xref: /linux/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c (revision 6fdcba32711044c35c0e1b094cbd8f3f0b4472c9)
1 /*
2  * Copyright 2014 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/kernel.h>
26 #include <linux/firmware.h>
27 #include <linux/module.h>
28 #include <linux/pci.h>
29 
30 #include "amdgpu.h"
31 #include "amdgpu_gfx.h"
32 #include "vi.h"
33 #include "vi_structs.h"
34 #include "vid.h"
35 #include "amdgpu_ucode.h"
36 #include "amdgpu_atombios.h"
37 #include "atombios_i2c.h"
38 #include "clearstate_vi.h"
39 
40 #include "gmc/gmc_8_2_d.h"
41 #include "gmc/gmc_8_2_sh_mask.h"
42 
43 #include "oss/oss_3_0_d.h"
44 #include "oss/oss_3_0_sh_mask.h"
45 
46 #include "bif/bif_5_0_d.h"
47 #include "bif/bif_5_0_sh_mask.h"
48 #include "gca/gfx_8_0_d.h"
49 #include "gca/gfx_8_0_enum.h"
50 #include "gca/gfx_8_0_sh_mask.h"
51 
52 #include "dce/dce_10_0_d.h"
53 #include "dce/dce_10_0_sh_mask.h"
54 
55 #include "smu/smu_7_1_3_d.h"
56 
57 #include "ivsrcid/ivsrcid_vislands30.h"
58 
59 #define GFX8_NUM_GFX_RINGS     1
60 #define GFX8_MEC_HPD_SIZE 4096
61 
62 #define TOPAZ_GB_ADDR_CONFIG_GOLDEN 0x22010001
63 #define CARRIZO_GB_ADDR_CONFIG_GOLDEN 0x22010001
64 #define POLARIS11_GB_ADDR_CONFIG_GOLDEN 0x22011002
65 #define TONGA_GB_ADDR_CONFIG_GOLDEN 0x22011003
66 
67 #define ARRAY_MODE(x)					((x) << GB_TILE_MODE0__ARRAY_MODE__SHIFT)
68 #define PIPE_CONFIG(x)					((x) << GB_TILE_MODE0__PIPE_CONFIG__SHIFT)
69 #define TILE_SPLIT(x)					((x) << GB_TILE_MODE0__TILE_SPLIT__SHIFT)
70 #define MICRO_TILE_MODE_NEW(x)				((x) << GB_TILE_MODE0__MICRO_TILE_MODE_NEW__SHIFT)
71 #define SAMPLE_SPLIT(x)					((x) << GB_TILE_MODE0__SAMPLE_SPLIT__SHIFT)
72 #define BANK_WIDTH(x)					((x) << GB_MACROTILE_MODE0__BANK_WIDTH__SHIFT)
73 #define BANK_HEIGHT(x)					((x) << GB_MACROTILE_MODE0__BANK_HEIGHT__SHIFT)
74 #define MACRO_TILE_ASPECT(x)				((x) << GB_MACROTILE_MODE0__MACRO_TILE_ASPECT__SHIFT)
75 #define NUM_BANKS(x)					((x) << GB_MACROTILE_MODE0__NUM_BANKS__SHIFT)
76 
77 #define RLC_CGTT_MGCG_OVERRIDE__CPF_MASK            0x00000001L
78 #define RLC_CGTT_MGCG_OVERRIDE__RLC_MASK            0x00000002L
79 #define RLC_CGTT_MGCG_OVERRIDE__MGCG_MASK           0x00000004L
80 #define RLC_CGTT_MGCG_OVERRIDE__CGCG_MASK           0x00000008L
81 #define RLC_CGTT_MGCG_OVERRIDE__CGLS_MASK           0x00000010L
82 #define RLC_CGTT_MGCG_OVERRIDE__GRBM_MASK           0x00000020L
83 
84 /* BPM SERDES CMD */
85 #define SET_BPM_SERDES_CMD    1
86 #define CLE_BPM_SERDES_CMD    0
87 
88 /* BPM Register Address*/
89 enum {
90 	BPM_REG_CGLS_EN = 0,        /* Enable/Disable CGLS */
91 	BPM_REG_CGLS_ON,            /* ON/OFF CGLS: shall be controlled by RLC FW */
92 	BPM_REG_CGCG_OVERRIDE,      /* Set/Clear CGCG Override */
93 	BPM_REG_MGCG_OVERRIDE,      /* Set/Clear MGCG Override */
94 	BPM_REG_FGCG_OVERRIDE,      /* Set/Clear FGCG Override */
95 	BPM_REG_FGCG_MAX
96 };
97 
98 #define RLC_FormatDirectRegListLength        14
99 
100 MODULE_FIRMWARE("amdgpu/carrizo_ce.bin");
101 MODULE_FIRMWARE("amdgpu/carrizo_pfp.bin");
102 MODULE_FIRMWARE("amdgpu/carrizo_me.bin");
103 MODULE_FIRMWARE("amdgpu/carrizo_mec.bin");
104 MODULE_FIRMWARE("amdgpu/carrizo_mec2.bin");
105 MODULE_FIRMWARE("amdgpu/carrizo_rlc.bin");
106 
107 MODULE_FIRMWARE("amdgpu/stoney_ce.bin");
108 MODULE_FIRMWARE("amdgpu/stoney_pfp.bin");
109 MODULE_FIRMWARE("amdgpu/stoney_me.bin");
110 MODULE_FIRMWARE("amdgpu/stoney_mec.bin");
111 MODULE_FIRMWARE("amdgpu/stoney_rlc.bin");
112 
113 MODULE_FIRMWARE("amdgpu/tonga_ce.bin");
114 MODULE_FIRMWARE("amdgpu/tonga_pfp.bin");
115 MODULE_FIRMWARE("amdgpu/tonga_me.bin");
116 MODULE_FIRMWARE("amdgpu/tonga_mec.bin");
117 MODULE_FIRMWARE("amdgpu/tonga_mec2.bin");
118 MODULE_FIRMWARE("amdgpu/tonga_rlc.bin");
119 
120 MODULE_FIRMWARE("amdgpu/topaz_ce.bin");
121 MODULE_FIRMWARE("amdgpu/topaz_pfp.bin");
122 MODULE_FIRMWARE("amdgpu/topaz_me.bin");
123 MODULE_FIRMWARE("amdgpu/topaz_mec.bin");
124 MODULE_FIRMWARE("amdgpu/topaz_rlc.bin");
125 
126 MODULE_FIRMWARE("amdgpu/fiji_ce.bin");
127 MODULE_FIRMWARE("amdgpu/fiji_pfp.bin");
128 MODULE_FIRMWARE("amdgpu/fiji_me.bin");
129 MODULE_FIRMWARE("amdgpu/fiji_mec.bin");
130 MODULE_FIRMWARE("amdgpu/fiji_mec2.bin");
131 MODULE_FIRMWARE("amdgpu/fiji_rlc.bin");
132 
133 MODULE_FIRMWARE("amdgpu/polaris10_ce.bin");
134 MODULE_FIRMWARE("amdgpu/polaris10_ce_2.bin");
135 MODULE_FIRMWARE("amdgpu/polaris10_pfp.bin");
136 MODULE_FIRMWARE("amdgpu/polaris10_pfp_2.bin");
137 MODULE_FIRMWARE("amdgpu/polaris10_me.bin");
138 MODULE_FIRMWARE("amdgpu/polaris10_me_2.bin");
139 MODULE_FIRMWARE("amdgpu/polaris10_mec.bin");
140 MODULE_FIRMWARE("amdgpu/polaris10_mec_2.bin");
141 MODULE_FIRMWARE("amdgpu/polaris10_mec2.bin");
142 MODULE_FIRMWARE("amdgpu/polaris10_mec2_2.bin");
143 MODULE_FIRMWARE("amdgpu/polaris10_rlc.bin");
144 
145 MODULE_FIRMWARE("amdgpu/polaris11_ce.bin");
146 MODULE_FIRMWARE("amdgpu/polaris11_ce_2.bin");
147 MODULE_FIRMWARE("amdgpu/polaris11_pfp.bin");
148 MODULE_FIRMWARE("amdgpu/polaris11_pfp_2.bin");
149 MODULE_FIRMWARE("amdgpu/polaris11_me.bin");
150 MODULE_FIRMWARE("amdgpu/polaris11_me_2.bin");
151 MODULE_FIRMWARE("amdgpu/polaris11_mec.bin");
152 MODULE_FIRMWARE("amdgpu/polaris11_mec_2.bin");
153 MODULE_FIRMWARE("amdgpu/polaris11_mec2.bin");
154 MODULE_FIRMWARE("amdgpu/polaris11_mec2_2.bin");
155 MODULE_FIRMWARE("amdgpu/polaris11_rlc.bin");
156 
157 MODULE_FIRMWARE("amdgpu/polaris12_ce.bin");
158 MODULE_FIRMWARE("amdgpu/polaris12_ce_2.bin");
159 MODULE_FIRMWARE("amdgpu/polaris12_pfp.bin");
160 MODULE_FIRMWARE("amdgpu/polaris12_pfp_2.bin");
161 MODULE_FIRMWARE("amdgpu/polaris12_me.bin");
162 MODULE_FIRMWARE("amdgpu/polaris12_me_2.bin");
163 MODULE_FIRMWARE("amdgpu/polaris12_mec.bin");
164 MODULE_FIRMWARE("amdgpu/polaris12_mec_2.bin");
165 MODULE_FIRMWARE("amdgpu/polaris12_mec2.bin");
166 MODULE_FIRMWARE("amdgpu/polaris12_mec2_2.bin");
167 MODULE_FIRMWARE("amdgpu/polaris12_rlc.bin");
168 
169 MODULE_FIRMWARE("amdgpu/vegam_ce.bin");
170 MODULE_FIRMWARE("amdgpu/vegam_pfp.bin");
171 MODULE_FIRMWARE("amdgpu/vegam_me.bin");
172 MODULE_FIRMWARE("amdgpu/vegam_mec.bin");
173 MODULE_FIRMWARE("amdgpu/vegam_mec2.bin");
174 MODULE_FIRMWARE("amdgpu/vegam_rlc.bin");
175 
176 static const struct amdgpu_gds_reg_offset amdgpu_gds_reg_offset[] =
177 {
178 	{mmGDS_VMID0_BASE, mmGDS_VMID0_SIZE, mmGDS_GWS_VMID0, mmGDS_OA_VMID0},
179 	{mmGDS_VMID1_BASE, mmGDS_VMID1_SIZE, mmGDS_GWS_VMID1, mmGDS_OA_VMID1},
180 	{mmGDS_VMID2_BASE, mmGDS_VMID2_SIZE, mmGDS_GWS_VMID2, mmGDS_OA_VMID2},
181 	{mmGDS_VMID3_BASE, mmGDS_VMID3_SIZE, mmGDS_GWS_VMID3, mmGDS_OA_VMID3},
182 	{mmGDS_VMID4_BASE, mmGDS_VMID4_SIZE, mmGDS_GWS_VMID4, mmGDS_OA_VMID4},
183 	{mmGDS_VMID5_BASE, mmGDS_VMID5_SIZE, mmGDS_GWS_VMID5, mmGDS_OA_VMID5},
184 	{mmGDS_VMID6_BASE, mmGDS_VMID6_SIZE, mmGDS_GWS_VMID6, mmGDS_OA_VMID6},
185 	{mmGDS_VMID7_BASE, mmGDS_VMID7_SIZE, mmGDS_GWS_VMID7, mmGDS_OA_VMID7},
186 	{mmGDS_VMID8_BASE, mmGDS_VMID8_SIZE, mmGDS_GWS_VMID8, mmGDS_OA_VMID8},
187 	{mmGDS_VMID9_BASE, mmGDS_VMID9_SIZE, mmGDS_GWS_VMID9, mmGDS_OA_VMID9},
188 	{mmGDS_VMID10_BASE, mmGDS_VMID10_SIZE, mmGDS_GWS_VMID10, mmGDS_OA_VMID10},
189 	{mmGDS_VMID11_BASE, mmGDS_VMID11_SIZE, mmGDS_GWS_VMID11, mmGDS_OA_VMID11},
190 	{mmGDS_VMID12_BASE, mmGDS_VMID12_SIZE, mmGDS_GWS_VMID12, mmGDS_OA_VMID12},
191 	{mmGDS_VMID13_BASE, mmGDS_VMID13_SIZE, mmGDS_GWS_VMID13, mmGDS_OA_VMID13},
192 	{mmGDS_VMID14_BASE, mmGDS_VMID14_SIZE, mmGDS_GWS_VMID14, mmGDS_OA_VMID14},
193 	{mmGDS_VMID15_BASE, mmGDS_VMID15_SIZE, mmGDS_GWS_VMID15, mmGDS_OA_VMID15}
194 };
195 
196 static const u32 golden_settings_tonga_a11[] =
197 {
198 	mmCB_HW_CONTROL, 0xfffdf3cf, 0x00007208,
199 	mmCB_HW_CONTROL_3, 0x00000040, 0x00000040,
200 	mmDB_DEBUG2, 0xf00fffff, 0x00000400,
201 	mmGB_GPU_ID, 0x0000000f, 0x00000000,
202 	mmPA_SC_ENHANCE, 0xffffffff, 0x20000001,
203 	mmPA_SC_FIFO_DEPTH_CNTL, 0x000003ff, 0x000000fc,
204 	mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000,
205 	mmRLC_CGCG_CGLS_CTRL, 0x00000003, 0x0000003c,
206 	mmSQ_RANDOM_WAVE_PRI, 0x001fffff, 0x000006fd,
207 	mmTA_CNTL_AUX, 0x000f000f, 0x000b0000,
208 	mmTCC_CTRL, 0x00100000, 0xf31fff7f,
209 	mmTCC_EXE_DISABLE, 0x00000002, 0x00000002,
210 	mmTCP_ADDR_CONFIG, 0x000003ff, 0x000002fb,
211 	mmTCP_CHAN_STEER_HI, 0xffffffff, 0x0000543b,
212 	mmTCP_CHAN_STEER_LO, 0xffffffff, 0xa9210876,
213 	mmVGT_RESET_DEBUG, 0x00000004, 0x00000004,
214 };
215 
216 static const u32 tonga_golden_common_all[] =
217 {
218 	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
219 	mmPA_SC_RASTER_CONFIG, 0xffffffff, 0x16000012,
220 	mmPA_SC_RASTER_CONFIG_1, 0xffffffff, 0x0000002A,
221 	mmGB_ADDR_CONFIG, 0xffffffff, 0x22011003,
222 	mmSPI_RESOURCE_RESERVE_CU_0, 0xffffffff, 0x00000800,
223 	mmSPI_RESOURCE_RESERVE_CU_1, 0xffffffff, 0x00000800,
224 	mmSPI_RESOURCE_RESERVE_EN_CU_0, 0xffffffff, 0x00FF7FBF,
225 	mmSPI_RESOURCE_RESERVE_EN_CU_1, 0xffffffff, 0x00FF7FAF
226 };
227 
228 static const u32 tonga_mgcg_cgcg_init[] =
229 {
230 	mmRLC_CGTT_MGCG_OVERRIDE, 0xffffffff, 0xffffffff,
231 	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
232 	mmCB_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100,
233 	mmCGTT_BCI_CLK_CTRL, 0xffffffff, 0x00000100,
234 	mmCGTT_CP_CLK_CTRL, 0xffffffff, 0x00000100,
235 	mmCGTT_CPC_CLK_CTRL, 0xffffffff, 0x00000100,
236 	mmCGTT_CPF_CLK_CTRL, 0xffffffff, 0x40000100,
237 	mmCGTT_GDS_CLK_CTRL, 0xffffffff, 0x00000100,
238 	mmCGTT_IA_CLK_CTRL, 0xffffffff, 0x06000100,
239 	mmCGTT_PA_CLK_CTRL, 0xffffffff, 0x00000100,
240 	mmCGTT_WD_CLK_CTRL, 0xffffffff, 0x06000100,
241 	mmCGTT_PC_CLK_CTRL, 0xffffffff, 0x00000100,
242 	mmCGTT_RLC_CLK_CTRL, 0xffffffff, 0x00000100,
243 	mmCGTT_SC_CLK_CTRL, 0xffffffff, 0x00000100,
244 	mmCGTT_SPI_CLK_CTRL, 0xffffffff, 0x00000100,
245 	mmCGTT_SQ_CLK_CTRL, 0xffffffff, 0x00000100,
246 	mmCGTT_SQG_CLK_CTRL, 0xffffffff, 0x00000100,
247 	mmCGTT_SX_CLK_CTRL0, 0xffffffff, 0x00000100,
248 	mmCGTT_SX_CLK_CTRL1, 0xffffffff, 0x00000100,
249 	mmCGTT_SX_CLK_CTRL2, 0xffffffff, 0x00000100,
250 	mmCGTT_SX_CLK_CTRL3, 0xffffffff, 0x00000100,
251 	mmCGTT_SX_CLK_CTRL4, 0xffffffff, 0x00000100,
252 	mmCGTT_TCI_CLK_CTRL, 0xffffffff, 0x00000100,
253 	mmCGTT_TCP_CLK_CTRL, 0xffffffff, 0x00000100,
254 	mmCGTT_VGT_CLK_CTRL, 0xffffffff, 0x06000100,
255 	mmDB_CGTT_CLK_CTRL_0, 0xffffffff, 0x00000100,
256 	mmTA_CGTT_CTRL, 0xffffffff, 0x00000100,
257 	mmTCA_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100,
258 	mmTCC_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100,
259 	mmTD_CGTT_CTRL, 0xffffffff, 0x00000100,
260 	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
261 	mmCGTS_CU0_SP0_CTRL_REG, 0xffffffff, 0x00010000,
262 	mmCGTS_CU0_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
263 	mmCGTS_CU0_TA_SQC_CTRL_REG, 0xffffffff, 0x00040007,
264 	mmCGTS_CU0_SP1_CTRL_REG, 0xffffffff, 0x00060005,
265 	mmCGTS_CU0_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
266 	mmCGTS_CU1_SP0_CTRL_REG, 0xffffffff, 0x00010000,
267 	mmCGTS_CU1_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
268 	mmCGTS_CU1_TA_CTRL_REG, 0xffffffff, 0x00040007,
269 	mmCGTS_CU1_SP1_CTRL_REG, 0xffffffff, 0x00060005,
270 	mmCGTS_CU1_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
271 	mmCGTS_CU2_SP0_CTRL_REG, 0xffffffff, 0x00010000,
272 	mmCGTS_CU2_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
273 	mmCGTS_CU2_TA_CTRL_REG, 0xffffffff, 0x00040007,
274 	mmCGTS_CU2_SP1_CTRL_REG, 0xffffffff, 0x00060005,
275 	mmCGTS_CU2_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
276 	mmCGTS_CU3_SP0_CTRL_REG, 0xffffffff, 0x00010000,
277 	mmCGTS_CU3_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
278 	mmCGTS_CU3_TA_CTRL_REG, 0xffffffff, 0x00040007,
279 	mmCGTS_CU3_SP1_CTRL_REG, 0xffffffff, 0x00060005,
280 	mmCGTS_CU3_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
281 	mmCGTS_CU4_SP0_CTRL_REG, 0xffffffff, 0x00010000,
282 	mmCGTS_CU4_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
283 	mmCGTS_CU4_TA_SQC_CTRL_REG, 0xffffffff, 0x00040007,
284 	mmCGTS_CU4_SP1_CTRL_REG, 0xffffffff, 0x00060005,
285 	mmCGTS_CU4_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
286 	mmCGTS_CU5_SP0_CTRL_REG, 0xffffffff, 0x00010000,
287 	mmCGTS_CU5_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
288 	mmCGTS_CU5_TA_CTRL_REG, 0xffffffff, 0x00040007,
289 	mmCGTS_CU5_SP1_CTRL_REG, 0xffffffff, 0x00060005,
290 	mmCGTS_CU5_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
291 	mmCGTS_CU6_SP0_CTRL_REG, 0xffffffff, 0x00010000,
292 	mmCGTS_CU6_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
293 	mmCGTS_CU6_TA_CTRL_REG, 0xffffffff, 0x00040007,
294 	mmCGTS_CU6_SP1_CTRL_REG, 0xffffffff, 0x00060005,
295 	mmCGTS_CU6_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
296 	mmCGTS_CU7_SP0_CTRL_REG, 0xffffffff, 0x00010000,
297 	mmCGTS_CU7_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
298 	mmCGTS_CU7_TA_CTRL_REG, 0xffffffff, 0x00040007,
299 	mmCGTS_CU7_SP1_CTRL_REG, 0xffffffff, 0x00060005,
300 	mmCGTS_CU7_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
301 	mmCGTS_SM_CTRL_REG, 0xffffffff, 0x96e00200,
302 	mmCP_RB_WPTR_POLL_CNTL, 0xffffffff, 0x00900100,
303 	mmRLC_CGCG_CGLS_CTRL, 0xffffffff, 0x0020003c,
304 	mmCP_MEM_SLP_CNTL, 0x00000001, 0x00000001,
305 };
306 
307 static const u32 golden_settings_vegam_a11[] =
308 {
309 	mmCB_HW_CONTROL, 0x0001f3cf, 0x00007208,
310 	mmCB_HW_CONTROL_2, 0x0f000000, 0x0d000000,
311 	mmCB_HW_CONTROL_3, 0x000001ff, 0x00000040,
312 	mmDB_DEBUG2, 0xf00fffff, 0x00000400,
313 	mmPA_SC_ENHANCE, 0xffffffff, 0x20000001,
314 	mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000,
315 	mmPA_SC_RASTER_CONFIG, 0x3f3fffff, 0x3a00161a,
316 	mmPA_SC_RASTER_CONFIG_1, 0x0000003f, 0x0000002e,
317 	mmRLC_CGCG_CGLS_CTRL, 0x00000003, 0x0001003c,
318 	mmRLC_CGCG_CGLS_CTRL_3D, 0xffffffff, 0x0001003c,
319 	mmSQ_CONFIG, 0x07f80000, 0x01180000,
320 	mmTA_CNTL_AUX, 0x000f000f, 0x000b0000,
321 	mmTCC_CTRL, 0x00100000, 0xf31fff7f,
322 	mmTCP_ADDR_CONFIG, 0x000003ff, 0x000000f7,
323 	mmTCP_CHAN_STEER_HI, 0xffffffff, 0x00000000,
324 	mmTCP_CHAN_STEER_LO, 0xffffffff, 0x32761054,
325 	mmVGT_RESET_DEBUG, 0x00000004, 0x00000004,
326 };
327 
328 static const u32 vegam_golden_common_all[] =
329 {
330 	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
331 	mmGB_ADDR_CONFIG, 0xffffffff, 0x22011003,
332 	mmSPI_RESOURCE_RESERVE_CU_0, 0xffffffff, 0x00000800,
333 	mmSPI_RESOURCE_RESERVE_CU_1, 0xffffffff, 0x00000800,
334 	mmSPI_RESOURCE_RESERVE_EN_CU_0, 0xffffffff, 0x00FF7FBF,
335 	mmSPI_RESOURCE_RESERVE_EN_CU_1, 0xffffffff, 0x00FF7FAF,
336 };
337 
338 static const u32 golden_settings_polaris11_a11[] =
339 {
340 	mmCB_HW_CONTROL, 0x0000f3cf, 0x00007208,
341 	mmCB_HW_CONTROL_2, 0x0f000000, 0x0f000000,
342 	mmCB_HW_CONTROL_3, 0x000001ff, 0x00000040,
343 	mmDB_DEBUG2, 0xf00fffff, 0x00000400,
344 	mmPA_SC_ENHANCE, 0xffffffff, 0x20000001,
345 	mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000,
346 	mmPA_SC_RASTER_CONFIG, 0x3f3fffff, 0x16000012,
347 	mmPA_SC_RASTER_CONFIG_1, 0x0000003f, 0x00000000,
348 	mmRLC_CGCG_CGLS_CTRL, 0x00000003, 0x0001003c,
349 	mmRLC_CGCG_CGLS_CTRL_3D, 0xffffffff, 0x0001003c,
350 	mmSQ_CONFIG, 0x07f80000, 0x01180000,
351 	mmTA_CNTL_AUX, 0x000f000f, 0x000b0000,
352 	mmTCC_CTRL, 0x00100000, 0xf31fff7f,
353 	mmTCP_ADDR_CONFIG, 0x000003ff, 0x000000f3,
354 	mmTCP_CHAN_STEER_HI, 0xffffffff, 0x00000000,
355 	mmTCP_CHAN_STEER_LO, 0xffffffff, 0x00003210,
356 	mmVGT_RESET_DEBUG, 0x00000004, 0x00000004,
357 };
358 
359 static const u32 polaris11_golden_common_all[] =
360 {
361 	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
362 	mmGB_ADDR_CONFIG, 0xffffffff, 0x22011002,
363 	mmSPI_RESOURCE_RESERVE_CU_0, 0xffffffff, 0x00000800,
364 	mmSPI_RESOURCE_RESERVE_CU_1, 0xffffffff, 0x00000800,
365 	mmSPI_RESOURCE_RESERVE_EN_CU_0, 0xffffffff, 0x00FF7FBF,
366 	mmSPI_RESOURCE_RESERVE_EN_CU_1, 0xffffffff, 0x00FF7FAF,
367 };
368 
369 static const u32 golden_settings_polaris10_a11[] =
370 {
371 	mmATC_MISC_CG, 0x000c0fc0, 0x000c0200,
372 	mmCB_HW_CONTROL, 0x0001f3cf, 0x00007208,
373 	mmCB_HW_CONTROL_2, 0x0f000000, 0x0f000000,
374 	mmCB_HW_CONTROL_3, 0x000001ff, 0x00000040,
375 	mmDB_DEBUG2, 0xf00fffff, 0x00000400,
376 	mmPA_SC_ENHANCE, 0xffffffff, 0x20000001,
377 	mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000,
378 	mmPA_SC_RASTER_CONFIG, 0x3f3fffff, 0x16000012,
379 	mmPA_SC_RASTER_CONFIG_1, 0x0000003f, 0x0000002a,
380 	mmRLC_CGCG_CGLS_CTRL, 0x00000003, 0x0001003c,
381 	mmRLC_CGCG_CGLS_CTRL_3D, 0xffffffff, 0x0001003c,
382 	mmSQ_CONFIG, 0x07f80000, 0x07180000,
383 	mmTA_CNTL_AUX, 0x000f000f, 0x000b0000,
384 	mmTCC_CTRL, 0x00100000, 0xf31fff7f,
385 	mmTCP_ADDR_CONFIG, 0x000003ff, 0x000000f7,
386 	mmTCP_CHAN_STEER_HI, 0xffffffff, 0x00000000,
387 	mmVGT_RESET_DEBUG, 0x00000004, 0x00000004,
388 };
389 
390 static const u32 polaris10_golden_common_all[] =
391 {
392 	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
393 	mmPA_SC_RASTER_CONFIG, 0xffffffff, 0x16000012,
394 	mmPA_SC_RASTER_CONFIG_1, 0xffffffff, 0x0000002A,
395 	mmGB_ADDR_CONFIG, 0xffffffff, 0x22011003,
396 	mmSPI_RESOURCE_RESERVE_CU_0, 0xffffffff, 0x00000800,
397 	mmSPI_RESOURCE_RESERVE_CU_1, 0xffffffff, 0x00000800,
398 	mmSPI_RESOURCE_RESERVE_EN_CU_0, 0xffffffff, 0x00FF7FBF,
399 	mmSPI_RESOURCE_RESERVE_EN_CU_1, 0xffffffff, 0x00FF7FAF,
400 };
401 
402 static const u32 fiji_golden_common_all[] =
403 {
404 	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
405 	mmPA_SC_RASTER_CONFIG, 0xffffffff, 0x3a00161a,
406 	mmPA_SC_RASTER_CONFIG_1, 0xffffffff, 0x0000002e,
407 	mmGB_ADDR_CONFIG, 0xffffffff, 0x22011003,
408 	mmSPI_RESOURCE_RESERVE_CU_0, 0xffffffff, 0x00000800,
409 	mmSPI_RESOURCE_RESERVE_CU_1, 0xffffffff, 0x00000800,
410 	mmSPI_RESOURCE_RESERVE_EN_CU_0, 0xffffffff, 0x00FF7FBF,
411 	mmSPI_RESOURCE_RESERVE_EN_CU_1, 0xffffffff, 0x00FF7FAF,
412 	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
413 	mmSPI_CONFIG_CNTL_1, 0x0000000f, 0x00000009,
414 };
415 
416 static const u32 golden_settings_fiji_a10[] =
417 {
418 	mmCB_HW_CONTROL_3, 0x000001ff, 0x00000040,
419 	mmDB_DEBUG2, 0xf00fffff, 0x00000400,
420 	mmPA_SC_ENHANCE, 0xffffffff, 0x20000001,
421 	mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000,
422 	mmRLC_CGCG_CGLS_CTRL, 0x00000003, 0x0001003c,
423 	mmSQ_RANDOM_WAVE_PRI, 0x001fffff, 0x000006fd,
424 	mmTA_CNTL_AUX, 0x000f000f, 0x000b0000,
425 	mmTCC_CTRL, 0x00100000, 0xf31fff7f,
426 	mmTCC_EXE_DISABLE, 0x00000002, 0x00000002,
427 	mmTCP_ADDR_CONFIG, 0x000003ff, 0x000000ff,
428 	mmVGT_RESET_DEBUG, 0x00000004, 0x00000004,
429 };
430 
431 static const u32 fiji_mgcg_cgcg_init[] =
432 {
433 	mmRLC_CGTT_MGCG_OVERRIDE, 0xffffffff, 0xffffffff,
434 	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
435 	mmCB_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100,
436 	mmCGTT_BCI_CLK_CTRL, 0xffffffff, 0x00000100,
437 	mmCGTT_CP_CLK_CTRL, 0xffffffff, 0x00000100,
438 	mmCGTT_CPC_CLK_CTRL, 0xffffffff, 0x00000100,
439 	mmCGTT_CPF_CLK_CTRL, 0xffffffff, 0x40000100,
440 	mmCGTT_GDS_CLK_CTRL, 0xffffffff, 0x00000100,
441 	mmCGTT_IA_CLK_CTRL, 0xffffffff, 0x06000100,
442 	mmCGTT_PA_CLK_CTRL, 0xffffffff, 0x00000100,
443 	mmCGTT_WD_CLK_CTRL, 0xffffffff, 0x06000100,
444 	mmCGTT_PC_CLK_CTRL, 0xffffffff, 0x00000100,
445 	mmCGTT_RLC_CLK_CTRL, 0xffffffff, 0x00000100,
446 	mmCGTT_SC_CLK_CTRL, 0xffffffff, 0x00000100,
447 	mmCGTT_SPI_CLK_CTRL, 0xffffffff, 0x00000100,
448 	mmCGTT_SQ_CLK_CTRL, 0xffffffff, 0x00000100,
449 	mmCGTT_SQG_CLK_CTRL, 0xffffffff, 0x00000100,
450 	mmCGTT_SX_CLK_CTRL0, 0xffffffff, 0x00000100,
451 	mmCGTT_SX_CLK_CTRL1, 0xffffffff, 0x00000100,
452 	mmCGTT_SX_CLK_CTRL2, 0xffffffff, 0x00000100,
453 	mmCGTT_SX_CLK_CTRL3, 0xffffffff, 0x00000100,
454 	mmCGTT_SX_CLK_CTRL4, 0xffffffff, 0x00000100,
455 	mmCGTT_TCI_CLK_CTRL, 0xffffffff, 0x00000100,
456 	mmCGTT_TCP_CLK_CTRL, 0xffffffff, 0x00000100,
457 	mmCGTT_VGT_CLK_CTRL, 0xffffffff, 0x06000100,
458 	mmDB_CGTT_CLK_CTRL_0, 0xffffffff, 0x00000100,
459 	mmTA_CGTT_CTRL, 0xffffffff, 0x00000100,
460 	mmTCA_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100,
461 	mmTCC_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100,
462 	mmTD_CGTT_CTRL, 0xffffffff, 0x00000100,
463 	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
464 	mmCGTS_SM_CTRL_REG, 0xffffffff, 0x96e00200,
465 	mmCP_RB_WPTR_POLL_CNTL, 0xffffffff, 0x00900100,
466 	mmRLC_CGCG_CGLS_CTRL, 0xffffffff, 0x0020003c,
467 	mmCP_MEM_SLP_CNTL, 0x00000001, 0x00000001,
468 };
469 
470 static const u32 golden_settings_iceland_a11[] =
471 {
472 	mmCB_HW_CONTROL_3, 0x00000040, 0x00000040,
473 	mmDB_DEBUG2, 0xf00fffff, 0x00000400,
474 	mmDB_DEBUG3, 0xc0000000, 0xc0000000,
475 	mmGB_GPU_ID, 0x0000000f, 0x00000000,
476 	mmPA_SC_ENHANCE, 0xffffffff, 0x20000001,
477 	mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000,
478 	mmPA_SC_RASTER_CONFIG, 0x3f3fffff, 0x00000002,
479 	mmPA_SC_RASTER_CONFIG_1, 0x0000003f, 0x00000000,
480 	mmRLC_CGCG_CGLS_CTRL, 0x00000003, 0x0000003c,
481 	mmSQ_RANDOM_WAVE_PRI, 0x001fffff, 0x000006fd,
482 	mmTA_CNTL_AUX, 0x000f000f, 0x000b0000,
483 	mmTCC_CTRL, 0x00100000, 0xf31fff7f,
484 	mmTCC_EXE_DISABLE, 0x00000002, 0x00000002,
485 	mmTCP_ADDR_CONFIG, 0x000003ff, 0x000000f1,
486 	mmTCP_CHAN_STEER_HI, 0xffffffff, 0x00000000,
487 	mmTCP_CHAN_STEER_LO, 0xffffffff, 0x00000010,
488 };
489 
490 static const u32 iceland_golden_common_all[] =
491 {
492 	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
493 	mmPA_SC_RASTER_CONFIG, 0xffffffff, 0x00000002,
494 	mmPA_SC_RASTER_CONFIG_1, 0xffffffff, 0x00000000,
495 	mmGB_ADDR_CONFIG, 0xffffffff, 0x22010001,
496 	mmSPI_RESOURCE_RESERVE_CU_0, 0xffffffff, 0x00000800,
497 	mmSPI_RESOURCE_RESERVE_CU_1, 0xffffffff, 0x00000800,
498 	mmSPI_RESOURCE_RESERVE_EN_CU_0, 0xffffffff, 0x00FF7FBF,
499 	mmSPI_RESOURCE_RESERVE_EN_CU_1, 0xffffffff, 0x00FF7FAF
500 };
501 
502 static const u32 iceland_mgcg_cgcg_init[] =
503 {
504 	mmRLC_CGTT_MGCG_OVERRIDE, 0xffffffff, 0xffffffff,
505 	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
506 	mmCB_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100,
507 	mmCGTT_BCI_CLK_CTRL, 0xffffffff, 0x00000100,
508 	mmCGTT_CP_CLK_CTRL, 0xffffffff, 0xc0000100,
509 	mmCGTT_CPC_CLK_CTRL, 0xffffffff, 0xc0000100,
510 	mmCGTT_CPF_CLK_CTRL, 0xffffffff, 0xc0000100,
511 	mmCGTT_GDS_CLK_CTRL, 0xffffffff, 0x00000100,
512 	mmCGTT_IA_CLK_CTRL, 0xffffffff, 0x06000100,
513 	mmCGTT_PA_CLK_CTRL, 0xffffffff, 0x00000100,
514 	mmCGTT_WD_CLK_CTRL, 0xffffffff, 0x06000100,
515 	mmCGTT_PC_CLK_CTRL, 0xffffffff, 0x00000100,
516 	mmCGTT_RLC_CLK_CTRL, 0xffffffff, 0x00000100,
517 	mmCGTT_SC_CLK_CTRL, 0xffffffff, 0x00000100,
518 	mmCGTT_SPI_CLK_CTRL, 0xffffffff, 0x00000100,
519 	mmCGTT_SQ_CLK_CTRL, 0xffffffff, 0x00000100,
520 	mmCGTT_SQG_CLK_CTRL, 0xffffffff, 0x00000100,
521 	mmCGTT_SX_CLK_CTRL0, 0xffffffff, 0x00000100,
522 	mmCGTT_SX_CLK_CTRL1, 0xffffffff, 0x00000100,
523 	mmCGTT_SX_CLK_CTRL2, 0xffffffff, 0x00000100,
524 	mmCGTT_SX_CLK_CTRL3, 0xffffffff, 0x00000100,
525 	mmCGTT_SX_CLK_CTRL4, 0xffffffff, 0x00000100,
526 	mmCGTT_TCI_CLK_CTRL, 0xffffffff, 0xff000100,
527 	mmCGTT_TCP_CLK_CTRL, 0xffffffff, 0x00000100,
528 	mmCGTT_VGT_CLK_CTRL, 0xffffffff, 0x06000100,
529 	mmDB_CGTT_CLK_CTRL_0, 0xffffffff, 0x00000100,
530 	mmTA_CGTT_CTRL, 0xffffffff, 0x00000100,
531 	mmTCA_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100,
532 	mmTCC_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100,
533 	mmTD_CGTT_CTRL, 0xffffffff, 0x00000100,
534 	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
535 	mmCGTS_CU0_SP0_CTRL_REG, 0xffffffff, 0x00010000,
536 	mmCGTS_CU0_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
537 	mmCGTS_CU0_TA_SQC_CTRL_REG, 0xffffffff, 0x0f840f87,
538 	mmCGTS_CU0_SP1_CTRL_REG, 0xffffffff, 0x00060005,
539 	mmCGTS_CU0_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
540 	mmCGTS_CU1_SP0_CTRL_REG, 0xffffffff, 0x00010000,
541 	mmCGTS_CU1_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
542 	mmCGTS_CU1_TA_CTRL_REG, 0xffffffff, 0x00040007,
543 	mmCGTS_CU1_SP1_CTRL_REG, 0xffffffff, 0x00060005,
544 	mmCGTS_CU1_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
545 	mmCGTS_CU2_SP0_CTRL_REG, 0xffffffff, 0x00010000,
546 	mmCGTS_CU2_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
547 	mmCGTS_CU2_TA_CTRL_REG, 0xffffffff, 0x00040007,
548 	mmCGTS_CU2_SP1_CTRL_REG, 0xffffffff, 0x00060005,
549 	mmCGTS_CU2_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
550 	mmCGTS_CU3_SP0_CTRL_REG, 0xffffffff, 0x00010000,
551 	mmCGTS_CU3_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
552 	mmCGTS_CU3_TA_CTRL_REG, 0xffffffff, 0x00040007,
553 	mmCGTS_CU3_SP1_CTRL_REG, 0xffffffff, 0x00060005,
554 	mmCGTS_CU3_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
555 	mmCGTS_CU4_SP0_CTRL_REG, 0xffffffff, 0x00010000,
556 	mmCGTS_CU4_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
557 	mmCGTS_CU4_TA_SQC_CTRL_REG, 0xffffffff, 0x0f840f87,
558 	mmCGTS_CU4_SP1_CTRL_REG, 0xffffffff, 0x00060005,
559 	mmCGTS_CU4_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
560 	mmCGTS_CU5_SP0_CTRL_REG, 0xffffffff, 0x00010000,
561 	mmCGTS_CU5_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
562 	mmCGTS_CU5_TA_CTRL_REG, 0xffffffff, 0x00040007,
563 	mmCGTS_CU5_SP1_CTRL_REG, 0xffffffff, 0x00060005,
564 	mmCGTS_CU5_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
565 	mmCGTS_SM_CTRL_REG, 0xffffffff, 0x96e00200,
566 	mmCP_RB_WPTR_POLL_CNTL, 0xffffffff, 0x00900100,
567 	mmRLC_CGCG_CGLS_CTRL, 0xffffffff, 0x0020003c,
568 };
569 
570 static const u32 cz_golden_settings_a11[] =
571 {
572 	mmCB_HW_CONTROL_3, 0x00000040, 0x00000040,
573 	mmDB_DEBUG2, 0xf00fffff, 0x00000400,
574 	mmGB_GPU_ID, 0x0000000f, 0x00000000,
575 	mmPA_SC_ENHANCE, 0xffffffff, 0x00000001,
576 	mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000,
577 	mmRLC_CGCG_CGLS_CTRL, 0x00000003, 0x0000003c,
578 	mmSQ_RANDOM_WAVE_PRI, 0x001fffff, 0x000006fd,
579 	mmTA_CNTL_AUX, 0x000f000f, 0x00010000,
580 	mmTCC_CTRL, 0x00100000, 0xf31fff7f,
581 	mmTCC_EXE_DISABLE, 0x00000002, 0x00000002,
582 	mmTCP_ADDR_CONFIG, 0x0000000f, 0x000000f3,
583 	mmTCP_CHAN_STEER_LO, 0xffffffff, 0x00001302
584 };
585 
586 static const u32 cz_golden_common_all[] =
587 {
588 	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
589 	mmPA_SC_RASTER_CONFIG, 0xffffffff, 0x00000002,
590 	mmPA_SC_RASTER_CONFIG_1, 0xffffffff, 0x00000000,
591 	mmGB_ADDR_CONFIG, 0xffffffff, 0x22010001,
592 	mmSPI_RESOURCE_RESERVE_CU_0, 0xffffffff, 0x00000800,
593 	mmSPI_RESOURCE_RESERVE_CU_1, 0xffffffff, 0x00000800,
594 	mmSPI_RESOURCE_RESERVE_EN_CU_0, 0xffffffff, 0x00FF7FBF,
595 	mmSPI_RESOURCE_RESERVE_EN_CU_1, 0xffffffff, 0x00FF7FAF
596 };
597 
598 static const u32 cz_mgcg_cgcg_init[] =
599 {
600 	mmRLC_CGTT_MGCG_OVERRIDE, 0xffffffff, 0xffffffff,
601 	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
602 	mmCB_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100,
603 	mmCGTT_BCI_CLK_CTRL, 0xffffffff, 0x00000100,
604 	mmCGTT_CP_CLK_CTRL, 0xffffffff, 0x00000100,
605 	mmCGTT_CPC_CLK_CTRL, 0xffffffff, 0x00000100,
606 	mmCGTT_CPF_CLK_CTRL, 0xffffffff, 0x00000100,
607 	mmCGTT_GDS_CLK_CTRL, 0xffffffff, 0x00000100,
608 	mmCGTT_IA_CLK_CTRL, 0xffffffff, 0x06000100,
609 	mmCGTT_PA_CLK_CTRL, 0xffffffff, 0x00000100,
610 	mmCGTT_WD_CLK_CTRL, 0xffffffff, 0x06000100,
611 	mmCGTT_PC_CLK_CTRL, 0xffffffff, 0x00000100,
612 	mmCGTT_RLC_CLK_CTRL, 0xffffffff, 0x00000100,
613 	mmCGTT_SC_CLK_CTRL, 0xffffffff, 0x00000100,
614 	mmCGTT_SPI_CLK_CTRL, 0xffffffff, 0x00000100,
615 	mmCGTT_SQ_CLK_CTRL, 0xffffffff, 0x00000100,
616 	mmCGTT_SQG_CLK_CTRL, 0xffffffff, 0x00000100,
617 	mmCGTT_SX_CLK_CTRL0, 0xffffffff, 0x00000100,
618 	mmCGTT_SX_CLK_CTRL1, 0xffffffff, 0x00000100,
619 	mmCGTT_SX_CLK_CTRL2, 0xffffffff, 0x00000100,
620 	mmCGTT_SX_CLK_CTRL3, 0xffffffff, 0x00000100,
621 	mmCGTT_SX_CLK_CTRL4, 0xffffffff, 0x00000100,
622 	mmCGTT_TCI_CLK_CTRL, 0xffffffff, 0x00000100,
623 	mmCGTT_TCP_CLK_CTRL, 0xffffffff, 0x00000100,
624 	mmCGTT_VGT_CLK_CTRL, 0xffffffff, 0x06000100,
625 	mmDB_CGTT_CLK_CTRL_0, 0xffffffff, 0x00000100,
626 	mmTA_CGTT_CTRL, 0xffffffff, 0x00000100,
627 	mmTCA_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100,
628 	mmTCC_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100,
629 	mmTD_CGTT_CTRL, 0xffffffff, 0x00000100,
630 	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
631 	mmCGTS_CU0_SP0_CTRL_REG, 0xffffffff, 0x00010000,
632 	mmCGTS_CU0_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
633 	mmCGTS_CU0_TA_SQC_CTRL_REG, 0xffffffff, 0x00040007,
634 	mmCGTS_CU0_SP1_CTRL_REG, 0xffffffff, 0x00060005,
635 	mmCGTS_CU0_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
636 	mmCGTS_CU1_SP0_CTRL_REG, 0xffffffff, 0x00010000,
637 	mmCGTS_CU1_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
638 	mmCGTS_CU1_TA_CTRL_REG, 0xffffffff, 0x00040007,
639 	mmCGTS_CU1_SP1_CTRL_REG, 0xffffffff, 0x00060005,
640 	mmCGTS_CU1_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
641 	mmCGTS_CU2_SP0_CTRL_REG, 0xffffffff, 0x00010000,
642 	mmCGTS_CU2_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
643 	mmCGTS_CU2_TA_CTRL_REG, 0xffffffff, 0x00040007,
644 	mmCGTS_CU2_SP1_CTRL_REG, 0xffffffff, 0x00060005,
645 	mmCGTS_CU2_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
646 	mmCGTS_CU3_SP0_CTRL_REG, 0xffffffff, 0x00010000,
647 	mmCGTS_CU3_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
648 	mmCGTS_CU3_TA_CTRL_REG, 0xffffffff, 0x00040007,
649 	mmCGTS_CU3_SP1_CTRL_REG, 0xffffffff, 0x00060005,
650 	mmCGTS_CU3_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
651 	mmCGTS_CU4_SP0_CTRL_REG, 0xffffffff, 0x00010000,
652 	mmCGTS_CU4_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
653 	mmCGTS_CU4_TA_SQC_CTRL_REG, 0xffffffff, 0x00040007,
654 	mmCGTS_CU4_SP1_CTRL_REG, 0xffffffff, 0x00060005,
655 	mmCGTS_CU4_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
656 	mmCGTS_CU5_SP0_CTRL_REG, 0xffffffff, 0x00010000,
657 	mmCGTS_CU5_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
658 	mmCGTS_CU5_TA_CTRL_REG, 0xffffffff, 0x00040007,
659 	mmCGTS_CU5_SP1_CTRL_REG, 0xffffffff, 0x00060005,
660 	mmCGTS_CU5_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
661 	mmCGTS_CU6_SP0_CTRL_REG, 0xffffffff, 0x00010000,
662 	mmCGTS_CU6_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
663 	mmCGTS_CU6_TA_CTRL_REG, 0xffffffff, 0x00040007,
664 	mmCGTS_CU6_SP1_CTRL_REG, 0xffffffff, 0x00060005,
665 	mmCGTS_CU6_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
666 	mmCGTS_CU7_SP0_CTRL_REG, 0xffffffff, 0x00010000,
667 	mmCGTS_CU7_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
668 	mmCGTS_CU7_TA_CTRL_REG, 0xffffffff, 0x00040007,
669 	mmCGTS_CU7_SP1_CTRL_REG, 0xffffffff, 0x00060005,
670 	mmCGTS_CU7_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
671 	mmCGTS_SM_CTRL_REG, 0xffffffff, 0x96e00200,
672 	mmCP_RB_WPTR_POLL_CNTL, 0xffffffff, 0x00900100,
673 	mmRLC_CGCG_CGLS_CTRL, 0xffffffff, 0x0020003f,
674 	mmCP_MEM_SLP_CNTL, 0x00000001, 0x00000001,
675 };
676 
677 static const u32 stoney_golden_settings_a11[] =
678 {
679 	mmDB_DEBUG2, 0xf00fffff, 0x00000400,
680 	mmGB_GPU_ID, 0x0000000f, 0x00000000,
681 	mmPA_SC_ENHANCE, 0xffffffff, 0x20000001,
682 	mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000,
683 	mmRLC_CGCG_CGLS_CTRL, 0x00000003, 0x0001003c,
684 	mmTA_CNTL_AUX, 0x000f000f, 0x000b0000,
685 	mmTCC_CTRL, 0x00100000, 0xf31fff7f,
686 	mmTCC_EXE_DISABLE, 0x00000002, 0x00000002,
687 	mmTCP_ADDR_CONFIG, 0x0000000f, 0x000000f1,
688 	mmTCP_CHAN_STEER_LO, 0xffffffff, 0x10101010,
689 };
690 
691 static const u32 stoney_golden_common_all[] =
692 {
693 	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
694 	mmPA_SC_RASTER_CONFIG, 0xffffffff, 0x00000000,
695 	mmPA_SC_RASTER_CONFIG_1, 0xffffffff, 0x00000000,
696 	mmGB_ADDR_CONFIG, 0xffffffff, 0x12010001,
697 	mmSPI_RESOURCE_RESERVE_CU_0, 0xffffffff, 0x00000800,
698 	mmSPI_RESOURCE_RESERVE_CU_1, 0xffffffff, 0x00000800,
699 	mmSPI_RESOURCE_RESERVE_EN_CU_0, 0xffffffff, 0x00FF7FBF,
700 	mmSPI_RESOURCE_RESERVE_EN_CU_1, 0xffffffff, 0x00FF7FAF,
701 };
702 
703 static const u32 stoney_mgcg_cgcg_init[] =
704 {
705 	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
706 	mmRLC_CGCG_CGLS_CTRL, 0xffffffff, 0x0020003f,
707 	mmCP_MEM_SLP_CNTL, 0xffffffff, 0x00020201,
708 	mmRLC_MEM_SLP_CNTL, 0xffffffff, 0x00020201,
709 	mmCGTS_SM_CTRL_REG, 0xffffffff, 0x96940200,
710 };
711 
712 
713 static const char * const sq_edc_source_names[] = {
714 	"SQ_EDC_INFO_SOURCE_INVALID: No EDC error has occurred",
715 	"SQ_EDC_INFO_SOURCE_INST: EDC source is Instruction Fetch",
716 	"SQ_EDC_INFO_SOURCE_SGPR: EDC source is SGPR or SQC data return",
717 	"SQ_EDC_INFO_SOURCE_VGPR: EDC source is VGPR",
718 	"SQ_EDC_INFO_SOURCE_LDS: EDC source is LDS",
719 	"SQ_EDC_INFO_SOURCE_GDS: EDC source is GDS",
720 	"SQ_EDC_INFO_SOURCE_TA: EDC source is TA",
721 };
722 
723 static void gfx_v8_0_set_ring_funcs(struct amdgpu_device *adev);
724 static void gfx_v8_0_set_irq_funcs(struct amdgpu_device *adev);
725 static void gfx_v8_0_set_gds_init(struct amdgpu_device *adev);
726 static void gfx_v8_0_set_rlc_funcs(struct amdgpu_device *adev);
727 static u32 gfx_v8_0_get_csb_size(struct amdgpu_device *adev);
728 static void gfx_v8_0_get_cu_info(struct amdgpu_device *adev);
729 static void gfx_v8_0_ring_emit_ce_meta(struct amdgpu_ring *ring);
730 static void gfx_v8_0_ring_emit_de_meta(struct amdgpu_ring *ring);
731 
732 static void gfx_v8_0_init_golden_registers(struct amdgpu_device *adev)
733 {
734 	switch (adev->asic_type) {
735 	case CHIP_TOPAZ:
736 		amdgpu_device_program_register_sequence(adev,
737 							iceland_mgcg_cgcg_init,
738 							ARRAY_SIZE(iceland_mgcg_cgcg_init));
739 		amdgpu_device_program_register_sequence(adev,
740 							golden_settings_iceland_a11,
741 							ARRAY_SIZE(golden_settings_iceland_a11));
742 		amdgpu_device_program_register_sequence(adev,
743 							iceland_golden_common_all,
744 							ARRAY_SIZE(iceland_golden_common_all));
745 		break;
746 	case CHIP_FIJI:
747 		amdgpu_device_program_register_sequence(adev,
748 							fiji_mgcg_cgcg_init,
749 							ARRAY_SIZE(fiji_mgcg_cgcg_init));
750 		amdgpu_device_program_register_sequence(adev,
751 							golden_settings_fiji_a10,
752 							ARRAY_SIZE(golden_settings_fiji_a10));
753 		amdgpu_device_program_register_sequence(adev,
754 							fiji_golden_common_all,
755 							ARRAY_SIZE(fiji_golden_common_all));
756 		break;
757 
758 	case CHIP_TONGA:
759 		amdgpu_device_program_register_sequence(adev,
760 							tonga_mgcg_cgcg_init,
761 							ARRAY_SIZE(tonga_mgcg_cgcg_init));
762 		amdgpu_device_program_register_sequence(adev,
763 							golden_settings_tonga_a11,
764 							ARRAY_SIZE(golden_settings_tonga_a11));
765 		amdgpu_device_program_register_sequence(adev,
766 							tonga_golden_common_all,
767 							ARRAY_SIZE(tonga_golden_common_all));
768 		break;
769 	case CHIP_VEGAM:
770 		amdgpu_device_program_register_sequence(adev,
771 							golden_settings_vegam_a11,
772 							ARRAY_SIZE(golden_settings_vegam_a11));
773 		amdgpu_device_program_register_sequence(adev,
774 							vegam_golden_common_all,
775 							ARRAY_SIZE(vegam_golden_common_all));
776 		break;
777 	case CHIP_POLARIS11:
778 	case CHIP_POLARIS12:
779 		amdgpu_device_program_register_sequence(adev,
780 							golden_settings_polaris11_a11,
781 							ARRAY_SIZE(golden_settings_polaris11_a11));
782 		amdgpu_device_program_register_sequence(adev,
783 							polaris11_golden_common_all,
784 							ARRAY_SIZE(polaris11_golden_common_all));
785 		break;
786 	case CHIP_POLARIS10:
787 		amdgpu_device_program_register_sequence(adev,
788 							golden_settings_polaris10_a11,
789 							ARRAY_SIZE(golden_settings_polaris10_a11));
790 		amdgpu_device_program_register_sequence(adev,
791 							polaris10_golden_common_all,
792 							ARRAY_SIZE(polaris10_golden_common_all));
793 		WREG32_SMC(ixCG_ACLK_CNTL, 0x0000001C);
794 		if (adev->pdev->revision == 0xc7 &&
795 		    ((adev->pdev->subsystem_device == 0xb37 && adev->pdev->subsystem_vendor == 0x1002) ||
796 		     (adev->pdev->subsystem_device == 0x4a8 && adev->pdev->subsystem_vendor == 0x1043) ||
797 		     (adev->pdev->subsystem_device == 0x9480 && adev->pdev->subsystem_vendor == 0x1682))) {
798 			amdgpu_atombios_i2c_channel_trans(adev, 0x10, 0x96, 0x1E, 0xDD);
799 			amdgpu_atombios_i2c_channel_trans(adev, 0x10, 0x96, 0x1F, 0xD0);
800 		}
801 		break;
802 	case CHIP_CARRIZO:
803 		amdgpu_device_program_register_sequence(adev,
804 							cz_mgcg_cgcg_init,
805 							ARRAY_SIZE(cz_mgcg_cgcg_init));
806 		amdgpu_device_program_register_sequence(adev,
807 							cz_golden_settings_a11,
808 							ARRAY_SIZE(cz_golden_settings_a11));
809 		amdgpu_device_program_register_sequence(adev,
810 							cz_golden_common_all,
811 							ARRAY_SIZE(cz_golden_common_all));
812 		break;
813 	case CHIP_STONEY:
814 		amdgpu_device_program_register_sequence(adev,
815 							stoney_mgcg_cgcg_init,
816 							ARRAY_SIZE(stoney_mgcg_cgcg_init));
817 		amdgpu_device_program_register_sequence(adev,
818 							stoney_golden_settings_a11,
819 							ARRAY_SIZE(stoney_golden_settings_a11));
820 		amdgpu_device_program_register_sequence(adev,
821 							stoney_golden_common_all,
822 							ARRAY_SIZE(stoney_golden_common_all));
823 		break;
824 	default:
825 		break;
826 	}
827 }
828 
829 static void gfx_v8_0_scratch_init(struct amdgpu_device *adev)
830 {
831 	adev->gfx.scratch.num_reg = 8;
832 	adev->gfx.scratch.reg_base = mmSCRATCH_REG0;
833 	adev->gfx.scratch.free_mask = (1u << adev->gfx.scratch.num_reg) - 1;
834 }
835 
836 static int gfx_v8_0_ring_test_ring(struct amdgpu_ring *ring)
837 {
838 	struct amdgpu_device *adev = ring->adev;
839 	uint32_t scratch;
840 	uint32_t tmp = 0;
841 	unsigned i;
842 	int r;
843 
844 	r = amdgpu_gfx_scratch_get(adev, &scratch);
845 	if (r)
846 		return r;
847 
848 	WREG32(scratch, 0xCAFEDEAD);
849 	r = amdgpu_ring_alloc(ring, 3);
850 	if (r)
851 		goto error_free_scratch;
852 
853 	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_UCONFIG_REG, 1));
854 	amdgpu_ring_write(ring, (scratch - PACKET3_SET_UCONFIG_REG_START));
855 	amdgpu_ring_write(ring, 0xDEADBEEF);
856 	amdgpu_ring_commit(ring);
857 
858 	for (i = 0; i < adev->usec_timeout; i++) {
859 		tmp = RREG32(scratch);
860 		if (tmp == 0xDEADBEEF)
861 			break;
862 		udelay(1);
863 	}
864 
865 	if (i >= adev->usec_timeout)
866 		r = -ETIMEDOUT;
867 
868 error_free_scratch:
869 	amdgpu_gfx_scratch_free(adev, scratch);
870 	return r;
871 }
872 
873 static int gfx_v8_0_ring_test_ib(struct amdgpu_ring *ring, long timeout)
874 {
875 	struct amdgpu_device *adev = ring->adev;
876 	struct amdgpu_ib ib;
877 	struct dma_fence *f = NULL;
878 
879 	unsigned int index;
880 	uint64_t gpu_addr;
881 	uint32_t tmp;
882 	long r;
883 
884 	r = amdgpu_device_wb_get(adev, &index);
885 	if (r)
886 		return r;
887 
888 	gpu_addr = adev->wb.gpu_addr + (index * 4);
889 	adev->wb.wb[index] = cpu_to_le32(0xCAFEDEAD);
890 	memset(&ib, 0, sizeof(ib));
891 	r = amdgpu_ib_get(adev, NULL, 16, &ib);
892 	if (r)
893 		goto err1;
894 
895 	ib.ptr[0] = PACKET3(PACKET3_WRITE_DATA, 3);
896 	ib.ptr[1] = WRITE_DATA_DST_SEL(5) | WR_CONFIRM;
897 	ib.ptr[2] = lower_32_bits(gpu_addr);
898 	ib.ptr[3] = upper_32_bits(gpu_addr);
899 	ib.ptr[4] = 0xDEADBEEF;
900 	ib.length_dw = 5;
901 
902 	r = amdgpu_ib_schedule(ring, 1, &ib, NULL, &f);
903 	if (r)
904 		goto err2;
905 
906 	r = dma_fence_wait_timeout(f, false, timeout);
907 	if (r == 0) {
908 		r = -ETIMEDOUT;
909 		goto err2;
910 	} else if (r < 0) {
911 		goto err2;
912 	}
913 
914 	tmp = adev->wb.wb[index];
915 	if (tmp == 0xDEADBEEF)
916 		r = 0;
917 	else
918 		r = -EINVAL;
919 
920 err2:
921 	amdgpu_ib_free(adev, &ib, NULL);
922 	dma_fence_put(f);
923 err1:
924 	amdgpu_device_wb_free(adev, index);
925 	return r;
926 }
927 
928 
929 static void gfx_v8_0_free_microcode(struct amdgpu_device *adev)
930 {
931 	release_firmware(adev->gfx.pfp_fw);
932 	adev->gfx.pfp_fw = NULL;
933 	release_firmware(adev->gfx.me_fw);
934 	adev->gfx.me_fw = NULL;
935 	release_firmware(adev->gfx.ce_fw);
936 	adev->gfx.ce_fw = NULL;
937 	release_firmware(adev->gfx.rlc_fw);
938 	adev->gfx.rlc_fw = NULL;
939 	release_firmware(adev->gfx.mec_fw);
940 	adev->gfx.mec_fw = NULL;
941 	if ((adev->asic_type != CHIP_STONEY) &&
942 	    (adev->asic_type != CHIP_TOPAZ))
943 		release_firmware(adev->gfx.mec2_fw);
944 	adev->gfx.mec2_fw = NULL;
945 
946 	kfree(adev->gfx.rlc.register_list_format);
947 }
948 
949 static int gfx_v8_0_init_microcode(struct amdgpu_device *adev)
950 {
951 	const char *chip_name;
952 	char fw_name[30];
953 	int err;
954 	struct amdgpu_firmware_info *info = NULL;
955 	const struct common_firmware_header *header = NULL;
956 	const struct gfx_firmware_header_v1_0 *cp_hdr;
957 	const struct rlc_firmware_header_v2_0 *rlc_hdr;
958 	unsigned int *tmp = NULL, i;
959 
960 	DRM_DEBUG("\n");
961 
962 	switch (adev->asic_type) {
963 	case CHIP_TOPAZ:
964 		chip_name = "topaz";
965 		break;
966 	case CHIP_TONGA:
967 		chip_name = "tonga";
968 		break;
969 	case CHIP_CARRIZO:
970 		chip_name = "carrizo";
971 		break;
972 	case CHIP_FIJI:
973 		chip_name = "fiji";
974 		break;
975 	case CHIP_STONEY:
976 		chip_name = "stoney";
977 		break;
978 	case CHIP_POLARIS10:
979 		chip_name = "polaris10";
980 		break;
981 	case CHIP_POLARIS11:
982 		chip_name = "polaris11";
983 		break;
984 	case CHIP_POLARIS12:
985 		chip_name = "polaris12";
986 		break;
987 	case CHIP_VEGAM:
988 		chip_name = "vegam";
989 		break;
990 	default:
991 		BUG();
992 	}
993 
994 	if (adev->asic_type >= CHIP_POLARIS10 && adev->asic_type <= CHIP_POLARIS12) {
995 		snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_pfp_2.bin", chip_name);
996 		err = request_firmware(&adev->gfx.pfp_fw, fw_name, adev->dev);
997 		if (err == -ENOENT) {
998 			snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_pfp.bin", chip_name);
999 			err = request_firmware(&adev->gfx.pfp_fw, fw_name, adev->dev);
1000 		}
1001 	} else {
1002 		snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_pfp.bin", chip_name);
1003 		err = request_firmware(&adev->gfx.pfp_fw, fw_name, adev->dev);
1004 	}
1005 	if (err)
1006 		goto out;
1007 	err = amdgpu_ucode_validate(adev->gfx.pfp_fw);
1008 	if (err)
1009 		goto out;
1010 	cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.pfp_fw->data;
1011 	adev->gfx.pfp_fw_version = le32_to_cpu(cp_hdr->header.ucode_version);
1012 	adev->gfx.pfp_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version);
1013 
1014 	if (adev->asic_type >= CHIP_POLARIS10 && adev->asic_type <= CHIP_POLARIS12) {
1015 		snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_me_2.bin", chip_name);
1016 		err = request_firmware(&adev->gfx.me_fw, fw_name, adev->dev);
1017 		if (err == -ENOENT) {
1018 			snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_me.bin", chip_name);
1019 			err = request_firmware(&adev->gfx.me_fw, fw_name, adev->dev);
1020 		}
1021 	} else {
1022 		snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_me.bin", chip_name);
1023 		err = request_firmware(&adev->gfx.me_fw, fw_name, adev->dev);
1024 	}
1025 	if (err)
1026 		goto out;
1027 	err = amdgpu_ucode_validate(adev->gfx.me_fw);
1028 	if (err)
1029 		goto out;
1030 	cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.me_fw->data;
1031 	adev->gfx.me_fw_version = le32_to_cpu(cp_hdr->header.ucode_version);
1032 
1033 	adev->gfx.me_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version);
1034 
1035 	if (adev->asic_type >= CHIP_POLARIS10 && adev->asic_type <= CHIP_POLARIS12) {
1036 		snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_ce_2.bin", chip_name);
1037 		err = request_firmware(&adev->gfx.ce_fw, fw_name, adev->dev);
1038 		if (err == -ENOENT) {
1039 			snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_ce.bin", chip_name);
1040 			err = request_firmware(&adev->gfx.ce_fw, fw_name, adev->dev);
1041 		}
1042 	} else {
1043 		snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_ce.bin", chip_name);
1044 		err = request_firmware(&adev->gfx.ce_fw, fw_name, adev->dev);
1045 	}
1046 	if (err)
1047 		goto out;
1048 	err = amdgpu_ucode_validate(adev->gfx.ce_fw);
1049 	if (err)
1050 		goto out;
1051 	cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.ce_fw->data;
1052 	adev->gfx.ce_fw_version = le32_to_cpu(cp_hdr->header.ucode_version);
1053 	adev->gfx.ce_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version);
1054 
1055 	/*
1056 	 * Support for MCBP/Virtualization in combination with chained IBs is
1057 	 * formal released on feature version #46
1058 	 */
1059 	if (adev->gfx.ce_feature_version >= 46 &&
1060 	    adev->gfx.pfp_feature_version >= 46) {
1061 		adev->virt.chained_ib_support = true;
1062 		DRM_INFO("Chained IB support enabled!\n");
1063 	} else
1064 		adev->virt.chained_ib_support = false;
1065 
1066 	snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_rlc.bin", chip_name);
1067 	err = request_firmware(&adev->gfx.rlc_fw, fw_name, adev->dev);
1068 	if (err)
1069 		goto out;
1070 	err = amdgpu_ucode_validate(adev->gfx.rlc_fw);
1071 	rlc_hdr = (const struct rlc_firmware_header_v2_0 *)adev->gfx.rlc_fw->data;
1072 	adev->gfx.rlc_fw_version = le32_to_cpu(rlc_hdr->header.ucode_version);
1073 	adev->gfx.rlc_feature_version = le32_to_cpu(rlc_hdr->ucode_feature_version);
1074 
1075 	adev->gfx.rlc.save_and_restore_offset =
1076 			le32_to_cpu(rlc_hdr->save_and_restore_offset);
1077 	adev->gfx.rlc.clear_state_descriptor_offset =
1078 			le32_to_cpu(rlc_hdr->clear_state_descriptor_offset);
1079 	adev->gfx.rlc.avail_scratch_ram_locations =
1080 			le32_to_cpu(rlc_hdr->avail_scratch_ram_locations);
1081 	adev->gfx.rlc.reg_restore_list_size =
1082 			le32_to_cpu(rlc_hdr->reg_restore_list_size);
1083 	adev->gfx.rlc.reg_list_format_start =
1084 			le32_to_cpu(rlc_hdr->reg_list_format_start);
1085 	adev->gfx.rlc.reg_list_format_separate_start =
1086 			le32_to_cpu(rlc_hdr->reg_list_format_separate_start);
1087 	adev->gfx.rlc.starting_offsets_start =
1088 			le32_to_cpu(rlc_hdr->starting_offsets_start);
1089 	adev->gfx.rlc.reg_list_format_size_bytes =
1090 			le32_to_cpu(rlc_hdr->reg_list_format_size_bytes);
1091 	adev->gfx.rlc.reg_list_size_bytes =
1092 			le32_to_cpu(rlc_hdr->reg_list_size_bytes);
1093 
1094 	adev->gfx.rlc.register_list_format =
1095 			kmalloc(adev->gfx.rlc.reg_list_format_size_bytes +
1096 					adev->gfx.rlc.reg_list_size_bytes, GFP_KERNEL);
1097 
1098 	if (!adev->gfx.rlc.register_list_format) {
1099 		err = -ENOMEM;
1100 		goto out;
1101 	}
1102 
1103 	tmp = (unsigned int *)((uintptr_t)rlc_hdr +
1104 			le32_to_cpu(rlc_hdr->reg_list_format_array_offset_bytes));
1105 	for (i = 0 ; i < (adev->gfx.rlc.reg_list_format_size_bytes >> 2); i++)
1106 		adev->gfx.rlc.register_list_format[i] =	le32_to_cpu(tmp[i]);
1107 
1108 	adev->gfx.rlc.register_restore = adev->gfx.rlc.register_list_format + i;
1109 
1110 	tmp = (unsigned int *)((uintptr_t)rlc_hdr +
1111 			le32_to_cpu(rlc_hdr->reg_list_array_offset_bytes));
1112 	for (i = 0 ; i < (adev->gfx.rlc.reg_list_size_bytes >> 2); i++)
1113 		adev->gfx.rlc.register_restore[i] = le32_to_cpu(tmp[i]);
1114 
1115 	if (adev->asic_type >= CHIP_POLARIS10 && adev->asic_type <= CHIP_POLARIS12) {
1116 		snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec_2.bin", chip_name);
1117 		err = request_firmware(&adev->gfx.mec_fw, fw_name, adev->dev);
1118 		if (err == -ENOENT) {
1119 			snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec.bin", chip_name);
1120 			err = request_firmware(&adev->gfx.mec_fw, fw_name, adev->dev);
1121 		}
1122 	} else {
1123 		snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec.bin", chip_name);
1124 		err = request_firmware(&adev->gfx.mec_fw, fw_name, adev->dev);
1125 	}
1126 	if (err)
1127 		goto out;
1128 	err = amdgpu_ucode_validate(adev->gfx.mec_fw);
1129 	if (err)
1130 		goto out;
1131 	cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data;
1132 	adev->gfx.mec_fw_version = le32_to_cpu(cp_hdr->header.ucode_version);
1133 	adev->gfx.mec_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version);
1134 
1135 	if ((adev->asic_type != CHIP_STONEY) &&
1136 	    (adev->asic_type != CHIP_TOPAZ)) {
1137 		if (adev->asic_type >= CHIP_POLARIS10 && adev->asic_type <= CHIP_POLARIS12) {
1138 			snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec2_2.bin", chip_name);
1139 			err = request_firmware(&adev->gfx.mec2_fw, fw_name, adev->dev);
1140 			if (err == -ENOENT) {
1141 				snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec2.bin", chip_name);
1142 				err = request_firmware(&adev->gfx.mec2_fw, fw_name, adev->dev);
1143 			}
1144 		} else {
1145 			snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec2.bin", chip_name);
1146 			err = request_firmware(&adev->gfx.mec2_fw, fw_name, adev->dev);
1147 		}
1148 		if (!err) {
1149 			err = amdgpu_ucode_validate(adev->gfx.mec2_fw);
1150 			if (err)
1151 				goto out;
1152 			cp_hdr = (const struct gfx_firmware_header_v1_0 *)
1153 				adev->gfx.mec2_fw->data;
1154 			adev->gfx.mec2_fw_version =
1155 				le32_to_cpu(cp_hdr->header.ucode_version);
1156 			adev->gfx.mec2_feature_version =
1157 				le32_to_cpu(cp_hdr->ucode_feature_version);
1158 		} else {
1159 			err = 0;
1160 			adev->gfx.mec2_fw = NULL;
1161 		}
1162 	}
1163 
1164 	info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_PFP];
1165 	info->ucode_id = AMDGPU_UCODE_ID_CP_PFP;
1166 	info->fw = adev->gfx.pfp_fw;
1167 	header = (const struct common_firmware_header *)info->fw->data;
1168 	adev->firmware.fw_size +=
1169 		ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
1170 
1171 	info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_ME];
1172 	info->ucode_id = AMDGPU_UCODE_ID_CP_ME;
1173 	info->fw = adev->gfx.me_fw;
1174 	header = (const struct common_firmware_header *)info->fw->data;
1175 	adev->firmware.fw_size +=
1176 		ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
1177 
1178 	info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_CE];
1179 	info->ucode_id = AMDGPU_UCODE_ID_CP_CE;
1180 	info->fw = adev->gfx.ce_fw;
1181 	header = (const struct common_firmware_header *)info->fw->data;
1182 	adev->firmware.fw_size +=
1183 		ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
1184 
1185 	info = &adev->firmware.ucode[AMDGPU_UCODE_ID_RLC_G];
1186 	info->ucode_id = AMDGPU_UCODE_ID_RLC_G;
1187 	info->fw = adev->gfx.rlc_fw;
1188 	header = (const struct common_firmware_header *)info->fw->data;
1189 	adev->firmware.fw_size +=
1190 		ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
1191 
1192 	info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_MEC1];
1193 	info->ucode_id = AMDGPU_UCODE_ID_CP_MEC1;
1194 	info->fw = adev->gfx.mec_fw;
1195 	header = (const struct common_firmware_header *)info->fw->data;
1196 	adev->firmware.fw_size +=
1197 		ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
1198 
1199 	/* we need account JT in */
1200 	cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data;
1201 	adev->firmware.fw_size +=
1202 		ALIGN(le32_to_cpu(cp_hdr->jt_size) << 2, PAGE_SIZE);
1203 
1204 	if (amdgpu_sriov_vf(adev)) {
1205 		info = &adev->firmware.ucode[AMDGPU_UCODE_ID_STORAGE];
1206 		info->ucode_id = AMDGPU_UCODE_ID_STORAGE;
1207 		info->fw = adev->gfx.mec_fw;
1208 		adev->firmware.fw_size +=
1209 			ALIGN(le32_to_cpu(64 * PAGE_SIZE), PAGE_SIZE);
1210 	}
1211 
1212 	if (adev->gfx.mec2_fw) {
1213 		info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_MEC2];
1214 		info->ucode_id = AMDGPU_UCODE_ID_CP_MEC2;
1215 		info->fw = adev->gfx.mec2_fw;
1216 		header = (const struct common_firmware_header *)info->fw->data;
1217 		adev->firmware.fw_size +=
1218 			ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
1219 	}
1220 
1221 out:
1222 	if (err) {
1223 		dev_err(adev->dev,
1224 			"gfx8: Failed to load firmware \"%s\"\n",
1225 			fw_name);
1226 		release_firmware(adev->gfx.pfp_fw);
1227 		adev->gfx.pfp_fw = NULL;
1228 		release_firmware(adev->gfx.me_fw);
1229 		adev->gfx.me_fw = NULL;
1230 		release_firmware(adev->gfx.ce_fw);
1231 		adev->gfx.ce_fw = NULL;
1232 		release_firmware(adev->gfx.rlc_fw);
1233 		adev->gfx.rlc_fw = NULL;
1234 		release_firmware(adev->gfx.mec_fw);
1235 		adev->gfx.mec_fw = NULL;
1236 		release_firmware(adev->gfx.mec2_fw);
1237 		adev->gfx.mec2_fw = NULL;
1238 	}
1239 	return err;
1240 }
1241 
1242 static void gfx_v8_0_get_csb_buffer(struct amdgpu_device *adev,
1243 				    volatile u32 *buffer)
1244 {
1245 	u32 count = 0, i;
1246 	const struct cs_section_def *sect = NULL;
1247 	const struct cs_extent_def *ext = NULL;
1248 
1249 	if (adev->gfx.rlc.cs_data == NULL)
1250 		return;
1251 	if (buffer == NULL)
1252 		return;
1253 
1254 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0));
1255 	buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
1256 
1257 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CONTEXT_CONTROL, 1));
1258 	buffer[count++] = cpu_to_le32(0x80000000);
1259 	buffer[count++] = cpu_to_le32(0x80000000);
1260 
1261 	for (sect = adev->gfx.rlc.cs_data; sect->section != NULL; ++sect) {
1262 		for (ext = sect->section; ext->extent != NULL; ++ext) {
1263 			if (sect->id == SECT_CONTEXT) {
1264 				buffer[count++] =
1265 					cpu_to_le32(PACKET3(PACKET3_SET_CONTEXT_REG, ext->reg_count));
1266 				buffer[count++] = cpu_to_le32(ext->reg_index -
1267 						PACKET3_SET_CONTEXT_REG_START);
1268 				for (i = 0; i < ext->reg_count; i++)
1269 					buffer[count++] = cpu_to_le32(ext->extent[i]);
1270 			} else {
1271 				return;
1272 			}
1273 		}
1274 	}
1275 
1276 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_SET_CONTEXT_REG, 2));
1277 	buffer[count++] = cpu_to_le32(mmPA_SC_RASTER_CONFIG -
1278 			PACKET3_SET_CONTEXT_REG_START);
1279 	buffer[count++] = cpu_to_le32(adev->gfx.config.rb_config[0][0].raster_config);
1280 	buffer[count++] = cpu_to_le32(adev->gfx.config.rb_config[0][0].raster_config_1);
1281 
1282 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0));
1283 	buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_END_CLEAR_STATE);
1284 
1285 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CLEAR_STATE, 0));
1286 	buffer[count++] = cpu_to_le32(0);
1287 }
1288 
1289 static int gfx_v8_0_cp_jump_table_num(struct amdgpu_device *adev)
1290 {
1291 	if (adev->asic_type == CHIP_CARRIZO)
1292 		return 5;
1293 	else
1294 		return 4;
1295 }
1296 
1297 static int gfx_v8_0_rlc_init(struct amdgpu_device *adev)
1298 {
1299 	const struct cs_section_def *cs_data;
1300 	int r;
1301 
1302 	adev->gfx.rlc.cs_data = vi_cs_data;
1303 
1304 	cs_data = adev->gfx.rlc.cs_data;
1305 
1306 	if (cs_data) {
1307 		/* init clear state block */
1308 		r = amdgpu_gfx_rlc_init_csb(adev);
1309 		if (r)
1310 			return r;
1311 	}
1312 
1313 	if ((adev->asic_type == CHIP_CARRIZO) ||
1314 	    (adev->asic_type == CHIP_STONEY)) {
1315 		adev->gfx.rlc.cp_table_size = ALIGN(96 * 5 * 4, 2048) + (64 * 1024); /* JT + GDS */
1316 		r = amdgpu_gfx_rlc_init_cpt(adev);
1317 		if (r)
1318 			return r;
1319 	}
1320 
1321 	return 0;
1322 }
1323 
1324 static void gfx_v8_0_mec_fini(struct amdgpu_device *adev)
1325 {
1326 	amdgpu_bo_free_kernel(&adev->gfx.mec.hpd_eop_obj, NULL, NULL);
1327 }
1328 
1329 static int gfx_v8_0_mec_init(struct amdgpu_device *adev)
1330 {
1331 	int r;
1332 	u32 *hpd;
1333 	size_t mec_hpd_size;
1334 
1335 	bitmap_zero(adev->gfx.mec.queue_bitmap, AMDGPU_MAX_COMPUTE_QUEUES);
1336 
1337 	/* take ownership of the relevant compute queues */
1338 	amdgpu_gfx_compute_queue_acquire(adev);
1339 
1340 	mec_hpd_size = adev->gfx.num_compute_rings * GFX8_MEC_HPD_SIZE;
1341 
1342 	r = amdgpu_bo_create_reserved(adev, mec_hpd_size, PAGE_SIZE,
1343 				      AMDGPU_GEM_DOMAIN_VRAM,
1344 				      &adev->gfx.mec.hpd_eop_obj,
1345 				      &adev->gfx.mec.hpd_eop_gpu_addr,
1346 				      (void **)&hpd);
1347 	if (r) {
1348 		dev_warn(adev->dev, "(%d) create HDP EOP bo failed\n", r);
1349 		return r;
1350 	}
1351 
1352 	memset(hpd, 0, mec_hpd_size);
1353 
1354 	amdgpu_bo_kunmap(adev->gfx.mec.hpd_eop_obj);
1355 	amdgpu_bo_unreserve(adev->gfx.mec.hpd_eop_obj);
1356 
1357 	return 0;
1358 }
1359 
1360 static const u32 vgpr_init_compute_shader[] =
1361 {
1362 	0x7e000209, 0x7e020208,
1363 	0x7e040207, 0x7e060206,
1364 	0x7e080205, 0x7e0a0204,
1365 	0x7e0c0203, 0x7e0e0202,
1366 	0x7e100201, 0x7e120200,
1367 	0x7e140209, 0x7e160208,
1368 	0x7e180207, 0x7e1a0206,
1369 	0x7e1c0205, 0x7e1e0204,
1370 	0x7e200203, 0x7e220202,
1371 	0x7e240201, 0x7e260200,
1372 	0x7e280209, 0x7e2a0208,
1373 	0x7e2c0207, 0x7e2e0206,
1374 	0x7e300205, 0x7e320204,
1375 	0x7e340203, 0x7e360202,
1376 	0x7e380201, 0x7e3a0200,
1377 	0x7e3c0209, 0x7e3e0208,
1378 	0x7e400207, 0x7e420206,
1379 	0x7e440205, 0x7e460204,
1380 	0x7e480203, 0x7e4a0202,
1381 	0x7e4c0201, 0x7e4e0200,
1382 	0x7e500209, 0x7e520208,
1383 	0x7e540207, 0x7e560206,
1384 	0x7e580205, 0x7e5a0204,
1385 	0x7e5c0203, 0x7e5e0202,
1386 	0x7e600201, 0x7e620200,
1387 	0x7e640209, 0x7e660208,
1388 	0x7e680207, 0x7e6a0206,
1389 	0x7e6c0205, 0x7e6e0204,
1390 	0x7e700203, 0x7e720202,
1391 	0x7e740201, 0x7e760200,
1392 	0x7e780209, 0x7e7a0208,
1393 	0x7e7c0207, 0x7e7e0206,
1394 	0xbf8a0000, 0xbf810000,
1395 };
1396 
1397 static const u32 sgpr_init_compute_shader[] =
1398 {
1399 	0xbe8a0100, 0xbe8c0102,
1400 	0xbe8e0104, 0xbe900106,
1401 	0xbe920108, 0xbe940100,
1402 	0xbe960102, 0xbe980104,
1403 	0xbe9a0106, 0xbe9c0108,
1404 	0xbe9e0100, 0xbea00102,
1405 	0xbea20104, 0xbea40106,
1406 	0xbea60108, 0xbea80100,
1407 	0xbeaa0102, 0xbeac0104,
1408 	0xbeae0106, 0xbeb00108,
1409 	0xbeb20100, 0xbeb40102,
1410 	0xbeb60104, 0xbeb80106,
1411 	0xbeba0108, 0xbebc0100,
1412 	0xbebe0102, 0xbec00104,
1413 	0xbec20106, 0xbec40108,
1414 	0xbec60100, 0xbec80102,
1415 	0xbee60004, 0xbee70005,
1416 	0xbeea0006, 0xbeeb0007,
1417 	0xbee80008, 0xbee90009,
1418 	0xbefc0000, 0xbf8a0000,
1419 	0xbf810000, 0x00000000,
1420 };
1421 
1422 static const u32 vgpr_init_regs[] =
1423 {
1424 	mmCOMPUTE_STATIC_THREAD_MGMT_SE0, 0xffffffff,
1425 	mmCOMPUTE_RESOURCE_LIMITS, 0x1000000, /* CU_GROUP_COUNT=1 */
1426 	mmCOMPUTE_NUM_THREAD_X, 256*4,
1427 	mmCOMPUTE_NUM_THREAD_Y, 1,
1428 	mmCOMPUTE_NUM_THREAD_Z, 1,
1429 	mmCOMPUTE_PGM_RSRC1, 0x100004f, /* VGPRS=15 (64 logical VGPRs), SGPRS=1 (16 SGPRs), BULKY=1 */
1430 	mmCOMPUTE_PGM_RSRC2, 20,
1431 	mmCOMPUTE_USER_DATA_0, 0xedcedc00,
1432 	mmCOMPUTE_USER_DATA_1, 0xedcedc01,
1433 	mmCOMPUTE_USER_DATA_2, 0xedcedc02,
1434 	mmCOMPUTE_USER_DATA_3, 0xedcedc03,
1435 	mmCOMPUTE_USER_DATA_4, 0xedcedc04,
1436 	mmCOMPUTE_USER_DATA_5, 0xedcedc05,
1437 	mmCOMPUTE_USER_DATA_6, 0xedcedc06,
1438 	mmCOMPUTE_USER_DATA_7, 0xedcedc07,
1439 	mmCOMPUTE_USER_DATA_8, 0xedcedc08,
1440 	mmCOMPUTE_USER_DATA_9, 0xedcedc09,
1441 };
1442 
1443 static const u32 sgpr1_init_regs[] =
1444 {
1445 	mmCOMPUTE_STATIC_THREAD_MGMT_SE0, 0x0f,
1446 	mmCOMPUTE_RESOURCE_LIMITS, 0x1000000, /* CU_GROUP_COUNT=1 */
1447 	mmCOMPUTE_NUM_THREAD_X, 256*5,
1448 	mmCOMPUTE_NUM_THREAD_Y, 1,
1449 	mmCOMPUTE_NUM_THREAD_Z, 1,
1450 	mmCOMPUTE_PGM_RSRC1, 0x240, /* SGPRS=9 (80 GPRS) */
1451 	mmCOMPUTE_PGM_RSRC2, 20,
1452 	mmCOMPUTE_USER_DATA_0, 0xedcedc00,
1453 	mmCOMPUTE_USER_DATA_1, 0xedcedc01,
1454 	mmCOMPUTE_USER_DATA_2, 0xedcedc02,
1455 	mmCOMPUTE_USER_DATA_3, 0xedcedc03,
1456 	mmCOMPUTE_USER_DATA_4, 0xedcedc04,
1457 	mmCOMPUTE_USER_DATA_5, 0xedcedc05,
1458 	mmCOMPUTE_USER_DATA_6, 0xedcedc06,
1459 	mmCOMPUTE_USER_DATA_7, 0xedcedc07,
1460 	mmCOMPUTE_USER_DATA_8, 0xedcedc08,
1461 	mmCOMPUTE_USER_DATA_9, 0xedcedc09,
1462 };
1463 
1464 static const u32 sgpr2_init_regs[] =
1465 {
1466 	mmCOMPUTE_STATIC_THREAD_MGMT_SE0, 0xf0,
1467 	mmCOMPUTE_RESOURCE_LIMITS, 0x1000000,
1468 	mmCOMPUTE_NUM_THREAD_X, 256*5,
1469 	mmCOMPUTE_NUM_THREAD_Y, 1,
1470 	mmCOMPUTE_NUM_THREAD_Z, 1,
1471 	mmCOMPUTE_PGM_RSRC1, 0x240, /* SGPRS=9 (80 GPRS) */
1472 	mmCOMPUTE_PGM_RSRC2, 20,
1473 	mmCOMPUTE_USER_DATA_0, 0xedcedc00,
1474 	mmCOMPUTE_USER_DATA_1, 0xedcedc01,
1475 	mmCOMPUTE_USER_DATA_2, 0xedcedc02,
1476 	mmCOMPUTE_USER_DATA_3, 0xedcedc03,
1477 	mmCOMPUTE_USER_DATA_4, 0xedcedc04,
1478 	mmCOMPUTE_USER_DATA_5, 0xedcedc05,
1479 	mmCOMPUTE_USER_DATA_6, 0xedcedc06,
1480 	mmCOMPUTE_USER_DATA_7, 0xedcedc07,
1481 	mmCOMPUTE_USER_DATA_8, 0xedcedc08,
1482 	mmCOMPUTE_USER_DATA_9, 0xedcedc09,
1483 };
1484 
1485 static const u32 sec_ded_counter_registers[] =
1486 {
1487 	mmCPC_EDC_ATC_CNT,
1488 	mmCPC_EDC_SCRATCH_CNT,
1489 	mmCPC_EDC_UCODE_CNT,
1490 	mmCPF_EDC_ATC_CNT,
1491 	mmCPF_EDC_ROQ_CNT,
1492 	mmCPF_EDC_TAG_CNT,
1493 	mmCPG_EDC_ATC_CNT,
1494 	mmCPG_EDC_DMA_CNT,
1495 	mmCPG_EDC_TAG_CNT,
1496 	mmDC_EDC_CSINVOC_CNT,
1497 	mmDC_EDC_RESTORE_CNT,
1498 	mmDC_EDC_STATE_CNT,
1499 	mmGDS_EDC_CNT,
1500 	mmGDS_EDC_GRBM_CNT,
1501 	mmGDS_EDC_OA_DED,
1502 	mmSPI_EDC_CNT,
1503 	mmSQC_ATC_EDC_GATCL1_CNT,
1504 	mmSQC_EDC_CNT,
1505 	mmSQ_EDC_DED_CNT,
1506 	mmSQ_EDC_INFO,
1507 	mmSQ_EDC_SEC_CNT,
1508 	mmTCC_EDC_CNT,
1509 	mmTCP_ATC_EDC_GATCL1_CNT,
1510 	mmTCP_EDC_CNT,
1511 	mmTD_EDC_CNT
1512 };
1513 
1514 static int gfx_v8_0_do_edc_gpr_workarounds(struct amdgpu_device *adev)
1515 {
1516 	struct amdgpu_ring *ring = &adev->gfx.compute_ring[0];
1517 	struct amdgpu_ib ib;
1518 	struct dma_fence *f = NULL;
1519 	int r, i;
1520 	u32 tmp;
1521 	unsigned total_size, vgpr_offset, sgpr_offset;
1522 	u64 gpu_addr;
1523 
1524 	/* only supported on CZ */
1525 	if (adev->asic_type != CHIP_CARRIZO)
1526 		return 0;
1527 
1528 	/* bail if the compute ring is not ready */
1529 	if (!ring->sched.ready)
1530 		return 0;
1531 
1532 	tmp = RREG32(mmGB_EDC_MODE);
1533 	WREG32(mmGB_EDC_MODE, 0);
1534 
1535 	total_size =
1536 		(((ARRAY_SIZE(vgpr_init_regs) / 2) * 3) + 4 + 5 + 2) * 4;
1537 	total_size +=
1538 		(((ARRAY_SIZE(sgpr1_init_regs) / 2) * 3) + 4 + 5 + 2) * 4;
1539 	total_size +=
1540 		(((ARRAY_SIZE(sgpr2_init_regs) / 2) * 3) + 4 + 5 + 2) * 4;
1541 	total_size = ALIGN(total_size, 256);
1542 	vgpr_offset = total_size;
1543 	total_size += ALIGN(sizeof(vgpr_init_compute_shader), 256);
1544 	sgpr_offset = total_size;
1545 	total_size += sizeof(sgpr_init_compute_shader);
1546 
1547 	/* allocate an indirect buffer to put the commands in */
1548 	memset(&ib, 0, sizeof(ib));
1549 	r = amdgpu_ib_get(adev, NULL, total_size, &ib);
1550 	if (r) {
1551 		DRM_ERROR("amdgpu: failed to get ib (%d).\n", r);
1552 		return r;
1553 	}
1554 
1555 	/* load the compute shaders */
1556 	for (i = 0; i < ARRAY_SIZE(vgpr_init_compute_shader); i++)
1557 		ib.ptr[i + (vgpr_offset / 4)] = vgpr_init_compute_shader[i];
1558 
1559 	for (i = 0; i < ARRAY_SIZE(sgpr_init_compute_shader); i++)
1560 		ib.ptr[i + (sgpr_offset / 4)] = sgpr_init_compute_shader[i];
1561 
1562 	/* init the ib length to 0 */
1563 	ib.length_dw = 0;
1564 
1565 	/* VGPR */
1566 	/* write the register state for the compute dispatch */
1567 	for (i = 0; i < ARRAY_SIZE(vgpr_init_regs); i += 2) {
1568 		ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 1);
1569 		ib.ptr[ib.length_dw++] = vgpr_init_regs[i] - PACKET3_SET_SH_REG_START;
1570 		ib.ptr[ib.length_dw++] = vgpr_init_regs[i + 1];
1571 	}
1572 	/* write the shader start address: mmCOMPUTE_PGM_LO, mmCOMPUTE_PGM_HI */
1573 	gpu_addr = (ib.gpu_addr + (u64)vgpr_offset) >> 8;
1574 	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 2);
1575 	ib.ptr[ib.length_dw++] = mmCOMPUTE_PGM_LO - PACKET3_SET_SH_REG_START;
1576 	ib.ptr[ib.length_dw++] = lower_32_bits(gpu_addr);
1577 	ib.ptr[ib.length_dw++] = upper_32_bits(gpu_addr);
1578 
1579 	/* write dispatch packet */
1580 	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_DISPATCH_DIRECT, 3);
1581 	ib.ptr[ib.length_dw++] = 8; /* x */
1582 	ib.ptr[ib.length_dw++] = 1; /* y */
1583 	ib.ptr[ib.length_dw++] = 1; /* z */
1584 	ib.ptr[ib.length_dw++] =
1585 		REG_SET_FIELD(0, COMPUTE_DISPATCH_INITIATOR, COMPUTE_SHADER_EN, 1);
1586 
1587 	/* write CS partial flush packet */
1588 	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_EVENT_WRITE, 0);
1589 	ib.ptr[ib.length_dw++] = EVENT_TYPE(7) | EVENT_INDEX(4);
1590 
1591 	/* SGPR1 */
1592 	/* write the register state for the compute dispatch */
1593 	for (i = 0; i < ARRAY_SIZE(sgpr1_init_regs); i += 2) {
1594 		ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 1);
1595 		ib.ptr[ib.length_dw++] = sgpr1_init_regs[i] - PACKET3_SET_SH_REG_START;
1596 		ib.ptr[ib.length_dw++] = sgpr1_init_regs[i + 1];
1597 	}
1598 	/* write the shader start address: mmCOMPUTE_PGM_LO, mmCOMPUTE_PGM_HI */
1599 	gpu_addr = (ib.gpu_addr + (u64)sgpr_offset) >> 8;
1600 	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 2);
1601 	ib.ptr[ib.length_dw++] = mmCOMPUTE_PGM_LO - PACKET3_SET_SH_REG_START;
1602 	ib.ptr[ib.length_dw++] = lower_32_bits(gpu_addr);
1603 	ib.ptr[ib.length_dw++] = upper_32_bits(gpu_addr);
1604 
1605 	/* write dispatch packet */
1606 	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_DISPATCH_DIRECT, 3);
1607 	ib.ptr[ib.length_dw++] = 8; /* x */
1608 	ib.ptr[ib.length_dw++] = 1; /* y */
1609 	ib.ptr[ib.length_dw++] = 1; /* z */
1610 	ib.ptr[ib.length_dw++] =
1611 		REG_SET_FIELD(0, COMPUTE_DISPATCH_INITIATOR, COMPUTE_SHADER_EN, 1);
1612 
1613 	/* write CS partial flush packet */
1614 	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_EVENT_WRITE, 0);
1615 	ib.ptr[ib.length_dw++] = EVENT_TYPE(7) | EVENT_INDEX(4);
1616 
1617 	/* SGPR2 */
1618 	/* write the register state for the compute dispatch */
1619 	for (i = 0; i < ARRAY_SIZE(sgpr2_init_regs); i += 2) {
1620 		ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 1);
1621 		ib.ptr[ib.length_dw++] = sgpr2_init_regs[i] - PACKET3_SET_SH_REG_START;
1622 		ib.ptr[ib.length_dw++] = sgpr2_init_regs[i + 1];
1623 	}
1624 	/* write the shader start address: mmCOMPUTE_PGM_LO, mmCOMPUTE_PGM_HI */
1625 	gpu_addr = (ib.gpu_addr + (u64)sgpr_offset) >> 8;
1626 	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 2);
1627 	ib.ptr[ib.length_dw++] = mmCOMPUTE_PGM_LO - PACKET3_SET_SH_REG_START;
1628 	ib.ptr[ib.length_dw++] = lower_32_bits(gpu_addr);
1629 	ib.ptr[ib.length_dw++] = upper_32_bits(gpu_addr);
1630 
1631 	/* write dispatch packet */
1632 	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_DISPATCH_DIRECT, 3);
1633 	ib.ptr[ib.length_dw++] = 8; /* x */
1634 	ib.ptr[ib.length_dw++] = 1; /* y */
1635 	ib.ptr[ib.length_dw++] = 1; /* z */
1636 	ib.ptr[ib.length_dw++] =
1637 		REG_SET_FIELD(0, COMPUTE_DISPATCH_INITIATOR, COMPUTE_SHADER_EN, 1);
1638 
1639 	/* write CS partial flush packet */
1640 	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_EVENT_WRITE, 0);
1641 	ib.ptr[ib.length_dw++] = EVENT_TYPE(7) | EVENT_INDEX(4);
1642 
1643 	/* shedule the ib on the ring */
1644 	r = amdgpu_ib_schedule(ring, 1, &ib, NULL, &f);
1645 	if (r) {
1646 		DRM_ERROR("amdgpu: ib submit failed (%d).\n", r);
1647 		goto fail;
1648 	}
1649 
1650 	/* wait for the GPU to finish processing the IB */
1651 	r = dma_fence_wait(f, false);
1652 	if (r) {
1653 		DRM_ERROR("amdgpu: fence wait failed (%d).\n", r);
1654 		goto fail;
1655 	}
1656 
1657 	tmp = REG_SET_FIELD(tmp, GB_EDC_MODE, DED_MODE, 2);
1658 	tmp = REG_SET_FIELD(tmp, GB_EDC_MODE, PROP_FED, 1);
1659 	WREG32(mmGB_EDC_MODE, tmp);
1660 
1661 	tmp = RREG32(mmCC_GC_EDC_CONFIG);
1662 	tmp = REG_SET_FIELD(tmp, CC_GC_EDC_CONFIG, DIS_EDC, 0) | 1;
1663 	WREG32(mmCC_GC_EDC_CONFIG, tmp);
1664 
1665 
1666 	/* read back registers to clear the counters */
1667 	for (i = 0; i < ARRAY_SIZE(sec_ded_counter_registers); i++)
1668 		RREG32(sec_ded_counter_registers[i]);
1669 
1670 fail:
1671 	amdgpu_ib_free(adev, &ib, NULL);
1672 	dma_fence_put(f);
1673 
1674 	return r;
1675 }
1676 
1677 static int gfx_v8_0_gpu_early_init(struct amdgpu_device *adev)
1678 {
1679 	u32 gb_addr_config;
1680 	u32 mc_arb_ramcfg;
1681 	u32 dimm00_addr_map, dimm01_addr_map, dimm10_addr_map, dimm11_addr_map;
1682 	u32 tmp;
1683 	int ret;
1684 
1685 	switch (adev->asic_type) {
1686 	case CHIP_TOPAZ:
1687 		adev->gfx.config.max_shader_engines = 1;
1688 		adev->gfx.config.max_tile_pipes = 2;
1689 		adev->gfx.config.max_cu_per_sh = 6;
1690 		adev->gfx.config.max_sh_per_se = 1;
1691 		adev->gfx.config.max_backends_per_se = 2;
1692 		adev->gfx.config.max_texture_channel_caches = 2;
1693 		adev->gfx.config.max_gprs = 256;
1694 		adev->gfx.config.max_gs_threads = 32;
1695 		adev->gfx.config.max_hw_contexts = 8;
1696 
1697 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1698 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1699 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1700 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
1701 		gb_addr_config = TOPAZ_GB_ADDR_CONFIG_GOLDEN;
1702 		break;
1703 	case CHIP_FIJI:
1704 		adev->gfx.config.max_shader_engines = 4;
1705 		adev->gfx.config.max_tile_pipes = 16;
1706 		adev->gfx.config.max_cu_per_sh = 16;
1707 		adev->gfx.config.max_sh_per_se = 1;
1708 		adev->gfx.config.max_backends_per_se = 4;
1709 		adev->gfx.config.max_texture_channel_caches = 16;
1710 		adev->gfx.config.max_gprs = 256;
1711 		adev->gfx.config.max_gs_threads = 32;
1712 		adev->gfx.config.max_hw_contexts = 8;
1713 
1714 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1715 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1716 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1717 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
1718 		gb_addr_config = TONGA_GB_ADDR_CONFIG_GOLDEN;
1719 		break;
1720 	case CHIP_POLARIS11:
1721 	case CHIP_POLARIS12:
1722 		ret = amdgpu_atombios_get_gfx_info(adev);
1723 		if (ret)
1724 			return ret;
1725 		adev->gfx.config.max_gprs = 256;
1726 		adev->gfx.config.max_gs_threads = 32;
1727 		adev->gfx.config.max_hw_contexts = 8;
1728 
1729 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1730 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1731 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1732 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
1733 		gb_addr_config = POLARIS11_GB_ADDR_CONFIG_GOLDEN;
1734 		break;
1735 	case CHIP_POLARIS10:
1736 	case CHIP_VEGAM:
1737 		ret = amdgpu_atombios_get_gfx_info(adev);
1738 		if (ret)
1739 			return ret;
1740 		adev->gfx.config.max_gprs = 256;
1741 		adev->gfx.config.max_gs_threads = 32;
1742 		adev->gfx.config.max_hw_contexts = 8;
1743 
1744 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1745 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1746 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1747 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
1748 		gb_addr_config = TONGA_GB_ADDR_CONFIG_GOLDEN;
1749 		break;
1750 	case CHIP_TONGA:
1751 		adev->gfx.config.max_shader_engines = 4;
1752 		adev->gfx.config.max_tile_pipes = 8;
1753 		adev->gfx.config.max_cu_per_sh = 8;
1754 		adev->gfx.config.max_sh_per_se = 1;
1755 		adev->gfx.config.max_backends_per_se = 2;
1756 		adev->gfx.config.max_texture_channel_caches = 8;
1757 		adev->gfx.config.max_gprs = 256;
1758 		adev->gfx.config.max_gs_threads = 32;
1759 		adev->gfx.config.max_hw_contexts = 8;
1760 
1761 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1762 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1763 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1764 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
1765 		gb_addr_config = TONGA_GB_ADDR_CONFIG_GOLDEN;
1766 		break;
1767 	case CHIP_CARRIZO:
1768 		adev->gfx.config.max_shader_engines = 1;
1769 		adev->gfx.config.max_tile_pipes = 2;
1770 		adev->gfx.config.max_sh_per_se = 1;
1771 		adev->gfx.config.max_backends_per_se = 2;
1772 		adev->gfx.config.max_cu_per_sh = 8;
1773 		adev->gfx.config.max_texture_channel_caches = 2;
1774 		adev->gfx.config.max_gprs = 256;
1775 		adev->gfx.config.max_gs_threads = 32;
1776 		adev->gfx.config.max_hw_contexts = 8;
1777 
1778 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1779 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1780 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1781 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
1782 		gb_addr_config = CARRIZO_GB_ADDR_CONFIG_GOLDEN;
1783 		break;
1784 	case CHIP_STONEY:
1785 		adev->gfx.config.max_shader_engines = 1;
1786 		adev->gfx.config.max_tile_pipes = 2;
1787 		adev->gfx.config.max_sh_per_se = 1;
1788 		adev->gfx.config.max_backends_per_se = 1;
1789 		adev->gfx.config.max_cu_per_sh = 3;
1790 		adev->gfx.config.max_texture_channel_caches = 2;
1791 		adev->gfx.config.max_gprs = 256;
1792 		adev->gfx.config.max_gs_threads = 16;
1793 		adev->gfx.config.max_hw_contexts = 8;
1794 
1795 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1796 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1797 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1798 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
1799 		gb_addr_config = CARRIZO_GB_ADDR_CONFIG_GOLDEN;
1800 		break;
1801 	default:
1802 		adev->gfx.config.max_shader_engines = 2;
1803 		adev->gfx.config.max_tile_pipes = 4;
1804 		adev->gfx.config.max_cu_per_sh = 2;
1805 		adev->gfx.config.max_sh_per_se = 1;
1806 		adev->gfx.config.max_backends_per_se = 2;
1807 		adev->gfx.config.max_texture_channel_caches = 4;
1808 		adev->gfx.config.max_gprs = 256;
1809 		adev->gfx.config.max_gs_threads = 32;
1810 		adev->gfx.config.max_hw_contexts = 8;
1811 
1812 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1813 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1814 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1815 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
1816 		gb_addr_config = TONGA_GB_ADDR_CONFIG_GOLDEN;
1817 		break;
1818 	}
1819 
1820 	adev->gfx.config.mc_arb_ramcfg = RREG32(mmMC_ARB_RAMCFG);
1821 	mc_arb_ramcfg = adev->gfx.config.mc_arb_ramcfg;
1822 
1823 	adev->gfx.config.num_tile_pipes = adev->gfx.config.max_tile_pipes;
1824 	adev->gfx.config.mem_max_burst_length_bytes = 256;
1825 	if (adev->flags & AMD_IS_APU) {
1826 		/* Get memory bank mapping mode. */
1827 		tmp = RREG32(mmMC_FUS_DRAM0_BANK_ADDR_MAPPING);
1828 		dimm00_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM0_BANK_ADDR_MAPPING, DIMM0ADDRMAP);
1829 		dimm01_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM0_BANK_ADDR_MAPPING, DIMM1ADDRMAP);
1830 
1831 		tmp = RREG32(mmMC_FUS_DRAM1_BANK_ADDR_MAPPING);
1832 		dimm10_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM1_BANK_ADDR_MAPPING, DIMM0ADDRMAP);
1833 		dimm11_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM1_BANK_ADDR_MAPPING, DIMM1ADDRMAP);
1834 
1835 		/* Validate settings in case only one DIMM installed. */
1836 		if ((dimm00_addr_map == 0) || (dimm00_addr_map == 3) || (dimm00_addr_map == 4) || (dimm00_addr_map > 12))
1837 			dimm00_addr_map = 0;
1838 		if ((dimm01_addr_map == 0) || (dimm01_addr_map == 3) || (dimm01_addr_map == 4) || (dimm01_addr_map > 12))
1839 			dimm01_addr_map = 0;
1840 		if ((dimm10_addr_map == 0) || (dimm10_addr_map == 3) || (dimm10_addr_map == 4) || (dimm10_addr_map > 12))
1841 			dimm10_addr_map = 0;
1842 		if ((dimm11_addr_map == 0) || (dimm11_addr_map == 3) || (dimm11_addr_map == 4) || (dimm11_addr_map > 12))
1843 			dimm11_addr_map = 0;
1844 
1845 		/* If DIMM Addr map is 8GB, ROW size should be 2KB. Otherwise 1KB. */
1846 		/* If ROW size(DIMM1) != ROW size(DMIMM0), ROW size should be larger one. */
1847 		if ((dimm00_addr_map == 11) || (dimm01_addr_map == 11) || (dimm10_addr_map == 11) || (dimm11_addr_map == 11))
1848 			adev->gfx.config.mem_row_size_in_kb = 2;
1849 		else
1850 			adev->gfx.config.mem_row_size_in_kb = 1;
1851 	} else {
1852 		tmp = REG_GET_FIELD(mc_arb_ramcfg, MC_ARB_RAMCFG, NOOFCOLS);
1853 		adev->gfx.config.mem_row_size_in_kb = (4 * (1 << (8 + tmp))) / 1024;
1854 		if (adev->gfx.config.mem_row_size_in_kb > 4)
1855 			adev->gfx.config.mem_row_size_in_kb = 4;
1856 	}
1857 
1858 	adev->gfx.config.shader_engine_tile_size = 32;
1859 	adev->gfx.config.num_gpus = 1;
1860 	adev->gfx.config.multi_gpu_tile_size = 64;
1861 
1862 	/* fix up row size */
1863 	switch (adev->gfx.config.mem_row_size_in_kb) {
1864 	case 1:
1865 	default:
1866 		gb_addr_config = REG_SET_FIELD(gb_addr_config, GB_ADDR_CONFIG, ROW_SIZE, 0);
1867 		break;
1868 	case 2:
1869 		gb_addr_config = REG_SET_FIELD(gb_addr_config, GB_ADDR_CONFIG, ROW_SIZE, 1);
1870 		break;
1871 	case 4:
1872 		gb_addr_config = REG_SET_FIELD(gb_addr_config, GB_ADDR_CONFIG, ROW_SIZE, 2);
1873 		break;
1874 	}
1875 	adev->gfx.config.gb_addr_config = gb_addr_config;
1876 
1877 	return 0;
1878 }
1879 
1880 static int gfx_v8_0_compute_ring_init(struct amdgpu_device *adev, int ring_id,
1881 					int mec, int pipe, int queue)
1882 {
1883 	int r;
1884 	unsigned irq_type;
1885 	struct amdgpu_ring *ring = &adev->gfx.compute_ring[ring_id];
1886 
1887 	ring = &adev->gfx.compute_ring[ring_id];
1888 
1889 	/* mec0 is me1 */
1890 	ring->me = mec + 1;
1891 	ring->pipe = pipe;
1892 	ring->queue = queue;
1893 
1894 	ring->ring_obj = NULL;
1895 	ring->use_doorbell = true;
1896 	ring->doorbell_index = adev->doorbell_index.mec_ring0 + ring_id;
1897 	ring->eop_gpu_addr = adev->gfx.mec.hpd_eop_gpu_addr
1898 				+ (ring_id * GFX8_MEC_HPD_SIZE);
1899 	sprintf(ring->name, "comp_%d.%d.%d", ring->me, ring->pipe, ring->queue);
1900 
1901 	irq_type = AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP
1902 		+ ((ring->me - 1) * adev->gfx.mec.num_pipe_per_mec)
1903 		+ ring->pipe;
1904 
1905 	/* type-2 packets are deprecated on MEC, use type-3 instead */
1906 	r = amdgpu_ring_init(adev, ring, 1024,
1907 			&adev->gfx.eop_irq, irq_type);
1908 	if (r)
1909 		return r;
1910 
1911 
1912 	return 0;
1913 }
1914 
1915 static void gfx_v8_0_sq_irq_work_func(struct work_struct *work);
1916 
1917 static int gfx_v8_0_sw_init(void *handle)
1918 {
1919 	int i, j, k, r, ring_id;
1920 	struct amdgpu_ring *ring;
1921 	struct amdgpu_kiq *kiq;
1922 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1923 
1924 	switch (adev->asic_type) {
1925 	case CHIP_TONGA:
1926 	case CHIP_CARRIZO:
1927 	case CHIP_FIJI:
1928 	case CHIP_POLARIS10:
1929 	case CHIP_POLARIS11:
1930 	case CHIP_POLARIS12:
1931 	case CHIP_VEGAM:
1932 		adev->gfx.mec.num_mec = 2;
1933 		break;
1934 	case CHIP_TOPAZ:
1935 	case CHIP_STONEY:
1936 	default:
1937 		adev->gfx.mec.num_mec = 1;
1938 		break;
1939 	}
1940 
1941 	adev->gfx.mec.num_pipe_per_mec = 4;
1942 	adev->gfx.mec.num_queue_per_pipe = 8;
1943 
1944 	/* EOP Event */
1945 	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, VISLANDS30_IV_SRCID_CP_END_OF_PIPE, &adev->gfx.eop_irq);
1946 	if (r)
1947 		return r;
1948 
1949 	/* Privileged reg */
1950 	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, VISLANDS30_IV_SRCID_CP_PRIV_REG_FAULT,
1951 			      &adev->gfx.priv_reg_irq);
1952 	if (r)
1953 		return r;
1954 
1955 	/* Privileged inst */
1956 	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, VISLANDS30_IV_SRCID_CP_PRIV_INSTR_FAULT,
1957 			      &adev->gfx.priv_inst_irq);
1958 	if (r)
1959 		return r;
1960 
1961 	/* Add CP EDC/ECC irq  */
1962 	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, VISLANDS30_IV_SRCID_CP_ECC_ERROR,
1963 			      &adev->gfx.cp_ecc_error_irq);
1964 	if (r)
1965 		return r;
1966 
1967 	/* SQ interrupts. */
1968 	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, VISLANDS30_IV_SRCID_SQ_INTERRUPT_MSG,
1969 			      &adev->gfx.sq_irq);
1970 	if (r) {
1971 		DRM_ERROR("amdgpu_irq_add() for SQ failed: %d\n", r);
1972 		return r;
1973 	}
1974 
1975 	INIT_WORK(&adev->gfx.sq_work.work, gfx_v8_0_sq_irq_work_func);
1976 
1977 	adev->gfx.gfx_current_status = AMDGPU_GFX_NORMAL_MODE;
1978 
1979 	gfx_v8_0_scratch_init(adev);
1980 
1981 	r = gfx_v8_0_init_microcode(adev);
1982 	if (r) {
1983 		DRM_ERROR("Failed to load gfx firmware!\n");
1984 		return r;
1985 	}
1986 
1987 	r = adev->gfx.rlc.funcs->init(adev);
1988 	if (r) {
1989 		DRM_ERROR("Failed to init rlc BOs!\n");
1990 		return r;
1991 	}
1992 
1993 	r = gfx_v8_0_mec_init(adev);
1994 	if (r) {
1995 		DRM_ERROR("Failed to init MEC BOs!\n");
1996 		return r;
1997 	}
1998 
1999 	/* set up the gfx ring */
2000 	for (i = 0; i < adev->gfx.num_gfx_rings; i++) {
2001 		ring = &adev->gfx.gfx_ring[i];
2002 		ring->ring_obj = NULL;
2003 		sprintf(ring->name, "gfx");
2004 		/* no gfx doorbells on iceland */
2005 		if (adev->asic_type != CHIP_TOPAZ) {
2006 			ring->use_doorbell = true;
2007 			ring->doorbell_index = adev->doorbell_index.gfx_ring0;
2008 		}
2009 
2010 		r = amdgpu_ring_init(adev, ring, 1024, &adev->gfx.eop_irq,
2011 				     AMDGPU_CP_IRQ_GFX_ME0_PIPE0_EOP);
2012 		if (r)
2013 			return r;
2014 	}
2015 
2016 
2017 	/* set up the compute queues - allocate horizontally across pipes */
2018 	ring_id = 0;
2019 	for (i = 0; i < adev->gfx.mec.num_mec; ++i) {
2020 		for (j = 0; j < adev->gfx.mec.num_queue_per_pipe; j++) {
2021 			for (k = 0; k < adev->gfx.mec.num_pipe_per_mec; k++) {
2022 				if (!amdgpu_gfx_is_mec_queue_enabled(adev, i, k, j))
2023 					continue;
2024 
2025 				r = gfx_v8_0_compute_ring_init(adev,
2026 								ring_id,
2027 								i, k, j);
2028 				if (r)
2029 					return r;
2030 
2031 				ring_id++;
2032 			}
2033 		}
2034 	}
2035 
2036 	r = amdgpu_gfx_kiq_init(adev, GFX8_MEC_HPD_SIZE);
2037 	if (r) {
2038 		DRM_ERROR("Failed to init KIQ BOs!\n");
2039 		return r;
2040 	}
2041 
2042 	kiq = &adev->gfx.kiq;
2043 	r = amdgpu_gfx_kiq_init_ring(adev, &kiq->ring, &kiq->irq);
2044 	if (r)
2045 		return r;
2046 
2047 	/* create MQD for all compute queues as well as KIQ for SRIOV case */
2048 	r = amdgpu_gfx_mqd_sw_init(adev, sizeof(struct vi_mqd_allocation));
2049 	if (r)
2050 		return r;
2051 
2052 	adev->gfx.ce_ram_size = 0x8000;
2053 
2054 	r = gfx_v8_0_gpu_early_init(adev);
2055 	if (r)
2056 		return r;
2057 
2058 	return 0;
2059 }
2060 
2061 static int gfx_v8_0_sw_fini(void *handle)
2062 {
2063 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2064 	int i;
2065 
2066 	for (i = 0; i < adev->gfx.num_gfx_rings; i++)
2067 		amdgpu_ring_fini(&adev->gfx.gfx_ring[i]);
2068 	for (i = 0; i < adev->gfx.num_compute_rings; i++)
2069 		amdgpu_ring_fini(&adev->gfx.compute_ring[i]);
2070 
2071 	amdgpu_gfx_mqd_sw_fini(adev);
2072 	amdgpu_gfx_kiq_free_ring(&adev->gfx.kiq.ring);
2073 	amdgpu_gfx_kiq_fini(adev);
2074 
2075 	gfx_v8_0_mec_fini(adev);
2076 	amdgpu_gfx_rlc_fini(adev);
2077 	amdgpu_bo_free_kernel(&adev->gfx.rlc.clear_state_obj,
2078 				&adev->gfx.rlc.clear_state_gpu_addr,
2079 				(void **)&adev->gfx.rlc.cs_ptr);
2080 	if ((adev->asic_type == CHIP_CARRIZO) ||
2081 	    (adev->asic_type == CHIP_STONEY)) {
2082 		amdgpu_bo_free_kernel(&adev->gfx.rlc.cp_table_obj,
2083 				&adev->gfx.rlc.cp_table_gpu_addr,
2084 				(void **)&adev->gfx.rlc.cp_table_ptr);
2085 	}
2086 	gfx_v8_0_free_microcode(adev);
2087 
2088 	return 0;
2089 }
2090 
2091 static void gfx_v8_0_tiling_mode_table_init(struct amdgpu_device *adev)
2092 {
2093 	uint32_t *modearray, *mod2array;
2094 	const u32 num_tile_mode_states = ARRAY_SIZE(adev->gfx.config.tile_mode_array);
2095 	const u32 num_secondary_tile_mode_states = ARRAY_SIZE(adev->gfx.config.macrotile_mode_array);
2096 	u32 reg_offset;
2097 
2098 	modearray = adev->gfx.config.tile_mode_array;
2099 	mod2array = adev->gfx.config.macrotile_mode_array;
2100 
2101 	for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
2102 		modearray[reg_offset] = 0;
2103 
2104 	for (reg_offset = 0; reg_offset <  num_secondary_tile_mode_states; reg_offset++)
2105 		mod2array[reg_offset] = 0;
2106 
2107 	switch (adev->asic_type) {
2108 	case CHIP_TOPAZ:
2109 		modearray[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2110 				PIPE_CONFIG(ADDR_SURF_P2) |
2111 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
2112 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2113 		modearray[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2114 				PIPE_CONFIG(ADDR_SURF_P2) |
2115 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
2116 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2117 		modearray[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2118 				PIPE_CONFIG(ADDR_SURF_P2) |
2119 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
2120 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2121 		modearray[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2122 				PIPE_CONFIG(ADDR_SURF_P2) |
2123 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
2124 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2125 		modearray[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2126 				PIPE_CONFIG(ADDR_SURF_P2) |
2127 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2128 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2129 		modearray[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2130 				PIPE_CONFIG(ADDR_SURF_P2) |
2131 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2132 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2133 		modearray[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2134 				PIPE_CONFIG(ADDR_SURF_P2) |
2135 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2136 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2137 		modearray[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
2138 				PIPE_CONFIG(ADDR_SURF_P2));
2139 		modearray[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2140 				PIPE_CONFIG(ADDR_SURF_P2) |
2141 				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2142 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2143 		modearray[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2144 				 PIPE_CONFIG(ADDR_SURF_P2) |
2145 				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2146 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2147 		modearray[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2148 				 PIPE_CONFIG(ADDR_SURF_P2) |
2149 				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2150 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2151 		modearray[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2152 				 PIPE_CONFIG(ADDR_SURF_P2) |
2153 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2154 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2155 		modearray[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2156 				 PIPE_CONFIG(ADDR_SURF_P2) |
2157 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2158 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2159 		modearray[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
2160 				 PIPE_CONFIG(ADDR_SURF_P2) |
2161 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2162 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2163 		modearray[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2164 				 PIPE_CONFIG(ADDR_SURF_P2) |
2165 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2166 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2167 		modearray[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2168 				 PIPE_CONFIG(ADDR_SURF_P2) |
2169 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2170 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2171 		modearray[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2172 				 PIPE_CONFIG(ADDR_SURF_P2) |
2173 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2174 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2175 		modearray[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2176 				 PIPE_CONFIG(ADDR_SURF_P2) |
2177 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2178 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2179 		modearray[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
2180 				 PIPE_CONFIG(ADDR_SURF_P2) |
2181 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2182 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2183 		modearray[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
2184 				 PIPE_CONFIG(ADDR_SURF_P2) |
2185 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2186 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2187 		modearray[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2188 				 PIPE_CONFIG(ADDR_SURF_P2) |
2189 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2190 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2191 		modearray[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
2192 				 PIPE_CONFIG(ADDR_SURF_P2) |
2193 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2194 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2195 		modearray[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
2196 				 PIPE_CONFIG(ADDR_SURF_P2) |
2197 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2198 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2199 		modearray[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2200 				 PIPE_CONFIG(ADDR_SURF_P2) |
2201 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2202 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2203 		modearray[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2204 				 PIPE_CONFIG(ADDR_SURF_P2) |
2205 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2206 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2207 		modearray[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2208 				 PIPE_CONFIG(ADDR_SURF_P2) |
2209 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2210 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2211 
2212 		mod2array[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
2213 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2214 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2215 				NUM_BANKS(ADDR_SURF_8_BANK));
2216 		mod2array[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
2217 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2218 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2219 				NUM_BANKS(ADDR_SURF_8_BANK));
2220 		mod2array[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
2221 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2222 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2223 				NUM_BANKS(ADDR_SURF_8_BANK));
2224 		mod2array[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2225 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2226 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2227 				NUM_BANKS(ADDR_SURF_8_BANK));
2228 		mod2array[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2229 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2230 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2231 				NUM_BANKS(ADDR_SURF_8_BANK));
2232 		mod2array[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2233 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2234 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2235 				NUM_BANKS(ADDR_SURF_8_BANK));
2236 		mod2array[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2237 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2238 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2239 				NUM_BANKS(ADDR_SURF_8_BANK));
2240 		mod2array[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
2241 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
2242 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2243 				NUM_BANKS(ADDR_SURF_16_BANK));
2244 		mod2array[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
2245 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2246 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2247 				NUM_BANKS(ADDR_SURF_16_BANK));
2248 		mod2array[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
2249 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2250 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2251 				 NUM_BANKS(ADDR_SURF_16_BANK));
2252 		mod2array[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
2253 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2254 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2255 				 NUM_BANKS(ADDR_SURF_16_BANK));
2256 		mod2array[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2257 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2258 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2259 				 NUM_BANKS(ADDR_SURF_16_BANK));
2260 		mod2array[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2261 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2262 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2263 				 NUM_BANKS(ADDR_SURF_16_BANK));
2264 		mod2array[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2265 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2266 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2267 				 NUM_BANKS(ADDR_SURF_8_BANK));
2268 
2269 		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
2270 			if (reg_offset != 7 && reg_offset != 12 && reg_offset != 17 &&
2271 			    reg_offset != 23)
2272 				WREG32(mmGB_TILE_MODE0 + reg_offset, modearray[reg_offset]);
2273 
2274 		for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
2275 			if (reg_offset != 7)
2276 				WREG32(mmGB_MACROTILE_MODE0 + reg_offset, mod2array[reg_offset]);
2277 
2278 		break;
2279 	case CHIP_FIJI:
2280 	case CHIP_VEGAM:
2281 		modearray[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2282 				PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2283 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
2284 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2285 		modearray[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2286 				PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2287 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
2288 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2289 		modearray[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2290 				PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2291 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
2292 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2293 		modearray[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2294 				PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2295 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
2296 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2297 		modearray[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2298 				PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2299 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2300 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2301 		modearray[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2302 				PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2303 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2304 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2305 		modearray[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2306 				PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2307 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2308 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2309 		modearray[7] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2310 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2311 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2312 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2313 		modearray[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
2314 				PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16));
2315 		modearray[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2316 				PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2317 				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2318 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2319 		modearray[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2320 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2321 				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2322 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2323 		modearray[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2324 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2325 				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2326 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2327 		modearray[12] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2328 				 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2329 				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2330 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2331 		modearray[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2332 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2333 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2334 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2335 		modearray[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2336 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2337 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2338 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2339 		modearray[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
2340 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2341 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2342 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2343 		modearray[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2344 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2345 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2346 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2347 		modearray[17] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2348 				 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2349 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2350 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2351 		modearray[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2352 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2353 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2354 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2355 		modearray[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2356 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2357 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2358 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2359 		modearray[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2360 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2361 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2362 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2363 		modearray[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
2364 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2365 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2366 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2367 		modearray[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
2368 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2369 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2370 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2371 		modearray[23] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
2372 				 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2373 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2374 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2375 		modearray[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2376 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2377 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2378 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2379 		modearray[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
2380 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2381 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2382 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2383 		modearray[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
2384 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2385 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2386 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2387 		modearray[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2388 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2389 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2390 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2391 		modearray[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2392 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2393 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2394 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2395 		modearray[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2396 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2397 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2398 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2399 		modearray[30] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2400 				 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2401 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2402 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2403 
2404 		mod2array[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2405 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2406 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2407 				NUM_BANKS(ADDR_SURF_8_BANK));
2408 		mod2array[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2409 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2410 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2411 				NUM_BANKS(ADDR_SURF_8_BANK));
2412 		mod2array[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2413 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2414 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2415 				NUM_BANKS(ADDR_SURF_8_BANK));
2416 		mod2array[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2417 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2418 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2419 				NUM_BANKS(ADDR_SURF_8_BANK));
2420 		mod2array[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2421 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2422 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2423 				NUM_BANKS(ADDR_SURF_8_BANK));
2424 		mod2array[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2425 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2426 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2427 				NUM_BANKS(ADDR_SURF_8_BANK));
2428 		mod2array[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2429 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2430 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2431 				NUM_BANKS(ADDR_SURF_8_BANK));
2432 		mod2array[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2433 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
2434 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2435 				NUM_BANKS(ADDR_SURF_8_BANK));
2436 		mod2array[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2437 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2438 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2439 				NUM_BANKS(ADDR_SURF_8_BANK));
2440 		mod2array[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2441 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2442 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2443 				 NUM_BANKS(ADDR_SURF_8_BANK));
2444 		mod2array[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2445 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2446 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2447 				 NUM_BANKS(ADDR_SURF_8_BANK));
2448 		mod2array[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2449 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2450 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2451 				 NUM_BANKS(ADDR_SURF_8_BANK));
2452 		mod2array[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2453 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2454 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2455 				 NUM_BANKS(ADDR_SURF_8_BANK));
2456 		mod2array[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2457 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2458 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2459 				 NUM_BANKS(ADDR_SURF_4_BANK));
2460 
2461 		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
2462 			WREG32(mmGB_TILE_MODE0 + reg_offset, modearray[reg_offset]);
2463 
2464 		for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
2465 			if (reg_offset != 7)
2466 				WREG32(mmGB_MACROTILE_MODE0 + reg_offset, mod2array[reg_offset]);
2467 
2468 		break;
2469 	case CHIP_TONGA:
2470 		modearray[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2471 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2472 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
2473 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2474 		modearray[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2475 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2476 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
2477 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2478 		modearray[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2479 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2480 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
2481 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2482 		modearray[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2483 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2484 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
2485 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2486 		modearray[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2487 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2488 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2489 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2490 		modearray[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2491 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2492 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2493 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2494 		modearray[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2495 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2496 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2497 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2498 		modearray[7] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2499 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2500 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2501 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2502 		modearray[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
2503 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16));
2504 		modearray[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2505 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2506 				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2507 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2508 		modearray[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2509 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2510 				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2511 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2512 		modearray[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2513 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2514 				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2515 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2516 		modearray[12] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2517 				 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2518 				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2519 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2520 		modearray[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2521 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2522 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2523 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2524 		modearray[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2525 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2526 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2527 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2528 		modearray[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
2529 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2530 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2531 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2532 		modearray[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2533 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2534 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2535 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2536 		modearray[17] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2537 				 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2538 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2539 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2540 		modearray[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2541 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2542 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2543 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2544 		modearray[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2545 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2546 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2547 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2548 		modearray[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2549 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2550 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2551 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2552 		modearray[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
2553 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2554 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2555 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2556 		modearray[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
2557 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2558 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2559 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2560 		modearray[23] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
2561 				 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2562 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2563 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2564 		modearray[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2565 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2566 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2567 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2568 		modearray[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
2569 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2570 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2571 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2572 		modearray[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
2573 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2574 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2575 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2576 		modearray[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2577 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2578 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2579 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2580 		modearray[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2581 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2582 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2583 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2584 		modearray[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2585 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2586 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2587 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2588 		modearray[30] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2589 				 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2590 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2591 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2592 
2593 		mod2array[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2594 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2595 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2596 				NUM_BANKS(ADDR_SURF_16_BANK));
2597 		mod2array[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2598 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2599 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2600 				NUM_BANKS(ADDR_SURF_16_BANK));
2601 		mod2array[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2602 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2603 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2604 				NUM_BANKS(ADDR_SURF_16_BANK));
2605 		mod2array[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2606 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2607 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2608 				NUM_BANKS(ADDR_SURF_16_BANK));
2609 		mod2array[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2610 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2611 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2612 				NUM_BANKS(ADDR_SURF_16_BANK));
2613 		mod2array[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2614 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2615 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2616 				NUM_BANKS(ADDR_SURF_16_BANK));
2617 		mod2array[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2618 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2619 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2620 				NUM_BANKS(ADDR_SURF_16_BANK));
2621 		mod2array[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2622 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
2623 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2624 				NUM_BANKS(ADDR_SURF_16_BANK));
2625 		mod2array[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2626 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2627 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2628 				NUM_BANKS(ADDR_SURF_16_BANK));
2629 		mod2array[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2630 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2631 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2632 				 NUM_BANKS(ADDR_SURF_16_BANK));
2633 		mod2array[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2634 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2635 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2636 				 NUM_BANKS(ADDR_SURF_16_BANK));
2637 		mod2array[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2638 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2639 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2640 				 NUM_BANKS(ADDR_SURF_8_BANK));
2641 		mod2array[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2642 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2643 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2644 				 NUM_BANKS(ADDR_SURF_4_BANK));
2645 		mod2array[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2646 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2647 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2648 				 NUM_BANKS(ADDR_SURF_4_BANK));
2649 
2650 		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
2651 			WREG32(mmGB_TILE_MODE0 + reg_offset, modearray[reg_offset]);
2652 
2653 		for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
2654 			if (reg_offset != 7)
2655 				WREG32(mmGB_MACROTILE_MODE0 + reg_offset, mod2array[reg_offset]);
2656 
2657 		break;
2658 	case CHIP_POLARIS11:
2659 	case CHIP_POLARIS12:
2660 		modearray[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2661 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2662 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
2663 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2664 		modearray[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2665 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2666 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
2667 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2668 		modearray[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2669 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2670 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
2671 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2672 		modearray[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2673 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2674 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
2675 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2676 		modearray[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2677 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2678 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2679 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2680 		modearray[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2681 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2682 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2683 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2684 		modearray[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2685 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2686 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2687 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2688 		modearray[7] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2689 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2690 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2691 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2692 		modearray[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
2693 				PIPE_CONFIG(ADDR_SURF_P4_16x16));
2694 		modearray[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2695 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2696 				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2697 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2698 		modearray[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2699 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2700 				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2701 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2702 		modearray[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2703 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2704 				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2705 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2706 		modearray[12] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2707 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2708 				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2709 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2710 		modearray[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2711 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2712 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2713 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2714 		modearray[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2715 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2716 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2717 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2718 		modearray[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
2719 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2720 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2721 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2722 		modearray[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2723 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2724 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2725 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2726 		modearray[17] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2727 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2728 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2729 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2730 		modearray[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2731 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2732 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2733 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2734 		modearray[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2735 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2736 				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2737 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2738 		modearray[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2739 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2740 				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2741 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2742 		modearray[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
2743 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2744 				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2745 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2746 		modearray[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
2747 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2748 				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2749 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2750 		modearray[23] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
2751 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2752 				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2753 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2754 		modearray[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2755 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2756 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2757 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2758 		modearray[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
2759 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2760 				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2761 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2762 		modearray[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
2763 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2764 				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2765 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2766 		modearray[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2767 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2768 				MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2769 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2770 		modearray[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2771 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2772 				MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2773 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2774 		modearray[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2775 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2776 				MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2777 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2778 		modearray[30] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2779 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2780 				MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2781 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2782 
2783 		mod2array[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2784 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2785 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2786 				NUM_BANKS(ADDR_SURF_16_BANK));
2787 
2788 		mod2array[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2789 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2790 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2791 				NUM_BANKS(ADDR_SURF_16_BANK));
2792 
2793 		mod2array[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2794 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2795 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2796 				NUM_BANKS(ADDR_SURF_16_BANK));
2797 
2798 		mod2array[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2799 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2800 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2801 				NUM_BANKS(ADDR_SURF_16_BANK));
2802 
2803 		mod2array[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2804 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2805 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2806 				NUM_BANKS(ADDR_SURF_16_BANK));
2807 
2808 		mod2array[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2809 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2810 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2811 				NUM_BANKS(ADDR_SURF_16_BANK));
2812 
2813 		mod2array[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2814 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2815 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2816 				NUM_BANKS(ADDR_SURF_16_BANK));
2817 
2818 		mod2array[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
2819 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
2820 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2821 				NUM_BANKS(ADDR_SURF_16_BANK));
2822 
2823 		mod2array[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
2824 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2825 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2826 				NUM_BANKS(ADDR_SURF_16_BANK));
2827 
2828 		mod2array[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2829 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2830 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2831 				NUM_BANKS(ADDR_SURF_16_BANK));
2832 
2833 		mod2array[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2834 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2835 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2836 				NUM_BANKS(ADDR_SURF_16_BANK));
2837 
2838 		mod2array[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2839 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2840 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2841 				NUM_BANKS(ADDR_SURF_16_BANK));
2842 
2843 		mod2array[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2844 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2845 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2846 				NUM_BANKS(ADDR_SURF_8_BANK));
2847 
2848 		mod2array[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2849 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2850 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2851 				NUM_BANKS(ADDR_SURF_4_BANK));
2852 
2853 		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
2854 			WREG32(mmGB_TILE_MODE0 + reg_offset, modearray[reg_offset]);
2855 
2856 		for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
2857 			if (reg_offset != 7)
2858 				WREG32(mmGB_MACROTILE_MODE0 + reg_offset, mod2array[reg_offset]);
2859 
2860 		break;
2861 	case CHIP_POLARIS10:
2862 		modearray[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2863 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2864 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
2865 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2866 		modearray[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2867 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2868 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
2869 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2870 		modearray[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2871 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2872 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
2873 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2874 		modearray[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2875 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2876 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
2877 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2878 		modearray[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2879 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2880 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2881 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2882 		modearray[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2883 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2884 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2885 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2886 		modearray[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2887 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2888 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2889 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2890 		modearray[7] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2891 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2892 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2893 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2894 		modearray[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
2895 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16));
2896 		modearray[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2897 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2898 				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2899 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2900 		modearray[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2901 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2902 				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2903 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2904 		modearray[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2905 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2906 				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2907 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2908 		modearray[12] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2909 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2910 				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2911 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2912 		modearray[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2913 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2914 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2915 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2916 		modearray[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2917 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2918 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2919 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2920 		modearray[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
2921 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2922 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2923 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2924 		modearray[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2925 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2926 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2927 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2928 		modearray[17] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2929 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2930 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2931 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2932 		modearray[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2933 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2934 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2935 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2936 		modearray[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2937 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2938 				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2939 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2940 		modearray[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2941 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2942 				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2943 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2944 		modearray[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
2945 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2946 				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2947 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2948 		modearray[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
2949 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2950 				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2951 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2952 		modearray[23] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
2953 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2954 				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2955 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2956 		modearray[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2957 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2958 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2959 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2960 		modearray[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
2961 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2962 				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2963 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2964 		modearray[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
2965 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2966 				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2967 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2968 		modearray[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2969 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2970 				MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2971 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2972 		modearray[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2973 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2974 				MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2975 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2976 		modearray[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2977 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2978 				MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2979 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2980 		modearray[30] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2981 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2982 				MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2983 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2984 
2985 		mod2array[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2986 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2987 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2988 				NUM_BANKS(ADDR_SURF_16_BANK));
2989 
2990 		mod2array[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2991 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2992 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2993 				NUM_BANKS(ADDR_SURF_16_BANK));
2994 
2995 		mod2array[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2996 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2997 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2998 				NUM_BANKS(ADDR_SURF_16_BANK));
2999 
3000 		mod2array[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3001 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
3002 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3003 				NUM_BANKS(ADDR_SURF_16_BANK));
3004 
3005 		mod2array[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3006 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
3007 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3008 				NUM_BANKS(ADDR_SURF_16_BANK));
3009 
3010 		mod2array[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3011 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3012 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
3013 				NUM_BANKS(ADDR_SURF_16_BANK));
3014 
3015 		mod2array[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3016 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3017 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
3018 				NUM_BANKS(ADDR_SURF_16_BANK));
3019 
3020 		mod2array[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3021 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
3022 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3023 				NUM_BANKS(ADDR_SURF_16_BANK));
3024 
3025 		mod2array[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3026 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
3027 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3028 				NUM_BANKS(ADDR_SURF_16_BANK));
3029 
3030 		mod2array[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3031 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
3032 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3033 				NUM_BANKS(ADDR_SURF_16_BANK));
3034 
3035 		mod2array[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3036 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3037 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3038 				NUM_BANKS(ADDR_SURF_16_BANK));
3039 
3040 		mod2array[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3041 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3042 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
3043 				NUM_BANKS(ADDR_SURF_8_BANK));
3044 
3045 		mod2array[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3046 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3047 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
3048 				NUM_BANKS(ADDR_SURF_4_BANK));
3049 
3050 		mod2array[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3051 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3052 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
3053 				NUM_BANKS(ADDR_SURF_4_BANK));
3054 
3055 		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
3056 			WREG32(mmGB_TILE_MODE0 + reg_offset, modearray[reg_offset]);
3057 
3058 		for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
3059 			if (reg_offset != 7)
3060 				WREG32(mmGB_MACROTILE_MODE0 + reg_offset, mod2array[reg_offset]);
3061 
3062 		break;
3063 	case CHIP_STONEY:
3064 		modearray[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3065 				PIPE_CONFIG(ADDR_SURF_P2) |
3066 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
3067 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3068 		modearray[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3069 				PIPE_CONFIG(ADDR_SURF_P2) |
3070 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
3071 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3072 		modearray[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3073 				PIPE_CONFIG(ADDR_SURF_P2) |
3074 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
3075 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3076 		modearray[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3077 				PIPE_CONFIG(ADDR_SURF_P2) |
3078 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
3079 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3080 		modearray[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3081 				PIPE_CONFIG(ADDR_SURF_P2) |
3082 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
3083 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3084 		modearray[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
3085 				PIPE_CONFIG(ADDR_SURF_P2) |
3086 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
3087 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3088 		modearray[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
3089 				PIPE_CONFIG(ADDR_SURF_P2) |
3090 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
3091 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3092 		modearray[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
3093 				PIPE_CONFIG(ADDR_SURF_P2));
3094 		modearray[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
3095 				PIPE_CONFIG(ADDR_SURF_P2) |
3096 				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
3097 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3098 		modearray[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3099 				 PIPE_CONFIG(ADDR_SURF_P2) |
3100 				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
3101 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3102 		modearray[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
3103 				 PIPE_CONFIG(ADDR_SURF_P2) |
3104 				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
3105 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
3106 		modearray[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
3107 				 PIPE_CONFIG(ADDR_SURF_P2) |
3108 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3109 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3110 		modearray[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3111 				 PIPE_CONFIG(ADDR_SURF_P2) |
3112 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3113 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3114 		modearray[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
3115 				 PIPE_CONFIG(ADDR_SURF_P2) |
3116 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3117 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3118 		modearray[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
3119 				 PIPE_CONFIG(ADDR_SURF_P2) |
3120 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3121 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
3122 		modearray[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
3123 				 PIPE_CONFIG(ADDR_SURF_P2) |
3124 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3125 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3126 		modearray[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
3127 				 PIPE_CONFIG(ADDR_SURF_P2) |
3128 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3129 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3130 		modearray[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
3131 				 PIPE_CONFIG(ADDR_SURF_P2) |
3132 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3133 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3134 		modearray[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
3135 				 PIPE_CONFIG(ADDR_SURF_P2) |
3136 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3137 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3138 		modearray[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
3139 				 PIPE_CONFIG(ADDR_SURF_P2) |
3140 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3141 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3142 		modearray[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
3143 				 PIPE_CONFIG(ADDR_SURF_P2) |
3144 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3145 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3146 		modearray[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
3147 				 PIPE_CONFIG(ADDR_SURF_P2) |
3148 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3149 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3150 		modearray[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
3151 				 PIPE_CONFIG(ADDR_SURF_P2) |
3152 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3153 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3154 		modearray[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
3155 				 PIPE_CONFIG(ADDR_SURF_P2) |
3156 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
3157 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3158 		modearray[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3159 				 PIPE_CONFIG(ADDR_SURF_P2) |
3160 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
3161 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3162 		modearray[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
3163 				 PIPE_CONFIG(ADDR_SURF_P2) |
3164 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
3165 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
3166 
3167 		mod2array[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3168 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
3169 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3170 				NUM_BANKS(ADDR_SURF_8_BANK));
3171 		mod2array[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3172 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
3173 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3174 				NUM_BANKS(ADDR_SURF_8_BANK));
3175 		mod2array[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3176 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3177 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3178 				NUM_BANKS(ADDR_SURF_8_BANK));
3179 		mod2array[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3180 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3181 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3182 				NUM_BANKS(ADDR_SURF_8_BANK));
3183 		mod2array[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3184 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3185 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3186 				NUM_BANKS(ADDR_SURF_8_BANK));
3187 		mod2array[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3188 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3189 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3190 				NUM_BANKS(ADDR_SURF_8_BANK));
3191 		mod2array[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3192 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3193 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3194 				NUM_BANKS(ADDR_SURF_8_BANK));
3195 		mod2array[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
3196 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
3197 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3198 				NUM_BANKS(ADDR_SURF_16_BANK));
3199 		mod2array[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
3200 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
3201 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3202 				NUM_BANKS(ADDR_SURF_16_BANK));
3203 		mod2array[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
3204 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
3205 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3206 				 NUM_BANKS(ADDR_SURF_16_BANK));
3207 		mod2array[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
3208 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
3209 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3210 				 NUM_BANKS(ADDR_SURF_16_BANK));
3211 		mod2array[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3212 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
3213 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3214 				 NUM_BANKS(ADDR_SURF_16_BANK));
3215 		mod2array[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3216 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3217 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3218 				 NUM_BANKS(ADDR_SURF_16_BANK));
3219 		mod2array[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3220 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3221 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3222 				 NUM_BANKS(ADDR_SURF_8_BANK));
3223 
3224 		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
3225 			if (reg_offset != 7 && reg_offset != 12 && reg_offset != 17 &&
3226 			    reg_offset != 23)
3227 				WREG32(mmGB_TILE_MODE0 + reg_offset, modearray[reg_offset]);
3228 
3229 		for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
3230 			if (reg_offset != 7)
3231 				WREG32(mmGB_MACROTILE_MODE0 + reg_offset, mod2array[reg_offset]);
3232 
3233 		break;
3234 	default:
3235 		dev_warn(adev->dev,
3236 			 "Unknown chip type (%d) in function gfx_v8_0_tiling_mode_table_init() falling through to CHIP_CARRIZO\n",
3237 			 adev->asic_type);
3238 		/* fall through */
3239 
3240 	case CHIP_CARRIZO:
3241 		modearray[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3242 				PIPE_CONFIG(ADDR_SURF_P2) |
3243 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
3244 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3245 		modearray[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3246 				PIPE_CONFIG(ADDR_SURF_P2) |
3247 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
3248 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3249 		modearray[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3250 				PIPE_CONFIG(ADDR_SURF_P2) |
3251 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
3252 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3253 		modearray[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3254 				PIPE_CONFIG(ADDR_SURF_P2) |
3255 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
3256 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3257 		modearray[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3258 				PIPE_CONFIG(ADDR_SURF_P2) |
3259 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
3260 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3261 		modearray[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
3262 				PIPE_CONFIG(ADDR_SURF_P2) |
3263 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
3264 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3265 		modearray[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
3266 				PIPE_CONFIG(ADDR_SURF_P2) |
3267 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
3268 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3269 		modearray[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
3270 				PIPE_CONFIG(ADDR_SURF_P2));
3271 		modearray[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
3272 				PIPE_CONFIG(ADDR_SURF_P2) |
3273 				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
3274 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3275 		modearray[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3276 				 PIPE_CONFIG(ADDR_SURF_P2) |
3277 				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
3278 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3279 		modearray[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
3280 				 PIPE_CONFIG(ADDR_SURF_P2) |
3281 				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
3282 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
3283 		modearray[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
3284 				 PIPE_CONFIG(ADDR_SURF_P2) |
3285 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3286 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3287 		modearray[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3288 				 PIPE_CONFIG(ADDR_SURF_P2) |
3289 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3290 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3291 		modearray[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
3292 				 PIPE_CONFIG(ADDR_SURF_P2) |
3293 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3294 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3295 		modearray[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
3296 				 PIPE_CONFIG(ADDR_SURF_P2) |
3297 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3298 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
3299 		modearray[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
3300 				 PIPE_CONFIG(ADDR_SURF_P2) |
3301 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3302 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3303 		modearray[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
3304 				 PIPE_CONFIG(ADDR_SURF_P2) |
3305 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3306 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3307 		modearray[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
3308 				 PIPE_CONFIG(ADDR_SURF_P2) |
3309 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3310 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3311 		modearray[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
3312 				 PIPE_CONFIG(ADDR_SURF_P2) |
3313 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3314 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3315 		modearray[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
3316 				 PIPE_CONFIG(ADDR_SURF_P2) |
3317 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3318 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3319 		modearray[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
3320 				 PIPE_CONFIG(ADDR_SURF_P2) |
3321 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3322 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3323 		modearray[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
3324 				 PIPE_CONFIG(ADDR_SURF_P2) |
3325 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3326 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3327 		modearray[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
3328 				 PIPE_CONFIG(ADDR_SURF_P2) |
3329 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3330 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3331 		modearray[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
3332 				 PIPE_CONFIG(ADDR_SURF_P2) |
3333 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
3334 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3335 		modearray[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3336 				 PIPE_CONFIG(ADDR_SURF_P2) |
3337 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
3338 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3339 		modearray[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
3340 				 PIPE_CONFIG(ADDR_SURF_P2) |
3341 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
3342 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
3343 
3344 		mod2array[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3345 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
3346 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3347 				NUM_BANKS(ADDR_SURF_8_BANK));
3348 		mod2array[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3349 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
3350 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3351 				NUM_BANKS(ADDR_SURF_8_BANK));
3352 		mod2array[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3353 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3354 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3355 				NUM_BANKS(ADDR_SURF_8_BANK));
3356 		mod2array[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3357 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3358 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3359 				NUM_BANKS(ADDR_SURF_8_BANK));
3360 		mod2array[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3361 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3362 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3363 				NUM_BANKS(ADDR_SURF_8_BANK));
3364 		mod2array[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3365 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3366 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3367 				NUM_BANKS(ADDR_SURF_8_BANK));
3368 		mod2array[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3369 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3370 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3371 				NUM_BANKS(ADDR_SURF_8_BANK));
3372 		mod2array[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
3373 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
3374 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3375 				NUM_BANKS(ADDR_SURF_16_BANK));
3376 		mod2array[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
3377 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
3378 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3379 				NUM_BANKS(ADDR_SURF_16_BANK));
3380 		mod2array[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
3381 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
3382 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3383 				 NUM_BANKS(ADDR_SURF_16_BANK));
3384 		mod2array[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
3385 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
3386 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3387 				 NUM_BANKS(ADDR_SURF_16_BANK));
3388 		mod2array[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3389 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
3390 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3391 				 NUM_BANKS(ADDR_SURF_16_BANK));
3392 		mod2array[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3393 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3394 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3395 				 NUM_BANKS(ADDR_SURF_16_BANK));
3396 		mod2array[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3397 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3398 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3399 				 NUM_BANKS(ADDR_SURF_8_BANK));
3400 
3401 		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
3402 			if (reg_offset != 7 && reg_offset != 12 && reg_offset != 17 &&
3403 			    reg_offset != 23)
3404 				WREG32(mmGB_TILE_MODE0 + reg_offset, modearray[reg_offset]);
3405 
3406 		for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
3407 			if (reg_offset != 7)
3408 				WREG32(mmGB_MACROTILE_MODE0 + reg_offset, mod2array[reg_offset]);
3409 
3410 		break;
3411 	}
3412 }
3413 
3414 static void gfx_v8_0_select_se_sh(struct amdgpu_device *adev,
3415 				  u32 se_num, u32 sh_num, u32 instance)
3416 {
3417 	u32 data;
3418 
3419 	if (instance == 0xffffffff)
3420 		data = REG_SET_FIELD(0, GRBM_GFX_INDEX, INSTANCE_BROADCAST_WRITES, 1);
3421 	else
3422 		data = REG_SET_FIELD(0, GRBM_GFX_INDEX, INSTANCE_INDEX, instance);
3423 
3424 	if (se_num == 0xffffffff)
3425 		data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SE_BROADCAST_WRITES, 1);
3426 	else
3427 		data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SE_INDEX, se_num);
3428 
3429 	if (sh_num == 0xffffffff)
3430 		data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SH_BROADCAST_WRITES, 1);
3431 	else
3432 		data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SH_INDEX, sh_num);
3433 
3434 	WREG32(mmGRBM_GFX_INDEX, data);
3435 }
3436 
3437 static void gfx_v8_0_select_me_pipe_q(struct amdgpu_device *adev,
3438 				  u32 me, u32 pipe, u32 q, u32 vm)
3439 {
3440 	vi_srbm_select(adev, me, pipe, q, vm);
3441 }
3442 
3443 static u32 gfx_v8_0_get_rb_active_bitmap(struct amdgpu_device *adev)
3444 {
3445 	u32 data, mask;
3446 
3447 	data =  RREG32(mmCC_RB_BACKEND_DISABLE) |
3448 		RREG32(mmGC_USER_RB_BACKEND_DISABLE);
3449 
3450 	data = REG_GET_FIELD(data, GC_USER_RB_BACKEND_DISABLE, BACKEND_DISABLE);
3451 
3452 	mask = amdgpu_gfx_create_bitmask(adev->gfx.config.max_backends_per_se /
3453 					 adev->gfx.config.max_sh_per_se);
3454 
3455 	return (~data) & mask;
3456 }
3457 
3458 static void
3459 gfx_v8_0_raster_config(struct amdgpu_device *adev, u32 *rconf, u32 *rconf1)
3460 {
3461 	switch (adev->asic_type) {
3462 	case CHIP_FIJI:
3463 	case CHIP_VEGAM:
3464 		*rconf |= RB_MAP_PKR0(2) | RB_MAP_PKR1(2) |
3465 			  RB_XSEL2(1) | PKR_MAP(2) |
3466 			  PKR_XSEL(1) | PKR_YSEL(1) |
3467 			  SE_MAP(2) | SE_XSEL(2) | SE_YSEL(3);
3468 		*rconf1 |= SE_PAIR_MAP(2) | SE_PAIR_XSEL(3) |
3469 			   SE_PAIR_YSEL(2);
3470 		break;
3471 	case CHIP_TONGA:
3472 	case CHIP_POLARIS10:
3473 		*rconf |= RB_MAP_PKR0(2) | RB_XSEL2(1) | SE_MAP(2) |
3474 			  SE_XSEL(1) | SE_YSEL(1);
3475 		*rconf1 |= SE_PAIR_MAP(2) | SE_PAIR_XSEL(2) |
3476 			   SE_PAIR_YSEL(2);
3477 		break;
3478 	case CHIP_TOPAZ:
3479 	case CHIP_CARRIZO:
3480 		*rconf |= RB_MAP_PKR0(2);
3481 		*rconf1 |= 0x0;
3482 		break;
3483 	case CHIP_POLARIS11:
3484 	case CHIP_POLARIS12:
3485 		*rconf |= RB_MAP_PKR0(2) | RB_XSEL2(1) | SE_MAP(2) |
3486 			  SE_XSEL(1) | SE_YSEL(1);
3487 		*rconf1 |= 0x0;
3488 		break;
3489 	case CHIP_STONEY:
3490 		*rconf |= 0x0;
3491 		*rconf1 |= 0x0;
3492 		break;
3493 	default:
3494 		DRM_ERROR("unknown asic: 0x%x\n", adev->asic_type);
3495 		break;
3496 	}
3497 }
3498 
3499 static void
3500 gfx_v8_0_write_harvested_raster_configs(struct amdgpu_device *adev,
3501 					u32 raster_config, u32 raster_config_1,
3502 					unsigned rb_mask, unsigned num_rb)
3503 {
3504 	unsigned sh_per_se = max_t(unsigned, adev->gfx.config.max_sh_per_se, 1);
3505 	unsigned num_se = max_t(unsigned, adev->gfx.config.max_shader_engines, 1);
3506 	unsigned rb_per_pkr = min_t(unsigned, num_rb / num_se / sh_per_se, 2);
3507 	unsigned rb_per_se = num_rb / num_se;
3508 	unsigned se_mask[4];
3509 	unsigned se;
3510 
3511 	se_mask[0] = ((1 << rb_per_se) - 1) & rb_mask;
3512 	se_mask[1] = (se_mask[0] << rb_per_se) & rb_mask;
3513 	se_mask[2] = (se_mask[1] << rb_per_se) & rb_mask;
3514 	se_mask[3] = (se_mask[2] << rb_per_se) & rb_mask;
3515 
3516 	WARN_ON(!(num_se == 1 || num_se == 2 || num_se == 4));
3517 	WARN_ON(!(sh_per_se == 1 || sh_per_se == 2));
3518 	WARN_ON(!(rb_per_pkr == 1 || rb_per_pkr == 2));
3519 
3520 	if ((num_se > 2) && ((!se_mask[0] && !se_mask[1]) ||
3521 			     (!se_mask[2] && !se_mask[3]))) {
3522 		raster_config_1 &= ~SE_PAIR_MAP_MASK;
3523 
3524 		if (!se_mask[0] && !se_mask[1]) {
3525 			raster_config_1 |=
3526 				SE_PAIR_MAP(RASTER_CONFIG_SE_PAIR_MAP_3);
3527 		} else {
3528 			raster_config_1 |=
3529 				SE_PAIR_MAP(RASTER_CONFIG_SE_PAIR_MAP_0);
3530 		}
3531 	}
3532 
3533 	for (se = 0; se < num_se; se++) {
3534 		unsigned raster_config_se = raster_config;
3535 		unsigned pkr0_mask = ((1 << rb_per_pkr) - 1) << (se * rb_per_se);
3536 		unsigned pkr1_mask = pkr0_mask << rb_per_pkr;
3537 		int idx = (se / 2) * 2;
3538 
3539 		if ((num_se > 1) && (!se_mask[idx] || !se_mask[idx + 1])) {
3540 			raster_config_se &= ~SE_MAP_MASK;
3541 
3542 			if (!se_mask[idx]) {
3543 				raster_config_se |= SE_MAP(RASTER_CONFIG_SE_MAP_3);
3544 			} else {
3545 				raster_config_se |= SE_MAP(RASTER_CONFIG_SE_MAP_0);
3546 			}
3547 		}
3548 
3549 		pkr0_mask &= rb_mask;
3550 		pkr1_mask &= rb_mask;
3551 		if (rb_per_se > 2 && (!pkr0_mask || !pkr1_mask)) {
3552 			raster_config_se &= ~PKR_MAP_MASK;
3553 
3554 			if (!pkr0_mask) {
3555 				raster_config_se |= PKR_MAP(RASTER_CONFIG_PKR_MAP_3);
3556 			} else {
3557 				raster_config_se |= PKR_MAP(RASTER_CONFIG_PKR_MAP_0);
3558 			}
3559 		}
3560 
3561 		if (rb_per_se >= 2) {
3562 			unsigned rb0_mask = 1 << (se * rb_per_se);
3563 			unsigned rb1_mask = rb0_mask << 1;
3564 
3565 			rb0_mask &= rb_mask;
3566 			rb1_mask &= rb_mask;
3567 			if (!rb0_mask || !rb1_mask) {
3568 				raster_config_se &= ~RB_MAP_PKR0_MASK;
3569 
3570 				if (!rb0_mask) {
3571 					raster_config_se |=
3572 						RB_MAP_PKR0(RASTER_CONFIG_RB_MAP_3);
3573 				} else {
3574 					raster_config_se |=
3575 						RB_MAP_PKR0(RASTER_CONFIG_RB_MAP_0);
3576 				}
3577 			}
3578 
3579 			if (rb_per_se > 2) {
3580 				rb0_mask = 1 << (se * rb_per_se + rb_per_pkr);
3581 				rb1_mask = rb0_mask << 1;
3582 				rb0_mask &= rb_mask;
3583 				rb1_mask &= rb_mask;
3584 				if (!rb0_mask || !rb1_mask) {
3585 					raster_config_se &= ~RB_MAP_PKR1_MASK;
3586 
3587 					if (!rb0_mask) {
3588 						raster_config_se |=
3589 							RB_MAP_PKR1(RASTER_CONFIG_RB_MAP_3);
3590 					} else {
3591 						raster_config_se |=
3592 							RB_MAP_PKR1(RASTER_CONFIG_RB_MAP_0);
3593 					}
3594 				}
3595 			}
3596 		}
3597 
3598 		/* GRBM_GFX_INDEX has a different offset on VI */
3599 		gfx_v8_0_select_se_sh(adev, se, 0xffffffff, 0xffffffff);
3600 		WREG32(mmPA_SC_RASTER_CONFIG, raster_config_se);
3601 		WREG32(mmPA_SC_RASTER_CONFIG_1, raster_config_1);
3602 	}
3603 
3604 	/* GRBM_GFX_INDEX has a different offset on VI */
3605 	gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
3606 }
3607 
3608 static void gfx_v8_0_setup_rb(struct amdgpu_device *adev)
3609 {
3610 	int i, j;
3611 	u32 data;
3612 	u32 raster_config = 0, raster_config_1 = 0;
3613 	u32 active_rbs = 0;
3614 	u32 rb_bitmap_width_per_sh = adev->gfx.config.max_backends_per_se /
3615 					adev->gfx.config.max_sh_per_se;
3616 	unsigned num_rb_pipes;
3617 
3618 	mutex_lock(&adev->grbm_idx_mutex);
3619 	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
3620 		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
3621 			gfx_v8_0_select_se_sh(adev, i, j, 0xffffffff);
3622 			data = gfx_v8_0_get_rb_active_bitmap(adev);
3623 			active_rbs |= data << ((i * adev->gfx.config.max_sh_per_se + j) *
3624 					       rb_bitmap_width_per_sh);
3625 		}
3626 	}
3627 	gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
3628 
3629 	adev->gfx.config.backend_enable_mask = active_rbs;
3630 	adev->gfx.config.num_rbs = hweight32(active_rbs);
3631 
3632 	num_rb_pipes = min_t(unsigned, adev->gfx.config.max_backends_per_se *
3633 			     adev->gfx.config.max_shader_engines, 16);
3634 
3635 	gfx_v8_0_raster_config(adev, &raster_config, &raster_config_1);
3636 
3637 	if (!adev->gfx.config.backend_enable_mask ||
3638 			adev->gfx.config.num_rbs >= num_rb_pipes) {
3639 		WREG32(mmPA_SC_RASTER_CONFIG, raster_config);
3640 		WREG32(mmPA_SC_RASTER_CONFIG_1, raster_config_1);
3641 	} else {
3642 		gfx_v8_0_write_harvested_raster_configs(adev, raster_config, raster_config_1,
3643 							adev->gfx.config.backend_enable_mask,
3644 							num_rb_pipes);
3645 	}
3646 
3647 	/* cache the values for userspace */
3648 	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
3649 		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
3650 			gfx_v8_0_select_se_sh(adev, i, j, 0xffffffff);
3651 			adev->gfx.config.rb_config[i][j].rb_backend_disable =
3652 				RREG32(mmCC_RB_BACKEND_DISABLE);
3653 			adev->gfx.config.rb_config[i][j].user_rb_backend_disable =
3654 				RREG32(mmGC_USER_RB_BACKEND_DISABLE);
3655 			adev->gfx.config.rb_config[i][j].raster_config =
3656 				RREG32(mmPA_SC_RASTER_CONFIG);
3657 			adev->gfx.config.rb_config[i][j].raster_config_1 =
3658 				RREG32(mmPA_SC_RASTER_CONFIG_1);
3659 		}
3660 	}
3661 	gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
3662 	mutex_unlock(&adev->grbm_idx_mutex);
3663 }
3664 
3665 /**
3666  * gfx_v8_0_init_compute_vmid - gart enable
3667  *
3668  * @adev: amdgpu_device pointer
3669  *
3670  * Initialize compute vmid sh_mem registers
3671  *
3672  */
3673 #define DEFAULT_SH_MEM_BASES	(0x6000)
3674 #define FIRST_COMPUTE_VMID	(8)
3675 #define LAST_COMPUTE_VMID	(16)
3676 static void gfx_v8_0_init_compute_vmid(struct amdgpu_device *adev)
3677 {
3678 	int i;
3679 	uint32_t sh_mem_config;
3680 	uint32_t sh_mem_bases;
3681 
3682 	/*
3683 	 * Configure apertures:
3684 	 * LDS:         0x60000000'00000000 - 0x60000001'00000000 (4GB)
3685 	 * Scratch:     0x60000001'00000000 - 0x60000002'00000000 (4GB)
3686 	 * GPUVM:       0x60010000'00000000 - 0x60020000'00000000 (1TB)
3687 	 */
3688 	sh_mem_bases = DEFAULT_SH_MEM_BASES | (DEFAULT_SH_MEM_BASES << 16);
3689 
3690 	sh_mem_config = SH_MEM_ADDRESS_MODE_HSA64 <<
3691 			SH_MEM_CONFIG__ADDRESS_MODE__SHIFT |
3692 			SH_MEM_ALIGNMENT_MODE_UNALIGNED <<
3693 			SH_MEM_CONFIG__ALIGNMENT_MODE__SHIFT |
3694 			MTYPE_CC << SH_MEM_CONFIG__DEFAULT_MTYPE__SHIFT |
3695 			SH_MEM_CONFIG__PRIVATE_ATC_MASK;
3696 
3697 	mutex_lock(&adev->srbm_mutex);
3698 	for (i = FIRST_COMPUTE_VMID; i < LAST_COMPUTE_VMID; i++) {
3699 		vi_srbm_select(adev, 0, 0, 0, i);
3700 		/* CP and shaders */
3701 		WREG32(mmSH_MEM_CONFIG, sh_mem_config);
3702 		WREG32(mmSH_MEM_APE1_BASE, 1);
3703 		WREG32(mmSH_MEM_APE1_LIMIT, 0);
3704 		WREG32(mmSH_MEM_BASES, sh_mem_bases);
3705 	}
3706 	vi_srbm_select(adev, 0, 0, 0, 0);
3707 	mutex_unlock(&adev->srbm_mutex);
3708 
3709 	/* Initialize all compute VMIDs to have no GDS, GWS, or OA
3710 	   acccess. These should be enabled by FW for target VMIDs. */
3711 	for (i = FIRST_COMPUTE_VMID; i < LAST_COMPUTE_VMID; i++) {
3712 		WREG32(amdgpu_gds_reg_offset[i].mem_base, 0);
3713 		WREG32(amdgpu_gds_reg_offset[i].mem_size, 0);
3714 		WREG32(amdgpu_gds_reg_offset[i].gws, 0);
3715 		WREG32(amdgpu_gds_reg_offset[i].oa, 0);
3716 	}
3717 }
3718 
3719 static void gfx_v8_0_init_gds_vmid(struct amdgpu_device *adev)
3720 {
3721 	int vmid;
3722 
3723 	/*
3724 	 * Initialize all compute and user-gfx VMIDs to have no GDS, GWS, or OA
3725 	 * access. Compute VMIDs should be enabled by FW for target VMIDs,
3726 	 * the driver can enable them for graphics. VMID0 should maintain
3727 	 * access so that HWS firmware can save/restore entries.
3728 	 */
3729 	for (vmid = 1; vmid < 16; vmid++) {
3730 		WREG32(amdgpu_gds_reg_offset[vmid].mem_base, 0);
3731 		WREG32(amdgpu_gds_reg_offset[vmid].mem_size, 0);
3732 		WREG32(amdgpu_gds_reg_offset[vmid].gws, 0);
3733 		WREG32(amdgpu_gds_reg_offset[vmid].oa, 0);
3734 	}
3735 }
3736 
3737 static void gfx_v8_0_config_init(struct amdgpu_device *adev)
3738 {
3739 	switch (adev->asic_type) {
3740 	default:
3741 		adev->gfx.config.double_offchip_lds_buf = 1;
3742 		break;
3743 	case CHIP_CARRIZO:
3744 	case CHIP_STONEY:
3745 		adev->gfx.config.double_offchip_lds_buf = 0;
3746 		break;
3747 	}
3748 }
3749 
3750 static void gfx_v8_0_constants_init(struct amdgpu_device *adev)
3751 {
3752 	u32 tmp, sh_static_mem_cfg;
3753 	int i;
3754 
3755 	WREG32_FIELD(GRBM_CNTL, READ_TIMEOUT, 0xFF);
3756 	WREG32(mmGB_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
3757 	WREG32(mmHDP_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
3758 	WREG32(mmDMIF_ADDR_CALC, adev->gfx.config.gb_addr_config);
3759 
3760 	gfx_v8_0_tiling_mode_table_init(adev);
3761 	gfx_v8_0_setup_rb(adev);
3762 	gfx_v8_0_get_cu_info(adev);
3763 	gfx_v8_0_config_init(adev);
3764 
3765 	/* XXX SH_MEM regs */
3766 	/* where to put LDS, scratch, GPUVM in FSA64 space */
3767 	sh_static_mem_cfg = REG_SET_FIELD(0, SH_STATIC_MEM_CONFIG,
3768 				   SWIZZLE_ENABLE, 1);
3769 	sh_static_mem_cfg = REG_SET_FIELD(sh_static_mem_cfg, SH_STATIC_MEM_CONFIG,
3770 				   ELEMENT_SIZE, 1);
3771 	sh_static_mem_cfg = REG_SET_FIELD(sh_static_mem_cfg, SH_STATIC_MEM_CONFIG,
3772 				   INDEX_STRIDE, 3);
3773 	WREG32(mmSH_STATIC_MEM_CONFIG, sh_static_mem_cfg);
3774 
3775 	mutex_lock(&adev->srbm_mutex);
3776 	for (i = 0; i < adev->vm_manager.id_mgr[0].num_ids; i++) {
3777 		vi_srbm_select(adev, 0, 0, 0, i);
3778 		/* CP and shaders */
3779 		if (i == 0) {
3780 			tmp = REG_SET_FIELD(0, SH_MEM_CONFIG, DEFAULT_MTYPE, MTYPE_UC);
3781 			tmp = REG_SET_FIELD(tmp, SH_MEM_CONFIG, APE1_MTYPE, MTYPE_UC);
3782 			tmp = REG_SET_FIELD(tmp, SH_MEM_CONFIG, ALIGNMENT_MODE,
3783 					    SH_MEM_ALIGNMENT_MODE_UNALIGNED);
3784 			WREG32(mmSH_MEM_CONFIG, tmp);
3785 			WREG32(mmSH_MEM_BASES, 0);
3786 		} else {
3787 			tmp = REG_SET_FIELD(0, SH_MEM_CONFIG, DEFAULT_MTYPE, MTYPE_NC);
3788 			tmp = REG_SET_FIELD(tmp, SH_MEM_CONFIG, APE1_MTYPE, MTYPE_UC);
3789 			tmp = REG_SET_FIELD(tmp, SH_MEM_CONFIG, ALIGNMENT_MODE,
3790 					    SH_MEM_ALIGNMENT_MODE_UNALIGNED);
3791 			WREG32(mmSH_MEM_CONFIG, tmp);
3792 			tmp = adev->gmc.shared_aperture_start >> 48;
3793 			WREG32(mmSH_MEM_BASES, tmp);
3794 		}
3795 
3796 		WREG32(mmSH_MEM_APE1_BASE, 1);
3797 		WREG32(mmSH_MEM_APE1_LIMIT, 0);
3798 	}
3799 	vi_srbm_select(adev, 0, 0, 0, 0);
3800 	mutex_unlock(&adev->srbm_mutex);
3801 
3802 	gfx_v8_0_init_compute_vmid(adev);
3803 	gfx_v8_0_init_gds_vmid(adev);
3804 
3805 	mutex_lock(&adev->grbm_idx_mutex);
3806 	/*
3807 	 * making sure that the following register writes will be broadcasted
3808 	 * to all the shaders
3809 	 */
3810 	gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
3811 
3812 	WREG32(mmPA_SC_FIFO_SIZE,
3813 		   (adev->gfx.config.sc_prim_fifo_size_frontend <<
3814 			PA_SC_FIFO_SIZE__SC_FRONTEND_PRIM_FIFO_SIZE__SHIFT) |
3815 		   (adev->gfx.config.sc_prim_fifo_size_backend <<
3816 			PA_SC_FIFO_SIZE__SC_BACKEND_PRIM_FIFO_SIZE__SHIFT) |
3817 		   (adev->gfx.config.sc_hiz_tile_fifo_size <<
3818 			PA_SC_FIFO_SIZE__SC_HIZ_TILE_FIFO_SIZE__SHIFT) |
3819 		   (adev->gfx.config.sc_earlyz_tile_fifo_size <<
3820 			PA_SC_FIFO_SIZE__SC_EARLYZ_TILE_FIFO_SIZE__SHIFT));
3821 
3822 	tmp = RREG32(mmSPI_ARB_PRIORITY);
3823 	tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS0, 2);
3824 	tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS1, 2);
3825 	tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS2, 2);
3826 	tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS3, 2);
3827 	WREG32(mmSPI_ARB_PRIORITY, tmp);
3828 
3829 	mutex_unlock(&adev->grbm_idx_mutex);
3830 
3831 }
3832 
3833 static void gfx_v8_0_wait_for_rlc_serdes(struct amdgpu_device *adev)
3834 {
3835 	u32 i, j, k;
3836 	u32 mask;
3837 
3838 	mutex_lock(&adev->grbm_idx_mutex);
3839 	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
3840 		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
3841 			gfx_v8_0_select_se_sh(adev, i, j, 0xffffffff);
3842 			for (k = 0; k < adev->usec_timeout; k++) {
3843 				if (RREG32(mmRLC_SERDES_CU_MASTER_BUSY) == 0)
3844 					break;
3845 				udelay(1);
3846 			}
3847 			if (k == adev->usec_timeout) {
3848 				gfx_v8_0_select_se_sh(adev, 0xffffffff,
3849 						      0xffffffff, 0xffffffff);
3850 				mutex_unlock(&adev->grbm_idx_mutex);
3851 				DRM_INFO("Timeout wait for RLC serdes %u,%u\n",
3852 					 i, j);
3853 				return;
3854 			}
3855 		}
3856 	}
3857 	gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
3858 	mutex_unlock(&adev->grbm_idx_mutex);
3859 
3860 	mask = RLC_SERDES_NONCU_MASTER_BUSY__SE_MASTER_BUSY_MASK |
3861 		RLC_SERDES_NONCU_MASTER_BUSY__GC_MASTER_BUSY_MASK |
3862 		RLC_SERDES_NONCU_MASTER_BUSY__TC0_MASTER_BUSY_MASK |
3863 		RLC_SERDES_NONCU_MASTER_BUSY__TC1_MASTER_BUSY_MASK;
3864 	for (k = 0; k < adev->usec_timeout; k++) {
3865 		if ((RREG32(mmRLC_SERDES_NONCU_MASTER_BUSY) & mask) == 0)
3866 			break;
3867 		udelay(1);
3868 	}
3869 }
3870 
3871 static void gfx_v8_0_enable_gui_idle_interrupt(struct amdgpu_device *adev,
3872 					       bool enable)
3873 {
3874 	u32 tmp = RREG32(mmCP_INT_CNTL_RING0);
3875 
3876 	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CNTX_BUSY_INT_ENABLE, enable ? 1 : 0);
3877 	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CNTX_EMPTY_INT_ENABLE, enable ? 1 : 0);
3878 	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CMP_BUSY_INT_ENABLE, enable ? 1 : 0);
3879 	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, GFX_IDLE_INT_ENABLE, enable ? 1 : 0);
3880 
3881 	WREG32(mmCP_INT_CNTL_RING0, tmp);
3882 }
3883 
3884 static void gfx_v8_0_init_csb(struct amdgpu_device *adev)
3885 {
3886 	adev->gfx.rlc.funcs->get_csb_buffer(adev, adev->gfx.rlc.cs_ptr);
3887 	/* csib */
3888 	WREG32(mmRLC_CSIB_ADDR_HI,
3889 			adev->gfx.rlc.clear_state_gpu_addr >> 32);
3890 	WREG32(mmRLC_CSIB_ADDR_LO,
3891 			adev->gfx.rlc.clear_state_gpu_addr & 0xfffffffc);
3892 	WREG32(mmRLC_CSIB_LENGTH,
3893 			adev->gfx.rlc.clear_state_size);
3894 }
3895 
3896 static void gfx_v8_0_parse_ind_reg_list(int *register_list_format,
3897 				int ind_offset,
3898 				int list_size,
3899 				int *unique_indices,
3900 				int *indices_count,
3901 				int max_indices,
3902 				int *ind_start_offsets,
3903 				int *offset_count,
3904 				int max_offset)
3905 {
3906 	int indices;
3907 	bool new_entry = true;
3908 
3909 	for (; ind_offset < list_size; ind_offset++) {
3910 
3911 		if (new_entry) {
3912 			new_entry = false;
3913 			ind_start_offsets[*offset_count] = ind_offset;
3914 			*offset_count = *offset_count + 1;
3915 			BUG_ON(*offset_count >= max_offset);
3916 		}
3917 
3918 		if (register_list_format[ind_offset] == 0xFFFFFFFF) {
3919 			new_entry = true;
3920 			continue;
3921 		}
3922 
3923 		ind_offset += 2;
3924 
3925 		/* look for the matching indice */
3926 		for (indices = 0;
3927 			indices < *indices_count;
3928 			indices++) {
3929 			if (unique_indices[indices] ==
3930 				register_list_format[ind_offset])
3931 				break;
3932 		}
3933 
3934 		if (indices >= *indices_count) {
3935 			unique_indices[*indices_count] =
3936 				register_list_format[ind_offset];
3937 			indices = *indices_count;
3938 			*indices_count = *indices_count + 1;
3939 			BUG_ON(*indices_count >= max_indices);
3940 		}
3941 
3942 		register_list_format[ind_offset] = indices;
3943 	}
3944 }
3945 
3946 static int gfx_v8_0_init_save_restore_list(struct amdgpu_device *adev)
3947 {
3948 	int i, temp, data;
3949 	int unique_indices[] = {0, 0, 0, 0, 0, 0, 0, 0};
3950 	int indices_count = 0;
3951 	int indirect_start_offsets[] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0};
3952 	int offset_count = 0;
3953 
3954 	int list_size;
3955 	unsigned int *register_list_format =
3956 		kmemdup(adev->gfx.rlc.register_list_format,
3957 			adev->gfx.rlc.reg_list_format_size_bytes, GFP_KERNEL);
3958 	if (!register_list_format)
3959 		return -ENOMEM;
3960 
3961 	gfx_v8_0_parse_ind_reg_list(register_list_format,
3962 				RLC_FormatDirectRegListLength,
3963 				adev->gfx.rlc.reg_list_format_size_bytes >> 2,
3964 				unique_indices,
3965 				&indices_count,
3966 				ARRAY_SIZE(unique_indices),
3967 				indirect_start_offsets,
3968 				&offset_count,
3969 				ARRAY_SIZE(indirect_start_offsets));
3970 
3971 	/* save and restore list */
3972 	WREG32_FIELD(RLC_SRM_CNTL, AUTO_INCR_ADDR, 1);
3973 
3974 	WREG32(mmRLC_SRM_ARAM_ADDR, 0);
3975 	for (i = 0; i < adev->gfx.rlc.reg_list_size_bytes >> 2; i++)
3976 		WREG32(mmRLC_SRM_ARAM_DATA, adev->gfx.rlc.register_restore[i]);
3977 
3978 	/* indirect list */
3979 	WREG32(mmRLC_GPM_SCRATCH_ADDR, adev->gfx.rlc.reg_list_format_start);
3980 	for (i = 0; i < adev->gfx.rlc.reg_list_format_size_bytes >> 2; i++)
3981 		WREG32(mmRLC_GPM_SCRATCH_DATA, register_list_format[i]);
3982 
3983 	list_size = adev->gfx.rlc.reg_list_size_bytes >> 2;
3984 	list_size = list_size >> 1;
3985 	WREG32(mmRLC_GPM_SCRATCH_ADDR, adev->gfx.rlc.reg_restore_list_size);
3986 	WREG32(mmRLC_GPM_SCRATCH_DATA, list_size);
3987 
3988 	/* starting offsets starts */
3989 	WREG32(mmRLC_GPM_SCRATCH_ADDR,
3990 		adev->gfx.rlc.starting_offsets_start);
3991 	for (i = 0; i < ARRAY_SIZE(indirect_start_offsets); i++)
3992 		WREG32(mmRLC_GPM_SCRATCH_DATA,
3993 				indirect_start_offsets[i]);
3994 
3995 	/* unique indices */
3996 	temp = mmRLC_SRM_INDEX_CNTL_ADDR_0;
3997 	data = mmRLC_SRM_INDEX_CNTL_DATA_0;
3998 	for (i = 0; i < ARRAY_SIZE(unique_indices); i++) {
3999 		if (unique_indices[i] != 0) {
4000 			WREG32(temp + i, unique_indices[i] & 0x3FFFF);
4001 			WREG32(data + i, unique_indices[i] >> 20);
4002 		}
4003 	}
4004 	kfree(register_list_format);
4005 
4006 	return 0;
4007 }
4008 
4009 static void gfx_v8_0_enable_save_restore_machine(struct amdgpu_device *adev)
4010 {
4011 	WREG32_FIELD(RLC_SRM_CNTL, SRM_ENABLE, 1);
4012 }
4013 
4014 static void gfx_v8_0_init_power_gating(struct amdgpu_device *adev)
4015 {
4016 	uint32_t data;
4017 
4018 	WREG32_FIELD(CP_RB_WPTR_POLL_CNTL, IDLE_POLL_COUNT, 0x60);
4019 
4020 	data = REG_SET_FIELD(0, RLC_PG_DELAY, POWER_UP_DELAY, 0x10);
4021 	data = REG_SET_FIELD(data, RLC_PG_DELAY, POWER_DOWN_DELAY, 0x10);
4022 	data = REG_SET_FIELD(data, RLC_PG_DELAY, CMD_PROPAGATE_DELAY, 0x10);
4023 	data = REG_SET_FIELD(data, RLC_PG_DELAY, MEM_SLEEP_DELAY, 0x10);
4024 	WREG32(mmRLC_PG_DELAY, data);
4025 
4026 	WREG32_FIELD(RLC_PG_DELAY_2, SERDES_CMD_DELAY, 0x3);
4027 	WREG32_FIELD(RLC_AUTO_PG_CTRL, GRBM_REG_SAVE_GFX_IDLE_THRESHOLD, 0x55f0);
4028 
4029 }
4030 
4031 static void cz_enable_sck_slow_down_on_power_up(struct amdgpu_device *adev,
4032 						bool enable)
4033 {
4034 	WREG32_FIELD(RLC_PG_CNTL, SMU_CLK_SLOWDOWN_ON_PU_ENABLE, enable ? 1 : 0);
4035 }
4036 
4037 static void cz_enable_sck_slow_down_on_power_down(struct amdgpu_device *adev,
4038 						  bool enable)
4039 {
4040 	WREG32_FIELD(RLC_PG_CNTL, SMU_CLK_SLOWDOWN_ON_PD_ENABLE, enable ? 1 : 0);
4041 }
4042 
4043 static void cz_enable_cp_power_gating(struct amdgpu_device *adev, bool enable)
4044 {
4045 	WREG32_FIELD(RLC_PG_CNTL, CP_PG_DISABLE, enable ? 0 : 1);
4046 }
4047 
4048 static void gfx_v8_0_init_pg(struct amdgpu_device *adev)
4049 {
4050 	if ((adev->asic_type == CHIP_CARRIZO) ||
4051 	    (adev->asic_type == CHIP_STONEY)) {
4052 		gfx_v8_0_init_csb(adev);
4053 		gfx_v8_0_init_save_restore_list(adev);
4054 		gfx_v8_0_enable_save_restore_machine(adev);
4055 		WREG32(mmRLC_JUMP_TABLE_RESTORE, adev->gfx.rlc.cp_table_gpu_addr >> 8);
4056 		gfx_v8_0_init_power_gating(adev);
4057 		WREG32(mmRLC_PG_ALWAYS_ON_CU_MASK, adev->gfx.cu_info.ao_cu_mask);
4058 	} else if ((adev->asic_type == CHIP_POLARIS11) ||
4059 		   (adev->asic_type == CHIP_POLARIS12) ||
4060 		   (adev->asic_type == CHIP_VEGAM)) {
4061 		gfx_v8_0_init_csb(adev);
4062 		gfx_v8_0_init_save_restore_list(adev);
4063 		gfx_v8_0_enable_save_restore_machine(adev);
4064 		gfx_v8_0_init_power_gating(adev);
4065 	}
4066 
4067 }
4068 
4069 static void gfx_v8_0_rlc_stop(struct amdgpu_device *adev)
4070 {
4071 	WREG32_FIELD(RLC_CNTL, RLC_ENABLE_F32, 0);
4072 
4073 	gfx_v8_0_enable_gui_idle_interrupt(adev, false);
4074 	gfx_v8_0_wait_for_rlc_serdes(adev);
4075 }
4076 
4077 static void gfx_v8_0_rlc_reset(struct amdgpu_device *adev)
4078 {
4079 	WREG32_FIELD(GRBM_SOFT_RESET, SOFT_RESET_RLC, 1);
4080 	udelay(50);
4081 
4082 	WREG32_FIELD(GRBM_SOFT_RESET, SOFT_RESET_RLC, 0);
4083 	udelay(50);
4084 }
4085 
4086 static void gfx_v8_0_rlc_start(struct amdgpu_device *adev)
4087 {
4088 	WREG32_FIELD(RLC_CNTL, RLC_ENABLE_F32, 1);
4089 
4090 	/* carrizo do enable cp interrupt after cp inited */
4091 	if (!(adev->flags & AMD_IS_APU))
4092 		gfx_v8_0_enable_gui_idle_interrupt(adev, true);
4093 
4094 	udelay(50);
4095 }
4096 
4097 static int gfx_v8_0_rlc_resume(struct amdgpu_device *adev)
4098 {
4099 	if (amdgpu_sriov_vf(adev)) {
4100 		gfx_v8_0_init_csb(adev);
4101 		return 0;
4102 	}
4103 
4104 	adev->gfx.rlc.funcs->stop(adev);
4105 	adev->gfx.rlc.funcs->reset(adev);
4106 	gfx_v8_0_init_pg(adev);
4107 	adev->gfx.rlc.funcs->start(adev);
4108 
4109 	return 0;
4110 }
4111 
4112 static void gfx_v8_0_cp_gfx_enable(struct amdgpu_device *adev, bool enable)
4113 {
4114 	int i;
4115 	u32 tmp = RREG32(mmCP_ME_CNTL);
4116 
4117 	if (enable) {
4118 		tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, ME_HALT, 0);
4119 		tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, PFP_HALT, 0);
4120 		tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, CE_HALT, 0);
4121 	} else {
4122 		tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, ME_HALT, 1);
4123 		tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, PFP_HALT, 1);
4124 		tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, CE_HALT, 1);
4125 		for (i = 0; i < adev->gfx.num_gfx_rings; i++)
4126 			adev->gfx.gfx_ring[i].sched.ready = false;
4127 	}
4128 	WREG32(mmCP_ME_CNTL, tmp);
4129 	udelay(50);
4130 }
4131 
4132 static u32 gfx_v8_0_get_csb_size(struct amdgpu_device *adev)
4133 {
4134 	u32 count = 0;
4135 	const struct cs_section_def *sect = NULL;
4136 	const struct cs_extent_def *ext = NULL;
4137 
4138 	/* begin clear state */
4139 	count += 2;
4140 	/* context control state */
4141 	count += 3;
4142 
4143 	for (sect = vi_cs_data; sect->section != NULL; ++sect) {
4144 		for (ext = sect->section; ext->extent != NULL; ++ext) {
4145 			if (sect->id == SECT_CONTEXT)
4146 				count += 2 + ext->reg_count;
4147 			else
4148 				return 0;
4149 		}
4150 	}
4151 	/* pa_sc_raster_config/pa_sc_raster_config1 */
4152 	count += 4;
4153 	/* end clear state */
4154 	count += 2;
4155 	/* clear state */
4156 	count += 2;
4157 
4158 	return count;
4159 }
4160 
4161 static int gfx_v8_0_cp_gfx_start(struct amdgpu_device *adev)
4162 {
4163 	struct amdgpu_ring *ring = &adev->gfx.gfx_ring[0];
4164 	const struct cs_section_def *sect = NULL;
4165 	const struct cs_extent_def *ext = NULL;
4166 	int r, i;
4167 
4168 	/* init the CP */
4169 	WREG32(mmCP_MAX_CONTEXT, adev->gfx.config.max_hw_contexts - 1);
4170 	WREG32(mmCP_ENDIAN_SWAP, 0);
4171 	WREG32(mmCP_DEVICE_ID, 1);
4172 
4173 	gfx_v8_0_cp_gfx_enable(adev, true);
4174 
4175 	r = amdgpu_ring_alloc(ring, gfx_v8_0_get_csb_size(adev) + 4);
4176 	if (r) {
4177 		DRM_ERROR("amdgpu: cp failed to lock ring (%d).\n", r);
4178 		return r;
4179 	}
4180 
4181 	/* clear state buffer */
4182 	amdgpu_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
4183 	amdgpu_ring_write(ring, PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
4184 
4185 	amdgpu_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1));
4186 	amdgpu_ring_write(ring, 0x80000000);
4187 	amdgpu_ring_write(ring, 0x80000000);
4188 
4189 	for (sect = vi_cs_data; sect->section != NULL; ++sect) {
4190 		for (ext = sect->section; ext->extent != NULL; ++ext) {
4191 			if (sect->id == SECT_CONTEXT) {
4192 				amdgpu_ring_write(ring,
4193 				       PACKET3(PACKET3_SET_CONTEXT_REG,
4194 					       ext->reg_count));
4195 				amdgpu_ring_write(ring,
4196 				       ext->reg_index - PACKET3_SET_CONTEXT_REG_START);
4197 				for (i = 0; i < ext->reg_count; i++)
4198 					amdgpu_ring_write(ring, ext->extent[i]);
4199 			}
4200 		}
4201 	}
4202 
4203 	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONTEXT_REG, 2));
4204 	amdgpu_ring_write(ring, mmPA_SC_RASTER_CONFIG - PACKET3_SET_CONTEXT_REG_START);
4205 	amdgpu_ring_write(ring, adev->gfx.config.rb_config[0][0].raster_config);
4206 	amdgpu_ring_write(ring, adev->gfx.config.rb_config[0][0].raster_config_1);
4207 
4208 	amdgpu_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
4209 	amdgpu_ring_write(ring, PACKET3_PREAMBLE_END_CLEAR_STATE);
4210 
4211 	amdgpu_ring_write(ring, PACKET3(PACKET3_CLEAR_STATE, 0));
4212 	amdgpu_ring_write(ring, 0);
4213 
4214 	/* init the CE partitions */
4215 	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_BASE, 2));
4216 	amdgpu_ring_write(ring, PACKET3_BASE_INDEX(CE_PARTITION_BASE));
4217 	amdgpu_ring_write(ring, 0x8000);
4218 	amdgpu_ring_write(ring, 0x8000);
4219 
4220 	amdgpu_ring_commit(ring);
4221 
4222 	return 0;
4223 }
4224 static void gfx_v8_0_set_cpg_door_bell(struct amdgpu_device *adev, struct amdgpu_ring *ring)
4225 {
4226 	u32 tmp;
4227 	/* no gfx doorbells on iceland */
4228 	if (adev->asic_type == CHIP_TOPAZ)
4229 		return;
4230 
4231 	tmp = RREG32(mmCP_RB_DOORBELL_CONTROL);
4232 
4233 	if (ring->use_doorbell) {
4234 		tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL,
4235 				DOORBELL_OFFSET, ring->doorbell_index);
4236 		tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL,
4237 						DOORBELL_HIT, 0);
4238 		tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL,
4239 					    DOORBELL_EN, 1);
4240 	} else {
4241 		tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL, DOORBELL_EN, 0);
4242 	}
4243 
4244 	WREG32(mmCP_RB_DOORBELL_CONTROL, tmp);
4245 
4246 	if (adev->flags & AMD_IS_APU)
4247 		return;
4248 
4249 	tmp = REG_SET_FIELD(0, CP_RB_DOORBELL_RANGE_LOWER,
4250 					DOORBELL_RANGE_LOWER,
4251 					adev->doorbell_index.gfx_ring0);
4252 	WREG32(mmCP_RB_DOORBELL_RANGE_LOWER, tmp);
4253 
4254 	WREG32(mmCP_RB_DOORBELL_RANGE_UPPER,
4255 		CP_RB_DOORBELL_RANGE_UPPER__DOORBELL_RANGE_UPPER_MASK);
4256 }
4257 
4258 static int gfx_v8_0_cp_gfx_resume(struct amdgpu_device *adev)
4259 {
4260 	struct amdgpu_ring *ring;
4261 	u32 tmp;
4262 	u32 rb_bufsz;
4263 	u64 rb_addr, rptr_addr, wptr_gpu_addr;
4264 
4265 	/* Set the write pointer delay */
4266 	WREG32(mmCP_RB_WPTR_DELAY, 0);
4267 
4268 	/* set the RB to use vmid 0 */
4269 	WREG32(mmCP_RB_VMID, 0);
4270 
4271 	/* Set ring buffer size */
4272 	ring = &adev->gfx.gfx_ring[0];
4273 	rb_bufsz = order_base_2(ring->ring_size / 8);
4274 	tmp = REG_SET_FIELD(0, CP_RB0_CNTL, RB_BUFSZ, rb_bufsz);
4275 	tmp = REG_SET_FIELD(tmp, CP_RB0_CNTL, RB_BLKSZ, rb_bufsz - 2);
4276 	tmp = REG_SET_FIELD(tmp, CP_RB0_CNTL, MTYPE, 3);
4277 	tmp = REG_SET_FIELD(tmp, CP_RB0_CNTL, MIN_IB_AVAILSZ, 1);
4278 #ifdef __BIG_ENDIAN
4279 	tmp = REG_SET_FIELD(tmp, CP_RB0_CNTL, BUF_SWAP, 1);
4280 #endif
4281 	WREG32(mmCP_RB0_CNTL, tmp);
4282 
4283 	/* Initialize the ring buffer's read and write pointers */
4284 	WREG32(mmCP_RB0_CNTL, tmp | CP_RB0_CNTL__RB_RPTR_WR_ENA_MASK);
4285 	ring->wptr = 0;
4286 	WREG32(mmCP_RB0_WPTR, lower_32_bits(ring->wptr));
4287 
4288 	/* set the wb address wether it's enabled or not */
4289 	rptr_addr = adev->wb.gpu_addr + (ring->rptr_offs * 4);
4290 	WREG32(mmCP_RB0_RPTR_ADDR, lower_32_bits(rptr_addr));
4291 	WREG32(mmCP_RB0_RPTR_ADDR_HI, upper_32_bits(rptr_addr) & 0xFF);
4292 
4293 	wptr_gpu_addr = adev->wb.gpu_addr + (ring->wptr_offs * 4);
4294 	WREG32(mmCP_RB_WPTR_POLL_ADDR_LO, lower_32_bits(wptr_gpu_addr));
4295 	WREG32(mmCP_RB_WPTR_POLL_ADDR_HI, upper_32_bits(wptr_gpu_addr));
4296 	mdelay(1);
4297 	WREG32(mmCP_RB0_CNTL, tmp);
4298 
4299 	rb_addr = ring->gpu_addr >> 8;
4300 	WREG32(mmCP_RB0_BASE, rb_addr);
4301 	WREG32(mmCP_RB0_BASE_HI, upper_32_bits(rb_addr));
4302 
4303 	gfx_v8_0_set_cpg_door_bell(adev, ring);
4304 	/* start the ring */
4305 	amdgpu_ring_clear_ring(ring);
4306 	gfx_v8_0_cp_gfx_start(adev);
4307 	ring->sched.ready = true;
4308 
4309 	return 0;
4310 }
4311 
4312 static void gfx_v8_0_cp_compute_enable(struct amdgpu_device *adev, bool enable)
4313 {
4314 	int i;
4315 
4316 	if (enable) {
4317 		WREG32(mmCP_MEC_CNTL, 0);
4318 	} else {
4319 		WREG32(mmCP_MEC_CNTL, (CP_MEC_CNTL__MEC_ME1_HALT_MASK | CP_MEC_CNTL__MEC_ME2_HALT_MASK));
4320 		for (i = 0; i < adev->gfx.num_compute_rings; i++)
4321 			adev->gfx.compute_ring[i].sched.ready = false;
4322 		adev->gfx.kiq.ring.sched.ready = false;
4323 	}
4324 	udelay(50);
4325 }
4326 
4327 /* KIQ functions */
4328 static void gfx_v8_0_kiq_setting(struct amdgpu_ring *ring)
4329 {
4330 	uint32_t tmp;
4331 	struct amdgpu_device *adev = ring->adev;
4332 
4333 	/* tell RLC which is KIQ queue */
4334 	tmp = RREG32(mmRLC_CP_SCHEDULERS);
4335 	tmp &= 0xffffff00;
4336 	tmp |= (ring->me << 5) | (ring->pipe << 3) | (ring->queue);
4337 	WREG32(mmRLC_CP_SCHEDULERS, tmp);
4338 	tmp |= 0x80;
4339 	WREG32(mmRLC_CP_SCHEDULERS, tmp);
4340 }
4341 
4342 static int gfx_v8_0_kiq_kcq_enable(struct amdgpu_device *adev)
4343 {
4344 	struct amdgpu_ring *kiq_ring = &adev->gfx.kiq.ring;
4345 	uint64_t queue_mask = 0;
4346 	int r, i;
4347 
4348 	for (i = 0; i < AMDGPU_MAX_COMPUTE_QUEUES; ++i) {
4349 		if (!test_bit(i, adev->gfx.mec.queue_bitmap))
4350 			continue;
4351 
4352 		/* This situation may be hit in the future if a new HW
4353 		 * generation exposes more than 64 queues. If so, the
4354 		 * definition of queue_mask needs updating */
4355 		if (WARN_ON(i >= (sizeof(queue_mask)*8))) {
4356 			DRM_ERROR("Invalid KCQ enabled: %d\n", i);
4357 			break;
4358 		}
4359 
4360 		queue_mask |= (1ull << i);
4361 	}
4362 
4363 	r = amdgpu_ring_alloc(kiq_ring, (8 * adev->gfx.num_compute_rings) + 8);
4364 	if (r) {
4365 		DRM_ERROR("Failed to lock KIQ (%d).\n", r);
4366 		return r;
4367 	}
4368 	/* set resources */
4369 	amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_SET_RESOURCES, 6));
4370 	amdgpu_ring_write(kiq_ring, 0);	/* vmid_mask:0 queue_type:0 (KIQ) */
4371 	amdgpu_ring_write(kiq_ring, lower_32_bits(queue_mask));	/* queue mask lo */
4372 	amdgpu_ring_write(kiq_ring, upper_32_bits(queue_mask));	/* queue mask hi */
4373 	amdgpu_ring_write(kiq_ring, 0);	/* gws mask lo */
4374 	amdgpu_ring_write(kiq_ring, 0);	/* gws mask hi */
4375 	amdgpu_ring_write(kiq_ring, 0);	/* oac mask */
4376 	amdgpu_ring_write(kiq_ring, 0);	/* gds heap base:0, gds heap size:0 */
4377 	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
4378 		struct amdgpu_ring *ring = &adev->gfx.compute_ring[i];
4379 		uint64_t mqd_addr = amdgpu_bo_gpu_offset(ring->mqd_obj);
4380 		uint64_t wptr_addr = adev->wb.gpu_addr + (ring->wptr_offs * 4);
4381 
4382 		/* map queues */
4383 		amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_MAP_QUEUES, 5));
4384 		/* Q_sel:0, vmid:0, vidmem: 1, engine:0, num_Q:1*/
4385 		amdgpu_ring_write(kiq_ring,
4386 				  PACKET3_MAP_QUEUES_NUM_QUEUES(1));
4387 		amdgpu_ring_write(kiq_ring,
4388 				  PACKET3_MAP_QUEUES_DOORBELL_OFFSET(ring->doorbell_index) |
4389 				  PACKET3_MAP_QUEUES_QUEUE(ring->queue) |
4390 				  PACKET3_MAP_QUEUES_PIPE(ring->pipe) |
4391 				  PACKET3_MAP_QUEUES_ME(ring->me == 1 ? 0 : 1)); /* doorbell */
4392 		amdgpu_ring_write(kiq_ring, lower_32_bits(mqd_addr));
4393 		amdgpu_ring_write(kiq_ring, upper_32_bits(mqd_addr));
4394 		amdgpu_ring_write(kiq_ring, lower_32_bits(wptr_addr));
4395 		amdgpu_ring_write(kiq_ring, upper_32_bits(wptr_addr));
4396 	}
4397 
4398 	amdgpu_ring_commit(kiq_ring);
4399 
4400 	return 0;
4401 }
4402 
4403 static int gfx_v8_0_deactivate_hqd(struct amdgpu_device *adev, u32 req)
4404 {
4405 	int i, r = 0;
4406 
4407 	if (RREG32(mmCP_HQD_ACTIVE) & CP_HQD_ACTIVE__ACTIVE_MASK) {
4408 		WREG32_FIELD(CP_HQD_DEQUEUE_REQUEST, DEQUEUE_REQ, req);
4409 		for (i = 0; i < adev->usec_timeout; i++) {
4410 			if (!(RREG32(mmCP_HQD_ACTIVE) & CP_HQD_ACTIVE__ACTIVE_MASK))
4411 				break;
4412 			udelay(1);
4413 		}
4414 		if (i == adev->usec_timeout)
4415 			r = -ETIMEDOUT;
4416 	}
4417 	WREG32(mmCP_HQD_DEQUEUE_REQUEST, 0);
4418 	WREG32(mmCP_HQD_PQ_RPTR, 0);
4419 	WREG32(mmCP_HQD_PQ_WPTR, 0);
4420 
4421 	return r;
4422 }
4423 
4424 static int gfx_v8_0_mqd_init(struct amdgpu_ring *ring)
4425 {
4426 	struct amdgpu_device *adev = ring->adev;
4427 	struct vi_mqd *mqd = ring->mqd_ptr;
4428 	uint64_t hqd_gpu_addr, wb_gpu_addr, eop_base_addr;
4429 	uint32_t tmp;
4430 
4431 	mqd->header = 0xC0310800;
4432 	mqd->compute_pipelinestat_enable = 0x00000001;
4433 	mqd->compute_static_thread_mgmt_se0 = 0xffffffff;
4434 	mqd->compute_static_thread_mgmt_se1 = 0xffffffff;
4435 	mqd->compute_static_thread_mgmt_se2 = 0xffffffff;
4436 	mqd->compute_static_thread_mgmt_se3 = 0xffffffff;
4437 	mqd->compute_misc_reserved = 0x00000003;
4438 	mqd->dynamic_cu_mask_addr_lo = lower_32_bits(ring->mqd_gpu_addr
4439 						     + offsetof(struct vi_mqd_allocation, dynamic_cu_mask));
4440 	mqd->dynamic_cu_mask_addr_hi = upper_32_bits(ring->mqd_gpu_addr
4441 						     + offsetof(struct vi_mqd_allocation, dynamic_cu_mask));
4442 	eop_base_addr = ring->eop_gpu_addr >> 8;
4443 	mqd->cp_hqd_eop_base_addr_lo = eop_base_addr;
4444 	mqd->cp_hqd_eop_base_addr_hi = upper_32_bits(eop_base_addr);
4445 
4446 	/* set the EOP size, register value is 2^(EOP_SIZE+1) dwords */
4447 	tmp = RREG32(mmCP_HQD_EOP_CONTROL);
4448 	tmp = REG_SET_FIELD(tmp, CP_HQD_EOP_CONTROL, EOP_SIZE,
4449 			(order_base_2(GFX8_MEC_HPD_SIZE / 4) - 1));
4450 
4451 	mqd->cp_hqd_eop_control = tmp;
4452 
4453 	/* enable doorbell? */
4454 	tmp = REG_SET_FIELD(RREG32(mmCP_HQD_PQ_DOORBELL_CONTROL),
4455 			    CP_HQD_PQ_DOORBELL_CONTROL,
4456 			    DOORBELL_EN,
4457 			    ring->use_doorbell ? 1 : 0);
4458 
4459 	mqd->cp_hqd_pq_doorbell_control = tmp;
4460 
4461 	/* set the pointer to the MQD */
4462 	mqd->cp_mqd_base_addr_lo = ring->mqd_gpu_addr & 0xfffffffc;
4463 	mqd->cp_mqd_base_addr_hi = upper_32_bits(ring->mqd_gpu_addr);
4464 
4465 	/* set MQD vmid to 0 */
4466 	tmp = RREG32(mmCP_MQD_CONTROL);
4467 	tmp = REG_SET_FIELD(tmp, CP_MQD_CONTROL, VMID, 0);
4468 	mqd->cp_mqd_control = tmp;
4469 
4470 	/* set the pointer to the HQD, this is similar CP_RB0_BASE/_HI */
4471 	hqd_gpu_addr = ring->gpu_addr >> 8;
4472 	mqd->cp_hqd_pq_base_lo = hqd_gpu_addr;
4473 	mqd->cp_hqd_pq_base_hi = upper_32_bits(hqd_gpu_addr);
4474 
4475 	/* set up the HQD, this is similar to CP_RB0_CNTL */
4476 	tmp = RREG32(mmCP_HQD_PQ_CONTROL);
4477 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, QUEUE_SIZE,
4478 			    (order_base_2(ring->ring_size / 4) - 1));
4479 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, RPTR_BLOCK_SIZE,
4480 			((order_base_2(AMDGPU_GPU_PAGE_SIZE / 4) - 1) << 8));
4481 #ifdef __BIG_ENDIAN
4482 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, ENDIAN_SWAP, 1);
4483 #endif
4484 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, UNORD_DISPATCH, 0);
4485 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, ROQ_PQ_IB_FLIP, 0);
4486 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, PRIV_STATE, 1);
4487 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, KMD_QUEUE, 1);
4488 	mqd->cp_hqd_pq_control = tmp;
4489 
4490 	/* set the wb address whether it's enabled or not */
4491 	wb_gpu_addr = adev->wb.gpu_addr + (ring->rptr_offs * 4);
4492 	mqd->cp_hqd_pq_rptr_report_addr_lo = wb_gpu_addr & 0xfffffffc;
4493 	mqd->cp_hqd_pq_rptr_report_addr_hi =
4494 		upper_32_bits(wb_gpu_addr) & 0xffff;
4495 
4496 	/* only used if CP_PQ_WPTR_POLL_CNTL.CP_PQ_WPTR_POLL_CNTL__EN_MASK=1 */
4497 	wb_gpu_addr = adev->wb.gpu_addr + (ring->wptr_offs * 4);
4498 	mqd->cp_hqd_pq_wptr_poll_addr_lo = wb_gpu_addr & 0xfffffffc;
4499 	mqd->cp_hqd_pq_wptr_poll_addr_hi = upper_32_bits(wb_gpu_addr) & 0xffff;
4500 
4501 	tmp = 0;
4502 	/* enable the doorbell if requested */
4503 	if (ring->use_doorbell) {
4504 		tmp = RREG32(mmCP_HQD_PQ_DOORBELL_CONTROL);
4505 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
4506 				DOORBELL_OFFSET, ring->doorbell_index);
4507 
4508 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
4509 					 DOORBELL_EN, 1);
4510 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
4511 					 DOORBELL_SOURCE, 0);
4512 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
4513 					 DOORBELL_HIT, 0);
4514 	}
4515 
4516 	mqd->cp_hqd_pq_doorbell_control = tmp;
4517 
4518 	/* reset read and write pointers, similar to CP_RB0_WPTR/_RPTR */
4519 	ring->wptr = 0;
4520 	mqd->cp_hqd_pq_wptr = ring->wptr;
4521 	mqd->cp_hqd_pq_rptr = RREG32(mmCP_HQD_PQ_RPTR);
4522 
4523 	/* set the vmid for the queue */
4524 	mqd->cp_hqd_vmid = 0;
4525 
4526 	tmp = RREG32(mmCP_HQD_PERSISTENT_STATE);
4527 	tmp = REG_SET_FIELD(tmp, CP_HQD_PERSISTENT_STATE, PRELOAD_SIZE, 0x53);
4528 	mqd->cp_hqd_persistent_state = tmp;
4529 
4530 	/* set MTYPE */
4531 	tmp = RREG32(mmCP_HQD_IB_CONTROL);
4532 	tmp = REG_SET_FIELD(tmp, CP_HQD_IB_CONTROL, MIN_IB_AVAIL_SIZE, 3);
4533 	tmp = REG_SET_FIELD(tmp, CP_HQD_IB_CONTROL, MTYPE, 3);
4534 	mqd->cp_hqd_ib_control = tmp;
4535 
4536 	tmp = RREG32(mmCP_HQD_IQ_TIMER);
4537 	tmp = REG_SET_FIELD(tmp, CP_HQD_IQ_TIMER, MTYPE, 3);
4538 	mqd->cp_hqd_iq_timer = tmp;
4539 
4540 	tmp = RREG32(mmCP_HQD_CTX_SAVE_CONTROL);
4541 	tmp = REG_SET_FIELD(tmp, CP_HQD_CTX_SAVE_CONTROL, MTYPE, 3);
4542 	mqd->cp_hqd_ctx_save_control = tmp;
4543 
4544 	/* defaults */
4545 	mqd->cp_hqd_eop_rptr = RREG32(mmCP_HQD_EOP_RPTR);
4546 	mqd->cp_hqd_eop_wptr = RREG32(mmCP_HQD_EOP_WPTR);
4547 	mqd->cp_hqd_pipe_priority = RREG32(mmCP_HQD_PIPE_PRIORITY);
4548 	mqd->cp_hqd_queue_priority = RREG32(mmCP_HQD_QUEUE_PRIORITY);
4549 	mqd->cp_hqd_quantum = RREG32(mmCP_HQD_QUANTUM);
4550 	mqd->cp_hqd_ctx_save_base_addr_lo = RREG32(mmCP_HQD_CTX_SAVE_BASE_ADDR_LO);
4551 	mqd->cp_hqd_ctx_save_base_addr_hi = RREG32(mmCP_HQD_CTX_SAVE_BASE_ADDR_HI);
4552 	mqd->cp_hqd_cntl_stack_offset = RREG32(mmCP_HQD_CNTL_STACK_OFFSET);
4553 	mqd->cp_hqd_cntl_stack_size = RREG32(mmCP_HQD_CNTL_STACK_SIZE);
4554 	mqd->cp_hqd_wg_state_offset = RREG32(mmCP_HQD_WG_STATE_OFFSET);
4555 	mqd->cp_hqd_ctx_save_size = RREG32(mmCP_HQD_CTX_SAVE_SIZE);
4556 	mqd->cp_hqd_eop_done_events = RREG32(mmCP_HQD_EOP_EVENTS);
4557 	mqd->cp_hqd_error = RREG32(mmCP_HQD_ERROR);
4558 	mqd->cp_hqd_eop_wptr_mem = RREG32(mmCP_HQD_EOP_WPTR_MEM);
4559 	mqd->cp_hqd_eop_dones = RREG32(mmCP_HQD_EOP_DONES);
4560 
4561 	/* activate the queue */
4562 	mqd->cp_hqd_active = 1;
4563 
4564 	return 0;
4565 }
4566 
4567 int gfx_v8_0_mqd_commit(struct amdgpu_device *adev,
4568 			struct vi_mqd *mqd)
4569 {
4570 	uint32_t mqd_reg;
4571 	uint32_t *mqd_data;
4572 
4573 	/* HQD registers extend from mmCP_MQD_BASE_ADDR to mmCP_HQD_ERROR */
4574 	mqd_data = &mqd->cp_mqd_base_addr_lo;
4575 
4576 	/* disable wptr polling */
4577 	WREG32_FIELD(CP_PQ_WPTR_POLL_CNTL, EN, 0);
4578 
4579 	/* program all HQD registers */
4580 	for (mqd_reg = mmCP_HQD_VMID; mqd_reg <= mmCP_HQD_EOP_CONTROL; mqd_reg++)
4581 		WREG32(mqd_reg, mqd_data[mqd_reg - mmCP_MQD_BASE_ADDR]);
4582 
4583 	/* Tonga errata: EOP RPTR/WPTR should be left unmodified.
4584 	 * This is safe since EOP RPTR==WPTR for any inactive HQD
4585 	 * on ASICs that do not support context-save.
4586 	 * EOP writes/reads can start anywhere in the ring.
4587 	 */
4588 	if (adev->asic_type != CHIP_TONGA) {
4589 		WREG32(mmCP_HQD_EOP_RPTR, mqd->cp_hqd_eop_rptr);
4590 		WREG32(mmCP_HQD_EOP_WPTR, mqd->cp_hqd_eop_wptr);
4591 		WREG32(mmCP_HQD_EOP_WPTR_MEM, mqd->cp_hqd_eop_wptr_mem);
4592 	}
4593 
4594 	for (mqd_reg = mmCP_HQD_EOP_EVENTS; mqd_reg <= mmCP_HQD_ERROR; mqd_reg++)
4595 		WREG32(mqd_reg, mqd_data[mqd_reg - mmCP_MQD_BASE_ADDR]);
4596 
4597 	/* activate the HQD */
4598 	for (mqd_reg = mmCP_MQD_BASE_ADDR; mqd_reg <= mmCP_HQD_ACTIVE; mqd_reg++)
4599 		WREG32(mqd_reg, mqd_data[mqd_reg - mmCP_MQD_BASE_ADDR]);
4600 
4601 	return 0;
4602 }
4603 
4604 static int gfx_v8_0_kiq_init_queue(struct amdgpu_ring *ring)
4605 {
4606 	struct amdgpu_device *adev = ring->adev;
4607 	struct vi_mqd *mqd = ring->mqd_ptr;
4608 	int mqd_idx = AMDGPU_MAX_COMPUTE_RINGS;
4609 
4610 	gfx_v8_0_kiq_setting(ring);
4611 
4612 	if (adev->in_gpu_reset) { /* for GPU_RESET case */
4613 		/* reset MQD to a clean status */
4614 		if (adev->gfx.mec.mqd_backup[mqd_idx])
4615 			memcpy(mqd, adev->gfx.mec.mqd_backup[mqd_idx], sizeof(struct vi_mqd_allocation));
4616 
4617 		/* reset ring buffer */
4618 		ring->wptr = 0;
4619 		amdgpu_ring_clear_ring(ring);
4620 		mutex_lock(&adev->srbm_mutex);
4621 		vi_srbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
4622 		gfx_v8_0_mqd_commit(adev, mqd);
4623 		vi_srbm_select(adev, 0, 0, 0, 0);
4624 		mutex_unlock(&adev->srbm_mutex);
4625 	} else {
4626 		memset((void *)mqd, 0, sizeof(struct vi_mqd_allocation));
4627 		((struct vi_mqd_allocation *)mqd)->dynamic_cu_mask = 0xFFFFFFFF;
4628 		((struct vi_mqd_allocation *)mqd)->dynamic_rb_mask = 0xFFFFFFFF;
4629 		mutex_lock(&adev->srbm_mutex);
4630 		vi_srbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
4631 		gfx_v8_0_mqd_init(ring);
4632 		gfx_v8_0_mqd_commit(adev, mqd);
4633 		vi_srbm_select(adev, 0, 0, 0, 0);
4634 		mutex_unlock(&adev->srbm_mutex);
4635 
4636 		if (adev->gfx.mec.mqd_backup[mqd_idx])
4637 			memcpy(adev->gfx.mec.mqd_backup[mqd_idx], mqd, sizeof(struct vi_mqd_allocation));
4638 	}
4639 
4640 	return 0;
4641 }
4642 
4643 static int gfx_v8_0_kcq_init_queue(struct amdgpu_ring *ring)
4644 {
4645 	struct amdgpu_device *adev = ring->adev;
4646 	struct vi_mqd *mqd = ring->mqd_ptr;
4647 	int mqd_idx = ring - &adev->gfx.compute_ring[0];
4648 
4649 	if (!adev->in_gpu_reset && !adev->in_suspend) {
4650 		memset((void *)mqd, 0, sizeof(struct vi_mqd_allocation));
4651 		((struct vi_mqd_allocation *)mqd)->dynamic_cu_mask = 0xFFFFFFFF;
4652 		((struct vi_mqd_allocation *)mqd)->dynamic_rb_mask = 0xFFFFFFFF;
4653 		mutex_lock(&adev->srbm_mutex);
4654 		vi_srbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
4655 		gfx_v8_0_mqd_init(ring);
4656 		vi_srbm_select(adev, 0, 0, 0, 0);
4657 		mutex_unlock(&adev->srbm_mutex);
4658 
4659 		if (adev->gfx.mec.mqd_backup[mqd_idx])
4660 			memcpy(adev->gfx.mec.mqd_backup[mqd_idx], mqd, sizeof(struct vi_mqd_allocation));
4661 	} else if (adev->in_gpu_reset) { /* for GPU_RESET case */
4662 		/* reset MQD to a clean status */
4663 		if (adev->gfx.mec.mqd_backup[mqd_idx])
4664 			memcpy(mqd, adev->gfx.mec.mqd_backup[mqd_idx], sizeof(struct vi_mqd_allocation));
4665 		/* reset ring buffer */
4666 		ring->wptr = 0;
4667 		amdgpu_ring_clear_ring(ring);
4668 	} else {
4669 		amdgpu_ring_clear_ring(ring);
4670 	}
4671 	return 0;
4672 }
4673 
4674 static void gfx_v8_0_set_mec_doorbell_range(struct amdgpu_device *adev)
4675 {
4676 	if (adev->asic_type > CHIP_TONGA) {
4677 		WREG32(mmCP_MEC_DOORBELL_RANGE_LOWER, adev->doorbell_index.kiq << 2);
4678 		WREG32(mmCP_MEC_DOORBELL_RANGE_UPPER, adev->doorbell_index.mec_ring7 << 2);
4679 	}
4680 	/* enable doorbells */
4681 	WREG32_FIELD(CP_PQ_STATUS, DOORBELL_ENABLE, 1);
4682 }
4683 
4684 static int gfx_v8_0_kiq_resume(struct amdgpu_device *adev)
4685 {
4686 	struct amdgpu_ring *ring;
4687 	int r;
4688 
4689 	ring = &adev->gfx.kiq.ring;
4690 
4691 	r = amdgpu_bo_reserve(ring->mqd_obj, false);
4692 	if (unlikely(r != 0))
4693 		return r;
4694 
4695 	r = amdgpu_bo_kmap(ring->mqd_obj, &ring->mqd_ptr);
4696 	if (unlikely(r != 0))
4697 		return r;
4698 
4699 	gfx_v8_0_kiq_init_queue(ring);
4700 	amdgpu_bo_kunmap(ring->mqd_obj);
4701 	ring->mqd_ptr = NULL;
4702 	amdgpu_bo_unreserve(ring->mqd_obj);
4703 	ring->sched.ready = true;
4704 	return 0;
4705 }
4706 
4707 static int gfx_v8_0_kcq_resume(struct amdgpu_device *adev)
4708 {
4709 	struct amdgpu_ring *ring = NULL;
4710 	int r = 0, i;
4711 
4712 	gfx_v8_0_cp_compute_enable(adev, true);
4713 
4714 	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
4715 		ring = &adev->gfx.compute_ring[i];
4716 
4717 		r = amdgpu_bo_reserve(ring->mqd_obj, false);
4718 		if (unlikely(r != 0))
4719 			goto done;
4720 		r = amdgpu_bo_kmap(ring->mqd_obj, &ring->mqd_ptr);
4721 		if (!r) {
4722 			r = gfx_v8_0_kcq_init_queue(ring);
4723 			amdgpu_bo_kunmap(ring->mqd_obj);
4724 			ring->mqd_ptr = NULL;
4725 		}
4726 		amdgpu_bo_unreserve(ring->mqd_obj);
4727 		if (r)
4728 			goto done;
4729 	}
4730 
4731 	gfx_v8_0_set_mec_doorbell_range(adev);
4732 
4733 	r = gfx_v8_0_kiq_kcq_enable(adev);
4734 	if (r)
4735 		goto done;
4736 
4737 done:
4738 	return r;
4739 }
4740 
4741 static int gfx_v8_0_cp_test_all_rings(struct amdgpu_device *adev)
4742 {
4743 	int r, i;
4744 	struct amdgpu_ring *ring;
4745 
4746 	/* collect all the ring_tests here, gfx, kiq, compute */
4747 	ring = &adev->gfx.gfx_ring[0];
4748 	r = amdgpu_ring_test_helper(ring);
4749 	if (r)
4750 		return r;
4751 
4752 	ring = &adev->gfx.kiq.ring;
4753 	r = amdgpu_ring_test_helper(ring);
4754 	if (r)
4755 		return r;
4756 
4757 	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
4758 		ring = &adev->gfx.compute_ring[i];
4759 		amdgpu_ring_test_helper(ring);
4760 	}
4761 
4762 	return 0;
4763 }
4764 
4765 static int gfx_v8_0_cp_resume(struct amdgpu_device *adev)
4766 {
4767 	int r;
4768 
4769 	if (!(adev->flags & AMD_IS_APU))
4770 		gfx_v8_0_enable_gui_idle_interrupt(adev, false);
4771 
4772 	r = gfx_v8_0_kiq_resume(adev);
4773 	if (r)
4774 		return r;
4775 
4776 	r = gfx_v8_0_cp_gfx_resume(adev);
4777 	if (r)
4778 		return r;
4779 
4780 	r = gfx_v8_0_kcq_resume(adev);
4781 	if (r)
4782 		return r;
4783 
4784 	r = gfx_v8_0_cp_test_all_rings(adev);
4785 	if (r)
4786 		return r;
4787 
4788 	gfx_v8_0_enable_gui_idle_interrupt(adev, true);
4789 
4790 	return 0;
4791 }
4792 
4793 static void gfx_v8_0_cp_enable(struct amdgpu_device *adev, bool enable)
4794 {
4795 	gfx_v8_0_cp_gfx_enable(adev, enable);
4796 	gfx_v8_0_cp_compute_enable(adev, enable);
4797 }
4798 
4799 static int gfx_v8_0_hw_init(void *handle)
4800 {
4801 	int r;
4802 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4803 
4804 	gfx_v8_0_init_golden_registers(adev);
4805 	gfx_v8_0_constants_init(adev);
4806 
4807 	r = adev->gfx.rlc.funcs->resume(adev);
4808 	if (r)
4809 		return r;
4810 
4811 	r = gfx_v8_0_cp_resume(adev);
4812 
4813 	return r;
4814 }
4815 
4816 static int gfx_v8_0_kcq_disable(struct amdgpu_device *adev)
4817 {
4818 	int r, i;
4819 	struct amdgpu_ring *kiq_ring = &adev->gfx.kiq.ring;
4820 
4821 	r = amdgpu_ring_alloc(kiq_ring, 6 * adev->gfx.num_compute_rings);
4822 	if (r)
4823 		DRM_ERROR("Failed to lock KIQ (%d).\n", r);
4824 
4825 	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
4826 		struct amdgpu_ring *ring = &adev->gfx.compute_ring[i];
4827 
4828 		amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_UNMAP_QUEUES, 4));
4829 		amdgpu_ring_write(kiq_ring, /* Q_sel: 0, vmid: 0, engine: 0, num_Q: 1 */
4830 						PACKET3_UNMAP_QUEUES_ACTION(1) | /* RESET_QUEUES */
4831 						PACKET3_UNMAP_QUEUES_QUEUE_SEL(0) |
4832 						PACKET3_UNMAP_QUEUES_ENGINE_SEL(0) |
4833 						PACKET3_UNMAP_QUEUES_NUM_QUEUES(1));
4834 		amdgpu_ring_write(kiq_ring, PACKET3_UNMAP_QUEUES_DOORBELL_OFFSET0(ring->doorbell_index));
4835 		amdgpu_ring_write(kiq_ring, 0);
4836 		amdgpu_ring_write(kiq_ring, 0);
4837 		amdgpu_ring_write(kiq_ring, 0);
4838 	}
4839 	r = amdgpu_ring_test_helper(kiq_ring);
4840 	if (r)
4841 		DRM_ERROR("KCQ disable failed\n");
4842 
4843 	return r;
4844 }
4845 
4846 static bool gfx_v8_0_is_idle(void *handle)
4847 {
4848 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4849 
4850 	if (REG_GET_FIELD(RREG32(mmGRBM_STATUS), GRBM_STATUS, GUI_ACTIVE)
4851 		|| RREG32(mmGRBM_STATUS2) != 0x8)
4852 		return false;
4853 	else
4854 		return true;
4855 }
4856 
4857 static bool gfx_v8_0_rlc_is_idle(void *handle)
4858 {
4859 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4860 
4861 	if (RREG32(mmGRBM_STATUS2) != 0x8)
4862 		return false;
4863 	else
4864 		return true;
4865 }
4866 
4867 static int gfx_v8_0_wait_for_rlc_idle(void *handle)
4868 {
4869 	unsigned int i;
4870 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4871 
4872 	for (i = 0; i < adev->usec_timeout; i++) {
4873 		if (gfx_v8_0_rlc_is_idle(handle))
4874 			return 0;
4875 
4876 		udelay(1);
4877 	}
4878 	return -ETIMEDOUT;
4879 }
4880 
4881 static int gfx_v8_0_wait_for_idle(void *handle)
4882 {
4883 	unsigned int i;
4884 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4885 
4886 	for (i = 0; i < adev->usec_timeout; i++) {
4887 		if (gfx_v8_0_is_idle(handle))
4888 			return 0;
4889 
4890 		udelay(1);
4891 	}
4892 	return -ETIMEDOUT;
4893 }
4894 
4895 static int gfx_v8_0_hw_fini(void *handle)
4896 {
4897 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4898 
4899 	amdgpu_irq_put(adev, &adev->gfx.priv_reg_irq, 0);
4900 	amdgpu_irq_put(adev, &adev->gfx.priv_inst_irq, 0);
4901 
4902 	amdgpu_irq_put(adev, &adev->gfx.cp_ecc_error_irq, 0);
4903 
4904 	amdgpu_irq_put(adev, &adev->gfx.sq_irq, 0);
4905 
4906 	/* disable KCQ to avoid CPC touch memory not valid anymore */
4907 	gfx_v8_0_kcq_disable(adev);
4908 
4909 	if (amdgpu_sriov_vf(adev)) {
4910 		pr_debug("For SRIOV client, shouldn't do anything.\n");
4911 		return 0;
4912 	}
4913 	amdgpu_gfx_rlc_enter_safe_mode(adev);
4914 	if (!gfx_v8_0_wait_for_idle(adev))
4915 		gfx_v8_0_cp_enable(adev, false);
4916 	else
4917 		pr_err("cp is busy, skip halt cp\n");
4918 	if (!gfx_v8_0_wait_for_rlc_idle(adev))
4919 		adev->gfx.rlc.funcs->stop(adev);
4920 	else
4921 		pr_err("rlc is busy, skip halt rlc\n");
4922 	amdgpu_gfx_rlc_exit_safe_mode(adev);
4923 
4924 	return 0;
4925 }
4926 
4927 static int gfx_v8_0_suspend(void *handle)
4928 {
4929 	return gfx_v8_0_hw_fini(handle);
4930 }
4931 
4932 static int gfx_v8_0_resume(void *handle)
4933 {
4934 	return gfx_v8_0_hw_init(handle);
4935 }
4936 
4937 static bool gfx_v8_0_check_soft_reset(void *handle)
4938 {
4939 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4940 	u32 grbm_soft_reset = 0, srbm_soft_reset = 0;
4941 	u32 tmp;
4942 
4943 	/* GRBM_STATUS */
4944 	tmp = RREG32(mmGRBM_STATUS);
4945 	if (tmp & (GRBM_STATUS__PA_BUSY_MASK | GRBM_STATUS__SC_BUSY_MASK |
4946 		   GRBM_STATUS__BCI_BUSY_MASK | GRBM_STATUS__SX_BUSY_MASK |
4947 		   GRBM_STATUS__TA_BUSY_MASK | GRBM_STATUS__VGT_BUSY_MASK |
4948 		   GRBM_STATUS__DB_BUSY_MASK | GRBM_STATUS__CB_BUSY_MASK |
4949 		   GRBM_STATUS__GDS_BUSY_MASK | GRBM_STATUS__SPI_BUSY_MASK |
4950 		   GRBM_STATUS__IA_BUSY_MASK | GRBM_STATUS__IA_BUSY_NO_DMA_MASK |
4951 		   GRBM_STATUS__CP_BUSY_MASK | GRBM_STATUS__CP_COHERENCY_BUSY_MASK)) {
4952 		grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset,
4953 						GRBM_SOFT_RESET, SOFT_RESET_CP, 1);
4954 		grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset,
4955 						GRBM_SOFT_RESET, SOFT_RESET_GFX, 1);
4956 		srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset,
4957 						SRBM_SOFT_RESET, SOFT_RESET_GRBM, 1);
4958 	}
4959 
4960 	/* GRBM_STATUS2 */
4961 	tmp = RREG32(mmGRBM_STATUS2);
4962 	if (REG_GET_FIELD(tmp, GRBM_STATUS2, RLC_BUSY))
4963 		grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset,
4964 						GRBM_SOFT_RESET, SOFT_RESET_RLC, 1);
4965 
4966 	if (REG_GET_FIELD(tmp, GRBM_STATUS2, CPF_BUSY) ||
4967 	    REG_GET_FIELD(tmp, GRBM_STATUS2, CPC_BUSY) ||
4968 	    REG_GET_FIELD(tmp, GRBM_STATUS2, CPG_BUSY)) {
4969 		grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET,
4970 						SOFT_RESET_CPF, 1);
4971 		grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET,
4972 						SOFT_RESET_CPC, 1);
4973 		grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET,
4974 						SOFT_RESET_CPG, 1);
4975 		srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset, SRBM_SOFT_RESET,
4976 						SOFT_RESET_GRBM, 1);
4977 	}
4978 
4979 	/* SRBM_STATUS */
4980 	tmp = RREG32(mmSRBM_STATUS);
4981 	if (REG_GET_FIELD(tmp, SRBM_STATUS, GRBM_RQ_PENDING))
4982 		srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset,
4983 						SRBM_SOFT_RESET, SOFT_RESET_GRBM, 1);
4984 	if (REG_GET_FIELD(tmp, SRBM_STATUS, SEM_BUSY))
4985 		srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset,
4986 						SRBM_SOFT_RESET, SOFT_RESET_SEM, 1);
4987 
4988 	if (grbm_soft_reset || srbm_soft_reset) {
4989 		adev->gfx.grbm_soft_reset = grbm_soft_reset;
4990 		adev->gfx.srbm_soft_reset = srbm_soft_reset;
4991 		return true;
4992 	} else {
4993 		adev->gfx.grbm_soft_reset = 0;
4994 		adev->gfx.srbm_soft_reset = 0;
4995 		return false;
4996 	}
4997 }
4998 
4999 static int gfx_v8_0_pre_soft_reset(void *handle)
5000 {
5001 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5002 	u32 grbm_soft_reset = 0;
5003 
5004 	if ((!adev->gfx.grbm_soft_reset) &&
5005 	    (!adev->gfx.srbm_soft_reset))
5006 		return 0;
5007 
5008 	grbm_soft_reset = adev->gfx.grbm_soft_reset;
5009 
5010 	/* stop the rlc */
5011 	adev->gfx.rlc.funcs->stop(adev);
5012 
5013 	if (REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CP) ||
5014 	    REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_GFX))
5015 		/* Disable GFX parsing/prefetching */
5016 		gfx_v8_0_cp_gfx_enable(adev, false);
5017 
5018 	if (REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CP) ||
5019 	    REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CPF) ||
5020 	    REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CPC) ||
5021 	    REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CPG)) {
5022 		int i;
5023 
5024 		for (i = 0; i < adev->gfx.num_compute_rings; i++) {
5025 			struct amdgpu_ring *ring = &adev->gfx.compute_ring[i];
5026 
5027 			mutex_lock(&adev->srbm_mutex);
5028 			vi_srbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
5029 			gfx_v8_0_deactivate_hqd(adev, 2);
5030 			vi_srbm_select(adev, 0, 0, 0, 0);
5031 			mutex_unlock(&adev->srbm_mutex);
5032 		}
5033 		/* Disable MEC parsing/prefetching */
5034 		gfx_v8_0_cp_compute_enable(adev, false);
5035 	}
5036 
5037        return 0;
5038 }
5039 
5040 static int gfx_v8_0_soft_reset(void *handle)
5041 {
5042 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5043 	u32 grbm_soft_reset = 0, srbm_soft_reset = 0;
5044 	u32 tmp;
5045 
5046 	if ((!adev->gfx.grbm_soft_reset) &&
5047 	    (!adev->gfx.srbm_soft_reset))
5048 		return 0;
5049 
5050 	grbm_soft_reset = adev->gfx.grbm_soft_reset;
5051 	srbm_soft_reset = adev->gfx.srbm_soft_reset;
5052 
5053 	if (grbm_soft_reset || srbm_soft_reset) {
5054 		tmp = RREG32(mmGMCON_DEBUG);
5055 		tmp = REG_SET_FIELD(tmp, GMCON_DEBUG, GFX_STALL, 1);
5056 		tmp = REG_SET_FIELD(tmp, GMCON_DEBUG, GFX_CLEAR, 1);
5057 		WREG32(mmGMCON_DEBUG, tmp);
5058 		udelay(50);
5059 	}
5060 
5061 	if (grbm_soft_reset) {
5062 		tmp = RREG32(mmGRBM_SOFT_RESET);
5063 		tmp |= grbm_soft_reset;
5064 		dev_info(adev->dev, "GRBM_SOFT_RESET=0x%08X\n", tmp);
5065 		WREG32(mmGRBM_SOFT_RESET, tmp);
5066 		tmp = RREG32(mmGRBM_SOFT_RESET);
5067 
5068 		udelay(50);
5069 
5070 		tmp &= ~grbm_soft_reset;
5071 		WREG32(mmGRBM_SOFT_RESET, tmp);
5072 		tmp = RREG32(mmGRBM_SOFT_RESET);
5073 	}
5074 
5075 	if (srbm_soft_reset) {
5076 		tmp = RREG32(mmSRBM_SOFT_RESET);
5077 		tmp |= srbm_soft_reset;
5078 		dev_info(adev->dev, "SRBM_SOFT_RESET=0x%08X\n", tmp);
5079 		WREG32(mmSRBM_SOFT_RESET, tmp);
5080 		tmp = RREG32(mmSRBM_SOFT_RESET);
5081 
5082 		udelay(50);
5083 
5084 		tmp &= ~srbm_soft_reset;
5085 		WREG32(mmSRBM_SOFT_RESET, tmp);
5086 		tmp = RREG32(mmSRBM_SOFT_RESET);
5087 	}
5088 
5089 	if (grbm_soft_reset || srbm_soft_reset) {
5090 		tmp = RREG32(mmGMCON_DEBUG);
5091 		tmp = REG_SET_FIELD(tmp, GMCON_DEBUG, GFX_STALL, 0);
5092 		tmp = REG_SET_FIELD(tmp, GMCON_DEBUG, GFX_CLEAR, 0);
5093 		WREG32(mmGMCON_DEBUG, tmp);
5094 	}
5095 
5096 	/* Wait a little for things to settle down */
5097 	udelay(50);
5098 
5099 	return 0;
5100 }
5101 
5102 static int gfx_v8_0_post_soft_reset(void *handle)
5103 {
5104 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5105 	u32 grbm_soft_reset = 0;
5106 
5107 	if ((!adev->gfx.grbm_soft_reset) &&
5108 	    (!adev->gfx.srbm_soft_reset))
5109 		return 0;
5110 
5111 	grbm_soft_reset = adev->gfx.grbm_soft_reset;
5112 
5113 	if (REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CP) ||
5114 	    REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CPF) ||
5115 	    REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CPC) ||
5116 	    REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CPG)) {
5117 		int i;
5118 
5119 		for (i = 0; i < adev->gfx.num_compute_rings; i++) {
5120 			struct amdgpu_ring *ring = &adev->gfx.compute_ring[i];
5121 
5122 			mutex_lock(&adev->srbm_mutex);
5123 			vi_srbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
5124 			gfx_v8_0_deactivate_hqd(adev, 2);
5125 			vi_srbm_select(adev, 0, 0, 0, 0);
5126 			mutex_unlock(&adev->srbm_mutex);
5127 		}
5128 		gfx_v8_0_kiq_resume(adev);
5129 		gfx_v8_0_kcq_resume(adev);
5130 	}
5131 
5132 	if (REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CP) ||
5133 	    REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_GFX))
5134 		gfx_v8_0_cp_gfx_resume(adev);
5135 
5136 	gfx_v8_0_cp_test_all_rings(adev);
5137 
5138 	adev->gfx.rlc.funcs->start(adev);
5139 
5140 	return 0;
5141 }
5142 
5143 /**
5144  * gfx_v8_0_get_gpu_clock_counter - return GPU clock counter snapshot
5145  *
5146  * @adev: amdgpu_device pointer
5147  *
5148  * Fetches a GPU clock counter snapshot.
5149  * Returns the 64 bit clock counter snapshot.
5150  */
5151 static uint64_t gfx_v8_0_get_gpu_clock_counter(struct amdgpu_device *adev)
5152 {
5153 	uint64_t clock;
5154 
5155 	mutex_lock(&adev->gfx.gpu_clock_mutex);
5156 	WREG32(mmRLC_CAPTURE_GPU_CLOCK_COUNT, 1);
5157 	clock = (uint64_t)RREG32(mmRLC_GPU_CLOCK_COUNT_LSB) |
5158 		((uint64_t)RREG32(mmRLC_GPU_CLOCK_COUNT_MSB) << 32ULL);
5159 	mutex_unlock(&adev->gfx.gpu_clock_mutex);
5160 	return clock;
5161 }
5162 
5163 static void gfx_v8_0_ring_emit_gds_switch(struct amdgpu_ring *ring,
5164 					  uint32_t vmid,
5165 					  uint32_t gds_base, uint32_t gds_size,
5166 					  uint32_t gws_base, uint32_t gws_size,
5167 					  uint32_t oa_base, uint32_t oa_size)
5168 {
5169 	/* GDS Base */
5170 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
5171 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
5172 				WRITE_DATA_DST_SEL(0)));
5173 	amdgpu_ring_write(ring, amdgpu_gds_reg_offset[vmid].mem_base);
5174 	amdgpu_ring_write(ring, 0);
5175 	amdgpu_ring_write(ring, gds_base);
5176 
5177 	/* GDS Size */
5178 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
5179 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
5180 				WRITE_DATA_DST_SEL(0)));
5181 	amdgpu_ring_write(ring, amdgpu_gds_reg_offset[vmid].mem_size);
5182 	amdgpu_ring_write(ring, 0);
5183 	amdgpu_ring_write(ring, gds_size);
5184 
5185 	/* GWS */
5186 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
5187 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
5188 				WRITE_DATA_DST_SEL(0)));
5189 	amdgpu_ring_write(ring, amdgpu_gds_reg_offset[vmid].gws);
5190 	amdgpu_ring_write(ring, 0);
5191 	amdgpu_ring_write(ring, gws_size << GDS_GWS_VMID0__SIZE__SHIFT | gws_base);
5192 
5193 	/* OA */
5194 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
5195 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
5196 				WRITE_DATA_DST_SEL(0)));
5197 	amdgpu_ring_write(ring, amdgpu_gds_reg_offset[vmid].oa);
5198 	amdgpu_ring_write(ring, 0);
5199 	amdgpu_ring_write(ring, (1 << (oa_size + oa_base)) - (1 << oa_base));
5200 }
5201 
5202 static uint32_t wave_read_ind(struct amdgpu_device *adev, uint32_t simd, uint32_t wave, uint32_t address)
5203 {
5204 	WREG32(mmSQ_IND_INDEX,
5205 		(wave << SQ_IND_INDEX__WAVE_ID__SHIFT) |
5206 		(simd << SQ_IND_INDEX__SIMD_ID__SHIFT) |
5207 		(address << SQ_IND_INDEX__INDEX__SHIFT) |
5208 		(SQ_IND_INDEX__FORCE_READ_MASK));
5209 	return RREG32(mmSQ_IND_DATA);
5210 }
5211 
5212 static void wave_read_regs(struct amdgpu_device *adev, uint32_t simd,
5213 			   uint32_t wave, uint32_t thread,
5214 			   uint32_t regno, uint32_t num, uint32_t *out)
5215 {
5216 	WREG32(mmSQ_IND_INDEX,
5217 		(wave << SQ_IND_INDEX__WAVE_ID__SHIFT) |
5218 		(simd << SQ_IND_INDEX__SIMD_ID__SHIFT) |
5219 		(regno << SQ_IND_INDEX__INDEX__SHIFT) |
5220 		(thread << SQ_IND_INDEX__THREAD_ID__SHIFT) |
5221 		(SQ_IND_INDEX__FORCE_READ_MASK) |
5222 		(SQ_IND_INDEX__AUTO_INCR_MASK));
5223 	while (num--)
5224 		*(out++) = RREG32(mmSQ_IND_DATA);
5225 }
5226 
5227 static void gfx_v8_0_read_wave_data(struct amdgpu_device *adev, uint32_t simd, uint32_t wave, uint32_t *dst, int *no_fields)
5228 {
5229 	/* type 0 wave data */
5230 	dst[(*no_fields)++] = 0;
5231 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_STATUS);
5232 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_PC_LO);
5233 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_PC_HI);
5234 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_EXEC_LO);
5235 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_EXEC_HI);
5236 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_HW_ID);
5237 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_INST_DW0);
5238 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_INST_DW1);
5239 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_GPR_ALLOC);
5240 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_LDS_ALLOC);
5241 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TRAPSTS);
5242 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_IB_STS);
5243 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TBA_LO);
5244 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TBA_HI);
5245 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TMA_LO);
5246 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TMA_HI);
5247 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_IB_DBG0);
5248 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_M0);
5249 }
5250 
5251 static void gfx_v8_0_read_wave_sgprs(struct amdgpu_device *adev, uint32_t simd,
5252 				     uint32_t wave, uint32_t start,
5253 				     uint32_t size, uint32_t *dst)
5254 {
5255 	wave_read_regs(
5256 		adev, simd, wave, 0,
5257 		start + SQIND_WAVE_SGPRS_OFFSET, size, dst);
5258 }
5259 
5260 
5261 static const struct amdgpu_gfx_funcs gfx_v8_0_gfx_funcs = {
5262 	.get_gpu_clock_counter = &gfx_v8_0_get_gpu_clock_counter,
5263 	.select_se_sh = &gfx_v8_0_select_se_sh,
5264 	.read_wave_data = &gfx_v8_0_read_wave_data,
5265 	.read_wave_sgprs = &gfx_v8_0_read_wave_sgprs,
5266 	.select_me_pipe_q = &gfx_v8_0_select_me_pipe_q
5267 };
5268 
5269 static int gfx_v8_0_early_init(void *handle)
5270 {
5271 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5272 
5273 	adev->gfx.num_gfx_rings = GFX8_NUM_GFX_RINGS;
5274 	adev->gfx.num_compute_rings = AMDGPU_MAX_COMPUTE_RINGS;
5275 	adev->gfx.funcs = &gfx_v8_0_gfx_funcs;
5276 	gfx_v8_0_set_ring_funcs(adev);
5277 	gfx_v8_0_set_irq_funcs(adev);
5278 	gfx_v8_0_set_gds_init(adev);
5279 	gfx_v8_0_set_rlc_funcs(adev);
5280 
5281 	return 0;
5282 }
5283 
5284 static int gfx_v8_0_late_init(void *handle)
5285 {
5286 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5287 	int r;
5288 
5289 	r = amdgpu_irq_get(adev, &adev->gfx.priv_reg_irq, 0);
5290 	if (r)
5291 		return r;
5292 
5293 	r = amdgpu_irq_get(adev, &adev->gfx.priv_inst_irq, 0);
5294 	if (r)
5295 		return r;
5296 
5297 	/* requires IBs so do in late init after IB pool is initialized */
5298 	r = gfx_v8_0_do_edc_gpr_workarounds(adev);
5299 	if (r)
5300 		return r;
5301 
5302 	r = amdgpu_irq_get(adev, &adev->gfx.cp_ecc_error_irq, 0);
5303 	if (r) {
5304 		DRM_ERROR("amdgpu_irq_get() failed to get IRQ for EDC, r: %d.\n", r);
5305 		return r;
5306 	}
5307 
5308 	r = amdgpu_irq_get(adev, &adev->gfx.sq_irq, 0);
5309 	if (r) {
5310 		DRM_ERROR(
5311 			"amdgpu_irq_get() failed to get IRQ for SQ, r: %d.\n",
5312 			r);
5313 		return r;
5314 	}
5315 
5316 	return 0;
5317 }
5318 
5319 static void gfx_v8_0_enable_gfx_static_mg_power_gating(struct amdgpu_device *adev,
5320 						       bool enable)
5321 {
5322 	if (((adev->asic_type == CHIP_POLARIS11) ||
5323 	    (adev->asic_type == CHIP_POLARIS12) ||
5324 	    (adev->asic_type == CHIP_VEGAM)) &&
5325 	    adev->powerplay.pp_funcs->set_powergating_by_smu)
5326 		/* Send msg to SMU via Powerplay */
5327 		amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_GFX, enable);
5328 
5329 	WREG32_FIELD(RLC_PG_CNTL, STATIC_PER_CU_PG_ENABLE, enable ? 1 : 0);
5330 }
5331 
5332 static void gfx_v8_0_enable_gfx_dynamic_mg_power_gating(struct amdgpu_device *adev,
5333 							bool enable)
5334 {
5335 	WREG32_FIELD(RLC_PG_CNTL, DYN_PER_CU_PG_ENABLE, enable ? 1 : 0);
5336 }
5337 
5338 static void polaris11_enable_gfx_quick_mg_power_gating(struct amdgpu_device *adev,
5339 		bool enable)
5340 {
5341 	WREG32_FIELD(RLC_PG_CNTL, QUICK_PG_ENABLE, enable ? 1 : 0);
5342 }
5343 
5344 static void cz_enable_gfx_cg_power_gating(struct amdgpu_device *adev,
5345 					  bool enable)
5346 {
5347 	WREG32_FIELD(RLC_PG_CNTL, GFX_POWER_GATING_ENABLE, enable ? 1 : 0);
5348 }
5349 
5350 static void cz_enable_gfx_pipeline_power_gating(struct amdgpu_device *adev,
5351 						bool enable)
5352 {
5353 	WREG32_FIELD(RLC_PG_CNTL, GFX_PIPELINE_PG_ENABLE, enable ? 1 : 0);
5354 
5355 	/* Read any GFX register to wake up GFX. */
5356 	if (!enable)
5357 		RREG32(mmDB_RENDER_CONTROL);
5358 }
5359 
5360 static void cz_update_gfx_cg_power_gating(struct amdgpu_device *adev,
5361 					  bool enable)
5362 {
5363 	if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_PG) && enable) {
5364 		cz_enable_gfx_cg_power_gating(adev, true);
5365 		if (adev->pg_flags & AMD_PG_SUPPORT_GFX_PIPELINE)
5366 			cz_enable_gfx_pipeline_power_gating(adev, true);
5367 	} else {
5368 		cz_enable_gfx_cg_power_gating(adev, false);
5369 		cz_enable_gfx_pipeline_power_gating(adev, false);
5370 	}
5371 }
5372 
5373 static int gfx_v8_0_set_powergating_state(void *handle,
5374 					  enum amd_powergating_state state)
5375 {
5376 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5377 	bool enable = (state == AMD_PG_STATE_GATE);
5378 
5379 	if (amdgpu_sriov_vf(adev))
5380 		return 0;
5381 
5382 	if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_SMG |
5383 				AMD_PG_SUPPORT_RLC_SMU_HS |
5384 				AMD_PG_SUPPORT_CP |
5385 				AMD_PG_SUPPORT_GFX_DMG))
5386 		amdgpu_gfx_rlc_enter_safe_mode(adev);
5387 	switch (adev->asic_type) {
5388 	case CHIP_CARRIZO:
5389 	case CHIP_STONEY:
5390 
5391 		if (adev->pg_flags & AMD_PG_SUPPORT_RLC_SMU_HS) {
5392 			cz_enable_sck_slow_down_on_power_up(adev, true);
5393 			cz_enable_sck_slow_down_on_power_down(adev, true);
5394 		} else {
5395 			cz_enable_sck_slow_down_on_power_up(adev, false);
5396 			cz_enable_sck_slow_down_on_power_down(adev, false);
5397 		}
5398 		if (adev->pg_flags & AMD_PG_SUPPORT_CP)
5399 			cz_enable_cp_power_gating(adev, true);
5400 		else
5401 			cz_enable_cp_power_gating(adev, false);
5402 
5403 		cz_update_gfx_cg_power_gating(adev, enable);
5404 
5405 		if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_SMG) && enable)
5406 			gfx_v8_0_enable_gfx_static_mg_power_gating(adev, true);
5407 		else
5408 			gfx_v8_0_enable_gfx_static_mg_power_gating(adev, false);
5409 
5410 		if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_DMG) && enable)
5411 			gfx_v8_0_enable_gfx_dynamic_mg_power_gating(adev, true);
5412 		else
5413 			gfx_v8_0_enable_gfx_dynamic_mg_power_gating(adev, false);
5414 		break;
5415 	case CHIP_POLARIS11:
5416 	case CHIP_POLARIS12:
5417 	case CHIP_VEGAM:
5418 		if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_SMG) && enable)
5419 			gfx_v8_0_enable_gfx_static_mg_power_gating(adev, true);
5420 		else
5421 			gfx_v8_0_enable_gfx_static_mg_power_gating(adev, false);
5422 
5423 		if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_DMG) && enable)
5424 			gfx_v8_0_enable_gfx_dynamic_mg_power_gating(adev, true);
5425 		else
5426 			gfx_v8_0_enable_gfx_dynamic_mg_power_gating(adev, false);
5427 
5428 		if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_QUICK_MG) && enable)
5429 			polaris11_enable_gfx_quick_mg_power_gating(adev, true);
5430 		else
5431 			polaris11_enable_gfx_quick_mg_power_gating(adev, false);
5432 		break;
5433 	default:
5434 		break;
5435 	}
5436 	if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_SMG |
5437 				AMD_PG_SUPPORT_RLC_SMU_HS |
5438 				AMD_PG_SUPPORT_CP |
5439 				AMD_PG_SUPPORT_GFX_DMG))
5440 		amdgpu_gfx_rlc_exit_safe_mode(adev);
5441 	return 0;
5442 }
5443 
5444 static void gfx_v8_0_get_clockgating_state(void *handle, u32 *flags)
5445 {
5446 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5447 	int data;
5448 
5449 	if (amdgpu_sriov_vf(adev))
5450 		*flags = 0;
5451 
5452 	/* AMD_CG_SUPPORT_GFX_MGCG */
5453 	data = RREG32(mmRLC_CGTT_MGCG_OVERRIDE);
5454 	if (!(data & RLC_CGTT_MGCG_OVERRIDE__CPF_MASK))
5455 		*flags |= AMD_CG_SUPPORT_GFX_MGCG;
5456 
5457 	/* AMD_CG_SUPPORT_GFX_CGLG */
5458 	data = RREG32(mmRLC_CGCG_CGLS_CTRL);
5459 	if (data & RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK)
5460 		*flags |= AMD_CG_SUPPORT_GFX_CGCG;
5461 
5462 	/* AMD_CG_SUPPORT_GFX_CGLS */
5463 	if (data & RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK)
5464 		*flags |= AMD_CG_SUPPORT_GFX_CGLS;
5465 
5466 	/* AMD_CG_SUPPORT_GFX_CGTS */
5467 	data = RREG32(mmCGTS_SM_CTRL_REG);
5468 	if (!(data & CGTS_SM_CTRL_REG__OVERRIDE_MASK))
5469 		*flags |= AMD_CG_SUPPORT_GFX_CGTS;
5470 
5471 	/* AMD_CG_SUPPORT_GFX_CGTS_LS */
5472 	if (!(data & CGTS_SM_CTRL_REG__LS_OVERRIDE_MASK))
5473 		*flags |= AMD_CG_SUPPORT_GFX_CGTS_LS;
5474 
5475 	/* AMD_CG_SUPPORT_GFX_RLC_LS */
5476 	data = RREG32(mmRLC_MEM_SLP_CNTL);
5477 	if (data & RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK)
5478 		*flags |= AMD_CG_SUPPORT_GFX_RLC_LS | AMD_CG_SUPPORT_GFX_MGLS;
5479 
5480 	/* AMD_CG_SUPPORT_GFX_CP_LS */
5481 	data = RREG32(mmCP_MEM_SLP_CNTL);
5482 	if (data & CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK)
5483 		*flags |= AMD_CG_SUPPORT_GFX_CP_LS | AMD_CG_SUPPORT_GFX_MGLS;
5484 }
5485 
5486 static void gfx_v8_0_send_serdes_cmd(struct amdgpu_device *adev,
5487 				     uint32_t reg_addr, uint32_t cmd)
5488 {
5489 	uint32_t data;
5490 
5491 	gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
5492 
5493 	WREG32(mmRLC_SERDES_WR_CU_MASTER_MASK, 0xffffffff);
5494 	WREG32(mmRLC_SERDES_WR_NONCU_MASTER_MASK, 0xffffffff);
5495 
5496 	data = RREG32(mmRLC_SERDES_WR_CTRL);
5497 	if (adev->asic_type == CHIP_STONEY)
5498 		data &= ~(RLC_SERDES_WR_CTRL__WRITE_COMMAND_MASK |
5499 			  RLC_SERDES_WR_CTRL__READ_COMMAND_MASK |
5500 			  RLC_SERDES_WR_CTRL__P1_SELECT_MASK |
5501 			  RLC_SERDES_WR_CTRL__P2_SELECT_MASK |
5502 			  RLC_SERDES_WR_CTRL__RDDATA_RESET_MASK |
5503 			  RLC_SERDES_WR_CTRL__POWER_DOWN_MASK |
5504 			  RLC_SERDES_WR_CTRL__POWER_UP_MASK |
5505 			  RLC_SERDES_WR_CTRL__SHORT_FORMAT_MASK |
5506 			  RLC_SERDES_WR_CTRL__SRBM_OVERRIDE_MASK);
5507 	else
5508 		data &= ~(RLC_SERDES_WR_CTRL__WRITE_COMMAND_MASK |
5509 			  RLC_SERDES_WR_CTRL__READ_COMMAND_MASK |
5510 			  RLC_SERDES_WR_CTRL__P1_SELECT_MASK |
5511 			  RLC_SERDES_WR_CTRL__P2_SELECT_MASK |
5512 			  RLC_SERDES_WR_CTRL__RDDATA_RESET_MASK |
5513 			  RLC_SERDES_WR_CTRL__POWER_DOWN_MASK |
5514 			  RLC_SERDES_WR_CTRL__POWER_UP_MASK |
5515 			  RLC_SERDES_WR_CTRL__SHORT_FORMAT_MASK |
5516 			  RLC_SERDES_WR_CTRL__BPM_DATA_MASK |
5517 			  RLC_SERDES_WR_CTRL__REG_ADDR_MASK |
5518 			  RLC_SERDES_WR_CTRL__SRBM_OVERRIDE_MASK);
5519 	data |= (RLC_SERDES_WR_CTRL__RSVD_BPM_ADDR_MASK |
5520 		 (cmd << RLC_SERDES_WR_CTRL__BPM_DATA__SHIFT) |
5521 		 (reg_addr << RLC_SERDES_WR_CTRL__REG_ADDR__SHIFT) |
5522 		 (0xff << RLC_SERDES_WR_CTRL__BPM_ADDR__SHIFT));
5523 
5524 	WREG32(mmRLC_SERDES_WR_CTRL, data);
5525 }
5526 
5527 #define MSG_ENTER_RLC_SAFE_MODE     1
5528 #define MSG_EXIT_RLC_SAFE_MODE      0
5529 #define RLC_GPR_REG2__REQ_MASK 0x00000001
5530 #define RLC_GPR_REG2__REQ__SHIFT 0
5531 #define RLC_GPR_REG2__MESSAGE__SHIFT 0x00000001
5532 #define RLC_GPR_REG2__MESSAGE_MASK 0x0000001e
5533 
5534 static bool gfx_v8_0_is_rlc_enabled(struct amdgpu_device *adev)
5535 {
5536 	uint32_t rlc_setting;
5537 
5538 	rlc_setting = RREG32(mmRLC_CNTL);
5539 	if (!(rlc_setting & RLC_CNTL__RLC_ENABLE_F32_MASK))
5540 		return false;
5541 
5542 	return true;
5543 }
5544 
5545 static void gfx_v8_0_set_safe_mode(struct amdgpu_device *adev)
5546 {
5547 	uint32_t data;
5548 	unsigned i;
5549 	data = RREG32(mmRLC_CNTL);
5550 	data |= RLC_SAFE_MODE__CMD_MASK;
5551 	data &= ~RLC_SAFE_MODE__MESSAGE_MASK;
5552 	data |= (1 << RLC_SAFE_MODE__MESSAGE__SHIFT);
5553 	WREG32(mmRLC_SAFE_MODE, data);
5554 
5555 	/* wait for RLC_SAFE_MODE */
5556 	for (i = 0; i < adev->usec_timeout; i++) {
5557 		if ((RREG32(mmRLC_GPM_STAT) &
5558 		     (RLC_GPM_STAT__GFX_CLOCK_STATUS_MASK |
5559 		      RLC_GPM_STAT__GFX_POWER_STATUS_MASK)) ==
5560 		    (RLC_GPM_STAT__GFX_CLOCK_STATUS_MASK |
5561 		     RLC_GPM_STAT__GFX_POWER_STATUS_MASK))
5562 			break;
5563 		udelay(1);
5564 	}
5565 	for (i = 0; i < adev->usec_timeout; i++) {
5566 		if (!REG_GET_FIELD(RREG32(mmRLC_SAFE_MODE), RLC_SAFE_MODE, CMD))
5567 			break;
5568 		udelay(1);
5569 	}
5570 }
5571 
5572 static void gfx_v8_0_unset_safe_mode(struct amdgpu_device *adev)
5573 {
5574 	uint32_t data;
5575 	unsigned i;
5576 
5577 	data = RREG32(mmRLC_CNTL);
5578 	data |= RLC_SAFE_MODE__CMD_MASK;
5579 	data &= ~RLC_SAFE_MODE__MESSAGE_MASK;
5580 	WREG32(mmRLC_SAFE_MODE, data);
5581 
5582 	for (i = 0; i < adev->usec_timeout; i++) {
5583 		if (!REG_GET_FIELD(RREG32(mmRLC_SAFE_MODE), RLC_SAFE_MODE, CMD))
5584 			break;
5585 		udelay(1);
5586 	}
5587 }
5588 
5589 static const struct amdgpu_rlc_funcs iceland_rlc_funcs = {
5590 	.is_rlc_enabled = gfx_v8_0_is_rlc_enabled,
5591 	.set_safe_mode = gfx_v8_0_set_safe_mode,
5592 	.unset_safe_mode = gfx_v8_0_unset_safe_mode,
5593 	.init = gfx_v8_0_rlc_init,
5594 	.get_csb_size = gfx_v8_0_get_csb_size,
5595 	.get_csb_buffer = gfx_v8_0_get_csb_buffer,
5596 	.get_cp_table_num = gfx_v8_0_cp_jump_table_num,
5597 	.resume = gfx_v8_0_rlc_resume,
5598 	.stop = gfx_v8_0_rlc_stop,
5599 	.reset = gfx_v8_0_rlc_reset,
5600 	.start = gfx_v8_0_rlc_start
5601 };
5602 
5603 static void gfx_v8_0_update_medium_grain_clock_gating(struct amdgpu_device *adev,
5604 						      bool enable)
5605 {
5606 	uint32_t temp, data;
5607 
5608 	amdgpu_gfx_rlc_enter_safe_mode(adev);
5609 
5610 	/* It is disabled by HW by default */
5611 	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGCG)) {
5612 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS) {
5613 			if (adev->cg_flags & AMD_CG_SUPPORT_GFX_RLC_LS)
5614 				/* 1 - RLC memory Light sleep */
5615 				WREG32_FIELD(RLC_MEM_SLP_CNTL, RLC_MEM_LS_EN, 1);
5616 
5617 			if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CP_LS)
5618 				WREG32_FIELD(CP_MEM_SLP_CNTL, CP_MEM_LS_EN, 1);
5619 		}
5620 
5621 		/* 3 - RLC_CGTT_MGCG_OVERRIDE */
5622 		temp = data = RREG32(mmRLC_CGTT_MGCG_OVERRIDE);
5623 		if (adev->flags & AMD_IS_APU)
5624 			data &= ~(RLC_CGTT_MGCG_OVERRIDE__CPF_MASK |
5625 				  RLC_CGTT_MGCG_OVERRIDE__RLC_MASK |
5626 				  RLC_CGTT_MGCG_OVERRIDE__MGCG_MASK);
5627 		else
5628 			data &= ~(RLC_CGTT_MGCG_OVERRIDE__CPF_MASK |
5629 				  RLC_CGTT_MGCG_OVERRIDE__RLC_MASK |
5630 				  RLC_CGTT_MGCG_OVERRIDE__MGCG_MASK |
5631 				  RLC_CGTT_MGCG_OVERRIDE__GRBM_MASK);
5632 
5633 		if (temp != data)
5634 			WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data);
5635 
5636 		/* 4 - wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */
5637 		gfx_v8_0_wait_for_rlc_serdes(adev);
5638 
5639 		/* 5 - clear mgcg override */
5640 		gfx_v8_0_send_serdes_cmd(adev, BPM_REG_MGCG_OVERRIDE, CLE_BPM_SERDES_CMD);
5641 
5642 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGTS) {
5643 			/* 6 - Enable CGTS(Tree Shade) MGCG /MGLS */
5644 			temp = data = RREG32(mmCGTS_SM_CTRL_REG);
5645 			data &= ~(CGTS_SM_CTRL_REG__SM_MODE_MASK);
5646 			data |= (0x2 << CGTS_SM_CTRL_REG__SM_MODE__SHIFT);
5647 			data |= CGTS_SM_CTRL_REG__SM_MODE_ENABLE_MASK;
5648 			data &= ~CGTS_SM_CTRL_REG__OVERRIDE_MASK;
5649 			if ((adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS) &&
5650 			    (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGTS_LS))
5651 				data &= ~CGTS_SM_CTRL_REG__LS_OVERRIDE_MASK;
5652 			data |= CGTS_SM_CTRL_REG__ON_MONITOR_ADD_EN_MASK;
5653 			data |= (0x96 << CGTS_SM_CTRL_REG__ON_MONITOR_ADD__SHIFT);
5654 			if (temp != data)
5655 				WREG32(mmCGTS_SM_CTRL_REG, data);
5656 		}
5657 		udelay(50);
5658 
5659 		/* 7 - wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */
5660 		gfx_v8_0_wait_for_rlc_serdes(adev);
5661 	} else {
5662 		/* 1 - MGCG_OVERRIDE[0] for CP and MGCG_OVERRIDE[1] for RLC */
5663 		temp = data = RREG32(mmRLC_CGTT_MGCG_OVERRIDE);
5664 		data |= (RLC_CGTT_MGCG_OVERRIDE__CPF_MASK |
5665 				RLC_CGTT_MGCG_OVERRIDE__RLC_MASK |
5666 				RLC_CGTT_MGCG_OVERRIDE__MGCG_MASK |
5667 				RLC_CGTT_MGCG_OVERRIDE__GRBM_MASK);
5668 		if (temp != data)
5669 			WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data);
5670 
5671 		/* 2 - disable MGLS in RLC */
5672 		data = RREG32(mmRLC_MEM_SLP_CNTL);
5673 		if (data & RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK) {
5674 			data &= ~RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK;
5675 			WREG32(mmRLC_MEM_SLP_CNTL, data);
5676 		}
5677 
5678 		/* 3 - disable MGLS in CP */
5679 		data = RREG32(mmCP_MEM_SLP_CNTL);
5680 		if (data & CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK) {
5681 			data &= ~CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK;
5682 			WREG32(mmCP_MEM_SLP_CNTL, data);
5683 		}
5684 
5685 		/* 4 - Disable CGTS(Tree Shade) MGCG and MGLS */
5686 		temp = data = RREG32(mmCGTS_SM_CTRL_REG);
5687 		data |= (CGTS_SM_CTRL_REG__OVERRIDE_MASK |
5688 				CGTS_SM_CTRL_REG__LS_OVERRIDE_MASK);
5689 		if (temp != data)
5690 			WREG32(mmCGTS_SM_CTRL_REG, data);
5691 
5692 		/* 5 - wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */
5693 		gfx_v8_0_wait_for_rlc_serdes(adev);
5694 
5695 		/* 6 - set mgcg override */
5696 		gfx_v8_0_send_serdes_cmd(adev, BPM_REG_MGCG_OVERRIDE, SET_BPM_SERDES_CMD);
5697 
5698 		udelay(50);
5699 
5700 		/* 7- wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */
5701 		gfx_v8_0_wait_for_rlc_serdes(adev);
5702 	}
5703 
5704 	amdgpu_gfx_rlc_exit_safe_mode(adev);
5705 }
5706 
5707 static void gfx_v8_0_update_coarse_grain_clock_gating(struct amdgpu_device *adev,
5708 						      bool enable)
5709 {
5710 	uint32_t temp, temp1, data, data1;
5711 
5712 	temp = data = RREG32(mmRLC_CGCG_CGLS_CTRL);
5713 
5714 	amdgpu_gfx_rlc_enter_safe_mode(adev);
5715 
5716 	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGCG)) {
5717 		temp1 = data1 =	RREG32(mmRLC_CGTT_MGCG_OVERRIDE);
5718 		data1 &= ~RLC_CGTT_MGCG_OVERRIDE__CGCG_MASK;
5719 		if (temp1 != data1)
5720 			WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data1);
5721 
5722 		/* : wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */
5723 		gfx_v8_0_wait_for_rlc_serdes(adev);
5724 
5725 		/* 2 - clear cgcg override */
5726 		gfx_v8_0_send_serdes_cmd(adev, BPM_REG_CGCG_OVERRIDE, CLE_BPM_SERDES_CMD);
5727 
5728 		/* wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */
5729 		gfx_v8_0_wait_for_rlc_serdes(adev);
5730 
5731 		/* 3 - write cmd to set CGLS */
5732 		gfx_v8_0_send_serdes_cmd(adev, BPM_REG_CGLS_EN, SET_BPM_SERDES_CMD);
5733 
5734 		/* 4 - enable cgcg */
5735 		data |= RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK;
5736 
5737 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGLS) {
5738 			/* enable cgls*/
5739 			data |= RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK;
5740 
5741 			temp1 = data1 =	RREG32(mmRLC_CGTT_MGCG_OVERRIDE);
5742 			data1 &= ~RLC_CGTT_MGCG_OVERRIDE__CGLS_MASK;
5743 
5744 			if (temp1 != data1)
5745 				WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data1);
5746 		} else {
5747 			data &= ~RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK;
5748 		}
5749 
5750 		if (temp != data)
5751 			WREG32(mmRLC_CGCG_CGLS_CTRL, data);
5752 
5753 		/* 5 enable cntx_empty_int_enable/cntx_busy_int_enable/
5754 		 * Cmp_busy/GFX_Idle interrupts
5755 		 */
5756 		gfx_v8_0_enable_gui_idle_interrupt(adev, true);
5757 	} else {
5758 		/* disable cntx_empty_int_enable & GFX Idle interrupt */
5759 		gfx_v8_0_enable_gui_idle_interrupt(adev, false);
5760 
5761 		/* TEST CGCG */
5762 		temp1 = data1 =	RREG32(mmRLC_CGTT_MGCG_OVERRIDE);
5763 		data1 |= (RLC_CGTT_MGCG_OVERRIDE__CGCG_MASK |
5764 				RLC_CGTT_MGCG_OVERRIDE__CGLS_MASK);
5765 		if (temp1 != data1)
5766 			WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data1);
5767 
5768 		/* read gfx register to wake up cgcg */
5769 		RREG32(mmCB_CGTT_SCLK_CTRL);
5770 		RREG32(mmCB_CGTT_SCLK_CTRL);
5771 		RREG32(mmCB_CGTT_SCLK_CTRL);
5772 		RREG32(mmCB_CGTT_SCLK_CTRL);
5773 
5774 		/* wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */
5775 		gfx_v8_0_wait_for_rlc_serdes(adev);
5776 
5777 		/* write cmd to Set CGCG Overrride */
5778 		gfx_v8_0_send_serdes_cmd(adev, BPM_REG_CGCG_OVERRIDE, SET_BPM_SERDES_CMD);
5779 
5780 		/* wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */
5781 		gfx_v8_0_wait_for_rlc_serdes(adev);
5782 
5783 		/* write cmd to Clear CGLS */
5784 		gfx_v8_0_send_serdes_cmd(adev, BPM_REG_CGLS_EN, CLE_BPM_SERDES_CMD);
5785 
5786 		/* disable cgcg, cgls should be disabled too. */
5787 		data &= ~(RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK |
5788 			  RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK);
5789 		if (temp != data)
5790 			WREG32(mmRLC_CGCG_CGLS_CTRL, data);
5791 		/* enable interrupts again for PG */
5792 		gfx_v8_0_enable_gui_idle_interrupt(adev, true);
5793 	}
5794 
5795 	gfx_v8_0_wait_for_rlc_serdes(adev);
5796 
5797 	amdgpu_gfx_rlc_exit_safe_mode(adev);
5798 }
5799 static int gfx_v8_0_update_gfx_clock_gating(struct amdgpu_device *adev,
5800 					    bool enable)
5801 {
5802 	if (enable) {
5803 		/* CGCG/CGLS should be enabled after MGCG/MGLS/TS(CG/LS)
5804 		 * ===  MGCG + MGLS + TS(CG/LS) ===
5805 		 */
5806 		gfx_v8_0_update_medium_grain_clock_gating(adev, enable);
5807 		gfx_v8_0_update_coarse_grain_clock_gating(adev, enable);
5808 	} else {
5809 		/* CGCG/CGLS should be disabled before MGCG/MGLS/TS(CG/LS)
5810 		 * ===  CGCG + CGLS ===
5811 		 */
5812 		gfx_v8_0_update_coarse_grain_clock_gating(adev, enable);
5813 		gfx_v8_0_update_medium_grain_clock_gating(adev, enable);
5814 	}
5815 	return 0;
5816 }
5817 
5818 static int gfx_v8_0_tonga_update_gfx_clock_gating(struct amdgpu_device *adev,
5819 					  enum amd_clockgating_state state)
5820 {
5821 	uint32_t msg_id, pp_state = 0;
5822 	uint32_t pp_support_state = 0;
5823 
5824 	if (adev->cg_flags & (AMD_CG_SUPPORT_GFX_CGCG | AMD_CG_SUPPORT_GFX_CGLS)) {
5825 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGLS) {
5826 			pp_support_state = PP_STATE_SUPPORT_LS;
5827 			pp_state = PP_STATE_LS;
5828 		}
5829 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGCG) {
5830 			pp_support_state |= PP_STATE_SUPPORT_CG;
5831 			pp_state |= PP_STATE_CG;
5832 		}
5833 		if (state == AMD_CG_STATE_UNGATE)
5834 			pp_state = 0;
5835 
5836 		msg_id = PP_CG_MSG_ID(PP_GROUP_GFX,
5837 				PP_BLOCK_GFX_CG,
5838 				pp_support_state,
5839 				pp_state);
5840 		if (adev->powerplay.pp_funcs->set_clockgating_by_smu)
5841 			amdgpu_dpm_set_clockgating_by_smu(adev, msg_id);
5842 	}
5843 
5844 	if (adev->cg_flags & (AMD_CG_SUPPORT_GFX_MGCG | AMD_CG_SUPPORT_GFX_MGLS)) {
5845 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS) {
5846 			pp_support_state = PP_STATE_SUPPORT_LS;
5847 			pp_state = PP_STATE_LS;
5848 		}
5849 
5850 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGCG) {
5851 			pp_support_state |= PP_STATE_SUPPORT_CG;
5852 			pp_state |= PP_STATE_CG;
5853 		}
5854 
5855 		if (state == AMD_CG_STATE_UNGATE)
5856 			pp_state = 0;
5857 
5858 		msg_id = PP_CG_MSG_ID(PP_GROUP_GFX,
5859 				PP_BLOCK_GFX_MG,
5860 				pp_support_state,
5861 				pp_state);
5862 		if (adev->powerplay.pp_funcs->set_clockgating_by_smu)
5863 			amdgpu_dpm_set_clockgating_by_smu(adev, msg_id);
5864 	}
5865 
5866 	return 0;
5867 }
5868 
5869 static int gfx_v8_0_polaris_update_gfx_clock_gating(struct amdgpu_device *adev,
5870 					  enum amd_clockgating_state state)
5871 {
5872 
5873 	uint32_t msg_id, pp_state = 0;
5874 	uint32_t pp_support_state = 0;
5875 
5876 	if (adev->cg_flags & (AMD_CG_SUPPORT_GFX_CGCG | AMD_CG_SUPPORT_GFX_CGLS)) {
5877 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGLS) {
5878 			pp_support_state = PP_STATE_SUPPORT_LS;
5879 			pp_state = PP_STATE_LS;
5880 		}
5881 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGCG) {
5882 			pp_support_state |= PP_STATE_SUPPORT_CG;
5883 			pp_state |= PP_STATE_CG;
5884 		}
5885 		if (state == AMD_CG_STATE_UNGATE)
5886 			pp_state = 0;
5887 
5888 		msg_id = PP_CG_MSG_ID(PP_GROUP_GFX,
5889 				PP_BLOCK_GFX_CG,
5890 				pp_support_state,
5891 				pp_state);
5892 		if (adev->powerplay.pp_funcs->set_clockgating_by_smu)
5893 			amdgpu_dpm_set_clockgating_by_smu(adev, msg_id);
5894 	}
5895 
5896 	if (adev->cg_flags & (AMD_CG_SUPPORT_GFX_3D_CGCG | AMD_CG_SUPPORT_GFX_3D_CGLS)) {
5897 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_3D_CGLS) {
5898 			pp_support_state = PP_STATE_SUPPORT_LS;
5899 			pp_state = PP_STATE_LS;
5900 		}
5901 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_3D_CGCG) {
5902 			pp_support_state |= PP_STATE_SUPPORT_CG;
5903 			pp_state |= PP_STATE_CG;
5904 		}
5905 		if (state == AMD_CG_STATE_UNGATE)
5906 			pp_state = 0;
5907 
5908 		msg_id = PP_CG_MSG_ID(PP_GROUP_GFX,
5909 				PP_BLOCK_GFX_3D,
5910 				pp_support_state,
5911 				pp_state);
5912 		if (adev->powerplay.pp_funcs->set_clockgating_by_smu)
5913 			amdgpu_dpm_set_clockgating_by_smu(adev, msg_id);
5914 	}
5915 
5916 	if (adev->cg_flags & (AMD_CG_SUPPORT_GFX_MGCG | AMD_CG_SUPPORT_GFX_MGLS)) {
5917 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS) {
5918 			pp_support_state = PP_STATE_SUPPORT_LS;
5919 			pp_state = PP_STATE_LS;
5920 		}
5921 
5922 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGCG) {
5923 			pp_support_state |= PP_STATE_SUPPORT_CG;
5924 			pp_state |= PP_STATE_CG;
5925 		}
5926 
5927 		if (state == AMD_CG_STATE_UNGATE)
5928 			pp_state = 0;
5929 
5930 		msg_id = PP_CG_MSG_ID(PP_GROUP_GFX,
5931 				PP_BLOCK_GFX_MG,
5932 				pp_support_state,
5933 				pp_state);
5934 		if (adev->powerplay.pp_funcs->set_clockgating_by_smu)
5935 			amdgpu_dpm_set_clockgating_by_smu(adev, msg_id);
5936 	}
5937 
5938 	if (adev->cg_flags & AMD_CG_SUPPORT_GFX_RLC_LS) {
5939 		pp_support_state = PP_STATE_SUPPORT_LS;
5940 
5941 		if (state == AMD_CG_STATE_UNGATE)
5942 			pp_state = 0;
5943 		else
5944 			pp_state = PP_STATE_LS;
5945 
5946 		msg_id = PP_CG_MSG_ID(PP_GROUP_GFX,
5947 				PP_BLOCK_GFX_RLC,
5948 				pp_support_state,
5949 				pp_state);
5950 		if (adev->powerplay.pp_funcs->set_clockgating_by_smu)
5951 			amdgpu_dpm_set_clockgating_by_smu(adev, msg_id);
5952 	}
5953 
5954 	if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CP_LS) {
5955 		pp_support_state = PP_STATE_SUPPORT_LS;
5956 
5957 		if (state == AMD_CG_STATE_UNGATE)
5958 			pp_state = 0;
5959 		else
5960 			pp_state = PP_STATE_LS;
5961 		msg_id = PP_CG_MSG_ID(PP_GROUP_GFX,
5962 			PP_BLOCK_GFX_CP,
5963 			pp_support_state,
5964 			pp_state);
5965 		if (adev->powerplay.pp_funcs->set_clockgating_by_smu)
5966 			amdgpu_dpm_set_clockgating_by_smu(adev, msg_id);
5967 	}
5968 
5969 	return 0;
5970 }
5971 
5972 static int gfx_v8_0_set_clockgating_state(void *handle,
5973 					  enum amd_clockgating_state state)
5974 {
5975 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5976 
5977 	if (amdgpu_sriov_vf(adev))
5978 		return 0;
5979 
5980 	switch (adev->asic_type) {
5981 	case CHIP_FIJI:
5982 	case CHIP_CARRIZO:
5983 	case CHIP_STONEY:
5984 		gfx_v8_0_update_gfx_clock_gating(adev,
5985 						 state == AMD_CG_STATE_GATE);
5986 		break;
5987 	case CHIP_TONGA:
5988 		gfx_v8_0_tonga_update_gfx_clock_gating(adev, state);
5989 		break;
5990 	case CHIP_POLARIS10:
5991 	case CHIP_POLARIS11:
5992 	case CHIP_POLARIS12:
5993 	case CHIP_VEGAM:
5994 		gfx_v8_0_polaris_update_gfx_clock_gating(adev, state);
5995 		break;
5996 	default:
5997 		break;
5998 	}
5999 	return 0;
6000 }
6001 
6002 static u64 gfx_v8_0_ring_get_rptr(struct amdgpu_ring *ring)
6003 {
6004 	return ring->adev->wb.wb[ring->rptr_offs];
6005 }
6006 
6007 static u64 gfx_v8_0_ring_get_wptr_gfx(struct amdgpu_ring *ring)
6008 {
6009 	struct amdgpu_device *adev = ring->adev;
6010 
6011 	if (ring->use_doorbell)
6012 		/* XXX check if swapping is necessary on BE */
6013 		return ring->adev->wb.wb[ring->wptr_offs];
6014 	else
6015 		return RREG32(mmCP_RB0_WPTR);
6016 }
6017 
6018 static void gfx_v8_0_ring_set_wptr_gfx(struct amdgpu_ring *ring)
6019 {
6020 	struct amdgpu_device *adev = ring->adev;
6021 
6022 	if (ring->use_doorbell) {
6023 		/* XXX check if swapping is necessary on BE */
6024 		adev->wb.wb[ring->wptr_offs] = lower_32_bits(ring->wptr);
6025 		WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
6026 	} else {
6027 		WREG32(mmCP_RB0_WPTR, lower_32_bits(ring->wptr));
6028 		(void)RREG32(mmCP_RB0_WPTR);
6029 	}
6030 }
6031 
6032 static void gfx_v8_0_ring_emit_hdp_flush(struct amdgpu_ring *ring)
6033 {
6034 	u32 ref_and_mask, reg_mem_engine;
6035 
6036 	if ((ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE) ||
6037 	    (ring->funcs->type == AMDGPU_RING_TYPE_KIQ)) {
6038 		switch (ring->me) {
6039 		case 1:
6040 			ref_and_mask = GPU_HDP_FLUSH_DONE__CP2_MASK << ring->pipe;
6041 			break;
6042 		case 2:
6043 			ref_and_mask = GPU_HDP_FLUSH_DONE__CP6_MASK << ring->pipe;
6044 			break;
6045 		default:
6046 			return;
6047 		}
6048 		reg_mem_engine = 0;
6049 	} else {
6050 		ref_and_mask = GPU_HDP_FLUSH_DONE__CP0_MASK;
6051 		reg_mem_engine = WAIT_REG_MEM_ENGINE(1); /* pfp */
6052 	}
6053 
6054 	amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
6055 	amdgpu_ring_write(ring, (WAIT_REG_MEM_OPERATION(1) | /* write, wait, write */
6056 				 WAIT_REG_MEM_FUNCTION(3) |  /* == */
6057 				 reg_mem_engine));
6058 	amdgpu_ring_write(ring, mmGPU_HDP_FLUSH_REQ);
6059 	amdgpu_ring_write(ring, mmGPU_HDP_FLUSH_DONE);
6060 	amdgpu_ring_write(ring, ref_and_mask);
6061 	amdgpu_ring_write(ring, ref_and_mask);
6062 	amdgpu_ring_write(ring, 0x20); /* poll interval */
6063 }
6064 
6065 static void gfx_v8_0_ring_emit_vgt_flush(struct amdgpu_ring *ring)
6066 {
6067 	amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE, 0));
6068 	amdgpu_ring_write(ring, EVENT_TYPE(VS_PARTIAL_FLUSH) |
6069 		EVENT_INDEX(4));
6070 
6071 	amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE, 0));
6072 	amdgpu_ring_write(ring, EVENT_TYPE(VGT_FLUSH) |
6073 		EVENT_INDEX(0));
6074 }
6075 
6076 static void gfx_v8_0_ring_emit_ib_gfx(struct amdgpu_ring *ring,
6077 					struct amdgpu_job *job,
6078 					struct amdgpu_ib *ib,
6079 					uint32_t flags)
6080 {
6081 	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
6082 	u32 header, control = 0;
6083 
6084 	if (ib->flags & AMDGPU_IB_FLAG_CE)
6085 		header = PACKET3(PACKET3_INDIRECT_BUFFER_CONST, 2);
6086 	else
6087 		header = PACKET3(PACKET3_INDIRECT_BUFFER, 2);
6088 
6089 	control |= ib->length_dw | (vmid << 24);
6090 
6091 	if (amdgpu_sriov_vf(ring->adev) && (ib->flags & AMDGPU_IB_FLAG_PREEMPT)) {
6092 		control |= INDIRECT_BUFFER_PRE_ENB(1);
6093 
6094 		if (!(ib->flags & AMDGPU_IB_FLAG_CE))
6095 			gfx_v8_0_ring_emit_de_meta(ring);
6096 	}
6097 
6098 	amdgpu_ring_write(ring, header);
6099 	amdgpu_ring_write(ring,
6100 #ifdef __BIG_ENDIAN
6101 			  (2 << 0) |
6102 #endif
6103 			  (ib->gpu_addr & 0xFFFFFFFC));
6104 	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr) & 0xFFFF);
6105 	amdgpu_ring_write(ring, control);
6106 }
6107 
6108 static void gfx_v8_0_ring_emit_ib_compute(struct amdgpu_ring *ring,
6109 					  struct amdgpu_job *job,
6110 					  struct amdgpu_ib *ib,
6111 					  uint32_t flags)
6112 {
6113 	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
6114 	u32 control = INDIRECT_BUFFER_VALID | ib->length_dw | (vmid << 24);
6115 
6116 	/* Currently, there is a high possibility to get wave ID mismatch
6117 	 * between ME and GDS, leading to a hw deadlock, because ME generates
6118 	 * different wave IDs than the GDS expects. This situation happens
6119 	 * randomly when at least 5 compute pipes use GDS ordered append.
6120 	 * The wave IDs generated by ME are also wrong after suspend/resume.
6121 	 * Those are probably bugs somewhere else in the kernel driver.
6122 	 *
6123 	 * Writing GDS_COMPUTE_MAX_WAVE_ID resets wave ID counters in ME and
6124 	 * GDS to 0 for this ring (me/pipe).
6125 	 */
6126 	if (ib->flags & AMDGPU_IB_FLAG_RESET_GDS_MAX_WAVE_ID) {
6127 		amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1));
6128 		amdgpu_ring_write(ring, mmGDS_COMPUTE_MAX_WAVE_ID - PACKET3_SET_CONFIG_REG_START);
6129 		amdgpu_ring_write(ring, ring->adev->gds.gds_compute_max_wave_id);
6130 	}
6131 
6132 	amdgpu_ring_write(ring, PACKET3(PACKET3_INDIRECT_BUFFER, 2));
6133 	amdgpu_ring_write(ring,
6134 #ifdef __BIG_ENDIAN
6135 				(2 << 0) |
6136 #endif
6137 				(ib->gpu_addr & 0xFFFFFFFC));
6138 	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr) & 0xFFFF);
6139 	amdgpu_ring_write(ring, control);
6140 }
6141 
6142 static void gfx_v8_0_ring_emit_fence_gfx(struct amdgpu_ring *ring, u64 addr,
6143 					 u64 seq, unsigned flags)
6144 {
6145 	bool write64bit = flags & AMDGPU_FENCE_FLAG_64BIT;
6146 	bool int_sel = flags & AMDGPU_FENCE_FLAG_INT;
6147 
6148 	/* Workaround for cache flush problems. First send a dummy EOP
6149 	 * event down the pipe with seq one below.
6150 	 */
6151 	amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE_EOP, 4));
6152 	amdgpu_ring_write(ring, (EOP_TCL1_ACTION_EN |
6153 				 EOP_TC_ACTION_EN |
6154 				 EOP_TC_WB_ACTION_EN |
6155 				 EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
6156 				 EVENT_INDEX(5)));
6157 	amdgpu_ring_write(ring, addr & 0xfffffffc);
6158 	amdgpu_ring_write(ring, (upper_32_bits(addr) & 0xffff) |
6159 				DATA_SEL(1) | INT_SEL(0));
6160 	amdgpu_ring_write(ring, lower_32_bits(seq - 1));
6161 	amdgpu_ring_write(ring, upper_32_bits(seq - 1));
6162 
6163 	/* Then send the real EOP event down the pipe:
6164 	 * EVENT_WRITE_EOP - flush caches, send int */
6165 	amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE_EOP, 4));
6166 	amdgpu_ring_write(ring, (EOP_TCL1_ACTION_EN |
6167 				 EOP_TC_ACTION_EN |
6168 				 EOP_TC_WB_ACTION_EN |
6169 				 EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
6170 				 EVENT_INDEX(5)));
6171 	amdgpu_ring_write(ring, addr & 0xfffffffc);
6172 	amdgpu_ring_write(ring, (upper_32_bits(addr) & 0xffff) |
6173 			  DATA_SEL(write64bit ? 2 : 1) | INT_SEL(int_sel ? 2 : 0));
6174 	amdgpu_ring_write(ring, lower_32_bits(seq));
6175 	amdgpu_ring_write(ring, upper_32_bits(seq));
6176 
6177 }
6178 
6179 static void gfx_v8_0_ring_emit_pipeline_sync(struct amdgpu_ring *ring)
6180 {
6181 	int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX);
6182 	uint32_t seq = ring->fence_drv.sync_seq;
6183 	uint64_t addr = ring->fence_drv.gpu_addr;
6184 
6185 	amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
6186 	amdgpu_ring_write(ring, (WAIT_REG_MEM_MEM_SPACE(1) | /* memory */
6187 				 WAIT_REG_MEM_FUNCTION(3) | /* equal */
6188 				 WAIT_REG_MEM_ENGINE(usepfp))); /* pfp or me */
6189 	amdgpu_ring_write(ring, addr & 0xfffffffc);
6190 	amdgpu_ring_write(ring, upper_32_bits(addr) & 0xffffffff);
6191 	amdgpu_ring_write(ring, seq);
6192 	amdgpu_ring_write(ring, 0xffffffff);
6193 	amdgpu_ring_write(ring, 4); /* poll interval */
6194 }
6195 
6196 static void gfx_v8_0_ring_emit_vm_flush(struct amdgpu_ring *ring,
6197 					unsigned vmid, uint64_t pd_addr)
6198 {
6199 	int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX);
6200 
6201 	amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr);
6202 
6203 	/* wait for the invalidate to complete */
6204 	amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
6205 	amdgpu_ring_write(ring, (WAIT_REG_MEM_OPERATION(0) | /* wait */
6206 				 WAIT_REG_MEM_FUNCTION(0) |  /* always */
6207 				 WAIT_REG_MEM_ENGINE(0))); /* me */
6208 	amdgpu_ring_write(ring, mmVM_INVALIDATE_REQUEST);
6209 	amdgpu_ring_write(ring, 0);
6210 	amdgpu_ring_write(ring, 0); /* ref */
6211 	amdgpu_ring_write(ring, 0); /* mask */
6212 	amdgpu_ring_write(ring, 0x20); /* poll interval */
6213 
6214 	/* compute doesn't have PFP */
6215 	if (usepfp) {
6216 		/* sync PFP to ME, otherwise we might get invalid PFP reads */
6217 		amdgpu_ring_write(ring, PACKET3(PACKET3_PFP_SYNC_ME, 0));
6218 		amdgpu_ring_write(ring, 0x0);
6219 	}
6220 }
6221 
6222 static u64 gfx_v8_0_ring_get_wptr_compute(struct amdgpu_ring *ring)
6223 {
6224 	return ring->adev->wb.wb[ring->wptr_offs];
6225 }
6226 
6227 static void gfx_v8_0_ring_set_wptr_compute(struct amdgpu_ring *ring)
6228 {
6229 	struct amdgpu_device *adev = ring->adev;
6230 
6231 	/* XXX check if swapping is necessary on BE */
6232 	adev->wb.wb[ring->wptr_offs] = lower_32_bits(ring->wptr);
6233 	WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
6234 }
6235 
6236 static void gfx_v8_0_ring_set_pipe_percent(struct amdgpu_ring *ring,
6237 					   bool acquire)
6238 {
6239 	struct amdgpu_device *adev = ring->adev;
6240 	int pipe_num, tmp, reg;
6241 	int pipe_percent = acquire ? SPI_WCL_PIPE_PERCENT_GFX__VALUE_MASK : 0x1;
6242 
6243 	pipe_num = ring->me * adev->gfx.mec.num_pipe_per_mec + ring->pipe;
6244 
6245 	/* first me only has 2 entries, GFX and HP3D */
6246 	if (ring->me > 0)
6247 		pipe_num -= 2;
6248 
6249 	reg = mmSPI_WCL_PIPE_PERCENT_GFX + pipe_num;
6250 	tmp = RREG32(reg);
6251 	tmp = REG_SET_FIELD(tmp, SPI_WCL_PIPE_PERCENT_GFX, VALUE, pipe_percent);
6252 	WREG32(reg, tmp);
6253 }
6254 
6255 static void gfx_v8_0_pipe_reserve_resources(struct amdgpu_device *adev,
6256 					    struct amdgpu_ring *ring,
6257 					    bool acquire)
6258 {
6259 	int i, pipe;
6260 	bool reserve;
6261 	struct amdgpu_ring *iring;
6262 
6263 	mutex_lock(&adev->gfx.pipe_reserve_mutex);
6264 	pipe = amdgpu_gfx_mec_queue_to_bit(adev, ring->me, ring->pipe, 0);
6265 	if (acquire)
6266 		set_bit(pipe, adev->gfx.pipe_reserve_bitmap);
6267 	else
6268 		clear_bit(pipe, adev->gfx.pipe_reserve_bitmap);
6269 
6270 	if (!bitmap_weight(adev->gfx.pipe_reserve_bitmap, AMDGPU_MAX_COMPUTE_QUEUES)) {
6271 		/* Clear all reservations - everyone reacquires all resources */
6272 		for (i = 0; i < adev->gfx.num_gfx_rings; ++i)
6273 			gfx_v8_0_ring_set_pipe_percent(&adev->gfx.gfx_ring[i],
6274 						       true);
6275 
6276 		for (i = 0; i < adev->gfx.num_compute_rings; ++i)
6277 			gfx_v8_0_ring_set_pipe_percent(&adev->gfx.compute_ring[i],
6278 						       true);
6279 	} else {
6280 		/* Lower all pipes without a current reservation */
6281 		for (i = 0; i < adev->gfx.num_gfx_rings; ++i) {
6282 			iring = &adev->gfx.gfx_ring[i];
6283 			pipe = amdgpu_gfx_mec_queue_to_bit(adev,
6284 							   iring->me,
6285 							   iring->pipe,
6286 							   0);
6287 			reserve = test_bit(pipe, adev->gfx.pipe_reserve_bitmap);
6288 			gfx_v8_0_ring_set_pipe_percent(iring, reserve);
6289 		}
6290 
6291 		for (i = 0; i < adev->gfx.num_compute_rings; ++i) {
6292 			iring = &adev->gfx.compute_ring[i];
6293 			pipe = amdgpu_gfx_mec_queue_to_bit(adev,
6294 							   iring->me,
6295 							   iring->pipe,
6296 							   0);
6297 			reserve = test_bit(pipe, adev->gfx.pipe_reserve_bitmap);
6298 			gfx_v8_0_ring_set_pipe_percent(iring, reserve);
6299 		}
6300 	}
6301 
6302 	mutex_unlock(&adev->gfx.pipe_reserve_mutex);
6303 }
6304 
6305 static void gfx_v8_0_hqd_set_priority(struct amdgpu_device *adev,
6306 				      struct amdgpu_ring *ring,
6307 				      bool acquire)
6308 {
6309 	uint32_t pipe_priority = acquire ? 0x2 : 0x0;
6310 	uint32_t queue_priority = acquire ? 0xf : 0x0;
6311 
6312 	mutex_lock(&adev->srbm_mutex);
6313 	vi_srbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
6314 
6315 	WREG32(mmCP_HQD_PIPE_PRIORITY, pipe_priority);
6316 	WREG32(mmCP_HQD_QUEUE_PRIORITY, queue_priority);
6317 
6318 	vi_srbm_select(adev, 0, 0, 0, 0);
6319 	mutex_unlock(&adev->srbm_mutex);
6320 }
6321 static void gfx_v8_0_ring_set_priority_compute(struct amdgpu_ring *ring,
6322 					       enum drm_sched_priority priority)
6323 {
6324 	struct amdgpu_device *adev = ring->adev;
6325 	bool acquire = priority == DRM_SCHED_PRIORITY_HIGH_HW;
6326 
6327 	if (ring->funcs->type != AMDGPU_RING_TYPE_COMPUTE)
6328 		return;
6329 
6330 	gfx_v8_0_hqd_set_priority(adev, ring, acquire);
6331 	gfx_v8_0_pipe_reserve_resources(adev, ring, acquire);
6332 }
6333 
6334 static void gfx_v8_0_ring_emit_fence_compute(struct amdgpu_ring *ring,
6335 					     u64 addr, u64 seq,
6336 					     unsigned flags)
6337 {
6338 	bool write64bit = flags & AMDGPU_FENCE_FLAG_64BIT;
6339 	bool int_sel = flags & AMDGPU_FENCE_FLAG_INT;
6340 
6341 	/* RELEASE_MEM - flush caches, send int */
6342 	amdgpu_ring_write(ring, PACKET3(PACKET3_RELEASE_MEM, 5));
6343 	amdgpu_ring_write(ring, (EOP_TCL1_ACTION_EN |
6344 				 EOP_TC_ACTION_EN |
6345 				 EOP_TC_WB_ACTION_EN |
6346 				 EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
6347 				 EVENT_INDEX(5)));
6348 	amdgpu_ring_write(ring, DATA_SEL(write64bit ? 2 : 1) | INT_SEL(int_sel ? 2 : 0));
6349 	amdgpu_ring_write(ring, addr & 0xfffffffc);
6350 	amdgpu_ring_write(ring, upper_32_bits(addr));
6351 	amdgpu_ring_write(ring, lower_32_bits(seq));
6352 	amdgpu_ring_write(ring, upper_32_bits(seq));
6353 }
6354 
6355 static void gfx_v8_0_ring_emit_fence_kiq(struct amdgpu_ring *ring, u64 addr,
6356 					 u64 seq, unsigned int flags)
6357 {
6358 	/* we only allocate 32bit for each seq wb address */
6359 	BUG_ON(flags & AMDGPU_FENCE_FLAG_64BIT);
6360 
6361 	/* write fence seq to the "addr" */
6362 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
6363 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
6364 				 WRITE_DATA_DST_SEL(5) | WR_CONFIRM));
6365 	amdgpu_ring_write(ring, lower_32_bits(addr));
6366 	amdgpu_ring_write(ring, upper_32_bits(addr));
6367 	amdgpu_ring_write(ring, lower_32_bits(seq));
6368 
6369 	if (flags & AMDGPU_FENCE_FLAG_INT) {
6370 		/* set register to trigger INT */
6371 		amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
6372 		amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
6373 					 WRITE_DATA_DST_SEL(0) | WR_CONFIRM));
6374 		amdgpu_ring_write(ring, mmCPC_INT_STATUS);
6375 		amdgpu_ring_write(ring, 0);
6376 		amdgpu_ring_write(ring, 0x20000000); /* src_id is 178 */
6377 	}
6378 }
6379 
6380 static void gfx_v8_ring_emit_sb(struct amdgpu_ring *ring)
6381 {
6382 	amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
6383 	amdgpu_ring_write(ring, 0);
6384 }
6385 
6386 static void gfx_v8_ring_emit_cntxcntl(struct amdgpu_ring *ring, uint32_t flags)
6387 {
6388 	uint32_t dw2 = 0;
6389 
6390 	if (amdgpu_sriov_vf(ring->adev))
6391 		gfx_v8_0_ring_emit_ce_meta(ring);
6392 
6393 	dw2 |= 0x80000000; /* set load_enable otherwise this package is just NOPs */
6394 	if (flags & AMDGPU_HAVE_CTX_SWITCH) {
6395 		gfx_v8_0_ring_emit_vgt_flush(ring);
6396 		/* set load_global_config & load_global_uconfig */
6397 		dw2 |= 0x8001;
6398 		/* set load_cs_sh_regs */
6399 		dw2 |= 0x01000000;
6400 		/* set load_per_context_state & load_gfx_sh_regs for GFX */
6401 		dw2 |= 0x10002;
6402 
6403 		/* set load_ce_ram if preamble presented */
6404 		if (AMDGPU_PREAMBLE_IB_PRESENT & flags)
6405 			dw2 |= 0x10000000;
6406 	} else {
6407 		/* still load_ce_ram if this is the first time preamble presented
6408 		 * although there is no context switch happens.
6409 		 */
6410 		if (AMDGPU_PREAMBLE_IB_PRESENT_FIRST & flags)
6411 			dw2 |= 0x10000000;
6412 	}
6413 
6414 	amdgpu_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1));
6415 	amdgpu_ring_write(ring, dw2);
6416 	amdgpu_ring_write(ring, 0);
6417 }
6418 
6419 static unsigned gfx_v8_0_ring_emit_init_cond_exec(struct amdgpu_ring *ring)
6420 {
6421 	unsigned ret;
6422 
6423 	amdgpu_ring_write(ring, PACKET3(PACKET3_COND_EXEC, 3));
6424 	amdgpu_ring_write(ring, lower_32_bits(ring->cond_exe_gpu_addr));
6425 	amdgpu_ring_write(ring, upper_32_bits(ring->cond_exe_gpu_addr));
6426 	amdgpu_ring_write(ring, 0); /* discard following DWs if *cond_exec_gpu_addr==0 */
6427 	ret = ring->wptr & ring->buf_mask;
6428 	amdgpu_ring_write(ring, 0x55aa55aa); /* patch dummy value later */
6429 	return ret;
6430 }
6431 
6432 static void gfx_v8_0_ring_emit_patch_cond_exec(struct amdgpu_ring *ring, unsigned offset)
6433 {
6434 	unsigned cur;
6435 
6436 	BUG_ON(offset > ring->buf_mask);
6437 	BUG_ON(ring->ring[offset] != 0x55aa55aa);
6438 
6439 	cur = (ring->wptr & ring->buf_mask) - 1;
6440 	if (likely(cur > offset))
6441 		ring->ring[offset] = cur - offset;
6442 	else
6443 		ring->ring[offset] = (ring->ring_size >> 2) - offset + cur;
6444 }
6445 
6446 static void gfx_v8_0_ring_emit_rreg(struct amdgpu_ring *ring, uint32_t reg)
6447 {
6448 	struct amdgpu_device *adev = ring->adev;
6449 
6450 	amdgpu_ring_write(ring, PACKET3(PACKET3_COPY_DATA, 4));
6451 	amdgpu_ring_write(ring, 0 |	/* src: register*/
6452 				(5 << 8) |	/* dst: memory */
6453 				(1 << 20));	/* write confirm */
6454 	amdgpu_ring_write(ring, reg);
6455 	amdgpu_ring_write(ring, 0);
6456 	amdgpu_ring_write(ring, lower_32_bits(adev->wb.gpu_addr +
6457 				adev->virt.reg_val_offs * 4));
6458 	amdgpu_ring_write(ring, upper_32_bits(adev->wb.gpu_addr +
6459 				adev->virt.reg_val_offs * 4));
6460 }
6461 
6462 static void gfx_v8_0_ring_emit_wreg(struct amdgpu_ring *ring, uint32_t reg,
6463 				  uint32_t val)
6464 {
6465 	uint32_t cmd;
6466 
6467 	switch (ring->funcs->type) {
6468 	case AMDGPU_RING_TYPE_GFX:
6469 		cmd = WRITE_DATA_ENGINE_SEL(1) | WR_CONFIRM;
6470 		break;
6471 	case AMDGPU_RING_TYPE_KIQ:
6472 		cmd = 1 << 16; /* no inc addr */
6473 		break;
6474 	default:
6475 		cmd = WR_CONFIRM;
6476 		break;
6477 	}
6478 
6479 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
6480 	amdgpu_ring_write(ring, cmd);
6481 	amdgpu_ring_write(ring, reg);
6482 	amdgpu_ring_write(ring, 0);
6483 	amdgpu_ring_write(ring, val);
6484 }
6485 
6486 static void gfx_v8_0_ring_soft_recovery(struct amdgpu_ring *ring, unsigned vmid)
6487 {
6488 	struct amdgpu_device *adev = ring->adev;
6489 	uint32_t value = 0;
6490 
6491 	value = REG_SET_FIELD(value, SQ_CMD, CMD, 0x03);
6492 	value = REG_SET_FIELD(value, SQ_CMD, MODE, 0x01);
6493 	value = REG_SET_FIELD(value, SQ_CMD, CHECK_VMID, 1);
6494 	value = REG_SET_FIELD(value, SQ_CMD, VM_ID, vmid);
6495 	WREG32(mmSQ_CMD, value);
6496 }
6497 
6498 static void gfx_v8_0_set_gfx_eop_interrupt_state(struct amdgpu_device *adev,
6499 						 enum amdgpu_interrupt_state state)
6500 {
6501 	WREG32_FIELD(CP_INT_CNTL_RING0, TIME_STAMP_INT_ENABLE,
6502 		     state == AMDGPU_IRQ_STATE_DISABLE ? 0 : 1);
6503 }
6504 
6505 static void gfx_v8_0_set_compute_eop_interrupt_state(struct amdgpu_device *adev,
6506 						     int me, int pipe,
6507 						     enum amdgpu_interrupt_state state)
6508 {
6509 	u32 mec_int_cntl, mec_int_cntl_reg;
6510 
6511 	/*
6512 	 * amdgpu controls only the first MEC. That's why this function only
6513 	 * handles the setting of interrupts for this specific MEC. All other
6514 	 * pipes' interrupts are set by amdkfd.
6515 	 */
6516 
6517 	if (me == 1) {
6518 		switch (pipe) {
6519 		case 0:
6520 			mec_int_cntl_reg = mmCP_ME1_PIPE0_INT_CNTL;
6521 			break;
6522 		case 1:
6523 			mec_int_cntl_reg = mmCP_ME1_PIPE1_INT_CNTL;
6524 			break;
6525 		case 2:
6526 			mec_int_cntl_reg = mmCP_ME1_PIPE2_INT_CNTL;
6527 			break;
6528 		case 3:
6529 			mec_int_cntl_reg = mmCP_ME1_PIPE3_INT_CNTL;
6530 			break;
6531 		default:
6532 			DRM_DEBUG("invalid pipe %d\n", pipe);
6533 			return;
6534 		}
6535 	} else {
6536 		DRM_DEBUG("invalid me %d\n", me);
6537 		return;
6538 	}
6539 
6540 	switch (state) {
6541 	case AMDGPU_IRQ_STATE_DISABLE:
6542 		mec_int_cntl = RREG32(mec_int_cntl_reg);
6543 		mec_int_cntl &= ~CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK;
6544 		WREG32(mec_int_cntl_reg, mec_int_cntl);
6545 		break;
6546 	case AMDGPU_IRQ_STATE_ENABLE:
6547 		mec_int_cntl = RREG32(mec_int_cntl_reg);
6548 		mec_int_cntl |= CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK;
6549 		WREG32(mec_int_cntl_reg, mec_int_cntl);
6550 		break;
6551 	default:
6552 		break;
6553 	}
6554 }
6555 
6556 static int gfx_v8_0_set_priv_reg_fault_state(struct amdgpu_device *adev,
6557 					     struct amdgpu_irq_src *source,
6558 					     unsigned type,
6559 					     enum amdgpu_interrupt_state state)
6560 {
6561 	WREG32_FIELD(CP_INT_CNTL_RING0, PRIV_REG_INT_ENABLE,
6562 		     state == AMDGPU_IRQ_STATE_DISABLE ? 0 : 1);
6563 
6564 	return 0;
6565 }
6566 
6567 static int gfx_v8_0_set_priv_inst_fault_state(struct amdgpu_device *adev,
6568 					      struct amdgpu_irq_src *source,
6569 					      unsigned type,
6570 					      enum amdgpu_interrupt_state state)
6571 {
6572 	WREG32_FIELD(CP_INT_CNTL_RING0, PRIV_INSTR_INT_ENABLE,
6573 		     state == AMDGPU_IRQ_STATE_DISABLE ? 0 : 1);
6574 
6575 	return 0;
6576 }
6577 
6578 static int gfx_v8_0_set_eop_interrupt_state(struct amdgpu_device *adev,
6579 					    struct amdgpu_irq_src *src,
6580 					    unsigned type,
6581 					    enum amdgpu_interrupt_state state)
6582 {
6583 	switch (type) {
6584 	case AMDGPU_CP_IRQ_GFX_ME0_PIPE0_EOP:
6585 		gfx_v8_0_set_gfx_eop_interrupt_state(adev, state);
6586 		break;
6587 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP:
6588 		gfx_v8_0_set_compute_eop_interrupt_state(adev, 1, 0, state);
6589 		break;
6590 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE1_EOP:
6591 		gfx_v8_0_set_compute_eop_interrupt_state(adev, 1, 1, state);
6592 		break;
6593 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE2_EOP:
6594 		gfx_v8_0_set_compute_eop_interrupt_state(adev, 1, 2, state);
6595 		break;
6596 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE3_EOP:
6597 		gfx_v8_0_set_compute_eop_interrupt_state(adev, 1, 3, state);
6598 		break;
6599 	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE0_EOP:
6600 		gfx_v8_0_set_compute_eop_interrupt_state(adev, 2, 0, state);
6601 		break;
6602 	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE1_EOP:
6603 		gfx_v8_0_set_compute_eop_interrupt_state(adev, 2, 1, state);
6604 		break;
6605 	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE2_EOP:
6606 		gfx_v8_0_set_compute_eop_interrupt_state(adev, 2, 2, state);
6607 		break;
6608 	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE3_EOP:
6609 		gfx_v8_0_set_compute_eop_interrupt_state(adev, 2, 3, state);
6610 		break;
6611 	default:
6612 		break;
6613 	}
6614 	return 0;
6615 }
6616 
6617 static int gfx_v8_0_set_cp_ecc_int_state(struct amdgpu_device *adev,
6618 					 struct amdgpu_irq_src *source,
6619 					 unsigned int type,
6620 					 enum amdgpu_interrupt_state state)
6621 {
6622 	int enable_flag;
6623 
6624 	switch (state) {
6625 	case AMDGPU_IRQ_STATE_DISABLE:
6626 		enable_flag = 0;
6627 		break;
6628 
6629 	case AMDGPU_IRQ_STATE_ENABLE:
6630 		enable_flag = 1;
6631 		break;
6632 
6633 	default:
6634 		return -EINVAL;
6635 	}
6636 
6637 	WREG32_FIELD(CP_INT_CNTL, CP_ECC_ERROR_INT_ENABLE, enable_flag);
6638 	WREG32_FIELD(CP_INT_CNTL_RING0, CP_ECC_ERROR_INT_ENABLE, enable_flag);
6639 	WREG32_FIELD(CP_INT_CNTL_RING1, CP_ECC_ERROR_INT_ENABLE, enable_flag);
6640 	WREG32_FIELD(CP_INT_CNTL_RING2, CP_ECC_ERROR_INT_ENABLE, enable_flag);
6641 	WREG32_FIELD(CPC_INT_CNTL, CP_ECC_ERROR_INT_ENABLE, enable_flag);
6642 	WREG32_FIELD(CP_ME1_PIPE0_INT_CNTL, CP_ECC_ERROR_INT_ENABLE,
6643 		     enable_flag);
6644 	WREG32_FIELD(CP_ME1_PIPE1_INT_CNTL, CP_ECC_ERROR_INT_ENABLE,
6645 		     enable_flag);
6646 	WREG32_FIELD(CP_ME1_PIPE2_INT_CNTL, CP_ECC_ERROR_INT_ENABLE,
6647 		     enable_flag);
6648 	WREG32_FIELD(CP_ME1_PIPE3_INT_CNTL, CP_ECC_ERROR_INT_ENABLE,
6649 		     enable_flag);
6650 	WREG32_FIELD(CP_ME2_PIPE0_INT_CNTL, CP_ECC_ERROR_INT_ENABLE,
6651 		     enable_flag);
6652 	WREG32_FIELD(CP_ME2_PIPE1_INT_CNTL, CP_ECC_ERROR_INT_ENABLE,
6653 		     enable_flag);
6654 	WREG32_FIELD(CP_ME2_PIPE2_INT_CNTL, CP_ECC_ERROR_INT_ENABLE,
6655 		     enable_flag);
6656 	WREG32_FIELD(CP_ME2_PIPE3_INT_CNTL, CP_ECC_ERROR_INT_ENABLE,
6657 		     enable_flag);
6658 
6659 	return 0;
6660 }
6661 
6662 static int gfx_v8_0_set_sq_int_state(struct amdgpu_device *adev,
6663 				     struct amdgpu_irq_src *source,
6664 				     unsigned int type,
6665 				     enum amdgpu_interrupt_state state)
6666 {
6667 	int enable_flag;
6668 
6669 	switch (state) {
6670 	case AMDGPU_IRQ_STATE_DISABLE:
6671 		enable_flag = 1;
6672 		break;
6673 
6674 	case AMDGPU_IRQ_STATE_ENABLE:
6675 		enable_flag = 0;
6676 		break;
6677 
6678 	default:
6679 		return -EINVAL;
6680 	}
6681 
6682 	WREG32_FIELD(SQ_INTERRUPT_MSG_CTRL, STALL,
6683 		     enable_flag);
6684 
6685 	return 0;
6686 }
6687 
6688 static int gfx_v8_0_eop_irq(struct amdgpu_device *adev,
6689 			    struct amdgpu_irq_src *source,
6690 			    struct amdgpu_iv_entry *entry)
6691 {
6692 	int i;
6693 	u8 me_id, pipe_id, queue_id;
6694 	struct amdgpu_ring *ring;
6695 
6696 	DRM_DEBUG("IH: CP EOP\n");
6697 	me_id = (entry->ring_id & 0x0c) >> 2;
6698 	pipe_id = (entry->ring_id & 0x03) >> 0;
6699 	queue_id = (entry->ring_id & 0x70) >> 4;
6700 
6701 	switch (me_id) {
6702 	case 0:
6703 		amdgpu_fence_process(&adev->gfx.gfx_ring[0]);
6704 		break;
6705 	case 1:
6706 	case 2:
6707 		for (i = 0; i < adev->gfx.num_compute_rings; i++) {
6708 			ring = &adev->gfx.compute_ring[i];
6709 			/* Per-queue interrupt is supported for MEC starting from VI.
6710 			  * The interrupt can only be enabled/disabled per pipe instead of per queue.
6711 			  */
6712 			if ((ring->me == me_id) && (ring->pipe == pipe_id) && (ring->queue == queue_id))
6713 				amdgpu_fence_process(ring);
6714 		}
6715 		break;
6716 	}
6717 	return 0;
6718 }
6719 
6720 static void gfx_v8_0_fault(struct amdgpu_device *adev,
6721 			   struct amdgpu_iv_entry *entry)
6722 {
6723 	u8 me_id, pipe_id, queue_id;
6724 	struct amdgpu_ring *ring;
6725 	int i;
6726 
6727 	me_id = (entry->ring_id & 0x0c) >> 2;
6728 	pipe_id = (entry->ring_id & 0x03) >> 0;
6729 	queue_id = (entry->ring_id & 0x70) >> 4;
6730 
6731 	switch (me_id) {
6732 	case 0:
6733 		drm_sched_fault(&adev->gfx.gfx_ring[0].sched);
6734 		break;
6735 	case 1:
6736 	case 2:
6737 		for (i = 0; i < adev->gfx.num_compute_rings; i++) {
6738 			ring = &adev->gfx.compute_ring[i];
6739 			if (ring->me == me_id && ring->pipe == pipe_id &&
6740 			    ring->queue == queue_id)
6741 				drm_sched_fault(&ring->sched);
6742 		}
6743 		break;
6744 	}
6745 }
6746 
6747 static int gfx_v8_0_priv_reg_irq(struct amdgpu_device *adev,
6748 				 struct amdgpu_irq_src *source,
6749 				 struct amdgpu_iv_entry *entry)
6750 {
6751 	DRM_ERROR("Illegal register access in command stream\n");
6752 	gfx_v8_0_fault(adev, entry);
6753 	return 0;
6754 }
6755 
6756 static int gfx_v8_0_priv_inst_irq(struct amdgpu_device *adev,
6757 				  struct amdgpu_irq_src *source,
6758 				  struct amdgpu_iv_entry *entry)
6759 {
6760 	DRM_ERROR("Illegal instruction in command stream\n");
6761 	gfx_v8_0_fault(adev, entry);
6762 	return 0;
6763 }
6764 
6765 static int gfx_v8_0_cp_ecc_error_irq(struct amdgpu_device *adev,
6766 				     struct amdgpu_irq_src *source,
6767 				     struct amdgpu_iv_entry *entry)
6768 {
6769 	DRM_ERROR("CP EDC/ECC error detected.");
6770 	return 0;
6771 }
6772 
6773 static void gfx_v8_0_parse_sq_irq(struct amdgpu_device *adev, unsigned ih_data)
6774 {
6775 	u32 enc, se_id, sh_id, cu_id;
6776 	char type[20];
6777 	int sq_edc_source = -1;
6778 
6779 	enc = REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_CMN, ENCODING);
6780 	se_id = REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_CMN, SE_ID);
6781 
6782 	switch (enc) {
6783 		case 0:
6784 			DRM_INFO("SQ general purpose intr detected:"
6785 					"se_id %d, immed_overflow %d, host_reg_overflow %d,"
6786 					"host_cmd_overflow %d, cmd_timestamp %d,"
6787 					"reg_timestamp %d, thread_trace_buff_full %d,"
6788 					"wlt %d, thread_trace %d.\n",
6789 					se_id,
6790 					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_AUTO, IMMED_OVERFLOW),
6791 					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_AUTO, HOST_REG_OVERFLOW),
6792 					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_AUTO, HOST_CMD_OVERFLOW),
6793 					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_AUTO, CMD_TIMESTAMP),
6794 					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_AUTO, REG_TIMESTAMP),
6795 					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_AUTO, THREAD_TRACE_BUF_FULL),
6796 					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_AUTO, WLT),
6797 					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_AUTO, THREAD_TRACE)
6798 					);
6799 			break;
6800 		case 1:
6801 		case 2:
6802 
6803 			cu_id = REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_WAVE, CU_ID);
6804 			sh_id = REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_WAVE, SH_ID);
6805 
6806 			/*
6807 			 * This function can be called either directly from ISR
6808 			 * or from BH in which case we can access SQ_EDC_INFO
6809 			 * instance
6810 			 */
6811 			if (in_task()) {
6812 				mutex_lock(&adev->grbm_idx_mutex);
6813 				gfx_v8_0_select_se_sh(adev, se_id, sh_id, cu_id);
6814 
6815 				sq_edc_source = REG_GET_FIELD(RREG32(mmSQ_EDC_INFO), SQ_EDC_INFO, SOURCE);
6816 
6817 				gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
6818 				mutex_unlock(&adev->grbm_idx_mutex);
6819 			}
6820 
6821 			if (enc == 1)
6822 				sprintf(type, "instruction intr");
6823 			else
6824 				sprintf(type, "EDC/ECC error");
6825 
6826 			DRM_INFO(
6827 				"SQ %s detected: "
6828 					"se_id %d, sh_id %d, cu_id %d, simd_id %d, wave_id %d, vm_id %d "
6829 					"trap %s, sq_ed_info.source %s.\n",
6830 					type, se_id, sh_id, cu_id,
6831 					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_WAVE, SIMD_ID),
6832 					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_WAVE, WAVE_ID),
6833 					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_WAVE, VM_ID),
6834 					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_WAVE, PRIV) ? "true" : "false",
6835 					(sq_edc_source != -1) ? sq_edc_source_names[sq_edc_source] : "unavailable"
6836 				);
6837 			break;
6838 		default:
6839 			DRM_ERROR("SQ invalid encoding type\n.");
6840 	}
6841 }
6842 
6843 static void gfx_v8_0_sq_irq_work_func(struct work_struct *work)
6844 {
6845 
6846 	struct amdgpu_device *adev = container_of(work, struct amdgpu_device, gfx.sq_work.work);
6847 	struct sq_work *sq_work = container_of(work, struct sq_work, work);
6848 
6849 	gfx_v8_0_parse_sq_irq(adev, sq_work->ih_data);
6850 }
6851 
6852 static int gfx_v8_0_sq_irq(struct amdgpu_device *adev,
6853 			   struct amdgpu_irq_src *source,
6854 			   struct amdgpu_iv_entry *entry)
6855 {
6856 	unsigned ih_data = entry->src_data[0];
6857 
6858 	/*
6859 	 * Try to submit work so SQ_EDC_INFO can be accessed from
6860 	 * BH. If previous work submission hasn't finished yet
6861 	 * just print whatever info is possible directly from the ISR.
6862 	 */
6863 	if (work_pending(&adev->gfx.sq_work.work)) {
6864 		gfx_v8_0_parse_sq_irq(adev, ih_data);
6865 	} else {
6866 		adev->gfx.sq_work.ih_data = ih_data;
6867 		schedule_work(&adev->gfx.sq_work.work);
6868 	}
6869 
6870 	return 0;
6871 }
6872 
6873 static const struct amd_ip_funcs gfx_v8_0_ip_funcs = {
6874 	.name = "gfx_v8_0",
6875 	.early_init = gfx_v8_0_early_init,
6876 	.late_init = gfx_v8_0_late_init,
6877 	.sw_init = gfx_v8_0_sw_init,
6878 	.sw_fini = gfx_v8_0_sw_fini,
6879 	.hw_init = gfx_v8_0_hw_init,
6880 	.hw_fini = gfx_v8_0_hw_fini,
6881 	.suspend = gfx_v8_0_suspend,
6882 	.resume = gfx_v8_0_resume,
6883 	.is_idle = gfx_v8_0_is_idle,
6884 	.wait_for_idle = gfx_v8_0_wait_for_idle,
6885 	.check_soft_reset = gfx_v8_0_check_soft_reset,
6886 	.pre_soft_reset = gfx_v8_0_pre_soft_reset,
6887 	.soft_reset = gfx_v8_0_soft_reset,
6888 	.post_soft_reset = gfx_v8_0_post_soft_reset,
6889 	.set_clockgating_state = gfx_v8_0_set_clockgating_state,
6890 	.set_powergating_state = gfx_v8_0_set_powergating_state,
6891 	.get_clockgating_state = gfx_v8_0_get_clockgating_state,
6892 };
6893 
6894 static const struct amdgpu_ring_funcs gfx_v8_0_ring_funcs_gfx = {
6895 	.type = AMDGPU_RING_TYPE_GFX,
6896 	.align_mask = 0xff,
6897 	.nop = PACKET3(PACKET3_NOP, 0x3FFF),
6898 	.support_64bit_ptrs = false,
6899 	.get_rptr = gfx_v8_0_ring_get_rptr,
6900 	.get_wptr = gfx_v8_0_ring_get_wptr_gfx,
6901 	.set_wptr = gfx_v8_0_ring_set_wptr_gfx,
6902 	.emit_frame_size = /* maximum 215dw if count 16 IBs in */
6903 		5 +  /* COND_EXEC */
6904 		7 +  /* PIPELINE_SYNC */
6905 		VI_FLUSH_GPU_TLB_NUM_WREG * 5 + 9 + /* VM_FLUSH */
6906 		12 +  /* FENCE for VM_FLUSH */
6907 		20 + /* GDS switch */
6908 		4 + /* double SWITCH_BUFFER,
6909 		       the first COND_EXEC jump to the place just
6910 			   prior to this double SWITCH_BUFFER  */
6911 		5 + /* COND_EXEC */
6912 		7 +	 /*	HDP_flush */
6913 		4 +	 /*	VGT_flush */
6914 		14 + /*	CE_META */
6915 		31 + /*	DE_META */
6916 		3 + /* CNTX_CTRL */
6917 		5 + /* HDP_INVL */
6918 		12 + 12 + /* FENCE x2 */
6919 		2, /* SWITCH_BUFFER */
6920 	.emit_ib_size =	4, /* gfx_v8_0_ring_emit_ib_gfx */
6921 	.emit_ib = gfx_v8_0_ring_emit_ib_gfx,
6922 	.emit_fence = gfx_v8_0_ring_emit_fence_gfx,
6923 	.emit_pipeline_sync = gfx_v8_0_ring_emit_pipeline_sync,
6924 	.emit_vm_flush = gfx_v8_0_ring_emit_vm_flush,
6925 	.emit_gds_switch = gfx_v8_0_ring_emit_gds_switch,
6926 	.emit_hdp_flush = gfx_v8_0_ring_emit_hdp_flush,
6927 	.test_ring = gfx_v8_0_ring_test_ring,
6928 	.test_ib = gfx_v8_0_ring_test_ib,
6929 	.insert_nop = amdgpu_ring_insert_nop,
6930 	.pad_ib = amdgpu_ring_generic_pad_ib,
6931 	.emit_switch_buffer = gfx_v8_ring_emit_sb,
6932 	.emit_cntxcntl = gfx_v8_ring_emit_cntxcntl,
6933 	.init_cond_exec = gfx_v8_0_ring_emit_init_cond_exec,
6934 	.patch_cond_exec = gfx_v8_0_ring_emit_patch_cond_exec,
6935 	.emit_wreg = gfx_v8_0_ring_emit_wreg,
6936 	.soft_recovery = gfx_v8_0_ring_soft_recovery,
6937 };
6938 
6939 static const struct amdgpu_ring_funcs gfx_v8_0_ring_funcs_compute = {
6940 	.type = AMDGPU_RING_TYPE_COMPUTE,
6941 	.align_mask = 0xff,
6942 	.nop = PACKET3(PACKET3_NOP, 0x3FFF),
6943 	.support_64bit_ptrs = false,
6944 	.get_rptr = gfx_v8_0_ring_get_rptr,
6945 	.get_wptr = gfx_v8_0_ring_get_wptr_compute,
6946 	.set_wptr = gfx_v8_0_ring_set_wptr_compute,
6947 	.emit_frame_size =
6948 		20 + /* gfx_v8_0_ring_emit_gds_switch */
6949 		7 + /* gfx_v8_0_ring_emit_hdp_flush */
6950 		5 + /* hdp_invalidate */
6951 		7 + /* gfx_v8_0_ring_emit_pipeline_sync */
6952 		VI_FLUSH_GPU_TLB_NUM_WREG * 5 + 7 + /* gfx_v8_0_ring_emit_vm_flush */
6953 		7 + 7 + 7, /* gfx_v8_0_ring_emit_fence_compute x3 for user fence, vm fence */
6954 	.emit_ib_size =	7, /* gfx_v8_0_ring_emit_ib_compute */
6955 	.emit_ib = gfx_v8_0_ring_emit_ib_compute,
6956 	.emit_fence = gfx_v8_0_ring_emit_fence_compute,
6957 	.emit_pipeline_sync = gfx_v8_0_ring_emit_pipeline_sync,
6958 	.emit_vm_flush = gfx_v8_0_ring_emit_vm_flush,
6959 	.emit_gds_switch = gfx_v8_0_ring_emit_gds_switch,
6960 	.emit_hdp_flush = gfx_v8_0_ring_emit_hdp_flush,
6961 	.test_ring = gfx_v8_0_ring_test_ring,
6962 	.test_ib = gfx_v8_0_ring_test_ib,
6963 	.insert_nop = amdgpu_ring_insert_nop,
6964 	.pad_ib = amdgpu_ring_generic_pad_ib,
6965 	.set_priority = gfx_v8_0_ring_set_priority_compute,
6966 	.emit_wreg = gfx_v8_0_ring_emit_wreg,
6967 };
6968 
6969 static const struct amdgpu_ring_funcs gfx_v8_0_ring_funcs_kiq = {
6970 	.type = AMDGPU_RING_TYPE_KIQ,
6971 	.align_mask = 0xff,
6972 	.nop = PACKET3(PACKET3_NOP, 0x3FFF),
6973 	.support_64bit_ptrs = false,
6974 	.get_rptr = gfx_v8_0_ring_get_rptr,
6975 	.get_wptr = gfx_v8_0_ring_get_wptr_compute,
6976 	.set_wptr = gfx_v8_0_ring_set_wptr_compute,
6977 	.emit_frame_size =
6978 		20 + /* gfx_v8_0_ring_emit_gds_switch */
6979 		7 + /* gfx_v8_0_ring_emit_hdp_flush */
6980 		5 + /* hdp_invalidate */
6981 		7 + /* gfx_v8_0_ring_emit_pipeline_sync */
6982 		17 + /* gfx_v8_0_ring_emit_vm_flush */
6983 		7 + 7 + 7, /* gfx_v8_0_ring_emit_fence_kiq x3 for user fence, vm fence */
6984 	.emit_ib_size =	7, /* gfx_v8_0_ring_emit_ib_compute */
6985 	.emit_fence = gfx_v8_0_ring_emit_fence_kiq,
6986 	.test_ring = gfx_v8_0_ring_test_ring,
6987 	.insert_nop = amdgpu_ring_insert_nop,
6988 	.pad_ib = amdgpu_ring_generic_pad_ib,
6989 	.emit_rreg = gfx_v8_0_ring_emit_rreg,
6990 	.emit_wreg = gfx_v8_0_ring_emit_wreg,
6991 };
6992 
6993 static void gfx_v8_0_set_ring_funcs(struct amdgpu_device *adev)
6994 {
6995 	int i;
6996 
6997 	adev->gfx.kiq.ring.funcs = &gfx_v8_0_ring_funcs_kiq;
6998 
6999 	for (i = 0; i < adev->gfx.num_gfx_rings; i++)
7000 		adev->gfx.gfx_ring[i].funcs = &gfx_v8_0_ring_funcs_gfx;
7001 
7002 	for (i = 0; i < adev->gfx.num_compute_rings; i++)
7003 		adev->gfx.compute_ring[i].funcs = &gfx_v8_0_ring_funcs_compute;
7004 }
7005 
7006 static const struct amdgpu_irq_src_funcs gfx_v8_0_eop_irq_funcs = {
7007 	.set = gfx_v8_0_set_eop_interrupt_state,
7008 	.process = gfx_v8_0_eop_irq,
7009 };
7010 
7011 static const struct amdgpu_irq_src_funcs gfx_v8_0_priv_reg_irq_funcs = {
7012 	.set = gfx_v8_0_set_priv_reg_fault_state,
7013 	.process = gfx_v8_0_priv_reg_irq,
7014 };
7015 
7016 static const struct amdgpu_irq_src_funcs gfx_v8_0_priv_inst_irq_funcs = {
7017 	.set = gfx_v8_0_set_priv_inst_fault_state,
7018 	.process = gfx_v8_0_priv_inst_irq,
7019 };
7020 
7021 static const struct amdgpu_irq_src_funcs gfx_v8_0_cp_ecc_error_irq_funcs = {
7022 	.set = gfx_v8_0_set_cp_ecc_int_state,
7023 	.process = gfx_v8_0_cp_ecc_error_irq,
7024 };
7025 
7026 static const struct amdgpu_irq_src_funcs gfx_v8_0_sq_irq_funcs = {
7027 	.set = gfx_v8_0_set_sq_int_state,
7028 	.process = gfx_v8_0_sq_irq,
7029 };
7030 
7031 static void gfx_v8_0_set_irq_funcs(struct amdgpu_device *adev)
7032 {
7033 	adev->gfx.eop_irq.num_types = AMDGPU_CP_IRQ_LAST;
7034 	adev->gfx.eop_irq.funcs = &gfx_v8_0_eop_irq_funcs;
7035 
7036 	adev->gfx.priv_reg_irq.num_types = 1;
7037 	adev->gfx.priv_reg_irq.funcs = &gfx_v8_0_priv_reg_irq_funcs;
7038 
7039 	adev->gfx.priv_inst_irq.num_types = 1;
7040 	adev->gfx.priv_inst_irq.funcs = &gfx_v8_0_priv_inst_irq_funcs;
7041 
7042 	adev->gfx.cp_ecc_error_irq.num_types = 1;
7043 	adev->gfx.cp_ecc_error_irq.funcs = &gfx_v8_0_cp_ecc_error_irq_funcs;
7044 
7045 	adev->gfx.sq_irq.num_types = 1;
7046 	adev->gfx.sq_irq.funcs = &gfx_v8_0_sq_irq_funcs;
7047 }
7048 
7049 static void gfx_v8_0_set_rlc_funcs(struct amdgpu_device *adev)
7050 {
7051 	adev->gfx.rlc.funcs = &iceland_rlc_funcs;
7052 }
7053 
7054 static void gfx_v8_0_set_gds_init(struct amdgpu_device *adev)
7055 {
7056 	/* init asci gds info */
7057 	adev->gds.gds_size = RREG32(mmGDS_VMID0_SIZE);
7058 	adev->gds.gws_size = 64;
7059 	adev->gds.oa_size = 16;
7060 	adev->gds.gds_compute_max_wave_id = RREG32(mmGDS_COMPUTE_MAX_WAVE_ID);
7061 }
7062 
7063 static void gfx_v8_0_set_user_cu_inactive_bitmap(struct amdgpu_device *adev,
7064 						 u32 bitmap)
7065 {
7066 	u32 data;
7067 
7068 	if (!bitmap)
7069 		return;
7070 
7071 	data = bitmap << GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_CUS__SHIFT;
7072 	data &= GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_CUS_MASK;
7073 
7074 	WREG32(mmGC_USER_SHADER_ARRAY_CONFIG, data);
7075 }
7076 
7077 static u32 gfx_v8_0_get_cu_active_bitmap(struct amdgpu_device *adev)
7078 {
7079 	u32 data, mask;
7080 
7081 	data =  RREG32(mmCC_GC_SHADER_ARRAY_CONFIG) |
7082 		RREG32(mmGC_USER_SHADER_ARRAY_CONFIG);
7083 
7084 	mask = amdgpu_gfx_create_bitmask(adev->gfx.config.max_cu_per_sh);
7085 
7086 	return ~REG_GET_FIELD(data, CC_GC_SHADER_ARRAY_CONFIG, INACTIVE_CUS) & mask;
7087 }
7088 
7089 static void gfx_v8_0_get_cu_info(struct amdgpu_device *adev)
7090 {
7091 	int i, j, k, counter, active_cu_number = 0;
7092 	u32 mask, bitmap, ao_bitmap, ao_cu_mask = 0;
7093 	struct amdgpu_cu_info *cu_info = &adev->gfx.cu_info;
7094 	unsigned disable_masks[4 * 2];
7095 	u32 ao_cu_num;
7096 
7097 	memset(cu_info, 0, sizeof(*cu_info));
7098 
7099 	if (adev->flags & AMD_IS_APU)
7100 		ao_cu_num = 2;
7101 	else
7102 		ao_cu_num = adev->gfx.config.max_cu_per_sh;
7103 
7104 	amdgpu_gfx_parse_disable_cu(disable_masks, 4, 2);
7105 
7106 	mutex_lock(&adev->grbm_idx_mutex);
7107 	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
7108 		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
7109 			mask = 1;
7110 			ao_bitmap = 0;
7111 			counter = 0;
7112 			gfx_v8_0_select_se_sh(adev, i, j, 0xffffffff);
7113 			if (i < 4 && j < 2)
7114 				gfx_v8_0_set_user_cu_inactive_bitmap(
7115 					adev, disable_masks[i * 2 + j]);
7116 			bitmap = gfx_v8_0_get_cu_active_bitmap(adev);
7117 			cu_info->bitmap[i][j] = bitmap;
7118 
7119 			for (k = 0; k < adev->gfx.config.max_cu_per_sh; k ++) {
7120 				if (bitmap & mask) {
7121 					if (counter < ao_cu_num)
7122 						ao_bitmap |= mask;
7123 					counter ++;
7124 				}
7125 				mask <<= 1;
7126 			}
7127 			active_cu_number += counter;
7128 			if (i < 2 && j < 2)
7129 				ao_cu_mask |= (ao_bitmap << (i * 16 + j * 8));
7130 			cu_info->ao_cu_bitmap[i][j] = ao_bitmap;
7131 		}
7132 	}
7133 	gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
7134 	mutex_unlock(&adev->grbm_idx_mutex);
7135 
7136 	cu_info->number = active_cu_number;
7137 	cu_info->ao_cu_mask = ao_cu_mask;
7138 	cu_info->simd_per_cu = NUM_SIMD_PER_CU;
7139 	cu_info->max_waves_per_simd = 10;
7140 	cu_info->max_scratch_slots_per_cu = 32;
7141 	cu_info->wave_front_size = 64;
7142 	cu_info->lds_size = 64;
7143 }
7144 
7145 const struct amdgpu_ip_block_version gfx_v8_0_ip_block =
7146 {
7147 	.type = AMD_IP_BLOCK_TYPE_GFX,
7148 	.major = 8,
7149 	.minor = 0,
7150 	.rev = 0,
7151 	.funcs = &gfx_v8_0_ip_funcs,
7152 };
7153 
7154 const struct amdgpu_ip_block_version gfx_v8_1_ip_block =
7155 {
7156 	.type = AMD_IP_BLOCK_TYPE_GFX,
7157 	.major = 8,
7158 	.minor = 1,
7159 	.rev = 0,
7160 	.funcs = &gfx_v8_0_ip_funcs,
7161 };
7162 
7163 static void gfx_v8_0_ring_emit_ce_meta(struct amdgpu_ring *ring)
7164 {
7165 	uint64_t ce_payload_addr;
7166 	int cnt_ce;
7167 	union {
7168 		struct vi_ce_ib_state regular;
7169 		struct vi_ce_ib_state_chained_ib chained;
7170 	} ce_payload = {};
7171 
7172 	if (ring->adev->virt.chained_ib_support) {
7173 		ce_payload_addr = amdgpu_csa_vaddr(ring->adev) +
7174 			offsetof(struct vi_gfx_meta_data_chained_ib, ce_payload);
7175 		cnt_ce = (sizeof(ce_payload.chained) >> 2) + 4 - 2;
7176 	} else {
7177 		ce_payload_addr = amdgpu_csa_vaddr(ring->adev) +
7178 			offsetof(struct vi_gfx_meta_data, ce_payload);
7179 		cnt_ce = (sizeof(ce_payload.regular) >> 2) + 4 - 2;
7180 	}
7181 
7182 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, cnt_ce));
7183 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(2) |
7184 				WRITE_DATA_DST_SEL(8) |
7185 				WR_CONFIRM) |
7186 				WRITE_DATA_CACHE_POLICY(0));
7187 	amdgpu_ring_write(ring, lower_32_bits(ce_payload_addr));
7188 	amdgpu_ring_write(ring, upper_32_bits(ce_payload_addr));
7189 	amdgpu_ring_write_multiple(ring, (void *)&ce_payload, cnt_ce - 2);
7190 }
7191 
7192 static void gfx_v8_0_ring_emit_de_meta(struct amdgpu_ring *ring)
7193 {
7194 	uint64_t de_payload_addr, gds_addr, csa_addr;
7195 	int cnt_de;
7196 	union {
7197 		struct vi_de_ib_state regular;
7198 		struct vi_de_ib_state_chained_ib chained;
7199 	} de_payload = {};
7200 
7201 	csa_addr = amdgpu_csa_vaddr(ring->adev);
7202 	gds_addr = csa_addr + 4096;
7203 	if (ring->adev->virt.chained_ib_support) {
7204 		de_payload.chained.gds_backup_addrlo = lower_32_bits(gds_addr);
7205 		de_payload.chained.gds_backup_addrhi = upper_32_bits(gds_addr);
7206 		de_payload_addr = csa_addr + offsetof(struct vi_gfx_meta_data_chained_ib, de_payload);
7207 		cnt_de = (sizeof(de_payload.chained) >> 2) + 4 - 2;
7208 	} else {
7209 		de_payload.regular.gds_backup_addrlo = lower_32_bits(gds_addr);
7210 		de_payload.regular.gds_backup_addrhi = upper_32_bits(gds_addr);
7211 		de_payload_addr = csa_addr + offsetof(struct vi_gfx_meta_data, de_payload);
7212 		cnt_de = (sizeof(de_payload.regular) >> 2) + 4 - 2;
7213 	}
7214 
7215 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, cnt_de));
7216 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(1) |
7217 				WRITE_DATA_DST_SEL(8) |
7218 				WR_CONFIRM) |
7219 				WRITE_DATA_CACHE_POLICY(0));
7220 	amdgpu_ring_write(ring, lower_32_bits(de_payload_addr));
7221 	amdgpu_ring_write(ring, upper_32_bits(de_payload_addr));
7222 	amdgpu_ring_write_multiple(ring, (void *)&de_payload, cnt_de - 2);
7223 }
7224