• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 "hdp/hdp_4_0_offset.h"
50 #include "sdma0/sdma0_4_1_default.h"
51 
52 #include "soc15_common.h"
53 #include "soc15.h"
54 #include "vega10_sdma_pkt_open.h"
55 
56 #include "ivsrcid/sdma0/irqsrcs_sdma0_4_0.h"
57 #include "ivsrcid/sdma1/irqsrcs_sdma1_4_0.h"
58 
59 #include "amdgpu_ras.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 
73 #define SDMA0_POWER_CNTL__ON_OFF_CONDITION_HOLD_TIME_MASK  0x000000F8L
74 #define SDMA0_POWER_CNTL__ON_OFF_STATUS_DURATION_TIME_MASK 0xFC000000L
75 
76 #define WREG32_SDMA(instance, offset, value) \
77 	WREG32(sdma_v4_0_get_reg_offset(adev, (instance), (offset)), value)
78 #define RREG32_SDMA(instance, offset) \
79 	RREG32(sdma_v4_0_get_reg_offset(adev, (instance), (offset)))
80 
81 static void sdma_v4_0_set_ring_funcs(struct amdgpu_device *adev);
82 static void sdma_v4_0_set_buffer_funcs(struct amdgpu_device *adev);
83 static void sdma_v4_0_set_vm_pte_funcs(struct amdgpu_device *adev);
84 static void sdma_v4_0_set_irq_funcs(struct amdgpu_device *adev);
85 
86 static const struct soc15_reg_golden golden_settings_sdma_4[] = {
87 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_CHICKEN_BITS, 0xfe931f07, 0x02831d07),
88 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_CLK_CTRL, 0xff000ff0, 0x3f000100),
89 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GFX_IB_CNTL, 0x800f0100, 0x00000100),
90 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GFX_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
91 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_PAGE_IB_CNTL, 0x800f0100, 0x00000100),
92 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_PAGE_RB_WPTR_POLL_CNTL, 0x0000fff0, 0x00403000),
93 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_POWER_CNTL, 0x003ff006, 0x0003c000),
94 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC0_IB_CNTL, 0x800f0100, 0x00000100),
95 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC0_RB_WPTR_POLL_CNTL, 0x0000fff0, 0x00403000),
96 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC1_IB_CNTL, 0x800f0100, 0x00000100),
97 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC1_RB_WPTR_POLL_CNTL, 0x0000fff0, 0x00403000),
98 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_UTCL1_PAGE, 0x000003ff, 0x000003c0),
99 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_UTCL1_WATERMK, 0xfc000000, 0x00000000),
100 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_CLK_CTRL, 0xffffffff, 0x3f000100),
101 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_GFX_IB_CNTL, 0x800f0100, 0x00000100),
102 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_GFX_RB_WPTR_POLL_CNTL, 0x0000fff0, 0x00403000),
103 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_PAGE_IB_CNTL, 0x800f0100, 0x00000100),
104 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_PAGE_RB_WPTR_POLL_CNTL, 0x0000fff0, 0x00403000),
105 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_POWER_CNTL, 0x003ff000, 0x0003c000),
106 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_RLC0_IB_CNTL, 0x800f0100, 0x00000100),
107 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_RLC0_RB_WPTR_POLL_CNTL, 0x0000fff0, 0x00403000),
108 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_RLC1_IB_CNTL, 0x800f0100, 0x00000100),
109 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_RLC1_RB_WPTR_POLL_CNTL, 0x0000fff0, 0x00403000),
110 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_UTCL1_PAGE, 0x000003ff, 0x000003c0),
111 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_UTCL1_WATERMK, 0xfc000000, 0x00000000)
112 };
113 
114 static const struct soc15_reg_golden golden_settings_sdma_vg10[] = {
115 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GB_ADDR_CONFIG, 0x0018773f, 0x00104002),
116 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GB_ADDR_CONFIG_READ, 0x0018773f, 0x00104002),
117 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_CHICKEN_BITS, 0xfe931f07, 0x02831d07),
118 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_GB_ADDR_CONFIG, 0x0018773f, 0x00104002),
119 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_GB_ADDR_CONFIG_READ, 0x0018773f, 0x00104002)
120 };
121 
122 static const struct soc15_reg_golden golden_settings_sdma_vg12[] = {
123 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GB_ADDR_CONFIG, 0x0018773f, 0x00104001),
124 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GB_ADDR_CONFIG_READ, 0x0018773f, 0x00104001),
125 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_CHICKEN_BITS, 0xfe931f07, 0x02831d07),
126 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_GB_ADDR_CONFIG, 0x0018773f, 0x00104001),
127 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_GB_ADDR_CONFIG_READ, 0x0018773f, 0x00104001)
128 };
129 
130 static const struct soc15_reg_golden golden_settings_sdma_4_1[] = {
131 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_CHICKEN_BITS, 0xfe931f07, 0x02831d07),
132 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_CLK_CTRL, 0xffffffff, 0x3f000100),
133 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GFX_IB_CNTL, 0x800f0111, 0x00000100),
134 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GFX_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
135 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_POWER_CNTL, 0xfc3fffff, 0x40000051),
136 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC0_IB_CNTL, 0x800f0111, 0x00000100),
137 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC0_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
138 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC1_IB_CNTL, 0x800f0111, 0x00000100),
139 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC1_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
140 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_UTCL1_PAGE, 0x000003ff, 0x000003c0),
141 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_UTCL1_WATERMK, 0xfc000000, 0x00000000)
142 };
143 
144 static const struct soc15_reg_golden golden_settings_sdma0_4_2_init[] = {
145 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC0_RB_WPTR_POLL_CNTL, 0xfffffff0, 0x00403000),
146 };
147 
148 static const struct soc15_reg_golden golden_settings_sdma0_4_2[] =
149 {
150 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_CHICKEN_BITS, 0xfe931f07, 0x02831f07),
151 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_CLK_CTRL, 0xffffffff, 0x3f000100),
152 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GB_ADDR_CONFIG, 0x0000773f, 0x00004002),
153 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GB_ADDR_CONFIG_READ, 0x0000773f, 0x00004002),
154 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GFX_RB_RPTR_ADDR_LO, 0xfffffffd, 0x00000001),
155 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GFX_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
156 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_PAGE_RB_RPTR_ADDR_LO, 0xfffffffd, 0x00000001),
157 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_PAGE_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
158 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RD_BURST_CNTL, 0x0000000f, 0x00000003),
159 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC0_RB_RPTR_ADDR_LO, 0xfffffffd, 0x00000001),
160 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC0_RB_WPTR_POLL_CNTL, 0xfffffff0, 0x00403000),
161 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC1_RB_RPTR_ADDR_LO, 0xfffffffd, 0x00000001),
162 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC1_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
163 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC2_RB_RPTR_ADDR_LO, 0xfffffffd, 0x00000001),
164 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC2_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
165 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC3_RB_RPTR_ADDR_LO, 0xfffffffd, 0x00000001),
166 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC3_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
167 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC4_RB_RPTR_ADDR_LO, 0xfffffffd, 0x00000001),
168 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC4_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
169 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC5_RB_RPTR_ADDR_LO, 0xfffffffd, 0x00000001),
170 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC5_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
171 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC6_RB_RPTR_ADDR_LO, 0xfffffffd, 0x00000001),
172 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC6_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
173 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC7_RB_RPTR_ADDR_LO, 0xfffffffd, 0x00000001),
174 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC7_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
175 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_UTCL1_PAGE, 0x000003ff, 0x000003c0),
176 };
177 
178 static const struct soc15_reg_golden golden_settings_sdma1_4_2[] = {
179 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_CHICKEN_BITS, 0xfe931f07, 0x02831f07),
180 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_CLK_CTRL, 0xffffffff, 0x3f000100),
181 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_GB_ADDR_CONFIG, 0x0000773f, 0x00004002),
182 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_GB_ADDR_CONFIG_READ, 0x0000773f, 0x00004002),
183 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_GFX_RB_RPTR_ADDR_LO, 0xfffffffd, 0x00000001),
184 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_GFX_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
185 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_PAGE_RB_RPTR_ADDR_LO, 0xfffffffd, 0x00000001),
186 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_PAGE_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
187 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_RD_BURST_CNTL, 0x0000000f, 0x00000003),
188 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_RLC0_RB_RPTR_ADDR_LO, 0xfffffffd, 0x00000001),
189 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_RLC0_RB_WPTR_POLL_CNTL, 0xfffffff0, 0x00403000),
190 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_RLC1_RB_RPTR_ADDR_LO, 0xfffffffd, 0x00000001),
191 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_RLC1_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
192 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_RLC2_RB_RPTR_ADDR_LO, 0xfffffffd, 0x00000001),
193 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_RLC2_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
194 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_RLC3_RB_RPTR_ADDR_LO, 0xfffffffd, 0x00000001),
195 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_RLC3_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
196 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_RLC4_RB_RPTR_ADDR_LO, 0xfffffffd, 0x00000001),
197 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_RLC4_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
198 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_RLC5_RB_RPTR_ADDR_LO, 0xfffffffd, 0x00000001),
199 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_RLC5_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
200 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_RLC6_RB_RPTR_ADDR_LO, 0xfffffffd, 0x00000001),
201 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_RLC6_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
202 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_RLC7_RB_RPTR_ADDR_LO, 0xfffffffd, 0x00000001),
203 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_RLC7_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
204 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_UTCL1_PAGE, 0x000003ff, 0x000003c0),
205 };
206 
207 static const struct soc15_reg_golden golden_settings_sdma_rv1[] =
208 {
209 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GB_ADDR_CONFIG, 0x0018773f, 0x00000002),
210 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GB_ADDR_CONFIG_READ, 0x0018773f, 0x00000002)
211 };
212 
213 static const struct soc15_reg_golden golden_settings_sdma_rv2[] =
214 {
215 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GB_ADDR_CONFIG, 0x0018773f, 0x00003001),
216 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GB_ADDR_CONFIG_READ, 0x0018773f, 0x00003001)
217 };
218 
219 static const struct soc15_reg_golden golden_settings_sdma_arct[] =
220 {
221 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_CHICKEN_BITS, 0xfe931f07, 0x02831f07),
222 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GB_ADDR_CONFIG, 0x0000773f, 0x00004002),
223 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GB_ADDR_CONFIG_READ, 0x0000773f, 0x00004002),
224 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_CHICKEN_BITS, 0xfe931f07, 0x02831f07),
225 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_GB_ADDR_CONFIG, 0x0000773f, 0x00004002),
226 	SOC15_REG_GOLDEN_VALUE(SDMA1, 0, mmSDMA1_GB_ADDR_CONFIG_READ, 0x0000773f, 0x00004002),
227 	SOC15_REG_GOLDEN_VALUE(SDMA2, 0, mmSDMA2_CHICKEN_BITS, 0xfe931f07, 0x02831f07),
228 	SOC15_REG_GOLDEN_VALUE(SDMA2, 0, mmSDMA2_GB_ADDR_CONFIG, 0x0000773f, 0x00004002),
229 	SOC15_REG_GOLDEN_VALUE(SDMA2, 0, mmSDMA2_GB_ADDR_CONFIG_READ, 0x0000773f, 0x00004002),
230 	SOC15_REG_GOLDEN_VALUE(SDMA3, 0, mmSDMA3_CHICKEN_BITS, 0xfe931f07, 0x02831f07),
231 	SOC15_REG_GOLDEN_VALUE(SDMA3, 0, mmSDMA3_GB_ADDR_CONFIG, 0x0000773f, 0x00004002),
232 	SOC15_REG_GOLDEN_VALUE(SDMA3, 0, mmSDMA3_GB_ADDR_CONFIG_READ, 0x0000773f, 0x00004002),
233 	SOC15_REG_GOLDEN_VALUE(SDMA4, 0, mmSDMA4_CHICKEN_BITS, 0xfe931f07, 0x02831f07),
234 	SOC15_REG_GOLDEN_VALUE(SDMA4, 0, mmSDMA4_GB_ADDR_CONFIG, 0x0000773f, 0x00004002),
235 	SOC15_REG_GOLDEN_VALUE(SDMA4, 0, mmSDMA4_GB_ADDR_CONFIG_READ, 0x0000773f, 0x00004002),
236 	SOC15_REG_GOLDEN_VALUE(SDMA5, 0, mmSDMA5_CHICKEN_BITS, 0xfe931f07, 0x02831f07),
237 	SOC15_REG_GOLDEN_VALUE(SDMA5, 0, mmSDMA5_GB_ADDR_CONFIG, 0x0000773f, 0x00004002),
238 	SOC15_REG_GOLDEN_VALUE(SDMA5, 0, mmSDMA5_GB_ADDR_CONFIG_READ, 0x0000773f, 0x00004002),
239 	SOC15_REG_GOLDEN_VALUE(SDMA6, 0, mmSDMA6_CHICKEN_BITS, 0xfe931f07, 0x02831f07),
240 	SOC15_REG_GOLDEN_VALUE(SDMA6, 0, mmSDMA6_GB_ADDR_CONFIG, 0x0000773f, 0x00004002),
241 	SOC15_REG_GOLDEN_VALUE(SDMA6, 0, mmSDMA6_GB_ADDR_CONFIG_READ, 0x0000773f, 0x00004002),
242 	SOC15_REG_GOLDEN_VALUE(SDMA7, 0, mmSDMA7_CHICKEN_BITS, 0xfe931f07, 0x02831f07),
243 	SOC15_REG_GOLDEN_VALUE(SDMA7, 0, mmSDMA7_GB_ADDR_CONFIG, 0x0000773f, 0x00004002),
244 	SOC15_REG_GOLDEN_VALUE(SDMA7, 0, mmSDMA7_GB_ADDR_CONFIG_READ, 0x0000773f, 0x00004002)
245 };
246 
247 static const struct soc15_reg_golden golden_settings_sdma_4_3[] = {
248 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_CHICKEN_BITS, 0xfe931f07, 0x02831f07),
249 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_CLK_CTRL, 0xffffffff, 0x3f000100),
250 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GB_ADDR_CONFIG, 0x0018773f, 0x00000002),
251 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GB_ADDR_CONFIG_READ, 0x0018773f, 0x00000002),
252 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_GFX_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
253 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_POWER_CNTL, 0x003fff07, 0x40000051),
254 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC0_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
255 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_RLC1_RB_WPTR_POLL_CNTL, 0xfffffff7, 0x00403000),
256 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_UTCL1_PAGE, 0x000003ff, 0x000003c0),
257 	SOC15_REG_GOLDEN_VALUE(SDMA0, 0, mmSDMA0_UTCL1_WATERMK, 0xfc000000, 0x00000000)
258 };
259 
sdma_v4_0_get_reg_offset(struct amdgpu_device * adev,u32 instance,u32 offset)260 static u32 sdma_v4_0_get_reg_offset(struct amdgpu_device *adev,
261 		u32 instance, u32 offset)
262 {
263 	switch (instance) {
264 	case 0:
265 		return (adev->reg_offset[SDMA0_HWIP][0][0] + offset);
266 	case 1:
267 		return (adev->reg_offset[SDMA1_HWIP][0][0] + offset);
268 	case 2:
269 		return (adev->reg_offset[SDMA2_HWIP][0][1] + offset);
270 	case 3:
271 		return (adev->reg_offset[SDMA3_HWIP][0][1] + offset);
272 	case 4:
273 		return (adev->reg_offset[SDMA4_HWIP][0][1] + offset);
274 	case 5:
275 		return (adev->reg_offset[SDMA5_HWIP][0][1] + offset);
276 	case 6:
277 		return (adev->reg_offset[SDMA6_HWIP][0][1] + offset);
278 	case 7:
279 		return (adev->reg_offset[SDMA7_HWIP][0][1] + offset);
280 	default:
281 		break;
282 	}
283 	return 0;
284 }
285 
sdma_v4_0_seq_to_irq_id(int seq_num)286 static unsigned sdma_v4_0_seq_to_irq_id(int seq_num)
287 {
288 	switch (seq_num) {
289 	case 0:
290 		return SOC15_IH_CLIENTID_SDMA0;
291 	case 1:
292 		return SOC15_IH_CLIENTID_SDMA1;
293 	case 2:
294 		return SOC15_IH_CLIENTID_SDMA2;
295 	case 3:
296 		return SOC15_IH_CLIENTID_SDMA3;
297 	case 4:
298 		return SOC15_IH_CLIENTID_SDMA4;
299 	case 5:
300 		return SOC15_IH_CLIENTID_SDMA5;
301 	case 6:
302 		return SOC15_IH_CLIENTID_SDMA6;
303 	case 7:
304 		return SOC15_IH_CLIENTID_SDMA7;
305 	default:
306 		break;
307 	}
308 	return -EINVAL;
309 }
310 
sdma_v4_0_irq_id_to_seq(unsigned client_id)311 static int sdma_v4_0_irq_id_to_seq(unsigned client_id)
312 {
313 	switch (client_id) {
314 	case SOC15_IH_CLIENTID_SDMA0:
315 		return 0;
316 	case SOC15_IH_CLIENTID_SDMA1:
317 		return 1;
318 	case SOC15_IH_CLIENTID_SDMA2:
319 		return 2;
320 	case SOC15_IH_CLIENTID_SDMA3:
321 		return 3;
322 	case SOC15_IH_CLIENTID_SDMA4:
323 		return 4;
324 	case SOC15_IH_CLIENTID_SDMA5:
325 		return 5;
326 	case SOC15_IH_CLIENTID_SDMA6:
327 		return 6;
328 	case SOC15_IH_CLIENTID_SDMA7:
329 		return 7;
330 	default:
331 		break;
332 	}
333 	return -EINVAL;
334 }
335 
sdma_v4_0_init_golden_registers(struct amdgpu_device * adev)336 static void sdma_v4_0_init_golden_registers(struct amdgpu_device *adev)
337 {
338 	switch (adev->asic_type) {
339 	case CHIP_VEGA10:
340 		soc15_program_register_sequence(adev,
341 						golden_settings_sdma_4,
342 						ARRAY_SIZE(golden_settings_sdma_4));
343 		soc15_program_register_sequence(adev,
344 						golden_settings_sdma_vg10,
345 						ARRAY_SIZE(golden_settings_sdma_vg10));
346 		break;
347 	case CHIP_VEGA12:
348 		soc15_program_register_sequence(adev,
349 						golden_settings_sdma_4,
350 						ARRAY_SIZE(golden_settings_sdma_4));
351 		soc15_program_register_sequence(adev,
352 						golden_settings_sdma_vg12,
353 						ARRAY_SIZE(golden_settings_sdma_vg12));
354 		break;
355 	case CHIP_VEGA20:
356 		soc15_program_register_sequence(adev,
357 						golden_settings_sdma0_4_2_init,
358 						ARRAY_SIZE(golden_settings_sdma0_4_2_init));
359 		soc15_program_register_sequence(adev,
360 						golden_settings_sdma0_4_2,
361 						ARRAY_SIZE(golden_settings_sdma0_4_2));
362 		soc15_program_register_sequence(adev,
363 						golden_settings_sdma1_4_2,
364 						ARRAY_SIZE(golden_settings_sdma1_4_2));
365 		break;
366 	case CHIP_ARCTURUS:
367 		soc15_program_register_sequence(adev,
368 						golden_settings_sdma_arct,
369 						ARRAY_SIZE(golden_settings_sdma_arct));
370 		break;
371 	case CHIP_RAVEN:
372 		soc15_program_register_sequence(adev,
373 						golden_settings_sdma_4_1,
374 						ARRAY_SIZE(golden_settings_sdma_4_1));
375 		if (adev->rev_id >= 8)
376 			soc15_program_register_sequence(adev,
377 							golden_settings_sdma_rv2,
378 							ARRAY_SIZE(golden_settings_sdma_rv2));
379 		else
380 			soc15_program_register_sequence(adev,
381 							golden_settings_sdma_rv1,
382 							ARRAY_SIZE(golden_settings_sdma_rv1));
383 		break;
384 	case CHIP_RENOIR:
385 		soc15_program_register_sequence(adev,
386 						golden_settings_sdma_4_3,
387 						ARRAY_SIZE(golden_settings_sdma_4_3));
388 		break;
389 	default:
390 		break;
391 	}
392 }
393 
sdma_v4_0_init_inst_ctx(struct amdgpu_sdma_instance * sdma_inst)394 static int sdma_v4_0_init_inst_ctx(struct amdgpu_sdma_instance *sdma_inst)
395 {
396 	int err = 0;
397 	const struct sdma_firmware_header_v1_0 *hdr;
398 
399 	err = amdgpu_ucode_validate(sdma_inst->fw);
400 	if (err)
401 		return err;
402 
403 	hdr = (const struct sdma_firmware_header_v1_0 *)sdma_inst->fw->data;
404 	sdma_inst->fw_version = le32_to_cpu(hdr->header.ucode_version);
405 	sdma_inst->feature_version = le32_to_cpu(hdr->ucode_feature_version);
406 
407 	if (sdma_inst->feature_version >= 20)
408 		sdma_inst->burst_nop = true;
409 
410 	return 0;
411 }
412 
sdma_v4_0_destroy_inst_ctx(struct amdgpu_device * adev)413 static void sdma_v4_0_destroy_inst_ctx(struct amdgpu_device *adev)
414 {
415 	int i;
416 
417 	for (i = 0; i < adev->sdma.num_instances; i++) {
418 		if (adev->sdma.instance[i].fw != NULL)
419 			release_firmware(adev->sdma.instance[i].fw);
420 
421 		/* arcturus shares the same FW memory across
422 		   all SDMA isntances */
423 		if (adev->asic_type == CHIP_ARCTURUS)
424 			break;
425 	}
426 
427 	memset((void*)adev->sdma.instance, 0,
428 		sizeof(struct amdgpu_sdma_instance) * AMDGPU_MAX_SDMA_INSTANCES);
429 }
430 
431 /**
432  * sdma_v4_0_init_microcode - load ucode images from disk
433  *
434  * @adev: amdgpu_device pointer
435  *
436  * Use the firmware interface to load the ucode images into
437  * the driver (not loaded into hw).
438  * Returns 0 on success, error on failure.
439  */
440 
441 // emulation only, won't work on real chip
442 // vega10 real chip need to use PSP to load firmware
sdma_v4_0_init_microcode(struct amdgpu_device * adev)443 static int sdma_v4_0_init_microcode(struct amdgpu_device *adev)
444 {
445 	const char *chip_name;
446 	char fw_name[30];
447 	int err = 0, i;
448 	struct amdgpu_firmware_info *info = NULL;
449 	const struct common_firmware_header *header = NULL;
450 
451 	DRM_DEBUG("\n");
452 
453 	switch (adev->asic_type) {
454 	case CHIP_VEGA10:
455 		chip_name = "vega10";
456 		break;
457 	case CHIP_VEGA12:
458 		chip_name = "vega12";
459 		break;
460 	case CHIP_VEGA20:
461 		chip_name = "vega20";
462 		break;
463 	case CHIP_RAVEN:
464 		if (adev->rev_id >= 8)
465 			chip_name = "raven2";
466 		else if (adev->pdev->device == 0x15d8)
467 			chip_name = "picasso";
468 		else
469 			chip_name = "raven";
470 		break;
471 	case CHIP_ARCTURUS:
472 		chip_name = "arcturus";
473 		break;
474 	case CHIP_RENOIR:
475 		chip_name = "renoir";
476 		break;
477 	default:
478 		BUG();
479 	}
480 
481 	snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_sdma.bin", chip_name);
482 
483 	err = request_firmware(&adev->sdma.instance[0].fw, fw_name, adev->dev);
484 	if (err)
485 		goto out;
486 
487 	err = sdma_v4_0_init_inst_ctx(&adev->sdma.instance[0]);
488 	if (err)
489 		goto out;
490 
491 	for (i = 1; i < adev->sdma.num_instances; i++) {
492 		if (adev->asic_type == CHIP_ARCTURUS) {
493 			/* Acturus will leverage the same FW memory
494 			   for every SDMA instance */
495 			memcpy((void*)&adev->sdma.instance[i],
496 			       (void*)&adev->sdma.instance[0],
497 			       sizeof(struct amdgpu_sdma_instance));
498 		}
499 		else {
500 			snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_sdma%d.bin", chip_name, i);
501 
502 			err = request_firmware(&adev->sdma.instance[i].fw, fw_name, adev->dev);
503 			if (err)
504 				goto out;
505 
506 			err = sdma_v4_0_init_inst_ctx(&adev->sdma.instance[i]);
507 			if (err)
508 				goto out;
509 		}
510 	}
511 
512 	DRM_DEBUG("psp_load == '%s'\n",
513 		adev->firmware.load_type == AMDGPU_FW_LOAD_PSP ? "true" : "false");
514 
515 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
516 		for (i = 0; i < adev->sdma.num_instances; i++) {
517 			info = &adev->firmware.ucode[AMDGPU_UCODE_ID_SDMA0 + i];
518 			info->ucode_id = AMDGPU_UCODE_ID_SDMA0 + i;
519 			info->fw = adev->sdma.instance[i].fw;
520 			header = (const struct common_firmware_header *)info->fw->data;
521 			adev->firmware.fw_size +=
522 				ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
523 		}
524 	}
525 
526 out:
527 	if (err) {
528 		DRM_ERROR("sdma_v4_0: Failed to load firmware \"%s\"\n", fw_name);
529 		sdma_v4_0_destroy_inst_ctx(adev);
530 	}
531 	return err;
532 }
533 
534 /**
535  * sdma_v4_0_ring_get_rptr - get the current read pointer
536  *
537  * @ring: amdgpu ring pointer
538  *
539  * Get the current rptr from the hardware (VEGA10+).
540  */
sdma_v4_0_ring_get_rptr(struct amdgpu_ring * ring)541 static uint64_t sdma_v4_0_ring_get_rptr(struct amdgpu_ring *ring)
542 {
543 	u64 *rptr;
544 
545 	/* XXX check if swapping is necessary on BE */
546 	rptr = ((u64 *)&ring->adev->wb.wb[ring->rptr_offs]);
547 
548 	DRM_DEBUG("rptr before shift == 0x%016llx\n", *rptr);
549 	return ((*rptr) >> 2);
550 }
551 
552 /**
553  * sdma_v4_0_ring_get_wptr - get the current write pointer
554  *
555  * @ring: amdgpu ring pointer
556  *
557  * Get the current wptr from the hardware (VEGA10+).
558  */
sdma_v4_0_ring_get_wptr(struct amdgpu_ring * ring)559 static uint64_t sdma_v4_0_ring_get_wptr(struct amdgpu_ring *ring)
560 {
561 	struct amdgpu_device *adev = ring->adev;
562 	u64 wptr;
563 
564 	if (ring->use_doorbell) {
565 		/* XXX check if swapping is necessary on BE */
566 		wptr = READ_ONCE(*((u64 *)&adev->wb.wb[ring->wptr_offs]));
567 		DRM_DEBUG("wptr/doorbell before shift == 0x%016llx\n", wptr);
568 	} else {
569 		wptr = RREG32_SDMA(ring->me, mmSDMA0_GFX_RB_WPTR_HI);
570 		wptr = wptr << 32;
571 		wptr |= RREG32_SDMA(ring->me, mmSDMA0_GFX_RB_WPTR);
572 		DRM_DEBUG("wptr before shift [%i] wptr == 0x%016llx\n",
573 				ring->me, wptr);
574 	}
575 
576 	return wptr >> 2;
577 }
578 
579 /**
580  * sdma_v4_0_ring_set_wptr - commit the write pointer
581  *
582  * @ring: amdgpu ring pointer
583  *
584  * Write the wptr back to the hardware (VEGA10+).
585  */
sdma_v4_0_ring_set_wptr(struct amdgpu_ring * ring)586 static void sdma_v4_0_ring_set_wptr(struct amdgpu_ring *ring)
587 {
588 	struct amdgpu_device *adev = ring->adev;
589 
590 	DRM_DEBUG("Setting write pointer\n");
591 	if (ring->use_doorbell) {
592 		u64 *wb = (u64 *)&adev->wb.wb[ring->wptr_offs];
593 
594 		DRM_DEBUG("Using doorbell -- "
595 				"wptr_offs == 0x%08x "
596 				"lower_32_bits(ring->wptr) << 2 == 0x%08x "
597 				"upper_32_bits(ring->wptr) << 2 == 0x%08x\n",
598 				ring->wptr_offs,
599 				lower_32_bits(ring->wptr << 2),
600 				upper_32_bits(ring->wptr << 2));
601 		/* XXX check if swapping is necessary on BE */
602 		WRITE_ONCE(*wb, (ring->wptr << 2));
603 		DRM_DEBUG("calling WDOORBELL64(0x%08x, 0x%016llx)\n",
604 				ring->doorbell_index, ring->wptr << 2);
605 		WDOORBELL64(ring->doorbell_index, ring->wptr << 2);
606 	} else {
607 		DRM_DEBUG("Not using doorbell -- "
608 				"mmSDMA%i_GFX_RB_WPTR == 0x%08x "
609 				"mmSDMA%i_GFX_RB_WPTR_HI == 0x%08x\n",
610 				ring->me,
611 				lower_32_bits(ring->wptr << 2),
612 				ring->me,
613 				upper_32_bits(ring->wptr << 2));
614 		WREG32_SDMA(ring->me, mmSDMA0_GFX_RB_WPTR,
615 			    lower_32_bits(ring->wptr << 2));
616 		WREG32_SDMA(ring->me, mmSDMA0_GFX_RB_WPTR_HI,
617 			    upper_32_bits(ring->wptr << 2));
618 	}
619 }
620 
621 /**
622  * sdma_v4_0_page_ring_get_wptr - get the current write pointer
623  *
624  * @ring: amdgpu ring pointer
625  *
626  * Get the current wptr from the hardware (VEGA10+).
627  */
sdma_v4_0_page_ring_get_wptr(struct amdgpu_ring * ring)628 static uint64_t sdma_v4_0_page_ring_get_wptr(struct amdgpu_ring *ring)
629 {
630 	struct amdgpu_device *adev = ring->adev;
631 	u64 wptr;
632 
633 	if (ring->use_doorbell) {
634 		/* XXX check if swapping is necessary on BE */
635 		wptr = READ_ONCE(*((u64 *)&adev->wb.wb[ring->wptr_offs]));
636 	} else {
637 		wptr = RREG32_SDMA(ring->me, mmSDMA0_PAGE_RB_WPTR_HI);
638 		wptr = wptr << 32;
639 		wptr |= RREG32_SDMA(ring->me, mmSDMA0_PAGE_RB_WPTR);
640 	}
641 
642 	return wptr >> 2;
643 }
644 
645 /**
646  * sdma_v4_0_ring_set_wptr - commit the write pointer
647  *
648  * @ring: amdgpu ring pointer
649  *
650  * Write the wptr back to the hardware (VEGA10+).
651  */
sdma_v4_0_page_ring_set_wptr(struct amdgpu_ring * ring)652 static void sdma_v4_0_page_ring_set_wptr(struct amdgpu_ring *ring)
653 {
654 	struct amdgpu_device *adev = ring->adev;
655 
656 	if (ring->use_doorbell) {
657 		u64 *wb = (u64 *)&adev->wb.wb[ring->wptr_offs];
658 
659 		/* XXX check if swapping is necessary on BE */
660 		WRITE_ONCE(*wb, (ring->wptr << 2));
661 		WDOORBELL64(ring->doorbell_index, ring->wptr << 2);
662 	} else {
663 		uint64_t wptr = ring->wptr << 2;
664 
665 		WREG32_SDMA(ring->me, mmSDMA0_PAGE_RB_WPTR,
666 			    lower_32_bits(wptr));
667 		WREG32_SDMA(ring->me, mmSDMA0_PAGE_RB_WPTR_HI,
668 			    upper_32_bits(wptr));
669 	}
670 }
671 
sdma_v4_0_ring_insert_nop(struct amdgpu_ring * ring,uint32_t count)672 static void sdma_v4_0_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count)
673 {
674 	struct amdgpu_sdma_instance *sdma = amdgpu_sdma_get_instance_from_ring(ring);
675 	int i;
676 
677 	for (i = 0; i < count; i++)
678 		if (sdma && sdma->burst_nop && (i == 0))
679 			amdgpu_ring_write(ring, ring->funcs->nop |
680 				SDMA_PKT_NOP_HEADER_COUNT(count - 1));
681 		else
682 			amdgpu_ring_write(ring, ring->funcs->nop);
683 }
684 
685 /**
686  * sdma_v4_0_ring_emit_ib - Schedule an IB on the DMA engine
687  *
688  * @ring: amdgpu ring pointer
689  * @ib: IB object to schedule
690  *
691  * Schedule an IB in the DMA ring (VEGA10).
692  */
sdma_v4_0_ring_emit_ib(struct amdgpu_ring * ring,struct amdgpu_job * job,struct amdgpu_ib * ib,uint32_t flags)693 static void sdma_v4_0_ring_emit_ib(struct amdgpu_ring *ring,
694 				   struct amdgpu_job *job,
695 				   struct amdgpu_ib *ib,
696 				   uint32_t flags)
697 {
698 	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
699 
700 	/* IB packet must end on a 8 DW boundary */
701 	sdma_v4_0_ring_insert_nop(ring, (10 - (lower_32_bits(ring->wptr) & 7)) % 8);
702 
703 	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_INDIRECT) |
704 			  SDMA_PKT_INDIRECT_HEADER_VMID(vmid & 0xf));
705 	/* base must be 32 byte aligned */
706 	amdgpu_ring_write(ring, lower_32_bits(ib->gpu_addr) & 0xffffffe0);
707 	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr));
708 	amdgpu_ring_write(ring, ib->length_dw);
709 	amdgpu_ring_write(ring, 0);
710 	amdgpu_ring_write(ring, 0);
711 
712 }
713 
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)714 static void sdma_v4_0_wait_reg_mem(struct amdgpu_ring *ring,
715 				   int mem_space, int hdp,
716 				   uint32_t addr0, uint32_t addr1,
717 				   uint32_t ref, uint32_t mask,
718 				   uint32_t inv)
719 {
720 	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_POLL_REGMEM) |
721 			  SDMA_PKT_POLL_REGMEM_HEADER_HDP_FLUSH(hdp) |
722 			  SDMA_PKT_POLL_REGMEM_HEADER_MEM_POLL(mem_space) |
723 			  SDMA_PKT_POLL_REGMEM_HEADER_FUNC(3)); /* == */
724 	if (mem_space) {
725 		/* memory */
726 		amdgpu_ring_write(ring, addr0);
727 		amdgpu_ring_write(ring, addr1);
728 	} else {
729 		/* registers */
730 		amdgpu_ring_write(ring, addr0 << 2);
731 		amdgpu_ring_write(ring, addr1 << 2);
732 	}
733 	amdgpu_ring_write(ring, ref); /* reference */
734 	amdgpu_ring_write(ring, mask); /* mask */
735 	amdgpu_ring_write(ring, SDMA_PKT_POLL_REGMEM_DW5_RETRY_COUNT(0xfff) |
736 			  SDMA_PKT_POLL_REGMEM_DW5_INTERVAL(inv)); /* retry count, poll interval */
737 }
738 
739 /**
740  * sdma_v4_0_ring_emit_hdp_flush - emit an hdp flush on the DMA ring
741  *
742  * @ring: amdgpu ring pointer
743  *
744  * Emit an hdp flush packet on the requested DMA ring.
745  */
sdma_v4_0_ring_emit_hdp_flush(struct amdgpu_ring * ring)746 static void sdma_v4_0_ring_emit_hdp_flush(struct amdgpu_ring *ring)
747 {
748 	struct amdgpu_device *adev = ring->adev;
749 	u32 ref_and_mask = 0;
750 	const struct nbio_hdp_flush_reg *nbio_hf_reg = adev->nbio_funcs->hdp_flush_reg;
751 
752 	ref_and_mask = nbio_hf_reg->ref_and_mask_sdma0 << ring->me;
753 
754 	sdma_v4_0_wait_reg_mem(ring, 0, 1,
755 			       adev->nbio_funcs->get_hdp_flush_done_offset(adev),
756 			       adev->nbio_funcs->get_hdp_flush_req_offset(adev),
757 			       ref_and_mask, ref_and_mask, 10);
758 }
759 
760 /**
761  * sdma_v4_0_ring_emit_fence - emit a fence on the DMA ring
762  *
763  * @ring: amdgpu ring pointer
764  * @fence: amdgpu fence object
765  *
766  * Add a DMA fence packet to the ring to write
767  * the fence seq number and DMA trap packet to generate
768  * an interrupt if needed (VEGA10).
769  */
sdma_v4_0_ring_emit_fence(struct amdgpu_ring * ring,u64 addr,u64 seq,unsigned flags)770 static void sdma_v4_0_ring_emit_fence(struct amdgpu_ring *ring, u64 addr, u64 seq,
771 				      unsigned flags)
772 {
773 	bool write64bit = flags & AMDGPU_FENCE_FLAG_64BIT;
774 	/* write the fence */
775 	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_FENCE));
776 	/* zero in first two bits */
777 	BUG_ON(addr & 0x3);
778 	amdgpu_ring_write(ring, lower_32_bits(addr));
779 	amdgpu_ring_write(ring, upper_32_bits(addr));
780 	amdgpu_ring_write(ring, lower_32_bits(seq));
781 
782 	/* optionally write high bits as well */
783 	if (write64bit) {
784 		addr += 4;
785 		amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_FENCE));
786 		/* zero in first two bits */
787 		BUG_ON(addr & 0x3);
788 		amdgpu_ring_write(ring, lower_32_bits(addr));
789 		amdgpu_ring_write(ring, upper_32_bits(addr));
790 		amdgpu_ring_write(ring, upper_32_bits(seq));
791 	}
792 
793 	/* generate an interrupt */
794 	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_TRAP));
795 	amdgpu_ring_write(ring, SDMA_PKT_TRAP_INT_CONTEXT_INT_CONTEXT(0));
796 }
797 
798 
799 /**
800  * sdma_v4_0_gfx_stop - stop the gfx async dma engines
801  *
802  * @adev: amdgpu_device pointer
803  *
804  * Stop the gfx async dma ring buffers (VEGA10).
805  */
sdma_v4_0_gfx_stop(struct amdgpu_device * adev)806 static void sdma_v4_0_gfx_stop(struct amdgpu_device *adev)
807 {
808 	struct amdgpu_ring *sdma[AMDGPU_MAX_SDMA_INSTANCES];
809 	u32 rb_cntl, ib_cntl;
810 	int i, unset = 0;
811 
812 	for (i = 0; i < adev->sdma.num_instances; i++) {
813 		sdma[i] = &adev->sdma.instance[i].ring;
814 
815 		if ((adev->mman.buffer_funcs_ring == sdma[i]) && unset != 1) {
816 			amdgpu_ttm_set_buffer_funcs_status(adev, false);
817 			unset = 1;
818 		}
819 
820 		rb_cntl = RREG32_SDMA(i, mmSDMA0_GFX_RB_CNTL);
821 		rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_GFX_RB_CNTL, RB_ENABLE, 0);
822 		WREG32_SDMA(i, mmSDMA0_GFX_RB_CNTL, rb_cntl);
823 		ib_cntl = RREG32_SDMA(i, mmSDMA0_GFX_IB_CNTL);
824 		ib_cntl = REG_SET_FIELD(ib_cntl, SDMA0_GFX_IB_CNTL, IB_ENABLE, 0);
825 		WREG32_SDMA(i, mmSDMA0_GFX_IB_CNTL, ib_cntl);
826 
827 		sdma[i]->sched.ready = false;
828 	}
829 }
830 
831 /**
832  * sdma_v4_0_rlc_stop - stop the compute async dma engines
833  *
834  * @adev: amdgpu_device pointer
835  *
836  * Stop the compute async dma queues (VEGA10).
837  */
sdma_v4_0_rlc_stop(struct amdgpu_device * adev)838 static void sdma_v4_0_rlc_stop(struct amdgpu_device *adev)
839 {
840 	/* XXX todo */
841 }
842 
843 /**
844  * sdma_v4_0_page_stop - stop the page async dma engines
845  *
846  * @adev: amdgpu_device pointer
847  *
848  * Stop the page async dma ring buffers (VEGA10).
849  */
sdma_v4_0_page_stop(struct amdgpu_device * adev)850 static void sdma_v4_0_page_stop(struct amdgpu_device *adev)
851 {
852 	struct amdgpu_ring *sdma[AMDGPU_MAX_SDMA_INSTANCES];
853 	u32 rb_cntl, ib_cntl;
854 	int i;
855 	bool unset = false;
856 
857 	for (i = 0; i < adev->sdma.num_instances; i++) {
858 		sdma[i] = &adev->sdma.instance[i].page;
859 
860 		if ((adev->mman.buffer_funcs_ring == sdma[i]) &&
861 			(unset == false)) {
862 			amdgpu_ttm_set_buffer_funcs_status(adev, false);
863 			unset = true;
864 		}
865 
866 		rb_cntl = RREG32_SDMA(i, mmSDMA0_PAGE_RB_CNTL);
867 		rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_PAGE_RB_CNTL,
868 					RB_ENABLE, 0);
869 		WREG32_SDMA(i, mmSDMA0_PAGE_RB_CNTL, rb_cntl);
870 		ib_cntl = RREG32_SDMA(i, mmSDMA0_PAGE_IB_CNTL);
871 		ib_cntl = REG_SET_FIELD(ib_cntl, SDMA0_PAGE_IB_CNTL,
872 					IB_ENABLE, 0);
873 		WREG32_SDMA(i, mmSDMA0_PAGE_IB_CNTL, ib_cntl);
874 
875 		sdma[i]->sched.ready = false;
876 	}
877 }
878 
879 /**
880  * sdma_v_0_ctx_switch_enable - stop the async dma engines context switch
881  *
882  * @adev: amdgpu_device pointer
883  * @enable: enable/disable the DMA MEs context switch.
884  *
885  * Halt or unhalt the async dma engines context switch (VEGA10).
886  */
sdma_v4_0_ctx_switch_enable(struct amdgpu_device * adev,bool enable)887 static void sdma_v4_0_ctx_switch_enable(struct amdgpu_device *adev, bool enable)
888 {
889 	u32 f32_cntl, phase_quantum = 0;
890 	int i;
891 
892 	if (amdgpu_sdma_phase_quantum) {
893 		unsigned value = amdgpu_sdma_phase_quantum;
894 		unsigned unit = 0;
895 
896 		while (value > (SDMA0_PHASE0_QUANTUM__VALUE_MASK >>
897 				SDMA0_PHASE0_QUANTUM__VALUE__SHIFT)) {
898 			value = (value + 1) >> 1;
899 			unit++;
900 		}
901 		if (unit > (SDMA0_PHASE0_QUANTUM__UNIT_MASK >>
902 			    SDMA0_PHASE0_QUANTUM__UNIT__SHIFT)) {
903 			value = (SDMA0_PHASE0_QUANTUM__VALUE_MASK >>
904 				 SDMA0_PHASE0_QUANTUM__VALUE__SHIFT);
905 			unit = (SDMA0_PHASE0_QUANTUM__UNIT_MASK >>
906 				SDMA0_PHASE0_QUANTUM__UNIT__SHIFT);
907 			WARN_ONCE(1,
908 			"clamping sdma_phase_quantum to %uK clock cycles\n",
909 				  value << unit);
910 		}
911 		phase_quantum =
912 			value << SDMA0_PHASE0_QUANTUM__VALUE__SHIFT |
913 			unit  << SDMA0_PHASE0_QUANTUM__UNIT__SHIFT;
914 	}
915 
916 	for (i = 0; i < adev->sdma.num_instances; i++) {
917 		f32_cntl = RREG32_SDMA(i, mmSDMA0_CNTL);
918 		f32_cntl = REG_SET_FIELD(f32_cntl, SDMA0_CNTL,
919 				AUTO_CTXSW_ENABLE, enable ? 1 : 0);
920 		if (enable && amdgpu_sdma_phase_quantum) {
921 			WREG32_SDMA(i, mmSDMA0_PHASE0_QUANTUM, phase_quantum);
922 			WREG32_SDMA(i, mmSDMA0_PHASE1_QUANTUM, phase_quantum);
923 			WREG32_SDMA(i, mmSDMA0_PHASE2_QUANTUM, phase_quantum);
924 		}
925 		WREG32_SDMA(i, mmSDMA0_CNTL, f32_cntl);
926 	}
927 
928 }
929 
930 /**
931  * sdma_v4_0_enable - stop the async dma engines
932  *
933  * @adev: amdgpu_device pointer
934  * @enable: enable/disable the DMA MEs.
935  *
936  * Halt or unhalt the async dma engines (VEGA10).
937  */
sdma_v4_0_enable(struct amdgpu_device * adev,bool enable)938 static void sdma_v4_0_enable(struct amdgpu_device *adev, bool enable)
939 {
940 	u32 f32_cntl;
941 	int i;
942 
943 	if (enable == false) {
944 		sdma_v4_0_gfx_stop(adev);
945 		sdma_v4_0_rlc_stop(adev);
946 		if (adev->sdma.has_page_queue)
947 			sdma_v4_0_page_stop(adev);
948 	}
949 
950 	for (i = 0; i < adev->sdma.num_instances; i++) {
951 		f32_cntl = RREG32_SDMA(i, mmSDMA0_F32_CNTL);
952 		f32_cntl = REG_SET_FIELD(f32_cntl, SDMA0_F32_CNTL, HALT, enable ? 0 : 1);
953 		WREG32_SDMA(i, mmSDMA0_F32_CNTL, f32_cntl);
954 	}
955 }
956 
957 /**
958  * sdma_v4_0_rb_cntl - get parameters for rb_cntl
959  */
sdma_v4_0_rb_cntl(struct amdgpu_ring * ring,uint32_t rb_cntl)960 static uint32_t sdma_v4_0_rb_cntl(struct amdgpu_ring *ring, uint32_t rb_cntl)
961 {
962 	/* Set ring buffer size in dwords */
963 	uint32_t rb_bufsz = order_base_2(ring->ring_size / 4);
964 
965 	rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_GFX_RB_CNTL, RB_SIZE, rb_bufsz);
966 #ifdef __BIG_ENDIAN
967 	rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_GFX_RB_CNTL, RB_SWAP_ENABLE, 1);
968 	rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_GFX_RB_CNTL,
969 				RPTR_WRITEBACK_SWAP_ENABLE, 1);
970 #endif
971 	return rb_cntl;
972 }
973 
974 /**
975  * sdma_v4_0_gfx_resume - setup and start the async dma engines
976  *
977  * @adev: amdgpu_device pointer
978  * @i: instance to resume
979  *
980  * Set up the gfx DMA ring buffers and enable them (VEGA10).
981  * Returns 0 for success, error for failure.
982  */
sdma_v4_0_gfx_resume(struct amdgpu_device * adev,unsigned int i)983 static void sdma_v4_0_gfx_resume(struct amdgpu_device *adev, unsigned int i)
984 {
985 	struct amdgpu_ring *ring = &adev->sdma.instance[i].ring;
986 	u32 rb_cntl, ib_cntl, wptr_poll_cntl;
987 	u32 wb_offset;
988 	u32 doorbell;
989 	u32 doorbell_offset;
990 	u64 wptr_gpu_addr;
991 
992 	wb_offset = (ring->rptr_offs * 4);
993 
994 	rb_cntl = RREG32_SDMA(i, mmSDMA0_GFX_RB_CNTL);
995 	rb_cntl = sdma_v4_0_rb_cntl(ring, rb_cntl);
996 	WREG32_SDMA(i, mmSDMA0_GFX_RB_CNTL, rb_cntl);
997 
998 	/* Initialize the ring buffer's read and write pointers */
999 	WREG32_SDMA(i, mmSDMA0_GFX_RB_RPTR, 0);
1000 	WREG32_SDMA(i, mmSDMA0_GFX_RB_RPTR_HI, 0);
1001 	WREG32_SDMA(i, mmSDMA0_GFX_RB_WPTR, 0);
1002 	WREG32_SDMA(i, mmSDMA0_GFX_RB_WPTR_HI, 0);
1003 
1004 	/* set the wb address whether it's enabled or not */
1005 	WREG32_SDMA(i, mmSDMA0_GFX_RB_RPTR_ADDR_HI,
1006 	       upper_32_bits(adev->wb.gpu_addr + wb_offset) & 0xFFFFFFFF);
1007 	WREG32_SDMA(i, mmSDMA0_GFX_RB_RPTR_ADDR_LO,
1008 	       lower_32_bits(adev->wb.gpu_addr + wb_offset) & 0xFFFFFFFC);
1009 
1010 	rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_GFX_RB_CNTL,
1011 				RPTR_WRITEBACK_ENABLE, 1);
1012 
1013 	WREG32_SDMA(i, mmSDMA0_GFX_RB_BASE, ring->gpu_addr >> 8);
1014 	WREG32_SDMA(i, mmSDMA0_GFX_RB_BASE_HI, ring->gpu_addr >> 40);
1015 
1016 	ring->wptr = 0;
1017 
1018 	/* before programing wptr to a less value, need set minor_ptr_update first */
1019 	WREG32_SDMA(i, mmSDMA0_GFX_MINOR_PTR_UPDATE, 1);
1020 
1021 	doorbell = RREG32_SDMA(i, mmSDMA0_GFX_DOORBELL);
1022 	doorbell_offset = RREG32_SDMA(i, mmSDMA0_GFX_DOORBELL_OFFSET);
1023 
1024 	doorbell = REG_SET_FIELD(doorbell, SDMA0_GFX_DOORBELL, ENABLE,
1025 				 ring->use_doorbell);
1026 	doorbell_offset = REG_SET_FIELD(doorbell_offset,
1027 					SDMA0_GFX_DOORBELL_OFFSET,
1028 					OFFSET, ring->doorbell_index);
1029 	WREG32_SDMA(i, mmSDMA0_GFX_DOORBELL, doorbell);
1030 	WREG32_SDMA(i, mmSDMA0_GFX_DOORBELL_OFFSET, doorbell_offset);
1031 
1032 	sdma_v4_0_ring_set_wptr(ring);
1033 
1034 	/* set minor_ptr_update to 0 after wptr programed */
1035 	WREG32_SDMA(i, mmSDMA0_GFX_MINOR_PTR_UPDATE, 0);
1036 
1037 	/* setup the wptr shadow polling */
1038 	wptr_gpu_addr = adev->wb.gpu_addr + (ring->wptr_offs * 4);
1039 	WREG32_SDMA(i, mmSDMA0_GFX_RB_WPTR_POLL_ADDR_LO,
1040 		    lower_32_bits(wptr_gpu_addr));
1041 	WREG32_SDMA(i, mmSDMA0_GFX_RB_WPTR_POLL_ADDR_HI,
1042 		    upper_32_bits(wptr_gpu_addr));
1043 	wptr_poll_cntl = RREG32_SDMA(i, mmSDMA0_GFX_RB_WPTR_POLL_CNTL);
1044 	wptr_poll_cntl = REG_SET_FIELD(wptr_poll_cntl,
1045 				       SDMA0_GFX_RB_WPTR_POLL_CNTL,
1046 				       F32_POLL_ENABLE, amdgpu_sriov_vf(adev)? 1 : 0);
1047 	WREG32_SDMA(i, mmSDMA0_GFX_RB_WPTR_POLL_CNTL, wptr_poll_cntl);
1048 
1049 	/* enable DMA RB */
1050 	rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_GFX_RB_CNTL, RB_ENABLE, 1);
1051 	WREG32_SDMA(i, mmSDMA0_GFX_RB_CNTL, rb_cntl);
1052 
1053 	ib_cntl = RREG32_SDMA(i, mmSDMA0_GFX_IB_CNTL);
1054 	ib_cntl = REG_SET_FIELD(ib_cntl, SDMA0_GFX_IB_CNTL, IB_ENABLE, 1);
1055 #ifdef __BIG_ENDIAN
1056 	ib_cntl = REG_SET_FIELD(ib_cntl, SDMA0_GFX_IB_CNTL, IB_SWAP_ENABLE, 1);
1057 #endif
1058 	/* enable DMA IBs */
1059 	WREG32_SDMA(i, mmSDMA0_GFX_IB_CNTL, ib_cntl);
1060 
1061 	ring->sched.ready = true;
1062 }
1063 
1064 /**
1065  * sdma_v4_0_page_resume - setup and start the async dma engines
1066  *
1067  * @adev: amdgpu_device pointer
1068  * @i: instance to resume
1069  *
1070  * Set up the page DMA ring buffers and enable them (VEGA10).
1071  * Returns 0 for success, error for failure.
1072  */
sdma_v4_0_page_resume(struct amdgpu_device * adev,unsigned int i)1073 static void sdma_v4_0_page_resume(struct amdgpu_device *adev, unsigned int i)
1074 {
1075 	struct amdgpu_ring *ring = &adev->sdma.instance[i].page;
1076 	u32 rb_cntl, ib_cntl, wptr_poll_cntl;
1077 	u32 wb_offset;
1078 	u32 doorbell;
1079 	u32 doorbell_offset;
1080 	u64 wptr_gpu_addr;
1081 
1082 	wb_offset = (ring->rptr_offs * 4);
1083 
1084 	rb_cntl = RREG32_SDMA(i, mmSDMA0_PAGE_RB_CNTL);
1085 	rb_cntl = sdma_v4_0_rb_cntl(ring, rb_cntl);
1086 	WREG32_SDMA(i, mmSDMA0_PAGE_RB_CNTL, rb_cntl);
1087 
1088 	/* Initialize the ring buffer's read and write pointers */
1089 	WREG32_SDMA(i, mmSDMA0_PAGE_RB_RPTR, 0);
1090 	WREG32_SDMA(i, mmSDMA0_PAGE_RB_RPTR_HI, 0);
1091 	WREG32_SDMA(i, mmSDMA0_PAGE_RB_WPTR, 0);
1092 	WREG32_SDMA(i, mmSDMA0_PAGE_RB_WPTR_HI, 0);
1093 
1094 	/* set the wb address whether it's enabled or not */
1095 	WREG32_SDMA(i, mmSDMA0_PAGE_RB_RPTR_ADDR_HI,
1096 	       upper_32_bits(adev->wb.gpu_addr + wb_offset) & 0xFFFFFFFF);
1097 	WREG32_SDMA(i, mmSDMA0_PAGE_RB_RPTR_ADDR_LO,
1098 	       lower_32_bits(adev->wb.gpu_addr + wb_offset) & 0xFFFFFFFC);
1099 
1100 	rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_PAGE_RB_CNTL,
1101 				RPTR_WRITEBACK_ENABLE, 1);
1102 
1103 	WREG32_SDMA(i, mmSDMA0_PAGE_RB_BASE, ring->gpu_addr >> 8);
1104 	WREG32_SDMA(i, mmSDMA0_PAGE_RB_BASE_HI, ring->gpu_addr >> 40);
1105 
1106 	ring->wptr = 0;
1107 
1108 	/* before programing wptr to a less value, need set minor_ptr_update first */
1109 	WREG32_SDMA(i, mmSDMA0_PAGE_MINOR_PTR_UPDATE, 1);
1110 
1111 	doorbell = RREG32_SDMA(i, mmSDMA0_PAGE_DOORBELL);
1112 	doorbell_offset = RREG32_SDMA(i, mmSDMA0_PAGE_DOORBELL_OFFSET);
1113 
1114 	doorbell = REG_SET_FIELD(doorbell, SDMA0_PAGE_DOORBELL, ENABLE,
1115 				 ring->use_doorbell);
1116 	doorbell_offset = REG_SET_FIELD(doorbell_offset,
1117 					SDMA0_PAGE_DOORBELL_OFFSET,
1118 					OFFSET, ring->doorbell_index);
1119 	WREG32_SDMA(i, mmSDMA0_PAGE_DOORBELL, doorbell);
1120 	WREG32_SDMA(i, mmSDMA0_PAGE_DOORBELL_OFFSET, doorbell_offset);
1121 
1122 	/* paging queue doorbell range is setup at sdma_v4_0_gfx_resume */
1123 	sdma_v4_0_page_ring_set_wptr(ring);
1124 
1125 	/* set minor_ptr_update to 0 after wptr programed */
1126 	WREG32_SDMA(i, mmSDMA0_PAGE_MINOR_PTR_UPDATE, 0);
1127 
1128 	/* setup the wptr shadow polling */
1129 	wptr_gpu_addr = adev->wb.gpu_addr + (ring->wptr_offs * 4);
1130 	WREG32_SDMA(i, mmSDMA0_PAGE_RB_WPTR_POLL_ADDR_LO,
1131 		    lower_32_bits(wptr_gpu_addr));
1132 	WREG32_SDMA(i, mmSDMA0_PAGE_RB_WPTR_POLL_ADDR_HI,
1133 		    upper_32_bits(wptr_gpu_addr));
1134 	wptr_poll_cntl = RREG32_SDMA(i, mmSDMA0_PAGE_RB_WPTR_POLL_CNTL);
1135 	wptr_poll_cntl = REG_SET_FIELD(wptr_poll_cntl,
1136 				       SDMA0_PAGE_RB_WPTR_POLL_CNTL,
1137 				       F32_POLL_ENABLE, amdgpu_sriov_vf(adev)? 1 : 0);
1138 	WREG32_SDMA(i, mmSDMA0_PAGE_RB_WPTR_POLL_CNTL, wptr_poll_cntl);
1139 
1140 	/* enable DMA RB */
1141 	rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_PAGE_RB_CNTL, RB_ENABLE, 1);
1142 	WREG32_SDMA(i, mmSDMA0_PAGE_RB_CNTL, rb_cntl);
1143 
1144 	ib_cntl = RREG32_SDMA(i, mmSDMA0_PAGE_IB_CNTL);
1145 	ib_cntl = REG_SET_FIELD(ib_cntl, SDMA0_PAGE_IB_CNTL, IB_ENABLE, 1);
1146 #ifdef __BIG_ENDIAN
1147 	ib_cntl = REG_SET_FIELD(ib_cntl, SDMA0_PAGE_IB_CNTL, IB_SWAP_ENABLE, 1);
1148 #endif
1149 	/* enable DMA IBs */
1150 	WREG32_SDMA(i, mmSDMA0_PAGE_IB_CNTL, ib_cntl);
1151 
1152 	ring->sched.ready = true;
1153 }
1154 
1155 static void
sdma_v4_1_update_power_gating(struct amdgpu_device * adev,bool enable)1156 sdma_v4_1_update_power_gating(struct amdgpu_device *adev, bool enable)
1157 {
1158 	uint32_t def, data;
1159 
1160 	if (enable && (adev->pg_flags & AMD_PG_SUPPORT_SDMA)) {
1161 		/* enable idle interrupt */
1162 		def = data = RREG32(SOC15_REG_OFFSET(SDMA0, 0, mmSDMA0_CNTL));
1163 		data |= SDMA0_CNTL__CTXEMPTY_INT_ENABLE_MASK;
1164 
1165 		if (data != def)
1166 			WREG32(SOC15_REG_OFFSET(SDMA0, 0, mmSDMA0_CNTL), data);
1167 	} else {
1168 		/* disable idle interrupt */
1169 		def = data = RREG32(SOC15_REG_OFFSET(SDMA0, 0, mmSDMA0_CNTL));
1170 		data &= ~SDMA0_CNTL__CTXEMPTY_INT_ENABLE_MASK;
1171 		if (data != def)
1172 			WREG32(SOC15_REG_OFFSET(SDMA0, 0, mmSDMA0_CNTL), data);
1173 	}
1174 }
1175 
sdma_v4_1_init_power_gating(struct amdgpu_device * adev)1176 static void sdma_v4_1_init_power_gating(struct amdgpu_device *adev)
1177 {
1178 	uint32_t def, data;
1179 
1180 	/* Enable HW based PG. */
1181 	def = data = RREG32(SOC15_REG_OFFSET(SDMA0, 0, mmSDMA0_POWER_CNTL));
1182 	data |= SDMA0_POWER_CNTL__PG_CNTL_ENABLE_MASK;
1183 	if (data != def)
1184 		WREG32(SOC15_REG_OFFSET(SDMA0, 0, mmSDMA0_POWER_CNTL), data);
1185 
1186 	/* enable interrupt */
1187 	def = data = RREG32(SOC15_REG_OFFSET(SDMA0, 0, mmSDMA0_CNTL));
1188 	data |= SDMA0_CNTL__CTXEMPTY_INT_ENABLE_MASK;
1189 	if (data != def)
1190 		WREG32(SOC15_REG_OFFSET(SDMA0, 0, mmSDMA0_CNTL), data);
1191 
1192 	/* Configure hold time to filter in-valid power on/off request. Use default right now */
1193 	def = data = RREG32(SOC15_REG_OFFSET(SDMA0, 0, mmSDMA0_POWER_CNTL));
1194 	data &= ~SDMA0_POWER_CNTL__ON_OFF_CONDITION_HOLD_TIME_MASK;
1195 	data |= (mmSDMA0_POWER_CNTL_DEFAULT & SDMA0_POWER_CNTL__ON_OFF_CONDITION_HOLD_TIME_MASK);
1196 	/* Configure switch time for hysteresis purpose. Use default right now */
1197 	data &= ~SDMA0_POWER_CNTL__ON_OFF_STATUS_DURATION_TIME_MASK;
1198 	data |= (mmSDMA0_POWER_CNTL_DEFAULT & SDMA0_POWER_CNTL__ON_OFF_STATUS_DURATION_TIME_MASK);
1199 	if(data != def)
1200 		WREG32(SOC15_REG_OFFSET(SDMA0, 0, mmSDMA0_POWER_CNTL), data);
1201 }
1202 
sdma_v4_0_init_pg(struct amdgpu_device * adev)1203 static void sdma_v4_0_init_pg(struct amdgpu_device *adev)
1204 {
1205 	if (!(adev->pg_flags & AMD_PG_SUPPORT_SDMA))
1206 		return;
1207 
1208 	switch (adev->asic_type) {
1209 	case CHIP_RAVEN:
1210 	case CHIP_RENOIR:
1211 		sdma_v4_1_init_power_gating(adev);
1212 		sdma_v4_1_update_power_gating(adev, true);
1213 		break;
1214 	default:
1215 		break;
1216 	}
1217 }
1218 
1219 /**
1220  * sdma_v4_0_rlc_resume - setup and start the async dma engines
1221  *
1222  * @adev: amdgpu_device pointer
1223  *
1224  * Set up the compute DMA queues and enable them (VEGA10).
1225  * Returns 0 for success, error for failure.
1226  */
sdma_v4_0_rlc_resume(struct amdgpu_device * adev)1227 static int sdma_v4_0_rlc_resume(struct amdgpu_device *adev)
1228 {
1229 	sdma_v4_0_init_pg(adev);
1230 
1231 	return 0;
1232 }
1233 
1234 /**
1235  * sdma_v4_0_load_microcode - load the sDMA ME ucode
1236  *
1237  * @adev: amdgpu_device pointer
1238  *
1239  * Loads the sDMA0/1 ucode.
1240  * Returns 0 for success, -EINVAL if the ucode is not available.
1241  */
sdma_v4_0_load_microcode(struct amdgpu_device * adev)1242 static int sdma_v4_0_load_microcode(struct amdgpu_device *adev)
1243 {
1244 	const struct sdma_firmware_header_v1_0 *hdr;
1245 	const __le32 *fw_data;
1246 	u32 fw_size;
1247 	int i, j;
1248 
1249 	/* halt the MEs */
1250 	sdma_v4_0_enable(adev, false);
1251 
1252 	for (i = 0; i < adev->sdma.num_instances; i++) {
1253 		if (!adev->sdma.instance[i].fw)
1254 			return -EINVAL;
1255 
1256 		hdr = (const struct sdma_firmware_header_v1_0 *)adev->sdma.instance[i].fw->data;
1257 		amdgpu_ucode_print_sdma_hdr(&hdr->header);
1258 		fw_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4;
1259 
1260 		fw_data = (const __le32 *)
1261 			(adev->sdma.instance[i].fw->data +
1262 				le32_to_cpu(hdr->header.ucode_array_offset_bytes));
1263 
1264 		WREG32_SDMA(i, mmSDMA0_UCODE_ADDR, 0);
1265 
1266 		for (j = 0; j < fw_size; j++)
1267 			WREG32_SDMA(i, mmSDMA0_UCODE_DATA,
1268 				    le32_to_cpup(fw_data++));
1269 
1270 		WREG32_SDMA(i, mmSDMA0_UCODE_ADDR,
1271 			    adev->sdma.instance[i].fw_version);
1272 	}
1273 
1274 	return 0;
1275 }
1276 
1277 /**
1278  * sdma_v4_0_start - setup and start the async dma engines
1279  *
1280  * @adev: amdgpu_device pointer
1281  *
1282  * Set up the DMA engines and enable them (VEGA10).
1283  * Returns 0 for success, error for failure.
1284  */
sdma_v4_0_start(struct amdgpu_device * adev)1285 static int sdma_v4_0_start(struct amdgpu_device *adev)
1286 {
1287 	struct amdgpu_ring *ring;
1288 	int i, r = 0;
1289 
1290 	if (amdgpu_sriov_vf(adev)) {
1291 		sdma_v4_0_ctx_switch_enable(adev, false);
1292 		sdma_v4_0_enable(adev, false);
1293 	} else {
1294 
1295 		if (adev->firmware.load_type != AMDGPU_FW_LOAD_PSP) {
1296 			r = sdma_v4_0_load_microcode(adev);
1297 			if (r)
1298 				return r;
1299 		}
1300 
1301 		/* unhalt the MEs */
1302 		sdma_v4_0_enable(adev, true);
1303 		/* enable sdma ring preemption */
1304 		sdma_v4_0_ctx_switch_enable(adev, true);
1305 	}
1306 
1307 	/* start the gfx rings and rlc compute queues */
1308 	for (i = 0; i < adev->sdma.num_instances; i++) {
1309 		uint32_t temp;
1310 
1311 		WREG32_SDMA(i, mmSDMA0_SEM_WAIT_FAIL_TIMER_CNTL, 0);
1312 		sdma_v4_0_gfx_resume(adev, i);
1313 		if (adev->sdma.has_page_queue)
1314 			sdma_v4_0_page_resume(adev, i);
1315 
1316 		/* set utc l1 enable flag always to 1 */
1317 		temp = RREG32_SDMA(i, mmSDMA0_CNTL);
1318 		temp = REG_SET_FIELD(temp, SDMA0_CNTL, UTC_L1_ENABLE, 1);
1319 		WREG32_SDMA(i, mmSDMA0_CNTL, temp);
1320 
1321 		if (!amdgpu_sriov_vf(adev)) {
1322 			/* unhalt engine */
1323 			temp = RREG32_SDMA(i, mmSDMA0_F32_CNTL);
1324 			temp = REG_SET_FIELD(temp, SDMA0_F32_CNTL, HALT, 0);
1325 			WREG32_SDMA(i, mmSDMA0_F32_CNTL, temp);
1326 		}
1327 	}
1328 
1329 	if (amdgpu_sriov_vf(adev)) {
1330 		sdma_v4_0_ctx_switch_enable(adev, true);
1331 		sdma_v4_0_enable(adev, true);
1332 	} else {
1333 		r = sdma_v4_0_rlc_resume(adev);
1334 		if (r)
1335 			return r;
1336 	}
1337 
1338 	for (i = 0; i < adev->sdma.num_instances; i++) {
1339 		ring = &adev->sdma.instance[i].ring;
1340 
1341 		r = amdgpu_ring_test_helper(ring);
1342 		if (r)
1343 			return r;
1344 
1345 		if (adev->sdma.has_page_queue) {
1346 			struct amdgpu_ring *page = &adev->sdma.instance[i].page;
1347 
1348 			r = amdgpu_ring_test_helper(page);
1349 			if (r)
1350 				return r;
1351 
1352 			if (adev->mman.buffer_funcs_ring == page)
1353 				amdgpu_ttm_set_buffer_funcs_status(adev, true);
1354 		}
1355 
1356 		if (adev->mman.buffer_funcs_ring == ring)
1357 			amdgpu_ttm_set_buffer_funcs_status(adev, true);
1358 	}
1359 
1360 	return r;
1361 }
1362 
1363 /**
1364  * sdma_v4_0_ring_test_ring - simple async dma engine test
1365  *
1366  * @ring: amdgpu_ring structure holding ring information
1367  *
1368  * Test the DMA engine by writing using it to write an
1369  * value to memory. (VEGA10).
1370  * Returns 0 for success, error for failure.
1371  */
sdma_v4_0_ring_test_ring(struct amdgpu_ring * ring)1372 static int sdma_v4_0_ring_test_ring(struct amdgpu_ring *ring)
1373 {
1374 	struct amdgpu_device *adev = ring->adev;
1375 	unsigned i;
1376 	unsigned index;
1377 	int r;
1378 	u32 tmp;
1379 	u64 gpu_addr;
1380 
1381 	r = amdgpu_device_wb_get(adev, &index);
1382 	if (r)
1383 		return r;
1384 
1385 	gpu_addr = adev->wb.gpu_addr + (index * 4);
1386 	tmp = 0xCAFEDEAD;
1387 	adev->wb.wb[index] = cpu_to_le32(tmp);
1388 
1389 	r = amdgpu_ring_alloc(ring, 5);
1390 	if (r)
1391 		goto error_free_wb;
1392 
1393 	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_WRITE) |
1394 			  SDMA_PKT_HEADER_SUB_OP(SDMA_SUBOP_WRITE_LINEAR));
1395 	amdgpu_ring_write(ring, lower_32_bits(gpu_addr));
1396 	amdgpu_ring_write(ring, upper_32_bits(gpu_addr));
1397 	amdgpu_ring_write(ring, SDMA_PKT_WRITE_UNTILED_DW_3_COUNT(0));
1398 	amdgpu_ring_write(ring, 0xDEADBEEF);
1399 	amdgpu_ring_commit(ring);
1400 
1401 	for (i = 0; i < adev->usec_timeout; i++) {
1402 		tmp = le32_to_cpu(adev->wb.wb[index]);
1403 		if (tmp == 0xDEADBEEF)
1404 			break;
1405 		udelay(1);
1406 	}
1407 
1408 	if (i >= adev->usec_timeout)
1409 		r = -ETIMEDOUT;
1410 
1411 error_free_wb:
1412 	amdgpu_device_wb_free(adev, index);
1413 	return r;
1414 }
1415 
1416 /**
1417  * sdma_v4_0_ring_test_ib - test an IB on the DMA engine
1418  *
1419  * @ring: amdgpu_ring structure holding ring information
1420  *
1421  * Test a simple IB in the DMA ring (VEGA10).
1422  * Returns 0 on success, error on failure.
1423  */
sdma_v4_0_ring_test_ib(struct amdgpu_ring * ring,long timeout)1424 static int sdma_v4_0_ring_test_ib(struct amdgpu_ring *ring, long timeout)
1425 {
1426 	struct amdgpu_device *adev = ring->adev;
1427 	struct amdgpu_ib ib;
1428 	struct dma_fence *f = NULL;
1429 	unsigned index;
1430 	long r;
1431 	u32 tmp = 0;
1432 	u64 gpu_addr;
1433 
1434 	r = amdgpu_device_wb_get(adev, &index);
1435 	if (r)
1436 		return r;
1437 
1438 	gpu_addr = adev->wb.gpu_addr + (index * 4);
1439 	tmp = 0xCAFEDEAD;
1440 	adev->wb.wb[index] = cpu_to_le32(tmp);
1441 	memset(&ib, 0, sizeof(ib));
1442 	r = amdgpu_ib_get(adev, NULL, 256, &ib);
1443 	if (r)
1444 		goto err0;
1445 
1446 	ib.ptr[0] = SDMA_PKT_HEADER_OP(SDMA_OP_WRITE) |
1447 		SDMA_PKT_HEADER_SUB_OP(SDMA_SUBOP_WRITE_LINEAR);
1448 	ib.ptr[1] = lower_32_bits(gpu_addr);
1449 	ib.ptr[2] = upper_32_bits(gpu_addr);
1450 	ib.ptr[3] = SDMA_PKT_WRITE_UNTILED_DW_3_COUNT(0);
1451 	ib.ptr[4] = 0xDEADBEEF;
1452 	ib.ptr[5] = SDMA_PKT_NOP_HEADER_OP(SDMA_OP_NOP);
1453 	ib.ptr[6] = SDMA_PKT_NOP_HEADER_OP(SDMA_OP_NOP);
1454 	ib.ptr[7] = SDMA_PKT_NOP_HEADER_OP(SDMA_OP_NOP);
1455 	ib.length_dw = 8;
1456 
1457 	r = amdgpu_ib_schedule(ring, 1, &ib, NULL, &f);
1458 	if (r)
1459 		goto err1;
1460 
1461 	r = dma_fence_wait_timeout(f, false, timeout);
1462 	if (r == 0) {
1463 		r = -ETIMEDOUT;
1464 		goto err1;
1465 	} else if (r < 0) {
1466 		goto err1;
1467 	}
1468 	tmp = le32_to_cpu(adev->wb.wb[index]);
1469 	if (tmp == 0xDEADBEEF)
1470 		r = 0;
1471 	else
1472 		r = -EINVAL;
1473 
1474 err1:
1475 	amdgpu_ib_free(adev, &ib, NULL);
1476 	dma_fence_put(f);
1477 err0:
1478 	amdgpu_device_wb_free(adev, index);
1479 	return r;
1480 }
1481 
1482 
1483 /**
1484  * sdma_v4_0_vm_copy_pte - update PTEs by copying them from the GART
1485  *
1486  * @ib: indirect buffer to fill with commands
1487  * @pe: addr of the page entry
1488  * @src: src addr to copy from
1489  * @count: number of page entries to update
1490  *
1491  * Update PTEs by copying them from the GART using sDMA (VEGA10).
1492  */
sdma_v4_0_vm_copy_pte(struct amdgpu_ib * ib,uint64_t pe,uint64_t src,unsigned count)1493 static void sdma_v4_0_vm_copy_pte(struct amdgpu_ib *ib,
1494 				  uint64_t pe, uint64_t src,
1495 				  unsigned count)
1496 {
1497 	unsigned bytes = count * 8;
1498 
1499 	ib->ptr[ib->length_dw++] = SDMA_PKT_HEADER_OP(SDMA_OP_COPY) |
1500 		SDMA_PKT_HEADER_SUB_OP(SDMA_SUBOP_COPY_LINEAR);
1501 	ib->ptr[ib->length_dw++] = bytes - 1;
1502 	ib->ptr[ib->length_dw++] = 0; /* src/dst endian swap */
1503 	ib->ptr[ib->length_dw++] = lower_32_bits(src);
1504 	ib->ptr[ib->length_dw++] = upper_32_bits(src);
1505 	ib->ptr[ib->length_dw++] = lower_32_bits(pe);
1506 	ib->ptr[ib->length_dw++] = upper_32_bits(pe);
1507 
1508 }
1509 
1510 /**
1511  * sdma_v4_0_vm_write_pte - update PTEs by writing them manually
1512  *
1513  * @ib: indirect buffer to fill with commands
1514  * @pe: addr of the page entry
1515  * @addr: dst addr to write into pe
1516  * @count: number of page entries to update
1517  * @incr: increase next addr by incr bytes
1518  * @flags: access flags
1519  *
1520  * Update PTEs by writing them manually using sDMA (VEGA10).
1521  */
sdma_v4_0_vm_write_pte(struct amdgpu_ib * ib,uint64_t pe,uint64_t value,unsigned count,uint32_t incr)1522 static void sdma_v4_0_vm_write_pte(struct amdgpu_ib *ib, uint64_t pe,
1523 				   uint64_t value, unsigned count,
1524 				   uint32_t incr)
1525 {
1526 	unsigned ndw = count * 2;
1527 
1528 	ib->ptr[ib->length_dw++] = SDMA_PKT_HEADER_OP(SDMA_OP_WRITE) |
1529 		SDMA_PKT_HEADER_SUB_OP(SDMA_SUBOP_WRITE_LINEAR);
1530 	ib->ptr[ib->length_dw++] = lower_32_bits(pe);
1531 	ib->ptr[ib->length_dw++] = upper_32_bits(pe);
1532 	ib->ptr[ib->length_dw++] = ndw - 1;
1533 	for (; ndw > 0; ndw -= 2) {
1534 		ib->ptr[ib->length_dw++] = lower_32_bits(value);
1535 		ib->ptr[ib->length_dw++] = upper_32_bits(value);
1536 		value += incr;
1537 	}
1538 }
1539 
1540 /**
1541  * sdma_v4_0_vm_set_pte_pde - update the page tables using sDMA
1542  *
1543  * @ib: indirect buffer to fill with commands
1544  * @pe: addr of the page entry
1545  * @addr: dst addr to write into pe
1546  * @count: number of page entries to update
1547  * @incr: increase next addr by incr bytes
1548  * @flags: access flags
1549  *
1550  * Update the page tables using sDMA (VEGA10).
1551  */
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)1552 static void sdma_v4_0_vm_set_pte_pde(struct amdgpu_ib *ib,
1553 				     uint64_t pe,
1554 				     uint64_t addr, unsigned count,
1555 				     uint32_t incr, uint64_t flags)
1556 {
1557 	/* for physically contiguous pages (vram) */
1558 	ib->ptr[ib->length_dw++] = SDMA_PKT_HEADER_OP(SDMA_OP_PTEPDE);
1559 	ib->ptr[ib->length_dw++] = lower_32_bits(pe); /* dst addr */
1560 	ib->ptr[ib->length_dw++] = upper_32_bits(pe);
1561 	ib->ptr[ib->length_dw++] = lower_32_bits(flags); /* mask */
1562 	ib->ptr[ib->length_dw++] = upper_32_bits(flags);
1563 	ib->ptr[ib->length_dw++] = lower_32_bits(addr); /* value */
1564 	ib->ptr[ib->length_dw++] = upper_32_bits(addr);
1565 	ib->ptr[ib->length_dw++] = incr; /* increment size */
1566 	ib->ptr[ib->length_dw++] = 0;
1567 	ib->ptr[ib->length_dw++] = count - 1; /* number of entries */
1568 }
1569 
1570 /**
1571  * sdma_v4_0_ring_pad_ib - pad the IB to the required number of dw
1572  *
1573  * @ib: indirect buffer to fill with padding
1574  *
1575  */
sdma_v4_0_ring_pad_ib(struct amdgpu_ring * ring,struct amdgpu_ib * ib)1576 static void sdma_v4_0_ring_pad_ib(struct amdgpu_ring *ring, struct amdgpu_ib *ib)
1577 {
1578 	struct amdgpu_sdma_instance *sdma = amdgpu_sdma_get_instance_from_ring(ring);
1579 	u32 pad_count;
1580 	int i;
1581 
1582 	pad_count = (8 - (ib->length_dw & 0x7)) % 8;
1583 	for (i = 0; i < pad_count; i++)
1584 		if (sdma && sdma->burst_nop && (i == 0))
1585 			ib->ptr[ib->length_dw++] =
1586 				SDMA_PKT_HEADER_OP(SDMA_OP_NOP) |
1587 				SDMA_PKT_NOP_HEADER_COUNT(pad_count - 1);
1588 		else
1589 			ib->ptr[ib->length_dw++] =
1590 				SDMA_PKT_HEADER_OP(SDMA_OP_NOP);
1591 }
1592 
1593 
1594 /**
1595  * sdma_v4_0_ring_emit_pipeline_sync - sync the pipeline
1596  *
1597  * @ring: amdgpu_ring pointer
1598  *
1599  * Make sure all previous operations are completed (CIK).
1600  */
sdma_v4_0_ring_emit_pipeline_sync(struct amdgpu_ring * ring)1601 static void sdma_v4_0_ring_emit_pipeline_sync(struct amdgpu_ring *ring)
1602 {
1603 	uint32_t seq = ring->fence_drv.sync_seq;
1604 	uint64_t addr = ring->fence_drv.gpu_addr;
1605 
1606 	/* wait for idle */
1607 	sdma_v4_0_wait_reg_mem(ring, 1, 0,
1608 			       addr & 0xfffffffc,
1609 			       upper_32_bits(addr) & 0xffffffff,
1610 			       seq, 0xffffffff, 4);
1611 }
1612 
1613 
1614 /**
1615  * sdma_v4_0_ring_emit_vm_flush - vm flush using sDMA
1616  *
1617  * @ring: amdgpu_ring pointer
1618  * @vm: amdgpu_vm pointer
1619  *
1620  * Update the page table base and flush the VM TLB
1621  * using sDMA (VEGA10).
1622  */
sdma_v4_0_ring_emit_vm_flush(struct amdgpu_ring * ring,unsigned vmid,uint64_t pd_addr)1623 static void sdma_v4_0_ring_emit_vm_flush(struct amdgpu_ring *ring,
1624 					 unsigned vmid, uint64_t pd_addr)
1625 {
1626 	amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr);
1627 }
1628 
sdma_v4_0_ring_emit_wreg(struct amdgpu_ring * ring,uint32_t reg,uint32_t val)1629 static void sdma_v4_0_ring_emit_wreg(struct amdgpu_ring *ring,
1630 				     uint32_t reg, uint32_t val)
1631 {
1632 	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_SRBM_WRITE) |
1633 			  SDMA_PKT_SRBM_WRITE_HEADER_BYTE_EN(0xf));
1634 	amdgpu_ring_write(ring, reg);
1635 	amdgpu_ring_write(ring, val);
1636 }
1637 
sdma_v4_0_ring_emit_reg_wait(struct amdgpu_ring * ring,uint32_t reg,uint32_t val,uint32_t mask)1638 static void sdma_v4_0_ring_emit_reg_wait(struct amdgpu_ring *ring, uint32_t reg,
1639 					 uint32_t val, uint32_t mask)
1640 {
1641 	sdma_v4_0_wait_reg_mem(ring, 0, 0, reg, 0, val, mask, 10);
1642 }
1643 
sdma_v4_0_fw_support_paging_queue(struct amdgpu_device * adev)1644 static bool sdma_v4_0_fw_support_paging_queue(struct amdgpu_device *adev)
1645 {
1646 	uint fw_version = adev->sdma.instance[0].fw_version;
1647 
1648 	switch (adev->asic_type) {
1649 	case CHIP_VEGA10:
1650 		return fw_version >= 430;
1651 	case CHIP_VEGA12:
1652 		/*return fw_version >= 31;*/
1653 		return false;
1654 	case CHIP_VEGA20:
1655 		return fw_version >= 123;
1656 	default:
1657 		return false;
1658 	}
1659 }
1660 
sdma_v4_0_early_init(void * handle)1661 static int sdma_v4_0_early_init(void *handle)
1662 {
1663 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1664 	int r;
1665 
1666 	if (adev->asic_type == CHIP_RAVEN || adev->asic_type == CHIP_RENOIR)
1667 		adev->sdma.num_instances = 1;
1668 	else if (adev->asic_type == CHIP_ARCTURUS)
1669 		adev->sdma.num_instances = 8;
1670 	else
1671 		adev->sdma.num_instances = 2;
1672 
1673 	r = sdma_v4_0_init_microcode(adev);
1674 	if (r) {
1675 		DRM_ERROR("Failed to load sdma firmware!\n");
1676 		return r;
1677 	}
1678 
1679 	/* TODO: Page queue breaks driver reload under SRIOV */
1680 	if ((adev->asic_type == CHIP_VEGA10) && amdgpu_sriov_vf((adev)))
1681 		adev->sdma.has_page_queue = false;
1682 	else if (sdma_v4_0_fw_support_paging_queue(adev))
1683 		adev->sdma.has_page_queue = true;
1684 
1685 	sdma_v4_0_set_ring_funcs(adev);
1686 	sdma_v4_0_set_buffer_funcs(adev);
1687 	sdma_v4_0_set_vm_pte_funcs(adev);
1688 	sdma_v4_0_set_irq_funcs(adev);
1689 
1690 	return 0;
1691 }
1692 
1693 static int sdma_v4_0_process_ras_data_cb(struct amdgpu_device *adev,
1694 		struct ras_err_data *err_data,
1695 		struct amdgpu_iv_entry *entry);
1696 
sdma_v4_0_late_init(void * handle)1697 static int sdma_v4_0_late_init(void *handle)
1698 {
1699 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1700 	struct ras_common_if **ras_if = &adev->sdma.ras_if;
1701 	struct ras_ih_if ih_info = {
1702 		.cb = sdma_v4_0_process_ras_data_cb,
1703 	};
1704 	struct ras_fs_if fs_info = {
1705 		.sysfs_name = "sdma_err_count",
1706 		.debugfs_name = "sdma_err_inject",
1707 	};
1708 	struct ras_common_if ras_block = {
1709 		.block = AMDGPU_RAS_BLOCK__SDMA,
1710 		.type = AMDGPU_RAS_ERROR__MULTI_UNCORRECTABLE,
1711 		.sub_block_index = 0,
1712 		.name = "sdma",
1713 	};
1714 	int r, i;
1715 
1716 	if (!amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__SDMA)) {
1717 		amdgpu_ras_feature_enable_on_boot(adev, &ras_block, 0);
1718 		return 0;
1719 	}
1720 
1721 	/* handle resume path. */
1722 	if (*ras_if) {
1723 		/* resend ras TA enable cmd during resume.
1724 		 * prepare to handle failure.
1725 		 */
1726 		ih_info.head = **ras_if;
1727 		r = amdgpu_ras_feature_enable_on_boot(adev, *ras_if, 1);
1728 		if (r) {
1729 			if (r == -EAGAIN) {
1730 				/* request a gpu reset. will run again. */
1731 				amdgpu_ras_request_reset_on_boot(adev,
1732 						AMDGPU_RAS_BLOCK__SDMA);
1733 				return 0;
1734 			}
1735 			/* fail to enable ras, cleanup all. */
1736 			goto irq;
1737 		}
1738 		/* enable successfully. continue. */
1739 		goto resume;
1740 	}
1741 
1742 	*ras_if = kmalloc(sizeof(**ras_if), GFP_KERNEL);
1743 	if (!*ras_if)
1744 		return -ENOMEM;
1745 
1746 	**ras_if = ras_block;
1747 
1748 	r = amdgpu_ras_feature_enable_on_boot(adev, *ras_if, 1);
1749 	if (r) {
1750 		if (r == -EAGAIN) {
1751 			amdgpu_ras_request_reset_on_boot(adev,
1752 					AMDGPU_RAS_BLOCK__SDMA);
1753 			r = 0;
1754 		}
1755 		goto feature;
1756 	}
1757 
1758 	ih_info.head = **ras_if;
1759 	fs_info.head = **ras_if;
1760 
1761 	r = amdgpu_ras_interrupt_add_handler(adev, &ih_info);
1762 	if (r)
1763 		goto interrupt;
1764 
1765 	amdgpu_ras_debugfs_create(adev, &fs_info);
1766 
1767 	r = amdgpu_ras_sysfs_create(adev, &fs_info);
1768 	if (r)
1769 		goto sysfs;
1770 resume:
1771 	for (i = 0; i < adev->sdma.num_instances; i++) {
1772 		r = amdgpu_irq_get(adev, &adev->sdma.ecc_irq,
1773 				   AMDGPU_SDMA_IRQ_INSTANCE0 + i);
1774 		if (r)
1775 			goto irq;
1776 	}
1777 
1778 	return 0;
1779 irq:
1780 	amdgpu_ras_sysfs_remove(adev, *ras_if);
1781 sysfs:
1782 	amdgpu_ras_debugfs_remove(adev, *ras_if);
1783 	amdgpu_ras_interrupt_remove_handler(adev, &ih_info);
1784 interrupt:
1785 	amdgpu_ras_feature_enable(adev, *ras_if, 0);
1786 feature:
1787 	kfree(*ras_if);
1788 	*ras_if = NULL;
1789 	return r;
1790 }
1791 
sdma_v4_0_sw_init(void * handle)1792 static int sdma_v4_0_sw_init(void *handle)
1793 {
1794 	struct amdgpu_ring *ring;
1795 	int r, i;
1796 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1797 
1798 	/* SDMA trap event */
1799 	for (i = 0; i < adev->sdma.num_instances; i++) {
1800 		r = amdgpu_irq_add_id(adev, sdma_v4_0_seq_to_irq_id(i),
1801 				      SDMA0_4_0__SRCID__SDMA_TRAP,
1802 				      &adev->sdma.trap_irq);
1803 		if (r)
1804 			return r;
1805 	}
1806 
1807 	/* SDMA SRAM ECC event */
1808 	for (i = 0; i < adev->sdma.num_instances; i++) {
1809 		r = amdgpu_irq_add_id(adev, sdma_v4_0_seq_to_irq_id(i),
1810 				      SDMA0_4_0__SRCID__SDMA_SRAM_ECC,
1811 				      &adev->sdma.ecc_irq);
1812 		if (r)
1813 			return r;
1814 	}
1815 
1816 	for (i = 0; i < adev->sdma.num_instances; i++) {
1817 		ring = &adev->sdma.instance[i].ring;
1818 		ring->ring_obj = NULL;
1819 		ring->use_doorbell = true;
1820 
1821 		DRM_INFO("use_doorbell being set to: [%s]\n",
1822 				ring->use_doorbell?"true":"false");
1823 
1824 		/* doorbell size is 2 dwords, get DWORD offset */
1825 		ring->doorbell_index = adev->doorbell_index.sdma_engine[i] << 1;
1826 
1827 		sprintf(ring->name, "sdma%d", i);
1828 		r = amdgpu_ring_init(adev, ring, 1024, &adev->sdma.trap_irq,
1829 				     AMDGPU_SDMA_IRQ_INSTANCE0 + i);
1830 		if (r)
1831 			return r;
1832 
1833 		if (adev->sdma.has_page_queue) {
1834 			ring = &adev->sdma.instance[i].page;
1835 			ring->ring_obj = NULL;
1836 			ring->use_doorbell = true;
1837 
1838 			/* paging queue use same doorbell index/routing as gfx queue
1839 			 * with 0x400 (4096 dwords) offset on second doorbell page
1840 			 */
1841 			ring->doorbell_index = adev->doorbell_index.sdma_engine[i] << 1;
1842 			ring->doorbell_index += 0x400;
1843 
1844 			sprintf(ring->name, "page%d", i);
1845 			r = amdgpu_ring_init(adev, ring, 1024,
1846 					     &adev->sdma.trap_irq,
1847 					     AMDGPU_SDMA_IRQ_INSTANCE0 + i);
1848 			if (r)
1849 				return r;
1850 		}
1851 	}
1852 
1853 	return r;
1854 }
1855 
sdma_v4_0_sw_fini(void * handle)1856 static int sdma_v4_0_sw_fini(void *handle)
1857 {
1858 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1859 	int i;
1860 
1861 	if (amdgpu_ras_is_supported(adev, AMDGPU_RAS_BLOCK__SDMA) &&
1862 			adev->sdma.ras_if) {
1863 		struct ras_common_if *ras_if = adev->sdma.ras_if;
1864 		struct ras_ih_if ih_info = {
1865 			.head = *ras_if,
1866 		};
1867 
1868 		/*remove fs first*/
1869 		amdgpu_ras_debugfs_remove(adev, ras_if);
1870 		amdgpu_ras_sysfs_remove(adev, ras_if);
1871 		/*remove the IH*/
1872 		amdgpu_ras_interrupt_remove_handler(adev, &ih_info);
1873 		amdgpu_ras_feature_enable(adev, ras_if, 0);
1874 		kfree(ras_if);
1875 	}
1876 
1877 	for (i = 0; i < adev->sdma.num_instances; i++) {
1878 		amdgpu_ring_fini(&adev->sdma.instance[i].ring);
1879 		if (adev->sdma.has_page_queue)
1880 			amdgpu_ring_fini(&adev->sdma.instance[i].page);
1881 	}
1882 
1883 	sdma_v4_0_destroy_inst_ctx(adev);
1884 
1885 	return 0;
1886 }
1887 
sdma_v4_0_hw_init(void * handle)1888 static int sdma_v4_0_hw_init(void *handle)
1889 {
1890 	int r;
1891 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1892 
1893 	if ((adev->asic_type == CHIP_RAVEN && adev->powerplay.pp_funcs &&
1894 			adev->powerplay.pp_funcs->set_powergating_by_smu) ||
1895 			adev->asic_type == CHIP_RENOIR)
1896 		amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_SDMA, false);
1897 
1898 	if (!amdgpu_sriov_vf(adev))
1899 		sdma_v4_0_init_golden_registers(adev);
1900 
1901 	r = sdma_v4_0_start(adev);
1902 
1903 	return r;
1904 }
1905 
sdma_v4_0_hw_fini(void * handle)1906 static int sdma_v4_0_hw_fini(void *handle)
1907 {
1908 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1909 	int i;
1910 
1911 	if (amdgpu_sriov_vf(adev))
1912 		return 0;
1913 
1914 	for (i = 0; i < adev->sdma.num_instances; i++) {
1915 		amdgpu_irq_put(adev, &adev->sdma.ecc_irq,
1916 			       AMDGPU_SDMA_IRQ_INSTANCE0 + i);
1917 	}
1918 
1919 	sdma_v4_0_ctx_switch_enable(adev, false);
1920 	sdma_v4_0_enable(adev, false);
1921 
1922 	if ((adev->asic_type == CHIP_RAVEN && adev->powerplay.pp_funcs
1923 			&& adev->powerplay.pp_funcs->set_powergating_by_smu) ||
1924 			adev->asic_type == CHIP_RENOIR)
1925 		amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_SDMA, true);
1926 
1927 	return 0;
1928 }
1929 
sdma_v4_0_suspend(void * handle)1930 static int sdma_v4_0_suspend(void *handle)
1931 {
1932 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1933 
1934 	return sdma_v4_0_hw_fini(adev);
1935 }
1936 
sdma_v4_0_resume(void * handle)1937 static int sdma_v4_0_resume(void *handle)
1938 {
1939 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1940 
1941 	return sdma_v4_0_hw_init(adev);
1942 }
1943 
sdma_v4_0_is_idle(void * handle)1944 static bool sdma_v4_0_is_idle(void *handle)
1945 {
1946 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1947 	u32 i;
1948 
1949 	for (i = 0; i < adev->sdma.num_instances; i++) {
1950 		u32 tmp = RREG32_SDMA(i, mmSDMA0_STATUS_REG);
1951 
1952 		if (!(tmp & SDMA0_STATUS_REG__IDLE_MASK))
1953 			return false;
1954 	}
1955 
1956 	return true;
1957 }
1958 
sdma_v4_0_wait_for_idle(void * handle)1959 static int sdma_v4_0_wait_for_idle(void *handle)
1960 {
1961 	unsigned i, j;
1962 	u32 sdma[AMDGPU_MAX_SDMA_INSTANCES];
1963 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1964 
1965 	for (i = 0; i < adev->usec_timeout; i++) {
1966 		for (j = 0; j < adev->sdma.num_instances; j++) {
1967 			sdma[j] = RREG32_SDMA(j, mmSDMA0_STATUS_REG);
1968 			if (!(sdma[j] & SDMA0_STATUS_REG__IDLE_MASK))
1969 				break;
1970 		}
1971 		if (j == adev->sdma.num_instances)
1972 			return 0;
1973 		udelay(1);
1974 	}
1975 	return -ETIMEDOUT;
1976 }
1977 
sdma_v4_0_soft_reset(void * handle)1978 static int sdma_v4_0_soft_reset(void *handle)
1979 {
1980 	/* todo */
1981 
1982 	return 0;
1983 }
1984 
sdma_v4_0_set_trap_irq_state(struct amdgpu_device * adev,struct amdgpu_irq_src * source,unsigned type,enum amdgpu_interrupt_state state)1985 static int sdma_v4_0_set_trap_irq_state(struct amdgpu_device *adev,
1986 					struct amdgpu_irq_src *source,
1987 					unsigned type,
1988 					enum amdgpu_interrupt_state state)
1989 {
1990 	u32 sdma_cntl;
1991 
1992 	sdma_cntl = RREG32_SDMA(type, mmSDMA0_CNTL);
1993 	sdma_cntl = REG_SET_FIELD(sdma_cntl, SDMA0_CNTL, TRAP_ENABLE,
1994 		       state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0);
1995 	WREG32_SDMA(type, mmSDMA0_CNTL, sdma_cntl);
1996 
1997 	return 0;
1998 }
1999 
sdma_v4_0_process_trap_irq(struct amdgpu_device * adev,struct amdgpu_irq_src * source,struct amdgpu_iv_entry * entry)2000 static int sdma_v4_0_process_trap_irq(struct amdgpu_device *adev,
2001 				      struct amdgpu_irq_src *source,
2002 				      struct amdgpu_iv_entry *entry)
2003 {
2004 	uint32_t instance;
2005 
2006 	DRM_DEBUG("IH: SDMA trap\n");
2007 	instance = sdma_v4_0_irq_id_to_seq(entry->client_id);
2008 	switch (entry->ring_id) {
2009 	case 0:
2010 		amdgpu_fence_process(&adev->sdma.instance[instance].ring);
2011 		break;
2012 	case 1:
2013 		if (adev->asic_type == CHIP_VEGA20)
2014 			amdgpu_fence_process(&adev->sdma.instance[instance].page);
2015 		break;
2016 	case 2:
2017 		/* XXX compute */
2018 		break;
2019 	case 3:
2020 		if (adev->asic_type != CHIP_VEGA20)
2021 			amdgpu_fence_process(&adev->sdma.instance[instance].page);
2022 		break;
2023 	}
2024 	return 0;
2025 }
2026 
sdma_v4_0_process_ras_data_cb(struct amdgpu_device * adev,struct ras_err_data * err_data,struct amdgpu_iv_entry * entry)2027 static int sdma_v4_0_process_ras_data_cb(struct amdgpu_device *adev,
2028 		struct ras_err_data *err_data,
2029 		struct amdgpu_iv_entry *entry)
2030 {
2031 	uint32_t err_source;
2032 	int instance;
2033 
2034 	instance = sdma_v4_0_irq_id_to_seq(entry->client_id);
2035 	if (instance < 0)
2036 		return 0;
2037 
2038 	switch (entry->src_id) {
2039 	case SDMA0_4_0__SRCID__SDMA_SRAM_ECC:
2040 		err_source = 0;
2041 		break;
2042 	case SDMA0_4_0__SRCID__SDMA_ECC:
2043 		err_source = 1;
2044 		break;
2045 	default:
2046 		return 0;
2047 	}
2048 
2049 	kgd2kfd_set_sram_ecc_flag(adev->kfd.dev);
2050 
2051 	amdgpu_ras_reset_gpu(adev, 0);
2052 
2053 	return AMDGPU_RAS_SUCCESS;
2054 }
2055 
sdma_v4_0_process_ecc_irq(struct amdgpu_device * adev,struct amdgpu_irq_src * source,struct amdgpu_iv_entry * entry)2056 static int sdma_v4_0_process_ecc_irq(struct amdgpu_device *adev,
2057 				      struct amdgpu_irq_src *source,
2058 				      struct amdgpu_iv_entry *entry)
2059 {
2060 	struct ras_common_if *ras_if = adev->sdma.ras_if;
2061 	struct ras_dispatch_if ih_data = {
2062 		.entry = entry,
2063 	};
2064 
2065 	if (!ras_if)
2066 		return 0;
2067 
2068 	ih_data.head = *ras_if;
2069 
2070 	amdgpu_ras_interrupt_dispatch(adev, &ih_data);
2071 	return 0;
2072 }
2073 
sdma_v4_0_process_illegal_inst_irq(struct amdgpu_device * adev,struct amdgpu_irq_src * source,struct amdgpu_iv_entry * entry)2074 static int sdma_v4_0_process_illegal_inst_irq(struct amdgpu_device *adev,
2075 					      struct amdgpu_irq_src *source,
2076 					      struct amdgpu_iv_entry *entry)
2077 {
2078 	int instance;
2079 
2080 	DRM_ERROR("Illegal instruction in SDMA command stream\n");
2081 
2082 	instance = sdma_v4_0_irq_id_to_seq(entry->client_id);
2083 	if (instance < 0)
2084 		return 0;
2085 
2086 	switch (entry->ring_id) {
2087 	case 0:
2088 		drm_sched_fault(&adev->sdma.instance[instance].ring.sched);
2089 		break;
2090 	}
2091 	return 0;
2092 }
2093 
sdma_v4_0_set_ecc_irq_state(struct amdgpu_device * adev,struct amdgpu_irq_src * source,unsigned type,enum amdgpu_interrupt_state state)2094 static int sdma_v4_0_set_ecc_irq_state(struct amdgpu_device *adev,
2095 					struct amdgpu_irq_src *source,
2096 					unsigned type,
2097 					enum amdgpu_interrupt_state state)
2098 {
2099 	u32 sdma_edc_config;
2100 
2101 	sdma_edc_config = RREG32_SDMA(type, mmSDMA0_EDC_CONFIG);
2102 	sdma_edc_config = REG_SET_FIELD(sdma_edc_config, SDMA0_EDC_CONFIG, ECC_INT_ENABLE,
2103 		       state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0);
2104 	WREG32_SDMA(type, mmSDMA0_EDC_CONFIG, sdma_edc_config);
2105 
2106 	return 0;
2107 }
2108 
sdma_v4_0_update_medium_grain_clock_gating(struct amdgpu_device * adev,bool enable)2109 static void sdma_v4_0_update_medium_grain_clock_gating(
2110 		struct amdgpu_device *adev,
2111 		bool enable)
2112 {
2113 	uint32_t data, def;
2114 	int i;
2115 
2116 	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_SDMA_MGCG)) {
2117 		for (i = 0; i < adev->sdma.num_instances; i++) {
2118 			def = data = RREG32_SDMA(i, mmSDMA0_CLK_CTRL);
2119 			data &= ~(SDMA0_CLK_CTRL__SOFT_OVERRIDE7_MASK |
2120 				  SDMA0_CLK_CTRL__SOFT_OVERRIDE6_MASK |
2121 				  SDMA0_CLK_CTRL__SOFT_OVERRIDE5_MASK |
2122 				  SDMA0_CLK_CTRL__SOFT_OVERRIDE4_MASK |
2123 				  SDMA0_CLK_CTRL__SOFT_OVERRIDE3_MASK |
2124 				  SDMA0_CLK_CTRL__SOFT_OVERRIDE2_MASK |
2125 				  SDMA0_CLK_CTRL__SOFT_OVERRIDE1_MASK |
2126 				  SDMA0_CLK_CTRL__SOFT_OVERRIDE0_MASK);
2127 			if (def != data)
2128 				WREG32_SDMA(i, mmSDMA0_CLK_CTRL, data);
2129 		}
2130 	} else {
2131 		for (i = 0; i < adev->sdma.num_instances; i++) {
2132 			def = data = RREG32_SDMA(i, mmSDMA0_CLK_CTRL);
2133 			data |= (SDMA0_CLK_CTRL__SOFT_OVERRIDE7_MASK |
2134 				 SDMA0_CLK_CTRL__SOFT_OVERRIDE6_MASK |
2135 				 SDMA0_CLK_CTRL__SOFT_OVERRIDE5_MASK |
2136 				 SDMA0_CLK_CTRL__SOFT_OVERRIDE4_MASK |
2137 				 SDMA0_CLK_CTRL__SOFT_OVERRIDE3_MASK |
2138 				 SDMA0_CLK_CTRL__SOFT_OVERRIDE2_MASK |
2139 				 SDMA0_CLK_CTRL__SOFT_OVERRIDE1_MASK |
2140 				 SDMA0_CLK_CTRL__SOFT_OVERRIDE0_MASK);
2141 			if (def != data)
2142 				WREG32_SDMA(i, mmSDMA0_CLK_CTRL, data);
2143 		}
2144 	}
2145 }
2146 
2147 
sdma_v4_0_update_medium_grain_light_sleep(struct amdgpu_device * adev,bool enable)2148 static void sdma_v4_0_update_medium_grain_light_sleep(
2149 		struct amdgpu_device *adev,
2150 		bool enable)
2151 {
2152 	uint32_t data, def;
2153 	int i;
2154 
2155 	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_SDMA_LS)) {
2156 		for (i = 0; i < adev->sdma.num_instances; i++) {
2157 			/* 1-not override: enable sdma mem light sleep */
2158 			def = data = RREG32_SDMA(0, mmSDMA0_POWER_CNTL);
2159 			data |= SDMA0_POWER_CNTL__MEM_POWER_OVERRIDE_MASK;
2160 			if (def != data)
2161 				WREG32_SDMA(0, mmSDMA0_POWER_CNTL, data);
2162 		}
2163 	} else {
2164 		for (i = 0; i < adev->sdma.num_instances; i++) {
2165 		/* 0-override:disable sdma mem light sleep */
2166 			def = data = RREG32_SDMA(0, mmSDMA0_POWER_CNTL);
2167 			data &= ~SDMA0_POWER_CNTL__MEM_POWER_OVERRIDE_MASK;
2168 			if (def != data)
2169 				WREG32_SDMA(0, mmSDMA0_POWER_CNTL, data);
2170 		}
2171 	}
2172 }
2173 
sdma_v4_0_set_clockgating_state(void * handle,enum amd_clockgating_state state)2174 static int sdma_v4_0_set_clockgating_state(void *handle,
2175 					  enum amd_clockgating_state state)
2176 {
2177 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2178 
2179 	if (amdgpu_sriov_vf(adev))
2180 		return 0;
2181 
2182 	switch (adev->asic_type) {
2183 	case CHIP_VEGA10:
2184 	case CHIP_VEGA12:
2185 	case CHIP_VEGA20:
2186 	case CHIP_RAVEN:
2187 	case CHIP_ARCTURUS:
2188 	case CHIP_RENOIR:
2189 		sdma_v4_0_update_medium_grain_clock_gating(adev,
2190 				state == AMD_CG_STATE_GATE ? true : false);
2191 		sdma_v4_0_update_medium_grain_light_sleep(adev,
2192 				state == AMD_CG_STATE_GATE ? true : false);
2193 		break;
2194 	default:
2195 		break;
2196 	}
2197 	return 0;
2198 }
2199 
sdma_v4_0_set_powergating_state(void * handle,enum amd_powergating_state state)2200 static int sdma_v4_0_set_powergating_state(void *handle,
2201 					  enum amd_powergating_state state)
2202 {
2203 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2204 
2205 	switch (adev->asic_type) {
2206 	case CHIP_RAVEN:
2207 		sdma_v4_1_update_power_gating(adev,
2208 				state == AMD_PG_STATE_GATE ? true : false);
2209 		break;
2210 	default:
2211 		break;
2212 	}
2213 
2214 	return 0;
2215 }
2216 
sdma_v4_0_get_clockgating_state(void * handle,u32 * flags)2217 static void sdma_v4_0_get_clockgating_state(void *handle, u32 *flags)
2218 {
2219 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2220 	int data;
2221 
2222 	if (amdgpu_sriov_vf(adev))
2223 		*flags = 0;
2224 
2225 	/* AMD_CG_SUPPORT_SDMA_MGCG */
2226 	data = RREG32(SOC15_REG_OFFSET(SDMA0, 0, mmSDMA0_CLK_CTRL));
2227 	if (!(data & SDMA0_CLK_CTRL__SOFT_OVERRIDE7_MASK))
2228 		*flags |= AMD_CG_SUPPORT_SDMA_MGCG;
2229 
2230 	/* AMD_CG_SUPPORT_SDMA_LS */
2231 	data = RREG32(SOC15_REG_OFFSET(SDMA0, 0, mmSDMA0_POWER_CNTL));
2232 	if (data & SDMA0_POWER_CNTL__MEM_POWER_OVERRIDE_MASK)
2233 		*flags |= AMD_CG_SUPPORT_SDMA_LS;
2234 }
2235 
2236 const struct amd_ip_funcs sdma_v4_0_ip_funcs = {
2237 	.name = "sdma_v4_0",
2238 	.early_init = sdma_v4_0_early_init,
2239 	.late_init = sdma_v4_0_late_init,
2240 	.sw_init = sdma_v4_0_sw_init,
2241 	.sw_fini = sdma_v4_0_sw_fini,
2242 	.hw_init = sdma_v4_0_hw_init,
2243 	.hw_fini = sdma_v4_0_hw_fini,
2244 	.suspend = sdma_v4_0_suspend,
2245 	.resume = sdma_v4_0_resume,
2246 	.is_idle = sdma_v4_0_is_idle,
2247 	.wait_for_idle = sdma_v4_0_wait_for_idle,
2248 	.soft_reset = sdma_v4_0_soft_reset,
2249 	.set_clockgating_state = sdma_v4_0_set_clockgating_state,
2250 	.set_powergating_state = sdma_v4_0_set_powergating_state,
2251 	.get_clockgating_state = sdma_v4_0_get_clockgating_state,
2252 };
2253 
2254 static const struct amdgpu_ring_funcs sdma_v4_0_ring_funcs = {
2255 	.type = AMDGPU_RING_TYPE_SDMA,
2256 	.align_mask = 0xf,
2257 	.nop = SDMA_PKT_NOP_HEADER_OP(SDMA_OP_NOP),
2258 	.support_64bit_ptrs = true,
2259 	.vmhub = AMDGPU_MMHUB_0,
2260 	.get_rptr = sdma_v4_0_ring_get_rptr,
2261 	.get_wptr = sdma_v4_0_ring_get_wptr,
2262 	.set_wptr = sdma_v4_0_ring_set_wptr,
2263 	.emit_frame_size =
2264 		6 + /* sdma_v4_0_ring_emit_hdp_flush */
2265 		3 + /* hdp invalidate */
2266 		6 + /* sdma_v4_0_ring_emit_pipeline_sync */
2267 		/* sdma_v4_0_ring_emit_vm_flush */
2268 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
2269 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 6 +
2270 		10 + 10 + 10, /* sdma_v4_0_ring_emit_fence x3 for user fence, vm fence */
2271 	.emit_ib_size = 7 + 6, /* sdma_v4_0_ring_emit_ib */
2272 	.emit_ib = sdma_v4_0_ring_emit_ib,
2273 	.emit_fence = sdma_v4_0_ring_emit_fence,
2274 	.emit_pipeline_sync = sdma_v4_0_ring_emit_pipeline_sync,
2275 	.emit_vm_flush = sdma_v4_0_ring_emit_vm_flush,
2276 	.emit_hdp_flush = sdma_v4_0_ring_emit_hdp_flush,
2277 	.test_ring = sdma_v4_0_ring_test_ring,
2278 	.test_ib = sdma_v4_0_ring_test_ib,
2279 	.insert_nop = sdma_v4_0_ring_insert_nop,
2280 	.pad_ib = sdma_v4_0_ring_pad_ib,
2281 	.emit_wreg = sdma_v4_0_ring_emit_wreg,
2282 	.emit_reg_wait = sdma_v4_0_ring_emit_reg_wait,
2283 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
2284 };
2285 
2286 /*
2287  * On Arcturus, SDMA instance 5~7 has a different vmhub type(AMDGPU_MMHUB_1).
2288  * So create a individual constant ring_funcs for those instances.
2289  */
2290 static const struct amdgpu_ring_funcs sdma_v4_0_ring_funcs_2nd_mmhub = {
2291 	.type = AMDGPU_RING_TYPE_SDMA,
2292 	.align_mask = 0xf,
2293 	.nop = SDMA_PKT_NOP_HEADER_OP(SDMA_OP_NOP),
2294 	.support_64bit_ptrs = true,
2295 	.vmhub = AMDGPU_MMHUB_1,
2296 	.get_rptr = sdma_v4_0_ring_get_rptr,
2297 	.get_wptr = sdma_v4_0_ring_get_wptr,
2298 	.set_wptr = sdma_v4_0_ring_set_wptr,
2299 	.emit_frame_size =
2300 		6 + /* sdma_v4_0_ring_emit_hdp_flush */
2301 		3 + /* hdp invalidate */
2302 		6 + /* sdma_v4_0_ring_emit_pipeline_sync */
2303 		/* sdma_v4_0_ring_emit_vm_flush */
2304 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
2305 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 6 +
2306 		10 + 10 + 10, /* sdma_v4_0_ring_emit_fence x3 for user fence, vm fence */
2307 	.emit_ib_size = 7 + 6, /* sdma_v4_0_ring_emit_ib */
2308 	.emit_ib = sdma_v4_0_ring_emit_ib,
2309 	.emit_fence = sdma_v4_0_ring_emit_fence,
2310 	.emit_pipeline_sync = sdma_v4_0_ring_emit_pipeline_sync,
2311 	.emit_vm_flush = sdma_v4_0_ring_emit_vm_flush,
2312 	.emit_hdp_flush = sdma_v4_0_ring_emit_hdp_flush,
2313 	.test_ring = sdma_v4_0_ring_test_ring,
2314 	.test_ib = sdma_v4_0_ring_test_ib,
2315 	.insert_nop = sdma_v4_0_ring_insert_nop,
2316 	.pad_ib = sdma_v4_0_ring_pad_ib,
2317 	.emit_wreg = sdma_v4_0_ring_emit_wreg,
2318 	.emit_reg_wait = sdma_v4_0_ring_emit_reg_wait,
2319 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
2320 };
2321 
2322 static const struct amdgpu_ring_funcs sdma_v4_0_page_ring_funcs = {
2323 	.type = AMDGPU_RING_TYPE_SDMA,
2324 	.align_mask = 0xf,
2325 	.nop = SDMA_PKT_NOP_HEADER_OP(SDMA_OP_NOP),
2326 	.support_64bit_ptrs = true,
2327 	.vmhub = AMDGPU_MMHUB_0,
2328 	.get_rptr = sdma_v4_0_ring_get_rptr,
2329 	.get_wptr = sdma_v4_0_page_ring_get_wptr,
2330 	.set_wptr = sdma_v4_0_page_ring_set_wptr,
2331 	.emit_frame_size =
2332 		6 + /* sdma_v4_0_ring_emit_hdp_flush */
2333 		3 + /* hdp invalidate */
2334 		6 + /* sdma_v4_0_ring_emit_pipeline_sync */
2335 		/* sdma_v4_0_ring_emit_vm_flush */
2336 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
2337 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 6 +
2338 		10 + 10 + 10, /* sdma_v4_0_ring_emit_fence x3 for user fence, vm fence */
2339 	.emit_ib_size = 7 + 6, /* sdma_v4_0_ring_emit_ib */
2340 	.emit_ib = sdma_v4_0_ring_emit_ib,
2341 	.emit_fence = sdma_v4_0_ring_emit_fence,
2342 	.emit_pipeline_sync = sdma_v4_0_ring_emit_pipeline_sync,
2343 	.emit_vm_flush = sdma_v4_0_ring_emit_vm_flush,
2344 	.emit_hdp_flush = sdma_v4_0_ring_emit_hdp_flush,
2345 	.test_ring = sdma_v4_0_ring_test_ring,
2346 	.test_ib = sdma_v4_0_ring_test_ib,
2347 	.insert_nop = sdma_v4_0_ring_insert_nop,
2348 	.pad_ib = sdma_v4_0_ring_pad_ib,
2349 	.emit_wreg = sdma_v4_0_ring_emit_wreg,
2350 	.emit_reg_wait = sdma_v4_0_ring_emit_reg_wait,
2351 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
2352 };
2353 
2354 static const struct amdgpu_ring_funcs sdma_v4_0_page_ring_funcs_2nd_mmhub = {
2355 	.type = AMDGPU_RING_TYPE_SDMA,
2356 	.align_mask = 0xf,
2357 	.nop = SDMA_PKT_NOP_HEADER_OP(SDMA_OP_NOP),
2358 	.support_64bit_ptrs = true,
2359 	.vmhub = AMDGPU_MMHUB_1,
2360 	.get_rptr = sdma_v4_0_ring_get_rptr,
2361 	.get_wptr = sdma_v4_0_page_ring_get_wptr,
2362 	.set_wptr = sdma_v4_0_page_ring_set_wptr,
2363 	.emit_frame_size =
2364 		6 + /* sdma_v4_0_ring_emit_hdp_flush */
2365 		3 + /* hdp invalidate */
2366 		6 + /* sdma_v4_0_ring_emit_pipeline_sync */
2367 		/* sdma_v4_0_ring_emit_vm_flush */
2368 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 3 +
2369 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 6 +
2370 		10 + 10 + 10, /* sdma_v4_0_ring_emit_fence x3 for user fence, vm fence */
2371 	.emit_ib_size = 7 + 6, /* sdma_v4_0_ring_emit_ib */
2372 	.emit_ib = sdma_v4_0_ring_emit_ib,
2373 	.emit_fence = sdma_v4_0_ring_emit_fence,
2374 	.emit_pipeline_sync = sdma_v4_0_ring_emit_pipeline_sync,
2375 	.emit_vm_flush = sdma_v4_0_ring_emit_vm_flush,
2376 	.emit_hdp_flush = sdma_v4_0_ring_emit_hdp_flush,
2377 	.test_ring = sdma_v4_0_ring_test_ring,
2378 	.test_ib = sdma_v4_0_ring_test_ib,
2379 	.insert_nop = sdma_v4_0_ring_insert_nop,
2380 	.pad_ib = sdma_v4_0_ring_pad_ib,
2381 	.emit_wreg = sdma_v4_0_ring_emit_wreg,
2382 	.emit_reg_wait = sdma_v4_0_ring_emit_reg_wait,
2383 	.emit_reg_write_reg_wait = amdgpu_ring_emit_reg_write_reg_wait_helper,
2384 };
2385 
sdma_v4_0_set_ring_funcs(struct amdgpu_device * adev)2386 static void sdma_v4_0_set_ring_funcs(struct amdgpu_device *adev)
2387 {
2388 	int i;
2389 
2390 	for (i = 0; i < adev->sdma.num_instances; i++) {
2391 		if (adev->asic_type == CHIP_ARCTURUS && i >= 5)
2392 			adev->sdma.instance[i].ring.funcs =
2393 					&sdma_v4_0_ring_funcs_2nd_mmhub;
2394 		else
2395 			adev->sdma.instance[i].ring.funcs =
2396 					&sdma_v4_0_ring_funcs;
2397 		adev->sdma.instance[i].ring.me = i;
2398 		if (adev->sdma.has_page_queue) {
2399 			if (adev->asic_type == CHIP_ARCTURUS && i >= 5)
2400 				adev->sdma.instance[i].page.funcs =
2401 					&sdma_v4_0_page_ring_funcs_2nd_mmhub;
2402 			else
2403 				adev->sdma.instance[i].page.funcs =
2404 					&sdma_v4_0_page_ring_funcs;
2405 			adev->sdma.instance[i].page.me = i;
2406 		}
2407 	}
2408 }
2409 
2410 static const struct amdgpu_irq_src_funcs sdma_v4_0_trap_irq_funcs = {
2411 	.set = sdma_v4_0_set_trap_irq_state,
2412 	.process = sdma_v4_0_process_trap_irq,
2413 };
2414 
2415 static const struct amdgpu_irq_src_funcs sdma_v4_0_illegal_inst_irq_funcs = {
2416 	.process = sdma_v4_0_process_illegal_inst_irq,
2417 };
2418 
2419 static const struct amdgpu_irq_src_funcs sdma_v4_0_ecc_irq_funcs = {
2420 	.set = sdma_v4_0_set_ecc_irq_state,
2421 	.process = sdma_v4_0_process_ecc_irq,
2422 };
2423 
2424 
2425 
sdma_v4_0_set_irq_funcs(struct amdgpu_device * adev)2426 static void sdma_v4_0_set_irq_funcs(struct amdgpu_device *adev)
2427 {
2428 	switch (adev->sdma.num_instances) {
2429 	case 1:
2430 		adev->sdma.trap_irq.num_types = AMDGPU_SDMA_IRQ_INSTANCE1;
2431 		adev->sdma.ecc_irq.num_types = AMDGPU_SDMA_IRQ_INSTANCE1;
2432 		break;
2433 	case 8:
2434 		adev->sdma.trap_irq.num_types = AMDGPU_SDMA_IRQ_LAST;
2435 		adev->sdma.ecc_irq.num_types = AMDGPU_SDMA_IRQ_LAST;
2436 		break;
2437 	case 2:
2438 	default:
2439 		adev->sdma.trap_irq.num_types = AMDGPU_SDMA_IRQ_INSTANCE2;
2440 		adev->sdma.ecc_irq.num_types = AMDGPU_SDMA_IRQ_INSTANCE2;
2441 		break;
2442 	}
2443 	adev->sdma.trap_irq.funcs = &sdma_v4_0_trap_irq_funcs;
2444 	adev->sdma.illegal_inst_irq.funcs = &sdma_v4_0_illegal_inst_irq_funcs;
2445 	adev->sdma.ecc_irq.funcs = &sdma_v4_0_ecc_irq_funcs;
2446 }
2447 
2448 /**
2449  * sdma_v4_0_emit_copy_buffer - copy buffer using the sDMA engine
2450  *
2451  * @ring: amdgpu_ring structure holding ring information
2452  * @src_offset: src GPU address
2453  * @dst_offset: dst GPU address
2454  * @byte_count: number of bytes to xfer
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)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 {
2465 	ib->ptr[ib->length_dw++] = SDMA_PKT_HEADER_OP(SDMA_OP_COPY) |
2466 		SDMA_PKT_HEADER_SUB_OP(SDMA_SUBOP_COPY_LINEAR);
2467 	ib->ptr[ib->length_dw++] = byte_count - 1;
2468 	ib->ptr[ib->length_dw++] = 0; /* src/dst endian swap */
2469 	ib->ptr[ib->length_dw++] = lower_32_bits(src_offset);
2470 	ib->ptr[ib->length_dw++] = upper_32_bits(src_offset);
2471 	ib->ptr[ib->length_dw++] = lower_32_bits(dst_offset);
2472 	ib->ptr[ib->length_dw++] = upper_32_bits(dst_offset);
2473 }
2474 
2475 /**
2476  * sdma_v4_0_emit_fill_buffer - fill buffer using the sDMA engine
2477  *
2478  * @ring: amdgpu_ring structure holding ring information
2479  * @src_data: value to write to buffer
2480  * @dst_offset: dst GPU address
2481  * @byte_count: number of bytes to xfer
2482  *
2483  * Fill GPU buffers using the DMA engine (VEGA10/12).
2484  */
sdma_v4_0_emit_fill_buffer(struct amdgpu_ib * ib,uint32_t src_data,uint64_t dst_offset,uint32_t byte_count)2485 static void sdma_v4_0_emit_fill_buffer(struct amdgpu_ib *ib,
2486 				       uint32_t src_data,
2487 				       uint64_t dst_offset,
2488 				       uint32_t byte_count)
2489 {
2490 	ib->ptr[ib->length_dw++] = SDMA_PKT_HEADER_OP(SDMA_OP_CONST_FILL);
2491 	ib->ptr[ib->length_dw++] = lower_32_bits(dst_offset);
2492 	ib->ptr[ib->length_dw++] = upper_32_bits(dst_offset);
2493 	ib->ptr[ib->length_dw++] = src_data;
2494 	ib->ptr[ib->length_dw++] = byte_count - 1;
2495 }
2496 
2497 static const struct amdgpu_buffer_funcs sdma_v4_0_buffer_funcs = {
2498 	.copy_max_bytes = 0x400000,
2499 	.copy_num_dw = 7,
2500 	.emit_copy_buffer = sdma_v4_0_emit_copy_buffer,
2501 
2502 	.fill_max_bytes = 0x400000,
2503 	.fill_num_dw = 5,
2504 	.emit_fill_buffer = sdma_v4_0_emit_fill_buffer,
2505 };
2506 
sdma_v4_0_set_buffer_funcs(struct amdgpu_device * adev)2507 static void sdma_v4_0_set_buffer_funcs(struct amdgpu_device *adev)
2508 {
2509 	adev->mman.buffer_funcs = &sdma_v4_0_buffer_funcs;
2510 	if (adev->sdma.has_page_queue)
2511 		adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].page;
2512 	else
2513 		adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].ring;
2514 }
2515 
2516 static const struct amdgpu_vm_pte_funcs sdma_v4_0_vm_pte_funcs = {
2517 	.copy_pte_num_dw = 7,
2518 	.copy_pte = sdma_v4_0_vm_copy_pte,
2519 
2520 	.write_pte = sdma_v4_0_vm_write_pte,
2521 	.set_pte_pde = sdma_v4_0_vm_set_pte_pde,
2522 };
2523 
sdma_v4_0_set_vm_pte_funcs(struct amdgpu_device * adev)2524 static void sdma_v4_0_set_vm_pte_funcs(struct amdgpu_device *adev)
2525 {
2526 	struct drm_gpu_scheduler *sched;
2527 	unsigned i;
2528 
2529 	adev->vm_manager.vm_pte_funcs = &sdma_v4_0_vm_pte_funcs;
2530 	for (i = 0; i < adev->sdma.num_instances; i++) {
2531 		if (adev->sdma.has_page_queue)
2532 			sched = &adev->sdma.instance[i].page.sched;
2533 		else
2534 			sched = &adev->sdma.instance[i].ring.sched;
2535 		adev->vm_manager.vm_pte_rqs[i] =
2536 			&sched->sched_rq[DRM_SCHED_PRIORITY_KERNEL];
2537 	}
2538 	adev->vm_manager.vm_pte_num_rqs = adev->sdma.num_instances;
2539 }
2540 
2541 const struct amdgpu_ip_block_version sdma_v4_0_ip_block = {
2542 	.type = AMD_IP_BLOCK_TYPE_SDMA,
2543 	.major = 4,
2544 	.minor = 0,
2545 	.rev = 0,
2546 	.funcs = &sdma_v4_0_ip_funcs,
2547 };
2548