xref: /openbmc/linux/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c (revision a72b9869)
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 "amdgpu_ring.h"
33 #include "vi.h"
34 #include "vi_structs.h"
35 #include "vid.h"
36 #include "amdgpu_ucode.h"
37 #include "amdgpu_atombios.h"
38 #include "atombios_i2c.h"
39 #include "clearstate_vi.h"
40 
41 #include "gmc/gmc_8_2_d.h"
42 #include "gmc/gmc_8_2_sh_mask.h"
43 
44 #include "oss/oss_3_0_d.h"
45 #include "oss/oss_3_0_sh_mask.h"
46 
47 #include "bif/bif_5_0_d.h"
48 #include "bif/bif_5_0_sh_mask.h"
49 #include "gca/gfx_8_0_d.h"
50 #include "gca/gfx_8_0_enum.h"
51 #include "gca/gfx_8_0_sh_mask.h"
52 
53 #include "dce/dce_10_0_d.h"
54 #include "dce/dce_10_0_sh_mask.h"
55 
56 #include "smu/smu_7_1_3_d.h"
57 
58 #include "ivsrcid/ivsrcid_vislands30.h"
59 
60 #define GFX8_NUM_GFX_RINGS     1
61 #define GFX8_MEC_HPD_SIZE 4096
62 
63 #define TOPAZ_GB_ADDR_CONFIG_GOLDEN 0x22010001
64 #define CARRIZO_GB_ADDR_CONFIG_GOLDEN 0x22010001
65 #define POLARIS11_GB_ADDR_CONFIG_GOLDEN 0x22011002
66 #define TONGA_GB_ADDR_CONFIG_GOLDEN 0x22011003
67 
68 #define ARRAY_MODE(x)					((x) << GB_TILE_MODE0__ARRAY_MODE__SHIFT)
69 #define PIPE_CONFIG(x)					((x) << GB_TILE_MODE0__PIPE_CONFIG__SHIFT)
70 #define TILE_SPLIT(x)					((x) << GB_TILE_MODE0__TILE_SPLIT__SHIFT)
71 #define MICRO_TILE_MODE_NEW(x)				((x) << GB_TILE_MODE0__MICRO_TILE_MODE_NEW__SHIFT)
72 #define SAMPLE_SPLIT(x)					((x) << GB_TILE_MODE0__SAMPLE_SPLIT__SHIFT)
73 #define BANK_WIDTH(x)					((x) << GB_MACROTILE_MODE0__BANK_WIDTH__SHIFT)
74 #define BANK_HEIGHT(x)					((x) << GB_MACROTILE_MODE0__BANK_HEIGHT__SHIFT)
75 #define MACRO_TILE_ASPECT(x)				((x) << GB_MACROTILE_MODE0__MACRO_TILE_ASPECT__SHIFT)
76 #define NUM_BANKS(x)					((x) << GB_MACROTILE_MODE0__NUM_BANKS__SHIFT)
77 
78 #define RLC_CGTT_MGCG_OVERRIDE__CPF_MASK            0x00000001L
79 #define RLC_CGTT_MGCG_OVERRIDE__RLC_MASK            0x00000002L
80 #define RLC_CGTT_MGCG_OVERRIDE__MGCG_MASK           0x00000004L
81 #define RLC_CGTT_MGCG_OVERRIDE__CGCG_MASK           0x00000008L
82 #define RLC_CGTT_MGCG_OVERRIDE__CGLS_MASK           0x00000010L
83 #define RLC_CGTT_MGCG_OVERRIDE__GRBM_MASK           0x00000020L
84 
85 /* BPM SERDES CMD */
86 #define SET_BPM_SERDES_CMD    1
87 #define CLE_BPM_SERDES_CMD    0
88 
89 /* BPM Register Address*/
90 enum {
91 	BPM_REG_CGLS_EN = 0,        /* Enable/Disable CGLS */
92 	BPM_REG_CGLS_ON,            /* ON/OFF CGLS: shall be controlled by RLC FW */
93 	BPM_REG_CGCG_OVERRIDE,      /* Set/Clear CGCG Override */
94 	BPM_REG_MGCG_OVERRIDE,      /* Set/Clear MGCG Override */
95 	BPM_REG_FGCG_OVERRIDE,      /* Set/Clear FGCG Override */
96 	BPM_REG_FGCG_MAX
97 };
98 
99 #define RLC_FormatDirectRegListLength        14
100 
101 MODULE_FIRMWARE("amdgpu/carrizo_ce.bin");
102 MODULE_FIRMWARE("amdgpu/carrizo_pfp.bin");
103 MODULE_FIRMWARE("amdgpu/carrizo_me.bin");
104 MODULE_FIRMWARE("amdgpu/carrizo_mec.bin");
105 MODULE_FIRMWARE("amdgpu/carrizo_mec2.bin");
106 MODULE_FIRMWARE("amdgpu/carrizo_rlc.bin");
107 
108 MODULE_FIRMWARE("amdgpu/stoney_ce.bin");
109 MODULE_FIRMWARE("amdgpu/stoney_pfp.bin");
110 MODULE_FIRMWARE("amdgpu/stoney_me.bin");
111 MODULE_FIRMWARE("amdgpu/stoney_mec.bin");
112 MODULE_FIRMWARE("amdgpu/stoney_rlc.bin");
113 
114 MODULE_FIRMWARE("amdgpu/tonga_ce.bin");
115 MODULE_FIRMWARE("amdgpu/tonga_pfp.bin");
116 MODULE_FIRMWARE("amdgpu/tonga_me.bin");
117 MODULE_FIRMWARE("amdgpu/tonga_mec.bin");
118 MODULE_FIRMWARE("amdgpu/tonga_mec2.bin");
119 MODULE_FIRMWARE("amdgpu/tonga_rlc.bin");
120 
121 MODULE_FIRMWARE("amdgpu/topaz_ce.bin");
122 MODULE_FIRMWARE("amdgpu/topaz_pfp.bin");
123 MODULE_FIRMWARE("amdgpu/topaz_me.bin");
124 MODULE_FIRMWARE("amdgpu/topaz_mec.bin");
125 MODULE_FIRMWARE("amdgpu/topaz_rlc.bin");
126 
127 MODULE_FIRMWARE("amdgpu/fiji_ce.bin");
128 MODULE_FIRMWARE("amdgpu/fiji_pfp.bin");
129 MODULE_FIRMWARE("amdgpu/fiji_me.bin");
130 MODULE_FIRMWARE("amdgpu/fiji_mec.bin");
131 MODULE_FIRMWARE("amdgpu/fiji_mec2.bin");
132 MODULE_FIRMWARE("amdgpu/fiji_rlc.bin");
133 
134 MODULE_FIRMWARE("amdgpu/polaris10_ce.bin");
135 MODULE_FIRMWARE("amdgpu/polaris10_ce_2.bin");
136 MODULE_FIRMWARE("amdgpu/polaris10_pfp.bin");
137 MODULE_FIRMWARE("amdgpu/polaris10_pfp_2.bin");
138 MODULE_FIRMWARE("amdgpu/polaris10_me.bin");
139 MODULE_FIRMWARE("amdgpu/polaris10_me_2.bin");
140 MODULE_FIRMWARE("amdgpu/polaris10_mec.bin");
141 MODULE_FIRMWARE("amdgpu/polaris10_mec_2.bin");
142 MODULE_FIRMWARE("amdgpu/polaris10_mec2.bin");
143 MODULE_FIRMWARE("amdgpu/polaris10_mec2_2.bin");
144 MODULE_FIRMWARE("amdgpu/polaris10_rlc.bin");
145 
146 MODULE_FIRMWARE("amdgpu/polaris11_ce.bin");
147 MODULE_FIRMWARE("amdgpu/polaris11_ce_2.bin");
148 MODULE_FIRMWARE("amdgpu/polaris11_pfp.bin");
149 MODULE_FIRMWARE("amdgpu/polaris11_pfp_2.bin");
150 MODULE_FIRMWARE("amdgpu/polaris11_me.bin");
151 MODULE_FIRMWARE("amdgpu/polaris11_me_2.bin");
152 MODULE_FIRMWARE("amdgpu/polaris11_mec.bin");
153 MODULE_FIRMWARE("amdgpu/polaris11_mec_2.bin");
154 MODULE_FIRMWARE("amdgpu/polaris11_mec2.bin");
155 MODULE_FIRMWARE("amdgpu/polaris11_mec2_2.bin");
156 MODULE_FIRMWARE("amdgpu/polaris11_rlc.bin");
157 
158 MODULE_FIRMWARE("amdgpu/polaris12_ce.bin");
159 MODULE_FIRMWARE("amdgpu/polaris12_ce_2.bin");
160 MODULE_FIRMWARE("amdgpu/polaris12_pfp.bin");
161 MODULE_FIRMWARE("amdgpu/polaris12_pfp_2.bin");
162 MODULE_FIRMWARE("amdgpu/polaris12_me.bin");
163 MODULE_FIRMWARE("amdgpu/polaris12_me_2.bin");
164 MODULE_FIRMWARE("amdgpu/polaris12_mec.bin");
165 MODULE_FIRMWARE("amdgpu/polaris12_mec_2.bin");
166 MODULE_FIRMWARE("amdgpu/polaris12_mec2.bin");
167 MODULE_FIRMWARE("amdgpu/polaris12_mec2_2.bin");
168 MODULE_FIRMWARE("amdgpu/polaris12_rlc.bin");
169 
170 MODULE_FIRMWARE("amdgpu/vegam_ce.bin");
171 MODULE_FIRMWARE("amdgpu/vegam_pfp.bin");
172 MODULE_FIRMWARE("amdgpu/vegam_me.bin");
173 MODULE_FIRMWARE("amdgpu/vegam_mec.bin");
174 MODULE_FIRMWARE("amdgpu/vegam_mec2.bin");
175 MODULE_FIRMWARE("amdgpu/vegam_rlc.bin");
176 
177 static const struct amdgpu_gds_reg_offset amdgpu_gds_reg_offset[] =
178 {
179 	{mmGDS_VMID0_BASE, mmGDS_VMID0_SIZE, mmGDS_GWS_VMID0, mmGDS_OA_VMID0},
180 	{mmGDS_VMID1_BASE, mmGDS_VMID1_SIZE, mmGDS_GWS_VMID1, mmGDS_OA_VMID1},
181 	{mmGDS_VMID2_BASE, mmGDS_VMID2_SIZE, mmGDS_GWS_VMID2, mmGDS_OA_VMID2},
182 	{mmGDS_VMID3_BASE, mmGDS_VMID3_SIZE, mmGDS_GWS_VMID3, mmGDS_OA_VMID3},
183 	{mmGDS_VMID4_BASE, mmGDS_VMID4_SIZE, mmGDS_GWS_VMID4, mmGDS_OA_VMID4},
184 	{mmGDS_VMID5_BASE, mmGDS_VMID5_SIZE, mmGDS_GWS_VMID5, mmGDS_OA_VMID5},
185 	{mmGDS_VMID6_BASE, mmGDS_VMID6_SIZE, mmGDS_GWS_VMID6, mmGDS_OA_VMID6},
186 	{mmGDS_VMID7_BASE, mmGDS_VMID7_SIZE, mmGDS_GWS_VMID7, mmGDS_OA_VMID7},
187 	{mmGDS_VMID8_BASE, mmGDS_VMID8_SIZE, mmGDS_GWS_VMID8, mmGDS_OA_VMID8},
188 	{mmGDS_VMID9_BASE, mmGDS_VMID9_SIZE, mmGDS_GWS_VMID9, mmGDS_OA_VMID9},
189 	{mmGDS_VMID10_BASE, mmGDS_VMID10_SIZE, mmGDS_GWS_VMID10, mmGDS_OA_VMID10},
190 	{mmGDS_VMID11_BASE, mmGDS_VMID11_SIZE, mmGDS_GWS_VMID11, mmGDS_OA_VMID11},
191 	{mmGDS_VMID12_BASE, mmGDS_VMID12_SIZE, mmGDS_GWS_VMID12, mmGDS_OA_VMID12},
192 	{mmGDS_VMID13_BASE, mmGDS_VMID13_SIZE, mmGDS_GWS_VMID13, mmGDS_OA_VMID13},
193 	{mmGDS_VMID14_BASE, mmGDS_VMID14_SIZE, mmGDS_GWS_VMID14, mmGDS_OA_VMID14},
194 	{mmGDS_VMID15_BASE, mmGDS_VMID15_SIZE, mmGDS_GWS_VMID15, mmGDS_OA_VMID15}
195 };
196 
197 static const u32 golden_settings_tonga_a11[] =
198 {
199 	mmCB_HW_CONTROL, 0xfffdf3cf, 0x00007208,
200 	mmCB_HW_CONTROL_3, 0x00000040, 0x00000040,
201 	mmDB_DEBUG2, 0xf00fffff, 0x00000400,
202 	mmGB_GPU_ID, 0x0000000f, 0x00000000,
203 	mmPA_SC_ENHANCE, 0xffffffff, 0x20000001,
204 	mmPA_SC_FIFO_DEPTH_CNTL, 0x000003ff, 0x000000fc,
205 	mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000,
206 	mmRLC_CGCG_CGLS_CTRL, 0x00000003, 0x0000003c,
207 	mmSQ_RANDOM_WAVE_PRI, 0x001fffff, 0x000006fd,
208 	mmTA_CNTL_AUX, 0x000f000f, 0x000b0000,
209 	mmTCC_CTRL, 0x00100000, 0xf31fff7f,
210 	mmTCC_EXE_DISABLE, 0x00000002, 0x00000002,
211 	mmTCP_ADDR_CONFIG, 0x000003ff, 0x000002fb,
212 	mmTCP_CHAN_STEER_HI, 0xffffffff, 0x0000543b,
213 	mmTCP_CHAN_STEER_LO, 0xffffffff, 0xa9210876,
214 	mmVGT_RESET_DEBUG, 0x00000004, 0x00000004,
215 };
216 
217 static const u32 tonga_golden_common_all[] =
218 {
219 	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
220 	mmPA_SC_RASTER_CONFIG, 0xffffffff, 0x16000012,
221 	mmPA_SC_RASTER_CONFIG_1, 0xffffffff, 0x0000002A,
222 	mmGB_ADDR_CONFIG, 0xffffffff, 0x22011003,
223 	mmSPI_RESOURCE_RESERVE_CU_0, 0xffffffff, 0x00000800,
224 	mmSPI_RESOURCE_RESERVE_CU_1, 0xffffffff, 0x00000800,
225 	mmSPI_RESOURCE_RESERVE_EN_CU_0, 0xffffffff, 0x00FF7FBF,
226 	mmSPI_RESOURCE_RESERVE_EN_CU_1, 0xffffffff, 0x00FF7FAF
227 };
228 
229 static const u32 tonga_mgcg_cgcg_init[] =
230 {
231 	mmRLC_CGTT_MGCG_OVERRIDE, 0xffffffff, 0xffffffff,
232 	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
233 	mmCB_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100,
234 	mmCGTT_BCI_CLK_CTRL, 0xffffffff, 0x00000100,
235 	mmCGTT_CP_CLK_CTRL, 0xffffffff, 0x00000100,
236 	mmCGTT_CPC_CLK_CTRL, 0xffffffff, 0x00000100,
237 	mmCGTT_CPF_CLK_CTRL, 0xffffffff, 0x40000100,
238 	mmCGTT_GDS_CLK_CTRL, 0xffffffff, 0x00000100,
239 	mmCGTT_IA_CLK_CTRL, 0xffffffff, 0x06000100,
240 	mmCGTT_PA_CLK_CTRL, 0xffffffff, 0x00000100,
241 	mmCGTT_WD_CLK_CTRL, 0xffffffff, 0x06000100,
242 	mmCGTT_PC_CLK_CTRL, 0xffffffff, 0x00000100,
243 	mmCGTT_RLC_CLK_CTRL, 0xffffffff, 0x00000100,
244 	mmCGTT_SC_CLK_CTRL, 0xffffffff, 0x00000100,
245 	mmCGTT_SPI_CLK_CTRL, 0xffffffff, 0x00000100,
246 	mmCGTT_SQ_CLK_CTRL, 0xffffffff, 0x00000100,
247 	mmCGTT_SQG_CLK_CTRL, 0xffffffff, 0x00000100,
248 	mmCGTT_SX_CLK_CTRL0, 0xffffffff, 0x00000100,
249 	mmCGTT_SX_CLK_CTRL1, 0xffffffff, 0x00000100,
250 	mmCGTT_SX_CLK_CTRL2, 0xffffffff, 0x00000100,
251 	mmCGTT_SX_CLK_CTRL3, 0xffffffff, 0x00000100,
252 	mmCGTT_SX_CLK_CTRL4, 0xffffffff, 0x00000100,
253 	mmCGTT_TCI_CLK_CTRL, 0xffffffff, 0x00000100,
254 	mmCGTT_TCP_CLK_CTRL, 0xffffffff, 0x00000100,
255 	mmCGTT_VGT_CLK_CTRL, 0xffffffff, 0x06000100,
256 	mmDB_CGTT_CLK_CTRL_0, 0xffffffff, 0x00000100,
257 	mmTA_CGTT_CTRL, 0xffffffff, 0x00000100,
258 	mmTCA_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100,
259 	mmTCC_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100,
260 	mmTD_CGTT_CTRL, 0xffffffff, 0x00000100,
261 	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
262 	mmCGTS_CU0_SP0_CTRL_REG, 0xffffffff, 0x00010000,
263 	mmCGTS_CU0_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
264 	mmCGTS_CU0_TA_SQC_CTRL_REG, 0xffffffff, 0x00040007,
265 	mmCGTS_CU0_SP1_CTRL_REG, 0xffffffff, 0x00060005,
266 	mmCGTS_CU0_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
267 	mmCGTS_CU1_SP0_CTRL_REG, 0xffffffff, 0x00010000,
268 	mmCGTS_CU1_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
269 	mmCGTS_CU1_TA_CTRL_REG, 0xffffffff, 0x00040007,
270 	mmCGTS_CU1_SP1_CTRL_REG, 0xffffffff, 0x00060005,
271 	mmCGTS_CU1_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
272 	mmCGTS_CU2_SP0_CTRL_REG, 0xffffffff, 0x00010000,
273 	mmCGTS_CU2_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
274 	mmCGTS_CU2_TA_CTRL_REG, 0xffffffff, 0x00040007,
275 	mmCGTS_CU2_SP1_CTRL_REG, 0xffffffff, 0x00060005,
276 	mmCGTS_CU2_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
277 	mmCGTS_CU3_SP0_CTRL_REG, 0xffffffff, 0x00010000,
278 	mmCGTS_CU3_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
279 	mmCGTS_CU3_TA_CTRL_REG, 0xffffffff, 0x00040007,
280 	mmCGTS_CU3_SP1_CTRL_REG, 0xffffffff, 0x00060005,
281 	mmCGTS_CU3_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
282 	mmCGTS_CU4_SP0_CTRL_REG, 0xffffffff, 0x00010000,
283 	mmCGTS_CU4_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
284 	mmCGTS_CU4_TA_SQC_CTRL_REG, 0xffffffff, 0x00040007,
285 	mmCGTS_CU4_SP1_CTRL_REG, 0xffffffff, 0x00060005,
286 	mmCGTS_CU4_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
287 	mmCGTS_CU5_SP0_CTRL_REG, 0xffffffff, 0x00010000,
288 	mmCGTS_CU5_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
289 	mmCGTS_CU5_TA_CTRL_REG, 0xffffffff, 0x00040007,
290 	mmCGTS_CU5_SP1_CTRL_REG, 0xffffffff, 0x00060005,
291 	mmCGTS_CU5_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
292 	mmCGTS_CU6_SP0_CTRL_REG, 0xffffffff, 0x00010000,
293 	mmCGTS_CU6_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
294 	mmCGTS_CU6_TA_CTRL_REG, 0xffffffff, 0x00040007,
295 	mmCGTS_CU6_SP1_CTRL_REG, 0xffffffff, 0x00060005,
296 	mmCGTS_CU6_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
297 	mmCGTS_CU7_SP0_CTRL_REG, 0xffffffff, 0x00010000,
298 	mmCGTS_CU7_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
299 	mmCGTS_CU7_TA_CTRL_REG, 0xffffffff, 0x00040007,
300 	mmCGTS_CU7_SP1_CTRL_REG, 0xffffffff, 0x00060005,
301 	mmCGTS_CU7_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
302 	mmCGTS_SM_CTRL_REG, 0xffffffff, 0x96e00200,
303 	mmCP_RB_WPTR_POLL_CNTL, 0xffffffff, 0x00900100,
304 	mmRLC_CGCG_CGLS_CTRL, 0xffffffff, 0x0020003c,
305 	mmCP_MEM_SLP_CNTL, 0x00000001, 0x00000001,
306 };
307 
308 static const u32 golden_settings_vegam_a11[] =
309 {
310 	mmCB_HW_CONTROL, 0x0001f3cf, 0x00007208,
311 	mmCB_HW_CONTROL_2, 0x0f000000, 0x0d000000,
312 	mmCB_HW_CONTROL_3, 0x000001ff, 0x00000040,
313 	mmDB_DEBUG2, 0xf00fffff, 0x00000400,
314 	mmPA_SC_ENHANCE, 0xffffffff, 0x20000001,
315 	mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000,
316 	mmPA_SC_RASTER_CONFIG, 0x3f3fffff, 0x3a00161a,
317 	mmPA_SC_RASTER_CONFIG_1, 0x0000003f, 0x0000002e,
318 	mmRLC_CGCG_CGLS_CTRL, 0x00000003, 0x0001003c,
319 	mmRLC_CGCG_CGLS_CTRL_3D, 0xffffffff, 0x0001003c,
320 	mmSQ_CONFIG, 0x07f80000, 0x01180000,
321 	mmTA_CNTL_AUX, 0x000f000f, 0x000b0000,
322 	mmTCC_CTRL, 0x00100000, 0xf31fff7f,
323 	mmTCP_ADDR_CONFIG, 0x000003ff, 0x000000f7,
324 	mmTCP_CHAN_STEER_HI, 0xffffffff, 0x00000000,
325 	mmTCP_CHAN_STEER_LO, 0xffffffff, 0x32761054,
326 	mmVGT_RESET_DEBUG, 0x00000004, 0x00000004,
327 };
328 
329 static const u32 vegam_golden_common_all[] =
330 {
331 	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
332 	mmGB_ADDR_CONFIG, 0xffffffff, 0x22011003,
333 	mmSPI_RESOURCE_RESERVE_CU_0, 0xffffffff, 0x00000800,
334 	mmSPI_RESOURCE_RESERVE_CU_1, 0xffffffff, 0x00000800,
335 	mmSPI_RESOURCE_RESERVE_EN_CU_0, 0xffffffff, 0x00FF7FBF,
336 	mmSPI_RESOURCE_RESERVE_EN_CU_1, 0xffffffff, 0x00FF7FAF,
337 };
338 
339 static const u32 golden_settings_polaris11_a11[] =
340 {
341 	mmCB_HW_CONTROL, 0x0000f3cf, 0x00007208,
342 	mmCB_HW_CONTROL_2, 0x0f000000, 0x0f000000,
343 	mmCB_HW_CONTROL_3, 0x000001ff, 0x00000040,
344 	mmDB_DEBUG2, 0xf00fffff, 0x00000400,
345 	mmPA_SC_ENHANCE, 0xffffffff, 0x20000001,
346 	mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000,
347 	mmPA_SC_RASTER_CONFIG, 0x3f3fffff, 0x16000012,
348 	mmPA_SC_RASTER_CONFIG_1, 0x0000003f, 0x00000000,
349 	mmRLC_CGCG_CGLS_CTRL, 0x00000003, 0x0001003c,
350 	mmRLC_CGCG_CGLS_CTRL_3D, 0xffffffff, 0x0001003c,
351 	mmSQ_CONFIG, 0x07f80000, 0x01180000,
352 	mmTA_CNTL_AUX, 0x000f000f, 0x000b0000,
353 	mmTCC_CTRL, 0x00100000, 0xf31fff7f,
354 	mmTCP_ADDR_CONFIG, 0x000003ff, 0x000000f3,
355 	mmTCP_CHAN_STEER_HI, 0xffffffff, 0x00000000,
356 	mmTCP_CHAN_STEER_LO, 0xffffffff, 0x00003210,
357 	mmVGT_RESET_DEBUG, 0x00000004, 0x00000004,
358 };
359 
360 static const u32 polaris11_golden_common_all[] =
361 {
362 	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
363 	mmGB_ADDR_CONFIG, 0xffffffff, 0x22011002,
364 	mmSPI_RESOURCE_RESERVE_CU_0, 0xffffffff, 0x00000800,
365 	mmSPI_RESOURCE_RESERVE_CU_1, 0xffffffff, 0x00000800,
366 	mmSPI_RESOURCE_RESERVE_EN_CU_0, 0xffffffff, 0x00FF7FBF,
367 	mmSPI_RESOURCE_RESERVE_EN_CU_1, 0xffffffff, 0x00FF7FAF,
368 };
369 
370 static const u32 golden_settings_polaris10_a11[] =
371 {
372 	mmATC_MISC_CG, 0x000c0fc0, 0x000c0200,
373 	mmCB_HW_CONTROL, 0x0001f3cf, 0x00007208,
374 	mmCB_HW_CONTROL_2, 0x0f000000, 0x0f000000,
375 	mmCB_HW_CONTROL_3, 0x000001ff, 0x00000040,
376 	mmDB_DEBUG2, 0xf00fffff, 0x00000400,
377 	mmPA_SC_ENHANCE, 0xffffffff, 0x20000001,
378 	mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000,
379 	mmPA_SC_RASTER_CONFIG, 0x3f3fffff, 0x16000012,
380 	mmPA_SC_RASTER_CONFIG_1, 0x0000003f, 0x0000002a,
381 	mmRLC_CGCG_CGLS_CTRL, 0x00000003, 0x0001003c,
382 	mmRLC_CGCG_CGLS_CTRL_3D, 0xffffffff, 0x0001003c,
383 	mmSQ_CONFIG, 0x07f80000, 0x07180000,
384 	mmTA_CNTL_AUX, 0x000f000f, 0x000b0000,
385 	mmTCC_CTRL, 0x00100000, 0xf31fff7f,
386 	mmTCP_ADDR_CONFIG, 0x000003ff, 0x000000f7,
387 	mmTCP_CHAN_STEER_HI, 0xffffffff, 0x00000000,
388 	mmVGT_RESET_DEBUG, 0x00000004, 0x00000004,
389 };
390 
391 static const u32 polaris10_golden_common_all[] =
392 {
393 	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
394 	mmPA_SC_RASTER_CONFIG, 0xffffffff, 0x16000012,
395 	mmPA_SC_RASTER_CONFIG_1, 0xffffffff, 0x0000002A,
396 	mmGB_ADDR_CONFIG, 0xffffffff, 0x22011003,
397 	mmSPI_RESOURCE_RESERVE_CU_0, 0xffffffff, 0x00000800,
398 	mmSPI_RESOURCE_RESERVE_CU_1, 0xffffffff, 0x00000800,
399 	mmSPI_RESOURCE_RESERVE_EN_CU_0, 0xffffffff, 0x00FF7FBF,
400 	mmSPI_RESOURCE_RESERVE_EN_CU_1, 0xffffffff, 0x00FF7FAF,
401 };
402 
403 static const u32 fiji_golden_common_all[] =
404 {
405 	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
406 	mmPA_SC_RASTER_CONFIG, 0xffffffff, 0x3a00161a,
407 	mmPA_SC_RASTER_CONFIG_1, 0xffffffff, 0x0000002e,
408 	mmGB_ADDR_CONFIG, 0xffffffff, 0x22011003,
409 	mmSPI_RESOURCE_RESERVE_CU_0, 0xffffffff, 0x00000800,
410 	mmSPI_RESOURCE_RESERVE_CU_1, 0xffffffff, 0x00000800,
411 	mmSPI_RESOURCE_RESERVE_EN_CU_0, 0xffffffff, 0x00FF7FBF,
412 	mmSPI_RESOURCE_RESERVE_EN_CU_1, 0xffffffff, 0x00FF7FAF,
413 	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
414 	mmSPI_CONFIG_CNTL_1, 0x0000000f, 0x00000009,
415 };
416 
417 static const u32 golden_settings_fiji_a10[] =
418 {
419 	mmCB_HW_CONTROL_3, 0x000001ff, 0x00000040,
420 	mmDB_DEBUG2, 0xf00fffff, 0x00000400,
421 	mmPA_SC_ENHANCE, 0xffffffff, 0x20000001,
422 	mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000,
423 	mmRLC_CGCG_CGLS_CTRL, 0x00000003, 0x0001003c,
424 	mmSQ_RANDOM_WAVE_PRI, 0x001fffff, 0x000006fd,
425 	mmTA_CNTL_AUX, 0x000f000f, 0x000b0000,
426 	mmTCC_CTRL, 0x00100000, 0xf31fff7f,
427 	mmTCC_EXE_DISABLE, 0x00000002, 0x00000002,
428 	mmTCP_ADDR_CONFIG, 0x000003ff, 0x000000ff,
429 	mmVGT_RESET_DEBUG, 0x00000004, 0x00000004,
430 };
431 
432 static const u32 fiji_mgcg_cgcg_init[] =
433 {
434 	mmRLC_CGTT_MGCG_OVERRIDE, 0xffffffff, 0xffffffff,
435 	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
436 	mmCB_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100,
437 	mmCGTT_BCI_CLK_CTRL, 0xffffffff, 0x00000100,
438 	mmCGTT_CP_CLK_CTRL, 0xffffffff, 0x00000100,
439 	mmCGTT_CPC_CLK_CTRL, 0xffffffff, 0x00000100,
440 	mmCGTT_CPF_CLK_CTRL, 0xffffffff, 0x40000100,
441 	mmCGTT_GDS_CLK_CTRL, 0xffffffff, 0x00000100,
442 	mmCGTT_IA_CLK_CTRL, 0xffffffff, 0x06000100,
443 	mmCGTT_PA_CLK_CTRL, 0xffffffff, 0x00000100,
444 	mmCGTT_WD_CLK_CTRL, 0xffffffff, 0x06000100,
445 	mmCGTT_PC_CLK_CTRL, 0xffffffff, 0x00000100,
446 	mmCGTT_RLC_CLK_CTRL, 0xffffffff, 0x00000100,
447 	mmCGTT_SC_CLK_CTRL, 0xffffffff, 0x00000100,
448 	mmCGTT_SPI_CLK_CTRL, 0xffffffff, 0x00000100,
449 	mmCGTT_SQ_CLK_CTRL, 0xffffffff, 0x00000100,
450 	mmCGTT_SQG_CLK_CTRL, 0xffffffff, 0x00000100,
451 	mmCGTT_SX_CLK_CTRL0, 0xffffffff, 0x00000100,
452 	mmCGTT_SX_CLK_CTRL1, 0xffffffff, 0x00000100,
453 	mmCGTT_SX_CLK_CTRL2, 0xffffffff, 0x00000100,
454 	mmCGTT_SX_CLK_CTRL3, 0xffffffff, 0x00000100,
455 	mmCGTT_SX_CLK_CTRL4, 0xffffffff, 0x00000100,
456 	mmCGTT_TCI_CLK_CTRL, 0xffffffff, 0x00000100,
457 	mmCGTT_TCP_CLK_CTRL, 0xffffffff, 0x00000100,
458 	mmCGTT_VGT_CLK_CTRL, 0xffffffff, 0x06000100,
459 	mmDB_CGTT_CLK_CTRL_0, 0xffffffff, 0x00000100,
460 	mmTA_CGTT_CTRL, 0xffffffff, 0x00000100,
461 	mmTCA_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100,
462 	mmTCC_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100,
463 	mmTD_CGTT_CTRL, 0xffffffff, 0x00000100,
464 	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
465 	mmCGTS_SM_CTRL_REG, 0xffffffff, 0x96e00200,
466 	mmCP_RB_WPTR_POLL_CNTL, 0xffffffff, 0x00900100,
467 	mmRLC_CGCG_CGLS_CTRL, 0xffffffff, 0x0020003c,
468 	mmCP_MEM_SLP_CNTL, 0x00000001, 0x00000001,
469 };
470 
471 static const u32 golden_settings_iceland_a11[] =
472 {
473 	mmCB_HW_CONTROL_3, 0x00000040, 0x00000040,
474 	mmDB_DEBUG2, 0xf00fffff, 0x00000400,
475 	mmDB_DEBUG3, 0xc0000000, 0xc0000000,
476 	mmGB_GPU_ID, 0x0000000f, 0x00000000,
477 	mmPA_SC_ENHANCE, 0xffffffff, 0x20000001,
478 	mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000,
479 	mmPA_SC_RASTER_CONFIG, 0x3f3fffff, 0x00000002,
480 	mmPA_SC_RASTER_CONFIG_1, 0x0000003f, 0x00000000,
481 	mmRLC_CGCG_CGLS_CTRL, 0x00000003, 0x0000003c,
482 	mmSQ_RANDOM_WAVE_PRI, 0x001fffff, 0x000006fd,
483 	mmTA_CNTL_AUX, 0x000f000f, 0x000b0000,
484 	mmTCC_CTRL, 0x00100000, 0xf31fff7f,
485 	mmTCC_EXE_DISABLE, 0x00000002, 0x00000002,
486 	mmTCP_ADDR_CONFIG, 0x000003ff, 0x000000f1,
487 	mmTCP_CHAN_STEER_HI, 0xffffffff, 0x00000000,
488 	mmTCP_CHAN_STEER_LO, 0xffffffff, 0x00000010,
489 };
490 
491 static const u32 iceland_golden_common_all[] =
492 {
493 	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
494 	mmPA_SC_RASTER_CONFIG, 0xffffffff, 0x00000002,
495 	mmPA_SC_RASTER_CONFIG_1, 0xffffffff, 0x00000000,
496 	mmGB_ADDR_CONFIG, 0xffffffff, 0x22010001,
497 	mmSPI_RESOURCE_RESERVE_CU_0, 0xffffffff, 0x00000800,
498 	mmSPI_RESOURCE_RESERVE_CU_1, 0xffffffff, 0x00000800,
499 	mmSPI_RESOURCE_RESERVE_EN_CU_0, 0xffffffff, 0x00FF7FBF,
500 	mmSPI_RESOURCE_RESERVE_EN_CU_1, 0xffffffff, 0x00FF7FAF
501 };
502 
503 static const u32 iceland_mgcg_cgcg_init[] =
504 {
505 	mmRLC_CGTT_MGCG_OVERRIDE, 0xffffffff, 0xffffffff,
506 	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
507 	mmCB_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100,
508 	mmCGTT_BCI_CLK_CTRL, 0xffffffff, 0x00000100,
509 	mmCGTT_CP_CLK_CTRL, 0xffffffff, 0xc0000100,
510 	mmCGTT_CPC_CLK_CTRL, 0xffffffff, 0xc0000100,
511 	mmCGTT_CPF_CLK_CTRL, 0xffffffff, 0xc0000100,
512 	mmCGTT_GDS_CLK_CTRL, 0xffffffff, 0x00000100,
513 	mmCGTT_IA_CLK_CTRL, 0xffffffff, 0x06000100,
514 	mmCGTT_PA_CLK_CTRL, 0xffffffff, 0x00000100,
515 	mmCGTT_WD_CLK_CTRL, 0xffffffff, 0x06000100,
516 	mmCGTT_PC_CLK_CTRL, 0xffffffff, 0x00000100,
517 	mmCGTT_RLC_CLK_CTRL, 0xffffffff, 0x00000100,
518 	mmCGTT_SC_CLK_CTRL, 0xffffffff, 0x00000100,
519 	mmCGTT_SPI_CLK_CTRL, 0xffffffff, 0x00000100,
520 	mmCGTT_SQ_CLK_CTRL, 0xffffffff, 0x00000100,
521 	mmCGTT_SQG_CLK_CTRL, 0xffffffff, 0x00000100,
522 	mmCGTT_SX_CLK_CTRL0, 0xffffffff, 0x00000100,
523 	mmCGTT_SX_CLK_CTRL1, 0xffffffff, 0x00000100,
524 	mmCGTT_SX_CLK_CTRL2, 0xffffffff, 0x00000100,
525 	mmCGTT_SX_CLK_CTRL3, 0xffffffff, 0x00000100,
526 	mmCGTT_SX_CLK_CTRL4, 0xffffffff, 0x00000100,
527 	mmCGTT_TCI_CLK_CTRL, 0xffffffff, 0xff000100,
528 	mmCGTT_TCP_CLK_CTRL, 0xffffffff, 0x00000100,
529 	mmCGTT_VGT_CLK_CTRL, 0xffffffff, 0x06000100,
530 	mmDB_CGTT_CLK_CTRL_0, 0xffffffff, 0x00000100,
531 	mmTA_CGTT_CTRL, 0xffffffff, 0x00000100,
532 	mmTCA_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100,
533 	mmTCC_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100,
534 	mmTD_CGTT_CTRL, 0xffffffff, 0x00000100,
535 	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
536 	mmCGTS_CU0_SP0_CTRL_REG, 0xffffffff, 0x00010000,
537 	mmCGTS_CU0_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
538 	mmCGTS_CU0_TA_SQC_CTRL_REG, 0xffffffff, 0x0f840f87,
539 	mmCGTS_CU0_SP1_CTRL_REG, 0xffffffff, 0x00060005,
540 	mmCGTS_CU0_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
541 	mmCGTS_CU1_SP0_CTRL_REG, 0xffffffff, 0x00010000,
542 	mmCGTS_CU1_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
543 	mmCGTS_CU1_TA_CTRL_REG, 0xffffffff, 0x00040007,
544 	mmCGTS_CU1_SP1_CTRL_REG, 0xffffffff, 0x00060005,
545 	mmCGTS_CU1_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
546 	mmCGTS_CU2_SP0_CTRL_REG, 0xffffffff, 0x00010000,
547 	mmCGTS_CU2_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
548 	mmCGTS_CU2_TA_CTRL_REG, 0xffffffff, 0x00040007,
549 	mmCGTS_CU2_SP1_CTRL_REG, 0xffffffff, 0x00060005,
550 	mmCGTS_CU2_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
551 	mmCGTS_CU3_SP0_CTRL_REG, 0xffffffff, 0x00010000,
552 	mmCGTS_CU3_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
553 	mmCGTS_CU3_TA_CTRL_REG, 0xffffffff, 0x00040007,
554 	mmCGTS_CU3_SP1_CTRL_REG, 0xffffffff, 0x00060005,
555 	mmCGTS_CU3_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
556 	mmCGTS_CU4_SP0_CTRL_REG, 0xffffffff, 0x00010000,
557 	mmCGTS_CU4_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
558 	mmCGTS_CU4_TA_SQC_CTRL_REG, 0xffffffff, 0x0f840f87,
559 	mmCGTS_CU4_SP1_CTRL_REG, 0xffffffff, 0x00060005,
560 	mmCGTS_CU4_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
561 	mmCGTS_CU5_SP0_CTRL_REG, 0xffffffff, 0x00010000,
562 	mmCGTS_CU5_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
563 	mmCGTS_CU5_TA_CTRL_REG, 0xffffffff, 0x00040007,
564 	mmCGTS_CU5_SP1_CTRL_REG, 0xffffffff, 0x00060005,
565 	mmCGTS_CU5_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
566 	mmCGTS_SM_CTRL_REG, 0xffffffff, 0x96e00200,
567 	mmCP_RB_WPTR_POLL_CNTL, 0xffffffff, 0x00900100,
568 	mmRLC_CGCG_CGLS_CTRL, 0xffffffff, 0x0020003c,
569 };
570 
571 static const u32 cz_golden_settings_a11[] =
572 {
573 	mmCB_HW_CONTROL_3, 0x00000040, 0x00000040,
574 	mmDB_DEBUG2, 0xf00fffff, 0x00000400,
575 	mmGB_GPU_ID, 0x0000000f, 0x00000000,
576 	mmPA_SC_ENHANCE, 0xffffffff, 0x00000001,
577 	mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000,
578 	mmRLC_CGCG_CGLS_CTRL, 0x00000003, 0x0000003c,
579 	mmSQ_RANDOM_WAVE_PRI, 0x001fffff, 0x000006fd,
580 	mmTA_CNTL_AUX, 0x000f000f, 0x00010000,
581 	mmTCC_CTRL, 0x00100000, 0xf31fff7f,
582 	mmTCC_EXE_DISABLE, 0x00000002, 0x00000002,
583 	mmTCP_ADDR_CONFIG, 0x0000000f, 0x000000f3,
584 	mmTCP_CHAN_STEER_LO, 0xffffffff, 0x00001302
585 };
586 
587 static const u32 cz_golden_common_all[] =
588 {
589 	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
590 	mmPA_SC_RASTER_CONFIG, 0xffffffff, 0x00000002,
591 	mmPA_SC_RASTER_CONFIG_1, 0xffffffff, 0x00000000,
592 	mmGB_ADDR_CONFIG, 0xffffffff, 0x22010001,
593 	mmSPI_RESOURCE_RESERVE_CU_0, 0xffffffff, 0x00000800,
594 	mmSPI_RESOURCE_RESERVE_CU_1, 0xffffffff, 0x00000800,
595 	mmSPI_RESOURCE_RESERVE_EN_CU_0, 0xffffffff, 0x00FF7FBF,
596 	mmSPI_RESOURCE_RESERVE_EN_CU_1, 0xffffffff, 0x00FF7FAF
597 };
598 
599 static const u32 cz_mgcg_cgcg_init[] =
600 {
601 	mmRLC_CGTT_MGCG_OVERRIDE, 0xffffffff, 0xffffffff,
602 	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
603 	mmCB_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100,
604 	mmCGTT_BCI_CLK_CTRL, 0xffffffff, 0x00000100,
605 	mmCGTT_CP_CLK_CTRL, 0xffffffff, 0x00000100,
606 	mmCGTT_CPC_CLK_CTRL, 0xffffffff, 0x00000100,
607 	mmCGTT_CPF_CLK_CTRL, 0xffffffff, 0x00000100,
608 	mmCGTT_GDS_CLK_CTRL, 0xffffffff, 0x00000100,
609 	mmCGTT_IA_CLK_CTRL, 0xffffffff, 0x06000100,
610 	mmCGTT_PA_CLK_CTRL, 0xffffffff, 0x00000100,
611 	mmCGTT_WD_CLK_CTRL, 0xffffffff, 0x06000100,
612 	mmCGTT_PC_CLK_CTRL, 0xffffffff, 0x00000100,
613 	mmCGTT_RLC_CLK_CTRL, 0xffffffff, 0x00000100,
614 	mmCGTT_SC_CLK_CTRL, 0xffffffff, 0x00000100,
615 	mmCGTT_SPI_CLK_CTRL, 0xffffffff, 0x00000100,
616 	mmCGTT_SQ_CLK_CTRL, 0xffffffff, 0x00000100,
617 	mmCGTT_SQG_CLK_CTRL, 0xffffffff, 0x00000100,
618 	mmCGTT_SX_CLK_CTRL0, 0xffffffff, 0x00000100,
619 	mmCGTT_SX_CLK_CTRL1, 0xffffffff, 0x00000100,
620 	mmCGTT_SX_CLK_CTRL2, 0xffffffff, 0x00000100,
621 	mmCGTT_SX_CLK_CTRL3, 0xffffffff, 0x00000100,
622 	mmCGTT_SX_CLK_CTRL4, 0xffffffff, 0x00000100,
623 	mmCGTT_TCI_CLK_CTRL, 0xffffffff, 0x00000100,
624 	mmCGTT_TCP_CLK_CTRL, 0xffffffff, 0x00000100,
625 	mmCGTT_VGT_CLK_CTRL, 0xffffffff, 0x06000100,
626 	mmDB_CGTT_CLK_CTRL_0, 0xffffffff, 0x00000100,
627 	mmTA_CGTT_CTRL, 0xffffffff, 0x00000100,
628 	mmTCA_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100,
629 	mmTCC_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100,
630 	mmTD_CGTT_CTRL, 0xffffffff, 0x00000100,
631 	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
632 	mmCGTS_CU0_SP0_CTRL_REG, 0xffffffff, 0x00010000,
633 	mmCGTS_CU0_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
634 	mmCGTS_CU0_TA_SQC_CTRL_REG, 0xffffffff, 0x00040007,
635 	mmCGTS_CU0_SP1_CTRL_REG, 0xffffffff, 0x00060005,
636 	mmCGTS_CU0_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
637 	mmCGTS_CU1_SP0_CTRL_REG, 0xffffffff, 0x00010000,
638 	mmCGTS_CU1_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
639 	mmCGTS_CU1_TA_CTRL_REG, 0xffffffff, 0x00040007,
640 	mmCGTS_CU1_SP1_CTRL_REG, 0xffffffff, 0x00060005,
641 	mmCGTS_CU1_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
642 	mmCGTS_CU2_SP0_CTRL_REG, 0xffffffff, 0x00010000,
643 	mmCGTS_CU2_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
644 	mmCGTS_CU2_TA_CTRL_REG, 0xffffffff, 0x00040007,
645 	mmCGTS_CU2_SP1_CTRL_REG, 0xffffffff, 0x00060005,
646 	mmCGTS_CU2_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
647 	mmCGTS_CU3_SP0_CTRL_REG, 0xffffffff, 0x00010000,
648 	mmCGTS_CU3_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
649 	mmCGTS_CU3_TA_CTRL_REG, 0xffffffff, 0x00040007,
650 	mmCGTS_CU3_SP1_CTRL_REG, 0xffffffff, 0x00060005,
651 	mmCGTS_CU3_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
652 	mmCGTS_CU4_SP0_CTRL_REG, 0xffffffff, 0x00010000,
653 	mmCGTS_CU4_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
654 	mmCGTS_CU4_TA_SQC_CTRL_REG, 0xffffffff, 0x00040007,
655 	mmCGTS_CU4_SP1_CTRL_REG, 0xffffffff, 0x00060005,
656 	mmCGTS_CU4_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
657 	mmCGTS_CU5_SP0_CTRL_REG, 0xffffffff, 0x00010000,
658 	mmCGTS_CU5_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
659 	mmCGTS_CU5_TA_CTRL_REG, 0xffffffff, 0x00040007,
660 	mmCGTS_CU5_SP1_CTRL_REG, 0xffffffff, 0x00060005,
661 	mmCGTS_CU5_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
662 	mmCGTS_CU6_SP0_CTRL_REG, 0xffffffff, 0x00010000,
663 	mmCGTS_CU6_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
664 	mmCGTS_CU6_TA_CTRL_REG, 0xffffffff, 0x00040007,
665 	mmCGTS_CU6_SP1_CTRL_REG, 0xffffffff, 0x00060005,
666 	mmCGTS_CU6_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
667 	mmCGTS_CU7_SP0_CTRL_REG, 0xffffffff, 0x00010000,
668 	mmCGTS_CU7_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
669 	mmCGTS_CU7_TA_CTRL_REG, 0xffffffff, 0x00040007,
670 	mmCGTS_CU7_SP1_CTRL_REG, 0xffffffff, 0x00060005,
671 	mmCGTS_CU7_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
672 	mmCGTS_SM_CTRL_REG, 0xffffffff, 0x96e00200,
673 	mmCP_RB_WPTR_POLL_CNTL, 0xffffffff, 0x00900100,
674 	mmRLC_CGCG_CGLS_CTRL, 0xffffffff, 0x0020003f,
675 	mmCP_MEM_SLP_CNTL, 0x00000001, 0x00000001,
676 };
677 
678 static const u32 stoney_golden_settings_a11[] =
679 {
680 	mmDB_DEBUG2, 0xf00fffff, 0x00000400,
681 	mmGB_GPU_ID, 0x0000000f, 0x00000000,
682 	mmPA_SC_ENHANCE, 0xffffffff, 0x20000001,
683 	mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000,
684 	mmRLC_CGCG_CGLS_CTRL, 0x00000003, 0x0001003c,
685 	mmTA_CNTL_AUX, 0x000f000f, 0x000b0000,
686 	mmTCC_CTRL, 0x00100000, 0xf31fff7f,
687 	mmTCC_EXE_DISABLE, 0x00000002, 0x00000002,
688 	mmTCP_ADDR_CONFIG, 0x0000000f, 0x000000f1,
689 	mmTCP_CHAN_STEER_LO, 0xffffffff, 0x10101010,
690 };
691 
692 static const u32 stoney_golden_common_all[] =
693 {
694 	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
695 	mmPA_SC_RASTER_CONFIG, 0xffffffff, 0x00000000,
696 	mmPA_SC_RASTER_CONFIG_1, 0xffffffff, 0x00000000,
697 	mmGB_ADDR_CONFIG, 0xffffffff, 0x12010001,
698 	mmSPI_RESOURCE_RESERVE_CU_0, 0xffffffff, 0x00000800,
699 	mmSPI_RESOURCE_RESERVE_CU_1, 0xffffffff, 0x00000800,
700 	mmSPI_RESOURCE_RESERVE_EN_CU_0, 0xffffffff, 0x00FF7FBF,
701 	mmSPI_RESOURCE_RESERVE_EN_CU_1, 0xffffffff, 0x00FF7FAF,
702 };
703 
704 static const u32 stoney_mgcg_cgcg_init[] =
705 {
706 	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
707 	mmRLC_CGCG_CGLS_CTRL, 0xffffffff, 0x0020003f,
708 	mmCP_MEM_SLP_CNTL, 0xffffffff, 0x00020201,
709 	mmRLC_MEM_SLP_CNTL, 0xffffffff, 0x00020201,
710 	mmCGTS_SM_CTRL_REG, 0xffffffff, 0x96940200,
711 };
712 
713 
714 static const char * const sq_edc_source_names[] = {
715 	"SQ_EDC_INFO_SOURCE_INVALID: No EDC error has occurred",
716 	"SQ_EDC_INFO_SOURCE_INST: EDC source is Instruction Fetch",
717 	"SQ_EDC_INFO_SOURCE_SGPR: EDC source is SGPR or SQC data return",
718 	"SQ_EDC_INFO_SOURCE_VGPR: EDC source is VGPR",
719 	"SQ_EDC_INFO_SOURCE_LDS: EDC source is LDS",
720 	"SQ_EDC_INFO_SOURCE_GDS: EDC source is GDS",
721 	"SQ_EDC_INFO_SOURCE_TA: EDC source is TA",
722 };
723 
724 static void gfx_v8_0_set_ring_funcs(struct amdgpu_device *adev);
725 static void gfx_v8_0_set_irq_funcs(struct amdgpu_device *adev);
726 static void gfx_v8_0_set_gds_init(struct amdgpu_device *adev);
727 static void gfx_v8_0_set_rlc_funcs(struct amdgpu_device *adev);
728 static u32 gfx_v8_0_get_csb_size(struct amdgpu_device *adev);
729 static void gfx_v8_0_get_cu_info(struct amdgpu_device *adev);
730 static void gfx_v8_0_ring_emit_ce_meta(struct amdgpu_ring *ring);
731 static void gfx_v8_0_ring_emit_de_meta(struct amdgpu_ring *ring);
732 
733 #define CG_ACLK_CNTL__ACLK_DIVIDER_MASK                    0x0000007fL
734 #define CG_ACLK_CNTL__ACLK_DIVIDER__SHIFT                  0x00000000L
735 
736 static void gfx_v8_0_init_golden_registers(struct amdgpu_device *adev)
737 {
738 	uint32_t data;
739 
740 	switch (adev->asic_type) {
741 	case CHIP_TOPAZ:
742 		amdgpu_device_program_register_sequence(adev,
743 							iceland_mgcg_cgcg_init,
744 							ARRAY_SIZE(iceland_mgcg_cgcg_init));
745 		amdgpu_device_program_register_sequence(adev,
746 							golden_settings_iceland_a11,
747 							ARRAY_SIZE(golden_settings_iceland_a11));
748 		amdgpu_device_program_register_sequence(adev,
749 							iceland_golden_common_all,
750 							ARRAY_SIZE(iceland_golden_common_all));
751 		break;
752 	case CHIP_FIJI:
753 		amdgpu_device_program_register_sequence(adev,
754 							fiji_mgcg_cgcg_init,
755 							ARRAY_SIZE(fiji_mgcg_cgcg_init));
756 		amdgpu_device_program_register_sequence(adev,
757 							golden_settings_fiji_a10,
758 							ARRAY_SIZE(golden_settings_fiji_a10));
759 		amdgpu_device_program_register_sequence(adev,
760 							fiji_golden_common_all,
761 							ARRAY_SIZE(fiji_golden_common_all));
762 		break;
763 
764 	case CHIP_TONGA:
765 		amdgpu_device_program_register_sequence(adev,
766 							tonga_mgcg_cgcg_init,
767 							ARRAY_SIZE(tonga_mgcg_cgcg_init));
768 		amdgpu_device_program_register_sequence(adev,
769 							golden_settings_tonga_a11,
770 							ARRAY_SIZE(golden_settings_tonga_a11));
771 		amdgpu_device_program_register_sequence(adev,
772 							tonga_golden_common_all,
773 							ARRAY_SIZE(tonga_golden_common_all));
774 		break;
775 	case CHIP_VEGAM:
776 		amdgpu_device_program_register_sequence(adev,
777 							golden_settings_vegam_a11,
778 							ARRAY_SIZE(golden_settings_vegam_a11));
779 		amdgpu_device_program_register_sequence(adev,
780 							vegam_golden_common_all,
781 							ARRAY_SIZE(vegam_golden_common_all));
782 		break;
783 	case CHIP_POLARIS11:
784 	case CHIP_POLARIS12:
785 		amdgpu_device_program_register_sequence(adev,
786 							golden_settings_polaris11_a11,
787 							ARRAY_SIZE(golden_settings_polaris11_a11));
788 		amdgpu_device_program_register_sequence(adev,
789 							polaris11_golden_common_all,
790 							ARRAY_SIZE(polaris11_golden_common_all));
791 		break;
792 	case CHIP_POLARIS10:
793 		amdgpu_device_program_register_sequence(adev,
794 							golden_settings_polaris10_a11,
795 							ARRAY_SIZE(golden_settings_polaris10_a11));
796 		amdgpu_device_program_register_sequence(adev,
797 							polaris10_golden_common_all,
798 							ARRAY_SIZE(polaris10_golden_common_all));
799 		data = RREG32_SMC(ixCG_ACLK_CNTL);
800 		data &= ~CG_ACLK_CNTL__ACLK_DIVIDER_MASK;
801 		data |= 0x18 << CG_ACLK_CNTL__ACLK_DIVIDER__SHIFT;
802 		WREG32_SMC(ixCG_ACLK_CNTL, data);
803 		if ((adev->pdev->device == 0x67DF) && (adev->pdev->revision == 0xc7) &&
804 		    ((adev->pdev->subsystem_device == 0xb37 && adev->pdev->subsystem_vendor == 0x1002) ||
805 		     (adev->pdev->subsystem_device == 0x4a8 && adev->pdev->subsystem_vendor == 0x1043) ||
806 		     (adev->pdev->subsystem_device == 0x9480 && adev->pdev->subsystem_vendor == 0x1680))) {
807 			amdgpu_atombios_i2c_channel_trans(adev, 0x10, 0x96, 0x1E, 0xDD);
808 			amdgpu_atombios_i2c_channel_trans(adev, 0x10, 0x96, 0x1F, 0xD0);
809 		}
810 		break;
811 	case CHIP_CARRIZO:
812 		amdgpu_device_program_register_sequence(adev,
813 							cz_mgcg_cgcg_init,
814 							ARRAY_SIZE(cz_mgcg_cgcg_init));
815 		amdgpu_device_program_register_sequence(adev,
816 							cz_golden_settings_a11,
817 							ARRAY_SIZE(cz_golden_settings_a11));
818 		amdgpu_device_program_register_sequence(adev,
819 							cz_golden_common_all,
820 							ARRAY_SIZE(cz_golden_common_all));
821 		break;
822 	case CHIP_STONEY:
823 		amdgpu_device_program_register_sequence(adev,
824 							stoney_mgcg_cgcg_init,
825 							ARRAY_SIZE(stoney_mgcg_cgcg_init));
826 		amdgpu_device_program_register_sequence(adev,
827 							stoney_golden_settings_a11,
828 							ARRAY_SIZE(stoney_golden_settings_a11));
829 		amdgpu_device_program_register_sequence(adev,
830 							stoney_golden_common_all,
831 							ARRAY_SIZE(stoney_golden_common_all));
832 		break;
833 	default:
834 		break;
835 	}
836 }
837 
838 static int gfx_v8_0_ring_test_ring(struct amdgpu_ring *ring)
839 {
840 	struct amdgpu_device *adev = ring->adev;
841 	uint32_t tmp = 0;
842 	unsigned i;
843 	int r;
844 
845 	WREG32(mmSCRATCH_REG0, 0xCAFEDEAD);
846 	r = amdgpu_ring_alloc(ring, 3);
847 	if (r)
848 		return r;
849 
850 	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_UCONFIG_REG, 1));
851 	amdgpu_ring_write(ring, mmSCRATCH_REG0 - PACKET3_SET_UCONFIG_REG_START);
852 	amdgpu_ring_write(ring, 0xDEADBEEF);
853 	amdgpu_ring_commit(ring);
854 
855 	for (i = 0; i < adev->usec_timeout; i++) {
856 		tmp = RREG32(mmSCRATCH_REG0);
857 		if (tmp == 0xDEADBEEF)
858 			break;
859 		udelay(1);
860 	}
861 
862 	if (i >= adev->usec_timeout)
863 		r = -ETIMEDOUT;
864 
865 	return r;
866 }
867 
868 static int gfx_v8_0_ring_test_ib(struct amdgpu_ring *ring, long timeout)
869 {
870 	struct amdgpu_device *adev = ring->adev;
871 	struct amdgpu_ib ib;
872 	struct dma_fence *f = NULL;
873 
874 	unsigned int index;
875 	uint64_t gpu_addr;
876 	uint32_t tmp;
877 	long r;
878 
879 	r = amdgpu_device_wb_get(adev, &index);
880 	if (r)
881 		return r;
882 
883 	gpu_addr = adev->wb.gpu_addr + (index * 4);
884 	adev->wb.wb[index] = cpu_to_le32(0xCAFEDEAD);
885 	memset(&ib, 0, sizeof(ib));
886 	r = amdgpu_ib_get(adev, NULL, 16,
887 					AMDGPU_IB_POOL_DIRECT, &ib);
888 	if (r)
889 		goto err1;
890 
891 	ib.ptr[0] = PACKET3(PACKET3_WRITE_DATA, 3);
892 	ib.ptr[1] = WRITE_DATA_DST_SEL(5) | WR_CONFIRM;
893 	ib.ptr[2] = lower_32_bits(gpu_addr);
894 	ib.ptr[3] = upper_32_bits(gpu_addr);
895 	ib.ptr[4] = 0xDEADBEEF;
896 	ib.length_dw = 5;
897 
898 	r = amdgpu_ib_schedule(ring, 1, &ib, NULL, &f);
899 	if (r)
900 		goto err2;
901 
902 	r = dma_fence_wait_timeout(f, false, timeout);
903 	if (r == 0) {
904 		r = -ETIMEDOUT;
905 		goto err2;
906 	} else if (r < 0) {
907 		goto err2;
908 	}
909 
910 	tmp = adev->wb.wb[index];
911 	if (tmp == 0xDEADBEEF)
912 		r = 0;
913 	else
914 		r = -EINVAL;
915 
916 err2:
917 	amdgpu_ib_free(adev, &ib, NULL);
918 	dma_fence_put(f);
919 err1:
920 	amdgpu_device_wb_free(adev, index);
921 	return r;
922 }
923 
924 
925 static void gfx_v8_0_free_microcode(struct amdgpu_device *adev)
926 {
927 	amdgpu_ucode_release(&adev->gfx.pfp_fw);
928 	amdgpu_ucode_release(&adev->gfx.me_fw);
929 	amdgpu_ucode_release(&adev->gfx.ce_fw);
930 	amdgpu_ucode_release(&adev->gfx.rlc_fw);
931 	amdgpu_ucode_release(&adev->gfx.mec_fw);
932 	if ((adev->asic_type != CHIP_STONEY) &&
933 	    (adev->asic_type != CHIP_TOPAZ))
934 		amdgpu_ucode_release(&adev->gfx.mec2_fw);
935 
936 	kfree(adev->gfx.rlc.register_list_format);
937 }
938 
939 static int gfx_v8_0_init_microcode(struct amdgpu_device *adev)
940 {
941 	const char *chip_name;
942 	char fw_name[30];
943 	int err;
944 	struct amdgpu_firmware_info *info = NULL;
945 	const struct common_firmware_header *header = NULL;
946 	const struct gfx_firmware_header_v1_0 *cp_hdr;
947 	const struct rlc_firmware_header_v2_0 *rlc_hdr;
948 	unsigned int *tmp = NULL, i;
949 
950 	DRM_DEBUG("\n");
951 
952 	switch (adev->asic_type) {
953 	case CHIP_TOPAZ:
954 		chip_name = "topaz";
955 		break;
956 	case CHIP_TONGA:
957 		chip_name = "tonga";
958 		break;
959 	case CHIP_CARRIZO:
960 		chip_name = "carrizo";
961 		break;
962 	case CHIP_FIJI:
963 		chip_name = "fiji";
964 		break;
965 	case CHIP_STONEY:
966 		chip_name = "stoney";
967 		break;
968 	case CHIP_POLARIS10:
969 		chip_name = "polaris10";
970 		break;
971 	case CHIP_POLARIS11:
972 		chip_name = "polaris11";
973 		break;
974 	case CHIP_POLARIS12:
975 		chip_name = "polaris12";
976 		break;
977 	case CHIP_VEGAM:
978 		chip_name = "vegam";
979 		break;
980 	default:
981 		BUG();
982 	}
983 
984 	if (adev->asic_type >= CHIP_POLARIS10 && adev->asic_type <= CHIP_POLARIS12) {
985 		snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_pfp_2.bin", chip_name);
986 		err = amdgpu_ucode_request(adev, &adev->gfx.pfp_fw, fw_name);
987 		if (err == -ENODEV) {
988 			snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_pfp.bin", chip_name);
989 			err = amdgpu_ucode_request(adev, &adev->gfx.pfp_fw, fw_name);
990 		}
991 	} else {
992 		snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_pfp.bin", chip_name);
993 		err = amdgpu_ucode_request(adev, &adev->gfx.pfp_fw, fw_name);
994 	}
995 	if (err)
996 		goto out;
997 	cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.pfp_fw->data;
998 	adev->gfx.pfp_fw_version = le32_to_cpu(cp_hdr->header.ucode_version);
999 	adev->gfx.pfp_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version);
1000 
1001 	if (adev->asic_type >= CHIP_POLARIS10 && adev->asic_type <= CHIP_POLARIS12) {
1002 		snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_me_2.bin", chip_name);
1003 		err = amdgpu_ucode_request(adev, &adev->gfx.me_fw, fw_name);
1004 		if (err == -ENODEV) {
1005 			snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_me.bin", chip_name);
1006 			err = amdgpu_ucode_request(adev, &adev->gfx.me_fw, fw_name);
1007 		}
1008 	} else {
1009 		snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_me.bin", chip_name);
1010 		err = amdgpu_ucode_request(adev, &adev->gfx.me_fw, fw_name);
1011 	}
1012 	if (err)
1013 		goto out;
1014 	cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.me_fw->data;
1015 	adev->gfx.me_fw_version = le32_to_cpu(cp_hdr->header.ucode_version);
1016 
1017 	adev->gfx.me_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version);
1018 
1019 	if (adev->asic_type >= CHIP_POLARIS10 && adev->asic_type <= CHIP_POLARIS12) {
1020 		snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_ce_2.bin", chip_name);
1021 		err = amdgpu_ucode_request(adev, &adev->gfx.ce_fw, fw_name);
1022 		if (err == -ENODEV) {
1023 			snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_ce.bin", chip_name);
1024 			err = amdgpu_ucode_request(adev, &adev->gfx.ce_fw, fw_name);
1025 		}
1026 	} else {
1027 		snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_ce.bin", chip_name);
1028 		err = amdgpu_ucode_request(adev, &adev->gfx.ce_fw, fw_name);
1029 	}
1030 	if (err)
1031 		goto out;
1032 	cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.ce_fw->data;
1033 	adev->gfx.ce_fw_version = le32_to_cpu(cp_hdr->header.ucode_version);
1034 	adev->gfx.ce_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version);
1035 
1036 	/*
1037 	 * Support for MCBP/Virtualization in combination with chained IBs is
1038 	 * formal released on feature version #46
1039 	 */
1040 	if (adev->gfx.ce_feature_version >= 46 &&
1041 	    adev->gfx.pfp_feature_version >= 46) {
1042 		adev->virt.chained_ib_support = true;
1043 		DRM_INFO("Chained IB support enabled!\n");
1044 	} else
1045 		adev->virt.chained_ib_support = false;
1046 
1047 	snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_rlc.bin", chip_name);
1048 	err = amdgpu_ucode_request(adev, &adev->gfx.rlc_fw, fw_name);
1049 	if (err)
1050 		goto out;
1051 	rlc_hdr = (const struct rlc_firmware_header_v2_0 *)adev->gfx.rlc_fw->data;
1052 	adev->gfx.rlc_fw_version = le32_to_cpu(rlc_hdr->header.ucode_version);
1053 	adev->gfx.rlc_feature_version = le32_to_cpu(rlc_hdr->ucode_feature_version);
1054 
1055 	adev->gfx.rlc.save_and_restore_offset =
1056 			le32_to_cpu(rlc_hdr->save_and_restore_offset);
1057 	adev->gfx.rlc.clear_state_descriptor_offset =
1058 			le32_to_cpu(rlc_hdr->clear_state_descriptor_offset);
1059 	adev->gfx.rlc.avail_scratch_ram_locations =
1060 			le32_to_cpu(rlc_hdr->avail_scratch_ram_locations);
1061 	adev->gfx.rlc.reg_restore_list_size =
1062 			le32_to_cpu(rlc_hdr->reg_restore_list_size);
1063 	adev->gfx.rlc.reg_list_format_start =
1064 			le32_to_cpu(rlc_hdr->reg_list_format_start);
1065 	adev->gfx.rlc.reg_list_format_separate_start =
1066 			le32_to_cpu(rlc_hdr->reg_list_format_separate_start);
1067 	adev->gfx.rlc.starting_offsets_start =
1068 			le32_to_cpu(rlc_hdr->starting_offsets_start);
1069 	adev->gfx.rlc.reg_list_format_size_bytes =
1070 			le32_to_cpu(rlc_hdr->reg_list_format_size_bytes);
1071 	adev->gfx.rlc.reg_list_size_bytes =
1072 			le32_to_cpu(rlc_hdr->reg_list_size_bytes);
1073 
1074 	adev->gfx.rlc.register_list_format =
1075 			kmalloc(adev->gfx.rlc.reg_list_format_size_bytes +
1076 					adev->gfx.rlc.reg_list_size_bytes, GFP_KERNEL);
1077 
1078 	if (!adev->gfx.rlc.register_list_format) {
1079 		err = -ENOMEM;
1080 		goto out;
1081 	}
1082 
1083 	tmp = (unsigned int *)((uintptr_t)rlc_hdr +
1084 			le32_to_cpu(rlc_hdr->reg_list_format_array_offset_bytes));
1085 	for (i = 0 ; i < (adev->gfx.rlc.reg_list_format_size_bytes >> 2); i++)
1086 		adev->gfx.rlc.register_list_format[i] =	le32_to_cpu(tmp[i]);
1087 
1088 	adev->gfx.rlc.register_restore = adev->gfx.rlc.register_list_format + i;
1089 
1090 	tmp = (unsigned int *)((uintptr_t)rlc_hdr +
1091 			le32_to_cpu(rlc_hdr->reg_list_array_offset_bytes));
1092 	for (i = 0 ; i < (adev->gfx.rlc.reg_list_size_bytes >> 2); i++)
1093 		adev->gfx.rlc.register_restore[i] = le32_to_cpu(tmp[i]);
1094 
1095 	if (adev->asic_type >= CHIP_POLARIS10 && adev->asic_type <= CHIP_POLARIS12) {
1096 		snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec_2.bin", chip_name);
1097 		err = amdgpu_ucode_request(adev, &adev->gfx.mec_fw, fw_name);
1098 		if (err == -ENODEV) {
1099 			snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec.bin", chip_name);
1100 			err = amdgpu_ucode_request(adev, &adev->gfx.mec_fw, fw_name);
1101 		}
1102 	} else {
1103 		snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec.bin", chip_name);
1104 		err = amdgpu_ucode_request(adev, &adev->gfx.mec_fw, fw_name);
1105 	}
1106 	if (err)
1107 		goto out;
1108 	cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data;
1109 	adev->gfx.mec_fw_version = le32_to_cpu(cp_hdr->header.ucode_version);
1110 	adev->gfx.mec_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version);
1111 
1112 	if ((adev->asic_type != CHIP_STONEY) &&
1113 	    (adev->asic_type != CHIP_TOPAZ)) {
1114 		if (adev->asic_type >= CHIP_POLARIS10 && adev->asic_type <= CHIP_POLARIS12) {
1115 			snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec2_2.bin", chip_name);
1116 			err = amdgpu_ucode_request(adev, &adev->gfx.mec2_fw, fw_name);
1117 			if (err == -ENODEV) {
1118 				snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec2.bin", chip_name);
1119 				err = amdgpu_ucode_request(adev, &adev->gfx.mec2_fw, fw_name);
1120 			}
1121 		} else {
1122 			snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec2.bin", chip_name);
1123 			err = amdgpu_ucode_request(adev, &adev->gfx.mec2_fw, fw_name);
1124 		}
1125 		if (!err) {
1126 			cp_hdr = (const struct gfx_firmware_header_v1_0 *)
1127 				adev->gfx.mec2_fw->data;
1128 			adev->gfx.mec2_fw_version =
1129 				le32_to_cpu(cp_hdr->header.ucode_version);
1130 			adev->gfx.mec2_feature_version =
1131 				le32_to_cpu(cp_hdr->ucode_feature_version);
1132 		} else {
1133 			err = 0;
1134 			adev->gfx.mec2_fw = NULL;
1135 		}
1136 	}
1137 
1138 	info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_PFP];
1139 	info->ucode_id = AMDGPU_UCODE_ID_CP_PFP;
1140 	info->fw = adev->gfx.pfp_fw;
1141 	header = (const struct common_firmware_header *)info->fw->data;
1142 	adev->firmware.fw_size +=
1143 		ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
1144 
1145 	info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_ME];
1146 	info->ucode_id = AMDGPU_UCODE_ID_CP_ME;
1147 	info->fw = adev->gfx.me_fw;
1148 	header = (const struct common_firmware_header *)info->fw->data;
1149 	adev->firmware.fw_size +=
1150 		ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
1151 
1152 	info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_CE];
1153 	info->ucode_id = AMDGPU_UCODE_ID_CP_CE;
1154 	info->fw = adev->gfx.ce_fw;
1155 	header = (const struct common_firmware_header *)info->fw->data;
1156 	adev->firmware.fw_size +=
1157 		ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
1158 
1159 	info = &adev->firmware.ucode[AMDGPU_UCODE_ID_RLC_G];
1160 	info->ucode_id = AMDGPU_UCODE_ID_RLC_G;
1161 	info->fw = adev->gfx.rlc_fw;
1162 	header = (const struct common_firmware_header *)info->fw->data;
1163 	adev->firmware.fw_size +=
1164 		ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
1165 
1166 	info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_MEC1];
1167 	info->ucode_id = AMDGPU_UCODE_ID_CP_MEC1;
1168 	info->fw = adev->gfx.mec_fw;
1169 	header = (const struct common_firmware_header *)info->fw->data;
1170 	adev->firmware.fw_size +=
1171 		ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
1172 
1173 	/* we need account JT in */
1174 	cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data;
1175 	adev->firmware.fw_size +=
1176 		ALIGN(le32_to_cpu(cp_hdr->jt_size) << 2, PAGE_SIZE);
1177 
1178 	if (amdgpu_sriov_vf(adev)) {
1179 		info = &adev->firmware.ucode[AMDGPU_UCODE_ID_STORAGE];
1180 		info->ucode_id = AMDGPU_UCODE_ID_STORAGE;
1181 		info->fw = adev->gfx.mec_fw;
1182 		adev->firmware.fw_size +=
1183 			ALIGN(le32_to_cpu(64 * PAGE_SIZE), PAGE_SIZE);
1184 	}
1185 
1186 	if (adev->gfx.mec2_fw) {
1187 		info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_MEC2];
1188 		info->ucode_id = AMDGPU_UCODE_ID_CP_MEC2;
1189 		info->fw = adev->gfx.mec2_fw;
1190 		header = (const struct common_firmware_header *)info->fw->data;
1191 		adev->firmware.fw_size +=
1192 			ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
1193 	}
1194 
1195 out:
1196 	if (err) {
1197 		dev_err(adev->dev,
1198 			"gfx8: Failed to load firmware \"%s\"\n",
1199 			fw_name);
1200 		amdgpu_ucode_release(&adev->gfx.pfp_fw);
1201 		amdgpu_ucode_release(&adev->gfx.me_fw);
1202 		amdgpu_ucode_release(&adev->gfx.ce_fw);
1203 		amdgpu_ucode_release(&adev->gfx.rlc_fw);
1204 		amdgpu_ucode_release(&adev->gfx.mec_fw);
1205 		amdgpu_ucode_release(&adev->gfx.mec2_fw);
1206 	}
1207 	return err;
1208 }
1209 
1210 static void gfx_v8_0_get_csb_buffer(struct amdgpu_device *adev,
1211 				    volatile u32 *buffer)
1212 {
1213 	u32 count = 0, i;
1214 	const struct cs_section_def *sect = NULL;
1215 	const struct cs_extent_def *ext = NULL;
1216 
1217 	if (adev->gfx.rlc.cs_data == NULL)
1218 		return;
1219 	if (buffer == NULL)
1220 		return;
1221 
1222 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0));
1223 	buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
1224 
1225 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CONTEXT_CONTROL, 1));
1226 	buffer[count++] = cpu_to_le32(0x80000000);
1227 	buffer[count++] = cpu_to_le32(0x80000000);
1228 
1229 	for (sect = adev->gfx.rlc.cs_data; sect->section != NULL; ++sect) {
1230 		for (ext = sect->section; ext->extent != NULL; ++ext) {
1231 			if (sect->id == SECT_CONTEXT) {
1232 				buffer[count++] =
1233 					cpu_to_le32(PACKET3(PACKET3_SET_CONTEXT_REG, ext->reg_count));
1234 				buffer[count++] = cpu_to_le32(ext->reg_index -
1235 						PACKET3_SET_CONTEXT_REG_START);
1236 				for (i = 0; i < ext->reg_count; i++)
1237 					buffer[count++] = cpu_to_le32(ext->extent[i]);
1238 			} else {
1239 				return;
1240 			}
1241 		}
1242 	}
1243 
1244 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_SET_CONTEXT_REG, 2));
1245 	buffer[count++] = cpu_to_le32(mmPA_SC_RASTER_CONFIG -
1246 			PACKET3_SET_CONTEXT_REG_START);
1247 	buffer[count++] = cpu_to_le32(adev->gfx.config.rb_config[0][0].raster_config);
1248 	buffer[count++] = cpu_to_le32(adev->gfx.config.rb_config[0][0].raster_config_1);
1249 
1250 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0));
1251 	buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_END_CLEAR_STATE);
1252 
1253 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CLEAR_STATE, 0));
1254 	buffer[count++] = cpu_to_le32(0);
1255 }
1256 
1257 static int gfx_v8_0_cp_jump_table_num(struct amdgpu_device *adev)
1258 {
1259 	if (adev->asic_type == CHIP_CARRIZO)
1260 		return 5;
1261 	else
1262 		return 4;
1263 }
1264 
1265 static int gfx_v8_0_rlc_init(struct amdgpu_device *adev)
1266 {
1267 	const struct cs_section_def *cs_data;
1268 	int r;
1269 
1270 	adev->gfx.rlc.cs_data = vi_cs_data;
1271 
1272 	cs_data = adev->gfx.rlc.cs_data;
1273 
1274 	if (cs_data) {
1275 		/* init clear state block */
1276 		r = amdgpu_gfx_rlc_init_csb(adev);
1277 		if (r)
1278 			return r;
1279 	}
1280 
1281 	if ((adev->asic_type == CHIP_CARRIZO) ||
1282 	    (adev->asic_type == CHIP_STONEY)) {
1283 		adev->gfx.rlc.cp_table_size = ALIGN(96 * 5 * 4, 2048) + (64 * 1024); /* JT + GDS */
1284 		r = amdgpu_gfx_rlc_init_cpt(adev);
1285 		if (r)
1286 			return r;
1287 	}
1288 
1289 	/* init spm vmid with 0xf */
1290 	if (adev->gfx.rlc.funcs->update_spm_vmid)
1291 		adev->gfx.rlc.funcs->update_spm_vmid(adev, 0xf);
1292 
1293 	return 0;
1294 }
1295 
1296 static void gfx_v8_0_mec_fini(struct amdgpu_device *adev)
1297 {
1298 	amdgpu_bo_free_kernel(&adev->gfx.mec.hpd_eop_obj, NULL, NULL);
1299 }
1300 
1301 static int gfx_v8_0_mec_init(struct amdgpu_device *adev)
1302 {
1303 	int r;
1304 	u32 *hpd;
1305 	size_t mec_hpd_size;
1306 
1307 	bitmap_zero(adev->gfx.mec.queue_bitmap, AMDGPU_MAX_COMPUTE_QUEUES);
1308 
1309 	/* take ownership of the relevant compute queues */
1310 	amdgpu_gfx_compute_queue_acquire(adev);
1311 
1312 	mec_hpd_size = adev->gfx.num_compute_rings * GFX8_MEC_HPD_SIZE;
1313 	if (mec_hpd_size) {
1314 		r = amdgpu_bo_create_reserved(adev, mec_hpd_size, PAGE_SIZE,
1315 					      AMDGPU_GEM_DOMAIN_VRAM |
1316 					      AMDGPU_GEM_DOMAIN_GTT,
1317 					      &adev->gfx.mec.hpd_eop_obj,
1318 					      &adev->gfx.mec.hpd_eop_gpu_addr,
1319 					      (void **)&hpd);
1320 		if (r) {
1321 			dev_warn(adev->dev, "(%d) create HDP EOP bo failed\n", r);
1322 			return r;
1323 		}
1324 
1325 		memset(hpd, 0, mec_hpd_size);
1326 
1327 		amdgpu_bo_kunmap(adev->gfx.mec.hpd_eop_obj);
1328 		amdgpu_bo_unreserve(adev->gfx.mec.hpd_eop_obj);
1329 	}
1330 
1331 	return 0;
1332 }
1333 
1334 static const u32 vgpr_init_compute_shader[] =
1335 {
1336 	0x7e000209, 0x7e020208,
1337 	0x7e040207, 0x7e060206,
1338 	0x7e080205, 0x7e0a0204,
1339 	0x7e0c0203, 0x7e0e0202,
1340 	0x7e100201, 0x7e120200,
1341 	0x7e140209, 0x7e160208,
1342 	0x7e180207, 0x7e1a0206,
1343 	0x7e1c0205, 0x7e1e0204,
1344 	0x7e200203, 0x7e220202,
1345 	0x7e240201, 0x7e260200,
1346 	0x7e280209, 0x7e2a0208,
1347 	0x7e2c0207, 0x7e2e0206,
1348 	0x7e300205, 0x7e320204,
1349 	0x7e340203, 0x7e360202,
1350 	0x7e380201, 0x7e3a0200,
1351 	0x7e3c0209, 0x7e3e0208,
1352 	0x7e400207, 0x7e420206,
1353 	0x7e440205, 0x7e460204,
1354 	0x7e480203, 0x7e4a0202,
1355 	0x7e4c0201, 0x7e4e0200,
1356 	0x7e500209, 0x7e520208,
1357 	0x7e540207, 0x7e560206,
1358 	0x7e580205, 0x7e5a0204,
1359 	0x7e5c0203, 0x7e5e0202,
1360 	0x7e600201, 0x7e620200,
1361 	0x7e640209, 0x7e660208,
1362 	0x7e680207, 0x7e6a0206,
1363 	0x7e6c0205, 0x7e6e0204,
1364 	0x7e700203, 0x7e720202,
1365 	0x7e740201, 0x7e760200,
1366 	0x7e780209, 0x7e7a0208,
1367 	0x7e7c0207, 0x7e7e0206,
1368 	0xbf8a0000, 0xbf810000,
1369 };
1370 
1371 static const u32 sgpr_init_compute_shader[] =
1372 {
1373 	0xbe8a0100, 0xbe8c0102,
1374 	0xbe8e0104, 0xbe900106,
1375 	0xbe920108, 0xbe940100,
1376 	0xbe960102, 0xbe980104,
1377 	0xbe9a0106, 0xbe9c0108,
1378 	0xbe9e0100, 0xbea00102,
1379 	0xbea20104, 0xbea40106,
1380 	0xbea60108, 0xbea80100,
1381 	0xbeaa0102, 0xbeac0104,
1382 	0xbeae0106, 0xbeb00108,
1383 	0xbeb20100, 0xbeb40102,
1384 	0xbeb60104, 0xbeb80106,
1385 	0xbeba0108, 0xbebc0100,
1386 	0xbebe0102, 0xbec00104,
1387 	0xbec20106, 0xbec40108,
1388 	0xbec60100, 0xbec80102,
1389 	0xbee60004, 0xbee70005,
1390 	0xbeea0006, 0xbeeb0007,
1391 	0xbee80008, 0xbee90009,
1392 	0xbefc0000, 0xbf8a0000,
1393 	0xbf810000, 0x00000000,
1394 };
1395 
1396 static const u32 vgpr_init_regs[] =
1397 {
1398 	mmCOMPUTE_STATIC_THREAD_MGMT_SE0, 0xffffffff,
1399 	mmCOMPUTE_RESOURCE_LIMITS, 0x1000000, /* CU_GROUP_COUNT=1 */
1400 	mmCOMPUTE_NUM_THREAD_X, 256*4,
1401 	mmCOMPUTE_NUM_THREAD_Y, 1,
1402 	mmCOMPUTE_NUM_THREAD_Z, 1,
1403 	mmCOMPUTE_PGM_RSRC1, 0x100004f, /* VGPRS=15 (64 logical VGPRs), SGPRS=1 (16 SGPRs), BULKY=1 */
1404 	mmCOMPUTE_PGM_RSRC2, 20,
1405 	mmCOMPUTE_USER_DATA_0, 0xedcedc00,
1406 	mmCOMPUTE_USER_DATA_1, 0xedcedc01,
1407 	mmCOMPUTE_USER_DATA_2, 0xedcedc02,
1408 	mmCOMPUTE_USER_DATA_3, 0xedcedc03,
1409 	mmCOMPUTE_USER_DATA_4, 0xedcedc04,
1410 	mmCOMPUTE_USER_DATA_5, 0xedcedc05,
1411 	mmCOMPUTE_USER_DATA_6, 0xedcedc06,
1412 	mmCOMPUTE_USER_DATA_7, 0xedcedc07,
1413 	mmCOMPUTE_USER_DATA_8, 0xedcedc08,
1414 	mmCOMPUTE_USER_DATA_9, 0xedcedc09,
1415 };
1416 
1417 static const u32 sgpr1_init_regs[] =
1418 {
1419 	mmCOMPUTE_STATIC_THREAD_MGMT_SE0, 0x0f,
1420 	mmCOMPUTE_RESOURCE_LIMITS, 0x1000000, /* CU_GROUP_COUNT=1 */
1421 	mmCOMPUTE_NUM_THREAD_X, 256*5,
1422 	mmCOMPUTE_NUM_THREAD_Y, 1,
1423 	mmCOMPUTE_NUM_THREAD_Z, 1,
1424 	mmCOMPUTE_PGM_RSRC1, 0x240, /* SGPRS=9 (80 GPRS) */
1425 	mmCOMPUTE_PGM_RSRC2, 20,
1426 	mmCOMPUTE_USER_DATA_0, 0xedcedc00,
1427 	mmCOMPUTE_USER_DATA_1, 0xedcedc01,
1428 	mmCOMPUTE_USER_DATA_2, 0xedcedc02,
1429 	mmCOMPUTE_USER_DATA_3, 0xedcedc03,
1430 	mmCOMPUTE_USER_DATA_4, 0xedcedc04,
1431 	mmCOMPUTE_USER_DATA_5, 0xedcedc05,
1432 	mmCOMPUTE_USER_DATA_6, 0xedcedc06,
1433 	mmCOMPUTE_USER_DATA_7, 0xedcedc07,
1434 	mmCOMPUTE_USER_DATA_8, 0xedcedc08,
1435 	mmCOMPUTE_USER_DATA_9, 0xedcedc09,
1436 };
1437 
1438 static const u32 sgpr2_init_regs[] =
1439 {
1440 	mmCOMPUTE_STATIC_THREAD_MGMT_SE0, 0xf0,
1441 	mmCOMPUTE_RESOURCE_LIMITS, 0x1000000,
1442 	mmCOMPUTE_NUM_THREAD_X, 256*5,
1443 	mmCOMPUTE_NUM_THREAD_Y, 1,
1444 	mmCOMPUTE_NUM_THREAD_Z, 1,
1445 	mmCOMPUTE_PGM_RSRC1, 0x240, /* SGPRS=9 (80 GPRS) */
1446 	mmCOMPUTE_PGM_RSRC2, 20,
1447 	mmCOMPUTE_USER_DATA_0, 0xedcedc00,
1448 	mmCOMPUTE_USER_DATA_1, 0xedcedc01,
1449 	mmCOMPUTE_USER_DATA_2, 0xedcedc02,
1450 	mmCOMPUTE_USER_DATA_3, 0xedcedc03,
1451 	mmCOMPUTE_USER_DATA_4, 0xedcedc04,
1452 	mmCOMPUTE_USER_DATA_5, 0xedcedc05,
1453 	mmCOMPUTE_USER_DATA_6, 0xedcedc06,
1454 	mmCOMPUTE_USER_DATA_7, 0xedcedc07,
1455 	mmCOMPUTE_USER_DATA_8, 0xedcedc08,
1456 	mmCOMPUTE_USER_DATA_9, 0xedcedc09,
1457 };
1458 
1459 static const u32 sec_ded_counter_registers[] =
1460 {
1461 	mmCPC_EDC_ATC_CNT,
1462 	mmCPC_EDC_SCRATCH_CNT,
1463 	mmCPC_EDC_UCODE_CNT,
1464 	mmCPF_EDC_ATC_CNT,
1465 	mmCPF_EDC_ROQ_CNT,
1466 	mmCPF_EDC_TAG_CNT,
1467 	mmCPG_EDC_ATC_CNT,
1468 	mmCPG_EDC_DMA_CNT,
1469 	mmCPG_EDC_TAG_CNT,
1470 	mmDC_EDC_CSINVOC_CNT,
1471 	mmDC_EDC_RESTORE_CNT,
1472 	mmDC_EDC_STATE_CNT,
1473 	mmGDS_EDC_CNT,
1474 	mmGDS_EDC_GRBM_CNT,
1475 	mmGDS_EDC_OA_DED,
1476 	mmSPI_EDC_CNT,
1477 	mmSQC_ATC_EDC_GATCL1_CNT,
1478 	mmSQC_EDC_CNT,
1479 	mmSQ_EDC_DED_CNT,
1480 	mmSQ_EDC_INFO,
1481 	mmSQ_EDC_SEC_CNT,
1482 	mmTCC_EDC_CNT,
1483 	mmTCP_ATC_EDC_GATCL1_CNT,
1484 	mmTCP_EDC_CNT,
1485 	mmTD_EDC_CNT
1486 };
1487 
1488 static int gfx_v8_0_do_edc_gpr_workarounds(struct amdgpu_device *adev)
1489 {
1490 	struct amdgpu_ring *ring = &adev->gfx.compute_ring[0];
1491 	struct amdgpu_ib ib;
1492 	struct dma_fence *f = NULL;
1493 	int r, i;
1494 	u32 tmp;
1495 	unsigned total_size, vgpr_offset, sgpr_offset;
1496 	u64 gpu_addr;
1497 
1498 	/* only supported on CZ */
1499 	if (adev->asic_type != CHIP_CARRIZO)
1500 		return 0;
1501 
1502 	/* bail if the compute ring is not ready */
1503 	if (!ring->sched.ready)
1504 		return 0;
1505 
1506 	tmp = RREG32(mmGB_EDC_MODE);
1507 	WREG32(mmGB_EDC_MODE, 0);
1508 
1509 	total_size =
1510 		(((ARRAY_SIZE(vgpr_init_regs) / 2) * 3) + 4 + 5 + 2) * 4;
1511 	total_size +=
1512 		(((ARRAY_SIZE(sgpr1_init_regs) / 2) * 3) + 4 + 5 + 2) * 4;
1513 	total_size +=
1514 		(((ARRAY_SIZE(sgpr2_init_regs) / 2) * 3) + 4 + 5 + 2) * 4;
1515 	total_size = ALIGN(total_size, 256);
1516 	vgpr_offset = total_size;
1517 	total_size += ALIGN(sizeof(vgpr_init_compute_shader), 256);
1518 	sgpr_offset = total_size;
1519 	total_size += sizeof(sgpr_init_compute_shader);
1520 
1521 	/* allocate an indirect buffer to put the commands in */
1522 	memset(&ib, 0, sizeof(ib));
1523 	r = amdgpu_ib_get(adev, NULL, total_size,
1524 					AMDGPU_IB_POOL_DIRECT, &ib);
1525 	if (r) {
1526 		DRM_ERROR("amdgpu: failed to get ib (%d).\n", r);
1527 		return r;
1528 	}
1529 
1530 	/* load the compute shaders */
1531 	for (i = 0; i < ARRAY_SIZE(vgpr_init_compute_shader); i++)
1532 		ib.ptr[i + (vgpr_offset / 4)] = vgpr_init_compute_shader[i];
1533 
1534 	for (i = 0; i < ARRAY_SIZE(sgpr_init_compute_shader); i++)
1535 		ib.ptr[i + (sgpr_offset / 4)] = sgpr_init_compute_shader[i];
1536 
1537 	/* init the ib length to 0 */
1538 	ib.length_dw = 0;
1539 
1540 	/* VGPR */
1541 	/* write the register state for the compute dispatch */
1542 	for (i = 0; i < ARRAY_SIZE(vgpr_init_regs); i += 2) {
1543 		ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 1);
1544 		ib.ptr[ib.length_dw++] = vgpr_init_regs[i] - PACKET3_SET_SH_REG_START;
1545 		ib.ptr[ib.length_dw++] = vgpr_init_regs[i + 1];
1546 	}
1547 	/* write the shader start address: mmCOMPUTE_PGM_LO, mmCOMPUTE_PGM_HI */
1548 	gpu_addr = (ib.gpu_addr + (u64)vgpr_offset) >> 8;
1549 	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 2);
1550 	ib.ptr[ib.length_dw++] = mmCOMPUTE_PGM_LO - PACKET3_SET_SH_REG_START;
1551 	ib.ptr[ib.length_dw++] = lower_32_bits(gpu_addr);
1552 	ib.ptr[ib.length_dw++] = upper_32_bits(gpu_addr);
1553 
1554 	/* write dispatch packet */
1555 	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_DISPATCH_DIRECT, 3);
1556 	ib.ptr[ib.length_dw++] = 8; /* x */
1557 	ib.ptr[ib.length_dw++] = 1; /* y */
1558 	ib.ptr[ib.length_dw++] = 1; /* z */
1559 	ib.ptr[ib.length_dw++] =
1560 		REG_SET_FIELD(0, COMPUTE_DISPATCH_INITIATOR, COMPUTE_SHADER_EN, 1);
1561 
1562 	/* write CS partial flush packet */
1563 	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_EVENT_WRITE, 0);
1564 	ib.ptr[ib.length_dw++] = EVENT_TYPE(7) | EVENT_INDEX(4);
1565 
1566 	/* SGPR1 */
1567 	/* write the register state for the compute dispatch */
1568 	for (i = 0; i < ARRAY_SIZE(sgpr1_init_regs); i += 2) {
1569 		ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 1);
1570 		ib.ptr[ib.length_dw++] = sgpr1_init_regs[i] - PACKET3_SET_SH_REG_START;
1571 		ib.ptr[ib.length_dw++] = sgpr1_init_regs[i + 1];
1572 	}
1573 	/* write the shader start address: mmCOMPUTE_PGM_LO, mmCOMPUTE_PGM_HI */
1574 	gpu_addr = (ib.gpu_addr + (u64)sgpr_offset) >> 8;
1575 	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 2);
1576 	ib.ptr[ib.length_dw++] = mmCOMPUTE_PGM_LO - PACKET3_SET_SH_REG_START;
1577 	ib.ptr[ib.length_dw++] = lower_32_bits(gpu_addr);
1578 	ib.ptr[ib.length_dw++] = upper_32_bits(gpu_addr);
1579 
1580 	/* write dispatch packet */
1581 	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_DISPATCH_DIRECT, 3);
1582 	ib.ptr[ib.length_dw++] = 8; /* x */
1583 	ib.ptr[ib.length_dw++] = 1; /* y */
1584 	ib.ptr[ib.length_dw++] = 1; /* z */
1585 	ib.ptr[ib.length_dw++] =
1586 		REG_SET_FIELD(0, COMPUTE_DISPATCH_INITIATOR, COMPUTE_SHADER_EN, 1);
1587 
1588 	/* write CS partial flush packet */
1589 	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_EVENT_WRITE, 0);
1590 	ib.ptr[ib.length_dw++] = EVENT_TYPE(7) | EVENT_INDEX(4);
1591 
1592 	/* SGPR2 */
1593 	/* write the register state for the compute dispatch */
1594 	for (i = 0; i < ARRAY_SIZE(sgpr2_init_regs); i += 2) {
1595 		ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 1);
1596 		ib.ptr[ib.length_dw++] = sgpr2_init_regs[i] - PACKET3_SET_SH_REG_START;
1597 		ib.ptr[ib.length_dw++] = sgpr2_init_regs[i + 1];
1598 	}
1599 	/* write the shader start address: mmCOMPUTE_PGM_LO, mmCOMPUTE_PGM_HI */
1600 	gpu_addr = (ib.gpu_addr + (u64)sgpr_offset) >> 8;
1601 	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 2);
1602 	ib.ptr[ib.length_dw++] = mmCOMPUTE_PGM_LO - PACKET3_SET_SH_REG_START;
1603 	ib.ptr[ib.length_dw++] = lower_32_bits(gpu_addr);
1604 	ib.ptr[ib.length_dw++] = upper_32_bits(gpu_addr);
1605 
1606 	/* write dispatch packet */
1607 	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_DISPATCH_DIRECT, 3);
1608 	ib.ptr[ib.length_dw++] = 8; /* x */
1609 	ib.ptr[ib.length_dw++] = 1; /* y */
1610 	ib.ptr[ib.length_dw++] = 1; /* z */
1611 	ib.ptr[ib.length_dw++] =
1612 		REG_SET_FIELD(0, COMPUTE_DISPATCH_INITIATOR, COMPUTE_SHADER_EN, 1);
1613 
1614 	/* write CS partial flush packet */
1615 	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_EVENT_WRITE, 0);
1616 	ib.ptr[ib.length_dw++] = EVENT_TYPE(7) | EVENT_INDEX(4);
1617 
1618 	/* shedule the ib on the ring */
1619 	r = amdgpu_ib_schedule(ring, 1, &ib, NULL, &f);
1620 	if (r) {
1621 		DRM_ERROR("amdgpu: ib submit failed (%d).\n", r);
1622 		goto fail;
1623 	}
1624 
1625 	/* wait for the GPU to finish processing the IB */
1626 	r = dma_fence_wait(f, false);
1627 	if (r) {
1628 		DRM_ERROR("amdgpu: fence wait failed (%d).\n", r);
1629 		goto fail;
1630 	}
1631 
1632 	tmp = REG_SET_FIELD(tmp, GB_EDC_MODE, DED_MODE, 2);
1633 	tmp = REG_SET_FIELD(tmp, GB_EDC_MODE, PROP_FED, 1);
1634 	WREG32(mmGB_EDC_MODE, tmp);
1635 
1636 	tmp = RREG32(mmCC_GC_EDC_CONFIG);
1637 	tmp = REG_SET_FIELD(tmp, CC_GC_EDC_CONFIG, DIS_EDC, 0) | 1;
1638 	WREG32(mmCC_GC_EDC_CONFIG, tmp);
1639 
1640 
1641 	/* read back registers to clear the counters */
1642 	for (i = 0; i < ARRAY_SIZE(sec_ded_counter_registers); i++)
1643 		RREG32(sec_ded_counter_registers[i]);
1644 
1645 fail:
1646 	amdgpu_ib_free(adev, &ib, NULL);
1647 	dma_fence_put(f);
1648 
1649 	return r;
1650 }
1651 
1652 static int gfx_v8_0_gpu_early_init(struct amdgpu_device *adev)
1653 {
1654 	u32 gb_addr_config;
1655 	u32 mc_arb_ramcfg;
1656 	u32 dimm00_addr_map, dimm01_addr_map, dimm10_addr_map, dimm11_addr_map;
1657 	u32 tmp;
1658 	int ret;
1659 
1660 	switch (adev->asic_type) {
1661 	case CHIP_TOPAZ:
1662 		adev->gfx.config.max_shader_engines = 1;
1663 		adev->gfx.config.max_tile_pipes = 2;
1664 		adev->gfx.config.max_cu_per_sh = 6;
1665 		adev->gfx.config.max_sh_per_se = 1;
1666 		adev->gfx.config.max_backends_per_se = 2;
1667 		adev->gfx.config.max_texture_channel_caches = 2;
1668 		adev->gfx.config.max_gprs = 256;
1669 		adev->gfx.config.max_gs_threads = 32;
1670 		adev->gfx.config.max_hw_contexts = 8;
1671 
1672 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1673 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1674 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1675 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
1676 		gb_addr_config = TOPAZ_GB_ADDR_CONFIG_GOLDEN;
1677 		break;
1678 	case CHIP_FIJI:
1679 		adev->gfx.config.max_shader_engines = 4;
1680 		adev->gfx.config.max_tile_pipes = 16;
1681 		adev->gfx.config.max_cu_per_sh = 16;
1682 		adev->gfx.config.max_sh_per_se = 1;
1683 		adev->gfx.config.max_backends_per_se = 4;
1684 		adev->gfx.config.max_texture_channel_caches = 16;
1685 		adev->gfx.config.max_gprs = 256;
1686 		adev->gfx.config.max_gs_threads = 32;
1687 		adev->gfx.config.max_hw_contexts = 8;
1688 
1689 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1690 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1691 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1692 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
1693 		gb_addr_config = TONGA_GB_ADDR_CONFIG_GOLDEN;
1694 		break;
1695 	case CHIP_POLARIS11:
1696 	case CHIP_POLARIS12:
1697 		ret = amdgpu_atombios_get_gfx_info(adev);
1698 		if (ret)
1699 			return ret;
1700 		adev->gfx.config.max_gprs = 256;
1701 		adev->gfx.config.max_gs_threads = 32;
1702 		adev->gfx.config.max_hw_contexts = 8;
1703 
1704 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1705 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1706 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1707 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
1708 		gb_addr_config = POLARIS11_GB_ADDR_CONFIG_GOLDEN;
1709 		break;
1710 	case CHIP_POLARIS10:
1711 	case CHIP_VEGAM:
1712 		ret = amdgpu_atombios_get_gfx_info(adev);
1713 		if (ret)
1714 			return ret;
1715 		adev->gfx.config.max_gprs = 256;
1716 		adev->gfx.config.max_gs_threads = 32;
1717 		adev->gfx.config.max_hw_contexts = 8;
1718 
1719 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1720 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1721 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1722 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
1723 		gb_addr_config = TONGA_GB_ADDR_CONFIG_GOLDEN;
1724 		break;
1725 	case CHIP_TONGA:
1726 		adev->gfx.config.max_shader_engines = 4;
1727 		adev->gfx.config.max_tile_pipes = 8;
1728 		adev->gfx.config.max_cu_per_sh = 8;
1729 		adev->gfx.config.max_sh_per_se = 1;
1730 		adev->gfx.config.max_backends_per_se = 2;
1731 		adev->gfx.config.max_texture_channel_caches = 8;
1732 		adev->gfx.config.max_gprs = 256;
1733 		adev->gfx.config.max_gs_threads = 32;
1734 		adev->gfx.config.max_hw_contexts = 8;
1735 
1736 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1737 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1738 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1739 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
1740 		gb_addr_config = TONGA_GB_ADDR_CONFIG_GOLDEN;
1741 		break;
1742 	case CHIP_CARRIZO:
1743 		adev->gfx.config.max_shader_engines = 1;
1744 		adev->gfx.config.max_tile_pipes = 2;
1745 		adev->gfx.config.max_sh_per_se = 1;
1746 		adev->gfx.config.max_backends_per_se = 2;
1747 		adev->gfx.config.max_cu_per_sh = 8;
1748 		adev->gfx.config.max_texture_channel_caches = 2;
1749 		adev->gfx.config.max_gprs = 256;
1750 		adev->gfx.config.max_gs_threads = 32;
1751 		adev->gfx.config.max_hw_contexts = 8;
1752 
1753 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1754 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1755 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1756 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
1757 		gb_addr_config = CARRIZO_GB_ADDR_CONFIG_GOLDEN;
1758 		break;
1759 	case CHIP_STONEY:
1760 		adev->gfx.config.max_shader_engines = 1;
1761 		adev->gfx.config.max_tile_pipes = 2;
1762 		adev->gfx.config.max_sh_per_se = 1;
1763 		adev->gfx.config.max_backends_per_se = 1;
1764 		adev->gfx.config.max_cu_per_sh = 3;
1765 		adev->gfx.config.max_texture_channel_caches = 2;
1766 		adev->gfx.config.max_gprs = 256;
1767 		adev->gfx.config.max_gs_threads = 16;
1768 		adev->gfx.config.max_hw_contexts = 8;
1769 
1770 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1771 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1772 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1773 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
1774 		gb_addr_config = CARRIZO_GB_ADDR_CONFIG_GOLDEN;
1775 		break;
1776 	default:
1777 		adev->gfx.config.max_shader_engines = 2;
1778 		adev->gfx.config.max_tile_pipes = 4;
1779 		adev->gfx.config.max_cu_per_sh = 2;
1780 		adev->gfx.config.max_sh_per_se = 1;
1781 		adev->gfx.config.max_backends_per_se = 2;
1782 		adev->gfx.config.max_texture_channel_caches = 4;
1783 		adev->gfx.config.max_gprs = 256;
1784 		adev->gfx.config.max_gs_threads = 32;
1785 		adev->gfx.config.max_hw_contexts = 8;
1786 
1787 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1788 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1789 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1790 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
1791 		gb_addr_config = TONGA_GB_ADDR_CONFIG_GOLDEN;
1792 		break;
1793 	}
1794 
1795 	adev->gfx.config.mc_arb_ramcfg = RREG32(mmMC_ARB_RAMCFG);
1796 	mc_arb_ramcfg = adev->gfx.config.mc_arb_ramcfg;
1797 
1798 	adev->gfx.config.num_banks = REG_GET_FIELD(mc_arb_ramcfg,
1799 				MC_ARB_RAMCFG, NOOFBANK);
1800 	adev->gfx.config.num_ranks = REG_GET_FIELD(mc_arb_ramcfg,
1801 				MC_ARB_RAMCFG, NOOFRANKS);
1802 
1803 	adev->gfx.config.num_tile_pipes = adev->gfx.config.max_tile_pipes;
1804 	adev->gfx.config.mem_max_burst_length_bytes = 256;
1805 	if (adev->flags & AMD_IS_APU) {
1806 		/* Get memory bank mapping mode. */
1807 		tmp = RREG32(mmMC_FUS_DRAM0_BANK_ADDR_MAPPING);
1808 		dimm00_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM0_BANK_ADDR_MAPPING, DIMM0ADDRMAP);
1809 		dimm01_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM0_BANK_ADDR_MAPPING, DIMM1ADDRMAP);
1810 
1811 		tmp = RREG32(mmMC_FUS_DRAM1_BANK_ADDR_MAPPING);
1812 		dimm10_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM1_BANK_ADDR_MAPPING, DIMM0ADDRMAP);
1813 		dimm11_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM1_BANK_ADDR_MAPPING, DIMM1ADDRMAP);
1814 
1815 		/* Validate settings in case only one DIMM installed. */
1816 		if ((dimm00_addr_map == 0) || (dimm00_addr_map == 3) || (dimm00_addr_map == 4) || (dimm00_addr_map > 12))
1817 			dimm00_addr_map = 0;
1818 		if ((dimm01_addr_map == 0) || (dimm01_addr_map == 3) || (dimm01_addr_map == 4) || (dimm01_addr_map > 12))
1819 			dimm01_addr_map = 0;
1820 		if ((dimm10_addr_map == 0) || (dimm10_addr_map == 3) || (dimm10_addr_map == 4) || (dimm10_addr_map > 12))
1821 			dimm10_addr_map = 0;
1822 		if ((dimm11_addr_map == 0) || (dimm11_addr_map == 3) || (dimm11_addr_map == 4) || (dimm11_addr_map > 12))
1823 			dimm11_addr_map = 0;
1824 
1825 		/* If DIMM Addr map is 8GB, ROW size should be 2KB. Otherwise 1KB. */
1826 		/* If ROW size(DIMM1) != ROW size(DMIMM0), ROW size should be larger one. */
1827 		if ((dimm00_addr_map == 11) || (dimm01_addr_map == 11) || (dimm10_addr_map == 11) || (dimm11_addr_map == 11))
1828 			adev->gfx.config.mem_row_size_in_kb = 2;
1829 		else
1830 			adev->gfx.config.mem_row_size_in_kb = 1;
1831 	} else {
1832 		tmp = REG_GET_FIELD(mc_arb_ramcfg, MC_ARB_RAMCFG, NOOFCOLS);
1833 		adev->gfx.config.mem_row_size_in_kb = (4 * (1 << (8 + tmp))) / 1024;
1834 		if (adev->gfx.config.mem_row_size_in_kb > 4)
1835 			adev->gfx.config.mem_row_size_in_kb = 4;
1836 	}
1837 
1838 	adev->gfx.config.shader_engine_tile_size = 32;
1839 	adev->gfx.config.num_gpus = 1;
1840 	adev->gfx.config.multi_gpu_tile_size = 64;
1841 
1842 	/* fix up row size */
1843 	switch (adev->gfx.config.mem_row_size_in_kb) {
1844 	case 1:
1845 	default:
1846 		gb_addr_config = REG_SET_FIELD(gb_addr_config, GB_ADDR_CONFIG, ROW_SIZE, 0);
1847 		break;
1848 	case 2:
1849 		gb_addr_config = REG_SET_FIELD(gb_addr_config, GB_ADDR_CONFIG, ROW_SIZE, 1);
1850 		break;
1851 	case 4:
1852 		gb_addr_config = REG_SET_FIELD(gb_addr_config, GB_ADDR_CONFIG, ROW_SIZE, 2);
1853 		break;
1854 	}
1855 	adev->gfx.config.gb_addr_config = gb_addr_config;
1856 
1857 	return 0;
1858 }
1859 
1860 static int gfx_v8_0_compute_ring_init(struct amdgpu_device *adev, int ring_id,
1861 					int mec, int pipe, int queue)
1862 {
1863 	int r;
1864 	unsigned irq_type;
1865 	struct amdgpu_ring *ring = &adev->gfx.compute_ring[ring_id];
1866 	unsigned int hw_prio;
1867 
1868 	ring = &adev->gfx.compute_ring[ring_id];
1869 
1870 	/* mec0 is me1 */
1871 	ring->me = mec + 1;
1872 	ring->pipe = pipe;
1873 	ring->queue = queue;
1874 
1875 	ring->ring_obj = NULL;
1876 	ring->use_doorbell = true;
1877 	ring->doorbell_index = adev->doorbell_index.mec_ring0 + ring_id;
1878 	ring->eop_gpu_addr = adev->gfx.mec.hpd_eop_gpu_addr
1879 				+ (ring_id * GFX8_MEC_HPD_SIZE);
1880 	sprintf(ring->name, "comp_%d.%d.%d", ring->me, ring->pipe, ring->queue);
1881 
1882 	irq_type = AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP
1883 		+ ((ring->me - 1) * adev->gfx.mec.num_pipe_per_mec)
1884 		+ ring->pipe;
1885 
1886 	hw_prio = amdgpu_gfx_is_high_priority_compute_queue(adev, ring) ?
1887 			AMDGPU_RING_PRIO_2 : AMDGPU_RING_PRIO_DEFAULT;
1888 	/* type-2 packets are deprecated on MEC, use type-3 instead */
1889 	r = amdgpu_ring_init(adev, ring, 1024, &adev->gfx.eop_irq, irq_type,
1890 			     hw_prio, NULL);
1891 	if (r)
1892 		return r;
1893 
1894 
1895 	return 0;
1896 }
1897 
1898 static void gfx_v8_0_sq_irq_work_func(struct work_struct *work);
1899 
1900 static int gfx_v8_0_sw_init(void *handle)
1901 {
1902 	int i, j, k, r, ring_id;
1903 	struct amdgpu_ring *ring;
1904 	struct amdgpu_kiq *kiq;
1905 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1906 
1907 	switch (adev->asic_type) {
1908 	case CHIP_TONGA:
1909 	case CHIP_CARRIZO:
1910 	case CHIP_FIJI:
1911 	case CHIP_POLARIS10:
1912 	case CHIP_POLARIS11:
1913 	case CHIP_POLARIS12:
1914 	case CHIP_VEGAM:
1915 		adev->gfx.mec.num_mec = 2;
1916 		break;
1917 	case CHIP_TOPAZ:
1918 	case CHIP_STONEY:
1919 	default:
1920 		adev->gfx.mec.num_mec = 1;
1921 		break;
1922 	}
1923 
1924 	adev->gfx.mec.num_pipe_per_mec = 4;
1925 	adev->gfx.mec.num_queue_per_pipe = 8;
1926 
1927 	/* EOP Event */
1928 	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, VISLANDS30_IV_SRCID_CP_END_OF_PIPE, &adev->gfx.eop_irq);
1929 	if (r)
1930 		return r;
1931 
1932 	/* Privileged reg */
1933 	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, VISLANDS30_IV_SRCID_CP_PRIV_REG_FAULT,
1934 			      &adev->gfx.priv_reg_irq);
1935 	if (r)
1936 		return r;
1937 
1938 	/* Privileged inst */
1939 	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, VISLANDS30_IV_SRCID_CP_PRIV_INSTR_FAULT,
1940 			      &adev->gfx.priv_inst_irq);
1941 	if (r)
1942 		return r;
1943 
1944 	/* Add CP EDC/ECC irq  */
1945 	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, VISLANDS30_IV_SRCID_CP_ECC_ERROR,
1946 			      &adev->gfx.cp_ecc_error_irq);
1947 	if (r)
1948 		return r;
1949 
1950 	/* SQ interrupts. */
1951 	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, VISLANDS30_IV_SRCID_SQ_INTERRUPT_MSG,
1952 			      &adev->gfx.sq_irq);
1953 	if (r) {
1954 		DRM_ERROR("amdgpu_irq_add() for SQ failed: %d\n", r);
1955 		return r;
1956 	}
1957 
1958 	INIT_WORK(&adev->gfx.sq_work.work, gfx_v8_0_sq_irq_work_func);
1959 
1960 	adev->gfx.gfx_current_status = AMDGPU_GFX_NORMAL_MODE;
1961 
1962 	r = gfx_v8_0_init_microcode(adev);
1963 	if (r) {
1964 		DRM_ERROR("Failed to load gfx firmware!\n");
1965 		return r;
1966 	}
1967 
1968 	r = adev->gfx.rlc.funcs->init(adev);
1969 	if (r) {
1970 		DRM_ERROR("Failed to init rlc BOs!\n");
1971 		return r;
1972 	}
1973 
1974 	r = gfx_v8_0_mec_init(adev);
1975 	if (r) {
1976 		DRM_ERROR("Failed to init MEC BOs!\n");
1977 		return r;
1978 	}
1979 
1980 	/* set up the gfx ring */
1981 	for (i = 0; i < adev->gfx.num_gfx_rings; i++) {
1982 		ring = &adev->gfx.gfx_ring[i];
1983 		ring->ring_obj = NULL;
1984 		sprintf(ring->name, "gfx");
1985 		/* no gfx doorbells on iceland */
1986 		if (adev->asic_type != CHIP_TOPAZ) {
1987 			ring->use_doorbell = true;
1988 			ring->doorbell_index = adev->doorbell_index.gfx_ring0;
1989 		}
1990 
1991 		r = amdgpu_ring_init(adev, ring, 1024, &adev->gfx.eop_irq,
1992 				     AMDGPU_CP_IRQ_GFX_ME0_PIPE0_EOP,
1993 				     AMDGPU_RING_PRIO_DEFAULT, NULL);
1994 		if (r)
1995 			return r;
1996 	}
1997 
1998 
1999 	/* set up the compute queues - allocate horizontally across pipes */
2000 	ring_id = 0;
2001 	for (i = 0; i < adev->gfx.mec.num_mec; ++i) {
2002 		for (j = 0; j < adev->gfx.mec.num_queue_per_pipe; j++) {
2003 			for (k = 0; k < adev->gfx.mec.num_pipe_per_mec; k++) {
2004 				if (!amdgpu_gfx_is_mec_queue_enabled(adev, i, k, j))
2005 					continue;
2006 
2007 				r = gfx_v8_0_compute_ring_init(adev,
2008 								ring_id,
2009 								i, k, j);
2010 				if (r)
2011 					return r;
2012 
2013 				ring_id++;
2014 			}
2015 		}
2016 	}
2017 
2018 	r = amdgpu_gfx_kiq_init(adev, GFX8_MEC_HPD_SIZE);
2019 	if (r) {
2020 		DRM_ERROR("Failed to init KIQ BOs!\n");
2021 		return r;
2022 	}
2023 
2024 	kiq = &adev->gfx.kiq;
2025 	r = amdgpu_gfx_kiq_init_ring(adev, &kiq->ring, &kiq->irq);
2026 	if (r)
2027 		return r;
2028 
2029 	/* create MQD for all compute queues as well as KIQ for SRIOV case */
2030 	r = amdgpu_gfx_mqd_sw_init(adev, sizeof(struct vi_mqd_allocation));
2031 	if (r)
2032 		return r;
2033 
2034 	adev->gfx.ce_ram_size = 0x8000;
2035 
2036 	r = gfx_v8_0_gpu_early_init(adev);
2037 	if (r)
2038 		return r;
2039 
2040 	return 0;
2041 }
2042 
2043 static int gfx_v8_0_sw_fini(void *handle)
2044 {
2045 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2046 	int i;
2047 
2048 	for (i = 0; i < adev->gfx.num_gfx_rings; i++)
2049 		amdgpu_ring_fini(&adev->gfx.gfx_ring[i]);
2050 	for (i = 0; i < adev->gfx.num_compute_rings; i++)
2051 		amdgpu_ring_fini(&adev->gfx.compute_ring[i]);
2052 
2053 	amdgpu_gfx_mqd_sw_fini(adev);
2054 	amdgpu_gfx_kiq_free_ring(&adev->gfx.kiq.ring);
2055 	amdgpu_gfx_kiq_fini(adev);
2056 
2057 	gfx_v8_0_mec_fini(adev);
2058 	amdgpu_gfx_rlc_fini(adev);
2059 	amdgpu_bo_free_kernel(&adev->gfx.rlc.clear_state_obj,
2060 				&adev->gfx.rlc.clear_state_gpu_addr,
2061 				(void **)&adev->gfx.rlc.cs_ptr);
2062 	if ((adev->asic_type == CHIP_CARRIZO) ||
2063 	    (adev->asic_type == CHIP_STONEY)) {
2064 		amdgpu_bo_free_kernel(&adev->gfx.rlc.cp_table_obj,
2065 				&adev->gfx.rlc.cp_table_gpu_addr,
2066 				(void **)&adev->gfx.rlc.cp_table_ptr);
2067 	}
2068 	gfx_v8_0_free_microcode(adev);
2069 
2070 	return 0;
2071 }
2072 
2073 static void gfx_v8_0_tiling_mode_table_init(struct amdgpu_device *adev)
2074 {
2075 	uint32_t *modearray, *mod2array;
2076 	const u32 num_tile_mode_states = ARRAY_SIZE(adev->gfx.config.tile_mode_array);
2077 	const u32 num_secondary_tile_mode_states = ARRAY_SIZE(adev->gfx.config.macrotile_mode_array);
2078 	u32 reg_offset;
2079 
2080 	modearray = adev->gfx.config.tile_mode_array;
2081 	mod2array = adev->gfx.config.macrotile_mode_array;
2082 
2083 	for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
2084 		modearray[reg_offset] = 0;
2085 
2086 	for (reg_offset = 0; reg_offset <  num_secondary_tile_mode_states; reg_offset++)
2087 		mod2array[reg_offset] = 0;
2088 
2089 	switch (adev->asic_type) {
2090 	case CHIP_TOPAZ:
2091 		modearray[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2092 				PIPE_CONFIG(ADDR_SURF_P2) |
2093 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
2094 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2095 		modearray[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2096 				PIPE_CONFIG(ADDR_SURF_P2) |
2097 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
2098 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2099 		modearray[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2100 				PIPE_CONFIG(ADDR_SURF_P2) |
2101 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
2102 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2103 		modearray[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2104 				PIPE_CONFIG(ADDR_SURF_P2) |
2105 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
2106 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2107 		modearray[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2108 				PIPE_CONFIG(ADDR_SURF_P2) |
2109 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2110 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2111 		modearray[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2112 				PIPE_CONFIG(ADDR_SURF_P2) |
2113 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2114 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2115 		modearray[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2116 				PIPE_CONFIG(ADDR_SURF_P2) |
2117 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2118 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2119 		modearray[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
2120 				PIPE_CONFIG(ADDR_SURF_P2));
2121 		modearray[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2122 				PIPE_CONFIG(ADDR_SURF_P2) |
2123 				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2124 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2125 		modearray[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2126 				 PIPE_CONFIG(ADDR_SURF_P2) |
2127 				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2128 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2129 		modearray[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2130 				 PIPE_CONFIG(ADDR_SURF_P2) |
2131 				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2132 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2133 		modearray[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2134 				 PIPE_CONFIG(ADDR_SURF_P2) |
2135 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2136 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2137 		modearray[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2138 				 PIPE_CONFIG(ADDR_SURF_P2) |
2139 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2140 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2141 		modearray[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
2142 				 PIPE_CONFIG(ADDR_SURF_P2) |
2143 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2144 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2145 		modearray[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2146 				 PIPE_CONFIG(ADDR_SURF_P2) |
2147 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2148 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2149 		modearray[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2150 				 PIPE_CONFIG(ADDR_SURF_P2) |
2151 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2152 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2153 		modearray[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2154 				 PIPE_CONFIG(ADDR_SURF_P2) |
2155 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2156 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2157 		modearray[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2158 				 PIPE_CONFIG(ADDR_SURF_P2) |
2159 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2160 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2161 		modearray[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
2162 				 PIPE_CONFIG(ADDR_SURF_P2) |
2163 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2164 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2165 		modearray[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
2166 				 PIPE_CONFIG(ADDR_SURF_P2) |
2167 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2168 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2169 		modearray[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2170 				 PIPE_CONFIG(ADDR_SURF_P2) |
2171 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2172 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2173 		modearray[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
2174 				 PIPE_CONFIG(ADDR_SURF_P2) |
2175 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2176 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2177 		modearray[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
2178 				 PIPE_CONFIG(ADDR_SURF_P2) |
2179 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2180 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2181 		modearray[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2182 				 PIPE_CONFIG(ADDR_SURF_P2) |
2183 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2184 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2185 		modearray[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2186 				 PIPE_CONFIG(ADDR_SURF_P2) |
2187 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2188 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2189 		modearray[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2190 				 PIPE_CONFIG(ADDR_SURF_P2) |
2191 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2192 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2193 
2194 		mod2array[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
2195 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2196 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2197 				NUM_BANKS(ADDR_SURF_8_BANK));
2198 		mod2array[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
2199 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2200 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2201 				NUM_BANKS(ADDR_SURF_8_BANK));
2202 		mod2array[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
2203 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2204 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2205 				NUM_BANKS(ADDR_SURF_8_BANK));
2206 		mod2array[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2207 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2208 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2209 				NUM_BANKS(ADDR_SURF_8_BANK));
2210 		mod2array[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2211 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2212 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2213 				NUM_BANKS(ADDR_SURF_8_BANK));
2214 		mod2array[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2215 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2216 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2217 				NUM_BANKS(ADDR_SURF_8_BANK));
2218 		mod2array[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2219 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2220 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2221 				NUM_BANKS(ADDR_SURF_8_BANK));
2222 		mod2array[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
2223 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
2224 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2225 				NUM_BANKS(ADDR_SURF_16_BANK));
2226 		mod2array[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
2227 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2228 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2229 				NUM_BANKS(ADDR_SURF_16_BANK));
2230 		mod2array[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
2231 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2232 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2233 				 NUM_BANKS(ADDR_SURF_16_BANK));
2234 		mod2array[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
2235 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2236 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2237 				 NUM_BANKS(ADDR_SURF_16_BANK));
2238 		mod2array[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2239 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2240 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2241 				 NUM_BANKS(ADDR_SURF_16_BANK));
2242 		mod2array[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2243 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2244 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2245 				 NUM_BANKS(ADDR_SURF_16_BANK));
2246 		mod2array[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2247 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2248 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2249 				 NUM_BANKS(ADDR_SURF_8_BANK));
2250 
2251 		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
2252 			if (reg_offset != 7 && reg_offset != 12 && reg_offset != 17 &&
2253 			    reg_offset != 23)
2254 				WREG32(mmGB_TILE_MODE0 + reg_offset, modearray[reg_offset]);
2255 
2256 		for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
2257 			if (reg_offset != 7)
2258 				WREG32(mmGB_MACROTILE_MODE0 + reg_offset, mod2array[reg_offset]);
2259 
2260 		break;
2261 	case CHIP_FIJI:
2262 	case CHIP_VEGAM:
2263 		modearray[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2264 				PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2265 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
2266 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2267 		modearray[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2268 				PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2269 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
2270 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2271 		modearray[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2272 				PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2273 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
2274 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2275 		modearray[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2276 				PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2277 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
2278 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2279 		modearray[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2280 				PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2281 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2282 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2283 		modearray[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2284 				PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2285 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2286 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2287 		modearray[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2288 				PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2289 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2290 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2291 		modearray[7] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2292 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2293 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2294 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2295 		modearray[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
2296 				PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16));
2297 		modearray[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2298 				PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2299 				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2300 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2301 		modearray[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2302 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2303 				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2304 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2305 		modearray[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2306 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2307 				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2308 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2309 		modearray[12] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2310 				 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2311 				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2312 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2313 		modearray[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2314 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2315 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2316 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2317 		modearray[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2318 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2319 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2320 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2321 		modearray[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
2322 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2323 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2324 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2325 		modearray[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2326 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2327 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2328 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2329 		modearray[17] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2330 				 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2331 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2332 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2333 		modearray[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2334 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2335 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2336 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2337 		modearray[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2338 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2339 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2340 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2341 		modearray[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2342 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2343 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2344 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2345 		modearray[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
2346 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2347 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2348 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2349 		modearray[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
2350 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2351 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2352 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2353 		modearray[23] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
2354 				 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2355 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2356 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2357 		modearray[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2358 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2359 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2360 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2361 		modearray[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
2362 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2363 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2364 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2365 		modearray[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
2366 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2367 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2368 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2369 		modearray[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2370 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2371 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2372 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2373 		modearray[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2374 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2375 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2376 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2377 		modearray[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2378 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2379 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2380 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2381 		modearray[30] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2382 				 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2383 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2384 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2385 
2386 		mod2array[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2387 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2388 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2389 				NUM_BANKS(ADDR_SURF_8_BANK));
2390 		mod2array[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2391 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2392 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2393 				NUM_BANKS(ADDR_SURF_8_BANK));
2394 		mod2array[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2395 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2396 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2397 				NUM_BANKS(ADDR_SURF_8_BANK));
2398 		mod2array[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2399 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2400 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2401 				NUM_BANKS(ADDR_SURF_8_BANK));
2402 		mod2array[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2403 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2404 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2405 				NUM_BANKS(ADDR_SURF_8_BANK));
2406 		mod2array[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2407 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2408 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2409 				NUM_BANKS(ADDR_SURF_8_BANK));
2410 		mod2array[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2411 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2412 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2413 				NUM_BANKS(ADDR_SURF_8_BANK));
2414 		mod2array[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2415 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
2416 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2417 				NUM_BANKS(ADDR_SURF_8_BANK));
2418 		mod2array[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2419 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2420 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2421 				NUM_BANKS(ADDR_SURF_8_BANK));
2422 		mod2array[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2423 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2424 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2425 				 NUM_BANKS(ADDR_SURF_8_BANK));
2426 		mod2array[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2427 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2428 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2429 				 NUM_BANKS(ADDR_SURF_8_BANK));
2430 		mod2array[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2431 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2432 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2433 				 NUM_BANKS(ADDR_SURF_8_BANK));
2434 		mod2array[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2435 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2436 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2437 				 NUM_BANKS(ADDR_SURF_8_BANK));
2438 		mod2array[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2439 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2440 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2441 				 NUM_BANKS(ADDR_SURF_4_BANK));
2442 
2443 		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
2444 			WREG32(mmGB_TILE_MODE0 + reg_offset, modearray[reg_offset]);
2445 
2446 		for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
2447 			if (reg_offset != 7)
2448 				WREG32(mmGB_MACROTILE_MODE0 + reg_offset, mod2array[reg_offset]);
2449 
2450 		break;
2451 	case CHIP_TONGA:
2452 		modearray[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2453 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2454 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
2455 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2456 		modearray[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2457 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2458 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
2459 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2460 		modearray[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2461 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2462 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
2463 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2464 		modearray[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2465 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2466 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
2467 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2468 		modearray[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2469 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2470 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2471 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2472 		modearray[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2473 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2474 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2475 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2476 		modearray[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2477 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2478 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2479 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2480 		modearray[7] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2481 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2482 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2483 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2484 		modearray[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
2485 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16));
2486 		modearray[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2487 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2488 				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2489 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2490 		modearray[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2491 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2492 				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2493 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2494 		modearray[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2495 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2496 				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2497 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2498 		modearray[12] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2499 				 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2500 				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2501 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2502 		modearray[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2503 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2504 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2505 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2506 		modearray[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2507 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2508 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2509 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2510 		modearray[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
2511 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2512 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2513 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2514 		modearray[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2515 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2516 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2517 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2518 		modearray[17] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2519 				 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2520 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2521 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2522 		modearray[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2523 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2524 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2525 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2526 		modearray[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2527 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2528 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2529 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2530 		modearray[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2531 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2532 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2533 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2534 		modearray[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
2535 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2536 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2537 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2538 		modearray[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
2539 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2540 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2541 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2542 		modearray[23] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
2543 				 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2544 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2545 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2546 		modearray[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2547 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2548 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2549 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2550 		modearray[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
2551 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2552 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2553 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2554 		modearray[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
2555 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2556 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2557 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2558 		modearray[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2559 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2560 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2561 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2562 		modearray[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2563 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2564 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2565 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2566 		modearray[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2567 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2568 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2569 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2570 		modearray[30] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2571 				 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2572 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2573 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2574 
2575 		mod2array[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2576 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2577 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2578 				NUM_BANKS(ADDR_SURF_16_BANK));
2579 		mod2array[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2580 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2581 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2582 				NUM_BANKS(ADDR_SURF_16_BANK));
2583 		mod2array[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2584 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2585 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2586 				NUM_BANKS(ADDR_SURF_16_BANK));
2587 		mod2array[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2588 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2589 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2590 				NUM_BANKS(ADDR_SURF_16_BANK));
2591 		mod2array[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2592 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2593 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2594 				NUM_BANKS(ADDR_SURF_16_BANK));
2595 		mod2array[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2596 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2597 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2598 				NUM_BANKS(ADDR_SURF_16_BANK));
2599 		mod2array[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2600 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2601 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2602 				NUM_BANKS(ADDR_SURF_16_BANK));
2603 		mod2array[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2604 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
2605 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2606 				NUM_BANKS(ADDR_SURF_16_BANK));
2607 		mod2array[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2608 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2609 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2610 				NUM_BANKS(ADDR_SURF_16_BANK));
2611 		mod2array[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2612 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2613 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2614 				 NUM_BANKS(ADDR_SURF_16_BANK));
2615 		mod2array[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2616 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2617 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2618 				 NUM_BANKS(ADDR_SURF_16_BANK));
2619 		mod2array[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2620 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2621 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2622 				 NUM_BANKS(ADDR_SURF_8_BANK));
2623 		mod2array[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2624 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2625 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2626 				 NUM_BANKS(ADDR_SURF_4_BANK));
2627 		mod2array[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2628 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2629 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2630 				 NUM_BANKS(ADDR_SURF_4_BANK));
2631 
2632 		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
2633 			WREG32(mmGB_TILE_MODE0 + reg_offset, modearray[reg_offset]);
2634 
2635 		for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
2636 			if (reg_offset != 7)
2637 				WREG32(mmGB_MACROTILE_MODE0 + reg_offset, mod2array[reg_offset]);
2638 
2639 		break;
2640 	case CHIP_POLARIS11:
2641 	case CHIP_POLARIS12:
2642 		modearray[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2643 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2644 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
2645 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2646 		modearray[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2647 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2648 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
2649 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2650 		modearray[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2651 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2652 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
2653 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2654 		modearray[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2655 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2656 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
2657 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2658 		modearray[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2659 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2660 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2661 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2662 		modearray[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2663 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2664 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2665 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2666 		modearray[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2667 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2668 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2669 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2670 		modearray[7] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2671 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2672 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2673 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2674 		modearray[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
2675 				PIPE_CONFIG(ADDR_SURF_P4_16x16));
2676 		modearray[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2677 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2678 				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2679 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2680 		modearray[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2681 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2682 				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2683 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2684 		modearray[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2685 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2686 				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2687 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2688 		modearray[12] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2689 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2690 				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2691 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2692 		modearray[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2693 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2694 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2695 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2696 		modearray[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2697 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2698 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2699 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2700 		modearray[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
2701 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2702 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2703 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2704 		modearray[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2705 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2706 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2707 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2708 		modearray[17] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2709 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2710 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2711 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2712 		modearray[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2713 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2714 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2715 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2716 		modearray[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2717 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2718 				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2719 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2720 		modearray[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2721 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2722 				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2723 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2724 		modearray[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
2725 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2726 				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2727 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2728 		modearray[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
2729 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2730 				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2731 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2732 		modearray[23] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
2733 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2734 				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2735 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2736 		modearray[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2737 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2738 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2739 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2740 		modearray[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
2741 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2742 				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2743 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2744 		modearray[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
2745 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2746 				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2747 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2748 		modearray[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2749 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2750 				MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2751 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2752 		modearray[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2753 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2754 				MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2755 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2756 		modearray[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2757 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2758 				MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2759 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2760 		modearray[30] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2761 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2762 				MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2763 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2764 
2765 		mod2array[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2766 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2767 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2768 				NUM_BANKS(ADDR_SURF_16_BANK));
2769 
2770 		mod2array[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2771 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2772 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2773 				NUM_BANKS(ADDR_SURF_16_BANK));
2774 
2775 		mod2array[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2776 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2777 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2778 				NUM_BANKS(ADDR_SURF_16_BANK));
2779 
2780 		mod2array[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2781 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2782 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2783 				NUM_BANKS(ADDR_SURF_16_BANK));
2784 
2785 		mod2array[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2786 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2787 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2788 				NUM_BANKS(ADDR_SURF_16_BANK));
2789 
2790 		mod2array[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2791 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2792 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2793 				NUM_BANKS(ADDR_SURF_16_BANK));
2794 
2795 		mod2array[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2796 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2797 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2798 				NUM_BANKS(ADDR_SURF_16_BANK));
2799 
2800 		mod2array[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
2801 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
2802 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2803 				NUM_BANKS(ADDR_SURF_16_BANK));
2804 
2805 		mod2array[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
2806 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2807 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2808 				NUM_BANKS(ADDR_SURF_16_BANK));
2809 
2810 		mod2array[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2811 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2812 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2813 				NUM_BANKS(ADDR_SURF_16_BANK));
2814 
2815 		mod2array[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2816 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2817 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2818 				NUM_BANKS(ADDR_SURF_16_BANK));
2819 
2820 		mod2array[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2821 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2822 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2823 				NUM_BANKS(ADDR_SURF_16_BANK));
2824 
2825 		mod2array[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2826 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2827 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2828 				NUM_BANKS(ADDR_SURF_8_BANK));
2829 
2830 		mod2array[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2831 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2832 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2833 				NUM_BANKS(ADDR_SURF_4_BANK));
2834 
2835 		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
2836 			WREG32(mmGB_TILE_MODE0 + reg_offset, modearray[reg_offset]);
2837 
2838 		for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
2839 			if (reg_offset != 7)
2840 				WREG32(mmGB_MACROTILE_MODE0 + reg_offset, mod2array[reg_offset]);
2841 
2842 		break;
2843 	case CHIP_POLARIS10:
2844 		modearray[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2845 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2846 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
2847 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2848 		modearray[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2849 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2850 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
2851 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2852 		modearray[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2853 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2854 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
2855 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2856 		modearray[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2857 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2858 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
2859 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2860 		modearray[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2861 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2862 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2863 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2864 		modearray[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2865 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2866 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2867 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2868 		modearray[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2869 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2870 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2871 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2872 		modearray[7] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2873 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2874 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2875 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2876 		modearray[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
2877 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16));
2878 		modearray[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2879 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2880 				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2881 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2882 		modearray[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2883 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2884 				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2885 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2886 		modearray[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2887 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2888 				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2889 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2890 		modearray[12] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2891 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2892 				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2893 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2894 		modearray[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2895 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2896 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2897 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2898 		modearray[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2899 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2900 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2901 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2902 		modearray[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
2903 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2904 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2905 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2906 		modearray[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2907 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2908 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2909 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2910 		modearray[17] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2911 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2912 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2913 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2914 		modearray[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2915 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2916 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2917 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2918 		modearray[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2919 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2920 				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2921 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2922 		modearray[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2923 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2924 				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2925 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2926 		modearray[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
2927 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2928 				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2929 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2930 		modearray[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
2931 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2932 				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2933 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2934 		modearray[23] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
2935 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2936 				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2937 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2938 		modearray[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2939 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2940 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2941 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2942 		modearray[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
2943 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2944 				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2945 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2946 		modearray[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
2947 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2948 				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2949 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2950 		modearray[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2951 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2952 				MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2953 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2954 		modearray[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2955 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2956 				MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2957 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2958 		modearray[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2959 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2960 				MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2961 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2962 		modearray[30] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2963 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2964 				MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2965 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2966 
2967 		mod2array[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2968 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2969 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2970 				NUM_BANKS(ADDR_SURF_16_BANK));
2971 
2972 		mod2array[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2973 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2974 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2975 				NUM_BANKS(ADDR_SURF_16_BANK));
2976 
2977 		mod2array[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2978 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2979 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2980 				NUM_BANKS(ADDR_SURF_16_BANK));
2981 
2982 		mod2array[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2983 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2984 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2985 				NUM_BANKS(ADDR_SURF_16_BANK));
2986 
2987 		mod2array[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2988 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2989 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2990 				NUM_BANKS(ADDR_SURF_16_BANK));
2991 
2992 		mod2array[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2993 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2994 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2995 				NUM_BANKS(ADDR_SURF_16_BANK));
2996 
2997 		mod2array[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2998 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2999 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
3000 				NUM_BANKS(ADDR_SURF_16_BANK));
3001 
3002 		mod2array[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3003 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
3004 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3005 				NUM_BANKS(ADDR_SURF_16_BANK));
3006 
3007 		mod2array[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3008 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
3009 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3010 				NUM_BANKS(ADDR_SURF_16_BANK));
3011 
3012 		mod2array[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3013 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
3014 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3015 				NUM_BANKS(ADDR_SURF_16_BANK));
3016 
3017 		mod2array[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3018 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3019 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3020 				NUM_BANKS(ADDR_SURF_16_BANK));
3021 
3022 		mod2array[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3023 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3024 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
3025 				NUM_BANKS(ADDR_SURF_8_BANK));
3026 
3027 		mod2array[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3028 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3029 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
3030 				NUM_BANKS(ADDR_SURF_4_BANK));
3031 
3032 		mod2array[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3033 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3034 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
3035 				NUM_BANKS(ADDR_SURF_4_BANK));
3036 
3037 		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
3038 			WREG32(mmGB_TILE_MODE0 + reg_offset, modearray[reg_offset]);
3039 
3040 		for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
3041 			if (reg_offset != 7)
3042 				WREG32(mmGB_MACROTILE_MODE0 + reg_offset, mod2array[reg_offset]);
3043 
3044 		break;
3045 	case CHIP_STONEY:
3046 		modearray[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3047 				PIPE_CONFIG(ADDR_SURF_P2) |
3048 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
3049 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3050 		modearray[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3051 				PIPE_CONFIG(ADDR_SURF_P2) |
3052 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
3053 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3054 		modearray[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3055 				PIPE_CONFIG(ADDR_SURF_P2) |
3056 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
3057 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3058 		modearray[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3059 				PIPE_CONFIG(ADDR_SURF_P2) |
3060 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
3061 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3062 		modearray[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3063 				PIPE_CONFIG(ADDR_SURF_P2) |
3064 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
3065 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3066 		modearray[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
3067 				PIPE_CONFIG(ADDR_SURF_P2) |
3068 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
3069 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3070 		modearray[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
3071 				PIPE_CONFIG(ADDR_SURF_P2) |
3072 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
3073 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3074 		modearray[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
3075 				PIPE_CONFIG(ADDR_SURF_P2));
3076 		modearray[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
3077 				PIPE_CONFIG(ADDR_SURF_P2) |
3078 				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
3079 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3080 		modearray[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3081 				 PIPE_CONFIG(ADDR_SURF_P2) |
3082 				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
3083 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3084 		modearray[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
3085 				 PIPE_CONFIG(ADDR_SURF_P2) |
3086 				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
3087 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
3088 		modearray[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
3089 				 PIPE_CONFIG(ADDR_SURF_P2) |
3090 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3091 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3092 		modearray[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3093 				 PIPE_CONFIG(ADDR_SURF_P2) |
3094 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3095 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3096 		modearray[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
3097 				 PIPE_CONFIG(ADDR_SURF_P2) |
3098 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3099 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3100 		modearray[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
3101 				 PIPE_CONFIG(ADDR_SURF_P2) |
3102 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3103 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
3104 		modearray[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
3105 				 PIPE_CONFIG(ADDR_SURF_P2) |
3106 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3107 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3108 		modearray[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
3109 				 PIPE_CONFIG(ADDR_SURF_P2) |
3110 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3111 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3112 		modearray[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
3113 				 PIPE_CONFIG(ADDR_SURF_P2) |
3114 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3115 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3116 		modearray[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
3117 				 PIPE_CONFIG(ADDR_SURF_P2) |
3118 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3119 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3120 		modearray[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
3121 				 PIPE_CONFIG(ADDR_SURF_P2) |
3122 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3123 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3124 		modearray[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
3125 				 PIPE_CONFIG(ADDR_SURF_P2) |
3126 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3127 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3128 		modearray[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
3129 				 PIPE_CONFIG(ADDR_SURF_P2) |
3130 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3131 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3132 		modearray[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
3133 				 PIPE_CONFIG(ADDR_SURF_P2) |
3134 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3135 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3136 		modearray[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
3137 				 PIPE_CONFIG(ADDR_SURF_P2) |
3138 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
3139 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3140 		modearray[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3141 				 PIPE_CONFIG(ADDR_SURF_P2) |
3142 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
3143 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3144 		modearray[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
3145 				 PIPE_CONFIG(ADDR_SURF_P2) |
3146 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
3147 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
3148 
3149 		mod2array[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3150 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
3151 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3152 				NUM_BANKS(ADDR_SURF_8_BANK));
3153 		mod2array[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3154 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
3155 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3156 				NUM_BANKS(ADDR_SURF_8_BANK));
3157 		mod2array[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3158 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3159 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3160 				NUM_BANKS(ADDR_SURF_8_BANK));
3161 		mod2array[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3162 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3163 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3164 				NUM_BANKS(ADDR_SURF_8_BANK));
3165 		mod2array[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3166 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3167 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3168 				NUM_BANKS(ADDR_SURF_8_BANK));
3169 		mod2array[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3170 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3171 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3172 				NUM_BANKS(ADDR_SURF_8_BANK));
3173 		mod2array[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3174 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3175 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3176 				NUM_BANKS(ADDR_SURF_8_BANK));
3177 		mod2array[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
3178 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
3179 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3180 				NUM_BANKS(ADDR_SURF_16_BANK));
3181 		mod2array[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
3182 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
3183 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3184 				NUM_BANKS(ADDR_SURF_16_BANK));
3185 		mod2array[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
3186 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
3187 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3188 				 NUM_BANKS(ADDR_SURF_16_BANK));
3189 		mod2array[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
3190 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
3191 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3192 				 NUM_BANKS(ADDR_SURF_16_BANK));
3193 		mod2array[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3194 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
3195 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3196 				 NUM_BANKS(ADDR_SURF_16_BANK));
3197 		mod2array[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3198 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3199 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3200 				 NUM_BANKS(ADDR_SURF_16_BANK));
3201 		mod2array[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3202 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3203 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3204 				 NUM_BANKS(ADDR_SURF_8_BANK));
3205 
3206 		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
3207 			if (reg_offset != 7 && reg_offset != 12 && reg_offset != 17 &&
3208 			    reg_offset != 23)
3209 				WREG32(mmGB_TILE_MODE0 + reg_offset, modearray[reg_offset]);
3210 
3211 		for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
3212 			if (reg_offset != 7)
3213 				WREG32(mmGB_MACROTILE_MODE0 + reg_offset, mod2array[reg_offset]);
3214 
3215 		break;
3216 	default:
3217 		dev_warn(adev->dev,
3218 			 "Unknown chip type (%d) in function gfx_v8_0_tiling_mode_table_init() falling through to CHIP_CARRIZO\n",
3219 			 adev->asic_type);
3220 		fallthrough;
3221 
3222 	case CHIP_CARRIZO:
3223 		modearray[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3224 				PIPE_CONFIG(ADDR_SURF_P2) |
3225 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
3226 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3227 		modearray[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3228 				PIPE_CONFIG(ADDR_SURF_P2) |
3229 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
3230 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3231 		modearray[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3232 				PIPE_CONFIG(ADDR_SURF_P2) |
3233 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
3234 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3235 		modearray[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3236 				PIPE_CONFIG(ADDR_SURF_P2) |
3237 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
3238 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3239 		modearray[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3240 				PIPE_CONFIG(ADDR_SURF_P2) |
3241 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
3242 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3243 		modearray[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
3244 				PIPE_CONFIG(ADDR_SURF_P2) |
3245 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
3246 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3247 		modearray[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
3248 				PIPE_CONFIG(ADDR_SURF_P2) |
3249 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
3250 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3251 		modearray[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
3252 				PIPE_CONFIG(ADDR_SURF_P2));
3253 		modearray[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
3254 				PIPE_CONFIG(ADDR_SURF_P2) |
3255 				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
3256 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3257 		modearray[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3258 				 PIPE_CONFIG(ADDR_SURF_P2) |
3259 				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
3260 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3261 		modearray[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
3262 				 PIPE_CONFIG(ADDR_SURF_P2) |
3263 				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
3264 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
3265 		modearray[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
3266 				 PIPE_CONFIG(ADDR_SURF_P2) |
3267 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3268 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3269 		modearray[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3270 				 PIPE_CONFIG(ADDR_SURF_P2) |
3271 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3272 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3273 		modearray[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
3274 				 PIPE_CONFIG(ADDR_SURF_P2) |
3275 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3276 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3277 		modearray[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
3278 				 PIPE_CONFIG(ADDR_SURF_P2) |
3279 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3280 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
3281 		modearray[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
3282 				 PIPE_CONFIG(ADDR_SURF_P2) |
3283 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3284 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3285 		modearray[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
3286 				 PIPE_CONFIG(ADDR_SURF_P2) |
3287 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3288 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3289 		modearray[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
3290 				 PIPE_CONFIG(ADDR_SURF_P2) |
3291 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3292 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3293 		modearray[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
3294 				 PIPE_CONFIG(ADDR_SURF_P2) |
3295 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3296 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3297 		modearray[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
3298 				 PIPE_CONFIG(ADDR_SURF_P2) |
3299 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3300 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3301 		modearray[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
3302 				 PIPE_CONFIG(ADDR_SURF_P2) |
3303 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3304 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3305 		modearray[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
3306 				 PIPE_CONFIG(ADDR_SURF_P2) |
3307 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3308 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3309 		modearray[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
3310 				 PIPE_CONFIG(ADDR_SURF_P2) |
3311 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3312 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3313 		modearray[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
3314 				 PIPE_CONFIG(ADDR_SURF_P2) |
3315 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
3316 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3317 		modearray[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3318 				 PIPE_CONFIG(ADDR_SURF_P2) |
3319 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
3320 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3321 		modearray[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
3322 				 PIPE_CONFIG(ADDR_SURF_P2) |
3323 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
3324 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
3325 
3326 		mod2array[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3327 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
3328 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3329 				NUM_BANKS(ADDR_SURF_8_BANK));
3330 		mod2array[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3331 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
3332 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3333 				NUM_BANKS(ADDR_SURF_8_BANK));
3334 		mod2array[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3335 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3336 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3337 				NUM_BANKS(ADDR_SURF_8_BANK));
3338 		mod2array[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3339 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3340 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3341 				NUM_BANKS(ADDR_SURF_8_BANK));
3342 		mod2array[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3343 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3344 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3345 				NUM_BANKS(ADDR_SURF_8_BANK));
3346 		mod2array[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3347 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3348 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3349 				NUM_BANKS(ADDR_SURF_8_BANK));
3350 		mod2array[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3351 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3352 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3353 				NUM_BANKS(ADDR_SURF_8_BANK));
3354 		mod2array[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
3355 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
3356 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3357 				NUM_BANKS(ADDR_SURF_16_BANK));
3358 		mod2array[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
3359 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
3360 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3361 				NUM_BANKS(ADDR_SURF_16_BANK));
3362 		mod2array[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
3363 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
3364 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3365 				 NUM_BANKS(ADDR_SURF_16_BANK));
3366 		mod2array[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
3367 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
3368 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3369 				 NUM_BANKS(ADDR_SURF_16_BANK));
3370 		mod2array[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3371 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
3372 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3373 				 NUM_BANKS(ADDR_SURF_16_BANK));
3374 		mod2array[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3375 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3376 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3377 				 NUM_BANKS(ADDR_SURF_16_BANK));
3378 		mod2array[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3379 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3380 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3381 				 NUM_BANKS(ADDR_SURF_8_BANK));
3382 
3383 		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
3384 			if (reg_offset != 7 && reg_offset != 12 && reg_offset != 17 &&
3385 			    reg_offset != 23)
3386 				WREG32(mmGB_TILE_MODE0 + reg_offset, modearray[reg_offset]);
3387 
3388 		for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
3389 			if (reg_offset != 7)
3390 				WREG32(mmGB_MACROTILE_MODE0 + reg_offset, mod2array[reg_offset]);
3391 
3392 		break;
3393 	}
3394 }
3395 
3396 static void gfx_v8_0_select_se_sh(struct amdgpu_device *adev,
3397 				  u32 se_num, u32 sh_num, u32 instance)
3398 {
3399 	u32 data;
3400 
3401 	if (instance == 0xffffffff)
3402 		data = REG_SET_FIELD(0, GRBM_GFX_INDEX, INSTANCE_BROADCAST_WRITES, 1);
3403 	else
3404 		data = REG_SET_FIELD(0, GRBM_GFX_INDEX, INSTANCE_INDEX, instance);
3405 
3406 	if (se_num == 0xffffffff)
3407 		data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SE_BROADCAST_WRITES, 1);
3408 	else
3409 		data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SE_INDEX, se_num);
3410 
3411 	if (sh_num == 0xffffffff)
3412 		data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SH_BROADCAST_WRITES, 1);
3413 	else
3414 		data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SH_INDEX, sh_num);
3415 
3416 	WREG32(mmGRBM_GFX_INDEX, data);
3417 }
3418 
3419 static void gfx_v8_0_select_me_pipe_q(struct amdgpu_device *adev,
3420 				  u32 me, u32 pipe, u32 q, u32 vm)
3421 {
3422 	vi_srbm_select(adev, me, pipe, q, vm);
3423 }
3424 
3425 static u32 gfx_v8_0_get_rb_active_bitmap(struct amdgpu_device *adev)
3426 {
3427 	u32 data, mask;
3428 
3429 	data =  RREG32(mmCC_RB_BACKEND_DISABLE) |
3430 		RREG32(mmGC_USER_RB_BACKEND_DISABLE);
3431 
3432 	data = REG_GET_FIELD(data, GC_USER_RB_BACKEND_DISABLE, BACKEND_DISABLE);
3433 
3434 	mask = amdgpu_gfx_create_bitmask(adev->gfx.config.max_backends_per_se /
3435 					 adev->gfx.config.max_sh_per_se);
3436 
3437 	return (~data) & mask;
3438 }
3439 
3440 static void
3441 gfx_v8_0_raster_config(struct amdgpu_device *adev, u32 *rconf, u32 *rconf1)
3442 {
3443 	switch (adev->asic_type) {
3444 	case CHIP_FIJI:
3445 	case CHIP_VEGAM:
3446 		*rconf |= RB_MAP_PKR0(2) | RB_MAP_PKR1(2) |
3447 			  RB_XSEL2(1) | PKR_MAP(2) |
3448 			  PKR_XSEL(1) | PKR_YSEL(1) |
3449 			  SE_MAP(2) | SE_XSEL(2) | SE_YSEL(3);
3450 		*rconf1 |= SE_PAIR_MAP(2) | SE_PAIR_XSEL(3) |
3451 			   SE_PAIR_YSEL(2);
3452 		break;
3453 	case CHIP_TONGA:
3454 	case CHIP_POLARIS10:
3455 		*rconf |= RB_MAP_PKR0(2) | RB_XSEL2(1) | SE_MAP(2) |
3456 			  SE_XSEL(1) | SE_YSEL(1);
3457 		*rconf1 |= SE_PAIR_MAP(2) | SE_PAIR_XSEL(2) |
3458 			   SE_PAIR_YSEL(2);
3459 		break;
3460 	case CHIP_TOPAZ:
3461 	case CHIP_CARRIZO:
3462 		*rconf |= RB_MAP_PKR0(2);
3463 		*rconf1 |= 0x0;
3464 		break;
3465 	case CHIP_POLARIS11:
3466 	case CHIP_POLARIS12:
3467 		*rconf |= RB_MAP_PKR0(2) | RB_XSEL2(1) | SE_MAP(2) |
3468 			  SE_XSEL(1) | SE_YSEL(1);
3469 		*rconf1 |= 0x0;
3470 		break;
3471 	case CHIP_STONEY:
3472 		*rconf |= 0x0;
3473 		*rconf1 |= 0x0;
3474 		break;
3475 	default:
3476 		DRM_ERROR("unknown asic: 0x%x\n", adev->asic_type);
3477 		break;
3478 	}
3479 }
3480 
3481 static void
3482 gfx_v8_0_write_harvested_raster_configs(struct amdgpu_device *adev,
3483 					u32 raster_config, u32 raster_config_1,
3484 					unsigned rb_mask, unsigned num_rb)
3485 {
3486 	unsigned sh_per_se = max_t(unsigned, adev->gfx.config.max_sh_per_se, 1);
3487 	unsigned num_se = max_t(unsigned, adev->gfx.config.max_shader_engines, 1);
3488 	unsigned rb_per_pkr = min_t(unsigned, num_rb / num_se / sh_per_se, 2);
3489 	unsigned rb_per_se = num_rb / num_se;
3490 	unsigned se_mask[4];
3491 	unsigned se;
3492 
3493 	se_mask[0] = ((1 << rb_per_se) - 1) & rb_mask;
3494 	se_mask[1] = (se_mask[0] << rb_per_se) & rb_mask;
3495 	se_mask[2] = (se_mask[1] << rb_per_se) & rb_mask;
3496 	se_mask[3] = (se_mask[2] << rb_per_se) & rb_mask;
3497 
3498 	WARN_ON(!(num_se == 1 || num_se == 2 || num_se == 4));
3499 	WARN_ON(!(sh_per_se == 1 || sh_per_se == 2));
3500 	WARN_ON(!(rb_per_pkr == 1 || rb_per_pkr == 2));
3501 
3502 	if ((num_se > 2) && ((!se_mask[0] && !se_mask[1]) ||
3503 			     (!se_mask[2] && !se_mask[3]))) {
3504 		raster_config_1 &= ~SE_PAIR_MAP_MASK;
3505 
3506 		if (!se_mask[0] && !se_mask[1]) {
3507 			raster_config_1 |=
3508 				SE_PAIR_MAP(RASTER_CONFIG_SE_PAIR_MAP_3);
3509 		} else {
3510 			raster_config_1 |=
3511 				SE_PAIR_MAP(RASTER_CONFIG_SE_PAIR_MAP_0);
3512 		}
3513 	}
3514 
3515 	for (se = 0; se < num_se; se++) {
3516 		unsigned raster_config_se = raster_config;
3517 		unsigned pkr0_mask = ((1 << rb_per_pkr) - 1) << (se * rb_per_se);
3518 		unsigned pkr1_mask = pkr0_mask << rb_per_pkr;
3519 		int idx = (se / 2) * 2;
3520 
3521 		if ((num_se > 1) && (!se_mask[idx] || !se_mask[idx + 1])) {
3522 			raster_config_se &= ~SE_MAP_MASK;
3523 
3524 			if (!se_mask[idx]) {
3525 				raster_config_se |= SE_MAP(RASTER_CONFIG_SE_MAP_3);
3526 			} else {
3527 				raster_config_se |= SE_MAP(RASTER_CONFIG_SE_MAP_0);
3528 			}
3529 		}
3530 
3531 		pkr0_mask &= rb_mask;
3532 		pkr1_mask &= rb_mask;
3533 		if (rb_per_se > 2 && (!pkr0_mask || !pkr1_mask)) {
3534 			raster_config_se &= ~PKR_MAP_MASK;
3535 
3536 			if (!pkr0_mask) {
3537 				raster_config_se |= PKR_MAP(RASTER_CONFIG_PKR_MAP_3);
3538 			} else {
3539 				raster_config_se |= PKR_MAP(RASTER_CONFIG_PKR_MAP_0);
3540 			}
3541 		}
3542 
3543 		if (rb_per_se >= 2) {
3544 			unsigned rb0_mask = 1 << (se * rb_per_se);
3545 			unsigned rb1_mask = rb0_mask << 1;
3546 
3547 			rb0_mask &= rb_mask;
3548 			rb1_mask &= rb_mask;
3549 			if (!rb0_mask || !rb1_mask) {
3550 				raster_config_se &= ~RB_MAP_PKR0_MASK;
3551 
3552 				if (!rb0_mask) {
3553 					raster_config_se |=
3554 						RB_MAP_PKR0(RASTER_CONFIG_RB_MAP_3);
3555 				} else {
3556 					raster_config_se |=
3557 						RB_MAP_PKR0(RASTER_CONFIG_RB_MAP_0);
3558 				}
3559 			}
3560 
3561 			if (rb_per_se > 2) {
3562 				rb0_mask = 1 << (se * rb_per_se + rb_per_pkr);
3563 				rb1_mask = rb0_mask << 1;
3564 				rb0_mask &= rb_mask;
3565 				rb1_mask &= rb_mask;
3566 				if (!rb0_mask || !rb1_mask) {
3567 					raster_config_se &= ~RB_MAP_PKR1_MASK;
3568 
3569 					if (!rb0_mask) {
3570 						raster_config_se |=
3571 							RB_MAP_PKR1(RASTER_CONFIG_RB_MAP_3);
3572 					} else {
3573 						raster_config_se |=
3574 							RB_MAP_PKR1(RASTER_CONFIG_RB_MAP_0);
3575 					}
3576 				}
3577 			}
3578 		}
3579 
3580 		/* GRBM_GFX_INDEX has a different offset on VI */
3581 		gfx_v8_0_select_se_sh(adev, se, 0xffffffff, 0xffffffff);
3582 		WREG32(mmPA_SC_RASTER_CONFIG, raster_config_se);
3583 		WREG32(mmPA_SC_RASTER_CONFIG_1, raster_config_1);
3584 	}
3585 
3586 	/* GRBM_GFX_INDEX has a different offset on VI */
3587 	gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
3588 }
3589 
3590 static void gfx_v8_0_setup_rb(struct amdgpu_device *adev)
3591 {
3592 	int i, j;
3593 	u32 data;
3594 	u32 raster_config = 0, raster_config_1 = 0;
3595 	u32 active_rbs = 0;
3596 	u32 rb_bitmap_width_per_sh = adev->gfx.config.max_backends_per_se /
3597 					adev->gfx.config.max_sh_per_se;
3598 	unsigned num_rb_pipes;
3599 
3600 	mutex_lock(&adev->grbm_idx_mutex);
3601 	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
3602 		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
3603 			gfx_v8_0_select_se_sh(adev, i, j, 0xffffffff);
3604 			data = gfx_v8_0_get_rb_active_bitmap(adev);
3605 			active_rbs |= data << ((i * adev->gfx.config.max_sh_per_se + j) *
3606 					       rb_bitmap_width_per_sh);
3607 		}
3608 	}
3609 	gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
3610 
3611 	adev->gfx.config.backend_enable_mask = active_rbs;
3612 	adev->gfx.config.num_rbs = hweight32(active_rbs);
3613 
3614 	num_rb_pipes = min_t(unsigned, adev->gfx.config.max_backends_per_se *
3615 			     adev->gfx.config.max_shader_engines, 16);
3616 
3617 	gfx_v8_0_raster_config(adev, &raster_config, &raster_config_1);
3618 
3619 	if (!adev->gfx.config.backend_enable_mask ||
3620 			adev->gfx.config.num_rbs >= num_rb_pipes) {
3621 		WREG32(mmPA_SC_RASTER_CONFIG, raster_config);
3622 		WREG32(mmPA_SC_RASTER_CONFIG_1, raster_config_1);
3623 	} else {
3624 		gfx_v8_0_write_harvested_raster_configs(adev, raster_config, raster_config_1,
3625 							adev->gfx.config.backend_enable_mask,
3626 							num_rb_pipes);
3627 	}
3628 
3629 	/* cache the values for userspace */
3630 	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
3631 		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
3632 			gfx_v8_0_select_se_sh(adev, i, j, 0xffffffff);
3633 			adev->gfx.config.rb_config[i][j].rb_backend_disable =
3634 				RREG32(mmCC_RB_BACKEND_DISABLE);
3635 			adev->gfx.config.rb_config[i][j].user_rb_backend_disable =
3636 				RREG32(mmGC_USER_RB_BACKEND_DISABLE);
3637 			adev->gfx.config.rb_config[i][j].raster_config =
3638 				RREG32(mmPA_SC_RASTER_CONFIG);
3639 			adev->gfx.config.rb_config[i][j].raster_config_1 =
3640 				RREG32(mmPA_SC_RASTER_CONFIG_1);
3641 		}
3642 	}
3643 	gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
3644 	mutex_unlock(&adev->grbm_idx_mutex);
3645 }
3646 
3647 #define DEFAULT_SH_MEM_BASES	(0x6000)
3648 /**
3649  * gfx_v8_0_init_compute_vmid - gart enable
3650  *
3651  * @adev: amdgpu_device pointer
3652  *
3653  * Initialize compute vmid sh_mem registers
3654  *
3655  */
3656 static void gfx_v8_0_init_compute_vmid(struct amdgpu_device *adev)
3657 {
3658 	int i;
3659 	uint32_t sh_mem_config;
3660 	uint32_t sh_mem_bases;
3661 
3662 	/*
3663 	 * Configure apertures:
3664 	 * LDS:         0x60000000'00000000 - 0x60000001'00000000 (4GB)
3665 	 * Scratch:     0x60000001'00000000 - 0x60000002'00000000 (4GB)
3666 	 * GPUVM:       0x60010000'00000000 - 0x60020000'00000000 (1TB)
3667 	 */
3668 	sh_mem_bases = DEFAULT_SH_MEM_BASES | (DEFAULT_SH_MEM_BASES << 16);
3669 
3670 	sh_mem_config = SH_MEM_ADDRESS_MODE_HSA64 <<
3671 			SH_MEM_CONFIG__ADDRESS_MODE__SHIFT |
3672 			SH_MEM_ALIGNMENT_MODE_UNALIGNED <<
3673 			SH_MEM_CONFIG__ALIGNMENT_MODE__SHIFT |
3674 			MTYPE_CC << SH_MEM_CONFIG__DEFAULT_MTYPE__SHIFT |
3675 			SH_MEM_CONFIG__PRIVATE_ATC_MASK;
3676 
3677 	mutex_lock(&adev->srbm_mutex);
3678 	for (i = adev->vm_manager.first_kfd_vmid; i < AMDGPU_NUM_VMID; i++) {
3679 		vi_srbm_select(adev, 0, 0, 0, i);
3680 		/* CP and shaders */
3681 		WREG32(mmSH_MEM_CONFIG, sh_mem_config);
3682 		WREG32(mmSH_MEM_APE1_BASE, 1);
3683 		WREG32(mmSH_MEM_APE1_LIMIT, 0);
3684 		WREG32(mmSH_MEM_BASES, sh_mem_bases);
3685 	}
3686 	vi_srbm_select(adev, 0, 0, 0, 0);
3687 	mutex_unlock(&adev->srbm_mutex);
3688 
3689 	/* Initialize all compute VMIDs to have no GDS, GWS, or OA
3690 	   access. These should be enabled by FW for target VMIDs. */
3691 	for (i = adev->vm_manager.first_kfd_vmid; i < AMDGPU_NUM_VMID; i++) {
3692 		WREG32(amdgpu_gds_reg_offset[i].mem_base, 0);
3693 		WREG32(amdgpu_gds_reg_offset[i].mem_size, 0);
3694 		WREG32(amdgpu_gds_reg_offset[i].gws, 0);
3695 		WREG32(amdgpu_gds_reg_offset[i].oa, 0);
3696 	}
3697 }
3698 
3699 static void gfx_v8_0_init_gds_vmid(struct amdgpu_device *adev)
3700 {
3701 	int vmid;
3702 
3703 	/*
3704 	 * Initialize all compute and user-gfx VMIDs to have no GDS, GWS, or OA
3705 	 * access. Compute VMIDs should be enabled by FW for target VMIDs,
3706 	 * the driver can enable them for graphics. VMID0 should maintain
3707 	 * access so that HWS firmware can save/restore entries.
3708 	 */
3709 	for (vmid = 1; vmid < AMDGPU_NUM_VMID; vmid++) {
3710 		WREG32(amdgpu_gds_reg_offset[vmid].mem_base, 0);
3711 		WREG32(amdgpu_gds_reg_offset[vmid].mem_size, 0);
3712 		WREG32(amdgpu_gds_reg_offset[vmid].gws, 0);
3713 		WREG32(amdgpu_gds_reg_offset[vmid].oa, 0);
3714 	}
3715 }
3716 
3717 static void gfx_v8_0_config_init(struct amdgpu_device *adev)
3718 {
3719 	switch (adev->asic_type) {
3720 	default:
3721 		adev->gfx.config.double_offchip_lds_buf = 1;
3722 		break;
3723 	case CHIP_CARRIZO:
3724 	case CHIP_STONEY:
3725 		adev->gfx.config.double_offchip_lds_buf = 0;
3726 		break;
3727 	}
3728 }
3729 
3730 static void gfx_v8_0_constants_init(struct amdgpu_device *adev)
3731 {
3732 	u32 tmp, sh_static_mem_cfg;
3733 	int i;
3734 
3735 	WREG32_FIELD(GRBM_CNTL, READ_TIMEOUT, 0xFF);
3736 	WREG32(mmGB_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
3737 	WREG32(mmHDP_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
3738 	WREG32(mmDMIF_ADDR_CALC, adev->gfx.config.gb_addr_config);
3739 
3740 	gfx_v8_0_tiling_mode_table_init(adev);
3741 	gfx_v8_0_setup_rb(adev);
3742 	gfx_v8_0_get_cu_info(adev);
3743 	gfx_v8_0_config_init(adev);
3744 
3745 	/* XXX SH_MEM regs */
3746 	/* where to put LDS, scratch, GPUVM in FSA64 space */
3747 	sh_static_mem_cfg = REG_SET_FIELD(0, SH_STATIC_MEM_CONFIG,
3748 				   SWIZZLE_ENABLE, 1);
3749 	sh_static_mem_cfg = REG_SET_FIELD(sh_static_mem_cfg, SH_STATIC_MEM_CONFIG,
3750 				   ELEMENT_SIZE, 1);
3751 	sh_static_mem_cfg = REG_SET_FIELD(sh_static_mem_cfg, SH_STATIC_MEM_CONFIG,
3752 				   INDEX_STRIDE, 3);
3753 	WREG32(mmSH_STATIC_MEM_CONFIG, sh_static_mem_cfg);
3754 
3755 	mutex_lock(&adev->srbm_mutex);
3756 	for (i = 0; i < adev->vm_manager.id_mgr[0].num_ids; i++) {
3757 		vi_srbm_select(adev, 0, 0, 0, i);
3758 		/* CP and shaders */
3759 		if (i == 0) {
3760 			tmp = REG_SET_FIELD(0, SH_MEM_CONFIG, DEFAULT_MTYPE, MTYPE_UC);
3761 			tmp = REG_SET_FIELD(tmp, SH_MEM_CONFIG, APE1_MTYPE, MTYPE_UC);
3762 			tmp = REG_SET_FIELD(tmp, SH_MEM_CONFIG, ALIGNMENT_MODE,
3763 					    SH_MEM_ALIGNMENT_MODE_UNALIGNED);
3764 			WREG32(mmSH_MEM_CONFIG, tmp);
3765 			WREG32(mmSH_MEM_BASES, 0);
3766 		} else {
3767 			tmp = REG_SET_FIELD(0, SH_MEM_CONFIG, DEFAULT_MTYPE, MTYPE_NC);
3768 			tmp = REG_SET_FIELD(tmp, SH_MEM_CONFIG, APE1_MTYPE, MTYPE_UC);
3769 			tmp = REG_SET_FIELD(tmp, SH_MEM_CONFIG, ALIGNMENT_MODE,
3770 					    SH_MEM_ALIGNMENT_MODE_UNALIGNED);
3771 			WREG32(mmSH_MEM_CONFIG, tmp);
3772 			tmp = adev->gmc.shared_aperture_start >> 48;
3773 			WREG32(mmSH_MEM_BASES, tmp);
3774 		}
3775 
3776 		WREG32(mmSH_MEM_APE1_BASE, 1);
3777 		WREG32(mmSH_MEM_APE1_LIMIT, 0);
3778 	}
3779 	vi_srbm_select(adev, 0, 0, 0, 0);
3780 	mutex_unlock(&adev->srbm_mutex);
3781 
3782 	gfx_v8_0_init_compute_vmid(adev);
3783 	gfx_v8_0_init_gds_vmid(adev);
3784 
3785 	mutex_lock(&adev->grbm_idx_mutex);
3786 	/*
3787 	 * making sure that the following register writes will be broadcasted
3788 	 * to all the shaders
3789 	 */
3790 	gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
3791 
3792 	WREG32(mmPA_SC_FIFO_SIZE,
3793 		   (adev->gfx.config.sc_prim_fifo_size_frontend <<
3794 			PA_SC_FIFO_SIZE__SC_FRONTEND_PRIM_FIFO_SIZE__SHIFT) |
3795 		   (adev->gfx.config.sc_prim_fifo_size_backend <<
3796 			PA_SC_FIFO_SIZE__SC_BACKEND_PRIM_FIFO_SIZE__SHIFT) |
3797 		   (adev->gfx.config.sc_hiz_tile_fifo_size <<
3798 			PA_SC_FIFO_SIZE__SC_HIZ_TILE_FIFO_SIZE__SHIFT) |
3799 		   (adev->gfx.config.sc_earlyz_tile_fifo_size <<
3800 			PA_SC_FIFO_SIZE__SC_EARLYZ_TILE_FIFO_SIZE__SHIFT));
3801 
3802 	tmp = RREG32(mmSPI_ARB_PRIORITY);
3803 	tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS0, 2);
3804 	tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS1, 2);
3805 	tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS2, 2);
3806 	tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS3, 2);
3807 	WREG32(mmSPI_ARB_PRIORITY, tmp);
3808 
3809 	mutex_unlock(&adev->grbm_idx_mutex);
3810 
3811 }
3812 
3813 static void gfx_v8_0_wait_for_rlc_serdes(struct amdgpu_device *adev)
3814 {
3815 	u32 i, j, k;
3816 	u32 mask;
3817 
3818 	mutex_lock(&adev->grbm_idx_mutex);
3819 	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
3820 		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
3821 			gfx_v8_0_select_se_sh(adev, i, j, 0xffffffff);
3822 			for (k = 0; k < adev->usec_timeout; k++) {
3823 				if (RREG32(mmRLC_SERDES_CU_MASTER_BUSY) == 0)
3824 					break;
3825 				udelay(1);
3826 			}
3827 			if (k == adev->usec_timeout) {
3828 				gfx_v8_0_select_se_sh(adev, 0xffffffff,
3829 						      0xffffffff, 0xffffffff);
3830 				mutex_unlock(&adev->grbm_idx_mutex);
3831 				DRM_INFO("Timeout wait for RLC serdes %u,%u\n",
3832 					 i, j);
3833 				return;
3834 			}
3835 		}
3836 	}
3837 	gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
3838 	mutex_unlock(&adev->grbm_idx_mutex);
3839 
3840 	mask = RLC_SERDES_NONCU_MASTER_BUSY__SE_MASTER_BUSY_MASK |
3841 		RLC_SERDES_NONCU_MASTER_BUSY__GC_MASTER_BUSY_MASK |
3842 		RLC_SERDES_NONCU_MASTER_BUSY__TC0_MASTER_BUSY_MASK |
3843 		RLC_SERDES_NONCU_MASTER_BUSY__TC1_MASTER_BUSY_MASK;
3844 	for (k = 0; k < adev->usec_timeout; k++) {
3845 		if ((RREG32(mmRLC_SERDES_NONCU_MASTER_BUSY) & mask) == 0)
3846 			break;
3847 		udelay(1);
3848 	}
3849 }
3850 
3851 static void gfx_v8_0_enable_gui_idle_interrupt(struct amdgpu_device *adev,
3852 					       bool enable)
3853 {
3854 	u32 tmp = RREG32(mmCP_INT_CNTL_RING0);
3855 
3856 	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CNTX_BUSY_INT_ENABLE, enable ? 1 : 0);
3857 	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CNTX_EMPTY_INT_ENABLE, enable ? 1 : 0);
3858 	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CMP_BUSY_INT_ENABLE, enable ? 1 : 0);
3859 	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, GFX_IDLE_INT_ENABLE, enable ? 1 : 0);
3860 
3861 	WREG32(mmCP_INT_CNTL_RING0, tmp);
3862 }
3863 
3864 static void gfx_v8_0_init_csb(struct amdgpu_device *adev)
3865 {
3866 	adev->gfx.rlc.funcs->get_csb_buffer(adev, adev->gfx.rlc.cs_ptr);
3867 	/* csib */
3868 	WREG32(mmRLC_CSIB_ADDR_HI,
3869 			adev->gfx.rlc.clear_state_gpu_addr >> 32);
3870 	WREG32(mmRLC_CSIB_ADDR_LO,
3871 			adev->gfx.rlc.clear_state_gpu_addr & 0xfffffffc);
3872 	WREG32(mmRLC_CSIB_LENGTH,
3873 			adev->gfx.rlc.clear_state_size);
3874 }
3875 
3876 static void gfx_v8_0_parse_ind_reg_list(int *register_list_format,
3877 				int ind_offset,
3878 				int list_size,
3879 				int *unique_indices,
3880 				int *indices_count,
3881 				int max_indices,
3882 				int *ind_start_offsets,
3883 				int *offset_count,
3884 				int max_offset)
3885 {
3886 	int indices;
3887 	bool new_entry = true;
3888 
3889 	for (; ind_offset < list_size; ind_offset++) {
3890 
3891 		if (new_entry) {
3892 			new_entry = false;
3893 			ind_start_offsets[*offset_count] = ind_offset;
3894 			*offset_count = *offset_count + 1;
3895 			BUG_ON(*offset_count >= max_offset);
3896 		}
3897 
3898 		if (register_list_format[ind_offset] == 0xFFFFFFFF) {
3899 			new_entry = true;
3900 			continue;
3901 		}
3902 
3903 		ind_offset += 2;
3904 
3905 		/* look for the matching indice */
3906 		for (indices = 0;
3907 			indices < *indices_count;
3908 			indices++) {
3909 			if (unique_indices[indices] ==
3910 				register_list_format[ind_offset])
3911 				break;
3912 		}
3913 
3914 		if (indices >= *indices_count) {
3915 			unique_indices[*indices_count] =
3916 				register_list_format[ind_offset];
3917 			indices = *indices_count;
3918 			*indices_count = *indices_count + 1;
3919 			BUG_ON(*indices_count >= max_indices);
3920 		}
3921 
3922 		register_list_format[ind_offset] = indices;
3923 	}
3924 }
3925 
3926 static int gfx_v8_0_init_save_restore_list(struct amdgpu_device *adev)
3927 {
3928 	int i, temp, data;
3929 	int unique_indices[] = {0, 0, 0, 0, 0, 0, 0, 0};
3930 	int indices_count = 0;
3931 	int indirect_start_offsets[] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0};
3932 	int offset_count = 0;
3933 
3934 	int list_size;
3935 	unsigned int *register_list_format =
3936 		kmemdup(adev->gfx.rlc.register_list_format,
3937 			adev->gfx.rlc.reg_list_format_size_bytes, GFP_KERNEL);
3938 	if (!register_list_format)
3939 		return -ENOMEM;
3940 
3941 	gfx_v8_0_parse_ind_reg_list(register_list_format,
3942 				RLC_FormatDirectRegListLength,
3943 				adev->gfx.rlc.reg_list_format_size_bytes >> 2,
3944 				unique_indices,
3945 				&indices_count,
3946 				ARRAY_SIZE(unique_indices),
3947 				indirect_start_offsets,
3948 				&offset_count,
3949 				ARRAY_SIZE(indirect_start_offsets));
3950 
3951 	/* save and restore list */
3952 	WREG32_FIELD(RLC_SRM_CNTL, AUTO_INCR_ADDR, 1);
3953 
3954 	WREG32(mmRLC_SRM_ARAM_ADDR, 0);
3955 	for (i = 0; i < adev->gfx.rlc.reg_list_size_bytes >> 2; i++)
3956 		WREG32(mmRLC_SRM_ARAM_DATA, adev->gfx.rlc.register_restore[i]);
3957 
3958 	/* indirect list */
3959 	WREG32(mmRLC_GPM_SCRATCH_ADDR, adev->gfx.rlc.reg_list_format_start);
3960 	for (i = 0; i < adev->gfx.rlc.reg_list_format_size_bytes >> 2; i++)
3961 		WREG32(mmRLC_GPM_SCRATCH_DATA, register_list_format[i]);
3962 
3963 	list_size = adev->gfx.rlc.reg_list_size_bytes >> 2;
3964 	list_size = list_size >> 1;
3965 	WREG32(mmRLC_GPM_SCRATCH_ADDR, adev->gfx.rlc.reg_restore_list_size);
3966 	WREG32(mmRLC_GPM_SCRATCH_DATA, list_size);
3967 
3968 	/* starting offsets starts */
3969 	WREG32(mmRLC_GPM_SCRATCH_ADDR,
3970 		adev->gfx.rlc.starting_offsets_start);
3971 	for (i = 0; i < ARRAY_SIZE(indirect_start_offsets); i++)
3972 		WREG32(mmRLC_GPM_SCRATCH_DATA,
3973 				indirect_start_offsets[i]);
3974 
3975 	/* unique indices */
3976 	temp = mmRLC_SRM_INDEX_CNTL_ADDR_0;
3977 	data = mmRLC_SRM_INDEX_CNTL_DATA_0;
3978 	for (i = 0; i < ARRAY_SIZE(unique_indices); i++) {
3979 		if (unique_indices[i] != 0) {
3980 			WREG32(temp + i, unique_indices[i] & 0x3FFFF);
3981 			WREG32(data + i, unique_indices[i] >> 20);
3982 		}
3983 	}
3984 	kfree(register_list_format);
3985 
3986 	return 0;
3987 }
3988 
3989 static void gfx_v8_0_enable_save_restore_machine(struct amdgpu_device *adev)
3990 {
3991 	WREG32_FIELD(RLC_SRM_CNTL, SRM_ENABLE, 1);
3992 }
3993 
3994 static void gfx_v8_0_init_power_gating(struct amdgpu_device *adev)
3995 {
3996 	uint32_t data;
3997 
3998 	WREG32_FIELD(CP_RB_WPTR_POLL_CNTL, IDLE_POLL_COUNT, 0x60);
3999 
4000 	data = REG_SET_FIELD(0, RLC_PG_DELAY, POWER_UP_DELAY, 0x10);
4001 	data = REG_SET_FIELD(data, RLC_PG_DELAY, POWER_DOWN_DELAY, 0x10);
4002 	data = REG_SET_FIELD(data, RLC_PG_DELAY, CMD_PROPAGATE_DELAY, 0x10);
4003 	data = REG_SET_FIELD(data, RLC_PG_DELAY, MEM_SLEEP_DELAY, 0x10);
4004 	WREG32(mmRLC_PG_DELAY, data);
4005 
4006 	WREG32_FIELD(RLC_PG_DELAY_2, SERDES_CMD_DELAY, 0x3);
4007 	WREG32_FIELD(RLC_AUTO_PG_CTRL, GRBM_REG_SAVE_GFX_IDLE_THRESHOLD, 0x55f0);
4008 
4009 }
4010 
4011 static void cz_enable_sck_slow_down_on_power_up(struct amdgpu_device *adev,
4012 						bool enable)
4013 {
4014 	WREG32_FIELD(RLC_PG_CNTL, SMU_CLK_SLOWDOWN_ON_PU_ENABLE, enable ? 1 : 0);
4015 }
4016 
4017 static void cz_enable_sck_slow_down_on_power_down(struct amdgpu_device *adev,
4018 						  bool enable)
4019 {
4020 	WREG32_FIELD(RLC_PG_CNTL, SMU_CLK_SLOWDOWN_ON_PD_ENABLE, enable ? 1 : 0);
4021 }
4022 
4023 static void cz_enable_cp_power_gating(struct amdgpu_device *adev, bool enable)
4024 {
4025 	WREG32_FIELD(RLC_PG_CNTL, CP_PG_DISABLE, enable ? 0 : 1);
4026 }
4027 
4028 static void gfx_v8_0_init_pg(struct amdgpu_device *adev)
4029 {
4030 	if ((adev->asic_type == CHIP_CARRIZO) ||
4031 	    (adev->asic_type == CHIP_STONEY)) {
4032 		gfx_v8_0_init_csb(adev);
4033 		gfx_v8_0_init_save_restore_list(adev);
4034 		gfx_v8_0_enable_save_restore_machine(adev);
4035 		WREG32(mmRLC_JUMP_TABLE_RESTORE, adev->gfx.rlc.cp_table_gpu_addr >> 8);
4036 		gfx_v8_0_init_power_gating(adev);
4037 		WREG32(mmRLC_PG_ALWAYS_ON_CU_MASK, adev->gfx.cu_info.ao_cu_mask);
4038 	} else if ((adev->asic_type == CHIP_POLARIS11) ||
4039 		   (adev->asic_type == CHIP_POLARIS12) ||
4040 		   (adev->asic_type == CHIP_VEGAM)) {
4041 		gfx_v8_0_init_csb(adev);
4042 		gfx_v8_0_init_save_restore_list(adev);
4043 		gfx_v8_0_enable_save_restore_machine(adev);
4044 		gfx_v8_0_init_power_gating(adev);
4045 	}
4046 
4047 }
4048 
4049 static void gfx_v8_0_rlc_stop(struct amdgpu_device *adev)
4050 {
4051 	WREG32_FIELD(RLC_CNTL, RLC_ENABLE_F32, 0);
4052 
4053 	gfx_v8_0_enable_gui_idle_interrupt(adev, false);
4054 	gfx_v8_0_wait_for_rlc_serdes(adev);
4055 }
4056 
4057 static void gfx_v8_0_rlc_reset(struct amdgpu_device *adev)
4058 {
4059 	WREG32_FIELD(GRBM_SOFT_RESET, SOFT_RESET_RLC, 1);
4060 	udelay(50);
4061 
4062 	WREG32_FIELD(GRBM_SOFT_RESET, SOFT_RESET_RLC, 0);
4063 	udelay(50);
4064 }
4065 
4066 static void gfx_v8_0_rlc_start(struct amdgpu_device *adev)
4067 {
4068 	WREG32_FIELD(RLC_CNTL, RLC_ENABLE_F32, 1);
4069 
4070 	/* carrizo do enable cp interrupt after cp inited */
4071 	if (!(adev->flags & AMD_IS_APU))
4072 		gfx_v8_0_enable_gui_idle_interrupt(adev, true);
4073 
4074 	udelay(50);
4075 }
4076 
4077 static int gfx_v8_0_rlc_resume(struct amdgpu_device *adev)
4078 {
4079 	if (amdgpu_sriov_vf(adev)) {
4080 		gfx_v8_0_init_csb(adev);
4081 		return 0;
4082 	}
4083 
4084 	adev->gfx.rlc.funcs->stop(adev);
4085 	adev->gfx.rlc.funcs->reset(adev);
4086 	gfx_v8_0_init_pg(adev);
4087 	adev->gfx.rlc.funcs->start(adev);
4088 
4089 	return 0;
4090 }
4091 
4092 static void gfx_v8_0_cp_gfx_enable(struct amdgpu_device *adev, bool enable)
4093 {
4094 	u32 tmp = RREG32(mmCP_ME_CNTL);
4095 
4096 	if (enable) {
4097 		tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, ME_HALT, 0);
4098 		tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, PFP_HALT, 0);
4099 		tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, CE_HALT, 0);
4100 	} else {
4101 		tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, ME_HALT, 1);
4102 		tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, PFP_HALT, 1);
4103 		tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, CE_HALT, 1);
4104 	}
4105 	WREG32(mmCP_ME_CNTL, tmp);
4106 	udelay(50);
4107 }
4108 
4109 static u32 gfx_v8_0_get_csb_size(struct amdgpu_device *adev)
4110 {
4111 	u32 count = 0;
4112 	const struct cs_section_def *sect = NULL;
4113 	const struct cs_extent_def *ext = NULL;
4114 
4115 	/* begin clear state */
4116 	count += 2;
4117 	/* context control state */
4118 	count += 3;
4119 
4120 	for (sect = vi_cs_data; sect->section != NULL; ++sect) {
4121 		for (ext = sect->section; ext->extent != NULL; ++ext) {
4122 			if (sect->id == SECT_CONTEXT)
4123 				count += 2 + ext->reg_count;
4124 			else
4125 				return 0;
4126 		}
4127 	}
4128 	/* pa_sc_raster_config/pa_sc_raster_config1 */
4129 	count += 4;
4130 	/* end clear state */
4131 	count += 2;
4132 	/* clear state */
4133 	count += 2;
4134 
4135 	return count;
4136 }
4137 
4138 static int gfx_v8_0_cp_gfx_start(struct amdgpu_device *adev)
4139 {
4140 	struct amdgpu_ring *ring = &adev->gfx.gfx_ring[0];
4141 	const struct cs_section_def *sect = NULL;
4142 	const struct cs_extent_def *ext = NULL;
4143 	int r, i;
4144 
4145 	/* init the CP */
4146 	WREG32(mmCP_MAX_CONTEXT, adev->gfx.config.max_hw_contexts - 1);
4147 	WREG32(mmCP_ENDIAN_SWAP, 0);
4148 	WREG32(mmCP_DEVICE_ID, 1);
4149 
4150 	gfx_v8_0_cp_gfx_enable(adev, true);
4151 
4152 	r = amdgpu_ring_alloc(ring, gfx_v8_0_get_csb_size(adev) + 4);
4153 	if (r) {
4154 		DRM_ERROR("amdgpu: cp failed to lock ring (%d).\n", r);
4155 		return r;
4156 	}
4157 
4158 	/* clear state buffer */
4159 	amdgpu_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
4160 	amdgpu_ring_write(ring, PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
4161 
4162 	amdgpu_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1));
4163 	amdgpu_ring_write(ring, 0x80000000);
4164 	amdgpu_ring_write(ring, 0x80000000);
4165 
4166 	for (sect = vi_cs_data; sect->section != NULL; ++sect) {
4167 		for (ext = sect->section; ext->extent != NULL; ++ext) {
4168 			if (sect->id == SECT_CONTEXT) {
4169 				amdgpu_ring_write(ring,
4170 				       PACKET3(PACKET3_SET_CONTEXT_REG,
4171 					       ext->reg_count));
4172 				amdgpu_ring_write(ring,
4173 				       ext->reg_index - PACKET3_SET_CONTEXT_REG_START);
4174 				for (i = 0; i < ext->reg_count; i++)
4175 					amdgpu_ring_write(ring, ext->extent[i]);
4176 			}
4177 		}
4178 	}
4179 
4180 	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONTEXT_REG, 2));
4181 	amdgpu_ring_write(ring, mmPA_SC_RASTER_CONFIG - PACKET3_SET_CONTEXT_REG_START);
4182 	amdgpu_ring_write(ring, adev->gfx.config.rb_config[0][0].raster_config);
4183 	amdgpu_ring_write(ring, adev->gfx.config.rb_config[0][0].raster_config_1);
4184 
4185 	amdgpu_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
4186 	amdgpu_ring_write(ring, PACKET3_PREAMBLE_END_CLEAR_STATE);
4187 
4188 	amdgpu_ring_write(ring, PACKET3(PACKET3_CLEAR_STATE, 0));
4189 	amdgpu_ring_write(ring, 0);
4190 
4191 	/* init the CE partitions */
4192 	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_BASE, 2));
4193 	amdgpu_ring_write(ring, PACKET3_BASE_INDEX(CE_PARTITION_BASE));
4194 	amdgpu_ring_write(ring, 0x8000);
4195 	amdgpu_ring_write(ring, 0x8000);
4196 
4197 	amdgpu_ring_commit(ring);
4198 
4199 	return 0;
4200 }
4201 static void gfx_v8_0_set_cpg_door_bell(struct amdgpu_device *adev, struct amdgpu_ring *ring)
4202 {
4203 	u32 tmp;
4204 	/* no gfx doorbells on iceland */
4205 	if (adev->asic_type == CHIP_TOPAZ)
4206 		return;
4207 
4208 	tmp = RREG32(mmCP_RB_DOORBELL_CONTROL);
4209 
4210 	if (ring->use_doorbell) {
4211 		tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL,
4212 				DOORBELL_OFFSET, ring->doorbell_index);
4213 		tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL,
4214 						DOORBELL_HIT, 0);
4215 		tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL,
4216 					    DOORBELL_EN, 1);
4217 	} else {
4218 		tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL, DOORBELL_EN, 0);
4219 	}
4220 
4221 	WREG32(mmCP_RB_DOORBELL_CONTROL, tmp);
4222 
4223 	if (adev->flags & AMD_IS_APU)
4224 		return;
4225 
4226 	tmp = REG_SET_FIELD(0, CP_RB_DOORBELL_RANGE_LOWER,
4227 					DOORBELL_RANGE_LOWER,
4228 					adev->doorbell_index.gfx_ring0);
4229 	WREG32(mmCP_RB_DOORBELL_RANGE_LOWER, tmp);
4230 
4231 	WREG32(mmCP_RB_DOORBELL_RANGE_UPPER,
4232 		CP_RB_DOORBELL_RANGE_UPPER__DOORBELL_RANGE_UPPER_MASK);
4233 }
4234 
4235 static int gfx_v8_0_cp_gfx_resume(struct amdgpu_device *adev)
4236 {
4237 	struct amdgpu_ring *ring;
4238 	u32 tmp;
4239 	u32 rb_bufsz;
4240 	u64 rb_addr, rptr_addr, wptr_gpu_addr;
4241 
4242 	/* Set the write pointer delay */
4243 	WREG32(mmCP_RB_WPTR_DELAY, 0);
4244 
4245 	/* set the RB to use vmid 0 */
4246 	WREG32(mmCP_RB_VMID, 0);
4247 
4248 	/* Set ring buffer size */
4249 	ring = &adev->gfx.gfx_ring[0];
4250 	rb_bufsz = order_base_2(ring->ring_size / 8);
4251 	tmp = REG_SET_FIELD(0, CP_RB0_CNTL, RB_BUFSZ, rb_bufsz);
4252 	tmp = REG_SET_FIELD(tmp, CP_RB0_CNTL, RB_BLKSZ, rb_bufsz - 2);
4253 	tmp = REG_SET_FIELD(tmp, CP_RB0_CNTL, MTYPE, 3);
4254 	tmp = REG_SET_FIELD(tmp, CP_RB0_CNTL, MIN_IB_AVAILSZ, 1);
4255 #ifdef __BIG_ENDIAN
4256 	tmp = REG_SET_FIELD(tmp, CP_RB0_CNTL, BUF_SWAP, 1);
4257 #endif
4258 	WREG32(mmCP_RB0_CNTL, tmp);
4259 
4260 	/* Initialize the ring buffer's read and write pointers */
4261 	WREG32(mmCP_RB0_CNTL, tmp | CP_RB0_CNTL__RB_RPTR_WR_ENA_MASK);
4262 	ring->wptr = 0;
4263 	WREG32(mmCP_RB0_WPTR, lower_32_bits(ring->wptr));
4264 
4265 	/* set the wb address wether it's enabled or not */
4266 	rptr_addr = ring->rptr_gpu_addr;
4267 	WREG32(mmCP_RB0_RPTR_ADDR, lower_32_bits(rptr_addr));
4268 	WREG32(mmCP_RB0_RPTR_ADDR_HI, upper_32_bits(rptr_addr) & 0xFF);
4269 
4270 	wptr_gpu_addr = ring->wptr_gpu_addr;
4271 	WREG32(mmCP_RB_WPTR_POLL_ADDR_LO, lower_32_bits(wptr_gpu_addr));
4272 	WREG32(mmCP_RB_WPTR_POLL_ADDR_HI, upper_32_bits(wptr_gpu_addr));
4273 	mdelay(1);
4274 	WREG32(mmCP_RB0_CNTL, tmp);
4275 
4276 	rb_addr = ring->gpu_addr >> 8;
4277 	WREG32(mmCP_RB0_BASE, rb_addr);
4278 	WREG32(mmCP_RB0_BASE_HI, upper_32_bits(rb_addr));
4279 
4280 	gfx_v8_0_set_cpg_door_bell(adev, ring);
4281 	/* start the ring */
4282 	amdgpu_ring_clear_ring(ring);
4283 	gfx_v8_0_cp_gfx_start(adev);
4284 	ring->sched.ready = true;
4285 
4286 	return 0;
4287 }
4288 
4289 static void gfx_v8_0_cp_compute_enable(struct amdgpu_device *adev, bool enable)
4290 {
4291 	if (enable) {
4292 		WREG32(mmCP_MEC_CNTL, 0);
4293 	} else {
4294 		WREG32(mmCP_MEC_CNTL, (CP_MEC_CNTL__MEC_ME1_HALT_MASK | CP_MEC_CNTL__MEC_ME2_HALT_MASK));
4295 		adev->gfx.kiq.ring.sched.ready = false;
4296 	}
4297 	udelay(50);
4298 }
4299 
4300 /* KIQ functions */
4301 static void gfx_v8_0_kiq_setting(struct amdgpu_ring *ring)
4302 {
4303 	uint32_t tmp;
4304 	struct amdgpu_device *adev = ring->adev;
4305 
4306 	/* tell RLC which is KIQ queue */
4307 	tmp = RREG32(mmRLC_CP_SCHEDULERS);
4308 	tmp &= 0xffffff00;
4309 	tmp |= (ring->me << 5) | (ring->pipe << 3) | (ring->queue);
4310 	WREG32(mmRLC_CP_SCHEDULERS, tmp);
4311 	tmp |= 0x80;
4312 	WREG32(mmRLC_CP_SCHEDULERS, tmp);
4313 }
4314 
4315 static int gfx_v8_0_kiq_kcq_enable(struct amdgpu_device *adev)
4316 {
4317 	struct amdgpu_ring *kiq_ring = &adev->gfx.kiq.ring;
4318 	uint64_t queue_mask = 0;
4319 	int r, i;
4320 
4321 	for (i = 0; i < AMDGPU_MAX_COMPUTE_QUEUES; ++i) {
4322 		if (!test_bit(i, adev->gfx.mec.queue_bitmap))
4323 			continue;
4324 
4325 		/* This situation may be hit in the future if a new HW
4326 		 * generation exposes more than 64 queues. If so, the
4327 		 * definition of queue_mask needs updating */
4328 		if (WARN_ON(i >= (sizeof(queue_mask)*8))) {
4329 			DRM_ERROR("Invalid KCQ enabled: %d\n", i);
4330 			break;
4331 		}
4332 
4333 		queue_mask |= (1ull << i);
4334 	}
4335 
4336 	r = amdgpu_ring_alloc(kiq_ring, (8 * adev->gfx.num_compute_rings) + 8);
4337 	if (r) {
4338 		DRM_ERROR("Failed to lock KIQ (%d).\n", r);
4339 		return r;
4340 	}
4341 	/* set resources */
4342 	amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_SET_RESOURCES, 6));
4343 	amdgpu_ring_write(kiq_ring, 0);	/* vmid_mask:0 queue_type:0 (KIQ) */
4344 	amdgpu_ring_write(kiq_ring, lower_32_bits(queue_mask));	/* queue mask lo */
4345 	amdgpu_ring_write(kiq_ring, upper_32_bits(queue_mask));	/* queue mask hi */
4346 	amdgpu_ring_write(kiq_ring, 0);	/* gws mask lo */
4347 	amdgpu_ring_write(kiq_ring, 0);	/* gws mask hi */
4348 	amdgpu_ring_write(kiq_ring, 0);	/* oac mask */
4349 	amdgpu_ring_write(kiq_ring, 0);	/* gds heap base:0, gds heap size:0 */
4350 	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
4351 		struct amdgpu_ring *ring = &adev->gfx.compute_ring[i];
4352 		uint64_t mqd_addr = amdgpu_bo_gpu_offset(ring->mqd_obj);
4353 		uint64_t wptr_addr = ring->wptr_gpu_addr;
4354 
4355 		/* map queues */
4356 		amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_MAP_QUEUES, 5));
4357 		/* Q_sel:0, vmid:0, vidmem: 1, engine:0, num_Q:1*/
4358 		amdgpu_ring_write(kiq_ring,
4359 				  PACKET3_MAP_QUEUES_NUM_QUEUES(1));
4360 		amdgpu_ring_write(kiq_ring,
4361 				  PACKET3_MAP_QUEUES_DOORBELL_OFFSET(ring->doorbell_index) |
4362 				  PACKET3_MAP_QUEUES_QUEUE(ring->queue) |
4363 				  PACKET3_MAP_QUEUES_PIPE(ring->pipe) |
4364 				  PACKET3_MAP_QUEUES_ME(ring->me == 1 ? 0 : 1)); /* doorbell */
4365 		amdgpu_ring_write(kiq_ring, lower_32_bits(mqd_addr));
4366 		amdgpu_ring_write(kiq_ring, upper_32_bits(mqd_addr));
4367 		amdgpu_ring_write(kiq_ring, lower_32_bits(wptr_addr));
4368 		amdgpu_ring_write(kiq_ring, upper_32_bits(wptr_addr));
4369 	}
4370 
4371 	amdgpu_ring_commit(kiq_ring);
4372 
4373 	return 0;
4374 }
4375 
4376 static int gfx_v8_0_deactivate_hqd(struct amdgpu_device *adev, u32 req)
4377 {
4378 	int i, r = 0;
4379 
4380 	if (RREG32(mmCP_HQD_ACTIVE) & CP_HQD_ACTIVE__ACTIVE_MASK) {
4381 		WREG32_FIELD(CP_HQD_DEQUEUE_REQUEST, DEQUEUE_REQ, req);
4382 		for (i = 0; i < adev->usec_timeout; i++) {
4383 			if (!(RREG32(mmCP_HQD_ACTIVE) & CP_HQD_ACTIVE__ACTIVE_MASK))
4384 				break;
4385 			udelay(1);
4386 		}
4387 		if (i == adev->usec_timeout)
4388 			r = -ETIMEDOUT;
4389 	}
4390 	WREG32(mmCP_HQD_DEQUEUE_REQUEST, 0);
4391 	WREG32(mmCP_HQD_PQ_RPTR, 0);
4392 	WREG32(mmCP_HQD_PQ_WPTR, 0);
4393 
4394 	return r;
4395 }
4396 
4397 static void gfx_v8_0_mqd_set_priority(struct amdgpu_ring *ring, struct vi_mqd *mqd)
4398 {
4399 	struct amdgpu_device *adev = ring->adev;
4400 
4401 	if (ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE) {
4402 		if (amdgpu_gfx_is_high_priority_compute_queue(adev, ring)) {
4403 			mqd->cp_hqd_pipe_priority = AMDGPU_GFX_PIPE_PRIO_HIGH;
4404 			mqd->cp_hqd_queue_priority =
4405 				AMDGPU_GFX_QUEUE_PRIORITY_MAXIMUM;
4406 		}
4407 	}
4408 }
4409 
4410 static int gfx_v8_0_mqd_init(struct amdgpu_ring *ring)
4411 {
4412 	struct amdgpu_device *adev = ring->adev;
4413 	struct vi_mqd *mqd = ring->mqd_ptr;
4414 	uint64_t hqd_gpu_addr, wb_gpu_addr, eop_base_addr;
4415 	uint32_t tmp;
4416 
4417 	mqd->header = 0xC0310800;
4418 	mqd->compute_pipelinestat_enable = 0x00000001;
4419 	mqd->compute_static_thread_mgmt_se0 = 0xffffffff;
4420 	mqd->compute_static_thread_mgmt_se1 = 0xffffffff;
4421 	mqd->compute_static_thread_mgmt_se2 = 0xffffffff;
4422 	mqd->compute_static_thread_mgmt_se3 = 0xffffffff;
4423 	mqd->compute_misc_reserved = 0x00000003;
4424 	mqd->dynamic_cu_mask_addr_lo = lower_32_bits(ring->mqd_gpu_addr
4425 						     + offsetof(struct vi_mqd_allocation, dynamic_cu_mask));
4426 	mqd->dynamic_cu_mask_addr_hi = upper_32_bits(ring->mqd_gpu_addr
4427 						     + offsetof(struct vi_mqd_allocation, dynamic_cu_mask));
4428 	eop_base_addr = ring->eop_gpu_addr >> 8;
4429 	mqd->cp_hqd_eop_base_addr_lo = eop_base_addr;
4430 	mqd->cp_hqd_eop_base_addr_hi = upper_32_bits(eop_base_addr);
4431 
4432 	/* set the EOP size, register value is 2^(EOP_SIZE+1) dwords */
4433 	tmp = RREG32(mmCP_HQD_EOP_CONTROL);
4434 	tmp = REG_SET_FIELD(tmp, CP_HQD_EOP_CONTROL, EOP_SIZE,
4435 			(order_base_2(GFX8_MEC_HPD_SIZE / 4) - 1));
4436 
4437 	mqd->cp_hqd_eop_control = tmp;
4438 
4439 	/* enable doorbell? */
4440 	tmp = REG_SET_FIELD(RREG32(mmCP_HQD_PQ_DOORBELL_CONTROL),
4441 			    CP_HQD_PQ_DOORBELL_CONTROL,
4442 			    DOORBELL_EN,
4443 			    ring->use_doorbell ? 1 : 0);
4444 
4445 	mqd->cp_hqd_pq_doorbell_control = tmp;
4446 
4447 	/* set the pointer to the MQD */
4448 	mqd->cp_mqd_base_addr_lo = ring->mqd_gpu_addr & 0xfffffffc;
4449 	mqd->cp_mqd_base_addr_hi = upper_32_bits(ring->mqd_gpu_addr);
4450 
4451 	/* set MQD vmid to 0 */
4452 	tmp = RREG32(mmCP_MQD_CONTROL);
4453 	tmp = REG_SET_FIELD(tmp, CP_MQD_CONTROL, VMID, 0);
4454 	mqd->cp_mqd_control = tmp;
4455 
4456 	/* set the pointer to the HQD, this is similar CP_RB0_BASE/_HI */
4457 	hqd_gpu_addr = ring->gpu_addr >> 8;
4458 	mqd->cp_hqd_pq_base_lo = hqd_gpu_addr;
4459 	mqd->cp_hqd_pq_base_hi = upper_32_bits(hqd_gpu_addr);
4460 
4461 	/* set up the HQD, this is similar to CP_RB0_CNTL */
4462 	tmp = RREG32(mmCP_HQD_PQ_CONTROL);
4463 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, QUEUE_SIZE,
4464 			    (order_base_2(ring->ring_size / 4) - 1));
4465 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, RPTR_BLOCK_SIZE,
4466 			(order_base_2(AMDGPU_GPU_PAGE_SIZE / 4) - 1));
4467 #ifdef __BIG_ENDIAN
4468 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, ENDIAN_SWAP, 1);
4469 #endif
4470 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, UNORD_DISPATCH, 0);
4471 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, ROQ_PQ_IB_FLIP, 0);
4472 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, PRIV_STATE, 1);
4473 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, KMD_QUEUE, 1);
4474 	mqd->cp_hqd_pq_control = tmp;
4475 
4476 	/* set the wb address whether it's enabled or not */
4477 	wb_gpu_addr = ring->rptr_gpu_addr;
4478 	mqd->cp_hqd_pq_rptr_report_addr_lo = wb_gpu_addr & 0xfffffffc;
4479 	mqd->cp_hqd_pq_rptr_report_addr_hi =
4480 		upper_32_bits(wb_gpu_addr) & 0xffff;
4481 
4482 	/* only used if CP_PQ_WPTR_POLL_CNTL.CP_PQ_WPTR_POLL_CNTL__EN_MASK=1 */
4483 	wb_gpu_addr = ring->wptr_gpu_addr;
4484 	mqd->cp_hqd_pq_wptr_poll_addr_lo = wb_gpu_addr & 0xfffffffc;
4485 	mqd->cp_hqd_pq_wptr_poll_addr_hi = upper_32_bits(wb_gpu_addr) & 0xffff;
4486 
4487 	tmp = 0;
4488 	/* enable the doorbell if requested */
4489 	if (ring->use_doorbell) {
4490 		tmp = RREG32(mmCP_HQD_PQ_DOORBELL_CONTROL);
4491 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
4492 				DOORBELL_OFFSET, ring->doorbell_index);
4493 
4494 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
4495 					 DOORBELL_EN, 1);
4496 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
4497 					 DOORBELL_SOURCE, 0);
4498 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
4499 					 DOORBELL_HIT, 0);
4500 	}
4501 
4502 	mqd->cp_hqd_pq_doorbell_control = tmp;
4503 
4504 	/* reset read and write pointers, similar to CP_RB0_WPTR/_RPTR */
4505 	ring->wptr = 0;
4506 	mqd->cp_hqd_pq_wptr = ring->wptr;
4507 	mqd->cp_hqd_pq_rptr = RREG32(mmCP_HQD_PQ_RPTR);
4508 
4509 	/* set the vmid for the queue */
4510 	mqd->cp_hqd_vmid = 0;
4511 
4512 	tmp = RREG32(mmCP_HQD_PERSISTENT_STATE);
4513 	tmp = REG_SET_FIELD(tmp, CP_HQD_PERSISTENT_STATE, PRELOAD_SIZE, 0x53);
4514 	mqd->cp_hqd_persistent_state = tmp;
4515 
4516 	/* set MTYPE */
4517 	tmp = RREG32(mmCP_HQD_IB_CONTROL);
4518 	tmp = REG_SET_FIELD(tmp, CP_HQD_IB_CONTROL, MIN_IB_AVAIL_SIZE, 3);
4519 	tmp = REG_SET_FIELD(tmp, CP_HQD_IB_CONTROL, MTYPE, 3);
4520 	mqd->cp_hqd_ib_control = tmp;
4521 
4522 	tmp = RREG32(mmCP_HQD_IQ_TIMER);
4523 	tmp = REG_SET_FIELD(tmp, CP_HQD_IQ_TIMER, MTYPE, 3);
4524 	mqd->cp_hqd_iq_timer = tmp;
4525 
4526 	tmp = RREG32(mmCP_HQD_CTX_SAVE_CONTROL);
4527 	tmp = REG_SET_FIELD(tmp, CP_HQD_CTX_SAVE_CONTROL, MTYPE, 3);
4528 	mqd->cp_hqd_ctx_save_control = tmp;
4529 
4530 	/* defaults */
4531 	mqd->cp_hqd_eop_rptr = RREG32(mmCP_HQD_EOP_RPTR);
4532 	mqd->cp_hqd_eop_wptr = RREG32(mmCP_HQD_EOP_WPTR);
4533 	mqd->cp_hqd_ctx_save_base_addr_lo = RREG32(mmCP_HQD_CTX_SAVE_BASE_ADDR_LO);
4534 	mqd->cp_hqd_ctx_save_base_addr_hi = RREG32(mmCP_HQD_CTX_SAVE_BASE_ADDR_HI);
4535 	mqd->cp_hqd_cntl_stack_offset = RREG32(mmCP_HQD_CNTL_STACK_OFFSET);
4536 	mqd->cp_hqd_cntl_stack_size = RREG32(mmCP_HQD_CNTL_STACK_SIZE);
4537 	mqd->cp_hqd_wg_state_offset = RREG32(mmCP_HQD_WG_STATE_OFFSET);
4538 	mqd->cp_hqd_ctx_save_size = RREG32(mmCP_HQD_CTX_SAVE_SIZE);
4539 	mqd->cp_hqd_eop_done_events = RREG32(mmCP_HQD_EOP_EVENTS);
4540 	mqd->cp_hqd_error = RREG32(mmCP_HQD_ERROR);
4541 	mqd->cp_hqd_eop_wptr_mem = RREG32(mmCP_HQD_EOP_WPTR_MEM);
4542 	mqd->cp_hqd_eop_dones = RREG32(mmCP_HQD_EOP_DONES);
4543 
4544 	/* set static priority for a queue/ring */
4545 	gfx_v8_0_mqd_set_priority(ring, mqd);
4546 	mqd->cp_hqd_quantum = RREG32(mmCP_HQD_QUANTUM);
4547 
4548 	/* map_queues packet doesn't need activate the queue,
4549 	 * so only kiq need set this field.
4550 	 */
4551 	if (ring->funcs->type == AMDGPU_RING_TYPE_KIQ)
4552 		mqd->cp_hqd_active = 1;
4553 
4554 	return 0;
4555 }
4556 
4557 static int gfx_v8_0_mqd_commit(struct amdgpu_device *adev,
4558 			struct vi_mqd *mqd)
4559 {
4560 	uint32_t mqd_reg;
4561 	uint32_t *mqd_data;
4562 
4563 	/* HQD registers extend from mmCP_MQD_BASE_ADDR to mmCP_HQD_ERROR */
4564 	mqd_data = &mqd->cp_mqd_base_addr_lo;
4565 
4566 	/* disable wptr polling */
4567 	WREG32_FIELD(CP_PQ_WPTR_POLL_CNTL, EN, 0);
4568 
4569 	/* program all HQD registers */
4570 	for (mqd_reg = mmCP_HQD_VMID; mqd_reg <= mmCP_HQD_EOP_CONTROL; mqd_reg++)
4571 		WREG32(mqd_reg, mqd_data[mqd_reg - mmCP_MQD_BASE_ADDR]);
4572 
4573 	/* Tonga errata: EOP RPTR/WPTR should be left unmodified.
4574 	 * This is safe since EOP RPTR==WPTR for any inactive HQD
4575 	 * on ASICs that do not support context-save.
4576 	 * EOP writes/reads can start anywhere in the ring.
4577 	 */
4578 	if (adev->asic_type != CHIP_TONGA) {
4579 		WREG32(mmCP_HQD_EOP_RPTR, mqd->cp_hqd_eop_rptr);
4580 		WREG32(mmCP_HQD_EOP_WPTR, mqd->cp_hqd_eop_wptr);
4581 		WREG32(mmCP_HQD_EOP_WPTR_MEM, mqd->cp_hqd_eop_wptr_mem);
4582 	}
4583 
4584 	for (mqd_reg = mmCP_HQD_EOP_EVENTS; mqd_reg <= mmCP_HQD_ERROR; mqd_reg++)
4585 		WREG32(mqd_reg, mqd_data[mqd_reg - mmCP_MQD_BASE_ADDR]);
4586 
4587 	/* activate the HQD */
4588 	for (mqd_reg = mmCP_MQD_BASE_ADDR; mqd_reg <= mmCP_HQD_ACTIVE; mqd_reg++)
4589 		WREG32(mqd_reg, mqd_data[mqd_reg - mmCP_MQD_BASE_ADDR]);
4590 
4591 	return 0;
4592 }
4593 
4594 static int gfx_v8_0_kiq_init_queue(struct amdgpu_ring *ring)
4595 {
4596 	struct amdgpu_device *adev = ring->adev;
4597 	struct vi_mqd *mqd = ring->mqd_ptr;
4598 	int mqd_idx = AMDGPU_MAX_COMPUTE_RINGS;
4599 
4600 	gfx_v8_0_kiq_setting(ring);
4601 
4602 	if (amdgpu_in_reset(adev)) { /* for GPU_RESET case */
4603 		/* reset MQD to a clean status */
4604 		if (adev->gfx.mec.mqd_backup[mqd_idx])
4605 			memcpy(mqd, adev->gfx.mec.mqd_backup[mqd_idx], sizeof(struct vi_mqd_allocation));
4606 
4607 		/* reset ring buffer */
4608 		ring->wptr = 0;
4609 		amdgpu_ring_clear_ring(ring);
4610 		mutex_lock(&adev->srbm_mutex);
4611 		vi_srbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
4612 		gfx_v8_0_mqd_commit(adev, mqd);
4613 		vi_srbm_select(adev, 0, 0, 0, 0);
4614 		mutex_unlock(&adev->srbm_mutex);
4615 	} else {
4616 		memset((void *)mqd, 0, sizeof(struct vi_mqd_allocation));
4617 		((struct vi_mqd_allocation *)mqd)->dynamic_cu_mask = 0xFFFFFFFF;
4618 		((struct vi_mqd_allocation *)mqd)->dynamic_rb_mask = 0xFFFFFFFF;
4619 		if (amdgpu_sriov_vf(adev) && adev->in_suspend)
4620 			amdgpu_ring_clear_ring(ring);
4621 		mutex_lock(&adev->srbm_mutex);
4622 		vi_srbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
4623 		gfx_v8_0_mqd_init(ring);
4624 		gfx_v8_0_mqd_commit(adev, mqd);
4625 		vi_srbm_select(adev, 0, 0, 0, 0);
4626 		mutex_unlock(&adev->srbm_mutex);
4627 
4628 		if (adev->gfx.mec.mqd_backup[mqd_idx])
4629 			memcpy(adev->gfx.mec.mqd_backup[mqd_idx], mqd, sizeof(struct vi_mqd_allocation));
4630 	}
4631 
4632 	return 0;
4633 }
4634 
4635 static int gfx_v8_0_kcq_init_queue(struct amdgpu_ring *ring)
4636 {
4637 	struct amdgpu_device *adev = ring->adev;
4638 	struct vi_mqd *mqd = ring->mqd_ptr;
4639 	int mqd_idx = ring - &adev->gfx.compute_ring[0];
4640 
4641 	if (!amdgpu_in_reset(adev) && !adev->in_suspend) {
4642 		memset((void *)mqd, 0, sizeof(struct vi_mqd_allocation));
4643 		((struct vi_mqd_allocation *)mqd)->dynamic_cu_mask = 0xFFFFFFFF;
4644 		((struct vi_mqd_allocation *)mqd)->dynamic_rb_mask = 0xFFFFFFFF;
4645 		mutex_lock(&adev->srbm_mutex);
4646 		vi_srbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
4647 		gfx_v8_0_mqd_init(ring);
4648 		vi_srbm_select(adev, 0, 0, 0, 0);
4649 		mutex_unlock(&adev->srbm_mutex);
4650 
4651 		if (adev->gfx.mec.mqd_backup[mqd_idx])
4652 			memcpy(adev->gfx.mec.mqd_backup[mqd_idx], mqd, sizeof(struct vi_mqd_allocation));
4653 	} else if (amdgpu_in_reset(adev)) { /* for GPU_RESET case */
4654 		/* reset MQD to a clean status */
4655 		if (adev->gfx.mec.mqd_backup[mqd_idx])
4656 			memcpy(mqd, adev->gfx.mec.mqd_backup[mqd_idx], sizeof(struct vi_mqd_allocation));
4657 		/* reset ring buffer */
4658 		ring->wptr = 0;
4659 		amdgpu_ring_clear_ring(ring);
4660 	} else {
4661 		amdgpu_ring_clear_ring(ring);
4662 	}
4663 	return 0;
4664 }
4665 
4666 static void gfx_v8_0_set_mec_doorbell_range(struct amdgpu_device *adev)
4667 {
4668 	if (adev->asic_type > CHIP_TONGA) {
4669 		WREG32(mmCP_MEC_DOORBELL_RANGE_LOWER, adev->doorbell_index.kiq << 2);
4670 		WREG32(mmCP_MEC_DOORBELL_RANGE_UPPER, adev->doorbell_index.mec_ring7 << 2);
4671 	}
4672 	/* enable doorbells */
4673 	WREG32_FIELD(CP_PQ_STATUS, DOORBELL_ENABLE, 1);
4674 }
4675 
4676 static int gfx_v8_0_kiq_resume(struct amdgpu_device *adev)
4677 {
4678 	struct amdgpu_ring *ring;
4679 	int r;
4680 
4681 	ring = &adev->gfx.kiq.ring;
4682 
4683 	r = amdgpu_bo_reserve(ring->mqd_obj, false);
4684 	if (unlikely(r != 0))
4685 		return r;
4686 
4687 	r = amdgpu_bo_kmap(ring->mqd_obj, &ring->mqd_ptr);
4688 	if (unlikely(r != 0))
4689 		return r;
4690 
4691 	gfx_v8_0_kiq_init_queue(ring);
4692 	amdgpu_bo_kunmap(ring->mqd_obj);
4693 	ring->mqd_ptr = NULL;
4694 	amdgpu_bo_unreserve(ring->mqd_obj);
4695 	ring->sched.ready = true;
4696 	return 0;
4697 }
4698 
4699 static int gfx_v8_0_kcq_resume(struct amdgpu_device *adev)
4700 {
4701 	struct amdgpu_ring *ring = NULL;
4702 	int r = 0, i;
4703 
4704 	gfx_v8_0_cp_compute_enable(adev, true);
4705 
4706 	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
4707 		ring = &adev->gfx.compute_ring[i];
4708 
4709 		r = amdgpu_bo_reserve(ring->mqd_obj, false);
4710 		if (unlikely(r != 0))
4711 			goto done;
4712 		r = amdgpu_bo_kmap(ring->mqd_obj, &ring->mqd_ptr);
4713 		if (!r) {
4714 			r = gfx_v8_0_kcq_init_queue(ring);
4715 			amdgpu_bo_kunmap(ring->mqd_obj);
4716 			ring->mqd_ptr = NULL;
4717 		}
4718 		amdgpu_bo_unreserve(ring->mqd_obj);
4719 		if (r)
4720 			goto done;
4721 	}
4722 
4723 	gfx_v8_0_set_mec_doorbell_range(adev);
4724 
4725 	r = gfx_v8_0_kiq_kcq_enable(adev);
4726 	if (r)
4727 		goto done;
4728 
4729 done:
4730 	return r;
4731 }
4732 
4733 static int gfx_v8_0_cp_test_all_rings(struct amdgpu_device *adev)
4734 {
4735 	int r, i;
4736 	struct amdgpu_ring *ring;
4737 
4738 	/* collect all the ring_tests here, gfx, kiq, compute */
4739 	ring = &adev->gfx.gfx_ring[0];
4740 	r = amdgpu_ring_test_helper(ring);
4741 	if (r)
4742 		return r;
4743 
4744 	ring = &adev->gfx.kiq.ring;
4745 	r = amdgpu_ring_test_helper(ring);
4746 	if (r)
4747 		return r;
4748 
4749 	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
4750 		ring = &adev->gfx.compute_ring[i];
4751 		amdgpu_ring_test_helper(ring);
4752 	}
4753 
4754 	return 0;
4755 }
4756 
4757 static int gfx_v8_0_cp_resume(struct amdgpu_device *adev)
4758 {
4759 	int r;
4760 
4761 	if (!(adev->flags & AMD_IS_APU))
4762 		gfx_v8_0_enable_gui_idle_interrupt(adev, false);
4763 
4764 	r = gfx_v8_0_kiq_resume(adev);
4765 	if (r)
4766 		return r;
4767 
4768 	r = gfx_v8_0_cp_gfx_resume(adev);
4769 	if (r)
4770 		return r;
4771 
4772 	r = gfx_v8_0_kcq_resume(adev);
4773 	if (r)
4774 		return r;
4775 
4776 	r = gfx_v8_0_cp_test_all_rings(adev);
4777 	if (r)
4778 		return r;
4779 
4780 	gfx_v8_0_enable_gui_idle_interrupt(adev, true);
4781 
4782 	return 0;
4783 }
4784 
4785 static void gfx_v8_0_cp_enable(struct amdgpu_device *adev, bool enable)
4786 {
4787 	gfx_v8_0_cp_gfx_enable(adev, enable);
4788 	gfx_v8_0_cp_compute_enable(adev, enable);
4789 }
4790 
4791 static int gfx_v8_0_hw_init(void *handle)
4792 {
4793 	int r;
4794 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4795 
4796 	gfx_v8_0_init_golden_registers(adev);
4797 	gfx_v8_0_constants_init(adev);
4798 
4799 	r = adev->gfx.rlc.funcs->resume(adev);
4800 	if (r)
4801 		return r;
4802 
4803 	r = gfx_v8_0_cp_resume(adev);
4804 
4805 	return r;
4806 }
4807 
4808 static int gfx_v8_0_kcq_disable(struct amdgpu_device *adev)
4809 {
4810 	int r, i;
4811 	struct amdgpu_ring *kiq_ring = &adev->gfx.kiq.ring;
4812 
4813 	r = amdgpu_ring_alloc(kiq_ring, 6 * adev->gfx.num_compute_rings);
4814 	if (r)
4815 		DRM_ERROR("Failed to lock KIQ (%d).\n", r);
4816 
4817 	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
4818 		struct amdgpu_ring *ring = &adev->gfx.compute_ring[i];
4819 
4820 		amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_UNMAP_QUEUES, 4));
4821 		amdgpu_ring_write(kiq_ring, /* Q_sel: 0, vmid: 0, engine: 0, num_Q: 1 */
4822 						PACKET3_UNMAP_QUEUES_ACTION(1) | /* RESET_QUEUES */
4823 						PACKET3_UNMAP_QUEUES_QUEUE_SEL(0) |
4824 						PACKET3_UNMAP_QUEUES_ENGINE_SEL(0) |
4825 						PACKET3_UNMAP_QUEUES_NUM_QUEUES(1));
4826 		amdgpu_ring_write(kiq_ring, PACKET3_UNMAP_QUEUES_DOORBELL_OFFSET0(ring->doorbell_index));
4827 		amdgpu_ring_write(kiq_ring, 0);
4828 		amdgpu_ring_write(kiq_ring, 0);
4829 		amdgpu_ring_write(kiq_ring, 0);
4830 	}
4831 	r = amdgpu_ring_test_helper(kiq_ring);
4832 	if (r)
4833 		DRM_ERROR("KCQ disable failed\n");
4834 
4835 	return r;
4836 }
4837 
4838 static bool gfx_v8_0_is_idle(void *handle)
4839 {
4840 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4841 
4842 	if (REG_GET_FIELD(RREG32(mmGRBM_STATUS), GRBM_STATUS, GUI_ACTIVE)
4843 		|| RREG32(mmGRBM_STATUS2) != 0x8)
4844 		return false;
4845 	else
4846 		return true;
4847 }
4848 
4849 static bool gfx_v8_0_rlc_is_idle(void *handle)
4850 {
4851 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4852 
4853 	if (RREG32(mmGRBM_STATUS2) != 0x8)
4854 		return false;
4855 	else
4856 		return true;
4857 }
4858 
4859 static int gfx_v8_0_wait_for_rlc_idle(void *handle)
4860 {
4861 	unsigned int i;
4862 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4863 
4864 	for (i = 0; i < adev->usec_timeout; i++) {
4865 		if (gfx_v8_0_rlc_is_idle(handle))
4866 			return 0;
4867 
4868 		udelay(1);
4869 	}
4870 	return -ETIMEDOUT;
4871 }
4872 
4873 static int gfx_v8_0_wait_for_idle(void *handle)
4874 {
4875 	unsigned int i;
4876 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4877 
4878 	for (i = 0; i < adev->usec_timeout; i++) {
4879 		if (gfx_v8_0_is_idle(handle))
4880 			return 0;
4881 
4882 		udelay(1);
4883 	}
4884 	return -ETIMEDOUT;
4885 }
4886 
4887 static int gfx_v8_0_hw_fini(void *handle)
4888 {
4889 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4890 
4891 	amdgpu_irq_put(adev, &adev->gfx.priv_reg_irq, 0);
4892 	amdgpu_irq_put(adev, &adev->gfx.priv_inst_irq, 0);
4893 
4894 	amdgpu_irq_put(adev, &adev->gfx.cp_ecc_error_irq, 0);
4895 
4896 	amdgpu_irq_put(adev, &adev->gfx.sq_irq, 0);
4897 
4898 	/* disable KCQ to avoid CPC touch memory not valid anymore */
4899 	gfx_v8_0_kcq_disable(adev);
4900 
4901 	if (amdgpu_sriov_vf(adev)) {
4902 		pr_debug("For SRIOV client, shouldn't do anything.\n");
4903 		return 0;
4904 	}
4905 	amdgpu_gfx_rlc_enter_safe_mode(adev);
4906 	if (!gfx_v8_0_wait_for_idle(adev))
4907 		gfx_v8_0_cp_enable(adev, false);
4908 	else
4909 		pr_err("cp is busy, skip halt cp\n");
4910 	if (!gfx_v8_0_wait_for_rlc_idle(adev))
4911 		adev->gfx.rlc.funcs->stop(adev);
4912 	else
4913 		pr_err("rlc is busy, skip halt rlc\n");
4914 	amdgpu_gfx_rlc_exit_safe_mode(adev);
4915 
4916 	return 0;
4917 }
4918 
4919 static int gfx_v8_0_suspend(void *handle)
4920 {
4921 	return gfx_v8_0_hw_fini(handle);
4922 }
4923 
4924 static int gfx_v8_0_resume(void *handle)
4925 {
4926 	return gfx_v8_0_hw_init(handle);
4927 }
4928 
4929 static bool gfx_v8_0_check_soft_reset(void *handle)
4930 {
4931 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4932 	u32 grbm_soft_reset = 0, srbm_soft_reset = 0;
4933 	u32 tmp;
4934 
4935 	/* GRBM_STATUS */
4936 	tmp = RREG32(mmGRBM_STATUS);
4937 	if (tmp & (GRBM_STATUS__PA_BUSY_MASK | GRBM_STATUS__SC_BUSY_MASK |
4938 		   GRBM_STATUS__BCI_BUSY_MASK | GRBM_STATUS__SX_BUSY_MASK |
4939 		   GRBM_STATUS__TA_BUSY_MASK | GRBM_STATUS__VGT_BUSY_MASK |
4940 		   GRBM_STATUS__DB_BUSY_MASK | GRBM_STATUS__CB_BUSY_MASK |
4941 		   GRBM_STATUS__GDS_BUSY_MASK | GRBM_STATUS__SPI_BUSY_MASK |
4942 		   GRBM_STATUS__IA_BUSY_MASK | GRBM_STATUS__IA_BUSY_NO_DMA_MASK |
4943 		   GRBM_STATUS__CP_BUSY_MASK | GRBM_STATUS__CP_COHERENCY_BUSY_MASK)) {
4944 		grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset,
4945 						GRBM_SOFT_RESET, SOFT_RESET_CP, 1);
4946 		grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset,
4947 						GRBM_SOFT_RESET, SOFT_RESET_GFX, 1);
4948 		srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset,
4949 						SRBM_SOFT_RESET, SOFT_RESET_GRBM, 1);
4950 	}
4951 
4952 	/* GRBM_STATUS2 */
4953 	tmp = RREG32(mmGRBM_STATUS2);
4954 	if (REG_GET_FIELD(tmp, GRBM_STATUS2, RLC_BUSY))
4955 		grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset,
4956 						GRBM_SOFT_RESET, SOFT_RESET_RLC, 1);
4957 
4958 	if (REG_GET_FIELD(tmp, GRBM_STATUS2, CPF_BUSY) ||
4959 	    REG_GET_FIELD(tmp, GRBM_STATUS2, CPC_BUSY) ||
4960 	    REG_GET_FIELD(tmp, GRBM_STATUS2, CPG_BUSY)) {
4961 		grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET,
4962 						SOFT_RESET_CPF, 1);
4963 		grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET,
4964 						SOFT_RESET_CPC, 1);
4965 		grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET,
4966 						SOFT_RESET_CPG, 1);
4967 		srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset, SRBM_SOFT_RESET,
4968 						SOFT_RESET_GRBM, 1);
4969 	}
4970 
4971 	/* SRBM_STATUS */
4972 	tmp = RREG32(mmSRBM_STATUS);
4973 	if (REG_GET_FIELD(tmp, SRBM_STATUS, GRBM_RQ_PENDING))
4974 		srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset,
4975 						SRBM_SOFT_RESET, SOFT_RESET_GRBM, 1);
4976 	if (REG_GET_FIELD(tmp, SRBM_STATUS, SEM_BUSY))
4977 		srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset,
4978 						SRBM_SOFT_RESET, SOFT_RESET_SEM, 1);
4979 
4980 	if (grbm_soft_reset || srbm_soft_reset) {
4981 		adev->gfx.grbm_soft_reset = grbm_soft_reset;
4982 		adev->gfx.srbm_soft_reset = srbm_soft_reset;
4983 		return true;
4984 	} else {
4985 		adev->gfx.grbm_soft_reset = 0;
4986 		adev->gfx.srbm_soft_reset = 0;
4987 		return false;
4988 	}
4989 }
4990 
4991 static int gfx_v8_0_pre_soft_reset(void *handle)
4992 {
4993 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4994 	u32 grbm_soft_reset = 0;
4995 
4996 	if ((!adev->gfx.grbm_soft_reset) &&
4997 	    (!adev->gfx.srbm_soft_reset))
4998 		return 0;
4999 
5000 	grbm_soft_reset = adev->gfx.grbm_soft_reset;
5001 
5002 	/* stop the rlc */
5003 	adev->gfx.rlc.funcs->stop(adev);
5004 
5005 	if (REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CP) ||
5006 	    REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_GFX))
5007 		/* Disable GFX parsing/prefetching */
5008 		gfx_v8_0_cp_gfx_enable(adev, false);
5009 
5010 	if (REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CP) ||
5011 	    REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CPF) ||
5012 	    REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CPC) ||
5013 	    REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CPG)) {
5014 		int i;
5015 
5016 		for (i = 0; i < adev->gfx.num_compute_rings; i++) {
5017 			struct amdgpu_ring *ring = &adev->gfx.compute_ring[i];
5018 
5019 			mutex_lock(&adev->srbm_mutex);
5020 			vi_srbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
5021 			gfx_v8_0_deactivate_hqd(adev, 2);
5022 			vi_srbm_select(adev, 0, 0, 0, 0);
5023 			mutex_unlock(&adev->srbm_mutex);
5024 		}
5025 		/* Disable MEC parsing/prefetching */
5026 		gfx_v8_0_cp_compute_enable(adev, false);
5027 	}
5028 
5029 	return 0;
5030 }
5031 
5032 static int gfx_v8_0_soft_reset(void *handle)
5033 {
5034 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5035 	u32 grbm_soft_reset = 0, srbm_soft_reset = 0;
5036 	u32 tmp;
5037 
5038 	if ((!adev->gfx.grbm_soft_reset) &&
5039 	    (!adev->gfx.srbm_soft_reset))
5040 		return 0;
5041 
5042 	grbm_soft_reset = adev->gfx.grbm_soft_reset;
5043 	srbm_soft_reset = adev->gfx.srbm_soft_reset;
5044 
5045 	if (grbm_soft_reset || srbm_soft_reset) {
5046 		tmp = RREG32(mmGMCON_DEBUG);
5047 		tmp = REG_SET_FIELD(tmp, GMCON_DEBUG, GFX_STALL, 1);
5048 		tmp = REG_SET_FIELD(tmp, GMCON_DEBUG, GFX_CLEAR, 1);
5049 		WREG32(mmGMCON_DEBUG, tmp);
5050 		udelay(50);
5051 	}
5052 
5053 	if (grbm_soft_reset) {
5054 		tmp = RREG32(mmGRBM_SOFT_RESET);
5055 		tmp |= grbm_soft_reset;
5056 		dev_info(adev->dev, "GRBM_SOFT_RESET=0x%08X\n", tmp);
5057 		WREG32(mmGRBM_SOFT_RESET, tmp);
5058 		tmp = RREG32(mmGRBM_SOFT_RESET);
5059 
5060 		udelay(50);
5061 
5062 		tmp &= ~grbm_soft_reset;
5063 		WREG32(mmGRBM_SOFT_RESET, tmp);
5064 		tmp = RREG32(mmGRBM_SOFT_RESET);
5065 	}
5066 
5067 	if (srbm_soft_reset) {
5068 		tmp = RREG32(mmSRBM_SOFT_RESET);
5069 		tmp |= srbm_soft_reset;
5070 		dev_info(adev->dev, "SRBM_SOFT_RESET=0x%08X\n", tmp);
5071 		WREG32(mmSRBM_SOFT_RESET, tmp);
5072 		tmp = RREG32(mmSRBM_SOFT_RESET);
5073 
5074 		udelay(50);
5075 
5076 		tmp &= ~srbm_soft_reset;
5077 		WREG32(mmSRBM_SOFT_RESET, tmp);
5078 		tmp = RREG32(mmSRBM_SOFT_RESET);
5079 	}
5080 
5081 	if (grbm_soft_reset || srbm_soft_reset) {
5082 		tmp = RREG32(mmGMCON_DEBUG);
5083 		tmp = REG_SET_FIELD(tmp, GMCON_DEBUG, GFX_STALL, 0);
5084 		tmp = REG_SET_FIELD(tmp, GMCON_DEBUG, GFX_CLEAR, 0);
5085 		WREG32(mmGMCON_DEBUG, tmp);
5086 	}
5087 
5088 	/* Wait a little for things to settle down */
5089 	udelay(50);
5090 
5091 	return 0;
5092 }
5093 
5094 static int gfx_v8_0_post_soft_reset(void *handle)
5095 {
5096 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5097 	u32 grbm_soft_reset = 0;
5098 
5099 	if ((!adev->gfx.grbm_soft_reset) &&
5100 	    (!adev->gfx.srbm_soft_reset))
5101 		return 0;
5102 
5103 	grbm_soft_reset = adev->gfx.grbm_soft_reset;
5104 
5105 	if (REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CP) ||
5106 	    REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CPF) ||
5107 	    REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CPC) ||
5108 	    REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CPG)) {
5109 		int i;
5110 
5111 		for (i = 0; i < adev->gfx.num_compute_rings; i++) {
5112 			struct amdgpu_ring *ring = &adev->gfx.compute_ring[i];
5113 
5114 			mutex_lock(&adev->srbm_mutex);
5115 			vi_srbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
5116 			gfx_v8_0_deactivate_hqd(adev, 2);
5117 			vi_srbm_select(adev, 0, 0, 0, 0);
5118 			mutex_unlock(&adev->srbm_mutex);
5119 		}
5120 		gfx_v8_0_kiq_resume(adev);
5121 		gfx_v8_0_kcq_resume(adev);
5122 	}
5123 
5124 	if (REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CP) ||
5125 	    REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_GFX))
5126 		gfx_v8_0_cp_gfx_resume(adev);
5127 
5128 	gfx_v8_0_cp_test_all_rings(adev);
5129 
5130 	adev->gfx.rlc.funcs->start(adev);
5131 
5132 	return 0;
5133 }
5134 
5135 /**
5136  * gfx_v8_0_get_gpu_clock_counter - return GPU clock counter snapshot
5137  *
5138  * @adev: amdgpu_device pointer
5139  *
5140  * Fetches a GPU clock counter snapshot.
5141  * Returns the 64 bit clock counter snapshot.
5142  */
5143 static uint64_t gfx_v8_0_get_gpu_clock_counter(struct amdgpu_device *adev)
5144 {
5145 	uint64_t clock;
5146 
5147 	mutex_lock(&adev->gfx.gpu_clock_mutex);
5148 	WREG32(mmRLC_CAPTURE_GPU_CLOCK_COUNT, 1);
5149 	clock = (uint64_t)RREG32(mmRLC_GPU_CLOCK_COUNT_LSB) |
5150 		((uint64_t)RREG32(mmRLC_GPU_CLOCK_COUNT_MSB) << 32ULL);
5151 	mutex_unlock(&adev->gfx.gpu_clock_mutex);
5152 	return clock;
5153 }
5154 
5155 static void gfx_v8_0_ring_emit_gds_switch(struct amdgpu_ring *ring,
5156 					  uint32_t vmid,
5157 					  uint32_t gds_base, uint32_t gds_size,
5158 					  uint32_t gws_base, uint32_t gws_size,
5159 					  uint32_t oa_base, uint32_t oa_size)
5160 {
5161 	/* GDS Base */
5162 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
5163 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
5164 				WRITE_DATA_DST_SEL(0)));
5165 	amdgpu_ring_write(ring, amdgpu_gds_reg_offset[vmid].mem_base);
5166 	amdgpu_ring_write(ring, 0);
5167 	amdgpu_ring_write(ring, gds_base);
5168 
5169 	/* GDS Size */
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_size);
5174 	amdgpu_ring_write(ring, 0);
5175 	amdgpu_ring_write(ring, gds_size);
5176 
5177 	/* GWS */
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].gws);
5182 	amdgpu_ring_write(ring, 0);
5183 	amdgpu_ring_write(ring, gws_size << GDS_GWS_VMID0__SIZE__SHIFT | gws_base);
5184 
5185 	/* OA */
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].oa);
5190 	amdgpu_ring_write(ring, 0);
5191 	amdgpu_ring_write(ring, (1 << (oa_size + oa_base)) - (1 << oa_base));
5192 }
5193 
5194 static uint32_t wave_read_ind(struct amdgpu_device *adev, uint32_t simd, uint32_t wave, uint32_t address)
5195 {
5196 	WREG32(mmSQ_IND_INDEX,
5197 		(wave << SQ_IND_INDEX__WAVE_ID__SHIFT) |
5198 		(simd << SQ_IND_INDEX__SIMD_ID__SHIFT) |
5199 		(address << SQ_IND_INDEX__INDEX__SHIFT) |
5200 		(SQ_IND_INDEX__FORCE_READ_MASK));
5201 	return RREG32(mmSQ_IND_DATA);
5202 }
5203 
5204 static void wave_read_regs(struct amdgpu_device *adev, uint32_t simd,
5205 			   uint32_t wave, uint32_t thread,
5206 			   uint32_t regno, uint32_t num, uint32_t *out)
5207 {
5208 	WREG32(mmSQ_IND_INDEX,
5209 		(wave << SQ_IND_INDEX__WAVE_ID__SHIFT) |
5210 		(simd << SQ_IND_INDEX__SIMD_ID__SHIFT) |
5211 		(regno << SQ_IND_INDEX__INDEX__SHIFT) |
5212 		(thread << SQ_IND_INDEX__THREAD_ID__SHIFT) |
5213 		(SQ_IND_INDEX__FORCE_READ_MASK) |
5214 		(SQ_IND_INDEX__AUTO_INCR_MASK));
5215 	while (num--)
5216 		*(out++) = RREG32(mmSQ_IND_DATA);
5217 }
5218 
5219 static void gfx_v8_0_read_wave_data(struct amdgpu_device *adev, uint32_t simd, uint32_t wave, uint32_t *dst, int *no_fields)
5220 {
5221 	/* type 0 wave data */
5222 	dst[(*no_fields)++] = 0;
5223 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_STATUS);
5224 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_PC_LO);
5225 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_PC_HI);
5226 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_EXEC_LO);
5227 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_EXEC_HI);
5228 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_HW_ID);
5229 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_INST_DW0);
5230 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_INST_DW1);
5231 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_GPR_ALLOC);
5232 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_LDS_ALLOC);
5233 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TRAPSTS);
5234 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_IB_STS);
5235 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TBA_LO);
5236 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TBA_HI);
5237 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TMA_LO);
5238 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TMA_HI);
5239 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_IB_DBG0);
5240 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_M0);
5241 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_MODE);
5242 }
5243 
5244 static void gfx_v8_0_read_wave_sgprs(struct amdgpu_device *adev, uint32_t simd,
5245 				     uint32_t wave, uint32_t start,
5246 				     uint32_t size, uint32_t *dst)
5247 {
5248 	wave_read_regs(
5249 		adev, simd, wave, 0,
5250 		start + SQIND_WAVE_SGPRS_OFFSET, size, dst);
5251 }
5252 
5253 
5254 static const struct amdgpu_gfx_funcs gfx_v8_0_gfx_funcs = {
5255 	.get_gpu_clock_counter = &gfx_v8_0_get_gpu_clock_counter,
5256 	.select_se_sh = &gfx_v8_0_select_se_sh,
5257 	.read_wave_data = &gfx_v8_0_read_wave_data,
5258 	.read_wave_sgprs = &gfx_v8_0_read_wave_sgprs,
5259 	.select_me_pipe_q = &gfx_v8_0_select_me_pipe_q
5260 };
5261 
5262 static int gfx_v8_0_early_init(void *handle)
5263 {
5264 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5265 
5266 	adev->gfx.num_gfx_rings = GFX8_NUM_GFX_RINGS;
5267 	adev->gfx.num_compute_rings = min(amdgpu_gfx_get_num_kcq(adev),
5268 					  AMDGPU_MAX_COMPUTE_RINGS);
5269 	adev->gfx.funcs = &gfx_v8_0_gfx_funcs;
5270 	gfx_v8_0_set_ring_funcs(adev);
5271 	gfx_v8_0_set_irq_funcs(adev);
5272 	gfx_v8_0_set_gds_init(adev);
5273 	gfx_v8_0_set_rlc_funcs(adev);
5274 
5275 	return 0;
5276 }
5277 
5278 static int gfx_v8_0_late_init(void *handle)
5279 {
5280 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5281 	int r;
5282 
5283 	r = amdgpu_irq_get(adev, &adev->gfx.priv_reg_irq, 0);
5284 	if (r)
5285 		return r;
5286 
5287 	r = amdgpu_irq_get(adev, &adev->gfx.priv_inst_irq, 0);
5288 	if (r)
5289 		return r;
5290 
5291 	/* requires IBs so do in late init after IB pool is initialized */
5292 	r = gfx_v8_0_do_edc_gpr_workarounds(adev);
5293 	if (r)
5294 		return r;
5295 
5296 	r = amdgpu_irq_get(adev, &adev->gfx.cp_ecc_error_irq, 0);
5297 	if (r) {
5298 		DRM_ERROR("amdgpu_irq_get() failed to get IRQ for EDC, r: %d.\n", r);
5299 		return r;
5300 	}
5301 
5302 	r = amdgpu_irq_get(adev, &adev->gfx.sq_irq, 0);
5303 	if (r) {
5304 		DRM_ERROR(
5305 			"amdgpu_irq_get() failed to get IRQ for SQ, r: %d.\n",
5306 			r);
5307 		return r;
5308 	}
5309 
5310 	return 0;
5311 }
5312 
5313 static void gfx_v8_0_enable_gfx_static_mg_power_gating(struct amdgpu_device *adev,
5314 						       bool enable)
5315 {
5316 	if ((adev->asic_type == CHIP_POLARIS11) ||
5317 	    (adev->asic_type == CHIP_POLARIS12) ||
5318 	    (adev->asic_type == CHIP_VEGAM))
5319 		/* Send msg to SMU via Powerplay */
5320 		amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_GFX, enable);
5321 
5322 	WREG32_FIELD(RLC_PG_CNTL, STATIC_PER_CU_PG_ENABLE, enable ? 1 : 0);
5323 }
5324 
5325 static void gfx_v8_0_enable_gfx_dynamic_mg_power_gating(struct amdgpu_device *adev,
5326 							bool enable)
5327 {
5328 	WREG32_FIELD(RLC_PG_CNTL, DYN_PER_CU_PG_ENABLE, enable ? 1 : 0);
5329 }
5330 
5331 static void polaris11_enable_gfx_quick_mg_power_gating(struct amdgpu_device *adev,
5332 		bool enable)
5333 {
5334 	WREG32_FIELD(RLC_PG_CNTL, QUICK_PG_ENABLE, enable ? 1 : 0);
5335 }
5336 
5337 static void cz_enable_gfx_cg_power_gating(struct amdgpu_device *adev,
5338 					  bool enable)
5339 {
5340 	WREG32_FIELD(RLC_PG_CNTL, GFX_POWER_GATING_ENABLE, enable ? 1 : 0);
5341 }
5342 
5343 static void cz_enable_gfx_pipeline_power_gating(struct amdgpu_device *adev,
5344 						bool enable)
5345 {
5346 	WREG32_FIELD(RLC_PG_CNTL, GFX_PIPELINE_PG_ENABLE, enable ? 1 : 0);
5347 
5348 	/* Read any GFX register to wake up GFX. */
5349 	if (!enable)
5350 		RREG32(mmDB_RENDER_CONTROL);
5351 }
5352 
5353 static void cz_update_gfx_cg_power_gating(struct amdgpu_device *adev,
5354 					  bool enable)
5355 {
5356 	if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_PG) && enable) {
5357 		cz_enable_gfx_cg_power_gating(adev, true);
5358 		if (adev->pg_flags & AMD_PG_SUPPORT_GFX_PIPELINE)
5359 			cz_enable_gfx_pipeline_power_gating(adev, true);
5360 	} else {
5361 		cz_enable_gfx_cg_power_gating(adev, false);
5362 		cz_enable_gfx_pipeline_power_gating(adev, false);
5363 	}
5364 }
5365 
5366 static int gfx_v8_0_set_powergating_state(void *handle,
5367 					  enum amd_powergating_state state)
5368 {
5369 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5370 	bool enable = (state == AMD_PG_STATE_GATE);
5371 
5372 	if (amdgpu_sriov_vf(adev))
5373 		return 0;
5374 
5375 	if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_SMG |
5376 				AMD_PG_SUPPORT_RLC_SMU_HS |
5377 				AMD_PG_SUPPORT_CP |
5378 				AMD_PG_SUPPORT_GFX_DMG))
5379 		amdgpu_gfx_rlc_enter_safe_mode(adev);
5380 	switch (adev->asic_type) {
5381 	case CHIP_CARRIZO:
5382 	case CHIP_STONEY:
5383 
5384 		if (adev->pg_flags & AMD_PG_SUPPORT_RLC_SMU_HS) {
5385 			cz_enable_sck_slow_down_on_power_up(adev, true);
5386 			cz_enable_sck_slow_down_on_power_down(adev, true);
5387 		} else {
5388 			cz_enable_sck_slow_down_on_power_up(adev, false);
5389 			cz_enable_sck_slow_down_on_power_down(adev, false);
5390 		}
5391 		if (adev->pg_flags & AMD_PG_SUPPORT_CP)
5392 			cz_enable_cp_power_gating(adev, true);
5393 		else
5394 			cz_enable_cp_power_gating(adev, false);
5395 
5396 		cz_update_gfx_cg_power_gating(adev, enable);
5397 
5398 		if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_SMG) && enable)
5399 			gfx_v8_0_enable_gfx_static_mg_power_gating(adev, true);
5400 		else
5401 			gfx_v8_0_enable_gfx_static_mg_power_gating(adev, false);
5402 
5403 		if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_DMG) && enable)
5404 			gfx_v8_0_enable_gfx_dynamic_mg_power_gating(adev, true);
5405 		else
5406 			gfx_v8_0_enable_gfx_dynamic_mg_power_gating(adev, false);
5407 		break;
5408 	case CHIP_POLARIS11:
5409 	case CHIP_POLARIS12:
5410 	case CHIP_VEGAM:
5411 		if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_SMG) && enable)
5412 			gfx_v8_0_enable_gfx_static_mg_power_gating(adev, true);
5413 		else
5414 			gfx_v8_0_enable_gfx_static_mg_power_gating(adev, false);
5415 
5416 		if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_DMG) && enable)
5417 			gfx_v8_0_enable_gfx_dynamic_mg_power_gating(adev, true);
5418 		else
5419 			gfx_v8_0_enable_gfx_dynamic_mg_power_gating(adev, false);
5420 
5421 		if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_QUICK_MG) && enable)
5422 			polaris11_enable_gfx_quick_mg_power_gating(adev, true);
5423 		else
5424 			polaris11_enable_gfx_quick_mg_power_gating(adev, false);
5425 		break;
5426 	default:
5427 		break;
5428 	}
5429 	if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_SMG |
5430 				AMD_PG_SUPPORT_RLC_SMU_HS |
5431 				AMD_PG_SUPPORT_CP |
5432 				AMD_PG_SUPPORT_GFX_DMG))
5433 		amdgpu_gfx_rlc_exit_safe_mode(adev);
5434 	return 0;
5435 }
5436 
5437 static void gfx_v8_0_get_clockgating_state(void *handle, u64 *flags)
5438 {
5439 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5440 	int data;
5441 
5442 	if (amdgpu_sriov_vf(adev))
5443 		*flags = 0;
5444 
5445 	/* AMD_CG_SUPPORT_GFX_MGCG */
5446 	data = RREG32(mmRLC_CGTT_MGCG_OVERRIDE);
5447 	if (!(data & RLC_CGTT_MGCG_OVERRIDE__CPF_MASK))
5448 		*flags |= AMD_CG_SUPPORT_GFX_MGCG;
5449 
5450 	/* AMD_CG_SUPPORT_GFX_CGLG */
5451 	data = RREG32(mmRLC_CGCG_CGLS_CTRL);
5452 	if (data & RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK)
5453 		*flags |= AMD_CG_SUPPORT_GFX_CGCG;
5454 
5455 	/* AMD_CG_SUPPORT_GFX_CGLS */
5456 	if (data & RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK)
5457 		*flags |= AMD_CG_SUPPORT_GFX_CGLS;
5458 
5459 	/* AMD_CG_SUPPORT_GFX_CGTS */
5460 	data = RREG32(mmCGTS_SM_CTRL_REG);
5461 	if (!(data & CGTS_SM_CTRL_REG__OVERRIDE_MASK))
5462 		*flags |= AMD_CG_SUPPORT_GFX_CGTS;
5463 
5464 	/* AMD_CG_SUPPORT_GFX_CGTS_LS */
5465 	if (!(data & CGTS_SM_CTRL_REG__LS_OVERRIDE_MASK))
5466 		*flags |= AMD_CG_SUPPORT_GFX_CGTS_LS;
5467 
5468 	/* AMD_CG_SUPPORT_GFX_RLC_LS */
5469 	data = RREG32(mmRLC_MEM_SLP_CNTL);
5470 	if (data & RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK)
5471 		*flags |= AMD_CG_SUPPORT_GFX_RLC_LS | AMD_CG_SUPPORT_GFX_MGLS;
5472 
5473 	/* AMD_CG_SUPPORT_GFX_CP_LS */
5474 	data = RREG32(mmCP_MEM_SLP_CNTL);
5475 	if (data & CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK)
5476 		*flags |= AMD_CG_SUPPORT_GFX_CP_LS | AMD_CG_SUPPORT_GFX_MGLS;
5477 }
5478 
5479 static void gfx_v8_0_send_serdes_cmd(struct amdgpu_device *adev,
5480 				     uint32_t reg_addr, uint32_t cmd)
5481 {
5482 	uint32_t data;
5483 
5484 	gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
5485 
5486 	WREG32(mmRLC_SERDES_WR_CU_MASTER_MASK, 0xffffffff);
5487 	WREG32(mmRLC_SERDES_WR_NONCU_MASTER_MASK, 0xffffffff);
5488 
5489 	data = RREG32(mmRLC_SERDES_WR_CTRL);
5490 	if (adev->asic_type == CHIP_STONEY)
5491 		data &= ~(RLC_SERDES_WR_CTRL__WRITE_COMMAND_MASK |
5492 			  RLC_SERDES_WR_CTRL__READ_COMMAND_MASK |
5493 			  RLC_SERDES_WR_CTRL__P1_SELECT_MASK |
5494 			  RLC_SERDES_WR_CTRL__P2_SELECT_MASK |
5495 			  RLC_SERDES_WR_CTRL__RDDATA_RESET_MASK |
5496 			  RLC_SERDES_WR_CTRL__POWER_DOWN_MASK |
5497 			  RLC_SERDES_WR_CTRL__POWER_UP_MASK |
5498 			  RLC_SERDES_WR_CTRL__SHORT_FORMAT_MASK |
5499 			  RLC_SERDES_WR_CTRL__SRBM_OVERRIDE_MASK);
5500 	else
5501 		data &= ~(RLC_SERDES_WR_CTRL__WRITE_COMMAND_MASK |
5502 			  RLC_SERDES_WR_CTRL__READ_COMMAND_MASK |
5503 			  RLC_SERDES_WR_CTRL__P1_SELECT_MASK |
5504 			  RLC_SERDES_WR_CTRL__P2_SELECT_MASK |
5505 			  RLC_SERDES_WR_CTRL__RDDATA_RESET_MASK |
5506 			  RLC_SERDES_WR_CTRL__POWER_DOWN_MASK |
5507 			  RLC_SERDES_WR_CTRL__POWER_UP_MASK |
5508 			  RLC_SERDES_WR_CTRL__SHORT_FORMAT_MASK |
5509 			  RLC_SERDES_WR_CTRL__BPM_DATA_MASK |
5510 			  RLC_SERDES_WR_CTRL__REG_ADDR_MASK |
5511 			  RLC_SERDES_WR_CTRL__SRBM_OVERRIDE_MASK);
5512 	data |= (RLC_SERDES_WR_CTRL__RSVD_BPM_ADDR_MASK |
5513 		 (cmd << RLC_SERDES_WR_CTRL__BPM_DATA__SHIFT) |
5514 		 (reg_addr << RLC_SERDES_WR_CTRL__REG_ADDR__SHIFT) |
5515 		 (0xff << RLC_SERDES_WR_CTRL__BPM_ADDR__SHIFT));
5516 
5517 	WREG32(mmRLC_SERDES_WR_CTRL, data);
5518 }
5519 
5520 #define MSG_ENTER_RLC_SAFE_MODE     1
5521 #define MSG_EXIT_RLC_SAFE_MODE      0
5522 #define RLC_GPR_REG2__REQ_MASK 0x00000001
5523 #define RLC_GPR_REG2__REQ__SHIFT 0
5524 #define RLC_GPR_REG2__MESSAGE__SHIFT 0x00000001
5525 #define RLC_GPR_REG2__MESSAGE_MASK 0x0000001e
5526 
5527 static bool gfx_v8_0_is_rlc_enabled(struct amdgpu_device *adev)
5528 {
5529 	uint32_t rlc_setting;
5530 
5531 	rlc_setting = RREG32(mmRLC_CNTL);
5532 	if (!(rlc_setting & RLC_CNTL__RLC_ENABLE_F32_MASK))
5533 		return false;
5534 
5535 	return true;
5536 }
5537 
5538 static void gfx_v8_0_set_safe_mode(struct amdgpu_device *adev)
5539 {
5540 	uint32_t data;
5541 	unsigned i;
5542 	data = RREG32(mmRLC_CNTL);
5543 	data |= RLC_SAFE_MODE__CMD_MASK;
5544 	data &= ~RLC_SAFE_MODE__MESSAGE_MASK;
5545 	data |= (1 << RLC_SAFE_MODE__MESSAGE__SHIFT);
5546 	WREG32(mmRLC_SAFE_MODE, data);
5547 
5548 	/* wait for RLC_SAFE_MODE */
5549 	for (i = 0; i < adev->usec_timeout; i++) {
5550 		if ((RREG32(mmRLC_GPM_STAT) &
5551 		     (RLC_GPM_STAT__GFX_CLOCK_STATUS_MASK |
5552 		      RLC_GPM_STAT__GFX_POWER_STATUS_MASK)) ==
5553 		    (RLC_GPM_STAT__GFX_CLOCK_STATUS_MASK |
5554 		     RLC_GPM_STAT__GFX_POWER_STATUS_MASK))
5555 			break;
5556 		udelay(1);
5557 	}
5558 	for (i = 0; i < adev->usec_timeout; i++) {
5559 		if (!REG_GET_FIELD(RREG32(mmRLC_SAFE_MODE), RLC_SAFE_MODE, CMD))
5560 			break;
5561 		udelay(1);
5562 	}
5563 }
5564 
5565 static void gfx_v8_0_unset_safe_mode(struct amdgpu_device *adev)
5566 {
5567 	uint32_t data;
5568 	unsigned i;
5569 
5570 	data = RREG32(mmRLC_CNTL);
5571 	data |= RLC_SAFE_MODE__CMD_MASK;
5572 	data &= ~RLC_SAFE_MODE__MESSAGE_MASK;
5573 	WREG32(mmRLC_SAFE_MODE, data);
5574 
5575 	for (i = 0; i < adev->usec_timeout; i++) {
5576 		if (!REG_GET_FIELD(RREG32(mmRLC_SAFE_MODE), RLC_SAFE_MODE, CMD))
5577 			break;
5578 		udelay(1);
5579 	}
5580 }
5581 
5582 static void gfx_v8_0_update_spm_vmid(struct amdgpu_device *adev, unsigned vmid)
5583 {
5584 	u32 data;
5585 
5586 	amdgpu_gfx_off_ctrl(adev, false);
5587 
5588 	if (amdgpu_sriov_is_pp_one_vf(adev))
5589 		data = RREG32_NO_KIQ(mmRLC_SPM_VMID);
5590 	else
5591 		data = RREG32(mmRLC_SPM_VMID);
5592 
5593 	data &= ~RLC_SPM_VMID__RLC_SPM_VMID_MASK;
5594 	data |= (vmid & RLC_SPM_VMID__RLC_SPM_VMID_MASK) << RLC_SPM_VMID__RLC_SPM_VMID__SHIFT;
5595 
5596 	if (amdgpu_sriov_is_pp_one_vf(adev))
5597 		WREG32_NO_KIQ(mmRLC_SPM_VMID, data);
5598 	else
5599 		WREG32(mmRLC_SPM_VMID, data);
5600 
5601 	amdgpu_gfx_off_ctrl(adev, true);
5602 }
5603 
5604 static const struct amdgpu_rlc_funcs iceland_rlc_funcs = {
5605 	.is_rlc_enabled = gfx_v8_0_is_rlc_enabled,
5606 	.set_safe_mode = gfx_v8_0_set_safe_mode,
5607 	.unset_safe_mode = gfx_v8_0_unset_safe_mode,
5608 	.init = gfx_v8_0_rlc_init,
5609 	.get_csb_size = gfx_v8_0_get_csb_size,
5610 	.get_csb_buffer = gfx_v8_0_get_csb_buffer,
5611 	.get_cp_table_num = gfx_v8_0_cp_jump_table_num,
5612 	.resume = gfx_v8_0_rlc_resume,
5613 	.stop = gfx_v8_0_rlc_stop,
5614 	.reset = gfx_v8_0_rlc_reset,
5615 	.start = gfx_v8_0_rlc_start,
5616 	.update_spm_vmid = gfx_v8_0_update_spm_vmid
5617 };
5618 
5619 static void gfx_v8_0_update_medium_grain_clock_gating(struct amdgpu_device *adev,
5620 						      bool enable)
5621 {
5622 	uint32_t temp, data;
5623 
5624 	amdgpu_gfx_rlc_enter_safe_mode(adev);
5625 
5626 	/* It is disabled by HW by default */
5627 	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGCG)) {
5628 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS) {
5629 			if (adev->cg_flags & AMD_CG_SUPPORT_GFX_RLC_LS)
5630 				/* 1 - RLC memory Light sleep */
5631 				WREG32_FIELD(RLC_MEM_SLP_CNTL, RLC_MEM_LS_EN, 1);
5632 
5633 			if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CP_LS)
5634 				WREG32_FIELD(CP_MEM_SLP_CNTL, CP_MEM_LS_EN, 1);
5635 		}
5636 
5637 		/* 3 - RLC_CGTT_MGCG_OVERRIDE */
5638 		temp = data = RREG32(mmRLC_CGTT_MGCG_OVERRIDE);
5639 		if (adev->flags & AMD_IS_APU)
5640 			data &= ~(RLC_CGTT_MGCG_OVERRIDE__CPF_MASK |
5641 				  RLC_CGTT_MGCG_OVERRIDE__RLC_MASK |
5642 				  RLC_CGTT_MGCG_OVERRIDE__MGCG_MASK);
5643 		else
5644 			data &= ~(RLC_CGTT_MGCG_OVERRIDE__CPF_MASK |
5645 				  RLC_CGTT_MGCG_OVERRIDE__RLC_MASK |
5646 				  RLC_CGTT_MGCG_OVERRIDE__MGCG_MASK |
5647 				  RLC_CGTT_MGCG_OVERRIDE__GRBM_MASK);
5648 
5649 		if (temp != data)
5650 			WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data);
5651 
5652 		/* 4 - wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */
5653 		gfx_v8_0_wait_for_rlc_serdes(adev);
5654 
5655 		/* 5 - clear mgcg override */
5656 		gfx_v8_0_send_serdes_cmd(adev, BPM_REG_MGCG_OVERRIDE, CLE_BPM_SERDES_CMD);
5657 
5658 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGTS) {
5659 			/* 6 - Enable CGTS(Tree Shade) MGCG /MGLS */
5660 			temp = data = RREG32(mmCGTS_SM_CTRL_REG);
5661 			data &= ~(CGTS_SM_CTRL_REG__SM_MODE_MASK);
5662 			data |= (0x2 << CGTS_SM_CTRL_REG__SM_MODE__SHIFT);
5663 			data |= CGTS_SM_CTRL_REG__SM_MODE_ENABLE_MASK;
5664 			data &= ~CGTS_SM_CTRL_REG__OVERRIDE_MASK;
5665 			if ((adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS) &&
5666 			    (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGTS_LS))
5667 				data &= ~CGTS_SM_CTRL_REG__LS_OVERRIDE_MASK;
5668 			data |= CGTS_SM_CTRL_REG__ON_MONITOR_ADD_EN_MASK;
5669 			data |= (0x96 << CGTS_SM_CTRL_REG__ON_MONITOR_ADD__SHIFT);
5670 			if (temp != data)
5671 				WREG32(mmCGTS_SM_CTRL_REG, data);
5672 		}
5673 		udelay(50);
5674 
5675 		/* 7 - wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */
5676 		gfx_v8_0_wait_for_rlc_serdes(adev);
5677 	} else {
5678 		/* 1 - MGCG_OVERRIDE[0] for CP and MGCG_OVERRIDE[1] for RLC */
5679 		temp = data = RREG32(mmRLC_CGTT_MGCG_OVERRIDE);
5680 		data |= (RLC_CGTT_MGCG_OVERRIDE__CPF_MASK |
5681 				RLC_CGTT_MGCG_OVERRIDE__RLC_MASK |
5682 				RLC_CGTT_MGCG_OVERRIDE__MGCG_MASK |
5683 				RLC_CGTT_MGCG_OVERRIDE__GRBM_MASK);
5684 		if (temp != data)
5685 			WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data);
5686 
5687 		/* 2 - disable MGLS in RLC */
5688 		data = RREG32(mmRLC_MEM_SLP_CNTL);
5689 		if (data & RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK) {
5690 			data &= ~RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK;
5691 			WREG32(mmRLC_MEM_SLP_CNTL, data);
5692 		}
5693 
5694 		/* 3 - disable MGLS in CP */
5695 		data = RREG32(mmCP_MEM_SLP_CNTL);
5696 		if (data & CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK) {
5697 			data &= ~CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK;
5698 			WREG32(mmCP_MEM_SLP_CNTL, data);
5699 		}
5700 
5701 		/* 4 - Disable CGTS(Tree Shade) MGCG and MGLS */
5702 		temp = data = RREG32(mmCGTS_SM_CTRL_REG);
5703 		data |= (CGTS_SM_CTRL_REG__OVERRIDE_MASK |
5704 				CGTS_SM_CTRL_REG__LS_OVERRIDE_MASK);
5705 		if (temp != data)
5706 			WREG32(mmCGTS_SM_CTRL_REG, data);
5707 
5708 		/* 5 - wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */
5709 		gfx_v8_0_wait_for_rlc_serdes(adev);
5710 
5711 		/* 6 - set mgcg override */
5712 		gfx_v8_0_send_serdes_cmd(adev, BPM_REG_MGCG_OVERRIDE, SET_BPM_SERDES_CMD);
5713 
5714 		udelay(50);
5715 
5716 		/* 7- wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */
5717 		gfx_v8_0_wait_for_rlc_serdes(adev);
5718 	}
5719 
5720 	amdgpu_gfx_rlc_exit_safe_mode(adev);
5721 }
5722 
5723 static void gfx_v8_0_update_coarse_grain_clock_gating(struct amdgpu_device *adev,
5724 						      bool enable)
5725 {
5726 	uint32_t temp, temp1, data, data1;
5727 
5728 	temp = data = RREG32(mmRLC_CGCG_CGLS_CTRL);
5729 
5730 	amdgpu_gfx_rlc_enter_safe_mode(adev);
5731 
5732 	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGCG)) {
5733 		temp1 = data1 =	RREG32(mmRLC_CGTT_MGCG_OVERRIDE);
5734 		data1 &= ~RLC_CGTT_MGCG_OVERRIDE__CGCG_MASK;
5735 		if (temp1 != data1)
5736 			WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data1);
5737 
5738 		/* : wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */
5739 		gfx_v8_0_wait_for_rlc_serdes(adev);
5740 
5741 		/* 2 - clear cgcg override */
5742 		gfx_v8_0_send_serdes_cmd(adev, BPM_REG_CGCG_OVERRIDE, CLE_BPM_SERDES_CMD);
5743 
5744 		/* wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */
5745 		gfx_v8_0_wait_for_rlc_serdes(adev);
5746 
5747 		/* 3 - write cmd to set CGLS */
5748 		gfx_v8_0_send_serdes_cmd(adev, BPM_REG_CGLS_EN, SET_BPM_SERDES_CMD);
5749 
5750 		/* 4 - enable cgcg */
5751 		data |= RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK;
5752 
5753 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGLS) {
5754 			/* enable cgls*/
5755 			data |= RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK;
5756 
5757 			temp1 = data1 =	RREG32(mmRLC_CGTT_MGCG_OVERRIDE);
5758 			data1 &= ~RLC_CGTT_MGCG_OVERRIDE__CGLS_MASK;
5759 
5760 			if (temp1 != data1)
5761 				WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data1);
5762 		} else {
5763 			data &= ~RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK;
5764 		}
5765 
5766 		if (temp != data)
5767 			WREG32(mmRLC_CGCG_CGLS_CTRL, data);
5768 
5769 		/* 5 enable cntx_empty_int_enable/cntx_busy_int_enable/
5770 		 * Cmp_busy/GFX_Idle interrupts
5771 		 */
5772 		gfx_v8_0_enable_gui_idle_interrupt(adev, true);
5773 	} else {
5774 		/* disable cntx_empty_int_enable & GFX Idle interrupt */
5775 		gfx_v8_0_enable_gui_idle_interrupt(adev, false);
5776 
5777 		/* TEST CGCG */
5778 		temp1 = data1 =	RREG32(mmRLC_CGTT_MGCG_OVERRIDE);
5779 		data1 |= (RLC_CGTT_MGCG_OVERRIDE__CGCG_MASK |
5780 				RLC_CGTT_MGCG_OVERRIDE__CGLS_MASK);
5781 		if (temp1 != data1)
5782 			WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data1);
5783 
5784 		/* read gfx register to wake up cgcg */
5785 		RREG32(mmCB_CGTT_SCLK_CTRL);
5786 		RREG32(mmCB_CGTT_SCLK_CTRL);
5787 		RREG32(mmCB_CGTT_SCLK_CTRL);
5788 		RREG32(mmCB_CGTT_SCLK_CTRL);
5789 
5790 		/* wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */
5791 		gfx_v8_0_wait_for_rlc_serdes(adev);
5792 
5793 		/* write cmd to Set CGCG Override */
5794 		gfx_v8_0_send_serdes_cmd(adev, BPM_REG_CGCG_OVERRIDE, SET_BPM_SERDES_CMD);
5795 
5796 		/* wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */
5797 		gfx_v8_0_wait_for_rlc_serdes(adev);
5798 
5799 		/* write cmd to Clear CGLS */
5800 		gfx_v8_0_send_serdes_cmd(adev, BPM_REG_CGLS_EN, CLE_BPM_SERDES_CMD);
5801 
5802 		/* disable cgcg, cgls should be disabled too. */
5803 		data &= ~(RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK |
5804 			  RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK);
5805 		if (temp != data)
5806 			WREG32(mmRLC_CGCG_CGLS_CTRL, data);
5807 		/* enable interrupts again for PG */
5808 		gfx_v8_0_enable_gui_idle_interrupt(adev, true);
5809 	}
5810 
5811 	gfx_v8_0_wait_for_rlc_serdes(adev);
5812 
5813 	amdgpu_gfx_rlc_exit_safe_mode(adev);
5814 }
5815 static int gfx_v8_0_update_gfx_clock_gating(struct amdgpu_device *adev,
5816 					    bool enable)
5817 {
5818 	if (enable) {
5819 		/* CGCG/CGLS should be enabled after MGCG/MGLS/TS(CG/LS)
5820 		 * ===  MGCG + MGLS + TS(CG/LS) ===
5821 		 */
5822 		gfx_v8_0_update_medium_grain_clock_gating(adev, enable);
5823 		gfx_v8_0_update_coarse_grain_clock_gating(adev, enable);
5824 	} else {
5825 		/* CGCG/CGLS should be disabled before MGCG/MGLS/TS(CG/LS)
5826 		 * ===  CGCG + CGLS ===
5827 		 */
5828 		gfx_v8_0_update_coarse_grain_clock_gating(adev, enable);
5829 		gfx_v8_0_update_medium_grain_clock_gating(adev, enable);
5830 	}
5831 	return 0;
5832 }
5833 
5834 static int gfx_v8_0_tonga_update_gfx_clock_gating(struct amdgpu_device *adev,
5835 					  enum amd_clockgating_state state)
5836 {
5837 	uint32_t msg_id, pp_state = 0;
5838 	uint32_t pp_support_state = 0;
5839 
5840 	if (adev->cg_flags & (AMD_CG_SUPPORT_GFX_CGCG | AMD_CG_SUPPORT_GFX_CGLS)) {
5841 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGLS) {
5842 			pp_support_state = PP_STATE_SUPPORT_LS;
5843 			pp_state = PP_STATE_LS;
5844 		}
5845 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGCG) {
5846 			pp_support_state |= PP_STATE_SUPPORT_CG;
5847 			pp_state |= PP_STATE_CG;
5848 		}
5849 		if (state == AMD_CG_STATE_UNGATE)
5850 			pp_state = 0;
5851 
5852 		msg_id = PP_CG_MSG_ID(PP_GROUP_GFX,
5853 				PP_BLOCK_GFX_CG,
5854 				pp_support_state,
5855 				pp_state);
5856 		amdgpu_dpm_set_clockgating_by_smu(adev, msg_id);
5857 	}
5858 
5859 	if (adev->cg_flags & (AMD_CG_SUPPORT_GFX_MGCG | AMD_CG_SUPPORT_GFX_MGLS)) {
5860 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS) {
5861 			pp_support_state = PP_STATE_SUPPORT_LS;
5862 			pp_state = PP_STATE_LS;
5863 		}
5864 
5865 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGCG) {
5866 			pp_support_state |= PP_STATE_SUPPORT_CG;
5867 			pp_state |= PP_STATE_CG;
5868 		}
5869 
5870 		if (state == AMD_CG_STATE_UNGATE)
5871 			pp_state = 0;
5872 
5873 		msg_id = PP_CG_MSG_ID(PP_GROUP_GFX,
5874 				PP_BLOCK_GFX_MG,
5875 				pp_support_state,
5876 				pp_state);
5877 		amdgpu_dpm_set_clockgating_by_smu(adev, msg_id);
5878 	}
5879 
5880 	return 0;
5881 }
5882 
5883 static int gfx_v8_0_polaris_update_gfx_clock_gating(struct amdgpu_device *adev,
5884 					  enum amd_clockgating_state state)
5885 {
5886 
5887 	uint32_t msg_id, pp_state = 0;
5888 	uint32_t pp_support_state = 0;
5889 
5890 	if (adev->cg_flags & (AMD_CG_SUPPORT_GFX_CGCG | AMD_CG_SUPPORT_GFX_CGLS)) {
5891 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGLS) {
5892 			pp_support_state = PP_STATE_SUPPORT_LS;
5893 			pp_state = PP_STATE_LS;
5894 		}
5895 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGCG) {
5896 			pp_support_state |= PP_STATE_SUPPORT_CG;
5897 			pp_state |= PP_STATE_CG;
5898 		}
5899 		if (state == AMD_CG_STATE_UNGATE)
5900 			pp_state = 0;
5901 
5902 		msg_id = PP_CG_MSG_ID(PP_GROUP_GFX,
5903 				PP_BLOCK_GFX_CG,
5904 				pp_support_state,
5905 				pp_state);
5906 		amdgpu_dpm_set_clockgating_by_smu(adev, msg_id);
5907 	}
5908 
5909 	if (adev->cg_flags & (AMD_CG_SUPPORT_GFX_3D_CGCG | AMD_CG_SUPPORT_GFX_3D_CGLS)) {
5910 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_3D_CGLS) {
5911 			pp_support_state = PP_STATE_SUPPORT_LS;
5912 			pp_state = PP_STATE_LS;
5913 		}
5914 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_3D_CGCG) {
5915 			pp_support_state |= PP_STATE_SUPPORT_CG;
5916 			pp_state |= PP_STATE_CG;
5917 		}
5918 		if (state == AMD_CG_STATE_UNGATE)
5919 			pp_state = 0;
5920 
5921 		msg_id = PP_CG_MSG_ID(PP_GROUP_GFX,
5922 				PP_BLOCK_GFX_3D,
5923 				pp_support_state,
5924 				pp_state);
5925 		amdgpu_dpm_set_clockgating_by_smu(adev, msg_id);
5926 	}
5927 
5928 	if (adev->cg_flags & (AMD_CG_SUPPORT_GFX_MGCG | AMD_CG_SUPPORT_GFX_MGLS)) {
5929 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS) {
5930 			pp_support_state = PP_STATE_SUPPORT_LS;
5931 			pp_state = PP_STATE_LS;
5932 		}
5933 
5934 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGCG) {
5935 			pp_support_state |= PP_STATE_SUPPORT_CG;
5936 			pp_state |= PP_STATE_CG;
5937 		}
5938 
5939 		if (state == AMD_CG_STATE_UNGATE)
5940 			pp_state = 0;
5941 
5942 		msg_id = PP_CG_MSG_ID(PP_GROUP_GFX,
5943 				PP_BLOCK_GFX_MG,
5944 				pp_support_state,
5945 				pp_state);
5946 		amdgpu_dpm_set_clockgating_by_smu(adev, msg_id);
5947 	}
5948 
5949 	if (adev->cg_flags & AMD_CG_SUPPORT_GFX_RLC_LS) {
5950 		pp_support_state = PP_STATE_SUPPORT_LS;
5951 
5952 		if (state == AMD_CG_STATE_UNGATE)
5953 			pp_state = 0;
5954 		else
5955 			pp_state = PP_STATE_LS;
5956 
5957 		msg_id = PP_CG_MSG_ID(PP_GROUP_GFX,
5958 				PP_BLOCK_GFX_RLC,
5959 				pp_support_state,
5960 				pp_state);
5961 		amdgpu_dpm_set_clockgating_by_smu(adev, msg_id);
5962 	}
5963 
5964 	if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CP_LS) {
5965 		pp_support_state = PP_STATE_SUPPORT_LS;
5966 
5967 		if (state == AMD_CG_STATE_UNGATE)
5968 			pp_state = 0;
5969 		else
5970 			pp_state = PP_STATE_LS;
5971 		msg_id = PP_CG_MSG_ID(PP_GROUP_GFX,
5972 			PP_BLOCK_GFX_CP,
5973 			pp_support_state,
5974 			pp_state);
5975 		amdgpu_dpm_set_clockgating_by_smu(adev, msg_id);
5976 	}
5977 
5978 	return 0;
5979 }
5980 
5981 static int gfx_v8_0_set_clockgating_state(void *handle,
5982 					  enum amd_clockgating_state state)
5983 {
5984 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5985 
5986 	if (amdgpu_sriov_vf(adev))
5987 		return 0;
5988 
5989 	switch (adev->asic_type) {
5990 	case CHIP_FIJI:
5991 	case CHIP_CARRIZO:
5992 	case CHIP_STONEY:
5993 		gfx_v8_0_update_gfx_clock_gating(adev,
5994 						 state == AMD_CG_STATE_GATE);
5995 		break;
5996 	case CHIP_TONGA:
5997 		gfx_v8_0_tonga_update_gfx_clock_gating(adev, state);
5998 		break;
5999 	case CHIP_POLARIS10:
6000 	case CHIP_POLARIS11:
6001 	case CHIP_POLARIS12:
6002 	case CHIP_VEGAM:
6003 		gfx_v8_0_polaris_update_gfx_clock_gating(adev, state);
6004 		break;
6005 	default:
6006 		break;
6007 	}
6008 	return 0;
6009 }
6010 
6011 static u64 gfx_v8_0_ring_get_rptr(struct amdgpu_ring *ring)
6012 {
6013 	return *ring->rptr_cpu_addr;
6014 }
6015 
6016 static u64 gfx_v8_0_ring_get_wptr_gfx(struct amdgpu_ring *ring)
6017 {
6018 	struct amdgpu_device *adev = ring->adev;
6019 
6020 	if (ring->use_doorbell)
6021 		/* XXX check if swapping is necessary on BE */
6022 		return *ring->wptr_cpu_addr;
6023 	else
6024 		return RREG32(mmCP_RB0_WPTR);
6025 }
6026 
6027 static void gfx_v8_0_ring_set_wptr_gfx(struct amdgpu_ring *ring)
6028 {
6029 	struct amdgpu_device *adev = ring->adev;
6030 
6031 	if (ring->use_doorbell) {
6032 		/* XXX check if swapping is necessary on BE */
6033 		*ring->wptr_cpu_addr = lower_32_bits(ring->wptr);
6034 		WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
6035 	} else {
6036 		WREG32(mmCP_RB0_WPTR, lower_32_bits(ring->wptr));
6037 		(void)RREG32(mmCP_RB0_WPTR);
6038 	}
6039 }
6040 
6041 static void gfx_v8_0_ring_emit_hdp_flush(struct amdgpu_ring *ring)
6042 {
6043 	u32 ref_and_mask, reg_mem_engine;
6044 
6045 	if ((ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE) ||
6046 	    (ring->funcs->type == AMDGPU_RING_TYPE_KIQ)) {
6047 		switch (ring->me) {
6048 		case 1:
6049 			ref_and_mask = GPU_HDP_FLUSH_DONE__CP2_MASK << ring->pipe;
6050 			break;
6051 		case 2:
6052 			ref_and_mask = GPU_HDP_FLUSH_DONE__CP6_MASK << ring->pipe;
6053 			break;
6054 		default:
6055 			return;
6056 		}
6057 		reg_mem_engine = 0;
6058 	} else {
6059 		ref_and_mask = GPU_HDP_FLUSH_DONE__CP0_MASK;
6060 		reg_mem_engine = WAIT_REG_MEM_ENGINE(1); /* pfp */
6061 	}
6062 
6063 	amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
6064 	amdgpu_ring_write(ring, (WAIT_REG_MEM_OPERATION(1) | /* write, wait, write */
6065 				 WAIT_REG_MEM_FUNCTION(3) |  /* == */
6066 				 reg_mem_engine));
6067 	amdgpu_ring_write(ring, mmGPU_HDP_FLUSH_REQ);
6068 	amdgpu_ring_write(ring, mmGPU_HDP_FLUSH_DONE);
6069 	amdgpu_ring_write(ring, ref_and_mask);
6070 	amdgpu_ring_write(ring, ref_and_mask);
6071 	amdgpu_ring_write(ring, 0x20); /* poll interval */
6072 }
6073 
6074 static void gfx_v8_0_ring_emit_vgt_flush(struct amdgpu_ring *ring)
6075 {
6076 	amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE, 0));
6077 	amdgpu_ring_write(ring, EVENT_TYPE(VS_PARTIAL_FLUSH) |
6078 		EVENT_INDEX(4));
6079 
6080 	amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE, 0));
6081 	amdgpu_ring_write(ring, EVENT_TYPE(VGT_FLUSH) |
6082 		EVENT_INDEX(0));
6083 }
6084 
6085 static void gfx_v8_0_ring_emit_ib_gfx(struct amdgpu_ring *ring,
6086 					struct amdgpu_job *job,
6087 					struct amdgpu_ib *ib,
6088 					uint32_t flags)
6089 {
6090 	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
6091 	u32 header, control = 0;
6092 
6093 	if (ib->flags & AMDGPU_IB_FLAG_CE)
6094 		header = PACKET3(PACKET3_INDIRECT_BUFFER_CONST, 2);
6095 	else
6096 		header = PACKET3(PACKET3_INDIRECT_BUFFER, 2);
6097 
6098 	control |= ib->length_dw | (vmid << 24);
6099 
6100 	if (amdgpu_sriov_vf(ring->adev) && (ib->flags & AMDGPU_IB_FLAG_PREEMPT)) {
6101 		control |= INDIRECT_BUFFER_PRE_ENB(1);
6102 
6103 		if (!(ib->flags & AMDGPU_IB_FLAG_CE) && vmid)
6104 			gfx_v8_0_ring_emit_de_meta(ring);
6105 	}
6106 
6107 	amdgpu_ring_write(ring, header);
6108 	amdgpu_ring_write(ring,
6109 #ifdef __BIG_ENDIAN
6110 			  (2 << 0) |
6111 #endif
6112 			  (ib->gpu_addr & 0xFFFFFFFC));
6113 	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr) & 0xFFFF);
6114 	amdgpu_ring_write(ring, control);
6115 }
6116 
6117 static void gfx_v8_0_ring_emit_ib_compute(struct amdgpu_ring *ring,
6118 					  struct amdgpu_job *job,
6119 					  struct amdgpu_ib *ib,
6120 					  uint32_t flags)
6121 {
6122 	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
6123 	u32 control = INDIRECT_BUFFER_VALID | ib->length_dw | (vmid << 24);
6124 
6125 	/* Currently, there is a high possibility to get wave ID mismatch
6126 	 * between ME and GDS, leading to a hw deadlock, because ME generates
6127 	 * different wave IDs than the GDS expects. This situation happens
6128 	 * randomly when at least 5 compute pipes use GDS ordered append.
6129 	 * The wave IDs generated by ME are also wrong after suspend/resume.
6130 	 * Those are probably bugs somewhere else in the kernel driver.
6131 	 *
6132 	 * Writing GDS_COMPUTE_MAX_WAVE_ID resets wave ID counters in ME and
6133 	 * GDS to 0 for this ring (me/pipe).
6134 	 */
6135 	if (ib->flags & AMDGPU_IB_FLAG_RESET_GDS_MAX_WAVE_ID) {
6136 		amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1));
6137 		amdgpu_ring_write(ring, mmGDS_COMPUTE_MAX_WAVE_ID - PACKET3_SET_CONFIG_REG_START);
6138 		amdgpu_ring_write(ring, ring->adev->gds.gds_compute_max_wave_id);
6139 	}
6140 
6141 	amdgpu_ring_write(ring, PACKET3(PACKET3_INDIRECT_BUFFER, 2));
6142 	amdgpu_ring_write(ring,
6143 #ifdef __BIG_ENDIAN
6144 				(2 << 0) |
6145 #endif
6146 				(ib->gpu_addr & 0xFFFFFFFC));
6147 	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr) & 0xFFFF);
6148 	amdgpu_ring_write(ring, control);
6149 }
6150 
6151 static void gfx_v8_0_ring_emit_fence_gfx(struct amdgpu_ring *ring, u64 addr,
6152 					 u64 seq, unsigned flags)
6153 {
6154 	bool write64bit = flags & AMDGPU_FENCE_FLAG_64BIT;
6155 	bool int_sel = flags & AMDGPU_FENCE_FLAG_INT;
6156 
6157 	/* Workaround for cache flush problems. First send a dummy EOP
6158 	 * event down the pipe with seq one below.
6159 	 */
6160 	amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE_EOP, 4));
6161 	amdgpu_ring_write(ring, (EOP_TCL1_ACTION_EN |
6162 				 EOP_TC_ACTION_EN |
6163 				 EOP_TC_WB_ACTION_EN |
6164 				 EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
6165 				 EVENT_INDEX(5)));
6166 	amdgpu_ring_write(ring, addr & 0xfffffffc);
6167 	amdgpu_ring_write(ring, (upper_32_bits(addr) & 0xffff) |
6168 				DATA_SEL(1) | INT_SEL(0));
6169 	amdgpu_ring_write(ring, lower_32_bits(seq - 1));
6170 	amdgpu_ring_write(ring, upper_32_bits(seq - 1));
6171 
6172 	/* Then send the real EOP event down the pipe:
6173 	 * EVENT_WRITE_EOP - flush caches, send int */
6174 	amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE_EOP, 4));
6175 	amdgpu_ring_write(ring, (EOP_TCL1_ACTION_EN |
6176 				 EOP_TC_ACTION_EN |
6177 				 EOP_TC_WB_ACTION_EN |
6178 				 EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
6179 				 EVENT_INDEX(5)));
6180 	amdgpu_ring_write(ring, addr & 0xfffffffc);
6181 	amdgpu_ring_write(ring, (upper_32_bits(addr) & 0xffff) |
6182 			  DATA_SEL(write64bit ? 2 : 1) | INT_SEL(int_sel ? 2 : 0));
6183 	amdgpu_ring_write(ring, lower_32_bits(seq));
6184 	amdgpu_ring_write(ring, upper_32_bits(seq));
6185 
6186 }
6187 
6188 static void gfx_v8_0_ring_emit_pipeline_sync(struct amdgpu_ring *ring)
6189 {
6190 	int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX);
6191 	uint32_t seq = ring->fence_drv.sync_seq;
6192 	uint64_t addr = ring->fence_drv.gpu_addr;
6193 
6194 	amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
6195 	amdgpu_ring_write(ring, (WAIT_REG_MEM_MEM_SPACE(1) | /* memory */
6196 				 WAIT_REG_MEM_FUNCTION(3) | /* equal */
6197 				 WAIT_REG_MEM_ENGINE(usepfp))); /* pfp or me */
6198 	amdgpu_ring_write(ring, addr & 0xfffffffc);
6199 	amdgpu_ring_write(ring, upper_32_bits(addr) & 0xffffffff);
6200 	amdgpu_ring_write(ring, seq);
6201 	amdgpu_ring_write(ring, 0xffffffff);
6202 	amdgpu_ring_write(ring, 4); /* poll interval */
6203 }
6204 
6205 static void gfx_v8_0_ring_emit_vm_flush(struct amdgpu_ring *ring,
6206 					unsigned vmid, uint64_t pd_addr)
6207 {
6208 	int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX);
6209 
6210 	amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr);
6211 
6212 	/* wait for the invalidate to complete */
6213 	amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
6214 	amdgpu_ring_write(ring, (WAIT_REG_MEM_OPERATION(0) | /* wait */
6215 				 WAIT_REG_MEM_FUNCTION(0) |  /* always */
6216 				 WAIT_REG_MEM_ENGINE(0))); /* me */
6217 	amdgpu_ring_write(ring, mmVM_INVALIDATE_REQUEST);
6218 	amdgpu_ring_write(ring, 0);
6219 	amdgpu_ring_write(ring, 0); /* ref */
6220 	amdgpu_ring_write(ring, 0); /* mask */
6221 	amdgpu_ring_write(ring, 0x20); /* poll interval */
6222 
6223 	/* compute doesn't have PFP */
6224 	if (usepfp) {
6225 		/* sync PFP to ME, otherwise we might get invalid PFP reads */
6226 		amdgpu_ring_write(ring, PACKET3(PACKET3_PFP_SYNC_ME, 0));
6227 		amdgpu_ring_write(ring, 0x0);
6228 	}
6229 }
6230 
6231 static u64 gfx_v8_0_ring_get_wptr_compute(struct amdgpu_ring *ring)
6232 {
6233 	return *ring->wptr_cpu_addr;
6234 }
6235 
6236 static void gfx_v8_0_ring_set_wptr_compute(struct amdgpu_ring *ring)
6237 {
6238 	struct amdgpu_device *adev = ring->adev;
6239 
6240 	/* XXX check if swapping is necessary on BE */
6241 	*ring->wptr_cpu_addr = lower_32_bits(ring->wptr);
6242 	WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
6243 }
6244 
6245 static void gfx_v8_0_ring_emit_fence_compute(struct amdgpu_ring *ring,
6246 					     u64 addr, u64 seq,
6247 					     unsigned flags)
6248 {
6249 	bool write64bit = flags & AMDGPU_FENCE_FLAG_64BIT;
6250 	bool int_sel = flags & AMDGPU_FENCE_FLAG_INT;
6251 
6252 	/* RELEASE_MEM - flush caches, send int */
6253 	amdgpu_ring_write(ring, PACKET3(PACKET3_RELEASE_MEM, 5));
6254 	amdgpu_ring_write(ring, (EOP_TCL1_ACTION_EN |
6255 				 EOP_TC_ACTION_EN |
6256 				 EOP_TC_WB_ACTION_EN |
6257 				 EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
6258 				 EVENT_INDEX(5)));
6259 	amdgpu_ring_write(ring, DATA_SEL(write64bit ? 2 : 1) | INT_SEL(int_sel ? 2 : 0));
6260 	amdgpu_ring_write(ring, addr & 0xfffffffc);
6261 	amdgpu_ring_write(ring, upper_32_bits(addr));
6262 	amdgpu_ring_write(ring, lower_32_bits(seq));
6263 	amdgpu_ring_write(ring, upper_32_bits(seq));
6264 }
6265 
6266 static void gfx_v8_0_ring_emit_fence_kiq(struct amdgpu_ring *ring, u64 addr,
6267 					 u64 seq, unsigned int flags)
6268 {
6269 	/* we only allocate 32bit for each seq wb address */
6270 	BUG_ON(flags & AMDGPU_FENCE_FLAG_64BIT);
6271 
6272 	/* write fence seq to the "addr" */
6273 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
6274 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
6275 				 WRITE_DATA_DST_SEL(5) | WR_CONFIRM));
6276 	amdgpu_ring_write(ring, lower_32_bits(addr));
6277 	amdgpu_ring_write(ring, upper_32_bits(addr));
6278 	amdgpu_ring_write(ring, lower_32_bits(seq));
6279 
6280 	if (flags & AMDGPU_FENCE_FLAG_INT) {
6281 		/* set register to trigger INT */
6282 		amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
6283 		amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
6284 					 WRITE_DATA_DST_SEL(0) | WR_CONFIRM));
6285 		amdgpu_ring_write(ring, mmCPC_INT_STATUS);
6286 		amdgpu_ring_write(ring, 0);
6287 		amdgpu_ring_write(ring, 0x20000000); /* src_id is 178 */
6288 	}
6289 }
6290 
6291 static void gfx_v8_ring_emit_sb(struct amdgpu_ring *ring)
6292 {
6293 	amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
6294 	amdgpu_ring_write(ring, 0);
6295 }
6296 
6297 static void gfx_v8_ring_emit_cntxcntl(struct amdgpu_ring *ring, uint32_t flags)
6298 {
6299 	uint32_t dw2 = 0;
6300 
6301 	if (amdgpu_sriov_vf(ring->adev))
6302 		gfx_v8_0_ring_emit_ce_meta(ring);
6303 
6304 	dw2 |= 0x80000000; /* set load_enable otherwise this package is just NOPs */
6305 	if (flags & AMDGPU_HAVE_CTX_SWITCH) {
6306 		gfx_v8_0_ring_emit_vgt_flush(ring);
6307 		/* set load_global_config & load_global_uconfig */
6308 		dw2 |= 0x8001;
6309 		/* set load_cs_sh_regs */
6310 		dw2 |= 0x01000000;
6311 		/* set load_per_context_state & load_gfx_sh_regs for GFX */
6312 		dw2 |= 0x10002;
6313 
6314 		/* set load_ce_ram if preamble presented */
6315 		if (AMDGPU_PREAMBLE_IB_PRESENT & flags)
6316 			dw2 |= 0x10000000;
6317 	} else {
6318 		/* still load_ce_ram if this is the first time preamble presented
6319 		 * although there is no context switch happens.
6320 		 */
6321 		if (AMDGPU_PREAMBLE_IB_PRESENT_FIRST & flags)
6322 			dw2 |= 0x10000000;
6323 	}
6324 
6325 	amdgpu_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1));
6326 	amdgpu_ring_write(ring, dw2);
6327 	amdgpu_ring_write(ring, 0);
6328 }
6329 
6330 static unsigned gfx_v8_0_ring_emit_init_cond_exec(struct amdgpu_ring *ring)
6331 {
6332 	unsigned ret;
6333 
6334 	amdgpu_ring_write(ring, PACKET3(PACKET3_COND_EXEC, 3));
6335 	amdgpu_ring_write(ring, lower_32_bits(ring->cond_exe_gpu_addr));
6336 	amdgpu_ring_write(ring, upper_32_bits(ring->cond_exe_gpu_addr));
6337 	amdgpu_ring_write(ring, 0); /* discard following DWs if *cond_exec_gpu_addr==0 */
6338 	ret = ring->wptr & ring->buf_mask;
6339 	amdgpu_ring_write(ring, 0x55aa55aa); /* patch dummy value later */
6340 	return ret;
6341 }
6342 
6343 static void gfx_v8_0_ring_emit_patch_cond_exec(struct amdgpu_ring *ring, unsigned offset)
6344 {
6345 	unsigned cur;
6346 
6347 	BUG_ON(offset > ring->buf_mask);
6348 	BUG_ON(ring->ring[offset] != 0x55aa55aa);
6349 
6350 	cur = (ring->wptr & ring->buf_mask) - 1;
6351 	if (likely(cur > offset))
6352 		ring->ring[offset] = cur - offset;
6353 	else
6354 		ring->ring[offset] = (ring->ring_size >> 2) - offset + cur;
6355 }
6356 
6357 static void gfx_v8_0_ring_emit_rreg(struct amdgpu_ring *ring, uint32_t reg,
6358 				    uint32_t reg_val_offs)
6359 {
6360 	struct amdgpu_device *adev = ring->adev;
6361 
6362 	amdgpu_ring_write(ring, PACKET3(PACKET3_COPY_DATA, 4));
6363 	amdgpu_ring_write(ring, 0 |	/* src: register*/
6364 				(5 << 8) |	/* dst: memory */
6365 				(1 << 20));	/* write confirm */
6366 	amdgpu_ring_write(ring, reg);
6367 	amdgpu_ring_write(ring, 0);
6368 	amdgpu_ring_write(ring, lower_32_bits(adev->wb.gpu_addr +
6369 				reg_val_offs * 4));
6370 	amdgpu_ring_write(ring, upper_32_bits(adev->wb.gpu_addr +
6371 				reg_val_offs * 4));
6372 }
6373 
6374 static void gfx_v8_0_ring_emit_wreg(struct amdgpu_ring *ring, uint32_t reg,
6375 				  uint32_t val)
6376 {
6377 	uint32_t cmd;
6378 
6379 	switch (ring->funcs->type) {
6380 	case AMDGPU_RING_TYPE_GFX:
6381 		cmd = WRITE_DATA_ENGINE_SEL(1) | WR_CONFIRM;
6382 		break;
6383 	case AMDGPU_RING_TYPE_KIQ:
6384 		cmd = 1 << 16; /* no inc addr */
6385 		break;
6386 	default:
6387 		cmd = WR_CONFIRM;
6388 		break;
6389 	}
6390 
6391 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
6392 	amdgpu_ring_write(ring, cmd);
6393 	amdgpu_ring_write(ring, reg);
6394 	amdgpu_ring_write(ring, 0);
6395 	amdgpu_ring_write(ring, val);
6396 }
6397 
6398 static void gfx_v8_0_ring_soft_recovery(struct amdgpu_ring *ring, unsigned vmid)
6399 {
6400 	struct amdgpu_device *adev = ring->adev;
6401 	uint32_t value = 0;
6402 
6403 	value = REG_SET_FIELD(value, SQ_CMD, CMD, 0x03);
6404 	value = REG_SET_FIELD(value, SQ_CMD, MODE, 0x01);
6405 	value = REG_SET_FIELD(value, SQ_CMD, CHECK_VMID, 1);
6406 	value = REG_SET_FIELD(value, SQ_CMD, VM_ID, vmid);
6407 	WREG32(mmSQ_CMD, value);
6408 }
6409 
6410 static void gfx_v8_0_set_gfx_eop_interrupt_state(struct amdgpu_device *adev,
6411 						 enum amdgpu_interrupt_state state)
6412 {
6413 	WREG32_FIELD(CP_INT_CNTL_RING0, TIME_STAMP_INT_ENABLE,
6414 		     state == AMDGPU_IRQ_STATE_DISABLE ? 0 : 1);
6415 }
6416 
6417 static void gfx_v8_0_set_compute_eop_interrupt_state(struct amdgpu_device *adev,
6418 						     int me, int pipe,
6419 						     enum amdgpu_interrupt_state state)
6420 {
6421 	u32 mec_int_cntl, mec_int_cntl_reg;
6422 
6423 	/*
6424 	 * amdgpu controls only the first MEC. That's why this function only
6425 	 * handles the setting of interrupts for this specific MEC. All other
6426 	 * pipes' interrupts are set by amdkfd.
6427 	 */
6428 
6429 	if (me == 1) {
6430 		switch (pipe) {
6431 		case 0:
6432 			mec_int_cntl_reg = mmCP_ME1_PIPE0_INT_CNTL;
6433 			break;
6434 		case 1:
6435 			mec_int_cntl_reg = mmCP_ME1_PIPE1_INT_CNTL;
6436 			break;
6437 		case 2:
6438 			mec_int_cntl_reg = mmCP_ME1_PIPE2_INT_CNTL;
6439 			break;
6440 		case 3:
6441 			mec_int_cntl_reg = mmCP_ME1_PIPE3_INT_CNTL;
6442 			break;
6443 		default:
6444 			DRM_DEBUG("invalid pipe %d\n", pipe);
6445 			return;
6446 		}
6447 	} else {
6448 		DRM_DEBUG("invalid me %d\n", me);
6449 		return;
6450 	}
6451 
6452 	switch (state) {
6453 	case AMDGPU_IRQ_STATE_DISABLE:
6454 		mec_int_cntl = RREG32(mec_int_cntl_reg);
6455 		mec_int_cntl &= ~CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK;
6456 		WREG32(mec_int_cntl_reg, mec_int_cntl);
6457 		break;
6458 	case AMDGPU_IRQ_STATE_ENABLE:
6459 		mec_int_cntl = RREG32(mec_int_cntl_reg);
6460 		mec_int_cntl |= CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK;
6461 		WREG32(mec_int_cntl_reg, mec_int_cntl);
6462 		break;
6463 	default:
6464 		break;
6465 	}
6466 }
6467 
6468 static int gfx_v8_0_set_priv_reg_fault_state(struct amdgpu_device *adev,
6469 					     struct amdgpu_irq_src *source,
6470 					     unsigned type,
6471 					     enum amdgpu_interrupt_state state)
6472 {
6473 	WREG32_FIELD(CP_INT_CNTL_RING0, PRIV_REG_INT_ENABLE,
6474 		     state == AMDGPU_IRQ_STATE_DISABLE ? 0 : 1);
6475 
6476 	return 0;
6477 }
6478 
6479 static int gfx_v8_0_set_priv_inst_fault_state(struct amdgpu_device *adev,
6480 					      struct amdgpu_irq_src *source,
6481 					      unsigned type,
6482 					      enum amdgpu_interrupt_state state)
6483 {
6484 	WREG32_FIELD(CP_INT_CNTL_RING0, PRIV_INSTR_INT_ENABLE,
6485 		     state == AMDGPU_IRQ_STATE_DISABLE ? 0 : 1);
6486 
6487 	return 0;
6488 }
6489 
6490 static int gfx_v8_0_set_eop_interrupt_state(struct amdgpu_device *adev,
6491 					    struct amdgpu_irq_src *src,
6492 					    unsigned type,
6493 					    enum amdgpu_interrupt_state state)
6494 {
6495 	switch (type) {
6496 	case AMDGPU_CP_IRQ_GFX_ME0_PIPE0_EOP:
6497 		gfx_v8_0_set_gfx_eop_interrupt_state(adev, state);
6498 		break;
6499 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP:
6500 		gfx_v8_0_set_compute_eop_interrupt_state(adev, 1, 0, state);
6501 		break;
6502 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE1_EOP:
6503 		gfx_v8_0_set_compute_eop_interrupt_state(adev, 1, 1, state);
6504 		break;
6505 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE2_EOP:
6506 		gfx_v8_0_set_compute_eop_interrupt_state(adev, 1, 2, state);
6507 		break;
6508 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE3_EOP:
6509 		gfx_v8_0_set_compute_eop_interrupt_state(adev, 1, 3, state);
6510 		break;
6511 	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE0_EOP:
6512 		gfx_v8_0_set_compute_eop_interrupt_state(adev, 2, 0, state);
6513 		break;
6514 	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE1_EOP:
6515 		gfx_v8_0_set_compute_eop_interrupt_state(adev, 2, 1, state);
6516 		break;
6517 	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE2_EOP:
6518 		gfx_v8_0_set_compute_eop_interrupt_state(adev, 2, 2, state);
6519 		break;
6520 	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE3_EOP:
6521 		gfx_v8_0_set_compute_eop_interrupt_state(adev, 2, 3, state);
6522 		break;
6523 	default:
6524 		break;
6525 	}
6526 	return 0;
6527 }
6528 
6529 static int gfx_v8_0_set_cp_ecc_int_state(struct amdgpu_device *adev,
6530 					 struct amdgpu_irq_src *source,
6531 					 unsigned int type,
6532 					 enum amdgpu_interrupt_state state)
6533 {
6534 	int enable_flag;
6535 
6536 	switch (state) {
6537 	case AMDGPU_IRQ_STATE_DISABLE:
6538 		enable_flag = 0;
6539 		break;
6540 
6541 	case AMDGPU_IRQ_STATE_ENABLE:
6542 		enable_flag = 1;
6543 		break;
6544 
6545 	default:
6546 		return -EINVAL;
6547 	}
6548 
6549 	WREG32_FIELD(CP_INT_CNTL, CP_ECC_ERROR_INT_ENABLE, enable_flag);
6550 	WREG32_FIELD(CP_INT_CNTL_RING0, CP_ECC_ERROR_INT_ENABLE, enable_flag);
6551 	WREG32_FIELD(CP_INT_CNTL_RING1, CP_ECC_ERROR_INT_ENABLE, enable_flag);
6552 	WREG32_FIELD(CP_INT_CNTL_RING2, CP_ECC_ERROR_INT_ENABLE, enable_flag);
6553 	WREG32_FIELD(CPC_INT_CNTL, CP_ECC_ERROR_INT_ENABLE, enable_flag);
6554 	WREG32_FIELD(CP_ME1_PIPE0_INT_CNTL, CP_ECC_ERROR_INT_ENABLE,
6555 		     enable_flag);
6556 	WREG32_FIELD(CP_ME1_PIPE1_INT_CNTL, CP_ECC_ERROR_INT_ENABLE,
6557 		     enable_flag);
6558 	WREG32_FIELD(CP_ME1_PIPE2_INT_CNTL, CP_ECC_ERROR_INT_ENABLE,
6559 		     enable_flag);
6560 	WREG32_FIELD(CP_ME1_PIPE3_INT_CNTL, CP_ECC_ERROR_INT_ENABLE,
6561 		     enable_flag);
6562 	WREG32_FIELD(CP_ME2_PIPE0_INT_CNTL, CP_ECC_ERROR_INT_ENABLE,
6563 		     enable_flag);
6564 	WREG32_FIELD(CP_ME2_PIPE1_INT_CNTL, CP_ECC_ERROR_INT_ENABLE,
6565 		     enable_flag);
6566 	WREG32_FIELD(CP_ME2_PIPE2_INT_CNTL, CP_ECC_ERROR_INT_ENABLE,
6567 		     enable_flag);
6568 	WREG32_FIELD(CP_ME2_PIPE3_INT_CNTL, CP_ECC_ERROR_INT_ENABLE,
6569 		     enable_flag);
6570 
6571 	return 0;
6572 }
6573 
6574 static int gfx_v8_0_set_sq_int_state(struct amdgpu_device *adev,
6575 				     struct amdgpu_irq_src *source,
6576 				     unsigned int type,
6577 				     enum amdgpu_interrupt_state state)
6578 {
6579 	int enable_flag;
6580 
6581 	switch (state) {
6582 	case AMDGPU_IRQ_STATE_DISABLE:
6583 		enable_flag = 1;
6584 		break;
6585 
6586 	case AMDGPU_IRQ_STATE_ENABLE:
6587 		enable_flag = 0;
6588 		break;
6589 
6590 	default:
6591 		return -EINVAL;
6592 	}
6593 
6594 	WREG32_FIELD(SQ_INTERRUPT_MSG_CTRL, STALL,
6595 		     enable_flag);
6596 
6597 	return 0;
6598 }
6599 
6600 static int gfx_v8_0_eop_irq(struct amdgpu_device *adev,
6601 			    struct amdgpu_irq_src *source,
6602 			    struct amdgpu_iv_entry *entry)
6603 {
6604 	int i;
6605 	u8 me_id, pipe_id, queue_id;
6606 	struct amdgpu_ring *ring;
6607 
6608 	DRM_DEBUG("IH: CP EOP\n");
6609 	me_id = (entry->ring_id & 0x0c) >> 2;
6610 	pipe_id = (entry->ring_id & 0x03) >> 0;
6611 	queue_id = (entry->ring_id & 0x70) >> 4;
6612 
6613 	switch (me_id) {
6614 	case 0:
6615 		amdgpu_fence_process(&adev->gfx.gfx_ring[0]);
6616 		break;
6617 	case 1:
6618 	case 2:
6619 		for (i = 0; i < adev->gfx.num_compute_rings; i++) {
6620 			ring = &adev->gfx.compute_ring[i];
6621 			/* Per-queue interrupt is supported for MEC starting from VI.
6622 			  * The interrupt can only be enabled/disabled per pipe instead of per queue.
6623 			  */
6624 			if ((ring->me == me_id) && (ring->pipe == pipe_id) && (ring->queue == queue_id))
6625 				amdgpu_fence_process(ring);
6626 		}
6627 		break;
6628 	}
6629 	return 0;
6630 }
6631 
6632 static void gfx_v8_0_fault(struct amdgpu_device *adev,
6633 			   struct amdgpu_iv_entry *entry)
6634 {
6635 	u8 me_id, pipe_id, queue_id;
6636 	struct amdgpu_ring *ring;
6637 	int i;
6638 
6639 	me_id = (entry->ring_id & 0x0c) >> 2;
6640 	pipe_id = (entry->ring_id & 0x03) >> 0;
6641 	queue_id = (entry->ring_id & 0x70) >> 4;
6642 
6643 	switch (me_id) {
6644 	case 0:
6645 		drm_sched_fault(&adev->gfx.gfx_ring[0].sched);
6646 		break;
6647 	case 1:
6648 	case 2:
6649 		for (i = 0; i < adev->gfx.num_compute_rings; i++) {
6650 			ring = &adev->gfx.compute_ring[i];
6651 			if (ring->me == me_id && ring->pipe == pipe_id &&
6652 			    ring->queue == queue_id)
6653 				drm_sched_fault(&ring->sched);
6654 		}
6655 		break;
6656 	}
6657 }
6658 
6659 static int gfx_v8_0_priv_reg_irq(struct amdgpu_device *adev,
6660 				 struct amdgpu_irq_src *source,
6661 				 struct amdgpu_iv_entry *entry)
6662 {
6663 	DRM_ERROR("Illegal register access in command stream\n");
6664 	gfx_v8_0_fault(adev, entry);
6665 	return 0;
6666 }
6667 
6668 static int gfx_v8_0_priv_inst_irq(struct amdgpu_device *adev,
6669 				  struct amdgpu_irq_src *source,
6670 				  struct amdgpu_iv_entry *entry)
6671 {
6672 	DRM_ERROR("Illegal instruction in command stream\n");
6673 	gfx_v8_0_fault(adev, entry);
6674 	return 0;
6675 }
6676 
6677 static int gfx_v8_0_cp_ecc_error_irq(struct amdgpu_device *adev,
6678 				     struct amdgpu_irq_src *source,
6679 				     struct amdgpu_iv_entry *entry)
6680 {
6681 	DRM_ERROR("CP EDC/ECC error detected.");
6682 	return 0;
6683 }
6684 
6685 static void gfx_v8_0_parse_sq_irq(struct amdgpu_device *adev, unsigned ih_data,
6686 				  bool from_wq)
6687 {
6688 	u32 enc, se_id, sh_id, cu_id;
6689 	char type[20];
6690 	int sq_edc_source = -1;
6691 
6692 	enc = REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_CMN, ENCODING);
6693 	se_id = REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_CMN, SE_ID);
6694 
6695 	switch (enc) {
6696 		case 0:
6697 			DRM_INFO("SQ general purpose intr detected:"
6698 					"se_id %d, immed_overflow %d, host_reg_overflow %d,"
6699 					"host_cmd_overflow %d, cmd_timestamp %d,"
6700 					"reg_timestamp %d, thread_trace_buff_full %d,"
6701 					"wlt %d, thread_trace %d.\n",
6702 					se_id,
6703 					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_AUTO, IMMED_OVERFLOW),
6704 					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_AUTO, HOST_REG_OVERFLOW),
6705 					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_AUTO, HOST_CMD_OVERFLOW),
6706 					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_AUTO, CMD_TIMESTAMP),
6707 					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_AUTO, REG_TIMESTAMP),
6708 					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_AUTO, THREAD_TRACE_BUF_FULL),
6709 					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_AUTO, WLT),
6710 					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_AUTO, THREAD_TRACE)
6711 					);
6712 			break;
6713 		case 1:
6714 		case 2:
6715 
6716 			cu_id = REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_WAVE, CU_ID);
6717 			sh_id = REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_WAVE, SH_ID);
6718 
6719 			/*
6720 			 * This function can be called either directly from ISR
6721 			 * or from BH in which case we can access SQ_EDC_INFO
6722 			 * instance
6723 			 */
6724 			if (from_wq) {
6725 				mutex_lock(&adev->grbm_idx_mutex);
6726 				gfx_v8_0_select_se_sh(adev, se_id, sh_id, cu_id);
6727 
6728 				sq_edc_source = REG_GET_FIELD(RREG32(mmSQ_EDC_INFO), SQ_EDC_INFO, SOURCE);
6729 
6730 				gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
6731 				mutex_unlock(&adev->grbm_idx_mutex);
6732 			}
6733 
6734 			if (enc == 1)
6735 				sprintf(type, "instruction intr");
6736 			else
6737 				sprintf(type, "EDC/ECC error");
6738 
6739 			DRM_INFO(
6740 				"SQ %s detected: "
6741 					"se_id %d, sh_id %d, cu_id %d, simd_id %d, wave_id %d, vm_id %d "
6742 					"trap %s, sq_ed_info.source %s.\n",
6743 					type, se_id, sh_id, cu_id,
6744 					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_WAVE, SIMD_ID),
6745 					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_WAVE, WAVE_ID),
6746 					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_WAVE, VM_ID),
6747 					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_WAVE, PRIV) ? "true" : "false",
6748 					(sq_edc_source != -1) ? sq_edc_source_names[sq_edc_source] : "unavailable"
6749 				);
6750 			break;
6751 		default:
6752 			DRM_ERROR("SQ invalid encoding type\n.");
6753 	}
6754 }
6755 
6756 static void gfx_v8_0_sq_irq_work_func(struct work_struct *work)
6757 {
6758 
6759 	struct amdgpu_device *adev = container_of(work, struct amdgpu_device, gfx.sq_work.work);
6760 	struct sq_work *sq_work = container_of(work, struct sq_work, work);
6761 
6762 	gfx_v8_0_parse_sq_irq(adev, sq_work->ih_data, true);
6763 }
6764 
6765 static int gfx_v8_0_sq_irq(struct amdgpu_device *adev,
6766 			   struct amdgpu_irq_src *source,
6767 			   struct amdgpu_iv_entry *entry)
6768 {
6769 	unsigned ih_data = entry->src_data[0];
6770 
6771 	/*
6772 	 * Try to submit work so SQ_EDC_INFO can be accessed from
6773 	 * BH. If previous work submission hasn't finished yet
6774 	 * just print whatever info is possible directly from the ISR.
6775 	 */
6776 	if (work_pending(&adev->gfx.sq_work.work)) {
6777 		gfx_v8_0_parse_sq_irq(adev, ih_data, false);
6778 	} else {
6779 		adev->gfx.sq_work.ih_data = ih_data;
6780 		schedule_work(&adev->gfx.sq_work.work);
6781 	}
6782 
6783 	return 0;
6784 }
6785 
6786 static void gfx_v8_0_emit_mem_sync(struct amdgpu_ring *ring)
6787 {
6788 	amdgpu_ring_write(ring, PACKET3(PACKET3_SURFACE_SYNC, 3));
6789 	amdgpu_ring_write(ring, PACKET3_TCL1_ACTION_ENA |
6790 			  PACKET3_TC_ACTION_ENA |
6791 			  PACKET3_SH_KCACHE_ACTION_ENA |
6792 			  PACKET3_SH_ICACHE_ACTION_ENA |
6793 			  PACKET3_TC_WB_ACTION_ENA);  /* CP_COHER_CNTL */
6794 	amdgpu_ring_write(ring, 0xffffffff);  /* CP_COHER_SIZE */
6795 	amdgpu_ring_write(ring, 0);  /* CP_COHER_BASE */
6796 	amdgpu_ring_write(ring, 0x0000000A); /* poll interval */
6797 }
6798 
6799 static void gfx_v8_0_emit_mem_sync_compute(struct amdgpu_ring *ring)
6800 {
6801 	amdgpu_ring_write(ring, PACKET3(PACKET3_ACQUIRE_MEM, 5));
6802 	amdgpu_ring_write(ring, PACKET3_TCL1_ACTION_ENA |
6803 			  PACKET3_TC_ACTION_ENA |
6804 			  PACKET3_SH_KCACHE_ACTION_ENA |
6805 			  PACKET3_SH_ICACHE_ACTION_ENA |
6806 			  PACKET3_TC_WB_ACTION_ENA);  /* CP_COHER_CNTL */
6807 	amdgpu_ring_write(ring, 0xffffffff);	/* CP_COHER_SIZE */
6808 	amdgpu_ring_write(ring, 0xff);		/* CP_COHER_SIZE_HI */
6809 	amdgpu_ring_write(ring, 0);		/* CP_COHER_BASE */
6810 	amdgpu_ring_write(ring, 0);		/* CP_COHER_BASE_HI */
6811 	amdgpu_ring_write(ring, 0x0000000A);	/* poll interval */
6812 }
6813 
6814 
6815 /* mmSPI_WCL_PIPE_PERCENT_CS[0-7]_DEFAULT values are same */
6816 #define mmSPI_WCL_PIPE_PERCENT_CS_DEFAULT	0x0000007f
6817 static void gfx_v8_0_emit_wave_limit_cs(struct amdgpu_ring *ring,
6818 					uint32_t pipe, bool enable)
6819 {
6820 	uint32_t val;
6821 	uint32_t wcl_cs_reg;
6822 
6823 	val = enable ? 0x1 : mmSPI_WCL_PIPE_PERCENT_CS_DEFAULT;
6824 
6825 	switch (pipe) {
6826 	case 0:
6827 		wcl_cs_reg = mmSPI_WCL_PIPE_PERCENT_CS0;
6828 		break;
6829 	case 1:
6830 		wcl_cs_reg = mmSPI_WCL_PIPE_PERCENT_CS1;
6831 		break;
6832 	case 2:
6833 		wcl_cs_reg = mmSPI_WCL_PIPE_PERCENT_CS2;
6834 		break;
6835 	case 3:
6836 		wcl_cs_reg = mmSPI_WCL_PIPE_PERCENT_CS3;
6837 		break;
6838 	default:
6839 		DRM_DEBUG("invalid pipe %d\n", pipe);
6840 		return;
6841 	}
6842 
6843 	amdgpu_ring_emit_wreg(ring, wcl_cs_reg, val);
6844 
6845 }
6846 
6847 #define mmSPI_WCL_PIPE_PERCENT_GFX_DEFAULT	0x07ffffff
6848 static void gfx_v8_0_emit_wave_limit(struct amdgpu_ring *ring, bool enable)
6849 {
6850 	struct amdgpu_device *adev = ring->adev;
6851 	uint32_t val;
6852 	int i;
6853 
6854 	/* mmSPI_WCL_PIPE_PERCENT_GFX is 7 bit multiplier register to limit
6855 	 * number of gfx waves. Setting 5 bit will make sure gfx only gets
6856 	 * around 25% of gpu resources.
6857 	 */
6858 	val = enable ? 0x1f : mmSPI_WCL_PIPE_PERCENT_GFX_DEFAULT;
6859 	amdgpu_ring_emit_wreg(ring, mmSPI_WCL_PIPE_PERCENT_GFX, val);
6860 
6861 	/* Restrict waves for normal/low priority compute queues as well
6862 	 * to get best QoS for high priority compute jobs.
6863 	 *
6864 	 * amdgpu controls only 1st ME(0-3 CS pipes).
6865 	 */
6866 	for (i = 0; i < adev->gfx.mec.num_pipe_per_mec; i++) {
6867 		if (i != ring->pipe)
6868 			gfx_v8_0_emit_wave_limit_cs(ring, i, enable);
6869 
6870 	}
6871 
6872 }
6873 
6874 static const struct amd_ip_funcs gfx_v8_0_ip_funcs = {
6875 	.name = "gfx_v8_0",
6876 	.early_init = gfx_v8_0_early_init,
6877 	.late_init = gfx_v8_0_late_init,
6878 	.sw_init = gfx_v8_0_sw_init,
6879 	.sw_fini = gfx_v8_0_sw_fini,
6880 	.hw_init = gfx_v8_0_hw_init,
6881 	.hw_fini = gfx_v8_0_hw_fini,
6882 	.suspend = gfx_v8_0_suspend,
6883 	.resume = gfx_v8_0_resume,
6884 	.is_idle = gfx_v8_0_is_idle,
6885 	.wait_for_idle = gfx_v8_0_wait_for_idle,
6886 	.check_soft_reset = gfx_v8_0_check_soft_reset,
6887 	.pre_soft_reset = gfx_v8_0_pre_soft_reset,
6888 	.soft_reset = gfx_v8_0_soft_reset,
6889 	.post_soft_reset = gfx_v8_0_post_soft_reset,
6890 	.set_clockgating_state = gfx_v8_0_set_clockgating_state,
6891 	.set_powergating_state = gfx_v8_0_set_powergating_state,
6892 	.get_clockgating_state = gfx_v8_0_get_clockgating_state,
6893 };
6894 
6895 static const struct amdgpu_ring_funcs gfx_v8_0_ring_funcs_gfx = {
6896 	.type = AMDGPU_RING_TYPE_GFX,
6897 	.align_mask = 0xff,
6898 	.nop = PACKET3(PACKET3_NOP, 0x3FFF),
6899 	.support_64bit_ptrs = false,
6900 	.get_rptr = gfx_v8_0_ring_get_rptr,
6901 	.get_wptr = gfx_v8_0_ring_get_wptr_gfx,
6902 	.set_wptr = gfx_v8_0_ring_set_wptr_gfx,
6903 	.emit_frame_size = /* maximum 215dw if count 16 IBs in */
6904 		5 +  /* COND_EXEC */
6905 		7 +  /* PIPELINE_SYNC */
6906 		VI_FLUSH_GPU_TLB_NUM_WREG * 5 + 9 + /* VM_FLUSH */
6907 		12 +  /* FENCE for VM_FLUSH */
6908 		20 + /* GDS switch */
6909 		4 + /* double SWITCH_BUFFER,
6910 		       the first COND_EXEC jump to the place just
6911 			   prior to this double SWITCH_BUFFER  */
6912 		5 + /* COND_EXEC */
6913 		7 +	 /*	HDP_flush */
6914 		4 +	 /*	VGT_flush */
6915 		14 + /*	CE_META */
6916 		31 + /*	DE_META */
6917 		3 + /* CNTX_CTRL */
6918 		5 + /* HDP_INVL */
6919 		12 + 12 + /* FENCE x2 */
6920 		2 + /* SWITCH_BUFFER */
6921 		5, /* SURFACE_SYNC */
6922 	.emit_ib_size =	4, /* gfx_v8_0_ring_emit_ib_gfx */
6923 	.emit_ib = gfx_v8_0_ring_emit_ib_gfx,
6924 	.emit_fence = gfx_v8_0_ring_emit_fence_gfx,
6925 	.emit_pipeline_sync = gfx_v8_0_ring_emit_pipeline_sync,
6926 	.emit_vm_flush = gfx_v8_0_ring_emit_vm_flush,
6927 	.emit_gds_switch = gfx_v8_0_ring_emit_gds_switch,
6928 	.emit_hdp_flush = gfx_v8_0_ring_emit_hdp_flush,
6929 	.test_ring = gfx_v8_0_ring_test_ring,
6930 	.test_ib = gfx_v8_0_ring_test_ib,
6931 	.insert_nop = amdgpu_ring_insert_nop,
6932 	.pad_ib = amdgpu_ring_generic_pad_ib,
6933 	.emit_switch_buffer = gfx_v8_ring_emit_sb,
6934 	.emit_cntxcntl = gfx_v8_ring_emit_cntxcntl,
6935 	.init_cond_exec = gfx_v8_0_ring_emit_init_cond_exec,
6936 	.patch_cond_exec = gfx_v8_0_ring_emit_patch_cond_exec,
6937 	.emit_wreg = gfx_v8_0_ring_emit_wreg,
6938 	.soft_recovery = gfx_v8_0_ring_soft_recovery,
6939 	.emit_mem_sync = gfx_v8_0_emit_mem_sync,
6940 };
6941 
6942 static const struct amdgpu_ring_funcs gfx_v8_0_ring_funcs_compute = {
6943 	.type = AMDGPU_RING_TYPE_COMPUTE,
6944 	.align_mask = 0xff,
6945 	.nop = PACKET3(PACKET3_NOP, 0x3FFF),
6946 	.support_64bit_ptrs = false,
6947 	.get_rptr = gfx_v8_0_ring_get_rptr,
6948 	.get_wptr = gfx_v8_0_ring_get_wptr_compute,
6949 	.set_wptr = gfx_v8_0_ring_set_wptr_compute,
6950 	.emit_frame_size =
6951 		20 + /* gfx_v8_0_ring_emit_gds_switch */
6952 		7 + /* gfx_v8_0_ring_emit_hdp_flush */
6953 		5 + /* hdp_invalidate */
6954 		7 + /* gfx_v8_0_ring_emit_pipeline_sync */
6955 		VI_FLUSH_GPU_TLB_NUM_WREG * 5 + 7 + /* gfx_v8_0_ring_emit_vm_flush */
6956 		7 + 7 + 7 + /* gfx_v8_0_ring_emit_fence_compute x3 for user fence, vm fence */
6957 		7 + /* gfx_v8_0_emit_mem_sync_compute */
6958 		5 + /* gfx_v8_0_emit_wave_limit for updating mmSPI_WCL_PIPE_PERCENT_GFX register */
6959 		15, /* for updating 3 mmSPI_WCL_PIPE_PERCENT_CS registers */
6960 	.emit_ib_size =	7, /* gfx_v8_0_ring_emit_ib_compute */
6961 	.emit_ib = gfx_v8_0_ring_emit_ib_compute,
6962 	.emit_fence = gfx_v8_0_ring_emit_fence_compute,
6963 	.emit_pipeline_sync = gfx_v8_0_ring_emit_pipeline_sync,
6964 	.emit_vm_flush = gfx_v8_0_ring_emit_vm_flush,
6965 	.emit_gds_switch = gfx_v8_0_ring_emit_gds_switch,
6966 	.emit_hdp_flush = gfx_v8_0_ring_emit_hdp_flush,
6967 	.test_ring = gfx_v8_0_ring_test_ring,
6968 	.test_ib = gfx_v8_0_ring_test_ib,
6969 	.insert_nop = amdgpu_ring_insert_nop,
6970 	.pad_ib = amdgpu_ring_generic_pad_ib,
6971 	.emit_wreg = gfx_v8_0_ring_emit_wreg,
6972 	.emit_mem_sync = gfx_v8_0_emit_mem_sync_compute,
6973 	.emit_wave_limit = gfx_v8_0_emit_wave_limit,
6974 };
6975 
6976 static const struct amdgpu_ring_funcs gfx_v8_0_ring_funcs_kiq = {
6977 	.type = AMDGPU_RING_TYPE_KIQ,
6978 	.align_mask = 0xff,
6979 	.nop = PACKET3(PACKET3_NOP, 0x3FFF),
6980 	.support_64bit_ptrs = false,
6981 	.get_rptr = gfx_v8_0_ring_get_rptr,
6982 	.get_wptr = gfx_v8_0_ring_get_wptr_compute,
6983 	.set_wptr = gfx_v8_0_ring_set_wptr_compute,
6984 	.emit_frame_size =
6985 		20 + /* gfx_v8_0_ring_emit_gds_switch */
6986 		7 + /* gfx_v8_0_ring_emit_hdp_flush */
6987 		5 + /* hdp_invalidate */
6988 		7 + /* gfx_v8_0_ring_emit_pipeline_sync */
6989 		17 + /* gfx_v8_0_ring_emit_vm_flush */
6990 		7 + 7 + 7, /* gfx_v8_0_ring_emit_fence_kiq x3 for user fence, vm fence */
6991 	.emit_ib_size =	7, /* gfx_v8_0_ring_emit_ib_compute */
6992 	.emit_fence = gfx_v8_0_ring_emit_fence_kiq,
6993 	.test_ring = gfx_v8_0_ring_test_ring,
6994 	.insert_nop = amdgpu_ring_insert_nop,
6995 	.pad_ib = amdgpu_ring_generic_pad_ib,
6996 	.emit_rreg = gfx_v8_0_ring_emit_rreg,
6997 	.emit_wreg = gfx_v8_0_ring_emit_wreg,
6998 };
6999 
7000 static void gfx_v8_0_set_ring_funcs(struct amdgpu_device *adev)
7001 {
7002 	int i;
7003 
7004 	adev->gfx.kiq.ring.funcs = &gfx_v8_0_ring_funcs_kiq;
7005 
7006 	for (i = 0; i < adev->gfx.num_gfx_rings; i++)
7007 		adev->gfx.gfx_ring[i].funcs = &gfx_v8_0_ring_funcs_gfx;
7008 
7009 	for (i = 0; i < adev->gfx.num_compute_rings; i++)
7010 		adev->gfx.compute_ring[i].funcs = &gfx_v8_0_ring_funcs_compute;
7011 }
7012 
7013 static const struct amdgpu_irq_src_funcs gfx_v8_0_eop_irq_funcs = {
7014 	.set = gfx_v8_0_set_eop_interrupt_state,
7015 	.process = gfx_v8_0_eop_irq,
7016 };
7017 
7018 static const struct amdgpu_irq_src_funcs gfx_v8_0_priv_reg_irq_funcs = {
7019 	.set = gfx_v8_0_set_priv_reg_fault_state,
7020 	.process = gfx_v8_0_priv_reg_irq,
7021 };
7022 
7023 static const struct amdgpu_irq_src_funcs gfx_v8_0_priv_inst_irq_funcs = {
7024 	.set = gfx_v8_0_set_priv_inst_fault_state,
7025 	.process = gfx_v8_0_priv_inst_irq,
7026 };
7027 
7028 static const struct amdgpu_irq_src_funcs gfx_v8_0_cp_ecc_error_irq_funcs = {
7029 	.set = gfx_v8_0_set_cp_ecc_int_state,
7030 	.process = gfx_v8_0_cp_ecc_error_irq,
7031 };
7032 
7033 static const struct amdgpu_irq_src_funcs gfx_v8_0_sq_irq_funcs = {
7034 	.set = gfx_v8_0_set_sq_int_state,
7035 	.process = gfx_v8_0_sq_irq,
7036 };
7037 
7038 static void gfx_v8_0_set_irq_funcs(struct amdgpu_device *adev)
7039 {
7040 	adev->gfx.eop_irq.num_types = AMDGPU_CP_IRQ_LAST;
7041 	adev->gfx.eop_irq.funcs = &gfx_v8_0_eop_irq_funcs;
7042 
7043 	adev->gfx.priv_reg_irq.num_types = 1;
7044 	adev->gfx.priv_reg_irq.funcs = &gfx_v8_0_priv_reg_irq_funcs;
7045 
7046 	adev->gfx.priv_inst_irq.num_types = 1;
7047 	adev->gfx.priv_inst_irq.funcs = &gfx_v8_0_priv_inst_irq_funcs;
7048 
7049 	adev->gfx.cp_ecc_error_irq.num_types = 1;
7050 	adev->gfx.cp_ecc_error_irq.funcs = &gfx_v8_0_cp_ecc_error_irq_funcs;
7051 
7052 	adev->gfx.sq_irq.num_types = 1;
7053 	adev->gfx.sq_irq.funcs = &gfx_v8_0_sq_irq_funcs;
7054 }
7055 
7056 static void gfx_v8_0_set_rlc_funcs(struct amdgpu_device *adev)
7057 {
7058 	adev->gfx.rlc.funcs = &iceland_rlc_funcs;
7059 }
7060 
7061 static void gfx_v8_0_set_gds_init(struct amdgpu_device *adev)
7062 {
7063 	/* init asci gds info */
7064 	adev->gds.gds_size = RREG32(mmGDS_VMID0_SIZE);
7065 	adev->gds.gws_size = 64;
7066 	adev->gds.oa_size = 16;
7067 	adev->gds.gds_compute_max_wave_id = RREG32(mmGDS_COMPUTE_MAX_WAVE_ID);
7068 }
7069 
7070 static void gfx_v8_0_set_user_cu_inactive_bitmap(struct amdgpu_device *adev,
7071 						 u32 bitmap)
7072 {
7073 	u32 data;
7074 
7075 	if (!bitmap)
7076 		return;
7077 
7078 	data = bitmap << GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_CUS__SHIFT;
7079 	data &= GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_CUS_MASK;
7080 
7081 	WREG32(mmGC_USER_SHADER_ARRAY_CONFIG, data);
7082 }
7083 
7084 static u32 gfx_v8_0_get_cu_active_bitmap(struct amdgpu_device *adev)
7085 {
7086 	u32 data, mask;
7087 
7088 	data =  RREG32(mmCC_GC_SHADER_ARRAY_CONFIG) |
7089 		RREG32(mmGC_USER_SHADER_ARRAY_CONFIG);
7090 
7091 	mask = amdgpu_gfx_create_bitmask(adev->gfx.config.max_cu_per_sh);
7092 
7093 	return ~REG_GET_FIELD(data, CC_GC_SHADER_ARRAY_CONFIG, INACTIVE_CUS) & mask;
7094 }
7095 
7096 static void gfx_v8_0_get_cu_info(struct amdgpu_device *adev)
7097 {
7098 	int i, j, k, counter, active_cu_number = 0;
7099 	u32 mask, bitmap, ao_bitmap, ao_cu_mask = 0;
7100 	struct amdgpu_cu_info *cu_info = &adev->gfx.cu_info;
7101 	unsigned disable_masks[4 * 2];
7102 	u32 ao_cu_num;
7103 
7104 	memset(cu_info, 0, sizeof(*cu_info));
7105 
7106 	if (adev->flags & AMD_IS_APU)
7107 		ao_cu_num = 2;
7108 	else
7109 		ao_cu_num = adev->gfx.config.max_cu_per_sh;
7110 
7111 	amdgpu_gfx_parse_disable_cu(disable_masks, 4, 2);
7112 
7113 	mutex_lock(&adev->grbm_idx_mutex);
7114 	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
7115 		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
7116 			mask = 1;
7117 			ao_bitmap = 0;
7118 			counter = 0;
7119 			gfx_v8_0_select_se_sh(adev, i, j, 0xffffffff);
7120 			if (i < 4 && j < 2)
7121 				gfx_v8_0_set_user_cu_inactive_bitmap(
7122 					adev, disable_masks[i * 2 + j]);
7123 			bitmap = gfx_v8_0_get_cu_active_bitmap(adev);
7124 			cu_info->bitmap[i][j] = bitmap;
7125 
7126 			for (k = 0; k < adev->gfx.config.max_cu_per_sh; k ++) {
7127 				if (bitmap & mask) {
7128 					if (counter < ao_cu_num)
7129 						ao_bitmap |= mask;
7130 					counter ++;
7131 				}
7132 				mask <<= 1;
7133 			}
7134 			active_cu_number += counter;
7135 			if (i < 2 && j < 2)
7136 				ao_cu_mask |= (ao_bitmap << (i * 16 + j * 8));
7137 			cu_info->ao_cu_bitmap[i][j] = ao_bitmap;
7138 		}
7139 	}
7140 	gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
7141 	mutex_unlock(&adev->grbm_idx_mutex);
7142 
7143 	cu_info->number = active_cu_number;
7144 	cu_info->ao_cu_mask = ao_cu_mask;
7145 	cu_info->simd_per_cu = NUM_SIMD_PER_CU;
7146 	cu_info->max_waves_per_simd = 10;
7147 	cu_info->max_scratch_slots_per_cu = 32;
7148 	cu_info->wave_front_size = 64;
7149 	cu_info->lds_size = 64;
7150 }
7151 
7152 const struct amdgpu_ip_block_version gfx_v8_0_ip_block =
7153 {
7154 	.type = AMD_IP_BLOCK_TYPE_GFX,
7155 	.major = 8,
7156 	.minor = 0,
7157 	.rev = 0,
7158 	.funcs = &gfx_v8_0_ip_funcs,
7159 };
7160 
7161 const struct amdgpu_ip_block_version gfx_v8_1_ip_block =
7162 {
7163 	.type = AMD_IP_BLOCK_TYPE_GFX,
7164 	.major = 8,
7165 	.minor = 1,
7166 	.rev = 0,
7167 	.funcs = &gfx_v8_0_ip_funcs,
7168 };
7169 
7170 static void gfx_v8_0_ring_emit_ce_meta(struct amdgpu_ring *ring)
7171 {
7172 	uint64_t ce_payload_addr;
7173 	int cnt_ce;
7174 	union {
7175 		struct vi_ce_ib_state regular;
7176 		struct vi_ce_ib_state_chained_ib chained;
7177 	} ce_payload = {};
7178 
7179 	if (ring->adev->virt.chained_ib_support) {
7180 		ce_payload_addr = amdgpu_csa_vaddr(ring->adev) +
7181 			offsetof(struct vi_gfx_meta_data_chained_ib, ce_payload);
7182 		cnt_ce = (sizeof(ce_payload.chained) >> 2) + 4 - 2;
7183 	} else {
7184 		ce_payload_addr = amdgpu_csa_vaddr(ring->adev) +
7185 			offsetof(struct vi_gfx_meta_data, ce_payload);
7186 		cnt_ce = (sizeof(ce_payload.regular) >> 2) + 4 - 2;
7187 	}
7188 
7189 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, cnt_ce));
7190 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(2) |
7191 				WRITE_DATA_DST_SEL(8) |
7192 				WR_CONFIRM) |
7193 				WRITE_DATA_CACHE_POLICY(0));
7194 	amdgpu_ring_write(ring, lower_32_bits(ce_payload_addr));
7195 	amdgpu_ring_write(ring, upper_32_bits(ce_payload_addr));
7196 	amdgpu_ring_write_multiple(ring, (void *)&ce_payload, cnt_ce - 2);
7197 }
7198 
7199 static void gfx_v8_0_ring_emit_de_meta(struct amdgpu_ring *ring)
7200 {
7201 	uint64_t de_payload_addr, gds_addr, csa_addr;
7202 	int cnt_de;
7203 	union {
7204 		struct vi_de_ib_state regular;
7205 		struct vi_de_ib_state_chained_ib chained;
7206 	} de_payload = {};
7207 
7208 	csa_addr = amdgpu_csa_vaddr(ring->adev);
7209 	gds_addr = csa_addr + 4096;
7210 	if (ring->adev->virt.chained_ib_support) {
7211 		de_payload.chained.gds_backup_addrlo = lower_32_bits(gds_addr);
7212 		de_payload.chained.gds_backup_addrhi = upper_32_bits(gds_addr);
7213 		de_payload_addr = csa_addr + offsetof(struct vi_gfx_meta_data_chained_ib, de_payload);
7214 		cnt_de = (sizeof(de_payload.chained) >> 2) + 4 - 2;
7215 	} else {
7216 		de_payload.regular.gds_backup_addrlo = lower_32_bits(gds_addr);
7217 		de_payload.regular.gds_backup_addrhi = upper_32_bits(gds_addr);
7218 		de_payload_addr = csa_addr + offsetof(struct vi_gfx_meta_data, de_payload);
7219 		cnt_de = (sizeof(de_payload.regular) >> 2) + 4 - 2;
7220 	}
7221 
7222 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, cnt_de));
7223 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(1) |
7224 				WRITE_DATA_DST_SEL(8) |
7225 				WR_CONFIRM) |
7226 				WRITE_DATA_CACHE_POLICY(0));
7227 	amdgpu_ring_write(ring, lower_32_bits(de_payload_addr));
7228 	amdgpu_ring_write(ring, upper_32_bits(de_payload_addr));
7229 	amdgpu_ring_write_multiple(ring, (void *)&de_payload, cnt_de - 2);
7230 }
7231