_backup[mqd_idx], sizeof(*mqd));
3688 		/* reset the ring */
3689 		ring->wptr = 0;
3690 		*ring->wptr_cpu_addr = 0;
3691 		amdgpu_ring_clear_ring(ring);
3692 	}
3693 
3694 	return 0;
3695 }
3696 
3697 static int gfx_v11_0_cp_async_gfx_ring_resume(struct amdgpu_device *adev)
3698 {
3699 	int r, i;
3700 	struct amdgpu_ring *ring;
3701 
3702 	for (i = 0; i < adev->gfx.num_gfx_rings; i++) {
3703 		ring = &adev->gfx.gfx_ring[i];
3704 
3705 		r = amdgpu_bo_reserve(ring->mqd_obj, false);
3706 		if (unlikely(r != 0))
3707 			return r;
3708 
3709 		r = amdgpu_bo_kmap(ring->mqd_obj, (void **)&ring->mqd_ptr);
3710 		if (!r) {
3711 			r = gfx_v11_0_gfx_init_queue(ring);
3712 			amdgpu_bo_kunmap(ring->mqd_obj);
3713 			ring->mqd_ptr = NULL;
3714 		}
3715 		amdgpu_bo_unreserve(ring->mqd_obj);
3716 		if (r)
3717 			return r;
3718 	}
3719 
3720 	r = amdgpu_gfx_enable_kgq(adev, 0);
3721 	if (r)
3722 		return r;
3723 
3724 	return gfx_v11_0_cp_gfx_start(adev);
3725 }
3726 
3727 static int gfx_v11_0_compute_mqd_init(struct amdgpu_device *adev, void *m,
3728 				      struct amdgpu_mqd_prop *prop)
3729 {
3730 	struct v11_compute_mqd *mqd = m;
3731 	uint64_t hqd_gpu_addr, wb_gpu_addr, eop_base_addr;
3732 	uint32_t tmp;
3733 
3734 	mqd->header = 0xC0310800;
3735 	mqd->compute_pipelinestat_enable = 0x00000001;
3736 	mqd->compute_static_thread_mgmt_se0 = 0xffffffff;
3737 	mqd->compute_static_thread_mgmt_se1 = 0xffffffff;
3738 	mqd->compute_static_thread_mgmt_se2 = 0xffffffff;
3739 	mqd->compute_static_thread_mgmt_se3 = 0xffffffff;
3740 	mqd->compute_misc_reserved = 0x00000007;
3741 
3742 	eop_base_addr = prop->eop_gpu_addr >> 8;
3743 	mqd->cp_hqd_eop_base_addr_lo = eop_base_addr;
3744 	mqd->cp_hqd_eop_base_addr_hi = upper_32_bits(eop_base_addr);
3745 
3746 	/* set the EOP size, register value is 2^(EOP_SIZE+1) dwords */
3747 	tmp = RREG32_SOC15(GC, 0, regCP_HQD_EOP_CONTROL);
3748 	tmp = REG_SET_FIELD(tmp, CP_HQD_EOP_CONTROL, EOP_SIZE,
3749 			(order_base_2(GFX11_MEC_HPD_SIZE / 4) - 1));
3750 
3751 	mqd->cp_hqd_eop_control = tmp;
3752 
3753 	/* enable doorbell? */
3754 	tmp = RREG32_SOC15(GC, 0, regCP_HQD_PQ_DOORBELL_CONTROL);
3755 
3756 	if (prop->use_doorbell) {
3757 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3758 				    DOORBELL_OFFSET, prop->doorbell_index);
3759 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3760 				    DOORBELL_EN, 1);
3761 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3762 				    DOORBELL_SOURCE, 0);
3763 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3764 				    DOORBELL_HIT, 0);
3765 	} else {
3766 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3767 				    DOORBELL_EN, 0);
3768 	}
3769 
3770 	mqd->cp_hqd_pq_doorbell_control = tmp;
3771 
3772 	/* disable the queue if it's active */
3773 	mqd->cp_hqd_dequeue_request = 0;
3774 	mqd->cp_hqd_pq_rptr = 0;
3775 	mqd->cp_hqd_pq_wptr_lo = 0;
3776 	mqd->cp_hqd_pq_wptr_hi = 0;
3777 
3778 	/* set the pointer to the MQD */
3779 	mqd->cp_mqd_base_addr_lo = prop->mqd_gpu_addr & 0xfffffffc;
3780 	mqd->cp_mqd_base_addr_hi = upper_32_bits(prop->mqd_gpu_addr);
3781 
3782 	/* set MQD vmid to 0 */
3783 	tmp = RREG32_SOC15(GC, 0, regCP_MQD_CONTROL);
3784 	tmp = REG_SET_FIELD(tmp, CP_MQD_CONTROL, VMID, 0);
3785 	mqd->cp_mqd_control = tmp;
3786 
3787 	/* set the pointer to the HQD, this is similar CP_RB0_BASE/_HI */
3788 	hqd_gpu_addr = prop->hqd_base_gpu_addr >> 8;
3789 	mqd->cp_hqd_pq_base_lo = hqd_gpu_addr;
3790 	mqd->cp_hqd_pq_base_hi = upper_32_bits(hqd_gpu_addr);
3791 
3792 	/* set up the HQD, this is similar to CP_RB0_CNTL */
3793 	tmp = RREG32_SOC15(GC, 0, regCP_HQD_PQ_CONTROL);
3794 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, QUEUE_SIZE,
3795 			    (order_base_2(prop->queue_size / 4) - 1));
3796 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, RPTR_BLOCK_SIZE,
3797 			    (order_base_2(AMDGPU_GPU_PAGE_SIZE / 4) - 1));
3798 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, UNORD_DISPATCH, 0);
3799 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, TUNNEL_DISPATCH, 0);
3800 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, PRIV_STATE, 1);
3801 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, KMD_QUEUE, 1);
3802 	mqd->cp_hqd_pq_control = tmp;
3803 
3804 	/* set the wb address whether it's enabled or not */
3805 	wb_gpu_addr = prop->rptr_gpu_addr;
3806 	mqd->cp_hqd_pq_rptr_report_addr_lo = wb_gpu_addr & 0xfffffffc;
3807 	mqd->cp_hqd_pq_rptr_report_addr_hi =
3808 		upper_32_bits(wb_gpu_addr) & 0xffff;
3809 
3810 	/* only used if CP_PQ_WPTR_POLL_CNTL.CP_PQ_WPTR_POLL_CNTL__EN_MASK=1 */
3811 	wb_gpu_addr = prop->wptr_gpu_addr;
3812 	mqd->cp_hqd_pq_wptr_poll_addr_lo = wb_gpu_addr & 0xfffffffc;
3813 	mqd->cp_hqd_pq_wptr_poll_addr_hi = upper_32_bits(wb_gpu_addr) & 0xffff;
3814 
3815 	tmp = 0;
3816 	/* enable the doorbell if requested */
3817 	if (prop->use_doorbell) {
3818 		tmp = RREG32_SOC15(GC, 0, regCP_HQD_PQ_DOORBELL_CONTROL);
3819 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3820 				DOORBELL_OFFSET, prop->doorbell_index);
3821 
3822 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3823 				    DOORBELL_EN, 1);
3824 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3825 				    DOORBELL_SOURCE, 0);
3826 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3827 				    DOORBELL_HIT, 0);
3828 	}
3829 
3830 	mqd->cp_hqd_pq_doorbell_control = tmp;
3831 
3832 	/* reset read and write pointers, similar to CP_RB0_WPTR/_RPTR */
3833 	mqd->cp_hqd_pq_rptr = RREG32_SOC15(GC, 0, regCP_HQD_PQ_RPTR);
3834 
3835 	/* set the vmid for the queue */
3836 	mqd->cp_hqd_vmid = 0;
3837 
3838 	tmp = RREG32_SOC15(GC, 0, regCP_HQD_PERSISTENT_STATE);
3839 	tmp = REG_SET_FIELD(tmp, CP_HQD_PERSISTENT_STATE, PRELOAD_SIZE, 0x55);
3840 	mqd->cp_hqd_persistent_state = tmp;
3841 
3842 	/* set MIN_IB_AVAIL_SIZE */
3843 	tmp = RREG32_SOC15(GC, 0, regCP_HQD_IB_CONTROL);
3844 	tmp = REG_SET_FIELD(tmp, CP_HQD_IB_CONTROL, MIN_IB_AVAIL_SIZE, 3);
3845 	mqd->cp_hqd_ib_control = tmp;
3846 
3847 	/* set static priority for a compute queue/ring */
3848 	mqd->cp_hqd_pipe_priority = prop->hqd_pipe_priority;
3849 	mqd->cp_hqd_queue_priority = prop->hqd_queue_priority;
3850 
3851 	mqd->cp_hqd_active = prop->hqd_active;
3852 
3853 	return 0;
3854 }
3855 
3856 static int gfx_v11_0_kiq_init_register(struct amdgpu_ring *ring)
3857 {
3858 	struct amdgpu_device *adev = ring->adev;
3859 	struct v11_compute_mqd *mqd = ring->mqd_ptr;
3860 	int j;
3861 
3862 	/* inactivate the queue */
3863 	if (amdgpu_sriov_vf(adev))
3864 		WREG32_SOC15(GC, 0, regCP_HQD_ACTIVE, 0);
3865 
3866 	/* disable wptr polling */
3867 	WREG32_FIELD15_PREREG(GC, 0, CP_PQ_WPTR_POLL_CNTL, EN, 0);
3868 
3869 	/* write the EOP addr */
3870 	WREG32_SOC15(GC, 0, regCP_HQD_EOP_BASE_ADDR,
3871 	       mqd->cp_hqd_eop_base_addr_lo);
3872 	WREG32_SOC15(GC, 0, regCP_HQD_EOP_BASE_ADDR_HI,
3873 	       mqd->cp_hqd_eop_base_addr_hi);
3874 
3875 	/* set the EOP size, register value is 2^(EOP_SIZE+1) dwords */
3876 	WREG32_SOC15(GC, 0, regCP_HQD_EOP_CONTROL,
3877 	       mqd->cp_hqd_eop_control);
3878 
3879 	/* enable doorbell? */
3880 	WREG32_SOC15(GC, 0, regCP_HQD_PQ_DOORBELL_CONTROL,
3881 	       mqd->cp_hqd_pq_doorbell_control);
3882 
3883 	/* disable the queue if it's active */
3884 	if (RREG32_SOC15(GC, 0, regCP_HQD_ACTIVE) & 1) {
3885 		WREG32_SOC15(GC, 0, regCP_HQD_DEQUEUE_REQUEST, 1);
3886 		for (j = 0; j < adev->usec_timeout; j++) {
3887 			if (!(RREG32_SOC15(GC, 0, regCP_HQD_ACTIVE) & 1))
3888 				break;
3889 			udelay(1);
3890 		}
3891 		WREG32_SOC15(GC, 0, regCP_HQD_DEQUEUE_REQUEST,
3892 		       mqd->cp_hqd_dequeue_request);
3893 		WREG32_SOC15(GC, 0, regCP_HQD_PQ_RPTR,
3894 		       mqd->cp_hqd_pq_rptr);
3895 		WREG32_SOC15(GC, 0, regCP_HQD_PQ_WPTR_LO,
3896 		       mqd->cp_hqd_pq_wptr_lo);
3897 		WREG32_SOC15(GC, 0, regCP_HQD_PQ_WPTR_HI,
3898 		       mqd->cp_hqd_pq_wptr_hi);
3899 	}
3900 
3901 	/* set the pointer to the MQD */
3902 	WREG32_SOC15(GC, 0, regCP_MQD_BASE_ADDR,
3903 	       mqd->cp_mqd_base_addr_lo);
3904 	WREG32_SOC15(GC, 0, regCP_MQD_BASE_ADDR_HI,
3905 	       mqd->cp_mqd_base_addr_hi);
3906 
3907 	/* set MQD vmid to 0 */
3908 	WREG32_SOC15(GC, 0, regCP_MQD_CONTROL,
3909 	       mqd->cp_mqd_control);
3910 
3911 	/* set the pointer to the HQD, this is similar CP_RB0_BASE/_HI */
3912 	WREG32_SOC15(GC, 0, regCP_HQD_PQ_BASE,
3913 	       mqd->cp_hqd_pq_base_lo);
3914 	WREG32_SOC15(GC, 0, regCP_HQD_PQ_BASE_HI,
3915 	       mqd->cp_hqd_pq_base_hi);
3916 
3917 	/* set up the HQD, this is similar to CP_RB0_CNTL */
3918 	WREG32_SOC15(GC, 0, regCP_HQD_PQ_CONTROL,
3919 	       mqd->cp_hqd_pq_control);
3920 
3921 	/* set the wb address whether it's enabled or not */
3922 	WREG32_SOC15(GC, 0, regCP_HQD_PQ_RPTR_REPORT_ADDR,
3923 		mqd->cp_hqd_pq_rptr_report_addr_lo);
3924 	WREG32_SOC15(GC, 0, regCP_HQD_PQ_RPTR_REPORT_ADDR_HI,
3925 		mqd->cp_hqd_pq_rptr_report_addr_hi);
3926 
3927 	/* only used if CP_PQ_WPTR_POLL_CNTL.CP_PQ_WPTR_POLL_CNTL__EN_MASK=1 */
3928 	WREG32_SOC15(GC, 0, regCP_HQD_PQ_WPTR_POLL_ADDR,
3929 	       mqd->cp_hqd_pq_wptr_poll_addr_lo);
3930 	WREG32_SOC15(GC, 0, regCP_HQD_PQ_WPTR_POLL_ADDR_HI,
3931 	       mqd->cp_hqd_pq_wptr_poll_addr_hi);
3932 
3933 	/* enable the doorbell if requested */
3934 	if (ring->use_doorbell) {
3935 		WREG32_SOC15(GC, 0, regCP_MEC_DOORBELL_RANGE_LOWER,
3936 			(adev->doorbell_index.kiq * 2) << 2);
3937 		WREG32_SOC15(GC, 0, regCP_MEC_DOORBELL_RANGE_UPPER,
3938 			(adev->doorbell_index.userqueue_end * 2) << 2);
3939 	}
3940 
3941 	WREG32_SOC15(GC, 0, regCP_HQD_PQ_DOORBELL_CONTROL,
3942 	       mqd->cp_hqd_pq_doorbell_control);
3943 
3944 	/* reset read and write pointers, similar to CP_RB0_WPTR/_RPTR */
3945 	WREG32_SOC15(GC, 0, regCP_HQD_PQ_WPTR_LO,
3946 	       mqd->cp_hqd_pq_wptr_lo);
3947 	WREG32_SOC15(GC, 0, regCP_HQD_PQ_WPTR_HI,
3948 	       mqd->cp_hqd_pq_wptr_hi);
3949 
3950 	/* set the vmid for the queue */
3951 	WREG32_SOC15(GC, 0, regCP_HQD_VMID, mqd->cp_hqd_vmid);
3952 
3953 	WREG32_SOC15(GC, 0, regCP_HQD_PERSISTENT_STATE,
3954 	       mqd->cp_hqd_persistent_state);
3955 
3956 	/* activate the queue */
3957 	WREG32_SOC15(GC, 0, regCP_HQD_ACTIVE,
3958 	       mqd->cp_hqd_active);
3959 
3960 	if (ring->use_doorbell)
3961 		WREG32_FIELD15_PREREG(GC, 0, CP_PQ_STATUS, DOORBELL_ENABLE, 1);
3962 
3963 	return 0;
3964 }
3965 
3966 static int gfx_v11_0_kiq_init_queue(struct amdgpu_ring *ring)
3967 {
3968 	struct amdgpu_device *adev = ring->adev;
3969 	struct v11_compute_mqd *mqd = ring->mqd_ptr;
3970 
3971 	gfx_v11_0_kiq_setting(ring);
3972 
3973 	if (amdgpu_in_reset(adev)) { /* for GPU_RESET case */
3974 		/* reset MQD to a clean status */
3975 		if (adev->gfx.kiq[0].mqd_backup)
3976 			memcpy(mqd, adev->gfx.kiq[0].mqd_backup, sizeof(*mqd));
3977 
3978 		/* reset ring buffer */
3979 		ring->wptr = 0;
3980 		amdgpu_ring_clear_ring(ring);
3981 
3982 		mutex_lock(&adev->srbm_mutex);
3983 		soc21_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
3984 		gfx_v11_0_kiq_init_register(ring);
3985 		soc21_grbm_select(adev, 0, 0, 0, 0);
3986 		mutex_unlock(&adev->srbm_mutex);
3987 	} else {
3988 		memset((void *)mqd, 0, sizeof(*mqd));
3989 		if (amdgpu_sriov_vf(adev) && adev->in_suspend)
3990 			amdgpu_ring_clear_ring(ring);
3991 		mutex_lock(&adev->srbm_mutex);
3992 		soc21_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
3993 		amdgpu_ring_init_mqd(ring);
3994 		gfx_v11_0_kiq_init_register(ring);
3995 		soc21_grbm_select(adev, 0, 0, 0, 0);
3996 		mutex_unlock(&adev->srbm_mutex);
3997 
3998 		if (adev->gfx.kiq[0].mqd_backup)
3999 			memcpy(adev->gfx.kiq[0].mqd_backup, mqd, sizeof(*mqd));
4000 	}
4001 
4002 	return 0;
4003 }
4004 
4005 static int gfx_v11_0_kcq_init_queue(struct amdgpu_ring *ring)
4006 {
4007 	struct amdgpu_device *adev = ring->adev;
4008 	struct v11_compute_mqd *mqd = ring->mqd_ptr;
4009 	int mqd_idx = ring - &adev->gfx.compute_ring[0];
4010 
4011 	if (!amdgpu_in_reset(adev) && !adev->in_suspend) {
4012 		memset((void *)mqd, 0, sizeof(*mqd));
4013 		mutex_lock(&adev->srbm_mutex);
4014 		soc21_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
4015 		amdgpu_ring_init_mqd(ring);
4016 		soc21_grbm_select(adev, 0, 0, 0, 0);
4017 		mutex_unlock(&adev->srbm_mutex);
4018 
4019 		if (adev->gfx.mec.mqd_backup[mqd_idx])
4020 			memcpy(adev->gfx.mec.mqd_backup[mqd_idx], mqd, sizeof(*mqd));
4021 	} else {
4022 		/* restore MQD to a clean status */
4023 		if (adev->gfx.mec.mqd_backup[mqd_idx])
4024 			memcpy(mqd, adev->gfx.mec.mqd_backup[mqd_idx], sizeof(*mqd));
4025 		/* reset ring buffer */
4026 		ring->wptr = 0;
4027 		atomic64_set((atomic64_t *)ring->wptr_cpu_addr, 0);
4028 		amdgpu_ring_clear_ring(ring);
4029 	}
4030 
4031 	return 0;
4032 }
4033 
4034 static int gfx_v11_0_kiq_resume(struct amdgpu_device *adev)
4035 {
4036 	struct amdgpu_ring *ring;
4037 	int r;
4038 
4039 	ring = &adev->gfx.kiq[0].ring;
4040 
4041 	r = amdgpu_bo_reserve(ring->mqd_obj, false);
4042 	if (unlikely(r != 0))
4043 		return r;
4044 
4045 	r = amdgpu_bo_kmap(ring->mqd_obj, (void **)&ring->mqd_ptr);
4046 	if (unlikely(r != 0)) {
4047 		amdgpu_bo_unreserve(ring->mqd_obj);
4048 		return r;
4049 	}
4050 
4051 	gfx_v11_0_kiq_init_queue(ring);
4052 	amdgpu_bo_kunmap(ring->mqd_obj);
4053 	ring->mqd_ptr = NULL;
4054 	amdgpu_bo_unreserve(ring->mqd_obj);
4055 	ring->sched.ready = true;
4056 	return 0;
4057 }
4058 
4059 static int gfx_v11_0_kcq_resume(struct amdgpu_device *adev)
4060 {
4061 	struct amdgpu_ring *ring = NULL;
4062 	int r = 0, i;
4063 
4064 	if (!amdgpu_async_gfx_ring)
4065 		gfx_v11_0_cp_compute_enable(adev, true);
4066 
4067 	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
4068 		ring = &adev->gfx.compute_ring[i];
4069 
4070 		r = amdgpu_bo_reserve(ring->mqd_obj, false);
4071 		if (unlikely(r != 0))
4072 			goto done;
4073 		r = amdgpu_bo_kmap(ring->mqd_obj, (void **)&ring->mqd_ptr);
4074 		if (!r) {
4075 			r = gfx_v11_0_kcq_init_queue(ring);
4076 			amdgpu_bo_kunmap(ring->mqd_obj);
4077 			ring->mqd_ptr = NULL;
4078 		}
4079 		amdgpu_bo_unreserve(ring->mqd_obj);
4080 		if (r)
4081 			goto done;
4082 	}
4083 
4084 	r = amdgpu_gfx_enable_kcq(adev, 0);
4085 done:
4086 	return r;
4087 }
4088 
4089 static int gfx_v11_0_cp_resume(struct amdgpu_device *adev)
4090 {
4091 	int r, i;
4092 	struct amdgpu_ring *ring;
4093 
4094 	if (!(adev->flags & AMD_IS_APU))
4095 		gfx_v11_0_enable_gui_idle_interrupt(adev, false);
4096 
4097 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_DIRECT) {
4098 		/* legacy firmware loading */
4099 		r = gfx_v11_0_cp_gfx_load_microcode(adev);
4100 		if (r)
4101 			return r;
4102 
4103 		if (adev->gfx.rs64_enable)
4104 			r = gfx_v11_0_cp_compute_load_microcode_rs64(adev);
4105 		else
4106 			r = gfx_v11_0_cp_compute_load_microcode(adev);
4107 		if (r)
4108 			return r;
4109 	}
4110 
4111 	gfx_v11_0_cp_set_doorbell_range(adev);
4112 
4113 	if (amdgpu_async_gfx_ring) {
4114 		gfx_v11_0_cp_compute_enable(adev, true);
4115 		gfx_v11_0_cp_gfx_enable(adev, true);
4116 	}
4117 
4118 	if (adev->enable_mes_kiq && adev->mes.kiq_hw_init)
4119 		r = amdgpu_mes_kiq_hw_init(adev);
4120 	else
4121 		r = gfx_v11_0_kiq_resume(adev);
4122 	if (r)
4123 		return r;
4124 
4125 	r = gfx_v11_0_kcq_resume(adev);
4126 	if (r)
4127 		return r;
4128 
4129 	if (!amdgpu_async_gfx_ring) {
4130 		r = gfx_v11_0_cp_gfx_resume(adev);
4131 		if (r)
4132 			return r;
4133 	} else {
4134 		r = gfx_v11_0_cp_async_gfx_ring_resume(adev);
4135 		if (r)
4136 			return r;
4137 	}
4138 
4139 	for (i = 0; i < adev->gfx.num_gfx_rings; i++) {
4140 		ring = &adev->gfx.gfx_ring[i];
4141 		r = amdgpu_ring_test_helper(ring);
4142 		if (r)
4143 			return r;
4144 	}
4145 
4146 	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
4147 		ring = &adev->gfx.compute_ring[i];
4148 		r = amdgpu_ring_test_helper(ring);
4149 		if (r)
4150 			return r;
4151 	}
4152 
4153 	return 0;
4154 }
4155 
4156 static void gfx_v11_0_cp_enable(struct amdgpu_device *adev, bool enable)
4157 {
4158 	gfx_v11_0_cp_gfx_enable(adev, enable);
4159 	gfx_v11_0_cp_compute_enable(adev, enable);
4160 }
4161 
4162 static int gfx_v11_0_gfxhub_enable(struct amdgpu_device *adev)
4163 {
4164 	int r;
4165 	bool value;
4166 
4167 	r = adev->gfxhub.funcs->gart_enable(adev);
4168 	if (r)
4169 		return r;
4170 
4171 	adev->hdp.funcs->flush_hdp(adev, NULL);
4172 
4173 	value = (amdgpu_vm_fault_stop == AMDGPU_VM_FAULT_STOP_ALWAYS) ?
4174 		false : true;
4175 
4176 	adev->gfxhub.funcs->set_fault_enable_default(adev, value);
4177 	amdgpu_gmc_flush_gpu_tlb(adev, 0, AMDGPU_GFXHUB(0), 0);
4178 
4179 	return 0;
4180 }
4181 
4182 static void gfx_v11_0_select_cp_fw_arch(struct amdgpu_device *adev)
4183 {
4184 	u32 tmp;
4185 
4186 	/* select RS64 */
4187 	if (adev->gfx.rs64_enable) {
4188 		tmp = RREG32_SOC15(GC, 0, regCP_GFX_CNTL);
4189 		tmp = REG_SET_FIELD(tmp, CP_GFX_CNTL, ENGINE_SEL, 1);
4190 		WREG32_SOC15(GC, 0, regCP_GFX_CNTL, tmp);
4191 
4192 		tmp = RREG32_SOC15(GC, 0, regCP_MEC_ISA_CNTL);
4193 		tmp = REG_SET_FIELD(tmp, CP_MEC_ISA_CNTL, ISA_MODE, 1);
4194 		WREG32_SOC15(GC, 0, regCP_MEC_ISA_CNTL, tmp);
4195 	}
4196 
4197 	if (amdgpu_emu_mode == 1)
4198 		msleep(100);
4199 }
4200 
4201 static int get_gb_addr_config(struct amdgpu_device * adev)
4202 {
4203 	u32 gb_addr_config;
4204 
4205 	gb_addr_config = RREG32_SOC15(GC, 0, regGB_ADDR_CONFIG);
4206 	if (gb_addr_config == 0)
4207 		return -EINVAL;
4208 
4209 	adev->gfx.config.gb_addr_config_fields.num_pkrs =
4210 		1 << REG_GET_FIELD(gb_addr_config, GB_ADDR_CONFIG, NUM_PKRS);
4211 
4212 	adev->gfx.config.gb_addr_config = gb_addr_config;
4213 
4214 	adev->gfx.config.gb_addr_config_fields.num_pipes = 1 <<
4215 			REG_GET_FIELD(adev->gfx.config.gb_addr_config,
4216 				      GB_ADDR_CONFIG, NUM_PIPES);
4217 
4218 	adev->gfx.config.max_tile_pipes =
4219 		adev->gfx.config.gb_addr_config_fields.num_pipes;
4220 
4221 	adev->gfx.config.gb_addr_config_fields.max_compress_frags = 1 <<
4222 			REG_GET_FIELD(adev->gfx.config.gb_addr_config,
4223 				      GB_ADDR_CONFIG, MAX_COMPRESSED_FRAGS);
4224 	adev->gfx.config.gb_addr_config_fields.num_rb_per_se = 1 <<
4225 			REG_GET_FIELD(adev->gfx.config.gb_addr_config,
4226 				      GB_ADDR_CONFIG, NUM_RB_PER_SE);
4227 	adev->gfx.config.gb_addr_config_fields.num_se = 1 <<
4228 			REG_GET_FIELD(adev->gfx.config.gb_addr_config,
4229 				      GB_ADDR_CONFIG, NUM_SHADER_ENGINES);
4230 	adev->gfx.config.gb_addr_config_fields.pipe_interleave_size = 1 << (8 +
4231 			REG_GET_FIELD(adev->gfx.config.gb_addr_config,
4232 				      GB_ADDR_CONFIG, PIPE_INTERLEAVE_SIZE));
4233 
4234 	return 0;
4235 }
4236 
4237 static void gfx_v11_0_disable_gpa_mode(struct amdgpu_device *adev)
4238 {
4239 	uint32_t data;
4240 
4241 	data = RREG32_SOC15(GC, 0, regCPC_PSP_DEBUG);
4242 	data |= CPC_PSP_DEBUG__GPA_OVERRIDE_MASK;
4243 	WREG32_SOC15(GC, 0, regCPC_PSP_DEBUG, data);
4244 
4245 	data = RREG32_SOC15(GC, 0, regCPG_PSP_DEBUG);
4246 	data |= CPG_PSP_DEBUG__GPA_OVERRIDE_MASK;
4247 	WREG32_SOC15(GC, 0, regCPG_PSP_DEBUG, data);
4248 }
4249 
4250 static int gfx_v11_0_hw_init(void *handle)
4251 {
4252 	int r;
4253 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4254 
4255 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_RLC_BACKDOOR_AUTO) {
4256 		if (adev->gfx.imu.funcs) {
4257 			/* RLC autoload sequence 1: Program rlc ram */
4258 			if (adev->gfx.imu.funcs->program_rlc_ram)
4259 				adev->gfx.imu.funcs->program_rlc_ram(adev);
4260 		}
4261 		/* rlc autoload firmware */
4262 		r = gfx_v11_0_rlc_backdoor_autoload_enable(adev);
4263 		if (r)
4264 			return r;
4265 	} else {
4266 		if (adev->firmware.load_type == AMDGPU_FW_LOAD_DIRECT) {
4267 			if (adev->gfx.imu.funcs && (amdgpu_dpm > 0)) {
4268 				if (adev->gfx.imu.funcs->load_microcode)
4269 					adev->gfx.imu.funcs->load_microcode(adev);
4270 				if (adev->gfx.imu.funcs->setup_imu)
4271 					adev->gfx.imu.funcs->setup_imu(adev);
4272 				if (adev->gfx.imu.funcs->start_imu)
4273 					adev->gfx.imu.funcs->start_imu(adev);
4274 			}
4275 
4276 			/* disable gpa mode in backdoor loading */
4277 			gfx_v11_0_disable_gpa_mode(adev);
4278 		}
4279 	}
4280 
4281 	if ((adev->firmware.load_type == AMDGPU_FW_LOAD_RLC_BACKDOOR_AUTO) ||
4282 	    (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP)) {
4283 		r = gfx_v11_0_wait_for_rlc_autoload_complete(adev);
4284 		if (r) {
4285 			dev_err(adev->dev, "(%d) failed to wait rlc autoload complete\n", r);
4286 			return r;
4287 		}
4288 	}
4289 
4290 	adev->gfx.is_poweron = true;
4291 
4292 	if(get_gb_addr_config(adev))
4293 		DRM_WARN("Invalid gb_addr_config !\n");
4294 
4295 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP &&
4296 	    adev->gfx.rs64_enable)
4297 		gfx_v11_0_config_gfx_rs64(adev);
4298 
4299 	r = gfx_v11_0_gfxhub_enable(adev);
4300 	if (r)
4301 		return r;
4302 
4303 	if (!amdgpu_emu_mode)
4304 		gfx_v11_0_init_golden_registers(adev);
4305 
4306 	if ((adev->firmware.load_type == AMDGPU_FW_LOAD_DIRECT) ||
4307 	    (adev->firmware.load_type == AMDGPU_FW_LOAD_RLC_BACKDOOR_AUTO && amdgpu_dpm == 1)) {
4308 		/**
4309 		 * For gfx 11, rlc firmware loading relies on smu firmware is
4310 		 * loaded firstly, so in direct type, it has to load smc ucode
4311 		 * here before rlc.
4312 		 */
4313 		if (!(adev->flags & AMD_IS_APU)) {
4314 			r = amdgpu_pm_load_smu_firmware(adev, NULL);
4315 			if (r)
4316 				return r;
4317 		}
4318 	}
4319 
4320 	gfx_v11_0_constants_init(adev);
4321 
4322 	if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP)
4323 		gfx_v11_0_select_cp_fw_arch(adev);
4324 
4325 	if (adev->nbio.funcs->gc_doorbell_init)
4326 		adev->nbio.funcs->gc_doorbell_init(adev);
4327 
4328 	r = gfx_v11_0_rlc_resume(adev);
4329 	if (r)
4330 		return r;
4331 
4332 	/*
4333 	 * init golden registers and rlc resume may override some registers,
4334 	 * reconfig them here
4335 	 */
4336 	gfx_v11_0_tcp_harvest(adev);
4337 
4338 	r = gfx_v11_0_cp_resume(adev);
4339 	if (r)
4340 		return r;
4341 
4342 	return r;
4343 }
4344 
4345 static int gfx_v11_0_hw_fini(void *handle)
4346 {
4347 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4348 
4349 	amdgpu_irq_put(adev, &adev->gfx.priv_reg_irq, 0);
4350 	amdgpu_irq_put(adev, &adev->gfx.priv_inst_irq, 0);
4351 
4352 	if (!adev->no_hw_access) {
4353 		if (amdgpu_async_gfx_ring) {
4354 			if (amdgpu_gfx_disable_kgq(adev, 0))
4355 				DRM_ERROR("KGQ disable failed\n");
4356 		}
4357 
4358 		if (amdgpu_gfx_disable_kcq(adev, 0))
4359 			DRM_ERROR("KCQ disable failed\n");
4360 
4361 		amdgpu_mes_kiq_hw_fini(adev);
4362 	}
4363 
4364 	if (amdgpu_sriov_vf(adev))
4365 		/* Remove the steps disabling CPG and clearing KIQ position,
4366 		 * so that CP could perform IDLE-SAVE during switch. Those
4367 		 * steps are necessary to avoid a DMAR error in gfx9 but it is
4368 		 * not reproduced on gfx11.
4369 		 */
4370 		return 0;
4371 
4372 	gfx_v11_0_cp_enable(adev, false);
4373 	gfx_v11_0_enable_gui_idle_interrupt(adev, false);
4374 
4375 	adev->gfxhub.funcs->gart_disable(adev);
4376 
4377 	adev->gfx.is_poweron = false;
4378 
4379 	return 0;
4380 }
4381 
4382 static int gfx_v11_0_suspend(void *handle)
4383 {
4384 	return gfx_v11_0_hw_fini(handle);
4385 }
4386 
4387 static int gfx_v11_0_resume(void *handle)
4388 {
4389 	return gfx_v11_0_hw_init(handle);
4390 }
4391 
4392 static bool gfx_v11_0_is_idle(void *handle)
4393 {
4394 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4395 
4396 	if (REG_GET_FIELD(RREG32_SOC15(GC, 0, regGRBM_STATUS),
4397 				GRBM_STATUS, GUI_ACTIVE))
4398 		return false;
4399 	else
4400 		return true;
4401 }
4402 
4403 static int gfx_v11_0_wait_for_idle(void *handle)
4404 {
4405 	unsigned i;
4406 	u32 tmp;
4407 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4408 
4409 	for (i = 0; i < adev->usec_timeout; i++) {
4410 		/* read MC_STATUS */
4411 		tmp = RREG32_SOC15(GC, 0, regGRBM_STATUS) &
4412 			GRBM_STATUS__GUI_ACTIVE_MASK;
4413 
4414 		if (!REG_GET_FIELD(tmp, GRBM_STATUS, GUI_ACTIVE))
4415 			return 0;
4416 		udelay(1);
4417 	}
4418 	return -ETIMEDOUT;
4419 }
4420 
4421 static int gfx_v11_0_soft_reset(void *handle)
4422 {
4423 	u32 grbm_soft_reset = 0;
4424 	u32 tmp;
4425 	int i, j, k;
4426 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4427 
4428 	tmp = RREG32_SOC15(GC, 0, regCP_INT_CNTL);
4429 	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL, CMP_BUSY_INT_ENABLE, 0);
4430 	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL, CNTX_BUSY_INT_ENABLE, 0);
4431 	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL, CNTX_EMPTY_INT_ENABLE, 0);
4432 	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL, GFX_IDLE_INT_ENABLE, 0);
4433 	WREG32_SOC15(GC, 0, regCP_INT_CNTL, tmp);
4434 
4435 	gfx_v11_0_set_safe_mode(adev, 0);
4436 
4437 	for (i = 0; i < adev->gfx.mec.num_mec; ++i) {
4438 		for (j = 0; j < adev->gfx.mec.num_queue_per_pipe; j++) {
4439 			for (k = 0; k < adev->gfx.mec.num_pipe_per_mec; k++) {
4440 				tmp = RREG32_SOC15(GC, 0, regGRBM_GFX_CNTL);
4441 				tmp = REG_SET_FIELD(tmp, GRBM_GFX_CNTL, MEID, i);
4442 				tmp = REG_SET_FIELD(tmp, GRBM_GFX_CNTL, QUEUEID, j);
4443 				tmp = REG_SET_FIELD(tmp, GRBM_GFX_CNTL, PIPEID, k);
4444 				WREG32_SOC15(GC, 0, regGRBM_GFX_CNTL, tmp);
4445 
4446 				WREG32_SOC15(GC, 0, regCP_HQD_DEQUEUE_REQUEST, 0x2);
4447 				WREG32_SOC15(GC, 0, regSPI_COMPUTE_QUEUE_RESET, 0x1);
4448 			}
4449 		}
4450 	}
4451 	for (i = 0; i < adev->gfx.me.num_me; ++i) {
4452 		for (j = 0; j < adev->gfx.me.num_queue_per_pipe; j++) {
4453 			for (k = 0; k < adev->gfx.me.num_pipe_per_me; k++) {
4454 				tmp = RREG32_SOC15(GC, 0, regGRBM_GFX_CNTL);
4455 				tmp = REG_SET_FIELD(tmp, GRBM_GFX_CNTL, MEID, i);
4456 				tmp = REG_SET_FIELD(tmp, GRBM_GFX_CNTL, QUEUEID, j);
4457 				tmp = REG_SET_FIELD(tmp, GRBM_GFX_CNTL, PIPEID, k);
4458 				WREG32_SOC15(GC, 0, regGRBM_GFX_CNTL, tmp);
4459 
4460 				WREG32_SOC15(GC, 0, regCP_GFX_HQD_DEQUEUE_REQUEST, 0x1);
4461 			}
4462 		}
4463 	}
4464 
4465 	WREG32_SOC15(GC, 0, regCP_VMID_RESET, 0xfffffffe);
4466 
4467 	// Read CP_VMID_RESET register three times.
4468 	// to get sufficient time for GFX_HQD_ACTIVE reach 0
4469 	RREG32_SOC15(GC, 0, regCP_VMID_RESET);
4470 	RREG32_SOC15(GC, 0, regCP_VMID_RESET);
4471 	RREG32_SOC15(GC, 0, regCP_VMID_RESET);
4472 
4473 	for (i = 0; i < adev->usec_timeout; i++) {
4474 		if (!RREG32_SOC15(GC, 0, regCP_HQD_ACTIVE) &&
4475 		    !RREG32_SOC15(GC, 0, regCP_GFX_HQD_ACTIVE))
4476 			break;
4477 		udelay(1);
4478 	}
4479 	if (i >= adev->usec_timeout) {
4480 		printk("Failed to wait all pipes clean\n");
4481 		return -EINVAL;
4482 	}
4483 
4484 	/**********  trigger soft reset  ***********/
4485 	grbm_soft_reset = RREG32_SOC15(GC, 0, regGRBM_SOFT_RESET);
4486 	grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET,
4487 					SOFT_RESET_CP, 1);
4488 	grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET,
4489 					SOFT_RESET_GFX, 1);
4490 	grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET,
4491 					SOFT_RESET_CPF, 1);
4492 	grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET,
4493 					SOFT_RESET_CPC, 1);
4494 	grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET,
4495 					SOFT_RESET_CPG, 1);
4496 	WREG32_SOC15(GC, 0, regGRBM_SOFT_RESET, grbm_soft_reset);
4497 	/**********  exit soft reset  ***********/
4498 	grbm_soft_reset = RREG32_SOC15(GC, 0, regGRBM_SOFT_RESET);
4499 	grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET,
4500 					SOFT_RESET_CP, 0);
4501 	grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET,
4502 					SOFT_RESET_GFX, 0);
4503 	grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET,
4504 					SOFT_RESET_CPF, 0);
4505 	grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET,
4506 					SOFT_RESET_CPC, 0);
4507 	grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET,
4508 					SOFT_RESET_CPG, 0);
4509 	WREG32_SOC15(GC, 0, regGRBM_SOFT_RESET, grbm_soft_reset);
4510 
4511 	tmp = RREG32_SOC15(GC, 0, regCP_SOFT_RESET_CNTL);
4512 	tmp = REG_SET_FIELD(tmp, CP_SOFT_RESET_CNTL, CMP_HQD_REG_RESET, 0x1);
4513 	WREG32_SOC15(GC, 0, regCP_SOFT_RESET_CNTL, tmp);
4514 
4515 	WREG32_SOC15(GC, 0, regCP_ME_CNTL, 0x0);
4516 	WREG32_SOC15(GC, 0, regCP_MEC_RS64_CNTL, 0x0);
4517 
4518 	for (i = 0; i < adev->usec_timeout; i++) {
4519 		if (!RREG32_SOC15(GC, 0, regCP_VMID_RESET))
4520 			break;
4521 		udelay(1);
4522 	}
4523 	if (i >= adev->usec_timeout) {
4524 		printk("Failed to wait CP_VMID_RESET to 0\n");
4525 		return -EINVAL;
4526 	}
4527 
4528 	tmp = RREG32_SOC15(GC, 0, regCP_INT_CNTL);
4529 	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL, CMP_BUSY_INT_ENABLE, 1);
4530 	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL, CNTX_BUSY_INT_ENABLE, 1);
4531 	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL, CNTX_EMPTY_INT_ENABLE, 1);
4532 	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL, GFX_IDLE_INT_ENABLE, 1);
4533 	WREG32_SOC15(GC, 0, regCP_INT_CNTL, tmp);
4534 
4535 	gfx_v11_0_unset_safe_mode(adev, 0);
4536 
4537 	return gfx_v11_0_cp_resume(adev);
4538 }
4539 
4540 static bool gfx_v11_0_check_soft_reset(void *handle)
4541 {
4542 	int i, r;
4543 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4544 	struct amdgpu_ring *ring;
4545 	long tmo = msecs_to_jiffies(1000);
4546 
4547 	for (i = 0; i < adev->gfx.num_gfx_rings; i++) {
4548 		ring = &adev->gfx.gfx_ring[i];
4549 		r = amdgpu_ring_test_ib(ring, tmo);
4550 		if (r)
4551 			return true;
4552 	}
4553 
4554 	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
4555 		ring = &adev->gfx.compute_ring[i];
4556 		r = amdgpu_ring_test_ib(ring, tmo);
4557 		if (r)
4558 			return true;
4559 	}
4560 
4561 	return false;
4562 }
4563 
4564 static int gfx_v11_0_post_soft_reset(void *handle)
4565 {
4566 	/**
4567 	 * GFX soft reset will impact MES, need resume MES when do GFX soft reset
4568 	 */
4569 	return amdgpu_mes_resume((struct amdgpu_device *)handle);
4570 }
4571 
4572 static uint64_t gfx_v11_0_get_gpu_clock_counter(struct amdgpu_device *adev)
4573 {
4574 	uint64_t clock;
4575 	uint64_t clock_counter_lo, clock_counter_hi_pre, clock_counter_hi_after;
4576 
4577 	if (amdgpu_sriov_vf(adev)) {
4578 		amdgpu_gfx_off_ctrl(adev, false);
4579 		mutex_lock(&adev->gfx.gpu_clock_mutex);
4580 		clock_counter_hi_pre = (uint64_t)RREG32_SOC15(GC, 0, regCP_MES_MTIME_HI);
4581 		clock_counter_lo = (uint64_t)RREG32_SOC15(GC, 0, regCP_MES_MTIME_LO);
4582 		clock_counter_hi_after = (uint64_t)RREG32_SOC15(GC, 0, regCP_MES_MTIME_HI);
4583 		if (clock_counter_hi_pre != clock_counter_hi_after)
4584 			clock_counter_lo = (uint64_t)RREG32_SOC15(GC, 0, regCP_MES_MTIME_LO);
4585 		mutex_unlock(&adev->gfx.gpu_clock_mutex);
4586 		amdgpu_gfx_off_ctrl(adev, true);
4587 	} else {
4588 		preempt_disable();
4589 		clock_counter_hi_pre = (uint64_t)RREG32_SOC15(SMUIO, 0, regGOLDEN_TSC_COUNT_UPPER);
4590 		clock_counter_lo = (uint64_t)RREG32_SOC15(SMUIO, 0, regGOLDEN_TSC_COUNT_LOWER);
4591 		clock_counter_hi_after = (uint64_t)RREG32_SOC15(SMUIO, 0, regGOLDEN_TSC_COUNT_UPPER);
4592 		if (clock_counter_hi_pre != clock_counter_hi_after)
4593 			clock_counter_lo = (uint64_t)RREG32_SOC15(SMUIO, 0, regGOLDEN_TSC_COUNT_LOWER);
4594 		preempt_enable();
4595 	}
4596 	clock = clock_counter_lo | (clock_counter_hi_after << 32ULL);
4597 
4598 	return clock;
4599 }
4600 
4601 static void gfx_v11_0_ring_emit_gds_switch(struct amdgpu_ring *ring,
4602 					   uint32_t vmid,
4603 					   uint32_t gds_base, uint32_t gds_size,
4604 					   uint32_t gws_base, uint32_t gws_size,
4605 					   uint32_t oa_base, uint32_t oa_size)
4606 {
4607 	struct amdgpu_device *adev = ring->adev;
4608 
4609 	/* GDS Base */
4610 	gfx_v11_0_write_data_to_reg(ring, 0, false,
4611 				    SOC15_REG_OFFSET(GC, 0, regGDS_VMID0_BASE) + 2 * vmid,
4612 				    gds_base);
4613 
4614 	/* GDS Size */
4615 	gfx_v11_0_write_data_to_reg(ring, 0, false,
4616 				    SOC15_REG_OFFSET(GC, 0, regGDS_VMID0_SIZE) + 2 * vmid,
4617 				    gds_size);
4618 
4619 	/* GWS */
4620 	gfx_v11_0_write_data_to_reg(ring, 0, false,
4621 				    SOC15_REG_OFFSET(GC, 0, regGDS_GWS_VMID0) + vmid,
4622 				    gws_size << GDS_GWS_VMID0__SIZE__SHIFT | gws_base);
4623 
4624 	/* OA */
4625 	gfx_v11_0_write_data_to_reg(ring, 0, false,
4626 				    SOC15_REG_OFFSET(GC, 0, regGDS_OA_VMID0) + vmid,
4627 				    (1 << (oa_size + oa_base)) - (1 << oa_base));
4628 }
4629 
4630 static int gfx_v11_0_early_init(void *handle)
4631 {
4632 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4633 
4634 	adev->gfx.funcs = &gfx_v11_0_gfx_funcs;
4635 
4636 	adev->gfx.num_gfx_rings = GFX11_NUM_GFX_RINGS;
4637 	adev->gfx.num_compute_rings = min(amdgpu_gfx_get_num_kcq(adev),
4638 					  AMDGPU_MAX_COMPUTE_RINGS);
4639 
4640 	gfx_v11_0_set_kiq_pm4_funcs(adev);
4641 	gfx_v11_0_set_ring_funcs(adev);
4642 	gfx_v11_0_set_irq_funcs(adev);
4643 	gfx_v11_0_set_gds_init(adev);
4644 	gfx_v11_0_set_rlc_funcs(adev);
4645 	gfx_v11_0_set_mqd_funcs(adev);
4646 	gfx_v11_0_set_imu_funcs(adev);
4647 
4648 	gfx_v11_0_init_rlcg_reg_access_ctrl(adev);
4649 
4650 	return gfx_v11_0_init_microcode(adev);
4651 }
4652 
4653 static int gfx_v11_0_late_init(void *handle)
4654 {
4655 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4656 	int r;
4657 
4658 	r = amdgpu_irq_get(adev, &adev->gfx.priv_reg_irq, 0);
4659 	if (r)
4660 		return r;
4661 
4662 	r = amdgpu_irq_get(adev, &adev->gfx.priv_inst_irq, 0);
4663 	if (r)
4664 		return r;
4665 
4666 	return 0;
4667 }
4668 
4669 static bool gfx_v11_0_is_rlc_enabled(struct amdgpu_device *adev)
4670 {
4671 	uint32_t rlc_cntl;
4672 
4673 	/* if RLC is not enabled, do nothing */
4674 	rlc_cntl = RREG32_SOC15(GC, 0, regRLC_CNTL);
4675 	return (REG_GET_FIELD(rlc_cntl, RLC_CNTL, RLC_ENABLE_F32)) ? true : false;
4676 }
4677 
4678 static void gfx_v11_0_set_safe_mode(struct amdgpu_device *adev, int xcc_id)
4679 {
4680 	uint32_t data;
4681 	unsigned i;
4682 
4683 	data = RLC_SAFE_MODE__CMD_MASK;
4684 	data |= (1 << RLC_SAFE_MODE__MESSAGE__SHIFT);
4685 
4686 	WREG32_SOC15(GC, 0, regRLC_SAFE_MODE, data);
4687 
4688 	/* wait for RLC_SAFE_MODE */
4689 	for (i = 0; i < adev->usec_timeout; i++) {
4690 		if (!REG_GET_FIELD(RREG32_SOC15(GC, 0, regRLC_SAFE_MODE),
4691 				   RLC_SAFE_MODE, CMD))
4692 			break;
4693 		udelay(1);
4694 	}
4695 }
4696 
4697 static void gfx_v11_0_unset_safe_mode(struct amdgpu_device *adev, int xcc_id)
4698 {
4699 	WREG32_SOC15(GC, 0, regRLC_SAFE_MODE, RLC_SAFE_MODE__CMD_MASK);
4700 }
4701 
4702 static void gfx_v11_0_update_perf_clk(struct amdgpu_device *adev,
4703 				      bool enable)
4704 {
4705 	uint32_t def, data;
4706 
4707 	if (!(adev->cg_flags & AMD_CG_SUPPORT_GFX_PERF_CLK))
4708 		return;
4709 
4710 	def = data = RREG32_SOC15(GC, 0, regRLC_CGTT_MGCG_OVERRIDE);
4711 
4712 	if (enable)
4713 		data &= ~RLC_CGTT_MGCG_OVERRIDE__PERFMON_CLOCK_STATE_MASK;
4714 	else
4715 		data |= RLC_CGTT_MGCG_OVERRIDE__PERFMON_CLOCK_STATE_MASK;
4716 
4717 	if (def != data)
4718 		WREG32_SOC15(GC, 0, regRLC_CGTT_MGCG_OVERRIDE, data);
4719 }
4720 
4721 static void gfx_v11_0_update_sram_fgcg(struct amdgpu_device *adev,
4722 				       bool enable)
4723 {
4724 	uint32_t def, data;
4725 
4726 	if (!(adev->cg_flags & AMD_CG_SUPPORT_GFX_FGCG))
4727 		return;
4728 
4729 	def = data = RREG32_SOC15(GC, 0, regRLC_CGTT_MGCG_OVERRIDE);
4730 
4731 	if (enable)
4732 		data &= ~RLC_CGTT_MGCG_OVERRIDE__GFXIP_FGCG_OVERRIDE_MASK;
4733 	else
4734 		data |= RLC_CGTT_MGCG_OVERRIDE__GFXIP_FGCG_OVERRIDE_MASK;
4735 
4736 	if (def != data)
4737 		WREG32_SOC15(GC, 0, regRLC_CGTT_MGCG_OVERRIDE, data);
4738 }
4739 
4740 static void gfx_v11_0_update_repeater_fgcg(struct amdgpu_device *adev,
4741 					   bool enable)
4742 {
4743 	uint32_t def, data;
4744 
4745 	if (!(adev->cg_flags & AMD_CG_SUPPORT_REPEATER_FGCG))
4746 		return;
4747 
4748 	def = data = RREG32_SOC15(GC, 0, regRLC_CGTT_MGCG_OVERRIDE);
4749 
4750 	if (enable)
4751 		data &= ~RLC_CGTT_MGCG_OVERRIDE__GFXIP_REPEATER_FGCG_OVERRIDE_MASK;
4752 	else
4753 		data |= RLC_CGTT_MGCG_OVERRIDE__GFXIP_REPEATER_FGCG_OVERRIDE_MASK;
4754 
4755 	if (def != data)
4756 		WREG32_SOC15(GC, 0, regRLC_CGTT_MGCG_OVERRIDE, data);
4757 }
4758 
4759 static void gfx_v11_0_update_medium_grain_clock_gating(struct amdgpu_device *adev,
4760 						       bool enable)
4761 {
4762 	uint32_t data, def;
4763 
4764 	if (!(adev->cg_flags & (AMD_CG_SUPPORT_GFX_MGCG | AMD_CG_SUPPORT_GFX_MGLS)))
4765 		return;
4766 
4767 	/* It is disabled by HW by default */
4768 	if (enable) {
4769 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGCG) {
4770 			/* 1 - RLC_CGTT_MGCG_OVERRIDE */
4771 			def = data = RREG32_SOC15(GC, 0, regRLC_CGTT_MGCG_OVERRIDE);
4772 
4773 			data &= ~(RLC_CGTT_MGCG_OVERRIDE__GRBM_CGTT_SCLK_OVERRIDE_MASK |
4774 				  RLC_CGTT_MGCG_OVERRIDE__RLC_CGTT_SCLK_OVERRIDE_MASK |
4775 				  RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGCG_OVERRIDE_MASK);
4776 
4777 			if (def != data)
4778 				WREG32_SOC15(GC, 0, regRLC_CGTT_MGCG_OVERRIDE, data);
4779 		}
4780 	} else {
4781 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGCG) {
4782 			def = data = RREG32_SOC15(GC, 0, regRLC_CGTT_MGCG_OVERRIDE);
4783 
4784 			data |= (RLC_CGTT_MGCG_OVERRIDE__RLC_CGTT_SCLK_OVERRIDE_MASK |
4785 				 RLC_CGTT_MGCG_OVERRIDE__GRBM_CGTT_SCLK_OVERRIDE_MASK |
4786 				 RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGCG_OVERRIDE_MASK);
4787 
4788 			if (def != data)
4789 				WREG32_SOC15(GC, 0, regRLC_CGTT_MGCG_OVERRIDE, data);
4790 		}
4791 	}
4792 }
4793 
4794 static void gfx_v11_0_update_coarse_grain_clock_gating(struct amdgpu_device *adev,
4795 						       bool enable)
4796 {
4797 	uint32_t def, data;
4798 
4799 	if (!(adev->cg_flags &
4800 	      (AMD_CG_SUPPORT_GFX_CGCG |
4801 	      AMD_CG_SUPPORT_GFX_CGLS |
4802 	      AMD_CG_SUPPORT_GFX_3D_CGCG |
4803 	      AMD_CG_SUPPORT_GFX_3D_CGLS)))
4804 		return;
4805 
4806 	if (enable) {
4807 		def = data = RREG32_SOC15(GC, 0, regRLC_CGTT_MGCG_OVERRIDE);
4808 
4809 		/* unset CGCG override */
4810 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGCG)
4811 			data &= ~RLC_CGTT_MGCG_OVERRIDE__GFXIP_CGCG_OVERRIDE_MASK;
4812 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGLS)
4813 			data &= ~RLC_CGTT_MGCG_OVERRIDE__GFXIP_CGLS_OVERRIDE_MASK;
4814 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_3D_CGCG ||
4815 		    adev->cg_flags & AMD_CG_SUPPORT_GFX_3D_CGLS)
4816 			data &= ~RLC_CGTT_MGCG_OVERRIDE__GFXIP_GFX3D_CG_OVERRIDE_MASK;
4817 
4818 		/* update CGCG override bits */
4819 		if (def != data)
4820 			WREG32_SOC15(GC, 0, regRLC_CGTT_MGCG_OVERRIDE, data);
4821 
4822 		/* enable cgcg FSM(0x0000363F) */
4823 		def = data = RREG32_SOC15(GC, 0, regRLC_CGCG_CGLS_CTRL);
4824 
4825 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGCG) {
4826 			data &= ~RLC_CGCG_CGLS_CTRL__CGCG_GFX_IDLE_THRESHOLD_MASK;
4827 			data |= (0x36 << RLC_CGCG_CGLS_CTRL__CGCG_GFX_IDLE_THRESHOLD__SHIFT) |
4828 				 RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK;
4829 		}
4830 
4831 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGLS) {
4832 			data &= ~RLC_CGCG_CGLS_CTRL__CGLS_REP_COMPANSAT_DELAY_MASK;
4833 			data |= (0x000F << RLC_CGCG_CGLS_CTRL__CGLS_REP_COMPANSAT_DELAY__SHIFT) |
4834 				 RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK;
4835 		}
4836 
4837 		if (def != data)
4838 			WREG32_SOC15(GC, 0, regRLC_CGCG_CGLS_CTRL, data);
4839 
4840 		/* Program RLC_CGCG_CGLS_CTRL_3D */
4841 		def = data = RREG32_SOC15(GC, 0, regRLC_CGCG_CGLS_CTRL_3D);
4842 
4843 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_3D_CGCG) {
4844 			data &= ~RLC_CGCG_CGLS_CTRL_3D__CGCG_GFX_IDLE_THRESHOLD_MASK;
4845 			data |= (0x36 << RLC_CGCG_CGLS_CTRL_3D__CGCG_GFX_IDLE_THRESHOLD__SHIFT) |
4846 				 RLC_CGCG_CGLS_CTRL_3D__CGCG_EN_MASK;
4847 		}
4848 
4849 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_3D_CGLS) {
4850 			data &= ~RLC_CGCG_CGLS_CTRL_3D__CGLS_REP_COMPANSAT_DELAY_MASK;
4851 			data |= (0xf << RLC_CGCG_CGLS_CTRL_3D__CGLS_REP_COMPANSAT_DELAY__SHIFT) |
4852 				 RLC_CGCG_CGLS_CTRL_3D__CGLS_EN_MASK;
4853 		}
4854 
4855 		if (def != data)
4856 			WREG32_SOC15(GC, 0, regRLC_CGCG_CGLS_CTRL_3D, data);
4857 
4858 		/* set IDLE_POLL_COUNT(0x00900100) */
4859 		def = data = RREG32_SOC15(GC, 0, regCP_RB_WPTR_POLL_CNTL);
4860 
4861 		data &= ~(CP_RB_WPTR_POLL_CNTL__POLL_FREQUENCY_MASK | CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT_MASK);
4862 		data |= (0x0100 << CP_RB_WPTR_POLL_CNTL__POLL_FREQUENCY__SHIFT) |
4863 			(0x0090 << CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT__SHIFT);
4864 
4865 		if (def != data)
4866 			WREG32_SOC15(GC, 0, regCP_RB_WPTR_POLL_CNTL, data);
4867 
4868 		data = RREG32_SOC15(GC, 0, regCP_INT_CNTL);
4869 		data = REG_SET_FIELD(data, CP_INT_CNTL, CNTX_BUSY_INT_ENABLE, 1);
4870 		data = REG_SET_FIELD(data, CP_INT_CNTL, CNTX_EMPTY_INT_ENABLE, 1);
4871 		data = REG_SET_FIELD(data, CP_INT_CNTL, CMP_BUSY_INT_ENABLE, 1);
4872 		data = REG_SET_FIELD(data, CP_INT_CNTL, GFX_IDLE_INT_ENABLE, 1);
4873 		WREG32_SOC15(GC, 0, regCP_INT_CNTL, data);
4874 
4875 		data = RREG32_SOC15(GC, 0, regSDMA0_RLC_CGCG_CTRL);
4876 		data = REG_SET_FIELD(data, SDMA0_RLC_CGCG_CTRL, CGCG_INT_ENABLE, 1);
4877 		WREG32_SOC15(GC, 0, regSDMA0_RLC_CGCG_CTRL, data);
4878 
4879 		/* Some ASICs only have one SDMA instance, not need to configure SDMA1 */
4880 		if (adev->sdma.num_instances > 1) {
4881 			data = RREG32_SOC15(GC, 0, regSDMA1_RLC_CGCG_CTRL);
4882 			data = REG_SET_FIELD(data, SDMA1_RLC_CGCG_CTRL, CGCG_INT_ENABLE, 1);
4883 			WREG32_SOC15(GC, 0, regSDMA1_RLC_CGCG_CTRL, data);
4884 		}
4885 	} else {
4886 		/* Program RLC_CGCG_CGLS_CTRL */
4887 		def = data = RREG32_SOC15(GC, 0, regRLC_CGCG_CGLS_CTRL);
4888 
4889 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGCG)
4890 			data &= ~RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK;
4891 
4892 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGLS)
4893 			data &= ~RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK;
4894 
4895 		if (def != data)
4896 			WREG32_SOC15(GC, 0, regRLC_CGCG_CGLS_CTRL, data);
4897 
4898 		/* Program RLC_CGCG_CGLS_CTRL_3D */
4899 		def = data = RREG32_SOC15(GC, 0, regRLC_CGCG_CGLS_CTRL_3D);
4900 
4901 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_3D_CGCG)
4902 			data &= ~RLC_CGCG_CGLS_CTRL_3D__CGCG_EN_MASK;
4903 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_3D_CGLS)
4904 			data &= ~RLC_CGCG_CGLS_CTRL_3D__CGLS_EN_MASK;
4905 
4906 		if (def != data)
4907 			WREG32_SOC15(GC, 0, regRLC_CGCG_CGLS_CTRL_3D, data);
4908 
4909 		data = RREG32_SOC15(GC, 0, regSDMA0_RLC_CGCG_CTRL);
4910 		data &= ~SDMA0_RLC_CGCG_CTRL__CGCG_INT_ENABLE_MASK;
4911 		WREG32_SOC15(GC, 0, regSDMA0_RLC_CGCG_CTRL, data);
4912 
4913 		/* Some ASICs only have one SDMA instance, not need to configure SDMA1 */
4914 		if (adev->sdma.num_instances > 1) {
4915 			data = RREG32_SOC15(GC, 0, regSDMA1_RLC_CGCG_CTRL);
4916 			data &= ~SDMA1_RLC_CGCG_CTRL__CGCG_INT_ENABLE_MASK;
4917 			WREG32_SOC15(GC, 0, regSDMA1_RLC_CGCG_CTRL, data);
4918 		}
4919 	}
4920 }
4921 
4922 static int gfx_v11_0_update_gfx_clock_gating(struct amdgpu_device *adev,
4923 					    bool enable)
4924 {
4925 	amdgpu_gfx_rlc_enter_safe_mode(adev, 0);
4926 
4927 	gfx_v11_0_update_coarse_grain_clock_gating(adev, enable);
4928 
4929 	gfx_v11_0_update_medium_grain_clock_gating(adev, enable);
4930 
4931 	gfx_v11_0_update_repeater_fgcg(adev, enable);
4932 
4933 	gfx_v11_0_update_sram_fgcg(adev, enable);
4934 
4935 	gfx_v11_0_update_perf_clk(adev, enable);
4936 
4937 	if (adev->cg_flags &
4938 	    (AMD_CG_SUPPORT_GFX_MGCG |
4939 	     AMD_CG_SUPPORT_GFX_CGLS |
4940 	     AMD_CG_SUPPORT_GFX_CGCG |
4941 	     AMD_CG_SUPPORT_GFX_3D_CGCG |
4942 	     AMD_CG_SUPPORT_GFX_3D_CGLS))
4943 	        gfx_v11_0_enable_gui_idle_interrupt(adev, enable);
4944 
4945 	amdgpu_gfx_rlc_exit_safe_mode(adev, 0);
4946 
4947 	return 0;
4948 }
4949 
4950 static void gfx_v11_0_update_spm_vmid(struct amdgpu_device *adev, unsigned vmid)
4951 {
4952 	u32 reg, data;
4953 
4954 	amdgpu_gfx_off_ctrl(adev, false);
4955 
4956 	reg = SOC15_REG_OFFSET(GC, 0, regRLC_SPM_MC_CNTL);
4957 	if (amdgpu_sriov_is_pp_one_vf(adev))
4958 		data = RREG32_NO_KIQ(reg);
4959 	else
4960 		data = RREG32(reg);
4961 
4962 	data &= ~RLC_SPM_MC_CNTL__RLC_SPM_VMID_MASK;
4963 	data |= (vmid & RLC_SPM_MC_CNTL__RLC_SPM_VMID_MASK) << RLC_SPM_MC_CNTL__RLC_SPM_VMID__SHIFT;
4964 
4965 	if (amdgpu_sriov_is_pp_one_vf(adev))
4966 		WREG32_SOC15_NO_KIQ(GC, 0, regRLC_SPM_MC_CNTL, data);
4967 	else
4968 		WREG32_SOC15(GC, 0, regRLC_SPM_MC_CNTL, data);
4969 
4970 	amdgpu_gfx_off_ctrl(adev, true);
4971 }
4972 
4973 static const struct amdgpu_rlc_funcs gfx_v11_0_rlc_funcs = {
4974 	.is_rlc_enabled = gfx_v11_0_is_rlc_enabled,
4975 	.set_safe_mode = gfx_v11_0_set_safe_mode,
4976 	.unset_safe_mode = gfx_v11_0_unset_safe_mode,
4977 	.init = gfx_v11_0_rlc_init,
4978 	.get_csb_size = gfx_v11_0_get_csb_size,
4979 	.get_csb_buffer = gfx_v11_0_get_csb_buffer,
4980 	.resume = gfx_v11_0_rlc_resume,
4981 	.stop = gfx_v11_0_rlc_stop,
4982 	.reset = gfx_v11_0_rlc_reset,
4983 	.start = gfx_v11_0_rlc_start,
4984 	.update_spm_vmid = gfx_v11_0_update_spm_vmid,
4985 };
4986 
4987 static void gfx_v11_cntl_power_gating(struct amdgpu_device *adev, bool enable)
4988 {
4989 	u32 data = RREG32_SOC15(GC, 0, regRLC_PG_CNTL);
4990 
4991 	if (enable && (adev->pg_flags & AMD_PG_SUPPORT_GFX_PG))
4992 		data |= RLC_PG_CNTL__GFX_POWER_GATING_ENABLE_MASK;
4993 	else
4994 		data &= ~RLC_PG_CNTL__GFX_POWER_GATING_ENABLE_MASK;
4995 
4996 	WREG32_SOC15(GC, 0, regRLC_PG_CNTL, data);
4997 
4998 	// Program RLC_PG_DELAY3 for CGPG hysteresis
4999 	if (enable && (adev->pg_flags & AMD_PG_SUPPORT_GFX_PG)) {
5000 		switch (adev->ip_versions[GC_HWIP][0]) {
5001 		case IP_VERSION(11, 0, 1):
5002 		case IP_VERSION(11, 0, 4):
5003 			WREG32_SOC15(GC, 0, regRLC_PG_DELAY_3, RLC_PG_DELAY_3_DEFAULT_GC_11_0_1);
5004 			break;
5005 		default:
5006 			break;
5007 		}
5008 	}
5009 }
5010 
5011 static void gfx_v11_cntl_pg(struct amdgpu_device *adev, bool enable)
5012 {
5013 	amdgpu_gfx_rlc_enter_safe_mode(adev, 0);
5014 
5015 	gfx_v11_cntl_power_gating(adev, enable);
5016 
5017 	amdgpu_gfx_rlc_exit_safe_mode(adev, 0);
5018 }
5019 
5020 static int gfx_v11_0_set_powergating_state(void *handle,
5021 					   enum amd_powergating_state state)
5022 {
5023 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5024 	bool enable = (state == AMD_PG_STATE_GATE);
5025 
5026 	if (amdgpu_sriov_vf(adev))
5027 		return 0;
5028 
5029 	switch (adev->ip_versions[GC_HWIP][0]) {
5030 	case IP_VERSION(11, 0, 0):
5031 	case IP_VERSION(11, 0, 2):
5032 	case IP_VERSION(11, 0, 3):
5033 		amdgpu_gfx_off_ctrl(adev, enable);
5034 		break;
5035 	case IP_VERSION(11, 0, 1):
5036 	case IP_VERSION(11, 0, 4):
5037 		if (!enable)
5038 			amdgpu_gfx_off_ctrl(adev, false);
5039 
5040 		gfx_v11_cntl_pg(adev, enable);
5041 
5042 		if (enable)
5043 			amdgpu_gfx_off_ctrl(adev, true);
5044 
5045 		break;
5046 	default:
5047 		break;
5048 	}
5049 
5050 	return 0;
5051 }
5052 
5053 static int gfx_v11_0_set_clockgating_state(void *handle,
5054 					  enum amd_clockgating_state state)
5055 {
5056 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5057 
5058 	if (amdgpu_sriov_vf(adev))
5059 	        return 0;
5060 
5061 	switch (adev->ip_versions[GC_HWIP][0]) {
5062 	case IP_VERSION(11, 0, 0):
5063 	case IP_VERSION(11, 0, 1):
5064 	case IP_VERSION(11, 0, 2):
5065 	case IP_VERSION(11, 0, 3):
5066 	case IP_VERSION(11, 0, 4):
5067 	        gfx_v11_0_update_gfx_clock_gating(adev,
5068 	                        state ==  AMD_CG_STATE_GATE);
5069 	        break;
5070 	default:
5071 	        break;
5072 	}
5073 
5074 	return 0;
5075 }
5076 
5077 static void gfx_v11_0_get_clockgating_state(void *handle, u64 *flags)
5078 {
5079 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5080 	int data;
5081 
5082 	/* AMD_CG_SUPPORT_GFX_MGCG */
5083 	data = RREG32_SOC15(GC, 0, regRLC_CGTT_MGCG_OVERRIDE);
5084 	if (!(data & RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGCG_OVERRIDE_MASK))
5085 		*flags |= AMD_CG_SUPPORT_GFX_MGCG;
5086 
5087 	/* AMD_CG_SUPPORT_REPEATER_FGCG */
5088 	if (!(data & RLC_CGTT_MGCG_OVERRIDE__GFXIP_REPEATER_FGCG_OVERRIDE_MASK))
5089 		*flags |= AMD_CG_SUPPORT_REPEATER_FGCG;
5090 
5091 	/* AMD_CG_SUPPORT_GFX_FGCG */
5092 	if (!(data & RLC_CGTT_MGCG_OVERRIDE__GFXIP_FGCG_OVERRIDE_MASK))
5093 		*flags |= AMD_CG_SUPPORT_GFX_FGCG;
5094 
5095 	/* AMD_CG_SUPPORT_GFX_PERF_CLK */
5096 	if (!(data & RLC_CGTT_MGCG_OVERRIDE__PERFMON_CLOCK_STATE_MASK))
5097 		*flags |= AMD_CG_SUPPORT_GFX_PERF_CLK;
5098 
5099 	/* AMD_CG_SUPPORT_GFX_CGCG */
5100 	data = RREG32_SOC15(GC, 0, regRLC_CGCG_CGLS_CTRL);
5101 	if (data & RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK)
5102 		*flags |= AMD_CG_SUPPORT_GFX_CGCG;
5103 
5104 	/* AMD_CG_SUPPORT_GFX_CGLS */
5105 	if (data & RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK)
5106 		*flags |= AMD_CG_SUPPORT_GFX_CGLS;
5107 
5108 	/* AMD_CG_SUPPORT_GFX_3D_CGCG */
5109 	data = RREG32_SOC15(GC, 0, regRLC_CGCG_CGLS_CTRL_3D);
5110 	if (data & RLC_CGCG_CGLS_CTRL_3D__CGCG_EN_MASK)
5111 		*flags |= AMD_CG_SUPPORT_GFX_3D_CGCG;
5112 
5113 	/* AMD_CG_SUPPORT_GFX_3D_CGLS */
5114 	if (data & RLC_CGCG_CGLS_CTRL_3D__CGLS_EN_MASK)
5115 		*flags |= AMD_CG_SUPPORT_GFX_3D_CGLS;
5116 }
5117 
5118 static u64 gfx_v11_0_ring_get_rptr_gfx(struct amdgpu_ring *ring)
5119 {
5120 	/* gfx11 is 32bit rptr*/
5121 	return *(uint32_t *)ring->rptr_cpu_addr;
5122 }
5123 
5124 static u64 gfx_v11_0_ring_get_wptr_gfx(struct amdgpu_ring *ring)
5125 {
5126 	struct amdgpu_device *adev = ring->adev;
5127 	u64 wptr;
5128 
5129 	/* XXX check if swapping is necessary on BE */
5130 	if (ring->use_doorbell) {
5131 		wptr = atomic64_read((atomic64_t *)ring->wptr_cpu_addr);
5132 	} else {
5133 		wptr = RREG32_SOC15(GC, 0, regCP_RB0_WPTR);
5134 		wptr += (u64)RREG32_SOC15(GC, 0, regCP_RB0_WPTR_HI) << 32;
5135 	}
5136 
5137 	return wptr;
5138 }
5139 
5140 static void gfx_v11_0_ring_set_wptr_gfx(struct amdgpu_ring *ring)
5141 {
5142 	struct amdgpu_device *adev = ring->adev;
5143 	uint32_t *wptr_saved;
5144 	uint32_t *is_queue_unmap;
5145 	uint64_t aggregated_db_index;
5146 	uint32_t mqd_size = adev->mqds[AMDGPU_HW_IP_GFX].mqd_size;
5147 	uint64_t wptr_tmp;
5148 
5149 	if (ring->is_mes_queue) {
5150 		wptr_saved = (uint32_t *)(ring->mqd_ptr + mqd_size);
5151 		is_queue_unmap = (uint32_t *)(ring->mqd_ptr + mqd_size +
5152 					      sizeof(uint32_t));
5153 		aggregated_db_index =
5154 			amdgpu_mes_get_aggregated_doorbell_index(adev,
5155 								 ring->hw_prio);
5156 
5157 		wptr_tmp = ring->wptr & ring->buf_mask;
5158 		atomic64_set((atomic64_t *)ring->wptr_cpu_addr, wptr_tmp);
5159 		*wptr_saved = wptr_tmp;
5160 		/* assume doorbell always being used by mes mapped queue */
5161 		if (*is_queue_unmap) {
5162 			WDOORBELL64(aggregated_db_index, wptr_tmp);
5163 			WDOORBELL64(ring->doorbell_index, wptr_tmp);
5164 		} else {
5165 			WDOORBELL64(ring->doorbell_index, wptr_tmp);
5166 
5167 			if (*is_queue_unmap)
5168 				WDOORBELL64(aggregated_db_index, wptr_tmp);
5169 		}
5170 	} else {
5171 		if (ring->use_doorbell) {
5172 			/* XXX check if swapping is necessary on BE */
5173 			atomic64_set((atomic64_t *)ring->wptr_cpu_addr,
5174 				     ring->wptr);
5175 			WDOORBELL64(ring->doorbell_index, ring->wptr);
5176 		} else {
5177 			WREG32_SOC15(GC, 0, regCP_RB0_WPTR,
5178 				     lower_32_bits(ring->wptr));
5179 			WREG32_SOC15(GC, 0, regCP_RB0_WPTR_HI,
5180 				     upper_32_bits(ring->wptr));
5181 		}
5182 	}
5183 }
5184 
5185 static u64 gfx_v11_0_ring_get_rptr_compute(struct amdgpu_ring *ring)
5186 {
5187 	/* gfx11 hardware is 32bit rptr */
5188 	return *(uint32_t *)ring->rptr_cpu_addr;
5189 }
5190 
5191 static u64 gfx_v11_0_ring_get_wptr_compute(struct amdgpu_ring *ring)
5192 {
5193 	u64 wptr;
5194 
5195 	/* XXX check if swapping is necessary on BE */
5196 	if (ring->use_doorbell)
5197 		wptr = atomic64_read((atomic64_t *)ring->wptr_cpu_addr);
5198 	else
5199 		BUG();
5200 	return wptr;
5201 }
5202 
5203 static void gfx_v11_0_ring_set_wptr_compute(struct amdgpu_ring *ring)
5204 {
5205 	struct amdgpu_device *adev = ring->adev;
5206 	uint32_t *wptr_saved;
5207 	uint32_t *is_queue_unmap;
5208 	uint64_t aggregated_db_index;
5209 	uint32_t mqd_size = adev->mqds[AMDGPU_HW_IP_COMPUTE].mqd_size;
5210 	uint64_t wptr_tmp;
5211 
5212 	if (ring->is_mes_queue) {
5213 		wptr_saved = (uint32_t *)(ring->mqd_ptr + mqd_size);
5214 		is_queue_unmap = (uint32_t *)(ring->mqd_ptr + mqd_size +
5215 					      sizeof(uint32_t));
5216 		aggregated_db_index =
5217 			amdgpu_mes_get_aggregated_doorbell_index(adev,
5218 								 ring->hw_prio);
5219 
5220 		wptr_tmp = ring->wptr & ring->buf_mask;
5221 		atomic64_set((atomic64_t *)ring->wptr_cpu_addr, wptr_tmp);
5222 		*wptr_saved = wptr_tmp;
5223 		/* assume doorbell always used by mes mapped queue */
5224 		if (*is_queue_unmap) {
5225 			WDOORBELL64(aggregated_db_index, wptr_tmp);
5226 			WDOORBELL64(ring->doorbell_index, wptr_tmp);
5227 		} else {
5228 			WDOORBELL64(ring->doorbell_index, wptr_tmp);
5229 
5230 			if (*is_queue_unmap)
5231 				WDOORBELL64(aggregated_db_index, wptr_tmp);
5232 		}
5233 	} else {
5234 		/* XXX check if swapping is necessary on BE */
5235 		if (ring->use_doorbell) {
5236 			atomic64_set((atomic64_t *)ring->wptr_cpu_addr,
5237 				     ring->wptr);
5238 			WDOORBELL64(ring->doorbell_index, ring->wptr);
5239 		} else {
5240 			BUG(); /* only DOORBELL method supported on gfx11 now */
5241 		}
5242 	}
5243 }
5244 
5245 static void gfx_v11_0_ring_emit_hdp_flush(struct amdgpu_ring *ring)
5246 {
5247 	struct amdgpu_device *adev = ring->adev;
5248 	u32 ref_and_mask, reg_mem_engine;
5249 	const struct nbio_hdp_flush_reg *nbio_hf_reg = adev->nbio.hdp_flush_reg;
5250 
5251 	if (ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE) {
5252 		switch (ring->me) {
5253 		case 1:
5254 			ref_and_mask = nbio_hf_reg->ref_and_mask_cp2 << ring->pipe;
5255 			break;
5256 		case 2:
5257 			ref_and_mask = nbio_hf_reg->ref_and_mask_cp6 << ring->pipe;
5258 			break;
5259 		default:
5260 			return;
5261 		}
5262 		reg_mem_engine = 0;
5263 	} else {
5264 		ref_and_mask = nbio_hf_reg->ref_and_mask_cp0;
5265 		reg_mem_engine = 1; /* pfp */
5266 	}
5267 
5268 	gfx_v11_0_wait_reg_mem(ring, reg_mem_engine, 0, 1,
5269 			       adev->nbio.funcs->get_hdp_flush_req_offset(adev),
5270 			       adev->nbio.funcs->get_hdp_flush_done_offset(adev),
5271 			       ref_and_mask, ref_and_mask, 0x20);
5272 }
5273 
5274 static void gfx_v11_0_ring_emit_ib_gfx(struct amdgpu_ring *ring,
5275 				       struct amdgpu_job *job,
5276 				       struct amdgpu_ib *ib,
5277 				       uint32_t flags)
5278 {
5279 	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
5280 	u32 header, control = 0;
5281 
5282 	BUG_ON(ib->flags & AMDGPU_IB_FLAG_CE);
5283 
5284 	header = PACKET3(PACKET3_INDIRECT_BUFFER, 2);
5285 
5286 	control |= ib->length_dw | (vmid << 24);
5287 
5288 	if (ring->adev->gfx.mcbp && (ib->flags & AMDGPU_IB_FLAG_PREEMPT)) {
5289 		control |= INDIRECT_BUFFER_PRE_ENB(1);
5290 
5291 		if (flags & AMDGPU_IB_PREEMPTED)
5292 			control |= INDIRECT_BUFFER_PRE_RESUME(1);
5293 
5294 		if (vmid)
5295 			gfx_v11_0_ring_emit_de_meta(ring,
5296 				    (!amdgpu_sriov_vf(ring->adev) && flags & AMDGPU_IB_PREEMPTED) ? true : false);
5297 	}
5298 
5299 	if (ring->is_mes_queue)
5300 		/* inherit vmid from mqd */
5301 		control |= 0x400000;
5302 
5303 	amdgpu_ring_write(ring, header);
5304 	BUG_ON(ib->gpu_addr & 0x3); /* Dword align */
5305 	amdgpu_ring_write(ring,
5306 #ifdef __BIG_ENDIAN
5307 		(2 << 0) |
5308 #endif
5309 		lower_32_bits(ib->gpu_addr));
5310 	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr));
5311 	amdgpu_ring_write(ring, control);
5312 }
5313 
5314 static void gfx_v11_0_ring_emit_ib_compute(struct amdgpu_ring *ring,
5315 					   struct amdgpu_job *job,
5316 					   struct amdgpu_ib *ib,
5317 					   uint32_t flags)
5318 {
5319 	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
5320 	u32 control = INDIRECT_BUFFER_VALID | ib->length_dw | (vmid << 24);
5321 
5322 	if (ring->is_mes_queue)
5323 		/* inherit vmid from mqd */
5324 		control |= 0x40000000;
5325 
5326 	/* Currently, there is a high possibility to get wave ID mismatch
5327 	 * between ME and GDS, leading to a hw deadlock, because ME generates
5328 	 * different wave IDs than the GDS expects. This situation happens
5329 	 * randomly when at least 5 compute pipes use GDS ordered append.
5330 	 * The wave IDs generated by ME are also wrong after suspend/resume.
5331 	 * Those are probably bugs somewhere else in the kernel driver.
5332 	 *
5333 	 * Writing GDS_COMPUTE_MAX_WAVE_ID resets wave ID counters in ME and
5334 	 * GDS to 0 for this ring (me/pipe).
5335 	 */
5336 	if (ib->flags & AMDGPU_IB_FLAG_RESET_GDS_MAX_WAVE_ID) {
5337 		amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1));
5338 		amdgpu_ring_write(ring, regGDS_COMPUTE_MAX_WAVE_ID);
5339 		amdgpu_ring_write(ring, ring->adev->gds.gds_compute_max_wave_id);
5340 	}
5341 
5342 	amdgpu_ring_write(ring, PACKET3(PACKET3_INDIRECT_BUFFER, 2));
5343 	BUG_ON(ib->gpu_addr & 0x3); /* Dword align */
5344 	amdgpu_ring_write(ring,
5345 #ifdef __BIG_ENDIAN
5346 				(2 << 0) |
5347 #endif
5348 				lower_32_bits(ib->gpu_addr));
5349 	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr));
5350 	amdgpu_ring_write(ring, control);
5351 }
5352 
5353 static void gfx_v11_0_ring_emit_fence(struct amdgpu_ring *ring, u64 addr,
5354 				     u64 seq, unsigned flags)
5355 {
5356 	bool write64bit = flags & AMDGPU_FENCE_FLAG_64BIT;
5357 	bool int_sel = flags & AMDGPU_FENCE_FLAG_INT;
5358 
5359 	/* RELEASE_MEM - flush caches, send int */
5360 	amdgpu_ring_write(ring, PACKET3(PACKET3_RELEASE_MEM, 6));
5361 	amdgpu_ring_write(ring, (PACKET3_RELEASE_MEM_GCR_SEQ |
5362 				 PACKET3_RELEASE_MEM_GCR_GL2_WB |
5363 				 PACKET3_RELEASE_MEM_GCR_GL2_INV |
5364 				 PACKET3_RELEASE_MEM_GCR_GL2_US |
5365 				 PACKET3_RELEASE_MEM_GCR_GL1_INV |
5366 				 PACKET3_RELEASE_MEM_GCR_GLV_INV |
5367 				 PACKET3_RELEASE_MEM_GCR_GLM_INV |
5368 				 PACKET3_RELEASE_MEM_GCR_GLM_WB |
5369 				 PACKET3_RELEASE_MEM_CACHE_POLICY(3) |
5370 				 PACKET3_RELEASE_MEM_EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
5371 				 PACKET3_RELEASE_MEM_EVENT_INDEX(5)));
5372 	amdgpu_ring_write(ring, (PACKET3_RELEASE_MEM_DATA_SEL(write64bit ? 2 : 1) |
5373 				 PACKET3_RELEASE_MEM_INT_SEL(int_sel ? 2 : 0)));
5374 
5375 	/*
5376 	 * the address should be Qword aligned if 64bit write, Dword
5377 	 * aligned if only send 32bit data low (discard data high)
5378 	 */
5379 	if (write64bit)
5380 		BUG_ON(addr & 0x7);
5381 	else
5382 		BUG_ON(addr & 0x3);
5383 	amdgpu_ring_write(ring, lower_32_bits(addr));
5384 	amdgpu_ring_write(ring, upper_32_bits(addr));
5385 	amdgpu_ring_write(ring, lower_32_bits(seq));
5386 	amdgpu_ring_write(ring, upper_32_bits(seq));
5387 	amdgpu_ring_write(ring, ring->is_mes_queue ?
5388 			 (ring->hw_queue_id | AMDGPU_FENCE_MES_QUEUE_FLAG) : 0);
5389 }
5390 
5391 static void gfx_v11_0_ring_emit_pipeline_sync(struct amdgpu_ring *ring)
5392 {
5393 	int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX);
5394 	uint32_t seq = ring->fence_drv.sync_seq;
5395 	uint64_t addr = ring->fence_drv.gpu_addr;
5396 
5397 	gfx_v11_0_wait_reg_mem(ring, usepfp, 1, 0, lower_32_bits(addr),
5398 			       upper_32_bits(addr), seq, 0xffffffff, 4);
5399 }
5400 
5401 static void gfx_v11_0_ring_invalidate_tlbs(struct amdgpu_ring *ring,
5402 				   uint16_t pasid, uint32_t flush_type,
5403 				   bool all_hub, uint8_t dst_sel)
5404 {
5405 	amdgpu_ring_write(ring, PACKET3(PACKET3_INVALIDATE_TLBS, 0));
5406 	amdgpu_ring_write(ring,
5407 			  PACKET3_INVALIDATE_TLBS_DST_SEL(dst_sel) |
5408 			  PACKET3_INVALIDATE_TLBS_ALL_HUB(all_hub) |
5409 			  PACKET3_INVALIDATE_TLBS_PASID(pasid) |
5410 			  PACKET3_INVALIDATE_TLBS_FLUSH_TYPE(flush_type));
5411 }
5412 
5413 static void gfx_v11_0_ring_emit_vm_flush(struct amdgpu_ring *ring,
5414 					 unsigned vmid, uint64_t pd_addr)
5415 {
5416 	if (ring->is_mes_queue)
5417 		gfx_v11_0_ring_invalidate_tlbs(ring, 0, 0, false, 0);
5418 	else
5419 		amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr);
5420 
5421 	/* compute doesn't have PFP */
5422 	if (ring->funcs->type == AMDGPU_RING_TYPE_GFX) {
5423 		/* sync PFP to ME, otherwise we might get invalid PFP reads */
5424 		amdgpu_ring_write(ring, PACKET3(PACKET3_PFP_SYNC_ME, 0));
5425 		amdgpu_ring_write(ring, 0x0);
5426 	}
5427 }
5428 
5429 static void gfx_v11_0_ring_emit_fence_kiq(struct amdgpu_ring *ring, u64 addr,
5430 					  u64 seq, unsigned int flags)
5431 {
5432 	struct amdgpu_device *adev = ring->adev;
5433 
5434 	/* we only allocate 32bit for each seq wb address */
5435 	BUG_ON(flags & AMDGPU_FENCE_FLAG_64BIT);
5436 
5437 	/* write fence seq to the "addr" */
5438 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
5439 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
5440 				 WRITE_DATA_DST_SEL(5) | WR_CONFIRM));
5441 	amdgpu_ring_write(ring, lower_32_bits(addr));
5442 	amdgpu_ring_write(ring, upper_32_bits(addr));
5443 	amdgpu_ring_write(ring, lower_32_bits(seq));
5444 
5445 	if (flags & AMDGPU_FENCE_FLAG_INT) {
5446 		/* set register to trigger INT */
5447 		amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
5448 		amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
5449 					 WRITE_DATA_DST_SEL(0) | WR_CONFIRM));
5450 		amdgpu_ring_write(ring, SOC15_REG_OFFSET(GC, 0, regCPC_INT_STATUS));
5451 		amdgpu_ring_write(ring, 0);
5452 		amdgpu_ring_write(ring, 0x20000000); /* src_id is 178 */
5453 	}
5454 }
5455 
5456 static void gfx_v11_0_ring_emit_cntxcntl(struct amdgpu_ring *ring,
5457 					 uint32_t flags)
5458 {
5459 	uint32_t dw2 = 0;
5460 
5461 	dw2 |= 0x80000000; /* set load_enable otherwise this package is just NOPs */
5462 	if (flags & AMDGPU_HAVE_CTX_SWITCH) {
5463 		/* set load_global_config & load_global_uconfig */
5464 		dw2 |= 0x8001;
5465 		/* set load_cs_sh_regs */
5466 		dw2 |= 0x01000000;
5467 		/* set load_per_context_state & load_gfx_sh_regs for GFX */
5468 		dw2 |= 0x10002;
5469 	}
5470 
5471 	amdgpu_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1));
5472 	amdgpu_ring_write(ring, dw2);
5473 	amdgpu_ring_write(ring, 0);
5474 }
5475 
5476 static void gfx_v11_0_ring_emit_gfx_shadow(struct amdgpu_ring *ring,
5477 					   u64 shadow_va, u64 csa_va,
5478 					   u64 gds_va, bool init_shadow,
5479 					   int vmid)
5480 {
5481 	struct amdgpu_device *adev = ring->adev;
5482 
5483 	if (!adev->gfx.cp_gfx_shadow)
5484 		return;
5485 
5486 	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_Q_PREEMPTION_MODE, 7));
5487 	amdgpu_ring_write(ring, lower_32_bits(shadow_va));
5488 	amdgpu_ring_write(ring, upper_32_bits(shadow_va));
5489 	amdgpu_ring_write(ring, lower_32_bits(gds_va));
5490 	amdgpu_ring_write(ring, upper_32_bits(gds_va));
5491 	amdgpu_ring_write(ring, lower_32_bits(csa_va));
5492 	amdgpu_ring_write(ring, upper_32_bits(csa_va));
5493 	amdgpu_ring_write(ring, shadow_va ?
5494 			  PACKET3_SET_Q_PREEMPTION_MODE_IB_VMID(vmid) : 0);
5495 	amdgpu_ring_write(ring, init_shadow ?
5496 			  PACKET3_SET_Q_PREEMPTION_MODE_INIT_SHADOW_MEM : 0);
5497 }
5498 
5499 static unsigned gfx_v11_0_ring_emit_init_cond_exec(struct amdgpu_ring *ring)
5500 {
5501 	unsigned ret;
5502 
5503 	amdgpu_ring_write(ring, PACKET3(PACKET3_COND_EXEC, 3));
5504 	amdgpu_ring_write(ring, lower_32_bits(ring->cond_exe_gpu_addr));
5505 	amdgpu_ring_write(ring, upper_32_bits(ring->cond_exe_gpu_addr));
5506 	amdgpu_ring_write(ring, 0); /* discard following DWs if *cond_exec_gpu_addr==0 */
5507 	ret = ring->wptr & ring->buf_mask;
5508 	amdgpu_ring_write(ring, 0x55aa55aa); /* patch dummy value later */
5509 
5510 	return ret;
5511 }
5512 
5513 static void gfx_v11_0_ring_emit_patch_cond_exec(struct amdgpu_ring *ring, unsigned offset)
5514 {
5515 	unsigned cur;
5516 	BUG_ON(offset > ring->buf_mask);
5517 	BUG_ON(ring->ring[offset] != 0x55aa55aa);
5518 
5519 	cur = (ring->wptr - 1) & ring->buf_mask;
5520 	if (likely(cur > offset))
5521 		ring->ring[offset] = cur - offset;
5522 	else
5523 		ring->ring[offset] = (ring->buf_mask + 1) - offset + cur;
5524 }
5525 
5526 static int gfx_v11_0_ring_preempt_ib(struct amdgpu_ring *ring)
5527 {
5528 	int i, r = 0;
5529 	struct amdgpu_device *adev = ring->adev;
5530 	struct amdgpu_kiq *kiq = &adev->gfx.kiq[0];
5531 	struct amdgpu_ring *kiq_ring = &kiq->ring;
5532 	unsigned long flags;
5533 
5534 	if (!kiq->pmf || !kiq->pmf->kiq_unmap_queues)
5535 		return -EINVAL;
5536 
5537 	spin_lock_irqsave(&kiq->ring_lock, flags);
5538 
5539 	if (amdgpu_ring_alloc(kiq_ring, kiq->pmf->unmap_queues_size)) {
5540 		spin_unlock_irqrestore(&kiq->ring_lock, flags);
5541 		return -ENOMEM;
5542 	}
5543 
5544 	/* assert preemption condition */
5545 	amdgpu_ring_set_preempt_cond_exec(ring, false);
5546 
5547 	/* assert IB preemption, emit the trailing fence */
5548 	kiq->pmf->kiq_unmap_queues(kiq_ring, ring, PREEMPT_QUEUES_NO_UNMAP,
5549 				   ring->trail_fence_gpu_addr,
5550 				   ++ring->trail_seq);
5551 	amdgpu_ring_commit(kiq_ring);
5552 
5553 	spin_unlock_irqrestore(&kiq->ring_lock, flags);
5554 
5555 	/* poll the trailing fence */
5556 	for (i = 0; i < adev->usec_timeout; i++) {
5557 		if (ring->trail_seq ==
5558 		    le32_to_cpu(*(ring->trail_fence_cpu_addr)))
5559 			break;
5560 		udelay(1);
5561 	}
5562 
5563 	if (i >= adev->usec_timeout) {
5564 		r = -EINVAL;
5565 		DRM_ERROR("ring %d failed to preempt ib\n", ring->idx);
5566 	}
5567 
5568 	/* deassert preemption condition */
5569 	amdgpu_ring_set_preempt_cond_exec(ring, true);
5570 	return r;
5571 }
5572 
5573 static void gfx_v11_0_ring_emit_de_meta(struct amdgpu_ring *ring, bool resume)
5574 {
5575 	struct amdgpu_device *adev = ring->adev;
5576 	struct v10_de_ib_state de_payload = {0};
5577 	uint64_t offset, gds_addr, de_payload_gpu_addr;
5578 	void *de_payload_cpu_addr;
5579 	int cnt;
5580 
5581 	if (ring->is_mes_queue) {
5582 		offset = offsetof(struct amdgpu_mes_ctx_meta_data,
5583 				  gfx[0].gfx_meta_data) +
5584 			offsetof(struct v10_gfx_meta_data, de_payload);
5585 		de_payload_gpu_addr =
5586 			amdgpu_mes_ctx_get_offs_gpu_addr(ring, offset);
5587 		de_payload_cpu_addr =
5588 			amdgpu_mes_ctx_get_offs_cpu_addr(ring, offset);
5589 
5590 		offset = offsetof(struct amdgpu_mes_ctx_meta_data,
5591 				  gfx[0].gds_backup) +
5592 			offsetof(struct v10_gfx_meta_data, de_payload);
5593 		gds_addr = amdgpu_mes_ctx_get_offs_gpu_addr(ring, offset);
5594 	} else {
5595 		offset = offsetof(struct v10_gfx_meta_data, de_payload);
5596 		de_payload_gpu_addr = amdgpu_csa_vaddr(ring->adev) + offset;
5597 		de_payload_cpu_addr = adev->virt.csa_cpu_addr + offset;
5598 
5599 		gds_addr = ALIGN(amdgpu_csa_vaddr(ring->adev) +
5600 				 AMDGPU_CSA_SIZE - adev->gds.gds_size,
5601 				 PAGE_SIZE);
5602 	}
5603 
5604 	de_payload.gds_backup_addrlo = lower_32_bits(gds_addr);
5605 	de_payload.gds_backup_addrhi = upper_32_bits(gds_addr);
5606 
5607 	cnt = (sizeof(de_payload) >> 2) + 4 - 2;
5608 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, cnt));
5609 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(1) |
5610 				 WRITE_DATA_DST_SEL(8) |
5611 				 WR_CONFIRM) |
5612 				 WRITE_DATA_CACHE_POLICY(0));
5613 	amdgpu_ring_write(ring, lower_32_bits(de_payload_gpu_addr));
5614 	amdgpu_ring_write(ring, upper_32_bits(de_payload_gpu_addr));
5615 
5616 	if (resume)
5617 		amdgpu_ring_write_multiple(ring, de_payload_cpu_addr,
5618 					   sizeof(de_payload) >> 2);
5619 	else
5620 		amdgpu_ring_write_multiple(ring, (void *)&de_payload,
5621 					   sizeof(de_payload) >> 2);
5622 }
5623 
5624 static void gfx_v11_0_ring_emit_frame_cntl(struct amdgpu_ring *ring, bool start,
5625 				    bool secure)
5626 {
5627 	uint32_t v = secure ? FRAME_TMZ : 0;
5628 
5629 	amdgpu_ring_write(ring, PACKET3(PACKET3_FRAME_CONTROL, 0));
5630 	amdgpu_ring_write(ring, v | FRAME_CMD(start ? 0 : 1));
5631 }
5632 
5633 static void gfx_v11_0_ring_emit_rreg(struct amdgpu_ring *ring, uint32_t reg,
5634 				     uint32_t reg_val_offs)
5635 {
5636 	struct amdgpu_device *adev = ring->adev;
5637 
5638 	amdgpu_ring_write(ring, PACKET3(PACKET3_COPY_DATA, 4));
5639 	amdgpu_ring_write(ring, 0 |	/* src: register*/
5640 				(5 << 8) |	/* dst: memory */
5641 				(1 << 20));	/* write confirm */
5642 	amdgpu_ring_write(ring, reg);
5643 	amdgpu_ring_write(ring, 0);
5644 	amdgpu_ring_write(ring, lower_32_bits(adev->wb.gpu_addr +
5645 				reg_val_offs * 4));
5646 	amdgpu_ring_write(ring, upper_32_bits(adev->wb.gpu_addr +
5647 				reg_val_offs * 4));
5648 }
5649 
5650 static void gfx_v11_0_ring_emit_wreg(struct amdgpu_ring *ring, uint32_t reg,
5651 				   uint32_t val)
5652 {
5653 	uint32_t cmd = 0;
5654 
5655 	switch (ring->funcs->type) {
5656 	case AMDGPU_RING_TYPE_GFX:
5657 		cmd = WRITE_DATA_ENGINE_SEL(1) | WR_CONFIRM;
5658 		break;
5659 	case AMDGPU_RING_TYPE_KIQ:
5660 		cmd = (1 << 16); /* no inc addr */
5661 		break;
5662 	default:
5663 		cmd = WR_CONFIRM;
5664 		break;
5665 	}
5666 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
5667 	amdgpu_ring_write(ring, cmd);
5668 	amdgpu_ring_write(ring, reg);
5669 	amdgpu_ring_write(ring, 0);
5670 	amdgpu_ring_write(ring, val);
5671 }
5672 
5673 static void gfx_v11_0_ring_emit_reg_wait(struct amdgpu_ring *ring, uint32_t reg,
5674 					uint32_t val, uint32_t mask)
5675 {
5676 	gfx_v11_0_wait_reg_mem(ring, 0, 0, 0, reg, 0, val, mask, 0x20);
5677 }
5678 
5679 static void gfx_v11_0_ring_emit_reg_write_reg_wait(struct amdgpu_ring *ring,
5680 						   uint32_t reg0, uint32_t reg1,
5681 						   uint32_t ref, uint32_t mask)
5682 {
5683 	int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX);
5684 
5685 	gfx_v11_0_wait_reg_mem(ring, usepfp, 0, 1, reg0, reg1,
5686 			       ref, mask, 0x20);
5687 }
5688 
5689 static void gfx_v11_0_ring_soft_recovery(struct amdgpu_ring *ring,
5690 					 unsigned vmid)
5691 {
5692 	struct amdgpu_device *adev = ring->adev;
5693 	uint32_t value = 0;
5694 
5695 	value = REG_SET_FIELD(value, SQ_CMD, CMD, 0x03);
5696 	value = REG_SET_FIELD(value, SQ_CMD, MODE, 0x01);
5697 	value = REG_SET_FIELD(value, SQ_CMD, CHECK_VMID, 1);
5698 	value = REG_SET_FIELD(value, SQ_CMD, VM_ID, vmid);
5699 	WREG32_SOC15(GC, 0, regSQ_CMD, value);
5700 }
5701 
5702 static void
5703 gfx_v11_0_set_gfx_eop_interrupt_state(struct amdgpu_device *adev,
5704 				      uint32_t me, uint32_t pipe,
5705 				      enum amdgpu_interrupt_state state)
5706 {
5707 	uint32_t cp_int_cntl, cp_int_cntl_reg;
5708 
5709 	if (!me) {
5710 		switch (pipe) {
5711 		case 0:
5712 			cp_int_cntl_reg = SOC15_REG_OFFSET(GC, 0, regCP_INT_CNTL_RING0);
5713 			break;
5714 		case 1:
5715 			cp_int_cntl_reg = SOC15_REG_OFFSET(GC, 0, regCP_INT_CNTL_RING1);
5716 			break;
5717 		default:
5718 			DRM_DEBUG("invalid pipe %d\n", pipe);
5719 			return;
5720 		}
5721 	} else {
5722 		DRM_DEBUG("invalid me %d\n", me);
5723 		return;
5724 	}
5725 
5726 	switch (state) {
5727 	case AMDGPU_IRQ_STATE_DISABLE:
5728 		cp_int_cntl = RREG32_SOC15_IP(GC, cp_int_cntl_reg);
5729 		cp_int_cntl = REG_SET_FIELD(cp_int_cntl, CP_INT_CNTL_RING0,
5730 					    TIME_STAMP_INT_ENABLE, 0);
5731 		cp_int_cntl = REG_SET_FIELD(cp_int_cntl, CP_INT_CNTL_RING0,
5732 					    GENERIC0_INT_ENABLE, 0);
5733 		WREG32_SOC15_IP(GC, cp_int_cntl_reg, cp_int_cntl);
5734 		break;
5735 	case AMDGPU_IRQ_STATE_ENABLE:
5736 		cp_int_cntl = RREG32_SOC15_IP(GC, cp_int_cntl_reg);
5737 		cp_int_cntl = REG_SET_FIELD(cp_int_cntl, CP_INT_CNTL_RING0,
5738 					    TIME_STAMP_INT_ENABLE, 1);
5739 		cp_int_cntl = REG_SET_FIELD(cp_int_cntl, CP_INT_CNTL_RING0,
5740 					    GENERIC0_INT_ENABLE, 1);
5741 		WREG32_SOC15_IP(GC, cp_int_cntl_reg, cp_int_cntl);
5742 		break;
5743 	default:
5744 		break;
5745 	}
5746 }
5747 
5748 static void gfx_v11_0_set_compute_eop_interrupt_state(struct amdgpu_device *adev,
5749 						     int me, int pipe,
5750 						     enum amdgpu_interrupt_state state)
5751 {
5752 	u32 mec_int_cntl, mec_int_cntl_reg;
5753 
5754 	/*
5755 	 * amdgpu controls only the first MEC. That's why this function only
5756 	 * handles the setting of interrupts for this specific MEC. All other
5757 	 * pipes' interrupts are set by amdkfd.
5758 	 */
5759 
5760 	if (me == 1) {
5761 		switch (pipe) {
5762 		case 0:
5763 			mec_int_cntl_reg = SOC15_REG_OFFSET(GC, 0, regCP_ME1_PIPE0_INT_CNTL);
5764 			break;
5765 		case 1:
5766 			mec_int_cntl_reg = SOC15_REG_OFFSET(GC, 0, regCP_ME1_PIPE1_INT_CNTL);
5767 			break;
5768 		case 2:
5769 			mec_int_cntl_reg = SOC15_REG_OFFSET(GC, 0, regCP_ME1_PIPE2_INT_CNTL);
5770 			break;
5771 		case 3:
5772 			mec_int_cntl_reg = SOC15_REG_OFFSET(GC, 0, regCP_ME1_PIPE3_INT_CNTL);
5773 			break;
5774 		default:
5775 			DRM_DEBUG("invalid pipe %d\n", pipe);
5776 			return;
5777 		}
5778 	} else {
5779 		DRM_DEBUG("invalid me %d\n", me);
5780 		return;
5781 	}
5782 
5783 	switch (state) {
5784 	case AMDGPU_IRQ_STATE_DISABLE:
5785 		mec_int_cntl = RREG32_SOC15_IP(GC, mec_int_cntl_reg);
5786 		mec_int_cntl = REG_SET_FIELD(mec_int_cntl, CP_ME1_PIPE0_INT_CNTL,
5787 					     TIME_STAMP_INT_ENABLE, 0);
5788 		mec_int_cntl = REG_SET_FIELD(mec_int_cntl, CP_ME1_PIPE0_INT_CNTL,
5789 					     GENERIC0_INT_ENABLE, 0);
5790 		WREG32_SOC15_IP(GC, mec_int_cntl_reg, mec_int_cntl);
5791 		break;
5792 	case AMDGPU_IRQ_STATE_ENABLE:
5793 		mec_int_cntl = RREG32_SOC15_IP(GC, mec_int_cntl_reg);
5794 		mec_int_cntl = REG_SET_FIELD(mec_int_cntl, CP_ME1_PIPE0_INT_CNTL,
5795 					     TIME_STAMP_INT_ENABLE, 1);
5796 		mec_int_cntl = REG_SET_FIELD(mec_int_cntl, CP_ME1_PIPE0_INT_CNTL,
5797 					     GENERIC0_INT_ENABLE, 1);
5798 		WREG32_SOC15_IP(GC, mec_int_cntl_reg, mec_int_cntl);
5799 		break;
5800 	default:
5801 		break;
5802 	}
5803 }
5804 
5805 static int gfx_v11_0_set_eop_interrupt_state(struct amdgpu_device *adev,
5806 					    struct amdgpu_irq_src *src,
5807 					    unsigned type,
5808 					    enum amdgpu_interrupt_state state)
5809 {
5810 	switch (type) {
5811 	case AMDGPU_CP_IRQ_GFX_ME0_PIPE0_EOP:
5812 		gfx_v11_0_set_gfx_eop_interrupt_state(adev, 0, 0, state);
5813 		break;
5814 	case AMDGPU_CP_IRQ_GFX_ME0_PIPE1_EOP:
5815 		gfx_v11_0_set_gfx_eop_interrupt_state(adev, 0, 1, state);
5816 		break;
5817 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP:
5818 		gfx_v11_0_set_compute_eop_interrupt_state(adev, 1, 0, state);
5819 		break;
5820 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE1_EOP:
5821 		gfx_v11_0_set_compute_eop_interrupt_state(adev, 1, 1, state);
5822 		break;
5823 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE2_EOP:
5824 		gfx_v11_0_set_compute_eop_interrupt_state(adev, 1, 2, state);
5825 		break;
5826 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE3_EOP:
5827 		gfx_v11_0_set_compute_eop_interrupt_state(adev, 1, 3, state);
5828 		break;
5829 	default:
5830 		break;
5831 	}
5832 	return 0;
5833 }
5834 
5835 static int gfx_v11_0_eop_irq(struct amdgpu_device *adev,
5836 			     struct amdgpu_irq_src *source,
5837 			     struct amdgpu_iv_entry *entry)
5838 {
5839 	int i;
5840 	u8 me_id, pipe_id, queue_id;
5841 	struct amdgpu_ring *ring;
5842 	uint32_t mes_queue_id = entry->src_data[0];
5843 
5844 	DRM_DEBUG("IH: CP EOP\n");
5845 
5846 	if (adev->enable_mes && (mes_queue_id & AMDGPU_FENCE_MES_QUEUE_FLAG)) {
5847 		struct amdgpu_mes_queue *queue;
5848 
5849 		mes_queue_id &= AMDGPU_FENCE_MES_QUEUE_ID_MASK;
5850 
5851 		spin_lock(&adev->mes.queue_id_lock);
5852 		queue = idr_find(&adev->mes.queue_id_idr, mes_queue_id);
5853 		if (queue) {
5854 			DRM_DEBUG("process mes queue id = %d\n", mes_queue_id);
5855 			amdgpu_fence_process(queue->ring);
5856 		}
5857 		spin_unlock(&adev->mes.queue_id_lock);
5858 	} else {
5859 		me_id = (entry->ring_id & 0x0c) >> 2;
5860 		pipe_id = (entry->ring_id & 0x03) >> 0;
5861 		queue_id = (entry->ring_id & 0x70) >> 4;
5862 
5863 		switch (me_id) {
5864 		case 0:
5865 			if (pipe_id == 0)
5866 				amdgpu_fence_process(&adev->gfx.gfx_ring[0]);
5867 			else
5868 				amdgpu_fence_process(&adev->gfx.gfx_ring[1]);
5869 			break;
5870 		case 1:
5871 		case 2:
5872 			for (i = 0; i < adev->gfx.num_compute_rings; i++) {
5873 				ring = &adev->gfx.compute_ring[i];
5874 				/* Per-queue interrupt is supported for MEC starting from VI.
5875 				 * The interrupt can only be enabled/disabled per pipe instead
5876 				 * of per queue.
5877 				 */
5878 				if ((ring->me == me_id) &&
5879 				    (ring->pipe == pipe_id) &&
5880 				    (ring->queue == queue_id))
5881 					amdgpu_fence_process(ring);
5882 			}
5883 			break;
5884 		}
5885 	}
5886 
5887 	return 0;
5888 }
5889 
5890 static int gfx_v11_0_set_priv_reg_fault_state(struct amdgpu_device *adev,
5891 					      struct amdgpu_irq_src *source,
5892 					      unsigned type,
5893 					      enum amdgpu_interrupt_state state)
5894 {
5895 	switch (state) {
5896 	case AMDGPU_IRQ_STATE_DISABLE:
5897 	case AMDGPU_IRQ_STATE_ENABLE:
5898 		WREG32_FIELD15_PREREG(GC, 0, CP_INT_CNTL_RING0,
5899 			       PRIV_REG_INT_ENABLE,
5900 			       state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0);
5901 		break;
5902 	default:
5903 		break;
5904 	}
5905 
5906 	return 0;
5907 }
5908 
5909 static int gfx_v11_0_set_priv_inst_fault_state(struct amdgpu_device *adev,
5910 					       struct amdgpu_irq_src *source,
5911 					       unsigned type,
5912 					       enum amdgpu_interrupt_state state)
5913 {
5914 	switch (state) {
5915 	case AMDGPU_IRQ_STATE_DISABLE:
5916 	case AMDGPU_IRQ_STATE_ENABLE:
5917 		WREG32_FIELD15_PREREG(GC, 0, CP_INT_CNTL_RING0,
5918 			       PRIV_INSTR_INT_ENABLE,
5919 			       state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0);
5920 		break;
5921 	default:
5922 		break;
5923 	}
5924 
5925 	return 0;
5926 }
5927 
5928 static void gfx_v11_0_handle_priv_fault(struct amdgpu_device *adev,
5929 					struct amdgpu_iv_entry *entry)
5930 {
5931 	u8 me_id, pipe_id, queue_id;
5932 	struct amdgpu_ring *ring;
5933 	int i;
5934 
5935 	me_id = (entry->ring_id & 0x0c) >> 2;
5936 	pipe_id = (entry->ring_id & 0x03) >> 0;
5937 	queue_id = (entry->ring_id & 0x70) >> 4;
5938 
5939 	switch (me_id) {
5940 	case 0:
5941 		for (i = 0; i < adev->gfx.num_gfx_rings; i++) {
5942 			ring = &adev->gfx.gfx_ring[i];
5943 			/* we only enabled 1 gfx queue per pipe for now */
5944 			if (ring->me == me_id && ring->pipe == pipe_id)
5945 				drm_sched_fault(&ring->sched);
5946 		}
5947 		break;
5948 	case 1:
5949 	case 2:
5950 		for (i = 0; i < adev->gfx.num_compute_rings; i++) {
5951 			ring = &adev->gfx.compute_ring[i];
5952 			if (ring->me == me_id && ring->pipe == pipe_id &&
5953 			    ring->queue == queue_id)
5954 				drm_sched_fault(&ring->sched);
5955 		}
5956 		break;
5957 	default:
5958 		BUG();
5959 		break;
5960 	}
5961 }
5962 
5963 static int gfx_v11_0_priv_reg_irq(struct amdgpu_device *adev,
5964 				  struct amdgpu_irq_src *source,
5965 				  struct amdgpu_iv_entry *entry)
5966 {
5967 	DRM_ERROR("Illegal register access in command stream\n");
5968 	gfx_v11_0_handle_priv_fault(adev, entry);
5969 	return 0;
5970 }
5971 
5972 static int gfx_v11_0_priv_inst_irq(struct amdgpu_device *adev,
5973 				   struct amdgpu_irq_src *source,
5974 				   struct amdgpu_iv_entry *entry)
5975 {
5976 	DRM_ERROR("Illegal instruction in command stream\n");
5977 	gfx_v11_0_handle_priv_fault(adev, entry);
5978 	return 0;
5979 }
5980 
5981 static int gfx_v11_0_rlc_gc_fed_irq(struct amdgpu_device *adev,
5982 				  struct amdgpu_irq_src *source,
5983 				  struct amdgpu_iv_entry *entry)
5984 {
5985 	if (adev->gfx.ras && adev->gfx.ras->rlc_gc_fed_irq)
5986 		return adev->gfx.ras->rlc_gc_fed_irq(adev, source, entry);
5987 
5988 	return 0;
5989 }
5990 
5991 #if 0
5992 static int gfx_v11_0_kiq_set_interrupt_state(struct amdgpu_device *adev,
5993 					     struct amdgpu_irq_src *src,
5994 					     unsigned int type,
5995 					     enum amdgpu_interrupt_state state)
5996 {
5997 	uint32_t tmp, target;
5998 	struct amdgpu_ring *ring = &(adev->gfx.kiq[0].ring);
5999 
6000 	target = SOC15_REG_OFFSET(GC, 0, regCP_ME1_PIPE0_INT_CNTL);
6001 	target += ring->pipe;
6002 
6003 	switch (type) {
6004 	case AMDGPU_CP_KIQ_IRQ_DRIVER0:
6005 		if (state == AMDGPU_IRQ_STATE_DISABLE) {
6006 			tmp = RREG32_SOC15(GC, 0, regCPC_INT_CNTL);
6007 			tmp = REG_SET_FIELD(tmp, CPC_INT_CNTL,
6008 					    GENERIC2_INT_ENABLE, 0);
6009 			WREG32_SOC15(GC, 0, regCPC_INT_CNTL, tmp);
6010 
6011 			tmp = RREG32_SOC15_IP(GC, target);
6012 			tmp = REG_SET_FIELD(tmp, CP_ME1_PIPE0_INT_CNTL,
6013 					    GENERIC2_INT_ENABLE, 0);
6014 			WREG32_SOC15_IP(GC, target, tmp);
6015 		} else {
6016 			tmp = RREG32_SOC15(GC, 0, regCPC_INT_CNTL);
6017 			tmp = REG_SET_FIELD(tmp, CPC_INT_CNTL,
6018 					    GENERIC2_INT_ENABLE, 1);
6019 			WREG32_SOC15(GC, 0, regCPC_INT_CNTL, tmp);
6020 
6021 			tmp = RREG32_SOC15_IP(GC, target);
6022 			tmp = REG_SET_FIELD(tmp, CP_ME1_PIPE0_INT_CNTL,
6023 					    GENERIC2_INT_ENABLE, 1);
6024 			WREG32_SOC15_IP(GC, target, tmp);
6025 		}
6026 		break;
6027 	default:
6028 		BUG(); /* kiq only support GENERIC2_INT now */
6029 		break;
6030 	}
6031 	return 0;
6032 }
6033 #endif
6034 
6035 static void gfx_v11_0_emit_mem_sync(struct amdgpu_ring *ring)
6036 {
6037 	const unsigned int gcr_cntl =
6038 			PACKET3_ACQUIRE_MEM_GCR_CNTL_GL2_INV(1) |
6039 			PACKET3_ACQUIRE_MEM_GCR_CNTL_GL2_WB(1) |
6040 			PACKET3_ACQUIRE_MEM_GCR_CNTL_GLM_INV(1) |
6041 			PACKET3_ACQUIRE_MEM_GCR_CNTL_GLM_WB(1) |
6042 			PACKET3_ACQUIRE_MEM_GCR_CNTL_GL1_INV(1) |
6043 			PACKET3_ACQUIRE_MEM_GCR_CNTL_GLV_INV(1) |
6044 			PACKET3_ACQUIRE_MEM_GCR_CNTL_GLK_INV(1) |
6045 			PACKET3_ACQUIRE_MEM_GCR_CNTL_GLI_INV(1);
6046 
6047 	/* ACQUIRE_MEM - make one or more surfaces valid for use by the subsequent operations */
6048 	amdgpu_ring_write(ring, PACKET3(PACKET3_ACQUIRE_MEM, 6));
6049 	amdgpu_ring_write(ring, 0); /* CP_COHER_CNTL */
6050 	amdgpu_ring_write(ring, 0xffffffff);  /* CP_COHER_SIZE */
6051 	amdgpu_ring_write(ring, 0xffffff);  /* CP_COHER_SIZE_HI */
6052 	amdgpu_ring_write(ring, 0); /* CP_COHER_BASE */
6053 	amdgpu_ring_write(ring, 0);  /* CP_COHER_BASE_HI */
6054 	amdgpu_ring_write(ring, 0x0000000A); /* POLL_INTERVAL */
6055 	amdgpu_ring_write(ring, gcr_cntl); /* GCR_CNTL */
6056 }
6057 
6058 static const struct amd_ip_funcs gfx_v11_0_ip_funcs = {
6059 	.name = "gfx_v11_0",
6060 	.early_init = gfx_v11_0_early_init,
6061 	.late_init = gfx_v11_0_late_init,
6062 	.sw_init = gfx_v11_0_sw_init,
6063 	.sw_fini = gfx_v11_0_sw_fini,
6064 	.hw_init = gfx_v11_0_hw_init,
6065 	.hw_fini = gfx_v11_0_hw_fini,
6066 	.suspend = gfx_v11_0_suspend,
6067 	.resume = gfx_v11_0_resume,
6068 	.is_idle = gfx_v11_0_is_idle,
6069 	.wait_for_idle = gfx_v11_0_wait_for_idle,
6070 	.soft_reset = gfx_v11_0_soft_reset,
6071 	.check_soft_reset = gfx_v11_0_check_soft_reset,
6072 	.post_soft_reset = gfx_v11_0_post_soft_reset,
6073 	.set_clockgating_state = gfx_v11_0_set_clockgating_state,
6074 	.set_powergating_state = gfx_v11_0_set_powergating_state,
6075 	.get_clockgating_state = gfx_v11_0_get_clockgating_state,
6076 };
6077 
6078 static const struct amdgpu_ring_funcs gfx_v11_0_ring_funcs_gfx = {
6079 	.type = AMDGPU_RING_TYPE_GFX,
6080 	.align_mask = 0xff,
6081 	.nop = PACKET3(PACKET3_NOP, 0x3FFF),
6082 	.support_64bit_ptrs = true,
6083 	.secure_submission_supported = true,
6084 	.get_rptr = gfx_v11_0_ring_get_rptr_gfx,
6085 	.get_wptr = gfx_v11_0_ring_get_wptr_gfx,
6086 	.set_wptr = gfx_v11_0_ring_set_wptr_gfx,
6087 	.emit_frame_size = /* totally 242 maximum if 16 IBs */
6088 		5 + /* COND_EXEC */
6089 		9 + /* SET_Q_PREEMPTION_MODE */
6090 		7 + /* PIPELINE_SYNC */
6091 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 5 +
6092 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 7 +
6093 		2 + /* VM_FLUSH */
6094 		8 + /* FENCE for VM_FLUSH */
6095 		20 + /* GDS switch */
6096 		5 + /* COND_EXEC */
6097 		7 + /* HDP_flush */
6098 		4 + /* VGT_flush */
6099 		31 + /*	DE_META */
6100 		3 + /* CNTX_CTRL */
6101 		5 + /* HDP_INVL */
6102 		8 + 8 + /* FENCE x2 */
6103 		8, /* gfx_v11_0_emit_mem_sync */
6104 	.emit_ib_size =	4, /* gfx_v11_0_ring_emit_ib_gfx */
6105 	.emit_ib = gfx_v11_0_ring_emit_ib_gfx,
6106 	.emit_fence = gfx_v11_0_ring_emit_fence,
6107 	.emit_pipeline_sync = gfx_v11_0_ring_emit_pipeline_sync,
6108 	.emit_vm_flush = gfx_v11_0_ring_emit_vm_flush,
6109 	.emit_gds_switch = gfx_v11_0_ring_emit_gds_switch,
6110 	.emit_hdp_flush = gfx_v11_0_ring_emit_hdp_flush,
6111 	.test_ring = gfx_v11_0_ring_test_ring,
6112 	.test_ib = gfx_v11_0_ring_test_ib,
6113 	.insert_nop = amdgpu_ring_insert_nop,
6114 	.pad_ib = amdgpu_ring_generic_pad_ib,
6115 	.emit_cntxcntl = gfx_v11_0_ring_emit_cntxcntl,
6116 	.emit_gfx_shadow = gfx_v11_0_ring_emit_gfx_shadow,
6117 	.init_cond_exec = gfx_v11_0_ring_emit_init_cond_exec,
6118 	.patch_cond_exec = gfx_v11_0_ring_emit_patch_cond_exec,
6119 	.preempt_ib = gfx_v11_0_ring_preempt_ib,
6120 	.emit_frame_cntl = gfx_v11_0_ring_emit_frame_cntl,
6121 	.emit_wreg = gfx_v11_0_ring_emit_wreg,
6122 	.emit_reg_wait = gfx_v11_0_ring_emit_reg_wait,
6123 	.emit_reg_write_reg_wait = gfx_v11_0_ring_emit_reg_write_reg_wait,
6124 	.soft_recovery = gfx_v11_0_ring_soft_recovery,
6125 	.emit_mem_sync = gfx_v11_0_emit_mem_sync,
6126 };
6127 
6128 static const struct amdgpu_ring_funcs gfx_v11_0_ring_funcs_compute = {
6129 	.type = AMDGPU_RING_TYPE_COMPUTE,
6130 	.align_mask = 0xff,
6131 	.nop = PACKET3(PACKET3_NOP, 0x3FFF),
6132 	.support_64bit_ptrs = true,
6133 	.get_rptr = gfx_v11_0_ring_get_rptr_compute,
6134 	.get_wptr = gfx_v11_0_ring_get_wptr_compute,
6135 	.set_wptr = gfx_v11_0_ring_set_wptr_compute,
6136 	.emit_frame_size =
6137 		20 + /* gfx_v11_0_ring_emit_gds_switch */
6138 		7 + /* gfx_v11_0_ring_emit_hdp_flush */
6139 		5 + /* hdp invalidate */
6140 		7 + /* gfx_v11_0_ring_emit_pipeline_sync */
6141 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 5 +
6142 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 7 +
6143 		2 + /* gfx_v11_0_ring_emit_vm_flush */
6144 		8 + 8 + 8 + /* gfx_v11_0_ring_emit_fence x3 for user fence, vm fence */
6145 		8, /* gfx_v11_0_emit_mem_sync */
6146 	.emit_ib_size =	7, /* gfx_v11_0_ring_emit_ib_compute */
6147 	.emit_ib = gfx_v11_0_ring_emit_ib_compute,
6148 	.emit_fence = gfx_v11_0_ring_emit_fence,
6149 	.emit_pipeline_sync = gfx_v11_0_ring_emit_pipeline_sync,
6150 	.emit_vm_flush = gfx_v11_0_ring_emit_vm_flush,
6151 	.emit_gds_switch = gfx_v11_0_ring_emit_gds_switch,
6152 	.emit_hdp_flush = gfx_v11_0_ring_emit_hdp_flush,
6153 	.test_ring = gfx_v11_0_ring_test_ring,
6154 	.test_ib = gfx_v11_0_ring_test_ib,
6155 	.insert_nop = amdgpu_ring_insert_nop,
6156 	.pad_ib = amdgpu_ring_generic_pad_ib,
6157 	.emit_wreg = gfx_v11_0_ring_emit_wreg,
6158 	.emit_reg_wait = gfx_v11_0_ring_emit_reg_wait,
6159 	.emit_reg_write_reg_wait = gfx_v11_0_ring_emit_reg_write_reg_wait,
6160 	.emit_mem_sync = gfx_v11_0_emit_mem_sync,
6161 };
6162 
6163 static const struct amdgpu_ring_funcs gfx_v11_0_ring_funcs_kiq = {
6164 	.type = AMDGPU_RING_TYPE_KIQ,
6165 	.align_mask = 0xff,
6166 	.nop = PACKET3(PACKET3_NOP, 0x3FFF),
6167 	.support_64bit_ptrs = true,
6168 	.get_rptr = gfx_v11_0_ring_get_rptr_compute,
6169 	.get_wptr = gfx_v11_0_ring_get_wptr_compute,
6170 	.set_wptr = gfx_v11_0_ring_set_wptr_compute,
6171 	.emit_frame_size =
6172 		20 + /* gfx_v11_0_ring_emit_gds_switch */
6173 		7 + /* gfx_v11_0_ring_emit_hdp_flush */
6174 		5 + /*hdp invalidate */
6175 		7 + /* gfx_v11_0_ring_emit_pipeline_sync */
6176 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 5 +
6177 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 7 +
6178 		2 + /* gfx_v11_0_ring_emit_vm_flush */
6179 		8 + 8 + 8, /* gfx_v11_0_ring_emit_fence_kiq x3 for user fence, vm fence */
6180 	.emit_ib_size =	7, /* gfx_v11_0_ring_emit_ib_compute */
6181 	.emit_ib = gfx_v11_0_ring_emit_ib_compute,
6182 	.emit_fence = gfx_v11_0_ring_emit_fence_kiq,
6183 	.test_ring = gfx_v11_0_ring_test_ring,
6184 	.test_ib = gfx_v11_0_ring_test_ib,
6185 	.insert_nop = amdgpu_ring_insert_nop,
6186 	.pad_ib = amdgpu_ring_generic_pad_ib,
6187 	.emit_rreg = gfx_v11_0_ring_emit_rreg,
6188 	.emit_wreg = gfx_v11_0_ring_emit_wreg,
6189 	.emit_reg_wait = gfx_v11_0_ring_emit_reg_wait,
6190 	.emit_reg_write_reg_wait = gfx_v11_0_ring_emit_reg_write_reg_wait,
6191 };
6192 
6193 static void gfx_v11_0_set_ring_funcs(struct amdgpu_device *adev)
6194 {
6195 	int i;
6196 
6197 	adev->gfx.kiq[0].ring.funcs = &gfx_v11_0_ring_funcs_kiq;
6198 
6199 	for (i = 0; i < adev->gfx.num_gfx_rings; i++)
6200 		adev->gfx.gfx_ring[i].funcs = &gfx_v11_0_ring_funcs_gfx;
6201 
6202 	for (i = 0; i < adev->gfx.num_compute_rings; i++)
6203 		adev->gfx.compute_ring[i].funcs = &gfx_v11_0_ring_funcs_compute;
6204 }
6205 
6206 static const struct amdgpu_irq_src_funcs gfx_v11_0_eop_irq_funcs = {
6207 	.set = gfx_v11_0_set_eop_interrupt_state,
6208 	.process = gfx_v11_0_eop_irq,
6209 };
6210 
6211 static const struct amdgpu_irq_src_funcs gfx_v11_0_priv_reg_irq_funcs = {
6212 	.set = gfx_v11_0_set_priv_reg_fault_state,
6213 	.process = gfx_v11_0_priv_reg_irq,
6214 };
6215 
6216 static const struct amdgpu_irq_src_funcs gfx_v11_0_priv_inst_irq_funcs = {
6217 	.set = gfx_v11_0_set_priv_inst_fault_state,
6218 	.process = gfx_v11_0_priv_inst_irq,
6219 };
6220 
6221 static const struct amdgpu_irq_src_funcs gfx_v11_0_rlc_gc_fed_irq_funcs = {
6222 	.process = gfx_v11_0_rlc_gc_fed_irq,
6223 };
6224 
6225 static void gfx_v11_0_set_irq_funcs(struct amdgpu_device *adev)
6226 {
6227 	adev->gfx.eop_irq.num_types = AMDGPU_CP_IRQ_LAST;
6228 	adev->gfx.eop_irq.funcs = &gfx_v11_0_eop_irq_funcs;
6229 
6230 	adev->gfx.priv_reg_irq.num_types = 1;
6231 	adev->gfx.priv_reg_irq.funcs = &gfx_v11_0_priv_reg_irq_funcs;
6232 
6233 	adev->gfx.priv_inst_irq.num_types = 1;
6234 	adev->gfx.priv_inst_irq.funcs = &gfx_v11_0_priv_inst_irq_funcs;
6235 
6236 	adev->gfx.rlc_gc_fed_irq.num_types = 1; /* 0x80 FED error */
6237 	adev->gfx.rlc_gc_fed_irq.funcs = &gfx_v11_0_rlc_gc_fed_irq_funcs;
6238 
6239 }
6240 
6241 static void gfx_v11_0_set_imu_funcs(struct amdgpu_device *adev)
6242 {
6243 	if (adev->flags & AMD_IS_APU)
6244 		adev->gfx.imu.mode = MISSION_MODE;
6245 	else
6246 		adev->gfx.imu.mode = DEBUG_MODE;
6247 
6248 	adev->gfx.imu.funcs = &gfx_v11_0_imu_funcs;
6249 }
6250 
6251 static void gfx_v11_0_set_rlc_funcs(struct amdgpu_device *adev)
6252 {
6253 	adev->gfx.rlc.funcs = &gfx_v11_0_rlc_funcs;
6254 }
6255 
6256 static void gfx_v11_0_set_gds_init(struct amdgpu_device *adev)
6257 {
6258 	unsigned total_cu = adev->gfx.config.max_cu_per_sh *
6259 			    adev->gfx.config.max_sh_per_se *
6260 			    adev->gfx.config.max_shader_engines;
6261 
6262 	adev->gds.gds_size = 0x1000;
6263 	adev->gds.gds_compute_max_wave_id = total_cu * 32 - 1;
6264 	adev->gds.gws_size = 64;
6265 	adev->gds.oa_size = 16;
6266 }
6267 
6268 static void gfx_v11_0_set_mqd_funcs(struct amdgpu_device *adev)
6269 {
6270 	/* set gfx eng mqd */
6271 	adev->mqds[AMDGPU_HW_IP_GFX].mqd_size =
6272 		sizeof(struct v11_gfx_mqd);
6273 	adev->mqds[AMDGPU_HW_IP_GFX].init_mqd =
6274 		gfx_v11_0_gfx_mqd_init;
6275 	/* set compute eng mqd */
6276 	adev->mqds[AMDGPU_HW_IP_COMPUTE].mqd_size =
6277 		sizeof(struct v11_compute_mqd);
6278 	adev->mqds[AMDGPU_HW_IP_COMPUTE].init_mqd =
6279 		gfx_v11_0_compute_mqd_init;
6280 }
6281 
6282 static void gfx_v11_0_set_user_wgp_inactive_bitmap_per_sh(struct amdgpu_device *adev,
6283 							  u32 bitmap)
6284 {
6285 	u32 data;
6286 
6287 	if (!bitmap)
6288 		return;
6289 
6290 	data = bitmap << GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_WGPS__SHIFT;
6291 	data &= GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_WGPS_MASK;
6292 
6293 	WREG32_SOC15(GC, 0, regGC_USER_SHADER_ARRAY_CONFIG, data);
6294 }
6295 
6296 static u32 gfx_v11_0_get_wgp_active_bitmap_per_sh(struct amdgpu_device *adev)
6297 {
6298 	u32 data, wgp_bitmask;
6299 	data = RREG32_SOC15(GC, 0, regCC_GC_SHADER_ARRAY_CONFIG);
6300 	data |= RREG32_SOC15(GC, 0, regGC_USER_SHADER_ARRAY_CONFIG);
6301 
6302 	data &= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_WGPS_MASK;
6303 	data >>= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_WGPS__SHIFT;
6304 
6305 	wgp_bitmask =
6306 		amdgpu_gfx_create_bitmask(adev->gfx.config.max_cu_per_sh >> 1);
6307 
6308 	return (~data) & wgp_bitmask;
6309 }
6310 
6311 static u32 gfx_v11_0_get_cu_active_bitmap_per_sh(struct amdgpu_device *adev)
6312 {
6313 	u32 wgp_idx, wgp_active_bitmap;
6314 	u32 cu_bitmap_per_wgp, cu_active_bitmap;
6315 
6316 	wgp_active_bitmap = gfx_v11_0_get_wgp_active_bitmap_per_sh(adev);
6317 	cu_active_bitmap = 0;
6318 
6319 	for (wgp_idx = 0; wgp_idx < 16; wgp_idx++) {
6320 		/* if there is one WGP enabled, it means 2 CUs will be enabled */
6321 		cu_bitmap_per_wgp = 3 << (2 * wgp_idx);
6322 		if (wgp_active_bitmap & (1 << wgp_idx))
6323 			cu_active_bitmap |= cu_bitmap_per_wgp;
6324 	}
6325 
6326 	return cu_active_bitmap;
6327 }
6328 
6329 static int gfx_v11_0_get_cu_info(struct amdgpu_device *adev,
6330 				 struct amdgpu_cu_info *cu_info)
6331 {
6332 	int i, j, k, counter, active_cu_number = 0;
6333 	u32 mask, bitmap;
6334 	unsigned disable_masks[8 * 2];
6335 
6336 	if (!adev || !cu_info)
6337 		return -EINVAL;
6338 
6339 	amdgpu_gfx_parse_disable_cu(disable_masks, 8, 2);
6340 
6341 	mutex_lock(&adev->grbm_idx_mutex);
6342 	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
6343 		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
6344 			mask = 1;
6345 			counter = 0;
6346 			gfx_v11_0_select_se_sh(adev, i, j, 0xffffffff, 0);
6347 			if (i < 8 && j < 2)
6348 				gfx_v11_0_set_user_wgp_inactive_bitmap_per_sh(
6349 					adev, disable_masks[i * 2 + j]);
6350 			bitmap = gfx_v11_0_get_cu_active_bitmap_per_sh(adev);
6351 
6352 			/**
6353 			 * GFX11 could support more than 4 SEs, while the bitmap
6354 			 * in cu_info struct is 4x4 and ioctl interface struct
6355 			 * drm_amdgpu_info_device should keep stable.
6356 			 * So we use last two columns of bitmap to store cu mask for
6357 			 * SEs 4 to 7, the layout of the bitmap is as below:
6358 			 *    SE0: {SH0,SH1} --> {bitmap[0][0], bitmap[0][1]}
6359 			 *    SE1: {SH0,SH1} --> {bitmap[1][0], bitmap[1][1]}
6360 			 *    SE2: {SH0,SH1} --> {bitmap[2][0], bitmap[2][1]}
6361 			 *    SE3: {SH0,SH1} --> {bitmap[3][0], bitmap[3][1]}
6362 			 *    SE4: {SH0,SH1} --> {bitmap[0][2], bitmap[0][3]}
6363 			 *    SE5: {SH0,SH1} --> {bitmap[1][2], bitmap[1][3]}
6364 			 *    SE6: {SH0,SH1} --> {bitmap[2][2], bitmap[2][3]}
6365 			 *    SE7: {SH0,SH1} --> {bitmap[3][2], bitmap[3][3]}
6366 			 */
6367 			cu_info->bitmap[i % 4][j + (i / 4) * 2] = bitmap;
6368 
6369 			for (k = 0; k < adev->gfx.config.max_cu_per_sh; k++) {
6370 				if (bitmap & mask)
6371 					counter++;
6372 
6373 				mask <<= 1;
6374 			}
6375 			active_cu_number += counter;
6376 		}
6377 	}
6378 	gfx_v11_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 0);
6379 	mutex_unlock(&adev->grbm_idx_mutex);
6380 
6381 	cu_info->number = active_cu_number;
6382 	cu_info->simd_per_cu = NUM_SIMD_PER_CU;
6383 
6384 	return 0;
6385 }
6386 
6387 const struct amdgpu_ip_block_version gfx_v11_0_ip_block =
6388 {
6389 	.type = AMD_IP_BLOCK_TYPE_GFX,
6390 	.major = 11,
6391 	.minor = 0,
6392 	.rev = 0,
6393 	.funcs = &gfx_v11_0_ip_funcs,
6394 };
6395