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/delay.h>
25 #include <linux/firmware.h>
26 #include <linux/module.h>
27 #include <linux/pci.h>
28 
29 #include "amdgpu.h"
30 #include "amdgpu_ucode.h"
31 #include "amdgpu_trace.h"
32 
33 #include "sdma0/sdma0_4_2_offset.h"
34 #include "sdma0/sdma0_4_2_sh_mask.h"
35 #include "sdma1/sdma1_4_2_offset.h"
36 #include "sdma1/sdma1_4_2_sh_mask.h"
37 #include "sdma2/sdma2_4_2_2_offset.h"
38 #include "sdma2/sdma2_4_2_2_sh_mask.h"
39 #include "sdma3/sdma3_4_2_2_offset.h"
40 #include "sdma3/sdma3_4_2_2_sh_mask.h"
41 #include "sdma4/sdma4_4_2_2_offset.h"
42 #include "sdma4/sdma4_4_2_2_sh_mask.h"
43 #include "sdma5/sdma5_4_2_2_offset.h"
44 #include "sdma5/sdma5_4_2_2_sh_mask.h"
45 #include "sdma6/sdma6_4_2_2_offset.h"
46 #include "sdma6/sdma6_4_2_2_sh_mask.h"
47 #include "sdma7/sdma7_4_2_2_offset.h"
48 #include "sdma7/sdma7_4_2_2_sh_mask.h"
49 #include "sdma0/sdma0_4_1_default.h"
50 
51 #include "soc15_common.h"
52 #include "soc15.h"
53 #include "vega10_sdma_pkt_open.h"
54 
55 #include "ivsrcid/sdma0/irqsrcs_sdma0_4_0.h"
56 #include "ivsrcid/sdma1/irqsrcs_sdma1_4_0.h"
57 
58 #include "amdgpu_ras.h"
59 #include "sdma_v4_4.h"
60 
61 MODULE_FIRMWARE("amdgpu/vega10_sdma.bin");
62 MODULE_FIRMWARE("amdgpu/vega10_sdma1.bin");
63 MODULE_FIRMWARE("amdgpu/vega12_sdma.bin");
64 MODULE_FIRMWARE("amdgpu/vega12_sdma1.bin");
65 MODULE_FIRMWARE("amdgpu/vega20_sdma.bin");
66 MODULE_FIRMWARE("amdgpu/vega20_sdma1.bin");
67 MODULE_FIRMWARE("amdgpu/raven_sdma.bin");
68 MODULE_FIRMWARE("amdgpu/picasso_sdma.bin");
69 MODULE_FIRMWARE("amdgpu/raven2_sdma.bin");
70 MODULE_FIRMWARE("amdgpu/arcturus_sdma.bin");
71 MODULE_FIRMWARE("amdgpu/renoir_sdma.bin");
72 MODULE_FIRMWARE("amdgpu/green_sardine_sdma.bin");
73 MODULE_FIRMWARE("amdgpu/aldebaran_sdma.bin");
74 
75 #define SDMA0_POWER_CNTL__ON_OFF_CONDITION_HOLD_TIME_MASK  0x000000F8L
76 #define SDMA0_POWER_CNTL__ON_OFF_STATUS_DURATION_TIME_MASK 0xFC000000L
77 
78 #define WREG32_SDMA(instance, offset, value) \
79 	WREG32(sdma_v4_0_get_reg_offset(adev, (instance), (offset)), value)
80 #define RREG32_SDMA(instance, offset) \
81 	RREG32(sdma_v4_0_get_reg_offset(adev, (instance), (offset)))
82 
83 static void sdma_v4_0_set_ring_funcs(struct amdgpu_device *adev);
84 static void sdma_v4_0_set_buffer_funcs(struct amdgpu_device *adev);
85 static void sdma_v4_0_set_vm_pte_funcs(struct amdgpu_device *adev);
86 static void sdma_v4_0_set_irq_funcs(struct amdgpu_device *adev);
87 static void sdma_v4_0_set_ras_funcs(struct amdgpu_device *adev);
88 
89 static const struct soc15_reg_golden golden_settings_sdma_4[] = {
90 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_CHICKEN_BITS, 0xfe931f07, 0x02831d07),
91 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_CLK_CTRL, 0xff000ff0, 0x3f000100),
92 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GFX_IB_CNTL, 0x800f0100, 0x00000100),
93 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GFX_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
94 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_PAGE_IB_CNTL, 0x800f0100, 0x00000100),
95 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_PAGE_RB_WPTR_POLL_CNTL, 0x0000fff0, 0x00403000),
96 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_POWER_CNTL, 0x003ff006, 0x0003c000),
97 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC0_IB_CNTL, 0x800f0100, 0x00000100),
98 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC0_RB_WPTR_POLL_CNTL, 0x0000fff0, 0x00403000),
99 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC1_IB_CNTL, 0x800f0100, 0x00000100),
100 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC1_RB_WPTR_POLL_CNTL, 0x0000fff0, 0x00403000),
101 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_UTCL1_PAGE, 0x000003ff, 0x000003c0),
102 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_UTCL1_WATERMK, 0xfc000000, 0x00000000),
103 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_CLK_CTRL, 0xffffffff, 0x3f000100),
104 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_GFX_IB_CNTL, 0x800f0100, 0x00000100),
105 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_GFX_RB_WPTR_POLL_CNTL, 0x0000fff0, 0x00403000),
106 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_PAGE_IB_CNTL, 0x800f0100, 0x00000100),
107 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_PAGE_RB_WPTR_POLL_CNTL, 0x0000fff0, 0x00403000),
108 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_POWER_CNTL, 0x003ff000, 0x0003c000),
109 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_RLC0_IB_CNTL, 0x800f0100, 0x00000100),
110 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_RLC0_RB_WPTR_POLL_CNTL, 0x0000fff0, 0x00403000),
111 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_RLC1_IB_CNTL, 0x800f0100, 0x00000100),
112 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_RLC1_RB_WPTR_POLL_CNTL, 0x0000fff0, 0x00403000),
113 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_UTCL1_PAGE, 0x000003ff, 0x000003c0),
114 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_UTCL1_WATERMK, 0xfc000000, 0x00000000)
115 };
116 
117 static const struct soc15_reg_golden golden_settings_sdma_vg10[] = {
118 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GB_ADDR_CONFIG, 0x0018773f, 0x00104002),
119 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GB_ADDR_CONFIG_READ, 0x0018773f, 0x00104002),
120 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_UTCL1_TIMEOUT, 0xffffffff, 0x00010001),
121 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_CHICKEN_BITS, 0xfe931f07, 0x02831d07),
122 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_GB_ADDR_CONFIG, 0x0018773f, 0x00104002),
123 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_GB_ADDR_CONFIG_READ, 0x0018773f, 0x00104002),
124 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_UTCL1_TIMEOUT, 0xffffffff, 0x00010001),
125 };
126 
127 static const struct soc15_reg_golden golden_settings_sdma_vg12[] = {
128 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GB_ADDR_CONFIG, 0x0018773f, 0x00104001),
129 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GB_ADDR_CONFIG_READ, 0x0018773f, 0x00104001),
130 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_UTCL1_TIMEOUT, 0xffffffff, 0x00010001),
131 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_CHICKEN_BITS, 0xfe931f07, 0x02831d07),
132 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_GB_ADDR_CONFIG, 0x0018773f, 0x00104001),
133 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_GB_ADDR_CONFIG_READ, 0x0018773f, 0x00104001),
134 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_UTCL1_TIMEOUT, 0xffffffff, 0x00010001),
135 };
136 
137 static const struct soc15_reg_golden golden_settings_sdma_4_1[] = {
138 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_CHICKEN_BITS, 0xfe931f07, 0x02831d07),
139 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_CLK_CTRL, 0xffffffff, 0x3f000100),
140 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GFX_IB_CNTL, 0x800f0111, 0x00000100),
141 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GFX_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
142 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_POWER_CNTL, 0xfc3fffff, 0x40000051),
143 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC0_IB_CNTL, 0x800f0111, 0x00000100),
144 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC0_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
145 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC1_IB_CNTL, 0x800f0111, 0x00000100),
146 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC1_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
147 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_UTCL1_PAGE, 0x000003ff, 0x000003e0),
148 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_UTCL1_WATERMK, 0xfc000000, 0x00000000)
149 };
150 
151 static const struct soc15_reg_golden golden_settings_sdma0_4_2_init[] = {
152 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC0_RB_WPTR_POLL_CNTL, 0xfffffff0, 0x00403000),
153 };
154 
155 static const struct soc15_reg_golden golden_settings_sdma0_4_2[] =
156 {
157 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_CHICKEN_BITS, 0xfe931f07, 0x02831f07),
158 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_CLK_CTRL, 0xffffffff, 0x3f000100),
159 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GB_ADDR_CONFIG, 0x0000773f, 0x00004002),
160 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GB_ADDR_CONFIG_READ, 0x0000773f, 0x00004002),
161 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GFX_RB_RPTR_ADDR_LO, 0xfffffffd, 0x00000001),
162 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GFX_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
163 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_PAGE_RB_RPTR_ADDR_LO, 0xfffffffd, 0x00000001),
164 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_PAGE_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
165 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RD_BURST_CNTL, 0x0000000f, 0x00000003),
166 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC0_RB_RPTR_ADDR_LO, 0xfffffffd, 0x00000001),
167 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC0_RB_WPTR_POLL_CNTL, 0xfffffff0, 0x00403000),
168 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC1_RB_RPTR_ADDR_LO, 0xfffffffd, 0x00000001),
169 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC1_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
170 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC2_RB_RPTR_ADDR_LO, 0xfffffffd, 0x00000001),
171 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC2_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
172 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC3_RB_RPTR_ADDR_LO, 0xfffffffd, 0x00000001),
173 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC3_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
174 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC4_RB_RPTR_ADDR_LO, 0xfffffffd, 0x00000001),
175 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC4_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
176 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC5_RB_RPTR_ADDR_LO, 0xfffffffd, 0x00000001),
177 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC5_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
178 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC6_RB_RPTR_ADDR_LO, 0xfffffffd, 0x00000001),
179 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC6_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
180 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC7_RB_RPTR_ADDR_LO, 0xfffffffd, 0x00000001),
181 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC7_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
182 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_UTCL1_PAGE, 0x000003ff, 0x000003c0),
183 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_UTCL1_TIMEOUT, 0xffffffff, 0x00010001),
184 };
185 
186 static const struct soc15_reg_golden golden_settings_sdma1_4_2[] = {
187 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_CHICKEN_BITS, 0xfe931f07, 0x02831f07),
188 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_CLK_CTRL, 0xffffffff, 0x3f000100),
189 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_GB_ADDR_CONFIG, 0x0000773f, 0x00004002),
190 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_GB_ADDR_CONFIG_READ, 0x0000773f, 0x00004002),
191 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_GFX_RB_RPTR_ADDR_LO, 0xfffffffd, 0x00000001),
192 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_GFX_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
193 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_PAGE_RB_RPTR_ADDR_LO, 0xfffffffd, 0x00000001),
194 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_PAGE_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
195 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_RD_BURST_CNTL, 0x0000000f, 0x00000003),
196 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_RLC0_RB_RPTR_ADDR_LO, 0xfffffffd, 0x00000001),
197 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_RLC0_RB_WPTR_POLL_CNTL, 0xfffffff0, 0x00403000),
198 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_RLC1_RB_RPTR_ADDR_LO, 0xfffffffd, 0x00000001),
199 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_RLC1_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
200 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_RLC2_RB_RPTR_ADDR_LO, 0xfffffffd, 0x00000001),
201 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_RLC2_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
202 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_RLC3_RB_RPTR_ADDR_LO, 0xfffffffd, 0x00000001),
203 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_RLC3_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
204 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_RLC4_RB_RPTR_ADDR_LO, 0xfffffffd, 0x00000001),
205 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_RLC4_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
206 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_RLC5_RB_RPTR_ADDR_LO, 0xfffffffd, 0x00000001),
207 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_RLC5_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
208 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_RLC6_RB_RPTR_ADDR_LO, 0xfffffffd, 0x00000001),
209 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_RLC6_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
210 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_RLC7_RB_RPTR_ADDR_LO, 0xfffffffd, 0x00000001),
211 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_RLC7_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
212 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_UTCL1_PAGE, 0x000003ff, 0x000003c0),
213 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_UTCL1_TIMEOUT, 0xffffffff, 0x00010001),
214 };
215 
216 static const struct soc15_reg_golden golden_settings_sdma_rv1[] =
217 {
218 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GB_ADDR_CONFIG, 0x0018773f, 0x00000002),
219 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GB_ADDR_CONFIG_READ, 0x0018773f, 0x00000002)
220 };
221 
222 static const struct soc15_reg_golden golden_settings_sdma_rv2[] =
223 {
224 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GB_ADDR_CONFIG, 0x0018773f, 0x00003001),
225 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GB_ADDR_CONFIG_READ, 0x0018773f, 0x00003001)
226 };
227 
228 static const struct soc15_reg_golden golden_settings_sdma_arct[] =
229 {
230 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_CHICKEN_BITS, 0xfe931f07, 0x02831f07),
231 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GB_ADDR_CONFIG, 0x0000773f, 0x00004002),
232 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GB_ADDR_CONFIG_READ, 0x0000773f, 0x00004002),
233 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_UTCL1_TIMEOUT, 0xffffffff, 0x00010001),
234 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_CHICKEN_BITS, 0xfe931f07, 0x02831f07),
235 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_GB_ADDR_CONFIG, 0x0000773f, 0x00004002),
236 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_GB_ADDR_CONFIG_READ, 0x0000773f, 0x00004002),
237 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_UTCL1_TIMEOUT, 0xffffffff, 0x00010001),
238 	SOC15_REG_GOLDEN_VALUE(SDMA2, 0, mmSDMA2_CHICKEN_BITS, 0xfe931f07, 0x02831f07),
239 	SOC15_REG_GOLDEN_VALUE(SDMA2, 0, mmSDMA2_GB_ADDR_CONFIG, 0x0000773f, 0x00004002),
240 	SOC15_REG_GOLDEN_VALUE(SDMA2, 0, mmSDMA2_GB_ADDR_CONFIG_READ, 0x0000773f, 0x00004002),
241 	SOC15_REG_GOLDEN_VALUE(SDMA2, 0, mmSDMA2_UTCL1_TIMEOUT, 0xffffffff, 0x00010001),
242 	SOC15_REG_GOLDEN_VALUE(SDMA3, 0, mmSDMA3_CHICKEN_BITS, 0xfe931f07, 0x02831f07),
243 	SOC15_REG_GOLDEN_VALUE(SDMA3, 0, mmSDMA3_GB_ADDR_CONFIG, 0x0000773f, 0x00004002),
244 	SOC15_REG_GOLDEN_VALUE(SDMA3, 0, mmSDMA3_GB_ADDR_CONFIG_READ, 0x0000773f, 0x00004002),
245 	SOC15_REG_GOLDEN_VALUE(SDMA3, 0, mmSDMA3_UTCL1_TIMEOUT, 0xffffffff, 0x00010001),
246 	SOC15_REG_GOLDEN_VALUE(SDMA4, 0, mmSDMA4_CHICKEN_BITS, 0xfe931f07, 0x02831f07),
247 	SOC15_REG_GOLDEN_VALUE(SDMA4, 0, mmSDMA4_GB_ADDR_CONFIG, 0x0000773f, 0x00004002),
248 	SOC15_REG_GOLDEN_VALUE(SDMA4, 0, mmSDMA4_GB_ADDR_CONFIG_READ, 0x0000773f, 0x00004002),
249 	SOC15_REG_GOLDEN_VALUE(SDMA4, 0, mmSDMA4_UTCL1_TIMEOUT, 0xffffffff, 0x00010001),
250 	SOC15_REG_GOLDEN_VALUE(SDMA5, 0, mmSDMA5_CHICKEN_BITS, 0xfe931f07, 0x02831f07),
251 	SOC15_REG_GOLDEN_VALUE(SDMA5, 0, mmSDMA5_GB_ADDR_CONFIG, 0x0000773f, 0x00004002),
252 	SOC15_REG_GOLDEN_VALUE(SDMA5, 0, mmSDMA5_GB_ADDR_CONFIG_READ, 0x0000773f, 0x00004002),
253 	SOC15_REG_GOLDEN_VALUE(SDMA5, 0, mmSDMA5_UTCL1_TIMEOUT, 0xffffffff, 0x00010001),
254 	SOC15_REG_GOLDEN_VALUE(SDMA6, 0, mmSDMA6_CHICKEN_BITS, 0xfe931f07, 0x02831f07),
255 	SOC15_REG_GOLDEN_VALUE(SDMA6, 0, mmSDMA6_GB_ADDR_CONFIG, 0x0000773f, 0x00004002),
256 	SOC15_REG_GOLDEN_VALUE(SDMA6, 0, mmSDMA6_GB_ADDR_CONFIG_READ, 0x0000773f, 0x00004002),
257 	SOC15_REG_GOLDEN_VALUE(SDMA6, 0, mmSDMA6_UTCL1_TIMEOUT, 0xffffffff, 0x00010001),
258 	SOC15_REG_GOLDEN_VALUE(SDMA7, 0, mmSDMA7_CHICKEN_BITS, 0xfe931f07, 0x02831f07),
259 	SOC15_REG_GOLDEN_VALUE(SDMA7, 0, mmSDMA7_GB_ADDR_CONFIG, 0x0000773f, 0x00004002),
260 	SOC15_REG_GOLDEN_VALUE(SDMA7, 0, mmSDMA7_GB_ADDR_CONFIG_READ, 0x0000773f, 0x00004002),
261 	SOC15_REG_GOLDEN_VALUE(SDMA7, 0, mmSDMA7_UTCL1_TIMEOUT, 0xffffffff, 0x00010001)
262 };
263 
264 static const struct soc15_reg_golden golden_settings_sdma_aldebaran[] = {
265 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GB_ADDR_CONFIG, 0x0018773f, 0x00104002),
266 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GB_ADDR_CONFIG_READ, 0x0018773f, 0x00104002),
267 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_UTCL1_TIMEOUT, 0xffffffff, 0x00010001),
268 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_GB_ADDR_CONFIG, 0x0018773f, 0x00104002),
269 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_GB_ADDR_CONFIG_READ, 0x0018773f, 0x00104002),
270 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_UTCL1_TIMEOUT, 0xffffffff, 0x00010001),
271 	SOC15_REG_GOLDEN_VALUE(SDMA2, 0, mmSDMA2_GB_ADDR_CONFIG, 0x0018773f, 0x00104002),
272 	SOC15_REG_GOLDEN_VALUE(SDMA2, 0, mmSDMA2_GB_ADDR_CONFIG_READ, 0x0018773f, 0x00104002),
273 	SOC15_REG_GOLDEN_VALUE(SDMA3, 0, mmSDMA2_UTCL1_TIMEOUT, 0xffffffff, 0x00010001),
274 	SOC15_REG_GOLDEN_VALUE(SDMA3, 0, mmSDMA3_GB_ADDR_CONFIG, 0x0018773f, 0x00104002),
275 	SOC15_REG_GOLDEN_VALUE(SDMA3, 0, mmSDMA3_GB_ADDR_CONFIG_READ, 0x0018773f, 0x00104002),
276 	SOC15_REG_GOLDEN_VALUE(SDMA3, 0, mmSDMA3_UTCL1_TIMEOUT, 0xffffffff, 0x00010001),
277 	SOC15_REG_GOLDEN_VALUE(SDMA4, 0, mmSDMA4_GB_ADDR_CONFIG, 0x0018773f, 0x00104002),
278 	SOC15_REG_GOLDEN_VALUE(SDMA4, 0, mmSDMA4_GB_ADDR_CONFIG_READ, 0x0018773f, 0x00104002),
279 	SOC15_REG_GOLDEN_VALUE(SDMA4, 0, mmSDMA4_UTCL1_TIMEOUT, 0xffffffff, 0x00010001),
280 };
281 
282 static const struct soc15_reg_golden golden_settings_sdma_4_3[] = {
283 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_CHICKEN_BITS, 0xfe931f07, 0x02831f07),
284 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_CLK_CTRL, 0xffffffff, 0x3f000100),
285 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GB_ADDR_CONFIG, 0x0018773f, 0x00000002),
286 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GB_ADDR_CONFIG_READ, 0x0018773f, 0x00000002),
287 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GFX_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
288 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_POWER_CNTL, 0x003fff07, 0x40000051),
289 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC0_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
290 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC1_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
291 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_UTCL1_PAGE, 0x000003ff, 0x000003e0),
292 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_UTCL1_WATERMK, 0xfc000000, 0x03fbe1fe)
293 };
294 
295 static const struct soc15_ras_field_entry sdma_v4_0_ras_fields[] = {
296 	{ "SDMA_UCODE_BUF_SED", SOC15_REG_ENTRY(SDMA0, 0, mmSDMA0_EDC_COUNTER),
297 	SOC15_REG_FIELD(SDMA0_EDC_COUNTER, SDMA_UCODE_BUF_SED),
298 	0, 0,
299 	},
300 	{ "SDMA_RB_CMD_BUF_SED", SOC15_REG_ENTRY(SDMA0, 0, mmSDMA0_EDC_COUNTER),
301 	SOC15_REG_FIELD(SDMA0_EDC_COUNTER, SDMA_RB_CMD_BUF_SED),
302 	0, 0,
303 	},
304 	{ "SDMA_IB_CMD_BUF_SED", SOC15_REG_ENTRY(SDMA0, 0, mmSDMA0_EDC_COUNTER),
305 	SOC15_REG_FIELD(SDMA0_EDC_COUNTER, SDMA_IB_CMD_BUF_SED),
306 	0, 0,
307 	},
308 	{ "SDMA_UTCL1_RD_FIFO_SED", SOC15_REG_ENTRY(SDMA0, 0, mmSDMA0_EDC_COUNTER),
309 	SOC15_REG_FIELD(SDMA0_EDC_COUNTER, SDMA_UTCL1_RD_FIFO_SED),
310 	0, 0,
311 	},
312 	{ "SDMA_UTCL1_RDBST_FIFO_SED", SOC15_REG_ENTRY(SDMA0, 0, mmSDMA0_EDC_COUNTER),
313 	SOC15_REG_FIELD(SDMA0_EDC_COUNTER, SDMA_UTCL1_RDBST_FIFO_SED),
314 	0, 0,
315 	},
316 	{ "SDMA_DATA_LUT_FIFO_SED", SOC15_REG_ENTRY(SDMA0, 0, mmSDMA0_EDC_COUNTER),
317 	SOC15_REG_FIELD(SDMA0_EDC_COUNTER, SDMA_DATA_LUT_FIFO_SED),
318 	0, 0,
319 	},
320 	{ "SDMA_MBANK_DATA_BUF0_SED", SOC15_REG_ENTRY(SDMA0, 0, mmSDMA0_EDC_COUNTER),
321 	SOC15_REG_FIELD(SDMA0_EDC_COUNTER, SDMA_MBANK_DATA_BUF0_SED),
322 	0, 0,
323 	},
324 	{ "SDMA_MBANK_DATA_BUF1_SED", SOC15_REG_ENTRY(SDMA0, 0, mmSDMA0_EDC_COUNTER),
325 	SOC15_REG_FIELD(SDMA0_EDC_COUNTER, SDMA_MBANK_DATA_BUF1_SED),
326 	0, 0,
327 	},
328 	{ "SDMA_MBANK_DATA_BUF2_SED", SOC15_REG_ENTRY(SDMA0, 0, mmSDMA0_EDC_COUNTER),
329 	SOC15_REG_FIELD(SDMA0_EDC_COUNTER, SDMA_MBANK_DATA_BUF2_SED),
330 	0, 0,
331 	},
332 	{ "SDMA_MBANK_DATA_BUF3_SED", SOC15_REG_ENTRY(SDMA0, 0, mmSDMA0_EDC_COUNTER),
333 	SOC15_REG_FIELD(SDMA0_EDC_COUNTER, SDMA_MBANK_DATA_BUF3_SED),
334 	0, 0,
335 	},
336 	{ "SDMA_MBANK_DATA_BUF4_SED", SOC15_REG_ENTRY(SDMA0, 0, mmSDMA0_EDC_COUNTER),
337 	SOC15_REG_FIELD(SDMA0_EDC_COUNTER, SDMA_MBANK_DATA_BUF4_SED),
338 	0, 0,
339 	},
340 	{ "SDMA_MBANK_DATA_BUF5_SED", SOC15_REG_ENTRY(SDMA0, 0, mmSDMA0_EDC_COUNTER),
341 	SOC15_REG_FIELD(SDMA0_EDC_COUNTER, SDMA_MBANK_DATA_BUF5_SED),
342 	0, 0,
343 	},
344 	{ "SDMA_MBANK_DATA_BUF6_SED", SOC15_REG_ENTRY(SDMA0, 0, mmSDMA0_EDC_COUNTER),
345 	SOC15_REG_FIELD(SDMA0_EDC_COUNTER, SDMA_MBANK_DATA_BUF6_SED),
346 	0, 0,
347 	},
348 	{ "SDMA_MBANK_DATA_BUF7_SED", SOC15_REG_ENTRY(SDMA0, 0, mmSDMA0_EDC_COUNTER),
349 	SOC15_REG_FIELD(SDMA0_EDC_COUNTER, SDMA_MBANK_DATA_BUF7_SED),
350 	0, 0,
351 	},
352 	{ "SDMA_MBANK_DATA_BUF8_SED", SOC15_REG_ENTRY(SDMA0, 0, mmSDMA0_EDC_COUNTER),
353 	SOC15_REG_FIELD(SDMA0_EDC_COUNTER, SDMA_MBANK_DATA_BUF8_SED),
354 	0, 0,
355 	},
356 	{ "SDMA_MBANK_DATA_BUF9_SED", SOC15_REG_ENTRY(SDMA0, 0, mmSDMA0_EDC_COUNTER),
357 	SOC15_REG_FIELD(SDMA0_EDC_COUNTER, SDMA_MBANK_DATA_BUF9_SED),
358 	0, 0,
359 	},
360 	{ "SDMA_MBANK_DATA_BUF10_SED", SOC15_REG_ENTRY(SDMA0, 0, mmSDMA0_EDC_COUNTER),
361 	SOC15_REG_FIELD(SDMA0_EDC_COUNTER, SDMA_MBANK_DATA_BUF10_SED),
362 	0, 0,
363 	},
364 	{ "SDMA_MBANK_DATA_BUF11_SED", SOC15_REG_ENTRY(SDMA0, 0, mmSDMA0_EDC_COUNTER),
365 	SOC15_REG_FIELD(SDMA0_EDC_COUNTER, SDMA_MBANK_DATA_BUF11_SED),
366 	0, 0,
367 	},
368 	{ "SDMA_MBANK_DATA_BUF12_SED", SOC15_REG_ENTRY(SDMA0, 0, mmSDMA0_EDC_COUNTER),
369 	SOC15_REG_FIELD(SDMA0_EDC_COUNTER, SDMA_MBANK_DATA_BUF12_SED),
370 	0, 0,
371 	},
372 	{ "SDMA_MBANK_DATA_BUF13_SED", SOC15_REG_ENTRY(SDMA0, 0, mmSDMA0_EDC_COUNTER),
373 	SOC15_REG_FIELD(SDMA0_EDC_COUNTER, SDMA_MBANK_DATA_BUF13_SED),
374 	0, 0,
375 	},
376 	{ "SDMA_MBANK_DATA_BUF14_SED", SOC15_REG_ENTRY(SDMA0, 0, mmSDMA0_EDC_COUNTER),
377 	SOC15_REG_FIELD(SDMA0_EDC_COUNTER, SDMA_MBANK_DATA_BUF14_SED),
378 	0, 0,
379 	},
380 	{ "SDMA_MBANK_DATA_BUF15_SED", SOC15_REG_ENTRY(SDMA0, 0, mmSDMA0_EDC_COUNTER),
381 	SOC15_REG_FIELD(SDMA0_EDC_COUNTER, SDMA_MBANK_DATA_BUF15_SED),
382 	0, 0,
383 	},
384 	{ "SDMA_SPLIT_DAT_BUF_SED", SOC15_REG_ENTRY(SDMA0, 0, mmSDMA0_EDC_COUNTER),
385 	SOC15_REG_FIELD(SDMA0_EDC_COUNTER, SDMA_SPLIT_DAT_BUF_SED),
386 	0, 0,
387 	},
388 	{ "SDMA_MC_WR_ADDR_FIFO_SED", SOC15_REG_ENTRY(SDMA0, 0, mmSDMA0_EDC_COUNTER),
389 	SOC15_REG_FIELD(SDMA0_EDC_COUNTER, SDMA_MC_WR_ADDR_FIFO_SED),
390 	0, 0,
391 	},
392 };
393 
394 static u32 sdma_v4_0_get_reg_offset(struct amdgpu_device *adev,
395 		u32 instance, u32 offset)
396 {
397 	switch (instance) {
398 	case 0:
399 		return (adev->reg_offset[SDMA0_HWIP][0][0] + offset);
400 	case 1:
401 		return (adev->reg_offset[SDMA1_HWIP][0][0] + offset);
402 	case 2:
403 		return (adev->reg_offset[SDMA2_HWIP][0][1] + offset);
404 	case 3:
405 		return (adev->reg_offset[SDMA3_HWIP][0][1] + offset);
406 	case 4:
407 		return (adev->reg_offset[SDMA4_HWIP][0][1] + offset);
408 	case 5:
409 		return (adev->reg_offset[SDMA5_HWIP][0][1] + offset);
410 	case 6:
411 		return (adev->reg_offset[SDMA6_HWIP][0][1] + offset);
412 	case 7:
413 		return (adev->reg_offset[SDMA7_HWIP][0][1] + offset);
414 	default:
415 		break;
416 	}
417 	return 0;
418 }
419 
420 static unsigned sdma_v4_0_seq_to_irq_id(int seq_num)
421 {
422 	switch (seq_num) {
423 	case 0:
424 		return SOC15_IH_CLIENTID_SDMA0;
425 	case 1:
426 		return SOC15_IH_CLIENTID_SDMA1;
427 	case 2:
428 		return SOC15_IH_CLIENTID_SDMA2;
429 	case 3:
430 		return SOC15_IH_CLIENTID_SDMA3;
431 	case 4:
432 		return SOC15_IH_CLIENTID_SDMA4;
433 	case 5:
434 		return SOC15_IH_CLIENTID_SDMA5;
435 	case 6:
436 		return SOC15_IH_CLIENTID_SDMA6;
437 	case 7:
438 		return SOC15_IH_CLIENTID_SDMA7;
439 	default:
440 		break;
441 	}
442 	return -EINVAL;
443 }
444 
445 static int sdma_v4_0_irq_id_to_seq(unsigned client_id)
446 {
447 	switch (client_id) {
448 	case SOC15_IH_CLIENTID_SDMA0:
449 		return 0;
450 	case SOC15_IH_CLIENTID_SDMA1:
451 		return 1;
452 	case SOC15_IH_CLIENTID_SDMA2:
453 		return 2;
454 	case SOC15_IH_CLIENTID_SDMA3:
455 		return 3;
456 	case SOC15_IH_CLIENTID_SDMA4:
457 		return 4;
458 	case SOC15_IH_CLIENTID_SDMA5:
459 		return 5;
460 	case SOC15_IH_CLIENTID_SDMA6:
461 		return 6;
462 	case SOC15_IH_CLIENTID_SDMA7:
463 		return 7;
464 	default:
465 		break;
466 	}
467 	return -EINVAL;
468 }
469 
470 static void sdma_v4_0_init_golden_registers(struct amdgpu_device *adev)
471 {
472 	switch (adev->ip_versions[SDMA0_HWIP][0]) {
473 	case IP_VERSION(4, 0, 0):
474 		soc15_program_register_sequence(adev,
475 						golden_settings_sdma_4,
476 						ARRAY_SIZE(golden_settings_sdma_4));
477 		soc15_program_register_sequence(adev,
478 						golden_settings_sdma_vg10,
479 						ARRAY_SIZE(golden_settings_sdma_vg10));
480 		break;
481 	case IP_VERSION(4, 0, 1):
482 		soc15_program_register_sequence(adev,
483 						golden_settings_sdma_4,
484 						ARRAY_SIZE(golden_settings_sdma_4));
485 		soc15_program_register_sequence(adev,
486 						golden_settings_sdma_vg12,
487 						ARRAY_SIZE(golden_settings_sdma_vg12));
488 		break;
489 	case IP_VERSION(4, 2, 0):
490 		soc15_program_register_sequence(adev,
491 						golden_settings_sdma0_4_2_init,
492 						ARRAY_SIZE(golden_settings_sdma0_4_2_init));
493 		soc15_program_register_sequence(adev,
494 						golden_settings_sdma0_4_2,
495 						ARRAY_SIZE(golden_settings_sdma0_4_2));
496 		soc15_program_register_sequence(adev,
497 						golden_settings_sdma1_4_2,
498 						ARRAY_SIZE(golden_settings_sdma1_4_2));
499 		break;
500 	case IP_VERSION(4, 2, 2):
501 		soc15_program_register_sequence(adev,
502 						golden_settings_sdma_arct,
503 						ARRAY_SIZE(golden_settings_sdma_arct));
504 		break;
505 	case IP_VERSION(4, 4, 0):
506 		soc15_program_register_sequence(adev,
507 						golden_settings_sdma_aldebaran,
508 						ARRAY_SIZE(golden_settings_sdma_aldebaran));
509 		break;
510 	case IP_VERSION(4, 1, 0):
511 	case IP_VERSION(4, 1, 1):
512 		soc15_program_register_sequence(adev,
513 						golden_settings_sdma_4_1,
514 						ARRAY_SIZE(golden_settings_sdma_4_1));
515 		if (adev->apu_flags & AMD_APU_IS_RAVEN2)
516 			soc15_program_register_sequence(adev,
517 							golden_settings_sdma_rv2,
518 							ARRAY_SIZE(golden_settings_sdma_rv2));
519 		else
520 			soc15_program_register_sequence(adev,
521 							golden_settings_sdma_rv1,
522 							ARRAY_SIZE(golden_settings_sdma_rv1));
523 		break;
524 	case IP_VERSION(4, 1, 2):
525 		soc15_program_register_sequence(adev,
526 						golden_settings_sdma_4_3,
527 						ARRAY_SIZE(golden_settings_sdma_4_3));
528 		break;
529 	default:
530 		break;
531 	}
532 }
533 
534 static void sdma_v4_0_setup_ulv(struct amdgpu_device *adev)
535 {
536 	int i;
537 
538 	/*
539 	 * The only chips with SDMAv4 and ULV are VG10 and VG20.
540 	 * Server SKUs take a different hysteresis setting from other SKUs.
541 	 */
542 	switch (adev->ip_versions[SDMA0_HWIP][0]) {
543 	case IP_VERSION(4, 0, 0):
544 		if (adev->pdev->device == 0x6860)
545 			break;
546 		return;
547 	case IP_VERSION(4, 2, 0):
548 		if (adev->pdev->device == 0x66a1)
549 			break;
550 		return;
551 	default:
552 		return;
553 	}
554 
555 	for (i = 0; i < adev->sdma.num_instances; i++) {
556 		uint32_t temp;
557 
558 		temp = RREG32_SDMA(i, mmSDMA0_ULV_CNTL);
559 		temp = REG_SET_FIELD(temp, SDMA0_ULV_CNTL, HYSTERESIS, 0x0);
560 		WREG32_SDMA(i, mmSDMA0_ULV_CNTL, temp);
561 	}
562 }
563 
564 static int sdma_v4_0_init_inst_ctx(struct amdgpu_sdma_instance *sdma_inst)
565 {
566 	int err = 0;
567 	const struct sdma_firmware_header_v1_0 *hdr;
568 
569 	err = amdgpu_ucode_validate(sdma_inst->fw);
570 	if (err)
571 		return err;
572 
573 	hdr = (const struct sdma_firmware_header_v1_0 *)sdma_inst->fw->data;
574 	sdma_inst->fw_version = le32_to_cpu(hdr->header.ucode_version);
575 	sdma_inst->feature_version = le32_to_cpu(hdr->ucode_feature_version);
576 
577 	if (sdma_inst->feature_version >= 20)
578 		sdma_inst->burst_nop = true;
579 
580 	return 0;
581 }
582 
583 static void sdma_v4_0_destroy_inst_ctx(struct amdgpu_device *adev)
584 {
585 	int i;
586 
587 	for (i = 0; i < adev->sdma.num_instances; i++) {
588 		release_firmware(adev->sdma.instance[i].fw);
589 		adev->sdma.instance[i].fw = NULL;
590 
591 		/* arcturus shares the same FW memory across
592 		   all SDMA isntances */
593 		if (adev->ip_versions[SDMA0_HWIP][0] == IP_VERSION(4, 2, 2) ||
594 		    adev->ip_versions[SDMA0_HWIP][0] == IP_VERSION(4, 4, 0))
595 			break;
596 	}
597 
598 	memset((void *)adev->sdma.instance, 0,
599 		sizeof(struct amdgpu_sdma_instance) * AMDGPU_MAX_SDMA_INSTANCES);
600 }
601 
602 /**
603  * sdma_v4_0_init_microcode - load ucode images from disk
604  *
605  * @adev: amdgpu_device pointer
606  *
607  * Use the firmware interface to load the ucode images into
608  * the driver (not loaded into hw).
609  * Returns 0 on success, error on failure.
610  */
611 
612 // emulation only, won't work on real chip
613 // vega10 real chip need to use PSP to load firmware
614 static int sdma_v4_0_init_microcode(struct amdgpu_device *adev)
615 {
616 	const char *chip_name;
617 	char fw_name[30];
618 	int err = 0, i;
619 	struct amdgpu_firmware_info *info = NULL;
620 	const struct common_firmware_header *header = NULL;
621 
622 	DRM_DEBUG("\n");
623 
624 	switch (adev->ip_versions[SDMA0_HWIP][0]) {
625 	case IP_VERSION(4, 0, 0):
626 		chip_name = "vega10";
627 		break;
628 	case IP_VERSION(4, 0, 1):
629 		chip_name = "vega12";
630 		break;
631 	case IP_VERSION(4, 2, 0):
632 		chip_name = "vega20";
633 		break;
634 	case IP_VERSION(4, 1, 0):
635 	case IP_VERSION(4, 1, 1):
636 		if (adev->apu_flags & AMD_APU_IS_RAVEN2)
637 			chip_name = "raven2";
638 		else if (adev->apu_flags & AMD_APU_IS_PICASSO)
639 			chip_name = "picasso";
640 		else
641 			chip_name = "raven";
642 		break;
643 	case IP_VERSION(4, 2, 2):
644 		chip_name = "arcturus";
645 		break;
646 	case IP_VERSION(4, 1, 2):
647 		if (adev->apu_flags & AMD_APU_IS_RENOIR)
648 			chip_name = "renoir";
649 		else
650 			chip_name = "green_sardine";
651 		break;
652 	case IP_VERSION(4, 4, 0):
653 		chip_name = "aldebaran";
654 		break;
655 	default:
656 		BUG();
657 	}
658 
659 	snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_sdma.bin", chip_name);
660 
661 	err = request_firmware(&adev->sdma.instance[0].fw, fw_name, adev->dev);
662 	if (err)
663 		goto out;
664 
665 	err = sdma_v4_0_init_inst_ctx(&adev->sdma.instance[0]);
666 	if (err)
667 		goto out;
668 
669 	for (i = 1; i < adev->sdma.num_instances; i++) {
670 		if (adev->ip_versions[SDMA0_HWIP][0] == IP_VERSION(4, 2, 2) ||
671                     adev->ip_versions[SDMA0_HWIP][0] == IP_VERSION(4, 4, 0)) {
672 			/* Acturus & Aldebaran will leverage the same FW memory
673 			   for every SDMA instance */
674 			memcpy((void *)&adev->sdma.instance[i],
675 			       (void *)&adev->sdma.instance[0],
676 			       sizeof(struct amdgpu_sdma_instance));
677 		}
678 		else {
679 			snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_sdma%d.bin", chip_name, i);
680 
681 			err = request_firmware(&adev->sdma.instance[i].fw, fw_name, adev->dev);
682 			if (err)
683 				goto out;
684 
685 			err = sdma_v4_0_init_inst_ctx(&adev->sdma.instance[i]);
686 			if (err)
687 				goto out;
688 		}
689 	}
690 
691 	DRM_DEBUG("psp_load == '%s'\n",
692 		adev->firmware.load_type == AMDGPU_FW_LOAD_PSP ? "true" : "false");
693 
694 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
695 		for (i = 0; i < adev->sdma.num_instances; i++) {
696 			info = &adev->firmware.ucode[AMDGPU_UCODE_ID_SDMA0 + i];
697 			info->ucode_id = AMDGPU_UCODE_ID_SDMA0 + i;
698 			info->fw = adev->sdma.instance[i].fw;
699 			header = (const struct common_firmware_header *)info->fw->data;
700 			adev->firmware.fw_size +=
701 				ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
702 		}
703 	}
704 
705 out:
706 	if (err) {
707 		DRM_ERROR("sdma_v4_0: Failed to load firmware \"%s\"\n", fw_name);
708 		sdma_v4_0_destroy_inst_ctx(adev);
709 	}
710 	return err;
711 }
712 
713 /**
714  * sdma_v4_0_ring_get_rptr - get the current read pointer
715  *
716  * @ring: amdgpu ring pointer
717  *
718  * Get the current rptr from the hardware (VEGA10+).
719  */
720 static uint64_t sdma_v4_0_ring_get_rptr(struct amdgpu_ring *ring)
721 {
722 	u64 *rptr;
723 
724 	/* XXX check if swapping is necessary on BE */
725 	rptr = ((u64 *)ring->rptr_cpu_addr);
726 
727 	DRM_DEBUG("rptr before shift == 0x%016llx\n", *rptr);
728 	return ((*rptr) >> 2);
729 }
730 
731 /**
732  * sdma_v4_0_ring_get_wptr - get the current write pointer
733  *
734  * @ring: amdgpu ring pointer
735  *
736  * Get the current wptr from the hardware (VEGA10+).
737  */
738 static uint64_t sdma_v4_0_ring_get_wptr(struct amdgpu_ring *ring)
739 {
740 	struct amdgpu_device *adev = ring->adev;
741 	u64 wptr;
742 
743 	if (ring->use_doorbell) {
744 		/* XXX check if swapping is necessary on BE */
745 		wptr = READ_ONCE(*((u64 *)ring->wptr_cpu_addr));
746 		DRM_DEBUG("wptr/doorbell before shift == 0x%016llx\n", wptr);
747 	} else {
748 		wptr = RREG32_SDMA(ring->me, mmSDMA0_GFX_RB_WPTR_HI);
749 		wptr = wptr << 32;
750 		wptr |= RREG32_SDMA(ring->me, mmSDMA0_GFX_RB_WPTR);
751 		DRM_DEBUG("wptr before shift [%i] wptr == 0x%016llx\n",
752 				ring->me, wptr);
753 	}
754 
755 	return wptr >> 2;
756 }
757 
758 /**
759  * sdma_v4_0_ring_set_wptr - commit the write pointer
760  *
761  * @ring: amdgpu ring pointer
762  *
763  * Write the wptr back to the hardware (VEGA10+).
764  */
765 static void sdma_v4_0_ring_set_wptr(struct amdgpu_ring *ring)
766 {
767 	struct amdgpu_device *adev = ring->adev;
768 
769 	DRM_DEBUG("Setting write pointer\n");
770 	if (ring->use_doorbell) {
771 		u64 *wb = (u64 *)ring->wptr_cpu_addr;
772 
773 		DRM_DEBUG("Using doorbell -- "
774 				"wptr_offs == 0x%08x "
775 				"lower_32_bits(ring->wptr << 2) == 0x%08x "
776 				"upper_32_bits(ring->wptr << 2) == 0x%08x\n",
777 				ring->wptr_offs,
778 				lower_32_bits(ring->wptr << 2),
779 				upper_32_bits(ring->wptr << 2));
780 		/* XXX check if swapping is necessary on BE */
781 		WRITE_ONCE(*wb, (ring->wptr << 2));
782 		DRM_DEBUG("calling WDOORBELL64(0x%08x, 0x%016llx)\n",
783 				ring->doorbell_index, ring->wptr << 2);
784 		WDOORBELL64(ring->doorbell_index, ring->wptr << 2);
785 	} else {
786 		DRM_DEBUG("Not using doorbell -- "
787 				"mmSDMA%i_GFX_RB_WPTR == 0x%08x "
788 				"mmSDMA%i_GFX_RB_WPTR_HI == 0x%08x\n",
789 				ring->me,
790 				lower_32_bits(ring->wptr << 2),
791 				ring->me,
792 				upper_32_bits(ring->wptr << 2));
793 		WREG32_SDMA(ring->me, mmSDMA0_GFX_RB_WPTR,
794 			    lower_32_bits(ring->wptr << 2));
795 		WREG32_SDMA(ring->me, mmSDMA0_GFX_RB_WPTR_HI,
796 			    upper_32_bits(ring->wptr << 2));
797 	}
798 }
799 
800 /**
801  * sdma_v4_0_page_ring_get_wptr - get the current write pointer
802  *
803  * @ring: amdgpu ring pointer
804  *
805  * Get the current wptr from the hardware (VEGA10+).
806  */
807 static uint64_t sdma_v4_0_page_ring_get_wptr(struct amdgpu_ring *ring)
808 {
809 	struct amdgpu_device *adev = ring->adev;
810 	u64 wptr;
811 
812 	if (ring->use_doorbell) {
813 		/* XXX check if swapping is necessary on BE */
814 		wptr = READ_ONCE(*((u64 *)ring->wptr_cpu_addr));
815 	} else {
816 		wptr = RREG32_SDMA(ring->me, mmSDMA0_PAGE_RB_WPTR_HI);
817 		wptr = wptr << 32;
818 		wptr |= RREG32_SDMA(ring->me, mmSDMA0_PAGE_RB_WPTR);
819 	}
820 
821 	return wptr >> 2;
822 }
823 
824 /**
825  * sdma_v4_0_page_ring_set_wptr - commit the write pointer
826  *
827  * @ring: amdgpu ring pointer
828  *
829  * Write the wptr back to the hardware (VEGA10+).
830  */
831 static void sdma_v4_0_page_ring_set_wptr(struct amdgpu_ring *ring)
832 {
833 	struct amdgpu_device *adev = ring->adev;
834 
835 	if (ring->use_doorbell) {
836 		u64 *wb = (u64 *)ring->wptr_cpu_addr;
837 
838 		/* XXX check if swapping is necessary on BE */
839 		WRITE_ONCE(*wb, (ring->wptr << 2));
840 		WDOORBELL64(ring->doorbell_index, ring->wptr << 2);
841 	} else {
842 		uint64_t wptr = ring->wptr << 2;
843 
844 		WREG32_SDMA(ring->me, mmSDMA0_PAGE_RB_WPTR,
845 			    lower_32_bits(wptr));
846 		WREG32_SDMA(ring->me, mmSDMA0_PAGE_RB_WPTR_HI,
847 			    upper_32_bits(wptr));
848 	}
849 }
850 
851 static void sdma_v4_0_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count)
852 {
853 	struct amdgpu_sdma_instance *sdma = amdgpu_sdma_get_instance_from_ring(ring);
854 	int i;
855 
856 	for (i = 0; i < count; i++)
857 		if (sdma && sdma->burst_nop && (i == 0))
858 			amdgpu_ring_write(ring, ring->funcs->nop |
859 				SDMA_PKT_NOP_HEADER_COUNT(count - 1));
860 		else
861 			amdgpu_ring_write(ring, ring->funcs->nop);
862 }
863 
864 /**
865  * sdma_v4_0_ring_emit_ib - Schedule an IB on the DMA engine
866  *
867  * @ring: amdgpu ring pointer
868  * @job: job to retrieve vmid from
869  * @ib: IB object to schedule
870  * @flags: unused
871  *
872  * Schedule an IB in the DMA ring (VEGA10).
873  */
874 static void sdma_v4_0_ring_emit_ib(struct amdgpu_ring *ring,
875 				   struct amdgpu_job *job,
876 				   struct amdgpu_ib *ib,
877 				   uint32_t flags)
878 {
879 	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
880 
881 	/* IB packet must end on a 8 DW boundary */
882 	sdma_v4_0_ring_insert_nop(ring, (2 - lower_32_bits(ring->wptr)) & 7);
883 
884 	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_INDIRECT) |
885 			  SDMA_PKT_INDIRECT_HEADER_VMID(vmid & 0xf));
886 	/* base must be 32 byte aligned */
887 	amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr) & 0xffffffe0);
888 	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr));
889 	amdgpu_ring_write(ring, ib->length_dw);
890 	amdgpu_ring_write(ring, 0);
891 	amdgpu_ring_write(ring, 0);
892 
893 }
894 
895 static void sdma_v4_0_wait_reg_mem(struct amdgpu_ring *ring,
896 				   int mem_space, int hdp,
897 				   uint32_t addr0, uint32_t addr1,
898 				   uint32_t ref, uint32_t mask,
899 				   uint32_t inv)
900 {
901 	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_POLL_REGMEM) |
902 			  SDMA_PKT_POLL_REGMEM_HEADER_HDP_FLUSH(hdp) |
903 			  SDMA_PKT_POLL_REGMEM_HEADER_MEM_POLL(mem_space) |
904 			  SDMA_PKT_POLL_REGMEM_HEADER_FUNC(3)); /* == */
905 	if (mem_space) {
906 		/* memory */
907 		amdgpu_ring_write(ring, addr0);
908 		amdgpu_ring_write(ring, addr1);
909 	} else {
910 		/* registers */
911 		amdgpu_ring_write(ring, addr0 << 2);
912 		amdgpu_ring_write(ring, addr1 << 2);
913 	}
914 	amdgpu_ring_write(ring, ref); /* reference */
915 	amdgpu_ring_write(ring, mask); /* mask */
916 	amdgpu_ring_write(ring, SDMA_PKT_POLL_REGMEM_DW5_RETRY_COUNT(0xfff) |
917 			  SDMA_PKT_POLL_REGMEM_DW5_INTERVAL(inv)); /* retry count, poll interval */
918 }
919 
920 /**
921  * sdma_v4_0_ring_emit_hdp_flush - emit an hdp flush on the DMA ring
922  *
923  * @ring: amdgpu ring pointer
924  *
925  * Emit an hdp flush packet on the requested DMA ring.
926  */
927 static void sdma_v4_0_ring_emit_hdp_flush(struct amdgpu_ring *ring)
928 {
929 	struct amdgpu_device *adev = ring->adev;
930 	u32 ref_and_mask = 0;
931 	const struct nbio_hdp_flush_reg *nbio_hf_reg = adev->nbio.hdp_flush_reg;
932 
933 	ref_and_mask = nbio_hf_reg->ref_and_mask_sdma0 << ring->me;
934 
935 	sdma_v4_0_wait_reg_mem(ring, 0, 1,
936 			       adev->nbio.funcs->get_hdp_flush_done_offset(adev),
937 			       adev->nbio.funcs->get_hdp_flush_req_offset(adev),
938 			       ref_and_mask, ref_and_mask, 10);
939 }
940 
941 /**
942  * sdma_v4_0_ring_emit_fence - emit a fence on the DMA ring
943  *
944  * @ring: amdgpu ring pointer
945  * @addr: address
946  * @seq: sequence number
947  * @flags: fence related flags
948  *
949  * Add a DMA fence packet to the ring to write
950  * the fence seq number and DMA trap packet to generate
951  * an interrupt if needed (VEGA10).
952  */
953 static void sdma_v4_0_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, u64 seq,
954 				      unsigned flags)
955 {
956 	bool write64bit = flags & AMDGPU_FENCE_FLAG_64BIT;
957 	/* write the fence */
958 	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_FENCE));
959 	/* zero in first two bits */
960 	BUG_ON(addr & 0x3);
961 	amdgpu_ring_write(ring, lower_32_bits(addr));
962 	amdgpu_ring_write(ring, upper_32_bits(addr));
963 	amdgpu_ring_write(ring, lower_32_bits(seq));
964 
965 	/* optionally write high bits as well */
966 	if (write64bit) {
967 		addr += 4;
968 		amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_FENCE));
969 		/* zero in first two bits */
970 		BUG_ON(addr & 0x3);
971 		amdgpu_ring_write(ring, lower_32_bits(addr));
972 		amdgpu_ring_write(ring, upper_32_bits(addr));
973 		amdgpu_ring_write(ring, upper_32_bits(seq));
974 	}
975 
976 	/* generate an interrupt */
977 	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_TRAP));
978 	amdgpu_ring_write(ring, SDMA_PKT_TRAP_INT_CONTEXT_INT_CONTEXT(0));
979 }
980 
981 
982 /**
983  * sdma_v4_0_gfx_stop - stop the gfx async dma engines
984  *
985  * @adev: amdgpu_device pointer
986  *
987  * Stop the gfx async dma ring buffers (VEGA10).
988  */
989 static void sdma_v4_0_gfx_stop(struct amdgpu_device *adev)
990 {
991 	struct amdgpu_ring *sdma[AMDGPU_MAX_SDMA_INSTANCES];
992 	u32 rb_cntl, ib_cntl;
993 	int i, unset = 0;
994 
995 	for (i = 0; i < adev->sdma.num_instances; i++) {
996 		sdma[i] = &adev->sdma.instance[i].ring;
997 
998 		if ((adev->mman.buffer_funcs_ring == sdma[i]) && unset != 1) {
999 			amdgpu_ttm_set_buffer_funcs_status(adev, false);
1000 			unset = 1;
1001 		}
1002 
1003 		rb_cntl = RREG32_SDMA(i, mmSDMA0_GFX_RB_CNTL);
1004 		rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_GFX_RB_CNTL, RB_ENABLE, 0);
1005 		WREG32_SDMA(i, mmSDMA0_GFX_RB_CNTL, rb_cntl);
1006 		ib_cntl = RREG32_SDMA(i, mmSDMA0_GFX_IB_CNTL);
1007 		ib_cntl = REG_SET_FIELD(ib_cntl, SDMA0_GFX_IB_CNTL, IB_ENABLE, 0);
1008 		WREG32_SDMA(i, mmSDMA0_GFX_IB_CNTL, ib_cntl);
1009 	}
1010 }
1011 
1012 /**
1013  * sdma_v4_0_rlc_stop - stop the compute async dma engines
1014  *
1015  * @adev: amdgpu_device pointer
1016  *
1017  * Stop the compute async dma queues (VEGA10).
1018  */
1019 static void sdma_v4_0_rlc_stop(struct amdgpu_device *adev)
1020 {
1021 	/* XXX todo */
1022 }
1023 
1024 /**
1025  * sdma_v4_0_page_stop - stop the page async dma engines
1026  *
1027  * @adev: amdgpu_device pointer
1028  *
1029  * Stop the page async dma ring buffers (VEGA10).
1030  */
1031 static void sdma_v4_0_page_stop(struct amdgpu_device *adev)
1032 {
1033 	struct amdgpu_ring *sdma[AMDGPU_MAX_SDMA_INSTANCES];
1034 	u32 rb_cntl, ib_cntl;
1035 	int i;
1036 	bool unset = false;
1037 
1038 	for (i = 0; i < adev->sdma.num_instances; i++) {
1039 		sdma[i] = &adev->sdma.instance[i].page;
1040 
1041 		if ((adev->mman.buffer_funcs_ring == sdma[i]) &&
1042 			(!unset)) {
1043 			amdgpu_ttm_set_buffer_funcs_status(adev, false);
1044 			unset = true;
1045 		}
1046 
1047 		rb_cntl = RREG32_SDMA(i, mmSDMA0_PAGE_RB_CNTL);
1048 		rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_PAGE_RB_CNTL,
1049 					RB_ENABLE, 0);
1050 		WREG32_SDMA(i, mmSDMA0_PAGE_RB_CNTL, rb_cntl);
1051 		ib_cntl = RREG32_SDMA(i, mmSDMA0_PAGE_IB_CNTL);
1052 		ib_cntl = REG_SET_FIELD(ib_cntl, SDMA0_PAGE_IB_CNTL,
1053 					IB_ENABLE, 0);
1054 		WREG32_SDMA(i, mmSDMA0_PAGE_IB_CNTL, ib_cntl);
1055 	}
1056 }
1057 
1058 /**
1059  * sdma_v4_0_ctx_switch_enable - stop the async dma engines context switch
1060  *
1061  * @adev: amdgpu_device pointer
1062  * @enable: enable/disable the DMA MEs context switch.
1063  *
1064  * Halt or unhalt the async dma engines context switch (VEGA10).
1065  */
1066 static void sdma_v4_0_ctx_switch_enable(struct amdgpu_device *adev, bool enable)
1067 {
1068 	u32 f32_cntl, phase_quantum = 0;
1069 	int i;
1070 
1071 	if (amdgpu_sdma_phase_quantum) {
1072 		unsigned value = amdgpu_sdma_phase_quantum;
1073 		unsigned unit = 0;
1074 
1075 		while (value > (SDMA0_PHASE0_QUANTUM__VALUE_MASK >>
1076 				SDMA0_PHASE0_QUANTUM__VALUE__SHIFT)) {
1077 			value = (value + 1) >> 1;
1078 			unit++;
1079 		}
1080 		if (unit > (SDMA0_PHASE0_QUANTUM__UNIT_MASK >>
1081 			    SDMA0_PHASE0_QUANTUM__UNIT__SHIFT)) {
1082 			value = (SDMA0_PHASE0_QUANTUM__VALUE_MASK >>
1083 				 SDMA0_PHASE0_QUANTUM__VALUE__SHIFT);
1084 			unit = (SDMA0_PHASE0_QUANTUM__UNIT_MASK >>
1085 				SDMA0_PHASE0_QUANTUM__UNIT__SHIFT);
1086 			WARN_ONCE(1,
1087 			"clamping sdma_phase_quantum to %uK clock cycles\n",
1088 				  value << unit);
1089 		}
1090 		phase_quantum =
1091 			value << SDMA0_PHASE0_QUANTUM__VALUE__SHIFT |
1092 			unit  << SDMA0_PHASE0_QUANTUM__UNIT__SHIFT;
1093 	}
1094 
1095 	for (i = 0; i < adev->sdma.num_instances; i++) {
1096 		f32_cntl = RREG32_SDMA(i, mmSDMA0_CNTL);
1097 		f32_cntl = REG_SET_FIELD(f32_cntl, SDMA0_CNTL,
1098 				AUTO_CTXSW_ENABLE, enable ? 1 : 0);
1099 		if (enable && amdgpu_sdma_phase_quantum) {
1100 			WREG32_SDMA(i, mmSDMA0_PHASE0_QUANTUM, phase_quantum);
1101 			WREG32_SDMA(i, mmSDMA0_PHASE1_QUANTUM, phase_quantum);
1102 			WREG32_SDMA(i, mmSDMA0_PHASE2_QUANTUM, phase_quantum);
1103 		}
1104 		WREG32_SDMA(i, mmSDMA0_CNTL, f32_cntl);
1105 
1106 		/*
1107 		 * Enable SDMA utilization. Its only supported on
1108 		 * Arcturus for the moment and firmware version 14
1109 		 * and above.
1110 		 */
1111 		if (adev->ip_versions[SDMA0_HWIP][0] == IP_VERSION(4, 2, 2) &&
1112 		    adev->sdma.instance[i].fw_version >= 14)
1113 			WREG32_SDMA(i, mmSDMA0_PUB_DUMMY_REG2, enable);
1114 		/* Extend page fault timeout to avoid interrupt storm */
1115 		WREG32_SDMA(i, mmSDMA0_UTCL1_TIMEOUT, 0x00800080);
1116 	}
1117 
1118 }
1119 
1120 /**
1121  * sdma_v4_0_enable - stop the async dma engines
1122  *
1123  * @adev: amdgpu_device pointer
1124  * @enable: enable/disable the DMA MEs.
1125  *
1126  * Halt or unhalt the async dma engines (VEGA10).
1127  */
1128 static void sdma_v4_0_enable(struct amdgpu_device *adev, bool enable)
1129 {
1130 	u32 f32_cntl;
1131 	int i;
1132 
1133 	if (!enable) {
1134 		sdma_v4_0_gfx_stop(adev);
1135 		sdma_v4_0_rlc_stop(adev);
1136 		if (adev->sdma.has_page_queue)
1137 			sdma_v4_0_page_stop(adev);
1138 	}
1139 
1140 	for (i = 0; i < adev->sdma.num_instances; i++) {
1141 		f32_cntl = RREG32_SDMA(i, mmSDMA0_F32_CNTL);
1142 		f32_cntl = REG_SET_FIELD(f32_cntl, SDMA0_F32_CNTL, HALT, enable ? 0 : 1);
1143 		WREG32_SDMA(i, mmSDMA0_F32_CNTL, f32_cntl);
1144 	}
1145 }
1146 
1147 /*
1148  * sdma_v4_0_rb_cntl - get parameters for rb_cntl
1149  */
1150 static uint32_t sdma_v4_0_rb_cntl(struct amdgpu_ring *ring, uint32_t rb_cntl)
1151 {
1152 	/* Set ring buffer size in dwords */
1153 	uint32_t rb_bufsz = order_base_2(ring->ring_size / 4);
1154 
1155 	rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_GFX_RB_CNTL, RB_SIZE, rb_bufsz);
1156 #ifdef __BIG_ENDIAN
1157 	rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_GFX_RB_CNTL, RB_SWAP_ENABLE, 1);
1158 	rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_GFX_RB_CNTL,
1159 				RPTR_WRITEBACK_SWAP_ENABLE, 1);
1160 #endif
1161 	return rb_cntl;
1162 }
1163 
1164 /**
1165  * sdma_v4_0_gfx_resume - setup and start the async dma engines
1166  *
1167  * @adev: amdgpu_device pointer
1168  * @i: instance to resume
1169  *
1170  * Set up the gfx DMA ring buffers and enable them (VEGA10).
1171  * Returns 0 for success, error for failure.
1172  */
1173 static void sdma_v4_0_gfx_resume(struct amdgpu_device *adev, unsigned int i)
1174 {
1175 	struct amdgpu_ring *ring = &adev->sdma.instance[i].ring;
1176 	u32 rb_cntl, ib_cntl, wptr_poll_cntl;
1177 	u32 doorbell;
1178 	u32 doorbell_offset;
1179 	u64 wptr_gpu_addr;
1180 
1181 	rb_cntl = RREG32_SDMA(i, mmSDMA0_GFX_RB_CNTL);
1182 	rb_cntl = sdma_v4_0_rb_cntl(ring, rb_cntl);
1183 	WREG32_SDMA(i, mmSDMA0_GFX_RB_CNTL, rb_cntl);
1184 
1185 	/* Initialize the ring buffer's read and write pointers */
1186 	WREG32_SDMA(i, mmSDMA0_GFX_RB_RPTR, 0);
1187 	WREG32_SDMA(i, mmSDMA0_GFX_RB_RPTR_HI, 0);
1188 	WREG32_SDMA(i, mmSDMA0_GFX_RB_WPTR, 0);
1189 	WREG32_SDMA(i, mmSDMA0_GFX_RB_WPTR_HI, 0);
1190 
1191 	/* set the wb address whether it's enabled or not */
1192 	WREG32_SDMA(i, mmSDMA0_GFX_RB_RPTR_ADDR_HI,
1193 	       upper_32_bits(ring->rptr_gpu_addr) & 0xFFFFFFFF);
1194 	WREG32_SDMA(i, mmSDMA0_GFX_RB_RPTR_ADDR_LO,
1195 	       lower_32_bits(ring->rptr_gpu_addr) & 0xFFFFFFFC);
1196 
1197 	rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_GFX_RB_CNTL,
1198 				RPTR_WRITEBACK_ENABLE, 1);
1199 
1200 	WREG32_SDMA(i, mmSDMA0_GFX_RB_BASE, ring->gpu_addr >> 8);
1201 	WREG32_SDMA(i, mmSDMA0_GFX_RB_BASE_HI, ring->gpu_addr >> 40);
1202 
1203 	ring->wptr = 0;
1204 
1205 	/* before programing wptr to a less value, need set minor_ptr_update first */
1206 	WREG32_SDMA(i, mmSDMA0_GFX_MINOR_PTR_UPDATE, 1);
1207 
1208 	doorbell = RREG32_SDMA(i, mmSDMA0_GFX_DOORBELL);
1209 	doorbell_offset = RREG32_SDMA(i, mmSDMA0_GFX_DOORBELL_OFFSET);
1210 
1211 	doorbell = REG_SET_FIELD(doorbell, SDMA0_GFX_DOORBELL, ENABLE,
1212 				 ring->use_doorbell);
1213 	doorbell_offset = REG_SET_FIELD(doorbell_offset,
1214 					SDMA0_GFX_DOORBELL_OFFSET,
1215 					OFFSET, ring->doorbell_index);
1216 	WREG32_SDMA(i, mmSDMA0_GFX_DOORBELL, doorbell);
1217 	WREG32_SDMA(i, mmSDMA0_GFX_DOORBELL_OFFSET, doorbell_offset);
1218 
1219 	sdma_v4_0_ring_set_wptr(ring);
1220 
1221 	/* set minor_ptr_update to 0 after wptr programed */
1222 	WREG32_SDMA(i, mmSDMA0_GFX_MINOR_PTR_UPDATE, 0);
1223 
1224 	/* setup the wptr shadow polling */
1225 	wptr_gpu_addr = ring->wptr_gpu_addr;
1226 	WREG32_SDMA(i, mmSDMA0_GFX_RB_WPTR_POLL_ADDR_LO,
1227 		    lower_32_bits(wptr_gpu_addr));
1228 	WREG32_SDMA(i, mmSDMA0_GFX_RB_WPTR_POLL_ADDR_HI,
1229 		    upper_32_bits(wptr_gpu_addr));
1230 	wptr_poll_cntl = RREG32_SDMA(i, mmSDMA0_GFX_RB_WPTR_POLL_CNTL);
1231 	wptr_poll_cntl = REG_SET_FIELD(wptr_poll_cntl,
1232 				       SDMA0_GFX_RB_WPTR_POLL_CNTL,
1233 				       F32_POLL_ENABLE, amdgpu_sriov_vf(adev)? 1 : 0);
1234 	WREG32_SDMA(i, mmSDMA0_GFX_RB_WPTR_POLL_CNTL, wptr_poll_cntl);
1235 
1236 	/* enable DMA RB */
1237 	rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_GFX_RB_CNTL, RB_ENABLE, 1);
1238 	WREG32_SDMA(i, mmSDMA0_GFX_RB_CNTL, rb_cntl);
1239 
1240 	ib_cntl = RREG32_SDMA(i, mmSDMA0_GFX_IB_CNTL);
1241 	ib_cntl = REG_SET_FIELD(ib_cntl, SDMA0_GFX_IB_CNTL, IB_ENABLE, 1);
1242 #ifdef __BIG_ENDIAN
1243 	ib_cntl = REG_SET_FIELD(ib_cntl, SDMA0_GFX_IB_CNTL, IB_SWAP_ENABLE, 1);
1244 #endif
1245 	/* enable DMA IBs */
1246 	WREG32_SDMA(i, mmSDMA0_GFX_IB_CNTL, ib_cntl);
1247 
1248 	ring->sched.ready = true;
1249 }
1250 
1251 /**
1252  * sdma_v4_0_page_resume - setup and start the async dma engines
1253  *
1254  * @adev: amdgpu_device pointer
1255  * @i: instance to resume
1256  *
1257  * Set up the page DMA ring buffers and enable them (VEGA10).
1258  * Returns 0 for success, error for failure.
1259  */
1260 static void sdma_v4_0_page_resume(struct amdgpu_device *adev, unsigned int i)
1261 {
1262 	struct amdgpu_ring *ring = &adev->sdma.instance[i].page;
1263 	u32 rb_cntl, ib_cntl, wptr_poll_cntl;
1264 	u32 doorbell;
1265 	u32 doorbell_offset;
1266 	u64 wptr_gpu_addr;
1267 
1268 	rb_cntl = RREG32_SDMA(i, mmSDMA0_PAGE_RB_CNTL);
1269 	rb_cntl = sdma_v4_0_rb_cntl(ring, rb_cntl);
1270 	WREG32_SDMA(i, mmSDMA0_PAGE_RB_CNTL, rb_cntl);
1271 
1272 	/* Initialize the ring buffer's read and write pointers */
1273 	WREG32_SDMA(i, mmSDMA0_PAGE_RB_RPTR, 0);
1274 	WREG32_SDMA(i, mmSDMA0_PAGE_RB_RPTR_HI, 0);
1275 	WREG32_SDMA(i, mmSDMA0_PAGE_RB_WPTR, 0);
1276 	WREG32_SDMA(i, mmSDMA0_PAGE_RB_WPTR_HI, 0);
1277 
1278 	/* set the wb address whether it's enabled or not */
1279 	WREG32_SDMA(i, mmSDMA0_PAGE_RB_RPTR_ADDR_HI,
1280 	       upper_32_bits(ring->rptr_gpu_addr) & 0xFFFFFFFF);
1281 	WREG32_SDMA(i, mmSDMA0_PAGE_RB_RPTR_ADDR_LO,
1282 	       lower_32_bits(ring->rptr_gpu_addr) & 0xFFFFFFFC);
1283 
1284 	rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_PAGE_RB_CNTL,
1285 				RPTR_WRITEBACK_ENABLE, 1);
1286 
1287 	WREG32_SDMA(i, mmSDMA0_PAGE_RB_BASE, ring->gpu_addr >> 8);
1288 	WREG32_SDMA(i, mmSDMA0_PAGE_RB_BASE_HI, ring->gpu_addr >> 40);
1289 
1290 	ring->wptr = 0;
1291 
1292 	/* before programing wptr to a less value, need set minor_ptr_update first */
1293 	WREG32_SDMA(i, mmSDMA0_PAGE_MINOR_PTR_UPDATE, 1);
1294 
1295 	doorbell = RREG32_SDMA(i, mmSDMA0_PAGE_DOORBELL);
1296 	doorbell_offset = RREG32_SDMA(i, mmSDMA0_PAGE_DOORBELL_OFFSET);
1297 
1298 	doorbell = REG_SET_FIELD(doorbell, SDMA0_PAGE_DOORBELL, ENABLE,
1299 				 ring->use_doorbell);
1300 	doorbell_offset = REG_SET_FIELD(doorbell_offset,
1301 					SDMA0_PAGE_DOORBELL_OFFSET,
1302 					OFFSET, ring->doorbell_index);
1303 	WREG32_SDMA(i, mmSDMA0_PAGE_DOORBELL, doorbell);
1304 	WREG32_SDMA(i, mmSDMA0_PAGE_DOORBELL_OFFSET, doorbell_offset);
1305 
1306 	/* paging queue doorbell range is setup at sdma_v4_0_gfx_resume */
1307 	sdma_v4_0_page_ring_set_wptr(ring);
1308 
1309 	/* set minor_ptr_update to 0 after wptr programed */
1310 	WREG32_SDMA(i, mmSDMA0_PAGE_MINOR_PTR_UPDATE, 0);
1311 
1312 	/* setup the wptr shadow polling */
1313 	wptr_gpu_addr = ring->wptr_gpu_addr;
1314 	WREG32_SDMA(i, mmSDMA0_PAGE_RB_WPTR_POLL_ADDR_LO,
1315 		    lower_32_bits(wptr_gpu_addr));
1316 	WREG32_SDMA(i, mmSDMA0_PAGE_RB_WPTR_POLL_ADDR_HI,
1317 		    upper_32_bits(wptr_gpu_addr));
1318 	wptr_poll_cntl = RREG32_SDMA(i, mmSDMA0_PAGE_RB_WPTR_POLL_CNTL);
1319 	wptr_poll_cntl = REG_SET_FIELD(wptr_poll_cntl,
1320 				       SDMA0_PAGE_RB_WPTR_POLL_CNTL,
1321 				       F32_POLL_ENABLE, amdgpu_sriov_vf(adev)? 1 : 0);
1322 	WREG32_SDMA(i, mmSDMA0_PAGE_RB_WPTR_POLL_CNTL, wptr_poll_cntl);
1323 
1324 	/* enable DMA RB */
1325 	rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_PAGE_RB_CNTL, RB_ENABLE, 1);
1326 	WREG32_SDMA(i, mmSDMA0_PAGE_RB_CNTL, rb_cntl);
1327 
1328 	ib_cntl = RREG32_SDMA(i, mmSDMA0_PAGE_IB_CNTL);
1329 	ib_cntl = REG_SET_FIELD(ib_cntl, SDMA0_PAGE_IB_CNTL, IB_ENABLE, 1);
1330 #ifdef __BIG_ENDIAN
1331 	ib_cntl = REG_SET_FIELD(ib_cntl, SDMA0_PAGE_IB_CNTL, IB_SWAP_ENABLE, 1);
1332 #endif
1333 	/* enable DMA IBs */
1334 	WREG32_SDMA(i, mmSDMA0_PAGE_IB_CNTL, ib_cntl);
1335 
1336 	ring->sched.ready = true;
1337 }
1338 
1339 static void
1340 sdma_v4_1_update_power_gating(struct amdgpu_device *adev, bool enable)
1341 {
1342 	uint32_t def, data;
1343 
1344 	if (enable && (adev->pg_flags & AMD_PG_SUPPORT_SDMA)) {
1345 		/* enable idle interrupt */
1346 		def = data = RREG32(SOC15_REG_OFFSET(SDMA0, 0, mmSDMA0_CNTL));
1347 		data |= SDMA0_CNTL__CTXEMPTY_INT_ENABLE_MASK;
1348 
1349 		if (data != def)
1350 			WREG32(SOC15_REG_OFFSET(SDMA0, 0, mmSDMA0_CNTL), data);
1351 	} else {
1352 		/* disable idle interrupt */
1353 		def = data = RREG32(SOC15_REG_OFFSET(SDMA0, 0, mmSDMA0_CNTL));
1354 		data &= ~SDMA0_CNTL__CTXEMPTY_INT_ENABLE_MASK;
1355 		if (data != def)
1356 			WREG32(SOC15_REG_OFFSET(SDMA0, 0, mmSDMA0_CNTL), data);
1357 	}
1358 }
1359 
1360 static void sdma_v4_1_init_power_gating(struct amdgpu_device *adev)
1361 {
1362 	uint32_t def, data;
1363 
1364 	/* Enable HW based PG. */
1365 	def = data = RREG32(SOC15_REG_OFFSET(SDMA0, 0, mmSDMA0_POWER_CNTL));
1366 	data |= SDMA0_POWER_CNTL__PG_CNTL_ENABLE_MASK;
1367 	if (data != def)
1368 		WREG32(SOC15_REG_OFFSET(SDMA0, 0, mmSDMA0_POWER_CNTL), data);
1369 
1370 	/* enable interrupt */
1371 	def = data = RREG32(SOC15_REG_OFFSET(SDMA0, 0, mmSDMA0_CNTL));
1372 	data |= SDMA0_CNTL__CTXEMPTY_INT_ENABLE_MASK;
1373 	if (data != def)
1374 		WREG32(SOC15_REG_OFFSET(SDMA0, 0, mmSDMA0_CNTL), data);
1375 
1376 	/* Configure hold time to filter in-valid power on/off request. Use default right now */
1377 	def = data = RREG32(SOC15_REG_OFFSET(SDMA0, 0, mmSDMA0_POWER_CNTL));
1378 	data &= ~SDMA0_POWER_CNTL__ON_OFF_CONDITION_HOLD_TIME_MASK;
1379 	data |= (mmSDMA0_POWER_CNTL_DEFAULT & SDMA0_POWER_CNTL__ON_OFF_CONDITION_HOLD_TIME_MASK);
1380 	/* Configure switch time for hysteresis purpose. Use default right now */
1381 	data &= ~SDMA0_POWER_CNTL__ON_OFF_STATUS_DURATION_TIME_MASK;
1382 	data |= (mmSDMA0_POWER_CNTL_DEFAULT & SDMA0_POWER_CNTL__ON_OFF_STATUS_DURATION_TIME_MASK);
1383 	if(data != def)
1384 		WREG32(SOC15_REG_OFFSET(SDMA0, 0, mmSDMA0_POWER_CNTL), data);
1385 }
1386 
1387 static void sdma_v4_0_init_pg(struct amdgpu_device *adev)
1388 {
1389 	if (!(adev->pg_flags & AMD_PG_SUPPORT_SDMA))
1390 		return;
1391 
1392 	switch (adev->ip_versions[SDMA0_HWIP][0]) {
1393 	case IP_VERSION(4, 1, 0):
1394         case IP_VERSION(4, 1, 1):
1395 	case IP_VERSION(4, 1, 2):
1396 		sdma_v4_1_init_power_gating(adev);
1397 		sdma_v4_1_update_power_gating(adev, true);
1398 		break;
1399 	default:
1400 		break;
1401 	}
1402 }
1403 
1404 /**
1405  * sdma_v4_0_rlc_resume - setup and start the async dma engines
1406  *
1407  * @adev: amdgpu_device pointer
1408  *
1409  * Set up the compute DMA queues and enable them (VEGA10).
1410  * Returns 0 for success, error for failure.
1411  */
1412 static int sdma_v4_0_rlc_resume(struct amdgpu_device *adev)
1413 {
1414 	sdma_v4_0_init_pg(adev);
1415 
1416 	return 0;
1417 }
1418 
1419 /**
1420  * sdma_v4_0_load_microcode - load the sDMA ME ucode
1421  *
1422  * @adev: amdgpu_device pointer
1423  *
1424  * Loads the sDMA0/1 ucode.
1425  * Returns 0 for success, -EINVAL if the ucode is not available.
1426  */
1427 static int sdma_v4_0_load_microcode(struct amdgpu_device *adev)
1428 {
1429 	const struct sdma_firmware_header_v1_0 *hdr;
1430 	const __le32 *fw_data;
1431 	u32 fw_size;
1432 	int i, j;
1433 
1434 	/* halt the MEs */
1435 	sdma_v4_0_enable(adev, false);
1436 
1437 	for (i = 0; i < adev->sdma.num_instances; i++) {
1438 		if (!adev->sdma.instance[i].fw)
1439 			return -EINVAL;
1440 
1441 		hdr = (const struct sdma_firmware_header_v1_0 *)adev->sdma.instance[i].fw->data;
1442 		amdgpu_ucode_print_sdma_hdr(&hdr->header);
1443 		fw_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4;
1444 
1445 		fw_data = (const __le32 *)
1446 			(adev->sdma.instance[i].fw->data +
1447 				le32_to_cpu(hdr->header.ucode_array_offset_bytes));
1448 
1449 		WREG32_SDMA(i, mmSDMA0_UCODE_ADDR, 0);
1450 
1451 		for (j = 0; j < fw_size; j++)
1452 			WREG32_SDMA(i, mmSDMA0_UCODE_DATA,
1453 				    le32_to_cpup(fw_data++));
1454 
1455 		WREG32_SDMA(i, mmSDMA0_UCODE_ADDR,
1456 			    adev->sdma.instance[i].fw_version);
1457 	}
1458 
1459 	return 0;
1460 }
1461 
1462 /**
1463  * sdma_v4_0_start - setup and start the async dma engines
1464  *
1465  * @adev: amdgpu_device pointer
1466  *
1467  * Set up the DMA engines and enable them (VEGA10).
1468  * Returns 0 for success, error for failure.
1469  */
1470 static int sdma_v4_0_start(struct amdgpu_device *adev)
1471 {
1472 	struct amdgpu_ring *ring;
1473 	int i, r = 0;
1474 
1475 	if (amdgpu_sriov_vf(adev)) {
1476 		sdma_v4_0_ctx_switch_enable(adev, false);
1477 		sdma_v4_0_enable(adev, false);
1478 	} else {
1479 
1480 		if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP) {
1481 			r = sdma_v4_0_load_microcode(adev);
1482 			if (r)
1483 				return r;
1484 		}
1485 
1486 		/* unhalt the MEs */
1487 		sdma_v4_0_enable(adev, true);
1488 		/* enable sdma ring preemption */
1489 		sdma_v4_0_ctx_switch_enable(adev, true);
1490 	}
1491 
1492 	/* start the gfx rings and rlc compute queues */
1493 	for (i = 0; i < adev->sdma.num_instances; i++) {
1494 		uint32_t temp;
1495 
1496 		WREG32_SDMA(i, mmSDMA0_SEM_WAIT_FAIL_TIMER_CNTL, 0);
1497 		sdma_v4_0_gfx_resume(adev, i);
1498 		if (adev->sdma.has_page_queue)
1499 			sdma_v4_0_page_resume(adev, i);
1500 
1501 		/* set utc l1 enable flag always to 1 */
1502 		temp = RREG32_SDMA(i, mmSDMA0_CNTL);
1503 		temp = REG_SET_FIELD(temp, SDMA0_CNTL, UTC_L1_ENABLE, 1);
1504 		WREG32_SDMA(i, mmSDMA0_CNTL, temp);
1505 
1506 		if (!amdgpu_sriov_vf(adev)) {
1507 			ring = &adev->sdma.instance[i].ring;
1508 			adev->nbio.funcs->sdma_doorbell_range(adev, i,
1509 				ring->use_doorbell, ring->doorbell_index,
1510 				adev->doorbell_index.sdma_doorbell_range);
1511 
1512 			/* unhalt engine */
1513 			temp = RREG32_SDMA(i, mmSDMA0_F32_CNTL);
1514 			temp = REG_SET_FIELD(temp, SDMA0_F32_CNTL, HALT, 0);
1515 			WREG32_SDMA(i, mmSDMA0_F32_CNTL, temp);
1516 		}
1517 	}
1518 
1519 	if (amdgpu_sriov_vf(adev)) {
1520 		sdma_v4_0_ctx_switch_enable(adev, true);
1521 		sdma_v4_0_enable(adev, true);
1522 	} else {
1523 		r = sdma_v4_0_rlc_resume(adev);
1524 		if (r)
1525 			return r;
1526 	}
1527 
1528 	for (i = 0; i < adev->sdma.num_instances; i++) {
1529 		ring = &adev->sdma.instance[i].ring;
1530 
1531 		r = amdgpu_ring_test_helper(ring);
1532 		if (r)
1533 			return r;
1534 
1535 		if (adev->sdma.has_page_queue) {
1536 			struct amdgpu_ring *page = &adev->sdma.instance[i].page;
1537 
1538 			r = amdgpu_ring_test_helper(page);
1539 			if (r)
1540 				return r;
1541 
1542 			if (adev->mman.buffer_funcs_ring == page)
1543 				amdgpu_ttm_set_buffer_funcs_status(adev, true);
1544 		}
1545 
1546 		if (adev->mman.buffer_funcs_ring == ring)
1547 			amdgpu_ttm_set_buffer_funcs_status(adev, true);
1548 	}
1549 
1550 	return r;
1551 }
1552 
1553 /**
1554  * sdma_v4_0_ring_test_ring - simple async dma engine test
1555  *
1556  * @ring: amdgpu_ring structure holding ring information
1557  *
1558  * Test the DMA engine by writing using it to write an
1559  * value to memory. (VEGA10).
1560  * Returns 0 for success, error for failure.
1561  */
1562 static int sdma_v4_0_ring_test_ring(struct amdgpu_ring *ring)
1563 {
1564 	struct amdgpu_device *adev = ring->adev;
1565 	unsigned i;
1566 	unsigned index;
1567 	int r;
1568 	u32 tmp;
1569 	u64 gpu_addr;
1570 
1571 	r = amdgpu_device_wb_get(adev, &index);
1572 	if (r)
1573 		return r;
1574 
1575 	gpu_addr = adev->wb.gpu_addr + (index * 4);
1576 	tmp = 0xCAFEDEAD;
1577 	adev->wb.wb[index] = cpu_to_le32(tmp);
1578 
1579 	r = amdgpu_ring_alloc(ring, 5);
1580 	if (r)
1581 		goto error_free_wb;
1582 
1583 	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_WRITE) |
1584 			  SDMA_PKT_HEADER_SUB_OP(SDMA_SUBOP_WRITE_LINEAR));
1585 	amdgpu_ring_write(ring, lower_32_bits(gpu_addr));
1586 	amdgpu_ring_write(ring, upper_32_bits(gpu_addr));
1587 	amdgpu_ring_write(ring, SDMA_PKT_WRITE_UNTILED_DW_3_COUNT(0));
1588 	amdgpu_ring_write(ring, 0xDEADBEEF);
1589 	amdgpu_ring_commit(ring);
1590 
1591 	for (i = 0; i < adev->usec_timeout; i++) {
1592 		tmp = le32_to_cpu(adev->wb.wb[index]);
1593 		if (tmp == 0xDEADBEEF)
1594 			break;
1595 		udelay(1);
1596 	}
1597 
1598 	if (i >= adev->usec_timeout)
1599 		r = -ETIMEDOUT;
1600 
1601 error_free_wb:
1602 	amdgpu_device_wb_free(adev, index);
1603 	return r;
1604 }
1605 
1606 /**
1607  * sdma_v4_0_ring_test_ib - test an IB on the DMA engine
1608  *
1609  * @ring: amdgpu_ring structure holding ring information
1610  * @timeout: timeout value in jiffies, or MAX_SCHEDULE_TIMEOUT
1611  *
1612  * Test a simple IB in the DMA ring (VEGA10).
1613  * Returns 0 on success, error on failure.
1614  */
1615 static int sdma_v4_0_ring_test_ib(struct amdgpu_ring *ring, long timeout)
1616 {
1617 	struct amdgpu_device *adev = ring->adev;
1618 	struct amdgpu_ib ib;
1619 	struct dma_fence *f = NULL;
1620 	unsigned index;
1621 	long r;
1622 	u32 tmp = 0;
1623 	u64 gpu_addr;
1624 
1625 	r = amdgpu_device_wb_get(adev, &index);
1626 	if (r)
1627 		return r;
1628 
1629 	gpu_addr = adev->wb.gpu_addr + (index * 4);
1630 	tmp = 0xCAFEDEAD;
1631 	adev->wb.wb[index] = cpu_to_le32(tmp);
1632 	memset(&ib, 0, sizeof(ib));
1633 	r = amdgpu_ib_get(adev, NULL, 256,
1634 					AMDGPU_IB_POOL_DIRECT, &ib);
1635 	if (r)
1636 		goto err0;
1637 
1638 	ib.ptr[0] = SDMA_PKT_HEADER_OP(SDMA_OP_WRITE) |
1639 		SDMA_PKT_HEADER_SUB_OP(SDMA_SUBOP_WRITE_LINEAR);
1640 	ib.ptr[1] = lower_32_bits(gpu_addr);
1641 	ib.ptr[2] = upper_32_bits(gpu_addr);
1642 	ib.ptr[3] = SDMA_PKT_WRITE_UNTILED_DW_3_COUNT(0);
1643 	ib.ptr[4] = 0xDEADBEEF;
1644 	ib.ptr[5] = SDMA_PKT_NOP_HEADER_OP(SDMA_OP_NOP);
1645 	ib.ptr[6] = SDMA_PKT_NOP_HEADER_OP(SDMA_OP_NOP);
1646 	ib.ptr[7] = SDMA_PKT_NOP_HEADER_OP(SDMA_OP_NOP);
1647 	ib.length_dw = 8;
1648 
1649 	r = amdgpu_ib_schedule(ring, 1, &ib, NULL, &f);
1650 	if (r)
1651 		goto err1;
1652 
1653 	r = dma_fence_wait_timeout(f, false, timeout);
1654 	if (r == 0) {
1655 		r = -ETIMEDOUT;
1656 		goto err1;
1657 	} else if (r < 0) {
1658 		goto err1;
1659 	}
1660 	tmp = le32_to_cpu(adev->wb.wb[index]);
1661 	if (tmp == 0xDEADBEEF)
1662 		r = 0;
1663 	else
1664 		r = -EINVAL;
1665 
1666 err1:
1667 	amdgpu_ib_free(adev, &ib, NULL);
1668 	dma_fence_put(f);
1669 err0:
1670 	amdgpu_device_wb_free(adev, index);
1671 	return r;
1672 }
1673 
1674 
1675 /**
1676  * sdma_v4_0_vm_copy_pte - update PTEs by copying them from the GART
1677  *
1678  * @ib: indirect buffer to fill with commands
1679  * @pe: addr of the page entry
1680  * @src: src addr to copy from
1681  * @count: number of page entries to update
1682  *
1683  * Update PTEs by copying them from the GART using sDMA (VEGA10).
1684  */
1685 static void sdma_v4_0_vm_copy_pte(struct amdgpu_ib *ib,
1686 				  uint64_t pe, uint64_t src,
1687 				  unsigned count)
1688 {
1689 	unsigned bytes = count * 8;
1690 
1691 	ib->ptr[ib->length_dw++] = SDMA_PKT_HEADER_OP(SDMA_OP_COPY) |
1692 		SDMA_PKT_HEADER_SUB_OP(SDMA_SUBOP_COPY_LINEAR);
1693 	ib->ptr[ib->length_dw++] = bytes - 1;
1694 	ib->ptr[ib->length_dw++] = 0; /* src/dst endian swap */
1695 	ib->ptr[ib->length_dw++] = lower_32_bits(src);
1696 	ib->ptr[ib->length_dw++] = upper_32_bits(src);
1697 	ib->ptr[ib->length_dw++] = lower_32_bits(pe);
1698 	ib->ptr[ib->length_dw++] = upper_32_bits(pe);
1699 
1700 }
1701 
1702 /**
1703  * sdma_v4_0_vm_write_pte - update PTEs by writing them manually
1704  *
1705  * @ib: indirect buffer to fill with commands
1706  * @pe: addr of the page entry
1707  * @value: dst addr to write into pe
1708  * @count: number of page entries to update
1709  * @incr: increase next addr by incr bytes
1710  *
1711  * Update PTEs by writing them manually using sDMA (VEGA10).
1712  */
1713 static void sdma_v4_0_vm_write_pte(struct amdgpu_ib *ib, uint64_t pe,
1714 				   uint64_t value, unsigned count,
1715 				   uint32_t incr)
1716 {
1717 	unsigned ndw = count * 2;
1718 
1719 	ib->ptr[ib->length_dw++] = SDMA_PKT_HEADER_OP(SDMA_OP_WRITE) |
1720 		SDMA_PKT_HEADER_SUB_OP(SDMA_SUBOP_WRITE_LINEAR);
1721 	ib->ptr[ib->length_dw++] = lower_32_bits(pe);
1722 	ib->ptr[ib->length_dw++] = upper_32_bits(pe);
1723 	ib->ptr[ib->length_dw++] = ndw - 1;
1724 	for (; ndw > 0; ndw -= 2) {
1725 		ib->ptr[ib->length_dw++] = lower_32_bits(value);
1726 		ib->ptr[ib->length_dw++] = upper_32_bits(value);
1727 		value += incr;
1728 	}
1729 }
1730 
1731 /**
1732  * sdma_v4_0_vm_set_pte_pde - update the page tables using sDMA
1733  *
1734  * @ib: indirect buffer to fill with commands
1735  * @pe: addr of the page entry
1736  * @addr: dst addr to write into pe
1737  * @count: number of page entries to update
1738  * @incr: increase next addr by incr bytes
1739  * @flags: access flags
1740  *
1741  * Update the page tables using sDMA (VEGA10).
1742  */
1743 static void sdma_v4_0_vm_set_pte_pde(struct amdgpu_ib *ib,
1744 				     uint64_t pe,
1745 				     uint64_t addr, unsigned count,
1746 				     uint32_t incr, uint64_t flags)
1747 {
1748 	/* for physically contiguous pages (vram) */
1749 	ib->ptr[ib->length_dw++] = SDMA_PKT_HEADER_OP(SDMA_OP_PTEPDE);
1750 	ib->ptr[ib->length_dw++] = lower_32_bits(pe); /* dst addr */
1751 	ib->ptr[ib->length_dw++] = upper_32_bits(pe);
1752 	ib->ptr[ib->length_dw++] = lower_32_bits(flags); /* mask */
1753 	ib->ptr[ib->length_dw++] = upper_32_bits(flags);
1754 	ib->ptr[ib->length_dw++] = lower_32_bits(addr); /* value */
1755 	ib->ptr[ib->length_dw++] = upper_32_bits(addr);
1756 	ib->ptr[ib->length_dw++] = incr; /* increment size */
1757 	ib->ptr[ib->length_dw++] = 0;
1758 	ib->ptr[ib->length_dw++] = count - 1; /* number of entries */
1759 }
1760 
1761 /**
1762  * sdma_v4_0_ring_pad_ib - pad the IB to the required number of dw
1763  *
1764  * @ring: amdgpu_ring structure holding ring information
1765  * @ib: indirect buffer to fill with padding
1766  */
1767 static void sdma_v4_0_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib)
1768 {
1769 	struct amdgpu_sdma_instance *sdma = amdgpu_sdma_get_instance_from_ring(ring);
1770 	u32 pad_count;
1771 	int i;
1772 
1773 	pad_count = (-ib->length_dw) & 7;
1774 	for (i = 0; i < pad_count; i++)
1775 		if (sdma && sdma->burst_nop && (i == 0))
1776 			ib->ptr[ib->length_dw++] =
1777 				SDMA_PKT_HEADER_OP(SDMA_OP_NOP) |
1778 				SDMA_PKT_NOP_HEADER_COUNT(pad_count - 1);
1779 		else
1780 			ib->ptr[ib->length_dw++] =
1781 				SDMA_PKT_HEADER_OP(SDMA_OP_NOP);
1782 }
1783 
1784 
1785 /**
1786  * sdma_v4_0_ring_emit_pipeline_sync - sync the pipeline
1787  *
1788  * @ring: amdgpu_ring pointer
1789  *
1790  * Make sure all previous operations are completed (CIK).
1791  */
1792 static void sdma_v4_0_ring_emit_pipeline_sync(struct amdgpu_ring *ring)
1793 {
1794 	uint32_t seq = ring->fence_drv.sync_seq;
1795 	uint64_t addr = ring->fence_drv.gpu_addr;
1796 
1797 	/* wait for idle */
1798 	sdma_v4_0_wait_reg_mem(ring, 1, 0,
1799 			       addr & 0xfffffffc,
1800 			       upper_32_bits(addr) & 0xffffffff,
1801 			       seq, 0xffffffff, 4);
1802 }
1803 
1804 
1805 /**
1806  * sdma_v4_0_ring_emit_vm_flush - vm flush using sDMA
1807  *
1808  * @ring: amdgpu_ring pointer
1809  * @vmid: vmid number to use
1810  * @pd_addr: address
1811  *
1812  * Update the page table base and flush the VM TLB
1813  * using sDMA (VEGA10).
1814  */
1815 static void sdma_v4_0_ring_emit_vm_flush(struct amdgpu_ring *ring,
1816 					 unsigned vmid, uint64_t pd_addr)
1817 {
1818 	amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr);
1819 }
1820 
1821 static void sdma_v4_0_ring_emit_wreg(struct amdgpu_ring *ring,
1822 				     uint32_t reg, uint32_t val)
1823 {
1824 	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_SRBM_WRITE) |
1825 			  SDMA_PKT_SRBM_WRITE_HEADER_BYTE_EN(0xf));
1826 	amdgpu_ring_write(ring, reg);
1827 	amdgpu_ring_write(ring, val);
1828 }
1829 
1830 static void sdma_v4_0_ring_emit_reg_wait(struct amdgpu_ring *ring, uint32_t reg,
1831 					 uint32_t val, uint32_t mask)
1832 {
1833 	sdma_v4_0_wait_reg_mem(ring, 0, 0, reg, 0, val, mask, 10);
1834 }
1835 
1836 static bool sdma_v4_0_fw_support_paging_queue(struct amdgpu_device *adev)
1837 {
1838 	uint fw_version = adev->sdma.instance[0].fw_version;
1839 
1840 	switch (adev->ip_versions[SDMA0_HWIP][0]) {
1841 	case IP_VERSION(4, 0, 0):
1842 		return fw_version >= 430;
1843 	case IP_VERSION(4, 0, 1):
1844 		/*return fw_version >= 31;*/
1845 		return false;
1846 	case IP_VERSION(4, 2, 0):
1847 		return fw_version >= 123;
1848 	default:
1849 		return false;
1850 	}
1851 }
1852 
1853 static int sdma_v4_0_early_init(void *handle)
1854 {
1855 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1856 	int r;
1857 
1858 	r = sdma_v4_0_init_microcode(adev);
1859 	if (r) {
1860 		DRM_ERROR("Failed to load sdma firmware!\n");
1861 		return r;
1862 	}
1863 
1864 	/* TODO: Page queue breaks driver reload under SRIOV */
1865 	if ((adev->ip_versions[SDMA0_HWIP][0] == IP_VERSION(4, 0, 0)) &&
1866 	    amdgpu_sriov_vf((adev)))
1867 		adev->sdma.has_page_queue = false;
1868 	else if (sdma_v4_0_fw_support_paging_queue(adev))
1869 		adev->sdma.has_page_queue = true;
1870 
1871 	sdma_v4_0_set_ring_funcs(adev);
1872 	sdma_v4_0_set_buffer_funcs(adev);
1873 	sdma_v4_0_set_vm_pte_funcs(adev);
1874 	sdma_v4_0_set_irq_funcs(adev);
1875 	sdma_v4_0_set_ras_funcs(adev);
1876 
1877 	return 0;
1878 }
1879 
1880 static int sdma_v4_0_process_ras_data_cb(struct amdgpu_device *adev,
1881 		void *err_data,
1882 		struct amdgpu_iv_entry *entry);
1883 
1884 static int sdma_v4_0_late_init(void *handle)
1885 {
1886 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1887 
1888 	sdma_v4_0_setup_ulv(adev);
1889 
1890 	if (!amdgpu_persistent_edc_harvesting_supported(adev)) {
1891 		if (adev->sdma.ras && adev->sdma.ras->ras_block.hw_ops &&
1892 		    adev->sdma.ras->ras_block.hw_ops->reset_ras_error_count)
1893 			adev->sdma.ras->ras_block.hw_ops->reset_ras_error_count(adev);
1894 	}
1895 
1896 	return 0;
1897 }
1898 
1899 static int sdma_v4_0_sw_init(void *handle)
1900 {
1901 	struct amdgpu_ring *ring;
1902 	int r, i;
1903 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1904 
1905 	/* SDMA trap event */
1906 	for (i = 0; i < adev->sdma.num_instances; i++) {
1907 		r = amdgpu_irq_add_id(adev, sdma_v4_0_seq_to_irq_id(i),
1908 				      SDMA0_4_0__SRCID__SDMA_TRAP,
1909 				      &adev->sdma.trap_irq);
1910 		if (r)
1911 			return r;
1912 	}
1913 
1914 	/* SDMA SRAM ECC event */
1915 	for (i = 0; i < adev->sdma.num_instances; i++) {
1916 		r = amdgpu_irq_add_id(adev, sdma_v4_0_seq_to_irq_id(i),
1917 				      SDMA0_4_0__SRCID__SDMA_SRAM_ECC,
1918 				      &adev->sdma.ecc_irq);
1919 		if (r)
1920 			return r;
1921 	}
1922 
1923 	/* SDMA VM_HOLE/DOORBELL_INV/POLL_TIMEOUT/SRBM_WRITE_PROTECTION event*/
1924 	for (i = 0; i < adev->sdma.num_instances; i++) {
1925 		r = amdgpu_irq_add_id(adev, sdma_v4_0_seq_to_irq_id(i),
1926 				      SDMA0_4_0__SRCID__SDMA_VM_HOLE,
1927 				      &adev->sdma.vm_hole_irq);
1928 		if (r)
1929 			return r;
1930 
1931 		r = amdgpu_irq_add_id(adev, sdma_v4_0_seq_to_irq_id(i),
1932 				      SDMA0_4_0__SRCID__SDMA_DOORBELL_INVALID,
1933 				      &adev->sdma.doorbell_invalid_irq);
1934 		if (r)
1935 			return r;
1936 
1937 		r = amdgpu_irq_add_id(adev, sdma_v4_0_seq_to_irq_id(i),
1938 				      SDMA0_4_0__SRCID__SDMA_POLL_TIMEOUT,
1939 				      &adev->sdma.pool_timeout_irq);
1940 		if (r)
1941 			return r;
1942 
1943 		r = amdgpu_irq_add_id(adev, sdma_v4_0_seq_to_irq_id(i),
1944 				      SDMA0_4_0__SRCID__SDMA_SRBMWRITE,
1945 				      &adev->sdma.srbm_write_irq);
1946 		if (r)
1947 			return r;
1948 	}
1949 
1950 	for (i = 0; i < adev->sdma.num_instances; i++) {
1951 		ring = &adev->sdma.instance[i].ring;
1952 		ring->ring_obj = NULL;
1953 		ring->use_doorbell = true;
1954 
1955 		DRM_DEBUG("SDMA %d use_doorbell being set to: [%s]\n", i,
1956 				ring->use_doorbell?"true":"false");
1957 
1958 		/* doorbell size is 2 dwords, get DWORD offset */
1959 		ring->doorbell_index = adev->doorbell_index.sdma_engine[i] << 1;
1960 
1961 		sprintf(ring->name, "sdma%d", i);
1962 		r = amdgpu_ring_init(adev, ring, 1024, &adev->sdma.trap_irq,
1963 				     AMDGPU_SDMA_IRQ_INSTANCE0 + i,
1964 				     AMDGPU_RING_PRIO_DEFAULT, NULL);
1965 		if (r)
1966 			return r;
1967 
1968 		if (adev->sdma.has_page_queue) {
1969 			ring = &adev->sdma.instance[i].page;
1970 			ring->ring_obj = NULL;
1971 			ring->use_doorbell = true;
1972 
1973 			/* paging queue use same doorbell index/routing as gfx queue
1974 			 * with 0x400 (4096 dwords) offset on second doorbell page
1975 			 */
1976 			ring->doorbell_index = adev->doorbell_index.sdma_engine[i] << 1;
1977 			ring->doorbell_index += 0x400;
1978 
1979 			sprintf(ring->name, "page%d", i);
1980 			r = amdgpu_ring_init(adev, ring, 1024,
1981 					     &adev->sdma.trap_irq,
1982 					     AMDGPU_SDMA_IRQ_INSTANCE0 + i,
1983 					     AMDGPU_RING_PRIO_DEFAULT, NULL);
1984 			if (r)
1985 				return r;
1986 		}
1987 	}
1988 
1989 	return r;
1990 }
1991 
1992 static int sdma_v4_0_sw_fini(void *handle)
1993 {
1994 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1995 	int i;
1996 
1997 	for (i = 0; i < adev->sdma.num_instances; i++) {
1998 		amdgpu_ring_fini(&adev->sdma.instance[i].ring);
1999 		if (adev->sdma.has_page_queue)
2000 			amdgpu_ring_fini(&adev->sdma.instance[i].page);
2001 	}
2002 
2003 	sdma_v4_0_destroy_inst_ctx(adev);
2004 
2005 	return 0;
2006 }
2007 
2008 static int sdma_v4_0_hw_init(void *handle)
2009 {
2010 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2011 
2012 	if (adev->flags & AMD_IS_APU)
2013 		amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_SDMA, false);
2014 
2015 	if (!amdgpu_sriov_vf(adev))
2016 		sdma_v4_0_init_golden_registers(adev);
2017 
2018 	return sdma_v4_0_start(adev);
2019 }
2020 
2021 static int sdma_v4_0_hw_fini(void *handle)
2022 {
2023 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2024 	int i;
2025 
2026 	if (amdgpu_sriov_vf(adev))
2027 		return 0;
2028 
2029 	for (i = 0; i < adev->sdma.num_instances; i++) {
2030 		amdgpu_irq_put(adev, &adev->sdma.ecc_irq,
2031 			       AMDGPU_SDMA_IRQ_INSTANCE0 + i);
2032 	}
2033 
2034 	sdma_v4_0_ctx_switch_enable(adev, false);
2035 	sdma_v4_0_enable(adev, false);
2036 
2037 	if (adev->flags & AMD_IS_APU)
2038 		amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_SDMA, true);
2039 
2040 	return 0;
2041 }
2042 
2043 static int sdma_v4_0_suspend(void *handle)
2044 {
2045 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2046 
2047 	/* SMU saves SDMA state for us */
2048 	if (adev->in_s0ix)
2049 		return 0;
2050 
2051 	return sdma_v4_0_hw_fini(adev);
2052 }
2053 
2054 static int sdma_v4_0_resume(void *handle)
2055 {
2056 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2057 
2058 	/* SMU restores SDMA state for us */
2059 	if (adev->in_s0ix)
2060 		return 0;
2061 
2062 	return sdma_v4_0_hw_init(adev);
2063 }
2064 
2065 static bool sdma_v4_0_is_idle(void *handle)
2066 {
2067 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2068 	u32 i;
2069 
2070 	for (i = 0; i < adev->sdma.num_instances; i++) {
2071 		u32 tmp = RREG32_SDMA(i, mmSDMA0_STATUS_REG);
2072 
2073 		if (!(tmp & SDMA0_STATUS_REG__IDLE_MASK))
2074 			return false;
2075 	}
2076 
2077 	return true;
2078 }
2079 
2080 static int sdma_v4_0_wait_for_idle(void *handle)
2081 {
2082 	unsigned i, j;
2083 	u32 sdma[AMDGPU_MAX_SDMA_INSTANCES];
2084 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2085 
2086 	for (i = 0; i < adev->usec_timeout; i++) {
2087 		for (j = 0; j < adev->sdma.num_instances; j++) {
2088 			sdma[j] = RREG32_SDMA(j, mmSDMA0_STATUS_REG);
2089 			if (!(sdma[j] & SDMA0_STATUS_REG__IDLE_MASK))
2090 				break;
2091 		}
2092 		if (j == adev->sdma.num_instances)
2093 			return 0;
2094 		udelay(1);
2095 	}
2096 	return -ETIMEDOUT;
2097 }
2098 
2099 static int sdma_v4_0_soft_reset(void *handle)
2100 {
2101 	/* todo */
2102 
2103 	return 0;
2104 }
2105 
2106 static int sdma_v4_0_set_trap_irq_state(struct amdgpu_device *adev,
2107 					struct amdgpu_irq_src *source,
2108 					unsigned type,
2109 					enum amdgpu_interrupt_state state)
2110 {
2111 	u32 sdma_cntl;
2112 
2113 	sdma_cntl = RREG32_SDMA(type, mmSDMA0_CNTL);
2114 	sdma_cntl = REG_SET_FIELD(sdma_cntl, SDMA0_CNTL, TRAP_ENABLE,
2115 		       state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0);
2116 	WREG32_SDMA(type, mmSDMA0_CNTL, sdma_cntl);
2117 
2118 	return 0;
2119 }
2120 
2121 static int sdma_v4_0_process_trap_irq(struct amdgpu_device *adev,
2122 				      struct amdgpu_irq_src *source,
2123 				      struct amdgpu_iv_entry *entry)
2124 {
2125 	uint32_t instance;
2126 
2127 	DRM_DEBUG("IH: SDMA trap\n");
2128 	instance = sdma_v4_0_irq_id_to_seq(entry->client_id);
2129 	switch (entry->ring_id) {
2130 	case 0:
2131 		amdgpu_fence_process(&adev->sdma.instance[instance].ring);
2132 		break;
2133 	case 1:
2134 		if (adev->ip_versions[SDMA0_HWIP][0] == IP_VERSION(4, 2, 0))
2135 			amdgpu_fence_process(&adev->sdma.instance[instance].page);
2136 		break;
2137 	case 2:
2138 		/* XXX compute */
2139 		break;
2140 	case 3:
2141 		if (adev->ip_versions[SDMA0_HWIP][0] != IP_VERSION(4, 2, 0))
2142 			amdgpu_fence_process(&adev->sdma.instance[instance].page);
2143 		break;
2144 	}
2145 	return 0;
2146 }
2147 
2148 static int sdma_v4_0_process_ras_data_cb(struct amdgpu_device *adev,
2149 		void *err_data,
2150 		struct amdgpu_iv_entry *entry)
2151 {
2152 	int instance;
2153 
2154 	/* When “Full RAS” is enabled, the per-IP interrupt sources should
2155 	 * be disabled and the driver should only look for the aggregated
2156 	 * interrupt via sync flood
2157 	 */
2158 	if (amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__GFX))
2159 		goto out;
2160 
2161 	instance = sdma_v4_0_irq_id_to_seq(entry->client_id);
2162 	if (instance < 0)
2163 		goto out;
2164 
2165 	amdgpu_sdma_process_ras_data_cb(adev, err_data, entry);
2166 
2167 out:
2168 	return AMDGPU_RAS_SUCCESS;
2169 }
2170 
2171 static int sdma_v4_0_process_illegal_inst_irq(struct amdgpu_device *adev,
2172 					      struct amdgpu_irq_src *source,
2173 					      struct amdgpu_iv_entry *entry)
2174 {
2175 	int instance;
2176 
2177 	DRM_ERROR("Illegal instruction in SDMA command stream\n");
2178 
2179 	instance = sdma_v4_0_irq_id_to_seq(entry->client_id);
2180 	if (instance < 0)
2181 		return 0;
2182 
2183 	switch (entry->ring_id) {
2184 	case 0:
2185 		drm_sched_fault(&adev->sdma.instance[instance].ring.sched);
2186 		break;
2187 	}
2188 	return 0;
2189 }
2190 
2191 static int sdma_v4_0_set_ecc_irq_state(struct amdgpu_device *adev,
2192 					struct amdgpu_irq_src *source,
2193 					unsigned type,
2194 					enum amdgpu_interrupt_state state)
2195 {
2196 	u32 sdma_edc_config;
2197 
2198 	sdma_edc_config = RREG32_SDMA(type, mmSDMA0_EDC_CONFIG);
2199 	sdma_edc_config = REG_SET_FIELD(sdma_edc_config, SDMA0_EDC_CONFIG, ECC_INT_ENABLE,
2200 		       state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0);
2201 	WREG32_SDMA(type, mmSDMA0_EDC_CONFIG, sdma_edc_config);
2202 
2203 	return 0;
2204 }
2205 
2206 static int sdma_v4_0_print_iv_entry(struct amdgpu_device *adev,
2207 					      struct amdgpu_iv_entry *entry)
2208 {
2209 	int instance;
2210 	struct amdgpu_task_info task_info;
2211 	u64 addr;
2212 
2213 	instance = sdma_v4_0_irq_id_to_seq(entry->client_id);
2214 	if (instance < 0 || instance >= adev->sdma.num_instances) {
2215 		dev_err(adev->dev, "sdma instance invalid %d\n", instance);
2216 		return -EINVAL;
2217 	}
2218 
2219 	addr = (u64)entry->src_data[0] << 12;
2220 	addr |= ((u64)entry->src_data[1] & 0xf) << 44;
2221 
2222 	memset(&task_info, 0, sizeof(struct amdgpu_task_info));
2223 	amdgpu_vm_get_task_info(adev, entry->pasid, &task_info);
2224 
2225 	dev_dbg_ratelimited(adev->dev,
2226 		   "[sdma%d] address:0x%016llx src_id:%u ring:%u vmid:%u "
2227 		   "pasid:%u, for process %s pid %d thread %s pid %d\n",
2228 		   instance, addr, entry->src_id, entry->ring_id, entry->vmid,
2229 		   entry->pasid, task_info.process_name, task_info.tgid,
2230 		   task_info.task_name, task_info.pid);
2231 	return 0;
2232 }
2233 
2234 static int sdma_v4_0_process_vm_hole_irq(struct amdgpu_device *adev,
2235 					      struct amdgpu_irq_src *source,
2236 					      struct amdgpu_iv_entry *entry)
2237 {
2238 	dev_dbg_ratelimited(adev->dev, "MC or SEM address in VM hole\n");
2239 	sdma_v4_0_print_iv_entry(adev, entry);
2240 	return 0;
2241 }
2242 
2243 static int sdma_v4_0_process_doorbell_invalid_irq(struct amdgpu_device *adev,
2244 					      struct amdgpu_irq_src *source,
2245 					      struct amdgpu_iv_entry *entry)
2246 {
2247 	dev_dbg_ratelimited(adev->dev, "SDMA received a doorbell from BIF with byte_enable !=0xff\n");
2248 	sdma_v4_0_print_iv_entry(adev, entry);
2249 	return 0;
2250 }
2251 
2252 static int sdma_v4_0_process_pool_timeout_irq(struct amdgpu_device *adev,
2253 					      struct amdgpu_irq_src *source,
2254 					      struct amdgpu_iv_entry *entry)
2255 {
2256 	dev_dbg_ratelimited(adev->dev,
2257 		"Polling register/memory timeout executing POLL_REG/MEM with finite timer\n");
2258 	sdma_v4_0_print_iv_entry(adev, entry);
2259 	return 0;
2260 }
2261 
2262 static int sdma_v4_0_process_srbm_write_irq(struct amdgpu_device *adev,
2263 					      struct amdgpu_irq_src *source,
2264 					      struct amdgpu_iv_entry *entry)
2265 {
2266 	dev_dbg_ratelimited(adev->dev,
2267 		"SDMA gets an Register Write SRBM_WRITE command in non-privilege command buffer\n");
2268 	sdma_v4_0_print_iv_entry(adev, entry);
2269 	return 0;
2270 }
2271 
2272 static void sdma_v4_0_update_medium_grain_clock_gating(
2273 		struct amdgpu_device *adev,
2274 		bool enable)
2275 {
2276 	uint32_t data, def;
2277 	int i;
2278 
2279 	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_SDMA_MGCG)) {
2280 		for (i = 0; i < adev->sdma.num_instances; i++) {
2281 			def = data = RREG32_SDMA(i, mmSDMA0_CLK_CTRL);
2282 			data &= ~(SDMA0_CLK_CTRL__SOFT_OVERRIDE7_MASK |
2283 				  SDMA0_CLK_CTRL__SOFT_OVERRIDE6_MASK |
2284 				  SDMA0_CLK_CTRL__SOFT_OVERRIDE5_MASK |
2285 				  SDMA0_CLK_CTRL__SOFT_OVERRIDE4_MASK |
2286 				  SDMA0_CLK_CTRL__SOFT_OVERRIDE3_MASK |
2287 				  SDMA0_CLK_CTRL__SOFT_OVERRIDE2_MASK |
2288 				  SDMA0_CLK_CTRL__SOFT_OVERRIDE1_MASK |
2289 				  SDMA0_CLK_CTRL__SOFT_OVERRIDE0_MASK);
2290 			if (def != data)
2291 				WREG32_SDMA(i, mmSDMA0_CLK_CTRL, data);
2292 		}
2293 	} else {
2294 		for (i = 0; i < adev->sdma.num_instances; i++) {
2295 			def = data = RREG32_SDMA(i, mmSDMA0_CLK_CTRL);
2296 			data |= (SDMA0_CLK_CTRL__SOFT_OVERRIDE7_MASK |
2297 				 SDMA0_CLK_CTRL__SOFT_OVERRIDE6_MASK |
2298 				 SDMA0_CLK_CTRL__SOFT_OVERRIDE5_MASK |
2299 				 SDMA0_CLK_CTRL__SOFT_OVERRIDE4_MASK |
2300 				 SDMA0_CLK_CTRL__SOFT_OVERRIDE3_MASK |
2301 				 SDMA0_CLK_CTRL__SOFT_OVERRIDE2_MASK |
2302 				 SDMA0_CLK_CTRL__SOFT_OVERRIDE1_MASK |
2303 				 SDMA0_CLK_CTRL__SOFT_OVERRIDE0_MASK);
2304 			if (def != data)
2305 				WREG32_SDMA(i, mmSDMA0_CLK_CTRL, data);
2306 		}
2307 	}
2308 }
2309 
2310 
2311 static void sdma_v4_0_update_medium_grain_light_sleep(
2312 		struct amdgpu_device *adev,
2313 		bool enable)
2314 {
2315 	uint32_t data, def;
2316 	int i;
2317 
2318 	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_SDMA_LS)) {
2319 		for (i = 0; i < adev->sdma.num_instances; i++) {
2320 			/* 1-not override: enable sdma mem light sleep */
2321 			def = data = RREG32_SDMA(0, mmSDMA0_POWER_CNTL);
2322 			data |= SDMA0_POWER_CNTL__MEM_POWER_OVERRIDE_MASK;
2323 			if (def != data)
2324 				WREG32_SDMA(0, mmSDMA0_POWER_CNTL, data);
2325 		}
2326 	} else {
2327 		for (i = 0; i < adev->sdma.num_instances; i++) {
2328 		/* 0-override:disable sdma mem light sleep */
2329 			def = data = RREG32_SDMA(0, mmSDMA0_POWER_CNTL);
2330 			data &= ~SDMA0_POWER_CNTL__MEM_POWER_OVERRIDE_MASK;
2331 			if (def != data)
2332 				WREG32_SDMA(0, mmSDMA0_POWER_CNTL, data);
2333 		}
2334 	}
2335 }
2336 
2337 static int sdma_v4_0_set_clockgating_state(void *handle,
2338 					  enum amd_clockgating_state state)
2339 {
2340 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2341 
2342 	if (amdgpu_sriov_vf(adev))
2343 		return 0;
2344 
2345 	sdma_v4_0_update_medium_grain_clock_gating(adev,
2346 			state == AMD_CG_STATE_GATE);
2347 	sdma_v4_0_update_medium_grain_light_sleep(adev,
2348 			state == AMD_CG_STATE_GATE);
2349 	return 0;
2350 }
2351 
2352 static int sdma_v4_0_set_powergating_state(void *handle,
2353 					  enum amd_powergating_state state)
2354 {
2355 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2356 
2357 	switch (adev->ip_versions[SDMA0_HWIP][0]) {
2358 	case IP_VERSION(4, 1, 0):
2359 	case IP_VERSION(4, 1, 1):
2360 	case IP_VERSION(4, 1, 2):
2361 		sdma_v4_1_update_power_gating(adev,
2362 				state == AMD_PG_STATE_GATE);
2363 		break;
2364 	default:
2365 		break;
2366 	}
2367 
2368 	return 0;
2369 }
2370 
2371 static void sdma_v4_0_get_clockgating_state(void *handle, u64 *flags)
2372 {
2373 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2374 	int data;
2375 
2376 	if (amdgpu_sriov_vf(adev))
2377 		*flags = 0;
2378 
2379 	/* AMD_CG_SUPPORT_SDMA_MGCG */
2380 	data = RREG32(SOC15_REG_OFFSET(SDMA0, 0, mmSDMA0_CLK_CTRL));
2381 	if (!(data & SDMA0_CLK_CTRL__SOFT_OVERRIDE7_MASK))
2382 		*flags |= AMD_CG_SUPPORT_SDMA_MGCG;
2383 
2384 	/* AMD_CG_SUPPORT_SDMA_LS */
2385 	data = RREG32(SOC15_REG_OFFSET(SDMA0, 0, mmSDMA0_POWER_CNTL));
2386 	if (data & SDMA0_POWER_CNTL__MEM_POWER_OVERRIDE_MASK)
2387 		*flags |= AMD_CG_SUPPORT_SDMA_LS;
2388 }
2389 
2390 const struct amd_ip_funcs sdma_v4_0_ip_funcs = {
2391 	.name = "sdma_v4_0",
2392 	.early_init = sdma_v4_0_early_init,
2393 	.late_init = sdma_v4_0_late_init,
2394 	.sw_init = sdma_v4_0_sw_init,
2395 	.sw_fini = sdma_v4_0_sw_fini,
2396 	.hw_init = sdma_v4_0_hw_init,
2397 	.hw_fini = sdma_v4_0_hw_fini,
2398 	.suspend = sdma_v4_0_suspend,
2399 	.resume = sdma_v4_0_resume,
2400 	.is_idle = sdma_v4_0_is_idle,
2401 	.wait_for_idle = sdma_v4_0_wait_for_idle,
2402 	.soft_reset = sdma_v4_0_soft_reset,
2403 	.set_clockgating_state = sdma_v4_0_set_clockgating_state,
2404 	.set_powergating_state = sdma_v4_0_set_powergating_state,
2405 	.get_clockgating_state = sdma_v4_0_get_clockgating_state,
2406 };
2407 
2408 static const struct amdgpu_ring_funcs sdma_v4_0_ring_funcs = {
2409 	.type = AMDGPU_RING_TYPE_SDMA,
2410 	.align_mask = 0xf,
2411 	.nop = SDMA_PKT_NOP_HEADER_OP(SDMA_OP_NOP),
2412 	.support_64bit_ptrs = true,
2413 	.secure_submission_supported = true,
2414 	.vmhub = AMDGPU_MMHUB_0,
2415 	.get_rptr = sdma_v4_0_ring_get_rptr,
2416 	.get_wptr = sdma_v4_0_ring_get_wptr,
2417 	.set_wptr = sdma_v4_0_ring_set_wptr,
2418 	.emit_frame_size =
2419 		6 + /* sdma_v4_0_ring_emit_hdp_flush */
2420 		3 + /* hdp invalidate */
2421 		6 + /* sdma_v4_0_ring_emit_pipeline_sync */
2422 		/* sdma_v4_0_ring_emit_vm_flush */
2423 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
2424 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 6 +
2425 		10 + 10 + 10, /* sdma_v4_0_ring_emit_fence x3 for user fence, vm fence */
2426 	.emit_ib_size = 7 + 6, /* sdma_v4_0_ring_emit_ib */
2427 	.emit_ib = sdma_v4_0_ring_emit_ib,
2428 	.emit_fence = sdma_v4_0_ring_emit_fence,
2429 	.emit_pipeline_sync = sdma_v4_0_ring_emit_pipeline_sync,
2430 	.emit_vm_flush = sdma_v4_0_ring_emit_vm_flush,
2431 	.emit_hdp_flush = sdma_v4_0_ring_emit_hdp_flush,
2432 	.test_ring = sdma_v4_0_ring_test_ring,
2433 	.test_ib = sdma_v4_0_ring_test_ib,
2434 	.insert_nop = sdma_v4_0_ring_insert_nop,
2435 	.pad_ib = sdma_v4_0_ring_pad_ib,
2436 	.emit_wreg = sdma_v4_0_ring_emit_wreg,
2437 	.emit_reg_wait = sdma_v4_0_ring_emit_reg_wait,
2438 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
2439 };
2440 
2441 /*
2442  * On Arcturus, SDMA instance 5~7 has a different vmhub type(AMDGPU_MMHUB_1).
2443  * So create a individual constant ring_funcs for those instances.
2444  */
2445 static const struct amdgpu_ring_funcs sdma_v4_0_ring_funcs_2nd_mmhub = {
2446 	.type = AMDGPU_RING_TYPE_SDMA,
2447 	.align_mask = 0xf,
2448 	.nop = SDMA_PKT_NOP_HEADER_OP(SDMA_OP_NOP),
2449 	.support_64bit_ptrs = true,
2450 	.secure_submission_supported = true,
2451 	.vmhub = AMDGPU_MMHUB_1,
2452 	.get_rptr = sdma_v4_0_ring_get_rptr,
2453 	.get_wptr = sdma_v4_0_ring_get_wptr,
2454 	.set_wptr = sdma_v4_0_ring_set_wptr,
2455 	.emit_frame_size =
2456 		6 + /* sdma_v4_0_ring_emit_hdp_flush */
2457 		3 + /* hdp invalidate */
2458 		6 + /* sdma_v4_0_ring_emit_pipeline_sync */
2459 		/* sdma_v4_0_ring_emit_vm_flush */
2460 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
2461 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 6 +
2462 		10 + 10 + 10, /* sdma_v4_0_ring_emit_fence x3 for user fence, vm fence */
2463 	.emit_ib_size = 7 + 6, /* sdma_v4_0_ring_emit_ib */
2464 	.emit_ib = sdma_v4_0_ring_emit_ib,
2465 	.emit_fence = sdma_v4_0_ring_emit_fence,
2466 	.emit_pipeline_sync = sdma_v4_0_ring_emit_pipeline_sync,
2467 	.emit_vm_flush = sdma_v4_0_ring_emit_vm_flush,
2468 	.emit_hdp_flush = sdma_v4_0_ring_emit_hdp_flush,
2469 	.test_ring = sdma_v4_0_ring_test_ring,
2470 	.test_ib = sdma_v4_0_ring_test_ib,
2471 	.insert_nop = sdma_v4_0_ring_insert_nop,
2472 	.pad_ib = sdma_v4_0_ring_pad_ib,
2473 	.emit_wreg = sdma_v4_0_ring_emit_wreg,
2474 	.emit_reg_wait = sdma_v4_0_ring_emit_reg_wait,
2475 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
2476 };
2477 
2478 static const struct amdgpu_ring_funcs sdma_v4_0_page_ring_funcs = {
2479 	.type = AMDGPU_RING_TYPE_SDMA,
2480 	.align_mask = 0xf,
2481 	.nop = SDMA_PKT_NOP_HEADER_OP(SDMA_OP_NOP),
2482 	.support_64bit_ptrs = true,
2483 	.secure_submission_supported = true,
2484 	.vmhub = AMDGPU_MMHUB_0,
2485 	.get_rptr = sdma_v4_0_ring_get_rptr,
2486 	.get_wptr = sdma_v4_0_page_ring_get_wptr,
2487 	.set_wptr = sdma_v4_0_page_ring_set_wptr,
2488 	.emit_frame_size =
2489 		6 + /* sdma_v4_0_ring_emit_hdp_flush */
2490 		3 + /* hdp invalidate */
2491 		6 + /* sdma_v4_0_ring_emit_pipeline_sync */
2492 		/* sdma_v4_0_ring_emit_vm_flush */
2493 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
2494 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 6 +
2495 		10 + 10 + 10, /* sdma_v4_0_ring_emit_fence x3 for user fence, vm fence */
2496 	.emit_ib_size = 7 + 6, /* sdma_v4_0_ring_emit_ib */
2497 	.emit_ib = sdma_v4_0_ring_emit_ib,
2498 	.emit_fence = sdma_v4_0_ring_emit_fence,
2499 	.emit_pipeline_sync = sdma_v4_0_ring_emit_pipeline_sync,
2500 	.emit_vm_flush = sdma_v4_0_ring_emit_vm_flush,
2501 	.emit_hdp_flush = sdma_v4_0_ring_emit_hdp_flush,
2502 	.test_ring = sdma_v4_0_ring_test_ring,
2503 	.test_ib = sdma_v4_0_ring_test_ib,
2504 	.insert_nop = sdma_v4_0_ring_insert_nop,
2505 	.pad_ib = sdma_v4_0_ring_pad_ib,
2506 	.emit_wreg = sdma_v4_0_ring_emit_wreg,
2507 	.emit_reg_wait = sdma_v4_0_ring_emit_reg_wait,
2508 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
2509 };
2510 
2511 static const struct amdgpu_ring_funcs sdma_v4_0_page_ring_funcs_2nd_mmhub = {
2512 	.type = AMDGPU_RING_TYPE_SDMA,
2513 	.align_mask = 0xf,
2514 	.nop = SDMA_PKT_NOP_HEADER_OP(SDMA_OP_NOP),
2515 	.support_64bit_ptrs = true,
2516 	.secure_submission_supported = true,
2517 	.vmhub = AMDGPU_MMHUB_1,
2518 	.get_rptr = sdma_v4_0_ring_get_rptr,
2519 	.get_wptr = sdma_v4_0_page_ring_get_wptr,
2520 	.set_wptr = sdma_v4_0_page_ring_set_wptr,
2521 	.emit_frame_size =
2522 		6 + /* sdma_v4_0_ring_emit_hdp_flush */
2523 		3 + /* hdp invalidate */
2524 		6 + /* sdma_v4_0_ring_emit_pipeline_sync */
2525 		/* sdma_v4_0_ring_emit_vm_flush */
2526 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
2527 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 6 +
2528 		10 + 10 + 10, /* sdma_v4_0_ring_emit_fence x3 for user fence, vm fence */
2529 	.emit_ib_size = 7 + 6, /* sdma_v4_0_ring_emit_ib */
2530 	.emit_ib = sdma_v4_0_ring_emit_ib,
2531 	.emit_fence = sdma_v4_0_ring_emit_fence,
2532 	.emit_pipeline_sync = sdma_v4_0_ring_emit_pipeline_sync,
2533 	.emit_vm_flush = sdma_v4_0_ring_emit_vm_flush,
2534 	.emit_hdp_flush = sdma_v4_0_ring_emit_hdp_flush,
2535 	.test_ring = sdma_v4_0_ring_test_ring,
2536 	.test_ib = sdma_v4_0_ring_test_ib,
2537 	.insert_nop = sdma_v4_0_ring_insert_nop,
2538 	.pad_ib = sdma_v4_0_ring_pad_ib,
2539 	.emit_wreg = sdma_v4_0_ring_emit_wreg,
2540 	.emit_reg_wait = sdma_v4_0_ring_emit_reg_wait,
2541 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
2542 };
2543 
2544 static void sdma_v4_0_set_ring_funcs(struct amdgpu_device *adev)
2545 {
2546 	int i;
2547 
2548 	for (i = 0; i < adev->sdma.num_instances; i++) {
2549 		if (adev->ip_versions[SDMA0_HWIP][0] == IP_VERSION(4, 2, 2) && i >= 5)
2550 			adev->sdma.instance[i].ring.funcs =
2551 					&sdma_v4_0_ring_funcs_2nd_mmhub;
2552 		else
2553 			adev->sdma.instance[i].ring.funcs =
2554 					&sdma_v4_0_ring_funcs;
2555 		adev->sdma.instance[i].ring.me = i;
2556 		if (adev->sdma.has_page_queue) {
2557 			if (adev->ip_versions[SDMA0_HWIP][0] == IP_VERSION(4, 2, 2) && i >= 5)
2558 				adev->sdma.instance[i].page.funcs =
2559 					&sdma_v4_0_page_ring_funcs_2nd_mmhub;
2560 			else
2561 				adev->sdma.instance[i].page.funcs =
2562 					&sdma_v4_0_page_ring_funcs;
2563 			adev->sdma.instance[i].page.me = i;
2564 		}
2565 	}
2566 }
2567 
2568 static const struct amdgpu_irq_src_funcs sdma_v4_0_trap_irq_funcs = {
2569 	.set = sdma_v4_0_set_trap_irq_state,
2570 	.process = sdma_v4_0_process_trap_irq,
2571 };
2572 
2573 static const struct amdgpu_irq_src_funcs sdma_v4_0_illegal_inst_irq_funcs = {
2574 	.process = sdma_v4_0_process_illegal_inst_irq,
2575 };
2576 
2577 static const struct amdgpu_irq_src_funcs sdma_v4_0_ecc_irq_funcs = {
2578 	.set = sdma_v4_0_set_ecc_irq_state,
2579 	.process = amdgpu_sdma_process_ecc_irq,
2580 };
2581 
2582 static const struct amdgpu_irq_src_funcs sdma_v4_0_vm_hole_irq_funcs = {
2583 	.process = sdma_v4_0_process_vm_hole_irq,
2584 };
2585 
2586 static const struct amdgpu_irq_src_funcs sdma_v4_0_doorbell_invalid_irq_funcs = {
2587 	.process = sdma_v4_0_process_doorbell_invalid_irq,
2588 };
2589 
2590 static const struct amdgpu_irq_src_funcs sdma_v4_0_pool_timeout_irq_funcs = {
2591 	.process = sdma_v4_0_process_pool_timeout_irq,
2592 };
2593 
2594 static const struct amdgpu_irq_src_funcs sdma_v4_0_srbm_write_irq_funcs = {
2595 	.process = sdma_v4_0_process_srbm_write_irq,
2596 };
2597 
2598 static void sdma_v4_0_set_irq_funcs(struct amdgpu_device *adev)
2599 {
2600 	adev->sdma.trap_irq.num_types = adev->sdma.num_instances;
2601 	adev->sdma.ecc_irq.num_types = adev->sdma.num_instances;
2602 	/*For Arcturus and Aldebaran, add another 4 irq handler*/
2603 	switch (adev->sdma.num_instances) {
2604 	case 5:
2605 	case 8:
2606 		adev->sdma.vm_hole_irq.num_types = adev->sdma.num_instances;
2607 		adev->sdma.doorbell_invalid_irq.num_types = adev->sdma.num_instances;
2608 		adev->sdma.pool_timeout_irq.num_types = adev->sdma.num_instances;
2609 		adev->sdma.srbm_write_irq.num_types = adev->sdma.num_instances;
2610 		break;
2611 	default:
2612 		break;
2613 	}
2614 	adev->sdma.trap_irq.funcs = &sdma_v4_0_trap_irq_funcs;
2615 	adev->sdma.illegal_inst_irq.funcs = &sdma_v4_0_illegal_inst_irq_funcs;
2616 	adev->sdma.ecc_irq.funcs = &sdma_v4_0_ecc_irq_funcs;
2617 	adev->sdma.vm_hole_irq.funcs = &sdma_v4_0_vm_hole_irq_funcs;
2618 	adev->sdma.doorbell_invalid_irq.funcs = &sdma_v4_0_doorbell_invalid_irq_funcs;
2619 	adev->sdma.pool_timeout_irq.funcs = &sdma_v4_0_pool_timeout_irq_funcs;
2620 	adev->sdma.srbm_write_irq.funcs = &sdma_v4_0_srbm_write_irq_funcs;
2621 }
2622 
2623 /**
2624  * sdma_v4_0_emit_copy_buffer - copy buffer using the sDMA engine
2625  *
2626  * @ib: indirect buffer to copy to
2627  * @src_offset: src GPU address
2628  * @dst_offset: dst GPU address
2629  * @byte_count: number of bytes to xfer
2630  * @tmz: if a secure copy should be used
2631  *
2632  * Copy GPU buffers using the DMA engine (VEGA10/12).
2633  * Used by the amdgpu ttm implementation to move pages if
2634  * registered as the asic copy callback.
2635  */
2636 static void sdma_v4_0_emit_copy_buffer(struct amdgpu_ib *ib,
2637 				       uint64_t src_offset,
2638 				       uint64_t dst_offset,
2639 				       uint32_t byte_count,
2640 				       bool tmz)
2641 {
2642 	ib->ptr[ib->length_dw++] = SDMA_PKT_HEADER_OP(SDMA_OP_COPY) |
2643 		SDMA_PKT_HEADER_SUB_OP(SDMA_SUBOP_COPY_LINEAR) |
2644 		SDMA_PKT_COPY_LINEAR_HEADER_TMZ(tmz ? 1 : 0);
2645 	ib->ptr[ib->length_dw++] = byte_count - 1;
2646 	ib->ptr[ib->length_dw++] = 0; /* src/dst endian swap */
2647 	ib->ptr[ib->length_dw++] = lower_32_bits(src_offset);
2648 	ib->ptr[ib->length_dw++] = upper_32_bits(src_offset);
2649 	ib->ptr[ib->length_dw++] = lower_32_bits(dst_offset);
2650 	ib->ptr[ib->length_dw++] = upper_32_bits(dst_offset);
2651 }
2652 
2653 /**
2654  * sdma_v4_0_emit_fill_buffer - fill buffer using the sDMA engine
2655  *
2656  * @ib: indirect buffer to copy to
2657  * @src_data: value to write to buffer
2658  * @dst_offset: dst GPU address
2659  * @byte_count: number of bytes to xfer
2660  *
2661  * Fill GPU buffers using the DMA engine (VEGA10/12).
2662  */
2663 static void sdma_v4_0_emit_fill_buffer(struct amdgpu_ib *ib,
2664 				       uint32_t src_data,
2665 				       uint64_t dst_offset,
2666 				       uint32_t byte_count)
2667 {
2668 	ib->ptr[ib->length_dw++] = SDMA_PKT_HEADER_OP(SDMA_OP_CONST_FILL);
2669 	ib->ptr[ib->length_dw++] = lower_32_bits(dst_offset);
2670 	ib->ptr[ib->length_dw++] = upper_32_bits(dst_offset);
2671 	ib->ptr[ib->length_dw++] = src_data;
2672 	ib->ptr[ib->length_dw++] = byte_count - 1;
2673 }
2674 
2675 static const struct amdgpu_buffer_funcs sdma_v4_0_buffer_funcs = {
2676 	.copy_max_bytes = 0x400000,
2677 	.copy_num_dw = 7,
2678 	.emit_copy_buffer = sdma_v4_0_emit_copy_buffer,
2679 
2680 	.fill_max_bytes = 0x400000,
2681 	.fill_num_dw = 5,
2682 	.emit_fill_buffer = sdma_v4_0_emit_fill_buffer,
2683 };
2684 
2685 static void sdma_v4_0_set_buffer_funcs(struct amdgpu_device *adev)
2686 {
2687 	adev->mman.buffer_funcs = &sdma_v4_0_buffer_funcs;
2688 	if (adev->sdma.has_page_queue)
2689 		adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].page;
2690 	else
2691 		adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].ring;
2692 }
2693 
2694 static const struct amdgpu_vm_pte_funcs sdma_v4_0_vm_pte_funcs = {
2695 	.copy_pte_num_dw = 7,
2696 	.copy_pte = sdma_v4_0_vm_copy_pte,
2697 
2698 	.write_pte = sdma_v4_0_vm_write_pte,
2699 	.set_pte_pde = sdma_v4_0_vm_set_pte_pde,
2700 };
2701 
2702 static void sdma_v4_0_set_vm_pte_funcs(struct amdgpu_device *adev)
2703 {
2704 	struct drm_gpu_scheduler *sched;
2705 	unsigned i;
2706 
2707 	adev->vm_manager.vm_pte_funcs = &sdma_v4_0_vm_pte_funcs;
2708 	for (i = 0; i < adev->sdma.num_instances; i++) {
2709 		if (adev->sdma.has_page_queue)
2710 			sched = &adev->sdma.instance[i].page.sched;
2711 		else
2712 			sched = &adev->sdma.instance[i].ring.sched;
2713 		adev->vm_manager.vm_pte_scheds[i] = sched;
2714 	}
2715 	adev->vm_manager.vm_pte_num_scheds = adev->sdma.num_instances;
2716 }
2717 
2718 static void sdma_v4_0_get_ras_error_count(uint32_t value,
2719 					uint32_t instance,
2720 					uint32_t *sec_count)
2721 {
2722 	uint32_t i;
2723 	uint32_t sec_cnt;
2724 
2725 	/* double bits error (multiple bits) error detection is not supported */
2726 	for (i = 0; i < ARRAY_SIZE(sdma_v4_0_ras_fields); i++) {
2727 		/* the SDMA_EDC_COUNTER register in each sdma instance
2728 		 * shares the same sed shift_mask
2729 		 * */
2730 		sec_cnt = (value &
2731 			sdma_v4_0_ras_fields[i].sec_count_mask) >>
2732 			sdma_v4_0_ras_fields[i].sec_count_shift;
2733 		if (sec_cnt) {
2734 			DRM_INFO("Detected %s in SDMA%d, SED %d\n",
2735 				sdma_v4_0_ras_fields[i].name,
2736 				instance, sec_cnt);
2737 			*sec_count += sec_cnt;
2738 		}
2739 	}
2740 }
2741 
2742 static int sdma_v4_0_query_ras_error_count_by_instance(struct amdgpu_device *adev,
2743 			uint32_t instance, void *ras_error_status)
2744 {
2745 	struct ras_err_data *err_data = (struct ras_err_data *)ras_error_status;
2746 	uint32_t sec_count = 0;
2747 	uint32_t reg_value = 0;
2748 
2749 	reg_value = RREG32_SDMA(instance, mmSDMA0_EDC_COUNTER);
2750 	/* double bit error is not supported */
2751 	if (reg_value)
2752 		sdma_v4_0_get_ras_error_count(reg_value,
2753 				instance, &sec_count);
2754 	/* err_data->ce_count should be initialized to 0
2755 	 * before calling into this function */
2756 	err_data->ce_count += sec_count;
2757 	/* double bit error is not supported
2758 	 * set ue count to 0 */
2759 	err_data->ue_count = 0;
2760 
2761 	return 0;
2762 };
2763 
2764 static void sdma_v4_0_query_ras_error_count(struct amdgpu_device *adev,  void *ras_error_status)
2765 {
2766 	int i = 0;
2767 
2768 	for (i = 0; i < adev->sdma.num_instances; i++) {
2769 		if (sdma_v4_0_query_ras_error_count_by_instance(adev, i, ras_error_status)) {
2770 			dev_err(adev->dev, "Query ras error count failed in SDMA%d\n", i);
2771 			return;
2772 		}
2773 	}
2774 }
2775 
2776 static void sdma_v4_0_reset_ras_error_count(struct amdgpu_device *adev)
2777 {
2778 	int i;
2779 
2780 	/* read back edc counter registers to clear the counters */
2781 	if (amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__SDMA)) {
2782 		for (i = 0; i < adev->sdma.num_instances; i++)
2783 			RREG32_SDMA(i, mmSDMA0_EDC_COUNTER);
2784 	}
2785 }
2786 
2787 const struct amdgpu_ras_block_hw_ops sdma_v4_0_ras_hw_ops = {
2788 	.query_ras_error_count = sdma_v4_0_query_ras_error_count,
2789 	.reset_ras_error_count = sdma_v4_0_reset_ras_error_count,
2790 };
2791 
2792 static struct amdgpu_sdma_ras sdma_v4_0_ras = {
2793 	.ras_block = {
2794 		.hw_ops = &sdma_v4_0_ras_hw_ops,
2795 		.ras_cb = sdma_v4_0_process_ras_data_cb,
2796 	},
2797 };
2798 
2799 static void sdma_v4_0_set_ras_funcs(struct amdgpu_device *adev)
2800 {
2801 	switch (adev->ip_versions[SDMA0_HWIP][0]) {
2802 	case IP_VERSION(4, 2, 0):
2803 	case IP_VERSION(4, 2, 2):
2804 		adev->sdma.ras = &sdma_v4_0_ras;
2805 		break;
2806 	case IP_VERSION(4, 4, 0):
2807 		adev->sdma.ras = &sdma_v4_4_ras;
2808 		break;
2809 	default:
2810 		break;
2811 	}
2812 
2813 	if (adev->sdma.ras) {
2814 		amdgpu_ras_register_ras_block(adev, &adev->sdma.ras->ras_block);
2815 
2816 		strcpy(adev->sdma.ras->ras_block.ras_comm.name, "sdma");
2817 		adev->sdma.ras->ras_block.ras_comm.block = AMDGPU_RAS_BLOCK__SDMA;
2818 		adev->sdma.ras->ras_block.ras_comm.type = AMDGPU_RAS_ERROR__MULTI_UNCORRECTABLE;
2819 		adev->sdma.ras_if = &adev->sdma.ras->ras_block.ras_comm;
2820 
2821 		/* If don't define special ras_late_init function, use default ras_late_init */
2822 		if (!adev->sdma.ras->ras_block.ras_late_init)
2823 			adev->sdma.ras->ras_block.ras_late_init = amdgpu_sdma_ras_late_init;
2824 
2825 		/* If not defined special ras_cb function, use default ras_cb */
2826 		if (!adev->sdma.ras->ras_block.ras_cb)
2827 			adev->sdma.ras->ras_block.ras_cb = amdgpu_sdma_process_ras_data_cb;
2828 	}
2829 }
2830 
2831 const struct amdgpu_ip_block_version sdma_v4_0_ip_block = {
2832 	.type = AMD_IP_BLOCK_TYPE_SDMA,
2833 	.major = 4,
2834 	.minor = 0,
2835 	.rev = 0,
2836 	.funcs = &sdma_v4_0_ip_funcs,
2837 };
2838