1 /*
2  * Copyright 2016 Advanced Micro Devices, Inc.
3  *
4  * Permission is hereby granted, free of charge, to any person obtaining a
5  * copy of this software and associated documentation files (the "Software"),
6  * to deal in the Software without restriction, including without limitation
7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8  * and/or sell copies of the Software, and to permit persons to whom the
9  * Software is furnished to do so, subject to the following conditions:
10  *
11  * The above copyright notice and this permission notice shall be included in
12  * all copies or substantial portions of the Software.
13  *
14  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
17  * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20  * OTHER DEALINGS IN THE SOFTWARE.
21  *
22  */
23 
24 #include <linux/firmware.h>
25 #include <drm/drmP.h>
26 #include "amdgpu.h"
27 #include "amdgpu_ucode.h"
28 #include "amdgpu_trace.h"
29 
30 #include "sdma0/sdma0_4_2_offset.h"
31 #include "sdma0/sdma0_4_2_sh_mask.h"
32 #include "sdma1/sdma1_4_2_offset.h"
33 #include "sdma1/sdma1_4_2_sh_mask.h"
34 #include "hdp/hdp_4_0_offset.h"
35 #include "sdma0/sdma0_4_1_default.h"
36 
37 #include "soc15_common.h"
38 #include "soc15.h"
39 #include "vega10_sdma_pkt_open.h"
40 
41 #include "ivsrcid/sdma0/irqsrcs_sdma0_4_0.h"
42 #include "ivsrcid/sdma1/irqsrcs_sdma1_4_0.h"
43 
44 MODULE_FIRMWARE("amdgpu/vega10_sdma.bin");
45 MODULE_FIRMWARE("amdgpu/vega10_sdma1.bin");
46 MODULE_FIRMWARE("amdgpu/vega12_sdma.bin");
47 MODULE_FIRMWARE("amdgpu/vega12_sdma1.bin");
48 MODULE_FIRMWARE("amdgpu/vega20_sdma.bin");
49 MODULE_FIRMWARE("amdgpu/vega20_sdma1.bin");
50 MODULE_FIRMWARE("amdgpu/raven_sdma.bin");
51 MODULE_FIRMWARE("amdgpu/picasso_sdma.bin");
52 MODULE_FIRMWARE("amdgpu/raven2_sdma.bin");
53 
54 #define SDMA0_POWER_CNTL__ON_OFF_CONDITION_HOLD_TIME_MASK  0x000000F8L
55 #define SDMA0_POWER_CNTL__ON_OFF_STATUS_DURATION_TIME_MASK 0xFC000000L
56 
57 static void sdma_v4_0_set_ring_funcs(struct amdgpu_device *adev);
58 static void sdma_v4_0_set_buffer_funcs(struct amdgpu_device *adev);
59 static void sdma_v4_0_set_vm_pte_funcs(struct amdgpu_device *adev);
60 static void sdma_v4_0_set_irq_funcs(struct amdgpu_device *adev);
61 
62 static const struct soc15_reg_golden golden_settings_sdma_4[] = {
63 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_CHICKEN_BITS, 0xfe931f07, 0x02831d07),
64 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_CLK_CTRL, 0xff000ff0, 0x3f000100),
65 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GFX_IB_CNTL, 0x800f0100, 0x00000100),
66 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GFX_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
67 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_PAGE_IB_CNTL, 0x800f0100, 0x00000100),
68 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_PAGE_RB_WPTR_POLL_CNTL, 0x0000fff0, 0x00403000),
69 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_POWER_CNTL, 0x003ff006, 0x0003c000),
70 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC0_IB_CNTL, 0x800f0100, 0x00000100),
71 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC0_RB_WPTR_POLL_CNTL, 0x0000fff0, 0x00403000),
72 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC1_IB_CNTL, 0x800f0100, 0x00000100),
73 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC1_RB_WPTR_POLL_CNTL, 0x0000fff0, 0x00403000),
74 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_UTCL1_PAGE, 0x000003ff, 0x000003c0),
75 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_UTCL1_WATERMK, 0xfc000000, 0x00000000),
76 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_CHICKEN_BITS, 0xfe931f07, 0x02831f07),
77 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_CLK_CTRL, 0xffffffff, 0x3f000100),
78 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_GFX_IB_CNTL, 0x800f0100, 0x00000100),
79 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_GFX_RB_WPTR_POLL_CNTL, 0x0000fff0, 0x00403000),
80 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_PAGE_IB_CNTL, 0x800f0100, 0x00000100),
81 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_PAGE_RB_WPTR_POLL_CNTL, 0x0000fff0, 0x00403000),
82 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_POWER_CNTL, 0x003ff000, 0x0003c000),
83 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_RLC0_IB_CNTL, 0x800f0100, 0x00000100),
84 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_RLC0_RB_WPTR_POLL_CNTL, 0x0000fff0, 0x00403000),
85 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_RLC1_IB_CNTL, 0x800f0100, 0x00000100),
86 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_RLC1_RB_WPTR_POLL_CNTL, 0x0000fff0, 0x00403000),
87 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_UTCL1_PAGE, 0x000003ff, 0x000003c0),
88 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_UTCL1_WATERMK, 0xfc000000, 0x00000000)
89 };
90 
91 static const struct soc15_reg_golden golden_settings_sdma_vg10[] = {
92 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GB_ADDR_CONFIG, 0x0018773f, 0x00104002),
93 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GB_ADDR_CONFIG_READ, 0x0018773f, 0x00104002),
94 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_GB_ADDR_CONFIG, 0x0018773f, 0x00104002),
95 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_GB_ADDR_CONFIG_READ, 0x0018773f, 0x00104002)
96 };
97 
98 static const struct soc15_reg_golden golden_settings_sdma_vg12[] = {
99 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GB_ADDR_CONFIG, 0x0018773f, 0x00104001),
100 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GB_ADDR_CONFIG_READ, 0x0018773f, 0x00104001),
101 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_GB_ADDR_CONFIG, 0x0018773f, 0x00104001),
102 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_GB_ADDR_CONFIG_READ, 0x0018773f, 0x00104001)
103 };
104 
105 static const struct soc15_reg_golden golden_settings_sdma_4_1[] = {
106 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_CHICKEN_BITS, 0xfe931f07, 0x02831d07),
107 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_CLK_CTRL, 0xffffffff, 0x3f000100),
108 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GFX_IB_CNTL, 0x800f0111, 0x00000100),
109 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GFX_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
110 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_POWER_CNTL, 0xfc3fffff, 0x40000051),
111 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC0_IB_CNTL, 0x800f0111, 0x00000100),
112 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC0_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
113 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC1_IB_CNTL, 0x800f0111, 0x00000100),
114 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC1_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
115 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_UTCL1_PAGE, 0x000003ff, 0x000003c0),
116 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_UTCL1_WATERMK, 0xfc000000, 0x00000000)
117 };
118 
119 static const struct soc15_reg_golden golden_settings_sdma0_4_2_init[] = {
120 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC0_RB_WPTR_POLL_CNTL, 0xfffffff0, 0x00403000),
121 };
122 
123 static const struct soc15_reg_golden golden_settings_sdma0_4_2[] =
124 {
125 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_CHICKEN_BITS, 0xfe931f07, 0x02831d07),
126 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_CLK_CTRL, 0xffffffff, 0x3f000100),
127 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GB_ADDR_CONFIG, 0x0000773f, 0x00004002),
128 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GB_ADDR_CONFIG_READ, 0x0000773f, 0x00004002),
129 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GFX_RB_RPTR_ADDR_LO, 0xfffffffd, 0x00000001),
130 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GFX_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
131 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_PAGE_RB_RPTR_ADDR_LO, 0xfffffffd, 0x00000001),
132 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_PAGE_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
133 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RD_BURST_CNTL, 0x0000000f, 0x00000003),
134 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC0_RB_RPTR_ADDR_LO, 0xfffffffd, 0x00000001),
135 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC0_RB_WPTR_POLL_CNTL, 0xfffffff0, 0x00403000),
136 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC1_RB_RPTR_ADDR_LO, 0xfffffffd, 0x00000001),
137 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC1_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
138 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC2_RB_RPTR_ADDR_LO, 0xfffffffd, 0x00000001),
139 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC2_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
140 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC3_RB_RPTR_ADDR_LO, 0xfffffffd, 0x00000001),
141 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC3_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
142 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC4_RB_RPTR_ADDR_LO, 0xfffffffd, 0x00000001),
143 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC4_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
144 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC5_RB_RPTR_ADDR_LO, 0xfffffffd, 0x00000001),
145 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC5_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
146 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC6_RB_RPTR_ADDR_LO, 0xfffffffd, 0x00000001),
147 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC6_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
148 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC7_RB_RPTR_ADDR_LO, 0xfffffffd, 0x00000001),
149 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC7_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
150 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_UTCL1_PAGE, 0x000003ff, 0x000003c0),
151 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_UTCL1_WATERMK, 0xFE000000, 0x00000000),
152 };
153 
154 static const struct soc15_reg_golden golden_settings_sdma1_4_2[] = {
155 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_CHICKEN_BITS, 0xfe931f07, 0x02831d07),
156 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_CLK_CTRL, 0xffffffff, 0x3f000100),
157 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_GB_ADDR_CONFIG, 0x0000773f, 0x00004002),
158 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_GB_ADDR_CONFIG_READ, 0x0000773f, 0x00004002),
159 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_GFX_RB_RPTR_ADDR_LO, 0xfffffffd, 0x00000001),
160 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_GFX_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
161 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_PAGE_RB_RPTR_ADDR_LO, 0xfffffffd, 0x00000001),
162 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_PAGE_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
163 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_RD_BURST_CNTL, 0x0000000f, 0x00000003),
164 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_RLC0_RB_RPTR_ADDR_LO, 0xfffffffd, 0x00000001),
165 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_RLC0_RB_WPTR_POLL_CNTL, 0xfffffff0, 0x00403000),
166 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_RLC1_RB_RPTR_ADDR_LO, 0xfffffffd, 0x00000001),
167 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_RLC1_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
168 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_RLC2_RB_RPTR_ADDR_LO, 0xfffffffd, 0x00000001),
169 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_RLC2_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
170 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_RLC3_RB_RPTR_ADDR_LO, 0xfffffffd, 0x00000001),
171 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_RLC3_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
172 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_RLC4_RB_RPTR_ADDR_LO, 0xfffffffd, 0x00000001),
173 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_RLC4_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
174 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_RLC5_RB_RPTR_ADDR_LO, 0xfffffffd, 0x00000001),
175 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_RLC5_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
176 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_RLC6_RB_RPTR_ADDR_LO, 0xfffffffd, 0x00000001),
177 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_RLC6_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
178 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_RLC7_RB_RPTR_ADDR_LO, 0xfffffffd, 0x00000001),
179 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_RLC7_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
180 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_UTCL1_PAGE, 0x000003ff, 0x000003c0),
181 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_UTCL1_WATERMK, 0xFE000000, 0x00000000),
182 };
183 
184 static const struct soc15_reg_golden golden_settings_sdma_rv1[] =
185 {
186 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GB_ADDR_CONFIG, 0x0018773f, 0x00000002),
187 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GB_ADDR_CONFIG_READ, 0x0018773f, 0x00000002)
188 };
189 
190 static const struct soc15_reg_golden golden_settings_sdma_rv2[] =
191 {
192 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GB_ADDR_CONFIG, 0x0018773f, 0x00003001),
193 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GB_ADDR_CONFIG_READ, 0x0018773f, 0x00003001)
194 };
195 
196 static u32 sdma_v4_0_get_reg_offset(struct amdgpu_device *adev,
197 		u32 instance, u32 offset)
198 {
199 	return ( 0 == instance ? (adev->reg_offset[SDMA0_HWIP][0][0] + offset) :
200 			(adev->reg_offset[SDMA1_HWIP][0][0] + offset));
201 }
202 
203 static void sdma_v4_0_init_golden_registers(struct amdgpu_device *adev)
204 {
205 	switch (adev->asic_type) {
206 	case CHIP_VEGA10:
207 		soc15_program_register_sequence(adev,
208 						 golden_settings_sdma_4,
209 						 ARRAY_SIZE(golden_settings_sdma_4));
210 		soc15_program_register_sequence(adev,
211 						 golden_settings_sdma_vg10,
212 						 ARRAY_SIZE(golden_settings_sdma_vg10));
213 		break;
214 	case CHIP_VEGA12:
215 		soc15_program_register_sequence(adev,
216 						golden_settings_sdma_4,
217 						ARRAY_SIZE(golden_settings_sdma_4));
218 		soc15_program_register_sequence(adev,
219 						golden_settings_sdma_vg12,
220 						ARRAY_SIZE(golden_settings_sdma_vg12));
221 		break;
222 	case CHIP_VEGA20:
223 		soc15_program_register_sequence(adev,
224 						golden_settings_sdma0_4_2_init,
225 						ARRAY_SIZE(golden_settings_sdma0_4_2_init));
226 		soc15_program_register_sequence(adev,
227 						golden_settings_sdma0_4_2,
228 						ARRAY_SIZE(golden_settings_sdma0_4_2));
229 		soc15_program_register_sequence(adev,
230 						golden_settings_sdma1_4_2,
231 						ARRAY_SIZE(golden_settings_sdma1_4_2));
232 		break;
233 	case CHIP_RAVEN:
234 		soc15_program_register_sequence(adev,
235 						golden_settings_sdma_4_1,
236 						ARRAY_SIZE(golden_settings_sdma_4_1));
237 		if (adev->rev_id >= 8)
238 			soc15_program_register_sequence(adev,
239 							golden_settings_sdma_rv2,
240 							ARRAY_SIZE(golden_settings_sdma_rv2));
241 		else
242 			soc15_program_register_sequence(adev,
243 							golden_settings_sdma_rv1,
244 							ARRAY_SIZE(golden_settings_sdma_rv1));
245 		break;
246 	default:
247 		break;
248 	}
249 }
250 
251 /**
252  * sdma_v4_0_init_microcode - load ucode images from disk
253  *
254  * @adev: amdgpu_device pointer
255  *
256  * Use the firmware interface to load the ucode images into
257  * the driver (not loaded into hw).
258  * Returns 0 on success, error on failure.
259  */
260 
261 // emulation only, won't work on real chip
262 // vega10 real chip need to use PSP to load firmware
263 static int sdma_v4_0_init_microcode(struct amdgpu_device *adev)
264 {
265 	const char *chip_name;
266 	char fw_name[30];
267 	int err = 0, i;
268 	struct amdgpu_firmware_info *info = NULL;
269 	const struct common_firmware_header *header = NULL;
270 	const struct sdma_firmware_header_v1_0 *hdr;
271 
272 	DRM_DEBUG("\n");
273 
274 	switch (adev->asic_type) {
275 	case CHIP_VEGA10:
276 		chip_name = "vega10";
277 		break;
278 	case CHIP_VEGA12:
279 		chip_name = "vega12";
280 		break;
281 	case CHIP_VEGA20:
282 		chip_name = "vega20";
283 		break;
284 	case CHIP_RAVEN:
285 		if (adev->rev_id >= 8)
286 			chip_name = "raven2";
287 		else if (adev->pdev->device == 0x15d8)
288 			chip_name = "picasso";
289 		else
290 			chip_name = "raven";
291 		break;
292 	default:
293 		BUG();
294 	}
295 
296 	for (i = 0; i < adev->sdma.num_instances; i++) {
297 		if (i == 0)
298 			snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_sdma.bin", chip_name);
299 		else
300 			snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_sdma1.bin", chip_name);
301 		err = request_firmware(&adev->sdma.instance[i].fw, fw_name, adev->dev);
302 		if (err)
303 			goto out;
304 		err = amdgpu_ucode_validate(adev->sdma.instance[i].fw);
305 		if (err)
306 			goto out;
307 		hdr = (const struct sdma_firmware_header_v1_0 *)adev->sdma.instance[i].fw->data;
308 		adev->sdma.instance[i].fw_version = le32_to_cpu(hdr->header.ucode_version);
309 		adev->sdma.instance[i].feature_version = le32_to_cpu(hdr->ucode_feature_version);
310 		if (adev->sdma.instance[i].feature_version >= 20)
311 			adev->sdma.instance[i].burst_nop = true;
312 		DRM_DEBUG("psp_load == '%s'\n",
313 				adev->firmware.load_type == AMDGPU_FW_LOAD_PSP ? "true" : "false");
314 
315 		if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
316 			info = &adev->firmware.ucode[AMDGPU_UCODE_ID_SDMA0 + i];
317 			info->ucode_id = AMDGPU_UCODE_ID_SDMA0 + i;
318 			info->fw = adev->sdma.instance[i].fw;
319 			header = (const struct common_firmware_header *)info->fw->data;
320 			adev->firmware.fw_size +=
321 				ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
322 		}
323 	}
324 out:
325 	if (err) {
326 		DRM_ERROR("sdma_v4_0: Failed to load firmware \"%s\"\n", fw_name);
327 		for (i = 0; i < adev->sdma.num_instances; i++) {
328 			release_firmware(adev->sdma.instance[i].fw);
329 			adev->sdma.instance[i].fw = NULL;
330 		}
331 	}
332 	return err;
333 }
334 
335 /**
336  * sdma_v4_0_ring_get_rptr - get the current read pointer
337  *
338  * @ring: amdgpu ring pointer
339  *
340  * Get the current rptr from the hardware (VEGA10+).
341  */
342 static uint64_t sdma_v4_0_ring_get_rptr(struct amdgpu_ring *ring)
343 {
344 	u64 *rptr;
345 
346 	/* XXX check if swapping is necessary on BE */
347 	rptr = ((u64 *)&ring->adev->wb.wb[ring->rptr_offs]);
348 
349 	DRM_DEBUG("rptr before shift == 0x%016llx\n", *rptr);
350 	return ((*rptr) >> 2);
351 }
352 
353 /**
354  * sdma_v4_0_ring_get_wptr - get the current write pointer
355  *
356  * @ring: amdgpu ring pointer
357  *
358  * Get the current wptr from the hardware (VEGA10+).
359  */
360 static uint64_t sdma_v4_0_ring_get_wptr(struct amdgpu_ring *ring)
361 {
362 	struct amdgpu_device *adev = ring->adev;
363 	u64 wptr;
364 
365 	if (ring->use_doorbell) {
366 		/* XXX check if swapping is necessary on BE */
367 		wptr = READ_ONCE(*((u64 *)&adev->wb.wb[ring->wptr_offs]));
368 		DRM_DEBUG("wptr/doorbell before shift == 0x%016llx\n", wptr);
369 	} else {
370 		u32 lowbit, highbit;
371 
372 		lowbit = RREG32(sdma_v4_0_get_reg_offset(adev, ring->me, mmSDMA0_GFX_RB_WPTR)) >> 2;
373 		highbit = RREG32(sdma_v4_0_get_reg_offset(adev, ring->me, mmSDMA0_GFX_RB_WPTR_HI)) >> 2;
374 
375 		DRM_DEBUG("wptr [%i]high== 0x%08x low==0x%08x\n",
376 				ring->me, highbit, lowbit);
377 		wptr = highbit;
378 		wptr = wptr << 32;
379 		wptr |= lowbit;
380 	}
381 
382 	return wptr >> 2;
383 }
384 
385 /**
386  * sdma_v4_0_ring_set_wptr - commit the write pointer
387  *
388  * @ring: amdgpu ring pointer
389  *
390  * Write the wptr back to the hardware (VEGA10+).
391  */
392 static void sdma_v4_0_ring_set_wptr(struct amdgpu_ring *ring)
393 {
394 	struct amdgpu_device *adev = ring->adev;
395 
396 	DRM_DEBUG("Setting write pointer\n");
397 	if (ring->use_doorbell) {
398 		u64 *wb = (u64 *)&adev->wb.wb[ring->wptr_offs];
399 
400 		DRM_DEBUG("Using doorbell -- "
401 				"wptr_offs == 0x%08x "
402 				"lower_32_bits(ring->wptr) << 2 == 0x%08x "
403 				"upper_32_bits(ring->wptr) << 2 == 0x%08x\n",
404 				ring->wptr_offs,
405 				lower_32_bits(ring->wptr << 2),
406 				upper_32_bits(ring->wptr << 2));
407 		/* XXX check if swapping is necessary on BE */
408 		WRITE_ONCE(*wb, (ring->wptr << 2));
409 		DRM_DEBUG("calling WDOORBELL64(0x%08x, 0x%016llx)\n",
410 				ring->doorbell_index, ring->wptr << 2);
411 		WDOORBELL64(ring->doorbell_index, ring->wptr << 2);
412 	} else {
413 		DRM_DEBUG("Not using doorbell -- "
414 				"mmSDMA%i_GFX_RB_WPTR == 0x%08x "
415 				"mmSDMA%i_GFX_RB_WPTR_HI == 0x%08x\n",
416 				ring->me,
417 				lower_32_bits(ring->wptr << 2),
418 				ring->me,
419 				upper_32_bits(ring->wptr << 2));
420 		WREG32(sdma_v4_0_get_reg_offset(adev, ring->me, mmSDMA0_GFX_RB_WPTR), lower_32_bits(ring->wptr << 2));
421 		WREG32(sdma_v4_0_get_reg_offset(adev, ring->me, mmSDMA0_GFX_RB_WPTR_HI), upper_32_bits(ring->wptr << 2));
422 	}
423 }
424 
425 static void sdma_v4_0_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count)
426 {
427 	struct amdgpu_sdma_instance *sdma = amdgpu_get_sdma_instance(ring);
428 	int i;
429 
430 	for (i = 0; i < count; i++)
431 		if (sdma && sdma->burst_nop && (i == 0))
432 			amdgpu_ring_write(ring, ring->funcs->nop |
433 				SDMA_PKT_NOP_HEADER_COUNT(count - 1));
434 		else
435 			amdgpu_ring_write(ring, ring->funcs->nop);
436 }
437 
438 /**
439  * sdma_v4_0_ring_emit_ib - Schedule an IB on the DMA engine
440  *
441  * @ring: amdgpu ring pointer
442  * @ib: IB object to schedule
443  *
444  * Schedule an IB in the DMA ring (VEGA10).
445  */
446 static void sdma_v4_0_ring_emit_ib(struct amdgpu_ring *ring,
447 					struct amdgpu_ib *ib,
448 					unsigned vmid, bool ctx_switch)
449 {
450 	/* IB packet must end on a 8 DW boundary */
451 	sdma_v4_0_ring_insert_nop(ring, (10 - (lower_32_bits(ring->wptr) & 7)) % 8);
452 
453 	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_INDIRECT) |
454 			  SDMA_PKT_INDIRECT_HEADER_VMID(vmid & 0xf));
455 	/* base must be 32 byte aligned */
456 	amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr) & 0xffffffe0);
457 	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr));
458 	amdgpu_ring_write(ring, ib->length_dw);
459 	amdgpu_ring_write(ring, 0);
460 	amdgpu_ring_write(ring, 0);
461 
462 }
463 
464 static void sdma_v4_0_wait_reg_mem(struct amdgpu_ring *ring,
465 				   int mem_space, int hdp,
466 				   uint32_t addr0, uint32_t addr1,
467 				   uint32_t ref, uint32_t mask,
468 				   uint32_t inv)
469 {
470 	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_POLL_REGMEM) |
471 			  SDMA_PKT_POLL_REGMEM_HEADER_HDP_FLUSH(hdp) |
472 			  SDMA_PKT_POLL_REGMEM_HEADER_MEM_POLL(mem_space) |
473 			  SDMA_PKT_POLL_REGMEM_HEADER_FUNC(3)); /* == */
474 	if (mem_space) {
475 		/* memory */
476 		amdgpu_ring_write(ring, addr0);
477 		amdgpu_ring_write(ring, addr1);
478 	} else {
479 		/* registers */
480 		amdgpu_ring_write(ring, addr0 << 2);
481 		amdgpu_ring_write(ring, addr1 << 2);
482 	}
483 	amdgpu_ring_write(ring, ref); /* reference */
484 	amdgpu_ring_write(ring, mask); /* mask */
485 	amdgpu_ring_write(ring, SDMA_PKT_POLL_REGMEM_DW5_RETRY_COUNT(0xfff) |
486 			  SDMA_PKT_POLL_REGMEM_DW5_INTERVAL(inv)); /* retry count, poll interval */
487 }
488 
489 /**
490  * sdma_v4_0_ring_emit_hdp_flush - emit an hdp flush on the DMA ring
491  *
492  * @ring: amdgpu ring pointer
493  *
494  * Emit an hdp flush packet on the requested DMA ring.
495  */
496 static void sdma_v4_0_ring_emit_hdp_flush(struct amdgpu_ring *ring)
497 {
498 	struct amdgpu_device *adev = ring->adev;
499 	u32 ref_and_mask = 0;
500 	const struct nbio_hdp_flush_reg *nbio_hf_reg = adev->nbio_funcs->hdp_flush_reg;
501 
502 	if (ring->me == 0)
503 		ref_and_mask = nbio_hf_reg->ref_and_mask_sdma0;
504 	else
505 		ref_and_mask = nbio_hf_reg->ref_and_mask_sdma1;
506 
507 	sdma_v4_0_wait_reg_mem(ring, 0, 1,
508 			       adev->nbio_funcs->get_hdp_flush_done_offset(adev),
509 			       adev->nbio_funcs->get_hdp_flush_req_offset(adev),
510 			       ref_and_mask, ref_and_mask, 10);
511 }
512 
513 /**
514  * sdma_v4_0_ring_emit_fence - emit a fence on the DMA ring
515  *
516  * @ring: amdgpu ring pointer
517  * @fence: amdgpu fence object
518  *
519  * Add a DMA fence packet to the ring to write
520  * the fence seq number and DMA trap packet to generate
521  * an interrupt if needed (VEGA10).
522  */
523 static void sdma_v4_0_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, u64 seq,
524 				      unsigned flags)
525 {
526 	bool write64bit = flags & AMDGPU_FENCE_FLAG_64BIT;
527 	/* write the fence */
528 	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_FENCE));
529 	/* zero in first two bits */
530 	BUG_ON(addr & 0x3);
531 	amdgpu_ring_write(ring, lower_32_bits(addr));
532 	amdgpu_ring_write(ring, upper_32_bits(addr));
533 	amdgpu_ring_write(ring, lower_32_bits(seq));
534 
535 	/* optionally write high bits as well */
536 	if (write64bit) {
537 		addr += 4;
538 		amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_FENCE));
539 		/* zero in first two bits */
540 		BUG_ON(addr & 0x3);
541 		amdgpu_ring_write(ring, lower_32_bits(addr));
542 		amdgpu_ring_write(ring, upper_32_bits(addr));
543 		amdgpu_ring_write(ring, upper_32_bits(seq));
544 	}
545 
546 	/* generate an interrupt */
547 	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_TRAP));
548 	amdgpu_ring_write(ring, SDMA_PKT_TRAP_INT_CONTEXT_INT_CONTEXT(0));
549 }
550 
551 
552 /**
553  * sdma_v4_0_gfx_stop - stop the gfx async dma engines
554  *
555  * @adev: amdgpu_device pointer
556  *
557  * Stop the gfx async dma ring buffers (VEGA10).
558  */
559 static void sdma_v4_0_gfx_stop(struct amdgpu_device *adev)
560 {
561 	struct amdgpu_ring *sdma0 = &adev->sdma.instance[0].ring;
562 	struct amdgpu_ring *sdma1 = &adev->sdma.instance[1].ring;
563 	u32 rb_cntl, ib_cntl;
564 	int i;
565 
566 	if ((adev->mman.buffer_funcs_ring == sdma0) ||
567 	    (adev->mman.buffer_funcs_ring == sdma1))
568 			amdgpu_ttm_set_buffer_funcs_status(adev, false);
569 
570 	for (i = 0; i < adev->sdma.num_instances; i++) {
571 		rb_cntl = RREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_GFX_RB_CNTL));
572 		rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_GFX_RB_CNTL, RB_ENABLE, 0);
573 		WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_GFX_RB_CNTL), rb_cntl);
574 		ib_cntl = RREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_GFX_IB_CNTL));
575 		ib_cntl = REG_SET_FIELD(ib_cntl, SDMA0_GFX_IB_CNTL, IB_ENABLE, 0);
576 		WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_GFX_IB_CNTL), ib_cntl);
577 	}
578 
579 	sdma0->ready = false;
580 	sdma1->ready = false;
581 }
582 
583 /**
584  * sdma_v4_0_rlc_stop - stop the compute async dma engines
585  *
586  * @adev: amdgpu_device pointer
587  *
588  * Stop the compute async dma queues (VEGA10).
589  */
590 static void sdma_v4_0_rlc_stop(struct amdgpu_device *adev)
591 {
592 	/* XXX todo */
593 }
594 
595 /**
596  * sdma_v_0_ctx_switch_enable - stop the async dma engines context switch
597  *
598  * @adev: amdgpu_device pointer
599  * @enable: enable/disable the DMA MEs context switch.
600  *
601  * Halt or unhalt the async dma engines context switch (VEGA10).
602  */
603 static void sdma_v4_0_ctx_switch_enable(struct amdgpu_device *adev, bool enable)
604 {
605 	u32 f32_cntl, phase_quantum = 0;
606 	int i;
607 
608 	if (amdgpu_sdma_phase_quantum) {
609 		unsigned value = amdgpu_sdma_phase_quantum;
610 		unsigned unit = 0;
611 
612 		while (value > (SDMA0_PHASE0_QUANTUM__VALUE_MASK >>
613 				SDMA0_PHASE0_QUANTUM__VALUE__SHIFT)) {
614 			value = (value + 1) >> 1;
615 			unit++;
616 		}
617 		if (unit > (SDMA0_PHASE0_QUANTUM__UNIT_MASK >>
618 			    SDMA0_PHASE0_QUANTUM__UNIT__SHIFT)) {
619 			value = (SDMA0_PHASE0_QUANTUM__VALUE_MASK >>
620 				 SDMA0_PHASE0_QUANTUM__VALUE__SHIFT);
621 			unit = (SDMA0_PHASE0_QUANTUM__UNIT_MASK >>
622 				SDMA0_PHASE0_QUANTUM__UNIT__SHIFT);
623 			WARN_ONCE(1,
624 			"clamping sdma_phase_quantum to %uK clock cycles\n",
625 				  value << unit);
626 		}
627 		phase_quantum =
628 			value << SDMA0_PHASE0_QUANTUM__VALUE__SHIFT |
629 			unit  << SDMA0_PHASE0_QUANTUM__UNIT__SHIFT;
630 	}
631 
632 	for (i = 0; i < adev->sdma.num_instances; i++) {
633 		f32_cntl = RREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_CNTL));
634 		f32_cntl = REG_SET_FIELD(f32_cntl, SDMA0_CNTL,
635 				AUTO_CTXSW_ENABLE, enable ? 1 : 0);
636 		if (enable && amdgpu_sdma_phase_quantum) {
637 			WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_PHASE0_QUANTUM),
638 			       phase_quantum);
639 			WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_PHASE1_QUANTUM),
640 			       phase_quantum);
641 			WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_PHASE2_QUANTUM),
642 			       phase_quantum);
643 		}
644 		WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_CNTL), f32_cntl);
645 	}
646 
647 }
648 
649 /**
650  * sdma_v4_0_enable - stop the async dma engines
651  *
652  * @adev: amdgpu_device pointer
653  * @enable: enable/disable the DMA MEs.
654  *
655  * Halt or unhalt the async dma engines (VEGA10).
656  */
657 static void sdma_v4_0_enable(struct amdgpu_device *adev, bool enable)
658 {
659 	u32 f32_cntl;
660 	int i;
661 
662 	if (enable == false) {
663 		sdma_v4_0_gfx_stop(adev);
664 		sdma_v4_0_rlc_stop(adev);
665 	}
666 
667 	for (i = 0; i < adev->sdma.num_instances; i++) {
668 		f32_cntl = RREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_F32_CNTL));
669 		f32_cntl = REG_SET_FIELD(f32_cntl, SDMA0_F32_CNTL, HALT, enable ? 0 : 1);
670 		WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_F32_CNTL), f32_cntl);
671 	}
672 }
673 
674 /**
675  * sdma_v4_0_gfx_resume - setup and start the async dma engines
676  *
677  * @adev: amdgpu_device pointer
678  *
679  * Set up the gfx DMA ring buffers and enable them (VEGA10).
680  * Returns 0 for success, error for failure.
681  */
682 static int sdma_v4_0_gfx_resume(struct amdgpu_device *adev)
683 {
684 	struct amdgpu_ring *ring;
685 	u32 rb_cntl, ib_cntl, wptr_poll_cntl;
686 	u32 rb_bufsz;
687 	u32 wb_offset;
688 	u32 doorbell;
689 	u32 doorbell_offset;
690 	u32 temp;
691 	u64 wptr_gpu_addr;
692 	int i, r;
693 
694 	for (i = 0; i < adev->sdma.num_instances; i++) {
695 		ring = &adev->sdma.instance[i].ring;
696 		wb_offset = (ring->rptr_offs * 4);
697 
698 		WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_SEM_WAIT_FAIL_TIMER_CNTL), 0);
699 
700 		/* Set ring buffer size in dwords */
701 		rb_bufsz = order_base_2(ring->ring_size / 4);
702 		rb_cntl = RREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_GFX_RB_CNTL));
703 		rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_GFX_RB_CNTL, RB_SIZE, rb_bufsz);
704 #ifdef __BIG_ENDIAN
705 		rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_GFX_RB_CNTL, RB_SWAP_ENABLE, 1);
706 		rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_GFX_RB_CNTL,
707 					RPTR_WRITEBACK_SWAP_ENABLE, 1);
708 #endif
709 		WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_GFX_RB_CNTL), rb_cntl);
710 
711 		/* Initialize the ring buffer's read and write pointers */
712 		WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_GFX_RB_RPTR), 0);
713 		WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_GFX_RB_RPTR_HI), 0);
714 		WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_GFX_RB_WPTR), 0);
715 		WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_GFX_RB_WPTR_HI), 0);
716 
717 		/* set the wb address whether it's enabled or not */
718 		WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_GFX_RB_RPTR_ADDR_HI),
719 		       upper_32_bits(adev->wb.gpu_addr + wb_offset) & 0xFFFFFFFF);
720 		WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_GFX_RB_RPTR_ADDR_LO),
721 		       lower_32_bits(adev->wb.gpu_addr + wb_offset) & 0xFFFFFFFC);
722 
723 		rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_GFX_RB_CNTL, RPTR_WRITEBACK_ENABLE, 1);
724 
725 		WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_GFX_RB_BASE), ring->gpu_addr >> 8);
726 		WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_GFX_RB_BASE_HI), ring->gpu_addr >> 40);
727 
728 		ring->wptr = 0;
729 
730 		/* before programing wptr to a less value, need set minor_ptr_update first */
731 		WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_GFX_MINOR_PTR_UPDATE), 1);
732 
733 		if (!amdgpu_sriov_vf(adev)) { /* only bare-metal use register write for wptr */
734 			WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_GFX_RB_WPTR), lower_32_bits(ring->wptr) << 2);
735 			WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_GFX_RB_WPTR_HI), upper_32_bits(ring->wptr) << 2);
736 		}
737 
738 		doorbell = RREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_GFX_DOORBELL));
739 		doorbell_offset = RREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_GFX_DOORBELL_OFFSET));
740 
741 		if (ring->use_doorbell) {
742 			doorbell = REG_SET_FIELD(doorbell, SDMA0_GFX_DOORBELL, ENABLE, 1);
743 			doorbell_offset = REG_SET_FIELD(doorbell_offset, SDMA0_GFX_DOORBELL_OFFSET,
744 					OFFSET, ring->doorbell_index);
745 		} else {
746 			doorbell = REG_SET_FIELD(doorbell, SDMA0_GFX_DOORBELL, ENABLE, 0);
747 		}
748 		WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_GFX_DOORBELL), doorbell);
749 		WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_GFX_DOORBELL_OFFSET), doorbell_offset);
750 		adev->nbio_funcs->sdma_doorbell_range(adev, i, ring->use_doorbell,
751 						      ring->doorbell_index);
752 
753 		if (amdgpu_sriov_vf(adev))
754 			sdma_v4_0_ring_set_wptr(ring);
755 
756 		/* set minor_ptr_update to 0 after wptr programed */
757 		WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_GFX_MINOR_PTR_UPDATE), 0);
758 
759 		/* set utc l1 enable flag always to 1 */
760 		temp = RREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_CNTL));
761 		temp = REG_SET_FIELD(temp, SDMA0_CNTL, UTC_L1_ENABLE, 1);
762 		WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_CNTL), temp);
763 
764 		if (!amdgpu_sriov_vf(adev)) {
765 			/* unhalt engine */
766 			temp = RREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_F32_CNTL));
767 			temp = REG_SET_FIELD(temp, SDMA0_F32_CNTL, HALT, 0);
768 			WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_F32_CNTL), temp);
769 		}
770 
771 		/* setup the wptr shadow polling */
772 		wptr_gpu_addr = adev->wb.gpu_addr + (ring->wptr_offs * 4);
773 		WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_GFX_RB_WPTR_POLL_ADDR_LO),
774 		       lower_32_bits(wptr_gpu_addr));
775 		WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_GFX_RB_WPTR_POLL_ADDR_HI),
776 		       upper_32_bits(wptr_gpu_addr));
777 		wptr_poll_cntl = RREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_GFX_RB_WPTR_POLL_CNTL));
778 		if (amdgpu_sriov_vf(adev))
779 			wptr_poll_cntl = REG_SET_FIELD(wptr_poll_cntl, SDMA0_GFX_RB_WPTR_POLL_CNTL, F32_POLL_ENABLE, 1);
780 		else
781 			wptr_poll_cntl = REG_SET_FIELD(wptr_poll_cntl, SDMA0_GFX_RB_WPTR_POLL_CNTL, F32_POLL_ENABLE, 0);
782 		WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_GFX_RB_WPTR_POLL_CNTL), wptr_poll_cntl);
783 
784 		/* enable DMA RB */
785 		rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_GFX_RB_CNTL, RB_ENABLE, 1);
786 		WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_GFX_RB_CNTL), rb_cntl);
787 
788 		ib_cntl = RREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_GFX_IB_CNTL));
789 		ib_cntl = REG_SET_FIELD(ib_cntl, SDMA0_GFX_IB_CNTL, IB_ENABLE, 1);
790 #ifdef __BIG_ENDIAN
791 		ib_cntl = REG_SET_FIELD(ib_cntl, SDMA0_GFX_IB_CNTL, IB_SWAP_ENABLE, 1);
792 #endif
793 		/* enable DMA IBs */
794 		WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_GFX_IB_CNTL), ib_cntl);
795 
796 		ring->ready = true;
797 
798 		if (amdgpu_sriov_vf(adev)) { /* bare-metal sequence doesn't need below to lines */
799 			sdma_v4_0_ctx_switch_enable(adev, true);
800 			sdma_v4_0_enable(adev, true);
801 		}
802 
803 		r = amdgpu_ring_test_ring(ring);
804 		if (r) {
805 			ring->ready = false;
806 			return r;
807 		}
808 
809 		if (adev->mman.buffer_funcs_ring == ring)
810 			amdgpu_ttm_set_buffer_funcs_status(adev, true);
811 
812 	}
813 
814 	return 0;
815 }
816 
817 static void
818 sdma_v4_1_update_power_gating(struct amdgpu_device *adev, bool enable)
819 {
820 	uint32_t def, data;
821 
822 	if (enable && (adev->pg_flags & AMD_PG_SUPPORT_SDMA)) {
823 		/* enable idle interrupt */
824 		def = data = RREG32(SOC15_REG_OFFSET(SDMA0, 0, mmSDMA0_CNTL));
825 		data |= SDMA0_CNTL__CTXEMPTY_INT_ENABLE_MASK;
826 
827 		if (data != def)
828 			WREG32(SOC15_REG_OFFSET(SDMA0, 0, mmSDMA0_CNTL), data);
829 	} else {
830 		/* disable idle interrupt */
831 		def = data = RREG32(SOC15_REG_OFFSET(SDMA0, 0, mmSDMA0_CNTL));
832 		data &= ~SDMA0_CNTL__CTXEMPTY_INT_ENABLE_MASK;
833 		if (data != def)
834 			WREG32(SOC15_REG_OFFSET(SDMA0, 0, mmSDMA0_CNTL), data);
835 	}
836 }
837 
838 static void sdma_v4_1_init_power_gating(struct amdgpu_device *adev)
839 {
840 	uint32_t def, data;
841 
842 	/* Enable HW based PG. */
843 	def = data = RREG32(SOC15_REG_OFFSET(SDMA0, 0, mmSDMA0_POWER_CNTL));
844 	data |= SDMA0_POWER_CNTL__PG_CNTL_ENABLE_MASK;
845 	if (data != def)
846 		WREG32(SOC15_REG_OFFSET(SDMA0, 0, mmSDMA0_POWER_CNTL), data);
847 
848 	/* enable interrupt */
849 	def = data = RREG32(SOC15_REG_OFFSET(SDMA0, 0, mmSDMA0_CNTL));
850 	data |= SDMA0_CNTL__CTXEMPTY_INT_ENABLE_MASK;
851 	if (data != def)
852 		WREG32(SOC15_REG_OFFSET(SDMA0, 0, mmSDMA0_CNTL), data);
853 
854 	/* Configure hold time to filter in-valid power on/off request. Use default right now */
855 	def = data = RREG32(SOC15_REG_OFFSET(SDMA0, 0, mmSDMA0_POWER_CNTL));
856 	data &= ~SDMA0_POWER_CNTL__ON_OFF_CONDITION_HOLD_TIME_MASK;
857 	data |= (mmSDMA0_POWER_CNTL_DEFAULT & SDMA0_POWER_CNTL__ON_OFF_CONDITION_HOLD_TIME_MASK);
858 	/* Configure switch time for hysteresis purpose. Use default right now */
859 	data &= ~SDMA0_POWER_CNTL__ON_OFF_STATUS_DURATION_TIME_MASK;
860 	data |= (mmSDMA0_POWER_CNTL_DEFAULT & SDMA0_POWER_CNTL__ON_OFF_STATUS_DURATION_TIME_MASK);
861 	if(data != def)
862 		WREG32(SOC15_REG_OFFSET(SDMA0, 0, mmSDMA0_POWER_CNTL), data);
863 }
864 
865 static void sdma_v4_0_init_pg(struct amdgpu_device *adev)
866 {
867 	if (!(adev->pg_flags & AMD_PG_SUPPORT_SDMA))
868 		return;
869 
870 	switch (adev->asic_type) {
871 	case CHIP_RAVEN:
872 		sdma_v4_1_init_power_gating(adev);
873 		sdma_v4_1_update_power_gating(adev, true);
874 		break;
875 	default:
876 		break;
877 	}
878 }
879 
880 /**
881  * sdma_v4_0_rlc_resume - setup and start the async dma engines
882  *
883  * @adev: amdgpu_device pointer
884  *
885  * Set up the compute DMA queues and enable them (VEGA10).
886  * Returns 0 for success, error for failure.
887  */
888 static int sdma_v4_0_rlc_resume(struct amdgpu_device *adev)
889 {
890 	sdma_v4_0_init_pg(adev);
891 
892 	return 0;
893 }
894 
895 /**
896  * sdma_v4_0_load_microcode - load the sDMA ME ucode
897  *
898  * @adev: amdgpu_device pointer
899  *
900  * Loads the sDMA0/1 ucode.
901  * Returns 0 for success, -EINVAL if the ucode is not available.
902  */
903 static int sdma_v4_0_load_microcode(struct amdgpu_device *adev)
904 {
905 	const struct sdma_firmware_header_v1_0 *hdr;
906 	const __le32 *fw_data;
907 	u32 fw_size;
908 	int i, j;
909 
910 	/* halt the MEs */
911 	sdma_v4_0_enable(adev, false);
912 
913 	for (i = 0; i < adev->sdma.num_instances; i++) {
914 		if (!adev->sdma.instance[i].fw)
915 			return -EINVAL;
916 
917 		hdr = (const struct sdma_firmware_header_v1_0 *)adev->sdma.instance[i].fw->data;
918 		amdgpu_ucode_print_sdma_hdr(&hdr->header);
919 		fw_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4;
920 
921 		fw_data = (const __le32 *)
922 			(adev->sdma.instance[i].fw->data +
923 				le32_to_cpu(hdr->header.ucode_array_offset_bytes));
924 
925 		WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_UCODE_ADDR), 0);
926 
927 		for (j = 0; j < fw_size; j++)
928 			WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_UCODE_DATA), le32_to_cpup(fw_data++));
929 
930 		WREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_UCODE_ADDR), adev->sdma.instance[i].fw_version);
931 	}
932 
933 	return 0;
934 }
935 
936 /**
937  * sdma_v4_0_start - setup and start the async dma engines
938  *
939  * @adev: amdgpu_device pointer
940  *
941  * Set up the DMA engines and enable them (VEGA10).
942  * Returns 0 for success, error for failure.
943  */
944 static int sdma_v4_0_start(struct amdgpu_device *adev)
945 {
946 	int r = 0;
947 
948 	if (amdgpu_sriov_vf(adev)) {
949 		sdma_v4_0_ctx_switch_enable(adev, false);
950 		sdma_v4_0_enable(adev, false);
951 
952 		/* set RB registers */
953 		r = sdma_v4_0_gfx_resume(adev);
954 		return r;
955 	}
956 
957 	if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP) {
958 		r = sdma_v4_0_load_microcode(adev);
959 		if (r)
960 			return r;
961 	}
962 
963 	/* unhalt the MEs */
964 	sdma_v4_0_enable(adev, true);
965 	/* enable sdma ring preemption */
966 	sdma_v4_0_ctx_switch_enable(adev, true);
967 
968 	/* start the gfx rings and rlc compute queues */
969 	r = sdma_v4_0_gfx_resume(adev);
970 	if (r)
971 		return r;
972 	r = sdma_v4_0_rlc_resume(adev);
973 
974 	return r;
975 }
976 
977 /**
978  * sdma_v4_0_ring_test_ring - simple async dma engine test
979  *
980  * @ring: amdgpu_ring structure holding ring information
981  *
982  * Test the DMA engine by writing using it to write an
983  * value to memory. (VEGA10).
984  * Returns 0 for success, error for failure.
985  */
986 static int sdma_v4_0_ring_test_ring(struct amdgpu_ring *ring)
987 {
988 	struct amdgpu_device *adev = ring->adev;
989 	unsigned i;
990 	unsigned index;
991 	int r;
992 	u32 tmp;
993 	u64 gpu_addr;
994 
995 	r = amdgpu_device_wb_get(adev, &index);
996 	if (r) {
997 		dev_err(adev->dev, "(%d) failed to allocate wb slot\n", r);
998 		return r;
999 	}
1000 
1001 	gpu_addr = adev->wb.gpu_addr + (index * 4);
1002 	tmp = 0xCAFEDEAD;
1003 	adev->wb.wb[index] = cpu_to_le32(tmp);
1004 
1005 	r = amdgpu_ring_alloc(ring, 5);
1006 	if (r) {
1007 		DRM_ERROR("amdgpu: dma failed to lock ring %d (%d).\n", ring->idx, r);
1008 		amdgpu_device_wb_free(adev, index);
1009 		return r;
1010 	}
1011 
1012 	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_WRITE) |
1013 			  SDMA_PKT_HEADER_SUB_OP(SDMA_SUBOP_WRITE_LINEAR));
1014 	amdgpu_ring_write(ring, lower_32_bits(gpu_addr));
1015 	amdgpu_ring_write(ring, upper_32_bits(gpu_addr));
1016 	amdgpu_ring_write(ring, SDMA_PKT_WRITE_UNTILED_DW_3_COUNT(0));
1017 	amdgpu_ring_write(ring, 0xDEADBEEF);
1018 	amdgpu_ring_commit(ring);
1019 
1020 	for (i = 0; i < adev->usec_timeout; i++) {
1021 		tmp = le32_to_cpu(adev->wb.wb[index]);
1022 		if (tmp == 0xDEADBEEF)
1023 			break;
1024 		DRM_UDELAY(1);
1025 	}
1026 
1027 	if (i < adev->usec_timeout) {
1028 		DRM_DEBUG("ring test on %d succeeded in %d usecs\n", ring->idx, i);
1029 	} else {
1030 		DRM_ERROR("amdgpu: ring %d test failed (0x%08X)\n",
1031 			  ring->idx, tmp);
1032 		r = -EINVAL;
1033 	}
1034 	amdgpu_device_wb_free(adev, index);
1035 
1036 	return r;
1037 }
1038 
1039 /**
1040  * sdma_v4_0_ring_test_ib - test an IB on the DMA engine
1041  *
1042  * @ring: amdgpu_ring structure holding ring information
1043  *
1044  * Test a simple IB in the DMA ring (VEGA10).
1045  * Returns 0 on success, error on failure.
1046  */
1047 static int sdma_v4_0_ring_test_ib(struct amdgpu_ring *ring, long timeout)
1048 {
1049 	struct amdgpu_device *adev = ring->adev;
1050 	struct amdgpu_ib ib;
1051 	struct dma_fence *f = NULL;
1052 	unsigned index;
1053 	long r;
1054 	u32 tmp = 0;
1055 	u64 gpu_addr;
1056 
1057 	r = amdgpu_device_wb_get(adev, &index);
1058 	if (r) {
1059 		dev_err(adev->dev, "(%ld) failed to allocate wb slot\n", r);
1060 		return r;
1061 	}
1062 
1063 	gpu_addr = adev->wb.gpu_addr + (index * 4);
1064 	tmp = 0xCAFEDEAD;
1065 	adev->wb.wb[index] = cpu_to_le32(tmp);
1066 	memset(&ib, 0, sizeof(ib));
1067 	r = amdgpu_ib_get(adev, NULL, 256, &ib);
1068 	if (r) {
1069 		DRM_ERROR("amdgpu: failed to get ib (%ld).\n", r);
1070 		goto err0;
1071 	}
1072 
1073 	ib.ptr[0] = SDMA_PKT_HEADER_OP(SDMA_OP_WRITE) |
1074 		SDMA_PKT_HEADER_SUB_OP(SDMA_SUBOP_WRITE_LINEAR);
1075 	ib.ptr[1] = lower_32_bits(gpu_addr);
1076 	ib.ptr[2] = upper_32_bits(gpu_addr);
1077 	ib.ptr[3] = SDMA_PKT_WRITE_UNTILED_DW_3_COUNT(0);
1078 	ib.ptr[4] = 0xDEADBEEF;
1079 	ib.ptr[5] = SDMA_PKT_NOP_HEADER_OP(SDMA_OP_NOP);
1080 	ib.ptr[6] = SDMA_PKT_NOP_HEADER_OP(SDMA_OP_NOP);
1081 	ib.ptr[7] = SDMA_PKT_NOP_HEADER_OP(SDMA_OP_NOP);
1082 	ib.length_dw = 8;
1083 
1084 	r = amdgpu_ib_schedule(ring, 1, &ib, NULL, &f);
1085 	if (r)
1086 		goto err1;
1087 
1088 	r = dma_fence_wait_timeout(f, false, timeout);
1089 	if (r == 0) {
1090 		DRM_ERROR("amdgpu: IB test timed out\n");
1091 		r = -ETIMEDOUT;
1092 		goto err1;
1093 	} else if (r < 0) {
1094 		DRM_ERROR("amdgpu: fence wait failed (%ld).\n", r);
1095 		goto err1;
1096 	}
1097 	tmp = le32_to_cpu(adev->wb.wb[index]);
1098 	if (tmp == 0xDEADBEEF) {
1099 		DRM_DEBUG("ib test on ring %d succeeded\n", ring->idx);
1100 		r = 0;
1101 	} else {
1102 		DRM_ERROR("amdgpu: ib test failed (0x%08X)\n", tmp);
1103 		r = -EINVAL;
1104 	}
1105 err1:
1106 	amdgpu_ib_free(adev, &ib, NULL);
1107 	dma_fence_put(f);
1108 err0:
1109 	amdgpu_device_wb_free(adev, index);
1110 	return r;
1111 }
1112 
1113 
1114 /**
1115  * sdma_v4_0_vm_copy_pte - update PTEs by copying them from the GART
1116  *
1117  * @ib: indirect buffer to fill with commands
1118  * @pe: addr of the page entry
1119  * @src: src addr to copy from
1120  * @count: number of page entries to update
1121  *
1122  * Update PTEs by copying them from the GART using sDMA (VEGA10).
1123  */
1124 static void sdma_v4_0_vm_copy_pte(struct amdgpu_ib *ib,
1125 				  uint64_t pe, uint64_t src,
1126 				  unsigned count)
1127 {
1128 	unsigned bytes = count * 8;
1129 
1130 	ib->ptr[ib->length_dw++] = SDMA_PKT_HEADER_OP(SDMA_OP_COPY) |
1131 		SDMA_PKT_HEADER_SUB_OP(SDMA_SUBOP_COPY_LINEAR);
1132 	ib->ptr[ib->length_dw++] = bytes - 1;
1133 	ib->ptr[ib->length_dw++] = 0; /* src/dst endian swap */
1134 	ib->ptr[ib->length_dw++] = lower_32_bits(src);
1135 	ib->ptr[ib->length_dw++] = upper_32_bits(src);
1136 	ib->ptr[ib->length_dw++] = lower_32_bits(pe);
1137 	ib->ptr[ib->length_dw++] = upper_32_bits(pe);
1138 
1139 }
1140 
1141 /**
1142  * sdma_v4_0_vm_write_pte - update PTEs by writing them manually
1143  *
1144  * @ib: indirect buffer to fill with commands
1145  * @pe: addr of the page entry
1146  * @addr: dst addr to write into pe
1147  * @count: number of page entries to update
1148  * @incr: increase next addr by incr bytes
1149  * @flags: access flags
1150  *
1151  * Update PTEs by writing them manually using sDMA (VEGA10).
1152  */
1153 static void sdma_v4_0_vm_write_pte(struct amdgpu_ib *ib, uint64_t pe,
1154 				   uint64_t value, unsigned count,
1155 				   uint32_t incr)
1156 {
1157 	unsigned ndw = count * 2;
1158 
1159 	ib->ptr[ib->length_dw++] = SDMA_PKT_HEADER_OP(SDMA_OP_WRITE) |
1160 		SDMA_PKT_HEADER_SUB_OP(SDMA_SUBOP_WRITE_LINEAR);
1161 	ib->ptr[ib->length_dw++] = lower_32_bits(pe);
1162 	ib->ptr[ib->length_dw++] = upper_32_bits(pe);
1163 	ib->ptr[ib->length_dw++] = ndw - 1;
1164 	for (; ndw > 0; ndw -= 2) {
1165 		ib->ptr[ib->length_dw++] = lower_32_bits(value);
1166 		ib->ptr[ib->length_dw++] = upper_32_bits(value);
1167 		value += incr;
1168 	}
1169 }
1170 
1171 /**
1172  * sdma_v4_0_vm_set_pte_pde - update the page tables using sDMA
1173  *
1174  * @ib: indirect buffer to fill with commands
1175  * @pe: addr of the page entry
1176  * @addr: dst addr to write into pe
1177  * @count: number of page entries to update
1178  * @incr: increase next addr by incr bytes
1179  * @flags: access flags
1180  *
1181  * Update the page tables using sDMA (VEGA10).
1182  */
1183 static void sdma_v4_0_vm_set_pte_pde(struct amdgpu_ib *ib,
1184 				     uint64_t pe,
1185 				     uint64_t addr, unsigned count,
1186 				     uint32_t incr, uint64_t flags)
1187 {
1188 	/* for physically contiguous pages (vram) */
1189 	ib->ptr[ib->length_dw++] = SDMA_PKT_HEADER_OP(SDMA_OP_PTEPDE);
1190 	ib->ptr[ib->length_dw++] = lower_32_bits(pe); /* dst addr */
1191 	ib->ptr[ib->length_dw++] = upper_32_bits(pe);
1192 	ib->ptr[ib->length_dw++] = lower_32_bits(flags); /* mask */
1193 	ib->ptr[ib->length_dw++] = upper_32_bits(flags);
1194 	ib->ptr[ib->length_dw++] = lower_32_bits(addr); /* value */
1195 	ib->ptr[ib->length_dw++] = upper_32_bits(addr);
1196 	ib->ptr[ib->length_dw++] = incr; /* increment size */
1197 	ib->ptr[ib->length_dw++] = 0;
1198 	ib->ptr[ib->length_dw++] = count - 1; /* number of entries */
1199 }
1200 
1201 /**
1202  * sdma_v4_0_ring_pad_ib - pad the IB to the required number of dw
1203  *
1204  * @ib: indirect buffer to fill with padding
1205  *
1206  */
1207 static void sdma_v4_0_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib)
1208 {
1209 	struct amdgpu_sdma_instance *sdma = amdgpu_get_sdma_instance(ring);
1210 	u32 pad_count;
1211 	int i;
1212 
1213 	pad_count = (8 - (ib->length_dw & 0x7)) % 8;
1214 	for (i = 0; i < pad_count; i++)
1215 		if (sdma && sdma->burst_nop && (i == 0))
1216 			ib->ptr[ib->length_dw++] =
1217 				SDMA_PKT_HEADER_OP(SDMA_OP_NOP) |
1218 				SDMA_PKT_NOP_HEADER_COUNT(pad_count - 1);
1219 		else
1220 			ib->ptr[ib->length_dw++] =
1221 				SDMA_PKT_HEADER_OP(SDMA_OP_NOP);
1222 }
1223 
1224 
1225 /**
1226  * sdma_v4_0_ring_emit_pipeline_sync - sync the pipeline
1227  *
1228  * @ring: amdgpu_ring pointer
1229  *
1230  * Make sure all previous operations are completed (CIK).
1231  */
1232 static void sdma_v4_0_ring_emit_pipeline_sync(struct amdgpu_ring *ring)
1233 {
1234 	uint32_t seq = ring->fence_drv.sync_seq;
1235 	uint64_t addr = ring->fence_drv.gpu_addr;
1236 
1237 	/* wait for idle */
1238 	sdma_v4_0_wait_reg_mem(ring, 1, 0,
1239 			       addr & 0xfffffffc,
1240 			       upper_32_bits(addr) & 0xffffffff,
1241 			       seq, 0xffffffff, 4);
1242 }
1243 
1244 
1245 /**
1246  * sdma_v4_0_ring_emit_vm_flush - vm flush using sDMA
1247  *
1248  * @ring: amdgpu_ring pointer
1249  * @vm: amdgpu_vm pointer
1250  *
1251  * Update the page table base and flush the VM TLB
1252  * using sDMA (VEGA10).
1253  */
1254 static void sdma_v4_0_ring_emit_vm_flush(struct amdgpu_ring *ring,
1255 					 unsigned vmid, uint64_t pd_addr)
1256 {
1257 	amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr);
1258 }
1259 
1260 static void sdma_v4_0_ring_emit_wreg(struct amdgpu_ring *ring,
1261 				     uint32_t reg, uint32_t val)
1262 {
1263 	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_SRBM_WRITE) |
1264 			  SDMA_PKT_SRBM_WRITE_HEADER_BYTE_EN(0xf));
1265 	amdgpu_ring_write(ring, reg);
1266 	amdgpu_ring_write(ring, val);
1267 }
1268 
1269 static void sdma_v4_0_ring_emit_reg_wait(struct amdgpu_ring *ring, uint32_t reg,
1270 					 uint32_t val, uint32_t mask)
1271 {
1272 	sdma_v4_0_wait_reg_mem(ring, 0, 0, reg, 0, val, mask, 10);
1273 }
1274 
1275 static int sdma_v4_0_early_init(void *handle)
1276 {
1277 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1278 
1279 	if (adev->asic_type == CHIP_RAVEN)
1280 		adev->sdma.num_instances = 1;
1281 	else
1282 		adev->sdma.num_instances = 2;
1283 
1284 	sdma_v4_0_set_ring_funcs(adev);
1285 	sdma_v4_0_set_buffer_funcs(adev);
1286 	sdma_v4_0_set_vm_pte_funcs(adev);
1287 	sdma_v4_0_set_irq_funcs(adev);
1288 
1289 	return 0;
1290 }
1291 
1292 
1293 static int sdma_v4_0_sw_init(void *handle)
1294 {
1295 	struct amdgpu_ring *ring;
1296 	int r, i;
1297 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1298 
1299 	/* SDMA trap event */
1300 	r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_SDMA0, SDMA0_4_0__SRCID__SDMA_TRAP,
1301 			      &adev->sdma.trap_irq);
1302 	if (r)
1303 		return r;
1304 
1305 	/* SDMA trap event */
1306 	r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_SDMA1, SDMA1_4_0__SRCID__SDMA_TRAP,
1307 			      &adev->sdma.trap_irq);
1308 	if (r)
1309 		return r;
1310 
1311 	r = sdma_v4_0_init_microcode(adev);
1312 	if (r) {
1313 		DRM_ERROR("Failed to load sdma firmware!\n");
1314 		return r;
1315 	}
1316 
1317 	for (i = 0; i < adev->sdma.num_instances; i++) {
1318 		ring = &adev->sdma.instance[i].ring;
1319 		ring->ring_obj = NULL;
1320 		ring->use_doorbell = true;
1321 
1322 		DRM_INFO("use_doorbell being set to: [%s]\n",
1323 				ring->use_doorbell?"true":"false");
1324 
1325 		if (adev->asic_type == CHIP_VEGA10)
1326 			ring->doorbell_index = (i == 0) ?
1327 				(AMDGPU_VEGA10_DOORBELL64_sDMA_ENGINE0 << 1) //get DWORD offset
1328 				: (AMDGPU_VEGA10_DOORBELL64_sDMA_ENGINE1 << 1); // get DWORD offset
1329 		else
1330 			ring->doorbell_index = (i == 0) ?
1331 				(AMDGPU_DOORBELL64_sDMA_ENGINE0 << 1) //get DWORD offset
1332 				: (AMDGPU_DOORBELL64_sDMA_ENGINE1 << 1); // get DWORD offset
1333 
1334 
1335 		sprintf(ring->name, "sdma%d", i);
1336 		r = amdgpu_ring_init(adev, ring, 1024,
1337 				     &adev->sdma.trap_irq,
1338 				     (i == 0) ?
1339 				     AMDGPU_SDMA_IRQ_TRAP0 :
1340 				     AMDGPU_SDMA_IRQ_TRAP1);
1341 		if (r)
1342 			return r;
1343 	}
1344 
1345 	return r;
1346 }
1347 
1348 static int sdma_v4_0_sw_fini(void *handle)
1349 {
1350 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1351 	int i;
1352 
1353 	for (i = 0; i < adev->sdma.num_instances; i++)
1354 		amdgpu_ring_fini(&adev->sdma.instance[i].ring);
1355 
1356 	for (i = 0; i < adev->sdma.num_instances; i++) {
1357 		release_firmware(adev->sdma.instance[i].fw);
1358 		adev->sdma.instance[i].fw = NULL;
1359 	}
1360 
1361 	return 0;
1362 }
1363 
1364 static int sdma_v4_0_hw_init(void *handle)
1365 {
1366 	int r;
1367 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1368 
1369 	if (adev->asic_type == CHIP_RAVEN && adev->powerplay.pp_funcs->set_powergating_by_smu)
1370 		amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_SDMA, false);
1371 
1372 	sdma_v4_0_init_golden_registers(adev);
1373 
1374 	r = sdma_v4_0_start(adev);
1375 
1376 	return r;
1377 }
1378 
1379 static int sdma_v4_0_hw_fini(void *handle)
1380 {
1381 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1382 
1383 	if (amdgpu_sriov_vf(adev))
1384 		return 0;
1385 
1386 	sdma_v4_0_ctx_switch_enable(adev, false);
1387 	sdma_v4_0_enable(adev, false);
1388 
1389 	if (adev->asic_type == CHIP_RAVEN && adev->powerplay.pp_funcs->set_powergating_by_smu)
1390 		amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_SDMA, true);
1391 
1392 	return 0;
1393 }
1394 
1395 static int sdma_v4_0_suspend(void *handle)
1396 {
1397 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1398 
1399 	return sdma_v4_0_hw_fini(adev);
1400 }
1401 
1402 static int sdma_v4_0_resume(void *handle)
1403 {
1404 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1405 
1406 	return sdma_v4_0_hw_init(adev);
1407 }
1408 
1409 static bool sdma_v4_0_is_idle(void *handle)
1410 {
1411 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1412 	u32 i;
1413 
1414 	for (i = 0; i < adev->sdma.num_instances; i++) {
1415 		u32 tmp = RREG32(sdma_v4_0_get_reg_offset(adev, i, mmSDMA0_STATUS_REG));
1416 
1417 		if (!(tmp & SDMA0_STATUS_REG__IDLE_MASK))
1418 			return false;
1419 	}
1420 
1421 	return true;
1422 }
1423 
1424 static int sdma_v4_0_wait_for_idle(void *handle)
1425 {
1426 	unsigned i;
1427 	u32 sdma0, sdma1;
1428 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1429 
1430 	for (i = 0; i < adev->usec_timeout; i++) {
1431 		sdma0 = RREG32(sdma_v4_0_get_reg_offset(adev, 0, mmSDMA0_STATUS_REG));
1432 		sdma1 = RREG32(sdma_v4_0_get_reg_offset(adev, 1, mmSDMA0_STATUS_REG));
1433 
1434 		if (sdma0 & sdma1 & SDMA0_STATUS_REG__IDLE_MASK)
1435 			return 0;
1436 		udelay(1);
1437 	}
1438 	return -ETIMEDOUT;
1439 }
1440 
1441 static int sdma_v4_0_soft_reset(void *handle)
1442 {
1443 	/* todo */
1444 
1445 	return 0;
1446 }
1447 
1448 static int sdma_v4_0_set_trap_irq_state(struct amdgpu_device *adev,
1449 					struct amdgpu_irq_src *source,
1450 					unsigned type,
1451 					enum amdgpu_interrupt_state state)
1452 {
1453 	u32 sdma_cntl;
1454 
1455 	u32 reg_offset = (type == AMDGPU_SDMA_IRQ_TRAP0) ?
1456 		sdma_v4_0_get_reg_offset(adev, 0, mmSDMA0_CNTL) :
1457 		sdma_v4_0_get_reg_offset(adev, 1, mmSDMA0_CNTL);
1458 
1459 	sdma_cntl = RREG32(reg_offset);
1460 	sdma_cntl = REG_SET_FIELD(sdma_cntl, SDMA0_CNTL, TRAP_ENABLE,
1461 		       state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0);
1462 	WREG32(reg_offset, sdma_cntl);
1463 
1464 	return 0;
1465 }
1466 
1467 static int sdma_v4_0_process_trap_irq(struct amdgpu_device *adev,
1468 				      struct amdgpu_irq_src *source,
1469 				      struct amdgpu_iv_entry *entry)
1470 {
1471 	DRM_DEBUG("IH: SDMA trap\n");
1472 	switch (entry->client_id) {
1473 	case SOC15_IH_CLIENTID_SDMA0:
1474 		switch (entry->ring_id) {
1475 		case 0:
1476 			amdgpu_fence_process(&adev->sdma.instance[0].ring);
1477 			break;
1478 		case 1:
1479 			/* XXX compute */
1480 			break;
1481 		case 2:
1482 			/* XXX compute */
1483 			break;
1484 		case 3:
1485 			/* XXX page queue*/
1486 			break;
1487 		}
1488 		break;
1489 	case SOC15_IH_CLIENTID_SDMA1:
1490 		switch (entry->ring_id) {
1491 		case 0:
1492 			amdgpu_fence_process(&adev->sdma.instance[1].ring);
1493 			break;
1494 		case 1:
1495 			/* XXX compute */
1496 			break;
1497 		case 2:
1498 			/* XXX compute */
1499 			break;
1500 		case 3:
1501 			/* XXX page queue*/
1502 			break;
1503 		}
1504 		break;
1505 	}
1506 	return 0;
1507 }
1508 
1509 static int sdma_v4_0_process_illegal_inst_irq(struct amdgpu_device *adev,
1510 					      struct amdgpu_irq_src *source,
1511 					      struct amdgpu_iv_entry *entry)
1512 {
1513 	DRM_ERROR("Illegal instruction in SDMA command stream\n");
1514 	schedule_work(&adev->reset_work);
1515 	return 0;
1516 }
1517 
1518 
1519 static void sdma_v4_0_update_medium_grain_clock_gating(
1520 		struct amdgpu_device *adev,
1521 		bool enable)
1522 {
1523 	uint32_t data, def;
1524 
1525 	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_SDMA_MGCG)) {
1526 		/* enable sdma0 clock gating */
1527 		def = data = RREG32(SOC15_REG_OFFSET(SDMA0, 0, mmSDMA0_CLK_CTRL));
1528 		data &= ~(SDMA0_CLK_CTRL__SOFT_OVERRIDE7_MASK |
1529 			  SDMA0_CLK_CTRL__SOFT_OVERRIDE6_MASK |
1530 			  SDMA0_CLK_CTRL__SOFT_OVERRIDE5_MASK |
1531 			  SDMA0_CLK_CTRL__SOFT_OVERRIDE4_MASK |
1532 			  SDMA0_CLK_CTRL__SOFT_OVERRIDE3_MASK |
1533 			  SDMA0_CLK_CTRL__SOFT_OVERRIDE2_MASK |
1534 			  SDMA0_CLK_CTRL__SOFT_OVERRIDE1_MASK |
1535 			  SDMA0_CLK_CTRL__SOFT_OVERRIDE0_MASK);
1536 		if (def != data)
1537 			WREG32(SOC15_REG_OFFSET(SDMA0, 0, mmSDMA0_CLK_CTRL), data);
1538 
1539 		if (adev->sdma.num_instances > 1) {
1540 			def = data = RREG32(SOC15_REG_OFFSET(SDMA1, 0, mmSDMA1_CLK_CTRL));
1541 			data &= ~(SDMA1_CLK_CTRL__SOFT_OVERRIDE7_MASK |
1542 				  SDMA1_CLK_CTRL__SOFT_OVERRIDE6_MASK |
1543 				  SDMA1_CLK_CTRL__SOFT_OVERRIDE5_MASK |
1544 				  SDMA1_CLK_CTRL__SOFT_OVERRIDE4_MASK |
1545 				  SDMA1_CLK_CTRL__SOFT_OVERRIDE3_MASK |
1546 				  SDMA1_CLK_CTRL__SOFT_OVERRIDE2_MASK |
1547 				  SDMA1_CLK_CTRL__SOFT_OVERRIDE1_MASK |
1548 				  SDMA1_CLK_CTRL__SOFT_OVERRIDE0_MASK);
1549 			if (def != data)
1550 				WREG32(SOC15_REG_OFFSET(SDMA1, 0, mmSDMA1_CLK_CTRL), data);
1551 		}
1552 	} else {
1553 		/* disable sdma0 clock gating */
1554 		def = data = RREG32(SOC15_REG_OFFSET(SDMA0, 0, mmSDMA0_CLK_CTRL));
1555 		data |= (SDMA0_CLK_CTRL__SOFT_OVERRIDE7_MASK |
1556 			 SDMA0_CLK_CTRL__SOFT_OVERRIDE6_MASK |
1557 			 SDMA0_CLK_CTRL__SOFT_OVERRIDE5_MASK |
1558 			 SDMA0_CLK_CTRL__SOFT_OVERRIDE4_MASK |
1559 			 SDMA0_CLK_CTRL__SOFT_OVERRIDE3_MASK |
1560 			 SDMA0_CLK_CTRL__SOFT_OVERRIDE2_MASK |
1561 			 SDMA0_CLK_CTRL__SOFT_OVERRIDE1_MASK |
1562 			 SDMA0_CLK_CTRL__SOFT_OVERRIDE0_MASK);
1563 
1564 		if (def != data)
1565 			WREG32(SOC15_REG_OFFSET(SDMA0, 0, mmSDMA0_CLK_CTRL), data);
1566 
1567 		if (adev->sdma.num_instances > 1) {
1568 			def = data = RREG32(SOC15_REG_OFFSET(SDMA1, 0, mmSDMA1_CLK_CTRL));
1569 			data |= (SDMA1_CLK_CTRL__SOFT_OVERRIDE7_MASK |
1570 				 SDMA1_CLK_CTRL__SOFT_OVERRIDE6_MASK |
1571 				 SDMA1_CLK_CTRL__SOFT_OVERRIDE5_MASK |
1572 				 SDMA1_CLK_CTRL__SOFT_OVERRIDE4_MASK |
1573 				 SDMA1_CLK_CTRL__SOFT_OVERRIDE3_MASK |
1574 				 SDMA1_CLK_CTRL__SOFT_OVERRIDE2_MASK |
1575 				 SDMA1_CLK_CTRL__SOFT_OVERRIDE1_MASK |
1576 				 SDMA1_CLK_CTRL__SOFT_OVERRIDE0_MASK);
1577 			if (def != data)
1578 				WREG32(SOC15_REG_OFFSET(SDMA1, 0, mmSDMA1_CLK_CTRL), data);
1579 		}
1580 	}
1581 }
1582 
1583 
1584 static void sdma_v4_0_update_medium_grain_light_sleep(
1585 		struct amdgpu_device *adev,
1586 		bool enable)
1587 {
1588 	uint32_t data, def;
1589 
1590 	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_SDMA_LS)) {
1591 		/* 1-not override: enable sdma0 mem light sleep */
1592 		def = data = RREG32(SOC15_REG_OFFSET(SDMA0, 0, mmSDMA0_POWER_CNTL));
1593 		data |= SDMA0_POWER_CNTL__MEM_POWER_OVERRIDE_MASK;
1594 		if (def != data)
1595 			WREG32(SOC15_REG_OFFSET(SDMA0, 0, mmSDMA0_POWER_CNTL), data);
1596 
1597 		/* 1-not override: enable sdma1 mem light sleep */
1598 		if (adev->sdma.num_instances > 1) {
1599 			def = data = RREG32(SOC15_REG_OFFSET(SDMA1, 0, mmSDMA1_POWER_CNTL));
1600 			data |= SDMA1_POWER_CNTL__MEM_POWER_OVERRIDE_MASK;
1601 			if (def != data)
1602 				WREG32(SOC15_REG_OFFSET(SDMA1, 0, mmSDMA1_POWER_CNTL), data);
1603 		}
1604 	} else {
1605 		/* 0-override:disable sdma0 mem light sleep */
1606 		def = data = RREG32(SOC15_REG_OFFSET(SDMA0, 0, mmSDMA0_POWER_CNTL));
1607 		data &= ~SDMA0_POWER_CNTL__MEM_POWER_OVERRIDE_MASK;
1608 		if (def != data)
1609 			WREG32(SOC15_REG_OFFSET(SDMA0, 0, mmSDMA0_POWER_CNTL), data);
1610 
1611 		/* 0-override:disable sdma1 mem light sleep */
1612 		if (adev->sdma.num_instances > 1) {
1613 			def = data = RREG32(SOC15_REG_OFFSET(SDMA1, 0, mmSDMA1_POWER_CNTL));
1614 			data &= ~SDMA1_POWER_CNTL__MEM_POWER_OVERRIDE_MASK;
1615 			if (def != data)
1616 				WREG32(SOC15_REG_OFFSET(SDMA1, 0, mmSDMA1_POWER_CNTL), data);
1617 		}
1618 	}
1619 }
1620 
1621 static int sdma_v4_0_set_clockgating_state(void *handle,
1622 					  enum amd_clockgating_state state)
1623 {
1624 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1625 
1626 	if (amdgpu_sriov_vf(adev))
1627 		return 0;
1628 
1629 	switch (adev->asic_type) {
1630 	case CHIP_VEGA10:
1631 	case CHIP_VEGA12:
1632 	case CHIP_VEGA20:
1633 	case CHIP_RAVEN:
1634 		sdma_v4_0_update_medium_grain_clock_gating(adev,
1635 				state == AMD_CG_STATE_GATE ? true : false);
1636 		sdma_v4_0_update_medium_grain_light_sleep(adev,
1637 				state == AMD_CG_STATE_GATE ? true : false);
1638 		break;
1639 	default:
1640 		break;
1641 	}
1642 	return 0;
1643 }
1644 
1645 static int sdma_v4_0_set_powergating_state(void *handle,
1646 					  enum amd_powergating_state state)
1647 {
1648 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1649 
1650 	switch (adev->asic_type) {
1651 	case CHIP_RAVEN:
1652 		sdma_v4_1_update_power_gating(adev,
1653 				state == AMD_PG_STATE_GATE ? true : false);
1654 		break;
1655 	default:
1656 		break;
1657 	}
1658 
1659 	return 0;
1660 }
1661 
1662 static void sdma_v4_0_get_clockgating_state(void *handle, u32 *flags)
1663 {
1664 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1665 	int data;
1666 
1667 	if (amdgpu_sriov_vf(adev))
1668 		*flags = 0;
1669 
1670 	/* AMD_CG_SUPPORT_SDMA_MGCG */
1671 	data = RREG32(SOC15_REG_OFFSET(SDMA0, 0, mmSDMA0_CLK_CTRL));
1672 	if (!(data & SDMA0_CLK_CTRL__SOFT_OVERRIDE7_MASK))
1673 		*flags |= AMD_CG_SUPPORT_SDMA_MGCG;
1674 
1675 	/* AMD_CG_SUPPORT_SDMA_LS */
1676 	data = RREG32(SOC15_REG_OFFSET(SDMA0, 0, mmSDMA0_POWER_CNTL));
1677 	if (data & SDMA0_POWER_CNTL__MEM_POWER_OVERRIDE_MASK)
1678 		*flags |= AMD_CG_SUPPORT_SDMA_LS;
1679 }
1680 
1681 const struct amd_ip_funcs sdma_v4_0_ip_funcs = {
1682 	.name = "sdma_v4_0",
1683 	.early_init = sdma_v4_0_early_init,
1684 	.late_init = NULL,
1685 	.sw_init = sdma_v4_0_sw_init,
1686 	.sw_fini = sdma_v4_0_sw_fini,
1687 	.hw_init = sdma_v4_0_hw_init,
1688 	.hw_fini = sdma_v4_0_hw_fini,
1689 	.suspend = sdma_v4_0_suspend,
1690 	.resume = sdma_v4_0_resume,
1691 	.is_idle = sdma_v4_0_is_idle,
1692 	.wait_for_idle = sdma_v4_0_wait_for_idle,
1693 	.soft_reset = sdma_v4_0_soft_reset,
1694 	.set_clockgating_state = sdma_v4_0_set_clockgating_state,
1695 	.set_powergating_state = sdma_v4_0_set_powergating_state,
1696 	.get_clockgating_state = sdma_v4_0_get_clockgating_state,
1697 };
1698 
1699 static const struct amdgpu_ring_funcs sdma_v4_0_ring_funcs = {
1700 	.type = AMDGPU_RING_TYPE_SDMA,
1701 	.align_mask = 0xf,
1702 	.nop = SDMA_PKT_NOP_HEADER_OP(SDMA_OP_NOP),
1703 	.support_64bit_ptrs = true,
1704 	.vmhub = AMDGPU_MMHUB,
1705 	.get_rptr = sdma_v4_0_ring_get_rptr,
1706 	.get_wptr = sdma_v4_0_ring_get_wptr,
1707 	.set_wptr = sdma_v4_0_ring_set_wptr,
1708 	.emit_frame_size =
1709 		6 + /* sdma_v4_0_ring_emit_hdp_flush */
1710 		3 + /* hdp invalidate */
1711 		6 + /* sdma_v4_0_ring_emit_pipeline_sync */
1712 		/* sdma_v4_0_ring_emit_vm_flush */
1713 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
1714 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 6 +
1715 		10 + 10 + 10, /* sdma_v4_0_ring_emit_fence x3 for user fence, vm fence */
1716 	.emit_ib_size = 7 + 6, /* sdma_v4_0_ring_emit_ib */
1717 	.emit_ib = sdma_v4_0_ring_emit_ib,
1718 	.emit_fence = sdma_v4_0_ring_emit_fence,
1719 	.emit_pipeline_sync = sdma_v4_0_ring_emit_pipeline_sync,
1720 	.emit_vm_flush = sdma_v4_0_ring_emit_vm_flush,
1721 	.emit_hdp_flush = sdma_v4_0_ring_emit_hdp_flush,
1722 	.test_ring = sdma_v4_0_ring_test_ring,
1723 	.test_ib = sdma_v4_0_ring_test_ib,
1724 	.insert_nop = sdma_v4_0_ring_insert_nop,
1725 	.pad_ib = sdma_v4_0_ring_pad_ib,
1726 	.emit_wreg = sdma_v4_0_ring_emit_wreg,
1727 	.emit_reg_wait = sdma_v4_0_ring_emit_reg_wait,
1728 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
1729 };
1730 
1731 static void sdma_v4_0_set_ring_funcs(struct amdgpu_device *adev)
1732 {
1733 	int i;
1734 
1735 	for (i = 0; i < adev->sdma.num_instances; i++) {
1736 		adev->sdma.instance[i].ring.funcs = &sdma_v4_0_ring_funcs;
1737 		adev->sdma.instance[i].ring.me = i;
1738 	}
1739 }
1740 
1741 static const struct amdgpu_irq_src_funcs sdma_v4_0_trap_irq_funcs = {
1742 	.set = sdma_v4_0_set_trap_irq_state,
1743 	.process = sdma_v4_0_process_trap_irq,
1744 };
1745 
1746 static const struct amdgpu_irq_src_funcs sdma_v4_0_illegal_inst_irq_funcs = {
1747 	.process = sdma_v4_0_process_illegal_inst_irq,
1748 };
1749 
1750 static void sdma_v4_0_set_irq_funcs(struct amdgpu_device *adev)
1751 {
1752 	adev->sdma.trap_irq.num_types = AMDGPU_SDMA_IRQ_LAST;
1753 	adev->sdma.trap_irq.funcs = &sdma_v4_0_trap_irq_funcs;
1754 	adev->sdma.illegal_inst_irq.funcs = &sdma_v4_0_illegal_inst_irq_funcs;
1755 }
1756 
1757 /**
1758  * sdma_v4_0_emit_copy_buffer - copy buffer using the sDMA engine
1759  *
1760  * @ring: amdgpu_ring structure holding ring information
1761  * @src_offset: src GPU address
1762  * @dst_offset: dst GPU address
1763  * @byte_count: number of bytes to xfer
1764  *
1765  * Copy GPU buffers using the DMA engine (VEGA10/12).
1766  * Used by the amdgpu ttm implementation to move pages if
1767  * registered as the asic copy callback.
1768  */
1769 static void sdma_v4_0_emit_copy_buffer(struct amdgpu_ib *ib,
1770 				       uint64_t src_offset,
1771 				       uint64_t dst_offset,
1772 				       uint32_t byte_count)
1773 {
1774 	ib->ptr[ib->length_dw++] = SDMA_PKT_HEADER_OP(SDMA_OP_COPY) |
1775 		SDMA_PKT_HEADER_SUB_OP(SDMA_SUBOP_COPY_LINEAR);
1776 	ib->ptr[ib->length_dw++] = byte_count - 1;
1777 	ib->ptr[ib->length_dw++] = 0; /* src/dst endian swap */
1778 	ib->ptr[ib->length_dw++] = lower_32_bits(src_offset);
1779 	ib->ptr[ib->length_dw++] = upper_32_bits(src_offset);
1780 	ib->ptr[ib->length_dw++] = lower_32_bits(dst_offset);
1781 	ib->ptr[ib->length_dw++] = upper_32_bits(dst_offset);
1782 }
1783 
1784 /**
1785  * sdma_v4_0_emit_fill_buffer - fill buffer using the sDMA engine
1786  *
1787  * @ring: amdgpu_ring structure holding ring information
1788  * @src_data: value to write to buffer
1789  * @dst_offset: dst GPU address
1790  * @byte_count: number of bytes to xfer
1791  *
1792  * Fill GPU buffers using the DMA engine (VEGA10/12).
1793  */
1794 static void sdma_v4_0_emit_fill_buffer(struct amdgpu_ib *ib,
1795 				       uint32_t src_data,
1796 				       uint64_t dst_offset,
1797 				       uint32_t byte_count)
1798 {
1799 	ib->ptr[ib->length_dw++] = SDMA_PKT_HEADER_OP(SDMA_OP_CONST_FILL);
1800 	ib->ptr[ib->length_dw++] = lower_32_bits(dst_offset);
1801 	ib->ptr[ib->length_dw++] = upper_32_bits(dst_offset);
1802 	ib->ptr[ib->length_dw++] = src_data;
1803 	ib->ptr[ib->length_dw++] = byte_count - 1;
1804 }
1805 
1806 static const struct amdgpu_buffer_funcs sdma_v4_0_buffer_funcs = {
1807 	.copy_max_bytes = 0x400000,
1808 	.copy_num_dw = 7,
1809 	.emit_copy_buffer = sdma_v4_0_emit_copy_buffer,
1810 
1811 	.fill_max_bytes = 0x400000,
1812 	.fill_num_dw = 5,
1813 	.emit_fill_buffer = sdma_v4_0_emit_fill_buffer,
1814 };
1815 
1816 static void sdma_v4_0_set_buffer_funcs(struct amdgpu_device *adev)
1817 {
1818 	adev->mman.buffer_funcs = &sdma_v4_0_buffer_funcs;
1819 	adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].ring;
1820 }
1821 
1822 static const struct amdgpu_vm_pte_funcs sdma_v4_0_vm_pte_funcs = {
1823 	.copy_pte_num_dw = 7,
1824 	.copy_pte = sdma_v4_0_vm_copy_pte,
1825 
1826 	.write_pte = sdma_v4_0_vm_write_pte,
1827 	.set_pte_pde = sdma_v4_0_vm_set_pte_pde,
1828 };
1829 
1830 static void sdma_v4_0_set_vm_pte_funcs(struct amdgpu_device *adev)
1831 {
1832 	struct drm_gpu_scheduler *sched;
1833 	unsigned i;
1834 
1835 	adev->vm_manager.vm_pte_funcs = &sdma_v4_0_vm_pte_funcs;
1836 	for (i = 0; i < adev->sdma.num_instances; i++) {
1837 		sched = &adev->sdma.instance[i].ring.sched;
1838 		adev->vm_manager.vm_pte_rqs[i] =
1839 			&sched->sched_rq[DRM_SCHED_PRIORITY_KERNEL];
1840 	}
1841 	adev->vm_manager.vm_pte_num_rqs = adev->sdma.num_instances;
1842 }
1843 
1844 const struct amdgpu_ip_block_version sdma_v4_0_ip_block = {
1845 	.type = AMD_IP_BLOCK_TYPE_SDMA,
1846 	.major = 4,
1847 	.minor = 0,
1848 	.rev = 0,
1849 	.funcs = &sdma_v4_0_ip_funcs,
1850 };
1851