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 
564 /**
565  * sdma_v4_0_init_microcode - load ucode images from disk
566  *
567  * @adev: amdgpu_device pointer
568  *
569  * Use the firmware interface to load the ucode images into
570  * the driver (not loaded into hw).
571  * Returns 0 on success, error on failure.
572  */
573 
574 // emulation only, won't work on real chip
575 // vega10 real chip need to use PSP to load firmware
sdma_v4_0_init_microcode(struct amdgpu_device * adev)576 static int sdma_v4_0_init_microcode(struct amdgpu_device *adev)
577 {
578 	int ret, i;
579 
580 	for (i = 0; i < adev->sdma.num_instances; i++) {
581 		if (adev->ip_versions[SDMA0_HWIP][0] == IP_VERSION(4, 2, 2) ||
582                     adev->ip_versions[SDMA0_HWIP][0] == IP_VERSION(4, 4, 0)) {
583 			/* Acturus & Aldebaran will leverage the same FW memory
584 			   for every SDMA instance */
585 			ret = amdgpu_sdma_init_microcode(adev, 0, true);
586 			break;
587 		} else {
588 			ret = amdgpu_sdma_init_microcode(adev, i, false);
589 			if (ret)
590 				return ret;
591 		}
592 	}
593 
594 	return ret;
595 }
596 
597 /**
598  * sdma_v4_0_ring_get_rptr - get the current read pointer
599  *
600  * @ring: amdgpu ring pointer
601  *
602  * Get the current rptr from the hardware (VEGA10+).
603  */
sdma_v4_0_ring_get_rptr(struct amdgpu_ring * ring)604 static uint64_t sdma_v4_0_ring_get_rptr(struct amdgpu_ring *ring)
605 {
606 	u64 *rptr;
607 
608 	/* XXX check if swapping is necessary on BE */
609 	rptr = ((u64 *)ring->rptr_cpu_addr);
610 
611 	DRM_DEBUG("rptr before shift == 0x%016llx\n", *rptr);
612 	return ((*rptr) >> 2);
613 }
614 
615 /**
616  * sdma_v4_0_ring_get_wptr - get the current write pointer
617  *
618  * @ring: amdgpu ring pointer
619  *
620  * Get the current wptr from the hardware (VEGA10+).
621  */
sdma_v4_0_ring_get_wptr(struct amdgpu_ring * ring)622 static uint64_t sdma_v4_0_ring_get_wptr(struct amdgpu_ring *ring)
623 {
624 	struct amdgpu_device *adev = ring->adev;
625 	u64 wptr;
626 
627 	if (ring->use_doorbell) {
628 		/* XXX check if swapping is necessary on BE */
629 		wptr = READ_ONCE(*((u64 *)ring->wptr_cpu_addr));
630 		DRM_DEBUG("wptr/doorbell before shift == 0x%016llx\n", wptr);
631 	} else {
632 		wptr = RREG32_SDMA(ring->me, mmSDMA0_GFX_RB_WPTR_HI);
633 		wptr = wptr << 32;
634 		wptr |= RREG32_SDMA(ring->me, mmSDMA0_GFX_RB_WPTR);
635 		DRM_DEBUG("wptr before shift [%i] wptr == 0x%016llx\n",
636 				ring->me, wptr);
637 	}
638 
639 	return wptr >> 2;
640 }
641 
642 /**
643  * sdma_v4_0_ring_set_wptr - commit the write pointer
644  *
645  * @ring: amdgpu ring pointer
646  *
647  * Write the wptr back to the hardware (VEGA10+).
648  */
sdma_v4_0_ring_set_wptr(struct amdgpu_ring * ring)649 static void sdma_v4_0_ring_set_wptr(struct amdgpu_ring *ring)
650 {
651 	struct amdgpu_device *adev = ring->adev;
652 
653 	DRM_DEBUG("Setting write pointer\n");
654 	if (ring->use_doorbell) {
655 		u64 *wb = (u64 *)ring->wptr_cpu_addr;
656 
657 		DRM_DEBUG("Using doorbell -- "
658 				"wptr_offs == 0x%08x "
659 				"lower_32_bits(ring->wptr << 2) == 0x%08x "
660 				"upper_32_bits(ring->wptr << 2) == 0x%08x\n",
661 				ring->wptr_offs,
662 				lower_32_bits(ring->wptr << 2),
663 				upper_32_bits(ring->wptr << 2));
664 		/* XXX check if swapping is necessary on BE */
665 		WRITE_ONCE(*wb, (ring->wptr << 2));
666 		DRM_DEBUG("calling WDOORBELL64(0x%08x, 0x%016llx)\n",
667 				ring->doorbell_index, ring->wptr << 2);
668 		WDOORBELL64(ring->doorbell_index, ring->wptr << 2);
669 	} else {
670 		DRM_DEBUG("Not using doorbell -- "
671 				"mmSDMA%i_GFX_RB_WPTR == 0x%08x "
672 				"mmSDMA%i_GFX_RB_WPTR_HI == 0x%08x\n",
673 				ring->me,
674 				lower_32_bits(ring->wptr << 2),
675 				ring->me,
676 				upper_32_bits(ring->wptr << 2));
677 		WREG32_SDMA(ring->me, mmSDMA0_GFX_RB_WPTR,
678 			    lower_32_bits(ring->wptr << 2));
679 		WREG32_SDMA(ring->me, mmSDMA0_GFX_RB_WPTR_HI,
680 			    upper_32_bits(ring->wptr << 2));
681 	}
682 }
683 
684 /**
685  * sdma_v4_0_page_ring_get_wptr - get the current write pointer
686  *
687  * @ring: amdgpu ring pointer
688  *
689  * Get the current wptr from the hardware (VEGA10+).
690  */
sdma_v4_0_page_ring_get_wptr(struct amdgpu_ring * ring)691 static uint64_t sdma_v4_0_page_ring_get_wptr(struct amdgpu_ring *ring)
692 {
693 	struct amdgpu_device *adev = ring->adev;
694 	u64 wptr;
695 
696 	if (ring->use_doorbell) {
697 		/* XXX check if swapping is necessary on BE */
698 		wptr = READ_ONCE(*((u64 *)ring->wptr_cpu_addr));
699 	} else {
700 		wptr = RREG32_SDMA(ring->me, mmSDMA0_PAGE_RB_WPTR_HI);
701 		wptr = wptr << 32;
702 		wptr |= RREG32_SDMA(ring->me, mmSDMA0_PAGE_RB_WPTR);
703 	}
704 
705 	return wptr >> 2;
706 }
707 
708 /**
709  * sdma_v4_0_page_ring_set_wptr - commit the write pointer
710  *
711  * @ring: amdgpu ring pointer
712  *
713  * Write the wptr back to the hardware (VEGA10+).
714  */
sdma_v4_0_page_ring_set_wptr(struct amdgpu_ring * ring)715 static void sdma_v4_0_page_ring_set_wptr(struct amdgpu_ring *ring)
716 {
717 	struct amdgpu_device *adev = ring->adev;
718 
719 	if (ring->use_doorbell) {
720 		u64 *wb = (u64 *)ring->wptr_cpu_addr;
721 
722 		/* XXX check if swapping is necessary on BE */
723 		WRITE_ONCE(*wb, (ring->wptr << 2));
724 		WDOORBELL64(ring->doorbell_index, ring->wptr << 2);
725 	} else {
726 		uint64_t wptr = ring->wptr << 2;
727 
728 		WREG32_SDMA(ring->me, mmSDMA0_PAGE_RB_WPTR,
729 			    lower_32_bits(wptr));
730 		WREG32_SDMA(ring->me, mmSDMA0_PAGE_RB_WPTR_HI,
731 			    upper_32_bits(wptr));
732 	}
733 }
734 
sdma_v4_0_ring_insert_nop(struct amdgpu_ring * ring,uint32_t count)735 static void sdma_v4_0_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count)
736 {
737 	struct amdgpu_sdma_instance *sdma = amdgpu_sdma_get_instance_from_ring(ring);
738 	int i;
739 
740 	for (i = 0; i < count; i++)
741 		if (sdma && sdma->burst_nop && (i == 0))
742 			amdgpu_ring_write(ring, ring->funcs->nop |
743 				SDMA_PKT_NOP_HEADER_COUNT(count - 1));
744 		else
745 			amdgpu_ring_write(ring, ring->funcs->nop);
746 }
747 
748 /**
749  * sdma_v4_0_ring_emit_ib - Schedule an IB on the DMA engine
750  *
751  * @ring: amdgpu ring pointer
752  * @job: job to retrieve vmid from
753  * @ib: IB object to schedule
754  * @flags: unused
755  *
756  * Schedule an IB in the DMA ring (VEGA10).
757  */
sdma_v4_0_ring_emit_ib(struct amdgpu_ring * ring,struct amdgpu_job * job,struct amdgpu_ib * ib,uint32_t flags)758 static void sdma_v4_0_ring_emit_ib(struct amdgpu_ring *ring,
759 				   struct amdgpu_job *job,
760 				   struct amdgpu_ib *ib,
761 				   uint32_t flags)
762 {
763 	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
764 
765 	/* IB packet must end on a 8 DW boundary */
766 	sdma_v4_0_ring_insert_nop(ring, (2 - lower_32_bits(ring->wptr)) & 7);
767 
768 	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_INDIRECT) |
769 			  SDMA_PKT_INDIRECT_HEADER_VMID(vmid & 0xf));
770 	/* base must be 32 byte aligned */
771 	amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr) & 0xffffffe0);
772 	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr));
773 	amdgpu_ring_write(ring, ib->length_dw);
774 	amdgpu_ring_write(ring, 0);
775 	amdgpu_ring_write(ring, 0);
776 
777 }
778 
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)779 static void sdma_v4_0_wait_reg_mem(struct amdgpu_ring *ring,
780 				   int mem_space, int hdp,
781 				   uint32_t addr0, uint32_t addr1,
782 				   uint32_t ref, uint32_t mask,
783 				   uint32_t inv)
784 {
785 	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_POLL_REGMEM) |
786 			  SDMA_PKT_POLL_REGMEM_HEADER_HDP_FLUSH(hdp) |
787 			  SDMA_PKT_POLL_REGMEM_HEADER_MEM_POLL(mem_space) |
788 			  SDMA_PKT_POLL_REGMEM_HEADER_FUNC(3)); /* == */
789 	if (mem_space) {
790 		/* memory */
791 		amdgpu_ring_write(ring, addr0);
792 		amdgpu_ring_write(ring, addr1);
793 	} else {
794 		/* registers */
795 		amdgpu_ring_write(ring, addr0 << 2);
796 		amdgpu_ring_write(ring, addr1 << 2);
797 	}
798 	amdgpu_ring_write(ring, ref); /* reference */
799 	amdgpu_ring_write(ring, mask); /* mask */
800 	amdgpu_ring_write(ring, SDMA_PKT_POLL_REGMEM_DW5_RETRY_COUNT(0xfff) |
801 			  SDMA_PKT_POLL_REGMEM_DW5_INTERVAL(inv)); /* retry count, poll interval */
802 }
803 
804 /**
805  * sdma_v4_0_ring_emit_hdp_flush - emit an hdp flush on the DMA ring
806  *
807  * @ring: amdgpu ring pointer
808  *
809  * Emit an hdp flush packet on the requested DMA ring.
810  */
sdma_v4_0_ring_emit_hdp_flush(struct amdgpu_ring * ring)811 static void sdma_v4_0_ring_emit_hdp_flush(struct amdgpu_ring *ring)
812 {
813 	struct amdgpu_device *adev = ring->adev;
814 	u32 ref_and_mask = 0;
815 	const struct nbio_hdp_flush_reg *nbio_hf_reg = adev->nbio.hdp_flush_reg;
816 
817 	ref_and_mask = nbio_hf_reg->ref_and_mask_sdma0 << ring->me;
818 
819 	sdma_v4_0_wait_reg_mem(ring, 0, 1,
820 			       adev->nbio.funcs->get_hdp_flush_done_offset(adev),
821 			       adev->nbio.funcs->get_hdp_flush_req_offset(adev),
822 			       ref_and_mask, ref_and_mask, 10);
823 }
824 
825 /**
826  * sdma_v4_0_ring_emit_fence - emit a fence on the DMA ring
827  *
828  * @ring: amdgpu ring pointer
829  * @addr: address
830  * @seq: sequence number
831  * @flags: fence related flags
832  *
833  * Add a DMA fence packet to the ring to write
834  * the fence seq number and DMA trap packet to generate
835  * an interrupt if needed (VEGA10).
836  */
sdma_v4_0_ring_emit_fence(struct amdgpu_ring * ring,u64 addr,u64 seq,unsigned flags)837 static void sdma_v4_0_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, u64 seq,
838 				      unsigned flags)
839 {
840 	bool write64bit = flags & AMDGPU_FENCE_FLAG_64BIT;
841 	/* write the fence */
842 	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_FENCE));
843 	/* zero in first two bits */
844 	BUG_ON(addr & 0x3);
845 	amdgpu_ring_write(ring, lower_32_bits(addr));
846 	amdgpu_ring_write(ring, upper_32_bits(addr));
847 	amdgpu_ring_write(ring, lower_32_bits(seq));
848 
849 	/* optionally write high bits as well */
850 	if (write64bit) {
851 		addr += 4;
852 		amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_FENCE));
853 		/* zero in first two bits */
854 		BUG_ON(addr & 0x3);
855 		amdgpu_ring_write(ring, lower_32_bits(addr));
856 		amdgpu_ring_write(ring, upper_32_bits(addr));
857 		amdgpu_ring_write(ring, upper_32_bits(seq));
858 	}
859 
860 	/* generate an interrupt */
861 	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_TRAP));
862 	amdgpu_ring_write(ring, SDMA_PKT_TRAP_INT_CONTEXT_INT_CONTEXT(0));
863 }
864 
865 
866 /**
867  * sdma_v4_0_gfx_enable - enable the gfx async dma engines
868  *
869  * @adev: amdgpu_device pointer
870  * @enable: enable SDMA RB/IB
871  * control the gfx async dma ring buffers (VEGA10).
872  */
sdma_v4_0_gfx_enable(struct amdgpu_device * adev,bool enable)873 static void sdma_v4_0_gfx_enable(struct amdgpu_device *adev, bool enable)
874 {
875 	u32 rb_cntl, ib_cntl;
876 	int i;
877 
878 	amdgpu_sdma_unset_buffer_funcs_helper(adev);
879 
880 	for (i = 0; i < adev->sdma.num_instances; i++) {
881 		rb_cntl = RREG32_SDMA(i, mmSDMA0_GFX_RB_CNTL);
882 		rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_GFX_RB_CNTL, RB_ENABLE, enable ? 1 : 0);
883 		WREG32_SDMA(i, mmSDMA0_GFX_RB_CNTL, rb_cntl);
884 		ib_cntl = RREG32_SDMA(i, mmSDMA0_GFX_IB_CNTL);
885 		ib_cntl = REG_SET_FIELD(ib_cntl, SDMA0_GFX_IB_CNTL, IB_ENABLE, enable ? 1 : 0);
886 		WREG32_SDMA(i, mmSDMA0_GFX_IB_CNTL, ib_cntl);
887 	}
888 }
889 
890 /**
891  * sdma_v4_0_rlc_stop - stop the compute async dma engines
892  *
893  * @adev: amdgpu_device pointer
894  *
895  * Stop the compute async dma queues (VEGA10).
896  */
sdma_v4_0_rlc_stop(struct amdgpu_device * adev)897 static void sdma_v4_0_rlc_stop(struct amdgpu_device *adev)
898 {
899 	/* XXX todo */
900 }
901 
902 /**
903  * sdma_v4_0_page_stop - stop the page async dma engines
904  *
905  * @adev: amdgpu_device pointer
906  *
907  * Stop the page async dma ring buffers (VEGA10).
908  */
sdma_v4_0_page_stop(struct amdgpu_device * adev)909 static void sdma_v4_0_page_stop(struct amdgpu_device *adev)
910 {
911 	u32 rb_cntl, ib_cntl;
912 	int i;
913 
914 	amdgpu_sdma_unset_buffer_funcs_helper(adev);
915 
916 	for (i = 0; i < adev->sdma.num_instances; i++) {
917 		rb_cntl = RREG32_SDMA(i, mmSDMA0_PAGE_RB_CNTL);
918 		rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_PAGE_RB_CNTL,
919 					RB_ENABLE, 0);
920 		WREG32_SDMA(i, mmSDMA0_PAGE_RB_CNTL, rb_cntl);
921 		ib_cntl = RREG32_SDMA(i, mmSDMA0_PAGE_IB_CNTL);
922 		ib_cntl = REG_SET_FIELD(ib_cntl, SDMA0_PAGE_IB_CNTL,
923 					IB_ENABLE, 0);
924 		WREG32_SDMA(i, mmSDMA0_PAGE_IB_CNTL, ib_cntl);
925 	}
926 }
927 
928 /**
929  * sdma_v4_0_ctx_switch_enable - stop the async dma engines context switch
930  *
931  * @adev: amdgpu_device pointer
932  * @enable: enable/disable the DMA MEs context switch.
933  *
934  * Halt or unhalt the async dma engines context switch (VEGA10).
935  */
sdma_v4_0_ctx_switch_enable(struct amdgpu_device * adev,bool enable)936 static void sdma_v4_0_ctx_switch_enable(struct amdgpu_device *adev, bool enable)
937 {
938 	u32 f32_cntl, phase_quantum = 0;
939 	int i;
940 
941 	if (amdgpu_sdma_phase_quantum) {
942 		unsigned value = amdgpu_sdma_phase_quantum;
943 		unsigned unit = 0;
944 
945 		while (value > (SDMA0_PHASE0_QUANTUM__VALUE_MASK >>
946 				SDMA0_PHASE0_QUANTUM__VALUE__SHIFT)) {
947 			value = (value + 1) >> 1;
948 			unit++;
949 		}
950 		if (unit > (SDMA0_PHASE0_QUANTUM__UNIT_MASK >>
951 			    SDMA0_PHASE0_QUANTUM__UNIT__SHIFT)) {
952 			value = (SDMA0_PHASE0_QUANTUM__VALUE_MASK >>
953 				 SDMA0_PHASE0_QUANTUM__VALUE__SHIFT);
954 			unit = (SDMA0_PHASE0_QUANTUM__UNIT_MASK >>
955 				SDMA0_PHASE0_QUANTUM__UNIT__SHIFT);
956 			WARN_ONCE(1,
957 			"clamping sdma_phase_quantum to %uK clock cycles\n",
958 				  value << unit);
959 		}
960 		phase_quantum =
961 			value << SDMA0_PHASE0_QUANTUM__VALUE__SHIFT |
962 			unit  << SDMA0_PHASE0_QUANTUM__UNIT__SHIFT;
963 	}
964 
965 	for (i = 0; i < adev->sdma.num_instances; i++) {
966 		f32_cntl = RREG32_SDMA(i, mmSDMA0_CNTL);
967 		f32_cntl = REG_SET_FIELD(f32_cntl, SDMA0_CNTL,
968 				AUTO_CTXSW_ENABLE, enable ? 1 : 0);
969 		if (enable && amdgpu_sdma_phase_quantum) {
970 			WREG32_SDMA(i, mmSDMA0_PHASE0_QUANTUM, phase_quantum);
971 			WREG32_SDMA(i, mmSDMA0_PHASE1_QUANTUM, phase_quantum);
972 			WREG32_SDMA(i, mmSDMA0_PHASE2_QUANTUM, phase_quantum);
973 		}
974 		WREG32_SDMA(i, mmSDMA0_CNTL, f32_cntl);
975 
976 		/*
977 		 * Enable SDMA utilization. Its only supported on
978 		 * Arcturus for the moment and firmware version 14
979 		 * and above.
980 		 */
981 		if (adev->ip_versions[SDMA0_HWIP][0] == IP_VERSION(4, 2, 2) &&
982 		    adev->sdma.instance[i].fw_version >= 14)
983 			WREG32_SDMA(i, mmSDMA0_PUB_DUMMY_REG2, enable);
984 		/* Extend page fault timeout to avoid interrupt storm */
985 		WREG32_SDMA(i, mmSDMA0_UTCL1_TIMEOUT, 0x00800080);
986 	}
987 
988 }
989 
990 /**
991  * sdma_v4_0_enable - stop the async dma engines
992  *
993  * @adev: amdgpu_device pointer
994  * @enable: enable/disable the DMA MEs.
995  *
996  * Halt or unhalt the async dma engines (VEGA10).
997  */
sdma_v4_0_enable(struct amdgpu_device * adev,bool enable)998 static void sdma_v4_0_enable(struct amdgpu_device *adev, bool enable)
999 {
1000 	u32 f32_cntl;
1001 	int i;
1002 
1003 	if (!enable) {
1004 		sdma_v4_0_gfx_enable(adev, enable);
1005 		sdma_v4_0_rlc_stop(adev);
1006 		if (adev->sdma.has_page_queue)
1007 			sdma_v4_0_page_stop(adev);
1008 	}
1009 
1010 	for (i = 0; i < adev->sdma.num_instances; i++) {
1011 		f32_cntl = RREG32_SDMA(i, mmSDMA0_F32_CNTL);
1012 		f32_cntl = REG_SET_FIELD(f32_cntl, SDMA0_F32_CNTL, HALT, enable ? 0 : 1);
1013 		WREG32_SDMA(i, mmSDMA0_F32_CNTL, f32_cntl);
1014 	}
1015 }
1016 
1017 /*
1018  * sdma_v4_0_rb_cntl - get parameters for rb_cntl
1019  */
sdma_v4_0_rb_cntl(struct amdgpu_ring * ring,uint32_t rb_cntl)1020 static uint32_t sdma_v4_0_rb_cntl(struct amdgpu_ring *ring, uint32_t rb_cntl)
1021 {
1022 	/* Set ring buffer size in dwords */
1023 	uint32_t rb_bufsz = order_base_2(ring->ring_size / 4);
1024 
1025 	rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_GFX_RB_CNTL, RB_SIZE, rb_bufsz);
1026 #ifdef __BIG_ENDIAN
1027 	rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_GFX_RB_CNTL, RB_SWAP_ENABLE, 1);
1028 	rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_GFX_RB_CNTL,
1029 				RPTR_WRITEBACK_SWAP_ENABLE, 1);
1030 #endif
1031 	return rb_cntl;
1032 }
1033 
1034 /**
1035  * sdma_v4_0_gfx_resume - setup and start the async dma engines
1036  *
1037  * @adev: amdgpu_device pointer
1038  * @i: instance to resume
1039  *
1040  * Set up the gfx DMA ring buffers and enable them (VEGA10).
1041  * Returns 0 for success, error for failure.
1042  */
sdma_v4_0_gfx_resume(struct amdgpu_device * adev,unsigned int i)1043 static void sdma_v4_0_gfx_resume(struct amdgpu_device *adev, unsigned int i)
1044 {
1045 	struct amdgpu_ring *ring = &adev->sdma.instance[i].ring;
1046 	u32 rb_cntl, ib_cntl, wptr_poll_cntl;
1047 	u32 doorbell;
1048 	u32 doorbell_offset;
1049 	u64 wptr_gpu_addr;
1050 
1051 	rb_cntl = RREG32_SDMA(i, mmSDMA0_GFX_RB_CNTL);
1052 	rb_cntl = sdma_v4_0_rb_cntl(ring, rb_cntl);
1053 	WREG32_SDMA(i, mmSDMA0_GFX_RB_CNTL, rb_cntl);
1054 
1055 	/* Initialize the ring buffer's read and write pointers */
1056 	WREG32_SDMA(i, mmSDMA0_GFX_RB_RPTR, 0);
1057 	WREG32_SDMA(i, mmSDMA0_GFX_RB_RPTR_HI, 0);
1058 	WREG32_SDMA(i, mmSDMA0_GFX_RB_WPTR, 0);
1059 	WREG32_SDMA(i, mmSDMA0_GFX_RB_WPTR_HI, 0);
1060 
1061 	/* set the wb address whether it's enabled or not */
1062 	WREG32_SDMA(i, mmSDMA0_GFX_RB_RPTR_ADDR_HI,
1063 	       upper_32_bits(ring->rptr_gpu_addr) & 0xFFFFFFFF);
1064 	WREG32_SDMA(i, mmSDMA0_GFX_RB_RPTR_ADDR_LO,
1065 	       lower_32_bits(ring->rptr_gpu_addr) & 0xFFFFFFFC);
1066 
1067 	rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_GFX_RB_CNTL,
1068 				RPTR_WRITEBACK_ENABLE, 1);
1069 
1070 	WREG32_SDMA(i, mmSDMA0_GFX_RB_BASE, ring->gpu_addr >> 8);
1071 	WREG32_SDMA(i, mmSDMA0_GFX_RB_BASE_HI, ring->gpu_addr >> 40);
1072 
1073 	ring->wptr = 0;
1074 
1075 	/* before programing wptr to a less value, need set minor_ptr_update first */
1076 	WREG32_SDMA(i, mmSDMA0_GFX_MINOR_PTR_UPDATE, 1);
1077 
1078 	doorbell = RREG32_SDMA(i, mmSDMA0_GFX_DOORBELL);
1079 	doorbell_offset = RREG32_SDMA(i, mmSDMA0_GFX_DOORBELL_OFFSET);
1080 
1081 	doorbell = REG_SET_FIELD(doorbell, SDMA0_GFX_DOORBELL, ENABLE,
1082 				 ring->use_doorbell);
1083 	doorbell_offset = REG_SET_FIELD(doorbell_offset,
1084 					SDMA0_GFX_DOORBELL_OFFSET,
1085 					OFFSET, ring->doorbell_index);
1086 	WREG32_SDMA(i, mmSDMA0_GFX_DOORBELL, doorbell);
1087 	WREG32_SDMA(i, mmSDMA0_GFX_DOORBELL_OFFSET, doorbell_offset);
1088 
1089 	sdma_v4_0_ring_set_wptr(ring);
1090 
1091 	/* set minor_ptr_update to 0 after wptr programed */
1092 	WREG32_SDMA(i, mmSDMA0_GFX_MINOR_PTR_UPDATE, 0);
1093 
1094 	/* setup the wptr shadow polling */
1095 	wptr_gpu_addr = ring->wptr_gpu_addr;
1096 	WREG32_SDMA(i, mmSDMA0_GFX_RB_WPTR_POLL_ADDR_LO,
1097 		    lower_32_bits(wptr_gpu_addr));
1098 	WREG32_SDMA(i, mmSDMA0_GFX_RB_WPTR_POLL_ADDR_HI,
1099 		    upper_32_bits(wptr_gpu_addr));
1100 	wptr_poll_cntl = RREG32_SDMA(i, mmSDMA0_GFX_RB_WPTR_POLL_CNTL);
1101 	wptr_poll_cntl = REG_SET_FIELD(wptr_poll_cntl,
1102 				       SDMA0_GFX_RB_WPTR_POLL_CNTL,
1103 				       F32_POLL_ENABLE, amdgpu_sriov_vf(adev)? 1 : 0);
1104 	WREG32_SDMA(i, mmSDMA0_GFX_RB_WPTR_POLL_CNTL, wptr_poll_cntl);
1105 
1106 	/* enable DMA RB */
1107 	rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_GFX_RB_CNTL, RB_ENABLE, 1);
1108 	WREG32_SDMA(i, mmSDMA0_GFX_RB_CNTL, rb_cntl);
1109 
1110 	ib_cntl = RREG32_SDMA(i, mmSDMA0_GFX_IB_CNTL);
1111 	ib_cntl = REG_SET_FIELD(ib_cntl, SDMA0_GFX_IB_CNTL, IB_ENABLE, 1);
1112 #ifdef __BIG_ENDIAN
1113 	ib_cntl = REG_SET_FIELD(ib_cntl, SDMA0_GFX_IB_CNTL, IB_SWAP_ENABLE, 1);
1114 #endif
1115 	/* enable DMA IBs */
1116 	WREG32_SDMA(i, mmSDMA0_GFX_IB_CNTL, ib_cntl);
1117 }
1118 
1119 /**
1120  * sdma_v4_0_page_resume - setup and start the async dma engines
1121  *
1122  * @adev: amdgpu_device pointer
1123  * @i: instance to resume
1124  *
1125  * Set up the page DMA ring buffers and enable them (VEGA10).
1126  * Returns 0 for success, error for failure.
1127  */
sdma_v4_0_page_resume(struct amdgpu_device * adev,unsigned int i)1128 static void sdma_v4_0_page_resume(struct amdgpu_device *adev, unsigned int i)
1129 {
1130 	struct amdgpu_ring *ring = &adev->sdma.instance[i].page;
1131 	u32 rb_cntl, ib_cntl, wptr_poll_cntl;
1132 	u32 doorbell;
1133 	u32 doorbell_offset;
1134 	u64 wptr_gpu_addr;
1135 
1136 	rb_cntl = RREG32_SDMA(i, mmSDMA0_PAGE_RB_CNTL);
1137 	rb_cntl = sdma_v4_0_rb_cntl(ring, rb_cntl);
1138 	WREG32_SDMA(i, mmSDMA0_PAGE_RB_CNTL, rb_cntl);
1139 
1140 	/* Initialize the ring buffer's read and write pointers */
1141 	WREG32_SDMA(i, mmSDMA0_PAGE_RB_RPTR, 0);
1142 	WREG32_SDMA(i, mmSDMA0_PAGE_RB_RPTR_HI, 0);
1143 	WREG32_SDMA(i, mmSDMA0_PAGE_RB_WPTR, 0);
1144 	WREG32_SDMA(i, mmSDMA0_PAGE_RB_WPTR_HI, 0);
1145 
1146 	/* set the wb address whether it's enabled or not */
1147 	WREG32_SDMA(i, mmSDMA0_PAGE_RB_RPTR_ADDR_HI,
1148 	       upper_32_bits(ring->rptr_gpu_addr) & 0xFFFFFFFF);
1149 	WREG32_SDMA(i, mmSDMA0_PAGE_RB_RPTR_ADDR_LO,
1150 	       lower_32_bits(ring->rptr_gpu_addr) & 0xFFFFFFFC);
1151 
1152 	rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_PAGE_RB_CNTL,
1153 				RPTR_WRITEBACK_ENABLE, 1);
1154 
1155 	WREG32_SDMA(i, mmSDMA0_PAGE_RB_BASE, ring->gpu_addr >> 8);
1156 	WREG32_SDMA(i, mmSDMA0_PAGE_RB_BASE_HI, ring->gpu_addr >> 40);
1157 
1158 	ring->wptr = 0;
1159 
1160 	/* before programing wptr to a less value, need set minor_ptr_update first */
1161 	WREG32_SDMA(i, mmSDMA0_PAGE_MINOR_PTR_UPDATE, 1);
1162 
1163 	doorbell = RREG32_SDMA(i, mmSDMA0_PAGE_DOORBELL);
1164 	doorbell_offset = RREG32_SDMA(i, mmSDMA0_PAGE_DOORBELL_OFFSET);
1165 
1166 	doorbell = REG_SET_FIELD(doorbell, SDMA0_PAGE_DOORBELL, ENABLE,
1167 				 ring->use_doorbell);
1168 	doorbell_offset = REG_SET_FIELD(doorbell_offset,
1169 					SDMA0_PAGE_DOORBELL_OFFSET,
1170 					OFFSET, ring->doorbell_index);
1171 	WREG32_SDMA(i, mmSDMA0_PAGE_DOORBELL, doorbell);
1172 	WREG32_SDMA(i, mmSDMA0_PAGE_DOORBELL_OFFSET, doorbell_offset);
1173 
1174 	/* paging queue doorbell range is setup at sdma_v4_0_gfx_resume */
1175 	sdma_v4_0_page_ring_set_wptr(ring);
1176 
1177 	/* set minor_ptr_update to 0 after wptr programed */
1178 	WREG32_SDMA(i, mmSDMA0_PAGE_MINOR_PTR_UPDATE, 0);
1179 
1180 	/* setup the wptr shadow polling */
1181 	wptr_gpu_addr = ring->wptr_gpu_addr;
1182 	WREG32_SDMA(i, mmSDMA0_PAGE_RB_WPTR_POLL_ADDR_LO,
1183 		    lower_32_bits(wptr_gpu_addr));
1184 	WREG32_SDMA(i, mmSDMA0_PAGE_RB_WPTR_POLL_ADDR_HI,
1185 		    upper_32_bits(wptr_gpu_addr));
1186 	wptr_poll_cntl = RREG32_SDMA(i, mmSDMA0_PAGE_RB_WPTR_POLL_CNTL);
1187 	wptr_poll_cntl = REG_SET_FIELD(wptr_poll_cntl,
1188 				       SDMA0_PAGE_RB_WPTR_POLL_CNTL,
1189 				       F32_POLL_ENABLE, amdgpu_sriov_vf(adev)? 1 : 0);
1190 	WREG32_SDMA(i, mmSDMA0_PAGE_RB_WPTR_POLL_CNTL, wptr_poll_cntl);
1191 
1192 	/* enable DMA RB */
1193 	rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_PAGE_RB_CNTL, RB_ENABLE, 1);
1194 	WREG32_SDMA(i, mmSDMA0_PAGE_RB_CNTL, rb_cntl);
1195 
1196 	ib_cntl = RREG32_SDMA(i, mmSDMA0_PAGE_IB_CNTL);
1197 	ib_cntl = REG_SET_FIELD(ib_cntl, SDMA0_PAGE_IB_CNTL, IB_ENABLE, 1);
1198 #ifdef __BIG_ENDIAN
1199 	ib_cntl = REG_SET_FIELD(ib_cntl, SDMA0_PAGE_IB_CNTL, IB_SWAP_ENABLE, 1);
1200 #endif
1201 	/* enable DMA IBs */
1202 	WREG32_SDMA(i, mmSDMA0_PAGE_IB_CNTL, ib_cntl);
1203 }
1204 
1205 static void
sdma_v4_1_update_power_gating(struct amdgpu_device * adev,bool enable)1206 sdma_v4_1_update_power_gating(struct amdgpu_device *adev, bool enable)
1207 {
1208 	uint32_t def, data;
1209 
1210 	if (enable && (adev->pg_flags & AMD_PG_SUPPORT_SDMA)) {
1211 		/* enable idle interrupt */
1212 		def = data = RREG32(SOC15_REG_OFFSET(SDMA0, 0, mmSDMA0_CNTL));
1213 		data |= SDMA0_CNTL__CTXEMPTY_INT_ENABLE_MASK;
1214 
1215 		if (data != def)
1216 			WREG32(SOC15_REG_OFFSET(SDMA0, 0, mmSDMA0_CNTL), data);
1217 	} else {
1218 		/* disable idle interrupt */
1219 		def = data = RREG32(SOC15_REG_OFFSET(SDMA0, 0, mmSDMA0_CNTL));
1220 		data &= ~SDMA0_CNTL__CTXEMPTY_INT_ENABLE_MASK;
1221 		if (data != def)
1222 			WREG32(SOC15_REG_OFFSET(SDMA0, 0, mmSDMA0_CNTL), data);
1223 	}
1224 }
1225 
sdma_v4_1_init_power_gating(struct amdgpu_device * adev)1226 static void sdma_v4_1_init_power_gating(struct amdgpu_device *adev)
1227 {
1228 	uint32_t def, data;
1229 
1230 	/* Enable HW based PG. */
1231 	def = data = RREG32(SOC15_REG_OFFSET(SDMA0, 0, mmSDMA0_POWER_CNTL));
1232 	data |= SDMA0_POWER_CNTL__PG_CNTL_ENABLE_MASK;
1233 	if (data != def)
1234 		WREG32(SOC15_REG_OFFSET(SDMA0, 0, mmSDMA0_POWER_CNTL), data);
1235 
1236 	/* enable interrupt */
1237 	def = data = RREG32(SOC15_REG_OFFSET(SDMA0, 0, mmSDMA0_CNTL));
1238 	data |= SDMA0_CNTL__CTXEMPTY_INT_ENABLE_MASK;
1239 	if (data != def)
1240 		WREG32(SOC15_REG_OFFSET(SDMA0, 0, mmSDMA0_CNTL), data);
1241 
1242 	/* Configure hold time to filter in-valid power on/off request. Use default right now */
1243 	def = data = RREG32(SOC15_REG_OFFSET(SDMA0, 0, mmSDMA0_POWER_CNTL));
1244 	data &= ~SDMA0_POWER_CNTL__ON_OFF_CONDITION_HOLD_TIME_MASK;
1245 	data |= (mmSDMA0_POWER_CNTL_DEFAULT & SDMA0_POWER_CNTL__ON_OFF_CONDITION_HOLD_TIME_MASK);
1246 	/* Configure switch time for hysteresis purpose. Use default right now */
1247 	data &= ~SDMA0_POWER_CNTL__ON_OFF_STATUS_DURATION_TIME_MASK;
1248 	data |= (mmSDMA0_POWER_CNTL_DEFAULT & SDMA0_POWER_CNTL__ON_OFF_STATUS_DURATION_TIME_MASK);
1249 	if(data != def)
1250 		WREG32(SOC15_REG_OFFSET(SDMA0, 0, mmSDMA0_POWER_CNTL), data);
1251 }
1252 
sdma_v4_0_init_pg(struct amdgpu_device * adev)1253 static void sdma_v4_0_init_pg(struct amdgpu_device *adev)
1254 {
1255 	if (!(adev->pg_flags & AMD_PG_SUPPORT_SDMA))
1256 		return;
1257 
1258 	switch (adev->ip_versions[SDMA0_HWIP][0]) {
1259 	case IP_VERSION(4, 1, 0):
1260         case IP_VERSION(4, 1, 1):
1261 	case IP_VERSION(4, 1, 2):
1262 		sdma_v4_1_init_power_gating(adev);
1263 		sdma_v4_1_update_power_gating(adev, true);
1264 		break;
1265 	default:
1266 		break;
1267 	}
1268 }
1269 
1270 /**
1271  * sdma_v4_0_rlc_resume - setup and start the async dma engines
1272  *
1273  * @adev: amdgpu_device pointer
1274  *
1275  * Set up the compute DMA queues and enable them (VEGA10).
1276  * Returns 0 for success, error for failure.
1277  */
sdma_v4_0_rlc_resume(struct amdgpu_device * adev)1278 static int sdma_v4_0_rlc_resume(struct amdgpu_device *adev)
1279 {
1280 	sdma_v4_0_init_pg(adev);
1281 
1282 	return 0;
1283 }
1284 
1285 /**
1286  * sdma_v4_0_load_microcode - load the sDMA ME ucode
1287  *
1288  * @adev: amdgpu_device pointer
1289  *
1290  * Loads the sDMA0/1 ucode.
1291  * Returns 0 for success, -EINVAL if the ucode is not available.
1292  */
sdma_v4_0_load_microcode(struct amdgpu_device * adev)1293 static int sdma_v4_0_load_microcode(struct amdgpu_device *adev)
1294 {
1295 	const struct sdma_firmware_header_v1_0 *hdr;
1296 	const __le32 *fw_data;
1297 	u32 fw_size;
1298 	int i, j;
1299 
1300 	/* halt the MEs */
1301 	sdma_v4_0_enable(adev, false);
1302 
1303 	for (i = 0; i < adev->sdma.num_instances; i++) {
1304 		if (!adev->sdma.instance[i].fw)
1305 			return -EINVAL;
1306 
1307 		hdr = (const struct sdma_firmware_header_v1_0 *)adev->sdma.instance[i].fw->data;
1308 		amdgpu_ucode_print_sdma_hdr(&hdr->header);
1309 		fw_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4;
1310 
1311 		fw_data = (const __le32 *)
1312 			(adev->sdma.instance[i].fw->data +
1313 				le32_to_cpu(hdr->header.ucode_array_offset_bytes));
1314 
1315 		WREG32_SDMA(i, mmSDMA0_UCODE_ADDR, 0);
1316 
1317 		for (j = 0; j < fw_size; j++)
1318 			WREG32_SDMA(i, mmSDMA0_UCODE_DATA,
1319 				    le32_to_cpup(fw_data++));
1320 
1321 		WREG32_SDMA(i, mmSDMA0_UCODE_ADDR,
1322 			    adev->sdma.instance[i].fw_version);
1323 	}
1324 
1325 	return 0;
1326 }
1327 
1328 /**
1329  * sdma_v4_0_start - setup and start the async dma engines
1330  *
1331  * @adev: amdgpu_device pointer
1332  *
1333  * Set up the DMA engines and enable them (VEGA10).
1334  * Returns 0 for success, error for failure.
1335  */
sdma_v4_0_start(struct amdgpu_device * adev)1336 static int sdma_v4_0_start(struct amdgpu_device *adev)
1337 {
1338 	struct amdgpu_ring *ring;
1339 	int i, r = 0;
1340 
1341 	if (amdgpu_sriov_vf(adev)) {
1342 		sdma_v4_0_ctx_switch_enable(adev, false);
1343 		sdma_v4_0_enable(adev, false);
1344 	} else {
1345 
1346 		if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP) {
1347 			r = sdma_v4_0_load_microcode(adev);
1348 			if (r)
1349 				return r;
1350 		}
1351 
1352 		/* unhalt the MEs */
1353 		sdma_v4_0_enable(adev, true);
1354 		/* enable sdma ring preemption */
1355 		sdma_v4_0_ctx_switch_enable(adev, true);
1356 	}
1357 
1358 	/* start the gfx rings and rlc compute queues */
1359 	for (i = 0; i < adev->sdma.num_instances; i++) {
1360 		uint32_t temp;
1361 
1362 		WREG32_SDMA(i, mmSDMA0_SEM_WAIT_FAIL_TIMER_CNTL, 0);
1363 		sdma_v4_0_gfx_resume(adev, i);
1364 		if (adev->sdma.has_page_queue)
1365 			sdma_v4_0_page_resume(adev, i);
1366 
1367 		/* set utc l1 enable flag always to 1 */
1368 		temp = RREG32_SDMA(i, mmSDMA0_CNTL);
1369 		temp = REG_SET_FIELD(temp, SDMA0_CNTL, UTC_L1_ENABLE, 1);
1370 		WREG32_SDMA(i, mmSDMA0_CNTL, temp);
1371 
1372 		if (!amdgpu_sriov_vf(adev)) {
1373 			/* unhalt engine */
1374 			temp = RREG32_SDMA(i, mmSDMA0_F32_CNTL);
1375 			temp = REG_SET_FIELD(temp, SDMA0_F32_CNTL, HALT, 0);
1376 			WREG32_SDMA(i, mmSDMA0_F32_CNTL, temp);
1377 		}
1378 	}
1379 
1380 	if (amdgpu_sriov_vf(adev)) {
1381 		sdma_v4_0_ctx_switch_enable(adev, true);
1382 		sdma_v4_0_enable(adev, true);
1383 	} else {
1384 		r = sdma_v4_0_rlc_resume(adev);
1385 		if (r)
1386 			return r;
1387 	}
1388 
1389 	for (i = 0; i < adev->sdma.num_instances; i++) {
1390 		ring = &adev->sdma.instance[i].ring;
1391 
1392 		r = amdgpu_ring_test_helper(ring);
1393 		if (r)
1394 			return r;
1395 
1396 		if (adev->sdma.has_page_queue) {
1397 			struct amdgpu_ring *page = &adev->sdma.instance[i].page;
1398 
1399 			r = amdgpu_ring_test_helper(page);
1400 			if (r)
1401 				return r;
1402 
1403 			if (adev->mman.buffer_funcs_ring == page)
1404 				amdgpu_ttm_set_buffer_funcs_status(adev, true);
1405 		}
1406 
1407 		if (adev->mman.buffer_funcs_ring == ring)
1408 			amdgpu_ttm_set_buffer_funcs_status(adev, true);
1409 	}
1410 
1411 	return r;
1412 }
1413 
1414 /**
1415  * sdma_v4_0_ring_test_ring - simple async dma engine test
1416  *
1417  * @ring: amdgpu_ring structure holding ring information
1418  *
1419  * Test the DMA engine by writing using it to write an
1420  * value to memory. (VEGA10).
1421  * Returns 0 for success, error for failure.
1422  */
sdma_v4_0_ring_test_ring(struct amdgpu_ring * ring)1423 static int sdma_v4_0_ring_test_ring(struct amdgpu_ring *ring)
1424 {
1425 	struct amdgpu_device *adev = ring->adev;
1426 	unsigned i;
1427 	unsigned index;
1428 	int r;
1429 	u32 tmp;
1430 	u64 gpu_addr;
1431 
1432 	r = amdgpu_device_wb_get(adev, &index);
1433 	if (r)
1434 		return r;
1435 
1436 	gpu_addr = adev->wb.gpu_addr + (index * 4);
1437 	tmp = 0xCAFEDEAD;
1438 	adev->wb.wb[index] = cpu_to_le32(tmp);
1439 
1440 	r = amdgpu_ring_alloc(ring, 5);
1441 	if (r)
1442 		goto error_free_wb;
1443 
1444 	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_WRITE) |
1445 			  SDMA_PKT_HEADER_SUB_OP(SDMA_SUBOP_WRITE_LINEAR));
1446 	amdgpu_ring_write(ring, lower_32_bits(gpu_addr));
1447 	amdgpu_ring_write(ring, upper_32_bits(gpu_addr));
1448 	amdgpu_ring_write(ring, SDMA_PKT_WRITE_UNTILED_DW_3_COUNT(0));
1449 	amdgpu_ring_write(ring, 0xDEADBEEF);
1450 	amdgpu_ring_commit(ring);
1451 
1452 	for (i = 0; i < adev->usec_timeout; i++) {
1453 		tmp = le32_to_cpu(adev->wb.wb[index]);
1454 		if (tmp == 0xDEADBEEF)
1455 			break;
1456 		udelay(1);
1457 	}
1458 
1459 	if (i >= adev->usec_timeout)
1460 		r = -ETIMEDOUT;
1461 
1462 error_free_wb:
1463 	amdgpu_device_wb_free(adev, index);
1464 	return r;
1465 }
1466 
1467 /**
1468  * sdma_v4_0_ring_test_ib - test an IB on the DMA engine
1469  *
1470  * @ring: amdgpu_ring structure holding ring information
1471  * @timeout: timeout value in jiffies, or MAX_SCHEDULE_TIMEOUT
1472  *
1473  * Test a simple IB in the DMA ring (VEGA10).
1474  * Returns 0 on success, error on failure.
1475  */
sdma_v4_0_ring_test_ib(struct amdgpu_ring * ring,long timeout)1476 static int sdma_v4_0_ring_test_ib(struct amdgpu_ring *ring, long timeout)
1477 {
1478 	struct amdgpu_device *adev = ring->adev;
1479 	struct amdgpu_ib ib;
1480 	struct dma_fence *f = NULL;
1481 	unsigned index;
1482 	long r;
1483 	u32 tmp = 0;
1484 	u64 gpu_addr;
1485 
1486 	r = amdgpu_device_wb_get(adev, &index);
1487 	if (r)
1488 		return r;
1489 
1490 	gpu_addr = adev->wb.gpu_addr + (index * 4);
1491 	tmp = 0xCAFEDEAD;
1492 	adev->wb.wb[index] = cpu_to_le32(tmp);
1493 	memset(&ib, 0, sizeof(ib));
1494 	r = amdgpu_ib_get(adev, NULL, 256,
1495 					AMDGPU_IB_POOL_DIRECT, &ib);
1496 	if (r)
1497 		goto err0;
1498 
1499 	ib.ptr[0] = SDMA_PKT_HEADER_OP(SDMA_OP_WRITE) |
1500 		SDMA_PKT_HEADER_SUB_OP(SDMA_SUBOP_WRITE_LINEAR);
1501 	ib.ptr[1] = lower_32_bits(gpu_addr);
1502 	ib.ptr[2] = upper_32_bits(gpu_addr);
1503 	ib.ptr[3] = SDMA_PKT_WRITE_UNTILED_DW_3_COUNT(0);
1504 	ib.ptr[4] = 0xDEADBEEF;
1505 	ib.ptr[5] = SDMA_PKT_NOP_HEADER_OP(SDMA_OP_NOP);
1506 	ib.ptr[6] = SDMA_PKT_NOP_HEADER_OP(SDMA_OP_NOP);
1507 	ib.ptr[7] = SDMA_PKT_NOP_HEADER_OP(SDMA_OP_NOP);
1508 	ib.length_dw = 8;
1509 
1510 	r = amdgpu_ib_schedule(ring, 1, &ib, NULL, &f);
1511 	if (r)
1512 		goto err1;
1513 
1514 	r = dma_fence_wait_timeout(f, false, timeout);
1515 	if (r == 0) {
1516 		r = -ETIMEDOUT;
1517 		goto err1;
1518 	} else if (r < 0) {
1519 		goto err1;
1520 	}
1521 	tmp = le32_to_cpu(adev->wb.wb[index]);
1522 	if (tmp == 0xDEADBEEF)
1523 		r = 0;
1524 	else
1525 		r = -EINVAL;
1526 
1527 err1:
1528 	amdgpu_ib_free(adev, &ib, NULL);
1529 	dma_fence_put(f);
1530 err0:
1531 	amdgpu_device_wb_free(adev, index);
1532 	return r;
1533 }
1534 
1535 
1536 /**
1537  * sdma_v4_0_vm_copy_pte - update PTEs by copying them from the GART
1538  *
1539  * @ib: indirect buffer to fill with commands
1540  * @pe: addr of the page entry
1541  * @src: src addr to copy from
1542  * @count: number of page entries to update
1543  *
1544  * Update PTEs by copying them from the GART using sDMA (VEGA10).
1545  */
sdma_v4_0_vm_copy_pte(struct amdgpu_ib * ib,uint64_t pe,uint64_t src,unsigned count)1546 static void sdma_v4_0_vm_copy_pte(struct amdgpu_ib *ib,
1547 				  uint64_t pe, uint64_t src,
1548 				  unsigned count)
1549 {
1550 	unsigned bytes = count * 8;
1551 
1552 	ib->ptr[ib->length_dw++] = SDMA_PKT_HEADER_OP(SDMA_OP_COPY) |
1553 		SDMA_PKT_HEADER_SUB_OP(SDMA_SUBOP_COPY_LINEAR);
1554 	ib->ptr[ib->length_dw++] = bytes - 1;
1555 	ib->ptr[ib->length_dw++] = 0; /* src/dst endian swap */
1556 	ib->ptr[ib->length_dw++] = lower_32_bits(src);
1557 	ib->ptr[ib->length_dw++] = upper_32_bits(src);
1558 	ib->ptr[ib->length_dw++] = lower_32_bits(pe);
1559 	ib->ptr[ib->length_dw++] = upper_32_bits(pe);
1560 
1561 }
1562 
1563 /**
1564  * sdma_v4_0_vm_write_pte - update PTEs by writing them manually
1565  *
1566  * @ib: indirect buffer to fill with commands
1567  * @pe: addr of the page entry
1568  * @value: dst addr to write into pe
1569  * @count: number of page entries to update
1570  * @incr: increase next addr by incr bytes
1571  *
1572  * Update PTEs by writing them manually using sDMA (VEGA10).
1573  */
sdma_v4_0_vm_write_pte(struct amdgpu_ib * ib,uint64_t pe,uint64_t value,unsigned count,uint32_t incr)1574 static void sdma_v4_0_vm_write_pte(struct amdgpu_ib *ib, uint64_t pe,
1575 				   uint64_t value, unsigned count,
1576 				   uint32_t incr)
1577 {
1578 	unsigned ndw = count * 2;
1579 
1580 	ib->ptr[ib->length_dw++] = SDMA_PKT_HEADER_OP(SDMA_OP_WRITE) |
1581 		SDMA_PKT_HEADER_SUB_OP(SDMA_SUBOP_WRITE_LINEAR);
1582 	ib->ptr[ib->length_dw++] = lower_32_bits(pe);
1583 	ib->ptr[ib->length_dw++] = upper_32_bits(pe);
1584 	ib->ptr[ib->length_dw++] = ndw - 1;
1585 	for (; ndw > 0; ndw -= 2) {
1586 		ib->ptr[ib->length_dw++] = lower_32_bits(value);
1587 		ib->ptr[ib->length_dw++] = upper_32_bits(value);
1588 		value += incr;
1589 	}
1590 }
1591 
1592 /**
1593  * sdma_v4_0_vm_set_pte_pde - update the page tables using sDMA
1594  *
1595  * @ib: indirect buffer to fill with commands
1596  * @pe: addr of the page entry
1597  * @addr: dst addr to write into pe
1598  * @count: number of page entries to update
1599  * @incr: increase next addr by incr bytes
1600  * @flags: access flags
1601  *
1602  * Update the page tables using sDMA (VEGA10).
1603  */
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)1604 static void sdma_v4_0_vm_set_pte_pde(struct amdgpu_ib *ib,
1605 				     uint64_t pe,
1606 				     uint64_t addr, unsigned count,
1607 				     uint32_t incr, uint64_t flags)
1608 {
1609 	/* for physically contiguous pages (vram) */
1610 	ib->ptr[ib->length_dw++] = SDMA_PKT_HEADER_OP(SDMA_OP_PTEPDE);
1611 	ib->ptr[ib->length_dw++] = lower_32_bits(pe); /* dst addr */
1612 	ib->ptr[ib->length_dw++] = upper_32_bits(pe);
1613 	ib->ptr[ib->length_dw++] = lower_32_bits(flags); /* mask */
1614 	ib->ptr[ib->length_dw++] = upper_32_bits(flags);
1615 	ib->ptr[ib->length_dw++] = lower_32_bits(addr); /* value */
1616 	ib->ptr[ib->length_dw++] = upper_32_bits(addr);
1617 	ib->ptr[ib->length_dw++] = incr; /* increment size */
1618 	ib->ptr[ib->length_dw++] = 0;
1619 	ib->ptr[ib->length_dw++] = count - 1; /* number of entries */
1620 }
1621 
1622 /**
1623  * sdma_v4_0_ring_pad_ib - pad the IB to the required number of dw
1624  *
1625  * @ring: amdgpu_ring structure holding ring information
1626  * @ib: indirect buffer to fill with padding
1627  */
sdma_v4_0_ring_pad_ib(struct amdgpu_ring * ring,struct amdgpu_ib * ib)1628 static void sdma_v4_0_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib)
1629 {
1630 	struct amdgpu_sdma_instance *sdma = amdgpu_sdma_get_instance_from_ring(ring);
1631 	u32 pad_count;
1632 	int i;
1633 
1634 	pad_count = (-ib->length_dw) & 7;
1635 	for (i = 0; i < pad_count; i++)
1636 		if (sdma && sdma->burst_nop && (i == 0))
1637 			ib->ptr[ib->length_dw++] =
1638 				SDMA_PKT_HEADER_OP(SDMA_OP_NOP) |
1639 				SDMA_PKT_NOP_HEADER_COUNT(pad_count - 1);
1640 		else
1641 			ib->ptr[ib->length_dw++] =
1642 				SDMA_PKT_HEADER_OP(SDMA_OP_NOP);
1643 }
1644 
1645 
1646 /**
1647  * sdma_v4_0_ring_emit_pipeline_sync - sync the pipeline
1648  *
1649  * @ring: amdgpu_ring pointer
1650  *
1651  * Make sure all previous operations are completed (CIK).
1652  */
sdma_v4_0_ring_emit_pipeline_sync(struct amdgpu_ring * ring)1653 static void sdma_v4_0_ring_emit_pipeline_sync(struct amdgpu_ring *ring)
1654 {
1655 	uint32_t seq = ring->fence_drv.sync_seq;
1656 	uint64_t addr = ring->fence_drv.gpu_addr;
1657 
1658 	/* wait for idle */
1659 	sdma_v4_0_wait_reg_mem(ring, 1, 0,
1660 			       addr & 0xfffffffc,
1661 			       upper_32_bits(addr) & 0xffffffff,
1662 			       seq, 0xffffffff, 4);
1663 }
1664 
1665 
1666 /**
1667  * sdma_v4_0_ring_emit_vm_flush - vm flush using sDMA
1668  *
1669  * @ring: amdgpu_ring pointer
1670  * @vmid: vmid number to use
1671  * @pd_addr: address
1672  *
1673  * Update the page table base and flush the VM TLB
1674  * using sDMA (VEGA10).
1675  */
sdma_v4_0_ring_emit_vm_flush(struct amdgpu_ring * ring,unsigned vmid,uint64_t pd_addr)1676 static void sdma_v4_0_ring_emit_vm_flush(struct amdgpu_ring *ring,
1677 					 unsigned vmid, uint64_t pd_addr)
1678 {
1679 	amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr);
1680 }
1681 
sdma_v4_0_ring_emit_wreg(struct amdgpu_ring * ring,uint32_t reg,uint32_t val)1682 static void sdma_v4_0_ring_emit_wreg(struct amdgpu_ring *ring,
1683 				     uint32_t reg, uint32_t val)
1684 {
1685 	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_SRBM_WRITE) |
1686 			  SDMA_PKT_SRBM_WRITE_HEADER_BYTE_EN(0xf));
1687 	amdgpu_ring_write(ring, reg);
1688 	amdgpu_ring_write(ring, val);
1689 }
1690 
sdma_v4_0_ring_emit_reg_wait(struct amdgpu_ring * ring,uint32_t reg,uint32_t val,uint32_t mask)1691 static void sdma_v4_0_ring_emit_reg_wait(struct amdgpu_ring *ring, uint32_t reg,
1692 					 uint32_t val, uint32_t mask)
1693 {
1694 	sdma_v4_0_wait_reg_mem(ring, 0, 0, reg, 0, val, mask, 10);
1695 }
1696 
sdma_v4_0_fw_support_paging_queue(struct amdgpu_device * adev)1697 static bool sdma_v4_0_fw_support_paging_queue(struct amdgpu_device *adev)
1698 {
1699 	uint fw_version = adev->sdma.instance[0].fw_version;
1700 
1701 	switch (adev->ip_versions[SDMA0_HWIP][0]) {
1702 	case IP_VERSION(4, 0, 0):
1703 		return fw_version >= 430;
1704 	case IP_VERSION(4, 0, 1):
1705 		/*return fw_version >= 31;*/
1706 		return false;
1707 	case IP_VERSION(4, 2, 0):
1708 		return fw_version >= 123;
1709 	default:
1710 		return false;
1711 	}
1712 }
1713 
sdma_v4_0_early_init(void * handle)1714 static int sdma_v4_0_early_init(void *handle)
1715 {
1716 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1717 	int r;
1718 
1719 	r = sdma_v4_0_init_microcode(adev);
1720 	if (r) {
1721 		DRM_ERROR("Failed to load sdma firmware!\n");
1722 		return r;
1723 	}
1724 
1725 	/* TODO: Page queue breaks driver reload under SRIOV */
1726 	if ((adev->ip_versions[SDMA0_HWIP][0] == IP_VERSION(4, 0, 0)) &&
1727 	    amdgpu_sriov_vf((adev)))
1728 		adev->sdma.has_page_queue = false;
1729 	else if (sdma_v4_0_fw_support_paging_queue(adev))
1730 		adev->sdma.has_page_queue = true;
1731 
1732 	sdma_v4_0_set_ring_funcs(adev);
1733 	sdma_v4_0_set_buffer_funcs(adev);
1734 	sdma_v4_0_set_vm_pte_funcs(adev);
1735 	sdma_v4_0_set_irq_funcs(adev);
1736 	sdma_v4_0_set_ras_funcs(adev);
1737 
1738 	return 0;
1739 }
1740 
1741 static int sdma_v4_0_process_ras_data_cb(struct amdgpu_device *adev,
1742 		void *err_data,
1743 		struct amdgpu_iv_entry *entry);
1744 
sdma_v4_0_late_init(void * handle)1745 static int sdma_v4_0_late_init(void *handle)
1746 {
1747 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1748 
1749 	sdma_v4_0_setup_ulv(adev);
1750 
1751 	if (!amdgpu_persistent_edc_harvesting_supported(adev)) {
1752 		if (adev->sdma.ras && adev->sdma.ras->ras_block.hw_ops &&
1753 		    adev->sdma.ras->ras_block.hw_ops->reset_ras_error_count)
1754 			adev->sdma.ras->ras_block.hw_ops->reset_ras_error_count(adev);
1755 	}
1756 
1757 	return 0;
1758 }
1759 
sdma_v4_0_sw_init(void * handle)1760 static int sdma_v4_0_sw_init(void *handle)
1761 {
1762 	struct amdgpu_ring *ring;
1763 	int r, i;
1764 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1765 
1766 	/* SDMA trap event */
1767 	for (i = 0; i < adev->sdma.num_instances; i++) {
1768 		r = amdgpu_irq_add_id(adev, sdma_v4_0_seq_to_irq_id(i),
1769 				      SDMA0_4_0__SRCID__SDMA_TRAP,
1770 				      &adev->sdma.trap_irq);
1771 		if (r)
1772 			return r;
1773 	}
1774 
1775 	/* SDMA SRAM ECC event */
1776 	for (i = 0; i < adev->sdma.num_instances; i++) {
1777 		r = amdgpu_irq_add_id(adev, sdma_v4_0_seq_to_irq_id(i),
1778 				      SDMA0_4_0__SRCID__SDMA_SRAM_ECC,
1779 				      &adev->sdma.ecc_irq);
1780 		if (r)
1781 			return r;
1782 	}
1783 
1784 	/* SDMA VM_HOLE/DOORBELL_INV/POLL_TIMEOUT/SRBM_WRITE_PROTECTION event*/
1785 	for (i = 0; i < adev->sdma.num_instances; i++) {
1786 		r = amdgpu_irq_add_id(adev, sdma_v4_0_seq_to_irq_id(i),
1787 				      SDMA0_4_0__SRCID__SDMA_VM_HOLE,
1788 				      &adev->sdma.vm_hole_irq);
1789 		if (r)
1790 			return r;
1791 
1792 		r = amdgpu_irq_add_id(adev, sdma_v4_0_seq_to_irq_id(i),
1793 				      SDMA0_4_0__SRCID__SDMA_DOORBELL_INVALID,
1794 				      &adev->sdma.doorbell_invalid_irq);
1795 		if (r)
1796 			return r;
1797 
1798 		r = amdgpu_irq_add_id(adev, sdma_v4_0_seq_to_irq_id(i),
1799 				      SDMA0_4_0__SRCID__SDMA_POLL_TIMEOUT,
1800 				      &adev->sdma.pool_timeout_irq);
1801 		if (r)
1802 			return r;
1803 
1804 		r = amdgpu_irq_add_id(adev, sdma_v4_0_seq_to_irq_id(i),
1805 				      SDMA0_4_0__SRCID__SDMA_SRBMWRITE,
1806 				      &adev->sdma.srbm_write_irq);
1807 		if (r)
1808 			return r;
1809 	}
1810 
1811 	for (i = 0; i < adev->sdma.num_instances; i++) {
1812 		ring = &adev->sdma.instance[i].ring;
1813 		ring->ring_obj = NULL;
1814 		ring->use_doorbell = true;
1815 
1816 		DRM_DEBUG("SDMA %d use_doorbell being set to: [%s]\n", i,
1817 				ring->use_doorbell?"true":"false");
1818 
1819 		/* doorbell size is 2 dwords, get DWORD offset */
1820 		ring->doorbell_index = adev->doorbell_index.sdma_engine[i] << 1;
1821 
1822 		/*
1823 		 * On Arcturus, SDMA instance 5~7 has a different vmhub
1824 		 * type(AMDGPU_MMHUB1).
1825 		 */
1826 		if (adev->ip_versions[SDMA0_HWIP][0] == IP_VERSION(4, 2, 2) && i >= 5)
1827 			ring->vm_hub = AMDGPU_MMHUB1(0);
1828 		else
1829 			ring->vm_hub = AMDGPU_MMHUB0(0);
1830 
1831 		sprintf(ring->name, "sdma%d", i);
1832 		r = amdgpu_ring_init(adev, ring, 1024, &adev->sdma.trap_irq,
1833 				     AMDGPU_SDMA_IRQ_INSTANCE0 + i,
1834 				     AMDGPU_RING_PRIO_DEFAULT, NULL);
1835 		if (r)
1836 			return r;
1837 
1838 		if (adev->sdma.has_page_queue) {
1839 			ring = &adev->sdma.instance[i].page;
1840 			ring->ring_obj = NULL;
1841 			ring->use_doorbell = true;
1842 
1843 			/* paging queue use same doorbell index/routing as gfx queue
1844 			 * with 0x400 (4096 dwords) offset on second doorbell page
1845 			 */
1846 			if (adev->ip_versions[SDMA0_HWIP][0] >= IP_VERSION(4, 0, 0) &&
1847 			    adev->ip_versions[SDMA0_HWIP][0] < IP_VERSION(4, 2, 0)) {
1848 				ring->doorbell_index =
1849 					adev->doorbell_index.sdma_engine[i] << 1;
1850 				ring->doorbell_index += 0x400;
1851 			} else {
1852 				/* From vega20, the sdma_doorbell_range in 1st
1853 				 * doorbell page is reserved for page queue.
1854 				 */
1855 				ring->doorbell_index =
1856 					(adev->doorbell_index.sdma_engine[i] + 1) << 1;
1857 			}
1858 
1859 			if (adev->ip_versions[SDMA0_HWIP][0] == IP_VERSION(4, 2, 2) && i >= 5)
1860 				ring->vm_hub = AMDGPU_MMHUB1(0);
1861 			else
1862 				ring->vm_hub = AMDGPU_MMHUB0(0);
1863 
1864 			sprintf(ring->name, "page%d", i);
1865 			r = amdgpu_ring_init(adev, ring, 1024,
1866 					     &adev->sdma.trap_irq,
1867 					     AMDGPU_SDMA_IRQ_INSTANCE0 + i,
1868 					     AMDGPU_RING_PRIO_DEFAULT, NULL);
1869 			if (r)
1870 				return r;
1871 		}
1872 	}
1873 
1874 	if (amdgpu_sdma_ras_sw_init(adev)) {
1875 		dev_err(adev->dev, "Failed to initialize sdma ras block!\n");
1876 		return -EINVAL;
1877 	}
1878 
1879 	return r;
1880 }
1881 
sdma_v4_0_sw_fini(void * handle)1882 static int sdma_v4_0_sw_fini(void *handle)
1883 {
1884 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1885 	int i;
1886 
1887 	for (i = 0; i < adev->sdma.num_instances; i++) {
1888 		amdgpu_ring_fini(&adev->sdma.instance[i].ring);
1889 		if (adev->sdma.has_page_queue)
1890 			amdgpu_ring_fini(&adev->sdma.instance[i].page);
1891 	}
1892 
1893 	if (adev->ip_versions[SDMA0_HWIP][0] == IP_VERSION(4, 2, 2) ||
1894             adev->ip_versions[SDMA0_HWIP][0] == IP_VERSION(4, 4, 0))
1895 		amdgpu_sdma_destroy_inst_ctx(adev, true);
1896 	else
1897 		amdgpu_sdma_destroy_inst_ctx(adev, false);
1898 
1899 	return 0;
1900 }
1901 
sdma_v4_0_hw_init(void * handle)1902 static int sdma_v4_0_hw_init(void *handle)
1903 {
1904 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1905 
1906 	if (adev->flags & AMD_IS_APU)
1907 		amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_SDMA, false);
1908 
1909 	if (!amdgpu_sriov_vf(adev))
1910 		sdma_v4_0_init_golden_registers(adev);
1911 
1912 	return sdma_v4_0_start(adev);
1913 }
1914 
sdma_v4_0_hw_fini(void * handle)1915 static int sdma_v4_0_hw_fini(void *handle)
1916 {
1917 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1918 	int i;
1919 
1920 	if (amdgpu_sriov_vf(adev)) {
1921 		/* disable the scheduler for SDMA */
1922 		amdgpu_sdma_unset_buffer_funcs_helper(adev);
1923 		return 0;
1924 	}
1925 
1926 	if (amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__SDMA)) {
1927 		for (i = 0; i < adev->sdma.num_instances; i++) {
1928 			amdgpu_irq_put(adev, &adev->sdma.ecc_irq,
1929 				       AMDGPU_SDMA_IRQ_INSTANCE0 + i);
1930 		}
1931 	}
1932 
1933 	sdma_v4_0_ctx_switch_enable(adev, false);
1934 	sdma_v4_0_enable(adev, false);
1935 
1936 	if (adev->flags & AMD_IS_APU)
1937 		amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_SDMA, true);
1938 
1939 	return 0;
1940 }
1941 
sdma_v4_0_suspend(void * handle)1942 static int sdma_v4_0_suspend(void *handle)
1943 {
1944 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1945 
1946 	/* SMU saves SDMA state for us */
1947 	if (adev->in_s0ix) {
1948 		sdma_v4_0_gfx_enable(adev, false);
1949 		return 0;
1950 	}
1951 
1952 	return sdma_v4_0_hw_fini(adev);
1953 }
1954 
sdma_v4_0_resume(void * handle)1955 static int sdma_v4_0_resume(void *handle)
1956 {
1957 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1958 
1959 	/* SMU restores SDMA state for us */
1960 	if (adev->in_s0ix) {
1961 		sdma_v4_0_enable(adev, true);
1962 		sdma_v4_0_gfx_enable(adev, true);
1963 		amdgpu_ttm_set_buffer_funcs_status(adev, true);
1964 		return 0;
1965 	}
1966 
1967 	return sdma_v4_0_hw_init(adev);
1968 }
1969 
sdma_v4_0_is_idle(void * handle)1970 static bool sdma_v4_0_is_idle(void *handle)
1971 {
1972 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1973 	u32 i;
1974 
1975 	for (i = 0; i < adev->sdma.num_instances; i++) {
1976 		u32 tmp = RREG32_SDMA(i, mmSDMA0_STATUS_REG);
1977 
1978 		if (!(tmp & SDMA0_STATUS_REG__IDLE_MASK))
1979 			return false;
1980 	}
1981 
1982 	return true;
1983 }
1984 
sdma_v4_0_wait_for_idle(void * handle)1985 static int sdma_v4_0_wait_for_idle(void *handle)
1986 {
1987 	unsigned i, j;
1988 	u32 sdma[AMDGPU_MAX_SDMA_INSTANCES];
1989 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1990 
1991 	for (i = 0; i < adev->usec_timeout; i++) {
1992 		for (j = 0; j < adev->sdma.num_instances; j++) {
1993 			sdma[j] = RREG32_SDMA(j, mmSDMA0_STATUS_REG);
1994 			if (!(sdma[j] & SDMA0_STATUS_REG__IDLE_MASK))
1995 				break;
1996 		}
1997 		if (j == adev->sdma.num_instances)
1998 			return 0;
1999 		udelay(1);
2000 	}
2001 	return -ETIMEDOUT;
2002 }
2003 
sdma_v4_0_soft_reset(void * handle)2004 static int sdma_v4_0_soft_reset(void *handle)
2005 {
2006 	/* todo */
2007 
2008 	return 0;
2009 }
2010 
sdma_v4_0_set_trap_irq_state(struct amdgpu_device * adev,struct amdgpu_irq_src * source,unsigned type,enum amdgpu_interrupt_state state)2011 static int sdma_v4_0_set_trap_irq_state(struct amdgpu_device *adev,
2012 					struct amdgpu_irq_src *source,
2013 					unsigned type,
2014 					enum amdgpu_interrupt_state state)
2015 {
2016 	u32 sdma_cntl;
2017 
2018 	sdma_cntl = RREG32_SDMA(type, mmSDMA0_CNTL);
2019 	sdma_cntl = REG_SET_FIELD(sdma_cntl, SDMA0_CNTL, TRAP_ENABLE,
2020 		       state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0);
2021 	WREG32_SDMA(type, mmSDMA0_CNTL, sdma_cntl);
2022 
2023 	return 0;
2024 }
2025 
sdma_v4_0_process_trap_irq(struct amdgpu_device * adev,struct amdgpu_irq_src * source,struct amdgpu_iv_entry * entry)2026 static int sdma_v4_0_process_trap_irq(struct amdgpu_device *adev,
2027 				      struct amdgpu_irq_src *source,
2028 				      struct amdgpu_iv_entry *entry)
2029 {
2030 	uint32_t instance;
2031 
2032 	DRM_DEBUG("IH: SDMA trap\n");
2033 	instance = sdma_v4_0_irq_id_to_seq(entry->client_id);
2034 	switch (entry->ring_id) {
2035 	case 0:
2036 		amdgpu_fence_process(&adev->sdma.instance[instance].ring);
2037 		break;
2038 	case 1:
2039 		if (adev->ip_versions[SDMA0_HWIP][0] == IP_VERSION(4, 2, 0))
2040 			amdgpu_fence_process(&adev->sdma.instance[instance].page);
2041 		break;
2042 	case 2:
2043 		/* XXX compute */
2044 		break;
2045 	case 3:
2046 		if (adev->ip_versions[SDMA0_HWIP][0] != IP_VERSION(4, 2, 0))
2047 			amdgpu_fence_process(&adev->sdma.instance[instance].page);
2048 		break;
2049 	}
2050 	return 0;
2051 }
2052 
sdma_v4_0_process_ras_data_cb(struct amdgpu_device * adev,void * err_data,struct amdgpu_iv_entry * entry)2053 static int sdma_v4_0_process_ras_data_cb(struct amdgpu_device *adev,
2054 		void *err_data,
2055 		struct amdgpu_iv_entry *entry)
2056 {
2057 	int instance;
2058 
2059 	/* When “Full RAS” is enabled, the per-IP interrupt sources should
2060 	 * be disabled and the driver should only look for the aggregated
2061 	 * interrupt via sync flood
2062 	 */
2063 	if (amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__GFX))
2064 		goto out;
2065 
2066 	instance = sdma_v4_0_irq_id_to_seq(entry->client_id);
2067 	if (instance < 0)
2068 		goto out;
2069 
2070 	amdgpu_sdma_process_ras_data_cb(adev, err_data, entry);
2071 
2072 out:
2073 	return AMDGPU_RAS_SUCCESS;
2074 }
2075 
sdma_v4_0_process_illegal_inst_irq(struct amdgpu_device * adev,struct amdgpu_irq_src * source,struct amdgpu_iv_entry * entry)2076 static int sdma_v4_0_process_illegal_inst_irq(struct amdgpu_device *adev,
2077 					      struct amdgpu_irq_src *source,
2078 					      struct amdgpu_iv_entry *entry)
2079 {
2080 	int instance;
2081 
2082 	DRM_ERROR("Illegal instruction in SDMA command stream\n");
2083 
2084 	instance = sdma_v4_0_irq_id_to_seq(entry->client_id);
2085 	if (instance < 0)
2086 		return 0;
2087 
2088 	switch (entry->ring_id) {
2089 	case 0:
2090 		drm_sched_fault(&adev->sdma.instance[instance].ring.sched);
2091 		break;
2092 	}
2093 	return 0;
2094 }
2095 
sdma_v4_0_set_ecc_irq_state(struct amdgpu_device * adev,struct amdgpu_irq_src * source,unsigned type,enum amdgpu_interrupt_state state)2096 static int sdma_v4_0_set_ecc_irq_state(struct amdgpu_device *adev,
2097 					struct amdgpu_irq_src *source,
2098 					unsigned type,
2099 					enum amdgpu_interrupt_state state)
2100 {
2101 	u32 sdma_edc_config;
2102 
2103 	sdma_edc_config = RREG32_SDMA(type, mmSDMA0_EDC_CONFIG);
2104 	sdma_edc_config = REG_SET_FIELD(sdma_edc_config, SDMA0_EDC_CONFIG, ECC_INT_ENABLE,
2105 		       state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0);
2106 	WREG32_SDMA(type, mmSDMA0_EDC_CONFIG, sdma_edc_config);
2107 
2108 	return 0;
2109 }
2110 
sdma_v4_0_print_iv_entry(struct amdgpu_device * adev,struct amdgpu_iv_entry * entry)2111 static int sdma_v4_0_print_iv_entry(struct amdgpu_device *adev,
2112 					      struct amdgpu_iv_entry *entry)
2113 {
2114 	int instance;
2115 	struct amdgpu_task_info task_info;
2116 	u64 addr;
2117 
2118 	instance = sdma_v4_0_irq_id_to_seq(entry->client_id);
2119 	if (instance < 0 || instance >= adev->sdma.num_instances) {
2120 		dev_err(adev->dev, "sdma instance invalid %d\n", instance);
2121 		return -EINVAL;
2122 	}
2123 
2124 	addr = (u64)entry->src_data[0] << 12;
2125 	addr |= ((u64)entry->src_data[1] & 0xf) << 44;
2126 
2127 	memset(&task_info, 0, sizeof(struct amdgpu_task_info));
2128 	amdgpu_vm_get_task_info(adev, entry->pasid, &task_info);
2129 
2130 	dev_dbg_ratelimited(adev->dev,
2131 		   "[sdma%d] address:0x%016llx src_id:%u ring:%u vmid:%u "
2132 		   "pasid:%u, for process %s pid %d thread %s pid %d\n",
2133 		   instance, addr, entry->src_id, entry->ring_id, entry->vmid,
2134 		   entry->pasid, task_info.process_name, task_info.tgid,
2135 		   task_info.task_name, task_info.pid);
2136 	return 0;
2137 }
2138 
sdma_v4_0_process_vm_hole_irq(struct amdgpu_device * adev,struct amdgpu_irq_src * source,struct amdgpu_iv_entry * entry)2139 static int sdma_v4_0_process_vm_hole_irq(struct amdgpu_device *adev,
2140 					      struct amdgpu_irq_src *source,
2141 					      struct amdgpu_iv_entry *entry)
2142 {
2143 	dev_dbg_ratelimited(adev->dev, "MC or SEM address in VM hole\n");
2144 	sdma_v4_0_print_iv_entry(adev, entry);
2145 	return 0;
2146 }
2147 
sdma_v4_0_process_doorbell_invalid_irq(struct amdgpu_device * adev,struct amdgpu_irq_src * source,struct amdgpu_iv_entry * entry)2148 static int sdma_v4_0_process_doorbell_invalid_irq(struct amdgpu_device *adev,
2149 					      struct amdgpu_irq_src *source,
2150 					      struct amdgpu_iv_entry *entry)
2151 {
2152 	dev_dbg_ratelimited(adev->dev, "SDMA received a doorbell from BIF with byte_enable !=0xff\n");
2153 	sdma_v4_0_print_iv_entry(adev, entry);
2154 	return 0;
2155 }
2156 
sdma_v4_0_process_pool_timeout_irq(struct amdgpu_device * adev,struct amdgpu_irq_src * source,struct amdgpu_iv_entry * entry)2157 static int sdma_v4_0_process_pool_timeout_irq(struct amdgpu_device *adev,
2158 					      struct amdgpu_irq_src *source,
2159 					      struct amdgpu_iv_entry *entry)
2160 {
2161 	dev_dbg_ratelimited(adev->dev,
2162 		"Polling register/memory timeout executing POLL_REG/MEM with finite timer\n");
2163 	sdma_v4_0_print_iv_entry(adev, entry);
2164 	return 0;
2165 }
2166 
sdma_v4_0_process_srbm_write_irq(struct amdgpu_device * adev,struct amdgpu_irq_src * source,struct amdgpu_iv_entry * entry)2167 static int sdma_v4_0_process_srbm_write_irq(struct amdgpu_device *adev,
2168 					      struct amdgpu_irq_src *source,
2169 					      struct amdgpu_iv_entry *entry)
2170 {
2171 	dev_dbg_ratelimited(adev->dev,
2172 		"SDMA gets an Register Write SRBM_WRITE command in non-privilege command buffer\n");
2173 	sdma_v4_0_print_iv_entry(adev, entry);
2174 	return 0;
2175 }
2176 
sdma_v4_0_update_medium_grain_clock_gating(struct amdgpu_device * adev,bool enable)2177 static void sdma_v4_0_update_medium_grain_clock_gating(
2178 		struct amdgpu_device *adev,
2179 		bool enable)
2180 {
2181 	uint32_t data, def;
2182 	int i;
2183 
2184 	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_SDMA_MGCG)) {
2185 		for (i = 0; i < adev->sdma.num_instances; i++) {
2186 			def = data = RREG32_SDMA(i, mmSDMA0_CLK_CTRL);
2187 			data &= ~(SDMA0_CLK_CTRL__SOFT_OVERRIDE7_MASK |
2188 				  SDMA0_CLK_CTRL__SOFT_OVERRIDE6_MASK |
2189 				  SDMA0_CLK_CTRL__SOFT_OVERRIDE5_MASK |
2190 				  SDMA0_CLK_CTRL__SOFT_OVERRIDE4_MASK |
2191 				  SDMA0_CLK_CTRL__SOFT_OVERRIDE3_MASK |
2192 				  SDMA0_CLK_CTRL__SOFT_OVERRIDE2_MASK |
2193 				  SDMA0_CLK_CTRL__SOFT_OVERRIDE1_MASK |
2194 				  SDMA0_CLK_CTRL__SOFT_OVERRIDE0_MASK);
2195 			if (def != data)
2196 				WREG32_SDMA(i, mmSDMA0_CLK_CTRL, data);
2197 		}
2198 	} else {
2199 		for (i = 0; i < adev->sdma.num_instances; i++) {
2200 			def = data = RREG32_SDMA(i, mmSDMA0_CLK_CTRL);
2201 			data |= (SDMA0_CLK_CTRL__SOFT_OVERRIDE7_MASK |
2202 				 SDMA0_CLK_CTRL__SOFT_OVERRIDE6_MASK |
2203 				 SDMA0_CLK_CTRL__SOFT_OVERRIDE5_MASK |
2204 				 SDMA0_CLK_CTRL__SOFT_OVERRIDE4_MASK |
2205 				 SDMA0_CLK_CTRL__SOFT_OVERRIDE3_MASK |
2206 				 SDMA0_CLK_CTRL__SOFT_OVERRIDE2_MASK |
2207 				 SDMA0_CLK_CTRL__SOFT_OVERRIDE1_MASK |
2208 				 SDMA0_CLK_CTRL__SOFT_OVERRIDE0_MASK);
2209 			if (def != data)
2210 				WREG32_SDMA(i, mmSDMA0_CLK_CTRL, data);
2211 		}
2212 	}
2213 }
2214 
2215 
sdma_v4_0_update_medium_grain_light_sleep(struct amdgpu_device * adev,bool enable)2216 static void sdma_v4_0_update_medium_grain_light_sleep(
2217 		struct amdgpu_device *adev,
2218 		bool enable)
2219 {
2220 	uint32_t data, def;
2221 	int i;
2222 
2223 	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_SDMA_LS)) {
2224 		for (i = 0; i < adev->sdma.num_instances; i++) {
2225 			/* 1-not override: enable sdma mem light sleep */
2226 			def = data = RREG32_SDMA(0, mmSDMA0_POWER_CNTL);
2227 			data |= SDMA0_POWER_CNTL__MEM_POWER_OVERRIDE_MASK;
2228 			if (def != data)
2229 				WREG32_SDMA(0, mmSDMA0_POWER_CNTL, data);
2230 		}
2231 	} else {
2232 		for (i = 0; i < adev->sdma.num_instances; i++) {
2233 		/* 0-override:disable sdma mem light sleep */
2234 			def = data = RREG32_SDMA(0, mmSDMA0_POWER_CNTL);
2235 			data &= ~SDMA0_POWER_CNTL__MEM_POWER_OVERRIDE_MASK;
2236 			if (def != data)
2237 				WREG32_SDMA(0, mmSDMA0_POWER_CNTL, data);
2238 		}
2239 	}
2240 }
2241 
sdma_v4_0_set_clockgating_state(void * handle,enum amd_clockgating_state state)2242 static int sdma_v4_0_set_clockgating_state(void *handle,
2243 					  enum amd_clockgating_state state)
2244 {
2245 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2246 
2247 	if (amdgpu_sriov_vf(adev))
2248 		return 0;
2249 
2250 	sdma_v4_0_update_medium_grain_clock_gating(adev,
2251 			state == AMD_CG_STATE_GATE);
2252 	sdma_v4_0_update_medium_grain_light_sleep(adev,
2253 			state == AMD_CG_STATE_GATE);
2254 	return 0;
2255 }
2256 
sdma_v4_0_set_powergating_state(void * handle,enum amd_powergating_state state)2257 static int sdma_v4_0_set_powergating_state(void *handle,
2258 					  enum amd_powergating_state state)
2259 {
2260 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2261 
2262 	switch (adev->ip_versions[SDMA0_HWIP][0]) {
2263 	case IP_VERSION(4, 1, 0):
2264 	case IP_VERSION(4, 1, 1):
2265 	case IP_VERSION(4, 1, 2):
2266 		sdma_v4_1_update_power_gating(adev,
2267 				state == AMD_PG_STATE_GATE);
2268 		break;
2269 	default:
2270 		break;
2271 	}
2272 
2273 	return 0;
2274 }
2275 
sdma_v4_0_get_clockgating_state(void * handle,u64 * flags)2276 static void sdma_v4_0_get_clockgating_state(void *handle, u64 *flags)
2277 {
2278 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2279 	int data;
2280 
2281 	if (amdgpu_sriov_vf(adev))
2282 		*flags = 0;
2283 
2284 	/* AMD_CG_SUPPORT_SDMA_MGCG */
2285 	data = RREG32(SOC15_REG_OFFSET(SDMA0, 0, mmSDMA0_CLK_CTRL));
2286 	if (!(data & SDMA0_CLK_CTRL__SOFT_OVERRIDE7_MASK))
2287 		*flags |= AMD_CG_SUPPORT_SDMA_MGCG;
2288 
2289 	/* AMD_CG_SUPPORT_SDMA_LS */
2290 	data = RREG32(SOC15_REG_OFFSET(SDMA0, 0, mmSDMA0_POWER_CNTL));
2291 	if (data & SDMA0_POWER_CNTL__MEM_POWER_OVERRIDE_MASK)
2292 		*flags |= AMD_CG_SUPPORT_SDMA_LS;
2293 }
2294 
2295 const struct amd_ip_funcs sdma_v4_0_ip_funcs = {
2296 	.name = "sdma_v4_0",
2297 	.early_init = sdma_v4_0_early_init,
2298 	.late_init = sdma_v4_0_late_init,
2299 	.sw_init = sdma_v4_0_sw_init,
2300 	.sw_fini = sdma_v4_0_sw_fini,
2301 	.hw_init = sdma_v4_0_hw_init,
2302 	.hw_fini = sdma_v4_0_hw_fini,
2303 	.suspend = sdma_v4_0_suspend,
2304 	.resume = sdma_v4_0_resume,
2305 	.is_idle = sdma_v4_0_is_idle,
2306 	.wait_for_idle = sdma_v4_0_wait_for_idle,
2307 	.soft_reset = sdma_v4_0_soft_reset,
2308 	.set_clockgating_state = sdma_v4_0_set_clockgating_state,
2309 	.set_powergating_state = sdma_v4_0_set_powergating_state,
2310 	.get_clockgating_state = sdma_v4_0_get_clockgating_state,
2311 };
2312 
2313 static const struct amdgpu_ring_funcs sdma_v4_0_ring_funcs = {
2314 	.type = AMDGPU_RING_TYPE_SDMA,
2315 	.align_mask = 0xff,
2316 	.nop = SDMA_PKT_NOP_HEADER_OP(SDMA_OP_NOP),
2317 	.support_64bit_ptrs = true,
2318 	.secure_submission_supported = true,
2319 	.get_rptr = sdma_v4_0_ring_get_rptr,
2320 	.get_wptr = sdma_v4_0_ring_get_wptr,
2321 	.set_wptr = sdma_v4_0_ring_set_wptr,
2322 	.emit_frame_size =
2323 		6 + /* sdma_v4_0_ring_emit_hdp_flush */
2324 		3 + /* hdp invalidate */
2325 		6 + /* sdma_v4_0_ring_emit_pipeline_sync */
2326 		/* sdma_v4_0_ring_emit_vm_flush */
2327 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
2328 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 6 +
2329 		10 + 10 + 10, /* sdma_v4_0_ring_emit_fence x3 for user fence, vm fence */
2330 	.emit_ib_size = 7 + 6, /* sdma_v4_0_ring_emit_ib */
2331 	.emit_ib = sdma_v4_0_ring_emit_ib,
2332 	.emit_fence = sdma_v4_0_ring_emit_fence,
2333 	.emit_pipeline_sync = sdma_v4_0_ring_emit_pipeline_sync,
2334 	.emit_vm_flush = sdma_v4_0_ring_emit_vm_flush,
2335 	.emit_hdp_flush = sdma_v4_0_ring_emit_hdp_flush,
2336 	.test_ring = sdma_v4_0_ring_test_ring,
2337 	.test_ib = sdma_v4_0_ring_test_ib,
2338 	.insert_nop = sdma_v4_0_ring_insert_nop,
2339 	.pad_ib = sdma_v4_0_ring_pad_ib,
2340 	.emit_wreg = sdma_v4_0_ring_emit_wreg,
2341 	.emit_reg_wait = sdma_v4_0_ring_emit_reg_wait,
2342 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
2343 };
2344 
2345 static const struct amdgpu_ring_funcs sdma_v4_0_page_ring_funcs = {
2346 	.type = AMDGPU_RING_TYPE_SDMA,
2347 	.align_mask = 0xff,
2348 	.nop = SDMA_PKT_NOP_HEADER_OP(SDMA_OP_NOP),
2349 	.support_64bit_ptrs = true,
2350 	.secure_submission_supported = true,
2351 	.get_rptr = sdma_v4_0_ring_get_rptr,
2352 	.get_wptr = sdma_v4_0_page_ring_get_wptr,
2353 	.set_wptr = sdma_v4_0_page_ring_set_wptr,
2354 	.emit_frame_size =
2355 		6 + /* sdma_v4_0_ring_emit_hdp_flush */
2356 		3 + /* hdp invalidate */
2357 		6 + /* sdma_v4_0_ring_emit_pipeline_sync */
2358 		/* sdma_v4_0_ring_emit_vm_flush */
2359 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
2360 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 6 +
2361 		10 + 10 + 10, /* sdma_v4_0_ring_emit_fence x3 for user fence, vm fence */
2362 	.emit_ib_size = 7 + 6, /* sdma_v4_0_ring_emit_ib */
2363 	.emit_ib = sdma_v4_0_ring_emit_ib,
2364 	.emit_fence = sdma_v4_0_ring_emit_fence,
2365 	.emit_pipeline_sync = sdma_v4_0_ring_emit_pipeline_sync,
2366 	.emit_vm_flush = sdma_v4_0_ring_emit_vm_flush,
2367 	.emit_hdp_flush = sdma_v4_0_ring_emit_hdp_flush,
2368 	.test_ring = sdma_v4_0_ring_test_ring,
2369 	.test_ib = sdma_v4_0_ring_test_ib,
2370 	.insert_nop = sdma_v4_0_ring_insert_nop,
2371 	.pad_ib = sdma_v4_0_ring_pad_ib,
2372 	.emit_wreg = sdma_v4_0_ring_emit_wreg,
2373 	.emit_reg_wait = sdma_v4_0_ring_emit_reg_wait,
2374 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
2375 };
2376 
sdma_v4_0_set_ring_funcs(struct amdgpu_device * adev)2377 static void sdma_v4_0_set_ring_funcs(struct amdgpu_device *adev)
2378 {
2379 	int i;
2380 
2381 	for (i = 0; i < adev->sdma.num_instances; i++) {
2382 		adev->sdma.instance[i].ring.funcs = &sdma_v4_0_ring_funcs;
2383 		adev->sdma.instance[i].ring.me = i;
2384 		if (adev->sdma.has_page_queue) {
2385 			adev->sdma.instance[i].page.funcs =
2386 					&sdma_v4_0_page_ring_funcs;
2387 			adev->sdma.instance[i].page.me = i;
2388 		}
2389 	}
2390 }
2391 
2392 static const struct amdgpu_irq_src_funcs sdma_v4_0_trap_irq_funcs = {
2393 	.set = sdma_v4_0_set_trap_irq_state,
2394 	.process = sdma_v4_0_process_trap_irq,
2395 };
2396 
2397 static const struct amdgpu_irq_src_funcs sdma_v4_0_illegal_inst_irq_funcs = {
2398 	.process = sdma_v4_0_process_illegal_inst_irq,
2399 };
2400 
2401 static const struct amdgpu_irq_src_funcs sdma_v4_0_ecc_irq_funcs = {
2402 	.set = sdma_v4_0_set_ecc_irq_state,
2403 	.process = amdgpu_sdma_process_ecc_irq,
2404 };
2405 
2406 static const struct amdgpu_irq_src_funcs sdma_v4_0_vm_hole_irq_funcs = {
2407 	.process = sdma_v4_0_process_vm_hole_irq,
2408 };
2409 
2410 static const struct amdgpu_irq_src_funcs sdma_v4_0_doorbell_invalid_irq_funcs = {
2411 	.process = sdma_v4_0_process_doorbell_invalid_irq,
2412 };
2413 
2414 static const struct amdgpu_irq_src_funcs sdma_v4_0_pool_timeout_irq_funcs = {
2415 	.process = sdma_v4_0_process_pool_timeout_irq,
2416 };
2417 
2418 static const struct amdgpu_irq_src_funcs sdma_v4_0_srbm_write_irq_funcs = {
2419 	.process = sdma_v4_0_process_srbm_write_irq,
2420 };
2421 
sdma_v4_0_set_irq_funcs(struct amdgpu_device * adev)2422 static void sdma_v4_0_set_irq_funcs(struct amdgpu_device *adev)
2423 {
2424 	adev->sdma.trap_irq.num_types = adev->sdma.num_instances;
2425 	adev->sdma.ecc_irq.num_types = adev->sdma.num_instances;
2426 	/*For Arcturus and Aldebaran, add another 4 irq handler*/
2427 	switch (adev->sdma.num_instances) {
2428 	case 5:
2429 	case 8:
2430 		adev->sdma.vm_hole_irq.num_types = adev->sdma.num_instances;
2431 		adev->sdma.doorbell_invalid_irq.num_types = adev->sdma.num_instances;
2432 		adev->sdma.pool_timeout_irq.num_types = adev->sdma.num_instances;
2433 		adev->sdma.srbm_write_irq.num_types = adev->sdma.num_instances;
2434 		break;
2435 	default:
2436 		break;
2437 	}
2438 	adev->sdma.trap_irq.funcs = &sdma_v4_0_trap_irq_funcs;
2439 	adev->sdma.illegal_inst_irq.funcs = &sdma_v4_0_illegal_inst_irq_funcs;
2440 	adev->sdma.ecc_irq.funcs = &sdma_v4_0_ecc_irq_funcs;
2441 	adev->sdma.vm_hole_irq.funcs = &sdma_v4_0_vm_hole_irq_funcs;
2442 	adev->sdma.doorbell_invalid_irq.funcs = &sdma_v4_0_doorbell_invalid_irq_funcs;
2443 	adev->sdma.pool_timeout_irq.funcs = &sdma_v4_0_pool_timeout_irq_funcs;
2444 	adev->sdma.srbm_write_irq.funcs = &sdma_v4_0_srbm_write_irq_funcs;
2445 }
2446 
2447 /**
2448  * sdma_v4_0_emit_copy_buffer - copy buffer using the sDMA engine
2449  *
2450  * @ib: indirect buffer to copy to
2451  * @src_offset: src GPU address
2452  * @dst_offset: dst GPU address
2453  * @byte_count: number of bytes to xfer
2454  * @tmz: if a secure copy should be used
2455  *
2456  * Copy GPU buffers using the DMA engine (VEGA10/12).
2457  * Used by the amdgpu ttm implementation to move pages if
2458  * registered as the asic copy callback.
2459  */
sdma_v4_0_emit_copy_buffer(struct amdgpu_ib * ib,uint64_t src_offset,uint64_t dst_offset,uint32_t byte_count,bool tmz)2460 static void sdma_v4_0_emit_copy_buffer(struct amdgpu_ib *ib,
2461 				       uint64_t src_offset,
2462 				       uint64_t dst_offset,
2463 				       uint32_t byte_count,
2464 				       bool tmz)
2465 {
2466 	ib->ptr[ib->length_dw++] = SDMA_PKT_HEADER_OP(SDMA_OP_COPY) |
2467 		SDMA_PKT_HEADER_SUB_OP(SDMA_SUBOP_COPY_LINEAR) |
2468 		SDMA_PKT_COPY_LINEAR_HEADER_TMZ(tmz ? 1 : 0);
2469 	ib->ptr[ib->length_dw++] = byte_count - 1;
2470 	ib->ptr[ib->length_dw++] = 0; /* src/dst endian swap */
2471 	ib->ptr[ib->length_dw++] = lower_32_bits(src_offset);
2472 	ib->ptr[ib->length_dw++] = upper_32_bits(src_offset);
2473 	ib->ptr[ib->length_dw++] = lower_32_bits(dst_offset);
2474 	ib->ptr[ib->length_dw++] = upper_32_bits(dst_offset);
2475 }
2476 
2477 /**
2478  * sdma_v4_0_emit_fill_buffer - fill buffer using the sDMA engine
2479  *
2480  * @ib: indirect buffer to copy to
2481  * @src_data: value to write to buffer
2482  * @dst_offset: dst GPU address
2483  * @byte_count: number of bytes to xfer
2484  *
2485  * Fill GPU buffers using the DMA engine (VEGA10/12).
2486  */
sdma_v4_0_emit_fill_buffer(struct amdgpu_ib * ib,uint32_t src_data,uint64_t dst_offset,uint32_t byte_count)2487 static void sdma_v4_0_emit_fill_buffer(struct amdgpu_ib *ib,
2488 				       uint32_t src_data,
2489 				       uint64_t dst_offset,
2490 				       uint32_t byte_count)
2491 {
2492 	ib->ptr[ib->length_dw++] = SDMA_PKT_HEADER_OP(SDMA_OP_CONST_FILL);
2493 	ib->ptr[ib->length_dw++] = lower_32_bits(dst_offset);
2494 	ib->ptr[ib->length_dw++] = upper_32_bits(dst_offset);
2495 	ib->ptr[ib->length_dw++] = src_data;
2496 	ib->ptr[ib->length_dw++] = byte_count - 1;
2497 }
2498 
2499 static const struct amdgpu_buffer_funcs sdma_v4_0_buffer_funcs = {
2500 	.copy_max_bytes = 0x400000,
2501 	.copy_num_dw = 7,
2502 	.emit_copy_buffer = sdma_v4_0_emit_copy_buffer,
2503 
2504 	.fill_max_bytes = 0x400000,
2505 	.fill_num_dw = 5,
2506 	.emit_fill_buffer = sdma_v4_0_emit_fill_buffer,
2507 };
2508 
sdma_v4_0_set_buffer_funcs(struct amdgpu_device * adev)2509 static void sdma_v4_0_set_buffer_funcs(struct amdgpu_device *adev)
2510 {
2511 	adev->mman.buffer_funcs = &sdma_v4_0_buffer_funcs;
2512 	if (adev->sdma.has_page_queue)
2513 		adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].page;
2514 	else
2515 		adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].ring;
2516 }
2517 
2518 static const struct amdgpu_vm_pte_funcs sdma_v4_0_vm_pte_funcs = {
2519 	.copy_pte_num_dw = 7,
2520 	.copy_pte = sdma_v4_0_vm_copy_pte,
2521 
2522 	.write_pte = sdma_v4_0_vm_write_pte,
2523 	.set_pte_pde = sdma_v4_0_vm_set_pte_pde,
2524 };
2525 
sdma_v4_0_set_vm_pte_funcs(struct amdgpu_device * adev)2526 static void sdma_v4_0_set_vm_pte_funcs(struct amdgpu_device *adev)
2527 {
2528 	struct drm_gpu_scheduler *sched;
2529 	unsigned i;
2530 
2531 	adev->vm_manager.vm_pte_funcs = &sdma_v4_0_vm_pte_funcs;
2532 	for (i = 0; i < adev->sdma.num_instances; i++) {
2533 		if (adev->sdma.has_page_queue)
2534 			sched = &adev->sdma.instance[i].page.sched;
2535 		else
2536 			sched = &adev->sdma.instance[i].ring.sched;
2537 		adev->vm_manager.vm_pte_scheds[i] = sched;
2538 	}
2539 	adev->vm_manager.vm_pte_num_scheds = adev->sdma.num_instances;
2540 }
2541 
sdma_v4_0_get_ras_error_count(uint32_t value,uint32_t instance,uint32_t * sec_count)2542 static void sdma_v4_0_get_ras_error_count(uint32_t value,
2543 					uint32_t instance,
2544 					uint32_t *sec_count)
2545 {
2546 	uint32_t i;
2547 	uint32_t sec_cnt;
2548 
2549 	/* double bits error (multiple bits) error detection is not supported */
2550 	for (i = 0; i < ARRAY_SIZE(sdma_v4_0_ras_fields); i++) {
2551 		/* the SDMA_EDC_COUNTER register in each sdma instance
2552 		 * shares the same sed shift_mask
2553 		 * */
2554 		sec_cnt = (value &
2555 			sdma_v4_0_ras_fields[i].sec_count_mask) >>
2556 			sdma_v4_0_ras_fields[i].sec_count_shift;
2557 		if (sec_cnt) {
2558 			DRM_INFO("Detected %s in SDMA%d, SED %d\n",
2559 				sdma_v4_0_ras_fields[i].name,
2560 				instance, sec_cnt);
2561 			*sec_count += sec_cnt;
2562 		}
2563 	}
2564 }
2565 
sdma_v4_0_query_ras_error_count_by_instance(struct amdgpu_device * adev,uint32_t instance,void * ras_error_status)2566 static int sdma_v4_0_query_ras_error_count_by_instance(struct amdgpu_device *adev,
2567 			uint32_t instance, void *ras_error_status)
2568 {
2569 	struct ras_err_data *err_data = (struct ras_err_data *)ras_error_status;
2570 	uint32_t sec_count = 0;
2571 	uint32_t reg_value = 0;
2572 
2573 	reg_value = RREG32_SDMA(instance, mmSDMA0_EDC_COUNTER);
2574 	/* double bit error is not supported */
2575 	if (reg_value)
2576 		sdma_v4_0_get_ras_error_count(reg_value,
2577 				instance, &sec_count);
2578 	/* err_data->ce_count should be initialized to 0
2579 	 * before calling into this function */
2580 	err_data->ce_count += sec_count;
2581 	/* double bit error is not supported
2582 	 * set ue count to 0 */
2583 	err_data->ue_count = 0;
2584 
2585 	return 0;
2586 };
2587 
sdma_v4_0_query_ras_error_count(struct amdgpu_device * adev,void * ras_error_status)2588 static void sdma_v4_0_query_ras_error_count(struct amdgpu_device *adev,  void *ras_error_status)
2589 {
2590 	int i = 0;
2591 
2592 	for (i = 0; i < adev->sdma.num_instances; i++) {
2593 		if (sdma_v4_0_query_ras_error_count_by_instance(adev, i, ras_error_status)) {
2594 			dev_err(adev->dev, "Query ras error count failed in SDMA%d\n", i);
2595 			return;
2596 		}
2597 	}
2598 }
2599 
sdma_v4_0_reset_ras_error_count(struct amdgpu_device * adev)2600 static void sdma_v4_0_reset_ras_error_count(struct amdgpu_device *adev)
2601 {
2602 	int i;
2603 
2604 	/* read back edc counter registers to clear the counters */
2605 	if (amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__SDMA)) {
2606 		for (i = 0; i < adev->sdma.num_instances; i++)
2607 			RREG32_SDMA(i, mmSDMA0_EDC_COUNTER);
2608 	}
2609 }
2610 
2611 const struct amdgpu_ras_block_hw_ops sdma_v4_0_ras_hw_ops = {
2612 	.query_ras_error_count = sdma_v4_0_query_ras_error_count,
2613 	.reset_ras_error_count = sdma_v4_0_reset_ras_error_count,
2614 };
2615 
2616 static struct amdgpu_sdma_ras sdma_v4_0_ras = {
2617 	.ras_block = {
2618 		.hw_ops = &sdma_v4_0_ras_hw_ops,
2619 		.ras_cb = sdma_v4_0_process_ras_data_cb,
2620 	},
2621 };
2622 
sdma_v4_0_set_ras_funcs(struct amdgpu_device * adev)2623 static void sdma_v4_0_set_ras_funcs(struct amdgpu_device *adev)
2624 {
2625 	switch (adev->ip_versions[SDMA0_HWIP][0]) {
2626 	case IP_VERSION(4, 2, 0):
2627 	case IP_VERSION(4, 2, 2):
2628 		adev->sdma.ras = &sdma_v4_0_ras;
2629 		break;
2630 	case IP_VERSION(4, 4, 0):
2631 		adev->sdma.ras = &sdma_v4_4_ras;
2632 		break;
2633 	default:
2634 		break;
2635 	}
2636 
2637 }
2638 
2639 const struct amdgpu_ip_block_version sdma_v4_0_ip_block = {
2640 	.type = AMD_IP_BLOCK_TYPE_SDMA,
2641 	.major = 4,
2642 	.minor = 0,
2643 	.rev = 0,
2644 	.funcs = &sdma_v4_0_ip_funcs,
2645 };
2646