xref: /openbmc/linux/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c (revision ecfb9f40)
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 	release_firmware(adev->gfx.pfp_fw);
928 	adev->gfx.pfp_fw = NULL;
929 	release_firmware(adev->gfx.me_fw);
930 	adev->gfx.me_fw = NULL;
931 	release_firmware(adev->gfx.ce_fw);
932 	adev->gfx.ce_fw = NULL;
933 	release_firmware(adev->gfx.rlc_fw);
934 	adev->gfx.rlc_fw = NULL;
935 	release_firmware(adev->gfx.mec_fw);
936 	adev->gfx.mec_fw = NULL;
937 	if ((adev->asic_type != CHIP_STONEY) &&
938 	    (adev->asic_type != CHIP_TOPAZ))
939 		release_firmware(adev->gfx.mec2_fw);
940 	adev->gfx.mec2_fw = NULL;
941 
942 	kfree(adev->gfx.rlc.register_list_format);
943 }
944 
945 static int gfx_v8_0_init_microcode(struct amdgpu_device *adev)
946 {
947 	const char *chip_name;
948 	char fw_name[30];
949 	int err;
950 	struct amdgpu_firmware_info *info = NULL;
951 	const struct common_firmware_header *header = NULL;
952 	const struct gfx_firmware_header_v1_0 *cp_hdr;
953 	const struct rlc_firmware_header_v2_0 *rlc_hdr;
954 	unsigned int *tmp = NULL, i;
955 
956 	DRM_DEBUG("\n");
957 
958 	switch (adev->asic_type) {
959 	case CHIP_TOPAZ:
960 		chip_name = "topaz";
961 		break;
962 	case CHIP_TONGA:
963 		chip_name = "tonga";
964 		break;
965 	case CHIP_CARRIZO:
966 		chip_name = "carrizo";
967 		break;
968 	case CHIP_FIJI:
969 		chip_name = "fiji";
970 		break;
971 	case CHIP_STONEY:
972 		chip_name = "stoney";
973 		break;
974 	case CHIP_POLARIS10:
975 		chip_name = "polaris10";
976 		break;
977 	case CHIP_POLARIS11:
978 		chip_name = "polaris11";
979 		break;
980 	case CHIP_POLARIS12:
981 		chip_name = "polaris12";
982 		break;
983 	case CHIP_VEGAM:
984 		chip_name = "vegam";
985 		break;
986 	default:
987 		BUG();
988 	}
989 
990 	if (adev->asic_type >= CHIP_POLARIS10 && adev->asic_type <= CHIP_POLARIS12) {
991 		snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_pfp_2.bin", chip_name);
992 		err = request_firmware(&adev->gfx.pfp_fw, fw_name, adev->dev);
993 		if (err == -ENOENT) {
994 			snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_pfp.bin", chip_name);
995 			err = request_firmware(&adev->gfx.pfp_fw, fw_name, adev->dev);
996 		}
997 	} else {
998 		snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_pfp.bin", chip_name);
999 		err = request_firmware(&adev->gfx.pfp_fw, fw_name, adev->dev);
1000 	}
1001 	if (err)
1002 		goto out;
1003 	err = amdgpu_ucode_validate(adev->gfx.pfp_fw);
1004 	if (err)
1005 		goto out;
1006 	cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.pfp_fw->data;
1007 	adev->gfx.pfp_fw_version = le32_to_cpu(cp_hdr->header.ucode_version);
1008 	adev->gfx.pfp_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version);
1009 
1010 	if (adev->asic_type >= CHIP_POLARIS10 && adev->asic_type <= CHIP_POLARIS12) {
1011 		snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_me_2.bin", chip_name);
1012 		err = request_firmware(&adev->gfx.me_fw, fw_name, adev->dev);
1013 		if (err == -ENOENT) {
1014 			snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_me.bin", chip_name);
1015 			err = request_firmware(&adev->gfx.me_fw, fw_name, adev->dev);
1016 		}
1017 	} else {
1018 		snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_me.bin", chip_name);
1019 		err = request_firmware(&adev->gfx.me_fw, fw_name, adev->dev);
1020 	}
1021 	if (err)
1022 		goto out;
1023 	err = amdgpu_ucode_validate(adev->gfx.me_fw);
1024 	if (err)
1025 		goto out;
1026 	cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.me_fw->data;
1027 	adev->gfx.me_fw_version = le32_to_cpu(cp_hdr->header.ucode_version);
1028 
1029 	adev->gfx.me_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version);
1030 
1031 	if (adev->asic_type >= CHIP_POLARIS10 && adev->asic_type <= CHIP_POLARIS12) {
1032 		snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_ce_2.bin", chip_name);
1033 		err = request_firmware(&adev->gfx.ce_fw, fw_name, adev->dev);
1034 		if (err == -ENOENT) {
1035 			snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_ce.bin", chip_name);
1036 			err = request_firmware(&adev->gfx.ce_fw, fw_name, adev->dev);
1037 		}
1038 	} else {
1039 		snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_ce.bin", chip_name);
1040 		err = request_firmware(&adev->gfx.ce_fw, fw_name, adev->dev);
1041 	}
1042 	if (err)
1043 		goto out;
1044 	err = amdgpu_ucode_validate(adev->gfx.ce_fw);
1045 	if (err)
1046 		goto out;
1047 	cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.ce_fw->data;
1048 	adev->gfx.ce_fw_version = le32_to_cpu(cp_hdr->header.ucode_version);
1049 	adev->gfx.ce_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version);
1050 
1051 	/*
1052 	 * Support for MCBP/Virtualization in combination with chained IBs is
1053 	 * formal released on feature version #46
1054 	 */
1055 	if (adev->gfx.ce_feature_version >= 46 &&
1056 	    adev->gfx.pfp_feature_version >= 46) {
1057 		adev->virt.chained_ib_support = true;
1058 		DRM_INFO("Chained IB support enabled!\n");
1059 	} else
1060 		adev->virt.chained_ib_support = false;
1061 
1062 	snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_rlc.bin", chip_name);
1063 	err = request_firmware(&adev->gfx.rlc_fw, fw_name, adev->dev);
1064 	if (err)
1065 		goto out;
1066 	err = amdgpu_ucode_validate(adev->gfx.rlc_fw);
1067 	rlc_hdr = (const struct rlc_firmware_header_v2_0 *)adev->gfx.rlc_fw->data;
1068 	adev->gfx.rlc_fw_version = le32_to_cpu(rlc_hdr->header.ucode_version);
1069 	adev->gfx.rlc_feature_version = le32_to_cpu(rlc_hdr->ucode_feature_version);
1070 
1071 	adev->gfx.rlc.save_and_restore_offset =
1072 			le32_to_cpu(rlc_hdr->save_and_restore_offset);
1073 	adev->gfx.rlc.clear_state_descriptor_offset =
1074 			le32_to_cpu(rlc_hdr->clear_state_descriptor_offset);
1075 	adev->gfx.rlc.avail_scratch_ram_locations =
1076 			le32_to_cpu(rlc_hdr->avail_scratch_ram_locations);
1077 	adev->gfx.rlc.reg_restore_list_size =
1078 			le32_to_cpu(rlc_hdr->reg_restore_list_size);
1079 	adev->gfx.rlc.reg_list_format_start =
1080 			le32_to_cpu(rlc_hdr->reg_list_format_start);
1081 	adev->gfx.rlc.reg_list_format_separate_start =
1082 			le32_to_cpu(rlc_hdr->reg_list_format_separate_start);
1083 	adev->gfx.rlc.starting_offsets_start =
1084 			le32_to_cpu(rlc_hdr->starting_offsets_start);
1085 	adev->gfx.rlc.reg_list_format_size_bytes =
1086 			le32_to_cpu(rlc_hdr->reg_list_format_size_bytes);
1087 	adev->gfx.rlc.reg_list_size_bytes =
1088 			le32_to_cpu(rlc_hdr->reg_list_size_bytes);
1089 
1090 	adev->gfx.rlc.register_list_format =
1091 			kmalloc(adev->gfx.rlc.reg_list_format_size_bytes +
1092 					adev->gfx.rlc.reg_list_size_bytes, GFP_KERNEL);
1093 
1094 	if (!adev->gfx.rlc.register_list_format) {
1095 		err = -ENOMEM;
1096 		goto out;
1097 	}
1098 
1099 	tmp = (unsigned int *)((uintptr_t)rlc_hdr +
1100 			le32_to_cpu(rlc_hdr->reg_list_format_array_offset_bytes));
1101 	for (i = 0 ; i < (adev->gfx.rlc.reg_list_format_size_bytes >> 2); i++)
1102 		adev->gfx.rlc.register_list_format[i] =	le32_to_cpu(tmp[i]);
1103 
1104 	adev->gfx.rlc.register_restore = adev->gfx.rlc.register_list_format + i;
1105 
1106 	tmp = (unsigned int *)((uintptr_t)rlc_hdr +
1107 			le32_to_cpu(rlc_hdr->reg_list_array_offset_bytes));
1108 	for (i = 0 ; i < (adev->gfx.rlc.reg_list_size_bytes >> 2); i++)
1109 		adev->gfx.rlc.register_restore[i] = le32_to_cpu(tmp[i]);
1110 
1111 	if (adev->asic_type >= CHIP_POLARIS10 && adev->asic_type <= CHIP_POLARIS12) {
1112 		snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec_2.bin", chip_name);
1113 		err = request_firmware(&adev->gfx.mec_fw, fw_name, adev->dev);
1114 		if (err == -ENOENT) {
1115 			snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec.bin", chip_name);
1116 			err = request_firmware(&adev->gfx.mec_fw, fw_name, adev->dev);
1117 		}
1118 	} else {
1119 		snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec.bin", chip_name);
1120 		err = request_firmware(&adev->gfx.mec_fw, fw_name, adev->dev);
1121 	}
1122 	if (err)
1123 		goto out;
1124 	err = amdgpu_ucode_validate(adev->gfx.mec_fw);
1125 	if (err)
1126 		goto out;
1127 	cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data;
1128 	adev->gfx.mec_fw_version = le32_to_cpu(cp_hdr->header.ucode_version);
1129 	adev->gfx.mec_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version);
1130 
1131 	if ((adev->asic_type != CHIP_STONEY) &&
1132 	    (adev->asic_type != CHIP_TOPAZ)) {
1133 		if (adev->asic_type >= CHIP_POLARIS10 && adev->asic_type <= CHIP_POLARIS12) {
1134 			snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec2_2.bin", chip_name);
1135 			err = request_firmware(&adev->gfx.mec2_fw, fw_name, adev->dev);
1136 			if (err == -ENOENT) {
1137 				snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec2.bin", chip_name);
1138 				err = request_firmware(&adev->gfx.mec2_fw, fw_name, adev->dev);
1139 			}
1140 		} else {
1141 			snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec2.bin", chip_name);
1142 			err = request_firmware(&adev->gfx.mec2_fw, fw_name, adev->dev);
1143 		}
1144 		if (!err) {
1145 			err = amdgpu_ucode_validate(adev->gfx.mec2_fw);
1146 			if (err)
1147 				goto out;
1148 			cp_hdr = (const struct gfx_firmware_header_v1_0 *)
1149 				adev->gfx.mec2_fw->data;
1150 			adev->gfx.mec2_fw_version =
1151 				le32_to_cpu(cp_hdr->header.ucode_version);
1152 			adev->gfx.mec2_feature_version =
1153 				le32_to_cpu(cp_hdr->ucode_feature_version);
1154 		} else {
1155 			err = 0;
1156 			adev->gfx.mec2_fw = NULL;
1157 		}
1158 	}
1159 
1160 	info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_PFP];
1161 	info->ucode_id = AMDGPU_UCODE_ID_CP_PFP;
1162 	info->fw = adev->gfx.pfp_fw;
1163 	header = (const struct common_firmware_header *)info->fw->data;
1164 	adev->firmware.fw_size +=
1165 		ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
1166 
1167 	info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_ME];
1168 	info->ucode_id = AMDGPU_UCODE_ID_CP_ME;
1169 	info->fw = adev->gfx.me_fw;
1170 	header = (const struct common_firmware_header *)info->fw->data;
1171 	adev->firmware.fw_size +=
1172 		ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
1173 
1174 	info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_CE];
1175 	info->ucode_id = AMDGPU_UCODE_ID_CP_CE;
1176 	info->fw = adev->gfx.ce_fw;
1177 	header = (const struct common_firmware_header *)info->fw->data;
1178 	adev->firmware.fw_size +=
1179 		ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
1180 
1181 	info = &adev->firmware.ucode[AMDGPU_UCODE_ID_RLC_G];
1182 	info->ucode_id = AMDGPU_UCODE_ID_RLC_G;
1183 	info->fw = adev->gfx.rlc_fw;
1184 	header = (const struct common_firmware_header *)info->fw->data;
1185 	adev->firmware.fw_size +=
1186 		ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
1187 
1188 	info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_MEC1];
1189 	info->ucode_id = AMDGPU_UCODE_ID_CP_MEC1;
1190 	info->fw = adev->gfx.mec_fw;
1191 	header = (const struct common_firmware_header *)info->fw->data;
1192 	adev->firmware.fw_size +=
1193 		ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
1194 
1195 	/* we need account JT in */
1196 	cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data;
1197 	adev->firmware.fw_size +=
1198 		ALIGN(le32_to_cpu(cp_hdr->jt_size) << 2, PAGE_SIZE);
1199 
1200 	if (amdgpu_sriov_vf(adev)) {
1201 		info = &adev->firmware.ucode[AMDGPU_UCODE_ID_STORAGE];
1202 		info->ucode_id = AMDGPU_UCODE_ID_STORAGE;
1203 		info->fw = adev->gfx.mec_fw;
1204 		adev->firmware.fw_size +=
1205 			ALIGN(le32_to_cpu(64 * PAGE_SIZE), PAGE_SIZE);
1206 	}
1207 
1208 	if (adev->gfx.mec2_fw) {
1209 		info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_MEC2];
1210 		info->ucode_id = AMDGPU_UCODE_ID_CP_MEC2;
1211 		info->fw = adev->gfx.mec2_fw;
1212 		header = (const struct common_firmware_header *)info->fw->data;
1213 		adev->firmware.fw_size +=
1214 			ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
1215 	}
1216 
1217 out:
1218 	if (err) {
1219 		dev_err(adev->dev,
1220 			"gfx8: Failed to load firmware \"%s\"\n",
1221 			fw_name);
1222 		release_firmware(adev->gfx.pfp_fw);
1223 		adev->gfx.pfp_fw = NULL;
1224 		release_firmware(adev->gfx.me_fw);
1225 		adev->gfx.me_fw = NULL;
1226 		release_firmware(adev->gfx.ce_fw);
1227 		adev->gfx.ce_fw = NULL;
1228 		release_firmware(adev->gfx.rlc_fw);
1229 		adev->gfx.rlc_fw = NULL;
1230 		release_firmware(adev->gfx.mec_fw);
1231 		adev->gfx.mec_fw = NULL;
1232 		release_firmware(adev->gfx.mec2_fw);
1233 		adev->gfx.mec2_fw = NULL;
1234 	}
1235 	return err;
1236 }
1237 
1238 static void gfx_v8_0_get_csb_buffer(struct amdgpu_device *adev,
1239 				    volatile u32 *buffer)
1240 {
1241 	u32 count = 0, i;
1242 	const struct cs_section_def *sect = NULL;
1243 	const struct cs_extent_def *ext = NULL;
1244 
1245 	if (adev->gfx.rlc.cs_data == NULL)
1246 		return;
1247 	if (buffer == NULL)
1248 		return;
1249 
1250 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0));
1251 	buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
1252 
1253 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CONTEXT_CONTROL, 1));
1254 	buffer[count++] = cpu_to_le32(0x80000000);
1255 	buffer[count++] = cpu_to_le32(0x80000000);
1256 
1257 	for (sect = adev->gfx.rlc.cs_data; sect->section != NULL; ++sect) {
1258 		for (ext = sect->section; ext->extent != NULL; ++ext) {
1259 			if (sect->id == SECT_CONTEXT) {
1260 				buffer[count++] =
1261 					cpu_to_le32(PACKET3(PACKET3_SET_CONTEXT_REG, ext->reg_count));
1262 				buffer[count++] = cpu_to_le32(ext->reg_index -
1263 						PACKET3_SET_CONTEXT_REG_START);
1264 				for (i = 0; i < ext->reg_count; i++)
1265 					buffer[count++] = cpu_to_le32(ext->extent[i]);
1266 			} else {
1267 				return;
1268 			}
1269 		}
1270 	}
1271 
1272 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_SET_CONTEXT_REG, 2));
1273 	buffer[count++] = cpu_to_le32(mmPA_SC_RASTER_CONFIG -
1274 			PACKET3_SET_CONTEXT_REG_START);
1275 	buffer[count++] = cpu_to_le32(adev->gfx.config.rb_config[0][0].raster_config);
1276 	buffer[count++] = cpu_to_le32(adev->gfx.config.rb_config[0][0].raster_config_1);
1277 
1278 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0));
1279 	buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_END_CLEAR_STATE);
1280 
1281 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CLEAR_STATE, 0));
1282 	buffer[count++] = cpu_to_le32(0);
1283 }
1284 
1285 static int gfx_v8_0_cp_jump_table_num(struct amdgpu_device *adev)
1286 {
1287 	if (adev->asic_type == CHIP_CARRIZO)
1288 		return 5;
1289 	else
1290 		return 4;
1291 }
1292 
1293 static int gfx_v8_0_rlc_init(struct amdgpu_device *adev)
1294 {
1295 	const struct cs_section_def *cs_data;
1296 	int r;
1297 
1298 	adev->gfx.rlc.cs_data = vi_cs_data;
1299 
1300 	cs_data = adev->gfx.rlc.cs_data;
1301 
1302 	if (cs_data) {
1303 		/* init clear state block */
1304 		r = amdgpu_gfx_rlc_init_csb(adev);
1305 		if (r)
1306 			return r;
1307 	}
1308 
1309 	if ((adev->asic_type == CHIP_CARRIZO) ||
1310 	    (adev->asic_type == CHIP_STONEY)) {
1311 		adev->gfx.rlc.cp_table_size = ALIGN(96 * 5 * 4, 2048) + (64 * 1024); /* JT + GDS */
1312 		r = amdgpu_gfx_rlc_init_cpt(adev);
1313 		if (r)
1314 			return r;
1315 	}
1316 
1317 	/* init spm vmid with 0xf */
1318 	if (adev->gfx.rlc.funcs->update_spm_vmid)
1319 		adev->gfx.rlc.funcs->update_spm_vmid(adev, 0xf);
1320 
1321 	return 0;
1322 }
1323 
1324 static void gfx_v8_0_mec_fini(struct amdgpu_device *adev)
1325 {
1326 	amdgpu_bo_free_kernel(&adev->gfx.mec.hpd_eop_obj, NULL, NULL);
1327 }
1328 
1329 static int gfx_v8_0_mec_init(struct amdgpu_device *adev)
1330 {
1331 	int r;
1332 	u32 *hpd;
1333 	size_t mec_hpd_size;
1334 
1335 	bitmap_zero(adev->gfx.mec.queue_bitmap, AMDGPU_MAX_COMPUTE_QUEUES);
1336 
1337 	/* take ownership of the relevant compute queues */
1338 	amdgpu_gfx_compute_queue_acquire(adev);
1339 
1340 	mec_hpd_size = adev->gfx.num_compute_rings * GFX8_MEC_HPD_SIZE;
1341 	if (mec_hpd_size) {
1342 		r = amdgpu_bo_create_reserved(adev, mec_hpd_size, PAGE_SIZE,
1343 					      AMDGPU_GEM_DOMAIN_VRAM,
1344 					      &adev->gfx.mec.hpd_eop_obj,
1345 					      &adev->gfx.mec.hpd_eop_gpu_addr,
1346 					      (void **)&hpd);
1347 		if (r) {
1348 			dev_warn(adev->dev, "(%d) create HDP EOP bo failed\n", r);
1349 			return r;
1350 		}
1351 
1352 		memset(hpd, 0, mec_hpd_size);
1353 
1354 		amdgpu_bo_kunmap(adev->gfx.mec.hpd_eop_obj);
1355 		amdgpu_bo_unreserve(adev->gfx.mec.hpd_eop_obj);
1356 	}
1357 
1358 	return 0;
1359 }
1360 
1361 static const u32 vgpr_init_compute_shader[] =
1362 {
1363 	0x7e000209, 0x7e020208,
1364 	0x7e040207, 0x7e060206,
1365 	0x7e080205, 0x7e0a0204,
1366 	0x7e0c0203, 0x7e0e0202,
1367 	0x7e100201, 0x7e120200,
1368 	0x7e140209, 0x7e160208,
1369 	0x7e180207, 0x7e1a0206,
1370 	0x7e1c0205, 0x7e1e0204,
1371 	0x7e200203, 0x7e220202,
1372 	0x7e240201, 0x7e260200,
1373 	0x7e280209, 0x7e2a0208,
1374 	0x7e2c0207, 0x7e2e0206,
1375 	0x7e300205, 0x7e320204,
1376 	0x7e340203, 0x7e360202,
1377 	0x7e380201, 0x7e3a0200,
1378 	0x7e3c0209, 0x7e3e0208,
1379 	0x7e400207, 0x7e420206,
1380 	0x7e440205, 0x7e460204,
1381 	0x7e480203, 0x7e4a0202,
1382 	0x7e4c0201, 0x7e4e0200,
1383 	0x7e500209, 0x7e520208,
1384 	0x7e540207, 0x7e560206,
1385 	0x7e580205, 0x7e5a0204,
1386 	0x7e5c0203, 0x7e5e0202,
1387 	0x7e600201, 0x7e620200,
1388 	0x7e640209, 0x7e660208,
1389 	0x7e680207, 0x7e6a0206,
1390 	0x7e6c0205, 0x7e6e0204,
1391 	0x7e700203, 0x7e720202,
1392 	0x7e740201, 0x7e760200,
1393 	0x7e780209, 0x7e7a0208,
1394 	0x7e7c0207, 0x7e7e0206,
1395 	0xbf8a0000, 0xbf810000,
1396 };
1397 
1398 static const u32 sgpr_init_compute_shader[] =
1399 {
1400 	0xbe8a0100, 0xbe8c0102,
1401 	0xbe8e0104, 0xbe900106,
1402 	0xbe920108, 0xbe940100,
1403 	0xbe960102, 0xbe980104,
1404 	0xbe9a0106, 0xbe9c0108,
1405 	0xbe9e0100, 0xbea00102,
1406 	0xbea20104, 0xbea40106,
1407 	0xbea60108, 0xbea80100,
1408 	0xbeaa0102, 0xbeac0104,
1409 	0xbeae0106, 0xbeb00108,
1410 	0xbeb20100, 0xbeb40102,
1411 	0xbeb60104, 0xbeb80106,
1412 	0xbeba0108, 0xbebc0100,
1413 	0xbebe0102, 0xbec00104,
1414 	0xbec20106, 0xbec40108,
1415 	0xbec60100, 0xbec80102,
1416 	0xbee60004, 0xbee70005,
1417 	0xbeea0006, 0xbeeb0007,
1418 	0xbee80008, 0xbee90009,
1419 	0xbefc0000, 0xbf8a0000,
1420 	0xbf810000, 0x00000000,
1421 };
1422 
1423 static const u32 vgpr_init_regs[] =
1424 {
1425 	mmCOMPUTE_STATIC_THREAD_MGMT_SE0, 0xffffffff,
1426 	mmCOMPUTE_RESOURCE_LIMITS, 0x1000000, /* CU_GROUP_COUNT=1 */
1427 	mmCOMPUTE_NUM_THREAD_X, 256*4,
1428 	mmCOMPUTE_NUM_THREAD_Y, 1,
1429 	mmCOMPUTE_NUM_THREAD_Z, 1,
1430 	mmCOMPUTE_PGM_RSRC1, 0x100004f, /* VGPRS=15 (64 logical VGPRs), SGPRS=1 (16 SGPRs), BULKY=1 */
1431 	mmCOMPUTE_PGM_RSRC2, 20,
1432 	mmCOMPUTE_USER_DATA_0, 0xedcedc00,
1433 	mmCOMPUTE_USER_DATA_1, 0xedcedc01,
1434 	mmCOMPUTE_USER_DATA_2, 0xedcedc02,
1435 	mmCOMPUTE_USER_DATA_3, 0xedcedc03,
1436 	mmCOMPUTE_USER_DATA_4, 0xedcedc04,
1437 	mmCOMPUTE_USER_DATA_5, 0xedcedc05,
1438 	mmCOMPUTE_USER_DATA_6, 0xedcedc06,
1439 	mmCOMPUTE_USER_DATA_7, 0xedcedc07,
1440 	mmCOMPUTE_USER_DATA_8, 0xedcedc08,
1441 	mmCOMPUTE_USER_DATA_9, 0xedcedc09,
1442 };
1443 
1444 static const u32 sgpr1_init_regs[] =
1445 {
1446 	mmCOMPUTE_STATIC_THREAD_MGMT_SE0, 0x0f,
1447 	mmCOMPUTE_RESOURCE_LIMITS, 0x1000000, /* CU_GROUP_COUNT=1 */
1448 	mmCOMPUTE_NUM_THREAD_X, 256*5,
1449 	mmCOMPUTE_NUM_THREAD_Y, 1,
1450 	mmCOMPUTE_NUM_THREAD_Z, 1,
1451 	mmCOMPUTE_PGM_RSRC1, 0x240, /* SGPRS=9 (80 GPRS) */
1452 	mmCOMPUTE_PGM_RSRC2, 20,
1453 	mmCOMPUTE_USER_DATA_0, 0xedcedc00,
1454 	mmCOMPUTE_USER_DATA_1, 0xedcedc01,
1455 	mmCOMPUTE_USER_DATA_2, 0xedcedc02,
1456 	mmCOMPUTE_USER_DATA_3, 0xedcedc03,
1457 	mmCOMPUTE_USER_DATA_4, 0xedcedc04,
1458 	mmCOMPUTE_USER_DATA_5, 0xedcedc05,
1459 	mmCOMPUTE_USER_DATA_6, 0xedcedc06,
1460 	mmCOMPUTE_USER_DATA_7, 0xedcedc07,
1461 	mmCOMPUTE_USER_DATA_8, 0xedcedc08,
1462 	mmCOMPUTE_USER_DATA_9, 0xedcedc09,
1463 };
1464 
1465 static const u32 sgpr2_init_regs[] =
1466 {
1467 	mmCOMPUTE_STATIC_THREAD_MGMT_SE0, 0xf0,
1468 	mmCOMPUTE_RESOURCE_LIMITS, 0x1000000,
1469 	mmCOMPUTE_NUM_THREAD_X, 256*5,
1470 	mmCOMPUTE_NUM_THREAD_Y, 1,
1471 	mmCOMPUTE_NUM_THREAD_Z, 1,
1472 	mmCOMPUTE_PGM_RSRC1, 0x240, /* SGPRS=9 (80 GPRS) */
1473 	mmCOMPUTE_PGM_RSRC2, 20,
1474 	mmCOMPUTE_USER_DATA_0, 0xedcedc00,
1475 	mmCOMPUTE_USER_DATA_1, 0xedcedc01,
1476 	mmCOMPUTE_USER_DATA_2, 0xedcedc02,
1477 	mmCOMPUTE_USER_DATA_3, 0xedcedc03,
1478 	mmCOMPUTE_USER_DATA_4, 0xedcedc04,
1479 	mmCOMPUTE_USER_DATA_5, 0xedcedc05,
1480 	mmCOMPUTE_USER_DATA_6, 0xedcedc06,
1481 	mmCOMPUTE_USER_DATA_7, 0xedcedc07,
1482 	mmCOMPUTE_USER_DATA_8, 0xedcedc08,
1483 	mmCOMPUTE_USER_DATA_9, 0xedcedc09,
1484 };
1485 
1486 static const u32 sec_ded_counter_registers[] =
1487 {
1488 	mmCPC_EDC_ATC_CNT,
1489 	mmCPC_EDC_SCRATCH_CNT,
1490 	mmCPC_EDC_UCODE_CNT,
1491 	mmCPF_EDC_ATC_CNT,
1492 	mmCPF_EDC_ROQ_CNT,
1493 	mmCPF_EDC_TAG_CNT,
1494 	mmCPG_EDC_ATC_CNT,
1495 	mmCPG_EDC_DMA_CNT,
1496 	mmCPG_EDC_TAG_CNT,
1497 	mmDC_EDC_CSINVOC_CNT,
1498 	mmDC_EDC_RESTORE_CNT,
1499 	mmDC_EDC_STATE_CNT,
1500 	mmGDS_EDC_CNT,
1501 	mmGDS_EDC_GRBM_CNT,
1502 	mmGDS_EDC_OA_DED,
1503 	mmSPI_EDC_CNT,
1504 	mmSQC_ATC_EDC_GATCL1_CNT,
1505 	mmSQC_EDC_CNT,
1506 	mmSQ_EDC_DED_CNT,
1507 	mmSQ_EDC_INFO,
1508 	mmSQ_EDC_SEC_CNT,
1509 	mmTCC_EDC_CNT,
1510 	mmTCP_ATC_EDC_GATCL1_CNT,
1511 	mmTCP_EDC_CNT,
1512 	mmTD_EDC_CNT
1513 };
1514 
1515 static int gfx_v8_0_do_edc_gpr_workarounds(struct amdgpu_device *adev)
1516 {
1517 	struct amdgpu_ring *ring = &adev->gfx.compute_ring[0];
1518 	struct amdgpu_ib ib;
1519 	struct dma_fence *f = NULL;
1520 	int r, i;
1521 	u32 tmp;
1522 	unsigned total_size, vgpr_offset, sgpr_offset;
1523 	u64 gpu_addr;
1524 
1525 	/* only supported on CZ */
1526 	if (adev->asic_type != CHIP_CARRIZO)
1527 		return 0;
1528 
1529 	/* bail if the compute ring is not ready */
1530 	if (!ring->sched.ready)
1531 		return 0;
1532 
1533 	tmp = RREG32(mmGB_EDC_MODE);
1534 	WREG32(mmGB_EDC_MODE, 0);
1535 
1536 	total_size =
1537 		(((ARRAY_SIZE(vgpr_init_regs) / 2) * 3) + 4 + 5 + 2) * 4;
1538 	total_size +=
1539 		(((ARRAY_SIZE(sgpr1_init_regs) / 2) * 3) + 4 + 5 + 2) * 4;
1540 	total_size +=
1541 		(((ARRAY_SIZE(sgpr2_init_regs) / 2) * 3) + 4 + 5 + 2) * 4;
1542 	total_size = ALIGN(total_size, 256);
1543 	vgpr_offset = total_size;
1544 	total_size += ALIGN(sizeof(vgpr_init_compute_shader), 256);
1545 	sgpr_offset = total_size;
1546 	total_size += sizeof(sgpr_init_compute_shader);
1547 
1548 	/* allocate an indirect buffer to put the commands in */
1549 	memset(&ib, 0, sizeof(ib));
1550 	r = amdgpu_ib_get(adev, NULL, total_size,
1551 					AMDGPU_IB_POOL_DIRECT, &ib);
1552 	if (r) {
1553 		DRM_ERROR("amdgpu: failed to get ib (%d).\n", r);
1554 		return r;
1555 	}
1556 
1557 	/* load the compute shaders */
1558 	for (i = 0; i < ARRAY_SIZE(vgpr_init_compute_shader); i++)
1559 		ib.ptr[i + (vgpr_offset / 4)] = vgpr_init_compute_shader[i];
1560 
1561 	for (i = 0; i < ARRAY_SIZE(sgpr_init_compute_shader); i++)
1562 		ib.ptr[i + (sgpr_offset / 4)] = sgpr_init_compute_shader[i];
1563 
1564 	/* init the ib length to 0 */
1565 	ib.length_dw = 0;
1566 
1567 	/* VGPR */
1568 	/* write the register state for the compute dispatch */
1569 	for (i = 0; i < ARRAY_SIZE(vgpr_init_regs); i += 2) {
1570 		ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 1);
1571 		ib.ptr[ib.length_dw++] = vgpr_init_regs[i] - PACKET3_SET_SH_REG_START;
1572 		ib.ptr[ib.length_dw++] = vgpr_init_regs[i + 1];
1573 	}
1574 	/* write the shader start address: mmCOMPUTE_PGM_LO, mmCOMPUTE_PGM_HI */
1575 	gpu_addr = (ib.gpu_addr + (u64)vgpr_offset) >> 8;
1576 	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 2);
1577 	ib.ptr[ib.length_dw++] = mmCOMPUTE_PGM_LO - PACKET3_SET_SH_REG_START;
1578 	ib.ptr[ib.length_dw++] = lower_32_bits(gpu_addr);
1579 	ib.ptr[ib.length_dw++] = upper_32_bits(gpu_addr);
1580 
1581 	/* write dispatch packet */
1582 	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_DISPATCH_DIRECT, 3);
1583 	ib.ptr[ib.length_dw++] = 8; /* x */
1584 	ib.ptr[ib.length_dw++] = 1; /* y */
1585 	ib.ptr[ib.length_dw++] = 1; /* z */
1586 	ib.ptr[ib.length_dw++] =
1587 		REG_SET_FIELD(0, COMPUTE_DISPATCH_INITIATOR, COMPUTE_SHADER_EN, 1);
1588 
1589 	/* write CS partial flush packet */
1590 	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_EVENT_WRITE, 0);
1591 	ib.ptr[ib.length_dw++] = EVENT_TYPE(7) | EVENT_INDEX(4);
1592 
1593 	/* SGPR1 */
1594 	/* write the register state for the compute dispatch */
1595 	for (i = 0; i < ARRAY_SIZE(sgpr1_init_regs); i += 2) {
1596 		ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 1);
1597 		ib.ptr[ib.length_dw++] = sgpr1_init_regs[i] - PACKET3_SET_SH_REG_START;
1598 		ib.ptr[ib.length_dw++] = sgpr1_init_regs[i + 1];
1599 	}
1600 	/* write the shader start address: mmCOMPUTE_PGM_LO, mmCOMPUTE_PGM_HI */
1601 	gpu_addr = (ib.gpu_addr + (u64)sgpr_offset) >> 8;
1602 	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 2);
1603 	ib.ptr[ib.length_dw++] = mmCOMPUTE_PGM_LO - PACKET3_SET_SH_REG_START;
1604 	ib.ptr[ib.length_dw++] = lower_32_bits(gpu_addr);
1605 	ib.ptr[ib.length_dw++] = upper_32_bits(gpu_addr);
1606 
1607 	/* write dispatch packet */
1608 	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_DISPATCH_DIRECT, 3);
1609 	ib.ptr[ib.length_dw++] = 8; /* x */
1610 	ib.ptr[ib.length_dw++] = 1; /* y */
1611 	ib.ptr[ib.length_dw++] = 1; /* z */
1612 	ib.ptr[ib.length_dw++] =
1613 		REG_SET_FIELD(0, COMPUTE_DISPATCH_INITIATOR, COMPUTE_SHADER_EN, 1);
1614 
1615 	/* write CS partial flush packet */
1616 	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_EVENT_WRITE, 0);
1617 	ib.ptr[ib.length_dw++] = EVENT_TYPE(7) | EVENT_INDEX(4);
1618 
1619 	/* SGPR2 */
1620 	/* write the register state for the compute dispatch */
1621 	for (i = 0; i < ARRAY_SIZE(sgpr2_init_regs); i += 2) {
1622 		ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 1);
1623 		ib.ptr[ib.length_dw++] = sgpr2_init_regs[i] - PACKET3_SET_SH_REG_START;
1624 		ib.ptr[ib.length_dw++] = sgpr2_init_regs[i + 1];
1625 	}
1626 	/* write the shader start address: mmCOMPUTE_PGM_LO, mmCOMPUTE_PGM_HI */
1627 	gpu_addr = (ib.gpu_addr + (u64)sgpr_offset) >> 8;
1628 	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 2);
1629 	ib.ptr[ib.length_dw++] = mmCOMPUTE_PGM_LO - PACKET3_SET_SH_REG_START;
1630 	ib.ptr[ib.length_dw++] = lower_32_bits(gpu_addr);
1631 	ib.ptr[ib.length_dw++] = upper_32_bits(gpu_addr);
1632 
1633 	/* write dispatch packet */
1634 	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_DISPATCH_DIRECT, 3);
1635 	ib.ptr[ib.length_dw++] = 8; /* x */
1636 	ib.ptr[ib.length_dw++] = 1; /* y */
1637 	ib.ptr[ib.length_dw++] = 1; /* z */
1638 	ib.ptr[ib.length_dw++] =
1639 		REG_SET_FIELD(0, COMPUTE_DISPATCH_INITIATOR, COMPUTE_SHADER_EN, 1);
1640 
1641 	/* write CS partial flush packet */
1642 	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_EVENT_WRITE, 0);
1643 	ib.ptr[ib.length_dw++] = EVENT_TYPE(7) | EVENT_INDEX(4);
1644 
1645 	/* shedule the ib on the ring */
1646 	r = amdgpu_ib_schedule(ring, 1, &ib, NULL, &f);
1647 	if (r) {
1648 		DRM_ERROR("amdgpu: ib submit failed (%d).\n", r);
1649 		goto fail;
1650 	}
1651 
1652 	/* wait for the GPU to finish processing the IB */
1653 	r = dma_fence_wait(f, false);
1654 	if (r) {
1655 		DRM_ERROR("amdgpu: fence wait failed (%d).\n", r);
1656 		goto fail;
1657 	}
1658 
1659 	tmp = REG_SET_FIELD(tmp, GB_EDC_MODE, DED_MODE, 2);
1660 	tmp = REG_SET_FIELD(tmp, GB_EDC_MODE, PROP_FED, 1);
1661 	WREG32(mmGB_EDC_MODE, tmp);
1662 
1663 	tmp = RREG32(mmCC_GC_EDC_CONFIG);
1664 	tmp = REG_SET_FIELD(tmp, CC_GC_EDC_CONFIG, DIS_EDC, 0) | 1;
1665 	WREG32(mmCC_GC_EDC_CONFIG, tmp);
1666 
1667 
1668 	/* read back registers to clear the counters */
1669 	for (i = 0; i < ARRAY_SIZE(sec_ded_counter_registers); i++)
1670 		RREG32(sec_ded_counter_registers[i]);
1671 
1672 fail:
1673 	amdgpu_ib_free(adev, &ib, NULL);
1674 	dma_fence_put(f);
1675 
1676 	return r;
1677 }
1678 
1679 static int gfx_v8_0_gpu_early_init(struct amdgpu_device *adev)
1680 {
1681 	u32 gb_addr_config;
1682 	u32 mc_arb_ramcfg;
1683 	u32 dimm00_addr_map, dimm01_addr_map, dimm10_addr_map, dimm11_addr_map;
1684 	u32 tmp;
1685 	int ret;
1686 
1687 	switch (adev->asic_type) {
1688 	case CHIP_TOPAZ:
1689 		adev->gfx.config.max_shader_engines = 1;
1690 		adev->gfx.config.max_tile_pipes = 2;
1691 		adev->gfx.config.max_cu_per_sh = 6;
1692 		adev->gfx.config.max_sh_per_se = 1;
1693 		adev->gfx.config.max_backends_per_se = 2;
1694 		adev->gfx.config.max_texture_channel_caches = 2;
1695 		adev->gfx.config.max_gprs = 256;
1696 		adev->gfx.config.max_gs_threads = 32;
1697 		adev->gfx.config.max_hw_contexts = 8;
1698 
1699 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1700 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1701 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1702 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
1703 		gb_addr_config = TOPAZ_GB_ADDR_CONFIG_GOLDEN;
1704 		break;
1705 	case CHIP_FIJI:
1706 		adev->gfx.config.max_shader_engines = 4;
1707 		adev->gfx.config.max_tile_pipes = 16;
1708 		adev->gfx.config.max_cu_per_sh = 16;
1709 		adev->gfx.config.max_sh_per_se = 1;
1710 		adev->gfx.config.max_backends_per_se = 4;
1711 		adev->gfx.config.max_texture_channel_caches = 16;
1712 		adev->gfx.config.max_gprs = 256;
1713 		adev->gfx.config.max_gs_threads = 32;
1714 		adev->gfx.config.max_hw_contexts = 8;
1715 
1716 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1717 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1718 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1719 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
1720 		gb_addr_config = TONGA_GB_ADDR_CONFIG_GOLDEN;
1721 		break;
1722 	case CHIP_POLARIS11:
1723 	case CHIP_POLARIS12:
1724 		ret = amdgpu_atombios_get_gfx_info(adev);
1725 		if (ret)
1726 			return ret;
1727 		adev->gfx.config.max_gprs = 256;
1728 		adev->gfx.config.max_gs_threads = 32;
1729 		adev->gfx.config.max_hw_contexts = 8;
1730 
1731 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1732 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1733 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1734 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
1735 		gb_addr_config = POLARIS11_GB_ADDR_CONFIG_GOLDEN;
1736 		break;
1737 	case CHIP_POLARIS10:
1738 	case CHIP_VEGAM:
1739 		ret = amdgpu_atombios_get_gfx_info(adev);
1740 		if (ret)
1741 			return ret;
1742 		adev->gfx.config.max_gprs = 256;
1743 		adev->gfx.config.max_gs_threads = 32;
1744 		adev->gfx.config.max_hw_contexts = 8;
1745 
1746 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1747 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1748 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1749 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
1750 		gb_addr_config = TONGA_GB_ADDR_CONFIG_GOLDEN;
1751 		break;
1752 	case CHIP_TONGA:
1753 		adev->gfx.config.max_shader_engines = 4;
1754 		adev->gfx.config.max_tile_pipes = 8;
1755 		adev->gfx.config.max_cu_per_sh = 8;
1756 		adev->gfx.config.max_sh_per_se = 1;
1757 		adev->gfx.config.max_backends_per_se = 2;
1758 		adev->gfx.config.max_texture_channel_caches = 8;
1759 		adev->gfx.config.max_gprs = 256;
1760 		adev->gfx.config.max_gs_threads = 32;
1761 		adev->gfx.config.max_hw_contexts = 8;
1762 
1763 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1764 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1765 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1766 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
1767 		gb_addr_config = TONGA_GB_ADDR_CONFIG_GOLDEN;
1768 		break;
1769 	case CHIP_CARRIZO:
1770 		adev->gfx.config.max_shader_engines = 1;
1771 		adev->gfx.config.max_tile_pipes = 2;
1772 		adev->gfx.config.max_sh_per_se = 1;
1773 		adev->gfx.config.max_backends_per_se = 2;
1774 		adev->gfx.config.max_cu_per_sh = 8;
1775 		adev->gfx.config.max_texture_channel_caches = 2;
1776 		adev->gfx.config.max_gprs = 256;
1777 		adev->gfx.config.max_gs_threads = 32;
1778 		adev->gfx.config.max_hw_contexts = 8;
1779 
1780 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1781 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1782 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1783 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
1784 		gb_addr_config = CARRIZO_GB_ADDR_CONFIG_GOLDEN;
1785 		break;
1786 	case CHIP_STONEY:
1787 		adev->gfx.config.max_shader_engines = 1;
1788 		adev->gfx.config.max_tile_pipes = 2;
1789 		adev->gfx.config.max_sh_per_se = 1;
1790 		adev->gfx.config.max_backends_per_se = 1;
1791 		adev->gfx.config.max_cu_per_sh = 3;
1792 		adev->gfx.config.max_texture_channel_caches = 2;
1793 		adev->gfx.config.max_gprs = 256;
1794 		adev->gfx.config.max_gs_threads = 16;
1795 		adev->gfx.config.max_hw_contexts = 8;
1796 
1797 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1798 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1799 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1800 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
1801 		gb_addr_config = CARRIZO_GB_ADDR_CONFIG_GOLDEN;
1802 		break;
1803 	default:
1804 		adev->gfx.config.max_shader_engines = 2;
1805 		adev->gfx.config.max_tile_pipes = 4;
1806 		adev->gfx.config.max_cu_per_sh = 2;
1807 		adev->gfx.config.max_sh_per_se = 1;
1808 		adev->gfx.config.max_backends_per_se = 2;
1809 		adev->gfx.config.max_texture_channel_caches = 4;
1810 		adev->gfx.config.max_gprs = 256;
1811 		adev->gfx.config.max_gs_threads = 32;
1812 		adev->gfx.config.max_hw_contexts = 8;
1813 
1814 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1815 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1816 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1817 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
1818 		gb_addr_config = TONGA_GB_ADDR_CONFIG_GOLDEN;
1819 		break;
1820 	}
1821 
1822 	adev->gfx.config.mc_arb_ramcfg = RREG32(mmMC_ARB_RAMCFG);
1823 	mc_arb_ramcfg = adev->gfx.config.mc_arb_ramcfg;
1824 
1825 	adev->gfx.config.num_banks = REG_GET_FIELD(mc_arb_ramcfg,
1826 				MC_ARB_RAMCFG, NOOFBANK);
1827 	adev->gfx.config.num_ranks = REG_GET_FIELD(mc_arb_ramcfg,
1828 				MC_ARB_RAMCFG, NOOFRANKS);
1829 
1830 	adev->gfx.config.num_tile_pipes = adev->gfx.config.max_tile_pipes;
1831 	adev->gfx.config.mem_max_burst_length_bytes = 256;
1832 	if (adev->flags & AMD_IS_APU) {
1833 		/* Get memory bank mapping mode. */
1834 		tmp = RREG32(mmMC_FUS_DRAM0_BANK_ADDR_MAPPING);
1835 		dimm00_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM0_BANK_ADDR_MAPPING, DIMM0ADDRMAP);
1836 		dimm01_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM0_BANK_ADDR_MAPPING, DIMM1ADDRMAP);
1837 
1838 		tmp = RREG32(mmMC_FUS_DRAM1_BANK_ADDR_MAPPING);
1839 		dimm10_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM1_BANK_ADDR_MAPPING, DIMM0ADDRMAP);
1840 		dimm11_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM1_BANK_ADDR_MAPPING, DIMM1ADDRMAP);
1841 
1842 		/* Validate settings in case only one DIMM installed. */
1843 		if ((dimm00_addr_map == 0) || (dimm00_addr_map == 3) || (dimm00_addr_map == 4) || (dimm00_addr_map > 12))
1844 			dimm00_addr_map = 0;
1845 		if ((dimm01_addr_map == 0) || (dimm01_addr_map == 3) || (dimm01_addr_map == 4) || (dimm01_addr_map > 12))
1846 			dimm01_addr_map = 0;
1847 		if ((dimm10_addr_map == 0) || (dimm10_addr_map == 3) || (dimm10_addr_map == 4) || (dimm10_addr_map > 12))
1848 			dimm10_addr_map = 0;
1849 		if ((dimm11_addr_map == 0) || (dimm11_addr_map == 3) || (dimm11_addr_map == 4) || (dimm11_addr_map > 12))
1850 			dimm11_addr_map = 0;
1851 
1852 		/* If DIMM Addr map is 8GB, ROW size should be 2KB. Otherwise 1KB. */
1853 		/* If ROW size(DIMM1) != ROW size(DMIMM0), ROW size should be larger one. */
1854 		if ((dimm00_addr_map == 11) || (dimm01_addr_map == 11) || (dimm10_addr_map == 11) || (dimm11_addr_map == 11))
1855 			adev->gfx.config.mem_row_size_in_kb = 2;
1856 		else
1857 			adev->gfx.config.mem_row_size_in_kb = 1;
1858 	} else {
1859 		tmp = REG_GET_FIELD(mc_arb_ramcfg, MC_ARB_RAMCFG, NOOFCOLS);
1860 		adev->gfx.config.mem_row_size_in_kb = (4 * (1 << (8 + tmp))) / 1024;
1861 		if (adev->gfx.config.mem_row_size_in_kb > 4)
1862 			adev->gfx.config.mem_row_size_in_kb = 4;
1863 	}
1864 
1865 	adev->gfx.config.shader_engine_tile_size = 32;
1866 	adev->gfx.config.num_gpus = 1;
1867 	adev->gfx.config.multi_gpu_tile_size = 64;
1868 
1869 	/* fix up row size */
1870 	switch (adev->gfx.config.mem_row_size_in_kb) {
1871 	case 1:
1872 	default:
1873 		gb_addr_config = REG_SET_FIELD(gb_addr_config, GB_ADDR_CONFIG, ROW_SIZE, 0);
1874 		break;
1875 	case 2:
1876 		gb_addr_config = REG_SET_FIELD(gb_addr_config, GB_ADDR_CONFIG, ROW_SIZE, 1);
1877 		break;
1878 	case 4:
1879 		gb_addr_config = REG_SET_FIELD(gb_addr_config, GB_ADDR_CONFIG, ROW_SIZE, 2);
1880 		break;
1881 	}
1882 	adev->gfx.config.gb_addr_config = gb_addr_config;
1883 
1884 	return 0;
1885 }
1886 
1887 static int gfx_v8_0_compute_ring_init(struct amdgpu_device *adev, int ring_id,
1888 					int mec, int pipe, int queue)
1889 {
1890 	int r;
1891 	unsigned irq_type;
1892 	struct amdgpu_ring *ring = &adev->gfx.compute_ring[ring_id];
1893 	unsigned int hw_prio;
1894 
1895 	ring = &adev->gfx.compute_ring[ring_id];
1896 
1897 	/* mec0 is me1 */
1898 	ring->me = mec + 1;
1899 	ring->pipe = pipe;
1900 	ring->queue = queue;
1901 
1902 	ring->ring_obj = NULL;
1903 	ring->use_doorbell = true;
1904 	ring->doorbell_index = adev->doorbell_index.mec_ring0 + ring_id;
1905 	ring->eop_gpu_addr = adev->gfx.mec.hpd_eop_gpu_addr
1906 				+ (ring_id * GFX8_MEC_HPD_SIZE);
1907 	sprintf(ring->name, "comp_%d.%d.%d", ring->me, ring->pipe, ring->queue);
1908 
1909 	irq_type = AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP
1910 		+ ((ring->me - 1) * adev->gfx.mec.num_pipe_per_mec)
1911 		+ ring->pipe;
1912 
1913 	hw_prio = amdgpu_gfx_is_high_priority_compute_queue(adev, ring) ?
1914 			AMDGPU_RING_PRIO_2 : AMDGPU_RING_PRIO_DEFAULT;
1915 	/* type-2 packets are deprecated on MEC, use type-3 instead */
1916 	r = amdgpu_ring_init(adev, ring, 1024, &adev->gfx.eop_irq, irq_type,
1917 			     hw_prio, NULL);
1918 	if (r)
1919 		return r;
1920 
1921 
1922 	return 0;
1923 }
1924 
1925 static void gfx_v8_0_sq_irq_work_func(struct work_struct *work);
1926 
1927 static int gfx_v8_0_sw_init(void *handle)
1928 {
1929 	int i, j, k, r, ring_id;
1930 	struct amdgpu_ring *ring;
1931 	struct amdgpu_kiq *kiq;
1932 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1933 
1934 	switch (adev->asic_type) {
1935 	case CHIP_TONGA:
1936 	case CHIP_CARRIZO:
1937 	case CHIP_FIJI:
1938 	case CHIP_POLARIS10:
1939 	case CHIP_POLARIS11:
1940 	case CHIP_POLARIS12:
1941 	case CHIP_VEGAM:
1942 		adev->gfx.mec.num_mec = 2;
1943 		break;
1944 	case CHIP_TOPAZ:
1945 	case CHIP_STONEY:
1946 	default:
1947 		adev->gfx.mec.num_mec = 1;
1948 		break;
1949 	}
1950 
1951 	adev->gfx.mec.num_pipe_per_mec = 4;
1952 	adev->gfx.mec.num_queue_per_pipe = 8;
1953 
1954 	/* EOP Event */
1955 	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, VISLANDS30_IV_SRCID_CP_END_OF_PIPE, &adev->gfx.eop_irq);
1956 	if (r)
1957 		return r;
1958 
1959 	/* Privileged reg */
1960 	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, VISLANDS30_IV_SRCID_CP_PRIV_REG_FAULT,
1961 			      &adev->gfx.priv_reg_irq);
1962 	if (r)
1963 		return r;
1964 
1965 	/* Privileged inst */
1966 	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, VISLANDS30_IV_SRCID_CP_PRIV_INSTR_FAULT,
1967 			      &adev->gfx.priv_inst_irq);
1968 	if (r)
1969 		return r;
1970 
1971 	/* Add CP EDC/ECC irq  */
1972 	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, VISLANDS30_IV_SRCID_CP_ECC_ERROR,
1973 			      &adev->gfx.cp_ecc_error_irq);
1974 	if (r)
1975 		return r;
1976 
1977 	/* SQ interrupts. */
1978 	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, VISLANDS30_IV_SRCID_SQ_INTERRUPT_MSG,
1979 			      &adev->gfx.sq_irq);
1980 	if (r) {
1981 		DRM_ERROR("amdgpu_irq_add() for SQ failed: %d\n", r);
1982 		return r;
1983 	}
1984 
1985 	INIT_WORK(&adev->gfx.sq_work.work, gfx_v8_0_sq_irq_work_func);
1986 
1987 	adev->gfx.gfx_current_status = AMDGPU_GFX_NORMAL_MODE;
1988 
1989 	r = gfx_v8_0_init_microcode(adev);
1990 	if (r) {
1991 		DRM_ERROR("Failed to load gfx firmware!\n");
1992 		return r;
1993 	}
1994 
1995 	r = adev->gfx.rlc.funcs->init(adev);
1996 	if (r) {
1997 		DRM_ERROR("Failed to init rlc BOs!\n");
1998 		return r;
1999 	}
2000 
2001 	r = gfx_v8_0_mec_init(adev);
2002 	if (r) {
2003 		DRM_ERROR("Failed to init MEC BOs!\n");
2004 		return r;
2005 	}
2006 
2007 	/* set up the gfx ring */
2008 	for (i = 0; i < adev->gfx.num_gfx_rings; i++) {
2009 		ring = &adev->gfx.gfx_ring[i];
2010 		ring->ring_obj = NULL;
2011 		sprintf(ring->name, "gfx");
2012 		/* no gfx doorbells on iceland */
2013 		if (adev->asic_type != CHIP_TOPAZ) {
2014 			ring->use_doorbell = true;
2015 			ring->doorbell_index = adev->doorbell_index.gfx_ring0;
2016 		}
2017 
2018 		r = amdgpu_ring_init(adev, ring, 1024, &adev->gfx.eop_irq,
2019 				     AMDGPU_CP_IRQ_GFX_ME0_PIPE0_EOP,
2020 				     AMDGPU_RING_PRIO_DEFAULT, NULL);
2021 		if (r)
2022 			return r;
2023 	}
2024 
2025 
2026 	/* set up the compute queues - allocate horizontally across pipes */
2027 	ring_id = 0;
2028 	for (i = 0; i < adev->gfx.mec.num_mec; ++i) {
2029 		for (j = 0; j < adev->gfx.mec.num_queue_per_pipe; j++) {
2030 			for (k = 0; k < adev->gfx.mec.num_pipe_per_mec; k++) {
2031 				if (!amdgpu_gfx_is_mec_queue_enabled(adev, i, k, j))
2032 					continue;
2033 
2034 				r = gfx_v8_0_compute_ring_init(adev,
2035 								ring_id,
2036 								i, k, j);
2037 				if (r)
2038 					return r;
2039 
2040 				ring_id++;
2041 			}
2042 		}
2043 	}
2044 
2045 	r = amdgpu_gfx_kiq_init(adev, GFX8_MEC_HPD_SIZE);
2046 	if (r) {
2047 		DRM_ERROR("Failed to init KIQ BOs!\n");
2048 		return r;
2049 	}
2050 
2051 	kiq = &adev->gfx.kiq;
2052 	r = amdgpu_gfx_kiq_init_ring(adev, &kiq->ring, &kiq->irq);
2053 	if (r)
2054 		return r;
2055 
2056 	/* create MQD for all compute queues as well as KIQ for SRIOV case */
2057 	r = amdgpu_gfx_mqd_sw_init(adev, sizeof(struct vi_mqd_allocation));
2058 	if (r)
2059 		return r;
2060 
2061 	adev->gfx.ce_ram_size = 0x8000;
2062 
2063 	r = gfx_v8_0_gpu_early_init(adev);
2064 	if (r)
2065 		return r;
2066 
2067 	return 0;
2068 }
2069 
2070 static int gfx_v8_0_sw_fini(void *handle)
2071 {
2072 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2073 	int i;
2074 
2075 	for (i = 0; i < adev->gfx.num_gfx_rings; i++)
2076 		amdgpu_ring_fini(&adev->gfx.gfx_ring[i]);
2077 	for (i = 0; i < adev->gfx.num_compute_rings; i++)
2078 		amdgpu_ring_fini(&adev->gfx.compute_ring[i]);
2079 
2080 	amdgpu_gfx_mqd_sw_fini(adev);
2081 	amdgpu_gfx_kiq_free_ring(&adev->gfx.kiq.ring);
2082 	amdgpu_gfx_kiq_fini(adev);
2083 
2084 	gfx_v8_0_mec_fini(adev);
2085 	amdgpu_gfx_rlc_fini(adev);
2086 	amdgpu_bo_free_kernel(&adev->gfx.rlc.clear_state_obj,
2087 				&adev->gfx.rlc.clear_state_gpu_addr,
2088 				(void **)&adev->gfx.rlc.cs_ptr);
2089 	if ((adev->asic_type == CHIP_CARRIZO) ||
2090 	    (adev->asic_type == CHIP_STONEY)) {
2091 		amdgpu_bo_free_kernel(&adev->gfx.rlc.cp_table_obj,
2092 				&adev->gfx.rlc.cp_table_gpu_addr,
2093 				(void **)&adev->gfx.rlc.cp_table_ptr);
2094 	}
2095 	gfx_v8_0_free_microcode(adev);
2096 
2097 	return 0;
2098 }
2099 
2100 static void gfx_v8_0_tiling_mode_table_init(struct amdgpu_device *adev)
2101 {
2102 	uint32_t *modearray, *mod2array;
2103 	const u32 num_tile_mode_states = ARRAY_SIZE(adev->gfx.config.tile_mode_array);
2104 	const u32 num_secondary_tile_mode_states = ARRAY_SIZE(adev->gfx.config.macrotile_mode_array);
2105 	u32 reg_offset;
2106 
2107 	modearray = adev->gfx.config.tile_mode_array;
2108 	mod2array = adev->gfx.config.macrotile_mode_array;
2109 
2110 	for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
2111 		modearray[reg_offset] = 0;
2112 
2113 	for (reg_offset = 0; reg_offset <  num_secondary_tile_mode_states; reg_offset++)
2114 		mod2array[reg_offset] = 0;
2115 
2116 	switch (adev->asic_type) {
2117 	case CHIP_TOPAZ:
2118 		modearray[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2119 				PIPE_CONFIG(ADDR_SURF_P2) |
2120 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
2121 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2122 		modearray[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2123 				PIPE_CONFIG(ADDR_SURF_P2) |
2124 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
2125 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2126 		modearray[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2127 				PIPE_CONFIG(ADDR_SURF_P2) |
2128 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
2129 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2130 		modearray[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2131 				PIPE_CONFIG(ADDR_SURF_P2) |
2132 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
2133 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2134 		modearray[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2135 				PIPE_CONFIG(ADDR_SURF_P2) |
2136 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2137 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2138 		modearray[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2139 				PIPE_CONFIG(ADDR_SURF_P2) |
2140 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2141 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2142 		modearray[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2143 				PIPE_CONFIG(ADDR_SURF_P2) |
2144 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2145 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2146 		modearray[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
2147 				PIPE_CONFIG(ADDR_SURF_P2));
2148 		modearray[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2149 				PIPE_CONFIG(ADDR_SURF_P2) |
2150 				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2151 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2152 		modearray[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2153 				 PIPE_CONFIG(ADDR_SURF_P2) |
2154 				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2155 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2156 		modearray[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2157 				 PIPE_CONFIG(ADDR_SURF_P2) |
2158 				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2159 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2160 		modearray[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2161 				 PIPE_CONFIG(ADDR_SURF_P2) |
2162 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2163 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2164 		modearray[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2165 				 PIPE_CONFIG(ADDR_SURF_P2) |
2166 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2167 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2168 		modearray[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
2169 				 PIPE_CONFIG(ADDR_SURF_P2) |
2170 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2171 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2172 		modearray[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2173 				 PIPE_CONFIG(ADDR_SURF_P2) |
2174 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2175 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2176 		modearray[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2177 				 PIPE_CONFIG(ADDR_SURF_P2) |
2178 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2179 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2180 		modearray[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2181 				 PIPE_CONFIG(ADDR_SURF_P2) |
2182 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2183 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2184 		modearray[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2185 				 PIPE_CONFIG(ADDR_SURF_P2) |
2186 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2187 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2188 		modearray[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
2189 				 PIPE_CONFIG(ADDR_SURF_P2) |
2190 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2191 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2192 		modearray[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
2193 				 PIPE_CONFIG(ADDR_SURF_P2) |
2194 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2195 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2196 		modearray[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2197 				 PIPE_CONFIG(ADDR_SURF_P2) |
2198 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2199 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2200 		modearray[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
2201 				 PIPE_CONFIG(ADDR_SURF_P2) |
2202 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2203 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2204 		modearray[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
2205 				 PIPE_CONFIG(ADDR_SURF_P2) |
2206 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2207 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2208 		modearray[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2209 				 PIPE_CONFIG(ADDR_SURF_P2) |
2210 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2211 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2212 		modearray[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2213 				 PIPE_CONFIG(ADDR_SURF_P2) |
2214 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2215 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2216 		modearray[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2217 				 PIPE_CONFIG(ADDR_SURF_P2) |
2218 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2219 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2220 
2221 		mod2array[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
2222 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2223 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2224 				NUM_BANKS(ADDR_SURF_8_BANK));
2225 		mod2array[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
2226 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2227 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2228 				NUM_BANKS(ADDR_SURF_8_BANK));
2229 		mod2array[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
2230 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2231 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2232 				NUM_BANKS(ADDR_SURF_8_BANK));
2233 		mod2array[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2234 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2235 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2236 				NUM_BANKS(ADDR_SURF_8_BANK));
2237 		mod2array[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2238 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2239 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2240 				NUM_BANKS(ADDR_SURF_8_BANK));
2241 		mod2array[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2242 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2243 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2244 				NUM_BANKS(ADDR_SURF_8_BANK));
2245 		mod2array[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2246 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2247 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2248 				NUM_BANKS(ADDR_SURF_8_BANK));
2249 		mod2array[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
2250 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
2251 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2252 				NUM_BANKS(ADDR_SURF_16_BANK));
2253 		mod2array[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
2254 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2255 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2256 				NUM_BANKS(ADDR_SURF_16_BANK));
2257 		mod2array[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
2258 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2259 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2260 				 NUM_BANKS(ADDR_SURF_16_BANK));
2261 		mod2array[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
2262 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2263 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2264 				 NUM_BANKS(ADDR_SURF_16_BANK));
2265 		mod2array[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2266 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2267 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2268 				 NUM_BANKS(ADDR_SURF_16_BANK));
2269 		mod2array[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2270 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2271 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2272 				 NUM_BANKS(ADDR_SURF_16_BANK));
2273 		mod2array[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2274 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2275 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2276 				 NUM_BANKS(ADDR_SURF_8_BANK));
2277 
2278 		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
2279 			if (reg_offset != 7 && reg_offset != 12 && reg_offset != 17 &&
2280 			    reg_offset != 23)
2281 				WREG32(mmGB_TILE_MODE0 + reg_offset, modearray[reg_offset]);
2282 
2283 		for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
2284 			if (reg_offset != 7)
2285 				WREG32(mmGB_MACROTILE_MODE0 + reg_offset, mod2array[reg_offset]);
2286 
2287 		break;
2288 	case CHIP_FIJI:
2289 	case CHIP_VEGAM:
2290 		modearray[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2291 				PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2292 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
2293 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2294 		modearray[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2295 				PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2296 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
2297 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2298 		modearray[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2299 				PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2300 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
2301 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2302 		modearray[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2303 				PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2304 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
2305 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2306 		modearray[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2307 				PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2308 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2309 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2310 		modearray[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2311 				PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2312 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2313 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2314 		modearray[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2315 				PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2316 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2317 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2318 		modearray[7] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2319 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2320 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2321 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2322 		modearray[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
2323 				PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16));
2324 		modearray[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2325 				PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2326 				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2327 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2328 		modearray[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2329 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2330 				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2331 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2332 		modearray[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2333 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2334 				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2335 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2336 		modearray[12] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2337 				 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2338 				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2339 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2340 		modearray[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2341 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2342 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2343 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2344 		modearray[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2345 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2346 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2347 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2348 		modearray[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
2349 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2350 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2351 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2352 		modearray[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2353 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2354 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2355 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2356 		modearray[17] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2357 				 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2358 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2359 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2360 		modearray[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2361 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2362 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2363 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2364 		modearray[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2365 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2366 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2367 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2368 		modearray[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2369 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2370 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2371 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2372 		modearray[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
2373 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2374 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2375 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2376 		modearray[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
2377 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2378 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2379 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2380 		modearray[23] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
2381 				 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2382 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2383 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2384 		modearray[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2385 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2386 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2387 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2388 		modearray[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
2389 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2390 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2391 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2392 		modearray[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
2393 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2394 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2395 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2396 		modearray[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2397 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2398 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2399 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2400 		modearray[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2401 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2402 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2403 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2404 		modearray[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2405 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2406 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2407 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2408 		modearray[30] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2409 				 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2410 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2411 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2412 
2413 		mod2array[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2414 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2415 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2416 				NUM_BANKS(ADDR_SURF_8_BANK));
2417 		mod2array[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2418 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2419 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2420 				NUM_BANKS(ADDR_SURF_8_BANK));
2421 		mod2array[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2422 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2423 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2424 				NUM_BANKS(ADDR_SURF_8_BANK));
2425 		mod2array[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2426 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2427 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2428 				NUM_BANKS(ADDR_SURF_8_BANK));
2429 		mod2array[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2430 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2431 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2432 				NUM_BANKS(ADDR_SURF_8_BANK));
2433 		mod2array[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2434 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2435 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2436 				NUM_BANKS(ADDR_SURF_8_BANK));
2437 		mod2array[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2438 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2439 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2440 				NUM_BANKS(ADDR_SURF_8_BANK));
2441 		mod2array[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2442 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
2443 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2444 				NUM_BANKS(ADDR_SURF_8_BANK));
2445 		mod2array[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2446 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2447 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2448 				NUM_BANKS(ADDR_SURF_8_BANK));
2449 		mod2array[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2450 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2451 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2452 				 NUM_BANKS(ADDR_SURF_8_BANK));
2453 		mod2array[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2454 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2455 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2456 				 NUM_BANKS(ADDR_SURF_8_BANK));
2457 		mod2array[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2458 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2459 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2460 				 NUM_BANKS(ADDR_SURF_8_BANK));
2461 		mod2array[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2462 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2463 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2464 				 NUM_BANKS(ADDR_SURF_8_BANK));
2465 		mod2array[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2466 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2467 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2468 				 NUM_BANKS(ADDR_SURF_4_BANK));
2469 
2470 		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
2471 			WREG32(mmGB_TILE_MODE0 + reg_offset, modearray[reg_offset]);
2472 
2473 		for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
2474 			if (reg_offset != 7)
2475 				WREG32(mmGB_MACROTILE_MODE0 + reg_offset, mod2array[reg_offset]);
2476 
2477 		break;
2478 	case CHIP_TONGA:
2479 		modearray[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2480 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2481 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
2482 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2483 		modearray[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2484 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2485 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
2486 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2487 		modearray[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2488 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2489 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
2490 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2491 		modearray[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2492 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2493 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
2494 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2495 		modearray[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2496 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2497 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2498 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2499 		modearray[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2500 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2501 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2502 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2503 		modearray[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2504 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2505 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2506 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2507 		modearray[7] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2508 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2509 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2510 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2511 		modearray[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
2512 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16));
2513 		modearray[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2514 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2515 				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2516 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2517 		modearray[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2518 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2519 				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2520 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2521 		modearray[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2522 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2523 				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2524 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2525 		modearray[12] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2526 				 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2527 				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2528 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2529 		modearray[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2530 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2531 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2532 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2533 		modearray[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2534 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2535 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2536 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2537 		modearray[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
2538 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2539 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2540 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2541 		modearray[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2542 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2543 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2544 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2545 		modearray[17] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2546 				 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2547 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2548 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2549 		modearray[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2550 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2551 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2552 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2553 		modearray[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2554 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2555 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2556 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2557 		modearray[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2558 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2559 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2560 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2561 		modearray[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
2562 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2563 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2564 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2565 		modearray[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
2566 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2567 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2568 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2569 		modearray[23] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
2570 				 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2571 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2572 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2573 		modearray[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2574 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2575 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2576 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2577 		modearray[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
2578 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2579 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2580 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2581 		modearray[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
2582 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2583 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2584 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2585 		modearray[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2586 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2587 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2588 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2589 		modearray[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2590 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2591 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2592 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2593 		modearray[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2594 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2595 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2596 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2597 		modearray[30] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2598 				 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2599 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2600 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2601 
2602 		mod2array[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2603 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2604 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2605 				NUM_BANKS(ADDR_SURF_16_BANK));
2606 		mod2array[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2607 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2608 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2609 				NUM_BANKS(ADDR_SURF_16_BANK));
2610 		mod2array[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2611 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2612 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2613 				NUM_BANKS(ADDR_SURF_16_BANK));
2614 		mod2array[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2615 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2616 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2617 				NUM_BANKS(ADDR_SURF_16_BANK));
2618 		mod2array[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2619 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2620 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2621 				NUM_BANKS(ADDR_SURF_16_BANK));
2622 		mod2array[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2623 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2624 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2625 				NUM_BANKS(ADDR_SURF_16_BANK));
2626 		mod2array[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2627 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2628 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2629 				NUM_BANKS(ADDR_SURF_16_BANK));
2630 		mod2array[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2631 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
2632 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2633 				NUM_BANKS(ADDR_SURF_16_BANK));
2634 		mod2array[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2635 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2636 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2637 				NUM_BANKS(ADDR_SURF_16_BANK));
2638 		mod2array[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2639 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2640 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2641 				 NUM_BANKS(ADDR_SURF_16_BANK));
2642 		mod2array[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2643 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2644 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2645 				 NUM_BANKS(ADDR_SURF_16_BANK));
2646 		mod2array[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2647 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2648 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2649 				 NUM_BANKS(ADDR_SURF_8_BANK));
2650 		mod2array[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2651 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2652 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2653 				 NUM_BANKS(ADDR_SURF_4_BANK));
2654 		mod2array[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2655 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2656 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2657 				 NUM_BANKS(ADDR_SURF_4_BANK));
2658 
2659 		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
2660 			WREG32(mmGB_TILE_MODE0 + reg_offset, modearray[reg_offset]);
2661 
2662 		for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
2663 			if (reg_offset != 7)
2664 				WREG32(mmGB_MACROTILE_MODE0 + reg_offset, mod2array[reg_offset]);
2665 
2666 		break;
2667 	case CHIP_POLARIS11:
2668 	case CHIP_POLARIS12:
2669 		modearray[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2670 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2671 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
2672 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2673 		modearray[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2674 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2675 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
2676 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2677 		modearray[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2678 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2679 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
2680 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2681 		modearray[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2682 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2683 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
2684 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2685 		modearray[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2686 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2687 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2688 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2689 		modearray[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2690 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2691 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2692 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2693 		modearray[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2694 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2695 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2696 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2697 		modearray[7] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2698 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2699 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2700 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2701 		modearray[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
2702 				PIPE_CONFIG(ADDR_SURF_P4_16x16));
2703 		modearray[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2704 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2705 				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2706 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2707 		modearray[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2708 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2709 				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2710 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2711 		modearray[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2712 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2713 				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2714 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2715 		modearray[12] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2716 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2717 				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2718 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2719 		modearray[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2720 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2721 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2722 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2723 		modearray[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2724 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2725 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2726 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2727 		modearray[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
2728 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2729 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2730 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2731 		modearray[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2732 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2733 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2734 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2735 		modearray[17] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2736 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2737 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2738 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2739 		modearray[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2740 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2741 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2742 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2743 		modearray[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2744 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2745 				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2746 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2747 		modearray[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2748 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2749 				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2750 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2751 		modearray[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
2752 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2753 				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2754 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2755 		modearray[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
2756 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2757 				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2758 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2759 		modearray[23] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
2760 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2761 				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2762 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2763 		modearray[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2764 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2765 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2766 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2767 		modearray[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
2768 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2769 				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2770 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2771 		modearray[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
2772 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2773 				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2774 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2775 		modearray[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2776 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2777 				MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2778 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2779 		modearray[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2780 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2781 				MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2782 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2783 		modearray[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2784 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2785 				MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2786 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2787 		modearray[30] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2788 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2789 				MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2790 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2791 
2792 		mod2array[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2793 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2794 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2795 				NUM_BANKS(ADDR_SURF_16_BANK));
2796 
2797 		mod2array[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2798 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2799 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2800 				NUM_BANKS(ADDR_SURF_16_BANK));
2801 
2802 		mod2array[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2803 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2804 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2805 				NUM_BANKS(ADDR_SURF_16_BANK));
2806 
2807 		mod2array[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2808 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2809 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2810 				NUM_BANKS(ADDR_SURF_16_BANK));
2811 
2812 		mod2array[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2813 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2814 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2815 				NUM_BANKS(ADDR_SURF_16_BANK));
2816 
2817 		mod2array[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2818 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2819 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2820 				NUM_BANKS(ADDR_SURF_16_BANK));
2821 
2822 		mod2array[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2823 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2824 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2825 				NUM_BANKS(ADDR_SURF_16_BANK));
2826 
2827 		mod2array[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
2828 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
2829 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2830 				NUM_BANKS(ADDR_SURF_16_BANK));
2831 
2832 		mod2array[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
2833 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2834 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2835 				NUM_BANKS(ADDR_SURF_16_BANK));
2836 
2837 		mod2array[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2838 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2839 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2840 				NUM_BANKS(ADDR_SURF_16_BANK));
2841 
2842 		mod2array[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2843 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2844 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2845 				NUM_BANKS(ADDR_SURF_16_BANK));
2846 
2847 		mod2array[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2848 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2849 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2850 				NUM_BANKS(ADDR_SURF_16_BANK));
2851 
2852 		mod2array[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2853 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2854 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2855 				NUM_BANKS(ADDR_SURF_8_BANK));
2856 
2857 		mod2array[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2858 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2859 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2860 				NUM_BANKS(ADDR_SURF_4_BANK));
2861 
2862 		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
2863 			WREG32(mmGB_TILE_MODE0 + reg_offset, modearray[reg_offset]);
2864 
2865 		for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
2866 			if (reg_offset != 7)
2867 				WREG32(mmGB_MACROTILE_MODE0 + reg_offset, mod2array[reg_offset]);
2868 
2869 		break;
2870 	case CHIP_POLARIS10:
2871 		modearray[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2872 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2873 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
2874 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2875 		modearray[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2876 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2877 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
2878 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2879 		modearray[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2880 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2881 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
2882 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2883 		modearray[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2884 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2885 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
2886 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2887 		modearray[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2888 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2889 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2890 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2891 		modearray[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2892 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2893 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2894 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2895 		modearray[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2896 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2897 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2898 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2899 		modearray[7] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2900 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2901 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2902 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2903 		modearray[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
2904 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16));
2905 		modearray[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2906 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2907 				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2908 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2909 		modearray[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2910 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2911 				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2912 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2913 		modearray[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2914 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2915 				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2916 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2917 		modearray[12] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2918 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2919 				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2920 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2921 		modearray[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2922 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2923 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2924 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2925 		modearray[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2926 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2927 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2928 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2929 		modearray[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
2930 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2931 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2932 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2933 		modearray[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2934 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2935 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2936 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2937 		modearray[17] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2938 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2939 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2940 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2941 		modearray[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2942 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2943 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2944 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2945 		modearray[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2946 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2947 				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2948 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2949 		modearray[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2950 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2951 				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2952 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2953 		modearray[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
2954 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2955 				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2956 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2957 		modearray[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
2958 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2959 				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2960 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2961 		modearray[23] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
2962 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2963 				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2964 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2965 		modearray[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2966 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2967 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2968 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2969 		modearray[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
2970 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2971 				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2972 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2973 		modearray[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
2974 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2975 				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2976 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2977 		modearray[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2978 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2979 				MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2980 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2981 		modearray[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2982 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2983 				MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2984 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2985 		modearray[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2986 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2987 				MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2988 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2989 		modearray[30] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2990 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2991 				MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2992 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2993 
2994 		mod2array[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2995 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2996 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2997 				NUM_BANKS(ADDR_SURF_16_BANK));
2998 
2999 		mod2array[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3000 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
3001 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3002 				NUM_BANKS(ADDR_SURF_16_BANK));
3003 
3004 		mod2array[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3005 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
3006 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3007 				NUM_BANKS(ADDR_SURF_16_BANK));
3008 
3009 		mod2array[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3010 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
3011 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3012 				NUM_BANKS(ADDR_SURF_16_BANK));
3013 
3014 		mod2array[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3015 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
3016 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3017 				NUM_BANKS(ADDR_SURF_16_BANK));
3018 
3019 		mod2array[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3020 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3021 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
3022 				NUM_BANKS(ADDR_SURF_16_BANK));
3023 
3024 		mod2array[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3025 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3026 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
3027 				NUM_BANKS(ADDR_SURF_16_BANK));
3028 
3029 		mod2array[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3030 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
3031 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3032 				NUM_BANKS(ADDR_SURF_16_BANK));
3033 
3034 		mod2array[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3035 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
3036 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3037 				NUM_BANKS(ADDR_SURF_16_BANK));
3038 
3039 		mod2array[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3040 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
3041 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3042 				NUM_BANKS(ADDR_SURF_16_BANK));
3043 
3044 		mod2array[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3045 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3046 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3047 				NUM_BANKS(ADDR_SURF_16_BANK));
3048 
3049 		mod2array[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3050 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3051 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
3052 				NUM_BANKS(ADDR_SURF_8_BANK));
3053 
3054 		mod2array[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3055 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3056 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
3057 				NUM_BANKS(ADDR_SURF_4_BANK));
3058 
3059 		mod2array[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3060 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3061 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
3062 				NUM_BANKS(ADDR_SURF_4_BANK));
3063 
3064 		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
3065 			WREG32(mmGB_TILE_MODE0 + reg_offset, modearray[reg_offset]);
3066 
3067 		for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
3068 			if (reg_offset != 7)
3069 				WREG32(mmGB_MACROTILE_MODE0 + reg_offset, mod2array[reg_offset]);
3070 
3071 		break;
3072 	case CHIP_STONEY:
3073 		modearray[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3074 				PIPE_CONFIG(ADDR_SURF_P2) |
3075 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
3076 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3077 		modearray[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3078 				PIPE_CONFIG(ADDR_SURF_P2) |
3079 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
3080 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3081 		modearray[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3082 				PIPE_CONFIG(ADDR_SURF_P2) |
3083 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
3084 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3085 		modearray[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3086 				PIPE_CONFIG(ADDR_SURF_P2) |
3087 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
3088 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3089 		modearray[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3090 				PIPE_CONFIG(ADDR_SURF_P2) |
3091 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
3092 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3093 		modearray[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
3094 				PIPE_CONFIG(ADDR_SURF_P2) |
3095 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
3096 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3097 		modearray[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
3098 				PIPE_CONFIG(ADDR_SURF_P2) |
3099 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
3100 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3101 		modearray[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
3102 				PIPE_CONFIG(ADDR_SURF_P2));
3103 		modearray[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
3104 				PIPE_CONFIG(ADDR_SURF_P2) |
3105 				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
3106 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3107 		modearray[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3108 				 PIPE_CONFIG(ADDR_SURF_P2) |
3109 				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
3110 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3111 		modearray[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
3112 				 PIPE_CONFIG(ADDR_SURF_P2) |
3113 				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
3114 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
3115 		modearray[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
3116 				 PIPE_CONFIG(ADDR_SURF_P2) |
3117 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3118 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3119 		modearray[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3120 				 PIPE_CONFIG(ADDR_SURF_P2) |
3121 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3122 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3123 		modearray[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
3124 				 PIPE_CONFIG(ADDR_SURF_P2) |
3125 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3126 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3127 		modearray[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
3128 				 PIPE_CONFIG(ADDR_SURF_P2) |
3129 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3130 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
3131 		modearray[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
3132 				 PIPE_CONFIG(ADDR_SURF_P2) |
3133 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3134 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3135 		modearray[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
3136 				 PIPE_CONFIG(ADDR_SURF_P2) |
3137 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3138 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3139 		modearray[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
3140 				 PIPE_CONFIG(ADDR_SURF_P2) |
3141 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3142 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3143 		modearray[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
3144 				 PIPE_CONFIG(ADDR_SURF_P2) |
3145 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3146 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3147 		modearray[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
3148 				 PIPE_CONFIG(ADDR_SURF_P2) |
3149 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3150 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3151 		modearray[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
3152 				 PIPE_CONFIG(ADDR_SURF_P2) |
3153 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3154 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3155 		modearray[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
3156 				 PIPE_CONFIG(ADDR_SURF_P2) |
3157 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3158 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3159 		modearray[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
3160 				 PIPE_CONFIG(ADDR_SURF_P2) |
3161 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3162 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3163 		modearray[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
3164 				 PIPE_CONFIG(ADDR_SURF_P2) |
3165 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
3166 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3167 		modearray[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3168 				 PIPE_CONFIG(ADDR_SURF_P2) |
3169 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
3170 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3171 		modearray[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
3172 				 PIPE_CONFIG(ADDR_SURF_P2) |
3173 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
3174 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
3175 
3176 		mod2array[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3177 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
3178 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3179 				NUM_BANKS(ADDR_SURF_8_BANK));
3180 		mod2array[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3181 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
3182 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3183 				NUM_BANKS(ADDR_SURF_8_BANK));
3184 		mod2array[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3185 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3186 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3187 				NUM_BANKS(ADDR_SURF_8_BANK));
3188 		mod2array[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3189 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3190 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3191 				NUM_BANKS(ADDR_SURF_8_BANK));
3192 		mod2array[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3193 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3194 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3195 				NUM_BANKS(ADDR_SURF_8_BANK));
3196 		mod2array[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3197 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3198 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3199 				NUM_BANKS(ADDR_SURF_8_BANK));
3200 		mod2array[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3201 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3202 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3203 				NUM_BANKS(ADDR_SURF_8_BANK));
3204 		mod2array[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
3205 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
3206 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3207 				NUM_BANKS(ADDR_SURF_16_BANK));
3208 		mod2array[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
3209 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
3210 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3211 				NUM_BANKS(ADDR_SURF_16_BANK));
3212 		mod2array[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
3213 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
3214 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3215 				 NUM_BANKS(ADDR_SURF_16_BANK));
3216 		mod2array[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
3217 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
3218 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3219 				 NUM_BANKS(ADDR_SURF_16_BANK));
3220 		mod2array[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3221 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
3222 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3223 				 NUM_BANKS(ADDR_SURF_16_BANK));
3224 		mod2array[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3225 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3226 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3227 				 NUM_BANKS(ADDR_SURF_16_BANK));
3228 		mod2array[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3229 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3230 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3231 				 NUM_BANKS(ADDR_SURF_8_BANK));
3232 
3233 		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
3234 			if (reg_offset != 7 && reg_offset != 12 && reg_offset != 17 &&
3235 			    reg_offset != 23)
3236 				WREG32(mmGB_TILE_MODE0 + reg_offset, modearray[reg_offset]);
3237 
3238 		for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
3239 			if (reg_offset != 7)
3240 				WREG32(mmGB_MACROTILE_MODE0 + reg_offset, mod2array[reg_offset]);
3241 
3242 		break;
3243 	default:
3244 		dev_warn(adev->dev,
3245 			 "Unknown chip type (%d) in function gfx_v8_0_tiling_mode_table_init() falling through to CHIP_CARRIZO\n",
3246 			 adev->asic_type);
3247 		fallthrough;
3248 
3249 	case CHIP_CARRIZO:
3250 		modearray[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3251 				PIPE_CONFIG(ADDR_SURF_P2) |
3252 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
3253 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3254 		modearray[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3255 				PIPE_CONFIG(ADDR_SURF_P2) |
3256 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
3257 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3258 		modearray[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3259 				PIPE_CONFIG(ADDR_SURF_P2) |
3260 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
3261 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3262 		modearray[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3263 				PIPE_CONFIG(ADDR_SURF_P2) |
3264 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
3265 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3266 		modearray[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3267 				PIPE_CONFIG(ADDR_SURF_P2) |
3268 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
3269 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3270 		modearray[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
3271 				PIPE_CONFIG(ADDR_SURF_P2) |
3272 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
3273 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3274 		modearray[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
3275 				PIPE_CONFIG(ADDR_SURF_P2) |
3276 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
3277 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3278 		modearray[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
3279 				PIPE_CONFIG(ADDR_SURF_P2));
3280 		modearray[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
3281 				PIPE_CONFIG(ADDR_SURF_P2) |
3282 				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
3283 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3284 		modearray[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3285 				 PIPE_CONFIG(ADDR_SURF_P2) |
3286 				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
3287 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3288 		modearray[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
3289 				 PIPE_CONFIG(ADDR_SURF_P2) |
3290 				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
3291 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
3292 		modearray[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
3293 				 PIPE_CONFIG(ADDR_SURF_P2) |
3294 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3295 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3296 		modearray[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3297 				 PIPE_CONFIG(ADDR_SURF_P2) |
3298 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3299 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3300 		modearray[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
3301 				 PIPE_CONFIG(ADDR_SURF_P2) |
3302 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3303 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3304 		modearray[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
3305 				 PIPE_CONFIG(ADDR_SURF_P2) |
3306 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3307 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
3308 		modearray[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
3309 				 PIPE_CONFIG(ADDR_SURF_P2) |
3310 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3311 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3312 		modearray[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
3313 				 PIPE_CONFIG(ADDR_SURF_P2) |
3314 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3315 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3316 		modearray[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
3317 				 PIPE_CONFIG(ADDR_SURF_P2) |
3318 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3319 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3320 		modearray[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
3321 				 PIPE_CONFIG(ADDR_SURF_P2) |
3322 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3323 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3324 		modearray[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
3325 				 PIPE_CONFIG(ADDR_SURF_P2) |
3326 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3327 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3328 		modearray[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
3329 				 PIPE_CONFIG(ADDR_SURF_P2) |
3330 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3331 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3332 		modearray[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
3333 				 PIPE_CONFIG(ADDR_SURF_P2) |
3334 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3335 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3336 		modearray[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
3337 				 PIPE_CONFIG(ADDR_SURF_P2) |
3338 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3339 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3340 		modearray[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
3341 				 PIPE_CONFIG(ADDR_SURF_P2) |
3342 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
3343 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3344 		modearray[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3345 				 PIPE_CONFIG(ADDR_SURF_P2) |
3346 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
3347 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3348 		modearray[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
3349 				 PIPE_CONFIG(ADDR_SURF_P2) |
3350 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
3351 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
3352 
3353 		mod2array[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3354 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
3355 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3356 				NUM_BANKS(ADDR_SURF_8_BANK));
3357 		mod2array[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3358 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
3359 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3360 				NUM_BANKS(ADDR_SURF_8_BANK));
3361 		mod2array[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3362 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3363 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3364 				NUM_BANKS(ADDR_SURF_8_BANK));
3365 		mod2array[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3366 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3367 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3368 				NUM_BANKS(ADDR_SURF_8_BANK));
3369 		mod2array[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3370 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3371 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3372 				NUM_BANKS(ADDR_SURF_8_BANK));
3373 		mod2array[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3374 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3375 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3376 				NUM_BANKS(ADDR_SURF_8_BANK));
3377 		mod2array[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3378 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3379 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3380 				NUM_BANKS(ADDR_SURF_8_BANK));
3381 		mod2array[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
3382 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
3383 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3384 				NUM_BANKS(ADDR_SURF_16_BANK));
3385 		mod2array[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
3386 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
3387 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3388 				NUM_BANKS(ADDR_SURF_16_BANK));
3389 		mod2array[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
3390 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
3391 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3392 				 NUM_BANKS(ADDR_SURF_16_BANK));
3393 		mod2array[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
3394 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
3395 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3396 				 NUM_BANKS(ADDR_SURF_16_BANK));
3397 		mod2array[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3398 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
3399 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3400 				 NUM_BANKS(ADDR_SURF_16_BANK));
3401 		mod2array[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3402 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3403 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3404 				 NUM_BANKS(ADDR_SURF_16_BANK));
3405 		mod2array[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3406 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3407 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3408 				 NUM_BANKS(ADDR_SURF_8_BANK));
3409 
3410 		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
3411 			if (reg_offset != 7 && reg_offset != 12 && reg_offset != 17 &&
3412 			    reg_offset != 23)
3413 				WREG32(mmGB_TILE_MODE0 + reg_offset, modearray[reg_offset]);
3414 
3415 		for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
3416 			if (reg_offset != 7)
3417 				WREG32(mmGB_MACROTILE_MODE0 + reg_offset, mod2array[reg_offset]);
3418 
3419 		break;
3420 	}
3421 }
3422 
3423 static void gfx_v8_0_select_se_sh(struct amdgpu_device *adev,
3424 				  u32 se_num, u32 sh_num, u32 instance)
3425 {
3426 	u32 data;
3427 
3428 	if (instance == 0xffffffff)
3429 		data = REG_SET_FIELD(0, GRBM_GFX_INDEX, INSTANCE_BROADCAST_WRITES, 1);
3430 	else
3431 		data = REG_SET_FIELD(0, GRBM_GFX_INDEX, INSTANCE_INDEX, instance);
3432 
3433 	if (se_num == 0xffffffff)
3434 		data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SE_BROADCAST_WRITES, 1);
3435 	else
3436 		data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SE_INDEX, se_num);
3437 
3438 	if (sh_num == 0xffffffff)
3439 		data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SH_BROADCAST_WRITES, 1);
3440 	else
3441 		data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SH_INDEX, sh_num);
3442 
3443 	WREG32(mmGRBM_GFX_INDEX, data);
3444 }
3445 
3446 static void gfx_v8_0_select_me_pipe_q(struct amdgpu_device *adev,
3447 				  u32 me, u32 pipe, u32 q, u32 vm)
3448 {
3449 	vi_srbm_select(adev, me, pipe, q, vm);
3450 }
3451 
3452 static u32 gfx_v8_0_get_rb_active_bitmap(struct amdgpu_device *adev)
3453 {
3454 	u32 data, mask;
3455 
3456 	data =  RREG32(mmCC_RB_BACKEND_DISABLE) |
3457 		RREG32(mmGC_USER_RB_BACKEND_DISABLE);
3458 
3459 	data = REG_GET_FIELD(data, GC_USER_RB_BACKEND_DISABLE, BACKEND_DISABLE);
3460 
3461 	mask = amdgpu_gfx_create_bitmask(adev->gfx.config.max_backends_per_se /
3462 					 adev->gfx.config.max_sh_per_se);
3463 
3464 	return (~data) & mask;
3465 }
3466 
3467 static void
3468 gfx_v8_0_raster_config(struct amdgpu_device *adev, u32 *rconf, u32 *rconf1)
3469 {
3470 	switch (adev->asic_type) {
3471 	case CHIP_FIJI:
3472 	case CHIP_VEGAM:
3473 		*rconf |= RB_MAP_PKR0(2) | RB_MAP_PKR1(2) |
3474 			  RB_XSEL2(1) | PKR_MAP(2) |
3475 			  PKR_XSEL(1) | PKR_YSEL(1) |
3476 			  SE_MAP(2) | SE_XSEL(2) | SE_YSEL(3);
3477 		*rconf1 |= SE_PAIR_MAP(2) | SE_PAIR_XSEL(3) |
3478 			   SE_PAIR_YSEL(2);
3479 		break;
3480 	case CHIP_TONGA:
3481 	case CHIP_POLARIS10:
3482 		*rconf |= RB_MAP_PKR0(2) | RB_XSEL2(1) | SE_MAP(2) |
3483 			  SE_XSEL(1) | SE_YSEL(1);
3484 		*rconf1 |= SE_PAIR_MAP(2) | SE_PAIR_XSEL(2) |
3485 			   SE_PAIR_YSEL(2);
3486 		break;
3487 	case CHIP_TOPAZ:
3488 	case CHIP_CARRIZO:
3489 		*rconf |= RB_MAP_PKR0(2);
3490 		*rconf1 |= 0x0;
3491 		break;
3492 	case CHIP_POLARIS11:
3493 	case CHIP_POLARIS12:
3494 		*rconf |= RB_MAP_PKR0(2) | RB_XSEL2(1) | SE_MAP(2) |
3495 			  SE_XSEL(1) | SE_YSEL(1);
3496 		*rconf1 |= 0x0;
3497 		break;
3498 	case CHIP_STONEY:
3499 		*rconf |= 0x0;
3500 		*rconf1 |= 0x0;
3501 		break;
3502 	default:
3503 		DRM_ERROR("unknown asic: 0x%x\n", adev->asic_type);
3504 		break;
3505 	}
3506 }
3507 
3508 static void
3509 gfx_v8_0_write_harvested_raster_configs(struct amdgpu_device *adev,
3510 					u32 raster_config, u32 raster_config_1,
3511 					unsigned rb_mask, unsigned num_rb)
3512 {
3513 	unsigned sh_per_se = max_t(unsigned, adev->gfx.config.max_sh_per_se, 1);
3514 	unsigned num_se = max_t(unsigned, adev->gfx.config.max_shader_engines, 1);
3515 	unsigned rb_per_pkr = min_t(unsigned, num_rb / num_se / sh_per_se, 2);
3516 	unsigned rb_per_se = num_rb / num_se;
3517 	unsigned se_mask[4];
3518 	unsigned se;
3519 
3520 	se_mask[0] = ((1 << rb_per_se) - 1) & rb_mask;
3521 	se_mask[1] = (se_mask[0] << rb_per_se) & rb_mask;
3522 	se_mask[2] = (se_mask[1] << rb_per_se) & rb_mask;
3523 	se_mask[3] = (se_mask[2] << rb_per_se) & rb_mask;
3524 
3525 	WARN_ON(!(num_se == 1 || num_se == 2 || num_se == 4));
3526 	WARN_ON(!(sh_per_se == 1 || sh_per_se == 2));
3527 	WARN_ON(!(rb_per_pkr == 1 || rb_per_pkr == 2));
3528 
3529 	if ((num_se > 2) && ((!se_mask[0] && !se_mask[1]) ||
3530 			     (!se_mask[2] && !se_mask[3]))) {
3531 		raster_config_1 &= ~SE_PAIR_MAP_MASK;
3532 
3533 		if (!se_mask[0] && !se_mask[1]) {
3534 			raster_config_1 |=
3535 				SE_PAIR_MAP(RASTER_CONFIG_SE_PAIR_MAP_3);
3536 		} else {
3537 			raster_config_1 |=
3538 				SE_PAIR_MAP(RASTER_CONFIG_SE_PAIR_MAP_0);
3539 		}
3540 	}
3541 
3542 	for (se = 0; se < num_se; se++) {
3543 		unsigned raster_config_se = raster_config;
3544 		unsigned pkr0_mask = ((1 << rb_per_pkr) - 1) << (se * rb_per_se);
3545 		unsigned pkr1_mask = pkr0_mask << rb_per_pkr;
3546 		int idx = (se / 2) * 2;
3547 
3548 		if ((num_se > 1) && (!se_mask[idx] || !se_mask[idx + 1])) {
3549 			raster_config_se &= ~SE_MAP_MASK;
3550 
3551 			if (!se_mask[idx]) {
3552 				raster_config_se |= SE_MAP(RASTER_CONFIG_SE_MAP_3);
3553 			} else {
3554 				raster_config_se |= SE_MAP(RASTER_CONFIG_SE_MAP_0);
3555 			}
3556 		}
3557 
3558 		pkr0_mask &= rb_mask;
3559 		pkr1_mask &= rb_mask;
3560 		if (rb_per_se > 2 && (!pkr0_mask || !pkr1_mask)) {
3561 			raster_config_se &= ~PKR_MAP_MASK;
3562 
3563 			if (!pkr0_mask) {
3564 				raster_config_se |= PKR_MAP(RASTER_CONFIG_PKR_MAP_3);
3565 			} else {
3566 				raster_config_se |= PKR_MAP(RASTER_CONFIG_PKR_MAP_0);
3567 			}
3568 		}
3569 
3570 		if (rb_per_se >= 2) {
3571 			unsigned rb0_mask = 1 << (se * rb_per_se);
3572 			unsigned rb1_mask = rb0_mask << 1;
3573 
3574 			rb0_mask &= rb_mask;
3575 			rb1_mask &= rb_mask;
3576 			if (!rb0_mask || !rb1_mask) {
3577 				raster_config_se &= ~RB_MAP_PKR0_MASK;
3578 
3579 				if (!rb0_mask) {
3580 					raster_config_se |=
3581 						RB_MAP_PKR0(RASTER_CONFIG_RB_MAP_3);
3582 				} else {
3583 					raster_config_se |=
3584 						RB_MAP_PKR0(RASTER_CONFIG_RB_MAP_0);
3585 				}
3586 			}
3587 
3588 			if (rb_per_se > 2) {
3589 				rb0_mask = 1 << (se * rb_per_se + rb_per_pkr);
3590 				rb1_mask = rb0_mask << 1;
3591 				rb0_mask &= rb_mask;
3592 				rb1_mask &= rb_mask;
3593 				if (!rb0_mask || !rb1_mask) {
3594 					raster_config_se &= ~RB_MAP_PKR1_MASK;
3595 
3596 					if (!rb0_mask) {
3597 						raster_config_se |=
3598 							RB_MAP_PKR1(RASTER_CONFIG_RB_MAP_3);
3599 					} else {
3600 						raster_config_se |=
3601 							RB_MAP_PKR1(RASTER_CONFIG_RB_MAP_0);
3602 					}
3603 				}
3604 			}
3605 		}
3606 
3607 		/* GRBM_GFX_INDEX has a different offset on VI */
3608 		gfx_v8_0_select_se_sh(adev, se, 0xffffffff, 0xffffffff);
3609 		WREG32(mmPA_SC_RASTER_CONFIG, raster_config_se);
3610 		WREG32(mmPA_SC_RASTER_CONFIG_1, raster_config_1);
3611 	}
3612 
3613 	/* GRBM_GFX_INDEX has a different offset on VI */
3614 	gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
3615 }
3616 
3617 static void gfx_v8_0_setup_rb(struct amdgpu_device *adev)
3618 {
3619 	int i, j;
3620 	u32 data;
3621 	u32 raster_config = 0, raster_config_1 = 0;
3622 	u32 active_rbs = 0;
3623 	u32 rb_bitmap_width_per_sh = adev->gfx.config.max_backends_per_se /
3624 					adev->gfx.config.max_sh_per_se;
3625 	unsigned num_rb_pipes;
3626 
3627 	mutex_lock(&adev->grbm_idx_mutex);
3628 	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
3629 		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
3630 			gfx_v8_0_select_se_sh(adev, i, j, 0xffffffff);
3631 			data = gfx_v8_0_get_rb_active_bitmap(adev);
3632 			active_rbs |= data << ((i * adev->gfx.config.max_sh_per_se + j) *
3633 					       rb_bitmap_width_per_sh);
3634 		}
3635 	}
3636 	gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
3637 
3638 	adev->gfx.config.backend_enable_mask = active_rbs;
3639 	adev->gfx.config.num_rbs = hweight32(active_rbs);
3640 
3641 	num_rb_pipes = min_t(unsigned, adev->gfx.config.max_backends_per_se *
3642 			     adev->gfx.config.max_shader_engines, 16);
3643 
3644 	gfx_v8_0_raster_config(adev, &raster_config, &raster_config_1);
3645 
3646 	if (!adev->gfx.config.backend_enable_mask ||
3647 			adev->gfx.config.num_rbs >= num_rb_pipes) {
3648 		WREG32(mmPA_SC_RASTER_CONFIG, raster_config);
3649 		WREG32(mmPA_SC_RASTER_CONFIG_1, raster_config_1);
3650 	} else {
3651 		gfx_v8_0_write_harvested_raster_configs(adev, raster_config, raster_config_1,
3652 							adev->gfx.config.backend_enable_mask,
3653 							num_rb_pipes);
3654 	}
3655 
3656 	/* cache the values for userspace */
3657 	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
3658 		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
3659 			gfx_v8_0_select_se_sh(adev, i, j, 0xffffffff);
3660 			adev->gfx.config.rb_config[i][j].rb_backend_disable =
3661 				RREG32(mmCC_RB_BACKEND_DISABLE);
3662 			adev->gfx.config.rb_config[i][j].user_rb_backend_disable =
3663 				RREG32(mmGC_USER_RB_BACKEND_DISABLE);
3664 			adev->gfx.config.rb_config[i][j].raster_config =
3665 				RREG32(mmPA_SC_RASTER_CONFIG);
3666 			adev->gfx.config.rb_config[i][j].raster_config_1 =
3667 				RREG32(mmPA_SC_RASTER_CONFIG_1);
3668 		}
3669 	}
3670 	gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
3671 	mutex_unlock(&adev->grbm_idx_mutex);
3672 }
3673 
3674 #define DEFAULT_SH_MEM_BASES	(0x6000)
3675 /**
3676  * gfx_v8_0_init_compute_vmid - gart enable
3677  *
3678  * @adev: amdgpu_device pointer
3679  *
3680  * Initialize compute vmid sh_mem registers
3681  *
3682  */
3683 static void gfx_v8_0_init_compute_vmid(struct amdgpu_device *adev)
3684 {
3685 	int i;
3686 	uint32_t sh_mem_config;
3687 	uint32_t sh_mem_bases;
3688 
3689 	/*
3690 	 * Configure apertures:
3691 	 * LDS:         0x60000000'00000000 - 0x60000001'00000000 (4GB)
3692 	 * Scratch:     0x60000001'00000000 - 0x60000002'00000000 (4GB)
3693 	 * GPUVM:       0x60010000'00000000 - 0x60020000'00000000 (1TB)
3694 	 */
3695 	sh_mem_bases = DEFAULT_SH_MEM_BASES | (DEFAULT_SH_MEM_BASES << 16);
3696 
3697 	sh_mem_config = SH_MEM_ADDRESS_MODE_HSA64 <<
3698 			SH_MEM_CONFIG__ADDRESS_MODE__SHIFT |
3699 			SH_MEM_ALIGNMENT_MODE_UNALIGNED <<
3700 			SH_MEM_CONFIG__ALIGNMENT_MODE__SHIFT |
3701 			MTYPE_CC << SH_MEM_CONFIG__DEFAULT_MTYPE__SHIFT |
3702 			SH_MEM_CONFIG__PRIVATE_ATC_MASK;
3703 
3704 	mutex_lock(&adev->srbm_mutex);
3705 	for (i = adev->vm_manager.first_kfd_vmid; i < AMDGPU_NUM_VMID; i++) {
3706 		vi_srbm_select(adev, 0, 0, 0, i);
3707 		/* CP and shaders */
3708 		WREG32(mmSH_MEM_CONFIG, sh_mem_config);
3709 		WREG32(mmSH_MEM_APE1_BASE, 1);
3710 		WREG32(mmSH_MEM_APE1_LIMIT, 0);
3711 		WREG32(mmSH_MEM_BASES, sh_mem_bases);
3712 	}
3713 	vi_srbm_select(adev, 0, 0, 0, 0);
3714 	mutex_unlock(&adev->srbm_mutex);
3715 
3716 	/* Initialize all compute VMIDs to have no GDS, GWS, or OA
3717 	   access. These should be enabled by FW for target VMIDs. */
3718 	for (i = adev->vm_manager.first_kfd_vmid; i < AMDGPU_NUM_VMID; i++) {
3719 		WREG32(amdgpu_gds_reg_offset[i].mem_base, 0);
3720 		WREG32(amdgpu_gds_reg_offset[i].mem_size, 0);
3721 		WREG32(amdgpu_gds_reg_offset[i].gws, 0);
3722 		WREG32(amdgpu_gds_reg_offset[i].oa, 0);
3723 	}
3724 }
3725 
3726 static void gfx_v8_0_init_gds_vmid(struct amdgpu_device *adev)
3727 {
3728 	int vmid;
3729 
3730 	/*
3731 	 * Initialize all compute and user-gfx VMIDs to have no GDS, GWS, or OA
3732 	 * access. Compute VMIDs should be enabled by FW for target VMIDs,
3733 	 * the driver can enable them for graphics. VMID0 should maintain
3734 	 * access so that HWS firmware can save/restore entries.
3735 	 */
3736 	for (vmid = 1; vmid < AMDGPU_NUM_VMID; vmid++) {
3737 		WREG32(amdgpu_gds_reg_offset[vmid].mem_base, 0);
3738 		WREG32(amdgpu_gds_reg_offset[vmid].mem_size, 0);
3739 		WREG32(amdgpu_gds_reg_offset[vmid].gws, 0);
3740 		WREG32(amdgpu_gds_reg_offset[vmid].oa, 0);
3741 	}
3742 }
3743 
3744 static void gfx_v8_0_config_init(struct amdgpu_device *adev)
3745 {
3746 	switch (adev->asic_type) {
3747 	default:
3748 		adev->gfx.config.double_offchip_lds_buf = 1;
3749 		break;
3750 	case CHIP_CARRIZO:
3751 	case CHIP_STONEY:
3752 		adev->gfx.config.double_offchip_lds_buf = 0;
3753 		break;
3754 	}
3755 }
3756 
3757 static void gfx_v8_0_constants_init(struct amdgpu_device *adev)
3758 {
3759 	u32 tmp, sh_static_mem_cfg;
3760 	int i;
3761 
3762 	WREG32_FIELD(GRBM_CNTL, READ_TIMEOUT, 0xFF);
3763 	WREG32(mmGB_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
3764 	WREG32(mmHDP_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
3765 	WREG32(mmDMIF_ADDR_CALC, adev->gfx.config.gb_addr_config);
3766 
3767 	gfx_v8_0_tiling_mode_table_init(adev);
3768 	gfx_v8_0_setup_rb(adev);
3769 	gfx_v8_0_get_cu_info(adev);
3770 	gfx_v8_0_config_init(adev);
3771 
3772 	/* XXX SH_MEM regs */
3773 	/* where to put LDS, scratch, GPUVM in FSA64 space */
3774 	sh_static_mem_cfg = REG_SET_FIELD(0, SH_STATIC_MEM_CONFIG,
3775 				   SWIZZLE_ENABLE, 1);
3776 	sh_static_mem_cfg = REG_SET_FIELD(sh_static_mem_cfg, SH_STATIC_MEM_CONFIG,
3777 				   ELEMENT_SIZE, 1);
3778 	sh_static_mem_cfg = REG_SET_FIELD(sh_static_mem_cfg, SH_STATIC_MEM_CONFIG,
3779 				   INDEX_STRIDE, 3);
3780 	WREG32(mmSH_STATIC_MEM_CONFIG, sh_static_mem_cfg);
3781 
3782 	mutex_lock(&adev->srbm_mutex);
3783 	for (i = 0; i < adev->vm_manager.id_mgr[0].num_ids; i++) {
3784 		vi_srbm_select(adev, 0, 0, 0, i);
3785 		/* CP and shaders */
3786 		if (i == 0) {
3787 			tmp = REG_SET_FIELD(0, SH_MEM_CONFIG, DEFAULT_MTYPE, MTYPE_UC);
3788 			tmp = REG_SET_FIELD(tmp, SH_MEM_CONFIG, APE1_MTYPE, MTYPE_UC);
3789 			tmp = REG_SET_FIELD(tmp, SH_MEM_CONFIG, ALIGNMENT_MODE,
3790 					    SH_MEM_ALIGNMENT_MODE_UNALIGNED);
3791 			WREG32(mmSH_MEM_CONFIG, tmp);
3792 			WREG32(mmSH_MEM_BASES, 0);
3793 		} else {
3794 			tmp = REG_SET_FIELD(0, SH_MEM_CONFIG, DEFAULT_MTYPE, MTYPE_NC);
3795 			tmp = REG_SET_FIELD(tmp, SH_MEM_CONFIG, APE1_MTYPE, MTYPE_UC);
3796 			tmp = REG_SET_FIELD(tmp, SH_MEM_CONFIG, ALIGNMENT_MODE,
3797 					    SH_MEM_ALIGNMENT_MODE_UNALIGNED);
3798 			WREG32(mmSH_MEM_CONFIG, tmp);
3799 			tmp = adev->gmc.shared_aperture_start >> 48;
3800 			WREG32(mmSH_MEM_BASES, tmp);
3801 		}
3802 
3803 		WREG32(mmSH_MEM_APE1_BASE, 1);
3804 		WREG32(mmSH_MEM_APE1_LIMIT, 0);
3805 	}
3806 	vi_srbm_select(adev, 0, 0, 0, 0);
3807 	mutex_unlock(&adev->srbm_mutex);
3808 
3809 	gfx_v8_0_init_compute_vmid(adev);
3810 	gfx_v8_0_init_gds_vmid(adev);
3811 
3812 	mutex_lock(&adev->grbm_idx_mutex);
3813 	/*
3814 	 * making sure that the following register writes will be broadcasted
3815 	 * to all the shaders
3816 	 */
3817 	gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
3818 
3819 	WREG32(mmPA_SC_FIFO_SIZE,
3820 		   (adev->gfx.config.sc_prim_fifo_size_frontend <<
3821 			PA_SC_FIFO_SIZE__SC_FRONTEND_PRIM_FIFO_SIZE__SHIFT) |
3822 		   (adev->gfx.config.sc_prim_fifo_size_backend <<
3823 			PA_SC_FIFO_SIZE__SC_BACKEND_PRIM_FIFO_SIZE__SHIFT) |
3824 		   (adev->gfx.config.sc_hiz_tile_fifo_size <<
3825 			PA_SC_FIFO_SIZE__SC_HIZ_TILE_FIFO_SIZE__SHIFT) |
3826 		   (adev->gfx.config.sc_earlyz_tile_fifo_size <<
3827 			PA_SC_FIFO_SIZE__SC_EARLYZ_TILE_FIFO_SIZE__SHIFT));
3828 
3829 	tmp = RREG32(mmSPI_ARB_PRIORITY);
3830 	tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS0, 2);
3831 	tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS1, 2);
3832 	tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS2, 2);
3833 	tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS3, 2);
3834 	WREG32(mmSPI_ARB_PRIORITY, tmp);
3835 
3836 	mutex_unlock(&adev->grbm_idx_mutex);
3837 
3838 }
3839 
3840 static void gfx_v8_0_wait_for_rlc_serdes(struct amdgpu_device *adev)
3841 {
3842 	u32 i, j, k;
3843 	u32 mask;
3844 
3845 	mutex_lock(&adev->grbm_idx_mutex);
3846 	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
3847 		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
3848 			gfx_v8_0_select_se_sh(adev, i, j, 0xffffffff);
3849 			for (k = 0; k < adev->usec_timeout; k++) {
3850 				if (RREG32(mmRLC_SERDES_CU_MASTER_BUSY) == 0)
3851 					break;
3852 				udelay(1);
3853 			}
3854 			if (k == adev->usec_timeout) {
3855 				gfx_v8_0_select_se_sh(adev, 0xffffffff,
3856 						      0xffffffff, 0xffffffff);
3857 				mutex_unlock(&adev->grbm_idx_mutex);
3858 				DRM_INFO("Timeout wait for RLC serdes %u,%u\n",
3859 					 i, j);
3860 				return;
3861 			}
3862 		}
3863 	}
3864 	gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
3865 	mutex_unlock(&adev->grbm_idx_mutex);
3866 
3867 	mask = RLC_SERDES_NONCU_MASTER_BUSY__SE_MASTER_BUSY_MASK |
3868 		RLC_SERDES_NONCU_MASTER_BUSY__GC_MASTER_BUSY_MASK |
3869 		RLC_SERDES_NONCU_MASTER_BUSY__TC0_MASTER_BUSY_MASK |
3870 		RLC_SERDES_NONCU_MASTER_BUSY__TC1_MASTER_BUSY_MASK;
3871 	for (k = 0; k < adev->usec_timeout; k++) {
3872 		if ((RREG32(mmRLC_SERDES_NONCU_MASTER_BUSY) & mask) == 0)
3873 			break;
3874 		udelay(1);
3875 	}
3876 }
3877 
3878 static void gfx_v8_0_enable_gui_idle_interrupt(struct amdgpu_device *adev,
3879 					       bool enable)
3880 {
3881 	u32 tmp = RREG32(mmCP_INT_CNTL_RING0);
3882 
3883 	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CNTX_BUSY_INT_ENABLE, enable ? 1 : 0);
3884 	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CNTX_EMPTY_INT_ENABLE, enable ? 1 : 0);
3885 	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CMP_BUSY_INT_ENABLE, enable ? 1 : 0);
3886 	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, GFX_IDLE_INT_ENABLE, enable ? 1 : 0);
3887 
3888 	WREG32(mmCP_INT_CNTL_RING0, tmp);
3889 }
3890 
3891 static void gfx_v8_0_init_csb(struct amdgpu_device *adev)
3892 {
3893 	adev->gfx.rlc.funcs->get_csb_buffer(adev, adev->gfx.rlc.cs_ptr);
3894 	/* csib */
3895 	WREG32(mmRLC_CSIB_ADDR_HI,
3896 			adev->gfx.rlc.clear_state_gpu_addr >> 32);
3897 	WREG32(mmRLC_CSIB_ADDR_LO,
3898 			adev->gfx.rlc.clear_state_gpu_addr & 0xfffffffc);
3899 	WREG32(mmRLC_CSIB_LENGTH,
3900 			adev->gfx.rlc.clear_state_size);
3901 }
3902 
3903 static void gfx_v8_0_parse_ind_reg_list(int *register_list_format,
3904 				int ind_offset,
3905 				int list_size,
3906 				int *unique_indices,
3907 				int *indices_count,
3908 				int max_indices,
3909 				int *ind_start_offsets,
3910 				int *offset_count,
3911 				int max_offset)
3912 {
3913 	int indices;
3914 	bool new_entry = true;
3915 
3916 	for (; ind_offset < list_size; ind_offset++) {
3917 
3918 		if (new_entry) {
3919 			new_entry = false;
3920 			ind_start_offsets[*offset_count] = ind_offset;
3921 			*offset_count = *offset_count + 1;
3922 			BUG_ON(*offset_count >= max_offset);
3923 		}
3924 
3925 		if (register_list_format[ind_offset] == 0xFFFFFFFF) {
3926 			new_entry = true;
3927 			continue;
3928 		}
3929 
3930 		ind_offset += 2;
3931 
3932 		/* look for the matching indice */
3933 		for (indices = 0;
3934 			indices < *indices_count;
3935 			indices++) {
3936 			if (unique_indices[indices] ==
3937 				register_list_format[ind_offset])
3938 				break;
3939 		}
3940 
3941 		if (indices >= *indices_count) {
3942 			unique_indices[*indices_count] =
3943 				register_list_format[ind_offset];
3944 			indices = *indices_count;
3945 			*indices_count = *indices_count + 1;
3946 			BUG_ON(*indices_count >= max_indices);
3947 		}
3948 
3949 		register_list_format[ind_offset] = indices;
3950 	}
3951 }
3952 
3953 static int gfx_v8_0_init_save_restore_list(struct amdgpu_device *adev)
3954 {
3955 	int i, temp, data;
3956 	int unique_indices[] = {0, 0, 0, 0, 0, 0, 0, 0};
3957 	int indices_count = 0;
3958 	int indirect_start_offsets[] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0};
3959 	int offset_count = 0;
3960 
3961 	int list_size;
3962 	unsigned int *register_list_format =
3963 		kmemdup(adev->gfx.rlc.register_list_format,
3964 			adev->gfx.rlc.reg_list_format_size_bytes, GFP_KERNEL);
3965 	if (!register_list_format)
3966 		return -ENOMEM;
3967 
3968 	gfx_v8_0_parse_ind_reg_list(register_list_format,
3969 				RLC_FormatDirectRegListLength,
3970 				adev->gfx.rlc.reg_list_format_size_bytes >> 2,
3971 				unique_indices,
3972 				&indices_count,
3973 				ARRAY_SIZE(unique_indices),
3974 				indirect_start_offsets,
3975 				&offset_count,
3976 				ARRAY_SIZE(indirect_start_offsets));
3977 
3978 	/* save and restore list */
3979 	WREG32_FIELD(RLC_SRM_CNTL, AUTO_INCR_ADDR, 1);
3980 
3981 	WREG32(mmRLC_SRM_ARAM_ADDR, 0);
3982 	for (i = 0; i < adev->gfx.rlc.reg_list_size_bytes >> 2; i++)
3983 		WREG32(mmRLC_SRM_ARAM_DATA, adev->gfx.rlc.register_restore[i]);
3984 
3985 	/* indirect list */
3986 	WREG32(mmRLC_GPM_SCRATCH_ADDR, adev->gfx.rlc.reg_list_format_start);
3987 	for (i = 0; i < adev->gfx.rlc.reg_list_format_size_bytes >> 2; i++)
3988 		WREG32(mmRLC_GPM_SCRATCH_DATA, register_list_format[i]);
3989 
3990 	list_size = adev->gfx.rlc.reg_list_size_bytes >> 2;
3991 	list_size = list_size >> 1;
3992 	WREG32(mmRLC_GPM_SCRATCH_ADDR, adev->gfx.rlc.reg_restore_list_size);
3993 	WREG32(mmRLC_GPM_SCRATCH_DATA, list_size);
3994 
3995 	/* starting offsets starts */
3996 	WREG32(mmRLC_GPM_SCRATCH_ADDR,
3997 		adev->gfx.rlc.starting_offsets_start);
3998 	for (i = 0; i < ARRAY_SIZE(indirect_start_offsets); i++)
3999 		WREG32(mmRLC_GPM_SCRATCH_DATA,
4000 				indirect_start_offsets[i]);
4001 
4002 	/* unique indices */
4003 	temp = mmRLC_SRM_INDEX_CNTL_ADDR_0;
4004 	data = mmRLC_SRM_INDEX_CNTL_DATA_0;
4005 	for (i = 0; i < ARRAY_SIZE(unique_indices); i++) {
4006 		if (unique_indices[i] != 0) {
4007 			WREG32(temp + i, unique_indices[i] & 0x3FFFF);
4008 			WREG32(data + i, unique_indices[i] >> 20);
4009 		}
4010 	}
4011 	kfree(register_list_format);
4012 
4013 	return 0;
4014 }
4015 
4016 static void gfx_v8_0_enable_save_restore_machine(struct amdgpu_device *adev)
4017 {
4018 	WREG32_FIELD(RLC_SRM_CNTL, SRM_ENABLE, 1);
4019 }
4020 
4021 static void gfx_v8_0_init_power_gating(struct amdgpu_device *adev)
4022 {
4023 	uint32_t data;
4024 
4025 	WREG32_FIELD(CP_RB_WPTR_POLL_CNTL, IDLE_POLL_COUNT, 0x60);
4026 
4027 	data = REG_SET_FIELD(0, RLC_PG_DELAY, POWER_UP_DELAY, 0x10);
4028 	data = REG_SET_FIELD(data, RLC_PG_DELAY, POWER_DOWN_DELAY, 0x10);
4029 	data = REG_SET_FIELD(data, RLC_PG_DELAY, CMD_PROPAGATE_DELAY, 0x10);
4030 	data = REG_SET_FIELD(data, RLC_PG_DELAY, MEM_SLEEP_DELAY, 0x10);
4031 	WREG32(mmRLC_PG_DELAY, data);
4032 
4033 	WREG32_FIELD(RLC_PG_DELAY_2, SERDES_CMD_DELAY, 0x3);
4034 	WREG32_FIELD(RLC_AUTO_PG_CTRL, GRBM_REG_SAVE_GFX_IDLE_THRESHOLD, 0x55f0);
4035 
4036 }
4037 
4038 static void cz_enable_sck_slow_down_on_power_up(struct amdgpu_device *adev,
4039 						bool enable)
4040 {
4041 	WREG32_FIELD(RLC_PG_CNTL, SMU_CLK_SLOWDOWN_ON_PU_ENABLE, enable ? 1 : 0);
4042 }
4043 
4044 static void cz_enable_sck_slow_down_on_power_down(struct amdgpu_device *adev,
4045 						  bool enable)
4046 {
4047 	WREG32_FIELD(RLC_PG_CNTL, SMU_CLK_SLOWDOWN_ON_PD_ENABLE, enable ? 1 : 0);
4048 }
4049 
4050 static void cz_enable_cp_power_gating(struct amdgpu_device *adev, bool enable)
4051 {
4052 	WREG32_FIELD(RLC_PG_CNTL, CP_PG_DISABLE, enable ? 0 : 1);
4053 }
4054 
4055 static void gfx_v8_0_init_pg(struct amdgpu_device *adev)
4056 {
4057 	if ((adev->asic_type == CHIP_CARRIZO) ||
4058 	    (adev->asic_type == CHIP_STONEY)) {
4059 		gfx_v8_0_init_csb(adev);
4060 		gfx_v8_0_init_save_restore_list(adev);
4061 		gfx_v8_0_enable_save_restore_machine(adev);
4062 		WREG32(mmRLC_JUMP_TABLE_RESTORE, adev->gfx.rlc.cp_table_gpu_addr >> 8);
4063 		gfx_v8_0_init_power_gating(adev);
4064 		WREG32(mmRLC_PG_ALWAYS_ON_CU_MASK, adev->gfx.cu_info.ao_cu_mask);
4065 	} else if ((adev->asic_type == CHIP_POLARIS11) ||
4066 		   (adev->asic_type == CHIP_POLARIS12) ||
4067 		   (adev->asic_type == CHIP_VEGAM)) {
4068 		gfx_v8_0_init_csb(adev);
4069 		gfx_v8_0_init_save_restore_list(adev);
4070 		gfx_v8_0_enable_save_restore_machine(adev);
4071 		gfx_v8_0_init_power_gating(adev);
4072 	}
4073 
4074 }
4075 
4076 static void gfx_v8_0_rlc_stop(struct amdgpu_device *adev)
4077 {
4078 	WREG32_FIELD(RLC_CNTL, RLC_ENABLE_F32, 0);
4079 
4080 	gfx_v8_0_enable_gui_idle_interrupt(adev, false);
4081 	gfx_v8_0_wait_for_rlc_serdes(adev);
4082 }
4083 
4084 static void gfx_v8_0_rlc_reset(struct amdgpu_device *adev)
4085 {
4086 	WREG32_FIELD(GRBM_SOFT_RESET, SOFT_RESET_RLC, 1);
4087 	udelay(50);
4088 
4089 	WREG32_FIELD(GRBM_SOFT_RESET, SOFT_RESET_RLC, 0);
4090 	udelay(50);
4091 }
4092 
4093 static void gfx_v8_0_rlc_start(struct amdgpu_device *adev)
4094 {
4095 	WREG32_FIELD(RLC_CNTL, RLC_ENABLE_F32, 1);
4096 
4097 	/* carrizo do enable cp interrupt after cp inited */
4098 	if (!(adev->flags & AMD_IS_APU))
4099 		gfx_v8_0_enable_gui_idle_interrupt(adev, true);
4100 
4101 	udelay(50);
4102 }
4103 
4104 static int gfx_v8_0_rlc_resume(struct amdgpu_device *adev)
4105 {
4106 	if (amdgpu_sriov_vf(adev)) {
4107 		gfx_v8_0_init_csb(adev);
4108 		return 0;
4109 	}
4110 
4111 	adev->gfx.rlc.funcs->stop(adev);
4112 	adev->gfx.rlc.funcs->reset(adev);
4113 	gfx_v8_0_init_pg(adev);
4114 	adev->gfx.rlc.funcs->start(adev);
4115 
4116 	return 0;
4117 }
4118 
4119 static void gfx_v8_0_cp_gfx_enable(struct amdgpu_device *adev, bool enable)
4120 {
4121 	u32 tmp = RREG32(mmCP_ME_CNTL);
4122 
4123 	if (enable) {
4124 		tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, ME_HALT, 0);
4125 		tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, PFP_HALT, 0);
4126 		tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, CE_HALT, 0);
4127 	} else {
4128 		tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, ME_HALT, 1);
4129 		tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, PFP_HALT, 1);
4130 		tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, CE_HALT, 1);
4131 	}
4132 	WREG32(mmCP_ME_CNTL, tmp);
4133 	udelay(50);
4134 }
4135 
4136 static u32 gfx_v8_0_get_csb_size(struct amdgpu_device *adev)
4137 {
4138 	u32 count = 0;
4139 	const struct cs_section_def *sect = NULL;
4140 	const struct cs_extent_def *ext = NULL;
4141 
4142 	/* begin clear state */
4143 	count += 2;
4144 	/* context control state */
4145 	count += 3;
4146 
4147 	for (sect = vi_cs_data; sect->section != NULL; ++sect) {
4148 		for (ext = sect->section; ext->extent != NULL; ++ext) {
4149 			if (sect->id == SECT_CONTEXT)
4150 				count += 2 + ext->reg_count;
4151 			else
4152 				return 0;
4153 		}
4154 	}
4155 	/* pa_sc_raster_config/pa_sc_raster_config1 */
4156 	count += 4;
4157 	/* end clear state */
4158 	count += 2;
4159 	/* clear state */
4160 	count += 2;
4161 
4162 	return count;
4163 }
4164 
4165 static int gfx_v8_0_cp_gfx_start(struct amdgpu_device *adev)
4166 {
4167 	struct amdgpu_ring *ring = &adev->gfx.gfx_ring[0];
4168 	const struct cs_section_def *sect = NULL;
4169 	const struct cs_extent_def *ext = NULL;
4170 	int r, i;
4171 
4172 	/* init the CP */
4173 	WREG32(mmCP_MAX_CONTEXT, adev->gfx.config.max_hw_contexts - 1);
4174 	WREG32(mmCP_ENDIAN_SWAP, 0);
4175 	WREG32(mmCP_DEVICE_ID, 1);
4176 
4177 	gfx_v8_0_cp_gfx_enable(adev, true);
4178 
4179 	r = amdgpu_ring_alloc(ring, gfx_v8_0_get_csb_size(adev) + 4);
4180 	if (r) {
4181 		DRM_ERROR("amdgpu: cp failed to lock ring (%d).\n", r);
4182 		return r;
4183 	}
4184 
4185 	/* clear state buffer */
4186 	amdgpu_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
4187 	amdgpu_ring_write(ring, PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
4188 
4189 	amdgpu_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1));
4190 	amdgpu_ring_write(ring, 0x80000000);
4191 	amdgpu_ring_write(ring, 0x80000000);
4192 
4193 	for (sect = vi_cs_data; sect->section != NULL; ++sect) {
4194 		for (ext = sect->section; ext->extent != NULL; ++ext) {
4195 			if (sect->id == SECT_CONTEXT) {
4196 				amdgpu_ring_write(ring,
4197 				       PACKET3(PACKET3_SET_CONTEXT_REG,
4198 					       ext->reg_count));
4199 				amdgpu_ring_write(ring,
4200 				       ext->reg_index - PACKET3_SET_CONTEXT_REG_START);
4201 				for (i = 0; i < ext->reg_count; i++)
4202 					amdgpu_ring_write(ring, ext->extent[i]);
4203 			}
4204 		}
4205 	}
4206 
4207 	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONTEXT_REG, 2));
4208 	amdgpu_ring_write(ring, mmPA_SC_RASTER_CONFIG - PACKET3_SET_CONTEXT_REG_START);
4209 	amdgpu_ring_write(ring, adev->gfx.config.rb_config[0][0].raster_config);
4210 	amdgpu_ring_write(ring, adev->gfx.config.rb_config[0][0].raster_config_1);
4211 
4212 	amdgpu_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
4213 	amdgpu_ring_write(ring, PACKET3_PREAMBLE_END_CLEAR_STATE);
4214 
4215 	amdgpu_ring_write(ring, PACKET3(PACKET3_CLEAR_STATE, 0));
4216 	amdgpu_ring_write(ring, 0);
4217 
4218 	/* init the CE partitions */
4219 	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_BASE, 2));
4220 	amdgpu_ring_write(ring, PACKET3_BASE_INDEX(CE_PARTITION_BASE));
4221 	amdgpu_ring_write(ring, 0x8000);
4222 	amdgpu_ring_write(ring, 0x8000);
4223 
4224 	amdgpu_ring_commit(ring);
4225 
4226 	return 0;
4227 }
4228 static void gfx_v8_0_set_cpg_door_bell(struct amdgpu_device *adev, struct amdgpu_ring *ring)
4229 {
4230 	u32 tmp;
4231 	/* no gfx doorbells on iceland */
4232 	if (adev->asic_type == CHIP_TOPAZ)
4233 		return;
4234 
4235 	tmp = RREG32(mmCP_RB_DOORBELL_CONTROL);
4236 
4237 	if (ring->use_doorbell) {
4238 		tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL,
4239 				DOORBELL_OFFSET, ring->doorbell_index);
4240 		tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL,
4241 						DOORBELL_HIT, 0);
4242 		tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL,
4243 					    DOORBELL_EN, 1);
4244 	} else {
4245 		tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL, DOORBELL_EN, 0);
4246 	}
4247 
4248 	WREG32(mmCP_RB_DOORBELL_CONTROL, tmp);
4249 
4250 	if (adev->flags & AMD_IS_APU)
4251 		return;
4252 
4253 	tmp = REG_SET_FIELD(0, CP_RB_DOORBELL_RANGE_LOWER,
4254 					DOORBELL_RANGE_LOWER,
4255 					adev->doorbell_index.gfx_ring0);
4256 	WREG32(mmCP_RB_DOORBELL_RANGE_LOWER, tmp);
4257 
4258 	WREG32(mmCP_RB_DOORBELL_RANGE_UPPER,
4259 		CP_RB_DOORBELL_RANGE_UPPER__DOORBELL_RANGE_UPPER_MASK);
4260 }
4261 
4262 static int gfx_v8_0_cp_gfx_resume(struct amdgpu_device *adev)
4263 {
4264 	struct amdgpu_ring *ring;
4265 	u32 tmp;
4266 	u32 rb_bufsz;
4267 	u64 rb_addr, rptr_addr, wptr_gpu_addr;
4268 
4269 	/* Set the write pointer delay */
4270 	WREG32(mmCP_RB_WPTR_DELAY, 0);
4271 
4272 	/* set the RB to use vmid 0 */
4273 	WREG32(mmCP_RB_VMID, 0);
4274 
4275 	/* Set ring buffer size */
4276 	ring = &adev->gfx.gfx_ring[0];
4277 	rb_bufsz = order_base_2(ring->ring_size / 8);
4278 	tmp = REG_SET_FIELD(0, CP_RB0_CNTL, RB_BUFSZ, rb_bufsz);
4279 	tmp = REG_SET_FIELD(tmp, CP_RB0_CNTL, RB_BLKSZ, rb_bufsz - 2);
4280 	tmp = REG_SET_FIELD(tmp, CP_RB0_CNTL, MTYPE, 3);
4281 	tmp = REG_SET_FIELD(tmp, CP_RB0_CNTL, MIN_IB_AVAILSZ, 1);
4282 #ifdef __BIG_ENDIAN
4283 	tmp = REG_SET_FIELD(tmp, CP_RB0_CNTL, BUF_SWAP, 1);
4284 #endif
4285 	WREG32(mmCP_RB0_CNTL, tmp);
4286 
4287 	/* Initialize the ring buffer's read and write pointers */
4288 	WREG32(mmCP_RB0_CNTL, tmp | CP_RB0_CNTL__RB_RPTR_WR_ENA_MASK);
4289 	ring->wptr = 0;
4290 	WREG32(mmCP_RB0_WPTR, lower_32_bits(ring->wptr));
4291 
4292 	/* set the wb address wether it's enabled or not */
4293 	rptr_addr = ring->rptr_gpu_addr;
4294 	WREG32(mmCP_RB0_RPTR_ADDR, lower_32_bits(rptr_addr));
4295 	WREG32(mmCP_RB0_RPTR_ADDR_HI, upper_32_bits(rptr_addr) & 0xFF);
4296 
4297 	wptr_gpu_addr = ring->wptr_gpu_addr;
4298 	WREG32(mmCP_RB_WPTR_POLL_ADDR_LO, lower_32_bits(wptr_gpu_addr));
4299 	WREG32(mmCP_RB_WPTR_POLL_ADDR_HI, upper_32_bits(wptr_gpu_addr));
4300 	mdelay(1);
4301 	WREG32(mmCP_RB0_CNTL, tmp);
4302 
4303 	rb_addr = ring->gpu_addr >> 8;
4304 	WREG32(mmCP_RB0_BASE, rb_addr);
4305 	WREG32(mmCP_RB0_BASE_HI, upper_32_bits(rb_addr));
4306 
4307 	gfx_v8_0_set_cpg_door_bell(adev, ring);
4308 	/* start the ring */
4309 	amdgpu_ring_clear_ring(ring);
4310 	gfx_v8_0_cp_gfx_start(adev);
4311 	ring->sched.ready = true;
4312 
4313 	return 0;
4314 }
4315 
4316 static void gfx_v8_0_cp_compute_enable(struct amdgpu_device *adev, bool enable)
4317 {
4318 	if (enable) {
4319 		WREG32(mmCP_MEC_CNTL, 0);
4320 	} else {
4321 		WREG32(mmCP_MEC_CNTL, (CP_MEC_CNTL__MEC_ME1_HALT_MASK | CP_MEC_CNTL__MEC_ME2_HALT_MASK));
4322 		adev->gfx.kiq.ring.sched.ready = false;
4323 	}
4324 	udelay(50);
4325 }
4326 
4327 /* KIQ functions */
4328 static void gfx_v8_0_kiq_setting(struct amdgpu_ring *ring)
4329 {
4330 	uint32_t tmp;
4331 	struct amdgpu_device *adev = ring->adev;
4332 
4333 	/* tell RLC which is KIQ queue */
4334 	tmp = RREG32(mmRLC_CP_SCHEDULERS);
4335 	tmp &= 0xffffff00;
4336 	tmp |= (ring->me << 5) | (ring->pipe << 3) | (ring->queue);
4337 	WREG32(mmRLC_CP_SCHEDULERS, tmp);
4338 	tmp |= 0x80;
4339 	WREG32(mmRLC_CP_SCHEDULERS, tmp);
4340 }
4341 
4342 static int gfx_v8_0_kiq_kcq_enable(struct amdgpu_device *adev)
4343 {
4344 	struct amdgpu_ring *kiq_ring = &adev->gfx.kiq.ring;
4345 	uint64_t queue_mask = 0;
4346 	int r, i;
4347 
4348 	for (i = 0; i < AMDGPU_MAX_COMPUTE_QUEUES; ++i) {
4349 		if (!test_bit(i, adev->gfx.mec.queue_bitmap))
4350 			continue;
4351 
4352 		/* This situation may be hit in the future if a new HW
4353 		 * generation exposes more than 64 queues. If so, the
4354 		 * definition of queue_mask needs updating */
4355 		if (WARN_ON(i >= (sizeof(queue_mask)*8))) {
4356 			DRM_ERROR("Invalid KCQ enabled: %d\n", i);
4357 			break;
4358 		}
4359 
4360 		queue_mask |= (1ull << i);
4361 	}
4362 
4363 	r = amdgpu_ring_alloc(kiq_ring, (8 * adev->gfx.num_compute_rings) + 8);
4364 	if (r) {
4365 		DRM_ERROR("Failed to lock KIQ (%d).\n", r);
4366 		return r;
4367 	}
4368 	/* set resources */
4369 	amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_SET_RESOURCES, 6));
4370 	amdgpu_ring_write(kiq_ring, 0);	/* vmid_mask:0 queue_type:0 (KIQ) */
4371 	amdgpu_ring_write(kiq_ring, lower_32_bits(queue_mask));	/* queue mask lo */
4372 	amdgpu_ring_write(kiq_ring, upper_32_bits(queue_mask));	/* queue mask hi */
4373 	amdgpu_ring_write(kiq_ring, 0);	/* gws mask lo */
4374 	amdgpu_ring_write(kiq_ring, 0);	/* gws mask hi */
4375 	amdgpu_ring_write(kiq_ring, 0);	/* oac mask */
4376 	amdgpu_ring_write(kiq_ring, 0);	/* gds heap base:0, gds heap size:0 */
4377 	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
4378 		struct amdgpu_ring *ring = &adev->gfx.compute_ring[i];
4379 		uint64_t mqd_addr = amdgpu_bo_gpu_offset(ring->mqd_obj);
4380 		uint64_t wptr_addr = ring->wptr_gpu_addr;
4381 
4382 		/* map queues */
4383 		amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_MAP_QUEUES, 5));
4384 		/* Q_sel:0, vmid:0, vidmem: 1, engine:0, num_Q:1*/
4385 		amdgpu_ring_write(kiq_ring,
4386 				  PACKET3_MAP_QUEUES_NUM_QUEUES(1));
4387 		amdgpu_ring_write(kiq_ring,
4388 				  PACKET3_MAP_QUEUES_DOORBELL_OFFSET(ring->doorbell_index) |
4389 				  PACKET3_MAP_QUEUES_QUEUE(ring->queue) |
4390 				  PACKET3_MAP_QUEUES_PIPE(ring->pipe) |
4391 				  PACKET3_MAP_QUEUES_ME(ring->me == 1 ? 0 : 1)); /* doorbell */
4392 		amdgpu_ring_write(kiq_ring, lower_32_bits(mqd_addr));
4393 		amdgpu_ring_write(kiq_ring, upper_32_bits(mqd_addr));
4394 		amdgpu_ring_write(kiq_ring, lower_32_bits(wptr_addr));
4395 		amdgpu_ring_write(kiq_ring, upper_32_bits(wptr_addr));
4396 	}
4397 
4398 	amdgpu_ring_commit(kiq_ring);
4399 
4400 	return 0;
4401 }
4402 
4403 static int gfx_v8_0_deactivate_hqd(struct amdgpu_device *adev, u32 req)
4404 {
4405 	int i, r = 0;
4406 
4407 	if (RREG32(mmCP_HQD_ACTIVE) & CP_HQD_ACTIVE__ACTIVE_MASK) {
4408 		WREG32_FIELD(CP_HQD_DEQUEUE_REQUEST, DEQUEUE_REQ, req);
4409 		for (i = 0; i < adev->usec_timeout; i++) {
4410 			if (!(RREG32(mmCP_HQD_ACTIVE) & CP_HQD_ACTIVE__ACTIVE_MASK))
4411 				break;
4412 			udelay(1);
4413 		}
4414 		if (i == adev->usec_timeout)
4415 			r = -ETIMEDOUT;
4416 	}
4417 	WREG32(mmCP_HQD_DEQUEUE_REQUEST, 0);
4418 	WREG32(mmCP_HQD_PQ_RPTR, 0);
4419 	WREG32(mmCP_HQD_PQ_WPTR, 0);
4420 
4421 	return r;
4422 }
4423 
4424 static void gfx_v8_0_mqd_set_priority(struct amdgpu_ring *ring, struct vi_mqd *mqd)
4425 {
4426 	struct amdgpu_device *adev = ring->adev;
4427 
4428 	if (ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE) {
4429 		if (amdgpu_gfx_is_high_priority_compute_queue(adev, ring)) {
4430 			mqd->cp_hqd_pipe_priority = AMDGPU_GFX_PIPE_PRIO_HIGH;
4431 			mqd->cp_hqd_queue_priority =
4432 				AMDGPU_GFX_QUEUE_PRIORITY_MAXIMUM;
4433 		}
4434 	}
4435 }
4436 
4437 static int gfx_v8_0_mqd_init(struct amdgpu_ring *ring)
4438 {
4439 	struct amdgpu_device *adev = ring->adev;
4440 	struct vi_mqd *mqd = ring->mqd_ptr;
4441 	uint64_t hqd_gpu_addr, wb_gpu_addr, eop_base_addr;
4442 	uint32_t tmp;
4443 
4444 	mqd->header = 0xC0310800;
4445 	mqd->compute_pipelinestat_enable = 0x00000001;
4446 	mqd->compute_static_thread_mgmt_se0 = 0xffffffff;
4447 	mqd->compute_static_thread_mgmt_se1 = 0xffffffff;
4448 	mqd->compute_static_thread_mgmt_se2 = 0xffffffff;
4449 	mqd->compute_static_thread_mgmt_se3 = 0xffffffff;
4450 	mqd->compute_misc_reserved = 0x00000003;
4451 	mqd->dynamic_cu_mask_addr_lo = lower_32_bits(ring->mqd_gpu_addr
4452 						     + offsetof(struct vi_mqd_allocation, dynamic_cu_mask));
4453 	mqd->dynamic_cu_mask_addr_hi = upper_32_bits(ring->mqd_gpu_addr
4454 						     + offsetof(struct vi_mqd_allocation, dynamic_cu_mask));
4455 	eop_base_addr = ring->eop_gpu_addr >> 8;
4456 	mqd->cp_hqd_eop_base_addr_lo = eop_base_addr;
4457 	mqd->cp_hqd_eop_base_addr_hi = upper_32_bits(eop_base_addr);
4458 
4459 	/* set the EOP size, register value is 2^(EOP_SIZE+1) dwords */
4460 	tmp = RREG32(mmCP_HQD_EOP_CONTROL);
4461 	tmp = REG_SET_FIELD(tmp, CP_HQD_EOP_CONTROL, EOP_SIZE,
4462 			(order_base_2(GFX8_MEC_HPD_SIZE / 4) - 1));
4463 
4464 	mqd->cp_hqd_eop_control = tmp;
4465 
4466 	/* enable doorbell? */
4467 	tmp = REG_SET_FIELD(RREG32(mmCP_HQD_PQ_DOORBELL_CONTROL),
4468 			    CP_HQD_PQ_DOORBELL_CONTROL,
4469 			    DOORBELL_EN,
4470 			    ring->use_doorbell ? 1 : 0);
4471 
4472 	mqd->cp_hqd_pq_doorbell_control = tmp;
4473 
4474 	/* set the pointer to the MQD */
4475 	mqd->cp_mqd_base_addr_lo = ring->mqd_gpu_addr & 0xfffffffc;
4476 	mqd->cp_mqd_base_addr_hi = upper_32_bits(ring->mqd_gpu_addr);
4477 
4478 	/* set MQD vmid to 0 */
4479 	tmp = RREG32(mmCP_MQD_CONTROL);
4480 	tmp = REG_SET_FIELD(tmp, CP_MQD_CONTROL, VMID, 0);
4481 	mqd->cp_mqd_control = tmp;
4482 
4483 	/* set the pointer to the HQD, this is similar CP_RB0_BASE/_HI */
4484 	hqd_gpu_addr = ring->gpu_addr >> 8;
4485 	mqd->cp_hqd_pq_base_lo = hqd_gpu_addr;
4486 	mqd->cp_hqd_pq_base_hi = upper_32_bits(hqd_gpu_addr);
4487 
4488 	/* set up the HQD, this is similar to CP_RB0_CNTL */
4489 	tmp = RREG32(mmCP_HQD_PQ_CONTROL);
4490 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, QUEUE_SIZE,
4491 			    (order_base_2(ring->ring_size / 4) - 1));
4492 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, RPTR_BLOCK_SIZE,
4493 			(order_base_2(AMDGPU_GPU_PAGE_SIZE / 4) - 1));
4494 #ifdef __BIG_ENDIAN
4495 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, ENDIAN_SWAP, 1);
4496 #endif
4497 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, UNORD_DISPATCH, 0);
4498 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, ROQ_PQ_IB_FLIP, 0);
4499 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, PRIV_STATE, 1);
4500 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, KMD_QUEUE, 1);
4501 	mqd->cp_hqd_pq_control = tmp;
4502 
4503 	/* set the wb address whether it's enabled or not */
4504 	wb_gpu_addr = ring->rptr_gpu_addr;
4505 	mqd->cp_hqd_pq_rptr_report_addr_lo = wb_gpu_addr & 0xfffffffc;
4506 	mqd->cp_hqd_pq_rptr_report_addr_hi =
4507 		upper_32_bits(wb_gpu_addr) & 0xffff;
4508 
4509 	/* only used if CP_PQ_WPTR_POLL_CNTL.CP_PQ_WPTR_POLL_CNTL__EN_MASK=1 */
4510 	wb_gpu_addr = ring->wptr_gpu_addr;
4511 	mqd->cp_hqd_pq_wptr_poll_addr_lo = wb_gpu_addr & 0xfffffffc;
4512 	mqd->cp_hqd_pq_wptr_poll_addr_hi = upper_32_bits(wb_gpu_addr) & 0xffff;
4513 
4514 	tmp = 0;
4515 	/* enable the doorbell if requested */
4516 	if (ring->use_doorbell) {
4517 		tmp = RREG32(mmCP_HQD_PQ_DOORBELL_CONTROL);
4518 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
4519 				DOORBELL_OFFSET, ring->doorbell_index);
4520 
4521 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
4522 					 DOORBELL_EN, 1);
4523 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
4524 					 DOORBELL_SOURCE, 0);
4525 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
4526 					 DOORBELL_HIT, 0);
4527 	}
4528 
4529 	mqd->cp_hqd_pq_doorbell_control = tmp;
4530 
4531 	/* reset read and write pointers, similar to CP_RB0_WPTR/_RPTR */
4532 	ring->wptr = 0;
4533 	mqd->cp_hqd_pq_wptr = ring->wptr;
4534 	mqd->cp_hqd_pq_rptr = RREG32(mmCP_HQD_PQ_RPTR);
4535 
4536 	/* set the vmid for the queue */
4537 	mqd->cp_hqd_vmid = 0;
4538 
4539 	tmp = RREG32(mmCP_HQD_PERSISTENT_STATE);
4540 	tmp = REG_SET_FIELD(tmp, CP_HQD_PERSISTENT_STATE, PRELOAD_SIZE, 0x53);
4541 	mqd->cp_hqd_persistent_state = tmp;
4542 
4543 	/* set MTYPE */
4544 	tmp = RREG32(mmCP_HQD_IB_CONTROL);
4545 	tmp = REG_SET_FIELD(tmp, CP_HQD_IB_CONTROL, MIN_IB_AVAIL_SIZE, 3);
4546 	tmp = REG_SET_FIELD(tmp, CP_HQD_IB_CONTROL, MTYPE, 3);
4547 	mqd->cp_hqd_ib_control = tmp;
4548 
4549 	tmp = RREG32(mmCP_HQD_IQ_TIMER);
4550 	tmp = REG_SET_FIELD(tmp, CP_HQD_IQ_TIMER, MTYPE, 3);
4551 	mqd->cp_hqd_iq_timer = tmp;
4552 
4553 	tmp = RREG32(mmCP_HQD_CTX_SAVE_CONTROL);
4554 	tmp = REG_SET_FIELD(tmp, CP_HQD_CTX_SAVE_CONTROL, MTYPE, 3);
4555 	mqd->cp_hqd_ctx_save_control = tmp;
4556 
4557 	/* defaults */
4558 	mqd->cp_hqd_eop_rptr = RREG32(mmCP_HQD_EOP_RPTR);
4559 	mqd->cp_hqd_eop_wptr = RREG32(mmCP_HQD_EOP_WPTR);
4560 	mqd->cp_hqd_ctx_save_base_addr_lo = RREG32(mmCP_HQD_CTX_SAVE_BASE_ADDR_LO);
4561 	mqd->cp_hqd_ctx_save_base_addr_hi = RREG32(mmCP_HQD_CTX_SAVE_BASE_ADDR_HI);
4562 	mqd->cp_hqd_cntl_stack_offset = RREG32(mmCP_HQD_CNTL_STACK_OFFSET);
4563 	mqd->cp_hqd_cntl_stack_size = RREG32(mmCP_HQD_CNTL_STACK_SIZE);
4564 	mqd->cp_hqd_wg_state_offset = RREG32(mmCP_HQD_WG_STATE_OFFSET);
4565 	mqd->cp_hqd_ctx_save_size = RREG32(mmCP_HQD_CTX_SAVE_SIZE);
4566 	mqd->cp_hqd_eop_done_events = RREG32(mmCP_HQD_EOP_EVENTS);
4567 	mqd->cp_hqd_error = RREG32(mmCP_HQD_ERROR);
4568 	mqd->cp_hqd_eop_wptr_mem = RREG32(mmCP_HQD_EOP_WPTR_MEM);
4569 	mqd->cp_hqd_eop_dones = RREG32(mmCP_HQD_EOP_DONES);
4570 
4571 	/* set static priority for a queue/ring */
4572 	gfx_v8_0_mqd_set_priority(ring, mqd);
4573 	mqd->cp_hqd_quantum = RREG32(mmCP_HQD_QUANTUM);
4574 
4575 	/* map_queues packet doesn't need activate the queue,
4576 	 * so only kiq need set this field.
4577 	 */
4578 	if (ring->funcs->type == AMDGPU_RING_TYPE_KIQ)
4579 		mqd->cp_hqd_active = 1;
4580 
4581 	return 0;
4582 }
4583 
4584 static int gfx_v8_0_mqd_commit(struct amdgpu_device *adev,
4585 			struct vi_mqd *mqd)
4586 {
4587 	uint32_t mqd_reg;
4588 	uint32_t *mqd_data;
4589 
4590 	/* HQD registers extend from mmCP_MQD_BASE_ADDR to mmCP_HQD_ERROR */
4591 	mqd_data = &mqd->cp_mqd_base_addr_lo;
4592 
4593 	/* disable wptr polling */
4594 	WREG32_FIELD(CP_PQ_WPTR_POLL_CNTL, EN, 0);
4595 
4596 	/* program all HQD registers */
4597 	for (mqd_reg = mmCP_HQD_VMID; mqd_reg <= mmCP_HQD_EOP_CONTROL; mqd_reg++)
4598 		WREG32(mqd_reg, mqd_data[mqd_reg - mmCP_MQD_BASE_ADDR]);
4599 
4600 	/* Tonga errata: EOP RPTR/WPTR should be left unmodified.
4601 	 * This is safe since EOP RPTR==WPTR for any inactive HQD
4602 	 * on ASICs that do not support context-save.
4603 	 * EOP writes/reads can start anywhere in the ring.
4604 	 */
4605 	if (adev->asic_type != CHIP_TONGA) {
4606 		WREG32(mmCP_HQD_EOP_RPTR, mqd->cp_hqd_eop_rptr);
4607 		WREG32(mmCP_HQD_EOP_WPTR, mqd->cp_hqd_eop_wptr);
4608 		WREG32(mmCP_HQD_EOP_WPTR_MEM, mqd->cp_hqd_eop_wptr_mem);
4609 	}
4610 
4611 	for (mqd_reg = mmCP_HQD_EOP_EVENTS; mqd_reg <= mmCP_HQD_ERROR; mqd_reg++)
4612 		WREG32(mqd_reg, mqd_data[mqd_reg - mmCP_MQD_BASE_ADDR]);
4613 
4614 	/* activate the HQD */
4615 	for (mqd_reg = mmCP_MQD_BASE_ADDR; mqd_reg <= mmCP_HQD_ACTIVE; mqd_reg++)
4616 		WREG32(mqd_reg, mqd_data[mqd_reg - mmCP_MQD_BASE_ADDR]);
4617 
4618 	return 0;
4619 }
4620 
4621 static int gfx_v8_0_kiq_init_queue(struct amdgpu_ring *ring)
4622 {
4623 	struct amdgpu_device *adev = ring->adev;
4624 	struct vi_mqd *mqd = ring->mqd_ptr;
4625 	int mqd_idx = AMDGPU_MAX_COMPUTE_RINGS;
4626 
4627 	gfx_v8_0_kiq_setting(ring);
4628 
4629 	if (amdgpu_in_reset(adev)) { /* for GPU_RESET case */
4630 		/* reset MQD to a clean status */
4631 		if (adev->gfx.mec.mqd_backup[mqd_idx])
4632 			memcpy(mqd, adev->gfx.mec.mqd_backup[mqd_idx], sizeof(struct vi_mqd_allocation));
4633 
4634 		/* reset ring buffer */
4635 		ring->wptr = 0;
4636 		amdgpu_ring_clear_ring(ring);
4637 		mutex_lock(&adev->srbm_mutex);
4638 		vi_srbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
4639 		gfx_v8_0_mqd_commit(adev, mqd);
4640 		vi_srbm_select(adev, 0, 0, 0, 0);
4641 		mutex_unlock(&adev->srbm_mutex);
4642 	} else {
4643 		memset((void *)mqd, 0, sizeof(struct vi_mqd_allocation));
4644 		((struct vi_mqd_allocation *)mqd)->dynamic_cu_mask = 0xFFFFFFFF;
4645 		((struct vi_mqd_allocation *)mqd)->dynamic_rb_mask = 0xFFFFFFFF;
4646 		if (amdgpu_sriov_vf(adev) && adev->in_suspend)
4647 			amdgpu_ring_clear_ring(ring);
4648 		mutex_lock(&adev->srbm_mutex);
4649 		vi_srbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
4650 		gfx_v8_0_mqd_init(ring);
4651 		gfx_v8_0_mqd_commit(adev, mqd);
4652 		vi_srbm_select(adev, 0, 0, 0, 0);
4653 		mutex_unlock(&adev->srbm_mutex);
4654 
4655 		if (adev->gfx.mec.mqd_backup[mqd_idx])
4656 			memcpy(adev->gfx.mec.mqd_backup[mqd_idx], mqd, sizeof(struct vi_mqd_allocation));
4657 	}
4658 
4659 	return 0;
4660 }
4661 
4662 static int gfx_v8_0_kcq_init_queue(struct amdgpu_ring *ring)
4663 {
4664 	struct amdgpu_device *adev = ring->adev;
4665 	struct vi_mqd *mqd = ring->mqd_ptr;
4666 	int mqd_idx = ring - &adev->gfx.compute_ring[0];
4667 
4668 	if (!amdgpu_in_reset(adev) && !adev->in_suspend) {
4669 		memset((void *)mqd, 0, sizeof(struct vi_mqd_allocation));
4670 		((struct vi_mqd_allocation *)mqd)->dynamic_cu_mask = 0xFFFFFFFF;
4671 		((struct vi_mqd_allocation *)mqd)->dynamic_rb_mask = 0xFFFFFFFF;
4672 		mutex_lock(&adev->srbm_mutex);
4673 		vi_srbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
4674 		gfx_v8_0_mqd_init(ring);
4675 		vi_srbm_select(adev, 0, 0, 0, 0);
4676 		mutex_unlock(&adev->srbm_mutex);
4677 
4678 		if (adev->gfx.mec.mqd_backup[mqd_idx])
4679 			memcpy(adev->gfx.mec.mqd_backup[mqd_idx], mqd, sizeof(struct vi_mqd_allocation));
4680 	} else if (amdgpu_in_reset(adev)) { /* for GPU_RESET case */
4681 		/* reset MQD to a clean status */
4682 		if (adev->gfx.mec.mqd_backup[mqd_idx])
4683 			memcpy(mqd, adev->gfx.mec.mqd_backup[mqd_idx], sizeof(struct vi_mqd_allocation));
4684 		/* reset ring buffer */
4685 		ring->wptr = 0;
4686 		amdgpu_ring_clear_ring(ring);
4687 	} else {
4688 		amdgpu_ring_clear_ring(ring);
4689 	}
4690 	return 0;
4691 }
4692 
4693 static void gfx_v8_0_set_mec_doorbell_range(struct amdgpu_device *adev)
4694 {
4695 	if (adev->asic_type > CHIP_TONGA) {
4696 		WREG32(mmCP_MEC_DOORBELL_RANGE_LOWER, adev->doorbell_index.kiq << 2);
4697 		WREG32(mmCP_MEC_DOORBELL_RANGE_UPPER, adev->doorbell_index.mec_ring7 << 2);
4698 	}
4699 	/* enable doorbells */
4700 	WREG32_FIELD(CP_PQ_STATUS, DOORBELL_ENABLE, 1);
4701 }
4702 
4703 static int gfx_v8_0_kiq_resume(struct amdgpu_device *adev)
4704 {
4705 	struct amdgpu_ring *ring;
4706 	int r;
4707 
4708 	ring = &adev->gfx.kiq.ring;
4709 
4710 	r = amdgpu_bo_reserve(ring->mqd_obj, false);
4711 	if (unlikely(r != 0))
4712 		return r;
4713 
4714 	r = amdgpu_bo_kmap(ring->mqd_obj, &ring->mqd_ptr);
4715 	if (unlikely(r != 0))
4716 		return r;
4717 
4718 	gfx_v8_0_kiq_init_queue(ring);
4719 	amdgpu_bo_kunmap(ring->mqd_obj);
4720 	ring->mqd_ptr = NULL;
4721 	amdgpu_bo_unreserve(ring->mqd_obj);
4722 	ring->sched.ready = true;
4723 	return 0;
4724 }
4725 
4726 static int gfx_v8_0_kcq_resume(struct amdgpu_device *adev)
4727 {
4728 	struct amdgpu_ring *ring = NULL;
4729 	int r = 0, i;
4730 
4731 	gfx_v8_0_cp_compute_enable(adev, true);
4732 
4733 	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
4734 		ring = &adev->gfx.compute_ring[i];
4735 
4736 		r = amdgpu_bo_reserve(ring->mqd_obj, false);
4737 		if (unlikely(r != 0))
4738 			goto done;
4739 		r = amdgpu_bo_kmap(ring->mqd_obj, &ring->mqd_ptr);
4740 		if (!r) {
4741 			r = gfx_v8_0_kcq_init_queue(ring);
4742 			amdgpu_bo_kunmap(ring->mqd_obj);
4743 			ring->mqd_ptr = NULL;
4744 		}
4745 		amdgpu_bo_unreserve(ring->mqd_obj);
4746 		if (r)
4747 			goto done;
4748 	}
4749 
4750 	gfx_v8_0_set_mec_doorbell_range(adev);
4751 
4752 	r = gfx_v8_0_kiq_kcq_enable(adev);
4753 	if (r)
4754 		goto done;
4755 
4756 done:
4757 	return r;
4758 }
4759 
4760 static int gfx_v8_0_cp_test_all_rings(struct amdgpu_device *adev)
4761 {
4762 	int r, i;
4763 	struct amdgpu_ring *ring;
4764 
4765 	/* collect all the ring_tests here, gfx, kiq, compute */
4766 	ring = &adev->gfx.gfx_ring[0];
4767 	r = amdgpu_ring_test_helper(ring);
4768 	if (r)
4769 		return r;
4770 
4771 	ring = &adev->gfx.kiq.ring;
4772 	r = amdgpu_ring_test_helper(ring);
4773 	if (r)
4774 		return r;
4775 
4776 	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
4777 		ring = &adev->gfx.compute_ring[i];
4778 		amdgpu_ring_test_helper(ring);
4779 	}
4780 
4781 	return 0;
4782 }
4783 
4784 static int gfx_v8_0_cp_resume(struct amdgpu_device *adev)
4785 {
4786 	int r;
4787 
4788 	if (!(adev->flags & AMD_IS_APU))
4789 		gfx_v8_0_enable_gui_idle_interrupt(adev, false);
4790 
4791 	r = gfx_v8_0_kiq_resume(adev);
4792 	if (r)
4793 		return r;
4794 
4795 	r = gfx_v8_0_cp_gfx_resume(adev);
4796 	if (r)
4797 		return r;
4798 
4799 	r = gfx_v8_0_kcq_resume(adev);
4800 	if (r)
4801 		return r;
4802 
4803 	r = gfx_v8_0_cp_test_all_rings(adev);
4804 	if (r)
4805 		return r;
4806 
4807 	gfx_v8_0_enable_gui_idle_interrupt(adev, true);
4808 
4809 	return 0;
4810 }
4811 
4812 static void gfx_v8_0_cp_enable(struct amdgpu_device *adev, bool enable)
4813 {
4814 	gfx_v8_0_cp_gfx_enable(adev, enable);
4815 	gfx_v8_0_cp_compute_enable(adev, enable);
4816 }
4817 
4818 static int gfx_v8_0_hw_init(void *handle)
4819 {
4820 	int r;
4821 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4822 
4823 	gfx_v8_0_init_golden_registers(adev);
4824 	gfx_v8_0_constants_init(adev);
4825 
4826 	r = adev->gfx.rlc.funcs->resume(adev);
4827 	if (r)
4828 		return r;
4829 
4830 	r = gfx_v8_0_cp_resume(adev);
4831 
4832 	return r;
4833 }
4834 
4835 static int gfx_v8_0_kcq_disable(struct amdgpu_device *adev)
4836 {
4837 	int r, i;
4838 	struct amdgpu_ring *kiq_ring = &adev->gfx.kiq.ring;
4839 
4840 	r = amdgpu_ring_alloc(kiq_ring, 6 * adev->gfx.num_compute_rings);
4841 	if (r)
4842 		DRM_ERROR("Failed to lock KIQ (%d).\n", r);
4843 
4844 	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
4845 		struct amdgpu_ring *ring = &adev->gfx.compute_ring[i];
4846 
4847 		amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_UNMAP_QUEUES, 4));
4848 		amdgpu_ring_write(kiq_ring, /* Q_sel: 0, vmid: 0, engine: 0, num_Q: 1 */
4849 						PACKET3_UNMAP_QUEUES_ACTION(1) | /* RESET_QUEUES */
4850 						PACKET3_UNMAP_QUEUES_QUEUE_SEL(0) |
4851 						PACKET3_UNMAP_QUEUES_ENGINE_SEL(0) |
4852 						PACKET3_UNMAP_QUEUES_NUM_QUEUES(1));
4853 		amdgpu_ring_write(kiq_ring, PACKET3_UNMAP_QUEUES_DOORBELL_OFFSET0(ring->doorbell_index));
4854 		amdgpu_ring_write(kiq_ring, 0);
4855 		amdgpu_ring_write(kiq_ring, 0);
4856 		amdgpu_ring_write(kiq_ring, 0);
4857 	}
4858 	r = amdgpu_ring_test_helper(kiq_ring);
4859 	if (r)
4860 		DRM_ERROR("KCQ disable failed\n");
4861 
4862 	return r;
4863 }
4864 
4865 static bool gfx_v8_0_is_idle(void *handle)
4866 {
4867 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4868 
4869 	if (REG_GET_FIELD(RREG32(mmGRBM_STATUS), GRBM_STATUS, GUI_ACTIVE)
4870 		|| RREG32(mmGRBM_STATUS2) != 0x8)
4871 		return false;
4872 	else
4873 		return true;
4874 }
4875 
4876 static bool gfx_v8_0_rlc_is_idle(void *handle)
4877 {
4878 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4879 
4880 	if (RREG32(mmGRBM_STATUS2) != 0x8)
4881 		return false;
4882 	else
4883 		return true;
4884 }
4885 
4886 static int gfx_v8_0_wait_for_rlc_idle(void *handle)
4887 {
4888 	unsigned int i;
4889 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4890 
4891 	for (i = 0; i < adev->usec_timeout; i++) {
4892 		if (gfx_v8_0_rlc_is_idle(handle))
4893 			return 0;
4894 
4895 		udelay(1);
4896 	}
4897 	return -ETIMEDOUT;
4898 }
4899 
4900 static int gfx_v8_0_wait_for_idle(void *handle)
4901 {
4902 	unsigned int i;
4903 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4904 
4905 	for (i = 0; i < adev->usec_timeout; i++) {
4906 		if (gfx_v8_0_is_idle(handle))
4907 			return 0;
4908 
4909 		udelay(1);
4910 	}
4911 	return -ETIMEDOUT;
4912 }
4913 
4914 static int gfx_v8_0_hw_fini(void *handle)
4915 {
4916 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4917 
4918 	amdgpu_irq_put(adev, &adev->gfx.priv_reg_irq, 0);
4919 	amdgpu_irq_put(adev, &adev->gfx.priv_inst_irq, 0);
4920 
4921 	amdgpu_irq_put(adev, &adev->gfx.cp_ecc_error_irq, 0);
4922 
4923 	amdgpu_irq_put(adev, &adev->gfx.sq_irq, 0);
4924 
4925 	/* disable KCQ to avoid CPC touch memory not valid anymore */
4926 	gfx_v8_0_kcq_disable(adev);
4927 
4928 	if (amdgpu_sriov_vf(adev)) {
4929 		pr_debug("For SRIOV client, shouldn't do anything.\n");
4930 		return 0;
4931 	}
4932 	amdgpu_gfx_rlc_enter_safe_mode(adev);
4933 	if (!gfx_v8_0_wait_for_idle(adev))
4934 		gfx_v8_0_cp_enable(adev, false);
4935 	else
4936 		pr_err("cp is busy, skip halt cp\n");
4937 	if (!gfx_v8_0_wait_for_rlc_idle(adev))
4938 		adev->gfx.rlc.funcs->stop(adev);
4939 	else
4940 		pr_err("rlc is busy, skip halt rlc\n");
4941 	amdgpu_gfx_rlc_exit_safe_mode(adev);
4942 
4943 	return 0;
4944 }
4945 
4946 static int gfx_v8_0_suspend(void *handle)
4947 {
4948 	return gfx_v8_0_hw_fini(handle);
4949 }
4950 
4951 static int gfx_v8_0_resume(void *handle)
4952 {
4953 	return gfx_v8_0_hw_init(handle);
4954 }
4955 
4956 static bool gfx_v8_0_check_soft_reset(void *handle)
4957 {
4958 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4959 	u32 grbm_soft_reset = 0, srbm_soft_reset = 0;
4960 	u32 tmp;
4961 
4962 	/* GRBM_STATUS */
4963 	tmp = RREG32(mmGRBM_STATUS);
4964 	if (tmp & (GRBM_STATUS__PA_BUSY_MASK | GRBM_STATUS__SC_BUSY_MASK |
4965 		   GRBM_STATUS__BCI_BUSY_MASK | GRBM_STATUS__SX_BUSY_MASK |
4966 		   GRBM_STATUS__TA_BUSY_MASK | GRBM_STATUS__VGT_BUSY_MASK |
4967 		   GRBM_STATUS__DB_BUSY_MASK | GRBM_STATUS__CB_BUSY_MASK |
4968 		   GRBM_STATUS__GDS_BUSY_MASK | GRBM_STATUS__SPI_BUSY_MASK |
4969 		   GRBM_STATUS__IA_BUSY_MASK | GRBM_STATUS__IA_BUSY_NO_DMA_MASK |
4970 		   GRBM_STATUS__CP_BUSY_MASK | GRBM_STATUS__CP_COHERENCY_BUSY_MASK)) {
4971 		grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset,
4972 						GRBM_SOFT_RESET, SOFT_RESET_CP, 1);
4973 		grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset,
4974 						GRBM_SOFT_RESET, SOFT_RESET_GFX, 1);
4975 		srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset,
4976 						SRBM_SOFT_RESET, SOFT_RESET_GRBM, 1);
4977 	}
4978 
4979 	/* GRBM_STATUS2 */
4980 	tmp = RREG32(mmGRBM_STATUS2);
4981 	if (REG_GET_FIELD(tmp, GRBM_STATUS2, RLC_BUSY))
4982 		grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset,
4983 						GRBM_SOFT_RESET, SOFT_RESET_RLC, 1);
4984 
4985 	if (REG_GET_FIELD(tmp, GRBM_STATUS2, CPF_BUSY) ||
4986 	    REG_GET_FIELD(tmp, GRBM_STATUS2, CPC_BUSY) ||
4987 	    REG_GET_FIELD(tmp, GRBM_STATUS2, CPG_BUSY)) {
4988 		grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET,
4989 						SOFT_RESET_CPF, 1);
4990 		grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET,
4991 						SOFT_RESET_CPC, 1);
4992 		grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET,
4993 						SOFT_RESET_CPG, 1);
4994 		srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset, SRBM_SOFT_RESET,
4995 						SOFT_RESET_GRBM, 1);
4996 	}
4997 
4998 	/* SRBM_STATUS */
4999 	tmp = RREG32(mmSRBM_STATUS);
5000 	if (REG_GET_FIELD(tmp, SRBM_STATUS, GRBM_RQ_PENDING))
5001 		srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset,
5002 						SRBM_SOFT_RESET, SOFT_RESET_GRBM, 1);
5003 	if (REG_GET_FIELD(tmp, SRBM_STATUS, SEM_BUSY))
5004 		srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset,
5005 						SRBM_SOFT_RESET, SOFT_RESET_SEM, 1);
5006 
5007 	if (grbm_soft_reset || srbm_soft_reset) {
5008 		adev->gfx.grbm_soft_reset = grbm_soft_reset;
5009 		adev->gfx.srbm_soft_reset = srbm_soft_reset;
5010 		return true;
5011 	} else {
5012 		adev->gfx.grbm_soft_reset = 0;
5013 		adev->gfx.srbm_soft_reset = 0;
5014 		return false;
5015 	}
5016 }
5017 
5018 static int gfx_v8_0_pre_soft_reset(void *handle)
5019 {
5020 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5021 	u32 grbm_soft_reset = 0;
5022 
5023 	if ((!adev->gfx.grbm_soft_reset) &&
5024 	    (!adev->gfx.srbm_soft_reset))
5025 		return 0;
5026 
5027 	grbm_soft_reset = adev->gfx.grbm_soft_reset;
5028 
5029 	/* stop the rlc */
5030 	adev->gfx.rlc.funcs->stop(adev);
5031 
5032 	if (REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CP) ||
5033 	    REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_GFX))
5034 		/* Disable GFX parsing/prefetching */
5035 		gfx_v8_0_cp_gfx_enable(adev, false);
5036 
5037 	if (REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CP) ||
5038 	    REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CPF) ||
5039 	    REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CPC) ||
5040 	    REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CPG)) {
5041 		int i;
5042 
5043 		for (i = 0; i < adev->gfx.num_compute_rings; i++) {
5044 			struct amdgpu_ring *ring = &adev->gfx.compute_ring[i];
5045 
5046 			mutex_lock(&adev->srbm_mutex);
5047 			vi_srbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
5048 			gfx_v8_0_deactivate_hqd(adev, 2);
5049 			vi_srbm_select(adev, 0, 0, 0, 0);
5050 			mutex_unlock(&adev->srbm_mutex);
5051 		}
5052 		/* Disable MEC parsing/prefetching */
5053 		gfx_v8_0_cp_compute_enable(adev, false);
5054 	}
5055 
5056 	return 0;
5057 }
5058 
5059 static int gfx_v8_0_soft_reset(void *handle)
5060 {
5061 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5062 	u32 grbm_soft_reset = 0, srbm_soft_reset = 0;
5063 	u32 tmp;
5064 
5065 	if ((!adev->gfx.grbm_soft_reset) &&
5066 	    (!adev->gfx.srbm_soft_reset))
5067 		return 0;
5068 
5069 	grbm_soft_reset = adev->gfx.grbm_soft_reset;
5070 	srbm_soft_reset = adev->gfx.srbm_soft_reset;
5071 
5072 	if (grbm_soft_reset || srbm_soft_reset) {
5073 		tmp = RREG32(mmGMCON_DEBUG);
5074 		tmp = REG_SET_FIELD(tmp, GMCON_DEBUG, GFX_STALL, 1);
5075 		tmp = REG_SET_FIELD(tmp, GMCON_DEBUG, GFX_CLEAR, 1);
5076 		WREG32(mmGMCON_DEBUG, tmp);
5077 		udelay(50);
5078 	}
5079 
5080 	if (grbm_soft_reset) {
5081 		tmp = RREG32(mmGRBM_SOFT_RESET);
5082 		tmp |= grbm_soft_reset;
5083 		dev_info(adev->dev, "GRBM_SOFT_RESET=0x%08X\n", tmp);
5084 		WREG32(mmGRBM_SOFT_RESET, tmp);
5085 		tmp = RREG32(mmGRBM_SOFT_RESET);
5086 
5087 		udelay(50);
5088 
5089 		tmp &= ~grbm_soft_reset;
5090 		WREG32(mmGRBM_SOFT_RESET, tmp);
5091 		tmp = RREG32(mmGRBM_SOFT_RESET);
5092 	}
5093 
5094 	if (srbm_soft_reset) {
5095 		tmp = RREG32(mmSRBM_SOFT_RESET);
5096 		tmp |= srbm_soft_reset;
5097 		dev_info(adev->dev, "SRBM_SOFT_RESET=0x%08X\n", tmp);
5098 		WREG32(mmSRBM_SOFT_RESET, tmp);
5099 		tmp = RREG32(mmSRBM_SOFT_RESET);
5100 
5101 		udelay(50);
5102 
5103 		tmp &= ~srbm_soft_reset;
5104 		WREG32(mmSRBM_SOFT_RESET, tmp);
5105 		tmp = RREG32(mmSRBM_SOFT_RESET);
5106 	}
5107 
5108 	if (grbm_soft_reset || srbm_soft_reset) {
5109 		tmp = RREG32(mmGMCON_DEBUG);
5110 		tmp = REG_SET_FIELD(tmp, GMCON_DEBUG, GFX_STALL, 0);
5111 		tmp = REG_SET_FIELD(tmp, GMCON_DEBUG, GFX_CLEAR, 0);
5112 		WREG32(mmGMCON_DEBUG, tmp);
5113 	}
5114 
5115 	/* Wait a little for things to settle down */
5116 	udelay(50);
5117 
5118 	return 0;
5119 }
5120 
5121 static int gfx_v8_0_post_soft_reset(void *handle)
5122 {
5123 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5124 	u32 grbm_soft_reset = 0;
5125 
5126 	if ((!adev->gfx.grbm_soft_reset) &&
5127 	    (!adev->gfx.srbm_soft_reset))
5128 		return 0;
5129 
5130 	grbm_soft_reset = adev->gfx.grbm_soft_reset;
5131 
5132 	if (REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CP) ||
5133 	    REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CPF) ||
5134 	    REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CPC) ||
5135 	    REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CPG)) {
5136 		int i;
5137 
5138 		for (i = 0; i < adev->gfx.num_compute_rings; i++) {
5139 			struct amdgpu_ring *ring = &adev->gfx.compute_ring[i];
5140 
5141 			mutex_lock(&adev->srbm_mutex);
5142 			vi_srbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
5143 			gfx_v8_0_deactivate_hqd(adev, 2);
5144 			vi_srbm_select(adev, 0, 0, 0, 0);
5145 			mutex_unlock(&adev->srbm_mutex);
5146 		}
5147 		gfx_v8_0_kiq_resume(adev);
5148 		gfx_v8_0_kcq_resume(adev);
5149 	}
5150 
5151 	if (REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CP) ||
5152 	    REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_GFX))
5153 		gfx_v8_0_cp_gfx_resume(adev);
5154 
5155 	gfx_v8_0_cp_test_all_rings(adev);
5156 
5157 	adev->gfx.rlc.funcs->start(adev);
5158 
5159 	return 0;
5160 }
5161 
5162 /**
5163  * gfx_v8_0_get_gpu_clock_counter - return GPU clock counter snapshot
5164  *
5165  * @adev: amdgpu_device pointer
5166  *
5167  * Fetches a GPU clock counter snapshot.
5168  * Returns the 64 bit clock counter snapshot.
5169  */
5170 static uint64_t gfx_v8_0_get_gpu_clock_counter(struct amdgpu_device *adev)
5171 {
5172 	uint64_t clock;
5173 
5174 	mutex_lock(&adev->gfx.gpu_clock_mutex);
5175 	WREG32(mmRLC_CAPTURE_GPU_CLOCK_COUNT, 1);
5176 	clock = (uint64_t)RREG32(mmRLC_GPU_CLOCK_COUNT_LSB) |
5177 		((uint64_t)RREG32(mmRLC_GPU_CLOCK_COUNT_MSB) << 32ULL);
5178 	mutex_unlock(&adev->gfx.gpu_clock_mutex);
5179 	return clock;
5180 }
5181 
5182 static void gfx_v8_0_ring_emit_gds_switch(struct amdgpu_ring *ring,
5183 					  uint32_t vmid,
5184 					  uint32_t gds_base, uint32_t gds_size,
5185 					  uint32_t gws_base, uint32_t gws_size,
5186 					  uint32_t oa_base, uint32_t oa_size)
5187 {
5188 	/* GDS Base */
5189 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
5190 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
5191 				WRITE_DATA_DST_SEL(0)));
5192 	amdgpu_ring_write(ring, amdgpu_gds_reg_offset[vmid].mem_base);
5193 	amdgpu_ring_write(ring, 0);
5194 	amdgpu_ring_write(ring, gds_base);
5195 
5196 	/* GDS Size */
5197 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
5198 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
5199 				WRITE_DATA_DST_SEL(0)));
5200 	amdgpu_ring_write(ring, amdgpu_gds_reg_offset[vmid].mem_size);
5201 	amdgpu_ring_write(ring, 0);
5202 	amdgpu_ring_write(ring, gds_size);
5203 
5204 	/* GWS */
5205 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
5206 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
5207 				WRITE_DATA_DST_SEL(0)));
5208 	amdgpu_ring_write(ring, amdgpu_gds_reg_offset[vmid].gws);
5209 	amdgpu_ring_write(ring, 0);
5210 	amdgpu_ring_write(ring, gws_size << GDS_GWS_VMID0__SIZE__SHIFT | gws_base);
5211 
5212 	/* OA */
5213 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
5214 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
5215 				WRITE_DATA_DST_SEL(0)));
5216 	amdgpu_ring_write(ring, amdgpu_gds_reg_offset[vmid].oa);
5217 	amdgpu_ring_write(ring, 0);
5218 	amdgpu_ring_write(ring, (1 << (oa_size + oa_base)) - (1 << oa_base));
5219 }
5220 
5221 static uint32_t wave_read_ind(struct amdgpu_device *adev, uint32_t simd, uint32_t wave, uint32_t address)
5222 {
5223 	WREG32(mmSQ_IND_INDEX,
5224 		(wave << SQ_IND_INDEX__WAVE_ID__SHIFT) |
5225 		(simd << SQ_IND_INDEX__SIMD_ID__SHIFT) |
5226 		(address << SQ_IND_INDEX__INDEX__SHIFT) |
5227 		(SQ_IND_INDEX__FORCE_READ_MASK));
5228 	return RREG32(mmSQ_IND_DATA);
5229 }
5230 
5231 static void wave_read_regs(struct amdgpu_device *adev, uint32_t simd,
5232 			   uint32_t wave, uint32_t thread,
5233 			   uint32_t regno, uint32_t num, uint32_t *out)
5234 {
5235 	WREG32(mmSQ_IND_INDEX,
5236 		(wave << SQ_IND_INDEX__WAVE_ID__SHIFT) |
5237 		(simd << SQ_IND_INDEX__SIMD_ID__SHIFT) |
5238 		(regno << SQ_IND_INDEX__INDEX__SHIFT) |
5239 		(thread << SQ_IND_INDEX__THREAD_ID__SHIFT) |
5240 		(SQ_IND_INDEX__FORCE_READ_MASK) |
5241 		(SQ_IND_INDEX__AUTO_INCR_MASK));
5242 	while (num--)
5243 		*(out++) = RREG32(mmSQ_IND_DATA);
5244 }
5245 
5246 static void gfx_v8_0_read_wave_data(struct amdgpu_device *adev, uint32_t simd, uint32_t wave, uint32_t *dst, int *no_fields)
5247 {
5248 	/* type 0 wave data */
5249 	dst[(*no_fields)++] = 0;
5250 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_STATUS);
5251 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_PC_LO);
5252 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_PC_HI);
5253 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_EXEC_LO);
5254 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_EXEC_HI);
5255 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_HW_ID);
5256 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_INST_DW0);
5257 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_INST_DW1);
5258 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_GPR_ALLOC);
5259 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_LDS_ALLOC);
5260 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TRAPSTS);
5261 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_IB_STS);
5262 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TBA_LO);
5263 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TBA_HI);
5264 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TMA_LO);
5265 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TMA_HI);
5266 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_IB_DBG0);
5267 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_M0);
5268 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_MODE);
5269 }
5270 
5271 static void gfx_v8_0_read_wave_sgprs(struct amdgpu_device *adev, uint32_t simd,
5272 				     uint32_t wave, uint32_t start,
5273 				     uint32_t size, uint32_t *dst)
5274 {
5275 	wave_read_regs(
5276 		adev, simd, wave, 0,
5277 		start + SQIND_WAVE_SGPRS_OFFSET, size, dst);
5278 }
5279 
5280 
5281 static const struct amdgpu_gfx_funcs gfx_v8_0_gfx_funcs = {
5282 	.get_gpu_clock_counter = &gfx_v8_0_get_gpu_clock_counter,
5283 	.select_se_sh = &gfx_v8_0_select_se_sh,
5284 	.read_wave_data = &gfx_v8_0_read_wave_data,
5285 	.read_wave_sgprs = &gfx_v8_0_read_wave_sgprs,
5286 	.select_me_pipe_q = &gfx_v8_0_select_me_pipe_q
5287 };
5288 
5289 static int gfx_v8_0_early_init(void *handle)
5290 {
5291 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5292 
5293 	adev->gfx.num_gfx_rings = GFX8_NUM_GFX_RINGS;
5294 	adev->gfx.num_compute_rings = min(amdgpu_gfx_get_num_kcq(adev),
5295 					  AMDGPU_MAX_COMPUTE_RINGS);
5296 	adev->gfx.funcs = &gfx_v8_0_gfx_funcs;
5297 	gfx_v8_0_set_ring_funcs(adev);
5298 	gfx_v8_0_set_irq_funcs(adev);
5299 	gfx_v8_0_set_gds_init(adev);
5300 	gfx_v8_0_set_rlc_funcs(adev);
5301 
5302 	return 0;
5303 }
5304 
5305 static int gfx_v8_0_late_init(void *handle)
5306 {
5307 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5308 	int r;
5309 
5310 	r = amdgpu_irq_get(adev, &adev->gfx.priv_reg_irq, 0);
5311 	if (r)
5312 		return r;
5313 
5314 	r = amdgpu_irq_get(adev, &adev->gfx.priv_inst_irq, 0);
5315 	if (r)
5316 		return r;
5317 
5318 	/* requires IBs so do in late init after IB pool is initialized */
5319 	r = gfx_v8_0_do_edc_gpr_workarounds(adev);
5320 	if (r)
5321 		return r;
5322 
5323 	r = amdgpu_irq_get(adev, &adev->gfx.cp_ecc_error_irq, 0);
5324 	if (r) {
5325 		DRM_ERROR("amdgpu_irq_get() failed to get IRQ for EDC, r: %d.\n", r);
5326 		return r;
5327 	}
5328 
5329 	r = amdgpu_irq_get(adev, &adev->gfx.sq_irq, 0);
5330 	if (r) {
5331 		DRM_ERROR(
5332 			"amdgpu_irq_get() failed to get IRQ for SQ, r: %d.\n",
5333 			r);
5334 		return r;
5335 	}
5336 
5337 	return 0;
5338 }
5339 
5340 static void gfx_v8_0_enable_gfx_static_mg_power_gating(struct amdgpu_device *adev,
5341 						       bool enable)
5342 {
5343 	if ((adev->asic_type == CHIP_POLARIS11) ||
5344 	    (adev->asic_type == CHIP_POLARIS12) ||
5345 	    (adev->asic_type == CHIP_VEGAM))
5346 		/* Send msg to SMU via Powerplay */
5347 		amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_GFX, enable);
5348 
5349 	WREG32_FIELD(RLC_PG_CNTL, STATIC_PER_CU_PG_ENABLE, enable ? 1 : 0);
5350 }
5351 
5352 static void gfx_v8_0_enable_gfx_dynamic_mg_power_gating(struct amdgpu_device *adev,
5353 							bool enable)
5354 {
5355 	WREG32_FIELD(RLC_PG_CNTL, DYN_PER_CU_PG_ENABLE, enable ? 1 : 0);
5356 }
5357 
5358 static void polaris11_enable_gfx_quick_mg_power_gating(struct amdgpu_device *adev,
5359 		bool enable)
5360 {
5361 	WREG32_FIELD(RLC_PG_CNTL, QUICK_PG_ENABLE, enable ? 1 : 0);
5362 }
5363 
5364 static void cz_enable_gfx_cg_power_gating(struct amdgpu_device *adev,
5365 					  bool enable)
5366 {
5367 	WREG32_FIELD(RLC_PG_CNTL, GFX_POWER_GATING_ENABLE, enable ? 1 : 0);
5368 }
5369 
5370 static void cz_enable_gfx_pipeline_power_gating(struct amdgpu_device *adev,
5371 						bool enable)
5372 {
5373 	WREG32_FIELD(RLC_PG_CNTL, GFX_PIPELINE_PG_ENABLE, enable ? 1 : 0);
5374 
5375 	/* Read any GFX register to wake up GFX. */
5376 	if (!enable)
5377 		RREG32(mmDB_RENDER_CONTROL);
5378 }
5379 
5380 static void cz_update_gfx_cg_power_gating(struct amdgpu_device *adev,
5381 					  bool enable)
5382 {
5383 	if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_PG) && enable) {
5384 		cz_enable_gfx_cg_power_gating(adev, true);
5385 		if (adev->pg_flags & AMD_PG_SUPPORT_GFX_PIPELINE)
5386 			cz_enable_gfx_pipeline_power_gating(adev, true);
5387 	} else {
5388 		cz_enable_gfx_cg_power_gating(adev, false);
5389 		cz_enable_gfx_pipeline_power_gating(adev, false);
5390 	}
5391 }
5392 
5393 static int gfx_v8_0_set_powergating_state(void *handle,
5394 					  enum amd_powergating_state state)
5395 {
5396 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5397 	bool enable = (state == AMD_PG_STATE_GATE);
5398 
5399 	if (amdgpu_sriov_vf(adev))
5400 		return 0;
5401 
5402 	if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_SMG |
5403 				AMD_PG_SUPPORT_RLC_SMU_HS |
5404 				AMD_PG_SUPPORT_CP |
5405 				AMD_PG_SUPPORT_GFX_DMG))
5406 		amdgpu_gfx_rlc_enter_safe_mode(adev);
5407 	switch (adev->asic_type) {
5408 	case CHIP_CARRIZO:
5409 	case CHIP_STONEY:
5410 
5411 		if (adev->pg_flags & AMD_PG_SUPPORT_RLC_SMU_HS) {
5412 			cz_enable_sck_slow_down_on_power_up(adev, true);
5413 			cz_enable_sck_slow_down_on_power_down(adev, true);
5414 		} else {
5415 			cz_enable_sck_slow_down_on_power_up(adev, false);
5416 			cz_enable_sck_slow_down_on_power_down(adev, false);
5417 		}
5418 		if (adev->pg_flags & AMD_PG_SUPPORT_CP)
5419 			cz_enable_cp_power_gating(adev, true);
5420 		else
5421 			cz_enable_cp_power_gating(adev, false);
5422 
5423 		cz_update_gfx_cg_power_gating(adev, enable);
5424 
5425 		if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_SMG) && enable)
5426 			gfx_v8_0_enable_gfx_static_mg_power_gating(adev, true);
5427 		else
5428 			gfx_v8_0_enable_gfx_static_mg_power_gating(adev, false);
5429 
5430 		if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_DMG) && enable)
5431 			gfx_v8_0_enable_gfx_dynamic_mg_power_gating(adev, true);
5432 		else
5433 			gfx_v8_0_enable_gfx_dynamic_mg_power_gating(adev, false);
5434 		break;
5435 	case CHIP_POLARIS11:
5436 	case CHIP_POLARIS12:
5437 	case CHIP_VEGAM:
5438 		if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_SMG) && enable)
5439 			gfx_v8_0_enable_gfx_static_mg_power_gating(adev, true);
5440 		else
5441 			gfx_v8_0_enable_gfx_static_mg_power_gating(adev, false);
5442 
5443 		if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_DMG) && enable)
5444 			gfx_v8_0_enable_gfx_dynamic_mg_power_gating(adev, true);
5445 		else
5446 			gfx_v8_0_enable_gfx_dynamic_mg_power_gating(adev, false);
5447 
5448 		if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_QUICK_MG) && enable)
5449 			polaris11_enable_gfx_quick_mg_power_gating(adev, true);
5450 		else
5451 			polaris11_enable_gfx_quick_mg_power_gating(adev, false);
5452 		break;
5453 	default:
5454 		break;
5455 	}
5456 	if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_SMG |
5457 				AMD_PG_SUPPORT_RLC_SMU_HS |
5458 				AMD_PG_SUPPORT_CP |
5459 				AMD_PG_SUPPORT_GFX_DMG))
5460 		amdgpu_gfx_rlc_exit_safe_mode(adev);
5461 	return 0;
5462 }
5463 
5464 static void gfx_v8_0_get_clockgating_state(void *handle, u64 *flags)
5465 {
5466 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5467 	int data;
5468 
5469 	if (amdgpu_sriov_vf(adev))
5470 		*flags = 0;
5471 
5472 	/* AMD_CG_SUPPORT_GFX_MGCG */
5473 	data = RREG32(mmRLC_CGTT_MGCG_OVERRIDE);
5474 	if (!(data & RLC_CGTT_MGCG_OVERRIDE__CPF_MASK))
5475 		*flags |= AMD_CG_SUPPORT_GFX_MGCG;
5476 
5477 	/* AMD_CG_SUPPORT_GFX_CGLG */
5478 	data = RREG32(mmRLC_CGCG_CGLS_CTRL);
5479 	if (data & RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK)
5480 		*flags |= AMD_CG_SUPPORT_GFX_CGCG;
5481 
5482 	/* AMD_CG_SUPPORT_GFX_CGLS */
5483 	if (data & RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK)
5484 		*flags |= AMD_CG_SUPPORT_GFX_CGLS;
5485 
5486 	/* AMD_CG_SUPPORT_GFX_CGTS */
5487 	data = RREG32(mmCGTS_SM_CTRL_REG);
5488 	if (!(data & CGTS_SM_CTRL_REG__OVERRIDE_MASK))
5489 		*flags |= AMD_CG_SUPPORT_GFX_CGTS;
5490 
5491 	/* AMD_CG_SUPPORT_GFX_CGTS_LS */
5492 	if (!(data & CGTS_SM_CTRL_REG__LS_OVERRIDE_MASK))
5493 		*flags |= AMD_CG_SUPPORT_GFX_CGTS_LS;
5494 
5495 	/* AMD_CG_SUPPORT_GFX_RLC_LS */
5496 	data = RREG32(mmRLC_MEM_SLP_CNTL);
5497 	if (data & RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK)
5498 		*flags |= AMD_CG_SUPPORT_GFX_RLC_LS | AMD_CG_SUPPORT_GFX_MGLS;
5499 
5500 	/* AMD_CG_SUPPORT_GFX_CP_LS */
5501 	data = RREG32(mmCP_MEM_SLP_CNTL);
5502 	if (data & CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK)
5503 		*flags |= AMD_CG_SUPPORT_GFX_CP_LS | AMD_CG_SUPPORT_GFX_MGLS;
5504 }
5505 
5506 static void gfx_v8_0_send_serdes_cmd(struct amdgpu_device *adev,
5507 				     uint32_t reg_addr, uint32_t cmd)
5508 {
5509 	uint32_t data;
5510 
5511 	gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
5512 
5513 	WREG32(mmRLC_SERDES_WR_CU_MASTER_MASK, 0xffffffff);
5514 	WREG32(mmRLC_SERDES_WR_NONCU_MASTER_MASK, 0xffffffff);
5515 
5516 	data = RREG32(mmRLC_SERDES_WR_CTRL);
5517 	if (adev->asic_type == CHIP_STONEY)
5518 		data &= ~(RLC_SERDES_WR_CTRL__WRITE_COMMAND_MASK |
5519 			  RLC_SERDES_WR_CTRL__READ_COMMAND_MASK |
5520 			  RLC_SERDES_WR_CTRL__P1_SELECT_MASK |
5521 			  RLC_SERDES_WR_CTRL__P2_SELECT_MASK |
5522 			  RLC_SERDES_WR_CTRL__RDDATA_RESET_MASK |
5523 			  RLC_SERDES_WR_CTRL__POWER_DOWN_MASK |
5524 			  RLC_SERDES_WR_CTRL__POWER_UP_MASK |
5525 			  RLC_SERDES_WR_CTRL__SHORT_FORMAT_MASK |
5526 			  RLC_SERDES_WR_CTRL__SRBM_OVERRIDE_MASK);
5527 	else
5528 		data &= ~(RLC_SERDES_WR_CTRL__WRITE_COMMAND_MASK |
5529 			  RLC_SERDES_WR_CTRL__READ_COMMAND_MASK |
5530 			  RLC_SERDES_WR_CTRL__P1_SELECT_MASK |
5531 			  RLC_SERDES_WR_CTRL__P2_SELECT_MASK |
5532 			  RLC_SERDES_WR_CTRL__RDDATA_RESET_MASK |
5533 			  RLC_SERDES_WR_CTRL__POWER_DOWN_MASK |
5534 			  RLC_SERDES_WR_CTRL__POWER_UP_MASK |
5535 			  RLC_SERDES_WR_CTRL__SHORT_FORMAT_MASK |
5536 			  RLC_SERDES_WR_CTRL__BPM_DATA_MASK |
5537 			  RLC_SERDES_WR_CTRL__REG_ADDR_MASK |
5538 			  RLC_SERDES_WR_CTRL__SRBM_OVERRIDE_MASK);
5539 	data |= (RLC_SERDES_WR_CTRL__RSVD_BPM_ADDR_MASK |
5540 		 (cmd << RLC_SERDES_WR_CTRL__BPM_DATA__SHIFT) |
5541 		 (reg_addr << RLC_SERDES_WR_CTRL__REG_ADDR__SHIFT) |
5542 		 (0xff << RLC_SERDES_WR_CTRL__BPM_ADDR__SHIFT));
5543 
5544 	WREG32(mmRLC_SERDES_WR_CTRL, data);
5545 }
5546 
5547 #define MSG_ENTER_RLC_SAFE_MODE     1
5548 #define MSG_EXIT_RLC_SAFE_MODE      0
5549 #define RLC_GPR_REG2__REQ_MASK 0x00000001
5550 #define RLC_GPR_REG2__REQ__SHIFT 0
5551 #define RLC_GPR_REG2__MESSAGE__SHIFT 0x00000001
5552 #define RLC_GPR_REG2__MESSAGE_MASK 0x0000001e
5553 
5554 static bool gfx_v8_0_is_rlc_enabled(struct amdgpu_device *adev)
5555 {
5556 	uint32_t rlc_setting;
5557 
5558 	rlc_setting = RREG32(mmRLC_CNTL);
5559 	if (!(rlc_setting & RLC_CNTL__RLC_ENABLE_F32_MASK))
5560 		return false;
5561 
5562 	return true;
5563 }
5564 
5565 static void gfx_v8_0_set_safe_mode(struct amdgpu_device *adev)
5566 {
5567 	uint32_t data;
5568 	unsigned i;
5569 	data = RREG32(mmRLC_CNTL);
5570 	data |= RLC_SAFE_MODE__CMD_MASK;
5571 	data &= ~RLC_SAFE_MODE__MESSAGE_MASK;
5572 	data |= (1 << RLC_SAFE_MODE__MESSAGE__SHIFT);
5573 	WREG32(mmRLC_SAFE_MODE, data);
5574 
5575 	/* wait for RLC_SAFE_MODE */
5576 	for (i = 0; i < adev->usec_timeout; i++) {
5577 		if ((RREG32(mmRLC_GPM_STAT) &
5578 		     (RLC_GPM_STAT__GFX_CLOCK_STATUS_MASK |
5579 		      RLC_GPM_STAT__GFX_POWER_STATUS_MASK)) ==
5580 		    (RLC_GPM_STAT__GFX_CLOCK_STATUS_MASK |
5581 		     RLC_GPM_STAT__GFX_POWER_STATUS_MASK))
5582 			break;
5583 		udelay(1);
5584 	}
5585 	for (i = 0; i < adev->usec_timeout; i++) {
5586 		if (!REG_GET_FIELD(RREG32(mmRLC_SAFE_MODE), RLC_SAFE_MODE, CMD))
5587 			break;
5588 		udelay(1);
5589 	}
5590 }
5591 
5592 static void gfx_v8_0_unset_safe_mode(struct amdgpu_device *adev)
5593 {
5594 	uint32_t data;
5595 	unsigned i;
5596 
5597 	data = RREG32(mmRLC_CNTL);
5598 	data |= RLC_SAFE_MODE__CMD_MASK;
5599 	data &= ~RLC_SAFE_MODE__MESSAGE_MASK;
5600 	WREG32(mmRLC_SAFE_MODE, data);
5601 
5602 	for (i = 0; i < adev->usec_timeout; i++) {
5603 		if (!REG_GET_FIELD(RREG32(mmRLC_SAFE_MODE), RLC_SAFE_MODE, CMD))
5604 			break;
5605 		udelay(1);
5606 	}
5607 }
5608 
5609 static void gfx_v8_0_update_spm_vmid(struct amdgpu_device *adev, unsigned vmid)
5610 {
5611 	u32 data;
5612 
5613 	amdgpu_gfx_off_ctrl(adev, false);
5614 
5615 	if (amdgpu_sriov_is_pp_one_vf(adev))
5616 		data = RREG32_NO_KIQ(mmRLC_SPM_VMID);
5617 	else
5618 		data = RREG32(mmRLC_SPM_VMID);
5619 
5620 	data &= ~RLC_SPM_VMID__RLC_SPM_VMID_MASK;
5621 	data |= (vmid & RLC_SPM_VMID__RLC_SPM_VMID_MASK) << RLC_SPM_VMID__RLC_SPM_VMID__SHIFT;
5622 
5623 	if (amdgpu_sriov_is_pp_one_vf(adev))
5624 		WREG32_NO_KIQ(mmRLC_SPM_VMID, data);
5625 	else
5626 		WREG32(mmRLC_SPM_VMID, data);
5627 
5628 	amdgpu_gfx_off_ctrl(adev, true);
5629 }
5630 
5631 static const struct amdgpu_rlc_funcs iceland_rlc_funcs = {
5632 	.is_rlc_enabled = gfx_v8_0_is_rlc_enabled,
5633 	.set_safe_mode = gfx_v8_0_set_safe_mode,
5634 	.unset_safe_mode = gfx_v8_0_unset_safe_mode,
5635 	.init = gfx_v8_0_rlc_init,
5636 	.get_csb_size = gfx_v8_0_get_csb_size,
5637 	.get_csb_buffer = gfx_v8_0_get_csb_buffer,
5638 	.get_cp_table_num = gfx_v8_0_cp_jump_table_num,
5639 	.resume = gfx_v8_0_rlc_resume,
5640 	.stop = gfx_v8_0_rlc_stop,
5641 	.reset = gfx_v8_0_rlc_reset,
5642 	.start = gfx_v8_0_rlc_start,
5643 	.update_spm_vmid = gfx_v8_0_update_spm_vmid
5644 };
5645 
5646 static void gfx_v8_0_update_medium_grain_clock_gating(struct amdgpu_device *adev,
5647 						      bool enable)
5648 {
5649 	uint32_t temp, data;
5650 
5651 	amdgpu_gfx_rlc_enter_safe_mode(adev);
5652 
5653 	/* It is disabled by HW by default */
5654 	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGCG)) {
5655 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS) {
5656 			if (adev->cg_flags & AMD_CG_SUPPORT_GFX_RLC_LS)
5657 				/* 1 - RLC memory Light sleep */
5658 				WREG32_FIELD(RLC_MEM_SLP_CNTL, RLC_MEM_LS_EN, 1);
5659 
5660 			if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CP_LS)
5661 				WREG32_FIELD(CP_MEM_SLP_CNTL, CP_MEM_LS_EN, 1);
5662 		}
5663 
5664 		/* 3 - RLC_CGTT_MGCG_OVERRIDE */
5665 		temp = data = RREG32(mmRLC_CGTT_MGCG_OVERRIDE);
5666 		if (adev->flags & AMD_IS_APU)
5667 			data &= ~(RLC_CGTT_MGCG_OVERRIDE__CPF_MASK |
5668 				  RLC_CGTT_MGCG_OVERRIDE__RLC_MASK |
5669 				  RLC_CGTT_MGCG_OVERRIDE__MGCG_MASK);
5670 		else
5671 			data &= ~(RLC_CGTT_MGCG_OVERRIDE__CPF_MASK |
5672 				  RLC_CGTT_MGCG_OVERRIDE__RLC_MASK |
5673 				  RLC_CGTT_MGCG_OVERRIDE__MGCG_MASK |
5674 				  RLC_CGTT_MGCG_OVERRIDE__GRBM_MASK);
5675 
5676 		if (temp != data)
5677 			WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data);
5678 
5679 		/* 4 - wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */
5680 		gfx_v8_0_wait_for_rlc_serdes(adev);
5681 
5682 		/* 5 - clear mgcg override */
5683 		gfx_v8_0_send_serdes_cmd(adev, BPM_REG_MGCG_OVERRIDE, CLE_BPM_SERDES_CMD);
5684 
5685 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGTS) {
5686 			/* 6 - Enable CGTS(Tree Shade) MGCG /MGLS */
5687 			temp = data = RREG32(mmCGTS_SM_CTRL_REG);
5688 			data &= ~(CGTS_SM_CTRL_REG__SM_MODE_MASK);
5689 			data |= (0x2 << CGTS_SM_CTRL_REG__SM_MODE__SHIFT);
5690 			data |= CGTS_SM_CTRL_REG__SM_MODE_ENABLE_MASK;
5691 			data &= ~CGTS_SM_CTRL_REG__OVERRIDE_MASK;
5692 			if ((adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS) &&
5693 			    (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGTS_LS))
5694 				data &= ~CGTS_SM_CTRL_REG__LS_OVERRIDE_MASK;
5695 			data |= CGTS_SM_CTRL_REG__ON_MONITOR_ADD_EN_MASK;
5696 			data |= (0x96 << CGTS_SM_CTRL_REG__ON_MONITOR_ADD__SHIFT);
5697 			if (temp != data)
5698 				WREG32(mmCGTS_SM_CTRL_REG, data);
5699 		}
5700 		udelay(50);
5701 
5702 		/* 7 - wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */
5703 		gfx_v8_0_wait_for_rlc_serdes(adev);
5704 	} else {
5705 		/* 1 - MGCG_OVERRIDE[0] for CP and MGCG_OVERRIDE[1] for RLC */
5706 		temp = data = RREG32(mmRLC_CGTT_MGCG_OVERRIDE);
5707 		data |= (RLC_CGTT_MGCG_OVERRIDE__CPF_MASK |
5708 				RLC_CGTT_MGCG_OVERRIDE__RLC_MASK |
5709 				RLC_CGTT_MGCG_OVERRIDE__MGCG_MASK |
5710 				RLC_CGTT_MGCG_OVERRIDE__GRBM_MASK);
5711 		if (temp != data)
5712 			WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data);
5713 
5714 		/* 2 - disable MGLS in RLC */
5715 		data = RREG32(mmRLC_MEM_SLP_CNTL);
5716 		if (data & RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK) {
5717 			data &= ~RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK;
5718 			WREG32(mmRLC_MEM_SLP_CNTL, data);
5719 		}
5720 
5721 		/* 3 - disable MGLS in CP */
5722 		data = RREG32(mmCP_MEM_SLP_CNTL);
5723 		if (data & CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK) {
5724 			data &= ~CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK;
5725 			WREG32(mmCP_MEM_SLP_CNTL, data);
5726 		}
5727 
5728 		/* 4 - Disable CGTS(Tree Shade) MGCG and MGLS */
5729 		temp = data = RREG32(mmCGTS_SM_CTRL_REG);
5730 		data |= (CGTS_SM_CTRL_REG__OVERRIDE_MASK |
5731 				CGTS_SM_CTRL_REG__LS_OVERRIDE_MASK);
5732 		if (temp != data)
5733 			WREG32(mmCGTS_SM_CTRL_REG, data);
5734 
5735 		/* 5 - wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */
5736 		gfx_v8_0_wait_for_rlc_serdes(adev);
5737 
5738 		/* 6 - set mgcg override */
5739 		gfx_v8_0_send_serdes_cmd(adev, BPM_REG_MGCG_OVERRIDE, SET_BPM_SERDES_CMD);
5740 
5741 		udelay(50);
5742 
5743 		/* 7- wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */
5744 		gfx_v8_0_wait_for_rlc_serdes(adev);
5745 	}
5746 
5747 	amdgpu_gfx_rlc_exit_safe_mode(adev);
5748 }
5749 
5750 static void gfx_v8_0_update_coarse_grain_clock_gating(struct amdgpu_device *adev,
5751 						      bool enable)
5752 {
5753 	uint32_t temp, temp1, data, data1;
5754 
5755 	temp = data = RREG32(mmRLC_CGCG_CGLS_CTRL);
5756 
5757 	amdgpu_gfx_rlc_enter_safe_mode(adev);
5758 
5759 	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGCG)) {
5760 		temp1 = data1 =	RREG32(mmRLC_CGTT_MGCG_OVERRIDE);
5761 		data1 &= ~RLC_CGTT_MGCG_OVERRIDE__CGCG_MASK;
5762 		if (temp1 != data1)
5763 			WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data1);
5764 
5765 		/* : wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */
5766 		gfx_v8_0_wait_for_rlc_serdes(adev);
5767 
5768 		/* 2 - clear cgcg override */
5769 		gfx_v8_0_send_serdes_cmd(adev, BPM_REG_CGCG_OVERRIDE, CLE_BPM_SERDES_CMD);
5770 
5771 		/* wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */
5772 		gfx_v8_0_wait_for_rlc_serdes(adev);
5773 
5774 		/* 3 - write cmd to set CGLS */
5775 		gfx_v8_0_send_serdes_cmd(adev, BPM_REG_CGLS_EN, SET_BPM_SERDES_CMD);
5776 
5777 		/* 4 - enable cgcg */
5778 		data |= RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK;
5779 
5780 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGLS) {
5781 			/* enable cgls*/
5782 			data |= RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK;
5783 
5784 			temp1 = data1 =	RREG32(mmRLC_CGTT_MGCG_OVERRIDE);
5785 			data1 &= ~RLC_CGTT_MGCG_OVERRIDE__CGLS_MASK;
5786 
5787 			if (temp1 != data1)
5788 				WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data1);
5789 		} else {
5790 			data &= ~RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK;
5791 		}
5792 
5793 		if (temp != data)
5794 			WREG32(mmRLC_CGCG_CGLS_CTRL, data);
5795 
5796 		/* 5 enable cntx_empty_int_enable/cntx_busy_int_enable/
5797 		 * Cmp_busy/GFX_Idle interrupts
5798 		 */
5799 		gfx_v8_0_enable_gui_idle_interrupt(adev, true);
5800 	} else {
5801 		/* disable cntx_empty_int_enable & GFX Idle interrupt */
5802 		gfx_v8_0_enable_gui_idle_interrupt(adev, false);
5803 
5804 		/* TEST CGCG */
5805 		temp1 = data1 =	RREG32(mmRLC_CGTT_MGCG_OVERRIDE);
5806 		data1 |= (RLC_CGTT_MGCG_OVERRIDE__CGCG_MASK |
5807 				RLC_CGTT_MGCG_OVERRIDE__CGLS_MASK);
5808 		if (temp1 != data1)
5809 			WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data1);
5810 
5811 		/* read gfx register to wake up cgcg */
5812 		RREG32(mmCB_CGTT_SCLK_CTRL);
5813 		RREG32(mmCB_CGTT_SCLK_CTRL);
5814 		RREG32(mmCB_CGTT_SCLK_CTRL);
5815 		RREG32(mmCB_CGTT_SCLK_CTRL);
5816 
5817 		/* wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */
5818 		gfx_v8_0_wait_for_rlc_serdes(adev);
5819 
5820 		/* write cmd to Set CGCG Override */
5821 		gfx_v8_0_send_serdes_cmd(adev, BPM_REG_CGCG_OVERRIDE, SET_BPM_SERDES_CMD);
5822 
5823 		/* wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */
5824 		gfx_v8_0_wait_for_rlc_serdes(adev);
5825 
5826 		/* write cmd to Clear CGLS */
5827 		gfx_v8_0_send_serdes_cmd(adev, BPM_REG_CGLS_EN, CLE_BPM_SERDES_CMD);
5828 
5829 		/* disable cgcg, cgls should be disabled too. */
5830 		data &= ~(RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK |
5831 			  RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK);
5832 		if (temp != data)
5833 			WREG32(mmRLC_CGCG_CGLS_CTRL, data);
5834 		/* enable interrupts again for PG */
5835 		gfx_v8_0_enable_gui_idle_interrupt(adev, true);
5836 	}
5837 
5838 	gfx_v8_0_wait_for_rlc_serdes(adev);
5839 
5840 	amdgpu_gfx_rlc_exit_safe_mode(adev);
5841 }
5842 static int gfx_v8_0_update_gfx_clock_gating(struct amdgpu_device *adev,
5843 					    bool enable)
5844 {
5845 	if (enable) {
5846 		/* CGCG/CGLS should be enabled after MGCG/MGLS/TS(CG/LS)
5847 		 * ===  MGCG + MGLS + TS(CG/LS) ===
5848 		 */
5849 		gfx_v8_0_update_medium_grain_clock_gating(adev, enable);
5850 		gfx_v8_0_update_coarse_grain_clock_gating(adev, enable);
5851 	} else {
5852 		/* CGCG/CGLS should be disabled before MGCG/MGLS/TS(CG/LS)
5853 		 * ===  CGCG + CGLS ===
5854 		 */
5855 		gfx_v8_0_update_coarse_grain_clock_gating(adev, enable);
5856 		gfx_v8_0_update_medium_grain_clock_gating(adev, enable);
5857 	}
5858 	return 0;
5859 }
5860 
5861 static int gfx_v8_0_tonga_update_gfx_clock_gating(struct amdgpu_device *adev,
5862 					  enum amd_clockgating_state state)
5863 {
5864 	uint32_t msg_id, pp_state = 0;
5865 	uint32_t pp_support_state = 0;
5866 
5867 	if (adev->cg_flags & (AMD_CG_SUPPORT_GFX_CGCG | AMD_CG_SUPPORT_GFX_CGLS)) {
5868 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGLS) {
5869 			pp_support_state = PP_STATE_SUPPORT_LS;
5870 			pp_state = PP_STATE_LS;
5871 		}
5872 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGCG) {
5873 			pp_support_state |= PP_STATE_SUPPORT_CG;
5874 			pp_state |= PP_STATE_CG;
5875 		}
5876 		if (state == AMD_CG_STATE_UNGATE)
5877 			pp_state = 0;
5878 
5879 		msg_id = PP_CG_MSG_ID(PP_GROUP_GFX,
5880 				PP_BLOCK_GFX_CG,
5881 				pp_support_state,
5882 				pp_state);
5883 		amdgpu_dpm_set_clockgating_by_smu(adev, msg_id);
5884 	}
5885 
5886 	if (adev->cg_flags & (AMD_CG_SUPPORT_GFX_MGCG | AMD_CG_SUPPORT_GFX_MGLS)) {
5887 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS) {
5888 			pp_support_state = PP_STATE_SUPPORT_LS;
5889 			pp_state = PP_STATE_LS;
5890 		}
5891 
5892 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGCG) {
5893 			pp_support_state |= PP_STATE_SUPPORT_CG;
5894 			pp_state |= PP_STATE_CG;
5895 		}
5896 
5897 		if (state == AMD_CG_STATE_UNGATE)
5898 			pp_state = 0;
5899 
5900 		msg_id = PP_CG_MSG_ID(PP_GROUP_GFX,
5901 				PP_BLOCK_GFX_MG,
5902 				pp_support_state,
5903 				pp_state);
5904 		amdgpu_dpm_set_clockgating_by_smu(adev, msg_id);
5905 	}
5906 
5907 	return 0;
5908 }
5909 
5910 static int gfx_v8_0_polaris_update_gfx_clock_gating(struct amdgpu_device *adev,
5911 					  enum amd_clockgating_state state)
5912 {
5913 
5914 	uint32_t msg_id, pp_state = 0;
5915 	uint32_t pp_support_state = 0;
5916 
5917 	if (adev->cg_flags & (AMD_CG_SUPPORT_GFX_CGCG | AMD_CG_SUPPORT_GFX_CGLS)) {
5918 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGLS) {
5919 			pp_support_state = PP_STATE_SUPPORT_LS;
5920 			pp_state = PP_STATE_LS;
5921 		}
5922 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGCG) {
5923 			pp_support_state |= PP_STATE_SUPPORT_CG;
5924 			pp_state |= PP_STATE_CG;
5925 		}
5926 		if (state == AMD_CG_STATE_UNGATE)
5927 			pp_state = 0;
5928 
5929 		msg_id = PP_CG_MSG_ID(PP_GROUP_GFX,
5930 				PP_BLOCK_GFX_CG,
5931 				pp_support_state,
5932 				pp_state);
5933 		amdgpu_dpm_set_clockgating_by_smu(adev, msg_id);
5934 	}
5935 
5936 	if (adev->cg_flags & (AMD_CG_SUPPORT_GFX_3D_CGCG | AMD_CG_SUPPORT_GFX_3D_CGLS)) {
5937 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_3D_CGLS) {
5938 			pp_support_state = PP_STATE_SUPPORT_LS;
5939 			pp_state = PP_STATE_LS;
5940 		}
5941 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_3D_CGCG) {
5942 			pp_support_state |= PP_STATE_SUPPORT_CG;
5943 			pp_state |= PP_STATE_CG;
5944 		}
5945 		if (state == AMD_CG_STATE_UNGATE)
5946 			pp_state = 0;
5947 
5948 		msg_id = PP_CG_MSG_ID(PP_GROUP_GFX,
5949 				PP_BLOCK_GFX_3D,
5950 				pp_support_state,
5951 				pp_state);
5952 		amdgpu_dpm_set_clockgating_by_smu(adev, msg_id);
5953 	}
5954 
5955 	if (adev->cg_flags & (AMD_CG_SUPPORT_GFX_MGCG | AMD_CG_SUPPORT_GFX_MGLS)) {
5956 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS) {
5957 			pp_support_state = PP_STATE_SUPPORT_LS;
5958 			pp_state = PP_STATE_LS;
5959 		}
5960 
5961 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGCG) {
5962 			pp_support_state |= PP_STATE_SUPPORT_CG;
5963 			pp_state |= PP_STATE_CG;
5964 		}
5965 
5966 		if (state == AMD_CG_STATE_UNGATE)
5967 			pp_state = 0;
5968 
5969 		msg_id = PP_CG_MSG_ID(PP_GROUP_GFX,
5970 				PP_BLOCK_GFX_MG,
5971 				pp_support_state,
5972 				pp_state);
5973 		amdgpu_dpm_set_clockgating_by_smu(adev, msg_id);
5974 	}
5975 
5976 	if (adev->cg_flags & AMD_CG_SUPPORT_GFX_RLC_LS) {
5977 		pp_support_state = PP_STATE_SUPPORT_LS;
5978 
5979 		if (state == AMD_CG_STATE_UNGATE)
5980 			pp_state = 0;
5981 		else
5982 			pp_state = PP_STATE_LS;
5983 
5984 		msg_id = PP_CG_MSG_ID(PP_GROUP_GFX,
5985 				PP_BLOCK_GFX_RLC,
5986 				pp_support_state,
5987 				pp_state);
5988 		amdgpu_dpm_set_clockgating_by_smu(adev, msg_id);
5989 	}
5990 
5991 	if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CP_LS) {
5992 		pp_support_state = PP_STATE_SUPPORT_LS;
5993 
5994 		if (state == AMD_CG_STATE_UNGATE)
5995 			pp_state = 0;
5996 		else
5997 			pp_state = PP_STATE_LS;
5998 		msg_id = PP_CG_MSG_ID(PP_GROUP_GFX,
5999 			PP_BLOCK_GFX_CP,
6000 			pp_support_state,
6001 			pp_state);
6002 		amdgpu_dpm_set_clockgating_by_smu(adev, msg_id);
6003 	}
6004 
6005 	return 0;
6006 }
6007 
6008 static int gfx_v8_0_set_clockgating_state(void *handle,
6009 					  enum amd_clockgating_state state)
6010 {
6011 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
6012 
6013 	if (amdgpu_sriov_vf(adev))
6014 		return 0;
6015 
6016 	switch (adev->asic_type) {
6017 	case CHIP_FIJI:
6018 	case CHIP_CARRIZO:
6019 	case CHIP_STONEY:
6020 		gfx_v8_0_update_gfx_clock_gating(adev,
6021 						 state == AMD_CG_STATE_GATE);
6022 		break;
6023 	case CHIP_TONGA:
6024 		gfx_v8_0_tonga_update_gfx_clock_gating(adev, state);
6025 		break;
6026 	case CHIP_POLARIS10:
6027 	case CHIP_POLARIS11:
6028 	case CHIP_POLARIS12:
6029 	case CHIP_VEGAM:
6030 		gfx_v8_0_polaris_update_gfx_clock_gating(adev, state);
6031 		break;
6032 	default:
6033 		break;
6034 	}
6035 	return 0;
6036 }
6037 
6038 static u64 gfx_v8_0_ring_get_rptr(struct amdgpu_ring *ring)
6039 {
6040 	return *ring->rptr_cpu_addr;
6041 }
6042 
6043 static u64 gfx_v8_0_ring_get_wptr_gfx(struct amdgpu_ring *ring)
6044 {
6045 	struct amdgpu_device *adev = ring->adev;
6046 
6047 	if (ring->use_doorbell)
6048 		/* XXX check if swapping is necessary on BE */
6049 		return *ring->wptr_cpu_addr;
6050 	else
6051 		return RREG32(mmCP_RB0_WPTR);
6052 }
6053 
6054 static void gfx_v8_0_ring_set_wptr_gfx(struct amdgpu_ring *ring)
6055 {
6056 	struct amdgpu_device *adev = ring->adev;
6057 
6058 	if (ring->use_doorbell) {
6059 		/* XXX check if swapping is necessary on BE */
6060 		*ring->wptr_cpu_addr = lower_32_bits(ring->wptr);
6061 		WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
6062 	} else {
6063 		WREG32(mmCP_RB0_WPTR, lower_32_bits(ring->wptr));
6064 		(void)RREG32(mmCP_RB0_WPTR);
6065 	}
6066 }
6067 
6068 static void gfx_v8_0_ring_emit_hdp_flush(struct amdgpu_ring *ring)
6069 {
6070 	u32 ref_and_mask, reg_mem_engine;
6071 
6072 	if ((ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE) ||
6073 	    (ring->funcs->type == AMDGPU_RING_TYPE_KIQ)) {
6074 		switch (ring->me) {
6075 		case 1:
6076 			ref_and_mask = GPU_HDP_FLUSH_DONE__CP2_MASK << ring->pipe;
6077 			break;
6078 		case 2:
6079 			ref_and_mask = GPU_HDP_FLUSH_DONE__CP6_MASK << ring->pipe;
6080 			break;
6081 		default:
6082 			return;
6083 		}
6084 		reg_mem_engine = 0;
6085 	} else {
6086 		ref_and_mask = GPU_HDP_FLUSH_DONE__CP0_MASK;
6087 		reg_mem_engine = WAIT_REG_MEM_ENGINE(1); /* pfp */
6088 	}
6089 
6090 	amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
6091 	amdgpu_ring_write(ring, (WAIT_REG_MEM_OPERATION(1) | /* write, wait, write */
6092 				 WAIT_REG_MEM_FUNCTION(3) |  /* == */
6093 				 reg_mem_engine));
6094 	amdgpu_ring_write(ring, mmGPU_HDP_FLUSH_REQ);
6095 	amdgpu_ring_write(ring, mmGPU_HDP_FLUSH_DONE);
6096 	amdgpu_ring_write(ring, ref_and_mask);
6097 	amdgpu_ring_write(ring, ref_and_mask);
6098 	amdgpu_ring_write(ring, 0x20); /* poll interval */
6099 }
6100 
6101 static void gfx_v8_0_ring_emit_vgt_flush(struct amdgpu_ring *ring)
6102 {
6103 	amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE, 0));
6104 	amdgpu_ring_write(ring, EVENT_TYPE(VS_PARTIAL_FLUSH) |
6105 		EVENT_INDEX(4));
6106 
6107 	amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE, 0));
6108 	amdgpu_ring_write(ring, EVENT_TYPE(VGT_FLUSH) |
6109 		EVENT_INDEX(0));
6110 }
6111 
6112 static void gfx_v8_0_ring_emit_ib_gfx(struct amdgpu_ring *ring,
6113 					struct amdgpu_job *job,
6114 					struct amdgpu_ib *ib,
6115 					uint32_t flags)
6116 {
6117 	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
6118 	u32 header, control = 0;
6119 
6120 	if (ib->flags & AMDGPU_IB_FLAG_CE)
6121 		header = PACKET3(PACKET3_INDIRECT_BUFFER_CONST, 2);
6122 	else
6123 		header = PACKET3(PACKET3_INDIRECT_BUFFER, 2);
6124 
6125 	control |= ib->length_dw | (vmid << 24);
6126 
6127 	if (amdgpu_sriov_vf(ring->adev) && (ib->flags & AMDGPU_IB_FLAG_PREEMPT)) {
6128 		control |= INDIRECT_BUFFER_PRE_ENB(1);
6129 
6130 		if (!(ib->flags & AMDGPU_IB_FLAG_CE) && vmid)
6131 			gfx_v8_0_ring_emit_de_meta(ring);
6132 	}
6133 
6134 	amdgpu_ring_write(ring, header);
6135 	amdgpu_ring_write(ring,
6136 #ifdef __BIG_ENDIAN
6137 			  (2 << 0) |
6138 #endif
6139 			  (ib->gpu_addr & 0xFFFFFFFC));
6140 	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr) & 0xFFFF);
6141 	amdgpu_ring_write(ring, control);
6142 }
6143 
6144 static void gfx_v8_0_ring_emit_ib_compute(struct amdgpu_ring *ring,
6145 					  struct amdgpu_job *job,
6146 					  struct amdgpu_ib *ib,
6147 					  uint32_t flags)
6148 {
6149 	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
6150 	u32 control = INDIRECT_BUFFER_VALID | ib->length_dw | (vmid << 24);
6151 
6152 	/* Currently, there is a high possibility to get wave ID mismatch
6153 	 * between ME and GDS, leading to a hw deadlock, because ME generates
6154 	 * different wave IDs than the GDS expects. This situation happens
6155 	 * randomly when at least 5 compute pipes use GDS ordered append.
6156 	 * The wave IDs generated by ME are also wrong after suspend/resume.
6157 	 * Those are probably bugs somewhere else in the kernel driver.
6158 	 *
6159 	 * Writing GDS_COMPUTE_MAX_WAVE_ID resets wave ID counters in ME and
6160 	 * GDS to 0 for this ring (me/pipe).
6161 	 */
6162 	if (ib->flags & AMDGPU_IB_FLAG_RESET_GDS_MAX_WAVE_ID) {
6163 		amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1));
6164 		amdgpu_ring_write(ring, mmGDS_COMPUTE_MAX_WAVE_ID - PACKET3_SET_CONFIG_REG_START);
6165 		amdgpu_ring_write(ring, ring->adev->gds.gds_compute_max_wave_id);
6166 	}
6167 
6168 	amdgpu_ring_write(ring, PACKET3(PACKET3_INDIRECT_BUFFER, 2));
6169 	amdgpu_ring_write(ring,
6170 #ifdef __BIG_ENDIAN
6171 				(2 << 0) |
6172 #endif
6173 				(ib->gpu_addr & 0xFFFFFFFC));
6174 	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr) & 0xFFFF);
6175 	amdgpu_ring_write(ring, control);
6176 }
6177 
6178 static void gfx_v8_0_ring_emit_fence_gfx(struct amdgpu_ring *ring, u64 addr,
6179 					 u64 seq, unsigned flags)
6180 {
6181 	bool write64bit = flags & AMDGPU_FENCE_FLAG_64BIT;
6182 	bool int_sel = flags & AMDGPU_FENCE_FLAG_INT;
6183 
6184 	/* Workaround for cache flush problems. First send a dummy EOP
6185 	 * event down the pipe with seq one below.
6186 	 */
6187 	amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE_EOP, 4));
6188 	amdgpu_ring_write(ring, (EOP_TCL1_ACTION_EN |
6189 				 EOP_TC_ACTION_EN |
6190 				 EOP_TC_WB_ACTION_EN |
6191 				 EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
6192 				 EVENT_INDEX(5)));
6193 	amdgpu_ring_write(ring, addr & 0xfffffffc);
6194 	amdgpu_ring_write(ring, (upper_32_bits(addr) & 0xffff) |
6195 				DATA_SEL(1) | INT_SEL(0));
6196 	amdgpu_ring_write(ring, lower_32_bits(seq - 1));
6197 	amdgpu_ring_write(ring, upper_32_bits(seq - 1));
6198 
6199 	/* Then send the real EOP event down the pipe:
6200 	 * EVENT_WRITE_EOP - flush caches, send int */
6201 	amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE_EOP, 4));
6202 	amdgpu_ring_write(ring, (EOP_TCL1_ACTION_EN |
6203 				 EOP_TC_ACTION_EN |
6204 				 EOP_TC_WB_ACTION_EN |
6205 				 EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
6206 				 EVENT_INDEX(5)));
6207 	amdgpu_ring_write(ring, addr & 0xfffffffc);
6208 	amdgpu_ring_write(ring, (upper_32_bits(addr) & 0xffff) |
6209 			  DATA_SEL(write64bit ? 2 : 1) | INT_SEL(int_sel ? 2 : 0));
6210 	amdgpu_ring_write(ring, lower_32_bits(seq));
6211 	amdgpu_ring_write(ring, upper_32_bits(seq));
6212 
6213 }
6214 
6215 static void gfx_v8_0_ring_emit_pipeline_sync(struct amdgpu_ring *ring)
6216 {
6217 	int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX);
6218 	uint32_t seq = ring->fence_drv.sync_seq;
6219 	uint64_t addr = ring->fence_drv.gpu_addr;
6220 
6221 	amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
6222 	amdgpu_ring_write(ring, (WAIT_REG_MEM_MEM_SPACE(1) | /* memory */
6223 				 WAIT_REG_MEM_FUNCTION(3) | /* equal */
6224 				 WAIT_REG_MEM_ENGINE(usepfp))); /* pfp or me */
6225 	amdgpu_ring_write(ring, addr & 0xfffffffc);
6226 	amdgpu_ring_write(ring, upper_32_bits(addr) & 0xffffffff);
6227 	amdgpu_ring_write(ring, seq);
6228 	amdgpu_ring_write(ring, 0xffffffff);
6229 	amdgpu_ring_write(ring, 4); /* poll interval */
6230 }
6231 
6232 static void gfx_v8_0_ring_emit_vm_flush(struct amdgpu_ring *ring,
6233 					unsigned vmid, uint64_t pd_addr)
6234 {
6235 	int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX);
6236 
6237 	amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr);
6238 
6239 	/* wait for the invalidate to complete */
6240 	amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
6241 	amdgpu_ring_write(ring, (WAIT_REG_MEM_OPERATION(0) | /* wait */
6242 				 WAIT_REG_MEM_FUNCTION(0) |  /* always */
6243 				 WAIT_REG_MEM_ENGINE(0))); /* me */
6244 	amdgpu_ring_write(ring, mmVM_INVALIDATE_REQUEST);
6245 	amdgpu_ring_write(ring, 0);
6246 	amdgpu_ring_write(ring, 0); /* ref */
6247 	amdgpu_ring_write(ring, 0); /* mask */
6248 	amdgpu_ring_write(ring, 0x20); /* poll interval */
6249 
6250 	/* compute doesn't have PFP */
6251 	if (usepfp) {
6252 		/* sync PFP to ME, otherwise we might get invalid PFP reads */
6253 		amdgpu_ring_write(ring, PACKET3(PACKET3_PFP_SYNC_ME, 0));
6254 		amdgpu_ring_write(ring, 0x0);
6255 	}
6256 }
6257 
6258 static u64 gfx_v8_0_ring_get_wptr_compute(struct amdgpu_ring *ring)
6259 {
6260 	return *ring->wptr_cpu_addr;
6261 }
6262 
6263 static void gfx_v8_0_ring_set_wptr_compute(struct amdgpu_ring *ring)
6264 {
6265 	struct amdgpu_device *adev = ring->adev;
6266 
6267 	/* XXX check if swapping is necessary on BE */
6268 	*ring->wptr_cpu_addr = lower_32_bits(ring->wptr);
6269 	WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
6270 }
6271 
6272 static void gfx_v8_0_ring_emit_fence_compute(struct amdgpu_ring *ring,
6273 					     u64 addr, u64 seq,
6274 					     unsigned flags)
6275 {
6276 	bool write64bit = flags & AMDGPU_FENCE_FLAG_64BIT;
6277 	bool int_sel = flags & AMDGPU_FENCE_FLAG_INT;
6278 
6279 	/* RELEASE_MEM - flush caches, send int */
6280 	amdgpu_ring_write(ring, PACKET3(PACKET3_RELEASE_MEM, 5));
6281 	amdgpu_ring_write(ring, (EOP_TCL1_ACTION_EN |
6282 				 EOP_TC_ACTION_EN |
6283 				 EOP_TC_WB_ACTION_EN |
6284 				 EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
6285 				 EVENT_INDEX(5)));
6286 	amdgpu_ring_write(ring, DATA_SEL(write64bit ? 2 : 1) | INT_SEL(int_sel ? 2 : 0));
6287 	amdgpu_ring_write(ring, addr & 0xfffffffc);
6288 	amdgpu_ring_write(ring, upper_32_bits(addr));
6289 	amdgpu_ring_write(ring, lower_32_bits(seq));
6290 	amdgpu_ring_write(ring, upper_32_bits(seq));
6291 }
6292 
6293 static void gfx_v8_0_ring_emit_fence_kiq(struct amdgpu_ring *ring, u64 addr,
6294 					 u64 seq, unsigned int flags)
6295 {
6296 	/* we only allocate 32bit for each seq wb address */
6297 	BUG_ON(flags & AMDGPU_FENCE_FLAG_64BIT);
6298 
6299 	/* write fence seq to the "addr" */
6300 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
6301 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
6302 				 WRITE_DATA_DST_SEL(5) | WR_CONFIRM));
6303 	amdgpu_ring_write(ring, lower_32_bits(addr));
6304 	amdgpu_ring_write(ring, upper_32_bits(addr));
6305 	amdgpu_ring_write(ring, lower_32_bits(seq));
6306 
6307 	if (flags & AMDGPU_FENCE_FLAG_INT) {
6308 		/* set register to trigger INT */
6309 		amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
6310 		amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
6311 					 WRITE_DATA_DST_SEL(0) | WR_CONFIRM));
6312 		amdgpu_ring_write(ring, mmCPC_INT_STATUS);
6313 		amdgpu_ring_write(ring, 0);
6314 		amdgpu_ring_write(ring, 0x20000000); /* src_id is 178 */
6315 	}
6316 }
6317 
6318 static void gfx_v8_ring_emit_sb(struct amdgpu_ring *ring)
6319 {
6320 	amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
6321 	amdgpu_ring_write(ring, 0);
6322 }
6323 
6324 static void gfx_v8_ring_emit_cntxcntl(struct amdgpu_ring *ring, uint32_t flags)
6325 {
6326 	uint32_t dw2 = 0;
6327 
6328 	if (amdgpu_sriov_vf(ring->adev))
6329 		gfx_v8_0_ring_emit_ce_meta(ring);
6330 
6331 	dw2 |= 0x80000000; /* set load_enable otherwise this package is just NOPs */
6332 	if (flags & AMDGPU_HAVE_CTX_SWITCH) {
6333 		gfx_v8_0_ring_emit_vgt_flush(ring);
6334 		/* set load_global_config & load_global_uconfig */
6335 		dw2 |= 0x8001;
6336 		/* set load_cs_sh_regs */
6337 		dw2 |= 0x01000000;
6338 		/* set load_per_context_state & load_gfx_sh_regs for GFX */
6339 		dw2 |= 0x10002;
6340 
6341 		/* set load_ce_ram if preamble presented */
6342 		if (AMDGPU_PREAMBLE_IB_PRESENT & flags)
6343 			dw2 |= 0x10000000;
6344 	} else {
6345 		/* still load_ce_ram if this is the first time preamble presented
6346 		 * although there is no context switch happens.
6347 		 */
6348 		if (AMDGPU_PREAMBLE_IB_PRESENT_FIRST & flags)
6349 			dw2 |= 0x10000000;
6350 	}
6351 
6352 	amdgpu_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1));
6353 	amdgpu_ring_write(ring, dw2);
6354 	amdgpu_ring_write(ring, 0);
6355 }
6356 
6357 static unsigned gfx_v8_0_ring_emit_init_cond_exec(struct amdgpu_ring *ring)
6358 {
6359 	unsigned ret;
6360 
6361 	amdgpu_ring_write(ring, PACKET3(PACKET3_COND_EXEC, 3));
6362 	amdgpu_ring_write(ring, lower_32_bits(ring->cond_exe_gpu_addr));
6363 	amdgpu_ring_write(ring, upper_32_bits(ring->cond_exe_gpu_addr));
6364 	amdgpu_ring_write(ring, 0); /* discard following DWs if *cond_exec_gpu_addr==0 */
6365 	ret = ring->wptr & ring->buf_mask;
6366 	amdgpu_ring_write(ring, 0x55aa55aa); /* patch dummy value later */
6367 	return ret;
6368 }
6369 
6370 static void gfx_v8_0_ring_emit_patch_cond_exec(struct amdgpu_ring *ring, unsigned offset)
6371 {
6372 	unsigned cur;
6373 
6374 	BUG_ON(offset > ring->buf_mask);
6375 	BUG_ON(ring->ring[offset] != 0x55aa55aa);
6376 
6377 	cur = (ring->wptr & ring->buf_mask) - 1;
6378 	if (likely(cur > offset))
6379 		ring->ring[offset] = cur - offset;
6380 	else
6381 		ring->ring[offset] = (ring->ring_size >> 2) - offset + cur;
6382 }
6383 
6384 static void gfx_v8_0_ring_emit_rreg(struct amdgpu_ring *ring, uint32_t reg,
6385 				    uint32_t reg_val_offs)
6386 {
6387 	struct amdgpu_device *adev = ring->adev;
6388 
6389 	amdgpu_ring_write(ring, PACKET3(PACKET3_COPY_DATA, 4));
6390 	amdgpu_ring_write(ring, 0 |	/* src: register*/
6391 				(5 << 8) |	/* dst: memory */
6392 				(1 << 20));	/* write confirm */
6393 	amdgpu_ring_write(ring, reg);
6394 	amdgpu_ring_write(ring, 0);
6395 	amdgpu_ring_write(ring, lower_32_bits(adev->wb.gpu_addr +
6396 				reg_val_offs * 4));
6397 	amdgpu_ring_write(ring, upper_32_bits(adev->wb.gpu_addr +
6398 				reg_val_offs * 4));
6399 }
6400 
6401 static void gfx_v8_0_ring_emit_wreg(struct amdgpu_ring *ring, uint32_t reg,
6402 				  uint32_t val)
6403 {
6404 	uint32_t cmd;
6405 
6406 	switch (ring->funcs->type) {
6407 	case AMDGPU_RING_TYPE_GFX:
6408 		cmd = WRITE_DATA_ENGINE_SEL(1) | WR_CONFIRM;
6409 		break;
6410 	case AMDGPU_RING_TYPE_KIQ:
6411 		cmd = 1 << 16; /* no inc addr */
6412 		break;
6413 	default:
6414 		cmd = WR_CONFIRM;
6415 		break;
6416 	}
6417 
6418 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
6419 	amdgpu_ring_write(ring, cmd);
6420 	amdgpu_ring_write(ring, reg);
6421 	amdgpu_ring_write(ring, 0);
6422 	amdgpu_ring_write(ring, val);
6423 }
6424 
6425 static void gfx_v8_0_ring_soft_recovery(struct amdgpu_ring *ring, unsigned vmid)
6426 {
6427 	struct amdgpu_device *adev = ring->adev;
6428 	uint32_t value = 0;
6429 
6430 	value = REG_SET_FIELD(value, SQ_CMD, CMD, 0x03);
6431 	value = REG_SET_FIELD(value, SQ_CMD, MODE, 0x01);
6432 	value = REG_SET_FIELD(value, SQ_CMD, CHECK_VMID, 1);
6433 	value = REG_SET_FIELD(value, SQ_CMD, VM_ID, vmid);
6434 	WREG32(mmSQ_CMD, value);
6435 }
6436 
6437 static void gfx_v8_0_set_gfx_eop_interrupt_state(struct amdgpu_device *adev,
6438 						 enum amdgpu_interrupt_state state)
6439 {
6440 	WREG32_FIELD(CP_INT_CNTL_RING0, TIME_STAMP_INT_ENABLE,
6441 		     state == AMDGPU_IRQ_STATE_DISABLE ? 0 : 1);
6442 }
6443 
6444 static void gfx_v8_0_set_compute_eop_interrupt_state(struct amdgpu_device *adev,
6445 						     int me, int pipe,
6446 						     enum amdgpu_interrupt_state state)
6447 {
6448 	u32 mec_int_cntl, mec_int_cntl_reg;
6449 
6450 	/*
6451 	 * amdgpu controls only the first MEC. That's why this function only
6452 	 * handles the setting of interrupts for this specific MEC. All other
6453 	 * pipes' interrupts are set by amdkfd.
6454 	 */
6455 
6456 	if (me == 1) {
6457 		switch (pipe) {
6458 		case 0:
6459 			mec_int_cntl_reg = mmCP_ME1_PIPE0_INT_CNTL;
6460 			break;
6461 		case 1:
6462 			mec_int_cntl_reg = mmCP_ME1_PIPE1_INT_CNTL;
6463 			break;
6464 		case 2:
6465 			mec_int_cntl_reg = mmCP_ME1_PIPE2_INT_CNTL;
6466 			break;
6467 		case 3:
6468 			mec_int_cntl_reg = mmCP_ME1_PIPE3_INT_CNTL;
6469 			break;
6470 		default:
6471 			DRM_DEBUG("invalid pipe %d\n", pipe);
6472 			return;
6473 		}
6474 	} else {
6475 		DRM_DEBUG("invalid me %d\n", me);
6476 		return;
6477 	}
6478 
6479 	switch (state) {
6480 	case AMDGPU_IRQ_STATE_DISABLE:
6481 		mec_int_cntl = RREG32(mec_int_cntl_reg);
6482 		mec_int_cntl &= ~CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK;
6483 		WREG32(mec_int_cntl_reg, mec_int_cntl);
6484 		break;
6485 	case AMDGPU_IRQ_STATE_ENABLE:
6486 		mec_int_cntl = RREG32(mec_int_cntl_reg);
6487 		mec_int_cntl |= CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK;
6488 		WREG32(mec_int_cntl_reg, mec_int_cntl);
6489 		break;
6490 	default:
6491 		break;
6492 	}
6493 }
6494 
6495 static int gfx_v8_0_set_priv_reg_fault_state(struct amdgpu_device *adev,
6496 					     struct amdgpu_irq_src *source,
6497 					     unsigned type,
6498 					     enum amdgpu_interrupt_state state)
6499 {
6500 	WREG32_FIELD(CP_INT_CNTL_RING0, PRIV_REG_INT_ENABLE,
6501 		     state == AMDGPU_IRQ_STATE_DISABLE ? 0 : 1);
6502 
6503 	return 0;
6504 }
6505 
6506 static int gfx_v8_0_set_priv_inst_fault_state(struct amdgpu_device *adev,
6507 					      struct amdgpu_irq_src *source,
6508 					      unsigned type,
6509 					      enum amdgpu_interrupt_state state)
6510 {
6511 	WREG32_FIELD(CP_INT_CNTL_RING0, PRIV_INSTR_INT_ENABLE,
6512 		     state == AMDGPU_IRQ_STATE_DISABLE ? 0 : 1);
6513 
6514 	return 0;
6515 }
6516 
6517 static int gfx_v8_0_set_eop_interrupt_state(struct amdgpu_device *adev,
6518 					    struct amdgpu_irq_src *src,
6519 					    unsigned type,
6520 					    enum amdgpu_interrupt_state state)
6521 {
6522 	switch (type) {
6523 	case AMDGPU_CP_IRQ_GFX_ME0_PIPE0_EOP:
6524 		gfx_v8_0_set_gfx_eop_interrupt_state(adev, state);
6525 		break;
6526 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP:
6527 		gfx_v8_0_set_compute_eop_interrupt_state(adev, 1, 0, state);
6528 		break;
6529 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE1_EOP:
6530 		gfx_v8_0_set_compute_eop_interrupt_state(adev, 1, 1, state);
6531 		break;
6532 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE2_EOP:
6533 		gfx_v8_0_set_compute_eop_interrupt_state(adev, 1, 2, state);
6534 		break;
6535 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE3_EOP:
6536 		gfx_v8_0_set_compute_eop_interrupt_state(adev, 1, 3, state);
6537 		break;
6538 	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE0_EOP:
6539 		gfx_v8_0_set_compute_eop_interrupt_state(adev, 2, 0, state);
6540 		break;
6541 	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE1_EOP:
6542 		gfx_v8_0_set_compute_eop_interrupt_state(adev, 2, 1, state);
6543 		break;
6544 	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE2_EOP:
6545 		gfx_v8_0_set_compute_eop_interrupt_state(adev, 2, 2, state);
6546 		break;
6547 	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE3_EOP:
6548 		gfx_v8_0_set_compute_eop_interrupt_state(adev, 2, 3, state);
6549 		break;
6550 	default:
6551 		break;
6552 	}
6553 	return 0;
6554 }
6555 
6556 static int gfx_v8_0_set_cp_ecc_int_state(struct amdgpu_device *adev,
6557 					 struct amdgpu_irq_src *source,
6558 					 unsigned int type,
6559 					 enum amdgpu_interrupt_state state)
6560 {
6561 	int enable_flag;
6562 
6563 	switch (state) {
6564 	case AMDGPU_IRQ_STATE_DISABLE:
6565 		enable_flag = 0;
6566 		break;
6567 
6568 	case AMDGPU_IRQ_STATE_ENABLE:
6569 		enable_flag = 1;
6570 		break;
6571 
6572 	default:
6573 		return -EINVAL;
6574 	}
6575 
6576 	WREG32_FIELD(CP_INT_CNTL, CP_ECC_ERROR_INT_ENABLE, enable_flag);
6577 	WREG32_FIELD(CP_INT_CNTL_RING0, CP_ECC_ERROR_INT_ENABLE, enable_flag);
6578 	WREG32_FIELD(CP_INT_CNTL_RING1, CP_ECC_ERROR_INT_ENABLE, enable_flag);
6579 	WREG32_FIELD(CP_INT_CNTL_RING2, CP_ECC_ERROR_INT_ENABLE, enable_flag);
6580 	WREG32_FIELD(CPC_INT_CNTL, CP_ECC_ERROR_INT_ENABLE, enable_flag);
6581 	WREG32_FIELD(CP_ME1_PIPE0_INT_CNTL, CP_ECC_ERROR_INT_ENABLE,
6582 		     enable_flag);
6583 	WREG32_FIELD(CP_ME1_PIPE1_INT_CNTL, CP_ECC_ERROR_INT_ENABLE,
6584 		     enable_flag);
6585 	WREG32_FIELD(CP_ME1_PIPE2_INT_CNTL, CP_ECC_ERROR_INT_ENABLE,
6586 		     enable_flag);
6587 	WREG32_FIELD(CP_ME1_PIPE3_INT_CNTL, CP_ECC_ERROR_INT_ENABLE,
6588 		     enable_flag);
6589 	WREG32_FIELD(CP_ME2_PIPE0_INT_CNTL, CP_ECC_ERROR_INT_ENABLE,
6590 		     enable_flag);
6591 	WREG32_FIELD(CP_ME2_PIPE1_INT_CNTL, CP_ECC_ERROR_INT_ENABLE,
6592 		     enable_flag);
6593 	WREG32_FIELD(CP_ME2_PIPE2_INT_CNTL, CP_ECC_ERROR_INT_ENABLE,
6594 		     enable_flag);
6595 	WREG32_FIELD(CP_ME2_PIPE3_INT_CNTL, CP_ECC_ERROR_INT_ENABLE,
6596 		     enable_flag);
6597 
6598 	return 0;
6599 }
6600 
6601 static int gfx_v8_0_set_sq_int_state(struct amdgpu_device *adev,
6602 				     struct amdgpu_irq_src *source,
6603 				     unsigned int type,
6604 				     enum amdgpu_interrupt_state state)
6605 {
6606 	int enable_flag;
6607 
6608 	switch (state) {
6609 	case AMDGPU_IRQ_STATE_DISABLE:
6610 		enable_flag = 1;
6611 		break;
6612 
6613 	case AMDGPU_IRQ_STATE_ENABLE:
6614 		enable_flag = 0;
6615 		break;
6616 
6617 	default:
6618 		return -EINVAL;
6619 	}
6620 
6621 	WREG32_FIELD(SQ_INTERRUPT_MSG_CTRL, STALL,
6622 		     enable_flag);
6623 
6624 	return 0;
6625 }
6626 
6627 static int gfx_v8_0_eop_irq(struct amdgpu_device *adev,
6628 			    struct amdgpu_irq_src *source,
6629 			    struct amdgpu_iv_entry *entry)
6630 {
6631 	int i;
6632 	u8 me_id, pipe_id, queue_id;
6633 	struct amdgpu_ring *ring;
6634 
6635 	DRM_DEBUG("IH: CP EOP\n");
6636 	me_id = (entry->ring_id & 0x0c) >> 2;
6637 	pipe_id = (entry->ring_id & 0x03) >> 0;
6638 	queue_id = (entry->ring_id & 0x70) >> 4;
6639 
6640 	switch (me_id) {
6641 	case 0:
6642 		amdgpu_fence_process(&adev->gfx.gfx_ring[0]);
6643 		break;
6644 	case 1:
6645 	case 2:
6646 		for (i = 0; i < adev->gfx.num_compute_rings; i++) {
6647 			ring = &adev->gfx.compute_ring[i];
6648 			/* Per-queue interrupt is supported for MEC starting from VI.
6649 			  * The interrupt can only be enabled/disabled per pipe instead of per queue.
6650 			  */
6651 			if ((ring->me == me_id) && (ring->pipe == pipe_id) && (ring->queue == queue_id))
6652 				amdgpu_fence_process(ring);
6653 		}
6654 		break;
6655 	}
6656 	return 0;
6657 }
6658 
6659 static void gfx_v8_0_fault(struct amdgpu_device *adev,
6660 			   struct amdgpu_iv_entry *entry)
6661 {
6662 	u8 me_id, pipe_id, queue_id;
6663 	struct amdgpu_ring *ring;
6664 	int i;
6665 
6666 	me_id = (entry->ring_id & 0x0c) >> 2;
6667 	pipe_id = (entry->ring_id & 0x03) >> 0;
6668 	queue_id = (entry->ring_id & 0x70) >> 4;
6669 
6670 	switch (me_id) {
6671 	case 0:
6672 		drm_sched_fault(&adev->gfx.gfx_ring[0].sched);
6673 		break;
6674 	case 1:
6675 	case 2:
6676 		for (i = 0; i < adev->gfx.num_compute_rings; i++) {
6677 			ring = &adev->gfx.compute_ring[i];
6678 			if (ring->me == me_id && ring->pipe == pipe_id &&
6679 			    ring->queue == queue_id)
6680 				drm_sched_fault(&ring->sched);
6681 		}
6682 		break;
6683 	}
6684 }
6685 
6686 static int gfx_v8_0_priv_reg_irq(struct amdgpu_device *adev,
6687 				 struct amdgpu_irq_src *source,
6688 				 struct amdgpu_iv_entry *entry)
6689 {
6690 	DRM_ERROR("Illegal register access in command stream\n");
6691 	gfx_v8_0_fault(adev, entry);
6692 	return 0;
6693 }
6694 
6695 static int gfx_v8_0_priv_inst_irq(struct amdgpu_device *adev,
6696 				  struct amdgpu_irq_src *source,
6697 				  struct amdgpu_iv_entry *entry)
6698 {
6699 	DRM_ERROR("Illegal instruction in command stream\n");
6700 	gfx_v8_0_fault(adev, entry);
6701 	return 0;
6702 }
6703 
6704 static int gfx_v8_0_cp_ecc_error_irq(struct amdgpu_device *adev,
6705 				     struct amdgpu_irq_src *source,
6706 				     struct amdgpu_iv_entry *entry)
6707 {
6708 	DRM_ERROR("CP EDC/ECC error detected.");
6709 	return 0;
6710 }
6711 
6712 static void gfx_v8_0_parse_sq_irq(struct amdgpu_device *adev, unsigned ih_data,
6713 				  bool from_wq)
6714 {
6715 	u32 enc, se_id, sh_id, cu_id;
6716 	char type[20];
6717 	int sq_edc_source = -1;
6718 
6719 	enc = REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_CMN, ENCODING);
6720 	se_id = REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_CMN, SE_ID);
6721 
6722 	switch (enc) {
6723 		case 0:
6724 			DRM_INFO("SQ general purpose intr detected:"
6725 					"se_id %d, immed_overflow %d, host_reg_overflow %d,"
6726 					"host_cmd_overflow %d, cmd_timestamp %d,"
6727 					"reg_timestamp %d, thread_trace_buff_full %d,"
6728 					"wlt %d, thread_trace %d.\n",
6729 					se_id,
6730 					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_AUTO, IMMED_OVERFLOW),
6731 					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_AUTO, HOST_REG_OVERFLOW),
6732 					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_AUTO, HOST_CMD_OVERFLOW),
6733 					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_AUTO, CMD_TIMESTAMP),
6734 					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_AUTO, REG_TIMESTAMP),
6735 					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_AUTO, THREAD_TRACE_BUF_FULL),
6736 					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_AUTO, WLT),
6737 					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_AUTO, THREAD_TRACE)
6738 					);
6739 			break;
6740 		case 1:
6741 		case 2:
6742 
6743 			cu_id = REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_WAVE, CU_ID);
6744 			sh_id = REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_WAVE, SH_ID);
6745 
6746 			/*
6747 			 * This function can be called either directly from ISR
6748 			 * or from BH in which case we can access SQ_EDC_INFO
6749 			 * instance
6750 			 */
6751 			if (from_wq) {
6752 				mutex_lock(&adev->grbm_idx_mutex);
6753 				gfx_v8_0_select_se_sh(adev, se_id, sh_id, cu_id);
6754 
6755 				sq_edc_source = REG_GET_FIELD(RREG32(mmSQ_EDC_INFO), SQ_EDC_INFO, SOURCE);
6756 
6757 				gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
6758 				mutex_unlock(&adev->grbm_idx_mutex);
6759 			}
6760 
6761 			if (enc == 1)
6762 				sprintf(type, "instruction intr");
6763 			else
6764 				sprintf(type, "EDC/ECC error");
6765 
6766 			DRM_INFO(
6767 				"SQ %s detected: "
6768 					"se_id %d, sh_id %d, cu_id %d, simd_id %d, wave_id %d, vm_id %d "
6769 					"trap %s, sq_ed_info.source %s.\n",
6770 					type, se_id, sh_id, cu_id,
6771 					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_WAVE, SIMD_ID),
6772 					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_WAVE, WAVE_ID),
6773 					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_WAVE, VM_ID),
6774 					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_WAVE, PRIV) ? "true" : "false",
6775 					(sq_edc_source != -1) ? sq_edc_source_names[sq_edc_source] : "unavailable"
6776 				);
6777 			break;
6778 		default:
6779 			DRM_ERROR("SQ invalid encoding type\n.");
6780 	}
6781 }
6782 
6783 static void gfx_v8_0_sq_irq_work_func(struct work_struct *work)
6784 {
6785 
6786 	struct amdgpu_device *adev = container_of(work, struct amdgpu_device, gfx.sq_work.work);
6787 	struct sq_work *sq_work = container_of(work, struct sq_work, work);
6788 
6789 	gfx_v8_0_parse_sq_irq(adev, sq_work->ih_data, true);
6790 }
6791 
6792 static int gfx_v8_0_sq_irq(struct amdgpu_device *adev,
6793 			   struct amdgpu_irq_src *source,
6794 			   struct amdgpu_iv_entry *entry)
6795 {
6796 	unsigned ih_data = entry->src_data[0];
6797 
6798 	/*
6799 	 * Try to submit work so SQ_EDC_INFO can be accessed from
6800 	 * BH. If previous work submission hasn't finished yet
6801 	 * just print whatever info is possible directly from the ISR.
6802 	 */
6803 	if (work_pending(&adev->gfx.sq_work.work)) {
6804 		gfx_v8_0_parse_sq_irq(adev, ih_data, false);
6805 	} else {
6806 		adev->gfx.sq_work.ih_data = ih_data;
6807 		schedule_work(&adev->gfx.sq_work.work);
6808 	}
6809 
6810 	return 0;
6811 }
6812 
6813 static void gfx_v8_0_emit_mem_sync(struct amdgpu_ring *ring)
6814 {
6815 	amdgpu_ring_write(ring, PACKET3(PACKET3_SURFACE_SYNC, 3));
6816 	amdgpu_ring_write(ring, PACKET3_TCL1_ACTION_ENA |
6817 			  PACKET3_TC_ACTION_ENA |
6818 			  PACKET3_SH_KCACHE_ACTION_ENA |
6819 			  PACKET3_SH_ICACHE_ACTION_ENA |
6820 			  PACKET3_TC_WB_ACTION_ENA);  /* CP_COHER_CNTL */
6821 	amdgpu_ring_write(ring, 0xffffffff);  /* CP_COHER_SIZE */
6822 	amdgpu_ring_write(ring, 0);  /* CP_COHER_BASE */
6823 	amdgpu_ring_write(ring, 0x0000000A); /* poll interval */
6824 }
6825 
6826 static void gfx_v8_0_emit_mem_sync_compute(struct amdgpu_ring *ring)
6827 {
6828 	amdgpu_ring_write(ring, PACKET3(PACKET3_ACQUIRE_MEM, 5));
6829 	amdgpu_ring_write(ring, PACKET3_TCL1_ACTION_ENA |
6830 			  PACKET3_TC_ACTION_ENA |
6831 			  PACKET3_SH_KCACHE_ACTION_ENA |
6832 			  PACKET3_SH_ICACHE_ACTION_ENA |
6833 			  PACKET3_TC_WB_ACTION_ENA);  /* CP_COHER_CNTL */
6834 	amdgpu_ring_write(ring, 0xffffffff);	/* CP_COHER_SIZE */
6835 	amdgpu_ring_write(ring, 0xff);		/* CP_COHER_SIZE_HI */
6836 	amdgpu_ring_write(ring, 0);		/* CP_COHER_BASE */
6837 	amdgpu_ring_write(ring, 0);		/* CP_COHER_BASE_HI */
6838 	amdgpu_ring_write(ring, 0x0000000A);	/* poll interval */
6839 }
6840 
6841 
6842 /* mmSPI_WCL_PIPE_PERCENT_CS[0-7]_DEFAULT values are same */
6843 #define mmSPI_WCL_PIPE_PERCENT_CS_DEFAULT	0x0000007f
6844 static void gfx_v8_0_emit_wave_limit_cs(struct amdgpu_ring *ring,
6845 					uint32_t pipe, bool enable)
6846 {
6847 	uint32_t val;
6848 	uint32_t wcl_cs_reg;
6849 
6850 	val = enable ? 0x1 : mmSPI_WCL_PIPE_PERCENT_CS_DEFAULT;
6851 
6852 	switch (pipe) {
6853 	case 0:
6854 		wcl_cs_reg = mmSPI_WCL_PIPE_PERCENT_CS0;
6855 		break;
6856 	case 1:
6857 		wcl_cs_reg = mmSPI_WCL_PIPE_PERCENT_CS1;
6858 		break;
6859 	case 2:
6860 		wcl_cs_reg = mmSPI_WCL_PIPE_PERCENT_CS2;
6861 		break;
6862 	case 3:
6863 		wcl_cs_reg = mmSPI_WCL_PIPE_PERCENT_CS3;
6864 		break;
6865 	default:
6866 		DRM_DEBUG("invalid pipe %d\n", pipe);
6867 		return;
6868 	}
6869 
6870 	amdgpu_ring_emit_wreg(ring, wcl_cs_reg, val);
6871 
6872 }
6873 
6874 #define mmSPI_WCL_PIPE_PERCENT_GFX_DEFAULT	0x07ffffff
6875 static void gfx_v8_0_emit_wave_limit(struct amdgpu_ring *ring, bool enable)
6876 {
6877 	struct amdgpu_device *adev = ring->adev;
6878 	uint32_t val;
6879 	int i;
6880 
6881 	/* mmSPI_WCL_PIPE_PERCENT_GFX is 7 bit multiplier register to limit
6882 	 * number of gfx waves. Setting 5 bit will make sure gfx only gets
6883 	 * around 25% of gpu resources.
6884 	 */
6885 	val = enable ? 0x1f : mmSPI_WCL_PIPE_PERCENT_GFX_DEFAULT;
6886 	amdgpu_ring_emit_wreg(ring, mmSPI_WCL_PIPE_PERCENT_GFX, val);
6887 
6888 	/* Restrict waves for normal/low priority compute queues as well
6889 	 * to get best QoS for high priority compute jobs.
6890 	 *
6891 	 * amdgpu controls only 1st ME(0-3 CS pipes).
6892 	 */
6893 	for (i = 0; i < adev->gfx.mec.num_pipe_per_mec; i++) {
6894 		if (i != ring->pipe)
6895 			gfx_v8_0_emit_wave_limit_cs(ring, i, enable);
6896 
6897 	}
6898 
6899 }
6900 
6901 static const struct amd_ip_funcs gfx_v8_0_ip_funcs = {
6902 	.name = "gfx_v8_0",
6903 	.early_init = gfx_v8_0_early_init,
6904 	.late_init = gfx_v8_0_late_init,
6905 	.sw_init = gfx_v8_0_sw_init,
6906 	.sw_fini = gfx_v8_0_sw_fini,
6907 	.hw_init = gfx_v8_0_hw_init,
6908 	.hw_fini = gfx_v8_0_hw_fini,
6909 	.suspend = gfx_v8_0_suspend,
6910 	.resume = gfx_v8_0_resume,
6911 	.is_idle = gfx_v8_0_is_idle,
6912 	.wait_for_idle = gfx_v8_0_wait_for_idle,
6913 	.check_soft_reset = gfx_v8_0_check_soft_reset,
6914 	.pre_soft_reset = gfx_v8_0_pre_soft_reset,
6915 	.soft_reset = gfx_v8_0_soft_reset,
6916 	.post_soft_reset = gfx_v8_0_post_soft_reset,
6917 	.set_clockgating_state = gfx_v8_0_set_clockgating_state,
6918 	.set_powergating_state = gfx_v8_0_set_powergating_state,
6919 	.get_clockgating_state = gfx_v8_0_get_clockgating_state,
6920 };
6921 
6922 static const struct amdgpu_ring_funcs gfx_v8_0_ring_funcs_gfx = {
6923 	.type = AMDGPU_RING_TYPE_GFX,
6924 	.align_mask = 0xff,
6925 	.nop = PACKET3(PACKET3_NOP, 0x3FFF),
6926 	.support_64bit_ptrs = false,
6927 	.get_rptr = gfx_v8_0_ring_get_rptr,
6928 	.get_wptr = gfx_v8_0_ring_get_wptr_gfx,
6929 	.set_wptr = gfx_v8_0_ring_set_wptr_gfx,
6930 	.emit_frame_size = /* maximum 215dw if count 16 IBs in */
6931 		5 +  /* COND_EXEC */
6932 		7 +  /* PIPELINE_SYNC */
6933 		VI_FLUSH_GPU_TLB_NUM_WREG * 5 + 9 + /* VM_FLUSH */
6934 		12 +  /* FENCE for VM_FLUSH */
6935 		20 + /* GDS switch */
6936 		4 + /* double SWITCH_BUFFER,
6937 		       the first COND_EXEC jump to the place just
6938 			   prior to this double SWITCH_BUFFER  */
6939 		5 + /* COND_EXEC */
6940 		7 +	 /*	HDP_flush */
6941 		4 +	 /*	VGT_flush */
6942 		14 + /*	CE_META */
6943 		31 + /*	DE_META */
6944 		3 + /* CNTX_CTRL */
6945 		5 + /* HDP_INVL */
6946 		12 + 12 + /* FENCE x2 */
6947 		2 + /* SWITCH_BUFFER */
6948 		5, /* SURFACE_SYNC */
6949 	.emit_ib_size =	4, /* gfx_v8_0_ring_emit_ib_gfx */
6950 	.emit_ib = gfx_v8_0_ring_emit_ib_gfx,
6951 	.emit_fence = gfx_v8_0_ring_emit_fence_gfx,
6952 	.emit_pipeline_sync = gfx_v8_0_ring_emit_pipeline_sync,
6953 	.emit_vm_flush = gfx_v8_0_ring_emit_vm_flush,
6954 	.emit_gds_switch = gfx_v8_0_ring_emit_gds_switch,
6955 	.emit_hdp_flush = gfx_v8_0_ring_emit_hdp_flush,
6956 	.test_ring = gfx_v8_0_ring_test_ring,
6957 	.test_ib = gfx_v8_0_ring_test_ib,
6958 	.insert_nop = amdgpu_ring_insert_nop,
6959 	.pad_ib = amdgpu_ring_generic_pad_ib,
6960 	.emit_switch_buffer = gfx_v8_ring_emit_sb,
6961 	.emit_cntxcntl = gfx_v8_ring_emit_cntxcntl,
6962 	.init_cond_exec = gfx_v8_0_ring_emit_init_cond_exec,
6963 	.patch_cond_exec = gfx_v8_0_ring_emit_patch_cond_exec,
6964 	.emit_wreg = gfx_v8_0_ring_emit_wreg,
6965 	.soft_recovery = gfx_v8_0_ring_soft_recovery,
6966 	.emit_mem_sync = gfx_v8_0_emit_mem_sync,
6967 };
6968 
6969 static const struct amdgpu_ring_funcs gfx_v8_0_ring_funcs_compute = {
6970 	.type = AMDGPU_RING_TYPE_COMPUTE,
6971 	.align_mask = 0xff,
6972 	.nop = PACKET3(PACKET3_NOP, 0x3FFF),
6973 	.support_64bit_ptrs = false,
6974 	.get_rptr = gfx_v8_0_ring_get_rptr,
6975 	.get_wptr = gfx_v8_0_ring_get_wptr_compute,
6976 	.set_wptr = gfx_v8_0_ring_set_wptr_compute,
6977 	.emit_frame_size =
6978 		20 + /* gfx_v8_0_ring_emit_gds_switch */
6979 		7 + /* gfx_v8_0_ring_emit_hdp_flush */
6980 		5 + /* hdp_invalidate */
6981 		7 + /* gfx_v8_0_ring_emit_pipeline_sync */
6982 		VI_FLUSH_GPU_TLB_NUM_WREG * 5 + 7 + /* gfx_v8_0_ring_emit_vm_flush */
6983 		7 + 7 + 7 + /* gfx_v8_0_ring_emit_fence_compute x3 for user fence, vm fence */
6984 		7 + /* gfx_v8_0_emit_mem_sync_compute */
6985 		5 + /* gfx_v8_0_emit_wave_limit for updating mmSPI_WCL_PIPE_PERCENT_GFX register */
6986 		15, /* for updating 3 mmSPI_WCL_PIPE_PERCENT_CS registers */
6987 	.emit_ib_size =	7, /* gfx_v8_0_ring_emit_ib_compute */
6988 	.emit_ib = gfx_v8_0_ring_emit_ib_compute,
6989 	.emit_fence = gfx_v8_0_ring_emit_fence_compute,
6990 	.emit_pipeline_sync = gfx_v8_0_ring_emit_pipeline_sync,
6991 	.emit_vm_flush = gfx_v8_0_ring_emit_vm_flush,
6992 	.emit_gds_switch = gfx_v8_0_ring_emit_gds_switch,
6993 	.emit_hdp_flush = gfx_v8_0_ring_emit_hdp_flush,
6994 	.test_ring = gfx_v8_0_ring_test_ring,
6995 	.test_ib = gfx_v8_0_ring_test_ib,
6996 	.insert_nop = amdgpu_ring_insert_nop,
6997 	.pad_ib = amdgpu_ring_generic_pad_ib,
6998 	.emit_wreg = gfx_v8_0_ring_emit_wreg,
6999 	.emit_mem_sync = gfx_v8_0_emit_mem_sync_compute,
7000 	.emit_wave_limit = gfx_v8_0_emit_wave_limit,
7001 };
7002 
7003 static const struct amdgpu_ring_funcs gfx_v8_0_ring_funcs_kiq = {
7004 	.type = AMDGPU_RING_TYPE_KIQ,
7005 	.align_mask = 0xff,
7006 	.nop = PACKET3(PACKET3_NOP, 0x3FFF),
7007 	.support_64bit_ptrs = false,
7008 	.get_rptr = gfx_v8_0_ring_get_rptr,
7009 	.get_wptr = gfx_v8_0_ring_get_wptr_compute,
7010 	.set_wptr = gfx_v8_0_ring_set_wptr_compute,
7011 	.emit_frame_size =
7012 		20 + /* gfx_v8_0_ring_emit_gds_switch */
7013 		7 + /* gfx_v8_0_ring_emit_hdp_flush */
7014 		5 + /* hdp_invalidate */
7015 		7 + /* gfx_v8_0_ring_emit_pipeline_sync */
7016 		17 + /* gfx_v8_0_ring_emit_vm_flush */
7017 		7 + 7 + 7, /* gfx_v8_0_ring_emit_fence_kiq x3 for user fence, vm fence */
7018 	.emit_ib_size =	7, /* gfx_v8_0_ring_emit_ib_compute */
7019 	.emit_fence = gfx_v8_0_ring_emit_fence_kiq,
7020 	.test_ring = gfx_v8_0_ring_test_ring,
7021 	.insert_nop = amdgpu_ring_insert_nop,
7022 	.pad_ib = amdgpu_ring_generic_pad_ib,
7023 	.emit_rreg = gfx_v8_0_ring_emit_rreg,
7024 	.emit_wreg = gfx_v8_0_ring_emit_wreg,
7025 };
7026 
7027 static void gfx_v8_0_set_ring_funcs(struct amdgpu_device *adev)
7028 {
7029 	int i;
7030 
7031 	adev->gfx.kiq.ring.funcs = &gfx_v8_0_ring_funcs_kiq;
7032 
7033 	for (i = 0; i < adev->gfx.num_gfx_rings; i++)
7034 		adev->gfx.gfx_ring[i].funcs = &gfx_v8_0_ring_funcs_gfx;
7035 
7036 	for (i = 0; i < adev->gfx.num_compute_rings; i++)
7037 		adev->gfx.compute_ring[i].funcs = &gfx_v8_0_ring_funcs_compute;
7038 }
7039 
7040 static const struct amdgpu_irq_src_funcs gfx_v8_0_eop_irq_funcs = {
7041 	.set = gfx_v8_0_set_eop_interrupt_state,
7042 	.process = gfx_v8_0_eop_irq,
7043 };
7044 
7045 static const struct amdgpu_irq_src_funcs gfx_v8_0_priv_reg_irq_funcs = {
7046 	.set = gfx_v8_0_set_priv_reg_fault_state,
7047 	.process = gfx_v8_0_priv_reg_irq,
7048 };
7049 
7050 static const struct amdgpu_irq_src_funcs gfx_v8_0_priv_inst_irq_funcs = {
7051 	.set = gfx_v8_0_set_priv_inst_fault_state,
7052 	.process = gfx_v8_0_priv_inst_irq,
7053 };
7054 
7055 static const struct amdgpu_irq_src_funcs gfx_v8_0_cp_ecc_error_irq_funcs = {
7056 	.set = gfx_v8_0_set_cp_ecc_int_state,
7057 	.process = gfx_v8_0_cp_ecc_error_irq,
7058 };
7059 
7060 static const struct amdgpu_irq_src_funcs gfx_v8_0_sq_irq_funcs = {
7061 	.set = gfx_v8_0_set_sq_int_state,
7062 	.process = gfx_v8_0_sq_irq,
7063 };
7064 
7065 static void gfx_v8_0_set_irq_funcs(struct amdgpu_device *adev)
7066 {
7067 	adev->gfx.eop_irq.num_types = AMDGPU_CP_IRQ_LAST;
7068 	adev->gfx.eop_irq.funcs = &gfx_v8_0_eop_irq_funcs;
7069 
7070 	adev->gfx.priv_reg_irq.num_types = 1;
7071 	adev->gfx.priv_reg_irq.funcs = &gfx_v8_0_priv_reg_irq_funcs;
7072 
7073 	adev->gfx.priv_inst_irq.num_types = 1;
7074 	adev->gfx.priv_inst_irq.funcs = &gfx_v8_0_priv_inst_irq_funcs;
7075 
7076 	adev->gfx.cp_ecc_error_irq.num_types = 1;
7077 	adev->gfx.cp_ecc_error_irq.funcs = &gfx_v8_0_cp_ecc_error_irq_funcs;
7078 
7079 	adev->gfx.sq_irq.num_types = 1;
7080 	adev->gfx.sq_irq.funcs = &gfx_v8_0_sq_irq_funcs;
7081 }
7082 
7083 static void gfx_v8_0_set_rlc_funcs(struct amdgpu_device *adev)
7084 {
7085 	adev->gfx.rlc.funcs = &iceland_rlc_funcs;
7086 }
7087 
7088 static void gfx_v8_0_set_gds_init(struct amdgpu_device *adev)
7089 {
7090 	/* init asci gds info */
7091 	adev->gds.gds_size = RREG32(mmGDS_VMID0_SIZE);
7092 	adev->gds.gws_size = 64;
7093 	adev->gds.oa_size = 16;
7094 	adev->gds.gds_compute_max_wave_id = RREG32(mmGDS_COMPUTE_MAX_WAVE_ID);
7095 }
7096 
7097 static void gfx_v8_0_set_user_cu_inactive_bitmap(struct amdgpu_device *adev,
7098 						 u32 bitmap)
7099 {
7100 	u32 data;
7101 
7102 	if (!bitmap)
7103 		return;
7104 
7105 	data = bitmap << GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_CUS__SHIFT;
7106 	data &= GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_CUS_MASK;
7107 
7108 	WREG32(mmGC_USER_SHADER_ARRAY_CONFIG, data);
7109 }
7110 
7111 static u32 gfx_v8_0_get_cu_active_bitmap(struct amdgpu_device *adev)
7112 {
7113 	u32 data, mask;
7114 
7115 	data =  RREG32(mmCC_GC_SHADER_ARRAY_CONFIG) |
7116 		RREG32(mmGC_USER_SHADER_ARRAY_CONFIG);
7117 
7118 	mask = amdgpu_gfx_create_bitmask(adev->gfx.config.max_cu_per_sh);
7119 
7120 	return ~REG_GET_FIELD(data, CC_GC_SHADER_ARRAY_CONFIG, INACTIVE_CUS) & mask;
7121 }
7122 
7123 static void gfx_v8_0_get_cu_info(struct amdgpu_device *adev)
7124 {
7125 	int i, j, k, counter, active_cu_number = 0;
7126 	u32 mask, bitmap, ao_bitmap, ao_cu_mask = 0;
7127 	struct amdgpu_cu_info *cu_info = &adev->gfx.cu_info;
7128 	unsigned disable_masks[4 * 2];
7129 	u32 ao_cu_num;
7130 
7131 	memset(cu_info, 0, sizeof(*cu_info));
7132 
7133 	if (adev->flags & AMD_IS_APU)
7134 		ao_cu_num = 2;
7135 	else
7136 		ao_cu_num = adev->gfx.config.max_cu_per_sh;
7137 
7138 	amdgpu_gfx_parse_disable_cu(disable_masks, 4, 2);
7139 
7140 	mutex_lock(&adev->grbm_idx_mutex);
7141 	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
7142 		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
7143 			mask = 1;
7144 			ao_bitmap = 0;
7145 			counter = 0;
7146 			gfx_v8_0_select_se_sh(adev, i, j, 0xffffffff);
7147 			if (i < 4 && j < 2)
7148 				gfx_v8_0_set_user_cu_inactive_bitmap(
7149 					adev, disable_masks[i * 2 + j]);
7150 			bitmap = gfx_v8_0_get_cu_active_bitmap(adev);
7151 			cu_info->bitmap[i][j] = bitmap;
7152 
7153 			for (k = 0; k < adev->gfx.config.max_cu_per_sh; k ++) {
7154 				if (bitmap & mask) {
7155 					if (counter < ao_cu_num)
7156 						ao_bitmap |= mask;
7157 					counter ++;
7158 				}
7159 				mask <<= 1;
7160 			}
7161 			active_cu_number += counter;
7162 			if (i < 2 && j < 2)
7163 				ao_cu_mask |= (ao_bitmap << (i * 16 + j * 8));
7164 			cu_info->ao_cu_bitmap[i][j] = ao_bitmap;
7165 		}
7166 	}
7167 	gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
7168 	mutex_unlock(&adev->grbm_idx_mutex);
7169 
7170 	cu_info->number = active_cu_number;
7171 	cu_info->ao_cu_mask = ao_cu_mask;
7172 	cu_info->simd_per_cu = NUM_SIMD_PER_CU;
7173 	cu_info->max_waves_per_simd = 10;
7174 	cu_info->max_scratch_slots_per_cu = 32;
7175 	cu_info->wave_front_size = 64;
7176 	cu_info->lds_size = 64;
7177 }
7178 
7179 const struct amdgpu_ip_block_version gfx_v8_0_ip_block =
7180 {
7181 	.type = AMD_IP_BLOCK_TYPE_GFX,
7182 	.major = 8,
7183 	.minor = 0,
7184 	.rev = 0,
7185 	.funcs = &gfx_v8_0_ip_funcs,
7186 };
7187 
7188 const struct amdgpu_ip_block_version gfx_v8_1_ip_block =
7189 {
7190 	.type = AMD_IP_BLOCK_TYPE_GFX,
7191 	.major = 8,
7192 	.minor = 1,
7193 	.rev = 0,
7194 	.funcs = &gfx_v8_0_ip_funcs,
7195 };
7196 
7197 static void gfx_v8_0_ring_emit_ce_meta(struct amdgpu_ring *ring)
7198 {
7199 	uint64_t ce_payload_addr;
7200 	int cnt_ce;
7201 	union {
7202 		struct vi_ce_ib_state regular;
7203 		struct vi_ce_ib_state_chained_ib chained;
7204 	} ce_payload = {};
7205 
7206 	if (ring->adev->virt.chained_ib_support) {
7207 		ce_payload_addr = amdgpu_csa_vaddr(ring->adev) +
7208 			offsetof(struct vi_gfx_meta_data_chained_ib, ce_payload);
7209 		cnt_ce = (sizeof(ce_payload.chained) >> 2) + 4 - 2;
7210 	} else {
7211 		ce_payload_addr = amdgpu_csa_vaddr(ring->adev) +
7212 			offsetof(struct vi_gfx_meta_data, ce_payload);
7213 		cnt_ce = (sizeof(ce_payload.regular) >> 2) + 4 - 2;
7214 	}
7215 
7216 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, cnt_ce));
7217 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(2) |
7218 				WRITE_DATA_DST_SEL(8) |
7219 				WR_CONFIRM) |
7220 				WRITE_DATA_CACHE_POLICY(0));
7221 	amdgpu_ring_write(ring, lower_32_bits(ce_payload_addr));
7222 	amdgpu_ring_write(ring, upper_32_bits(ce_payload_addr));
7223 	amdgpu_ring_write_multiple(ring, (void *)&ce_payload, cnt_ce - 2);
7224 }
7225 
7226 static void gfx_v8_0_ring_emit_de_meta(struct amdgpu_ring *ring)
7227 {
7228 	uint64_t de_payload_addr, gds_addr, csa_addr;
7229 	int cnt_de;
7230 	union {
7231 		struct vi_de_ib_state regular;
7232 		struct vi_de_ib_state_chained_ib chained;
7233 	} de_payload = {};
7234 
7235 	csa_addr = amdgpu_csa_vaddr(ring->adev);
7236 	gds_addr = csa_addr + 4096;
7237 	if (ring->adev->virt.chained_ib_support) {
7238 		de_payload.chained.gds_backup_addrlo = lower_32_bits(gds_addr);
7239 		de_payload.chained.gds_backup_addrhi = upper_32_bits(gds_addr);
7240 		de_payload_addr = csa_addr + offsetof(struct vi_gfx_meta_data_chained_ib, de_payload);
7241 		cnt_de = (sizeof(de_payload.chained) >> 2) + 4 - 2;
7242 	} else {
7243 		de_payload.regular.gds_backup_addrlo = lower_32_bits(gds_addr);
7244 		de_payload.regular.gds_backup_addrhi = upper_32_bits(gds_addr);
7245 		de_payload_addr = csa_addr + offsetof(struct vi_gfx_meta_data, de_payload);
7246 		cnt_de = (sizeof(de_payload.regular) >> 2) + 4 - 2;
7247 	}
7248 
7249 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, cnt_de));
7250 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(1) |
7251 				WRITE_DATA_DST_SEL(8) |
7252 				WR_CONFIRM) |
7253 				WRITE_DATA_CACHE_POLICY(0));
7254 	amdgpu_ring_write(ring, lower_32_bits(de_payload_addr));
7255 	amdgpu_ring_write(ring, upper_32_bits(de_payload_addr));
7256 	amdgpu_ring_write_multiple(ring, (void *)&de_payload, cnt_de - 2);
7257 }
7258