xref: /openbmc/linux/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c (revision 1406f0f3)
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 void gfx_v8_0_scratch_init(struct amdgpu_device *adev)
839 {
840 	adev->gfx.scratch.num_reg = 8;
841 	adev->gfx.scratch.reg_base = mmSCRATCH_REG0;
842 	adev->gfx.scratch.free_mask = (1u << adev->gfx.scratch.num_reg) - 1;
843 }
844 
845 static int gfx_v8_0_ring_test_ring(struct amdgpu_ring *ring)
846 {
847 	struct amdgpu_device *adev = ring->adev;
848 	uint32_t scratch;
849 	uint32_t tmp = 0;
850 	unsigned i;
851 	int r;
852 
853 	r = amdgpu_gfx_scratch_get(adev, &scratch);
854 	if (r)
855 		return r;
856 
857 	WREG32(scratch, 0xCAFEDEAD);
858 	r = amdgpu_ring_alloc(ring, 3);
859 	if (r)
860 		goto error_free_scratch;
861 
862 	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_UCONFIG_REG, 1));
863 	amdgpu_ring_write(ring, (scratch - PACKET3_SET_UCONFIG_REG_START));
864 	amdgpu_ring_write(ring, 0xDEADBEEF);
865 	amdgpu_ring_commit(ring);
866 
867 	for (i = 0; i < adev->usec_timeout; i++) {
868 		tmp = RREG32(scratch);
869 		if (tmp == 0xDEADBEEF)
870 			break;
871 		udelay(1);
872 	}
873 
874 	if (i >= adev->usec_timeout)
875 		r = -ETIMEDOUT;
876 
877 error_free_scratch:
878 	amdgpu_gfx_scratch_free(adev, scratch);
879 	return r;
880 }
881 
882 static int gfx_v8_0_ring_test_ib(struct amdgpu_ring *ring, long timeout)
883 {
884 	struct amdgpu_device *adev = ring->adev;
885 	struct amdgpu_ib ib;
886 	struct dma_fence *f = NULL;
887 
888 	unsigned int index;
889 	uint64_t gpu_addr;
890 	uint32_t tmp;
891 	long r;
892 
893 	r = amdgpu_device_wb_get(adev, &index);
894 	if (r)
895 		return r;
896 
897 	gpu_addr = adev->wb.gpu_addr + (index * 4);
898 	adev->wb.wb[index] = cpu_to_le32(0xCAFEDEAD);
899 	memset(&ib, 0, sizeof(ib));
900 	r = amdgpu_ib_get(adev, NULL, 16,
901 					AMDGPU_IB_POOL_DIRECT, &ib);
902 	if (r)
903 		goto err1;
904 
905 	ib.ptr[0] = PACKET3(PACKET3_WRITE_DATA, 3);
906 	ib.ptr[1] = WRITE_DATA_DST_SEL(5) | WR_CONFIRM;
907 	ib.ptr[2] = lower_32_bits(gpu_addr);
908 	ib.ptr[3] = upper_32_bits(gpu_addr);
909 	ib.ptr[4] = 0xDEADBEEF;
910 	ib.length_dw = 5;
911 
912 	r = amdgpu_ib_schedule(ring, 1, &ib, NULL, &f);
913 	if (r)
914 		goto err2;
915 
916 	r = dma_fence_wait_timeout(f, false, timeout);
917 	if (r == 0) {
918 		r = -ETIMEDOUT;
919 		goto err2;
920 	} else if (r < 0) {
921 		goto err2;
922 	}
923 
924 	tmp = adev->wb.wb[index];
925 	if (tmp == 0xDEADBEEF)
926 		r = 0;
927 	else
928 		r = -EINVAL;
929 
930 err2:
931 	amdgpu_ib_free(adev, &ib, NULL);
932 	dma_fence_put(f);
933 err1:
934 	amdgpu_device_wb_free(adev, index);
935 	return r;
936 }
937 
938 
939 static void gfx_v8_0_free_microcode(struct amdgpu_device *adev)
940 {
941 	release_firmware(adev->gfx.pfp_fw);
942 	adev->gfx.pfp_fw = NULL;
943 	release_firmware(adev->gfx.me_fw);
944 	adev->gfx.me_fw = NULL;
945 	release_firmware(adev->gfx.ce_fw);
946 	adev->gfx.ce_fw = NULL;
947 	release_firmware(adev->gfx.rlc_fw);
948 	adev->gfx.rlc_fw = NULL;
949 	release_firmware(adev->gfx.mec_fw);
950 	adev->gfx.mec_fw = NULL;
951 	if ((adev->asic_type != CHIP_STONEY) &&
952 	    (adev->asic_type != CHIP_TOPAZ))
953 		release_firmware(adev->gfx.mec2_fw);
954 	adev->gfx.mec2_fw = NULL;
955 
956 	kfree(adev->gfx.rlc.register_list_format);
957 }
958 
959 static int gfx_v8_0_init_microcode(struct amdgpu_device *adev)
960 {
961 	const char *chip_name;
962 	char fw_name[30];
963 	int err;
964 	struct amdgpu_firmware_info *info = NULL;
965 	const struct common_firmware_header *header = NULL;
966 	const struct gfx_firmware_header_v1_0 *cp_hdr;
967 	const struct rlc_firmware_header_v2_0 *rlc_hdr;
968 	unsigned int *tmp = NULL, i;
969 
970 	DRM_DEBUG("\n");
971 
972 	switch (adev->asic_type) {
973 	case CHIP_TOPAZ:
974 		chip_name = "topaz";
975 		break;
976 	case CHIP_TONGA:
977 		chip_name = "tonga";
978 		break;
979 	case CHIP_CARRIZO:
980 		chip_name = "carrizo";
981 		break;
982 	case CHIP_FIJI:
983 		chip_name = "fiji";
984 		break;
985 	case CHIP_STONEY:
986 		chip_name = "stoney";
987 		break;
988 	case CHIP_POLARIS10:
989 		chip_name = "polaris10";
990 		break;
991 	case CHIP_POLARIS11:
992 		chip_name = "polaris11";
993 		break;
994 	case CHIP_POLARIS12:
995 		chip_name = "polaris12";
996 		break;
997 	case CHIP_VEGAM:
998 		chip_name = "vegam";
999 		break;
1000 	default:
1001 		BUG();
1002 	}
1003 
1004 	if (adev->asic_type >= CHIP_POLARIS10 && adev->asic_type <= CHIP_POLARIS12) {
1005 		snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_pfp_2.bin", chip_name);
1006 		err = request_firmware(&adev->gfx.pfp_fw, fw_name, adev->dev);
1007 		if (err == -ENOENT) {
1008 			snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_pfp.bin", chip_name);
1009 			err = request_firmware(&adev->gfx.pfp_fw, fw_name, adev->dev);
1010 		}
1011 	} else {
1012 		snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_pfp.bin", chip_name);
1013 		err = request_firmware(&adev->gfx.pfp_fw, fw_name, adev->dev);
1014 	}
1015 	if (err)
1016 		goto out;
1017 	err = amdgpu_ucode_validate(adev->gfx.pfp_fw);
1018 	if (err)
1019 		goto out;
1020 	cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.pfp_fw->data;
1021 	adev->gfx.pfp_fw_version = le32_to_cpu(cp_hdr->header.ucode_version);
1022 	adev->gfx.pfp_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version);
1023 
1024 	if (adev->asic_type >= CHIP_POLARIS10 && adev->asic_type <= CHIP_POLARIS12) {
1025 		snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_me_2.bin", chip_name);
1026 		err = request_firmware(&adev->gfx.me_fw, fw_name, adev->dev);
1027 		if (err == -ENOENT) {
1028 			snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_me.bin", chip_name);
1029 			err = request_firmware(&adev->gfx.me_fw, fw_name, adev->dev);
1030 		}
1031 	} else {
1032 		snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_me.bin", chip_name);
1033 		err = request_firmware(&adev->gfx.me_fw, fw_name, adev->dev);
1034 	}
1035 	if (err)
1036 		goto out;
1037 	err = amdgpu_ucode_validate(adev->gfx.me_fw);
1038 	if (err)
1039 		goto out;
1040 	cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.me_fw->data;
1041 	adev->gfx.me_fw_version = le32_to_cpu(cp_hdr->header.ucode_version);
1042 
1043 	adev->gfx.me_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version);
1044 
1045 	if (adev->asic_type >= CHIP_POLARIS10 && adev->asic_type <= CHIP_POLARIS12) {
1046 		snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_ce_2.bin", chip_name);
1047 		err = request_firmware(&adev->gfx.ce_fw, fw_name, adev->dev);
1048 		if (err == -ENOENT) {
1049 			snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_ce.bin", chip_name);
1050 			err = request_firmware(&adev->gfx.ce_fw, fw_name, adev->dev);
1051 		}
1052 	} else {
1053 		snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_ce.bin", chip_name);
1054 		err = request_firmware(&adev->gfx.ce_fw, fw_name, adev->dev);
1055 	}
1056 	if (err)
1057 		goto out;
1058 	err = amdgpu_ucode_validate(adev->gfx.ce_fw);
1059 	if (err)
1060 		goto out;
1061 	cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.ce_fw->data;
1062 	adev->gfx.ce_fw_version = le32_to_cpu(cp_hdr->header.ucode_version);
1063 	adev->gfx.ce_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version);
1064 
1065 	/*
1066 	 * Support for MCBP/Virtualization in combination with chained IBs is
1067 	 * formal released on feature version #46
1068 	 */
1069 	if (adev->gfx.ce_feature_version >= 46 &&
1070 	    adev->gfx.pfp_feature_version >= 46) {
1071 		adev->virt.chained_ib_support = true;
1072 		DRM_INFO("Chained IB support enabled!\n");
1073 	} else
1074 		adev->virt.chained_ib_support = false;
1075 
1076 	snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_rlc.bin", chip_name);
1077 	err = request_firmware(&adev->gfx.rlc_fw, fw_name, adev->dev);
1078 	if (err)
1079 		goto out;
1080 	err = amdgpu_ucode_validate(adev->gfx.rlc_fw);
1081 	rlc_hdr = (const struct rlc_firmware_header_v2_0 *)adev->gfx.rlc_fw->data;
1082 	adev->gfx.rlc_fw_version = le32_to_cpu(rlc_hdr->header.ucode_version);
1083 	adev->gfx.rlc_feature_version = le32_to_cpu(rlc_hdr->ucode_feature_version);
1084 
1085 	adev->gfx.rlc.save_and_restore_offset =
1086 			le32_to_cpu(rlc_hdr->save_and_restore_offset);
1087 	adev->gfx.rlc.clear_state_descriptor_offset =
1088 			le32_to_cpu(rlc_hdr->clear_state_descriptor_offset);
1089 	adev->gfx.rlc.avail_scratch_ram_locations =
1090 			le32_to_cpu(rlc_hdr->avail_scratch_ram_locations);
1091 	adev->gfx.rlc.reg_restore_list_size =
1092 			le32_to_cpu(rlc_hdr->reg_restore_list_size);
1093 	adev->gfx.rlc.reg_list_format_start =
1094 			le32_to_cpu(rlc_hdr->reg_list_format_start);
1095 	adev->gfx.rlc.reg_list_format_separate_start =
1096 			le32_to_cpu(rlc_hdr->reg_list_format_separate_start);
1097 	adev->gfx.rlc.starting_offsets_start =
1098 			le32_to_cpu(rlc_hdr->starting_offsets_start);
1099 	adev->gfx.rlc.reg_list_format_size_bytes =
1100 			le32_to_cpu(rlc_hdr->reg_list_format_size_bytes);
1101 	adev->gfx.rlc.reg_list_size_bytes =
1102 			le32_to_cpu(rlc_hdr->reg_list_size_bytes);
1103 
1104 	adev->gfx.rlc.register_list_format =
1105 			kmalloc(adev->gfx.rlc.reg_list_format_size_bytes +
1106 					adev->gfx.rlc.reg_list_size_bytes, GFP_KERNEL);
1107 
1108 	if (!adev->gfx.rlc.register_list_format) {
1109 		err = -ENOMEM;
1110 		goto out;
1111 	}
1112 
1113 	tmp = (unsigned int *)((uintptr_t)rlc_hdr +
1114 			le32_to_cpu(rlc_hdr->reg_list_format_array_offset_bytes));
1115 	for (i = 0 ; i < (adev->gfx.rlc.reg_list_format_size_bytes >> 2); i++)
1116 		adev->gfx.rlc.register_list_format[i] =	le32_to_cpu(tmp[i]);
1117 
1118 	adev->gfx.rlc.register_restore = adev->gfx.rlc.register_list_format + i;
1119 
1120 	tmp = (unsigned int *)((uintptr_t)rlc_hdr +
1121 			le32_to_cpu(rlc_hdr->reg_list_array_offset_bytes));
1122 	for (i = 0 ; i < (adev->gfx.rlc.reg_list_size_bytes >> 2); i++)
1123 		adev->gfx.rlc.register_restore[i] = le32_to_cpu(tmp[i]);
1124 
1125 	if (adev->asic_type >= CHIP_POLARIS10 && adev->asic_type <= CHIP_POLARIS12) {
1126 		snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec_2.bin", chip_name);
1127 		err = request_firmware(&adev->gfx.mec_fw, fw_name, adev->dev);
1128 		if (err == -ENOENT) {
1129 			snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec.bin", chip_name);
1130 			err = request_firmware(&adev->gfx.mec_fw, fw_name, adev->dev);
1131 		}
1132 	} else {
1133 		snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec.bin", chip_name);
1134 		err = request_firmware(&adev->gfx.mec_fw, fw_name, adev->dev);
1135 	}
1136 	if (err)
1137 		goto out;
1138 	err = amdgpu_ucode_validate(adev->gfx.mec_fw);
1139 	if (err)
1140 		goto out;
1141 	cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data;
1142 	adev->gfx.mec_fw_version = le32_to_cpu(cp_hdr->header.ucode_version);
1143 	adev->gfx.mec_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version);
1144 
1145 	if ((adev->asic_type != CHIP_STONEY) &&
1146 	    (adev->asic_type != CHIP_TOPAZ)) {
1147 		if (adev->asic_type >= CHIP_POLARIS10 && adev->asic_type <= CHIP_POLARIS12) {
1148 			snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec2_2.bin", chip_name);
1149 			err = request_firmware(&adev->gfx.mec2_fw, fw_name, adev->dev);
1150 			if (err == -ENOENT) {
1151 				snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec2.bin", chip_name);
1152 				err = request_firmware(&adev->gfx.mec2_fw, fw_name, adev->dev);
1153 			}
1154 		} else {
1155 			snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec2.bin", chip_name);
1156 			err = request_firmware(&adev->gfx.mec2_fw, fw_name, adev->dev);
1157 		}
1158 		if (!err) {
1159 			err = amdgpu_ucode_validate(adev->gfx.mec2_fw);
1160 			if (err)
1161 				goto out;
1162 			cp_hdr = (const struct gfx_firmware_header_v1_0 *)
1163 				adev->gfx.mec2_fw->data;
1164 			adev->gfx.mec2_fw_version =
1165 				le32_to_cpu(cp_hdr->header.ucode_version);
1166 			adev->gfx.mec2_feature_version =
1167 				le32_to_cpu(cp_hdr->ucode_feature_version);
1168 		} else {
1169 			err = 0;
1170 			adev->gfx.mec2_fw = NULL;
1171 		}
1172 	}
1173 
1174 	info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_PFP];
1175 	info->ucode_id = AMDGPU_UCODE_ID_CP_PFP;
1176 	info->fw = adev->gfx.pfp_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_CP_ME];
1182 	info->ucode_id = AMDGPU_UCODE_ID_CP_ME;
1183 	info->fw = adev->gfx.me_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_CE];
1189 	info->ucode_id = AMDGPU_UCODE_ID_CP_CE;
1190 	info->fw = adev->gfx.ce_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 	info = &adev->firmware.ucode[AMDGPU_UCODE_ID_RLC_G];
1196 	info->ucode_id = AMDGPU_UCODE_ID_RLC_G;
1197 	info->fw = adev->gfx.rlc_fw;
1198 	header = (const struct common_firmware_header *)info->fw->data;
1199 	adev->firmware.fw_size +=
1200 		ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
1201 
1202 	info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_MEC1];
1203 	info->ucode_id = AMDGPU_UCODE_ID_CP_MEC1;
1204 	info->fw = adev->gfx.mec_fw;
1205 	header = (const struct common_firmware_header *)info->fw->data;
1206 	adev->firmware.fw_size +=
1207 		ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
1208 
1209 	/* we need account JT in */
1210 	cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data;
1211 	adev->firmware.fw_size +=
1212 		ALIGN(le32_to_cpu(cp_hdr->jt_size) << 2, PAGE_SIZE);
1213 
1214 	if (amdgpu_sriov_vf(adev)) {
1215 		info = &adev->firmware.ucode[AMDGPU_UCODE_ID_STORAGE];
1216 		info->ucode_id = AMDGPU_UCODE_ID_STORAGE;
1217 		info->fw = adev->gfx.mec_fw;
1218 		adev->firmware.fw_size +=
1219 			ALIGN(le32_to_cpu(64 * PAGE_SIZE), PAGE_SIZE);
1220 	}
1221 
1222 	if (adev->gfx.mec2_fw) {
1223 		info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_MEC2];
1224 		info->ucode_id = AMDGPU_UCODE_ID_CP_MEC2;
1225 		info->fw = adev->gfx.mec2_fw;
1226 		header = (const struct common_firmware_header *)info->fw->data;
1227 		adev->firmware.fw_size +=
1228 			ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
1229 	}
1230 
1231 out:
1232 	if (err) {
1233 		dev_err(adev->dev,
1234 			"gfx8: Failed to load firmware \"%s\"\n",
1235 			fw_name);
1236 		release_firmware(adev->gfx.pfp_fw);
1237 		adev->gfx.pfp_fw = NULL;
1238 		release_firmware(adev->gfx.me_fw);
1239 		adev->gfx.me_fw = NULL;
1240 		release_firmware(adev->gfx.ce_fw);
1241 		adev->gfx.ce_fw = NULL;
1242 		release_firmware(adev->gfx.rlc_fw);
1243 		adev->gfx.rlc_fw = NULL;
1244 		release_firmware(adev->gfx.mec_fw);
1245 		adev->gfx.mec_fw = NULL;
1246 		release_firmware(adev->gfx.mec2_fw);
1247 		adev->gfx.mec2_fw = NULL;
1248 	}
1249 	return err;
1250 }
1251 
1252 static void gfx_v8_0_get_csb_buffer(struct amdgpu_device *adev,
1253 				    volatile u32 *buffer)
1254 {
1255 	u32 count = 0, i;
1256 	const struct cs_section_def *sect = NULL;
1257 	const struct cs_extent_def *ext = NULL;
1258 
1259 	if (adev->gfx.rlc.cs_data == NULL)
1260 		return;
1261 	if (buffer == NULL)
1262 		return;
1263 
1264 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0));
1265 	buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
1266 
1267 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CONTEXT_CONTROL, 1));
1268 	buffer[count++] = cpu_to_le32(0x80000000);
1269 	buffer[count++] = cpu_to_le32(0x80000000);
1270 
1271 	for (sect = adev->gfx.rlc.cs_data; sect->section != NULL; ++sect) {
1272 		for (ext = sect->section; ext->extent != NULL; ++ext) {
1273 			if (sect->id == SECT_CONTEXT) {
1274 				buffer[count++] =
1275 					cpu_to_le32(PACKET3(PACKET3_SET_CONTEXT_REG, ext->reg_count));
1276 				buffer[count++] = cpu_to_le32(ext->reg_index -
1277 						PACKET3_SET_CONTEXT_REG_START);
1278 				for (i = 0; i < ext->reg_count; i++)
1279 					buffer[count++] = cpu_to_le32(ext->extent[i]);
1280 			} else {
1281 				return;
1282 			}
1283 		}
1284 	}
1285 
1286 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_SET_CONTEXT_REG, 2));
1287 	buffer[count++] = cpu_to_le32(mmPA_SC_RASTER_CONFIG -
1288 			PACKET3_SET_CONTEXT_REG_START);
1289 	buffer[count++] = cpu_to_le32(adev->gfx.config.rb_config[0][0].raster_config);
1290 	buffer[count++] = cpu_to_le32(adev->gfx.config.rb_config[0][0].raster_config_1);
1291 
1292 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0));
1293 	buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_END_CLEAR_STATE);
1294 
1295 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CLEAR_STATE, 0));
1296 	buffer[count++] = cpu_to_le32(0);
1297 }
1298 
1299 static int gfx_v8_0_cp_jump_table_num(struct amdgpu_device *adev)
1300 {
1301 	if (adev->asic_type == CHIP_CARRIZO)
1302 		return 5;
1303 	else
1304 		return 4;
1305 }
1306 
1307 static int gfx_v8_0_rlc_init(struct amdgpu_device *adev)
1308 {
1309 	const struct cs_section_def *cs_data;
1310 	int r;
1311 
1312 	adev->gfx.rlc.cs_data = vi_cs_data;
1313 
1314 	cs_data = adev->gfx.rlc.cs_data;
1315 
1316 	if (cs_data) {
1317 		/* init clear state block */
1318 		r = amdgpu_gfx_rlc_init_csb(adev);
1319 		if (r)
1320 			return r;
1321 	}
1322 
1323 	if ((adev->asic_type == CHIP_CARRIZO) ||
1324 	    (adev->asic_type == CHIP_STONEY)) {
1325 		adev->gfx.rlc.cp_table_size = ALIGN(96 * 5 * 4, 2048) + (64 * 1024); /* JT + GDS */
1326 		r = amdgpu_gfx_rlc_init_cpt(adev);
1327 		if (r)
1328 			return r;
1329 	}
1330 
1331 	/* init spm vmid with 0xf */
1332 	if (adev->gfx.rlc.funcs->update_spm_vmid)
1333 		adev->gfx.rlc.funcs->update_spm_vmid(adev, 0xf);
1334 
1335 	return 0;
1336 }
1337 
1338 static void gfx_v8_0_mec_fini(struct amdgpu_device *adev)
1339 {
1340 	amdgpu_bo_free_kernel(&adev->gfx.mec.hpd_eop_obj, NULL, NULL);
1341 }
1342 
1343 static int gfx_v8_0_mec_init(struct amdgpu_device *adev)
1344 {
1345 	int r;
1346 	u32 *hpd;
1347 	size_t mec_hpd_size;
1348 
1349 	bitmap_zero(adev->gfx.mec.queue_bitmap, AMDGPU_MAX_COMPUTE_QUEUES);
1350 
1351 	/* take ownership of the relevant compute queues */
1352 	amdgpu_gfx_compute_queue_acquire(adev);
1353 
1354 	mec_hpd_size = adev->gfx.num_compute_rings * GFX8_MEC_HPD_SIZE;
1355 	if (mec_hpd_size) {
1356 		r = amdgpu_bo_create_reserved(adev, mec_hpd_size, PAGE_SIZE,
1357 					      AMDGPU_GEM_DOMAIN_VRAM,
1358 					      &adev->gfx.mec.hpd_eop_obj,
1359 					      &adev->gfx.mec.hpd_eop_gpu_addr,
1360 					      (void **)&hpd);
1361 		if (r) {
1362 			dev_warn(adev->dev, "(%d) create HDP EOP bo failed\n", r);
1363 			return r;
1364 		}
1365 
1366 		memset(hpd, 0, mec_hpd_size);
1367 
1368 		amdgpu_bo_kunmap(adev->gfx.mec.hpd_eop_obj);
1369 		amdgpu_bo_unreserve(adev->gfx.mec.hpd_eop_obj);
1370 	}
1371 
1372 	return 0;
1373 }
1374 
1375 static const u32 vgpr_init_compute_shader[] =
1376 {
1377 	0x7e000209, 0x7e020208,
1378 	0x7e040207, 0x7e060206,
1379 	0x7e080205, 0x7e0a0204,
1380 	0x7e0c0203, 0x7e0e0202,
1381 	0x7e100201, 0x7e120200,
1382 	0x7e140209, 0x7e160208,
1383 	0x7e180207, 0x7e1a0206,
1384 	0x7e1c0205, 0x7e1e0204,
1385 	0x7e200203, 0x7e220202,
1386 	0x7e240201, 0x7e260200,
1387 	0x7e280209, 0x7e2a0208,
1388 	0x7e2c0207, 0x7e2e0206,
1389 	0x7e300205, 0x7e320204,
1390 	0x7e340203, 0x7e360202,
1391 	0x7e380201, 0x7e3a0200,
1392 	0x7e3c0209, 0x7e3e0208,
1393 	0x7e400207, 0x7e420206,
1394 	0x7e440205, 0x7e460204,
1395 	0x7e480203, 0x7e4a0202,
1396 	0x7e4c0201, 0x7e4e0200,
1397 	0x7e500209, 0x7e520208,
1398 	0x7e540207, 0x7e560206,
1399 	0x7e580205, 0x7e5a0204,
1400 	0x7e5c0203, 0x7e5e0202,
1401 	0x7e600201, 0x7e620200,
1402 	0x7e640209, 0x7e660208,
1403 	0x7e680207, 0x7e6a0206,
1404 	0x7e6c0205, 0x7e6e0204,
1405 	0x7e700203, 0x7e720202,
1406 	0x7e740201, 0x7e760200,
1407 	0x7e780209, 0x7e7a0208,
1408 	0x7e7c0207, 0x7e7e0206,
1409 	0xbf8a0000, 0xbf810000,
1410 };
1411 
1412 static const u32 sgpr_init_compute_shader[] =
1413 {
1414 	0xbe8a0100, 0xbe8c0102,
1415 	0xbe8e0104, 0xbe900106,
1416 	0xbe920108, 0xbe940100,
1417 	0xbe960102, 0xbe980104,
1418 	0xbe9a0106, 0xbe9c0108,
1419 	0xbe9e0100, 0xbea00102,
1420 	0xbea20104, 0xbea40106,
1421 	0xbea60108, 0xbea80100,
1422 	0xbeaa0102, 0xbeac0104,
1423 	0xbeae0106, 0xbeb00108,
1424 	0xbeb20100, 0xbeb40102,
1425 	0xbeb60104, 0xbeb80106,
1426 	0xbeba0108, 0xbebc0100,
1427 	0xbebe0102, 0xbec00104,
1428 	0xbec20106, 0xbec40108,
1429 	0xbec60100, 0xbec80102,
1430 	0xbee60004, 0xbee70005,
1431 	0xbeea0006, 0xbeeb0007,
1432 	0xbee80008, 0xbee90009,
1433 	0xbefc0000, 0xbf8a0000,
1434 	0xbf810000, 0x00000000,
1435 };
1436 
1437 static const u32 vgpr_init_regs[] =
1438 {
1439 	mmCOMPUTE_STATIC_THREAD_MGMT_SE0, 0xffffffff,
1440 	mmCOMPUTE_RESOURCE_LIMITS, 0x1000000, /* CU_GROUP_COUNT=1 */
1441 	mmCOMPUTE_NUM_THREAD_X, 256*4,
1442 	mmCOMPUTE_NUM_THREAD_Y, 1,
1443 	mmCOMPUTE_NUM_THREAD_Z, 1,
1444 	mmCOMPUTE_PGM_RSRC1, 0x100004f, /* VGPRS=15 (64 logical VGPRs), SGPRS=1 (16 SGPRs), BULKY=1 */
1445 	mmCOMPUTE_PGM_RSRC2, 20,
1446 	mmCOMPUTE_USER_DATA_0, 0xedcedc00,
1447 	mmCOMPUTE_USER_DATA_1, 0xedcedc01,
1448 	mmCOMPUTE_USER_DATA_2, 0xedcedc02,
1449 	mmCOMPUTE_USER_DATA_3, 0xedcedc03,
1450 	mmCOMPUTE_USER_DATA_4, 0xedcedc04,
1451 	mmCOMPUTE_USER_DATA_5, 0xedcedc05,
1452 	mmCOMPUTE_USER_DATA_6, 0xedcedc06,
1453 	mmCOMPUTE_USER_DATA_7, 0xedcedc07,
1454 	mmCOMPUTE_USER_DATA_8, 0xedcedc08,
1455 	mmCOMPUTE_USER_DATA_9, 0xedcedc09,
1456 };
1457 
1458 static const u32 sgpr1_init_regs[] =
1459 {
1460 	mmCOMPUTE_STATIC_THREAD_MGMT_SE0, 0x0f,
1461 	mmCOMPUTE_RESOURCE_LIMITS, 0x1000000, /* CU_GROUP_COUNT=1 */
1462 	mmCOMPUTE_NUM_THREAD_X, 256*5,
1463 	mmCOMPUTE_NUM_THREAD_Y, 1,
1464 	mmCOMPUTE_NUM_THREAD_Z, 1,
1465 	mmCOMPUTE_PGM_RSRC1, 0x240, /* SGPRS=9 (80 GPRS) */
1466 	mmCOMPUTE_PGM_RSRC2, 20,
1467 	mmCOMPUTE_USER_DATA_0, 0xedcedc00,
1468 	mmCOMPUTE_USER_DATA_1, 0xedcedc01,
1469 	mmCOMPUTE_USER_DATA_2, 0xedcedc02,
1470 	mmCOMPUTE_USER_DATA_3, 0xedcedc03,
1471 	mmCOMPUTE_USER_DATA_4, 0xedcedc04,
1472 	mmCOMPUTE_USER_DATA_5, 0xedcedc05,
1473 	mmCOMPUTE_USER_DATA_6, 0xedcedc06,
1474 	mmCOMPUTE_USER_DATA_7, 0xedcedc07,
1475 	mmCOMPUTE_USER_DATA_8, 0xedcedc08,
1476 	mmCOMPUTE_USER_DATA_9, 0xedcedc09,
1477 };
1478 
1479 static const u32 sgpr2_init_regs[] =
1480 {
1481 	mmCOMPUTE_STATIC_THREAD_MGMT_SE0, 0xf0,
1482 	mmCOMPUTE_RESOURCE_LIMITS, 0x1000000,
1483 	mmCOMPUTE_NUM_THREAD_X, 256*5,
1484 	mmCOMPUTE_NUM_THREAD_Y, 1,
1485 	mmCOMPUTE_NUM_THREAD_Z, 1,
1486 	mmCOMPUTE_PGM_RSRC1, 0x240, /* SGPRS=9 (80 GPRS) */
1487 	mmCOMPUTE_PGM_RSRC2, 20,
1488 	mmCOMPUTE_USER_DATA_0, 0xedcedc00,
1489 	mmCOMPUTE_USER_DATA_1, 0xedcedc01,
1490 	mmCOMPUTE_USER_DATA_2, 0xedcedc02,
1491 	mmCOMPUTE_USER_DATA_3, 0xedcedc03,
1492 	mmCOMPUTE_USER_DATA_4, 0xedcedc04,
1493 	mmCOMPUTE_USER_DATA_5, 0xedcedc05,
1494 	mmCOMPUTE_USER_DATA_6, 0xedcedc06,
1495 	mmCOMPUTE_USER_DATA_7, 0xedcedc07,
1496 	mmCOMPUTE_USER_DATA_8, 0xedcedc08,
1497 	mmCOMPUTE_USER_DATA_9, 0xedcedc09,
1498 };
1499 
1500 static const u32 sec_ded_counter_registers[] =
1501 {
1502 	mmCPC_EDC_ATC_CNT,
1503 	mmCPC_EDC_SCRATCH_CNT,
1504 	mmCPC_EDC_UCODE_CNT,
1505 	mmCPF_EDC_ATC_CNT,
1506 	mmCPF_EDC_ROQ_CNT,
1507 	mmCPF_EDC_TAG_CNT,
1508 	mmCPG_EDC_ATC_CNT,
1509 	mmCPG_EDC_DMA_CNT,
1510 	mmCPG_EDC_TAG_CNT,
1511 	mmDC_EDC_CSINVOC_CNT,
1512 	mmDC_EDC_RESTORE_CNT,
1513 	mmDC_EDC_STATE_CNT,
1514 	mmGDS_EDC_CNT,
1515 	mmGDS_EDC_GRBM_CNT,
1516 	mmGDS_EDC_OA_DED,
1517 	mmSPI_EDC_CNT,
1518 	mmSQC_ATC_EDC_GATCL1_CNT,
1519 	mmSQC_EDC_CNT,
1520 	mmSQ_EDC_DED_CNT,
1521 	mmSQ_EDC_INFO,
1522 	mmSQ_EDC_SEC_CNT,
1523 	mmTCC_EDC_CNT,
1524 	mmTCP_ATC_EDC_GATCL1_CNT,
1525 	mmTCP_EDC_CNT,
1526 	mmTD_EDC_CNT
1527 };
1528 
1529 static int gfx_v8_0_do_edc_gpr_workarounds(struct amdgpu_device *adev)
1530 {
1531 	struct amdgpu_ring *ring = &adev->gfx.compute_ring[0];
1532 	struct amdgpu_ib ib;
1533 	struct dma_fence *f = NULL;
1534 	int r, i;
1535 	u32 tmp;
1536 	unsigned total_size, vgpr_offset, sgpr_offset;
1537 	u64 gpu_addr;
1538 
1539 	/* only supported on CZ */
1540 	if (adev->asic_type != CHIP_CARRIZO)
1541 		return 0;
1542 
1543 	/* bail if the compute ring is not ready */
1544 	if (!ring->sched.ready)
1545 		return 0;
1546 
1547 	tmp = RREG32(mmGB_EDC_MODE);
1548 	WREG32(mmGB_EDC_MODE, 0);
1549 
1550 	total_size =
1551 		(((ARRAY_SIZE(vgpr_init_regs) / 2) * 3) + 4 + 5 + 2) * 4;
1552 	total_size +=
1553 		(((ARRAY_SIZE(sgpr1_init_regs) / 2) * 3) + 4 + 5 + 2) * 4;
1554 	total_size +=
1555 		(((ARRAY_SIZE(sgpr2_init_regs) / 2) * 3) + 4 + 5 + 2) * 4;
1556 	total_size = ALIGN(total_size, 256);
1557 	vgpr_offset = total_size;
1558 	total_size += ALIGN(sizeof(vgpr_init_compute_shader), 256);
1559 	sgpr_offset = total_size;
1560 	total_size += sizeof(sgpr_init_compute_shader);
1561 
1562 	/* allocate an indirect buffer to put the commands in */
1563 	memset(&ib, 0, sizeof(ib));
1564 	r = amdgpu_ib_get(adev, NULL, total_size,
1565 					AMDGPU_IB_POOL_DIRECT, &ib);
1566 	if (r) {
1567 		DRM_ERROR("amdgpu: failed to get ib (%d).\n", r);
1568 		return r;
1569 	}
1570 
1571 	/* load the compute shaders */
1572 	for (i = 0; i < ARRAY_SIZE(vgpr_init_compute_shader); i++)
1573 		ib.ptr[i + (vgpr_offset / 4)] = vgpr_init_compute_shader[i];
1574 
1575 	for (i = 0; i < ARRAY_SIZE(sgpr_init_compute_shader); i++)
1576 		ib.ptr[i + (sgpr_offset / 4)] = sgpr_init_compute_shader[i];
1577 
1578 	/* init the ib length to 0 */
1579 	ib.length_dw = 0;
1580 
1581 	/* VGPR */
1582 	/* write the register state for the compute dispatch */
1583 	for (i = 0; i < ARRAY_SIZE(vgpr_init_regs); i += 2) {
1584 		ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 1);
1585 		ib.ptr[ib.length_dw++] = vgpr_init_regs[i] - PACKET3_SET_SH_REG_START;
1586 		ib.ptr[ib.length_dw++] = vgpr_init_regs[i + 1];
1587 	}
1588 	/* write the shader start address: mmCOMPUTE_PGM_LO, mmCOMPUTE_PGM_HI */
1589 	gpu_addr = (ib.gpu_addr + (u64)vgpr_offset) >> 8;
1590 	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 2);
1591 	ib.ptr[ib.length_dw++] = mmCOMPUTE_PGM_LO - PACKET3_SET_SH_REG_START;
1592 	ib.ptr[ib.length_dw++] = lower_32_bits(gpu_addr);
1593 	ib.ptr[ib.length_dw++] = upper_32_bits(gpu_addr);
1594 
1595 	/* write dispatch packet */
1596 	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_DISPATCH_DIRECT, 3);
1597 	ib.ptr[ib.length_dw++] = 8; /* x */
1598 	ib.ptr[ib.length_dw++] = 1; /* y */
1599 	ib.ptr[ib.length_dw++] = 1; /* z */
1600 	ib.ptr[ib.length_dw++] =
1601 		REG_SET_FIELD(0, COMPUTE_DISPATCH_INITIATOR, COMPUTE_SHADER_EN, 1);
1602 
1603 	/* write CS partial flush packet */
1604 	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_EVENT_WRITE, 0);
1605 	ib.ptr[ib.length_dw++] = EVENT_TYPE(7) | EVENT_INDEX(4);
1606 
1607 	/* SGPR1 */
1608 	/* write the register state for the compute dispatch */
1609 	for (i = 0; i < ARRAY_SIZE(sgpr1_init_regs); i += 2) {
1610 		ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 1);
1611 		ib.ptr[ib.length_dw++] = sgpr1_init_regs[i] - PACKET3_SET_SH_REG_START;
1612 		ib.ptr[ib.length_dw++] = sgpr1_init_regs[i + 1];
1613 	}
1614 	/* write the shader start address: mmCOMPUTE_PGM_LO, mmCOMPUTE_PGM_HI */
1615 	gpu_addr = (ib.gpu_addr + (u64)sgpr_offset) >> 8;
1616 	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 2);
1617 	ib.ptr[ib.length_dw++] = mmCOMPUTE_PGM_LO - PACKET3_SET_SH_REG_START;
1618 	ib.ptr[ib.length_dw++] = lower_32_bits(gpu_addr);
1619 	ib.ptr[ib.length_dw++] = upper_32_bits(gpu_addr);
1620 
1621 	/* write dispatch packet */
1622 	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_DISPATCH_DIRECT, 3);
1623 	ib.ptr[ib.length_dw++] = 8; /* x */
1624 	ib.ptr[ib.length_dw++] = 1; /* y */
1625 	ib.ptr[ib.length_dw++] = 1; /* z */
1626 	ib.ptr[ib.length_dw++] =
1627 		REG_SET_FIELD(0, COMPUTE_DISPATCH_INITIATOR, COMPUTE_SHADER_EN, 1);
1628 
1629 	/* write CS partial flush packet */
1630 	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_EVENT_WRITE, 0);
1631 	ib.ptr[ib.length_dw++] = EVENT_TYPE(7) | EVENT_INDEX(4);
1632 
1633 	/* SGPR2 */
1634 	/* write the register state for the compute dispatch */
1635 	for (i = 0; i < ARRAY_SIZE(sgpr2_init_regs); i += 2) {
1636 		ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 1);
1637 		ib.ptr[ib.length_dw++] = sgpr2_init_regs[i] - PACKET3_SET_SH_REG_START;
1638 		ib.ptr[ib.length_dw++] = sgpr2_init_regs[i + 1];
1639 	}
1640 	/* write the shader start address: mmCOMPUTE_PGM_LO, mmCOMPUTE_PGM_HI */
1641 	gpu_addr = (ib.gpu_addr + (u64)sgpr_offset) >> 8;
1642 	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 2);
1643 	ib.ptr[ib.length_dw++] = mmCOMPUTE_PGM_LO - PACKET3_SET_SH_REG_START;
1644 	ib.ptr[ib.length_dw++] = lower_32_bits(gpu_addr);
1645 	ib.ptr[ib.length_dw++] = upper_32_bits(gpu_addr);
1646 
1647 	/* write dispatch packet */
1648 	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_DISPATCH_DIRECT, 3);
1649 	ib.ptr[ib.length_dw++] = 8; /* x */
1650 	ib.ptr[ib.length_dw++] = 1; /* y */
1651 	ib.ptr[ib.length_dw++] = 1; /* z */
1652 	ib.ptr[ib.length_dw++] =
1653 		REG_SET_FIELD(0, COMPUTE_DISPATCH_INITIATOR, COMPUTE_SHADER_EN, 1);
1654 
1655 	/* write CS partial flush packet */
1656 	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_EVENT_WRITE, 0);
1657 	ib.ptr[ib.length_dw++] = EVENT_TYPE(7) | EVENT_INDEX(4);
1658 
1659 	/* shedule the ib on the ring */
1660 	r = amdgpu_ib_schedule(ring, 1, &ib, NULL, &f);
1661 	if (r) {
1662 		DRM_ERROR("amdgpu: ib submit failed (%d).\n", r);
1663 		goto fail;
1664 	}
1665 
1666 	/* wait for the GPU to finish processing the IB */
1667 	r = dma_fence_wait(f, false);
1668 	if (r) {
1669 		DRM_ERROR("amdgpu: fence wait failed (%d).\n", r);
1670 		goto fail;
1671 	}
1672 
1673 	tmp = REG_SET_FIELD(tmp, GB_EDC_MODE, DED_MODE, 2);
1674 	tmp = REG_SET_FIELD(tmp, GB_EDC_MODE, PROP_FED, 1);
1675 	WREG32(mmGB_EDC_MODE, tmp);
1676 
1677 	tmp = RREG32(mmCC_GC_EDC_CONFIG);
1678 	tmp = REG_SET_FIELD(tmp, CC_GC_EDC_CONFIG, DIS_EDC, 0) | 1;
1679 	WREG32(mmCC_GC_EDC_CONFIG, tmp);
1680 
1681 
1682 	/* read back registers to clear the counters */
1683 	for (i = 0; i < ARRAY_SIZE(sec_ded_counter_registers); i++)
1684 		RREG32(sec_ded_counter_registers[i]);
1685 
1686 fail:
1687 	amdgpu_ib_free(adev, &ib, NULL);
1688 	dma_fence_put(f);
1689 
1690 	return r;
1691 }
1692 
1693 static int gfx_v8_0_gpu_early_init(struct amdgpu_device *adev)
1694 {
1695 	u32 gb_addr_config;
1696 	u32 mc_arb_ramcfg;
1697 	u32 dimm00_addr_map, dimm01_addr_map, dimm10_addr_map, dimm11_addr_map;
1698 	u32 tmp;
1699 	int ret;
1700 
1701 	switch (adev->asic_type) {
1702 	case CHIP_TOPAZ:
1703 		adev->gfx.config.max_shader_engines = 1;
1704 		adev->gfx.config.max_tile_pipes = 2;
1705 		adev->gfx.config.max_cu_per_sh = 6;
1706 		adev->gfx.config.max_sh_per_se = 1;
1707 		adev->gfx.config.max_backends_per_se = 2;
1708 		adev->gfx.config.max_texture_channel_caches = 2;
1709 		adev->gfx.config.max_gprs = 256;
1710 		adev->gfx.config.max_gs_threads = 32;
1711 		adev->gfx.config.max_hw_contexts = 8;
1712 
1713 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1714 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1715 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1716 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
1717 		gb_addr_config = TOPAZ_GB_ADDR_CONFIG_GOLDEN;
1718 		break;
1719 	case CHIP_FIJI:
1720 		adev->gfx.config.max_shader_engines = 4;
1721 		adev->gfx.config.max_tile_pipes = 16;
1722 		adev->gfx.config.max_cu_per_sh = 16;
1723 		adev->gfx.config.max_sh_per_se = 1;
1724 		adev->gfx.config.max_backends_per_se = 4;
1725 		adev->gfx.config.max_texture_channel_caches = 16;
1726 		adev->gfx.config.max_gprs = 256;
1727 		adev->gfx.config.max_gs_threads = 32;
1728 		adev->gfx.config.max_hw_contexts = 8;
1729 
1730 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1731 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1732 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1733 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
1734 		gb_addr_config = TONGA_GB_ADDR_CONFIG_GOLDEN;
1735 		break;
1736 	case CHIP_POLARIS11:
1737 	case CHIP_POLARIS12:
1738 		ret = amdgpu_atombios_get_gfx_info(adev);
1739 		if (ret)
1740 			return ret;
1741 		adev->gfx.config.max_gprs = 256;
1742 		adev->gfx.config.max_gs_threads = 32;
1743 		adev->gfx.config.max_hw_contexts = 8;
1744 
1745 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1746 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1747 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1748 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
1749 		gb_addr_config = POLARIS11_GB_ADDR_CONFIG_GOLDEN;
1750 		break;
1751 	case CHIP_POLARIS10:
1752 	case CHIP_VEGAM:
1753 		ret = amdgpu_atombios_get_gfx_info(adev);
1754 		if (ret)
1755 			return ret;
1756 		adev->gfx.config.max_gprs = 256;
1757 		adev->gfx.config.max_gs_threads = 32;
1758 		adev->gfx.config.max_hw_contexts = 8;
1759 
1760 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1761 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1762 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1763 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
1764 		gb_addr_config = TONGA_GB_ADDR_CONFIG_GOLDEN;
1765 		break;
1766 	case CHIP_TONGA:
1767 		adev->gfx.config.max_shader_engines = 4;
1768 		adev->gfx.config.max_tile_pipes = 8;
1769 		adev->gfx.config.max_cu_per_sh = 8;
1770 		adev->gfx.config.max_sh_per_se = 1;
1771 		adev->gfx.config.max_backends_per_se = 2;
1772 		adev->gfx.config.max_texture_channel_caches = 8;
1773 		adev->gfx.config.max_gprs = 256;
1774 		adev->gfx.config.max_gs_threads = 32;
1775 		adev->gfx.config.max_hw_contexts = 8;
1776 
1777 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1778 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1779 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1780 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
1781 		gb_addr_config = TONGA_GB_ADDR_CONFIG_GOLDEN;
1782 		break;
1783 	case CHIP_CARRIZO:
1784 		adev->gfx.config.max_shader_engines = 1;
1785 		adev->gfx.config.max_tile_pipes = 2;
1786 		adev->gfx.config.max_sh_per_se = 1;
1787 		adev->gfx.config.max_backends_per_se = 2;
1788 		adev->gfx.config.max_cu_per_sh = 8;
1789 		adev->gfx.config.max_texture_channel_caches = 2;
1790 		adev->gfx.config.max_gprs = 256;
1791 		adev->gfx.config.max_gs_threads = 32;
1792 		adev->gfx.config.max_hw_contexts = 8;
1793 
1794 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1795 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1796 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1797 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
1798 		gb_addr_config = CARRIZO_GB_ADDR_CONFIG_GOLDEN;
1799 		break;
1800 	case CHIP_STONEY:
1801 		adev->gfx.config.max_shader_engines = 1;
1802 		adev->gfx.config.max_tile_pipes = 2;
1803 		adev->gfx.config.max_sh_per_se = 1;
1804 		adev->gfx.config.max_backends_per_se = 1;
1805 		adev->gfx.config.max_cu_per_sh = 3;
1806 		adev->gfx.config.max_texture_channel_caches = 2;
1807 		adev->gfx.config.max_gprs = 256;
1808 		adev->gfx.config.max_gs_threads = 16;
1809 		adev->gfx.config.max_hw_contexts = 8;
1810 
1811 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1812 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1813 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1814 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
1815 		gb_addr_config = CARRIZO_GB_ADDR_CONFIG_GOLDEN;
1816 		break;
1817 	default:
1818 		adev->gfx.config.max_shader_engines = 2;
1819 		adev->gfx.config.max_tile_pipes = 4;
1820 		adev->gfx.config.max_cu_per_sh = 2;
1821 		adev->gfx.config.max_sh_per_se = 1;
1822 		adev->gfx.config.max_backends_per_se = 2;
1823 		adev->gfx.config.max_texture_channel_caches = 4;
1824 		adev->gfx.config.max_gprs = 256;
1825 		adev->gfx.config.max_gs_threads = 32;
1826 		adev->gfx.config.max_hw_contexts = 8;
1827 
1828 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1829 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1830 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1831 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
1832 		gb_addr_config = TONGA_GB_ADDR_CONFIG_GOLDEN;
1833 		break;
1834 	}
1835 
1836 	adev->gfx.config.mc_arb_ramcfg = RREG32(mmMC_ARB_RAMCFG);
1837 	mc_arb_ramcfg = adev->gfx.config.mc_arb_ramcfg;
1838 
1839 	adev->gfx.config.num_banks = REG_GET_FIELD(mc_arb_ramcfg,
1840 				MC_ARB_RAMCFG, NOOFBANK);
1841 	adev->gfx.config.num_ranks = REG_GET_FIELD(mc_arb_ramcfg,
1842 				MC_ARB_RAMCFG, NOOFRANKS);
1843 
1844 	adev->gfx.config.num_tile_pipes = adev->gfx.config.max_tile_pipes;
1845 	adev->gfx.config.mem_max_burst_length_bytes = 256;
1846 	if (adev->flags & AMD_IS_APU) {
1847 		/* Get memory bank mapping mode. */
1848 		tmp = RREG32(mmMC_FUS_DRAM0_BANK_ADDR_MAPPING);
1849 		dimm00_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM0_BANK_ADDR_MAPPING, DIMM0ADDRMAP);
1850 		dimm01_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM0_BANK_ADDR_MAPPING, DIMM1ADDRMAP);
1851 
1852 		tmp = RREG32(mmMC_FUS_DRAM1_BANK_ADDR_MAPPING);
1853 		dimm10_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM1_BANK_ADDR_MAPPING, DIMM0ADDRMAP);
1854 		dimm11_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM1_BANK_ADDR_MAPPING, DIMM1ADDRMAP);
1855 
1856 		/* Validate settings in case only one DIMM installed. */
1857 		if ((dimm00_addr_map == 0) || (dimm00_addr_map == 3) || (dimm00_addr_map == 4) || (dimm00_addr_map > 12))
1858 			dimm00_addr_map = 0;
1859 		if ((dimm01_addr_map == 0) || (dimm01_addr_map == 3) || (dimm01_addr_map == 4) || (dimm01_addr_map > 12))
1860 			dimm01_addr_map = 0;
1861 		if ((dimm10_addr_map == 0) || (dimm10_addr_map == 3) || (dimm10_addr_map == 4) || (dimm10_addr_map > 12))
1862 			dimm10_addr_map = 0;
1863 		if ((dimm11_addr_map == 0) || (dimm11_addr_map == 3) || (dimm11_addr_map == 4) || (dimm11_addr_map > 12))
1864 			dimm11_addr_map = 0;
1865 
1866 		/* If DIMM Addr map is 8GB, ROW size should be 2KB. Otherwise 1KB. */
1867 		/* If ROW size(DIMM1) != ROW size(DMIMM0), ROW size should be larger one. */
1868 		if ((dimm00_addr_map == 11) || (dimm01_addr_map == 11) || (dimm10_addr_map == 11) || (dimm11_addr_map == 11))
1869 			adev->gfx.config.mem_row_size_in_kb = 2;
1870 		else
1871 			adev->gfx.config.mem_row_size_in_kb = 1;
1872 	} else {
1873 		tmp = REG_GET_FIELD(mc_arb_ramcfg, MC_ARB_RAMCFG, NOOFCOLS);
1874 		adev->gfx.config.mem_row_size_in_kb = (4 * (1 << (8 + tmp))) / 1024;
1875 		if (adev->gfx.config.mem_row_size_in_kb > 4)
1876 			adev->gfx.config.mem_row_size_in_kb = 4;
1877 	}
1878 
1879 	adev->gfx.config.shader_engine_tile_size = 32;
1880 	adev->gfx.config.num_gpus = 1;
1881 	adev->gfx.config.multi_gpu_tile_size = 64;
1882 
1883 	/* fix up row size */
1884 	switch (adev->gfx.config.mem_row_size_in_kb) {
1885 	case 1:
1886 	default:
1887 		gb_addr_config = REG_SET_FIELD(gb_addr_config, GB_ADDR_CONFIG, ROW_SIZE, 0);
1888 		break;
1889 	case 2:
1890 		gb_addr_config = REG_SET_FIELD(gb_addr_config, GB_ADDR_CONFIG, ROW_SIZE, 1);
1891 		break;
1892 	case 4:
1893 		gb_addr_config = REG_SET_FIELD(gb_addr_config, GB_ADDR_CONFIG, ROW_SIZE, 2);
1894 		break;
1895 	}
1896 	adev->gfx.config.gb_addr_config = gb_addr_config;
1897 
1898 	return 0;
1899 }
1900 
1901 static int gfx_v8_0_compute_ring_init(struct amdgpu_device *adev, int ring_id,
1902 					int mec, int pipe, int queue)
1903 {
1904 	int r;
1905 	unsigned irq_type;
1906 	struct amdgpu_ring *ring = &adev->gfx.compute_ring[ring_id];
1907 	unsigned int hw_prio;
1908 
1909 	ring = &adev->gfx.compute_ring[ring_id];
1910 
1911 	/* mec0 is me1 */
1912 	ring->me = mec + 1;
1913 	ring->pipe = pipe;
1914 	ring->queue = queue;
1915 
1916 	ring->ring_obj = NULL;
1917 	ring->use_doorbell = true;
1918 	ring->doorbell_index = adev->doorbell_index.mec_ring0 + ring_id;
1919 	ring->eop_gpu_addr = adev->gfx.mec.hpd_eop_gpu_addr
1920 				+ (ring_id * GFX8_MEC_HPD_SIZE);
1921 	sprintf(ring->name, "comp_%d.%d.%d", ring->me, ring->pipe, ring->queue);
1922 
1923 	irq_type = AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP
1924 		+ ((ring->me - 1) * adev->gfx.mec.num_pipe_per_mec)
1925 		+ ring->pipe;
1926 
1927 	hw_prio = amdgpu_gfx_is_high_priority_compute_queue(adev, ring) ?
1928 			AMDGPU_GFX_PIPE_PRIO_HIGH : AMDGPU_RING_PRIO_DEFAULT;
1929 	/* type-2 packets are deprecated on MEC, use type-3 instead */
1930 	r = amdgpu_ring_init(adev, ring, 1024,
1931 			     &adev->gfx.eop_irq, irq_type, hw_prio);
1932 	if (r)
1933 		return r;
1934 
1935 
1936 	return 0;
1937 }
1938 
1939 static void gfx_v8_0_sq_irq_work_func(struct work_struct *work);
1940 
1941 static int gfx_v8_0_sw_init(void *handle)
1942 {
1943 	int i, j, k, r, ring_id;
1944 	struct amdgpu_ring *ring;
1945 	struct amdgpu_kiq *kiq;
1946 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1947 
1948 	switch (adev->asic_type) {
1949 	case CHIP_TONGA:
1950 	case CHIP_CARRIZO:
1951 	case CHIP_FIJI:
1952 	case CHIP_POLARIS10:
1953 	case CHIP_POLARIS11:
1954 	case CHIP_POLARIS12:
1955 	case CHIP_VEGAM:
1956 		adev->gfx.mec.num_mec = 2;
1957 		break;
1958 	case CHIP_TOPAZ:
1959 	case CHIP_STONEY:
1960 	default:
1961 		adev->gfx.mec.num_mec = 1;
1962 		break;
1963 	}
1964 
1965 	adev->gfx.mec.num_pipe_per_mec = 4;
1966 	adev->gfx.mec.num_queue_per_pipe = 8;
1967 
1968 	/* EOP Event */
1969 	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, VISLANDS30_IV_SRCID_CP_END_OF_PIPE, &adev->gfx.eop_irq);
1970 	if (r)
1971 		return r;
1972 
1973 	/* Privileged reg */
1974 	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, VISLANDS30_IV_SRCID_CP_PRIV_REG_FAULT,
1975 			      &adev->gfx.priv_reg_irq);
1976 	if (r)
1977 		return r;
1978 
1979 	/* Privileged inst */
1980 	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, VISLANDS30_IV_SRCID_CP_PRIV_INSTR_FAULT,
1981 			      &adev->gfx.priv_inst_irq);
1982 	if (r)
1983 		return r;
1984 
1985 	/* Add CP EDC/ECC irq  */
1986 	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, VISLANDS30_IV_SRCID_CP_ECC_ERROR,
1987 			      &adev->gfx.cp_ecc_error_irq);
1988 	if (r)
1989 		return r;
1990 
1991 	/* SQ interrupts. */
1992 	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, VISLANDS30_IV_SRCID_SQ_INTERRUPT_MSG,
1993 			      &adev->gfx.sq_irq);
1994 	if (r) {
1995 		DRM_ERROR("amdgpu_irq_add() for SQ failed: %d\n", r);
1996 		return r;
1997 	}
1998 
1999 	INIT_WORK(&adev->gfx.sq_work.work, gfx_v8_0_sq_irq_work_func);
2000 
2001 	adev->gfx.gfx_current_status = AMDGPU_GFX_NORMAL_MODE;
2002 
2003 	gfx_v8_0_scratch_init(adev);
2004 
2005 	r = gfx_v8_0_init_microcode(adev);
2006 	if (r) {
2007 		DRM_ERROR("Failed to load gfx firmware!\n");
2008 		return r;
2009 	}
2010 
2011 	r = adev->gfx.rlc.funcs->init(adev);
2012 	if (r) {
2013 		DRM_ERROR("Failed to init rlc BOs!\n");
2014 		return r;
2015 	}
2016 
2017 	r = gfx_v8_0_mec_init(adev);
2018 	if (r) {
2019 		DRM_ERROR("Failed to init MEC BOs!\n");
2020 		return r;
2021 	}
2022 
2023 	/* set up the gfx ring */
2024 	for (i = 0; i < adev->gfx.num_gfx_rings; i++) {
2025 		ring = &adev->gfx.gfx_ring[i];
2026 		ring->ring_obj = NULL;
2027 		sprintf(ring->name, "gfx");
2028 		/* no gfx doorbells on iceland */
2029 		if (adev->asic_type != CHIP_TOPAZ) {
2030 			ring->use_doorbell = true;
2031 			ring->doorbell_index = adev->doorbell_index.gfx_ring0;
2032 		}
2033 
2034 		r = amdgpu_ring_init(adev, ring, 1024, &adev->gfx.eop_irq,
2035 				     AMDGPU_CP_IRQ_GFX_ME0_PIPE0_EOP,
2036 				     AMDGPU_RING_PRIO_DEFAULT);
2037 		if (r)
2038 			return r;
2039 	}
2040 
2041 
2042 	/* set up the compute queues - allocate horizontally across pipes */
2043 	ring_id = 0;
2044 	for (i = 0; i < adev->gfx.mec.num_mec; ++i) {
2045 		for (j = 0; j < adev->gfx.mec.num_queue_per_pipe; j++) {
2046 			for (k = 0; k < adev->gfx.mec.num_pipe_per_mec; k++) {
2047 				if (!amdgpu_gfx_is_mec_queue_enabled(adev, i, k, j))
2048 					continue;
2049 
2050 				r = gfx_v8_0_compute_ring_init(adev,
2051 								ring_id,
2052 								i, k, j);
2053 				if (r)
2054 					return r;
2055 
2056 				ring_id++;
2057 			}
2058 		}
2059 	}
2060 
2061 	r = amdgpu_gfx_kiq_init(adev, GFX8_MEC_HPD_SIZE);
2062 	if (r) {
2063 		DRM_ERROR("Failed to init KIQ BOs!\n");
2064 		return r;
2065 	}
2066 
2067 	kiq = &adev->gfx.kiq;
2068 	r = amdgpu_gfx_kiq_init_ring(adev, &kiq->ring, &kiq->irq);
2069 	if (r)
2070 		return r;
2071 
2072 	/* create MQD for all compute queues as well as KIQ for SRIOV case */
2073 	r = amdgpu_gfx_mqd_sw_init(adev, sizeof(struct vi_mqd_allocation));
2074 	if (r)
2075 		return r;
2076 
2077 	adev->gfx.ce_ram_size = 0x8000;
2078 
2079 	r = gfx_v8_0_gpu_early_init(adev);
2080 	if (r)
2081 		return r;
2082 
2083 	return 0;
2084 }
2085 
2086 static int gfx_v8_0_sw_fini(void *handle)
2087 {
2088 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2089 	int i;
2090 
2091 	for (i = 0; i < adev->gfx.num_gfx_rings; i++)
2092 		amdgpu_ring_fini(&adev->gfx.gfx_ring[i]);
2093 	for (i = 0; i < adev->gfx.num_compute_rings; i++)
2094 		amdgpu_ring_fini(&adev->gfx.compute_ring[i]);
2095 
2096 	amdgpu_gfx_mqd_sw_fini(adev);
2097 	amdgpu_gfx_kiq_free_ring(&adev->gfx.kiq.ring);
2098 	amdgpu_gfx_kiq_fini(adev);
2099 
2100 	gfx_v8_0_mec_fini(adev);
2101 	amdgpu_gfx_rlc_fini(adev);
2102 	amdgpu_bo_free_kernel(&adev->gfx.rlc.clear_state_obj,
2103 				&adev->gfx.rlc.clear_state_gpu_addr,
2104 				(void **)&adev->gfx.rlc.cs_ptr);
2105 	if ((adev->asic_type == CHIP_CARRIZO) ||
2106 	    (adev->asic_type == CHIP_STONEY)) {
2107 		amdgpu_bo_free_kernel(&adev->gfx.rlc.cp_table_obj,
2108 				&adev->gfx.rlc.cp_table_gpu_addr,
2109 				(void **)&adev->gfx.rlc.cp_table_ptr);
2110 	}
2111 	gfx_v8_0_free_microcode(adev);
2112 
2113 	return 0;
2114 }
2115 
2116 static void gfx_v8_0_tiling_mode_table_init(struct amdgpu_device *adev)
2117 {
2118 	uint32_t *modearray, *mod2array;
2119 	const u32 num_tile_mode_states = ARRAY_SIZE(adev->gfx.config.tile_mode_array);
2120 	const u32 num_secondary_tile_mode_states = ARRAY_SIZE(adev->gfx.config.macrotile_mode_array);
2121 	u32 reg_offset;
2122 
2123 	modearray = adev->gfx.config.tile_mode_array;
2124 	mod2array = adev->gfx.config.macrotile_mode_array;
2125 
2126 	for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
2127 		modearray[reg_offset] = 0;
2128 
2129 	for (reg_offset = 0; reg_offset <  num_secondary_tile_mode_states; reg_offset++)
2130 		mod2array[reg_offset] = 0;
2131 
2132 	switch (adev->asic_type) {
2133 	case CHIP_TOPAZ:
2134 		modearray[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2135 				PIPE_CONFIG(ADDR_SURF_P2) |
2136 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
2137 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2138 		modearray[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2139 				PIPE_CONFIG(ADDR_SURF_P2) |
2140 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
2141 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2142 		modearray[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2143 				PIPE_CONFIG(ADDR_SURF_P2) |
2144 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
2145 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2146 		modearray[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2147 				PIPE_CONFIG(ADDR_SURF_P2) |
2148 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
2149 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2150 		modearray[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2151 				PIPE_CONFIG(ADDR_SURF_P2) |
2152 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2153 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2154 		modearray[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2155 				PIPE_CONFIG(ADDR_SURF_P2) |
2156 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2157 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2158 		modearray[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2159 				PIPE_CONFIG(ADDR_SURF_P2) |
2160 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2161 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2162 		modearray[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
2163 				PIPE_CONFIG(ADDR_SURF_P2));
2164 		modearray[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2165 				PIPE_CONFIG(ADDR_SURF_P2) |
2166 				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2167 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2168 		modearray[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2169 				 PIPE_CONFIG(ADDR_SURF_P2) |
2170 				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2171 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2172 		modearray[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2173 				 PIPE_CONFIG(ADDR_SURF_P2) |
2174 				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2175 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2176 		modearray[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2177 				 PIPE_CONFIG(ADDR_SURF_P2) |
2178 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2179 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2180 		modearray[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2181 				 PIPE_CONFIG(ADDR_SURF_P2) |
2182 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2183 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2184 		modearray[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
2185 				 PIPE_CONFIG(ADDR_SURF_P2) |
2186 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2187 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2188 		modearray[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2189 				 PIPE_CONFIG(ADDR_SURF_P2) |
2190 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2191 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2192 		modearray[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2193 				 PIPE_CONFIG(ADDR_SURF_P2) |
2194 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2195 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2196 		modearray[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2197 				 PIPE_CONFIG(ADDR_SURF_P2) |
2198 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2199 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2200 		modearray[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
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[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
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[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
2209 				 PIPE_CONFIG(ADDR_SURF_P2) |
2210 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2211 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2212 		modearray[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2213 				 PIPE_CONFIG(ADDR_SURF_P2) |
2214 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2215 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2216 		modearray[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
2217 				 PIPE_CONFIG(ADDR_SURF_P2) |
2218 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2219 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2220 		modearray[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
2221 				 PIPE_CONFIG(ADDR_SURF_P2) |
2222 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2223 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2224 		modearray[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2225 				 PIPE_CONFIG(ADDR_SURF_P2) |
2226 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2227 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2228 		modearray[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2229 				 PIPE_CONFIG(ADDR_SURF_P2) |
2230 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2231 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2232 		modearray[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2233 				 PIPE_CONFIG(ADDR_SURF_P2) |
2234 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2235 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2236 
2237 		mod2array[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
2238 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2239 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2240 				NUM_BANKS(ADDR_SURF_8_BANK));
2241 		mod2array[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
2242 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2243 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2244 				NUM_BANKS(ADDR_SURF_8_BANK));
2245 		mod2array[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
2246 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2247 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2248 				NUM_BANKS(ADDR_SURF_8_BANK));
2249 		mod2array[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2250 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2251 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2252 				NUM_BANKS(ADDR_SURF_8_BANK));
2253 		mod2array[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2254 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2255 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2256 				NUM_BANKS(ADDR_SURF_8_BANK));
2257 		mod2array[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2258 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2259 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2260 				NUM_BANKS(ADDR_SURF_8_BANK));
2261 		mod2array[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2262 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2263 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2264 				NUM_BANKS(ADDR_SURF_8_BANK));
2265 		mod2array[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
2266 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
2267 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2268 				NUM_BANKS(ADDR_SURF_16_BANK));
2269 		mod2array[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
2270 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2271 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2272 				NUM_BANKS(ADDR_SURF_16_BANK));
2273 		mod2array[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
2274 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2275 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2276 				 NUM_BANKS(ADDR_SURF_16_BANK));
2277 		mod2array[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
2278 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2279 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2280 				 NUM_BANKS(ADDR_SURF_16_BANK));
2281 		mod2array[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2282 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2283 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2284 				 NUM_BANKS(ADDR_SURF_16_BANK));
2285 		mod2array[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2286 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2287 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2288 				 NUM_BANKS(ADDR_SURF_16_BANK));
2289 		mod2array[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2290 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2291 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2292 				 NUM_BANKS(ADDR_SURF_8_BANK));
2293 
2294 		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
2295 			if (reg_offset != 7 && reg_offset != 12 && reg_offset != 17 &&
2296 			    reg_offset != 23)
2297 				WREG32(mmGB_TILE_MODE0 + reg_offset, modearray[reg_offset]);
2298 
2299 		for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
2300 			if (reg_offset != 7)
2301 				WREG32(mmGB_MACROTILE_MODE0 + reg_offset, mod2array[reg_offset]);
2302 
2303 		break;
2304 	case CHIP_FIJI:
2305 	case CHIP_VEGAM:
2306 		modearray[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2307 				PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2308 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
2309 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2310 		modearray[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2311 				PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2312 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
2313 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2314 		modearray[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2315 				PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2316 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
2317 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2318 		modearray[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2319 				PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2320 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
2321 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2322 		modearray[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2323 				PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2324 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2325 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2326 		modearray[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2327 				PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2328 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2329 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2330 		modearray[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2331 				PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2332 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2333 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2334 		modearray[7] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2335 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2336 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2337 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2338 		modearray[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
2339 				PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16));
2340 		modearray[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2341 				PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2342 				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2343 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2344 		modearray[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2345 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2346 				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2347 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2348 		modearray[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2349 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2350 				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2351 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2352 		modearray[12] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2353 				 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2354 				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2355 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2356 		modearray[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2357 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2358 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2359 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2360 		modearray[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
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_2));
2364 		modearray[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
2365 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2366 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2367 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2368 		modearray[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2369 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2370 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2371 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2372 		modearray[17] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2373 				 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2374 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2375 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2376 		modearray[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2377 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2378 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2379 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2380 		modearray[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2381 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2382 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2383 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2384 		modearray[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2385 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2386 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2387 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2388 		modearray[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
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[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
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[23] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
2397 				 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2398 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2399 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2400 		modearray[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2401 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2402 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2403 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2404 		modearray[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
2405 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2406 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2407 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2408 		modearray[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
2409 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2410 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2411 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2412 		modearray[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2413 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2414 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2415 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2416 		modearray[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2417 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2418 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2419 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2420 		modearray[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2421 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2422 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2423 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2424 		modearray[30] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2425 				 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2426 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2427 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2428 
2429 		mod2array[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2430 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2431 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2432 				NUM_BANKS(ADDR_SURF_8_BANK));
2433 		mod2array[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2434 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2435 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2436 				NUM_BANKS(ADDR_SURF_8_BANK));
2437 		mod2array[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2438 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2439 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2440 				NUM_BANKS(ADDR_SURF_8_BANK));
2441 		mod2array[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2442 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2443 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2444 				NUM_BANKS(ADDR_SURF_8_BANK));
2445 		mod2array[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2446 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2447 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2448 				NUM_BANKS(ADDR_SURF_8_BANK));
2449 		mod2array[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2450 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2451 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2452 				NUM_BANKS(ADDR_SURF_8_BANK));
2453 		mod2array[6] = (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[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2458 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
2459 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2460 				NUM_BANKS(ADDR_SURF_8_BANK));
2461 		mod2array[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2462 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2463 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2464 				NUM_BANKS(ADDR_SURF_8_BANK));
2465 		mod2array[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2466 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2467 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2468 				 NUM_BANKS(ADDR_SURF_8_BANK));
2469 		mod2array[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2470 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2471 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2472 				 NUM_BANKS(ADDR_SURF_8_BANK));
2473 		mod2array[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2474 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2475 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2476 				 NUM_BANKS(ADDR_SURF_8_BANK));
2477 		mod2array[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2478 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2479 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2480 				 NUM_BANKS(ADDR_SURF_8_BANK));
2481 		mod2array[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2482 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2483 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2484 				 NUM_BANKS(ADDR_SURF_4_BANK));
2485 
2486 		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
2487 			WREG32(mmGB_TILE_MODE0 + reg_offset, modearray[reg_offset]);
2488 
2489 		for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
2490 			if (reg_offset != 7)
2491 				WREG32(mmGB_MACROTILE_MODE0 + reg_offset, mod2array[reg_offset]);
2492 
2493 		break;
2494 	case CHIP_TONGA:
2495 		modearray[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2496 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2497 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
2498 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2499 		modearray[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2500 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2501 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
2502 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2503 		modearray[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2504 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2505 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
2506 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2507 		modearray[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2508 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2509 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
2510 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2511 		modearray[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2512 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2513 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2514 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2515 		modearray[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2516 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2517 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2518 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2519 		modearray[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2520 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2521 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2522 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2523 		modearray[7] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2524 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2525 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2526 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2527 		modearray[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
2528 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16));
2529 		modearray[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2530 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2531 				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2532 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2533 		modearray[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2534 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2535 				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2536 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2537 		modearray[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2538 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2539 				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2540 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2541 		modearray[12] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2542 				 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2543 				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2544 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2545 		modearray[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2546 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2547 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2548 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2549 		modearray[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
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_2));
2553 		modearray[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
2554 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2555 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2556 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2557 		modearray[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2558 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2559 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2560 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2561 		modearray[17] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2562 				 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2563 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2564 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2565 		modearray[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2566 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2567 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2568 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2569 		modearray[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2570 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2571 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2572 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2573 		modearray[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2574 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2575 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2576 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2577 		modearray[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
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[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
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[23] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
2586 				 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2587 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2588 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2589 		modearray[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2590 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2591 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2592 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2593 		modearray[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
2594 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2595 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2596 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2597 		modearray[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
2598 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2599 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2600 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2601 		modearray[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2602 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2603 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2604 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2605 		modearray[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2606 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2607 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2608 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2609 		modearray[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2610 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2611 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2612 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2613 		modearray[30] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2614 				 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2615 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2616 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2617 
2618 		mod2array[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2619 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2620 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2621 				NUM_BANKS(ADDR_SURF_16_BANK));
2622 		mod2array[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2623 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2624 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2625 				NUM_BANKS(ADDR_SURF_16_BANK));
2626 		mod2array[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2627 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2628 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2629 				NUM_BANKS(ADDR_SURF_16_BANK));
2630 		mod2array[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2631 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2632 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2633 				NUM_BANKS(ADDR_SURF_16_BANK));
2634 		mod2array[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2635 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2636 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2637 				NUM_BANKS(ADDR_SURF_16_BANK));
2638 		mod2array[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2639 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2640 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2641 				NUM_BANKS(ADDR_SURF_16_BANK));
2642 		mod2array[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2643 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2644 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2645 				NUM_BANKS(ADDR_SURF_16_BANK));
2646 		mod2array[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2647 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
2648 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2649 				NUM_BANKS(ADDR_SURF_16_BANK));
2650 		mod2array[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2651 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2652 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2653 				NUM_BANKS(ADDR_SURF_16_BANK));
2654 		mod2array[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2655 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2656 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2657 				 NUM_BANKS(ADDR_SURF_16_BANK));
2658 		mod2array[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2659 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2660 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2661 				 NUM_BANKS(ADDR_SURF_16_BANK));
2662 		mod2array[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2663 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2664 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2665 				 NUM_BANKS(ADDR_SURF_8_BANK));
2666 		mod2array[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2667 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2668 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2669 				 NUM_BANKS(ADDR_SURF_4_BANK));
2670 		mod2array[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2671 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2672 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2673 				 NUM_BANKS(ADDR_SURF_4_BANK));
2674 
2675 		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
2676 			WREG32(mmGB_TILE_MODE0 + reg_offset, modearray[reg_offset]);
2677 
2678 		for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
2679 			if (reg_offset != 7)
2680 				WREG32(mmGB_MACROTILE_MODE0 + reg_offset, mod2array[reg_offset]);
2681 
2682 		break;
2683 	case CHIP_POLARIS11:
2684 	case CHIP_POLARIS12:
2685 		modearray[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2686 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2687 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
2688 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2689 		modearray[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2690 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2691 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
2692 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2693 		modearray[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2694 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2695 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
2696 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2697 		modearray[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2698 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2699 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
2700 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2701 		modearray[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2702 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2703 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2704 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2705 		modearray[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2706 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2707 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2708 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2709 		modearray[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2710 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2711 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2712 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2713 		modearray[7] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2714 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2715 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2716 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2717 		modearray[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
2718 				PIPE_CONFIG(ADDR_SURF_P4_16x16));
2719 		modearray[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2720 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2721 				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2722 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2723 		modearray[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2724 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2725 				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2726 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2727 		modearray[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2728 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2729 				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2730 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2731 		modearray[12] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2732 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2733 				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2734 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2735 		modearray[13] = (ARRAY_MODE(ARRAY_1D_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_2));
2739 		modearray[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2740 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2741 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2742 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2743 		modearray[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
2744 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2745 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2746 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2747 		modearray[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2748 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2749 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2750 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2751 		modearray[17] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2752 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2753 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2754 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2755 		modearray[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2756 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2757 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2758 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2759 		modearray[19] = (ARRAY_MODE(ARRAY_1D_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[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2764 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2765 				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2766 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2767 		modearray[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
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[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
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[23] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
2776 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2777 				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2778 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2779 		modearray[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2780 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2781 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2782 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2783 		modearray[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
2784 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2785 				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2786 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2787 		modearray[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
2788 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2789 				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2790 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2791 		modearray[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2792 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2793 				MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2794 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2795 		modearray[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2796 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2797 				MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2798 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2799 		modearray[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2800 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2801 				MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2802 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2803 		modearray[30] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2804 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2805 				MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2806 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2807 
2808 		mod2array[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2809 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2810 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2811 				NUM_BANKS(ADDR_SURF_16_BANK));
2812 
2813 		mod2array[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2814 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2815 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2816 				NUM_BANKS(ADDR_SURF_16_BANK));
2817 
2818 		mod2array[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2819 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2820 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2821 				NUM_BANKS(ADDR_SURF_16_BANK));
2822 
2823 		mod2array[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2824 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2825 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2826 				NUM_BANKS(ADDR_SURF_16_BANK));
2827 
2828 		mod2array[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2829 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2830 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2831 				NUM_BANKS(ADDR_SURF_16_BANK));
2832 
2833 		mod2array[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2834 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2835 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2836 				NUM_BANKS(ADDR_SURF_16_BANK));
2837 
2838 		mod2array[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2839 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2840 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2841 				NUM_BANKS(ADDR_SURF_16_BANK));
2842 
2843 		mod2array[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
2844 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
2845 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2846 				NUM_BANKS(ADDR_SURF_16_BANK));
2847 
2848 		mod2array[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
2849 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2850 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2851 				NUM_BANKS(ADDR_SURF_16_BANK));
2852 
2853 		mod2array[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2854 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2855 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2856 				NUM_BANKS(ADDR_SURF_16_BANK));
2857 
2858 		mod2array[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2859 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2860 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2861 				NUM_BANKS(ADDR_SURF_16_BANK));
2862 
2863 		mod2array[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2864 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2865 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2866 				NUM_BANKS(ADDR_SURF_16_BANK));
2867 
2868 		mod2array[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2869 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2870 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2871 				NUM_BANKS(ADDR_SURF_8_BANK));
2872 
2873 		mod2array[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2874 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2875 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2876 				NUM_BANKS(ADDR_SURF_4_BANK));
2877 
2878 		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
2879 			WREG32(mmGB_TILE_MODE0 + reg_offset, modearray[reg_offset]);
2880 
2881 		for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
2882 			if (reg_offset != 7)
2883 				WREG32(mmGB_MACROTILE_MODE0 + reg_offset, mod2array[reg_offset]);
2884 
2885 		break;
2886 	case CHIP_POLARIS10:
2887 		modearray[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2888 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2889 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
2890 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2891 		modearray[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2892 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2893 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
2894 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2895 		modearray[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2896 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2897 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
2898 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2899 		modearray[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2900 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2901 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
2902 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2903 		modearray[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2904 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2905 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2906 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2907 		modearray[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2908 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2909 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2910 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2911 		modearray[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2912 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2913 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2914 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2915 		modearray[7] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2916 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2917 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2918 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2919 		modearray[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
2920 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16));
2921 		modearray[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2922 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2923 				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2924 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2925 		modearray[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2926 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2927 				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2928 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2929 		modearray[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2930 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2931 				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2932 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2933 		modearray[12] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2934 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2935 				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2936 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2937 		modearray[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2938 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2939 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2940 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2941 		modearray[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
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_2));
2945 		modearray[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
2946 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2947 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2948 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2949 		modearray[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2950 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2951 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2952 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2953 		modearray[17] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2954 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2955 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2956 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2957 		modearray[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2958 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2959 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2960 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2961 		modearray[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2962 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2963 				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2964 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2965 		modearray[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2966 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2967 				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2968 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2969 		modearray[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
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[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
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[23] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
2978 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2979 				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2980 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2981 		modearray[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2982 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2983 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2984 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2985 		modearray[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
2986 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2987 				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2988 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2989 		modearray[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
2990 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2991 				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2992 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2993 		modearray[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2994 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2995 				MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2996 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2997 		modearray[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2998 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2999 				MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
3000 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3001 		modearray[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
3002 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
3003 				MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
3004 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
3005 		modearray[30] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
3006 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
3007 				MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
3008 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
3009 
3010 		mod2array[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3011 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
3012 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3013 				NUM_BANKS(ADDR_SURF_16_BANK));
3014 
3015 		mod2array[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3016 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
3017 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3018 				NUM_BANKS(ADDR_SURF_16_BANK));
3019 
3020 		mod2array[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3021 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
3022 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3023 				NUM_BANKS(ADDR_SURF_16_BANK));
3024 
3025 		mod2array[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3026 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
3027 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3028 				NUM_BANKS(ADDR_SURF_16_BANK));
3029 
3030 		mod2array[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3031 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
3032 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3033 				NUM_BANKS(ADDR_SURF_16_BANK));
3034 
3035 		mod2array[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3036 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3037 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
3038 				NUM_BANKS(ADDR_SURF_16_BANK));
3039 
3040 		mod2array[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3041 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3042 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
3043 				NUM_BANKS(ADDR_SURF_16_BANK));
3044 
3045 		mod2array[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3046 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
3047 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3048 				NUM_BANKS(ADDR_SURF_16_BANK));
3049 
3050 		mod2array[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3051 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
3052 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3053 				NUM_BANKS(ADDR_SURF_16_BANK));
3054 
3055 		mod2array[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3056 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
3057 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3058 				NUM_BANKS(ADDR_SURF_16_BANK));
3059 
3060 		mod2array[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3061 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3062 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3063 				NUM_BANKS(ADDR_SURF_16_BANK));
3064 
3065 		mod2array[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3066 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3067 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
3068 				NUM_BANKS(ADDR_SURF_8_BANK));
3069 
3070 		mod2array[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3071 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3072 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
3073 				NUM_BANKS(ADDR_SURF_4_BANK));
3074 
3075 		mod2array[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3076 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3077 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
3078 				NUM_BANKS(ADDR_SURF_4_BANK));
3079 
3080 		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
3081 			WREG32(mmGB_TILE_MODE0 + reg_offset, modearray[reg_offset]);
3082 
3083 		for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
3084 			if (reg_offset != 7)
3085 				WREG32(mmGB_MACROTILE_MODE0 + reg_offset, mod2array[reg_offset]);
3086 
3087 		break;
3088 	case CHIP_STONEY:
3089 		modearray[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3090 				PIPE_CONFIG(ADDR_SURF_P2) |
3091 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
3092 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3093 		modearray[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3094 				PIPE_CONFIG(ADDR_SURF_P2) |
3095 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
3096 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3097 		modearray[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3098 				PIPE_CONFIG(ADDR_SURF_P2) |
3099 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
3100 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3101 		modearray[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3102 				PIPE_CONFIG(ADDR_SURF_P2) |
3103 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
3104 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3105 		modearray[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3106 				PIPE_CONFIG(ADDR_SURF_P2) |
3107 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
3108 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3109 		modearray[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
3110 				PIPE_CONFIG(ADDR_SURF_P2) |
3111 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
3112 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3113 		modearray[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
3114 				PIPE_CONFIG(ADDR_SURF_P2) |
3115 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
3116 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3117 		modearray[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
3118 				PIPE_CONFIG(ADDR_SURF_P2));
3119 		modearray[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
3120 				PIPE_CONFIG(ADDR_SURF_P2) |
3121 				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
3122 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3123 		modearray[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3124 				 PIPE_CONFIG(ADDR_SURF_P2) |
3125 				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
3126 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3127 		modearray[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
3128 				 PIPE_CONFIG(ADDR_SURF_P2) |
3129 				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
3130 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
3131 		modearray[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
3132 				 PIPE_CONFIG(ADDR_SURF_P2) |
3133 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3134 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3135 		modearray[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3136 				 PIPE_CONFIG(ADDR_SURF_P2) |
3137 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3138 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3139 		modearray[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
3140 				 PIPE_CONFIG(ADDR_SURF_P2) |
3141 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3142 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3143 		modearray[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
3144 				 PIPE_CONFIG(ADDR_SURF_P2) |
3145 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3146 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
3147 		modearray[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
3148 				 PIPE_CONFIG(ADDR_SURF_P2) |
3149 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3150 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3151 		modearray[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
3152 				 PIPE_CONFIG(ADDR_SURF_P2) |
3153 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3154 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3155 		modearray[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
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[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
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[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
3164 				 PIPE_CONFIG(ADDR_SURF_P2) |
3165 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3166 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3167 		modearray[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
3168 				 PIPE_CONFIG(ADDR_SURF_P2) |
3169 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3170 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3171 		modearray[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
3172 				 PIPE_CONFIG(ADDR_SURF_P2) |
3173 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3174 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3175 		modearray[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
3176 				 PIPE_CONFIG(ADDR_SURF_P2) |
3177 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3178 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3179 		modearray[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
3180 				 PIPE_CONFIG(ADDR_SURF_P2) |
3181 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
3182 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3183 		modearray[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3184 				 PIPE_CONFIG(ADDR_SURF_P2) |
3185 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
3186 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3187 		modearray[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
3188 				 PIPE_CONFIG(ADDR_SURF_P2) |
3189 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
3190 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
3191 
3192 		mod2array[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3193 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
3194 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3195 				NUM_BANKS(ADDR_SURF_8_BANK));
3196 		mod2array[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3197 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
3198 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3199 				NUM_BANKS(ADDR_SURF_8_BANK));
3200 		mod2array[2] = (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[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3205 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3206 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3207 				NUM_BANKS(ADDR_SURF_8_BANK));
3208 		mod2array[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3209 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3210 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3211 				NUM_BANKS(ADDR_SURF_8_BANK));
3212 		mod2array[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3213 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3214 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3215 				NUM_BANKS(ADDR_SURF_8_BANK));
3216 		mod2array[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3217 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3218 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3219 				NUM_BANKS(ADDR_SURF_8_BANK));
3220 		mod2array[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
3221 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
3222 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3223 				NUM_BANKS(ADDR_SURF_16_BANK));
3224 		mod2array[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
3225 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
3226 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3227 				NUM_BANKS(ADDR_SURF_16_BANK));
3228 		mod2array[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
3229 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
3230 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3231 				 NUM_BANKS(ADDR_SURF_16_BANK));
3232 		mod2array[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
3233 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
3234 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3235 				 NUM_BANKS(ADDR_SURF_16_BANK));
3236 		mod2array[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3237 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
3238 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3239 				 NUM_BANKS(ADDR_SURF_16_BANK));
3240 		mod2array[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3241 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3242 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3243 				 NUM_BANKS(ADDR_SURF_16_BANK));
3244 		mod2array[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3245 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3246 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3247 				 NUM_BANKS(ADDR_SURF_8_BANK));
3248 
3249 		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
3250 			if (reg_offset != 7 && reg_offset != 12 && reg_offset != 17 &&
3251 			    reg_offset != 23)
3252 				WREG32(mmGB_TILE_MODE0 + reg_offset, modearray[reg_offset]);
3253 
3254 		for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
3255 			if (reg_offset != 7)
3256 				WREG32(mmGB_MACROTILE_MODE0 + reg_offset, mod2array[reg_offset]);
3257 
3258 		break;
3259 	default:
3260 		dev_warn(adev->dev,
3261 			 "Unknown chip type (%d) in function gfx_v8_0_tiling_mode_table_init() falling through to CHIP_CARRIZO\n",
3262 			 adev->asic_type);
3263 		fallthrough;
3264 
3265 	case CHIP_CARRIZO:
3266 		modearray[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3267 				PIPE_CONFIG(ADDR_SURF_P2) |
3268 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
3269 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3270 		modearray[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3271 				PIPE_CONFIG(ADDR_SURF_P2) |
3272 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
3273 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3274 		modearray[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3275 				PIPE_CONFIG(ADDR_SURF_P2) |
3276 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
3277 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3278 		modearray[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3279 				PIPE_CONFIG(ADDR_SURF_P2) |
3280 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
3281 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3282 		modearray[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3283 				PIPE_CONFIG(ADDR_SURF_P2) |
3284 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
3285 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3286 		modearray[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
3287 				PIPE_CONFIG(ADDR_SURF_P2) |
3288 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
3289 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3290 		modearray[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
3291 				PIPE_CONFIG(ADDR_SURF_P2) |
3292 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
3293 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3294 		modearray[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
3295 				PIPE_CONFIG(ADDR_SURF_P2));
3296 		modearray[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
3297 				PIPE_CONFIG(ADDR_SURF_P2) |
3298 				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
3299 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3300 		modearray[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3301 				 PIPE_CONFIG(ADDR_SURF_P2) |
3302 				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
3303 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3304 		modearray[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
3305 				 PIPE_CONFIG(ADDR_SURF_P2) |
3306 				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
3307 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
3308 		modearray[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
3309 				 PIPE_CONFIG(ADDR_SURF_P2) |
3310 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3311 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3312 		modearray[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3313 				 PIPE_CONFIG(ADDR_SURF_P2) |
3314 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3315 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3316 		modearray[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
3317 				 PIPE_CONFIG(ADDR_SURF_P2) |
3318 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3319 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3320 		modearray[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
3321 				 PIPE_CONFIG(ADDR_SURF_P2) |
3322 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3323 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
3324 		modearray[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
3325 				 PIPE_CONFIG(ADDR_SURF_P2) |
3326 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3327 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3328 		modearray[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
3329 				 PIPE_CONFIG(ADDR_SURF_P2) |
3330 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3331 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3332 		modearray[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
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[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
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[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
3341 				 PIPE_CONFIG(ADDR_SURF_P2) |
3342 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3343 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3344 		modearray[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
3345 				 PIPE_CONFIG(ADDR_SURF_P2) |
3346 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3347 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3348 		modearray[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
3349 				 PIPE_CONFIG(ADDR_SURF_P2) |
3350 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3351 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3352 		modearray[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
3353 				 PIPE_CONFIG(ADDR_SURF_P2) |
3354 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3355 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3356 		modearray[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
3357 				 PIPE_CONFIG(ADDR_SURF_P2) |
3358 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
3359 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3360 		modearray[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3361 				 PIPE_CONFIG(ADDR_SURF_P2) |
3362 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
3363 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3364 		modearray[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
3365 				 PIPE_CONFIG(ADDR_SURF_P2) |
3366 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
3367 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
3368 
3369 		mod2array[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3370 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
3371 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3372 				NUM_BANKS(ADDR_SURF_8_BANK));
3373 		mod2array[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3374 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
3375 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3376 				NUM_BANKS(ADDR_SURF_8_BANK));
3377 		mod2array[2] = (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[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3382 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3383 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3384 				NUM_BANKS(ADDR_SURF_8_BANK));
3385 		mod2array[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3386 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3387 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3388 				NUM_BANKS(ADDR_SURF_8_BANK));
3389 		mod2array[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3390 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3391 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3392 				NUM_BANKS(ADDR_SURF_8_BANK));
3393 		mod2array[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3394 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3395 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3396 				NUM_BANKS(ADDR_SURF_8_BANK));
3397 		mod2array[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
3398 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
3399 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3400 				NUM_BANKS(ADDR_SURF_16_BANK));
3401 		mod2array[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
3402 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
3403 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3404 				NUM_BANKS(ADDR_SURF_16_BANK));
3405 		mod2array[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
3406 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
3407 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3408 				 NUM_BANKS(ADDR_SURF_16_BANK));
3409 		mod2array[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
3410 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
3411 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3412 				 NUM_BANKS(ADDR_SURF_16_BANK));
3413 		mod2array[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3414 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
3415 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3416 				 NUM_BANKS(ADDR_SURF_16_BANK));
3417 		mod2array[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3418 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3419 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3420 				 NUM_BANKS(ADDR_SURF_16_BANK));
3421 		mod2array[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3422 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3423 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3424 				 NUM_BANKS(ADDR_SURF_8_BANK));
3425 
3426 		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
3427 			if (reg_offset != 7 && reg_offset != 12 && reg_offset != 17 &&
3428 			    reg_offset != 23)
3429 				WREG32(mmGB_TILE_MODE0 + reg_offset, modearray[reg_offset]);
3430 
3431 		for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
3432 			if (reg_offset != 7)
3433 				WREG32(mmGB_MACROTILE_MODE0 + reg_offset, mod2array[reg_offset]);
3434 
3435 		break;
3436 	}
3437 }
3438 
3439 static void gfx_v8_0_select_se_sh(struct amdgpu_device *adev,
3440 				  u32 se_num, u32 sh_num, u32 instance)
3441 {
3442 	u32 data;
3443 
3444 	if (instance == 0xffffffff)
3445 		data = REG_SET_FIELD(0, GRBM_GFX_INDEX, INSTANCE_BROADCAST_WRITES, 1);
3446 	else
3447 		data = REG_SET_FIELD(0, GRBM_GFX_INDEX, INSTANCE_INDEX, instance);
3448 
3449 	if (se_num == 0xffffffff)
3450 		data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SE_BROADCAST_WRITES, 1);
3451 	else
3452 		data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SE_INDEX, se_num);
3453 
3454 	if (sh_num == 0xffffffff)
3455 		data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SH_BROADCAST_WRITES, 1);
3456 	else
3457 		data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SH_INDEX, sh_num);
3458 
3459 	WREG32(mmGRBM_GFX_INDEX, data);
3460 }
3461 
3462 static void gfx_v8_0_select_me_pipe_q(struct amdgpu_device *adev,
3463 				  u32 me, u32 pipe, u32 q, u32 vm)
3464 {
3465 	vi_srbm_select(adev, me, pipe, q, vm);
3466 }
3467 
3468 static u32 gfx_v8_0_get_rb_active_bitmap(struct amdgpu_device *adev)
3469 {
3470 	u32 data, mask;
3471 
3472 	data =  RREG32(mmCC_RB_BACKEND_DISABLE) |
3473 		RREG32(mmGC_USER_RB_BACKEND_DISABLE);
3474 
3475 	data = REG_GET_FIELD(data, GC_USER_RB_BACKEND_DISABLE, BACKEND_DISABLE);
3476 
3477 	mask = amdgpu_gfx_create_bitmask(adev->gfx.config.max_backends_per_se /
3478 					 adev->gfx.config.max_sh_per_se);
3479 
3480 	return (~data) & mask;
3481 }
3482 
3483 static void
3484 gfx_v8_0_raster_config(struct amdgpu_device *adev, u32 *rconf, u32 *rconf1)
3485 {
3486 	switch (adev->asic_type) {
3487 	case CHIP_FIJI:
3488 	case CHIP_VEGAM:
3489 		*rconf |= RB_MAP_PKR0(2) | RB_MAP_PKR1(2) |
3490 			  RB_XSEL2(1) | PKR_MAP(2) |
3491 			  PKR_XSEL(1) | PKR_YSEL(1) |
3492 			  SE_MAP(2) | SE_XSEL(2) | SE_YSEL(3);
3493 		*rconf1 |= SE_PAIR_MAP(2) | SE_PAIR_XSEL(3) |
3494 			   SE_PAIR_YSEL(2);
3495 		break;
3496 	case CHIP_TONGA:
3497 	case CHIP_POLARIS10:
3498 		*rconf |= RB_MAP_PKR0(2) | RB_XSEL2(1) | SE_MAP(2) |
3499 			  SE_XSEL(1) | SE_YSEL(1);
3500 		*rconf1 |= SE_PAIR_MAP(2) | SE_PAIR_XSEL(2) |
3501 			   SE_PAIR_YSEL(2);
3502 		break;
3503 	case CHIP_TOPAZ:
3504 	case CHIP_CARRIZO:
3505 		*rconf |= RB_MAP_PKR0(2);
3506 		*rconf1 |= 0x0;
3507 		break;
3508 	case CHIP_POLARIS11:
3509 	case CHIP_POLARIS12:
3510 		*rconf |= RB_MAP_PKR0(2) | RB_XSEL2(1) | SE_MAP(2) |
3511 			  SE_XSEL(1) | SE_YSEL(1);
3512 		*rconf1 |= 0x0;
3513 		break;
3514 	case CHIP_STONEY:
3515 		*rconf |= 0x0;
3516 		*rconf1 |= 0x0;
3517 		break;
3518 	default:
3519 		DRM_ERROR("unknown asic: 0x%x\n", adev->asic_type);
3520 		break;
3521 	}
3522 }
3523 
3524 static void
3525 gfx_v8_0_write_harvested_raster_configs(struct amdgpu_device *adev,
3526 					u32 raster_config, u32 raster_config_1,
3527 					unsigned rb_mask, unsigned num_rb)
3528 {
3529 	unsigned sh_per_se = max_t(unsigned, adev->gfx.config.max_sh_per_se, 1);
3530 	unsigned num_se = max_t(unsigned, adev->gfx.config.max_shader_engines, 1);
3531 	unsigned rb_per_pkr = min_t(unsigned, num_rb / num_se / sh_per_se, 2);
3532 	unsigned rb_per_se = num_rb / num_se;
3533 	unsigned se_mask[4];
3534 	unsigned se;
3535 
3536 	se_mask[0] = ((1 << rb_per_se) - 1) & rb_mask;
3537 	se_mask[1] = (se_mask[0] << rb_per_se) & rb_mask;
3538 	se_mask[2] = (se_mask[1] << rb_per_se) & rb_mask;
3539 	se_mask[3] = (se_mask[2] << rb_per_se) & rb_mask;
3540 
3541 	WARN_ON(!(num_se == 1 || num_se == 2 || num_se == 4));
3542 	WARN_ON(!(sh_per_se == 1 || sh_per_se == 2));
3543 	WARN_ON(!(rb_per_pkr == 1 || rb_per_pkr == 2));
3544 
3545 	if ((num_se > 2) && ((!se_mask[0] && !se_mask[1]) ||
3546 			     (!se_mask[2] && !se_mask[3]))) {
3547 		raster_config_1 &= ~SE_PAIR_MAP_MASK;
3548 
3549 		if (!se_mask[0] && !se_mask[1]) {
3550 			raster_config_1 |=
3551 				SE_PAIR_MAP(RASTER_CONFIG_SE_PAIR_MAP_3);
3552 		} else {
3553 			raster_config_1 |=
3554 				SE_PAIR_MAP(RASTER_CONFIG_SE_PAIR_MAP_0);
3555 		}
3556 	}
3557 
3558 	for (se = 0; se < num_se; se++) {
3559 		unsigned raster_config_se = raster_config;
3560 		unsigned pkr0_mask = ((1 << rb_per_pkr) - 1) << (se * rb_per_se);
3561 		unsigned pkr1_mask = pkr0_mask << rb_per_pkr;
3562 		int idx = (se / 2) * 2;
3563 
3564 		if ((num_se > 1) && (!se_mask[idx] || !se_mask[idx + 1])) {
3565 			raster_config_se &= ~SE_MAP_MASK;
3566 
3567 			if (!se_mask[idx]) {
3568 				raster_config_se |= SE_MAP(RASTER_CONFIG_SE_MAP_3);
3569 			} else {
3570 				raster_config_se |= SE_MAP(RASTER_CONFIG_SE_MAP_0);
3571 			}
3572 		}
3573 
3574 		pkr0_mask &= rb_mask;
3575 		pkr1_mask &= rb_mask;
3576 		if (rb_per_se > 2 && (!pkr0_mask || !pkr1_mask)) {
3577 			raster_config_se &= ~PKR_MAP_MASK;
3578 
3579 			if (!pkr0_mask) {
3580 				raster_config_se |= PKR_MAP(RASTER_CONFIG_PKR_MAP_3);
3581 			} else {
3582 				raster_config_se |= PKR_MAP(RASTER_CONFIG_PKR_MAP_0);
3583 			}
3584 		}
3585 
3586 		if (rb_per_se >= 2) {
3587 			unsigned rb0_mask = 1 << (se * rb_per_se);
3588 			unsigned rb1_mask = rb0_mask << 1;
3589 
3590 			rb0_mask &= rb_mask;
3591 			rb1_mask &= rb_mask;
3592 			if (!rb0_mask || !rb1_mask) {
3593 				raster_config_se &= ~RB_MAP_PKR0_MASK;
3594 
3595 				if (!rb0_mask) {
3596 					raster_config_se |=
3597 						RB_MAP_PKR0(RASTER_CONFIG_RB_MAP_3);
3598 				} else {
3599 					raster_config_se |=
3600 						RB_MAP_PKR0(RASTER_CONFIG_RB_MAP_0);
3601 				}
3602 			}
3603 
3604 			if (rb_per_se > 2) {
3605 				rb0_mask = 1 << (se * rb_per_se + rb_per_pkr);
3606 				rb1_mask = rb0_mask << 1;
3607 				rb0_mask &= rb_mask;
3608 				rb1_mask &= rb_mask;
3609 				if (!rb0_mask || !rb1_mask) {
3610 					raster_config_se &= ~RB_MAP_PKR1_MASK;
3611 
3612 					if (!rb0_mask) {
3613 						raster_config_se |=
3614 							RB_MAP_PKR1(RASTER_CONFIG_RB_MAP_3);
3615 					} else {
3616 						raster_config_se |=
3617 							RB_MAP_PKR1(RASTER_CONFIG_RB_MAP_0);
3618 					}
3619 				}
3620 			}
3621 		}
3622 
3623 		/* GRBM_GFX_INDEX has a different offset on VI */
3624 		gfx_v8_0_select_se_sh(adev, se, 0xffffffff, 0xffffffff);
3625 		WREG32(mmPA_SC_RASTER_CONFIG, raster_config_se);
3626 		WREG32(mmPA_SC_RASTER_CONFIG_1, raster_config_1);
3627 	}
3628 
3629 	/* GRBM_GFX_INDEX has a different offset on VI */
3630 	gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
3631 }
3632 
3633 static void gfx_v8_0_setup_rb(struct amdgpu_device *adev)
3634 {
3635 	int i, j;
3636 	u32 data;
3637 	u32 raster_config = 0, raster_config_1 = 0;
3638 	u32 active_rbs = 0;
3639 	u32 rb_bitmap_width_per_sh = adev->gfx.config.max_backends_per_se /
3640 					adev->gfx.config.max_sh_per_se;
3641 	unsigned num_rb_pipes;
3642 
3643 	mutex_lock(&adev->grbm_idx_mutex);
3644 	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
3645 		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
3646 			gfx_v8_0_select_se_sh(adev, i, j, 0xffffffff);
3647 			data = gfx_v8_0_get_rb_active_bitmap(adev);
3648 			active_rbs |= data << ((i * adev->gfx.config.max_sh_per_se + j) *
3649 					       rb_bitmap_width_per_sh);
3650 		}
3651 	}
3652 	gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
3653 
3654 	adev->gfx.config.backend_enable_mask = active_rbs;
3655 	adev->gfx.config.num_rbs = hweight32(active_rbs);
3656 
3657 	num_rb_pipes = min_t(unsigned, adev->gfx.config.max_backends_per_se *
3658 			     adev->gfx.config.max_shader_engines, 16);
3659 
3660 	gfx_v8_0_raster_config(adev, &raster_config, &raster_config_1);
3661 
3662 	if (!adev->gfx.config.backend_enable_mask ||
3663 			adev->gfx.config.num_rbs >= num_rb_pipes) {
3664 		WREG32(mmPA_SC_RASTER_CONFIG, raster_config);
3665 		WREG32(mmPA_SC_RASTER_CONFIG_1, raster_config_1);
3666 	} else {
3667 		gfx_v8_0_write_harvested_raster_configs(adev, raster_config, raster_config_1,
3668 							adev->gfx.config.backend_enable_mask,
3669 							num_rb_pipes);
3670 	}
3671 
3672 	/* cache the values for userspace */
3673 	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
3674 		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
3675 			gfx_v8_0_select_se_sh(adev, i, j, 0xffffffff);
3676 			adev->gfx.config.rb_config[i][j].rb_backend_disable =
3677 				RREG32(mmCC_RB_BACKEND_DISABLE);
3678 			adev->gfx.config.rb_config[i][j].user_rb_backend_disable =
3679 				RREG32(mmGC_USER_RB_BACKEND_DISABLE);
3680 			adev->gfx.config.rb_config[i][j].raster_config =
3681 				RREG32(mmPA_SC_RASTER_CONFIG);
3682 			adev->gfx.config.rb_config[i][j].raster_config_1 =
3683 				RREG32(mmPA_SC_RASTER_CONFIG_1);
3684 		}
3685 	}
3686 	gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
3687 	mutex_unlock(&adev->grbm_idx_mutex);
3688 }
3689 
3690 #define DEFAULT_SH_MEM_BASES	(0x6000)
3691 /**
3692  * gfx_v8_0_init_compute_vmid - gart enable
3693  *
3694  * @adev: amdgpu_device pointer
3695  *
3696  * Initialize compute vmid sh_mem registers
3697  *
3698  */
3699 static void gfx_v8_0_init_compute_vmid(struct amdgpu_device *adev)
3700 {
3701 	int i;
3702 	uint32_t sh_mem_config;
3703 	uint32_t sh_mem_bases;
3704 
3705 	/*
3706 	 * Configure apertures:
3707 	 * LDS:         0x60000000'00000000 - 0x60000001'00000000 (4GB)
3708 	 * Scratch:     0x60000001'00000000 - 0x60000002'00000000 (4GB)
3709 	 * GPUVM:       0x60010000'00000000 - 0x60020000'00000000 (1TB)
3710 	 */
3711 	sh_mem_bases = DEFAULT_SH_MEM_BASES | (DEFAULT_SH_MEM_BASES << 16);
3712 
3713 	sh_mem_config = SH_MEM_ADDRESS_MODE_HSA64 <<
3714 			SH_MEM_CONFIG__ADDRESS_MODE__SHIFT |
3715 			SH_MEM_ALIGNMENT_MODE_UNALIGNED <<
3716 			SH_MEM_CONFIG__ALIGNMENT_MODE__SHIFT |
3717 			MTYPE_CC << SH_MEM_CONFIG__DEFAULT_MTYPE__SHIFT |
3718 			SH_MEM_CONFIG__PRIVATE_ATC_MASK;
3719 
3720 	mutex_lock(&adev->srbm_mutex);
3721 	for (i = adev->vm_manager.first_kfd_vmid; i < AMDGPU_NUM_VMID; i++) {
3722 		vi_srbm_select(adev, 0, 0, 0, i);
3723 		/* CP and shaders */
3724 		WREG32(mmSH_MEM_CONFIG, sh_mem_config);
3725 		WREG32(mmSH_MEM_APE1_BASE, 1);
3726 		WREG32(mmSH_MEM_APE1_LIMIT, 0);
3727 		WREG32(mmSH_MEM_BASES, sh_mem_bases);
3728 	}
3729 	vi_srbm_select(adev, 0, 0, 0, 0);
3730 	mutex_unlock(&adev->srbm_mutex);
3731 
3732 	/* Initialize all compute VMIDs to have no GDS, GWS, or OA
3733 	   acccess. These should be enabled by FW for target VMIDs. */
3734 	for (i = adev->vm_manager.first_kfd_vmid; i < AMDGPU_NUM_VMID; i++) {
3735 		WREG32(amdgpu_gds_reg_offset[i].mem_base, 0);
3736 		WREG32(amdgpu_gds_reg_offset[i].mem_size, 0);
3737 		WREG32(amdgpu_gds_reg_offset[i].gws, 0);
3738 		WREG32(amdgpu_gds_reg_offset[i].oa, 0);
3739 	}
3740 }
3741 
3742 static void gfx_v8_0_init_gds_vmid(struct amdgpu_device *adev)
3743 {
3744 	int vmid;
3745 
3746 	/*
3747 	 * Initialize all compute and user-gfx VMIDs to have no GDS, GWS, or OA
3748 	 * access. Compute VMIDs should be enabled by FW for target VMIDs,
3749 	 * the driver can enable them for graphics. VMID0 should maintain
3750 	 * access so that HWS firmware can save/restore entries.
3751 	 */
3752 	for (vmid = 1; vmid < AMDGPU_NUM_VMID; vmid++) {
3753 		WREG32(amdgpu_gds_reg_offset[vmid].mem_base, 0);
3754 		WREG32(amdgpu_gds_reg_offset[vmid].mem_size, 0);
3755 		WREG32(amdgpu_gds_reg_offset[vmid].gws, 0);
3756 		WREG32(amdgpu_gds_reg_offset[vmid].oa, 0);
3757 	}
3758 }
3759 
3760 static void gfx_v8_0_config_init(struct amdgpu_device *adev)
3761 {
3762 	switch (adev->asic_type) {
3763 	default:
3764 		adev->gfx.config.double_offchip_lds_buf = 1;
3765 		break;
3766 	case CHIP_CARRIZO:
3767 	case CHIP_STONEY:
3768 		adev->gfx.config.double_offchip_lds_buf = 0;
3769 		break;
3770 	}
3771 }
3772 
3773 static void gfx_v8_0_constants_init(struct amdgpu_device *adev)
3774 {
3775 	u32 tmp, sh_static_mem_cfg;
3776 	int i;
3777 
3778 	WREG32_FIELD(GRBM_CNTL, READ_TIMEOUT, 0xFF);
3779 	WREG32(mmGB_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
3780 	WREG32(mmHDP_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
3781 	WREG32(mmDMIF_ADDR_CALC, adev->gfx.config.gb_addr_config);
3782 
3783 	gfx_v8_0_tiling_mode_table_init(adev);
3784 	gfx_v8_0_setup_rb(adev);
3785 	gfx_v8_0_get_cu_info(adev);
3786 	gfx_v8_0_config_init(adev);
3787 
3788 	/* XXX SH_MEM regs */
3789 	/* where to put LDS, scratch, GPUVM in FSA64 space */
3790 	sh_static_mem_cfg = REG_SET_FIELD(0, SH_STATIC_MEM_CONFIG,
3791 				   SWIZZLE_ENABLE, 1);
3792 	sh_static_mem_cfg = REG_SET_FIELD(sh_static_mem_cfg, SH_STATIC_MEM_CONFIG,
3793 				   ELEMENT_SIZE, 1);
3794 	sh_static_mem_cfg = REG_SET_FIELD(sh_static_mem_cfg, SH_STATIC_MEM_CONFIG,
3795 				   INDEX_STRIDE, 3);
3796 	WREG32(mmSH_STATIC_MEM_CONFIG, sh_static_mem_cfg);
3797 
3798 	mutex_lock(&adev->srbm_mutex);
3799 	for (i = 0; i < adev->vm_manager.id_mgr[0].num_ids; i++) {
3800 		vi_srbm_select(adev, 0, 0, 0, i);
3801 		/* CP and shaders */
3802 		if (i == 0) {
3803 			tmp = REG_SET_FIELD(0, SH_MEM_CONFIG, DEFAULT_MTYPE, MTYPE_UC);
3804 			tmp = REG_SET_FIELD(tmp, SH_MEM_CONFIG, APE1_MTYPE, MTYPE_UC);
3805 			tmp = REG_SET_FIELD(tmp, SH_MEM_CONFIG, ALIGNMENT_MODE,
3806 					    SH_MEM_ALIGNMENT_MODE_UNALIGNED);
3807 			WREG32(mmSH_MEM_CONFIG, tmp);
3808 			WREG32(mmSH_MEM_BASES, 0);
3809 		} else {
3810 			tmp = REG_SET_FIELD(0, SH_MEM_CONFIG, DEFAULT_MTYPE, MTYPE_NC);
3811 			tmp = REG_SET_FIELD(tmp, SH_MEM_CONFIG, APE1_MTYPE, MTYPE_UC);
3812 			tmp = REG_SET_FIELD(tmp, SH_MEM_CONFIG, ALIGNMENT_MODE,
3813 					    SH_MEM_ALIGNMENT_MODE_UNALIGNED);
3814 			WREG32(mmSH_MEM_CONFIG, tmp);
3815 			tmp = adev->gmc.shared_aperture_start >> 48;
3816 			WREG32(mmSH_MEM_BASES, tmp);
3817 		}
3818 
3819 		WREG32(mmSH_MEM_APE1_BASE, 1);
3820 		WREG32(mmSH_MEM_APE1_LIMIT, 0);
3821 	}
3822 	vi_srbm_select(adev, 0, 0, 0, 0);
3823 	mutex_unlock(&adev->srbm_mutex);
3824 
3825 	gfx_v8_0_init_compute_vmid(adev);
3826 	gfx_v8_0_init_gds_vmid(adev);
3827 
3828 	mutex_lock(&adev->grbm_idx_mutex);
3829 	/*
3830 	 * making sure that the following register writes will be broadcasted
3831 	 * to all the shaders
3832 	 */
3833 	gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
3834 
3835 	WREG32(mmPA_SC_FIFO_SIZE,
3836 		   (adev->gfx.config.sc_prim_fifo_size_frontend <<
3837 			PA_SC_FIFO_SIZE__SC_FRONTEND_PRIM_FIFO_SIZE__SHIFT) |
3838 		   (adev->gfx.config.sc_prim_fifo_size_backend <<
3839 			PA_SC_FIFO_SIZE__SC_BACKEND_PRIM_FIFO_SIZE__SHIFT) |
3840 		   (adev->gfx.config.sc_hiz_tile_fifo_size <<
3841 			PA_SC_FIFO_SIZE__SC_HIZ_TILE_FIFO_SIZE__SHIFT) |
3842 		   (adev->gfx.config.sc_earlyz_tile_fifo_size <<
3843 			PA_SC_FIFO_SIZE__SC_EARLYZ_TILE_FIFO_SIZE__SHIFT));
3844 
3845 	tmp = RREG32(mmSPI_ARB_PRIORITY);
3846 	tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS0, 2);
3847 	tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS1, 2);
3848 	tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS2, 2);
3849 	tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS3, 2);
3850 	WREG32(mmSPI_ARB_PRIORITY, tmp);
3851 
3852 	mutex_unlock(&adev->grbm_idx_mutex);
3853 
3854 }
3855 
3856 static void gfx_v8_0_wait_for_rlc_serdes(struct amdgpu_device *adev)
3857 {
3858 	u32 i, j, k;
3859 	u32 mask;
3860 
3861 	mutex_lock(&adev->grbm_idx_mutex);
3862 	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
3863 		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
3864 			gfx_v8_0_select_se_sh(adev, i, j, 0xffffffff);
3865 			for (k = 0; k < adev->usec_timeout; k++) {
3866 				if (RREG32(mmRLC_SERDES_CU_MASTER_BUSY) == 0)
3867 					break;
3868 				udelay(1);
3869 			}
3870 			if (k == adev->usec_timeout) {
3871 				gfx_v8_0_select_se_sh(adev, 0xffffffff,
3872 						      0xffffffff, 0xffffffff);
3873 				mutex_unlock(&adev->grbm_idx_mutex);
3874 				DRM_INFO("Timeout wait for RLC serdes %u,%u\n",
3875 					 i, j);
3876 				return;
3877 			}
3878 		}
3879 	}
3880 	gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
3881 	mutex_unlock(&adev->grbm_idx_mutex);
3882 
3883 	mask = RLC_SERDES_NONCU_MASTER_BUSY__SE_MASTER_BUSY_MASK |
3884 		RLC_SERDES_NONCU_MASTER_BUSY__GC_MASTER_BUSY_MASK |
3885 		RLC_SERDES_NONCU_MASTER_BUSY__TC0_MASTER_BUSY_MASK |
3886 		RLC_SERDES_NONCU_MASTER_BUSY__TC1_MASTER_BUSY_MASK;
3887 	for (k = 0; k < adev->usec_timeout; k++) {
3888 		if ((RREG32(mmRLC_SERDES_NONCU_MASTER_BUSY) & mask) == 0)
3889 			break;
3890 		udelay(1);
3891 	}
3892 }
3893 
3894 static void gfx_v8_0_enable_gui_idle_interrupt(struct amdgpu_device *adev,
3895 					       bool enable)
3896 {
3897 	u32 tmp = RREG32(mmCP_INT_CNTL_RING0);
3898 
3899 	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CNTX_BUSY_INT_ENABLE, enable ? 1 : 0);
3900 	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CNTX_EMPTY_INT_ENABLE, enable ? 1 : 0);
3901 	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CMP_BUSY_INT_ENABLE, enable ? 1 : 0);
3902 	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, GFX_IDLE_INT_ENABLE, enable ? 1 : 0);
3903 
3904 	WREG32(mmCP_INT_CNTL_RING0, tmp);
3905 }
3906 
3907 static void gfx_v8_0_init_csb(struct amdgpu_device *adev)
3908 {
3909 	adev->gfx.rlc.funcs->get_csb_buffer(adev, adev->gfx.rlc.cs_ptr);
3910 	/* csib */
3911 	WREG32(mmRLC_CSIB_ADDR_HI,
3912 			adev->gfx.rlc.clear_state_gpu_addr >> 32);
3913 	WREG32(mmRLC_CSIB_ADDR_LO,
3914 			adev->gfx.rlc.clear_state_gpu_addr & 0xfffffffc);
3915 	WREG32(mmRLC_CSIB_LENGTH,
3916 			adev->gfx.rlc.clear_state_size);
3917 }
3918 
3919 static void gfx_v8_0_parse_ind_reg_list(int *register_list_format,
3920 				int ind_offset,
3921 				int list_size,
3922 				int *unique_indices,
3923 				int *indices_count,
3924 				int max_indices,
3925 				int *ind_start_offsets,
3926 				int *offset_count,
3927 				int max_offset)
3928 {
3929 	int indices;
3930 	bool new_entry = true;
3931 
3932 	for (; ind_offset < list_size; ind_offset++) {
3933 
3934 		if (new_entry) {
3935 			new_entry = false;
3936 			ind_start_offsets[*offset_count] = ind_offset;
3937 			*offset_count = *offset_count + 1;
3938 			BUG_ON(*offset_count >= max_offset);
3939 		}
3940 
3941 		if (register_list_format[ind_offset] == 0xFFFFFFFF) {
3942 			new_entry = true;
3943 			continue;
3944 		}
3945 
3946 		ind_offset += 2;
3947 
3948 		/* look for the matching indice */
3949 		for (indices = 0;
3950 			indices < *indices_count;
3951 			indices++) {
3952 			if (unique_indices[indices] ==
3953 				register_list_format[ind_offset])
3954 				break;
3955 		}
3956 
3957 		if (indices >= *indices_count) {
3958 			unique_indices[*indices_count] =
3959 				register_list_format[ind_offset];
3960 			indices = *indices_count;
3961 			*indices_count = *indices_count + 1;
3962 			BUG_ON(*indices_count >= max_indices);
3963 		}
3964 
3965 		register_list_format[ind_offset] = indices;
3966 	}
3967 }
3968 
3969 static int gfx_v8_0_init_save_restore_list(struct amdgpu_device *adev)
3970 {
3971 	int i, temp, data;
3972 	int unique_indices[] = {0, 0, 0, 0, 0, 0, 0, 0};
3973 	int indices_count = 0;
3974 	int indirect_start_offsets[] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0};
3975 	int offset_count = 0;
3976 
3977 	int list_size;
3978 	unsigned int *register_list_format =
3979 		kmemdup(adev->gfx.rlc.register_list_format,
3980 			adev->gfx.rlc.reg_list_format_size_bytes, GFP_KERNEL);
3981 	if (!register_list_format)
3982 		return -ENOMEM;
3983 
3984 	gfx_v8_0_parse_ind_reg_list(register_list_format,
3985 				RLC_FormatDirectRegListLength,
3986 				adev->gfx.rlc.reg_list_format_size_bytes >> 2,
3987 				unique_indices,
3988 				&indices_count,
3989 				ARRAY_SIZE(unique_indices),
3990 				indirect_start_offsets,
3991 				&offset_count,
3992 				ARRAY_SIZE(indirect_start_offsets));
3993 
3994 	/* save and restore list */
3995 	WREG32_FIELD(RLC_SRM_CNTL, AUTO_INCR_ADDR, 1);
3996 
3997 	WREG32(mmRLC_SRM_ARAM_ADDR, 0);
3998 	for (i = 0; i < adev->gfx.rlc.reg_list_size_bytes >> 2; i++)
3999 		WREG32(mmRLC_SRM_ARAM_DATA, adev->gfx.rlc.register_restore[i]);
4000 
4001 	/* indirect list */
4002 	WREG32(mmRLC_GPM_SCRATCH_ADDR, adev->gfx.rlc.reg_list_format_start);
4003 	for (i = 0; i < adev->gfx.rlc.reg_list_format_size_bytes >> 2; i++)
4004 		WREG32(mmRLC_GPM_SCRATCH_DATA, register_list_format[i]);
4005 
4006 	list_size = adev->gfx.rlc.reg_list_size_bytes >> 2;
4007 	list_size = list_size >> 1;
4008 	WREG32(mmRLC_GPM_SCRATCH_ADDR, adev->gfx.rlc.reg_restore_list_size);
4009 	WREG32(mmRLC_GPM_SCRATCH_DATA, list_size);
4010 
4011 	/* starting offsets starts */
4012 	WREG32(mmRLC_GPM_SCRATCH_ADDR,
4013 		adev->gfx.rlc.starting_offsets_start);
4014 	for (i = 0; i < ARRAY_SIZE(indirect_start_offsets); i++)
4015 		WREG32(mmRLC_GPM_SCRATCH_DATA,
4016 				indirect_start_offsets[i]);
4017 
4018 	/* unique indices */
4019 	temp = mmRLC_SRM_INDEX_CNTL_ADDR_0;
4020 	data = mmRLC_SRM_INDEX_CNTL_DATA_0;
4021 	for (i = 0; i < ARRAY_SIZE(unique_indices); i++) {
4022 		if (unique_indices[i] != 0) {
4023 			WREG32(temp + i, unique_indices[i] & 0x3FFFF);
4024 			WREG32(data + i, unique_indices[i] >> 20);
4025 		}
4026 	}
4027 	kfree(register_list_format);
4028 
4029 	return 0;
4030 }
4031 
4032 static void gfx_v8_0_enable_save_restore_machine(struct amdgpu_device *adev)
4033 {
4034 	WREG32_FIELD(RLC_SRM_CNTL, SRM_ENABLE, 1);
4035 }
4036 
4037 static void gfx_v8_0_init_power_gating(struct amdgpu_device *adev)
4038 {
4039 	uint32_t data;
4040 
4041 	WREG32_FIELD(CP_RB_WPTR_POLL_CNTL, IDLE_POLL_COUNT, 0x60);
4042 
4043 	data = REG_SET_FIELD(0, RLC_PG_DELAY, POWER_UP_DELAY, 0x10);
4044 	data = REG_SET_FIELD(data, RLC_PG_DELAY, POWER_DOWN_DELAY, 0x10);
4045 	data = REG_SET_FIELD(data, RLC_PG_DELAY, CMD_PROPAGATE_DELAY, 0x10);
4046 	data = REG_SET_FIELD(data, RLC_PG_DELAY, MEM_SLEEP_DELAY, 0x10);
4047 	WREG32(mmRLC_PG_DELAY, data);
4048 
4049 	WREG32_FIELD(RLC_PG_DELAY_2, SERDES_CMD_DELAY, 0x3);
4050 	WREG32_FIELD(RLC_AUTO_PG_CTRL, GRBM_REG_SAVE_GFX_IDLE_THRESHOLD, 0x55f0);
4051 
4052 }
4053 
4054 static void cz_enable_sck_slow_down_on_power_up(struct amdgpu_device *adev,
4055 						bool enable)
4056 {
4057 	WREG32_FIELD(RLC_PG_CNTL, SMU_CLK_SLOWDOWN_ON_PU_ENABLE, enable ? 1 : 0);
4058 }
4059 
4060 static void cz_enable_sck_slow_down_on_power_down(struct amdgpu_device *adev,
4061 						  bool enable)
4062 {
4063 	WREG32_FIELD(RLC_PG_CNTL, SMU_CLK_SLOWDOWN_ON_PD_ENABLE, enable ? 1 : 0);
4064 }
4065 
4066 static void cz_enable_cp_power_gating(struct amdgpu_device *adev, bool enable)
4067 {
4068 	WREG32_FIELD(RLC_PG_CNTL, CP_PG_DISABLE, enable ? 0 : 1);
4069 }
4070 
4071 static void gfx_v8_0_init_pg(struct amdgpu_device *adev)
4072 {
4073 	if ((adev->asic_type == CHIP_CARRIZO) ||
4074 	    (adev->asic_type == CHIP_STONEY)) {
4075 		gfx_v8_0_init_csb(adev);
4076 		gfx_v8_0_init_save_restore_list(adev);
4077 		gfx_v8_0_enable_save_restore_machine(adev);
4078 		WREG32(mmRLC_JUMP_TABLE_RESTORE, adev->gfx.rlc.cp_table_gpu_addr >> 8);
4079 		gfx_v8_0_init_power_gating(adev);
4080 		WREG32(mmRLC_PG_ALWAYS_ON_CU_MASK, adev->gfx.cu_info.ao_cu_mask);
4081 	} else if ((adev->asic_type == CHIP_POLARIS11) ||
4082 		   (adev->asic_type == CHIP_POLARIS12) ||
4083 		   (adev->asic_type == CHIP_VEGAM)) {
4084 		gfx_v8_0_init_csb(adev);
4085 		gfx_v8_0_init_save_restore_list(adev);
4086 		gfx_v8_0_enable_save_restore_machine(adev);
4087 		gfx_v8_0_init_power_gating(adev);
4088 	}
4089 
4090 }
4091 
4092 static void gfx_v8_0_rlc_stop(struct amdgpu_device *adev)
4093 {
4094 	WREG32_FIELD(RLC_CNTL, RLC_ENABLE_F32, 0);
4095 
4096 	gfx_v8_0_enable_gui_idle_interrupt(adev, false);
4097 	gfx_v8_0_wait_for_rlc_serdes(adev);
4098 }
4099 
4100 static void gfx_v8_0_rlc_reset(struct amdgpu_device *adev)
4101 {
4102 	WREG32_FIELD(GRBM_SOFT_RESET, SOFT_RESET_RLC, 1);
4103 	udelay(50);
4104 
4105 	WREG32_FIELD(GRBM_SOFT_RESET, SOFT_RESET_RLC, 0);
4106 	udelay(50);
4107 }
4108 
4109 static void gfx_v8_0_rlc_start(struct amdgpu_device *adev)
4110 {
4111 	WREG32_FIELD(RLC_CNTL, RLC_ENABLE_F32, 1);
4112 
4113 	/* carrizo do enable cp interrupt after cp inited */
4114 	if (!(adev->flags & AMD_IS_APU))
4115 		gfx_v8_0_enable_gui_idle_interrupt(adev, true);
4116 
4117 	udelay(50);
4118 }
4119 
4120 static int gfx_v8_0_rlc_resume(struct amdgpu_device *adev)
4121 {
4122 	if (amdgpu_sriov_vf(adev)) {
4123 		gfx_v8_0_init_csb(adev);
4124 		return 0;
4125 	}
4126 
4127 	adev->gfx.rlc.funcs->stop(adev);
4128 	adev->gfx.rlc.funcs->reset(adev);
4129 	gfx_v8_0_init_pg(adev);
4130 	adev->gfx.rlc.funcs->start(adev);
4131 
4132 	return 0;
4133 }
4134 
4135 static void gfx_v8_0_cp_gfx_enable(struct amdgpu_device *adev, bool enable)
4136 {
4137 	u32 tmp = RREG32(mmCP_ME_CNTL);
4138 
4139 	if (enable) {
4140 		tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, ME_HALT, 0);
4141 		tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, PFP_HALT, 0);
4142 		tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, CE_HALT, 0);
4143 	} else {
4144 		tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, ME_HALT, 1);
4145 		tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, PFP_HALT, 1);
4146 		tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, CE_HALT, 1);
4147 	}
4148 	WREG32(mmCP_ME_CNTL, tmp);
4149 	udelay(50);
4150 }
4151 
4152 static u32 gfx_v8_0_get_csb_size(struct amdgpu_device *adev)
4153 {
4154 	u32 count = 0;
4155 	const struct cs_section_def *sect = NULL;
4156 	const struct cs_extent_def *ext = NULL;
4157 
4158 	/* begin clear state */
4159 	count += 2;
4160 	/* context control state */
4161 	count += 3;
4162 
4163 	for (sect = vi_cs_data; sect->section != NULL; ++sect) {
4164 		for (ext = sect->section; ext->extent != NULL; ++ext) {
4165 			if (sect->id == SECT_CONTEXT)
4166 				count += 2 + ext->reg_count;
4167 			else
4168 				return 0;
4169 		}
4170 	}
4171 	/* pa_sc_raster_config/pa_sc_raster_config1 */
4172 	count += 4;
4173 	/* end clear state */
4174 	count += 2;
4175 	/* clear state */
4176 	count += 2;
4177 
4178 	return count;
4179 }
4180 
4181 static int gfx_v8_0_cp_gfx_start(struct amdgpu_device *adev)
4182 {
4183 	struct amdgpu_ring *ring = &adev->gfx.gfx_ring[0];
4184 	const struct cs_section_def *sect = NULL;
4185 	const struct cs_extent_def *ext = NULL;
4186 	int r, i;
4187 
4188 	/* init the CP */
4189 	WREG32(mmCP_MAX_CONTEXT, adev->gfx.config.max_hw_contexts - 1);
4190 	WREG32(mmCP_ENDIAN_SWAP, 0);
4191 	WREG32(mmCP_DEVICE_ID, 1);
4192 
4193 	gfx_v8_0_cp_gfx_enable(adev, true);
4194 
4195 	r = amdgpu_ring_alloc(ring, gfx_v8_0_get_csb_size(adev) + 4);
4196 	if (r) {
4197 		DRM_ERROR("amdgpu: cp failed to lock ring (%d).\n", r);
4198 		return r;
4199 	}
4200 
4201 	/* clear state buffer */
4202 	amdgpu_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
4203 	amdgpu_ring_write(ring, PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
4204 
4205 	amdgpu_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1));
4206 	amdgpu_ring_write(ring, 0x80000000);
4207 	amdgpu_ring_write(ring, 0x80000000);
4208 
4209 	for (sect = vi_cs_data; sect->section != NULL; ++sect) {
4210 		for (ext = sect->section; ext->extent != NULL; ++ext) {
4211 			if (sect->id == SECT_CONTEXT) {
4212 				amdgpu_ring_write(ring,
4213 				       PACKET3(PACKET3_SET_CONTEXT_REG,
4214 					       ext->reg_count));
4215 				amdgpu_ring_write(ring,
4216 				       ext->reg_index - PACKET3_SET_CONTEXT_REG_START);
4217 				for (i = 0; i < ext->reg_count; i++)
4218 					amdgpu_ring_write(ring, ext->extent[i]);
4219 			}
4220 		}
4221 	}
4222 
4223 	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONTEXT_REG, 2));
4224 	amdgpu_ring_write(ring, mmPA_SC_RASTER_CONFIG - PACKET3_SET_CONTEXT_REG_START);
4225 	amdgpu_ring_write(ring, adev->gfx.config.rb_config[0][0].raster_config);
4226 	amdgpu_ring_write(ring, adev->gfx.config.rb_config[0][0].raster_config_1);
4227 
4228 	amdgpu_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
4229 	amdgpu_ring_write(ring, PACKET3_PREAMBLE_END_CLEAR_STATE);
4230 
4231 	amdgpu_ring_write(ring, PACKET3(PACKET3_CLEAR_STATE, 0));
4232 	amdgpu_ring_write(ring, 0);
4233 
4234 	/* init the CE partitions */
4235 	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_BASE, 2));
4236 	amdgpu_ring_write(ring, PACKET3_BASE_INDEX(CE_PARTITION_BASE));
4237 	amdgpu_ring_write(ring, 0x8000);
4238 	amdgpu_ring_write(ring, 0x8000);
4239 
4240 	amdgpu_ring_commit(ring);
4241 
4242 	return 0;
4243 }
4244 static void gfx_v8_0_set_cpg_door_bell(struct amdgpu_device *adev, struct amdgpu_ring *ring)
4245 {
4246 	u32 tmp;
4247 	/* no gfx doorbells on iceland */
4248 	if (adev->asic_type == CHIP_TOPAZ)
4249 		return;
4250 
4251 	tmp = RREG32(mmCP_RB_DOORBELL_CONTROL);
4252 
4253 	if (ring->use_doorbell) {
4254 		tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL,
4255 				DOORBELL_OFFSET, ring->doorbell_index);
4256 		tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL,
4257 						DOORBELL_HIT, 0);
4258 		tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL,
4259 					    DOORBELL_EN, 1);
4260 	} else {
4261 		tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL, DOORBELL_EN, 0);
4262 	}
4263 
4264 	WREG32(mmCP_RB_DOORBELL_CONTROL, tmp);
4265 
4266 	if (adev->flags & AMD_IS_APU)
4267 		return;
4268 
4269 	tmp = REG_SET_FIELD(0, CP_RB_DOORBELL_RANGE_LOWER,
4270 					DOORBELL_RANGE_LOWER,
4271 					adev->doorbell_index.gfx_ring0);
4272 	WREG32(mmCP_RB_DOORBELL_RANGE_LOWER, tmp);
4273 
4274 	WREG32(mmCP_RB_DOORBELL_RANGE_UPPER,
4275 		CP_RB_DOORBELL_RANGE_UPPER__DOORBELL_RANGE_UPPER_MASK);
4276 }
4277 
4278 static int gfx_v8_0_cp_gfx_resume(struct amdgpu_device *adev)
4279 {
4280 	struct amdgpu_ring *ring;
4281 	u32 tmp;
4282 	u32 rb_bufsz;
4283 	u64 rb_addr, rptr_addr, wptr_gpu_addr;
4284 
4285 	/* Set the write pointer delay */
4286 	WREG32(mmCP_RB_WPTR_DELAY, 0);
4287 
4288 	/* set the RB to use vmid 0 */
4289 	WREG32(mmCP_RB_VMID, 0);
4290 
4291 	/* Set ring buffer size */
4292 	ring = &adev->gfx.gfx_ring[0];
4293 	rb_bufsz = order_base_2(ring->ring_size / 8);
4294 	tmp = REG_SET_FIELD(0, CP_RB0_CNTL, RB_BUFSZ, rb_bufsz);
4295 	tmp = REG_SET_FIELD(tmp, CP_RB0_CNTL, RB_BLKSZ, rb_bufsz - 2);
4296 	tmp = REG_SET_FIELD(tmp, CP_RB0_CNTL, MTYPE, 3);
4297 	tmp = REG_SET_FIELD(tmp, CP_RB0_CNTL, MIN_IB_AVAILSZ, 1);
4298 #ifdef __BIG_ENDIAN
4299 	tmp = REG_SET_FIELD(tmp, CP_RB0_CNTL, BUF_SWAP, 1);
4300 #endif
4301 	WREG32(mmCP_RB0_CNTL, tmp);
4302 
4303 	/* Initialize the ring buffer's read and write pointers */
4304 	WREG32(mmCP_RB0_CNTL, tmp | CP_RB0_CNTL__RB_RPTR_WR_ENA_MASK);
4305 	ring->wptr = 0;
4306 	WREG32(mmCP_RB0_WPTR, lower_32_bits(ring->wptr));
4307 
4308 	/* set the wb address wether it's enabled or not */
4309 	rptr_addr = adev->wb.gpu_addr + (ring->rptr_offs * 4);
4310 	WREG32(mmCP_RB0_RPTR_ADDR, lower_32_bits(rptr_addr));
4311 	WREG32(mmCP_RB0_RPTR_ADDR_HI, upper_32_bits(rptr_addr) & 0xFF);
4312 
4313 	wptr_gpu_addr = adev->wb.gpu_addr + (ring->wptr_offs * 4);
4314 	WREG32(mmCP_RB_WPTR_POLL_ADDR_LO, lower_32_bits(wptr_gpu_addr));
4315 	WREG32(mmCP_RB_WPTR_POLL_ADDR_HI, upper_32_bits(wptr_gpu_addr));
4316 	mdelay(1);
4317 	WREG32(mmCP_RB0_CNTL, tmp);
4318 
4319 	rb_addr = ring->gpu_addr >> 8;
4320 	WREG32(mmCP_RB0_BASE, rb_addr);
4321 	WREG32(mmCP_RB0_BASE_HI, upper_32_bits(rb_addr));
4322 
4323 	gfx_v8_0_set_cpg_door_bell(adev, ring);
4324 	/* start the ring */
4325 	amdgpu_ring_clear_ring(ring);
4326 	gfx_v8_0_cp_gfx_start(adev);
4327 	ring->sched.ready = true;
4328 
4329 	return 0;
4330 }
4331 
4332 static void gfx_v8_0_cp_compute_enable(struct amdgpu_device *adev, bool enable)
4333 {
4334 	if (enable) {
4335 		WREG32(mmCP_MEC_CNTL, 0);
4336 	} else {
4337 		WREG32(mmCP_MEC_CNTL, (CP_MEC_CNTL__MEC_ME1_HALT_MASK | CP_MEC_CNTL__MEC_ME2_HALT_MASK));
4338 		adev->gfx.kiq.ring.sched.ready = false;
4339 	}
4340 	udelay(50);
4341 }
4342 
4343 /* KIQ functions */
4344 static void gfx_v8_0_kiq_setting(struct amdgpu_ring *ring)
4345 {
4346 	uint32_t tmp;
4347 	struct amdgpu_device *adev = ring->adev;
4348 
4349 	/* tell RLC which is KIQ queue */
4350 	tmp = RREG32(mmRLC_CP_SCHEDULERS);
4351 	tmp &= 0xffffff00;
4352 	tmp |= (ring->me << 5) | (ring->pipe << 3) | (ring->queue);
4353 	WREG32(mmRLC_CP_SCHEDULERS, tmp);
4354 	tmp |= 0x80;
4355 	WREG32(mmRLC_CP_SCHEDULERS, tmp);
4356 }
4357 
4358 static int gfx_v8_0_kiq_kcq_enable(struct amdgpu_device *adev)
4359 {
4360 	struct amdgpu_ring *kiq_ring = &adev->gfx.kiq.ring;
4361 	uint64_t queue_mask = 0;
4362 	int r, i;
4363 
4364 	for (i = 0; i < AMDGPU_MAX_COMPUTE_QUEUES; ++i) {
4365 		if (!test_bit(i, adev->gfx.mec.queue_bitmap))
4366 			continue;
4367 
4368 		/* This situation may be hit in the future if a new HW
4369 		 * generation exposes more than 64 queues. If so, the
4370 		 * definition of queue_mask needs updating */
4371 		if (WARN_ON(i >= (sizeof(queue_mask)*8))) {
4372 			DRM_ERROR("Invalid KCQ enabled: %d\n", i);
4373 			break;
4374 		}
4375 
4376 		queue_mask |= (1ull << i);
4377 	}
4378 
4379 	r = amdgpu_ring_alloc(kiq_ring, (8 * adev->gfx.num_compute_rings) + 8);
4380 	if (r) {
4381 		DRM_ERROR("Failed to lock KIQ (%d).\n", r);
4382 		return r;
4383 	}
4384 	/* set resources */
4385 	amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_SET_RESOURCES, 6));
4386 	amdgpu_ring_write(kiq_ring, 0);	/* vmid_mask:0 queue_type:0 (KIQ) */
4387 	amdgpu_ring_write(kiq_ring, lower_32_bits(queue_mask));	/* queue mask lo */
4388 	amdgpu_ring_write(kiq_ring, upper_32_bits(queue_mask));	/* queue mask hi */
4389 	amdgpu_ring_write(kiq_ring, 0);	/* gws mask lo */
4390 	amdgpu_ring_write(kiq_ring, 0);	/* gws mask hi */
4391 	amdgpu_ring_write(kiq_ring, 0);	/* oac mask */
4392 	amdgpu_ring_write(kiq_ring, 0);	/* gds heap base:0, gds heap size:0 */
4393 	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
4394 		struct amdgpu_ring *ring = &adev->gfx.compute_ring[i];
4395 		uint64_t mqd_addr = amdgpu_bo_gpu_offset(ring->mqd_obj);
4396 		uint64_t wptr_addr = adev->wb.gpu_addr + (ring->wptr_offs * 4);
4397 
4398 		/* map queues */
4399 		amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_MAP_QUEUES, 5));
4400 		/* Q_sel:0, vmid:0, vidmem: 1, engine:0, num_Q:1*/
4401 		amdgpu_ring_write(kiq_ring,
4402 				  PACKET3_MAP_QUEUES_NUM_QUEUES(1));
4403 		amdgpu_ring_write(kiq_ring,
4404 				  PACKET3_MAP_QUEUES_DOORBELL_OFFSET(ring->doorbell_index) |
4405 				  PACKET3_MAP_QUEUES_QUEUE(ring->queue) |
4406 				  PACKET3_MAP_QUEUES_PIPE(ring->pipe) |
4407 				  PACKET3_MAP_QUEUES_ME(ring->me == 1 ? 0 : 1)); /* doorbell */
4408 		amdgpu_ring_write(kiq_ring, lower_32_bits(mqd_addr));
4409 		amdgpu_ring_write(kiq_ring, upper_32_bits(mqd_addr));
4410 		amdgpu_ring_write(kiq_ring, lower_32_bits(wptr_addr));
4411 		amdgpu_ring_write(kiq_ring, upper_32_bits(wptr_addr));
4412 	}
4413 
4414 	amdgpu_ring_commit(kiq_ring);
4415 
4416 	return 0;
4417 }
4418 
4419 static int gfx_v8_0_deactivate_hqd(struct amdgpu_device *adev, u32 req)
4420 {
4421 	int i, r = 0;
4422 
4423 	if (RREG32(mmCP_HQD_ACTIVE) & CP_HQD_ACTIVE__ACTIVE_MASK) {
4424 		WREG32_FIELD(CP_HQD_DEQUEUE_REQUEST, DEQUEUE_REQ, req);
4425 		for (i = 0; i < adev->usec_timeout; i++) {
4426 			if (!(RREG32(mmCP_HQD_ACTIVE) & CP_HQD_ACTIVE__ACTIVE_MASK))
4427 				break;
4428 			udelay(1);
4429 		}
4430 		if (i == adev->usec_timeout)
4431 			r = -ETIMEDOUT;
4432 	}
4433 	WREG32(mmCP_HQD_DEQUEUE_REQUEST, 0);
4434 	WREG32(mmCP_HQD_PQ_RPTR, 0);
4435 	WREG32(mmCP_HQD_PQ_WPTR, 0);
4436 
4437 	return r;
4438 }
4439 
4440 static void gfx_v8_0_mqd_set_priority(struct amdgpu_ring *ring, struct vi_mqd *mqd)
4441 {
4442 	struct amdgpu_device *adev = ring->adev;
4443 
4444 	if (ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE) {
4445 		if (amdgpu_gfx_is_high_priority_compute_queue(adev, ring)) {
4446 			mqd->cp_hqd_pipe_priority = AMDGPU_GFX_PIPE_PRIO_HIGH;
4447 			mqd->cp_hqd_queue_priority =
4448 				AMDGPU_GFX_QUEUE_PRIORITY_MAXIMUM;
4449 		}
4450 	}
4451 }
4452 
4453 static int gfx_v8_0_mqd_init(struct amdgpu_ring *ring)
4454 {
4455 	struct amdgpu_device *adev = ring->adev;
4456 	struct vi_mqd *mqd = ring->mqd_ptr;
4457 	uint64_t hqd_gpu_addr, wb_gpu_addr, eop_base_addr;
4458 	uint32_t tmp;
4459 
4460 	mqd->header = 0xC0310800;
4461 	mqd->compute_pipelinestat_enable = 0x00000001;
4462 	mqd->compute_static_thread_mgmt_se0 = 0xffffffff;
4463 	mqd->compute_static_thread_mgmt_se1 = 0xffffffff;
4464 	mqd->compute_static_thread_mgmt_se2 = 0xffffffff;
4465 	mqd->compute_static_thread_mgmt_se3 = 0xffffffff;
4466 	mqd->compute_misc_reserved = 0x00000003;
4467 	mqd->dynamic_cu_mask_addr_lo = lower_32_bits(ring->mqd_gpu_addr
4468 						     + offsetof(struct vi_mqd_allocation, dynamic_cu_mask));
4469 	mqd->dynamic_cu_mask_addr_hi = upper_32_bits(ring->mqd_gpu_addr
4470 						     + offsetof(struct vi_mqd_allocation, dynamic_cu_mask));
4471 	eop_base_addr = ring->eop_gpu_addr >> 8;
4472 	mqd->cp_hqd_eop_base_addr_lo = eop_base_addr;
4473 	mqd->cp_hqd_eop_base_addr_hi = upper_32_bits(eop_base_addr);
4474 
4475 	/* set the EOP size, register value is 2^(EOP_SIZE+1) dwords */
4476 	tmp = RREG32(mmCP_HQD_EOP_CONTROL);
4477 	tmp = REG_SET_FIELD(tmp, CP_HQD_EOP_CONTROL, EOP_SIZE,
4478 			(order_base_2(GFX8_MEC_HPD_SIZE / 4) - 1));
4479 
4480 	mqd->cp_hqd_eop_control = tmp;
4481 
4482 	/* enable doorbell? */
4483 	tmp = REG_SET_FIELD(RREG32(mmCP_HQD_PQ_DOORBELL_CONTROL),
4484 			    CP_HQD_PQ_DOORBELL_CONTROL,
4485 			    DOORBELL_EN,
4486 			    ring->use_doorbell ? 1 : 0);
4487 
4488 	mqd->cp_hqd_pq_doorbell_control = tmp;
4489 
4490 	/* set the pointer to the MQD */
4491 	mqd->cp_mqd_base_addr_lo = ring->mqd_gpu_addr & 0xfffffffc;
4492 	mqd->cp_mqd_base_addr_hi = upper_32_bits(ring->mqd_gpu_addr);
4493 
4494 	/* set MQD vmid to 0 */
4495 	tmp = RREG32(mmCP_MQD_CONTROL);
4496 	tmp = REG_SET_FIELD(tmp, CP_MQD_CONTROL, VMID, 0);
4497 	mqd->cp_mqd_control = tmp;
4498 
4499 	/* set the pointer to the HQD, this is similar CP_RB0_BASE/_HI */
4500 	hqd_gpu_addr = ring->gpu_addr >> 8;
4501 	mqd->cp_hqd_pq_base_lo = hqd_gpu_addr;
4502 	mqd->cp_hqd_pq_base_hi = upper_32_bits(hqd_gpu_addr);
4503 
4504 	/* set up the HQD, this is similar to CP_RB0_CNTL */
4505 	tmp = RREG32(mmCP_HQD_PQ_CONTROL);
4506 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, QUEUE_SIZE,
4507 			    (order_base_2(ring->ring_size / 4) - 1));
4508 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, RPTR_BLOCK_SIZE,
4509 			((order_base_2(AMDGPU_GPU_PAGE_SIZE / 4) - 1) << 8));
4510 #ifdef __BIG_ENDIAN
4511 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, ENDIAN_SWAP, 1);
4512 #endif
4513 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, UNORD_DISPATCH, 0);
4514 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, ROQ_PQ_IB_FLIP, 0);
4515 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, PRIV_STATE, 1);
4516 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, KMD_QUEUE, 1);
4517 	mqd->cp_hqd_pq_control = tmp;
4518 
4519 	/* set the wb address whether it's enabled or not */
4520 	wb_gpu_addr = adev->wb.gpu_addr + (ring->rptr_offs * 4);
4521 	mqd->cp_hqd_pq_rptr_report_addr_lo = wb_gpu_addr & 0xfffffffc;
4522 	mqd->cp_hqd_pq_rptr_report_addr_hi =
4523 		upper_32_bits(wb_gpu_addr) & 0xffff;
4524 
4525 	/* only used if CP_PQ_WPTR_POLL_CNTL.CP_PQ_WPTR_POLL_CNTL__EN_MASK=1 */
4526 	wb_gpu_addr = adev->wb.gpu_addr + (ring->wptr_offs * 4);
4527 	mqd->cp_hqd_pq_wptr_poll_addr_lo = wb_gpu_addr & 0xfffffffc;
4528 	mqd->cp_hqd_pq_wptr_poll_addr_hi = upper_32_bits(wb_gpu_addr) & 0xffff;
4529 
4530 	tmp = 0;
4531 	/* enable the doorbell if requested */
4532 	if (ring->use_doorbell) {
4533 		tmp = RREG32(mmCP_HQD_PQ_DOORBELL_CONTROL);
4534 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
4535 				DOORBELL_OFFSET, ring->doorbell_index);
4536 
4537 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
4538 					 DOORBELL_EN, 1);
4539 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
4540 					 DOORBELL_SOURCE, 0);
4541 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
4542 					 DOORBELL_HIT, 0);
4543 	}
4544 
4545 	mqd->cp_hqd_pq_doorbell_control = tmp;
4546 
4547 	/* reset read and write pointers, similar to CP_RB0_WPTR/_RPTR */
4548 	ring->wptr = 0;
4549 	mqd->cp_hqd_pq_wptr = ring->wptr;
4550 	mqd->cp_hqd_pq_rptr = RREG32(mmCP_HQD_PQ_RPTR);
4551 
4552 	/* set the vmid for the queue */
4553 	mqd->cp_hqd_vmid = 0;
4554 
4555 	tmp = RREG32(mmCP_HQD_PERSISTENT_STATE);
4556 	tmp = REG_SET_FIELD(tmp, CP_HQD_PERSISTENT_STATE, PRELOAD_SIZE, 0x53);
4557 	mqd->cp_hqd_persistent_state = tmp;
4558 
4559 	/* set MTYPE */
4560 	tmp = RREG32(mmCP_HQD_IB_CONTROL);
4561 	tmp = REG_SET_FIELD(tmp, CP_HQD_IB_CONTROL, MIN_IB_AVAIL_SIZE, 3);
4562 	tmp = REG_SET_FIELD(tmp, CP_HQD_IB_CONTROL, MTYPE, 3);
4563 	mqd->cp_hqd_ib_control = tmp;
4564 
4565 	tmp = RREG32(mmCP_HQD_IQ_TIMER);
4566 	tmp = REG_SET_FIELD(tmp, CP_HQD_IQ_TIMER, MTYPE, 3);
4567 	mqd->cp_hqd_iq_timer = tmp;
4568 
4569 	tmp = RREG32(mmCP_HQD_CTX_SAVE_CONTROL);
4570 	tmp = REG_SET_FIELD(tmp, CP_HQD_CTX_SAVE_CONTROL, MTYPE, 3);
4571 	mqd->cp_hqd_ctx_save_control = tmp;
4572 
4573 	/* defaults */
4574 	mqd->cp_hqd_eop_rptr = RREG32(mmCP_HQD_EOP_RPTR);
4575 	mqd->cp_hqd_eop_wptr = RREG32(mmCP_HQD_EOP_WPTR);
4576 	mqd->cp_hqd_ctx_save_base_addr_lo = RREG32(mmCP_HQD_CTX_SAVE_BASE_ADDR_LO);
4577 	mqd->cp_hqd_ctx_save_base_addr_hi = RREG32(mmCP_HQD_CTX_SAVE_BASE_ADDR_HI);
4578 	mqd->cp_hqd_cntl_stack_offset = RREG32(mmCP_HQD_CNTL_STACK_OFFSET);
4579 	mqd->cp_hqd_cntl_stack_size = RREG32(mmCP_HQD_CNTL_STACK_SIZE);
4580 	mqd->cp_hqd_wg_state_offset = RREG32(mmCP_HQD_WG_STATE_OFFSET);
4581 	mqd->cp_hqd_ctx_save_size = RREG32(mmCP_HQD_CTX_SAVE_SIZE);
4582 	mqd->cp_hqd_eop_done_events = RREG32(mmCP_HQD_EOP_EVENTS);
4583 	mqd->cp_hqd_error = RREG32(mmCP_HQD_ERROR);
4584 	mqd->cp_hqd_eop_wptr_mem = RREG32(mmCP_HQD_EOP_WPTR_MEM);
4585 	mqd->cp_hqd_eop_dones = RREG32(mmCP_HQD_EOP_DONES);
4586 
4587 	/* set static priority for a queue/ring */
4588 	gfx_v8_0_mqd_set_priority(ring, mqd);
4589 	mqd->cp_hqd_quantum = RREG32(mmCP_HQD_QUANTUM);
4590 
4591 	/* map_queues packet doesn't need activate the queue,
4592 	 * so only kiq need set this field.
4593 	 */
4594 	if (ring->funcs->type == AMDGPU_RING_TYPE_KIQ)
4595 		mqd->cp_hqd_active = 1;
4596 
4597 	return 0;
4598 }
4599 
4600 static int gfx_v8_0_mqd_commit(struct amdgpu_device *adev,
4601 			struct vi_mqd *mqd)
4602 {
4603 	uint32_t mqd_reg;
4604 	uint32_t *mqd_data;
4605 
4606 	/* HQD registers extend from mmCP_MQD_BASE_ADDR to mmCP_HQD_ERROR */
4607 	mqd_data = &mqd->cp_mqd_base_addr_lo;
4608 
4609 	/* disable wptr polling */
4610 	WREG32_FIELD(CP_PQ_WPTR_POLL_CNTL, EN, 0);
4611 
4612 	/* program all HQD registers */
4613 	for (mqd_reg = mmCP_HQD_VMID; mqd_reg <= mmCP_HQD_EOP_CONTROL; mqd_reg++)
4614 		WREG32(mqd_reg, mqd_data[mqd_reg - mmCP_MQD_BASE_ADDR]);
4615 
4616 	/* Tonga errata: EOP RPTR/WPTR should be left unmodified.
4617 	 * This is safe since EOP RPTR==WPTR for any inactive HQD
4618 	 * on ASICs that do not support context-save.
4619 	 * EOP writes/reads can start anywhere in the ring.
4620 	 */
4621 	if (adev->asic_type != CHIP_TONGA) {
4622 		WREG32(mmCP_HQD_EOP_RPTR, mqd->cp_hqd_eop_rptr);
4623 		WREG32(mmCP_HQD_EOP_WPTR, mqd->cp_hqd_eop_wptr);
4624 		WREG32(mmCP_HQD_EOP_WPTR_MEM, mqd->cp_hqd_eop_wptr_mem);
4625 	}
4626 
4627 	for (mqd_reg = mmCP_HQD_EOP_EVENTS; mqd_reg <= mmCP_HQD_ERROR; mqd_reg++)
4628 		WREG32(mqd_reg, mqd_data[mqd_reg - mmCP_MQD_BASE_ADDR]);
4629 
4630 	/* activate the HQD */
4631 	for (mqd_reg = mmCP_MQD_BASE_ADDR; mqd_reg <= mmCP_HQD_ACTIVE; mqd_reg++)
4632 		WREG32(mqd_reg, mqd_data[mqd_reg - mmCP_MQD_BASE_ADDR]);
4633 
4634 	return 0;
4635 }
4636 
4637 static int gfx_v8_0_kiq_init_queue(struct amdgpu_ring *ring)
4638 {
4639 	struct amdgpu_device *adev = ring->adev;
4640 	struct vi_mqd *mqd = ring->mqd_ptr;
4641 	int mqd_idx = AMDGPU_MAX_COMPUTE_RINGS;
4642 
4643 	gfx_v8_0_kiq_setting(ring);
4644 
4645 	if (amdgpu_in_reset(adev)) { /* for GPU_RESET case */
4646 		/* reset MQD to a clean status */
4647 		if (adev->gfx.mec.mqd_backup[mqd_idx])
4648 			memcpy(mqd, adev->gfx.mec.mqd_backup[mqd_idx], sizeof(struct vi_mqd_allocation));
4649 
4650 		/* reset ring buffer */
4651 		ring->wptr = 0;
4652 		amdgpu_ring_clear_ring(ring);
4653 		mutex_lock(&adev->srbm_mutex);
4654 		vi_srbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
4655 		gfx_v8_0_mqd_commit(adev, mqd);
4656 		vi_srbm_select(adev, 0, 0, 0, 0);
4657 		mutex_unlock(&adev->srbm_mutex);
4658 	} else {
4659 		memset((void *)mqd, 0, sizeof(struct vi_mqd_allocation));
4660 		((struct vi_mqd_allocation *)mqd)->dynamic_cu_mask = 0xFFFFFFFF;
4661 		((struct vi_mqd_allocation *)mqd)->dynamic_rb_mask = 0xFFFFFFFF;
4662 		mutex_lock(&adev->srbm_mutex);
4663 		vi_srbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
4664 		gfx_v8_0_mqd_init(ring);
4665 		gfx_v8_0_mqd_commit(adev, mqd);
4666 		vi_srbm_select(adev, 0, 0, 0, 0);
4667 		mutex_unlock(&adev->srbm_mutex);
4668 
4669 		if (adev->gfx.mec.mqd_backup[mqd_idx])
4670 			memcpy(adev->gfx.mec.mqd_backup[mqd_idx], mqd, sizeof(struct vi_mqd_allocation));
4671 	}
4672 
4673 	return 0;
4674 }
4675 
4676 static int gfx_v8_0_kcq_init_queue(struct amdgpu_ring *ring)
4677 {
4678 	struct amdgpu_device *adev = ring->adev;
4679 	struct vi_mqd *mqd = ring->mqd_ptr;
4680 	int mqd_idx = ring - &adev->gfx.compute_ring[0];
4681 
4682 	if (!amdgpu_in_reset(adev) && !adev->in_suspend) {
4683 		memset((void *)mqd, 0, sizeof(struct vi_mqd_allocation));
4684 		((struct vi_mqd_allocation *)mqd)->dynamic_cu_mask = 0xFFFFFFFF;
4685 		((struct vi_mqd_allocation *)mqd)->dynamic_rb_mask = 0xFFFFFFFF;
4686 		mutex_lock(&adev->srbm_mutex);
4687 		vi_srbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
4688 		gfx_v8_0_mqd_init(ring);
4689 		vi_srbm_select(adev, 0, 0, 0, 0);
4690 		mutex_unlock(&adev->srbm_mutex);
4691 
4692 		if (adev->gfx.mec.mqd_backup[mqd_idx])
4693 			memcpy(adev->gfx.mec.mqd_backup[mqd_idx], mqd, sizeof(struct vi_mqd_allocation));
4694 	} else if (amdgpu_in_reset(adev)) { /* for GPU_RESET case */
4695 		/* reset MQD to a clean status */
4696 		if (adev->gfx.mec.mqd_backup[mqd_idx])
4697 			memcpy(mqd, adev->gfx.mec.mqd_backup[mqd_idx], sizeof(struct vi_mqd_allocation));
4698 		/* reset ring buffer */
4699 		ring->wptr = 0;
4700 		amdgpu_ring_clear_ring(ring);
4701 	} else {
4702 		amdgpu_ring_clear_ring(ring);
4703 	}
4704 	return 0;
4705 }
4706 
4707 static void gfx_v8_0_set_mec_doorbell_range(struct amdgpu_device *adev)
4708 {
4709 	if (adev->asic_type > CHIP_TONGA) {
4710 		WREG32(mmCP_MEC_DOORBELL_RANGE_LOWER, adev->doorbell_index.kiq << 2);
4711 		WREG32(mmCP_MEC_DOORBELL_RANGE_UPPER, adev->doorbell_index.mec_ring7 << 2);
4712 	}
4713 	/* enable doorbells */
4714 	WREG32_FIELD(CP_PQ_STATUS, DOORBELL_ENABLE, 1);
4715 }
4716 
4717 static int gfx_v8_0_kiq_resume(struct amdgpu_device *adev)
4718 {
4719 	struct amdgpu_ring *ring;
4720 	int r;
4721 
4722 	ring = &adev->gfx.kiq.ring;
4723 
4724 	r = amdgpu_bo_reserve(ring->mqd_obj, false);
4725 	if (unlikely(r != 0))
4726 		return r;
4727 
4728 	r = amdgpu_bo_kmap(ring->mqd_obj, &ring->mqd_ptr);
4729 	if (unlikely(r != 0))
4730 		return r;
4731 
4732 	gfx_v8_0_kiq_init_queue(ring);
4733 	amdgpu_bo_kunmap(ring->mqd_obj);
4734 	ring->mqd_ptr = NULL;
4735 	amdgpu_bo_unreserve(ring->mqd_obj);
4736 	ring->sched.ready = true;
4737 	return 0;
4738 }
4739 
4740 static int gfx_v8_0_kcq_resume(struct amdgpu_device *adev)
4741 {
4742 	struct amdgpu_ring *ring = NULL;
4743 	int r = 0, i;
4744 
4745 	gfx_v8_0_cp_compute_enable(adev, true);
4746 
4747 	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
4748 		ring = &adev->gfx.compute_ring[i];
4749 
4750 		r = amdgpu_bo_reserve(ring->mqd_obj, false);
4751 		if (unlikely(r != 0))
4752 			goto done;
4753 		r = amdgpu_bo_kmap(ring->mqd_obj, &ring->mqd_ptr);
4754 		if (!r) {
4755 			r = gfx_v8_0_kcq_init_queue(ring);
4756 			amdgpu_bo_kunmap(ring->mqd_obj);
4757 			ring->mqd_ptr = NULL;
4758 		}
4759 		amdgpu_bo_unreserve(ring->mqd_obj);
4760 		if (r)
4761 			goto done;
4762 	}
4763 
4764 	gfx_v8_0_set_mec_doorbell_range(adev);
4765 
4766 	r = gfx_v8_0_kiq_kcq_enable(adev);
4767 	if (r)
4768 		goto done;
4769 
4770 done:
4771 	return r;
4772 }
4773 
4774 static int gfx_v8_0_cp_test_all_rings(struct amdgpu_device *adev)
4775 {
4776 	int r, i;
4777 	struct amdgpu_ring *ring;
4778 
4779 	/* collect all the ring_tests here, gfx, kiq, compute */
4780 	ring = &adev->gfx.gfx_ring[0];
4781 	r = amdgpu_ring_test_helper(ring);
4782 	if (r)
4783 		return r;
4784 
4785 	ring = &adev->gfx.kiq.ring;
4786 	r = amdgpu_ring_test_helper(ring);
4787 	if (r)
4788 		return r;
4789 
4790 	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
4791 		ring = &adev->gfx.compute_ring[i];
4792 		amdgpu_ring_test_helper(ring);
4793 	}
4794 
4795 	return 0;
4796 }
4797 
4798 static int gfx_v8_0_cp_resume(struct amdgpu_device *adev)
4799 {
4800 	int r;
4801 
4802 	if (!(adev->flags & AMD_IS_APU))
4803 		gfx_v8_0_enable_gui_idle_interrupt(adev, false);
4804 
4805 	r = gfx_v8_0_kiq_resume(adev);
4806 	if (r)
4807 		return r;
4808 
4809 	r = gfx_v8_0_cp_gfx_resume(adev);
4810 	if (r)
4811 		return r;
4812 
4813 	r = gfx_v8_0_kcq_resume(adev);
4814 	if (r)
4815 		return r;
4816 
4817 	r = gfx_v8_0_cp_test_all_rings(adev);
4818 	if (r)
4819 		return r;
4820 
4821 	gfx_v8_0_enable_gui_idle_interrupt(adev, true);
4822 
4823 	return 0;
4824 }
4825 
4826 static void gfx_v8_0_cp_enable(struct amdgpu_device *adev, bool enable)
4827 {
4828 	gfx_v8_0_cp_gfx_enable(adev, enable);
4829 	gfx_v8_0_cp_compute_enable(adev, enable);
4830 }
4831 
4832 static int gfx_v8_0_hw_init(void *handle)
4833 {
4834 	int r;
4835 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4836 
4837 	gfx_v8_0_init_golden_registers(adev);
4838 	gfx_v8_0_constants_init(adev);
4839 
4840 	r = adev->gfx.rlc.funcs->resume(adev);
4841 	if (r)
4842 		return r;
4843 
4844 	r = gfx_v8_0_cp_resume(adev);
4845 
4846 	return r;
4847 }
4848 
4849 static int gfx_v8_0_kcq_disable(struct amdgpu_device *adev)
4850 {
4851 	int r, i;
4852 	struct amdgpu_ring *kiq_ring = &adev->gfx.kiq.ring;
4853 
4854 	r = amdgpu_ring_alloc(kiq_ring, 6 * adev->gfx.num_compute_rings);
4855 	if (r)
4856 		DRM_ERROR("Failed to lock KIQ (%d).\n", r);
4857 
4858 	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
4859 		struct amdgpu_ring *ring = &adev->gfx.compute_ring[i];
4860 
4861 		amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_UNMAP_QUEUES, 4));
4862 		amdgpu_ring_write(kiq_ring, /* Q_sel: 0, vmid: 0, engine: 0, num_Q: 1 */
4863 						PACKET3_UNMAP_QUEUES_ACTION(1) | /* RESET_QUEUES */
4864 						PACKET3_UNMAP_QUEUES_QUEUE_SEL(0) |
4865 						PACKET3_UNMAP_QUEUES_ENGINE_SEL(0) |
4866 						PACKET3_UNMAP_QUEUES_NUM_QUEUES(1));
4867 		amdgpu_ring_write(kiq_ring, PACKET3_UNMAP_QUEUES_DOORBELL_OFFSET0(ring->doorbell_index));
4868 		amdgpu_ring_write(kiq_ring, 0);
4869 		amdgpu_ring_write(kiq_ring, 0);
4870 		amdgpu_ring_write(kiq_ring, 0);
4871 	}
4872 	r = amdgpu_ring_test_helper(kiq_ring);
4873 	if (r)
4874 		DRM_ERROR("KCQ disable failed\n");
4875 
4876 	return r;
4877 }
4878 
4879 static bool gfx_v8_0_is_idle(void *handle)
4880 {
4881 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4882 
4883 	if (REG_GET_FIELD(RREG32(mmGRBM_STATUS), GRBM_STATUS, GUI_ACTIVE)
4884 		|| RREG32(mmGRBM_STATUS2) != 0x8)
4885 		return false;
4886 	else
4887 		return true;
4888 }
4889 
4890 static bool gfx_v8_0_rlc_is_idle(void *handle)
4891 {
4892 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4893 
4894 	if (RREG32(mmGRBM_STATUS2) != 0x8)
4895 		return false;
4896 	else
4897 		return true;
4898 }
4899 
4900 static int gfx_v8_0_wait_for_rlc_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_rlc_is_idle(handle))
4907 			return 0;
4908 
4909 		udelay(1);
4910 	}
4911 	return -ETIMEDOUT;
4912 }
4913 
4914 static int gfx_v8_0_wait_for_idle(void *handle)
4915 {
4916 	unsigned int i;
4917 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4918 
4919 	for (i = 0; i < adev->usec_timeout; i++) {
4920 		if (gfx_v8_0_is_idle(handle))
4921 			return 0;
4922 
4923 		udelay(1);
4924 	}
4925 	return -ETIMEDOUT;
4926 }
4927 
4928 static int gfx_v8_0_hw_fini(void *handle)
4929 {
4930 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4931 
4932 	amdgpu_irq_put(adev, &adev->gfx.priv_reg_irq, 0);
4933 	amdgpu_irq_put(adev, &adev->gfx.priv_inst_irq, 0);
4934 
4935 	amdgpu_irq_put(adev, &adev->gfx.cp_ecc_error_irq, 0);
4936 
4937 	amdgpu_irq_put(adev, &adev->gfx.sq_irq, 0);
4938 
4939 	/* disable KCQ to avoid CPC touch memory not valid anymore */
4940 	gfx_v8_0_kcq_disable(adev);
4941 
4942 	if (amdgpu_sriov_vf(adev)) {
4943 		pr_debug("For SRIOV client, shouldn't do anything.\n");
4944 		return 0;
4945 	}
4946 	amdgpu_gfx_rlc_enter_safe_mode(adev);
4947 	if (!gfx_v8_0_wait_for_idle(adev))
4948 		gfx_v8_0_cp_enable(adev, false);
4949 	else
4950 		pr_err("cp is busy, skip halt cp\n");
4951 	if (!gfx_v8_0_wait_for_rlc_idle(adev))
4952 		adev->gfx.rlc.funcs->stop(adev);
4953 	else
4954 		pr_err("rlc is busy, skip halt rlc\n");
4955 	amdgpu_gfx_rlc_exit_safe_mode(adev);
4956 
4957 	return 0;
4958 }
4959 
4960 static int gfx_v8_0_suspend(void *handle)
4961 {
4962 	return gfx_v8_0_hw_fini(handle);
4963 }
4964 
4965 static int gfx_v8_0_resume(void *handle)
4966 {
4967 	return gfx_v8_0_hw_init(handle);
4968 }
4969 
4970 static bool gfx_v8_0_check_soft_reset(void *handle)
4971 {
4972 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4973 	u32 grbm_soft_reset = 0, srbm_soft_reset = 0;
4974 	u32 tmp;
4975 
4976 	/* GRBM_STATUS */
4977 	tmp = RREG32(mmGRBM_STATUS);
4978 	if (tmp & (GRBM_STATUS__PA_BUSY_MASK | GRBM_STATUS__SC_BUSY_MASK |
4979 		   GRBM_STATUS__BCI_BUSY_MASK | GRBM_STATUS__SX_BUSY_MASK |
4980 		   GRBM_STATUS__TA_BUSY_MASK | GRBM_STATUS__VGT_BUSY_MASK |
4981 		   GRBM_STATUS__DB_BUSY_MASK | GRBM_STATUS__CB_BUSY_MASK |
4982 		   GRBM_STATUS__GDS_BUSY_MASK | GRBM_STATUS__SPI_BUSY_MASK |
4983 		   GRBM_STATUS__IA_BUSY_MASK | GRBM_STATUS__IA_BUSY_NO_DMA_MASK |
4984 		   GRBM_STATUS__CP_BUSY_MASK | GRBM_STATUS__CP_COHERENCY_BUSY_MASK)) {
4985 		grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset,
4986 						GRBM_SOFT_RESET, SOFT_RESET_CP, 1);
4987 		grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset,
4988 						GRBM_SOFT_RESET, SOFT_RESET_GFX, 1);
4989 		srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset,
4990 						SRBM_SOFT_RESET, SOFT_RESET_GRBM, 1);
4991 	}
4992 
4993 	/* GRBM_STATUS2 */
4994 	tmp = RREG32(mmGRBM_STATUS2);
4995 	if (REG_GET_FIELD(tmp, GRBM_STATUS2, RLC_BUSY))
4996 		grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset,
4997 						GRBM_SOFT_RESET, SOFT_RESET_RLC, 1);
4998 
4999 	if (REG_GET_FIELD(tmp, GRBM_STATUS2, CPF_BUSY) ||
5000 	    REG_GET_FIELD(tmp, GRBM_STATUS2, CPC_BUSY) ||
5001 	    REG_GET_FIELD(tmp, GRBM_STATUS2, CPG_BUSY)) {
5002 		grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET,
5003 						SOFT_RESET_CPF, 1);
5004 		grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET,
5005 						SOFT_RESET_CPC, 1);
5006 		grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET,
5007 						SOFT_RESET_CPG, 1);
5008 		srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset, SRBM_SOFT_RESET,
5009 						SOFT_RESET_GRBM, 1);
5010 	}
5011 
5012 	/* SRBM_STATUS */
5013 	tmp = RREG32(mmSRBM_STATUS);
5014 	if (REG_GET_FIELD(tmp, SRBM_STATUS, GRBM_RQ_PENDING))
5015 		srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset,
5016 						SRBM_SOFT_RESET, SOFT_RESET_GRBM, 1);
5017 	if (REG_GET_FIELD(tmp, SRBM_STATUS, SEM_BUSY))
5018 		srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset,
5019 						SRBM_SOFT_RESET, SOFT_RESET_SEM, 1);
5020 
5021 	if (grbm_soft_reset || srbm_soft_reset) {
5022 		adev->gfx.grbm_soft_reset = grbm_soft_reset;
5023 		adev->gfx.srbm_soft_reset = srbm_soft_reset;
5024 		return true;
5025 	} else {
5026 		adev->gfx.grbm_soft_reset = 0;
5027 		adev->gfx.srbm_soft_reset = 0;
5028 		return false;
5029 	}
5030 }
5031 
5032 static int gfx_v8_0_pre_soft_reset(void *handle)
5033 {
5034 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5035 	u32 grbm_soft_reset = 0;
5036 
5037 	if ((!adev->gfx.grbm_soft_reset) &&
5038 	    (!adev->gfx.srbm_soft_reset))
5039 		return 0;
5040 
5041 	grbm_soft_reset = adev->gfx.grbm_soft_reset;
5042 
5043 	/* stop the rlc */
5044 	adev->gfx.rlc.funcs->stop(adev);
5045 
5046 	if (REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CP) ||
5047 	    REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_GFX))
5048 		/* Disable GFX parsing/prefetching */
5049 		gfx_v8_0_cp_gfx_enable(adev, false);
5050 
5051 	if (REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CP) ||
5052 	    REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CPF) ||
5053 	    REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CPC) ||
5054 	    REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CPG)) {
5055 		int i;
5056 
5057 		for (i = 0; i < adev->gfx.num_compute_rings; i++) {
5058 			struct amdgpu_ring *ring = &adev->gfx.compute_ring[i];
5059 
5060 			mutex_lock(&adev->srbm_mutex);
5061 			vi_srbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
5062 			gfx_v8_0_deactivate_hqd(adev, 2);
5063 			vi_srbm_select(adev, 0, 0, 0, 0);
5064 			mutex_unlock(&adev->srbm_mutex);
5065 		}
5066 		/* Disable MEC parsing/prefetching */
5067 		gfx_v8_0_cp_compute_enable(adev, false);
5068 	}
5069 
5070 	return 0;
5071 }
5072 
5073 static int gfx_v8_0_soft_reset(void *handle)
5074 {
5075 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5076 	u32 grbm_soft_reset = 0, srbm_soft_reset = 0;
5077 	u32 tmp;
5078 
5079 	if ((!adev->gfx.grbm_soft_reset) &&
5080 	    (!adev->gfx.srbm_soft_reset))
5081 		return 0;
5082 
5083 	grbm_soft_reset = adev->gfx.grbm_soft_reset;
5084 	srbm_soft_reset = adev->gfx.srbm_soft_reset;
5085 
5086 	if (grbm_soft_reset || srbm_soft_reset) {
5087 		tmp = RREG32(mmGMCON_DEBUG);
5088 		tmp = REG_SET_FIELD(tmp, GMCON_DEBUG, GFX_STALL, 1);
5089 		tmp = REG_SET_FIELD(tmp, GMCON_DEBUG, GFX_CLEAR, 1);
5090 		WREG32(mmGMCON_DEBUG, tmp);
5091 		udelay(50);
5092 	}
5093 
5094 	if (grbm_soft_reset) {
5095 		tmp = RREG32(mmGRBM_SOFT_RESET);
5096 		tmp |= grbm_soft_reset;
5097 		dev_info(adev->dev, "GRBM_SOFT_RESET=0x%08X\n", tmp);
5098 		WREG32(mmGRBM_SOFT_RESET, tmp);
5099 		tmp = RREG32(mmGRBM_SOFT_RESET);
5100 
5101 		udelay(50);
5102 
5103 		tmp &= ~grbm_soft_reset;
5104 		WREG32(mmGRBM_SOFT_RESET, tmp);
5105 		tmp = RREG32(mmGRBM_SOFT_RESET);
5106 	}
5107 
5108 	if (srbm_soft_reset) {
5109 		tmp = RREG32(mmSRBM_SOFT_RESET);
5110 		tmp |= srbm_soft_reset;
5111 		dev_info(adev->dev, "SRBM_SOFT_RESET=0x%08X\n", tmp);
5112 		WREG32(mmSRBM_SOFT_RESET, tmp);
5113 		tmp = RREG32(mmSRBM_SOFT_RESET);
5114 
5115 		udelay(50);
5116 
5117 		tmp &= ~srbm_soft_reset;
5118 		WREG32(mmSRBM_SOFT_RESET, tmp);
5119 		tmp = RREG32(mmSRBM_SOFT_RESET);
5120 	}
5121 
5122 	if (grbm_soft_reset || srbm_soft_reset) {
5123 		tmp = RREG32(mmGMCON_DEBUG);
5124 		tmp = REG_SET_FIELD(tmp, GMCON_DEBUG, GFX_STALL, 0);
5125 		tmp = REG_SET_FIELD(tmp, GMCON_DEBUG, GFX_CLEAR, 0);
5126 		WREG32(mmGMCON_DEBUG, tmp);
5127 	}
5128 
5129 	/* Wait a little for things to settle down */
5130 	udelay(50);
5131 
5132 	return 0;
5133 }
5134 
5135 static int gfx_v8_0_post_soft_reset(void *handle)
5136 {
5137 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5138 	u32 grbm_soft_reset = 0;
5139 
5140 	if ((!adev->gfx.grbm_soft_reset) &&
5141 	    (!adev->gfx.srbm_soft_reset))
5142 		return 0;
5143 
5144 	grbm_soft_reset = adev->gfx.grbm_soft_reset;
5145 
5146 	if (REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CP) ||
5147 	    REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CPF) ||
5148 	    REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CPC) ||
5149 	    REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CPG)) {
5150 		int i;
5151 
5152 		for (i = 0; i < adev->gfx.num_compute_rings; i++) {
5153 			struct amdgpu_ring *ring = &adev->gfx.compute_ring[i];
5154 
5155 			mutex_lock(&adev->srbm_mutex);
5156 			vi_srbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
5157 			gfx_v8_0_deactivate_hqd(adev, 2);
5158 			vi_srbm_select(adev, 0, 0, 0, 0);
5159 			mutex_unlock(&adev->srbm_mutex);
5160 		}
5161 		gfx_v8_0_kiq_resume(adev);
5162 		gfx_v8_0_kcq_resume(adev);
5163 	}
5164 
5165 	if (REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CP) ||
5166 	    REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_GFX))
5167 		gfx_v8_0_cp_gfx_resume(adev);
5168 
5169 	gfx_v8_0_cp_test_all_rings(adev);
5170 
5171 	adev->gfx.rlc.funcs->start(adev);
5172 
5173 	return 0;
5174 }
5175 
5176 /**
5177  * gfx_v8_0_get_gpu_clock_counter - return GPU clock counter snapshot
5178  *
5179  * @adev: amdgpu_device pointer
5180  *
5181  * Fetches a GPU clock counter snapshot.
5182  * Returns the 64 bit clock counter snapshot.
5183  */
5184 static uint64_t gfx_v8_0_get_gpu_clock_counter(struct amdgpu_device *adev)
5185 {
5186 	uint64_t clock;
5187 
5188 	mutex_lock(&adev->gfx.gpu_clock_mutex);
5189 	WREG32(mmRLC_CAPTURE_GPU_CLOCK_COUNT, 1);
5190 	clock = (uint64_t)RREG32(mmRLC_GPU_CLOCK_COUNT_LSB) |
5191 		((uint64_t)RREG32(mmRLC_GPU_CLOCK_COUNT_MSB) << 32ULL);
5192 	mutex_unlock(&adev->gfx.gpu_clock_mutex);
5193 	return clock;
5194 }
5195 
5196 static void gfx_v8_0_ring_emit_gds_switch(struct amdgpu_ring *ring,
5197 					  uint32_t vmid,
5198 					  uint32_t gds_base, uint32_t gds_size,
5199 					  uint32_t gws_base, uint32_t gws_size,
5200 					  uint32_t oa_base, uint32_t oa_size)
5201 {
5202 	/* GDS Base */
5203 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
5204 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
5205 				WRITE_DATA_DST_SEL(0)));
5206 	amdgpu_ring_write(ring, amdgpu_gds_reg_offset[vmid].mem_base);
5207 	amdgpu_ring_write(ring, 0);
5208 	amdgpu_ring_write(ring, gds_base);
5209 
5210 	/* GDS Size */
5211 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
5212 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
5213 				WRITE_DATA_DST_SEL(0)));
5214 	amdgpu_ring_write(ring, amdgpu_gds_reg_offset[vmid].mem_size);
5215 	amdgpu_ring_write(ring, 0);
5216 	amdgpu_ring_write(ring, gds_size);
5217 
5218 	/* GWS */
5219 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
5220 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
5221 				WRITE_DATA_DST_SEL(0)));
5222 	amdgpu_ring_write(ring, amdgpu_gds_reg_offset[vmid].gws);
5223 	amdgpu_ring_write(ring, 0);
5224 	amdgpu_ring_write(ring, gws_size << GDS_GWS_VMID0__SIZE__SHIFT | gws_base);
5225 
5226 	/* OA */
5227 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
5228 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
5229 				WRITE_DATA_DST_SEL(0)));
5230 	amdgpu_ring_write(ring, amdgpu_gds_reg_offset[vmid].oa);
5231 	amdgpu_ring_write(ring, 0);
5232 	amdgpu_ring_write(ring, (1 << (oa_size + oa_base)) - (1 << oa_base));
5233 }
5234 
5235 static uint32_t wave_read_ind(struct amdgpu_device *adev, uint32_t simd, uint32_t wave, uint32_t address)
5236 {
5237 	WREG32(mmSQ_IND_INDEX,
5238 		(wave << SQ_IND_INDEX__WAVE_ID__SHIFT) |
5239 		(simd << SQ_IND_INDEX__SIMD_ID__SHIFT) |
5240 		(address << SQ_IND_INDEX__INDEX__SHIFT) |
5241 		(SQ_IND_INDEX__FORCE_READ_MASK));
5242 	return RREG32(mmSQ_IND_DATA);
5243 }
5244 
5245 static void wave_read_regs(struct amdgpu_device *adev, uint32_t simd,
5246 			   uint32_t wave, uint32_t thread,
5247 			   uint32_t regno, uint32_t num, uint32_t *out)
5248 {
5249 	WREG32(mmSQ_IND_INDEX,
5250 		(wave << SQ_IND_INDEX__WAVE_ID__SHIFT) |
5251 		(simd << SQ_IND_INDEX__SIMD_ID__SHIFT) |
5252 		(regno << SQ_IND_INDEX__INDEX__SHIFT) |
5253 		(thread << SQ_IND_INDEX__THREAD_ID__SHIFT) |
5254 		(SQ_IND_INDEX__FORCE_READ_MASK) |
5255 		(SQ_IND_INDEX__AUTO_INCR_MASK));
5256 	while (num--)
5257 		*(out++) = RREG32(mmSQ_IND_DATA);
5258 }
5259 
5260 static void gfx_v8_0_read_wave_data(struct amdgpu_device *adev, uint32_t simd, uint32_t wave, uint32_t *dst, int *no_fields)
5261 {
5262 	/* type 0 wave data */
5263 	dst[(*no_fields)++] = 0;
5264 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_STATUS);
5265 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_PC_LO);
5266 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_PC_HI);
5267 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_EXEC_LO);
5268 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_EXEC_HI);
5269 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_HW_ID);
5270 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_INST_DW0);
5271 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_INST_DW1);
5272 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_GPR_ALLOC);
5273 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_LDS_ALLOC);
5274 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TRAPSTS);
5275 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_IB_STS);
5276 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TBA_LO);
5277 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TBA_HI);
5278 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TMA_LO);
5279 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TMA_HI);
5280 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_IB_DBG0);
5281 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_M0);
5282 }
5283 
5284 static void gfx_v8_0_read_wave_sgprs(struct amdgpu_device *adev, uint32_t simd,
5285 				     uint32_t wave, uint32_t start,
5286 				     uint32_t size, uint32_t *dst)
5287 {
5288 	wave_read_regs(
5289 		adev, simd, wave, 0,
5290 		start + SQIND_WAVE_SGPRS_OFFSET, size, dst);
5291 }
5292 
5293 
5294 static const struct amdgpu_gfx_funcs gfx_v8_0_gfx_funcs = {
5295 	.get_gpu_clock_counter = &gfx_v8_0_get_gpu_clock_counter,
5296 	.select_se_sh = &gfx_v8_0_select_se_sh,
5297 	.read_wave_data = &gfx_v8_0_read_wave_data,
5298 	.read_wave_sgprs = &gfx_v8_0_read_wave_sgprs,
5299 	.select_me_pipe_q = &gfx_v8_0_select_me_pipe_q
5300 };
5301 
5302 static int gfx_v8_0_early_init(void *handle)
5303 {
5304 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5305 
5306 	adev->gfx.num_gfx_rings = GFX8_NUM_GFX_RINGS;
5307 	adev->gfx.num_compute_rings = min(amdgpu_gfx_get_num_kcq(adev),
5308 					  AMDGPU_MAX_COMPUTE_RINGS);
5309 	adev->gfx.funcs = &gfx_v8_0_gfx_funcs;
5310 	gfx_v8_0_set_ring_funcs(adev);
5311 	gfx_v8_0_set_irq_funcs(adev);
5312 	gfx_v8_0_set_gds_init(adev);
5313 	gfx_v8_0_set_rlc_funcs(adev);
5314 
5315 	return 0;
5316 }
5317 
5318 static int gfx_v8_0_late_init(void *handle)
5319 {
5320 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5321 	int r;
5322 
5323 	r = amdgpu_irq_get(adev, &adev->gfx.priv_reg_irq, 0);
5324 	if (r)
5325 		return r;
5326 
5327 	r = amdgpu_irq_get(adev, &adev->gfx.priv_inst_irq, 0);
5328 	if (r)
5329 		return r;
5330 
5331 	/* requires IBs so do in late init after IB pool is initialized */
5332 	r = gfx_v8_0_do_edc_gpr_workarounds(adev);
5333 	if (r)
5334 		return r;
5335 
5336 	r = amdgpu_irq_get(adev, &adev->gfx.cp_ecc_error_irq, 0);
5337 	if (r) {
5338 		DRM_ERROR("amdgpu_irq_get() failed to get IRQ for EDC, r: %d.\n", r);
5339 		return r;
5340 	}
5341 
5342 	r = amdgpu_irq_get(adev, &adev->gfx.sq_irq, 0);
5343 	if (r) {
5344 		DRM_ERROR(
5345 			"amdgpu_irq_get() failed to get IRQ for SQ, r: %d.\n",
5346 			r);
5347 		return r;
5348 	}
5349 
5350 	return 0;
5351 }
5352 
5353 static void gfx_v8_0_enable_gfx_static_mg_power_gating(struct amdgpu_device *adev,
5354 						       bool enable)
5355 {
5356 	if ((adev->asic_type == CHIP_POLARIS11) ||
5357 	    (adev->asic_type == CHIP_POLARIS12) ||
5358 	    (adev->asic_type == CHIP_VEGAM))
5359 		/* Send msg to SMU via Powerplay */
5360 		amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_GFX, enable);
5361 
5362 	WREG32_FIELD(RLC_PG_CNTL, STATIC_PER_CU_PG_ENABLE, enable ? 1 : 0);
5363 }
5364 
5365 static void gfx_v8_0_enable_gfx_dynamic_mg_power_gating(struct amdgpu_device *adev,
5366 							bool enable)
5367 {
5368 	WREG32_FIELD(RLC_PG_CNTL, DYN_PER_CU_PG_ENABLE, enable ? 1 : 0);
5369 }
5370 
5371 static void polaris11_enable_gfx_quick_mg_power_gating(struct amdgpu_device *adev,
5372 		bool enable)
5373 {
5374 	WREG32_FIELD(RLC_PG_CNTL, QUICK_PG_ENABLE, enable ? 1 : 0);
5375 }
5376 
5377 static void cz_enable_gfx_cg_power_gating(struct amdgpu_device *adev,
5378 					  bool enable)
5379 {
5380 	WREG32_FIELD(RLC_PG_CNTL, GFX_POWER_GATING_ENABLE, enable ? 1 : 0);
5381 }
5382 
5383 static void cz_enable_gfx_pipeline_power_gating(struct amdgpu_device *adev,
5384 						bool enable)
5385 {
5386 	WREG32_FIELD(RLC_PG_CNTL, GFX_PIPELINE_PG_ENABLE, enable ? 1 : 0);
5387 
5388 	/* Read any GFX register to wake up GFX. */
5389 	if (!enable)
5390 		RREG32(mmDB_RENDER_CONTROL);
5391 }
5392 
5393 static void cz_update_gfx_cg_power_gating(struct amdgpu_device *adev,
5394 					  bool enable)
5395 {
5396 	if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_PG) && enable) {
5397 		cz_enable_gfx_cg_power_gating(adev, true);
5398 		if (adev->pg_flags & AMD_PG_SUPPORT_GFX_PIPELINE)
5399 			cz_enable_gfx_pipeline_power_gating(adev, true);
5400 	} else {
5401 		cz_enable_gfx_cg_power_gating(adev, false);
5402 		cz_enable_gfx_pipeline_power_gating(adev, false);
5403 	}
5404 }
5405 
5406 static int gfx_v8_0_set_powergating_state(void *handle,
5407 					  enum amd_powergating_state state)
5408 {
5409 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5410 	bool enable = (state == AMD_PG_STATE_GATE);
5411 
5412 	if (amdgpu_sriov_vf(adev))
5413 		return 0;
5414 
5415 	if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_SMG |
5416 				AMD_PG_SUPPORT_RLC_SMU_HS |
5417 				AMD_PG_SUPPORT_CP |
5418 				AMD_PG_SUPPORT_GFX_DMG))
5419 		amdgpu_gfx_rlc_enter_safe_mode(adev);
5420 	switch (adev->asic_type) {
5421 	case CHIP_CARRIZO:
5422 	case CHIP_STONEY:
5423 
5424 		if (adev->pg_flags & AMD_PG_SUPPORT_RLC_SMU_HS) {
5425 			cz_enable_sck_slow_down_on_power_up(adev, true);
5426 			cz_enable_sck_slow_down_on_power_down(adev, true);
5427 		} else {
5428 			cz_enable_sck_slow_down_on_power_up(adev, false);
5429 			cz_enable_sck_slow_down_on_power_down(adev, false);
5430 		}
5431 		if (adev->pg_flags & AMD_PG_SUPPORT_CP)
5432 			cz_enable_cp_power_gating(adev, true);
5433 		else
5434 			cz_enable_cp_power_gating(adev, false);
5435 
5436 		cz_update_gfx_cg_power_gating(adev, enable);
5437 
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 		break;
5448 	case CHIP_POLARIS11:
5449 	case CHIP_POLARIS12:
5450 	case CHIP_VEGAM:
5451 		if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_SMG) && enable)
5452 			gfx_v8_0_enable_gfx_static_mg_power_gating(adev, true);
5453 		else
5454 			gfx_v8_0_enable_gfx_static_mg_power_gating(adev, false);
5455 
5456 		if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_DMG) && enable)
5457 			gfx_v8_0_enable_gfx_dynamic_mg_power_gating(adev, true);
5458 		else
5459 			gfx_v8_0_enable_gfx_dynamic_mg_power_gating(adev, false);
5460 
5461 		if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_QUICK_MG) && enable)
5462 			polaris11_enable_gfx_quick_mg_power_gating(adev, true);
5463 		else
5464 			polaris11_enable_gfx_quick_mg_power_gating(adev, false);
5465 		break;
5466 	default:
5467 		break;
5468 	}
5469 	if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_SMG |
5470 				AMD_PG_SUPPORT_RLC_SMU_HS |
5471 				AMD_PG_SUPPORT_CP |
5472 				AMD_PG_SUPPORT_GFX_DMG))
5473 		amdgpu_gfx_rlc_exit_safe_mode(adev);
5474 	return 0;
5475 }
5476 
5477 static void gfx_v8_0_get_clockgating_state(void *handle, u32 *flags)
5478 {
5479 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5480 	int data;
5481 
5482 	if (amdgpu_sriov_vf(adev))
5483 		*flags = 0;
5484 
5485 	/* AMD_CG_SUPPORT_GFX_MGCG */
5486 	data = RREG32(mmRLC_CGTT_MGCG_OVERRIDE);
5487 	if (!(data & RLC_CGTT_MGCG_OVERRIDE__CPF_MASK))
5488 		*flags |= AMD_CG_SUPPORT_GFX_MGCG;
5489 
5490 	/* AMD_CG_SUPPORT_GFX_CGLG */
5491 	data = RREG32(mmRLC_CGCG_CGLS_CTRL);
5492 	if (data & RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK)
5493 		*flags |= AMD_CG_SUPPORT_GFX_CGCG;
5494 
5495 	/* AMD_CG_SUPPORT_GFX_CGLS */
5496 	if (data & RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK)
5497 		*flags |= AMD_CG_SUPPORT_GFX_CGLS;
5498 
5499 	/* AMD_CG_SUPPORT_GFX_CGTS */
5500 	data = RREG32(mmCGTS_SM_CTRL_REG);
5501 	if (!(data & CGTS_SM_CTRL_REG__OVERRIDE_MASK))
5502 		*flags |= AMD_CG_SUPPORT_GFX_CGTS;
5503 
5504 	/* AMD_CG_SUPPORT_GFX_CGTS_LS */
5505 	if (!(data & CGTS_SM_CTRL_REG__LS_OVERRIDE_MASK))
5506 		*flags |= AMD_CG_SUPPORT_GFX_CGTS_LS;
5507 
5508 	/* AMD_CG_SUPPORT_GFX_RLC_LS */
5509 	data = RREG32(mmRLC_MEM_SLP_CNTL);
5510 	if (data & RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK)
5511 		*flags |= AMD_CG_SUPPORT_GFX_RLC_LS | AMD_CG_SUPPORT_GFX_MGLS;
5512 
5513 	/* AMD_CG_SUPPORT_GFX_CP_LS */
5514 	data = RREG32(mmCP_MEM_SLP_CNTL);
5515 	if (data & CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK)
5516 		*flags |= AMD_CG_SUPPORT_GFX_CP_LS | AMD_CG_SUPPORT_GFX_MGLS;
5517 }
5518 
5519 static void gfx_v8_0_send_serdes_cmd(struct amdgpu_device *adev,
5520 				     uint32_t reg_addr, uint32_t cmd)
5521 {
5522 	uint32_t data;
5523 
5524 	gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
5525 
5526 	WREG32(mmRLC_SERDES_WR_CU_MASTER_MASK, 0xffffffff);
5527 	WREG32(mmRLC_SERDES_WR_NONCU_MASTER_MASK, 0xffffffff);
5528 
5529 	data = RREG32(mmRLC_SERDES_WR_CTRL);
5530 	if (adev->asic_type == CHIP_STONEY)
5531 		data &= ~(RLC_SERDES_WR_CTRL__WRITE_COMMAND_MASK |
5532 			  RLC_SERDES_WR_CTRL__READ_COMMAND_MASK |
5533 			  RLC_SERDES_WR_CTRL__P1_SELECT_MASK |
5534 			  RLC_SERDES_WR_CTRL__P2_SELECT_MASK |
5535 			  RLC_SERDES_WR_CTRL__RDDATA_RESET_MASK |
5536 			  RLC_SERDES_WR_CTRL__POWER_DOWN_MASK |
5537 			  RLC_SERDES_WR_CTRL__POWER_UP_MASK |
5538 			  RLC_SERDES_WR_CTRL__SHORT_FORMAT_MASK |
5539 			  RLC_SERDES_WR_CTRL__SRBM_OVERRIDE_MASK);
5540 	else
5541 		data &= ~(RLC_SERDES_WR_CTRL__WRITE_COMMAND_MASK |
5542 			  RLC_SERDES_WR_CTRL__READ_COMMAND_MASK |
5543 			  RLC_SERDES_WR_CTRL__P1_SELECT_MASK |
5544 			  RLC_SERDES_WR_CTRL__P2_SELECT_MASK |
5545 			  RLC_SERDES_WR_CTRL__RDDATA_RESET_MASK |
5546 			  RLC_SERDES_WR_CTRL__POWER_DOWN_MASK |
5547 			  RLC_SERDES_WR_CTRL__POWER_UP_MASK |
5548 			  RLC_SERDES_WR_CTRL__SHORT_FORMAT_MASK |
5549 			  RLC_SERDES_WR_CTRL__BPM_DATA_MASK |
5550 			  RLC_SERDES_WR_CTRL__REG_ADDR_MASK |
5551 			  RLC_SERDES_WR_CTRL__SRBM_OVERRIDE_MASK);
5552 	data |= (RLC_SERDES_WR_CTRL__RSVD_BPM_ADDR_MASK |
5553 		 (cmd << RLC_SERDES_WR_CTRL__BPM_DATA__SHIFT) |
5554 		 (reg_addr << RLC_SERDES_WR_CTRL__REG_ADDR__SHIFT) |
5555 		 (0xff << RLC_SERDES_WR_CTRL__BPM_ADDR__SHIFT));
5556 
5557 	WREG32(mmRLC_SERDES_WR_CTRL, data);
5558 }
5559 
5560 #define MSG_ENTER_RLC_SAFE_MODE     1
5561 #define MSG_EXIT_RLC_SAFE_MODE      0
5562 #define RLC_GPR_REG2__REQ_MASK 0x00000001
5563 #define RLC_GPR_REG2__REQ__SHIFT 0
5564 #define RLC_GPR_REG2__MESSAGE__SHIFT 0x00000001
5565 #define RLC_GPR_REG2__MESSAGE_MASK 0x0000001e
5566 
5567 static bool gfx_v8_0_is_rlc_enabled(struct amdgpu_device *adev)
5568 {
5569 	uint32_t rlc_setting;
5570 
5571 	rlc_setting = RREG32(mmRLC_CNTL);
5572 	if (!(rlc_setting & RLC_CNTL__RLC_ENABLE_F32_MASK))
5573 		return false;
5574 
5575 	return true;
5576 }
5577 
5578 static void gfx_v8_0_set_safe_mode(struct amdgpu_device *adev)
5579 {
5580 	uint32_t data;
5581 	unsigned i;
5582 	data = RREG32(mmRLC_CNTL);
5583 	data |= RLC_SAFE_MODE__CMD_MASK;
5584 	data &= ~RLC_SAFE_MODE__MESSAGE_MASK;
5585 	data |= (1 << RLC_SAFE_MODE__MESSAGE__SHIFT);
5586 	WREG32(mmRLC_SAFE_MODE, data);
5587 
5588 	/* wait for RLC_SAFE_MODE */
5589 	for (i = 0; i < adev->usec_timeout; i++) {
5590 		if ((RREG32(mmRLC_GPM_STAT) &
5591 		     (RLC_GPM_STAT__GFX_CLOCK_STATUS_MASK |
5592 		      RLC_GPM_STAT__GFX_POWER_STATUS_MASK)) ==
5593 		    (RLC_GPM_STAT__GFX_CLOCK_STATUS_MASK |
5594 		     RLC_GPM_STAT__GFX_POWER_STATUS_MASK))
5595 			break;
5596 		udelay(1);
5597 	}
5598 	for (i = 0; i < adev->usec_timeout; i++) {
5599 		if (!REG_GET_FIELD(RREG32(mmRLC_SAFE_MODE), RLC_SAFE_MODE, CMD))
5600 			break;
5601 		udelay(1);
5602 	}
5603 }
5604 
5605 static void gfx_v8_0_unset_safe_mode(struct amdgpu_device *adev)
5606 {
5607 	uint32_t data;
5608 	unsigned i;
5609 
5610 	data = RREG32(mmRLC_CNTL);
5611 	data |= RLC_SAFE_MODE__CMD_MASK;
5612 	data &= ~RLC_SAFE_MODE__MESSAGE_MASK;
5613 	WREG32(mmRLC_SAFE_MODE, data);
5614 
5615 	for (i = 0; i < adev->usec_timeout; i++) {
5616 		if (!REG_GET_FIELD(RREG32(mmRLC_SAFE_MODE), RLC_SAFE_MODE, CMD))
5617 			break;
5618 		udelay(1);
5619 	}
5620 }
5621 
5622 static void gfx_v8_0_update_spm_vmid(struct amdgpu_device *adev, unsigned vmid)
5623 {
5624 	u32 data;
5625 
5626 	if (amdgpu_sriov_is_pp_one_vf(adev))
5627 		data = RREG32_NO_KIQ(mmRLC_SPM_VMID);
5628 	else
5629 		data = RREG32(mmRLC_SPM_VMID);
5630 
5631 	data &= ~RLC_SPM_VMID__RLC_SPM_VMID_MASK;
5632 	data |= (vmid & RLC_SPM_VMID__RLC_SPM_VMID_MASK) << RLC_SPM_VMID__RLC_SPM_VMID__SHIFT;
5633 
5634 	if (amdgpu_sriov_is_pp_one_vf(adev))
5635 		WREG32_NO_KIQ(mmRLC_SPM_VMID, data);
5636 	else
5637 		WREG32(mmRLC_SPM_VMID, data);
5638 }
5639 
5640 static const struct amdgpu_rlc_funcs iceland_rlc_funcs = {
5641 	.is_rlc_enabled = gfx_v8_0_is_rlc_enabled,
5642 	.set_safe_mode = gfx_v8_0_set_safe_mode,
5643 	.unset_safe_mode = gfx_v8_0_unset_safe_mode,
5644 	.init = gfx_v8_0_rlc_init,
5645 	.get_csb_size = gfx_v8_0_get_csb_size,
5646 	.get_csb_buffer = gfx_v8_0_get_csb_buffer,
5647 	.get_cp_table_num = gfx_v8_0_cp_jump_table_num,
5648 	.resume = gfx_v8_0_rlc_resume,
5649 	.stop = gfx_v8_0_rlc_stop,
5650 	.reset = gfx_v8_0_rlc_reset,
5651 	.start = gfx_v8_0_rlc_start,
5652 	.update_spm_vmid = gfx_v8_0_update_spm_vmid
5653 };
5654 
5655 static void gfx_v8_0_update_medium_grain_clock_gating(struct amdgpu_device *adev,
5656 						      bool enable)
5657 {
5658 	uint32_t temp, data;
5659 
5660 	amdgpu_gfx_rlc_enter_safe_mode(adev);
5661 
5662 	/* It is disabled by HW by default */
5663 	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGCG)) {
5664 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS) {
5665 			if (adev->cg_flags & AMD_CG_SUPPORT_GFX_RLC_LS)
5666 				/* 1 - RLC memory Light sleep */
5667 				WREG32_FIELD(RLC_MEM_SLP_CNTL, RLC_MEM_LS_EN, 1);
5668 
5669 			if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CP_LS)
5670 				WREG32_FIELD(CP_MEM_SLP_CNTL, CP_MEM_LS_EN, 1);
5671 		}
5672 
5673 		/* 3 - RLC_CGTT_MGCG_OVERRIDE */
5674 		temp = data = RREG32(mmRLC_CGTT_MGCG_OVERRIDE);
5675 		if (adev->flags & AMD_IS_APU)
5676 			data &= ~(RLC_CGTT_MGCG_OVERRIDE__CPF_MASK |
5677 				  RLC_CGTT_MGCG_OVERRIDE__RLC_MASK |
5678 				  RLC_CGTT_MGCG_OVERRIDE__MGCG_MASK);
5679 		else
5680 			data &= ~(RLC_CGTT_MGCG_OVERRIDE__CPF_MASK |
5681 				  RLC_CGTT_MGCG_OVERRIDE__RLC_MASK |
5682 				  RLC_CGTT_MGCG_OVERRIDE__MGCG_MASK |
5683 				  RLC_CGTT_MGCG_OVERRIDE__GRBM_MASK);
5684 
5685 		if (temp != data)
5686 			WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data);
5687 
5688 		/* 4 - wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */
5689 		gfx_v8_0_wait_for_rlc_serdes(adev);
5690 
5691 		/* 5 - clear mgcg override */
5692 		gfx_v8_0_send_serdes_cmd(adev, BPM_REG_MGCG_OVERRIDE, CLE_BPM_SERDES_CMD);
5693 
5694 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGTS) {
5695 			/* 6 - Enable CGTS(Tree Shade) MGCG /MGLS */
5696 			temp = data = RREG32(mmCGTS_SM_CTRL_REG);
5697 			data &= ~(CGTS_SM_CTRL_REG__SM_MODE_MASK);
5698 			data |= (0x2 << CGTS_SM_CTRL_REG__SM_MODE__SHIFT);
5699 			data |= CGTS_SM_CTRL_REG__SM_MODE_ENABLE_MASK;
5700 			data &= ~CGTS_SM_CTRL_REG__OVERRIDE_MASK;
5701 			if ((adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS) &&
5702 			    (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGTS_LS))
5703 				data &= ~CGTS_SM_CTRL_REG__LS_OVERRIDE_MASK;
5704 			data |= CGTS_SM_CTRL_REG__ON_MONITOR_ADD_EN_MASK;
5705 			data |= (0x96 << CGTS_SM_CTRL_REG__ON_MONITOR_ADD__SHIFT);
5706 			if (temp != data)
5707 				WREG32(mmCGTS_SM_CTRL_REG, data);
5708 		}
5709 		udelay(50);
5710 
5711 		/* 7 - wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */
5712 		gfx_v8_0_wait_for_rlc_serdes(adev);
5713 	} else {
5714 		/* 1 - MGCG_OVERRIDE[0] for CP and MGCG_OVERRIDE[1] for RLC */
5715 		temp = data = RREG32(mmRLC_CGTT_MGCG_OVERRIDE);
5716 		data |= (RLC_CGTT_MGCG_OVERRIDE__CPF_MASK |
5717 				RLC_CGTT_MGCG_OVERRIDE__RLC_MASK |
5718 				RLC_CGTT_MGCG_OVERRIDE__MGCG_MASK |
5719 				RLC_CGTT_MGCG_OVERRIDE__GRBM_MASK);
5720 		if (temp != data)
5721 			WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data);
5722 
5723 		/* 2 - disable MGLS in RLC */
5724 		data = RREG32(mmRLC_MEM_SLP_CNTL);
5725 		if (data & RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK) {
5726 			data &= ~RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK;
5727 			WREG32(mmRLC_MEM_SLP_CNTL, data);
5728 		}
5729 
5730 		/* 3 - disable MGLS in CP */
5731 		data = RREG32(mmCP_MEM_SLP_CNTL);
5732 		if (data & CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK) {
5733 			data &= ~CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK;
5734 			WREG32(mmCP_MEM_SLP_CNTL, data);
5735 		}
5736 
5737 		/* 4 - Disable CGTS(Tree Shade) MGCG and MGLS */
5738 		temp = data = RREG32(mmCGTS_SM_CTRL_REG);
5739 		data |= (CGTS_SM_CTRL_REG__OVERRIDE_MASK |
5740 				CGTS_SM_CTRL_REG__LS_OVERRIDE_MASK);
5741 		if (temp != data)
5742 			WREG32(mmCGTS_SM_CTRL_REG, data);
5743 
5744 		/* 5 - wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */
5745 		gfx_v8_0_wait_for_rlc_serdes(adev);
5746 
5747 		/* 6 - set mgcg override */
5748 		gfx_v8_0_send_serdes_cmd(adev, BPM_REG_MGCG_OVERRIDE, SET_BPM_SERDES_CMD);
5749 
5750 		udelay(50);
5751 
5752 		/* 7- wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */
5753 		gfx_v8_0_wait_for_rlc_serdes(adev);
5754 	}
5755 
5756 	amdgpu_gfx_rlc_exit_safe_mode(adev);
5757 }
5758 
5759 static void gfx_v8_0_update_coarse_grain_clock_gating(struct amdgpu_device *adev,
5760 						      bool enable)
5761 {
5762 	uint32_t temp, temp1, data, data1;
5763 
5764 	temp = data = RREG32(mmRLC_CGCG_CGLS_CTRL);
5765 
5766 	amdgpu_gfx_rlc_enter_safe_mode(adev);
5767 
5768 	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGCG)) {
5769 		temp1 = data1 =	RREG32(mmRLC_CGTT_MGCG_OVERRIDE);
5770 		data1 &= ~RLC_CGTT_MGCG_OVERRIDE__CGCG_MASK;
5771 		if (temp1 != data1)
5772 			WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data1);
5773 
5774 		/* : wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */
5775 		gfx_v8_0_wait_for_rlc_serdes(adev);
5776 
5777 		/* 2 - clear cgcg override */
5778 		gfx_v8_0_send_serdes_cmd(adev, BPM_REG_CGCG_OVERRIDE, CLE_BPM_SERDES_CMD);
5779 
5780 		/* wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */
5781 		gfx_v8_0_wait_for_rlc_serdes(adev);
5782 
5783 		/* 3 - write cmd to set CGLS */
5784 		gfx_v8_0_send_serdes_cmd(adev, BPM_REG_CGLS_EN, SET_BPM_SERDES_CMD);
5785 
5786 		/* 4 - enable cgcg */
5787 		data |= RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK;
5788 
5789 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGLS) {
5790 			/* enable cgls*/
5791 			data |= RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK;
5792 
5793 			temp1 = data1 =	RREG32(mmRLC_CGTT_MGCG_OVERRIDE);
5794 			data1 &= ~RLC_CGTT_MGCG_OVERRIDE__CGLS_MASK;
5795 
5796 			if (temp1 != data1)
5797 				WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data1);
5798 		} else {
5799 			data &= ~RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK;
5800 		}
5801 
5802 		if (temp != data)
5803 			WREG32(mmRLC_CGCG_CGLS_CTRL, data);
5804 
5805 		/* 5 enable cntx_empty_int_enable/cntx_busy_int_enable/
5806 		 * Cmp_busy/GFX_Idle interrupts
5807 		 */
5808 		gfx_v8_0_enable_gui_idle_interrupt(adev, true);
5809 	} else {
5810 		/* disable cntx_empty_int_enable & GFX Idle interrupt */
5811 		gfx_v8_0_enable_gui_idle_interrupt(adev, false);
5812 
5813 		/* TEST CGCG */
5814 		temp1 = data1 =	RREG32(mmRLC_CGTT_MGCG_OVERRIDE);
5815 		data1 |= (RLC_CGTT_MGCG_OVERRIDE__CGCG_MASK |
5816 				RLC_CGTT_MGCG_OVERRIDE__CGLS_MASK);
5817 		if (temp1 != data1)
5818 			WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data1);
5819 
5820 		/* read gfx register to wake up cgcg */
5821 		RREG32(mmCB_CGTT_SCLK_CTRL);
5822 		RREG32(mmCB_CGTT_SCLK_CTRL);
5823 		RREG32(mmCB_CGTT_SCLK_CTRL);
5824 		RREG32(mmCB_CGTT_SCLK_CTRL);
5825 
5826 		/* wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */
5827 		gfx_v8_0_wait_for_rlc_serdes(adev);
5828 
5829 		/* write cmd to Set CGCG Overrride */
5830 		gfx_v8_0_send_serdes_cmd(adev, BPM_REG_CGCG_OVERRIDE, SET_BPM_SERDES_CMD);
5831 
5832 		/* wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */
5833 		gfx_v8_0_wait_for_rlc_serdes(adev);
5834 
5835 		/* write cmd to Clear CGLS */
5836 		gfx_v8_0_send_serdes_cmd(adev, BPM_REG_CGLS_EN, CLE_BPM_SERDES_CMD);
5837 
5838 		/* disable cgcg, cgls should be disabled too. */
5839 		data &= ~(RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK |
5840 			  RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK);
5841 		if (temp != data)
5842 			WREG32(mmRLC_CGCG_CGLS_CTRL, data);
5843 		/* enable interrupts again for PG */
5844 		gfx_v8_0_enable_gui_idle_interrupt(adev, true);
5845 	}
5846 
5847 	gfx_v8_0_wait_for_rlc_serdes(adev);
5848 
5849 	amdgpu_gfx_rlc_exit_safe_mode(adev);
5850 }
5851 static int gfx_v8_0_update_gfx_clock_gating(struct amdgpu_device *adev,
5852 					    bool enable)
5853 {
5854 	if (enable) {
5855 		/* CGCG/CGLS should be enabled after MGCG/MGLS/TS(CG/LS)
5856 		 * ===  MGCG + MGLS + TS(CG/LS) ===
5857 		 */
5858 		gfx_v8_0_update_medium_grain_clock_gating(adev, enable);
5859 		gfx_v8_0_update_coarse_grain_clock_gating(adev, enable);
5860 	} else {
5861 		/* CGCG/CGLS should be disabled before MGCG/MGLS/TS(CG/LS)
5862 		 * ===  CGCG + CGLS ===
5863 		 */
5864 		gfx_v8_0_update_coarse_grain_clock_gating(adev, enable);
5865 		gfx_v8_0_update_medium_grain_clock_gating(adev, enable);
5866 	}
5867 	return 0;
5868 }
5869 
5870 static int gfx_v8_0_tonga_update_gfx_clock_gating(struct amdgpu_device *adev,
5871 					  enum amd_clockgating_state state)
5872 {
5873 	uint32_t msg_id, pp_state = 0;
5874 	uint32_t pp_support_state = 0;
5875 
5876 	if (adev->cg_flags & (AMD_CG_SUPPORT_GFX_CGCG | AMD_CG_SUPPORT_GFX_CGLS)) {
5877 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGLS) {
5878 			pp_support_state = PP_STATE_SUPPORT_LS;
5879 			pp_state = PP_STATE_LS;
5880 		}
5881 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGCG) {
5882 			pp_support_state |= PP_STATE_SUPPORT_CG;
5883 			pp_state |= PP_STATE_CG;
5884 		}
5885 		if (state == AMD_CG_STATE_UNGATE)
5886 			pp_state = 0;
5887 
5888 		msg_id = PP_CG_MSG_ID(PP_GROUP_GFX,
5889 				PP_BLOCK_GFX_CG,
5890 				pp_support_state,
5891 				pp_state);
5892 		amdgpu_dpm_set_clockgating_by_smu(adev, msg_id);
5893 	}
5894 
5895 	if (adev->cg_flags & (AMD_CG_SUPPORT_GFX_MGCG | AMD_CG_SUPPORT_GFX_MGLS)) {
5896 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS) {
5897 			pp_support_state = PP_STATE_SUPPORT_LS;
5898 			pp_state = PP_STATE_LS;
5899 		}
5900 
5901 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGCG) {
5902 			pp_support_state |= PP_STATE_SUPPORT_CG;
5903 			pp_state |= PP_STATE_CG;
5904 		}
5905 
5906 		if (state == AMD_CG_STATE_UNGATE)
5907 			pp_state = 0;
5908 
5909 		msg_id = PP_CG_MSG_ID(PP_GROUP_GFX,
5910 				PP_BLOCK_GFX_MG,
5911 				pp_support_state,
5912 				pp_state);
5913 		amdgpu_dpm_set_clockgating_by_smu(adev, msg_id);
5914 	}
5915 
5916 	return 0;
5917 }
5918 
5919 static int gfx_v8_0_polaris_update_gfx_clock_gating(struct amdgpu_device *adev,
5920 					  enum amd_clockgating_state state)
5921 {
5922 
5923 	uint32_t msg_id, pp_state = 0;
5924 	uint32_t pp_support_state = 0;
5925 
5926 	if (adev->cg_flags & (AMD_CG_SUPPORT_GFX_CGCG | AMD_CG_SUPPORT_GFX_CGLS)) {
5927 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGLS) {
5928 			pp_support_state = PP_STATE_SUPPORT_LS;
5929 			pp_state = PP_STATE_LS;
5930 		}
5931 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGCG) {
5932 			pp_support_state |= PP_STATE_SUPPORT_CG;
5933 			pp_state |= PP_STATE_CG;
5934 		}
5935 		if (state == AMD_CG_STATE_UNGATE)
5936 			pp_state = 0;
5937 
5938 		msg_id = PP_CG_MSG_ID(PP_GROUP_GFX,
5939 				PP_BLOCK_GFX_CG,
5940 				pp_support_state,
5941 				pp_state);
5942 		amdgpu_dpm_set_clockgating_by_smu(adev, msg_id);
5943 	}
5944 
5945 	if (adev->cg_flags & (AMD_CG_SUPPORT_GFX_3D_CGCG | AMD_CG_SUPPORT_GFX_3D_CGLS)) {
5946 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_3D_CGLS) {
5947 			pp_support_state = PP_STATE_SUPPORT_LS;
5948 			pp_state = PP_STATE_LS;
5949 		}
5950 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_3D_CGCG) {
5951 			pp_support_state |= PP_STATE_SUPPORT_CG;
5952 			pp_state |= PP_STATE_CG;
5953 		}
5954 		if (state == AMD_CG_STATE_UNGATE)
5955 			pp_state = 0;
5956 
5957 		msg_id = PP_CG_MSG_ID(PP_GROUP_GFX,
5958 				PP_BLOCK_GFX_3D,
5959 				pp_support_state,
5960 				pp_state);
5961 		amdgpu_dpm_set_clockgating_by_smu(adev, msg_id);
5962 	}
5963 
5964 	if (adev->cg_flags & (AMD_CG_SUPPORT_GFX_MGCG | AMD_CG_SUPPORT_GFX_MGLS)) {
5965 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS) {
5966 			pp_support_state = PP_STATE_SUPPORT_LS;
5967 			pp_state = PP_STATE_LS;
5968 		}
5969 
5970 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGCG) {
5971 			pp_support_state |= PP_STATE_SUPPORT_CG;
5972 			pp_state |= PP_STATE_CG;
5973 		}
5974 
5975 		if (state == AMD_CG_STATE_UNGATE)
5976 			pp_state = 0;
5977 
5978 		msg_id = PP_CG_MSG_ID(PP_GROUP_GFX,
5979 				PP_BLOCK_GFX_MG,
5980 				pp_support_state,
5981 				pp_state);
5982 		amdgpu_dpm_set_clockgating_by_smu(adev, msg_id);
5983 	}
5984 
5985 	if (adev->cg_flags & AMD_CG_SUPPORT_GFX_RLC_LS) {
5986 		pp_support_state = PP_STATE_SUPPORT_LS;
5987 
5988 		if (state == AMD_CG_STATE_UNGATE)
5989 			pp_state = 0;
5990 		else
5991 			pp_state = PP_STATE_LS;
5992 
5993 		msg_id = PP_CG_MSG_ID(PP_GROUP_GFX,
5994 				PP_BLOCK_GFX_RLC,
5995 				pp_support_state,
5996 				pp_state);
5997 		amdgpu_dpm_set_clockgating_by_smu(adev, msg_id);
5998 	}
5999 
6000 	if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CP_LS) {
6001 		pp_support_state = PP_STATE_SUPPORT_LS;
6002 
6003 		if (state == AMD_CG_STATE_UNGATE)
6004 			pp_state = 0;
6005 		else
6006 			pp_state = PP_STATE_LS;
6007 		msg_id = PP_CG_MSG_ID(PP_GROUP_GFX,
6008 			PP_BLOCK_GFX_CP,
6009 			pp_support_state,
6010 			pp_state);
6011 		amdgpu_dpm_set_clockgating_by_smu(adev, msg_id);
6012 	}
6013 
6014 	return 0;
6015 }
6016 
6017 static int gfx_v8_0_set_clockgating_state(void *handle,
6018 					  enum amd_clockgating_state state)
6019 {
6020 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
6021 
6022 	if (amdgpu_sriov_vf(adev))
6023 		return 0;
6024 
6025 	switch (adev->asic_type) {
6026 	case CHIP_FIJI:
6027 	case CHIP_CARRIZO:
6028 	case CHIP_STONEY:
6029 		gfx_v8_0_update_gfx_clock_gating(adev,
6030 						 state == AMD_CG_STATE_GATE);
6031 		break;
6032 	case CHIP_TONGA:
6033 		gfx_v8_0_tonga_update_gfx_clock_gating(adev, state);
6034 		break;
6035 	case CHIP_POLARIS10:
6036 	case CHIP_POLARIS11:
6037 	case CHIP_POLARIS12:
6038 	case CHIP_VEGAM:
6039 		gfx_v8_0_polaris_update_gfx_clock_gating(adev, state);
6040 		break;
6041 	default:
6042 		break;
6043 	}
6044 	return 0;
6045 }
6046 
6047 static u64 gfx_v8_0_ring_get_rptr(struct amdgpu_ring *ring)
6048 {
6049 	return ring->adev->wb.wb[ring->rptr_offs];
6050 }
6051 
6052 static u64 gfx_v8_0_ring_get_wptr_gfx(struct amdgpu_ring *ring)
6053 {
6054 	struct amdgpu_device *adev = ring->adev;
6055 
6056 	if (ring->use_doorbell)
6057 		/* XXX check if swapping is necessary on BE */
6058 		return ring->adev->wb.wb[ring->wptr_offs];
6059 	else
6060 		return RREG32(mmCP_RB0_WPTR);
6061 }
6062 
6063 static void gfx_v8_0_ring_set_wptr_gfx(struct amdgpu_ring *ring)
6064 {
6065 	struct amdgpu_device *adev = ring->adev;
6066 
6067 	if (ring->use_doorbell) {
6068 		/* XXX check if swapping is necessary on BE */
6069 		adev->wb.wb[ring->wptr_offs] = lower_32_bits(ring->wptr);
6070 		WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
6071 	} else {
6072 		WREG32(mmCP_RB0_WPTR, lower_32_bits(ring->wptr));
6073 		(void)RREG32(mmCP_RB0_WPTR);
6074 	}
6075 }
6076 
6077 static void gfx_v8_0_ring_emit_hdp_flush(struct amdgpu_ring *ring)
6078 {
6079 	u32 ref_and_mask, reg_mem_engine;
6080 
6081 	if ((ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE) ||
6082 	    (ring->funcs->type == AMDGPU_RING_TYPE_KIQ)) {
6083 		switch (ring->me) {
6084 		case 1:
6085 			ref_and_mask = GPU_HDP_FLUSH_DONE__CP2_MASK << ring->pipe;
6086 			break;
6087 		case 2:
6088 			ref_and_mask = GPU_HDP_FLUSH_DONE__CP6_MASK << ring->pipe;
6089 			break;
6090 		default:
6091 			return;
6092 		}
6093 		reg_mem_engine = 0;
6094 	} else {
6095 		ref_and_mask = GPU_HDP_FLUSH_DONE__CP0_MASK;
6096 		reg_mem_engine = WAIT_REG_MEM_ENGINE(1); /* pfp */
6097 	}
6098 
6099 	amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
6100 	amdgpu_ring_write(ring, (WAIT_REG_MEM_OPERATION(1) | /* write, wait, write */
6101 				 WAIT_REG_MEM_FUNCTION(3) |  /* == */
6102 				 reg_mem_engine));
6103 	amdgpu_ring_write(ring, mmGPU_HDP_FLUSH_REQ);
6104 	amdgpu_ring_write(ring, mmGPU_HDP_FLUSH_DONE);
6105 	amdgpu_ring_write(ring, ref_and_mask);
6106 	amdgpu_ring_write(ring, ref_and_mask);
6107 	amdgpu_ring_write(ring, 0x20); /* poll interval */
6108 }
6109 
6110 static void gfx_v8_0_ring_emit_vgt_flush(struct amdgpu_ring *ring)
6111 {
6112 	amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE, 0));
6113 	amdgpu_ring_write(ring, EVENT_TYPE(VS_PARTIAL_FLUSH) |
6114 		EVENT_INDEX(4));
6115 
6116 	amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE, 0));
6117 	amdgpu_ring_write(ring, EVENT_TYPE(VGT_FLUSH) |
6118 		EVENT_INDEX(0));
6119 }
6120 
6121 static void gfx_v8_0_ring_emit_ib_gfx(struct amdgpu_ring *ring,
6122 					struct amdgpu_job *job,
6123 					struct amdgpu_ib *ib,
6124 					uint32_t flags)
6125 {
6126 	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
6127 	u32 header, control = 0;
6128 
6129 	if (ib->flags & AMDGPU_IB_FLAG_CE)
6130 		header = PACKET3(PACKET3_INDIRECT_BUFFER_CONST, 2);
6131 	else
6132 		header = PACKET3(PACKET3_INDIRECT_BUFFER, 2);
6133 
6134 	control |= ib->length_dw | (vmid << 24);
6135 
6136 	if (amdgpu_sriov_vf(ring->adev) && (ib->flags & AMDGPU_IB_FLAG_PREEMPT)) {
6137 		control |= INDIRECT_BUFFER_PRE_ENB(1);
6138 
6139 		if (!(ib->flags & AMDGPU_IB_FLAG_CE) && vmid)
6140 			gfx_v8_0_ring_emit_de_meta(ring);
6141 	}
6142 
6143 	amdgpu_ring_write(ring, header);
6144 	amdgpu_ring_write(ring,
6145 #ifdef __BIG_ENDIAN
6146 			  (2 << 0) |
6147 #endif
6148 			  (ib->gpu_addr & 0xFFFFFFFC));
6149 	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr) & 0xFFFF);
6150 	amdgpu_ring_write(ring, control);
6151 }
6152 
6153 static void gfx_v8_0_ring_emit_ib_compute(struct amdgpu_ring *ring,
6154 					  struct amdgpu_job *job,
6155 					  struct amdgpu_ib *ib,
6156 					  uint32_t flags)
6157 {
6158 	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
6159 	u32 control = INDIRECT_BUFFER_VALID | ib->length_dw | (vmid << 24);
6160 
6161 	/* Currently, there is a high possibility to get wave ID mismatch
6162 	 * between ME and GDS, leading to a hw deadlock, because ME generates
6163 	 * different wave IDs than the GDS expects. This situation happens
6164 	 * randomly when at least 5 compute pipes use GDS ordered append.
6165 	 * The wave IDs generated by ME are also wrong after suspend/resume.
6166 	 * Those are probably bugs somewhere else in the kernel driver.
6167 	 *
6168 	 * Writing GDS_COMPUTE_MAX_WAVE_ID resets wave ID counters in ME and
6169 	 * GDS to 0 for this ring (me/pipe).
6170 	 */
6171 	if (ib->flags & AMDGPU_IB_FLAG_RESET_GDS_MAX_WAVE_ID) {
6172 		amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1));
6173 		amdgpu_ring_write(ring, mmGDS_COMPUTE_MAX_WAVE_ID - PACKET3_SET_CONFIG_REG_START);
6174 		amdgpu_ring_write(ring, ring->adev->gds.gds_compute_max_wave_id);
6175 	}
6176 
6177 	amdgpu_ring_write(ring, PACKET3(PACKET3_INDIRECT_BUFFER, 2));
6178 	amdgpu_ring_write(ring,
6179 #ifdef __BIG_ENDIAN
6180 				(2 << 0) |
6181 #endif
6182 				(ib->gpu_addr & 0xFFFFFFFC));
6183 	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr) & 0xFFFF);
6184 	amdgpu_ring_write(ring, control);
6185 }
6186 
6187 static void gfx_v8_0_ring_emit_fence_gfx(struct amdgpu_ring *ring, u64 addr,
6188 					 u64 seq, unsigned flags)
6189 {
6190 	bool write64bit = flags & AMDGPU_FENCE_FLAG_64BIT;
6191 	bool int_sel = flags & AMDGPU_FENCE_FLAG_INT;
6192 
6193 	/* Workaround for cache flush problems. First send a dummy EOP
6194 	 * event down the pipe with seq one below.
6195 	 */
6196 	amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE_EOP, 4));
6197 	amdgpu_ring_write(ring, (EOP_TCL1_ACTION_EN |
6198 				 EOP_TC_ACTION_EN |
6199 				 EOP_TC_WB_ACTION_EN |
6200 				 EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
6201 				 EVENT_INDEX(5)));
6202 	amdgpu_ring_write(ring, addr & 0xfffffffc);
6203 	amdgpu_ring_write(ring, (upper_32_bits(addr) & 0xffff) |
6204 				DATA_SEL(1) | INT_SEL(0));
6205 	amdgpu_ring_write(ring, lower_32_bits(seq - 1));
6206 	amdgpu_ring_write(ring, upper_32_bits(seq - 1));
6207 
6208 	/* Then send the real EOP event down the pipe:
6209 	 * EVENT_WRITE_EOP - flush caches, send int */
6210 	amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE_EOP, 4));
6211 	amdgpu_ring_write(ring, (EOP_TCL1_ACTION_EN |
6212 				 EOP_TC_ACTION_EN |
6213 				 EOP_TC_WB_ACTION_EN |
6214 				 EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
6215 				 EVENT_INDEX(5)));
6216 	amdgpu_ring_write(ring, addr & 0xfffffffc);
6217 	amdgpu_ring_write(ring, (upper_32_bits(addr) & 0xffff) |
6218 			  DATA_SEL(write64bit ? 2 : 1) | INT_SEL(int_sel ? 2 : 0));
6219 	amdgpu_ring_write(ring, lower_32_bits(seq));
6220 	amdgpu_ring_write(ring, upper_32_bits(seq));
6221 
6222 }
6223 
6224 static void gfx_v8_0_ring_emit_pipeline_sync(struct amdgpu_ring *ring)
6225 {
6226 	int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX);
6227 	uint32_t seq = ring->fence_drv.sync_seq;
6228 	uint64_t addr = ring->fence_drv.gpu_addr;
6229 
6230 	amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
6231 	amdgpu_ring_write(ring, (WAIT_REG_MEM_MEM_SPACE(1) | /* memory */
6232 				 WAIT_REG_MEM_FUNCTION(3) | /* equal */
6233 				 WAIT_REG_MEM_ENGINE(usepfp))); /* pfp or me */
6234 	amdgpu_ring_write(ring, addr & 0xfffffffc);
6235 	amdgpu_ring_write(ring, upper_32_bits(addr) & 0xffffffff);
6236 	amdgpu_ring_write(ring, seq);
6237 	amdgpu_ring_write(ring, 0xffffffff);
6238 	amdgpu_ring_write(ring, 4); /* poll interval */
6239 }
6240 
6241 static void gfx_v8_0_ring_emit_vm_flush(struct amdgpu_ring *ring,
6242 					unsigned vmid, uint64_t pd_addr)
6243 {
6244 	int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX);
6245 
6246 	amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr);
6247 
6248 	/* wait for the invalidate to complete */
6249 	amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
6250 	amdgpu_ring_write(ring, (WAIT_REG_MEM_OPERATION(0) | /* wait */
6251 				 WAIT_REG_MEM_FUNCTION(0) |  /* always */
6252 				 WAIT_REG_MEM_ENGINE(0))); /* me */
6253 	amdgpu_ring_write(ring, mmVM_INVALIDATE_REQUEST);
6254 	amdgpu_ring_write(ring, 0);
6255 	amdgpu_ring_write(ring, 0); /* ref */
6256 	amdgpu_ring_write(ring, 0); /* mask */
6257 	amdgpu_ring_write(ring, 0x20); /* poll interval */
6258 
6259 	/* compute doesn't have PFP */
6260 	if (usepfp) {
6261 		/* sync PFP to ME, otherwise we might get invalid PFP reads */
6262 		amdgpu_ring_write(ring, PACKET3(PACKET3_PFP_SYNC_ME, 0));
6263 		amdgpu_ring_write(ring, 0x0);
6264 	}
6265 }
6266 
6267 static u64 gfx_v8_0_ring_get_wptr_compute(struct amdgpu_ring *ring)
6268 {
6269 	return ring->adev->wb.wb[ring->wptr_offs];
6270 }
6271 
6272 static void gfx_v8_0_ring_set_wptr_compute(struct amdgpu_ring *ring)
6273 {
6274 	struct amdgpu_device *adev = ring->adev;
6275 
6276 	/* XXX check if swapping is necessary on BE */
6277 	adev->wb.wb[ring->wptr_offs] = lower_32_bits(ring->wptr);
6278 	WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
6279 }
6280 
6281 static void gfx_v8_0_ring_emit_fence_compute(struct amdgpu_ring *ring,
6282 					     u64 addr, u64 seq,
6283 					     unsigned flags)
6284 {
6285 	bool write64bit = flags & AMDGPU_FENCE_FLAG_64BIT;
6286 	bool int_sel = flags & AMDGPU_FENCE_FLAG_INT;
6287 
6288 	/* RELEASE_MEM - flush caches, send int */
6289 	amdgpu_ring_write(ring, PACKET3(PACKET3_RELEASE_MEM, 5));
6290 	amdgpu_ring_write(ring, (EOP_TCL1_ACTION_EN |
6291 				 EOP_TC_ACTION_EN |
6292 				 EOP_TC_WB_ACTION_EN |
6293 				 EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
6294 				 EVENT_INDEX(5)));
6295 	amdgpu_ring_write(ring, DATA_SEL(write64bit ? 2 : 1) | INT_SEL(int_sel ? 2 : 0));
6296 	amdgpu_ring_write(ring, addr & 0xfffffffc);
6297 	amdgpu_ring_write(ring, upper_32_bits(addr));
6298 	amdgpu_ring_write(ring, lower_32_bits(seq));
6299 	amdgpu_ring_write(ring, upper_32_bits(seq));
6300 }
6301 
6302 static void gfx_v8_0_ring_emit_fence_kiq(struct amdgpu_ring *ring, u64 addr,
6303 					 u64 seq, unsigned int flags)
6304 {
6305 	/* we only allocate 32bit for each seq wb address */
6306 	BUG_ON(flags & AMDGPU_FENCE_FLAG_64BIT);
6307 
6308 	/* write fence seq to the "addr" */
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(5) | WR_CONFIRM));
6312 	amdgpu_ring_write(ring, lower_32_bits(addr));
6313 	amdgpu_ring_write(ring, upper_32_bits(addr));
6314 	amdgpu_ring_write(ring, lower_32_bits(seq));
6315 
6316 	if (flags & AMDGPU_FENCE_FLAG_INT) {
6317 		/* set register to trigger INT */
6318 		amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
6319 		amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
6320 					 WRITE_DATA_DST_SEL(0) | WR_CONFIRM));
6321 		amdgpu_ring_write(ring, mmCPC_INT_STATUS);
6322 		amdgpu_ring_write(ring, 0);
6323 		amdgpu_ring_write(ring, 0x20000000); /* src_id is 178 */
6324 	}
6325 }
6326 
6327 static void gfx_v8_ring_emit_sb(struct amdgpu_ring *ring)
6328 {
6329 	amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
6330 	amdgpu_ring_write(ring, 0);
6331 }
6332 
6333 static void gfx_v8_ring_emit_cntxcntl(struct amdgpu_ring *ring, uint32_t flags)
6334 {
6335 	uint32_t dw2 = 0;
6336 
6337 	if (amdgpu_sriov_vf(ring->adev))
6338 		gfx_v8_0_ring_emit_ce_meta(ring);
6339 
6340 	dw2 |= 0x80000000; /* set load_enable otherwise this package is just NOPs */
6341 	if (flags & AMDGPU_HAVE_CTX_SWITCH) {
6342 		gfx_v8_0_ring_emit_vgt_flush(ring);
6343 		/* set load_global_config & load_global_uconfig */
6344 		dw2 |= 0x8001;
6345 		/* set load_cs_sh_regs */
6346 		dw2 |= 0x01000000;
6347 		/* set load_per_context_state & load_gfx_sh_regs for GFX */
6348 		dw2 |= 0x10002;
6349 
6350 		/* set load_ce_ram if preamble presented */
6351 		if (AMDGPU_PREAMBLE_IB_PRESENT & flags)
6352 			dw2 |= 0x10000000;
6353 	} else {
6354 		/* still load_ce_ram if this is the first time preamble presented
6355 		 * although there is no context switch happens.
6356 		 */
6357 		if (AMDGPU_PREAMBLE_IB_PRESENT_FIRST & flags)
6358 			dw2 |= 0x10000000;
6359 	}
6360 
6361 	amdgpu_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1));
6362 	amdgpu_ring_write(ring, dw2);
6363 	amdgpu_ring_write(ring, 0);
6364 }
6365 
6366 static unsigned gfx_v8_0_ring_emit_init_cond_exec(struct amdgpu_ring *ring)
6367 {
6368 	unsigned ret;
6369 
6370 	amdgpu_ring_write(ring, PACKET3(PACKET3_COND_EXEC, 3));
6371 	amdgpu_ring_write(ring, lower_32_bits(ring->cond_exe_gpu_addr));
6372 	amdgpu_ring_write(ring, upper_32_bits(ring->cond_exe_gpu_addr));
6373 	amdgpu_ring_write(ring, 0); /* discard following DWs if *cond_exec_gpu_addr==0 */
6374 	ret = ring->wptr & ring->buf_mask;
6375 	amdgpu_ring_write(ring, 0x55aa55aa); /* patch dummy value later */
6376 	return ret;
6377 }
6378 
6379 static void gfx_v8_0_ring_emit_patch_cond_exec(struct amdgpu_ring *ring, unsigned offset)
6380 {
6381 	unsigned cur;
6382 
6383 	BUG_ON(offset > ring->buf_mask);
6384 	BUG_ON(ring->ring[offset] != 0x55aa55aa);
6385 
6386 	cur = (ring->wptr & ring->buf_mask) - 1;
6387 	if (likely(cur > offset))
6388 		ring->ring[offset] = cur - offset;
6389 	else
6390 		ring->ring[offset] = (ring->ring_size >> 2) - offset + cur;
6391 }
6392 
6393 static void gfx_v8_0_ring_emit_rreg(struct amdgpu_ring *ring, uint32_t reg,
6394 				    uint32_t reg_val_offs)
6395 {
6396 	struct amdgpu_device *adev = ring->adev;
6397 
6398 	amdgpu_ring_write(ring, PACKET3(PACKET3_COPY_DATA, 4));
6399 	amdgpu_ring_write(ring, 0 |	/* src: register*/
6400 				(5 << 8) |	/* dst: memory */
6401 				(1 << 20));	/* write confirm */
6402 	amdgpu_ring_write(ring, reg);
6403 	amdgpu_ring_write(ring, 0);
6404 	amdgpu_ring_write(ring, lower_32_bits(adev->wb.gpu_addr +
6405 				reg_val_offs * 4));
6406 	amdgpu_ring_write(ring, upper_32_bits(adev->wb.gpu_addr +
6407 				reg_val_offs * 4));
6408 }
6409 
6410 static void gfx_v8_0_ring_emit_wreg(struct amdgpu_ring *ring, uint32_t reg,
6411 				  uint32_t val)
6412 {
6413 	uint32_t cmd;
6414 
6415 	switch (ring->funcs->type) {
6416 	case AMDGPU_RING_TYPE_GFX:
6417 		cmd = WRITE_DATA_ENGINE_SEL(1) | WR_CONFIRM;
6418 		break;
6419 	case AMDGPU_RING_TYPE_KIQ:
6420 		cmd = 1 << 16; /* no inc addr */
6421 		break;
6422 	default:
6423 		cmd = WR_CONFIRM;
6424 		break;
6425 	}
6426 
6427 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
6428 	amdgpu_ring_write(ring, cmd);
6429 	amdgpu_ring_write(ring, reg);
6430 	amdgpu_ring_write(ring, 0);
6431 	amdgpu_ring_write(ring, val);
6432 }
6433 
6434 static void gfx_v8_0_ring_soft_recovery(struct amdgpu_ring *ring, unsigned vmid)
6435 {
6436 	struct amdgpu_device *adev = ring->adev;
6437 	uint32_t value = 0;
6438 
6439 	value = REG_SET_FIELD(value, SQ_CMD, CMD, 0x03);
6440 	value = REG_SET_FIELD(value, SQ_CMD, MODE, 0x01);
6441 	value = REG_SET_FIELD(value, SQ_CMD, CHECK_VMID, 1);
6442 	value = REG_SET_FIELD(value, SQ_CMD, VM_ID, vmid);
6443 	WREG32(mmSQ_CMD, value);
6444 }
6445 
6446 static void gfx_v8_0_set_gfx_eop_interrupt_state(struct amdgpu_device *adev,
6447 						 enum amdgpu_interrupt_state state)
6448 {
6449 	WREG32_FIELD(CP_INT_CNTL_RING0, TIME_STAMP_INT_ENABLE,
6450 		     state == AMDGPU_IRQ_STATE_DISABLE ? 0 : 1);
6451 }
6452 
6453 static void gfx_v8_0_set_compute_eop_interrupt_state(struct amdgpu_device *adev,
6454 						     int me, int pipe,
6455 						     enum amdgpu_interrupt_state state)
6456 {
6457 	u32 mec_int_cntl, mec_int_cntl_reg;
6458 
6459 	/*
6460 	 * amdgpu controls only the first MEC. That's why this function only
6461 	 * handles the setting of interrupts for this specific MEC. All other
6462 	 * pipes' interrupts are set by amdkfd.
6463 	 */
6464 
6465 	if (me == 1) {
6466 		switch (pipe) {
6467 		case 0:
6468 			mec_int_cntl_reg = mmCP_ME1_PIPE0_INT_CNTL;
6469 			break;
6470 		case 1:
6471 			mec_int_cntl_reg = mmCP_ME1_PIPE1_INT_CNTL;
6472 			break;
6473 		case 2:
6474 			mec_int_cntl_reg = mmCP_ME1_PIPE2_INT_CNTL;
6475 			break;
6476 		case 3:
6477 			mec_int_cntl_reg = mmCP_ME1_PIPE3_INT_CNTL;
6478 			break;
6479 		default:
6480 			DRM_DEBUG("invalid pipe %d\n", pipe);
6481 			return;
6482 		}
6483 	} else {
6484 		DRM_DEBUG("invalid me %d\n", me);
6485 		return;
6486 	}
6487 
6488 	switch (state) {
6489 	case AMDGPU_IRQ_STATE_DISABLE:
6490 		mec_int_cntl = RREG32(mec_int_cntl_reg);
6491 		mec_int_cntl &= ~CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK;
6492 		WREG32(mec_int_cntl_reg, mec_int_cntl);
6493 		break;
6494 	case AMDGPU_IRQ_STATE_ENABLE:
6495 		mec_int_cntl = RREG32(mec_int_cntl_reg);
6496 		mec_int_cntl |= CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK;
6497 		WREG32(mec_int_cntl_reg, mec_int_cntl);
6498 		break;
6499 	default:
6500 		break;
6501 	}
6502 }
6503 
6504 static int gfx_v8_0_set_priv_reg_fault_state(struct amdgpu_device *adev,
6505 					     struct amdgpu_irq_src *source,
6506 					     unsigned type,
6507 					     enum amdgpu_interrupt_state state)
6508 {
6509 	WREG32_FIELD(CP_INT_CNTL_RING0, PRIV_REG_INT_ENABLE,
6510 		     state == AMDGPU_IRQ_STATE_DISABLE ? 0 : 1);
6511 
6512 	return 0;
6513 }
6514 
6515 static int gfx_v8_0_set_priv_inst_fault_state(struct amdgpu_device *adev,
6516 					      struct amdgpu_irq_src *source,
6517 					      unsigned type,
6518 					      enum amdgpu_interrupt_state state)
6519 {
6520 	WREG32_FIELD(CP_INT_CNTL_RING0, PRIV_INSTR_INT_ENABLE,
6521 		     state == AMDGPU_IRQ_STATE_DISABLE ? 0 : 1);
6522 
6523 	return 0;
6524 }
6525 
6526 static int gfx_v8_0_set_eop_interrupt_state(struct amdgpu_device *adev,
6527 					    struct amdgpu_irq_src *src,
6528 					    unsigned type,
6529 					    enum amdgpu_interrupt_state state)
6530 {
6531 	switch (type) {
6532 	case AMDGPU_CP_IRQ_GFX_ME0_PIPE0_EOP:
6533 		gfx_v8_0_set_gfx_eop_interrupt_state(adev, state);
6534 		break;
6535 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP:
6536 		gfx_v8_0_set_compute_eop_interrupt_state(adev, 1, 0, state);
6537 		break;
6538 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE1_EOP:
6539 		gfx_v8_0_set_compute_eop_interrupt_state(adev, 1, 1, state);
6540 		break;
6541 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE2_EOP:
6542 		gfx_v8_0_set_compute_eop_interrupt_state(adev, 1, 2, state);
6543 		break;
6544 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE3_EOP:
6545 		gfx_v8_0_set_compute_eop_interrupt_state(adev, 1, 3, state);
6546 		break;
6547 	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE0_EOP:
6548 		gfx_v8_0_set_compute_eop_interrupt_state(adev, 2, 0, state);
6549 		break;
6550 	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE1_EOP:
6551 		gfx_v8_0_set_compute_eop_interrupt_state(adev, 2, 1, state);
6552 		break;
6553 	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE2_EOP:
6554 		gfx_v8_0_set_compute_eop_interrupt_state(adev, 2, 2, state);
6555 		break;
6556 	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE3_EOP:
6557 		gfx_v8_0_set_compute_eop_interrupt_state(adev, 2, 3, state);
6558 		break;
6559 	default:
6560 		break;
6561 	}
6562 	return 0;
6563 }
6564 
6565 static int gfx_v8_0_set_cp_ecc_int_state(struct amdgpu_device *adev,
6566 					 struct amdgpu_irq_src *source,
6567 					 unsigned int type,
6568 					 enum amdgpu_interrupt_state state)
6569 {
6570 	int enable_flag;
6571 
6572 	switch (state) {
6573 	case AMDGPU_IRQ_STATE_DISABLE:
6574 		enable_flag = 0;
6575 		break;
6576 
6577 	case AMDGPU_IRQ_STATE_ENABLE:
6578 		enable_flag = 1;
6579 		break;
6580 
6581 	default:
6582 		return -EINVAL;
6583 	}
6584 
6585 	WREG32_FIELD(CP_INT_CNTL, CP_ECC_ERROR_INT_ENABLE, enable_flag);
6586 	WREG32_FIELD(CP_INT_CNTL_RING0, CP_ECC_ERROR_INT_ENABLE, enable_flag);
6587 	WREG32_FIELD(CP_INT_CNTL_RING1, CP_ECC_ERROR_INT_ENABLE, enable_flag);
6588 	WREG32_FIELD(CP_INT_CNTL_RING2, CP_ECC_ERROR_INT_ENABLE, enable_flag);
6589 	WREG32_FIELD(CPC_INT_CNTL, CP_ECC_ERROR_INT_ENABLE, enable_flag);
6590 	WREG32_FIELD(CP_ME1_PIPE0_INT_CNTL, CP_ECC_ERROR_INT_ENABLE,
6591 		     enable_flag);
6592 	WREG32_FIELD(CP_ME1_PIPE1_INT_CNTL, CP_ECC_ERROR_INT_ENABLE,
6593 		     enable_flag);
6594 	WREG32_FIELD(CP_ME1_PIPE2_INT_CNTL, CP_ECC_ERROR_INT_ENABLE,
6595 		     enable_flag);
6596 	WREG32_FIELD(CP_ME1_PIPE3_INT_CNTL, CP_ECC_ERROR_INT_ENABLE,
6597 		     enable_flag);
6598 	WREG32_FIELD(CP_ME2_PIPE0_INT_CNTL, CP_ECC_ERROR_INT_ENABLE,
6599 		     enable_flag);
6600 	WREG32_FIELD(CP_ME2_PIPE1_INT_CNTL, CP_ECC_ERROR_INT_ENABLE,
6601 		     enable_flag);
6602 	WREG32_FIELD(CP_ME2_PIPE2_INT_CNTL, CP_ECC_ERROR_INT_ENABLE,
6603 		     enable_flag);
6604 	WREG32_FIELD(CP_ME2_PIPE3_INT_CNTL, CP_ECC_ERROR_INT_ENABLE,
6605 		     enable_flag);
6606 
6607 	return 0;
6608 }
6609 
6610 static int gfx_v8_0_set_sq_int_state(struct amdgpu_device *adev,
6611 				     struct amdgpu_irq_src *source,
6612 				     unsigned int type,
6613 				     enum amdgpu_interrupt_state state)
6614 {
6615 	int enable_flag;
6616 
6617 	switch (state) {
6618 	case AMDGPU_IRQ_STATE_DISABLE:
6619 		enable_flag = 1;
6620 		break;
6621 
6622 	case AMDGPU_IRQ_STATE_ENABLE:
6623 		enable_flag = 0;
6624 		break;
6625 
6626 	default:
6627 		return -EINVAL;
6628 	}
6629 
6630 	WREG32_FIELD(SQ_INTERRUPT_MSG_CTRL, STALL,
6631 		     enable_flag);
6632 
6633 	return 0;
6634 }
6635 
6636 static int gfx_v8_0_eop_irq(struct amdgpu_device *adev,
6637 			    struct amdgpu_irq_src *source,
6638 			    struct amdgpu_iv_entry *entry)
6639 {
6640 	int i;
6641 	u8 me_id, pipe_id, queue_id;
6642 	struct amdgpu_ring *ring;
6643 
6644 	DRM_DEBUG("IH: CP EOP\n");
6645 	me_id = (entry->ring_id & 0x0c) >> 2;
6646 	pipe_id = (entry->ring_id & 0x03) >> 0;
6647 	queue_id = (entry->ring_id & 0x70) >> 4;
6648 
6649 	switch (me_id) {
6650 	case 0:
6651 		amdgpu_fence_process(&adev->gfx.gfx_ring[0]);
6652 		break;
6653 	case 1:
6654 	case 2:
6655 		for (i = 0; i < adev->gfx.num_compute_rings; i++) {
6656 			ring = &adev->gfx.compute_ring[i];
6657 			/* Per-queue interrupt is supported for MEC starting from VI.
6658 			  * The interrupt can only be enabled/disabled per pipe instead of per queue.
6659 			  */
6660 			if ((ring->me == me_id) && (ring->pipe == pipe_id) && (ring->queue == queue_id))
6661 				amdgpu_fence_process(ring);
6662 		}
6663 		break;
6664 	}
6665 	return 0;
6666 }
6667 
6668 static void gfx_v8_0_fault(struct amdgpu_device *adev,
6669 			   struct amdgpu_iv_entry *entry)
6670 {
6671 	u8 me_id, pipe_id, queue_id;
6672 	struct amdgpu_ring *ring;
6673 	int i;
6674 
6675 	me_id = (entry->ring_id & 0x0c) >> 2;
6676 	pipe_id = (entry->ring_id & 0x03) >> 0;
6677 	queue_id = (entry->ring_id & 0x70) >> 4;
6678 
6679 	switch (me_id) {
6680 	case 0:
6681 		drm_sched_fault(&adev->gfx.gfx_ring[0].sched);
6682 		break;
6683 	case 1:
6684 	case 2:
6685 		for (i = 0; i < adev->gfx.num_compute_rings; i++) {
6686 			ring = &adev->gfx.compute_ring[i];
6687 			if (ring->me == me_id && ring->pipe == pipe_id &&
6688 			    ring->queue == queue_id)
6689 				drm_sched_fault(&ring->sched);
6690 		}
6691 		break;
6692 	}
6693 }
6694 
6695 static int gfx_v8_0_priv_reg_irq(struct amdgpu_device *adev,
6696 				 struct amdgpu_irq_src *source,
6697 				 struct amdgpu_iv_entry *entry)
6698 {
6699 	DRM_ERROR("Illegal register access in command stream\n");
6700 	gfx_v8_0_fault(adev, entry);
6701 	return 0;
6702 }
6703 
6704 static int gfx_v8_0_priv_inst_irq(struct amdgpu_device *adev,
6705 				  struct amdgpu_irq_src *source,
6706 				  struct amdgpu_iv_entry *entry)
6707 {
6708 	DRM_ERROR("Illegal instruction in command stream\n");
6709 	gfx_v8_0_fault(adev, entry);
6710 	return 0;
6711 }
6712 
6713 static int gfx_v8_0_cp_ecc_error_irq(struct amdgpu_device *adev,
6714 				     struct amdgpu_irq_src *source,
6715 				     struct amdgpu_iv_entry *entry)
6716 {
6717 	DRM_ERROR("CP EDC/ECC error detected.");
6718 	return 0;
6719 }
6720 
6721 static void gfx_v8_0_parse_sq_irq(struct amdgpu_device *adev, unsigned ih_data)
6722 {
6723 	u32 enc, se_id, sh_id, cu_id;
6724 	char type[20];
6725 	int sq_edc_source = -1;
6726 
6727 	enc = REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_CMN, ENCODING);
6728 	se_id = REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_CMN, SE_ID);
6729 
6730 	switch (enc) {
6731 		case 0:
6732 			DRM_INFO("SQ general purpose intr detected:"
6733 					"se_id %d, immed_overflow %d, host_reg_overflow %d,"
6734 					"host_cmd_overflow %d, cmd_timestamp %d,"
6735 					"reg_timestamp %d, thread_trace_buff_full %d,"
6736 					"wlt %d, thread_trace %d.\n",
6737 					se_id,
6738 					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_AUTO, IMMED_OVERFLOW),
6739 					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_AUTO, HOST_REG_OVERFLOW),
6740 					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_AUTO, HOST_CMD_OVERFLOW),
6741 					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_AUTO, CMD_TIMESTAMP),
6742 					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_AUTO, REG_TIMESTAMP),
6743 					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_AUTO, THREAD_TRACE_BUF_FULL),
6744 					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_AUTO, WLT),
6745 					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_AUTO, THREAD_TRACE)
6746 					);
6747 			break;
6748 		case 1:
6749 		case 2:
6750 
6751 			cu_id = REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_WAVE, CU_ID);
6752 			sh_id = REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_WAVE, SH_ID);
6753 
6754 			/*
6755 			 * This function can be called either directly from ISR
6756 			 * or from BH in which case we can access SQ_EDC_INFO
6757 			 * instance
6758 			 */
6759 			if (in_task()) {
6760 				mutex_lock(&adev->grbm_idx_mutex);
6761 				gfx_v8_0_select_se_sh(adev, se_id, sh_id, cu_id);
6762 
6763 				sq_edc_source = REG_GET_FIELD(RREG32(mmSQ_EDC_INFO), SQ_EDC_INFO, SOURCE);
6764 
6765 				gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
6766 				mutex_unlock(&adev->grbm_idx_mutex);
6767 			}
6768 
6769 			if (enc == 1)
6770 				sprintf(type, "instruction intr");
6771 			else
6772 				sprintf(type, "EDC/ECC error");
6773 
6774 			DRM_INFO(
6775 				"SQ %s detected: "
6776 					"se_id %d, sh_id %d, cu_id %d, simd_id %d, wave_id %d, vm_id %d "
6777 					"trap %s, sq_ed_info.source %s.\n",
6778 					type, se_id, sh_id, cu_id,
6779 					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_WAVE, SIMD_ID),
6780 					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_WAVE, WAVE_ID),
6781 					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_WAVE, VM_ID),
6782 					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_WAVE, PRIV) ? "true" : "false",
6783 					(sq_edc_source != -1) ? sq_edc_source_names[sq_edc_source] : "unavailable"
6784 				);
6785 			break;
6786 		default:
6787 			DRM_ERROR("SQ invalid encoding type\n.");
6788 	}
6789 }
6790 
6791 static void gfx_v8_0_sq_irq_work_func(struct work_struct *work)
6792 {
6793 
6794 	struct amdgpu_device *adev = container_of(work, struct amdgpu_device, gfx.sq_work.work);
6795 	struct sq_work *sq_work = container_of(work, struct sq_work, work);
6796 
6797 	gfx_v8_0_parse_sq_irq(adev, sq_work->ih_data);
6798 }
6799 
6800 static int gfx_v8_0_sq_irq(struct amdgpu_device *adev,
6801 			   struct amdgpu_irq_src *source,
6802 			   struct amdgpu_iv_entry *entry)
6803 {
6804 	unsigned ih_data = entry->src_data[0];
6805 
6806 	/*
6807 	 * Try to submit work so SQ_EDC_INFO can be accessed from
6808 	 * BH. If previous work submission hasn't finished yet
6809 	 * just print whatever info is possible directly from the ISR.
6810 	 */
6811 	if (work_pending(&adev->gfx.sq_work.work)) {
6812 		gfx_v8_0_parse_sq_irq(adev, ih_data);
6813 	} else {
6814 		adev->gfx.sq_work.ih_data = ih_data;
6815 		schedule_work(&adev->gfx.sq_work.work);
6816 	}
6817 
6818 	return 0;
6819 }
6820 
6821 static void gfx_v8_0_emit_mem_sync(struct amdgpu_ring *ring)
6822 {
6823 	amdgpu_ring_write(ring, PACKET3(PACKET3_SURFACE_SYNC, 3));
6824 	amdgpu_ring_write(ring, PACKET3_TCL1_ACTION_ENA |
6825 			  PACKET3_TC_ACTION_ENA |
6826 			  PACKET3_SH_KCACHE_ACTION_ENA |
6827 			  PACKET3_SH_ICACHE_ACTION_ENA |
6828 			  PACKET3_TC_WB_ACTION_ENA);  /* CP_COHER_CNTL */
6829 	amdgpu_ring_write(ring, 0xffffffff);  /* CP_COHER_SIZE */
6830 	amdgpu_ring_write(ring, 0);  /* CP_COHER_BASE */
6831 	amdgpu_ring_write(ring, 0x0000000A); /* poll interval */
6832 }
6833 
6834 static void gfx_v8_0_emit_mem_sync_compute(struct amdgpu_ring *ring)
6835 {
6836 	amdgpu_ring_write(ring, PACKET3(PACKET3_ACQUIRE_MEM, 5));
6837 	amdgpu_ring_write(ring, PACKET3_TCL1_ACTION_ENA |
6838 			  PACKET3_TC_ACTION_ENA |
6839 			  PACKET3_SH_KCACHE_ACTION_ENA |
6840 			  PACKET3_SH_ICACHE_ACTION_ENA |
6841 			  PACKET3_TC_WB_ACTION_ENA);  /* CP_COHER_CNTL */
6842 	amdgpu_ring_write(ring, 0xffffffff);	/* CP_COHER_SIZE */
6843 	amdgpu_ring_write(ring, 0xff);		/* CP_COHER_SIZE_HI */
6844 	amdgpu_ring_write(ring, 0);		/* CP_COHER_BASE */
6845 	amdgpu_ring_write(ring, 0);		/* CP_COHER_BASE_HI */
6846 	amdgpu_ring_write(ring, 0x0000000A);	/* poll interval */
6847 }
6848 
6849 
6850 /* mmSPI_WCL_PIPE_PERCENT_CS[0-7]_DEFAULT values are same */
6851 #define mmSPI_WCL_PIPE_PERCENT_CS_DEFAULT	0x0000007f
6852 static void gfx_v8_0_emit_wave_limit_cs(struct amdgpu_ring *ring,
6853 					uint32_t pipe, bool enable)
6854 {
6855 	uint32_t val;
6856 	uint32_t wcl_cs_reg;
6857 
6858 	val = enable ? 0x1 : mmSPI_WCL_PIPE_PERCENT_CS_DEFAULT;
6859 
6860 	switch (pipe) {
6861 	case 0:
6862 		wcl_cs_reg = mmSPI_WCL_PIPE_PERCENT_CS0;
6863 		break;
6864 	case 1:
6865 		wcl_cs_reg = mmSPI_WCL_PIPE_PERCENT_CS1;
6866 		break;
6867 	case 2:
6868 		wcl_cs_reg = mmSPI_WCL_PIPE_PERCENT_CS2;
6869 		break;
6870 	case 3:
6871 		wcl_cs_reg = mmSPI_WCL_PIPE_PERCENT_CS3;
6872 		break;
6873 	default:
6874 		DRM_DEBUG("invalid pipe %d\n", pipe);
6875 		return;
6876 	}
6877 
6878 	amdgpu_ring_emit_wreg(ring, wcl_cs_reg, val);
6879 
6880 }
6881 
6882 #define mmSPI_WCL_PIPE_PERCENT_GFX_DEFAULT	0x07ffffff
6883 static void gfx_v8_0_emit_wave_limit(struct amdgpu_ring *ring, bool enable)
6884 {
6885 	struct amdgpu_device *adev = ring->adev;
6886 	uint32_t val;
6887 	int i;
6888 
6889 	/* mmSPI_WCL_PIPE_PERCENT_GFX is 7 bit multiplier register to limit
6890 	 * number of gfx waves. Setting 5 bit will make sure gfx only gets
6891 	 * around 25% of gpu resources.
6892 	 */
6893 	val = enable ? 0x1f : mmSPI_WCL_PIPE_PERCENT_GFX_DEFAULT;
6894 	amdgpu_ring_emit_wreg(ring, mmSPI_WCL_PIPE_PERCENT_GFX, val);
6895 
6896 	/* Restrict waves for normal/low priority compute queues as well
6897 	 * to get best QoS for high priority compute jobs.
6898 	 *
6899 	 * amdgpu controls only 1st ME(0-3 CS pipes).
6900 	 */
6901 	for (i = 0; i < adev->gfx.mec.num_pipe_per_mec; i++) {
6902 		if (i != ring->pipe)
6903 			gfx_v8_0_emit_wave_limit_cs(ring, i, enable);
6904 
6905 	}
6906 
6907 }
6908 
6909 static const struct amd_ip_funcs gfx_v8_0_ip_funcs = {
6910 	.name = "gfx_v8_0",
6911 	.early_init = gfx_v8_0_early_init,
6912 	.late_init = gfx_v8_0_late_init,
6913 	.sw_init = gfx_v8_0_sw_init,
6914 	.sw_fini = gfx_v8_0_sw_fini,
6915 	.hw_init = gfx_v8_0_hw_init,
6916 	.hw_fini = gfx_v8_0_hw_fini,
6917 	.suspend = gfx_v8_0_suspend,
6918 	.resume = gfx_v8_0_resume,
6919 	.is_idle = gfx_v8_0_is_idle,
6920 	.wait_for_idle = gfx_v8_0_wait_for_idle,
6921 	.check_soft_reset = gfx_v8_0_check_soft_reset,
6922 	.pre_soft_reset = gfx_v8_0_pre_soft_reset,
6923 	.soft_reset = gfx_v8_0_soft_reset,
6924 	.post_soft_reset = gfx_v8_0_post_soft_reset,
6925 	.set_clockgating_state = gfx_v8_0_set_clockgating_state,
6926 	.set_powergating_state = gfx_v8_0_set_powergating_state,
6927 	.get_clockgating_state = gfx_v8_0_get_clockgating_state,
6928 };
6929 
6930 static const struct amdgpu_ring_funcs gfx_v8_0_ring_funcs_gfx = {
6931 	.type = AMDGPU_RING_TYPE_GFX,
6932 	.align_mask = 0xff,
6933 	.nop = PACKET3(PACKET3_NOP, 0x3FFF),
6934 	.support_64bit_ptrs = false,
6935 	.get_rptr = gfx_v8_0_ring_get_rptr,
6936 	.get_wptr = gfx_v8_0_ring_get_wptr_gfx,
6937 	.set_wptr = gfx_v8_0_ring_set_wptr_gfx,
6938 	.emit_frame_size = /* maximum 215dw if count 16 IBs in */
6939 		5 +  /* COND_EXEC */
6940 		7 +  /* PIPELINE_SYNC */
6941 		VI_FLUSH_GPU_TLB_NUM_WREG * 5 + 9 + /* VM_FLUSH */
6942 		12 +  /* FENCE for VM_FLUSH */
6943 		20 + /* GDS switch */
6944 		4 + /* double SWITCH_BUFFER,
6945 		       the first COND_EXEC jump to the place just
6946 			   prior to this double SWITCH_BUFFER  */
6947 		5 + /* COND_EXEC */
6948 		7 +	 /*	HDP_flush */
6949 		4 +	 /*	VGT_flush */
6950 		14 + /*	CE_META */
6951 		31 + /*	DE_META */
6952 		3 + /* CNTX_CTRL */
6953 		5 + /* HDP_INVL */
6954 		12 + 12 + /* FENCE x2 */
6955 		2 + /* SWITCH_BUFFER */
6956 		5, /* SURFACE_SYNC */
6957 	.emit_ib_size =	4, /* gfx_v8_0_ring_emit_ib_gfx */
6958 	.emit_ib = gfx_v8_0_ring_emit_ib_gfx,
6959 	.emit_fence = gfx_v8_0_ring_emit_fence_gfx,
6960 	.emit_pipeline_sync = gfx_v8_0_ring_emit_pipeline_sync,
6961 	.emit_vm_flush = gfx_v8_0_ring_emit_vm_flush,
6962 	.emit_gds_switch = gfx_v8_0_ring_emit_gds_switch,
6963 	.emit_hdp_flush = gfx_v8_0_ring_emit_hdp_flush,
6964 	.test_ring = gfx_v8_0_ring_test_ring,
6965 	.test_ib = gfx_v8_0_ring_test_ib,
6966 	.insert_nop = amdgpu_ring_insert_nop,
6967 	.pad_ib = amdgpu_ring_generic_pad_ib,
6968 	.emit_switch_buffer = gfx_v8_ring_emit_sb,
6969 	.emit_cntxcntl = gfx_v8_ring_emit_cntxcntl,
6970 	.init_cond_exec = gfx_v8_0_ring_emit_init_cond_exec,
6971 	.patch_cond_exec = gfx_v8_0_ring_emit_patch_cond_exec,
6972 	.emit_wreg = gfx_v8_0_ring_emit_wreg,
6973 	.soft_recovery = gfx_v8_0_ring_soft_recovery,
6974 	.emit_mem_sync = gfx_v8_0_emit_mem_sync,
6975 };
6976 
6977 static const struct amdgpu_ring_funcs gfx_v8_0_ring_funcs_compute = {
6978 	.type = AMDGPU_RING_TYPE_COMPUTE,
6979 	.align_mask = 0xff,
6980 	.nop = PACKET3(PACKET3_NOP, 0x3FFF),
6981 	.support_64bit_ptrs = false,
6982 	.get_rptr = gfx_v8_0_ring_get_rptr,
6983 	.get_wptr = gfx_v8_0_ring_get_wptr_compute,
6984 	.set_wptr = gfx_v8_0_ring_set_wptr_compute,
6985 	.emit_frame_size =
6986 		20 + /* gfx_v8_0_ring_emit_gds_switch */
6987 		7 + /* gfx_v8_0_ring_emit_hdp_flush */
6988 		5 + /* hdp_invalidate */
6989 		7 + /* gfx_v8_0_ring_emit_pipeline_sync */
6990 		VI_FLUSH_GPU_TLB_NUM_WREG * 5 + 7 + /* gfx_v8_0_ring_emit_vm_flush */
6991 		7 + 7 + 7 + /* gfx_v8_0_ring_emit_fence_compute x3 for user fence, vm fence */
6992 		7 + /* gfx_v8_0_emit_mem_sync_compute */
6993 		5 + /* gfx_v8_0_emit_wave_limit for updating mmSPI_WCL_PIPE_PERCENT_GFX register */
6994 		15, /* for updating 3 mmSPI_WCL_PIPE_PERCENT_CS registers */
6995 	.emit_ib_size =	7, /* gfx_v8_0_ring_emit_ib_compute */
6996 	.emit_ib = gfx_v8_0_ring_emit_ib_compute,
6997 	.emit_fence = gfx_v8_0_ring_emit_fence_compute,
6998 	.emit_pipeline_sync = gfx_v8_0_ring_emit_pipeline_sync,
6999 	.emit_vm_flush = gfx_v8_0_ring_emit_vm_flush,
7000 	.emit_gds_switch = gfx_v8_0_ring_emit_gds_switch,
7001 	.emit_hdp_flush = gfx_v8_0_ring_emit_hdp_flush,
7002 	.test_ring = gfx_v8_0_ring_test_ring,
7003 	.test_ib = gfx_v8_0_ring_test_ib,
7004 	.insert_nop = amdgpu_ring_insert_nop,
7005 	.pad_ib = amdgpu_ring_generic_pad_ib,
7006 	.emit_wreg = gfx_v8_0_ring_emit_wreg,
7007 	.emit_mem_sync = gfx_v8_0_emit_mem_sync_compute,
7008 	.emit_wave_limit = gfx_v8_0_emit_wave_limit,
7009 };
7010 
7011 static const struct amdgpu_ring_funcs gfx_v8_0_ring_funcs_kiq = {
7012 	.type = AMDGPU_RING_TYPE_KIQ,
7013 	.align_mask = 0xff,
7014 	.nop = PACKET3(PACKET3_NOP, 0x3FFF),
7015 	.support_64bit_ptrs = false,
7016 	.get_rptr = gfx_v8_0_ring_get_rptr,
7017 	.get_wptr = gfx_v8_0_ring_get_wptr_compute,
7018 	.set_wptr = gfx_v8_0_ring_set_wptr_compute,
7019 	.emit_frame_size =
7020 		20 + /* gfx_v8_0_ring_emit_gds_switch */
7021 		7 + /* gfx_v8_0_ring_emit_hdp_flush */
7022 		5 + /* hdp_invalidate */
7023 		7 + /* gfx_v8_0_ring_emit_pipeline_sync */
7024 		17 + /* gfx_v8_0_ring_emit_vm_flush */
7025 		7 + 7 + 7, /* gfx_v8_0_ring_emit_fence_kiq x3 for user fence, vm fence */
7026 	.emit_ib_size =	7, /* gfx_v8_0_ring_emit_ib_compute */
7027 	.emit_fence = gfx_v8_0_ring_emit_fence_kiq,
7028 	.test_ring = gfx_v8_0_ring_test_ring,
7029 	.insert_nop = amdgpu_ring_insert_nop,
7030 	.pad_ib = amdgpu_ring_generic_pad_ib,
7031 	.emit_rreg = gfx_v8_0_ring_emit_rreg,
7032 	.emit_wreg = gfx_v8_0_ring_emit_wreg,
7033 };
7034 
7035 static void gfx_v8_0_set_ring_funcs(struct amdgpu_device *adev)
7036 {
7037 	int i;
7038 
7039 	adev->gfx.kiq.ring.funcs = &gfx_v8_0_ring_funcs_kiq;
7040 
7041 	for (i = 0; i < adev->gfx.num_gfx_rings; i++)
7042 		adev->gfx.gfx_ring[i].funcs = &gfx_v8_0_ring_funcs_gfx;
7043 
7044 	for (i = 0; i < adev->gfx.num_compute_rings; i++)
7045 		adev->gfx.compute_ring[i].funcs = &gfx_v8_0_ring_funcs_compute;
7046 }
7047 
7048 static const struct amdgpu_irq_src_funcs gfx_v8_0_eop_irq_funcs = {
7049 	.set = gfx_v8_0_set_eop_interrupt_state,
7050 	.process = gfx_v8_0_eop_irq,
7051 };
7052 
7053 static const struct amdgpu_irq_src_funcs gfx_v8_0_priv_reg_irq_funcs = {
7054 	.set = gfx_v8_0_set_priv_reg_fault_state,
7055 	.process = gfx_v8_0_priv_reg_irq,
7056 };
7057 
7058 static const struct amdgpu_irq_src_funcs gfx_v8_0_priv_inst_irq_funcs = {
7059 	.set = gfx_v8_0_set_priv_inst_fault_state,
7060 	.process = gfx_v8_0_priv_inst_irq,
7061 };
7062 
7063 static const struct amdgpu_irq_src_funcs gfx_v8_0_cp_ecc_error_irq_funcs = {
7064 	.set = gfx_v8_0_set_cp_ecc_int_state,
7065 	.process = gfx_v8_0_cp_ecc_error_irq,
7066 };
7067 
7068 static const struct amdgpu_irq_src_funcs gfx_v8_0_sq_irq_funcs = {
7069 	.set = gfx_v8_0_set_sq_int_state,
7070 	.process = gfx_v8_0_sq_irq,
7071 };
7072 
7073 static void gfx_v8_0_set_irq_funcs(struct amdgpu_device *adev)
7074 {
7075 	adev->gfx.eop_irq.num_types = AMDGPU_CP_IRQ_LAST;
7076 	adev->gfx.eop_irq.funcs = &gfx_v8_0_eop_irq_funcs;
7077 
7078 	adev->gfx.priv_reg_irq.num_types = 1;
7079 	adev->gfx.priv_reg_irq.funcs = &gfx_v8_0_priv_reg_irq_funcs;
7080 
7081 	adev->gfx.priv_inst_irq.num_types = 1;
7082 	adev->gfx.priv_inst_irq.funcs = &gfx_v8_0_priv_inst_irq_funcs;
7083 
7084 	adev->gfx.cp_ecc_error_irq.num_types = 1;
7085 	adev->gfx.cp_ecc_error_irq.funcs = &gfx_v8_0_cp_ecc_error_irq_funcs;
7086 
7087 	adev->gfx.sq_irq.num_types = 1;
7088 	adev->gfx.sq_irq.funcs = &gfx_v8_0_sq_irq_funcs;
7089 }
7090 
7091 static void gfx_v8_0_set_rlc_funcs(struct amdgpu_device *adev)
7092 {
7093 	adev->gfx.rlc.funcs = &iceland_rlc_funcs;
7094 }
7095 
7096 static void gfx_v8_0_set_gds_init(struct amdgpu_device *adev)
7097 {
7098 	/* init asci gds info */
7099 	adev->gds.gds_size = RREG32(mmGDS_VMID0_SIZE);
7100 	adev->gds.gws_size = 64;
7101 	adev->gds.oa_size = 16;
7102 	adev->gds.gds_compute_max_wave_id = RREG32(mmGDS_COMPUTE_MAX_WAVE_ID);
7103 }
7104 
7105 static void gfx_v8_0_set_user_cu_inactive_bitmap(struct amdgpu_device *adev,
7106 						 u32 bitmap)
7107 {
7108 	u32 data;
7109 
7110 	if (!bitmap)
7111 		return;
7112 
7113 	data = bitmap << GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_CUS__SHIFT;
7114 	data &= GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_CUS_MASK;
7115 
7116 	WREG32(mmGC_USER_SHADER_ARRAY_CONFIG, data);
7117 }
7118 
7119 static u32 gfx_v8_0_get_cu_active_bitmap(struct amdgpu_device *adev)
7120 {
7121 	u32 data, mask;
7122 
7123 	data =  RREG32(mmCC_GC_SHADER_ARRAY_CONFIG) |
7124 		RREG32(mmGC_USER_SHADER_ARRAY_CONFIG);
7125 
7126 	mask = amdgpu_gfx_create_bitmask(adev->gfx.config.max_cu_per_sh);
7127 
7128 	return ~REG_GET_FIELD(data, CC_GC_SHADER_ARRAY_CONFIG, INACTIVE_CUS) & mask;
7129 }
7130 
7131 static void gfx_v8_0_get_cu_info(struct amdgpu_device *adev)
7132 {
7133 	int i, j, k, counter, active_cu_number = 0;
7134 	u32 mask, bitmap, ao_bitmap, ao_cu_mask = 0;
7135 	struct amdgpu_cu_info *cu_info = &adev->gfx.cu_info;
7136 	unsigned disable_masks[4 * 2];
7137 	u32 ao_cu_num;
7138 
7139 	memset(cu_info, 0, sizeof(*cu_info));
7140 
7141 	if (adev->flags & AMD_IS_APU)
7142 		ao_cu_num = 2;
7143 	else
7144 		ao_cu_num = adev->gfx.config.max_cu_per_sh;
7145 
7146 	amdgpu_gfx_parse_disable_cu(disable_masks, 4, 2);
7147 
7148 	mutex_lock(&adev->grbm_idx_mutex);
7149 	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
7150 		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
7151 			mask = 1;
7152 			ao_bitmap = 0;
7153 			counter = 0;
7154 			gfx_v8_0_select_se_sh(adev, i, j, 0xffffffff);
7155 			if (i < 4 && j < 2)
7156 				gfx_v8_0_set_user_cu_inactive_bitmap(
7157 					adev, disable_masks[i * 2 + j]);
7158 			bitmap = gfx_v8_0_get_cu_active_bitmap(adev);
7159 			cu_info->bitmap[i][j] = bitmap;
7160 
7161 			for (k = 0; k < adev->gfx.config.max_cu_per_sh; k ++) {
7162 				if (bitmap & mask) {
7163 					if (counter < ao_cu_num)
7164 						ao_bitmap |= mask;
7165 					counter ++;
7166 				}
7167 				mask <<= 1;
7168 			}
7169 			active_cu_number += counter;
7170 			if (i < 2 && j < 2)
7171 				ao_cu_mask |= (ao_bitmap << (i * 16 + j * 8));
7172 			cu_info->ao_cu_bitmap[i][j] = ao_bitmap;
7173 		}
7174 	}
7175 	gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
7176 	mutex_unlock(&adev->grbm_idx_mutex);
7177 
7178 	cu_info->number = active_cu_number;
7179 	cu_info->ao_cu_mask = ao_cu_mask;
7180 	cu_info->simd_per_cu = NUM_SIMD_PER_CU;
7181 	cu_info->max_waves_per_simd = 10;
7182 	cu_info->max_scratch_slots_per_cu = 32;
7183 	cu_info->wave_front_size = 64;
7184 	cu_info->lds_size = 64;
7185 }
7186 
7187 const struct amdgpu_ip_block_version gfx_v8_0_ip_block =
7188 {
7189 	.type = AMD_IP_BLOCK_TYPE_GFX,
7190 	.major = 8,
7191 	.minor = 0,
7192 	.rev = 0,
7193 	.funcs = &gfx_v8_0_ip_funcs,
7194 };
7195 
7196 const struct amdgpu_ip_block_version gfx_v8_1_ip_block =
7197 {
7198 	.type = AMD_IP_BLOCK_TYPE_GFX,
7199 	.major = 8,
7200 	.minor = 1,
7201 	.rev = 0,
7202 	.funcs = &gfx_v8_0_ip_funcs,
7203 };
7204 
7205 static void gfx_v8_0_ring_emit_ce_meta(struct amdgpu_ring *ring)
7206 {
7207 	uint64_t ce_payload_addr;
7208 	int cnt_ce;
7209 	union {
7210 		struct vi_ce_ib_state regular;
7211 		struct vi_ce_ib_state_chained_ib chained;
7212 	} ce_payload = {};
7213 
7214 	if (ring->adev->virt.chained_ib_support) {
7215 		ce_payload_addr = amdgpu_csa_vaddr(ring->adev) +
7216 			offsetof(struct vi_gfx_meta_data_chained_ib, ce_payload);
7217 		cnt_ce = (sizeof(ce_payload.chained) >> 2) + 4 - 2;
7218 	} else {
7219 		ce_payload_addr = amdgpu_csa_vaddr(ring->adev) +
7220 			offsetof(struct vi_gfx_meta_data, ce_payload);
7221 		cnt_ce = (sizeof(ce_payload.regular) >> 2) + 4 - 2;
7222 	}
7223 
7224 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, cnt_ce));
7225 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(2) |
7226 				WRITE_DATA_DST_SEL(8) |
7227 				WR_CONFIRM) |
7228 				WRITE_DATA_CACHE_POLICY(0));
7229 	amdgpu_ring_write(ring, lower_32_bits(ce_payload_addr));
7230 	amdgpu_ring_write(ring, upper_32_bits(ce_payload_addr));
7231 	amdgpu_ring_write_multiple(ring, (void *)&ce_payload, cnt_ce - 2);
7232 }
7233 
7234 static void gfx_v8_0_ring_emit_de_meta(struct amdgpu_ring *ring)
7235 {
7236 	uint64_t de_payload_addr, gds_addr, csa_addr;
7237 	int cnt_de;
7238 	union {
7239 		struct vi_de_ib_state regular;
7240 		struct vi_de_ib_state_chained_ib chained;
7241 	} de_payload = {};
7242 
7243 	csa_addr = amdgpu_csa_vaddr(ring->adev);
7244 	gds_addr = csa_addr + 4096;
7245 	if (ring->adev->virt.chained_ib_support) {
7246 		de_payload.chained.gds_backup_addrlo = lower_32_bits(gds_addr);
7247 		de_payload.chained.gds_backup_addrhi = upper_32_bits(gds_addr);
7248 		de_payload_addr = csa_addr + offsetof(struct vi_gfx_meta_data_chained_ib, de_payload);
7249 		cnt_de = (sizeof(de_payload.chained) >> 2) + 4 - 2;
7250 	} else {
7251 		de_payload.regular.gds_backup_addrlo = lower_32_bits(gds_addr);
7252 		de_payload.regular.gds_backup_addrhi = upper_32_bits(gds_addr);
7253 		de_payload_addr = csa_addr + offsetof(struct vi_gfx_meta_data, de_payload);
7254 		cnt_de = (sizeof(de_payload.regular) >> 2) + 4 - 2;
7255 	}
7256 
7257 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, cnt_de));
7258 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(1) |
7259 				WRITE_DATA_DST_SEL(8) |
7260 				WR_CONFIRM) |
7261 				WRITE_DATA_CACHE_POLICY(0));
7262 	amdgpu_ring_write(ring, lower_32_bits(de_payload_addr));
7263 	amdgpu_ring_write(ring, upper_32_bits(de_payload_addr));
7264 	amdgpu_ring_write_multiple(ring, (void *)&de_payload, cnt_de - 2);
7265 }
7266