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 
sdma_v4_0_get_reg_offset(struct amdgpu_device * adev,u32 instance,u32 offset)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 
sdma_v4_0_seq_to_irq_id(int seq_num)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 
sdma_v4_0_irq_id_to_seq(unsigned client_id)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 
sdma_v4_0_init_golden_registers(struct amdgpu_device * adev)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 
sdma_v4_0_setup_ulv(struct amdgpu_device * adev)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 
sdma_v4_0_init_inst_ctx(struct amdgpu_sdma_instance * sdma_inst)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 
sdma_v4_0_destroy_inst_ctx(struct amdgpu_device * adev)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
sdma_v4_0_init_microcode(struct amdgpu_device * adev)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  */
sdma_v4_0_ring_get_rptr(struct amdgpu_ring * ring)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  */
sdma_v4_0_ring_get_wptr(struct amdgpu_ring * ring)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  */
sdma_v4_0_ring_set_wptr(struct amdgpu_ring * ring)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  */
sdma_v4_0_page_ring_get_wptr(struct amdgpu_ring * ring)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  */
sdma_v4_0_page_ring_set_wptr(struct amdgpu_ring * ring)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 
sdma_v4_0_ring_insert_nop(struct amdgpu_ring * ring,uint32_t count)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  */
sdma_v4_0_ring_emit_ib(struct amdgpu_ring * ring,struct amdgpu_job * job,struct amdgpu_ib * ib,uint32_t flags)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 
sdma_v4_0_wait_reg_mem(struct amdgpu_ring * ring,int mem_space,int hdp,uint32_t addr0,uint32_t addr1,uint32_t ref,uint32_t mask,uint32_t inv)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  */
sdma_v4_0_ring_emit_hdp_flush(struct amdgpu_ring * ring)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  */
sdma_v4_0_ring_emit_fence(struct amdgpu_ring * ring,u64 addr,u64 seq,unsigned flags)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  */
sdma_v4_0_gfx_stop(struct amdgpu_device * adev)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  */
sdma_v4_0_rlc_stop(struct amdgpu_device * adev)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  */
sdma_v4_0_page_stop(struct amdgpu_device * adev)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  */
sdma_v4_0_ctx_switch_enable(struct amdgpu_device * adev,bool enable)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  */
sdma_v4_0_enable(struct amdgpu_device * adev,bool enable)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  */
sdma_v4_0_rb_cntl(struct amdgpu_ring * ring,uint32_t rb_cntl)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  */
sdma_v4_0_gfx_resume(struct amdgpu_device * adev,unsigned int i)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  */
sdma_v4_0_page_resume(struct amdgpu_device * adev,unsigned int i)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
sdma_v4_1_update_power_gating(struct amdgpu_device * adev,bool enable)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 
sdma_v4_1_init_power_gating(struct amdgpu_device * adev)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 
sdma_v4_0_init_pg(struct amdgpu_device * adev)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  */
sdma_v4_0_rlc_resume(struct amdgpu_device * adev)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  */
sdma_v4_0_load_microcode(struct amdgpu_device * adev)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  */
sdma_v4_0_start(struct amdgpu_device * adev)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 			/* unhalt engine */
1508 			temp = RREG32_SDMA(i, mmSDMA0_F32_CNTL);
1509 			temp = REG_SET_FIELD(temp, SDMA0_F32_CNTL, HALT, 0);
1510 			WREG32_SDMA(i, mmSDMA0_F32_CNTL, temp);
1511 		}
1512 	}
1513 
1514 	if (amdgpu_sriov_vf(adev)) {
1515 		sdma_v4_0_ctx_switch_enable(adev, true);
1516 		sdma_v4_0_enable(adev, true);
1517 	} else {
1518 		r = sdma_v4_0_rlc_resume(adev);
1519 		if (r)
1520 			return r;
1521 	}
1522 
1523 	for (i = 0; i < adev->sdma.num_instances; i++) {
1524 		ring = &adev->sdma.instance[i].ring;
1525 
1526 		r = amdgpu_ring_test_helper(ring);
1527 		if (r)
1528 			return r;
1529 
1530 		if (adev->sdma.has_page_queue) {
1531 			struct amdgpu_ring *page = &adev->sdma.instance[i].page;
1532 
1533 			r = amdgpu_ring_test_helper(page);
1534 			if (r)
1535 				return r;
1536 
1537 			if (adev->mman.buffer_funcs_ring == page)
1538 				amdgpu_ttm_set_buffer_funcs_status(adev, true);
1539 		}
1540 
1541 		if (adev->mman.buffer_funcs_ring == ring)
1542 			amdgpu_ttm_set_buffer_funcs_status(adev, true);
1543 	}
1544 
1545 	return r;
1546 }
1547 
1548 /**
1549  * sdma_v4_0_ring_test_ring - simple async dma engine test
1550  *
1551  * @ring: amdgpu_ring structure holding ring information
1552  *
1553  * Test the DMA engine by writing using it to write an
1554  * value to memory. (VEGA10).
1555  * Returns 0 for success, error for failure.
1556  */
sdma_v4_0_ring_test_ring(struct amdgpu_ring * ring)1557 static int sdma_v4_0_ring_test_ring(struct amdgpu_ring *ring)
1558 {
1559 	struct amdgpu_device *adev = ring->adev;
1560 	unsigned i;
1561 	unsigned index;
1562 	int r;
1563 	u32 tmp;
1564 	u64 gpu_addr;
1565 
1566 	r = amdgpu_device_wb_get(adev, &index);
1567 	if (r)
1568 		return r;
1569 
1570 	gpu_addr = adev->wb.gpu_addr + (index * 4);
1571 	tmp = 0xCAFEDEAD;
1572 	adev->wb.wb[index] = cpu_to_le32(tmp);
1573 
1574 	r = amdgpu_ring_alloc(ring, 5);
1575 	if (r)
1576 		goto error_free_wb;
1577 
1578 	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_WRITE) |
1579 			  SDMA_PKT_HEADER_SUB_OP(SDMA_SUBOP_WRITE_LINEAR));
1580 	amdgpu_ring_write(ring, lower_32_bits(gpu_addr));
1581 	amdgpu_ring_write(ring, upper_32_bits(gpu_addr));
1582 	amdgpu_ring_write(ring, SDMA_PKT_WRITE_UNTILED_DW_3_COUNT(0));
1583 	amdgpu_ring_write(ring, 0xDEADBEEF);
1584 	amdgpu_ring_commit(ring);
1585 
1586 	for (i = 0; i < adev->usec_timeout; i++) {
1587 		tmp = le32_to_cpu(adev->wb.wb[index]);
1588 		if (tmp == 0xDEADBEEF)
1589 			break;
1590 		udelay(1);
1591 	}
1592 
1593 	if (i >= adev->usec_timeout)
1594 		r = -ETIMEDOUT;
1595 
1596 error_free_wb:
1597 	amdgpu_device_wb_free(adev, index);
1598 	return r;
1599 }
1600 
1601 /**
1602  * sdma_v4_0_ring_test_ib - test an IB on the DMA engine
1603  *
1604  * @ring: amdgpu_ring structure holding ring information
1605  * @timeout: timeout value in jiffies, or MAX_SCHEDULE_TIMEOUT
1606  *
1607  * Test a simple IB in the DMA ring (VEGA10).
1608  * Returns 0 on success, error on failure.
1609  */
sdma_v4_0_ring_test_ib(struct amdgpu_ring * ring,long timeout)1610 static int sdma_v4_0_ring_test_ib(struct amdgpu_ring *ring, long timeout)
1611 {
1612 	struct amdgpu_device *adev = ring->adev;
1613 	struct amdgpu_ib ib;
1614 	struct dma_fence *f = NULL;
1615 	unsigned index;
1616 	long r;
1617 	u32 tmp = 0;
1618 	u64 gpu_addr;
1619 
1620 	r = amdgpu_device_wb_get(adev, &index);
1621 	if (r)
1622 		return r;
1623 
1624 	gpu_addr = adev->wb.gpu_addr + (index * 4);
1625 	tmp = 0xCAFEDEAD;
1626 	adev->wb.wb[index] = cpu_to_le32(tmp);
1627 	memset(&ib, 0, sizeof(ib));
1628 	r = amdgpu_ib_get(adev, NULL, 256,
1629 					AMDGPU_IB_POOL_DIRECT, &ib);
1630 	if (r)
1631 		goto err0;
1632 
1633 	ib.ptr[0] = SDMA_PKT_HEADER_OP(SDMA_OP_WRITE) |
1634 		SDMA_PKT_HEADER_SUB_OP(SDMA_SUBOP_WRITE_LINEAR);
1635 	ib.ptr[1] = lower_32_bits(gpu_addr);
1636 	ib.ptr[2] = upper_32_bits(gpu_addr);
1637 	ib.ptr[3] = SDMA_PKT_WRITE_UNTILED_DW_3_COUNT(0);
1638 	ib.ptr[4] = 0xDEADBEEF;
1639 	ib.ptr[5] = SDMA_PKT_NOP_HEADER_OP(SDMA_OP_NOP);
1640 	ib.ptr[6] = SDMA_PKT_NOP_HEADER_OP(SDMA_OP_NOP);
1641 	ib.ptr[7] = SDMA_PKT_NOP_HEADER_OP(SDMA_OP_NOP);
1642 	ib.length_dw = 8;
1643 
1644 	r = amdgpu_ib_schedule(ring, 1, &ib, NULL, &f);
1645 	if (r)
1646 		goto err1;
1647 
1648 	r = dma_fence_wait_timeout(f, false, timeout);
1649 	if (r == 0) {
1650 		r = -ETIMEDOUT;
1651 		goto err1;
1652 	} else if (r < 0) {
1653 		goto err1;
1654 	}
1655 	tmp = le32_to_cpu(adev->wb.wb[index]);
1656 	if (tmp == 0xDEADBEEF)
1657 		r = 0;
1658 	else
1659 		r = -EINVAL;
1660 
1661 err1:
1662 	amdgpu_ib_free(adev, &ib, NULL);
1663 	dma_fence_put(f);
1664 err0:
1665 	amdgpu_device_wb_free(adev, index);
1666 	return r;
1667 }
1668 
1669 
1670 /**
1671  * sdma_v4_0_vm_copy_pte - update PTEs by copying them from the GART
1672  *
1673  * @ib: indirect buffer to fill with commands
1674  * @pe: addr of the page entry
1675  * @src: src addr to copy from
1676  * @count: number of page entries to update
1677  *
1678  * Update PTEs by copying them from the GART using sDMA (VEGA10).
1679  */
sdma_v4_0_vm_copy_pte(struct amdgpu_ib * ib,uint64_t pe,uint64_t src,unsigned count)1680 static void sdma_v4_0_vm_copy_pte(struct amdgpu_ib *ib,
1681 				  uint64_t pe, uint64_t src,
1682 				  unsigned count)
1683 {
1684 	unsigned bytes = count * 8;
1685 
1686 	ib->ptr[ib->length_dw++] = SDMA_PKT_HEADER_OP(SDMA_OP_COPY) |
1687 		SDMA_PKT_HEADER_SUB_OP(SDMA_SUBOP_COPY_LINEAR);
1688 	ib->ptr[ib->length_dw++] = bytes - 1;
1689 	ib->ptr[ib->length_dw++] = 0; /* src/dst endian swap */
1690 	ib->ptr[ib->length_dw++] = lower_32_bits(src);
1691 	ib->ptr[ib->length_dw++] = upper_32_bits(src);
1692 	ib->ptr[ib->length_dw++] = lower_32_bits(pe);
1693 	ib->ptr[ib->length_dw++] = upper_32_bits(pe);
1694 
1695 }
1696 
1697 /**
1698  * sdma_v4_0_vm_write_pte - update PTEs by writing them manually
1699  *
1700  * @ib: indirect buffer to fill with commands
1701  * @pe: addr of the page entry
1702  * @value: dst addr to write into pe
1703  * @count: number of page entries to update
1704  * @incr: increase next addr by incr bytes
1705  *
1706  * Update PTEs by writing them manually using sDMA (VEGA10).
1707  */
sdma_v4_0_vm_write_pte(struct amdgpu_ib * ib,uint64_t pe,uint64_t value,unsigned count,uint32_t incr)1708 static void sdma_v4_0_vm_write_pte(struct amdgpu_ib *ib, uint64_t pe,
1709 				   uint64_t value, unsigned count,
1710 				   uint32_t incr)
1711 {
1712 	unsigned ndw = count * 2;
1713 
1714 	ib->ptr[ib->length_dw++] = SDMA_PKT_HEADER_OP(SDMA_OP_WRITE) |
1715 		SDMA_PKT_HEADER_SUB_OP(SDMA_SUBOP_WRITE_LINEAR);
1716 	ib->ptr[ib->length_dw++] = lower_32_bits(pe);
1717 	ib->ptr[ib->length_dw++] = upper_32_bits(pe);
1718 	ib->ptr[ib->length_dw++] = ndw - 1;
1719 	for (; ndw > 0; ndw -= 2) {
1720 		ib->ptr[ib->length_dw++] = lower_32_bits(value);
1721 		ib->ptr[ib->length_dw++] = upper_32_bits(value);
1722 		value += incr;
1723 	}
1724 }
1725 
1726 /**
1727  * sdma_v4_0_vm_set_pte_pde - update the page tables using sDMA
1728  *
1729  * @ib: indirect buffer to fill with commands
1730  * @pe: addr of the page entry
1731  * @addr: dst addr to write into pe
1732  * @count: number of page entries to update
1733  * @incr: increase next addr by incr bytes
1734  * @flags: access flags
1735  *
1736  * Update the page tables using sDMA (VEGA10).
1737  */
sdma_v4_0_vm_set_pte_pde(struct amdgpu_ib * ib,uint64_t pe,uint64_t addr,unsigned count,uint32_t incr,uint64_t flags)1738 static void sdma_v4_0_vm_set_pte_pde(struct amdgpu_ib *ib,
1739 				     uint64_t pe,
1740 				     uint64_t addr, unsigned count,
1741 				     uint32_t incr, uint64_t flags)
1742 {
1743 	/* for physically contiguous pages (vram) */
1744 	ib->ptr[ib->length_dw++] = SDMA_PKT_HEADER_OP(SDMA_OP_PTEPDE);
1745 	ib->ptr[ib->length_dw++] = lower_32_bits(pe); /* dst addr */
1746 	ib->ptr[ib->length_dw++] = upper_32_bits(pe);
1747 	ib->ptr[ib->length_dw++] = lower_32_bits(flags); /* mask */
1748 	ib->ptr[ib->length_dw++] = upper_32_bits(flags);
1749 	ib->ptr[ib->length_dw++] = lower_32_bits(addr); /* value */
1750 	ib->ptr[ib->length_dw++] = upper_32_bits(addr);
1751 	ib->ptr[ib->length_dw++] = incr; /* increment size */
1752 	ib->ptr[ib->length_dw++] = 0;
1753 	ib->ptr[ib->length_dw++] = count - 1; /* number of entries */
1754 }
1755 
1756 /**
1757  * sdma_v4_0_ring_pad_ib - pad the IB to the required number of dw
1758  *
1759  * @ring: amdgpu_ring structure holding ring information
1760  * @ib: indirect buffer to fill with padding
1761  */
sdma_v4_0_ring_pad_ib(struct amdgpu_ring * ring,struct amdgpu_ib * ib)1762 static void sdma_v4_0_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib)
1763 {
1764 	struct amdgpu_sdma_instance *sdma = amdgpu_sdma_get_instance_from_ring(ring);
1765 	u32 pad_count;
1766 	int i;
1767 
1768 	pad_count = (-ib->length_dw) & 7;
1769 	for (i = 0; i < pad_count; i++)
1770 		if (sdma && sdma->burst_nop && (i == 0))
1771 			ib->ptr[ib->length_dw++] =
1772 				SDMA_PKT_HEADER_OP(SDMA_OP_NOP) |
1773 				SDMA_PKT_NOP_HEADER_COUNT(pad_count - 1);
1774 		else
1775 			ib->ptr[ib->length_dw++] =
1776 				SDMA_PKT_HEADER_OP(SDMA_OP_NOP);
1777 }
1778 
1779 
1780 /**
1781  * sdma_v4_0_ring_emit_pipeline_sync - sync the pipeline
1782  *
1783  * @ring: amdgpu_ring pointer
1784  *
1785  * Make sure all previous operations are completed (CIK).
1786  */
sdma_v4_0_ring_emit_pipeline_sync(struct amdgpu_ring * ring)1787 static void sdma_v4_0_ring_emit_pipeline_sync(struct amdgpu_ring *ring)
1788 {
1789 	uint32_t seq = ring->fence_drv.sync_seq;
1790 	uint64_t addr = ring->fence_drv.gpu_addr;
1791 
1792 	/* wait for idle */
1793 	sdma_v4_0_wait_reg_mem(ring, 1, 0,
1794 			       addr & 0xfffffffc,
1795 			       upper_32_bits(addr) & 0xffffffff,
1796 			       seq, 0xffffffff, 4);
1797 }
1798 
1799 
1800 /**
1801  * sdma_v4_0_ring_emit_vm_flush - vm flush using sDMA
1802  *
1803  * @ring: amdgpu_ring pointer
1804  * @vmid: vmid number to use
1805  * @pd_addr: address
1806  *
1807  * Update the page table base and flush the VM TLB
1808  * using sDMA (VEGA10).
1809  */
sdma_v4_0_ring_emit_vm_flush(struct amdgpu_ring * ring,unsigned vmid,uint64_t pd_addr)1810 static void sdma_v4_0_ring_emit_vm_flush(struct amdgpu_ring *ring,
1811 					 unsigned vmid, uint64_t pd_addr)
1812 {
1813 	amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr);
1814 }
1815 
sdma_v4_0_ring_emit_wreg(struct amdgpu_ring * ring,uint32_t reg,uint32_t val)1816 static void sdma_v4_0_ring_emit_wreg(struct amdgpu_ring *ring,
1817 				     uint32_t reg, uint32_t val)
1818 {
1819 	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_SRBM_WRITE) |
1820 			  SDMA_PKT_SRBM_WRITE_HEADER_BYTE_EN(0xf));
1821 	amdgpu_ring_write(ring, reg);
1822 	amdgpu_ring_write(ring, val);
1823 }
1824 
sdma_v4_0_ring_emit_reg_wait(struct amdgpu_ring * ring,uint32_t reg,uint32_t val,uint32_t mask)1825 static void sdma_v4_0_ring_emit_reg_wait(struct amdgpu_ring *ring, uint32_t reg,
1826 					 uint32_t val, uint32_t mask)
1827 {
1828 	sdma_v4_0_wait_reg_mem(ring, 0, 0, reg, 0, val, mask, 10);
1829 }
1830 
sdma_v4_0_fw_support_paging_queue(struct amdgpu_device * adev)1831 static bool sdma_v4_0_fw_support_paging_queue(struct amdgpu_device *adev)
1832 {
1833 	uint fw_version = adev->sdma.instance[0].fw_version;
1834 
1835 	switch (adev->ip_versions[SDMA0_HWIP][0]) {
1836 	case IP_VERSION(4, 0, 0):
1837 		return fw_version >= 430;
1838 	case IP_VERSION(4, 0, 1):
1839 		/*return fw_version >= 31;*/
1840 		return false;
1841 	case IP_VERSION(4, 2, 0):
1842 		return fw_version >= 123;
1843 	default:
1844 		return false;
1845 	}
1846 }
1847 
sdma_v4_0_early_init(void * handle)1848 static int sdma_v4_0_early_init(void *handle)
1849 {
1850 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1851 	int r;
1852 
1853 	r = sdma_v4_0_init_microcode(adev);
1854 	if (r) {
1855 		DRM_ERROR("Failed to load sdma firmware!\n");
1856 		return r;
1857 	}
1858 
1859 	/* TODO: Page queue breaks driver reload under SRIOV */
1860 	if ((adev->ip_versions[SDMA0_HWIP][0] == IP_VERSION(4, 0, 0)) &&
1861 	    amdgpu_sriov_vf((adev)))
1862 		adev->sdma.has_page_queue = false;
1863 	else if (sdma_v4_0_fw_support_paging_queue(adev))
1864 		adev->sdma.has_page_queue = true;
1865 
1866 	sdma_v4_0_set_ring_funcs(adev);
1867 	sdma_v4_0_set_buffer_funcs(adev);
1868 	sdma_v4_0_set_vm_pte_funcs(adev);
1869 	sdma_v4_0_set_irq_funcs(adev);
1870 	sdma_v4_0_set_ras_funcs(adev);
1871 
1872 	return 0;
1873 }
1874 
1875 static int sdma_v4_0_process_ras_data_cb(struct amdgpu_device *adev,
1876 		void *err_data,
1877 		struct amdgpu_iv_entry *entry);
1878 
sdma_v4_0_late_init(void * handle)1879 static int sdma_v4_0_late_init(void *handle)
1880 {
1881 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1882 
1883 	sdma_v4_0_setup_ulv(adev);
1884 
1885 	if (!amdgpu_persistent_edc_harvesting_supported(adev)) {
1886 		if (adev->sdma.ras && adev->sdma.ras->ras_block.hw_ops &&
1887 		    adev->sdma.ras->ras_block.hw_ops->reset_ras_error_count)
1888 			adev->sdma.ras->ras_block.hw_ops->reset_ras_error_count(adev);
1889 	}
1890 
1891 	return 0;
1892 }
1893 
sdma_v4_0_sw_init(void * handle)1894 static int sdma_v4_0_sw_init(void *handle)
1895 {
1896 	struct amdgpu_ring *ring;
1897 	int r, i;
1898 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1899 
1900 	/* SDMA trap event */
1901 	for (i = 0; i < adev->sdma.num_instances; i++) {
1902 		r = amdgpu_irq_add_id(adev, sdma_v4_0_seq_to_irq_id(i),
1903 				      SDMA0_4_0__SRCID__SDMA_TRAP,
1904 				      &adev->sdma.trap_irq);
1905 		if (r)
1906 			return r;
1907 	}
1908 
1909 	/* SDMA SRAM ECC event */
1910 	for (i = 0; i < adev->sdma.num_instances; i++) {
1911 		r = amdgpu_irq_add_id(adev, sdma_v4_0_seq_to_irq_id(i),
1912 				      SDMA0_4_0__SRCID__SDMA_SRAM_ECC,
1913 				      &adev->sdma.ecc_irq);
1914 		if (r)
1915 			return r;
1916 	}
1917 
1918 	/* SDMA VM_HOLE/DOORBELL_INV/POLL_TIMEOUT/SRBM_WRITE_PROTECTION event*/
1919 	for (i = 0; i < adev->sdma.num_instances; i++) {
1920 		r = amdgpu_irq_add_id(adev, sdma_v4_0_seq_to_irq_id(i),
1921 				      SDMA0_4_0__SRCID__SDMA_VM_HOLE,
1922 				      &adev->sdma.vm_hole_irq);
1923 		if (r)
1924 			return r;
1925 
1926 		r = amdgpu_irq_add_id(adev, sdma_v4_0_seq_to_irq_id(i),
1927 				      SDMA0_4_0__SRCID__SDMA_DOORBELL_INVALID,
1928 				      &adev->sdma.doorbell_invalid_irq);
1929 		if (r)
1930 			return r;
1931 
1932 		r = amdgpu_irq_add_id(adev, sdma_v4_0_seq_to_irq_id(i),
1933 				      SDMA0_4_0__SRCID__SDMA_POLL_TIMEOUT,
1934 				      &adev->sdma.pool_timeout_irq);
1935 		if (r)
1936 			return r;
1937 
1938 		r = amdgpu_irq_add_id(adev, sdma_v4_0_seq_to_irq_id(i),
1939 				      SDMA0_4_0__SRCID__SDMA_SRBMWRITE,
1940 				      &adev->sdma.srbm_write_irq);
1941 		if (r)
1942 			return r;
1943 	}
1944 
1945 	for (i = 0; i < adev->sdma.num_instances; i++) {
1946 		ring = &adev->sdma.instance[i].ring;
1947 		ring->ring_obj = NULL;
1948 		ring->use_doorbell = true;
1949 
1950 		DRM_DEBUG("SDMA %d use_doorbell being set to: [%s]\n", i,
1951 				ring->use_doorbell?"true":"false");
1952 
1953 		/* doorbell size is 2 dwords, get DWORD offset */
1954 		ring->doorbell_index = adev->doorbell_index.sdma_engine[i] << 1;
1955 
1956 		sprintf(ring->name, "sdma%d", i);
1957 		r = amdgpu_ring_init(adev, ring, 1024, &adev->sdma.trap_irq,
1958 				     AMDGPU_SDMA_IRQ_INSTANCE0 + i,
1959 				     AMDGPU_RING_PRIO_DEFAULT, NULL);
1960 		if (r)
1961 			return r;
1962 
1963 		if (adev->sdma.has_page_queue) {
1964 			ring = &adev->sdma.instance[i].page;
1965 			ring->ring_obj = NULL;
1966 			ring->use_doorbell = true;
1967 
1968 			/* paging queue use same doorbell index/routing as gfx queue
1969 			 * with 0x400 (4096 dwords) offset on second doorbell page
1970 			 */
1971 			ring->doorbell_index = adev->doorbell_index.sdma_engine[i] << 1;
1972 			ring->doorbell_index += 0x400;
1973 
1974 			sprintf(ring->name, "page%d", i);
1975 			r = amdgpu_ring_init(adev, ring, 1024,
1976 					     &adev->sdma.trap_irq,
1977 					     AMDGPU_SDMA_IRQ_INSTANCE0 + i,
1978 					     AMDGPU_RING_PRIO_DEFAULT, NULL);
1979 			if (r)
1980 				return r;
1981 		}
1982 	}
1983 
1984 	return r;
1985 }
1986 
sdma_v4_0_sw_fini(void * handle)1987 static int sdma_v4_0_sw_fini(void *handle)
1988 {
1989 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1990 	int i;
1991 
1992 	for (i = 0; i < adev->sdma.num_instances; i++) {
1993 		amdgpu_ring_fini(&adev->sdma.instance[i].ring);
1994 		if (adev->sdma.has_page_queue)
1995 			amdgpu_ring_fini(&adev->sdma.instance[i].page);
1996 	}
1997 
1998 	sdma_v4_0_destroy_inst_ctx(adev);
1999 
2000 	return 0;
2001 }
2002 
sdma_v4_0_hw_init(void * handle)2003 static int sdma_v4_0_hw_init(void *handle)
2004 {
2005 	int r;
2006 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2007 
2008 	if (adev->flags & AMD_IS_APU)
2009 		amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_SDMA, false);
2010 
2011 	if (!amdgpu_sriov_vf(adev))
2012 		sdma_v4_0_init_golden_registers(adev);
2013 
2014 	r = sdma_v4_0_start(adev);
2015 
2016 	return r;
2017 }
2018 
sdma_v4_0_hw_fini(void * handle)2019 static int sdma_v4_0_hw_fini(void *handle)
2020 {
2021 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2022 	int i;
2023 
2024 	if (amdgpu_sriov_vf(adev))
2025 		return 0;
2026 
2027 	for (i = 0; i < adev->sdma.num_instances; i++) {
2028 		amdgpu_irq_put(adev, &adev->sdma.ecc_irq,
2029 			       AMDGPU_SDMA_IRQ_INSTANCE0 + i);
2030 	}
2031 
2032 	sdma_v4_0_ctx_switch_enable(adev, false);
2033 	sdma_v4_0_enable(adev, false);
2034 
2035 	if (adev->flags & AMD_IS_APU)
2036 		amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_SDMA, true);
2037 
2038 	return 0;
2039 }
2040 
sdma_v4_0_suspend(void * handle)2041 static int sdma_v4_0_suspend(void *handle)
2042 {
2043 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2044 
2045 	/* SMU saves SDMA state for us */
2046 	if (adev->in_s0ix)
2047 		return 0;
2048 
2049 	return sdma_v4_0_hw_fini(adev);
2050 }
2051 
sdma_v4_0_resume(void * handle)2052 static int sdma_v4_0_resume(void *handle)
2053 {
2054 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2055 
2056 	/* SMU restores SDMA state for us */
2057 	if (adev->in_s0ix)
2058 		return 0;
2059 
2060 	return sdma_v4_0_hw_init(adev);
2061 }
2062 
sdma_v4_0_is_idle(void * handle)2063 static bool sdma_v4_0_is_idle(void *handle)
2064 {
2065 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2066 	u32 i;
2067 
2068 	for (i = 0; i < adev->sdma.num_instances; i++) {
2069 		u32 tmp = RREG32_SDMA(i, mmSDMA0_STATUS_REG);
2070 
2071 		if (!(tmp & SDMA0_STATUS_REG__IDLE_MASK))
2072 			return false;
2073 	}
2074 
2075 	return true;
2076 }
2077 
sdma_v4_0_wait_for_idle(void * handle)2078 static int sdma_v4_0_wait_for_idle(void *handle)
2079 {
2080 	unsigned i, j;
2081 	u32 sdma[AMDGPU_MAX_SDMA_INSTANCES];
2082 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2083 
2084 	for (i = 0; i < adev->usec_timeout; i++) {
2085 		for (j = 0; j < adev->sdma.num_instances; j++) {
2086 			sdma[j] = RREG32_SDMA(j, mmSDMA0_STATUS_REG);
2087 			if (!(sdma[j] & SDMA0_STATUS_REG__IDLE_MASK))
2088 				break;
2089 		}
2090 		if (j == adev->sdma.num_instances)
2091 			return 0;
2092 		udelay(1);
2093 	}
2094 	return -ETIMEDOUT;
2095 }
2096 
sdma_v4_0_soft_reset(void * handle)2097 static int sdma_v4_0_soft_reset(void *handle)
2098 {
2099 	/* todo */
2100 
2101 	return 0;
2102 }
2103 
sdma_v4_0_set_trap_irq_state(struct amdgpu_device * adev,struct amdgpu_irq_src * source,unsigned type,enum amdgpu_interrupt_state state)2104 static int sdma_v4_0_set_trap_irq_state(struct amdgpu_device *adev,
2105 					struct amdgpu_irq_src *source,
2106 					unsigned type,
2107 					enum amdgpu_interrupt_state state)
2108 {
2109 	u32 sdma_cntl;
2110 
2111 	sdma_cntl = RREG32_SDMA(type, mmSDMA0_CNTL);
2112 	sdma_cntl = REG_SET_FIELD(sdma_cntl, SDMA0_CNTL, TRAP_ENABLE,
2113 		       state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0);
2114 	WREG32_SDMA(type, mmSDMA0_CNTL, sdma_cntl);
2115 
2116 	return 0;
2117 }
2118 
sdma_v4_0_process_trap_irq(struct amdgpu_device * adev,struct amdgpu_irq_src * source,struct amdgpu_iv_entry * entry)2119 static int sdma_v4_0_process_trap_irq(struct amdgpu_device *adev,
2120 				      struct amdgpu_irq_src *source,
2121 				      struct amdgpu_iv_entry *entry)
2122 {
2123 	uint32_t instance;
2124 
2125 	DRM_DEBUG("IH: SDMA trap\n");
2126 	instance = sdma_v4_0_irq_id_to_seq(entry->client_id);
2127 	switch (entry->ring_id) {
2128 	case 0:
2129 		amdgpu_fence_process(&adev->sdma.instance[instance].ring);
2130 		break;
2131 	case 1:
2132 		if (adev->ip_versions[SDMA0_HWIP][0] == IP_VERSION(4, 2, 0))
2133 			amdgpu_fence_process(&adev->sdma.instance[instance].page);
2134 		break;
2135 	case 2:
2136 		/* XXX compute */
2137 		break;
2138 	case 3:
2139 		if (adev->ip_versions[SDMA0_HWIP][0] != IP_VERSION(4, 2, 0))
2140 			amdgpu_fence_process(&adev->sdma.instance[instance].page);
2141 		break;
2142 	}
2143 	return 0;
2144 }
2145 
sdma_v4_0_process_ras_data_cb(struct amdgpu_device * adev,void * err_data,struct amdgpu_iv_entry * entry)2146 static int sdma_v4_0_process_ras_data_cb(struct amdgpu_device *adev,
2147 		void *err_data,
2148 		struct amdgpu_iv_entry *entry)
2149 {
2150 	int instance;
2151 
2152 	/* When “Full RAS” is enabled, the per-IP interrupt sources should
2153 	 * be disabled and the driver should only look for the aggregated
2154 	 * interrupt via sync flood
2155 	 */
2156 	if (amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__GFX))
2157 		goto out;
2158 
2159 	instance = sdma_v4_0_irq_id_to_seq(entry->client_id);
2160 	if (instance < 0)
2161 		goto out;
2162 
2163 	amdgpu_sdma_process_ras_data_cb(adev, err_data, entry);
2164 
2165 out:
2166 	return AMDGPU_RAS_SUCCESS;
2167 }
2168 
sdma_v4_0_process_illegal_inst_irq(struct amdgpu_device * adev,struct amdgpu_irq_src * source,struct amdgpu_iv_entry * entry)2169 static int sdma_v4_0_process_illegal_inst_irq(struct amdgpu_device *adev,
2170 					      struct amdgpu_irq_src *source,
2171 					      struct amdgpu_iv_entry *entry)
2172 {
2173 	int instance;
2174 
2175 	DRM_ERROR("Illegal instruction in SDMA command stream\n");
2176 
2177 	instance = sdma_v4_0_irq_id_to_seq(entry->client_id);
2178 	if (instance < 0)
2179 		return 0;
2180 
2181 	switch (entry->ring_id) {
2182 	case 0:
2183 		drm_sched_fault(&adev->sdma.instance[instance].ring.sched);
2184 		break;
2185 	}
2186 	return 0;
2187 }
2188 
sdma_v4_0_set_ecc_irq_state(struct amdgpu_device * adev,struct amdgpu_irq_src * source,unsigned type,enum amdgpu_interrupt_state state)2189 static int sdma_v4_0_set_ecc_irq_state(struct amdgpu_device *adev,
2190 					struct amdgpu_irq_src *source,
2191 					unsigned type,
2192 					enum amdgpu_interrupt_state state)
2193 {
2194 	u32 sdma_edc_config;
2195 
2196 	sdma_edc_config = RREG32_SDMA(type, mmSDMA0_EDC_CONFIG);
2197 	sdma_edc_config = REG_SET_FIELD(sdma_edc_config, SDMA0_EDC_CONFIG, ECC_INT_ENABLE,
2198 		       state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0);
2199 	WREG32_SDMA(type, mmSDMA0_EDC_CONFIG, sdma_edc_config);
2200 
2201 	return 0;
2202 }
2203 
sdma_v4_0_print_iv_entry(struct amdgpu_device * adev,struct amdgpu_iv_entry * entry)2204 static int sdma_v4_0_print_iv_entry(struct amdgpu_device *adev,
2205 					      struct amdgpu_iv_entry *entry)
2206 {
2207 	int instance;
2208 	struct amdgpu_task_info task_info;
2209 	u64 addr;
2210 
2211 	instance = sdma_v4_0_irq_id_to_seq(entry->client_id);
2212 	if (instance < 0 || instance >= adev->sdma.num_instances) {
2213 		dev_err(adev->dev, "sdma instance invalid %d\n", instance);
2214 		return -EINVAL;
2215 	}
2216 
2217 	addr = (u64)entry->src_data[0] << 12;
2218 	addr |= ((u64)entry->src_data[1] & 0xf) << 44;
2219 
2220 	memset(&task_info, 0, sizeof(struct amdgpu_task_info));
2221 	amdgpu_vm_get_task_info(adev, entry->pasid, &task_info);
2222 
2223 	dev_dbg_ratelimited(adev->dev,
2224 		   "[sdma%d] address:0x%016llx src_id:%u ring:%u vmid:%u "
2225 		   "pasid:%u, for process %s pid %d thread %s pid %d\n",
2226 		   instance, addr, entry->src_id, entry->ring_id, entry->vmid,
2227 		   entry->pasid, task_info.process_name, task_info.tgid,
2228 		   task_info.task_name, task_info.pid);
2229 	return 0;
2230 }
2231 
sdma_v4_0_process_vm_hole_irq(struct amdgpu_device * adev,struct amdgpu_irq_src * source,struct amdgpu_iv_entry * entry)2232 static int sdma_v4_0_process_vm_hole_irq(struct amdgpu_device *adev,
2233 					      struct amdgpu_irq_src *source,
2234 					      struct amdgpu_iv_entry *entry)
2235 {
2236 	dev_dbg_ratelimited(adev->dev, "MC or SEM address in VM hole\n");
2237 	sdma_v4_0_print_iv_entry(adev, entry);
2238 	return 0;
2239 }
2240 
sdma_v4_0_process_doorbell_invalid_irq(struct amdgpu_device * adev,struct amdgpu_irq_src * source,struct amdgpu_iv_entry * entry)2241 static int sdma_v4_0_process_doorbell_invalid_irq(struct amdgpu_device *adev,
2242 					      struct amdgpu_irq_src *source,
2243 					      struct amdgpu_iv_entry *entry)
2244 {
2245 	dev_dbg_ratelimited(adev->dev, "SDMA received a doorbell from BIF with byte_enable !=0xff\n");
2246 	sdma_v4_0_print_iv_entry(adev, entry);
2247 	return 0;
2248 }
2249 
sdma_v4_0_process_pool_timeout_irq(struct amdgpu_device * adev,struct amdgpu_irq_src * source,struct amdgpu_iv_entry * entry)2250 static int sdma_v4_0_process_pool_timeout_irq(struct amdgpu_device *adev,
2251 					      struct amdgpu_irq_src *source,
2252 					      struct amdgpu_iv_entry *entry)
2253 {
2254 	dev_dbg_ratelimited(adev->dev,
2255 		"Polling register/memory timeout executing POLL_REG/MEM with finite timer\n");
2256 	sdma_v4_0_print_iv_entry(adev, entry);
2257 	return 0;
2258 }
2259 
sdma_v4_0_process_srbm_write_irq(struct amdgpu_device * adev,struct amdgpu_irq_src * source,struct amdgpu_iv_entry * entry)2260 static int sdma_v4_0_process_srbm_write_irq(struct amdgpu_device *adev,
2261 					      struct amdgpu_irq_src *source,
2262 					      struct amdgpu_iv_entry *entry)
2263 {
2264 	dev_dbg_ratelimited(adev->dev,
2265 		"SDMA gets an Register Write SRBM_WRITE command in non-privilege command buffer\n");
2266 	sdma_v4_0_print_iv_entry(adev, entry);
2267 	return 0;
2268 }
2269 
sdma_v4_0_update_medium_grain_clock_gating(struct amdgpu_device * adev,bool enable)2270 static void sdma_v4_0_update_medium_grain_clock_gating(
2271 		struct amdgpu_device *adev,
2272 		bool enable)
2273 {
2274 	uint32_t data, def;
2275 	int i;
2276 
2277 	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_SDMA_MGCG)) {
2278 		for (i = 0; i < adev->sdma.num_instances; i++) {
2279 			def = data = RREG32_SDMA(i, mmSDMA0_CLK_CTRL);
2280 			data &= ~(SDMA0_CLK_CTRL__SOFT_OVERRIDE7_MASK |
2281 				  SDMA0_CLK_CTRL__SOFT_OVERRIDE6_MASK |
2282 				  SDMA0_CLK_CTRL__SOFT_OVERRIDE5_MASK |
2283 				  SDMA0_CLK_CTRL__SOFT_OVERRIDE4_MASK |
2284 				  SDMA0_CLK_CTRL__SOFT_OVERRIDE3_MASK |
2285 				  SDMA0_CLK_CTRL__SOFT_OVERRIDE2_MASK |
2286 				  SDMA0_CLK_CTRL__SOFT_OVERRIDE1_MASK |
2287 				  SDMA0_CLK_CTRL__SOFT_OVERRIDE0_MASK);
2288 			if (def != data)
2289 				WREG32_SDMA(i, mmSDMA0_CLK_CTRL, data);
2290 		}
2291 	} else {
2292 		for (i = 0; i < adev->sdma.num_instances; i++) {
2293 			def = data = RREG32_SDMA(i, mmSDMA0_CLK_CTRL);
2294 			data |= (SDMA0_CLK_CTRL__SOFT_OVERRIDE7_MASK |
2295 				 SDMA0_CLK_CTRL__SOFT_OVERRIDE6_MASK |
2296 				 SDMA0_CLK_CTRL__SOFT_OVERRIDE5_MASK |
2297 				 SDMA0_CLK_CTRL__SOFT_OVERRIDE4_MASK |
2298 				 SDMA0_CLK_CTRL__SOFT_OVERRIDE3_MASK |
2299 				 SDMA0_CLK_CTRL__SOFT_OVERRIDE2_MASK |
2300 				 SDMA0_CLK_CTRL__SOFT_OVERRIDE1_MASK |
2301 				 SDMA0_CLK_CTRL__SOFT_OVERRIDE0_MASK);
2302 			if (def != data)
2303 				WREG32_SDMA(i, mmSDMA0_CLK_CTRL, data);
2304 		}
2305 	}
2306 }
2307 
2308 
sdma_v4_0_update_medium_grain_light_sleep(struct amdgpu_device * adev,bool enable)2309 static void sdma_v4_0_update_medium_grain_light_sleep(
2310 		struct amdgpu_device *adev,
2311 		bool enable)
2312 {
2313 	uint32_t data, def;
2314 	int i;
2315 
2316 	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_SDMA_LS)) {
2317 		for (i = 0; i < adev->sdma.num_instances; i++) {
2318 			/* 1-not override: enable sdma mem light sleep */
2319 			def = data = RREG32_SDMA(0, mmSDMA0_POWER_CNTL);
2320 			data |= SDMA0_POWER_CNTL__MEM_POWER_OVERRIDE_MASK;
2321 			if (def != data)
2322 				WREG32_SDMA(0, mmSDMA0_POWER_CNTL, data);
2323 		}
2324 	} else {
2325 		for (i = 0; i < adev->sdma.num_instances; i++) {
2326 		/* 0-override:disable sdma mem light sleep */
2327 			def = data = RREG32_SDMA(0, mmSDMA0_POWER_CNTL);
2328 			data &= ~SDMA0_POWER_CNTL__MEM_POWER_OVERRIDE_MASK;
2329 			if (def != data)
2330 				WREG32_SDMA(0, mmSDMA0_POWER_CNTL, data);
2331 		}
2332 	}
2333 }
2334 
sdma_v4_0_set_clockgating_state(void * handle,enum amd_clockgating_state state)2335 static int sdma_v4_0_set_clockgating_state(void *handle,
2336 					  enum amd_clockgating_state state)
2337 {
2338 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2339 
2340 	if (amdgpu_sriov_vf(adev))
2341 		return 0;
2342 
2343 	sdma_v4_0_update_medium_grain_clock_gating(adev,
2344 			state == AMD_CG_STATE_GATE);
2345 	sdma_v4_0_update_medium_grain_light_sleep(adev,
2346 			state == AMD_CG_STATE_GATE);
2347 	return 0;
2348 }
2349 
sdma_v4_0_set_powergating_state(void * handle,enum amd_powergating_state state)2350 static int sdma_v4_0_set_powergating_state(void *handle,
2351 					  enum amd_powergating_state state)
2352 {
2353 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2354 
2355 	switch (adev->ip_versions[SDMA0_HWIP][0]) {
2356 	case IP_VERSION(4, 1, 0):
2357 	case IP_VERSION(4, 1, 1):
2358 	case IP_VERSION(4, 1, 2):
2359 		sdma_v4_1_update_power_gating(adev,
2360 				state == AMD_PG_STATE_GATE);
2361 		break;
2362 	default:
2363 		break;
2364 	}
2365 
2366 	return 0;
2367 }
2368 
sdma_v4_0_get_clockgating_state(void * handle,u64 * flags)2369 static void sdma_v4_0_get_clockgating_state(void *handle, u64 *flags)
2370 {
2371 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2372 	int data;
2373 
2374 	if (amdgpu_sriov_vf(adev))
2375 		*flags = 0;
2376 
2377 	/* AMD_CG_SUPPORT_SDMA_MGCG */
2378 	data = RREG32(SOC15_REG_OFFSET(SDMA0, 0, mmSDMA0_CLK_CTRL));
2379 	if (!(data & SDMA0_CLK_CTRL__SOFT_OVERRIDE7_MASK))
2380 		*flags |= AMD_CG_SUPPORT_SDMA_MGCG;
2381 
2382 	/* AMD_CG_SUPPORT_SDMA_LS */
2383 	data = RREG32(SOC15_REG_OFFSET(SDMA0, 0, mmSDMA0_POWER_CNTL));
2384 	if (data & SDMA0_POWER_CNTL__MEM_POWER_OVERRIDE_MASK)
2385 		*flags |= AMD_CG_SUPPORT_SDMA_LS;
2386 }
2387 
2388 const struct amd_ip_funcs sdma_v4_0_ip_funcs = {
2389 	.name = "sdma_v4_0",
2390 	.early_init = sdma_v4_0_early_init,
2391 	.late_init = sdma_v4_0_late_init,
2392 	.sw_init = sdma_v4_0_sw_init,
2393 	.sw_fini = sdma_v4_0_sw_fini,
2394 	.hw_init = sdma_v4_0_hw_init,
2395 	.hw_fini = sdma_v4_0_hw_fini,
2396 	.suspend = sdma_v4_0_suspend,
2397 	.resume = sdma_v4_0_resume,
2398 	.is_idle = sdma_v4_0_is_idle,
2399 	.wait_for_idle = sdma_v4_0_wait_for_idle,
2400 	.soft_reset = sdma_v4_0_soft_reset,
2401 	.set_clockgating_state = sdma_v4_0_set_clockgating_state,
2402 	.set_powergating_state = sdma_v4_0_set_powergating_state,
2403 	.get_clockgating_state = sdma_v4_0_get_clockgating_state,
2404 };
2405 
2406 static const struct amdgpu_ring_funcs sdma_v4_0_ring_funcs = {
2407 	.type = AMDGPU_RING_TYPE_SDMA,
2408 	.align_mask = 0xf,
2409 	.nop = SDMA_PKT_NOP_HEADER_OP(SDMA_OP_NOP),
2410 	.support_64bit_ptrs = true,
2411 	.secure_submission_supported = true,
2412 	.vmhub = AMDGPU_MMHUB_0,
2413 	.get_rptr = sdma_v4_0_ring_get_rptr,
2414 	.get_wptr = sdma_v4_0_ring_get_wptr,
2415 	.set_wptr = sdma_v4_0_ring_set_wptr,
2416 	.emit_frame_size =
2417 		6 + /* sdma_v4_0_ring_emit_hdp_flush */
2418 		3 + /* hdp invalidate */
2419 		6 + /* sdma_v4_0_ring_emit_pipeline_sync */
2420 		/* sdma_v4_0_ring_emit_vm_flush */
2421 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
2422 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 6 +
2423 		10 + 10 + 10, /* sdma_v4_0_ring_emit_fence x3 for user fence, vm fence */
2424 	.emit_ib_size = 7 + 6, /* sdma_v4_0_ring_emit_ib */
2425 	.emit_ib = sdma_v4_0_ring_emit_ib,
2426 	.emit_fence = sdma_v4_0_ring_emit_fence,
2427 	.emit_pipeline_sync = sdma_v4_0_ring_emit_pipeline_sync,
2428 	.emit_vm_flush = sdma_v4_0_ring_emit_vm_flush,
2429 	.emit_hdp_flush = sdma_v4_0_ring_emit_hdp_flush,
2430 	.test_ring = sdma_v4_0_ring_test_ring,
2431 	.test_ib = sdma_v4_0_ring_test_ib,
2432 	.insert_nop = sdma_v4_0_ring_insert_nop,
2433 	.pad_ib = sdma_v4_0_ring_pad_ib,
2434 	.emit_wreg = sdma_v4_0_ring_emit_wreg,
2435 	.emit_reg_wait = sdma_v4_0_ring_emit_reg_wait,
2436 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
2437 };
2438 
2439 /*
2440  * On Arcturus, SDMA instance 5~7 has a different vmhub type(AMDGPU_MMHUB_1).
2441  * So create a individual constant ring_funcs for those instances.
2442  */
2443 static const struct amdgpu_ring_funcs sdma_v4_0_ring_funcs_2nd_mmhub = {
2444 	.type = AMDGPU_RING_TYPE_SDMA,
2445 	.align_mask = 0xf,
2446 	.nop = SDMA_PKT_NOP_HEADER_OP(SDMA_OP_NOP),
2447 	.support_64bit_ptrs = true,
2448 	.secure_submission_supported = true,
2449 	.vmhub = AMDGPU_MMHUB_1,
2450 	.get_rptr = sdma_v4_0_ring_get_rptr,
2451 	.get_wptr = sdma_v4_0_ring_get_wptr,
2452 	.set_wptr = sdma_v4_0_ring_set_wptr,
2453 	.emit_frame_size =
2454 		6 + /* sdma_v4_0_ring_emit_hdp_flush */
2455 		3 + /* hdp invalidate */
2456 		6 + /* sdma_v4_0_ring_emit_pipeline_sync */
2457 		/* sdma_v4_0_ring_emit_vm_flush */
2458 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
2459 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 6 +
2460 		10 + 10 + 10, /* sdma_v4_0_ring_emit_fence x3 for user fence, vm fence */
2461 	.emit_ib_size = 7 + 6, /* sdma_v4_0_ring_emit_ib */
2462 	.emit_ib = sdma_v4_0_ring_emit_ib,
2463 	.emit_fence = sdma_v4_0_ring_emit_fence,
2464 	.emit_pipeline_sync = sdma_v4_0_ring_emit_pipeline_sync,
2465 	.emit_vm_flush = sdma_v4_0_ring_emit_vm_flush,
2466 	.emit_hdp_flush = sdma_v4_0_ring_emit_hdp_flush,
2467 	.test_ring = sdma_v4_0_ring_test_ring,
2468 	.test_ib = sdma_v4_0_ring_test_ib,
2469 	.insert_nop = sdma_v4_0_ring_insert_nop,
2470 	.pad_ib = sdma_v4_0_ring_pad_ib,
2471 	.emit_wreg = sdma_v4_0_ring_emit_wreg,
2472 	.emit_reg_wait = sdma_v4_0_ring_emit_reg_wait,
2473 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
2474 };
2475 
2476 static const struct amdgpu_ring_funcs sdma_v4_0_page_ring_funcs = {
2477 	.type = AMDGPU_RING_TYPE_SDMA,
2478 	.align_mask = 0xf,
2479 	.nop = SDMA_PKT_NOP_HEADER_OP(SDMA_OP_NOP),
2480 	.support_64bit_ptrs = true,
2481 	.secure_submission_supported = true,
2482 	.vmhub = AMDGPU_MMHUB_0,
2483 	.get_rptr = sdma_v4_0_ring_get_rptr,
2484 	.get_wptr = sdma_v4_0_page_ring_get_wptr,
2485 	.set_wptr = sdma_v4_0_page_ring_set_wptr,
2486 	.emit_frame_size =
2487 		6 + /* sdma_v4_0_ring_emit_hdp_flush */
2488 		3 + /* hdp invalidate */
2489 		6 + /* sdma_v4_0_ring_emit_pipeline_sync */
2490 		/* sdma_v4_0_ring_emit_vm_flush */
2491 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
2492 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 6 +
2493 		10 + 10 + 10, /* sdma_v4_0_ring_emit_fence x3 for user fence, vm fence */
2494 	.emit_ib_size = 7 + 6, /* sdma_v4_0_ring_emit_ib */
2495 	.emit_ib = sdma_v4_0_ring_emit_ib,
2496 	.emit_fence = sdma_v4_0_ring_emit_fence,
2497 	.emit_pipeline_sync = sdma_v4_0_ring_emit_pipeline_sync,
2498 	.emit_vm_flush = sdma_v4_0_ring_emit_vm_flush,
2499 	.emit_hdp_flush = sdma_v4_0_ring_emit_hdp_flush,
2500 	.test_ring = sdma_v4_0_ring_test_ring,
2501 	.test_ib = sdma_v4_0_ring_test_ib,
2502 	.insert_nop = sdma_v4_0_ring_insert_nop,
2503 	.pad_ib = sdma_v4_0_ring_pad_ib,
2504 	.emit_wreg = sdma_v4_0_ring_emit_wreg,
2505 	.emit_reg_wait = sdma_v4_0_ring_emit_reg_wait,
2506 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
2507 };
2508 
2509 static const struct amdgpu_ring_funcs sdma_v4_0_page_ring_funcs_2nd_mmhub = {
2510 	.type = AMDGPU_RING_TYPE_SDMA,
2511 	.align_mask = 0xf,
2512 	.nop = SDMA_PKT_NOP_HEADER_OP(SDMA_OP_NOP),
2513 	.support_64bit_ptrs = true,
2514 	.secure_submission_supported = true,
2515 	.vmhub = AMDGPU_MMHUB_1,
2516 	.get_rptr = sdma_v4_0_ring_get_rptr,
2517 	.get_wptr = sdma_v4_0_page_ring_get_wptr,
2518 	.set_wptr = sdma_v4_0_page_ring_set_wptr,
2519 	.emit_frame_size =
2520 		6 + /* sdma_v4_0_ring_emit_hdp_flush */
2521 		3 + /* hdp invalidate */
2522 		6 + /* sdma_v4_0_ring_emit_pipeline_sync */
2523 		/* sdma_v4_0_ring_emit_vm_flush */
2524 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
2525 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 6 +
2526 		10 + 10 + 10, /* sdma_v4_0_ring_emit_fence x3 for user fence, vm fence */
2527 	.emit_ib_size = 7 + 6, /* sdma_v4_0_ring_emit_ib */
2528 	.emit_ib = sdma_v4_0_ring_emit_ib,
2529 	.emit_fence = sdma_v4_0_ring_emit_fence,
2530 	.emit_pipeline_sync = sdma_v4_0_ring_emit_pipeline_sync,
2531 	.emit_vm_flush = sdma_v4_0_ring_emit_vm_flush,
2532 	.emit_hdp_flush = sdma_v4_0_ring_emit_hdp_flush,
2533 	.test_ring = sdma_v4_0_ring_test_ring,
2534 	.test_ib = sdma_v4_0_ring_test_ib,
2535 	.insert_nop = sdma_v4_0_ring_insert_nop,
2536 	.pad_ib = sdma_v4_0_ring_pad_ib,
2537 	.emit_wreg = sdma_v4_0_ring_emit_wreg,
2538 	.emit_reg_wait = sdma_v4_0_ring_emit_reg_wait,
2539 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
2540 };
2541 
sdma_v4_0_set_ring_funcs(struct amdgpu_device * adev)2542 static void sdma_v4_0_set_ring_funcs(struct amdgpu_device *adev)
2543 {
2544 	int i;
2545 
2546 	for (i = 0; i < adev->sdma.num_instances; i++) {
2547 		if (adev->ip_versions[SDMA0_HWIP][0] == IP_VERSION(4, 2, 2) && i >= 5)
2548 			adev->sdma.instance[i].ring.funcs =
2549 					&sdma_v4_0_ring_funcs_2nd_mmhub;
2550 		else
2551 			adev->sdma.instance[i].ring.funcs =
2552 					&sdma_v4_0_ring_funcs;
2553 		adev->sdma.instance[i].ring.me = i;
2554 		if (adev->sdma.has_page_queue) {
2555 			if (adev->ip_versions[SDMA0_HWIP][0] == IP_VERSION(4, 2, 2) && i >= 5)
2556 				adev->sdma.instance[i].page.funcs =
2557 					&sdma_v4_0_page_ring_funcs_2nd_mmhub;
2558 			else
2559 				adev->sdma.instance[i].page.funcs =
2560 					&sdma_v4_0_page_ring_funcs;
2561 			adev->sdma.instance[i].page.me = i;
2562 		}
2563 	}
2564 }
2565 
2566 static const struct amdgpu_irq_src_funcs sdma_v4_0_trap_irq_funcs = {
2567 	.set = sdma_v4_0_set_trap_irq_state,
2568 	.process = sdma_v4_0_process_trap_irq,
2569 };
2570 
2571 static const struct amdgpu_irq_src_funcs sdma_v4_0_illegal_inst_irq_funcs = {
2572 	.process = sdma_v4_0_process_illegal_inst_irq,
2573 };
2574 
2575 static const struct amdgpu_irq_src_funcs sdma_v4_0_ecc_irq_funcs = {
2576 	.set = sdma_v4_0_set_ecc_irq_state,
2577 	.process = amdgpu_sdma_process_ecc_irq,
2578 };
2579 
2580 static const struct amdgpu_irq_src_funcs sdma_v4_0_vm_hole_irq_funcs = {
2581 	.process = sdma_v4_0_process_vm_hole_irq,
2582 };
2583 
2584 static const struct amdgpu_irq_src_funcs sdma_v4_0_doorbell_invalid_irq_funcs = {
2585 	.process = sdma_v4_0_process_doorbell_invalid_irq,
2586 };
2587 
2588 static const struct amdgpu_irq_src_funcs sdma_v4_0_pool_timeout_irq_funcs = {
2589 	.process = sdma_v4_0_process_pool_timeout_irq,
2590 };
2591 
2592 static const struct amdgpu_irq_src_funcs sdma_v4_0_srbm_write_irq_funcs = {
2593 	.process = sdma_v4_0_process_srbm_write_irq,
2594 };
2595 
sdma_v4_0_set_irq_funcs(struct amdgpu_device * adev)2596 static void sdma_v4_0_set_irq_funcs(struct amdgpu_device *adev)
2597 {
2598 	adev->sdma.trap_irq.num_types = adev->sdma.num_instances;
2599 	adev->sdma.ecc_irq.num_types = adev->sdma.num_instances;
2600 	/*For Arcturus and Aldebaran, add another 4 irq handler*/
2601 	switch (adev->sdma.num_instances) {
2602 	case 5:
2603 	case 8:
2604 		adev->sdma.vm_hole_irq.num_types = adev->sdma.num_instances;
2605 		adev->sdma.doorbell_invalid_irq.num_types = adev->sdma.num_instances;
2606 		adev->sdma.pool_timeout_irq.num_types = adev->sdma.num_instances;
2607 		adev->sdma.srbm_write_irq.num_types = adev->sdma.num_instances;
2608 		break;
2609 	default:
2610 		break;
2611 	}
2612 	adev->sdma.trap_irq.funcs = &sdma_v4_0_trap_irq_funcs;
2613 	adev->sdma.illegal_inst_irq.funcs = &sdma_v4_0_illegal_inst_irq_funcs;
2614 	adev->sdma.ecc_irq.funcs = &sdma_v4_0_ecc_irq_funcs;
2615 	adev->sdma.vm_hole_irq.funcs = &sdma_v4_0_vm_hole_irq_funcs;
2616 	adev->sdma.doorbell_invalid_irq.funcs = &sdma_v4_0_doorbell_invalid_irq_funcs;
2617 	adev->sdma.pool_timeout_irq.funcs = &sdma_v4_0_pool_timeout_irq_funcs;
2618 	adev->sdma.srbm_write_irq.funcs = &sdma_v4_0_srbm_write_irq_funcs;
2619 }
2620 
2621 /**
2622  * sdma_v4_0_emit_copy_buffer - copy buffer using the sDMA engine
2623  *
2624  * @ib: indirect buffer to copy to
2625  * @src_offset: src GPU address
2626  * @dst_offset: dst GPU address
2627  * @byte_count: number of bytes to xfer
2628  * @tmz: if a secure copy should be used
2629  *
2630  * Copy GPU buffers using the DMA engine (VEGA10/12).
2631  * Used by the amdgpu ttm implementation to move pages if
2632  * registered as the asic copy callback.
2633  */
sdma_v4_0_emit_copy_buffer(struct amdgpu_ib * ib,uint64_t src_offset,uint64_t dst_offset,uint32_t byte_count,bool tmz)2634 static void sdma_v4_0_emit_copy_buffer(struct amdgpu_ib *ib,
2635 				       uint64_t src_offset,
2636 				       uint64_t dst_offset,
2637 				       uint32_t byte_count,
2638 				       bool tmz)
2639 {
2640 	ib->ptr[ib->length_dw++] = SDMA_PKT_HEADER_OP(SDMA_OP_COPY) |
2641 		SDMA_PKT_HEADER_SUB_OP(SDMA_SUBOP_COPY_LINEAR) |
2642 		SDMA_PKT_COPY_LINEAR_HEADER_TMZ(tmz ? 1 : 0);
2643 	ib->ptr[ib->length_dw++] = byte_count - 1;
2644 	ib->ptr[ib->length_dw++] = 0; /* src/dst endian swap */
2645 	ib->ptr[ib->length_dw++] = lower_32_bits(src_offset);
2646 	ib->ptr[ib->length_dw++] = upper_32_bits(src_offset);
2647 	ib->ptr[ib->length_dw++] = lower_32_bits(dst_offset);
2648 	ib->ptr[ib->length_dw++] = upper_32_bits(dst_offset);
2649 }
2650 
2651 /**
2652  * sdma_v4_0_emit_fill_buffer - fill buffer using the sDMA engine
2653  *
2654  * @ib: indirect buffer to copy to
2655  * @src_data: value to write to buffer
2656  * @dst_offset: dst GPU address
2657  * @byte_count: number of bytes to xfer
2658  *
2659  * Fill GPU buffers using the DMA engine (VEGA10/12).
2660  */
sdma_v4_0_emit_fill_buffer(struct amdgpu_ib * ib,uint32_t src_data,uint64_t dst_offset,uint32_t byte_count)2661 static void sdma_v4_0_emit_fill_buffer(struct amdgpu_ib *ib,
2662 				       uint32_t src_data,
2663 				       uint64_t dst_offset,
2664 				       uint32_t byte_count)
2665 {
2666 	ib->ptr[ib->length_dw++] = SDMA_PKT_HEADER_OP(SDMA_OP_CONST_FILL);
2667 	ib->ptr[ib->length_dw++] = lower_32_bits(dst_offset);
2668 	ib->ptr[ib->length_dw++] = upper_32_bits(dst_offset);
2669 	ib->ptr[ib->length_dw++] = src_data;
2670 	ib->ptr[ib->length_dw++] = byte_count - 1;
2671 }
2672 
2673 static const struct amdgpu_buffer_funcs sdma_v4_0_buffer_funcs = {
2674 	.copy_max_bytes = 0x400000,
2675 	.copy_num_dw = 7,
2676 	.emit_copy_buffer = sdma_v4_0_emit_copy_buffer,
2677 
2678 	.fill_max_bytes = 0x400000,
2679 	.fill_num_dw = 5,
2680 	.emit_fill_buffer = sdma_v4_0_emit_fill_buffer,
2681 };
2682 
sdma_v4_0_set_buffer_funcs(struct amdgpu_device * adev)2683 static void sdma_v4_0_set_buffer_funcs(struct amdgpu_device *adev)
2684 {
2685 	adev->mman.buffer_funcs = &sdma_v4_0_buffer_funcs;
2686 	if (adev->sdma.has_page_queue)
2687 		adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].page;
2688 	else
2689 		adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].ring;
2690 }
2691 
2692 static const struct amdgpu_vm_pte_funcs sdma_v4_0_vm_pte_funcs = {
2693 	.copy_pte_num_dw = 7,
2694 	.copy_pte = sdma_v4_0_vm_copy_pte,
2695 
2696 	.write_pte = sdma_v4_0_vm_write_pte,
2697 	.set_pte_pde = sdma_v4_0_vm_set_pte_pde,
2698 };
2699 
sdma_v4_0_set_vm_pte_funcs(struct amdgpu_device * adev)2700 static void sdma_v4_0_set_vm_pte_funcs(struct amdgpu_device *adev)
2701 {
2702 	struct drm_gpu_scheduler *sched;
2703 	unsigned i;
2704 
2705 	adev->vm_manager.vm_pte_funcs = &sdma_v4_0_vm_pte_funcs;
2706 	for (i = 0; i < adev->sdma.num_instances; i++) {
2707 		if (adev->sdma.has_page_queue)
2708 			sched = &adev->sdma.instance[i].page.sched;
2709 		else
2710 			sched = &adev->sdma.instance[i].ring.sched;
2711 		adev->vm_manager.vm_pte_scheds[i] = sched;
2712 	}
2713 	adev->vm_manager.vm_pte_num_scheds = adev->sdma.num_instances;
2714 }
2715 
sdma_v4_0_get_ras_error_count(uint32_t value,uint32_t instance,uint32_t * sec_count)2716 static void sdma_v4_0_get_ras_error_count(uint32_t value,
2717 					uint32_t instance,
2718 					uint32_t *sec_count)
2719 {
2720 	uint32_t i;
2721 	uint32_t sec_cnt;
2722 
2723 	/* double bits error (multiple bits) error detection is not supported */
2724 	for (i = 0; i < ARRAY_SIZE(sdma_v4_0_ras_fields); i++) {
2725 		/* the SDMA_EDC_COUNTER register in each sdma instance
2726 		 * shares the same sed shift_mask
2727 		 * */
2728 		sec_cnt = (value &
2729 			sdma_v4_0_ras_fields[i].sec_count_mask) >>
2730 			sdma_v4_0_ras_fields[i].sec_count_shift;
2731 		if (sec_cnt) {
2732 			DRM_INFO("Detected %s in SDMA%d, SED %d\n",
2733 				sdma_v4_0_ras_fields[i].name,
2734 				instance, sec_cnt);
2735 			*sec_count += sec_cnt;
2736 		}
2737 	}
2738 }
2739 
sdma_v4_0_query_ras_error_count_by_instance(struct amdgpu_device * adev,uint32_t instance,void * ras_error_status)2740 static int sdma_v4_0_query_ras_error_count_by_instance(struct amdgpu_device *adev,
2741 			uint32_t instance, void *ras_error_status)
2742 {
2743 	struct ras_err_data *err_data = (struct ras_err_data *)ras_error_status;
2744 	uint32_t sec_count = 0;
2745 	uint32_t reg_value = 0;
2746 
2747 	reg_value = RREG32_SDMA(instance, mmSDMA0_EDC_COUNTER);
2748 	/* double bit error is not supported */
2749 	if (reg_value)
2750 		sdma_v4_0_get_ras_error_count(reg_value,
2751 				instance, &sec_count);
2752 	/* err_data->ce_count should be initialized to 0
2753 	 * before calling into this function */
2754 	err_data->ce_count += sec_count;
2755 	/* double bit error is not supported
2756 	 * set ue count to 0 */
2757 	err_data->ue_count = 0;
2758 
2759 	return 0;
2760 };
2761 
sdma_v4_0_query_ras_error_count(struct amdgpu_device * adev,void * ras_error_status)2762 static void sdma_v4_0_query_ras_error_count(struct amdgpu_device *adev,  void *ras_error_status)
2763 {
2764 	int i = 0;
2765 
2766 	for (i = 0; i < adev->sdma.num_instances; i++) {
2767 		if (sdma_v4_0_query_ras_error_count_by_instance(adev, i, ras_error_status)) {
2768 			dev_err(adev->dev, "Query ras error count failed in SDMA%d\n", i);
2769 			return;
2770 		}
2771 	}
2772 }
2773 
sdma_v4_0_reset_ras_error_count(struct amdgpu_device * adev)2774 static void sdma_v4_0_reset_ras_error_count(struct amdgpu_device *adev)
2775 {
2776 	int i;
2777 
2778 	/* read back edc counter registers to clear the counters */
2779 	if (amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__SDMA)) {
2780 		for (i = 0; i < adev->sdma.num_instances; i++)
2781 			RREG32_SDMA(i, mmSDMA0_EDC_COUNTER);
2782 	}
2783 }
2784 
2785 const struct amdgpu_ras_block_hw_ops sdma_v4_0_ras_hw_ops = {
2786 	.query_ras_error_count = sdma_v4_0_query_ras_error_count,
2787 	.reset_ras_error_count = sdma_v4_0_reset_ras_error_count,
2788 };
2789 
2790 static struct amdgpu_sdma_ras sdma_v4_0_ras = {
2791 	.ras_block = {
2792 		.hw_ops = &sdma_v4_0_ras_hw_ops,
2793 		.ras_cb = sdma_v4_0_process_ras_data_cb,
2794 	},
2795 };
2796 
sdma_v4_0_set_ras_funcs(struct amdgpu_device * adev)2797 static void sdma_v4_0_set_ras_funcs(struct amdgpu_device *adev)
2798 {
2799 	switch (adev->ip_versions[SDMA0_HWIP][0]) {
2800 	case IP_VERSION(4, 2, 0):
2801 	case IP_VERSION(4, 2, 2):
2802 		adev->sdma.ras = &sdma_v4_0_ras;
2803 		break;
2804 	case IP_VERSION(4, 4, 0):
2805 		adev->sdma.ras = &sdma_v4_4_ras;
2806 		break;
2807 	default:
2808 		break;
2809 	}
2810 
2811 	if (adev->sdma.ras) {
2812 		amdgpu_ras_register_ras_block(adev, &adev->sdma.ras->ras_block);
2813 
2814 		strcpy(adev->sdma.ras->ras_block.ras_comm.name, "sdma");
2815 		adev->sdma.ras->ras_block.ras_comm.block = AMDGPU_RAS_BLOCK__SDMA;
2816 		adev->sdma.ras->ras_block.ras_comm.type = AMDGPU_RAS_ERROR__MULTI_UNCORRECTABLE;
2817 		adev->sdma.ras_if = &adev->sdma.ras->ras_block.ras_comm;
2818 
2819 		/* If don't define special ras_late_init function, use default ras_late_init */
2820 		if (!adev->sdma.ras->ras_block.ras_late_init)
2821 			adev->sdma.ras->ras_block.ras_late_init = amdgpu_sdma_ras_late_init;
2822 
2823 		/* If not defined special ras_cb function, use default ras_cb */
2824 		if (!adev->sdma.ras->ras_block.ras_cb)
2825 			adev->sdma.ras->ras_block.ras_cb = amdgpu_sdma_process_ras_data_cb;
2826 	}
2827 }
2828 
2829 const struct amdgpu_ip_block_version sdma_v4_0_ip_block = {
2830 	.type = AMD_IP_BLOCK_TYPE_SDMA,
2831 	.major = 4,
2832 	.minor = 0,
2833 	.rev = 0,
2834 	.funcs = &sdma_v4_0_ip_funcs,
2835 };
2836