xref: /linux/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c (revision c532de5a67a70f8533d495f8f2aaa9a0491c3ad0)
1 /*
2  * Copyright 2014 Advanced Micro Devices, Inc.
3  *
4  * Permission is hereby granted, free of charge, to any person obtaining a
5  * copy of this software and associated documentation files (the "Software"),
6  * to deal in the Software without restriction, including without limitation
7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8  * and/or sell copies of the Software, and to permit persons to whom the
9  * Software is furnished to do so, subject to the following conditions:
10  *
11  * The above copyright notice and this permission notice shall be included in
12  * all copies or substantial portions of the Software.
13  *
14  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
17  * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20  * OTHER DEALINGS IN THE SOFTWARE.
21  *
22  */
23 
24 #include <linux/delay.h>
25 #include <linux/kernel.h>
26 #include <linux/firmware.h>
27 #include <linux/module.h>
28 #include <linux/pci.h>
29 
30 #include "amdgpu.h"
31 #include "amdgpu_gfx.h"
32 #include "amdgpu_ring.h"
33 #include "vi.h"
34 #include "vi_structs.h"
35 #include "vid.h"
36 #include "amdgpu_ucode.h"
37 #include "amdgpu_atombios.h"
38 #include "atombios_i2c.h"
39 #include "clearstate_vi.h"
40 
41 #include "gmc/gmc_8_2_d.h"
42 #include "gmc/gmc_8_2_sh_mask.h"
43 
44 #include "oss/oss_3_0_d.h"
45 #include "oss/oss_3_0_sh_mask.h"
46 
47 #include "bif/bif_5_0_d.h"
48 #include "bif/bif_5_0_sh_mask.h"
49 #include "gca/gfx_8_0_d.h"
50 #include "gca/gfx_8_0_enum.h"
51 #include "gca/gfx_8_0_sh_mask.h"
52 
53 #include "dce/dce_10_0_d.h"
54 #include "dce/dce_10_0_sh_mask.h"
55 
56 #include "smu/smu_7_1_3_d.h"
57 
58 #include "ivsrcid/ivsrcid_vislands30.h"
59 
60 #define GFX8_NUM_GFX_RINGS     1
61 #define GFX8_MEC_HPD_SIZE 4096
62 
63 #define TOPAZ_GB_ADDR_CONFIG_GOLDEN 0x22010001
64 #define CARRIZO_GB_ADDR_CONFIG_GOLDEN 0x22010001
65 #define POLARIS11_GB_ADDR_CONFIG_GOLDEN 0x22011002
66 #define TONGA_GB_ADDR_CONFIG_GOLDEN 0x22011003
67 
68 #define ARRAY_MODE(x)					((x) << GB_TILE_MODE0__ARRAY_MODE__SHIFT)
69 #define PIPE_CONFIG(x)					((x) << GB_TILE_MODE0__PIPE_CONFIG__SHIFT)
70 #define TILE_SPLIT(x)					((x) << GB_TILE_MODE0__TILE_SPLIT__SHIFT)
71 #define MICRO_TILE_MODE_NEW(x)				((x) << GB_TILE_MODE0__MICRO_TILE_MODE_NEW__SHIFT)
72 #define SAMPLE_SPLIT(x)					((x) << GB_TILE_MODE0__SAMPLE_SPLIT__SHIFT)
73 #define BANK_WIDTH(x)					((x) << GB_MACROTILE_MODE0__BANK_WIDTH__SHIFT)
74 #define BANK_HEIGHT(x)					((x) << GB_MACROTILE_MODE0__BANK_HEIGHT__SHIFT)
75 #define MACRO_TILE_ASPECT(x)				((x) << GB_MACROTILE_MODE0__MACRO_TILE_ASPECT__SHIFT)
76 #define NUM_BANKS(x)					((x) << GB_MACROTILE_MODE0__NUM_BANKS__SHIFT)
77 
78 #define RLC_CGTT_MGCG_OVERRIDE__CPF_MASK            0x00000001L
79 #define RLC_CGTT_MGCG_OVERRIDE__RLC_MASK            0x00000002L
80 #define RLC_CGTT_MGCG_OVERRIDE__MGCG_MASK           0x00000004L
81 #define RLC_CGTT_MGCG_OVERRIDE__CGCG_MASK           0x00000008L
82 #define RLC_CGTT_MGCG_OVERRIDE__CGLS_MASK           0x00000010L
83 #define RLC_CGTT_MGCG_OVERRIDE__GRBM_MASK           0x00000020L
84 
85 /* BPM SERDES CMD */
86 #define SET_BPM_SERDES_CMD    1
87 #define CLE_BPM_SERDES_CMD    0
88 
89 /* BPM Register Address*/
90 enum {
91 	BPM_REG_CGLS_EN = 0,        /* Enable/Disable CGLS */
92 	BPM_REG_CGLS_ON,            /* ON/OFF CGLS: shall be controlled by RLC FW */
93 	BPM_REG_CGCG_OVERRIDE,      /* Set/Clear CGCG Override */
94 	BPM_REG_MGCG_OVERRIDE,      /* Set/Clear MGCG Override */
95 	BPM_REG_FGCG_OVERRIDE,      /* Set/Clear FGCG Override */
96 	BPM_REG_FGCG_MAX
97 };
98 
99 #define RLC_FormatDirectRegListLength        14
100 
101 MODULE_FIRMWARE("amdgpu/carrizo_ce.bin");
102 MODULE_FIRMWARE("amdgpu/carrizo_pfp.bin");
103 MODULE_FIRMWARE("amdgpu/carrizo_me.bin");
104 MODULE_FIRMWARE("amdgpu/carrizo_mec.bin");
105 MODULE_FIRMWARE("amdgpu/carrizo_mec2.bin");
106 MODULE_FIRMWARE("amdgpu/carrizo_rlc.bin");
107 
108 MODULE_FIRMWARE("amdgpu/stoney_ce.bin");
109 MODULE_FIRMWARE("amdgpu/stoney_pfp.bin");
110 MODULE_FIRMWARE("amdgpu/stoney_me.bin");
111 MODULE_FIRMWARE("amdgpu/stoney_mec.bin");
112 MODULE_FIRMWARE("amdgpu/stoney_rlc.bin");
113 
114 MODULE_FIRMWARE("amdgpu/tonga_ce.bin");
115 MODULE_FIRMWARE("amdgpu/tonga_pfp.bin");
116 MODULE_FIRMWARE("amdgpu/tonga_me.bin");
117 MODULE_FIRMWARE("amdgpu/tonga_mec.bin");
118 MODULE_FIRMWARE("amdgpu/tonga_mec2.bin");
119 MODULE_FIRMWARE("amdgpu/tonga_rlc.bin");
120 
121 MODULE_FIRMWARE("amdgpu/topaz_ce.bin");
122 MODULE_FIRMWARE("amdgpu/topaz_pfp.bin");
123 MODULE_FIRMWARE("amdgpu/topaz_me.bin");
124 MODULE_FIRMWARE("amdgpu/topaz_mec.bin");
125 MODULE_FIRMWARE("amdgpu/topaz_rlc.bin");
126 
127 MODULE_FIRMWARE("amdgpu/fiji_ce.bin");
128 MODULE_FIRMWARE("amdgpu/fiji_pfp.bin");
129 MODULE_FIRMWARE("amdgpu/fiji_me.bin");
130 MODULE_FIRMWARE("amdgpu/fiji_mec.bin");
131 MODULE_FIRMWARE("amdgpu/fiji_mec2.bin");
132 MODULE_FIRMWARE("amdgpu/fiji_rlc.bin");
133 
134 MODULE_FIRMWARE("amdgpu/polaris10_ce.bin");
135 MODULE_FIRMWARE("amdgpu/polaris10_ce_2.bin");
136 MODULE_FIRMWARE("amdgpu/polaris10_pfp.bin");
137 MODULE_FIRMWARE("amdgpu/polaris10_pfp_2.bin");
138 MODULE_FIRMWARE("amdgpu/polaris10_me.bin");
139 MODULE_FIRMWARE("amdgpu/polaris10_me_2.bin");
140 MODULE_FIRMWARE("amdgpu/polaris10_mec.bin");
141 MODULE_FIRMWARE("amdgpu/polaris10_mec_2.bin");
142 MODULE_FIRMWARE("amdgpu/polaris10_mec2.bin");
143 MODULE_FIRMWARE("amdgpu/polaris10_mec2_2.bin");
144 MODULE_FIRMWARE("amdgpu/polaris10_rlc.bin");
145 
146 MODULE_FIRMWARE("amdgpu/polaris11_ce.bin");
147 MODULE_FIRMWARE("amdgpu/polaris11_ce_2.bin");
148 MODULE_FIRMWARE("amdgpu/polaris11_pfp.bin");
149 MODULE_FIRMWARE("amdgpu/polaris11_pfp_2.bin");
150 MODULE_FIRMWARE("amdgpu/polaris11_me.bin");
151 MODULE_FIRMWARE("amdgpu/polaris11_me_2.bin");
152 MODULE_FIRMWARE("amdgpu/polaris11_mec.bin");
153 MODULE_FIRMWARE("amdgpu/polaris11_mec_2.bin");
154 MODULE_FIRMWARE("amdgpu/polaris11_mec2.bin");
155 MODULE_FIRMWARE("amdgpu/polaris11_mec2_2.bin");
156 MODULE_FIRMWARE("amdgpu/polaris11_rlc.bin");
157 
158 MODULE_FIRMWARE("amdgpu/polaris12_ce.bin");
159 MODULE_FIRMWARE("amdgpu/polaris12_ce_2.bin");
160 MODULE_FIRMWARE("amdgpu/polaris12_pfp.bin");
161 MODULE_FIRMWARE("amdgpu/polaris12_pfp_2.bin");
162 MODULE_FIRMWARE("amdgpu/polaris12_me.bin");
163 MODULE_FIRMWARE("amdgpu/polaris12_me_2.bin");
164 MODULE_FIRMWARE("amdgpu/polaris12_mec.bin");
165 MODULE_FIRMWARE("amdgpu/polaris12_mec_2.bin");
166 MODULE_FIRMWARE("amdgpu/polaris12_mec2.bin");
167 MODULE_FIRMWARE("amdgpu/polaris12_mec2_2.bin");
168 MODULE_FIRMWARE("amdgpu/polaris12_rlc.bin");
169 
170 MODULE_FIRMWARE("amdgpu/vegam_ce.bin");
171 MODULE_FIRMWARE("amdgpu/vegam_pfp.bin");
172 MODULE_FIRMWARE("amdgpu/vegam_me.bin");
173 MODULE_FIRMWARE("amdgpu/vegam_mec.bin");
174 MODULE_FIRMWARE("amdgpu/vegam_mec2.bin");
175 MODULE_FIRMWARE("amdgpu/vegam_rlc.bin");
176 
177 static const struct amdgpu_gds_reg_offset amdgpu_gds_reg_offset[] =
178 {
179 	{mmGDS_VMID0_BASE, mmGDS_VMID0_SIZE, mmGDS_GWS_VMID0, mmGDS_OA_VMID0},
180 	{mmGDS_VMID1_BASE, mmGDS_VMID1_SIZE, mmGDS_GWS_VMID1, mmGDS_OA_VMID1},
181 	{mmGDS_VMID2_BASE, mmGDS_VMID2_SIZE, mmGDS_GWS_VMID2, mmGDS_OA_VMID2},
182 	{mmGDS_VMID3_BASE, mmGDS_VMID3_SIZE, mmGDS_GWS_VMID3, mmGDS_OA_VMID3},
183 	{mmGDS_VMID4_BASE, mmGDS_VMID4_SIZE, mmGDS_GWS_VMID4, mmGDS_OA_VMID4},
184 	{mmGDS_VMID5_BASE, mmGDS_VMID5_SIZE, mmGDS_GWS_VMID5, mmGDS_OA_VMID5},
185 	{mmGDS_VMID6_BASE, mmGDS_VMID6_SIZE, mmGDS_GWS_VMID6, mmGDS_OA_VMID6},
186 	{mmGDS_VMID7_BASE, mmGDS_VMID7_SIZE, mmGDS_GWS_VMID7, mmGDS_OA_VMID7},
187 	{mmGDS_VMID8_BASE, mmGDS_VMID8_SIZE, mmGDS_GWS_VMID8, mmGDS_OA_VMID8},
188 	{mmGDS_VMID9_BASE, mmGDS_VMID9_SIZE, mmGDS_GWS_VMID9, mmGDS_OA_VMID9},
189 	{mmGDS_VMID10_BASE, mmGDS_VMID10_SIZE, mmGDS_GWS_VMID10, mmGDS_OA_VMID10},
190 	{mmGDS_VMID11_BASE, mmGDS_VMID11_SIZE, mmGDS_GWS_VMID11, mmGDS_OA_VMID11},
191 	{mmGDS_VMID12_BASE, mmGDS_VMID12_SIZE, mmGDS_GWS_VMID12, mmGDS_OA_VMID12},
192 	{mmGDS_VMID13_BASE, mmGDS_VMID13_SIZE, mmGDS_GWS_VMID13, mmGDS_OA_VMID13},
193 	{mmGDS_VMID14_BASE, mmGDS_VMID14_SIZE, mmGDS_GWS_VMID14, mmGDS_OA_VMID14},
194 	{mmGDS_VMID15_BASE, mmGDS_VMID15_SIZE, mmGDS_GWS_VMID15, mmGDS_OA_VMID15}
195 };
196 
197 static const u32 golden_settings_tonga_a11[] =
198 {
199 	mmCB_HW_CONTROL, 0xfffdf3cf, 0x00007208,
200 	mmCB_HW_CONTROL_3, 0x00000040, 0x00000040,
201 	mmDB_DEBUG2, 0xf00fffff, 0x00000400,
202 	mmGB_GPU_ID, 0x0000000f, 0x00000000,
203 	mmPA_SC_ENHANCE, 0xffffffff, 0x20000001,
204 	mmPA_SC_FIFO_DEPTH_CNTL, 0x000003ff, 0x000000fc,
205 	mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000,
206 	mmRLC_CGCG_CGLS_CTRL, 0x00000003, 0x0000003c,
207 	mmSQ_RANDOM_WAVE_PRI, 0x001fffff, 0x000006fd,
208 	mmTA_CNTL_AUX, 0x000f000f, 0x000b0000,
209 	mmTCC_CTRL, 0x00100000, 0xf31fff7f,
210 	mmTCC_EXE_DISABLE, 0x00000002, 0x00000002,
211 	mmTCP_ADDR_CONFIG, 0x000003ff, 0x000002fb,
212 	mmTCP_CHAN_STEER_HI, 0xffffffff, 0x0000543b,
213 	mmTCP_CHAN_STEER_LO, 0xffffffff, 0xa9210876,
214 	mmVGT_RESET_DEBUG, 0x00000004, 0x00000004,
215 };
216 
217 static const u32 tonga_golden_common_all[] =
218 {
219 	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
220 	mmPA_SC_RASTER_CONFIG, 0xffffffff, 0x16000012,
221 	mmPA_SC_RASTER_CONFIG_1, 0xffffffff, 0x0000002A,
222 	mmGB_ADDR_CONFIG, 0xffffffff, 0x22011003,
223 	mmSPI_RESOURCE_RESERVE_CU_0, 0xffffffff, 0x00000800,
224 	mmSPI_RESOURCE_RESERVE_CU_1, 0xffffffff, 0x00000800,
225 	mmSPI_RESOURCE_RESERVE_EN_CU_0, 0xffffffff, 0x00FF7FBF,
226 	mmSPI_RESOURCE_RESERVE_EN_CU_1, 0xffffffff, 0x00FF7FAF
227 };
228 
229 static const u32 tonga_mgcg_cgcg_init[] =
230 {
231 	mmRLC_CGTT_MGCG_OVERRIDE, 0xffffffff, 0xffffffff,
232 	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
233 	mmCB_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100,
234 	mmCGTT_BCI_CLK_CTRL, 0xffffffff, 0x00000100,
235 	mmCGTT_CP_CLK_CTRL, 0xffffffff, 0x00000100,
236 	mmCGTT_CPC_CLK_CTRL, 0xffffffff, 0x00000100,
237 	mmCGTT_CPF_CLK_CTRL, 0xffffffff, 0x40000100,
238 	mmCGTT_GDS_CLK_CTRL, 0xffffffff, 0x00000100,
239 	mmCGTT_IA_CLK_CTRL, 0xffffffff, 0x06000100,
240 	mmCGTT_PA_CLK_CTRL, 0xffffffff, 0x00000100,
241 	mmCGTT_WD_CLK_CTRL, 0xffffffff, 0x06000100,
242 	mmCGTT_PC_CLK_CTRL, 0xffffffff, 0x00000100,
243 	mmCGTT_RLC_CLK_CTRL, 0xffffffff, 0x00000100,
244 	mmCGTT_SC_CLK_CTRL, 0xffffffff, 0x00000100,
245 	mmCGTT_SPI_CLK_CTRL, 0xffffffff, 0x00000100,
246 	mmCGTT_SQ_CLK_CTRL, 0xffffffff, 0x00000100,
247 	mmCGTT_SQG_CLK_CTRL, 0xffffffff, 0x00000100,
248 	mmCGTT_SX_CLK_CTRL0, 0xffffffff, 0x00000100,
249 	mmCGTT_SX_CLK_CTRL1, 0xffffffff, 0x00000100,
250 	mmCGTT_SX_CLK_CTRL2, 0xffffffff, 0x00000100,
251 	mmCGTT_SX_CLK_CTRL3, 0xffffffff, 0x00000100,
252 	mmCGTT_SX_CLK_CTRL4, 0xffffffff, 0x00000100,
253 	mmCGTT_TCI_CLK_CTRL, 0xffffffff, 0x00000100,
254 	mmCGTT_TCP_CLK_CTRL, 0xffffffff, 0x00000100,
255 	mmCGTT_VGT_CLK_CTRL, 0xffffffff, 0x06000100,
256 	mmDB_CGTT_CLK_CTRL_0, 0xffffffff, 0x00000100,
257 	mmTA_CGTT_CTRL, 0xffffffff, 0x00000100,
258 	mmTCA_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100,
259 	mmTCC_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100,
260 	mmTD_CGTT_CTRL, 0xffffffff, 0x00000100,
261 	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
262 	mmCGTS_CU0_SP0_CTRL_REG, 0xffffffff, 0x00010000,
263 	mmCGTS_CU0_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
264 	mmCGTS_CU0_TA_SQC_CTRL_REG, 0xffffffff, 0x00040007,
265 	mmCGTS_CU0_SP1_CTRL_REG, 0xffffffff, 0x00060005,
266 	mmCGTS_CU0_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
267 	mmCGTS_CU1_SP0_CTRL_REG, 0xffffffff, 0x00010000,
268 	mmCGTS_CU1_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
269 	mmCGTS_CU1_TA_CTRL_REG, 0xffffffff, 0x00040007,
270 	mmCGTS_CU1_SP1_CTRL_REG, 0xffffffff, 0x00060005,
271 	mmCGTS_CU1_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
272 	mmCGTS_CU2_SP0_CTRL_REG, 0xffffffff, 0x00010000,
273 	mmCGTS_CU2_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
274 	mmCGTS_CU2_TA_CTRL_REG, 0xffffffff, 0x00040007,
275 	mmCGTS_CU2_SP1_CTRL_REG, 0xffffffff, 0x00060005,
276 	mmCGTS_CU2_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
277 	mmCGTS_CU3_SP0_CTRL_REG, 0xffffffff, 0x00010000,
278 	mmCGTS_CU3_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
279 	mmCGTS_CU3_TA_CTRL_REG, 0xffffffff, 0x00040007,
280 	mmCGTS_CU3_SP1_CTRL_REG, 0xffffffff, 0x00060005,
281 	mmCGTS_CU3_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
282 	mmCGTS_CU4_SP0_CTRL_REG, 0xffffffff, 0x00010000,
283 	mmCGTS_CU4_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
284 	mmCGTS_CU4_TA_SQC_CTRL_REG, 0xffffffff, 0x00040007,
285 	mmCGTS_CU4_SP1_CTRL_REG, 0xffffffff, 0x00060005,
286 	mmCGTS_CU4_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
287 	mmCGTS_CU5_SP0_CTRL_REG, 0xffffffff, 0x00010000,
288 	mmCGTS_CU5_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
289 	mmCGTS_CU5_TA_CTRL_REG, 0xffffffff, 0x00040007,
290 	mmCGTS_CU5_SP1_CTRL_REG, 0xffffffff, 0x00060005,
291 	mmCGTS_CU5_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
292 	mmCGTS_CU6_SP0_CTRL_REG, 0xffffffff, 0x00010000,
293 	mmCGTS_CU6_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
294 	mmCGTS_CU6_TA_CTRL_REG, 0xffffffff, 0x00040007,
295 	mmCGTS_CU6_SP1_CTRL_REG, 0xffffffff, 0x00060005,
296 	mmCGTS_CU6_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
297 	mmCGTS_CU7_SP0_CTRL_REG, 0xffffffff, 0x00010000,
298 	mmCGTS_CU7_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
299 	mmCGTS_CU7_TA_CTRL_REG, 0xffffffff, 0x00040007,
300 	mmCGTS_CU7_SP1_CTRL_REG, 0xffffffff, 0x00060005,
301 	mmCGTS_CU7_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
302 	mmCGTS_SM_CTRL_REG, 0xffffffff, 0x96e00200,
303 	mmCP_RB_WPTR_POLL_CNTL, 0xffffffff, 0x00900100,
304 	mmRLC_CGCG_CGLS_CTRL, 0xffffffff, 0x0020003c,
305 	mmCP_MEM_SLP_CNTL, 0x00000001, 0x00000001,
306 };
307 
308 static const u32 golden_settings_vegam_a11[] =
309 {
310 	mmCB_HW_CONTROL, 0x0001f3cf, 0x00007208,
311 	mmCB_HW_CONTROL_2, 0x0f000000, 0x0d000000,
312 	mmCB_HW_CONTROL_3, 0x000001ff, 0x00000040,
313 	mmDB_DEBUG2, 0xf00fffff, 0x00000400,
314 	mmPA_SC_ENHANCE, 0xffffffff, 0x20000001,
315 	mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000,
316 	mmPA_SC_RASTER_CONFIG, 0x3f3fffff, 0x3a00161a,
317 	mmPA_SC_RASTER_CONFIG_1, 0x0000003f, 0x0000002e,
318 	mmRLC_CGCG_CGLS_CTRL, 0x00000003, 0x0001003c,
319 	mmRLC_CGCG_CGLS_CTRL_3D, 0xffffffff, 0x0001003c,
320 	mmSQ_CONFIG, 0x07f80000, 0x01180000,
321 	mmTA_CNTL_AUX, 0x000f000f, 0x000b0000,
322 	mmTCC_CTRL, 0x00100000, 0xf31fff7f,
323 	mmTCP_ADDR_CONFIG, 0x000003ff, 0x000000f7,
324 	mmTCP_CHAN_STEER_HI, 0xffffffff, 0x00000000,
325 	mmTCP_CHAN_STEER_LO, 0xffffffff, 0x32761054,
326 	mmVGT_RESET_DEBUG, 0x00000004, 0x00000004,
327 };
328 
329 static const u32 vegam_golden_common_all[] =
330 {
331 	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
332 	mmGB_ADDR_CONFIG, 0xffffffff, 0x22011003,
333 	mmSPI_RESOURCE_RESERVE_CU_0, 0xffffffff, 0x00000800,
334 	mmSPI_RESOURCE_RESERVE_CU_1, 0xffffffff, 0x00000800,
335 	mmSPI_RESOURCE_RESERVE_EN_CU_0, 0xffffffff, 0x00FF7FBF,
336 	mmSPI_RESOURCE_RESERVE_EN_CU_1, 0xffffffff, 0x00FF7FAF,
337 };
338 
339 static const u32 golden_settings_polaris11_a11[] =
340 {
341 	mmCB_HW_CONTROL, 0x0000f3cf, 0x00007208,
342 	mmCB_HW_CONTROL_2, 0x0f000000, 0x0f000000,
343 	mmCB_HW_CONTROL_3, 0x000001ff, 0x00000040,
344 	mmDB_DEBUG2, 0xf00fffff, 0x00000400,
345 	mmPA_SC_ENHANCE, 0xffffffff, 0x20000001,
346 	mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000,
347 	mmPA_SC_RASTER_CONFIG, 0x3f3fffff, 0x16000012,
348 	mmPA_SC_RASTER_CONFIG_1, 0x0000003f, 0x00000000,
349 	mmRLC_CGCG_CGLS_CTRL, 0x00000003, 0x0001003c,
350 	mmRLC_CGCG_CGLS_CTRL_3D, 0xffffffff, 0x0001003c,
351 	mmSQ_CONFIG, 0x07f80000, 0x01180000,
352 	mmTA_CNTL_AUX, 0x000f000f, 0x000b0000,
353 	mmTCC_CTRL, 0x00100000, 0xf31fff7f,
354 	mmTCP_ADDR_CONFIG, 0x000003ff, 0x000000f3,
355 	mmTCP_CHAN_STEER_HI, 0xffffffff, 0x00000000,
356 	mmTCP_CHAN_STEER_LO, 0xffffffff, 0x00003210,
357 	mmVGT_RESET_DEBUG, 0x00000004, 0x00000004,
358 };
359 
360 static const u32 polaris11_golden_common_all[] =
361 {
362 	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
363 	mmGB_ADDR_CONFIG, 0xffffffff, 0x22011002,
364 	mmSPI_RESOURCE_RESERVE_CU_0, 0xffffffff, 0x00000800,
365 	mmSPI_RESOURCE_RESERVE_CU_1, 0xffffffff, 0x00000800,
366 	mmSPI_RESOURCE_RESERVE_EN_CU_0, 0xffffffff, 0x00FF7FBF,
367 	mmSPI_RESOURCE_RESERVE_EN_CU_1, 0xffffffff, 0x00FF7FAF,
368 };
369 
370 static const u32 golden_settings_polaris10_a11[] =
371 {
372 	mmATC_MISC_CG, 0x000c0fc0, 0x000c0200,
373 	mmCB_HW_CONTROL, 0x0001f3cf, 0x00007208,
374 	mmCB_HW_CONTROL_2, 0x0f000000, 0x0f000000,
375 	mmCB_HW_CONTROL_3, 0x000001ff, 0x00000040,
376 	mmDB_DEBUG2, 0xf00fffff, 0x00000400,
377 	mmPA_SC_ENHANCE, 0xffffffff, 0x20000001,
378 	mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000,
379 	mmPA_SC_RASTER_CONFIG, 0x3f3fffff, 0x16000012,
380 	mmPA_SC_RASTER_CONFIG_1, 0x0000003f, 0x0000002a,
381 	mmRLC_CGCG_CGLS_CTRL, 0x00000003, 0x0001003c,
382 	mmRLC_CGCG_CGLS_CTRL_3D, 0xffffffff, 0x0001003c,
383 	mmSQ_CONFIG, 0x07f80000, 0x07180000,
384 	mmTA_CNTL_AUX, 0x000f000f, 0x000b0000,
385 	mmTCC_CTRL, 0x00100000, 0xf31fff7f,
386 	mmTCP_ADDR_CONFIG, 0x000003ff, 0x000000f7,
387 	mmTCP_CHAN_STEER_HI, 0xffffffff, 0x00000000,
388 	mmVGT_RESET_DEBUG, 0x00000004, 0x00000004,
389 };
390 
391 static const u32 polaris10_golden_common_all[] =
392 {
393 	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
394 	mmPA_SC_RASTER_CONFIG, 0xffffffff, 0x16000012,
395 	mmPA_SC_RASTER_CONFIG_1, 0xffffffff, 0x0000002A,
396 	mmGB_ADDR_CONFIG, 0xffffffff, 0x22011003,
397 	mmSPI_RESOURCE_RESERVE_CU_0, 0xffffffff, 0x00000800,
398 	mmSPI_RESOURCE_RESERVE_CU_1, 0xffffffff, 0x00000800,
399 	mmSPI_RESOURCE_RESERVE_EN_CU_0, 0xffffffff, 0x00FF7FBF,
400 	mmSPI_RESOURCE_RESERVE_EN_CU_1, 0xffffffff, 0x00FF7FAF,
401 };
402 
403 static const u32 fiji_golden_common_all[] =
404 {
405 	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
406 	mmPA_SC_RASTER_CONFIG, 0xffffffff, 0x3a00161a,
407 	mmPA_SC_RASTER_CONFIG_1, 0xffffffff, 0x0000002e,
408 	mmGB_ADDR_CONFIG, 0xffffffff, 0x22011003,
409 	mmSPI_RESOURCE_RESERVE_CU_0, 0xffffffff, 0x00000800,
410 	mmSPI_RESOURCE_RESERVE_CU_1, 0xffffffff, 0x00000800,
411 	mmSPI_RESOURCE_RESERVE_EN_CU_0, 0xffffffff, 0x00FF7FBF,
412 	mmSPI_RESOURCE_RESERVE_EN_CU_1, 0xffffffff, 0x00FF7FAF,
413 	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
414 	mmSPI_CONFIG_CNTL_1, 0x0000000f, 0x00000009,
415 };
416 
417 static const u32 golden_settings_fiji_a10[] =
418 {
419 	mmCB_HW_CONTROL_3, 0x000001ff, 0x00000040,
420 	mmDB_DEBUG2, 0xf00fffff, 0x00000400,
421 	mmPA_SC_ENHANCE, 0xffffffff, 0x20000001,
422 	mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000,
423 	mmRLC_CGCG_CGLS_CTRL, 0x00000003, 0x0001003c,
424 	mmSQ_RANDOM_WAVE_PRI, 0x001fffff, 0x000006fd,
425 	mmTA_CNTL_AUX, 0x000f000f, 0x000b0000,
426 	mmTCC_CTRL, 0x00100000, 0xf31fff7f,
427 	mmTCC_EXE_DISABLE, 0x00000002, 0x00000002,
428 	mmTCP_ADDR_CONFIG, 0x000003ff, 0x000000ff,
429 	mmVGT_RESET_DEBUG, 0x00000004, 0x00000004,
430 };
431 
432 static const u32 fiji_mgcg_cgcg_init[] =
433 {
434 	mmRLC_CGTT_MGCG_OVERRIDE, 0xffffffff, 0xffffffff,
435 	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
436 	mmCB_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100,
437 	mmCGTT_BCI_CLK_CTRL, 0xffffffff, 0x00000100,
438 	mmCGTT_CP_CLK_CTRL, 0xffffffff, 0x00000100,
439 	mmCGTT_CPC_CLK_CTRL, 0xffffffff, 0x00000100,
440 	mmCGTT_CPF_CLK_CTRL, 0xffffffff, 0x40000100,
441 	mmCGTT_GDS_CLK_CTRL, 0xffffffff, 0x00000100,
442 	mmCGTT_IA_CLK_CTRL, 0xffffffff, 0x06000100,
443 	mmCGTT_PA_CLK_CTRL, 0xffffffff, 0x00000100,
444 	mmCGTT_WD_CLK_CTRL, 0xffffffff, 0x06000100,
445 	mmCGTT_PC_CLK_CTRL, 0xffffffff, 0x00000100,
446 	mmCGTT_RLC_CLK_CTRL, 0xffffffff, 0x00000100,
447 	mmCGTT_SC_CLK_CTRL, 0xffffffff, 0x00000100,
448 	mmCGTT_SPI_CLK_CTRL, 0xffffffff, 0x00000100,
449 	mmCGTT_SQ_CLK_CTRL, 0xffffffff, 0x00000100,
450 	mmCGTT_SQG_CLK_CTRL, 0xffffffff, 0x00000100,
451 	mmCGTT_SX_CLK_CTRL0, 0xffffffff, 0x00000100,
452 	mmCGTT_SX_CLK_CTRL1, 0xffffffff, 0x00000100,
453 	mmCGTT_SX_CLK_CTRL2, 0xffffffff, 0x00000100,
454 	mmCGTT_SX_CLK_CTRL3, 0xffffffff, 0x00000100,
455 	mmCGTT_SX_CLK_CTRL4, 0xffffffff, 0x00000100,
456 	mmCGTT_TCI_CLK_CTRL, 0xffffffff, 0x00000100,
457 	mmCGTT_TCP_CLK_CTRL, 0xffffffff, 0x00000100,
458 	mmCGTT_VGT_CLK_CTRL, 0xffffffff, 0x06000100,
459 	mmDB_CGTT_CLK_CTRL_0, 0xffffffff, 0x00000100,
460 	mmTA_CGTT_CTRL, 0xffffffff, 0x00000100,
461 	mmTCA_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100,
462 	mmTCC_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100,
463 	mmTD_CGTT_CTRL, 0xffffffff, 0x00000100,
464 	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
465 	mmCGTS_SM_CTRL_REG, 0xffffffff, 0x96e00200,
466 	mmCP_RB_WPTR_POLL_CNTL, 0xffffffff, 0x00900100,
467 	mmRLC_CGCG_CGLS_CTRL, 0xffffffff, 0x0020003c,
468 	mmCP_MEM_SLP_CNTL, 0x00000001, 0x00000001,
469 };
470 
471 static const u32 golden_settings_iceland_a11[] =
472 {
473 	mmCB_HW_CONTROL_3, 0x00000040, 0x00000040,
474 	mmDB_DEBUG2, 0xf00fffff, 0x00000400,
475 	mmDB_DEBUG3, 0xc0000000, 0xc0000000,
476 	mmGB_GPU_ID, 0x0000000f, 0x00000000,
477 	mmPA_SC_ENHANCE, 0xffffffff, 0x20000001,
478 	mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000,
479 	mmPA_SC_RASTER_CONFIG, 0x3f3fffff, 0x00000002,
480 	mmPA_SC_RASTER_CONFIG_1, 0x0000003f, 0x00000000,
481 	mmRLC_CGCG_CGLS_CTRL, 0x00000003, 0x0000003c,
482 	mmSQ_RANDOM_WAVE_PRI, 0x001fffff, 0x000006fd,
483 	mmTA_CNTL_AUX, 0x000f000f, 0x000b0000,
484 	mmTCC_CTRL, 0x00100000, 0xf31fff7f,
485 	mmTCC_EXE_DISABLE, 0x00000002, 0x00000002,
486 	mmTCP_ADDR_CONFIG, 0x000003ff, 0x000000f1,
487 	mmTCP_CHAN_STEER_HI, 0xffffffff, 0x00000000,
488 	mmTCP_CHAN_STEER_LO, 0xffffffff, 0x00000010,
489 };
490 
491 static const u32 iceland_golden_common_all[] =
492 {
493 	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
494 	mmPA_SC_RASTER_CONFIG, 0xffffffff, 0x00000002,
495 	mmPA_SC_RASTER_CONFIG_1, 0xffffffff, 0x00000000,
496 	mmGB_ADDR_CONFIG, 0xffffffff, 0x22010001,
497 	mmSPI_RESOURCE_RESERVE_CU_0, 0xffffffff, 0x00000800,
498 	mmSPI_RESOURCE_RESERVE_CU_1, 0xffffffff, 0x00000800,
499 	mmSPI_RESOURCE_RESERVE_EN_CU_0, 0xffffffff, 0x00FF7FBF,
500 	mmSPI_RESOURCE_RESERVE_EN_CU_1, 0xffffffff, 0x00FF7FAF
501 };
502 
503 static const u32 iceland_mgcg_cgcg_init[] =
504 {
505 	mmRLC_CGTT_MGCG_OVERRIDE, 0xffffffff, 0xffffffff,
506 	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
507 	mmCB_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100,
508 	mmCGTT_BCI_CLK_CTRL, 0xffffffff, 0x00000100,
509 	mmCGTT_CP_CLK_CTRL, 0xffffffff, 0xc0000100,
510 	mmCGTT_CPC_CLK_CTRL, 0xffffffff, 0xc0000100,
511 	mmCGTT_CPF_CLK_CTRL, 0xffffffff, 0xc0000100,
512 	mmCGTT_GDS_CLK_CTRL, 0xffffffff, 0x00000100,
513 	mmCGTT_IA_CLK_CTRL, 0xffffffff, 0x06000100,
514 	mmCGTT_PA_CLK_CTRL, 0xffffffff, 0x00000100,
515 	mmCGTT_WD_CLK_CTRL, 0xffffffff, 0x06000100,
516 	mmCGTT_PC_CLK_CTRL, 0xffffffff, 0x00000100,
517 	mmCGTT_RLC_CLK_CTRL, 0xffffffff, 0x00000100,
518 	mmCGTT_SC_CLK_CTRL, 0xffffffff, 0x00000100,
519 	mmCGTT_SPI_CLK_CTRL, 0xffffffff, 0x00000100,
520 	mmCGTT_SQ_CLK_CTRL, 0xffffffff, 0x00000100,
521 	mmCGTT_SQG_CLK_CTRL, 0xffffffff, 0x00000100,
522 	mmCGTT_SX_CLK_CTRL0, 0xffffffff, 0x00000100,
523 	mmCGTT_SX_CLK_CTRL1, 0xffffffff, 0x00000100,
524 	mmCGTT_SX_CLK_CTRL2, 0xffffffff, 0x00000100,
525 	mmCGTT_SX_CLK_CTRL3, 0xffffffff, 0x00000100,
526 	mmCGTT_SX_CLK_CTRL4, 0xffffffff, 0x00000100,
527 	mmCGTT_TCI_CLK_CTRL, 0xffffffff, 0xff000100,
528 	mmCGTT_TCP_CLK_CTRL, 0xffffffff, 0x00000100,
529 	mmCGTT_VGT_CLK_CTRL, 0xffffffff, 0x06000100,
530 	mmDB_CGTT_CLK_CTRL_0, 0xffffffff, 0x00000100,
531 	mmTA_CGTT_CTRL, 0xffffffff, 0x00000100,
532 	mmTCA_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100,
533 	mmTCC_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100,
534 	mmTD_CGTT_CTRL, 0xffffffff, 0x00000100,
535 	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
536 	mmCGTS_CU0_SP0_CTRL_REG, 0xffffffff, 0x00010000,
537 	mmCGTS_CU0_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
538 	mmCGTS_CU0_TA_SQC_CTRL_REG, 0xffffffff, 0x0f840f87,
539 	mmCGTS_CU0_SP1_CTRL_REG, 0xffffffff, 0x00060005,
540 	mmCGTS_CU0_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
541 	mmCGTS_CU1_SP0_CTRL_REG, 0xffffffff, 0x00010000,
542 	mmCGTS_CU1_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
543 	mmCGTS_CU1_TA_CTRL_REG, 0xffffffff, 0x00040007,
544 	mmCGTS_CU1_SP1_CTRL_REG, 0xffffffff, 0x00060005,
545 	mmCGTS_CU1_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
546 	mmCGTS_CU2_SP0_CTRL_REG, 0xffffffff, 0x00010000,
547 	mmCGTS_CU2_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
548 	mmCGTS_CU2_TA_CTRL_REG, 0xffffffff, 0x00040007,
549 	mmCGTS_CU2_SP1_CTRL_REG, 0xffffffff, 0x00060005,
550 	mmCGTS_CU2_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
551 	mmCGTS_CU3_SP0_CTRL_REG, 0xffffffff, 0x00010000,
552 	mmCGTS_CU3_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
553 	mmCGTS_CU3_TA_CTRL_REG, 0xffffffff, 0x00040007,
554 	mmCGTS_CU3_SP1_CTRL_REG, 0xffffffff, 0x00060005,
555 	mmCGTS_CU3_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
556 	mmCGTS_CU4_SP0_CTRL_REG, 0xffffffff, 0x00010000,
557 	mmCGTS_CU4_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
558 	mmCGTS_CU4_TA_SQC_CTRL_REG, 0xffffffff, 0x0f840f87,
559 	mmCGTS_CU4_SP1_CTRL_REG, 0xffffffff, 0x00060005,
560 	mmCGTS_CU4_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
561 	mmCGTS_CU5_SP0_CTRL_REG, 0xffffffff, 0x00010000,
562 	mmCGTS_CU5_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
563 	mmCGTS_CU5_TA_CTRL_REG, 0xffffffff, 0x00040007,
564 	mmCGTS_CU5_SP1_CTRL_REG, 0xffffffff, 0x00060005,
565 	mmCGTS_CU5_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
566 	mmCGTS_SM_CTRL_REG, 0xffffffff, 0x96e00200,
567 	mmCP_RB_WPTR_POLL_CNTL, 0xffffffff, 0x00900100,
568 	mmRLC_CGCG_CGLS_CTRL, 0xffffffff, 0x0020003c,
569 };
570 
571 static const u32 cz_golden_settings_a11[] =
572 {
573 	mmCB_HW_CONTROL_3, 0x00000040, 0x00000040,
574 	mmDB_DEBUG2, 0xf00fffff, 0x00000400,
575 	mmGB_GPU_ID, 0x0000000f, 0x00000000,
576 	mmPA_SC_ENHANCE, 0xffffffff, 0x00000001,
577 	mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000,
578 	mmRLC_CGCG_CGLS_CTRL, 0x00000003, 0x0000003c,
579 	mmSQ_RANDOM_WAVE_PRI, 0x001fffff, 0x000006fd,
580 	mmTA_CNTL_AUX, 0x000f000f, 0x00010000,
581 	mmTCC_CTRL, 0x00100000, 0xf31fff7f,
582 	mmTCC_EXE_DISABLE, 0x00000002, 0x00000002,
583 	mmTCP_ADDR_CONFIG, 0x0000000f, 0x000000f3,
584 	mmTCP_CHAN_STEER_LO, 0xffffffff, 0x00001302
585 };
586 
587 static const u32 cz_golden_common_all[] =
588 {
589 	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
590 	mmPA_SC_RASTER_CONFIG, 0xffffffff, 0x00000002,
591 	mmPA_SC_RASTER_CONFIG_1, 0xffffffff, 0x00000000,
592 	mmGB_ADDR_CONFIG, 0xffffffff, 0x22010001,
593 	mmSPI_RESOURCE_RESERVE_CU_0, 0xffffffff, 0x00000800,
594 	mmSPI_RESOURCE_RESERVE_CU_1, 0xffffffff, 0x00000800,
595 	mmSPI_RESOURCE_RESERVE_EN_CU_0, 0xffffffff, 0x00FF7FBF,
596 	mmSPI_RESOURCE_RESERVE_EN_CU_1, 0xffffffff, 0x00FF7FAF
597 };
598 
599 static const u32 cz_mgcg_cgcg_init[] =
600 {
601 	mmRLC_CGTT_MGCG_OVERRIDE, 0xffffffff, 0xffffffff,
602 	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
603 	mmCB_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100,
604 	mmCGTT_BCI_CLK_CTRL, 0xffffffff, 0x00000100,
605 	mmCGTT_CP_CLK_CTRL, 0xffffffff, 0x00000100,
606 	mmCGTT_CPC_CLK_CTRL, 0xffffffff, 0x00000100,
607 	mmCGTT_CPF_CLK_CTRL, 0xffffffff, 0x00000100,
608 	mmCGTT_GDS_CLK_CTRL, 0xffffffff, 0x00000100,
609 	mmCGTT_IA_CLK_CTRL, 0xffffffff, 0x06000100,
610 	mmCGTT_PA_CLK_CTRL, 0xffffffff, 0x00000100,
611 	mmCGTT_WD_CLK_CTRL, 0xffffffff, 0x06000100,
612 	mmCGTT_PC_CLK_CTRL, 0xffffffff, 0x00000100,
613 	mmCGTT_RLC_CLK_CTRL, 0xffffffff, 0x00000100,
614 	mmCGTT_SC_CLK_CTRL, 0xffffffff, 0x00000100,
615 	mmCGTT_SPI_CLK_CTRL, 0xffffffff, 0x00000100,
616 	mmCGTT_SQ_CLK_CTRL, 0xffffffff, 0x00000100,
617 	mmCGTT_SQG_CLK_CTRL, 0xffffffff, 0x00000100,
618 	mmCGTT_SX_CLK_CTRL0, 0xffffffff, 0x00000100,
619 	mmCGTT_SX_CLK_CTRL1, 0xffffffff, 0x00000100,
620 	mmCGTT_SX_CLK_CTRL2, 0xffffffff, 0x00000100,
621 	mmCGTT_SX_CLK_CTRL3, 0xffffffff, 0x00000100,
622 	mmCGTT_SX_CLK_CTRL4, 0xffffffff, 0x00000100,
623 	mmCGTT_TCI_CLK_CTRL, 0xffffffff, 0x00000100,
624 	mmCGTT_TCP_CLK_CTRL, 0xffffffff, 0x00000100,
625 	mmCGTT_VGT_CLK_CTRL, 0xffffffff, 0x06000100,
626 	mmDB_CGTT_CLK_CTRL_0, 0xffffffff, 0x00000100,
627 	mmTA_CGTT_CTRL, 0xffffffff, 0x00000100,
628 	mmTCA_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100,
629 	mmTCC_CGTT_SCLK_CTRL, 0xffffffff, 0x00000100,
630 	mmTD_CGTT_CTRL, 0xffffffff, 0x00000100,
631 	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
632 	mmCGTS_CU0_SP0_CTRL_REG, 0xffffffff, 0x00010000,
633 	mmCGTS_CU0_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
634 	mmCGTS_CU0_TA_SQC_CTRL_REG, 0xffffffff, 0x00040007,
635 	mmCGTS_CU0_SP1_CTRL_REG, 0xffffffff, 0x00060005,
636 	mmCGTS_CU0_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
637 	mmCGTS_CU1_SP0_CTRL_REG, 0xffffffff, 0x00010000,
638 	mmCGTS_CU1_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
639 	mmCGTS_CU1_TA_CTRL_REG, 0xffffffff, 0x00040007,
640 	mmCGTS_CU1_SP1_CTRL_REG, 0xffffffff, 0x00060005,
641 	mmCGTS_CU1_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
642 	mmCGTS_CU2_SP0_CTRL_REG, 0xffffffff, 0x00010000,
643 	mmCGTS_CU2_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
644 	mmCGTS_CU2_TA_CTRL_REG, 0xffffffff, 0x00040007,
645 	mmCGTS_CU2_SP1_CTRL_REG, 0xffffffff, 0x00060005,
646 	mmCGTS_CU2_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
647 	mmCGTS_CU3_SP0_CTRL_REG, 0xffffffff, 0x00010000,
648 	mmCGTS_CU3_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
649 	mmCGTS_CU3_TA_CTRL_REG, 0xffffffff, 0x00040007,
650 	mmCGTS_CU3_SP1_CTRL_REG, 0xffffffff, 0x00060005,
651 	mmCGTS_CU3_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
652 	mmCGTS_CU4_SP0_CTRL_REG, 0xffffffff, 0x00010000,
653 	mmCGTS_CU4_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
654 	mmCGTS_CU4_TA_SQC_CTRL_REG, 0xffffffff, 0x00040007,
655 	mmCGTS_CU4_SP1_CTRL_REG, 0xffffffff, 0x00060005,
656 	mmCGTS_CU4_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
657 	mmCGTS_CU5_SP0_CTRL_REG, 0xffffffff, 0x00010000,
658 	mmCGTS_CU5_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
659 	mmCGTS_CU5_TA_CTRL_REG, 0xffffffff, 0x00040007,
660 	mmCGTS_CU5_SP1_CTRL_REG, 0xffffffff, 0x00060005,
661 	mmCGTS_CU5_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
662 	mmCGTS_CU6_SP0_CTRL_REG, 0xffffffff, 0x00010000,
663 	mmCGTS_CU6_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
664 	mmCGTS_CU6_TA_CTRL_REG, 0xffffffff, 0x00040007,
665 	mmCGTS_CU6_SP1_CTRL_REG, 0xffffffff, 0x00060005,
666 	mmCGTS_CU6_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
667 	mmCGTS_CU7_SP0_CTRL_REG, 0xffffffff, 0x00010000,
668 	mmCGTS_CU7_LDS_SQ_CTRL_REG, 0xffffffff, 0x00030002,
669 	mmCGTS_CU7_TA_CTRL_REG, 0xffffffff, 0x00040007,
670 	mmCGTS_CU7_SP1_CTRL_REG, 0xffffffff, 0x00060005,
671 	mmCGTS_CU7_TD_TCP_CTRL_REG, 0xffffffff, 0x00090008,
672 	mmCGTS_SM_CTRL_REG, 0xffffffff, 0x96e00200,
673 	mmCP_RB_WPTR_POLL_CNTL, 0xffffffff, 0x00900100,
674 	mmRLC_CGCG_CGLS_CTRL, 0xffffffff, 0x0020003f,
675 	mmCP_MEM_SLP_CNTL, 0x00000001, 0x00000001,
676 };
677 
678 static const u32 stoney_golden_settings_a11[] =
679 {
680 	mmDB_DEBUG2, 0xf00fffff, 0x00000400,
681 	mmGB_GPU_ID, 0x0000000f, 0x00000000,
682 	mmPA_SC_ENHANCE, 0xffffffff, 0x20000001,
683 	mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000,
684 	mmRLC_CGCG_CGLS_CTRL, 0x00000003, 0x0001003c,
685 	mmTA_CNTL_AUX, 0x000f000f, 0x000b0000,
686 	mmTCC_CTRL, 0x00100000, 0xf31fff7f,
687 	mmTCC_EXE_DISABLE, 0x00000002, 0x00000002,
688 	mmTCP_ADDR_CONFIG, 0x0000000f, 0x000000f1,
689 	mmTCP_CHAN_STEER_LO, 0xffffffff, 0x10101010,
690 };
691 
692 static const u32 stoney_golden_common_all[] =
693 {
694 	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
695 	mmPA_SC_RASTER_CONFIG, 0xffffffff, 0x00000000,
696 	mmPA_SC_RASTER_CONFIG_1, 0xffffffff, 0x00000000,
697 	mmGB_ADDR_CONFIG, 0xffffffff, 0x12010001,
698 	mmSPI_RESOURCE_RESERVE_CU_0, 0xffffffff, 0x00000800,
699 	mmSPI_RESOURCE_RESERVE_CU_1, 0xffffffff, 0x00000800,
700 	mmSPI_RESOURCE_RESERVE_EN_CU_0, 0xffffffff, 0x00FF7FBF,
701 	mmSPI_RESOURCE_RESERVE_EN_CU_1, 0xffffffff, 0x00FF7FAF,
702 };
703 
704 static const u32 stoney_mgcg_cgcg_init[] =
705 {
706 	mmGRBM_GFX_INDEX, 0xffffffff, 0xe0000000,
707 	mmRLC_CGCG_CGLS_CTRL, 0xffffffff, 0x0020003f,
708 	mmCP_MEM_SLP_CNTL, 0xffffffff, 0x00020201,
709 	mmRLC_MEM_SLP_CNTL, 0xffffffff, 0x00020201,
710 	mmCGTS_SM_CTRL_REG, 0xffffffff, 0x96940200,
711 };
712 
713 
714 static const char * const sq_edc_source_names[] = {
715 	"SQ_EDC_INFO_SOURCE_INVALID: No EDC error has occurred",
716 	"SQ_EDC_INFO_SOURCE_INST: EDC source is Instruction Fetch",
717 	"SQ_EDC_INFO_SOURCE_SGPR: EDC source is SGPR or SQC data return",
718 	"SQ_EDC_INFO_SOURCE_VGPR: EDC source is VGPR",
719 	"SQ_EDC_INFO_SOURCE_LDS: EDC source is LDS",
720 	"SQ_EDC_INFO_SOURCE_GDS: EDC source is GDS",
721 	"SQ_EDC_INFO_SOURCE_TA: EDC source is TA",
722 };
723 
724 static void gfx_v8_0_set_ring_funcs(struct amdgpu_device *adev);
725 static void gfx_v8_0_set_irq_funcs(struct amdgpu_device *adev);
726 static void gfx_v8_0_set_gds_init(struct amdgpu_device *adev);
727 static void gfx_v8_0_set_rlc_funcs(struct amdgpu_device *adev);
728 static u32 gfx_v8_0_get_csb_size(struct amdgpu_device *adev);
729 static void gfx_v8_0_get_cu_info(struct amdgpu_device *adev);
730 static void gfx_v8_0_ring_emit_ce_meta(struct amdgpu_ring *ring);
731 static void gfx_v8_0_ring_emit_de_meta(struct amdgpu_ring *ring);
732 
733 #define CG_ACLK_CNTL__ACLK_DIVIDER_MASK                    0x0000007fL
734 #define CG_ACLK_CNTL__ACLK_DIVIDER__SHIFT                  0x00000000L
735 
736 static void gfx_v8_0_init_golden_registers(struct amdgpu_device *adev)
737 {
738 	uint32_t data;
739 
740 	switch (adev->asic_type) {
741 	case CHIP_TOPAZ:
742 		amdgpu_device_program_register_sequence(adev,
743 							iceland_mgcg_cgcg_init,
744 							ARRAY_SIZE(iceland_mgcg_cgcg_init));
745 		amdgpu_device_program_register_sequence(adev,
746 							golden_settings_iceland_a11,
747 							ARRAY_SIZE(golden_settings_iceland_a11));
748 		amdgpu_device_program_register_sequence(adev,
749 							iceland_golden_common_all,
750 							ARRAY_SIZE(iceland_golden_common_all));
751 		break;
752 	case CHIP_FIJI:
753 		amdgpu_device_program_register_sequence(adev,
754 							fiji_mgcg_cgcg_init,
755 							ARRAY_SIZE(fiji_mgcg_cgcg_init));
756 		amdgpu_device_program_register_sequence(adev,
757 							golden_settings_fiji_a10,
758 							ARRAY_SIZE(golden_settings_fiji_a10));
759 		amdgpu_device_program_register_sequence(adev,
760 							fiji_golden_common_all,
761 							ARRAY_SIZE(fiji_golden_common_all));
762 		break;
763 
764 	case CHIP_TONGA:
765 		amdgpu_device_program_register_sequence(adev,
766 							tonga_mgcg_cgcg_init,
767 							ARRAY_SIZE(tonga_mgcg_cgcg_init));
768 		amdgpu_device_program_register_sequence(adev,
769 							golden_settings_tonga_a11,
770 							ARRAY_SIZE(golden_settings_tonga_a11));
771 		amdgpu_device_program_register_sequence(adev,
772 							tonga_golden_common_all,
773 							ARRAY_SIZE(tonga_golden_common_all));
774 		break;
775 	case CHIP_VEGAM:
776 		amdgpu_device_program_register_sequence(adev,
777 							golden_settings_vegam_a11,
778 							ARRAY_SIZE(golden_settings_vegam_a11));
779 		amdgpu_device_program_register_sequence(adev,
780 							vegam_golden_common_all,
781 							ARRAY_SIZE(vegam_golden_common_all));
782 		break;
783 	case CHIP_POLARIS11:
784 	case CHIP_POLARIS12:
785 		amdgpu_device_program_register_sequence(adev,
786 							golden_settings_polaris11_a11,
787 							ARRAY_SIZE(golden_settings_polaris11_a11));
788 		amdgpu_device_program_register_sequence(adev,
789 							polaris11_golden_common_all,
790 							ARRAY_SIZE(polaris11_golden_common_all));
791 		break;
792 	case CHIP_POLARIS10:
793 		amdgpu_device_program_register_sequence(adev,
794 							golden_settings_polaris10_a11,
795 							ARRAY_SIZE(golden_settings_polaris10_a11));
796 		amdgpu_device_program_register_sequence(adev,
797 							polaris10_golden_common_all,
798 							ARRAY_SIZE(polaris10_golden_common_all));
799 		data = RREG32_SMC(ixCG_ACLK_CNTL);
800 		data &= ~CG_ACLK_CNTL__ACLK_DIVIDER_MASK;
801 		data |= 0x18 << CG_ACLK_CNTL__ACLK_DIVIDER__SHIFT;
802 		WREG32_SMC(ixCG_ACLK_CNTL, data);
803 		if ((adev->pdev->device == 0x67DF) && (adev->pdev->revision == 0xc7) &&
804 		    ((adev->pdev->subsystem_device == 0xb37 && adev->pdev->subsystem_vendor == 0x1002) ||
805 		     (adev->pdev->subsystem_device == 0x4a8 && adev->pdev->subsystem_vendor == 0x1043) ||
806 		     (adev->pdev->subsystem_device == 0x9480 && adev->pdev->subsystem_vendor == 0x1680))) {
807 			amdgpu_atombios_i2c_channel_trans(adev, 0x10, 0x96, 0x1E, 0xDD);
808 			amdgpu_atombios_i2c_channel_trans(adev, 0x10, 0x96, 0x1F, 0xD0);
809 		}
810 		break;
811 	case CHIP_CARRIZO:
812 		amdgpu_device_program_register_sequence(adev,
813 							cz_mgcg_cgcg_init,
814 							ARRAY_SIZE(cz_mgcg_cgcg_init));
815 		amdgpu_device_program_register_sequence(adev,
816 							cz_golden_settings_a11,
817 							ARRAY_SIZE(cz_golden_settings_a11));
818 		amdgpu_device_program_register_sequence(adev,
819 							cz_golden_common_all,
820 							ARRAY_SIZE(cz_golden_common_all));
821 		break;
822 	case CHIP_STONEY:
823 		amdgpu_device_program_register_sequence(adev,
824 							stoney_mgcg_cgcg_init,
825 							ARRAY_SIZE(stoney_mgcg_cgcg_init));
826 		amdgpu_device_program_register_sequence(adev,
827 							stoney_golden_settings_a11,
828 							ARRAY_SIZE(stoney_golden_settings_a11));
829 		amdgpu_device_program_register_sequence(adev,
830 							stoney_golden_common_all,
831 							ARRAY_SIZE(stoney_golden_common_all));
832 		break;
833 	default:
834 		break;
835 	}
836 }
837 
838 static int gfx_v8_0_ring_test_ring(struct amdgpu_ring *ring)
839 {
840 	struct amdgpu_device *adev = ring->adev;
841 	uint32_t tmp = 0;
842 	unsigned i;
843 	int r;
844 
845 	WREG32(mmSCRATCH_REG0, 0xCAFEDEAD);
846 	r = amdgpu_ring_alloc(ring, 3);
847 	if (r)
848 		return r;
849 
850 	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_UCONFIG_REG, 1));
851 	amdgpu_ring_write(ring, mmSCRATCH_REG0 - PACKET3_SET_UCONFIG_REG_START);
852 	amdgpu_ring_write(ring, 0xDEADBEEF);
853 	amdgpu_ring_commit(ring);
854 
855 	for (i = 0; i < adev->usec_timeout; i++) {
856 		tmp = RREG32(mmSCRATCH_REG0);
857 		if (tmp == 0xDEADBEEF)
858 			break;
859 		udelay(1);
860 	}
861 
862 	if (i >= adev->usec_timeout)
863 		r = -ETIMEDOUT;
864 
865 	return r;
866 }
867 
868 static int gfx_v8_0_ring_test_ib(struct amdgpu_ring *ring, long timeout)
869 {
870 	struct amdgpu_device *adev = ring->adev;
871 	struct amdgpu_ib ib;
872 	struct dma_fence *f = NULL;
873 
874 	unsigned int index;
875 	uint64_t gpu_addr;
876 	uint32_t tmp;
877 	long r;
878 
879 	r = amdgpu_device_wb_get(adev, &index);
880 	if (r)
881 		return r;
882 
883 	gpu_addr = adev->wb.gpu_addr + (index * 4);
884 	adev->wb.wb[index] = cpu_to_le32(0xCAFEDEAD);
885 	memset(&ib, 0, sizeof(ib));
886 
887 	r = amdgpu_ib_get(adev, NULL, 20, AMDGPU_IB_POOL_DIRECT, &ib);
888 	if (r)
889 		goto err1;
890 
891 	ib.ptr[0] = PACKET3(PACKET3_WRITE_DATA, 3);
892 	ib.ptr[1] = WRITE_DATA_DST_SEL(5) | WR_CONFIRM;
893 	ib.ptr[2] = lower_32_bits(gpu_addr);
894 	ib.ptr[3] = upper_32_bits(gpu_addr);
895 	ib.ptr[4] = 0xDEADBEEF;
896 	ib.length_dw = 5;
897 
898 	r = amdgpu_ib_schedule(ring, 1, &ib, NULL, &f);
899 	if (r)
900 		goto err2;
901 
902 	r = dma_fence_wait_timeout(f, false, timeout);
903 	if (r == 0) {
904 		r = -ETIMEDOUT;
905 		goto err2;
906 	} else if (r < 0) {
907 		goto err2;
908 	}
909 
910 	tmp = adev->wb.wb[index];
911 	if (tmp == 0xDEADBEEF)
912 		r = 0;
913 	else
914 		r = -EINVAL;
915 
916 err2:
917 	amdgpu_ib_free(adev, &ib, NULL);
918 	dma_fence_put(f);
919 err1:
920 	amdgpu_device_wb_free(adev, index);
921 	return r;
922 }
923 
924 
925 static void gfx_v8_0_free_microcode(struct amdgpu_device *adev)
926 {
927 	amdgpu_ucode_release(&adev->gfx.pfp_fw);
928 	amdgpu_ucode_release(&adev->gfx.me_fw);
929 	amdgpu_ucode_release(&adev->gfx.ce_fw);
930 	amdgpu_ucode_release(&adev->gfx.rlc_fw);
931 	amdgpu_ucode_release(&adev->gfx.mec_fw);
932 	if ((adev->asic_type != CHIP_STONEY) &&
933 	    (adev->asic_type != CHIP_TOPAZ))
934 		amdgpu_ucode_release(&adev->gfx.mec2_fw);
935 
936 	kfree(adev->gfx.rlc.register_list_format);
937 }
938 
939 static int gfx_v8_0_init_microcode(struct amdgpu_device *adev)
940 {
941 	const char *chip_name;
942 	int err;
943 	struct amdgpu_firmware_info *info = NULL;
944 	const struct common_firmware_header *header = NULL;
945 	const struct gfx_firmware_header_v1_0 *cp_hdr;
946 	const struct rlc_firmware_header_v2_0 *rlc_hdr;
947 	unsigned int *tmp = NULL, i;
948 
949 	DRM_DEBUG("\n");
950 
951 	switch (adev->asic_type) {
952 	case CHIP_TOPAZ:
953 		chip_name = "topaz";
954 		break;
955 	case CHIP_TONGA:
956 		chip_name = "tonga";
957 		break;
958 	case CHIP_CARRIZO:
959 		chip_name = "carrizo";
960 		break;
961 	case CHIP_FIJI:
962 		chip_name = "fiji";
963 		break;
964 	case CHIP_STONEY:
965 		chip_name = "stoney";
966 		break;
967 	case CHIP_POLARIS10:
968 		chip_name = "polaris10";
969 		break;
970 	case CHIP_POLARIS11:
971 		chip_name = "polaris11";
972 		break;
973 	case CHIP_POLARIS12:
974 		chip_name = "polaris12";
975 		break;
976 	case CHIP_VEGAM:
977 		chip_name = "vegam";
978 		break;
979 	default:
980 		BUG();
981 	}
982 
983 	if (adev->asic_type >= CHIP_POLARIS10 && adev->asic_type <= CHIP_POLARIS12) {
984 		err = amdgpu_ucode_request(adev, &adev->gfx.pfp_fw,
985 					   "amdgpu/%s_pfp_2.bin", chip_name);
986 		if (err == -ENODEV) {
987 			err = amdgpu_ucode_request(adev, &adev->gfx.pfp_fw,
988 						   "amdgpu/%s_pfp.bin", chip_name);
989 		}
990 	} else {
991 		err = amdgpu_ucode_request(adev, &adev->gfx.pfp_fw,
992 					   "amdgpu/%s_pfp.bin", chip_name);
993 	}
994 	if (err)
995 		goto out;
996 	cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.pfp_fw->data;
997 	adev->gfx.pfp_fw_version = le32_to_cpu(cp_hdr->header.ucode_version);
998 	adev->gfx.pfp_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version);
999 
1000 	if (adev->asic_type >= CHIP_POLARIS10 && adev->asic_type <= CHIP_POLARIS12) {
1001 		err = amdgpu_ucode_request(adev, &adev->gfx.me_fw,
1002 					   "amdgpu/%s_me_2.bin", chip_name);
1003 		if (err == -ENODEV) {
1004 			err = amdgpu_ucode_request(adev, &adev->gfx.me_fw,
1005 						   "amdgpu/%s_me.bin", chip_name);
1006 		}
1007 	} else {
1008 		err = amdgpu_ucode_request(adev, &adev->gfx.me_fw,
1009 					   "amdgpu/%s_me.bin", chip_name);
1010 	}
1011 	if (err)
1012 		goto out;
1013 	cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.me_fw->data;
1014 	adev->gfx.me_fw_version = le32_to_cpu(cp_hdr->header.ucode_version);
1015 
1016 	adev->gfx.me_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version);
1017 
1018 	if (adev->asic_type >= CHIP_POLARIS10 && adev->asic_type <= CHIP_POLARIS12) {
1019 		err = amdgpu_ucode_request(adev, &adev->gfx.ce_fw,
1020 					   "amdgpu/%s_ce_2.bin", chip_name);
1021 		if (err == -ENODEV) {
1022 			err = amdgpu_ucode_request(adev, &adev->gfx.ce_fw,
1023 						   "amdgpu/%s_ce.bin", chip_name);
1024 		}
1025 	} else {
1026 		err = amdgpu_ucode_request(adev, &adev->gfx.ce_fw,
1027 					   "amdgpu/%s_ce.bin", chip_name);
1028 	}
1029 	if (err)
1030 		goto out;
1031 	cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.ce_fw->data;
1032 	adev->gfx.ce_fw_version = le32_to_cpu(cp_hdr->header.ucode_version);
1033 	adev->gfx.ce_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version);
1034 
1035 	/*
1036 	 * Support for MCBP/Virtualization in combination with chained IBs is
1037 	 * formal released on feature version #46
1038 	 */
1039 	if (adev->gfx.ce_feature_version >= 46 &&
1040 	    adev->gfx.pfp_feature_version >= 46) {
1041 		adev->virt.chained_ib_support = true;
1042 		DRM_INFO("Chained IB support enabled!\n");
1043 	} else
1044 		adev->virt.chained_ib_support = false;
1045 
1046 	err = amdgpu_ucode_request(adev, &adev->gfx.rlc_fw,
1047 				   "amdgpu/%s_rlc.bin", chip_name);
1048 	if (err)
1049 		goto out;
1050 	rlc_hdr = (const struct rlc_firmware_header_v2_0 *)adev->gfx.rlc_fw->data;
1051 	adev->gfx.rlc_fw_version = le32_to_cpu(rlc_hdr->header.ucode_version);
1052 	adev->gfx.rlc_feature_version = le32_to_cpu(rlc_hdr->ucode_feature_version);
1053 
1054 	adev->gfx.rlc.save_and_restore_offset =
1055 			le32_to_cpu(rlc_hdr->save_and_restore_offset);
1056 	adev->gfx.rlc.clear_state_descriptor_offset =
1057 			le32_to_cpu(rlc_hdr->clear_state_descriptor_offset);
1058 	adev->gfx.rlc.avail_scratch_ram_locations =
1059 			le32_to_cpu(rlc_hdr->avail_scratch_ram_locations);
1060 	adev->gfx.rlc.reg_restore_list_size =
1061 			le32_to_cpu(rlc_hdr->reg_restore_list_size);
1062 	adev->gfx.rlc.reg_list_format_start =
1063 			le32_to_cpu(rlc_hdr->reg_list_format_start);
1064 	adev->gfx.rlc.reg_list_format_separate_start =
1065 			le32_to_cpu(rlc_hdr->reg_list_format_separate_start);
1066 	adev->gfx.rlc.starting_offsets_start =
1067 			le32_to_cpu(rlc_hdr->starting_offsets_start);
1068 	adev->gfx.rlc.reg_list_format_size_bytes =
1069 			le32_to_cpu(rlc_hdr->reg_list_format_size_bytes);
1070 	adev->gfx.rlc.reg_list_size_bytes =
1071 			le32_to_cpu(rlc_hdr->reg_list_size_bytes);
1072 
1073 	adev->gfx.rlc.register_list_format =
1074 			kmalloc(adev->gfx.rlc.reg_list_format_size_bytes +
1075 					adev->gfx.rlc.reg_list_size_bytes, GFP_KERNEL);
1076 
1077 	if (!adev->gfx.rlc.register_list_format) {
1078 		err = -ENOMEM;
1079 		goto out;
1080 	}
1081 
1082 	tmp = (unsigned int *)((uintptr_t)rlc_hdr +
1083 			le32_to_cpu(rlc_hdr->reg_list_format_array_offset_bytes));
1084 	for (i = 0 ; i < (adev->gfx.rlc.reg_list_format_size_bytes >> 2); i++)
1085 		adev->gfx.rlc.register_list_format[i] =	le32_to_cpu(tmp[i]);
1086 
1087 	adev->gfx.rlc.register_restore = adev->gfx.rlc.register_list_format + i;
1088 
1089 	tmp = (unsigned int *)((uintptr_t)rlc_hdr +
1090 			le32_to_cpu(rlc_hdr->reg_list_array_offset_bytes));
1091 	for (i = 0 ; i < (adev->gfx.rlc.reg_list_size_bytes >> 2); i++)
1092 		adev->gfx.rlc.register_restore[i] = le32_to_cpu(tmp[i]);
1093 
1094 	if (adev->asic_type >= CHIP_POLARIS10 && adev->asic_type <= CHIP_POLARIS12) {
1095 		err = amdgpu_ucode_request(adev, &adev->gfx.mec_fw,
1096 					   "amdgpu/%s_mec_2.bin", chip_name);
1097 		if (err == -ENODEV) {
1098 			err = amdgpu_ucode_request(adev, &adev->gfx.mec_fw,
1099 						   "amdgpu/%s_mec.bin", chip_name);
1100 		}
1101 	} else {
1102 		err = amdgpu_ucode_request(adev, &adev->gfx.mec_fw,
1103 					   "amdgpu/%s_mec.bin", chip_name);
1104 	}
1105 	if (err)
1106 		goto out;
1107 	cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data;
1108 	adev->gfx.mec_fw_version = le32_to_cpu(cp_hdr->header.ucode_version);
1109 	adev->gfx.mec_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version);
1110 
1111 	if ((adev->asic_type != CHIP_STONEY) &&
1112 	    (adev->asic_type != CHIP_TOPAZ)) {
1113 		if (adev->asic_type >= CHIP_POLARIS10 && adev->asic_type <= CHIP_POLARIS12) {
1114 			err = amdgpu_ucode_request(adev, &adev->gfx.mec2_fw,
1115 						   "amdgpu/%s_mec2_2.bin", chip_name);
1116 			if (err == -ENODEV) {
1117 				err = amdgpu_ucode_request(adev, &adev->gfx.mec2_fw,
1118 							   "amdgpu/%s_mec2.bin", chip_name);
1119 			}
1120 		} else {
1121 			err = amdgpu_ucode_request(adev, &adev->gfx.mec2_fw,
1122 						   "amdgpu/%s_mec2.bin", chip_name);
1123 		}
1124 		if (!err) {
1125 			cp_hdr = (const struct gfx_firmware_header_v1_0 *)
1126 				adev->gfx.mec2_fw->data;
1127 			adev->gfx.mec2_fw_version =
1128 				le32_to_cpu(cp_hdr->header.ucode_version);
1129 			adev->gfx.mec2_feature_version =
1130 				le32_to_cpu(cp_hdr->ucode_feature_version);
1131 		} else {
1132 			err = 0;
1133 			adev->gfx.mec2_fw = NULL;
1134 		}
1135 	}
1136 
1137 	info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_PFP];
1138 	info->ucode_id = AMDGPU_UCODE_ID_CP_PFP;
1139 	info->fw = adev->gfx.pfp_fw;
1140 	header = (const struct common_firmware_header *)info->fw->data;
1141 	adev->firmware.fw_size +=
1142 		ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
1143 
1144 	info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_ME];
1145 	info->ucode_id = AMDGPU_UCODE_ID_CP_ME;
1146 	info->fw = adev->gfx.me_fw;
1147 	header = (const struct common_firmware_header *)info->fw->data;
1148 	adev->firmware.fw_size +=
1149 		ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
1150 
1151 	info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_CE];
1152 	info->ucode_id = AMDGPU_UCODE_ID_CP_CE;
1153 	info->fw = adev->gfx.ce_fw;
1154 	header = (const struct common_firmware_header *)info->fw->data;
1155 	adev->firmware.fw_size +=
1156 		ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
1157 
1158 	info = &adev->firmware.ucode[AMDGPU_UCODE_ID_RLC_G];
1159 	info->ucode_id = AMDGPU_UCODE_ID_RLC_G;
1160 	info->fw = adev->gfx.rlc_fw;
1161 	header = (const struct common_firmware_header *)info->fw->data;
1162 	adev->firmware.fw_size +=
1163 		ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
1164 
1165 	info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_MEC1];
1166 	info->ucode_id = AMDGPU_UCODE_ID_CP_MEC1;
1167 	info->fw = adev->gfx.mec_fw;
1168 	header = (const struct common_firmware_header *)info->fw->data;
1169 	adev->firmware.fw_size +=
1170 		ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
1171 
1172 	/* we need account JT in */
1173 	cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data;
1174 	adev->firmware.fw_size +=
1175 		ALIGN(le32_to_cpu(cp_hdr->jt_size) << 2, PAGE_SIZE);
1176 
1177 	if (amdgpu_sriov_vf(adev)) {
1178 		info = &adev->firmware.ucode[AMDGPU_UCODE_ID_STORAGE];
1179 		info->ucode_id = AMDGPU_UCODE_ID_STORAGE;
1180 		info->fw = adev->gfx.mec_fw;
1181 		adev->firmware.fw_size +=
1182 			ALIGN(le32_to_cpu(64 * PAGE_SIZE), PAGE_SIZE);
1183 	}
1184 
1185 	if (adev->gfx.mec2_fw) {
1186 		info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_MEC2];
1187 		info->ucode_id = AMDGPU_UCODE_ID_CP_MEC2;
1188 		info->fw = adev->gfx.mec2_fw;
1189 		header = (const struct common_firmware_header *)info->fw->data;
1190 		adev->firmware.fw_size +=
1191 			ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
1192 	}
1193 
1194 out:
1195 	if (err) {
1196 		dev_err(adev->dev, "gfx8: Failed to load firmware %s gfx firmware\n", chip_name);
1197 		amdgpu_ucode_release(&adev->gfx.pfp_fw);
1198 		amdgpu_ucode_release(&adev->gfx.me_fw);
1199 		amdgpu_ucode_release(&adev->gfx.ce_fw);
1200 		amdgpu_ucode_release(&adev->gfx.rlc_fw);
1201 		amdgpu_ucode_release(&adev->gfx.mec_fw);
1202 		amdgpu_ucode_release(&adev->gfx.mec2_fw);
1203 	}
1204 	return err;
1205 }
1206 
1207 static void gfx_v8_0_get_csb_buffer(struct amdgpu_device *adev,
1208 				    volatile u32 *buffer)
1209 {
1210 	u32 count = 0, i;
1211 	const struct cs_section_def *sect = NULL;
1212 	const struct cs_extent_def *ext = NULL;
1213 
1214 	if (adev->gfx.rlc.cs_data == NULL)
1215 		return;
1216 	if (buffer == NULL)
1217 		return;
1218 
1219 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0));
1220 	buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
1221 
1222 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CONTEXT_CONTROL, 1));
1223 	buffer[count++] = cpu_to_le32(0x80000000);
1224 	buffer[count++] = cpu_to_le32(0x80000000);
1225 
1226 	for (sect = adev->gfx.rlc.cs_data; sect->section != NULL; ++sect) {
1227 		for (ext = sect->section; ext->extent != NULL; ++ext) {
1228 			if (sect->id == SECT_CONTEXT) {
1229 				buffer[count++] =
1230 					cpu_to_le32(PACKET3(PACKET3_SET_CONTEXT_REG, ext->reg_count));
1231 				buffer[count++] = cpu_to_le32(ext->reg_index -
1232 						PACKET3_SET_CONTEXT_REG_START);
1233 				for (i = 0; i < ext->reg_count; i++)
1234 					buffer[count++] = cpu_to_le32(ext->extent[i]);
1235 			} else {
1236 				return;
1237 			}
1238 		}
1239 	}
1240 
1241 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_SET_CONTEXT_REG, 2));
1242 	buffer[count++] = cpu_to_le32(mmPA_SC_RASTER_CONFIG -
1243 			PACKET3_SET_CONTEXT_REG_START);
1244 	buffer[count++] = cpu_to_le32(adev->gfx.config.rb_config[0][0].raster_config);
1245 	buffer[count++] = cpu_to_le32(adev->gfx.config.rb_config[0][0].raster_config_1);
1246 
1247 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0));
1248 	buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_END_CLEAR_STATE);
1249 
1250 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CLEAR_STATE, 0));
1251 	buffer[count++] = cpu_to_le32(0);
1252 }
1253 
1254 static int gfx_v8_0_cp_jump_table_num(struct amdgpu_device *adev)
1255 {
1256 	if (adev->asic_type == CHIP_CARRIZO)
1257 		return 5;
1258 	else
1259 		return 4;
1260 }
1261 
1262 static int gfx_v8_0_rlc_init(struct amdgpu_device *adev)
1263 {
1264 	const struct cs_section_def *cs_data;
1265 	int r;
1266 
1267 	adev->gfx.rlc.cs_data = vi_cs_data;
1268 
1269 	cs_data = adev->gfx.rlc.cs_data;
1270 
1271 	if (cs_data) {
1272 		/* init clear state block */
1273 		r = amdgpu_gfx_rlc_init_csb(adev);
1274 		if (r)
1275 			return r;
1276 	}
1277 
1278 	if ((adev->asic_type == CHIP_CARRIZO) ||
1279 	    (adev->asic_type == CHIP_STONEY)) {
1280 		adev->gfx.rlc.cp_table_size = ALIGN(96 * 5 * 4, 2048) + (64 * 1024); /* JT + GDS */
1281 		r = amdgpu_gfx_rlc_init_cpt(adev);
1282 		if (r)
1283 			return r;
1284 	}
1285 
1286 	/* init spm vmid with 0xf */
1287 	if (adev->gfx.rlc.funcs->update_spm_vmid)
1288 		adev->gfx.rlc.funcs->update_spm_vmid(adev, NULL, 0xf);
1289 
1290 	return 0;
1291 }
1292 
1293 static void gfx_v8_0_mec_fini(struct amdgpu_device *adev)
1294 {
1295 	amdgpu_bo_free_kernel(&adev->gfx.mec.hpd_eop_obj, NULL, NULL);
1296 }
1297 
1298 static int gfx_v8_0_mec_init(struct amdgpu_device *adev)
1299 {
1300 	int r;
1301 	u32 *hpd;
1302 	size_t mec_hpd_size;
1303 
1304 	bitmap_zero(adev->gfx.mec_bitmap[0].queue_bitmap, AMDGPU_MAX_COMPUTE_QUEUES);
1305 
1306 	/* take ownership of the relevant compute queues */
1307 	amdgpu_gfx_compute_queue_acquire(adev);
1308 
1309 	mec_hpd_size = adev->gfx.num_compute_rings * GFX8_MEC_HPD_SIZE;
1310 	if (mec_hpd_size) {
1311 		r = amdgpu_bo_create_reserved(adev, mec_hpd_size, PAGE_SIZE,
1312 					      AMDGPU_GEM_DOMAIN_VRAM |
1313 					      AMDGPU_GEM_DOMAIN_GTT,
1314 					      &adev->gfx.mec.hpd_eop_obj,
1315 					      &adev->gfx.mec.hpd_eop_gpu_addr,
1316 					      (void **)&hpd);
1317 		if (r) {
1318 			dev_warn(adev->dev, "(%d) create HDP EOP bo failed\n", r);
1319 			return r;
1320 		}
1321 
1322 		memset(hpd, 0, mec_hpd_size);
1323 
1324 		amdgpu_bo_kunmap(adev->gfx.mec.hpd_eop_obj);
1325 		amdgpu_bo_unreserve(adev->gfx.mec.hpd_eop_obj);
1326 	}
1327 
1328 	return 0;
1329 }
1330 
1331 static const u32 vgpr_init_compute_shader[] =
1332 {
1333 	0x7e000209, 0x7e020208,
1334 	0x7e040207, 0x7e060206,
1335 	0x7e080205, 0x7e0a0204,
1336 	0x7e0c0203, 0x7e0e0202,
1337 	0x7e100201, 0x7e120200,
1338 	0x7e140209, 0x7e160208,
1339 	0x7e180207, 0x7e1a0206,
1340 	0x7e1c0205, 0x7e1e0204,
1341 	0x7e200203, 0x7e220202,
1342 	0x7e240201, 0x7e260200,
1343 	0x7e280209, 0x7e2a0208,
1344 	0x7e2c0207, 0x7e2e0206,
1345 	0x7e300205, 0x7e320204,
1346 	0x7e340203, 0x7e360202,
1347 	0x7e380201, 0x7e3a0200,
1348 	0x7e3c0209, 0x7e3e0208,
1349 	0x7e400207, 0x7e420206,
1350 	0x7e440205, 0x7e460204,
1351 	0x7e480203, 0x7e4a0202,
1352 	0x7e4c0201, 0x7e4e0200,
1353 	0x7e500209, 0x7e520208,
1354 	0x7e540207, 0x7e560206,
1355 	0x7e580205, 0x7e5a0204,
1356 	0x7e5c0203, 0x7e5e0202,
1357 	0x7e600201, 0x7e620200,
1358 	0x7e640209, 0x7e660208,
1359 	0x7e680207, 0x7e6a0206,
1360 	0x7e6c0205, 0x7e6e0204,
1361 	0x7e700203, 0x7e720202,
1362 	0x7e740201, 0x7e760200,
1363 	0x7e780209, 0x7e7a0208,
1364 	0x7e7c0207, 0x7e7e0206,
1365 	0xbf8a0000, 0xbf810000,
1366 };
1367 
1368 static const u32 sgpr_init_compute_shader[] =
1369 {
1370 	0xbe8a0100, 0xbe8c0102,
1371 	0xbe8e0104, 0xbe900106,
1372 	0xbe920108, 0xbe940100,
1373 	0xbe960102, 0xbe980104,
1374 	0xbe9a0106, 0xbe9c0108,
1375 	0xbe9e0100, 0xbea00102,
1376 	0xbea20104, 0xbea40106,
1377 	0xbea60108, 0xbea80100,
1378 	0xbeaa0102, 0xbeac0104,
1379 	0xbeae0106, 0xbeb00108,
1380 	0xbeb20100, 0xbeb40102,
1381 	0xbeb60104, 0xbeb80106,
1382 	0xbeba0108, 0xbebc0100,
1383 	0xbebe0102, 0xbec00104,
1384 	0xbec20106, 0xbec40108,
1385 	0xbec60100, 0xbec80102,
1386 	0xbee60004, 0xbee70005,
1387 	0xbeea0006, 0xbeeb0007,
1388 	0xbee80008, 0xbee90009,
1389 	0xbefc0000, 0xbf8a0000,
1390 	0xbf810000, 0x00000000,
1391 };
1392 
1393 static const u32 vgpr_init_regs[] =
1394 {
1395 	mmCOMPUTE_STATIC_THREAD_MGMT_SE0, 0xffffffff,
1396 	mmCOMPUTE_RESOURCE_LIMITS, 0x1000000, /* CU_GROUP_COUNT=1 */
1397 	mmCOMPUTE_NUM_THREAD_X, 256*4,
1398 	mmCOMPUTE_NUM_THREAD_Y, 1,
1399 	mmCOMPUTE_NUM_THREAD_Z, 1,
1400 	mmCOMPUTE_PGM_RSRC1, 0x100004f, /* VGPRS=15 (64 logical VGPRs), SGPRS=1 (16 SGPRs), BULKY=1 */
1401 	mmCOMPUTE_PGM_RSRC2, 20,
1402 	mmCOMPUTE_USER_DATA_0, 0xedcedc00,
1403 	mmCOMPUTE_USER_DATA_1, 0xedcedc01,
1404 	mmCOMPUTE_USER_DATA_2, 0xedcedc02,
1405 	mmCOMPUTE_USER_DATA_3, 0xedcedc03,
1406 	mmCOMPUTE_USER_DATA_4, 0xedcedc04,
1407 	mmCOMPUTE_USER_DATA_5, 0xedcedc05,
1408 	mmCOMPUTE_USER_DATA_6, 0xedcedc06,
1409 	mmCOMPUTE_USER_DATA_7, 0xedcedc07,
1410 	mmCOMPUTE_USER_DATA_8, 0xedcedc08,
1411 	mmCOMPUTE_USER_DATA_9, 0xedcedc09,
1412 };
1413 
1414 static const u32 sgpr1_init_regs[] =
1415 {
1416 	mmCOMPUTE_STATIC_THREAD_MGMT_SE0, 0x0f,
1417 	mmCOMPUTE_RESOURCE_LIMITS, 0x1000000, /* CU_GROUP_COUNT=1 */
1418 	mmCOMPUTE_NUM_THREAD_X, 256*5,
1419 	mmCOMPUTE_NUM_THREAD_Y, 1,
1420 	mmCOMPUTE_NUM_THREAD_Z, 1,
1421 	mmCOMPUTE_PGM_RSRC1, 0x240, /* SGPRS=9 (80 GPRS) */
1422 	mmCOMPUTE_PGM_RSRC2, 20,
1423 	mmCOMPUTE_USER_DATA_0, 0xedcedc00,
1424 	mmCOMPUTE_USER_DATA_1, 0xedcedc01,
1425 	mmCOMPUTE_USER_DATA_2, 0xedcedc02,
1426 	mmCOMPUTE_USER_DATA_3, 0xedcedc03,
1427 	mmCOMPUTE_USER_DATA_4, 0xedcedc04,
1428 	mmCOMPUTE_USER_DATA_5, 0xedcedc05,
1429 	mmCOMPUTE_USER_DATA_6, 0xedcedc06,
1430 	mmCOMPUTE_USER_DATA_7, 0xedcedc07,
1431 	mmCOMPUTE_USER_DATA_8, 0xedcedc08,
1432 	mmCOMPUTE_USER_DATA_9, 0xedcedc09,
1433 };
1434 
1435 static const u32 sgpr2_init_regs[] =
1436 {
1437 	mmCOMPUTE_STATIC_THREAD_MGMT_SE0, 0xf0,
1438 	mmCOMPUTE_RESOURCE_LIMITS, 0x1000000,
1439 	mmCOMPUTE_NUM_THREAD_X, 256*5,
1440 	mmCOMPUTE_NUM_THREAD_Y, 1,
1441 	mmCOMPUTE_NUM_THREAD_Z, 1,
1442 	mmCOMPUTE_PGM_RSRC1, 0x240, /* SGPRS=9 (80 GPRS) */
1443 	mmCOMPUTE_PGM_RSRC2, 20,
1444 	mmCOMPUTE_USER_DATA_0, 0xedcedc00,
1445 	mmCOMPUTE_USER_DATA_1, 0xedcedc01,
1446 	mmCOMPUTE_USER_DATA_2, 0xedcedc02,
1447 	mmCOMPUTE_USER_DATA_3, 0xedcedc03,
1448 	mmCOMPUTE_USER_DATA_4, 0xedcedc04,
1449 	mmCOMPUTE_USER_DATA_5, 0xedcedc05,
1450 	mmCOMPUTE_USER_DATA_6, 0xedcedc06,
1451 	mmCOMPUTE_USER_DATA_7, 0xedcedc07,
1452 	mmCOMPUTE_USER_DATA_8, 0xedcedc08,
1453 	mmCOMPUTE_USER_DATA_9, 0xedcedc09,
1454 };
1455 
1456 static const u32 sec_ded_counter_registers[] =
1457 {
1458 	mmCPC_EDC_ATC_CNT,
1459 	mmCPC_EDC_SCRATCH_CNT,
1460 	mmCPC_EDC_UCODE_CNT,
1461 	mmCPF_EDC_ATC_CNT,
1462 	mmCPF_EDC_ROQ_CNT,
1463 	mmCPF_EDC_TAG_CNT,
1464 	mmCPG_EDC_ATC_CNT,
1465 	mmCPG_EDC_DMA_CNT,
1466 	mmCPG_EDC_TAG_CNT,
1467 	mmDC_EDC_CSINVOC_CNT,
1468 	mmDC_EDC_RESTORE_CNT,
1469 	mmDC_EDC_STATE_CNT,
1470 	mmGDS_EDC_CNT,
1471 	mmGDS_EDC_GRBM_CNT,
1472 	mmGDS_EDC_OA_DED,
1473 	mmSPI_EDC_CNT,
1474 	mmSQC_ATC_EDC_GATCL1_CNT,
1475 	mmSQC_EDC_CNT,
1476 	mmSQ_EDC_DED_CNT,
1477 	mmSQ_EDC_INFO,
1478 	mmSQ_EDC_SEC_CNT,
1479 	mmTCC_EDC_CNT,
1480 	mmTCP_ATC_EDC_GATCL1_CNT,
1481 	mmTCP_EDC_CNT,
1482 	mmTD_EDC_CNT
1483 };
1484 
1485 static int gfx_v8_0_do_edc_gpr_workarounds(struct amdgpu_device *adev)
1486 {
1487 	struct amdgpu_ring *ring = &adev->gfx.compute_ring[0];
1488 	struct amdgpu_ib ib;
1489 	struct dma_fence *f = NULL;
1490 	int r, i;
1491 	u32 tmp;
1492 	unsigned total_size, vgpr_offset, sgpr_offset;
1493 	u64 gpu_addr;
1494 
1495 	/* only supported on CZ */
1496 	if (adev->asic_type != CHIP_CARRIZO)
1497 		return 0;
1498 
1499 	/* bail if the compute ring is not ready */
1500 	if (!ring->sched.ready)
1501 		return 0;
1502 
1503 	tmp = RREG32(mmGB_EDC_MODE);
1504 	WREG32(mmGB_EDC_MODE, 0);
1505 
1506 	total_size =
1507 		(((ARRAY_SIZE(vgpr_init_regs) / 2) * 3) + 4 + 5 + 2) * 4;
1508 	total_size +=
1509 		(((ARRAY_SIZE(sgpr1_init_regs) / 2) * 3) + 4 + 5 + 2) * 4;
1510 	total_size +=
1511 		(((ARRAY_SIZE(sgpr2_init_regs) / 2) * 3) + 4 + 5 + 2) * 4;
1512 	total_size = ALIGN(total_size, 256);
1513 	vgpr_offset = total_size;
1514 	total_size += ALIGN(sizeof(vgpr_init_compute_shader), 256);
1515 	sgpr_offset = total_size;
1516 	total_size += sizeof(sgpr_init_compute_shader);
1517 
1518 	/* allocate an indirect buffer to put the commands in */
1519 	memset(&ib, 0, sizeof(ib));
1520 	r = amdgpu_ib_get(adev, NULL, total_size,
1521 					AMDGPU_IB_POOL_DIRECT, &ib);
1522 	if (r) {
1523 		DRM_ERROR("amdgpu: failed to get ib (%d).\n", r);
1524 		return r;
1525 	}
1526 
1527 	/* load the compute shaders */
1528 	for (i = 0; i < ARRAY_SIZE(vgpr_init_compute_shader); i++)
1529 		ib.ptr[i + (vgpr_offset / 4)] = vgpr_init_compute_shader[i];
1530 
1531 	for (i = 0; i < ARRAY_SIZE(sgpr_init_compute_shader); i++)
1532 		ib.ptr[i + (sgpr_offset / 4)] = sgpr_init_compute_shader[i];
1533 
1534 	/* init the ib length to 0 */
1535 	ib.length_dw = 0;
1536 
1537 	/* VGPR */
1538 	/* write the register state for the compute dispatch */
1539 	for (i = 0; i < ARRAY_SIZE(vgpr_init_regs); i += 2) {
1540 		ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 1);
1541 		ib.ptr[ib.length_dw++] = vgpr_init_regs[i] - PACKET3_SET_SH_REG_START;
1542 		ib.ptr[ib.length_dw++] = vgpr_init_regs[i + 1];
1543 	}
1544 	/* write the shader start address: mmCOMPUTE_PGM_LO, mmCOMPUTE_PGM_HI */
1545 	gpu_addr = (ib.gpu_addr + (u64)vgpr_offset) >> 8;
1546 	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 2);
1547 	ib.ptr[ib.length_dw++] = mmCOMPUTE_PGM_LO - PACKET3_SET_SH_REG_START;
1548 	ib.ptr[ib.length_dw++] = lower_32_bits(gpu_addr);
1549 	ib.ptr[ib.length_dw++] = upper_32_bits(gpu_addr);
1550 
1551 	/* write dispatch packet */
1552 	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_DISPATCH_DIRECT, 3);
1553 	ib.ptr[ib.length_dw++] = 8; /* x */
1554 	ib.ptr[ib.length_dw++] = 1; /* y */
1555 	ib.ptr[ib.length_dw++] = 1; /* z */
1556 	ib.ptr[ib.length_dw++] =
1557 		REG_SET_FIELD(0, COMPUTE_DISPATCH_INITIATOR, COMPUTE_SHADER_EN, 1);
1558 
1559 	/* write CS partial flush packet */
1560 	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_EVENT_WRITE, 0);
1561 	ib.ptr[ib.length_dw++] = EVENT_TYPE(7) | EVENT_INDEX(4);
1562 
1563 	/* SGPR1 */
1564 	/* write the register state for the compute dispatch */
1565 	for (i = 0; i < ARRAY_SIZE(sgpr1_init_regs); i += 2) {
1566 		ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 1);
1567 		ib.ptr[ib.length_dw++] = sgpr1_init_regs[i] - PACKET3_SET_SH_REG_START;
1568 		ib.ptr[ib.length_dw++] = sgpr1_init_regs[i + 1];
1569 	}
1570 	/* write the shader start address: mmCOMPUTE_PGM_LO, mmCOMPUTE_PGM_HI */
1571 	gpu_addr = (ib.gpu_addr + (u64)sgpr_offset) >> 8;
1572 	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 2);
1573 	ib.ptr[ib.length_dw++] = mmCOMPUTE_PGM_LO - PACKET3_SET_SH_REG_START;
1574 	ib.ptr[ib.length_dw++] = lower_32_bits(gpu_addr);
1575 	ib.ptr[ib.length_dw++] = upper_32_bits(gpu_addr);
1576 
1577 	/* write dispatch packet */
1578 	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_DISPATCH_DIRECT, 3);
1579 	ib.ptr[ib.length_dw++] = 8; /* x */
1580 	ib.ptr[ib.length_dw++] = 1; /* y */
1581 	ib.ptr[ib.length_dw++] = 1; /* z */
1582 	ib.ptr[ib.length_dw++] =
1583 		REG_SET_FIELD(0, COMPUTE_DISPATCH_INITIATOR, COMPUTE_SHADER_EN, 1);
1584 
1585 	/* write CS partial flush packet */
1586 	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_EVENT_WRITE, 0);
1587 	ib.ptr[ib.length_dw++] = EVENT_TYPE(7) | EVENT_INDEX(4);
1588 
1589 	/* SGPR2 */
1590 	/* write the register state for the compute dispatch */
1591 	for (i = 0; i < ARRAY_SIZE(sgpr2_init_regs); i += 2) {
1592 		ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 1);
1593 		ib.ptr[ib.length_dw++] = sgpr2_init_regs[i] - PACKET3_SET_SH_REG_START;
1594 		ib.ptr[ib.length_dw++] = sgpr2_init_regs[i + 1];
1595 	}
1596 	/* write the shader start address: mmCOMPUTE_PGM_LO, mmCOMPUTE_PGM_HI */
1597 	gpu_addr = (ib.gpu_addr + (u64)sgpr_offset) >> 8;
1598 	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_SET_SH_REG, 2);
1599 	ib.ptr[ib.length_dw++] = mmCOMPUTE_PGM_LO - PACKET3_SET_SH_REG_START;
1600 	ib.ptr[ib.length_dw++] = lower_32_bits(gpu_addr);
1601 	ib.ptr[ib.length_dw++] = upper_32_bits(gpu_addr);
1602 
1603 	/* write dispatch packet */
1604 	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_DISPATCH_DIRECT, 3);
1605 	ib.ptr[ib.length_dw++] = 8; /* x */
1606 	ib.ptr[ib.length_dw++] = 1; /* y */
1607 	ib.ptr[ib.length_dw++] = 1; /* z */
1608 	ib.ptr[ib.length_dw++] =
1609 		REG_SET_FIELD(0, COMPUTE_DISPATCH_INITIATOR, COMPUTE_SHADER_EN, 1);
1610 
1611 	/* write CS partial flush packet */
1612 	ib.ptr[ib.length_dw++] = PACKET3(PACKET3_EVENT_WRITE, 0);
1613 	ib.ptr[ib.length_dw++] = EVENT_TYPE(7) | EVENT_INDEX(4);
1614 
1615 	/* shedule the ib on the ring */
1616 	r = amdgpu_ib_schedule(ring, 1, &ib, NULL, &f);
1617 	if (r) {
1618 		DRM_ERROR("amdgpu: ib submit failed (%d).\n", r);
1619 		goto fail;
1620 	}
1621 
1622 	/* wait for the GPU to finish processing the IB */
1623 	r = dma_fence_wait(f, false);
1624 	if (r) {
1625 		DRM_ERROR("amdgpu: fence wait failed (%d).\n", r);
1626 		goto fail;
1627 	}
1628 
1629 	tmp = REG_SET_FIELD(tmp, GB_EDC_MODE, DED_MODE, 2);
1630 	tmp = REG_SET_FIELD(tmp, GB_EDC_MODE, PROP_FED, 1);
1631 	WREG32(mmGB_EDC_MODE, tmp);
1632 
1633 	tmp = RREG32(mmCC_GC_EDC_CONFIG);
1634 	tmp = REG_SET_FIELD(tmp, CC_GC_EDC_CONFIG, DIS_EDC, 0) | 1;
1635 	WREG32(mmCC_GC_EDC_CONFIG, tmp);
1636 
1637 
1638 	/* read back registers to clear the counters */
1639 	for (i = 0; i < ARRAY_SIZE(sec_ded_counter_registers); i++)
1640 		RREG32(sec_ded_counter_registers[i]);
1641 
1642 fail:
1643 	amdgpu_ib_free(adev, &ib, NULL);
1644 	dma_fence_put(f);
1645 
1646 	return r;
1647 }
1648 
1649 static int gfx_v8_0_gpu_early_init(struct amdgpu_device *adev)
1650 {
1651 	u32 gb_addr_config;
1652 	u32 mc_arb_ramcfg;
1653 	u32 dimm00_addr_map, dimm01_addr_map, dimm10_addr_map, dimm11_addr_map;
1654 	u32 tmp;
1655 	int ret;
1656 
1657 	switch (adev->asic_type) {
1658 	case CHIP_TOPAZ:
1659 		adev->gfx.config.max_shader_engines = 1;
1660 		adev->gfx.config.max_tile_pipes = 2;
1661 		adev->gfx.config.max_cu_per_sh = 6;
1662 		adev->gfx.config.max_sh_per_se = 1;
1663 		adev->gfx.config.max_backends_per_se = 2;
1664 		adev->gfx.config.max_texture_channel_caches = 2;
1665 		adev->gfx.config.max_gprs = 256;
1666 		adev->gfx.config.max_gs_threads = 32;
1667 		adev->gfx.config.max_hw_contexts = 8;
1668 
1669 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1670 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1671 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1672 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
1673 		gb_addr_config = TOPAZ_GB_ADDR_CONFIG_GOLDEN;
1674 		break;
1675 	case CHIP_FIJI:
1676 		adev->gfx.config.max_shader_engines = 4;
1677 		adev->gfx.config.max_tile_pipes = 16;
1678 		adev->gfx.config.max_cu_per_sh = 16;
1679 		adev->gfx.config.max_sh_per_se = 1;
1680 		adev->gfx.config.max_backends_per_se = 4;
1681 		adev->gfx.config.max_texture_channel_caches = 16;
1682 		adev->gfx.config.max_gprs = 256;
1683 		adev->gfx.config.max_gs_threads = 32;
1684 		adev->gfx.config.max_hw_contexts = 8;
1685 
1686 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1687 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1688 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1689 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
1690 		gb_addr_config = TONGA_GB_ADDR_CONFIG_GOLDEN;
1691 		break;
1692 	case CHIP_POLARIS11:
1693 	case CHIP_POLARIS12:
1694 		ret = amdgpu_atombios_get_gfx_info(adev);
1695 		if (ret)
1696 			return ret;
1697 		adev->gfx.config.max_gprs = 256;
1698 		adev->gfx.config.max_gs_threads = 32;
1699 		adev->gfx.config.max_hw_contexts = 8;
1700 
1701 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1702 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1703 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1704 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
1705 		gb_addr_config = POLARIS11_GB_ADDR_CONFIG_GOLDEN;
1706 		break;
1707 	case CHIP_POLARIS10:
1708 	case CHIP_VEGAM:
1709 		ret = amdgpu_atombios_get_gfx_info(adev);
1710 		if (ret)
1711 			return ret;
1712 		adev->gfx.config.max_gprs = 256;
1713 		adev->gfx.config.max_gs_threads = 32;
1714 		adev->gfx.config.max_hw_contexts = 8;
1715 
1716 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1717 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1718 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1719 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
1720 		gb_addr_config = TONGA_GB_ADDR_CONFIG_GOLDEN;
1721 		break;
1722 	case CHIP_TONGA:
1723 		adev->gfx.config.max_shader_engines = 4;
1724 		adev->gfx.config.max_tile_pipes = 8;
1725 		adev->gfx.config.max_cu_per_sh = 8;
1726 		adev->gfx.config.max_sh_per_se = 1;
1727 		adev->gfx.config.max_backends_per_se = 2;
1728 		adev->gfx.config.max_texture_channel_caches = 8;
1729 		adev->gfx.config.max_gprs = 256;
1730 		adev->gfx.config.max_gs_threads = 32;
1731 		adev->gfx.config.max_hw_contexts = 8;
1732 
1733 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1734 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1735 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1736 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
1737 		gb_addr_config = TONGA_GB_ADDR_CONFIG_GOLDEN;
1738 		break;
1739 	case CHIP_CARRIZO:
1740 		adev->gfx.config.max_shader_engines = 1;
1741 		adev->gfx.config.max_tile_pipes = 2;
1742 		adev->gfx.config.max_sh_per_se = 1;
1743 		adev->gfx.config.max_backends_per_se = 2;
1744 		adev->gfx.config.max_cu_per_sh = 8;
1745 		adev->gfx.config.max_texture_channel_caches = 2;
1746 		adev->gfx.config.max_gprs = 256;
1747 		adev->gfx.config.max_gs_threads = 32;
1748 		adev->gfx.config.max_hw_contexts = 8;
1749 
1750 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1751 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1752 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1753 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
1754 		gb_addr_config = CARRIZO_GB_ADDR_CONFIG_GOLDEN;
1755 		break;
1756 	case CHIP_STONEY:
1757 		adev->gfx.config.max_shader_engines = 1;
1758 		adev->gfx.config.max_tile_pipes = 2;
1759 		adev->gfx.config.max_sh_per_se = 1;
1760 		adev->gfx.config.max_backends_per_se = 1;
1761 		adev->gfx.config.max_cu_per_sh = 3;
1762 		adev->gfx.config.max_texture_channel_caches = 2;
1763 		adev->gfx.config.max_gprs = 256;
1764 		adev->gfx.config.max_gs_threads = 16;
1765 		adev->gfx.config.max_hw_contexts = 8;
1766 
1767 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1768 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1769 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1770 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
1771 		gb_addr_config = CARRIZO_GB_ADDR_CONFIG_GOLDEN;
1772 		break;
1773 	default:
1774 		adev->gfx.config.max_shader_engines = 2;
1775 		adev->gfx.config.max_tile_pipes = 4;
1776 		adev->gfx.config.max_cu_per_sh = 2;
1777 		adev->gfx.config.max_sh_per_se = 1;
1778 		adev->gfx.config.max_backends_per_se = 2;
1779 		adev->gfx.config.max_texture_channel_caches = 4;
1780 		adev->gfx.config.max_gprs = 256;
1781 		adev->gfx.config.max_gs_threads = 32;
1782 		adev->gfx.config.max_hw_contexts = 8;
1783 
1784 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1785 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1786 		adev->gfx.config.sc_hiz_tile_fifo_size = 0x30;
1787 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x130;
1788 		gb_addr_config = TONGA_GB_ADDR_CONFIG_GOLDEN;
1789 		break;
1790 	}
1791 
1792 	adev->gfx.config.mc_arb_ramcfg = RREG32(mmMC_ARB_RAMCFG);
1793 	mc_arb_ramcfg = adev->gfx.config.mc_arb_ramcfg;
1794 
1795 	adev->gfx.config.num_banks = REG_GET_FIELD(mc_arb_ramcfg,
1796 				MC_ARB_RAMCFG, NOOFBANK);
1797 	adev->gfx.config.num_ranks = REG_GET_FIELD(mc_arb_ramcfg,
1798 				MC_ARB_RAMCFG, NOOFRANKS);
1799 
1800 	adev->gfx.config.num_tile_pipes = adev->gfx.config.max_tile_pipes;
1801 	adev->gfx.config.mem_max_burst_length_bytes = 256;
1802 	if (adev->flags & AMD_IS_APU) {
1803 		/* Get memory bank mapping mode. */
1804 		tmp = RREG32(mmMC_FUS_DRAM0_BANK_ADDR_MAPPING);
1805 		dimm00_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM0_BANK_ADDR_MAPPING, DIMM0ADDRMAP);
1806 		dimm01_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM0_BANK_ADDR_MAPPING, DIMM1ADDRMAP);
1807 
1808 		tmp = RREG32(mmMC_FUS_DRAM1_BANK_ADDR_MAPPING);
1809 		dimm10_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM1_BANK_ADDR_MAPPING, DIMM0ADDRMAP);
1810 		dimm11_addr_map = REG_GET_FIELD(tmp, MC_FUS_DRAM1_BANK_ADDR_MAPPING, DIMM1ADDRMAP);
1811 
1812 		/* Validate settings in case only one DIMM installed. */
1813 		if ((dimm00_addr_map == 0) || (dimm00_addr_map == 3) || (dimm00_addr_map == 4) || (dimm00_addr_map > 12))
1814 			dimm00_addr_map = 0;
1815 		if ((dimm01_addr_map == 0) || (dimm01_addr_map == 3) || (dimm01_addr_map == 4) || (dimm01_addr_map > 12))
1816 			dimm01_addr_map = 0;
1817 		if ((dimm10_addr_map == 0) || (dimm10_addr_map == 3) || (dimm10_addr_map == 4) || (dimm10_addr_map > 12))
1818 			dimm10_addr_map = 0;
1819 		if ((dimm11_addr_map == 0) || (dimm11_addr_map == 3) || (dimm11_addr_map == 4) || (dimm11_addr_map > 12))
1820 			dimm11_addr_map = 0;
1821 
1822 		/* If DIMM Addr map is 8GB, ROW size should be 2KB. Otherwise 1KB. */
1823 		/* If ROW size(DIMM1) != ROW size(DMIMM0), ROW size should be larger one. */
1824 		if ((dimm00_addr_map == 11) || (dimm01_addr_map == 11) || (dimm10_addr_map == 11) || (dimm11_addr_map == 11))
1825 			adev->gfx.config.mem_row_size_in_kb = 2;
1826 		else
1827 			adev->gfx.config.mem_row_size_in_kb = 1;
1828 	} else {
1829 		tmp = REG_GET_FIELD(mc_arb_ramcfg, MC_ARB_RAMCFG, NOOFCOLS);
1830 		adev->gfx.config.mem_row_size_in_kb = (4 * (1 << (8 + tmp))) / 1024;
1831 		if (adev->gfx.config.mem_row_size_in_kb > 4)
1832 			adev->gfx.config.mem_row_size_in_kb = 4;
1833 	}
1834 
1835 	adev->gfx.config.shader_engine_tile_size = 32;
1836 	adev->gfx.config.num_gpus = 1;
1837 	adev->gfx.config.multi_gpu_tile_size = 64;
1838 
1839 	/* fix up row size */
1840 	switch (adev->gfx.config.mem_row_size_in_kb) {
1841 	case 1:
1842 	default:
1843 		gb_addr_config = REG_SET_FIELD(gb_addr_config, GB_ADDR_CONFIG, ROW_SIZE, 0);
1844 		break;
1845 	case 2:
1846 		gb_addr_config = REG_SET_FIELD(gb_addr_config, GB_ADDR_CONFIG, ROW_SIZE, 1);
1847 		break;
1848 	case 4:
1849 		gb_addr_config = REG_SET_FIELD(gb_addr_config, GB_ADDR_CONFIG, ROW_SIZE, 2);
1850 		break;
1851 	}
1852 	adev->gfx.config.gb_addr_config = gb_addr_config;
1853 
1854 	return 0;
1855 }
1856 
1857 static int gfx_v8_0_compute_ring_init(struct amdgpu_device *adev, int ring_id,
1858 					int mec, int pipe, int queue)
1859 {
1860 	int r;
1861 	unsigned irq_type;
1862 	struct amdgpu_ring *ring = &adev->gfx.compute_ring[ring_id];
1863 	unsigned int hw_prio;
1864 
1865 	ring = &adev->gfx.compute_ring[ring_id];
1866 
1867 	/* mec0 is me1 */
1868 	ring->me = mec + 1;
1869 	ring->pipe = pipe;
1870 	ring->queue = queue;
1871 
1872 	ring->ring_obj = NULL;
1873 	ring->use_doorbell = true;
1874 	ring->doorbell_index = adev->doorbell_index.mec_ring0 + ring_id;
1875 	ring->eop_gpu_addr = adev->gfx.mec.hpd_eop_gpu_addr
1876 				+ (ring_id * GFX8_MEC_HPD_SIZE);
1877 	sprintf(ring->name, "comp_%d.%d.%d", ring->me, ring->pipe, ring->queue);
1878 
1879 	irq_type = AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP
1880 		+ ((ring->me - 1) * adev->gfx.mec.num_pipe_per_mec)
1881 		+ ring->pipe;
1882 
1883 	hw_prio = amdgpu_gfx_is_high_priority_compute_queue(adev, ring) ?
1884 			AMDGPU_RING_PRIO_2 : AMDGPU_RING_PRIO_DEFAULT;
1885 	/* type-2 packets are deprecated on MEC, use type-3 instead */
1886 	r = amdgpu_ring_init(adev, ring, 1024, &adev->gfx.eop_irq, irq_type,
1887 			     hw_prio, NULL);
1888 	if (r)
1889 		return r;
1890 
1891 
1892 	return 0;
1893 }
1894 
1895 static void gfx_v8_0_sq_irq_work_func(struct work_struct *work);
1896 
1897 static int gfx_v8_0_sw_init(void *handle)
1898 {
1899 	int i, j, k, r, ring_id;
1900 	int xcc_id = 0;
1901 	struct amdgpu_ring *ring;
1902 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1903 
1904 	switch (adev->asic_type) {
1905 	case CHIP_TONGA:
1906 	case CHIP_CARRIZO:
1907 	case CHIP_FIJI:
1908 	case CHIP_POLARIS10:
1909 	case CHIP_POLARIS11:
1910 	case CHIP_POLARIS12:
1911 	case CHIP_VEGAM:
1912 		adev->gfx.mec.num_mec = 2;
1913 		break;
1914 	case CHIP_TOPAZ:
1915 	case CHIP_STONEY:
1916 	default:
1917 		adev->gfx.mec.num_mec = 1;
1918 		break;
1919 	}
1920 
1921 	adev->gfx.mec.num_pipe_per_mec = 4;
1922 	adev->gfx.mec.num_queue_per_pipe = 8;
1923 
1924 	/* EOP Event */
1925 	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, VISLANDS30_IV_SRCID_CP_END_OF_PIPE, &adev->gfx.eop_irq);
1926 	if (r)
1927 		return r;
1928 
1929 	/* Privileged reg */
1930 	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, VISLANDS30_IV_SRCID_CP_PRIV_REG_FAULT,
1931 			      &adev->gfx.priv_reg_irq);
1932 	if (r)
1933 		return r;
1934 
1935 	/* Privileged inst */
1936 	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, VISLANDS30_IV_SRCID_CP_PRIV_INSTR_FAULT,
1937 			      &adev->gfx.priv_inst_irq);
1938 	if (r)
1939 		return r;
1940 
1941 	/* Add CP EDC/ECC irq  */
1942 	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, VISLANDS30_IV_SRCID_CP_ECC_ERROR,
1943 			      &adev->gfx.cp_ecc_error_irq);
1944 	if (r)
1945 		return r;
1946 
1947 	/* SQ interrupts. */
1948 	r = amdgpu_irq_add_id(adev, AMDGPU_IRQ_CLIENTID_LEGACY, VISLANDS30_IV_SRCID_SQ_INTERRUPT_MSG,
1949 			      &adev->gfx.sq_irq);
1950 	if (r) {
1951 		DRM_ERROR("amdgpu_irq_add() for SQ failed: %d\n", r);
1952 		return r;
1953 	}
1954 
1955 	INIT_WORK(&adev->gfx.sq_work.work, gfx_v8_0_sq_irq_work_func);
1956 
1957 	adev->gfx.gfx_current_status = AMDGPU_GFX_NORMAL_MODE;
1958 
1959 	r = gfx_v8_0_init_microcode(adev);
1960 	if (r) {
1961 		DRM_ERROR("Failed to load gfx firmware!\n");
1962 		return r;
1963 	}
1964 
1965 	r = adev->gfx.rlc.funcs->init(adev);
1966 	if (r) {
1967 		DRM_ERROR("Failed to init rlc BOs!\n");
1968 		return r;
1969 	}
1970 
1971 	r = gfx_v8_0_mec_init(adev);
1972 	if (r) {
1973 		DRM_ERROR("Failed to init MEC BOs!\n");
1974 		return r;
1975 	}
1976 
1977 	/* set up the gfx ring */
1978 	for (i = 0; i < adev->gfx.num_gfx_rings; i++) {
1979 		ring = &adev->gfx.gfx_ring[i];
1980 		ring->ring_obj = NULL;
1981 		sprintf(ring->name, "gfx");
1982 		/* no gfx doorbells on iceland */
1983 		if (adev->asic_type != CHIP_TOPAZ) {
1984 			ring->use_doorbell = true;
1985 			ring->doorbell_index = adev->doorbell_index.gfx_ring0;
1986 		}
1987 
1988 		r = amdgpu_ring_init(adev, ring, 1024, &adev->gfx.eop_irq,
1989 				     AMDGPU_CP_IRQ_GFX_ME0_PIPE0_EOP,
1990 				     AMDGPU_RING_PRIO_DEFAULT, NULL);
1991 		if (r)
1992 			return r;
1993 	}
1994 
1995 
1996 	/* set up the compute queues - allocate horizontally across pipes */
1997 	ring_id = 0;
1998 	for (i = 0; i < adev->gfx.mec.num_mec; ++i) {
1999 		for (j = 0; j < adev->gfx.mec.num_queue_per_pipe; j++) {
2000 			for (k = 0; k < adev->gfx.mec.num_pipe_per_mec; k++) {
2001 				if (!amdgpu_gfx_is_mec_queue_enabled(adev, 0, i,
2002 								     k, j))
2003 					continue;
2004 
2005 				r = gfx_v8_0_compute_ring_init(adev,
2006 								ring_id,
2007 								i, k, j);
2008 				if (r)
2009 					return r;
2010 
2011 				ring_id++;
2012 			}
2013 		}
2014 	}
2015 
2016 	r = amdgpu_gfx_kiq_init(adev, GFX8_MEC_HPD_SIZE, 0);
2017 	if (r) {
2018 		DRM_ERROR("Failed to init KIQ BOs!\n");
2019 		return r;
2020 	}
2021 
2022 	r = amdgpu_gfx_kiq_init_ring(adev, xcc_id);
2023 	if (r)
2024 		return r;
2025 
2026 	/* create MQD for all compute queues as well as KIQ for SRIOV case */
2027 	r = amdgpu_gfx_mqd_sw_init(adev, sizeof(struct vi_mqd_allocation), 0);
2028 	if (r)
2029 		return r;
2030 
2031 	adev->gfx.ce_ram_size = 0x8000;
2032 
2033 	r = gfx_v8_0_gpu_early_init(adev);
2034 	if (r)
2035 		return r;
2036 
2037 	return 0;
2038 }
2039 
2040 static int gfx_v8_0_sw_fini(void *handle)
2041 {
2042 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
2043 	int i;
2044 
2045 	for (i = 0; i < adev->gfx.num_gfx_rings; i++)
2046 		amdgpu_ring_fini(&adev->gfx.gfx_ring[i]);
2047 	for (i = 0; i < adev->gfx.num_compute_rings; i++)
2048 		amdgpu_ring_fini(&adev->gfx.compute_ring[i]);
2049 
2050 	amdgpu_gfx_mqd_sw_fini(adev, 0);
2051 	amdgpu_gfx_kiq_free_ring(&adev->gfx.kiq[0].ring);
2052 	amdgpu_gfx_kiq_fini(adev, 0);
2053 
2054 	gfx_v8_0_mec_fini(adev);
2055 	amdgpu_gfx_rlc_fini(adev);
2056 	amdgpu_bo_free_kernel(&adev->gfx.rlc.clear_state_obj,
2057 				&adev->gfx.rlc.clear_state_gpu_addr,
2058 				(void **)&adev->gfx.rlc.cs_ptr);
2059 	if ((adev->asic_type == CHIP_CARRIZO) ||
2060 	    (adev->asic_type == CHIP_STONEY)) {
2061 		amdgpu_bo_free_kernel(&adev->gfx.rlc.cp_table_obj,
2062 				&adev->gfx.rlc.cp_table_gpu_addr,
2063 				(void **)&adev->gfx.rlc.cp_table_ptr);
2064 	}
2065 	gfx_v8_0_free_microcode(adev);
2066 
2067 	return 0;
2068 }
2069 
2070 static void gfx_v8_0_tiling_mode_table_init(struct amdgpu_device *adev)
2071 {
2072 	uint32_t *modearray, *mod2array;
2073 	const u32 num_tile_mode_states = ARRAY_SIZE(adev->gfx.config.tile_mode_array);
2074 	const u32 num_secondary_tile_mode_states = ARRAY_SIZE(adev->gfx.config.macrotile_mode_array);
2075 	u32 reg_offset;
2076 
2077 	modearray = adev->gfx.config.tile_mode_array;
2078 	mod2array = adev->gfx.config.macrotile_mode_array;
2079 
2080 	for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
2081 		modearray[reg_offset] = 0;
2082 
2083 	for (reg_offset = 0; reg_offset <  num_secondary_tile_mode_states; reg_offset++)
2084 		mod2array[reg_offset] = 0;
2085 
2086 	switch (adev->asic_type) {
2087 	case CHIP_TOPAZ:
2088 		modearray[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2089 				PIPE_CONFIG(ADDR_SURF_P2) |
2090 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
2091 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2092 		modearray[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2093 				PIPE_CONFIG(ADDR_SURF_P2) |
2094 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
2095 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2096 		modearray[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2097 				PIPE_CONFIG(ADDR_SURF_P2) |
2098 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
2099 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2100 		modearray[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2101 				PIPE_CONFIG(ADDR_SURF_P2) |
2102 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
2103 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2104 		modearray[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2105 				PIPE_CONFIG(ADDR_SURF_P2) |
2106 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2107 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2108 		modearray[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2109 				PIPE_CONFIG(ADDR_SURF_P2) |
2110 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2111 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2112 		modearray[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2113 				PIPE_CONFIG(ADDR_SURF_P2) |
2114 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2115 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2116 		modearray[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
2117 				PIPE_CONFIG(ADDR_SURF_P2));
2118 		modearray[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2119 				PIPE_CONFIG(ADDR_SURF_P2) |
2120 				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2121 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2122 		modearray[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2123 				 PIPE_CONFIG(ADDR_SURF_P2) |
2124 				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2125 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2126 		modearray[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2127 				 PIPE_CONFIG(ADDR_SURF_P2) |
2128 				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2129 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2130 		modearray[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2131 				 PIPE_CONFIG(ADDR_SURF_P2) |
2132 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2133 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2134 		modearray[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2135 				 PIPE_CONFIG(ADDR_SURF_P2) |
2136 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2137 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2138 		modearray[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
2139 				 PIPE_CONFIG(ADDR_SURF_P2) |
2140 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2141 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2142 		modearray[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2143 				 PIPE_CONFIG(ADDR_SURF_P2) |
2144 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2145 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2146 		modearray[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2147 				 PIPE_CONFIG(ADDR_SURF_P2) |
2148 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2149 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2150 		modearray[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2151 				 PIPE_CONFIG(ADDR_SURF_P2) |
2152 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2153 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2154 		modearray[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2155 				 PIPE_CONFIG(ADDR_SURF_P2) |
2156 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2157 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2158 		modearray[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
2159 				 PIPE_CONFIG(ADDR_SURF_P2) |
2160 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2161 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2162 		modearray[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
2163 				 PIPE_CONFIG(ADDR_SURF_P2) |
2164 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2165 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2166 		modearray[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2167 				 PIPE_CONFIG(ADDR_SURF_P2) |
2168 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2169 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2170 		modearray[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
2171 				 PIPE_CONFIG(ADDR_SURF_P2) |
2172 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2173 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2174 		modearray[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
2175 				 PIPE_CONFIG(ADDR_SURF_P2) |
2176 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2177 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2178 		modearray[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2179 				 PIPE_CONFIG(ADDR_SURF_P2) |
2180 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2181 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2182 		modearray[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2183 				 PIPE_CONFIG(ADDR_SURF_P2) |
2184 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2185 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2186 		modearray[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2187 				 PIPE_CONFIG(ADDR_SURF_P2) |
2188 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2189 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2190 
2191 		mod2array[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
2192 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2193 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2194 				NUM_BANKS(ADDR_SURF_8_BANK));
2195 		mod2array[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
2196 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2197 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2198 				NUM_BANKS(ADDR_SURF_8_BANK));
2199 		mod2array[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
2200 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2201 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2202 				NUM_BANKS(ADDR_SURF_8_BANK));
2203 		mod2array[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2204 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2205 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2206 				NUM_BANKS(ADDR_SURF_8_BANK));
2207 		mod2array[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2208 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2209 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2210 				NUM_BANKS(ADDR_SURF_8_BANK));
2211 		mod2array[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2212 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2213 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2214 				NUM_BANKS(ADDR_SURF_8_BANK));
2215 		mod2array[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2216 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2217 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2218 				NUM_BANKS(ADDR_SURF_8_BANK));
2219 		mod2array[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
2220 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
2221 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2222 				NUM_BANKS(ADDR_SURF_16_BANK));
2223 		mod2array[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
2224 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2225 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2226 				NUM_BANKS(ADDR_SURF_16_BANK));
2227 		mod2array[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
2228 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2229 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2230 				 NUM_BANKS(ADDR_SURF_16_BANK));
2231 		mod2array[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
2232 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2233 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2234 				 NUM_BANKS(ADDR_SURF_16_BANK));
2235 		mod2array[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2236 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2237 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2238 				 NUM_BANKS(ADDR_SURF_16_BANK));
2239 		mod2array[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2240 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2241 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2242 				 NUM_BANKS(ADDR_SURF_16_BANK));
2243 		mod2array[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2244 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2245 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2246 				 NUM_BANKS(ADDR_SURF_8_BANK));
2247 
2248 		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
2249 			if (reg_offset != 7 && reg_offset != 12 && reg_offset != 17 &&
2250 			    reg_offset != 23)
2251 				WREG32(mmGB_TILE_MODE0 + reg_offset, modearray[reg_offset]);
2252 
2253 		for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
2254 			if (reg_offset != 7)
2255 				WREG32(mmGB_MACROTILE_MODE0 + reg_offset, mod2array[reg_offset]);
2256 
2257 		break;
2258 	case CHIP_FIJI:
2259 	case CHIP_VEGAM:
2260 		modearray[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2261 				PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2262 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
2263 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2264 		modearray[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2265 				PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2266 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
2267 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2268 		modearray[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2269 				PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2270 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
2271 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2272 		modearray[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2273 				PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2274 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
2275 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2276 		modearray[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2277 				PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2278 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2279 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2280 		modearray[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2281 				PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2282 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2283 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2284 		modearray[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2285 				PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2286 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2287 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2288 		modearray[7] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2289 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2290 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2291 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2292 		modearray[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
2293 				PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16));
2294 		modearray[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2295 				PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2296 				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2297 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2298 		modearray[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2299 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2300 				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2301 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2302 		modearray[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2303 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2304 				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2305 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2306 		modearray[12] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2307 				 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2308 				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2309 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2310 		modearray[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2311 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2312 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2313 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2314 		modearray[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2315 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2316 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2317 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2318 		modearray[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
2319 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2320 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2321 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2322 		modearray[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2323 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2324 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2325 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2326 		modearray[17] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2327 				 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2328 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2329 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2330 		modearray[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2331 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2332 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2333 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2334 		modearray[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2335 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2336 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2337 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2338 		modearray[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2339 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2340 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2341 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2342 		modearray[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
2343 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2344 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2345 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2346 		modearray[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
2347 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2348 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2349 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2350 		modearray[23] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
2351 				 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2352 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2353 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2354 		modearray[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2355 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2356 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2357 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2358 		modearray[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
2359 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2360 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2361 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2362 		modearray[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
2363 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2364 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2365 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2366 		modearray[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2367 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2368 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2369 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2370 		modearray[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2371 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2372 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2373 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2374 		modearray[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2375 				 PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) |
2376 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2377 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2378 		modearray[30] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2379 				 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2380 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2381 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2382 
2383 		mod2array[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2384 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2385 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2386 				NUM_BANKS(ADDR_SURF_8_BANK));
2387 		mod2array[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2388 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2389 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2390 				NUM_BANKS(ADDR_SURF_8_BANK));
2391 		mod2array[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2392 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2393 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2394 				NUM_BANKS(ADDR_SURF_8_BANK));
2395 		mod2array[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2396 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2397 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2398 				NUM_BANKS(ADDR_SURF_8_BANK));
2399 		mod2array[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2400 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2401 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2402 				NUM_BANKS(ADDR_SURF_8_BANK));
2403 		mod2array[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2404 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2405 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2406 				NUM_BANKS(ADDR_SURF_8_BANK));
2407 		mod2array[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2408 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2409 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2410 				NUM_BANKS(ADDR_SURF_8_BANK));
2411 		mod2array[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2412 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
2413 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2414 				NUM_BANKS(ADDR_SURF_8_BANK));
2415 		mod2array[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2416 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2417 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2418 				NUM_BANKS(ADDR_SURF_8_BANK));
2419 		mod2array[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2420 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2421 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2422 				 NUM_BANKS(ADDR_SURF_8_BANK));
2423 		mod2array[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2424 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2425 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2426 				 NUM_BANKS(ADDR_SURF_8_BANK));
2427 		mod2array[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2428 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2429 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2430 				 NUM_BANKS(ADDR_SURF_8_BANK));
2431 		mod2array[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2432 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2433 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2434 				 NUM_BANKS(ADDR_SURF_8_BANK));
2435 		mod2array[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2436 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2437 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2438 				 NUM_BANKS(ADDR_SURF_4_BANK));
2439 
2440 		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
2441 			WREG32(mmGB_TILE_MODE0 + reg_offset, modearray[reg_offset]);
2442 
2443 		for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
2444 			if (reg_offset != 7)
2445 				WREG32(mmGB_MACROTILE_MODE0 + reg_offset, mod2array[reg_offset]);
2446 
2447 		break;
2448 	case CHIP_TONGA:
2449 		modearray[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2450 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2451 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
2452 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2453 		modearray[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2454 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2455 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
2456 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2457 		modearray[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2458 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2459 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
2460 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2461 		modearray[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2462 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2463 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
2464 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2465 		modearray[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2466 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2467 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2468 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2469 		modearray[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2470 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2471 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2472 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2473 		modearray[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2474 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2475 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2476 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2477 		modearray[7] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2478 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2479 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2480 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2481 		modearray[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
2482 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16));
2483 		modearray[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2484 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2485 				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2486 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2487 		modearray[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2488 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2489 				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2490 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2491 		modearray[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2492 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2493 				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2494 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2495 		modearray[12] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2496 				 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2497 				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2498 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2499 		modearray[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2500 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2501 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2502 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2503 		modearray[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2504 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2505 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2506 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2507 		modearray[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
2508 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2509 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2510 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2511 		modearray[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2512 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2513 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2514 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2515 		modearray[17] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2516 				 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2517 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2518 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2519 		modearray[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2520 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2521 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2522 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2523 		modearray[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2524 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2525 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2526 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2527 		modearray[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2528 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2529 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2530 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2531 		modearray[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
2532 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2533 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2534 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2535 		modearray[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
2536 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2537 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2538 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2539 		modearray[23] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
2540 				 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2541 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2542 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2543 		modearray[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2544 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2545 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2546 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2547 		modearray[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
2548 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2549 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2550 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2551 		modearray[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
2552 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2553 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2554 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2555 		modearray[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2556 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2557 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2558 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2559 		modearray[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2560 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2561 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2562 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2563 		modearray[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2564 				 PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2565 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2566 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2567 		modearray[30] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2568 				 PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2569 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2570 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2571 
2572 		mod2array[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2573 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2574 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2575 				NUM_BANKS(ADDR_SURF_16_BANK));
2576 		mod2array[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2577 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2578 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2579 				NUM_BANKS(ADDR_SURF_16_BANK));
2580 		mod2array[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2581 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2582 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2583 				NUM_BANKS(ADDR_SURF_16_BANK));
2584 		mod2array[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2585 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2586 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2587 				NUM_BANKS(ADDR_SURF_16_BANK));
2588 		mod2array[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2589 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2590 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2591 				NUM_BANKS(ADDR_SURF_16_BANK));
2592 		mod2array[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2593 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2594 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2595 				NUM_BANKS(ADDR_SURF_16_BANK));
2596 		mod2array[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2597 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2598 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2599 				NUM_BANKS(ADDR_SURF_16_BANK));
2600 		mod2array[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2601 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
2602 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2603 				NUM_BANKS(ADDR_SURF_16_BANK));
2604 		mod2array[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2605 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2606 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2607 				NUM_BANKS(ADDR_SURF_16_BANK));
2608 		mod2array[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2609 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2610 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2611 				 NUM_BANKS(ADDR_SURF_16_BANK));
2612 		mod2array[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2613 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2614 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2615 				 NUM_BANKS(ADDR_SURF_16_BANK));
2616 		mod2array[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2617 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2618 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2619 				 NUM_BANKS(ADDR_SURF_8_BANK));
2620 		mod2array[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2621 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2622 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2623 				 NUM_BANKS(ADDR_SURF_4_BANK));
2624 		mod2array[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2625 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2626 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2627 				 NUM_BANKS(ADDR_SURF_4_BANK));
2628 
2629 		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
2630 			WREG32(mmGB_TILE_MODE0 + reg_offset, modearray[reg_offset]);
2631 
2632 		for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
2633 			if (reg_offset != 7)
2634 				WREG32(mmGB_MACROTILE_MODE0 + reg_offset, mod2array[reg_offset]);
2635 
2636 		break;
2637 	case CHIP_POLARIS11:
2638 	case CHIP_POLARIS12:
2639 		modearray[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2640 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2641 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
2642 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2643 		modearray[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2644 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2645 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
2646 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2647 		modearray[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2648 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2649 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
2650 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2651 		modearray[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2652 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2653 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
2654 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2655 		modearray[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2656 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2657 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2658 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2659 		modearray[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2660 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2661 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2662 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2663 		modearray[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2664 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2665 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2666 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2667 		modearray[7] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2668 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2669 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2670 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2671 		modearray[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
2672 				PIPE_CONFIG(ADDR_SURF_P4_16x16));
2673 		modearray[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2674 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2675 				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2676 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2677 		modearray[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2678 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2679 				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2680 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2681 		modearray[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2682 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2683 				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2684 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2685 		modearray[12] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2686 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2687 				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2688 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2689 		modearray[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2690 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2691 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2692 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2693 		modearray[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2694 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2695 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2696 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2697 		modearray[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
2698 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2699 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2700 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2701 		modearray[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2702 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2703 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2704 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2705 		modearray[17] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2706 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2707 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2708 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2709 		modearray[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2710 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2711 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2712 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2713 		modearray[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2714 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2715 				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2716 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2717 		modearray[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2718 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2719 				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2720 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2721 		modearray[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
2722 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2723 				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2724 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2725 		modearray[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
2726 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2727 				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2728 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2729 		modearray[23] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
2730 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2731 				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2732 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2733 		modearray[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2734 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2735 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2736 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2737 		modearray[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
2738 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2739 				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2740 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2741 		modearray[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
2742 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2743 				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2744 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2745 		modearray[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2746 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2747 				MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2748 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2749 		modearray[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2750 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2751 				MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2752 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2753 		modearray[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2754 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2755 				MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2756 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2757 		modearray[30] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2758 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2759 				MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2760 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2761 
2762 		mod2array[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2763 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2764 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2765 				NUM_BANKS(ADDR_SURF_16_BANK));
2766 
2767 		mod2array[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2768 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2769 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2770 				NUM_BANKS(ADDR_SURF_16_BANK));
2771 
2772 		mod2array[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2773 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2774 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2775 				NUM_BANKS(ADDR_SURF_16_BANK));
2776 
2777 		mod2array[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2778 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2779 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2780 				NUM_BANKS(ADDR_SURF_16_BANK));
2781 
2782 		mod2array[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2783 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2784 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2785 				NUM_BANKS(ADDR_SURF_16_BANK));
2786 
2787 		mod2array[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2788 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2789 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2790 				NUM_BANKS(ADDR_SURF_16_BANK));
2791 
2792 		mod2array[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2793 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2794 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2795 				NUM_BANKS(ADDR_SURF_16_BANK));
2796 
2797 		mod2array[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
2798 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
2799 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2800 				NUM_BANKS(ADDR_SURF_16_BANK));
2801 
2802 		mod2array[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
2803 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2804 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2805 				NUM_BANKS(ADDR_SURF_16_BANK));
2806 
2807 		mod2array[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2808 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2809 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2810 				NUM_BANKS(ADDR_SURF_16_BANK));
2811 
2812 		mod2array[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2813 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2814 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2815 				NUM_BANKS(ADDR_SURF_16_BANK));
2816 
2817 		mod2array[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2818 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2819 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2820 				NUM_BANKS(ADDR_SURF_16_BANK));
2821 
2822 		mod2array[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2823 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2824 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2825 				NUM_BANKS(ADDR_SURF_8_BANK));
2826 
2827 		mod2array[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2828 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2829 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2830 				NUM_BANKS(ADDR_SURF_4_BANK));
2831 
2832 		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
2833 			WREG32(mmGB_TILE_MODE0 + reg_offset, modearray[reg_offset]);
2834 
2835 		for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
2836 			if (reg_offset != 7)
2837 				WREG32(mmGB_MACROTILE_MODE0 + reg_offset, mod2array[reg_offset]);
2838 
2839 		break;
2840 	case CHIP_POLARIS10:
2841 		modearray[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2842 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2843 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
2844 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2845 		modearray[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2846 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2847 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
2848 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2849 		modearray[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2850 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2851 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
2852 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2853 		modearray[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2854 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2855 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
2856 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2857 		modearray[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2858 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2859 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2860 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2861 		modearray[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2862 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2863 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2864 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2865 		modearray[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2866 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2867 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2868 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2869 		modearray[7] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2870 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2871 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
2872 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
2873 		modearray[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
2874 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16));
2875 		modearray[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2876 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2877 				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2878 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2879 		modearray[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2880 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2881 				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2882 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2883 		modearray[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2884 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2885 				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2886 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2887 		modearray[12] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2888 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2889 				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
2890 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2891 		modearray[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2892 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2893 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2894 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2895 		modearray[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2896 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2897 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2898 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2899 		modearray[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
2900 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2901 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2902 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2903 		modearray[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2904 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2905 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2906 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2907 		modearray[17] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2908 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2909 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2910 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2911 		modearray[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2912 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2913 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2914 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2915 		modearray[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
2916 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2917 				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2918 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2919 		modearray[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2920 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2921 				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2922 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2923 		modearray[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
2924 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2925 				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2926 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2927 		modearray[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
2928 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2929 				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2930 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2931 		modearray[23] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
2932 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2933 				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2934 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2935 		modearray[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
2936 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2937 				MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
2938 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2939 		modearray[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
2940 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2941 				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2942 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2943 		modearray[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
2944 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2945 				MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
2946 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
2947 		modearray[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
2948 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2949 				MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2950 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2951 		modearray[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
2952 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2953 				MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2954 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
2955 		modearray[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2956 				PIPE_CONFIG(ADDR_SURF_P8_32x32_16x16) |
2957 				MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2958 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2959 		modearray[30] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
2960 				PIPE_CONFIG(ADDR_SURF_P4_16x16) |
2961 				MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
2962 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
2963 
2964 		mod2array[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2965 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2966 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2967 				NUM_BANKS(ADDR_SURF_16_BANK));
2968 
2969 		mod2array[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2970 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2971 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2972 				NUM_BANKS(ADDR_SURF_16_BANK));
2973 
2974 		mod2array[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2975 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2976 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2977 				NUM_BANKS(ADDR_SURF_16_BANK));
2978 
2979 		mod2array[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2980 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
2981 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
2982 				NUM_BANKS(ADDR_SURF_16_BANK));
2983 
2984 		mod2array[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2985 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
2986 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
2987 				NUM_BANKS(ADDR_SURF_16_BANK));
2988 
2989 		mod2array[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2990 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2991 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2992 				NUM_BANKS(ADDR_SURF_16_BANK));
2993 
2994 		mod2array[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
2995 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
2996 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
2997 				NUM_BANKS(ADDR_SURF_16_BANK));
2998 
2999 		mod2array[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3000 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
3001 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3002 				NUM_BANKS(ADDR_SURF_16_BANK));
3003 
3004 		mod2array[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3005 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
3006 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3007 				NUM_BANKS(ADDR_SURF_16_BANK));
3008 
3009 		mod2array[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3010 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
3011 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3012 				NUM_BANKS(ADDR_SURF_16_BANK));
3013 
3014 		mod2array[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3015 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3016 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3017 				NUM_BANKS(ADDR_SURF_16_BANK));
3018 
3019 		mod2array[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3020 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3021 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
3022 				NUM_BANKS(ADDR_SURF_8_BANK));
3023 
3024 		mod2array[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3025 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3026 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
3027 				NUM_BANKS(ADDR_SURF_4_BANK));
3028 
3029 		mod2array[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3030 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3031 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) |
3032 				NUM_BANKS(ADDR_SURF_4_BANK));
3033 
3034 		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
3035 			WREG32(mmGB_TILE_MODE0 + reg_offset, modearray[reg_offset]);
3036 
3037 		for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
3038 			if (reg_offset != 7)
3039 				WREG32(mmGB_MACROTILE_MODE0 + reg_offset, mod2array[reg_offset]);
3040 
3041 		break;
3042 	case CHIP_STONEY:
3043 		modearray[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3044 				PIPE_CONFIG(ADDR_SURF_P2) |
3045 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
3046 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3047 		modearray[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3048 				PIPE_CONFIG(ADDR_SURF_P2) |
3049 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
3050 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3051 		modearray[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3052 				PIPE_CONFIG(ADDR_SURF_P2) |
3053 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
3054 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3055 		modearray[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3056 				PIPE_CONFIG(ADDR_SURF_P2) |
3057 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
3058 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3059 		modearray[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3060 				PIPE_CONFIG(ADDR_SURF_P2) |
3061 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
3062 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3063 		modearray[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
3064 				PIPE_CONFIG(ADDR_SURF_P2) |
3065 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
3066 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3067 		modearray[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
3068 				PIPE_CONFIG(ADDR_SURF_P2) |
3069 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
3070 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3071 		modearray[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
3072 				PIPE_CONFIG(ADDR_SURF_P2));
3073 		modearray[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
3074 				PIPE_CONFIG(ADDR_SURF_P2) |
3075 				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
3076 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3077 		modearray[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3078 				 PIPE_CONFIG(ADDR_SURF_P2) |
3079 				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
3080 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3081 		modearray[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
3082 				 PIPE_CONFIG(ADDR_SURF_P2) |
3083 				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
3084 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
3085 		modearray[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
3086 				 PIPE_CONFIG(ADDR_SURF_P2) |
3087 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3088 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3089 		modearray[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3090 				 PIPE_CONFIG(ADDR_SURF_P2) |
3091 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3092 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3093 		modearray[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
3094 				 PIPE_CONFIG(ADDR_SURF_P2) |
3095 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3096 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3097 		modearray[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
3098 				 PIPE_CONFIG(ADDR_SURF_P2) |
3099 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3100 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
3101 		modearray[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
3102 				 PIPE_CONFIG(ADDR_SURF_P2) |
3103 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3104 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3105 		modearray[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
3106 				 PIPE_CONFIG(ADDR_SURF_P2) |
3107 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3108 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3109 		modearray[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
3110 				 PIPE_CONFIG(ADDR_SURF_P2) |
3111 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3112 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3113 		modearray[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
3114 				 PIPE_CONFIG(ADDR_SURF_P2) |
3115 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3116 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3117 		modearray[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
3118 				 PIPE_CONFIG(ADDR_SURF_P2) |
3119 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3120 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3121 		modearray[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
3122 				 PIPE_CONFIG(ADDR_SURF_P2) |
3123 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3124 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3125 		modearray[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
3126 				 PIPE_CONFIG(ADDR_SURF_P2) |
3127 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3128 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3129 		modearray[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
3130 				 PIPE_CONFIG(ADDR_SURF_P2) |
3131 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3132 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3133 		modearray[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
3134 				 PIPE_CONFIG(ADDR_SURF_P2) |
3135 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
3136 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3137 		modearray[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3138 				 PIPE_CONFIG(ADDR_SURF_P2) |
3139 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
3140 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3141 		modearray[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
3142 				 PIPE_CONFIG(ADDR_SURF_P2) |
3143 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
3144 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
3145 
3146 		mod2array[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3147 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
3148 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3149 				NUM_BANKS(ADDR_SURF_8_BANK));
3150 		mod2array[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3151 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
3152 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3153 				NUM_BANKS(ADDR_SURF_8_BANK));
3154 		mod2array[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3155 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3156 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3157 				NUM_BANKS(ADDR_SURF_8_BANK));
3158 		mod2array[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3159 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3160 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3161 				NUM_BANKS(ADDR_SURF_8_BANK));
3162 		mod2array[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3163 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3164 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3165 				NUM_BANKS(ADDR_SURF_8_BANK));
3166 		mod2array[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3167 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3168 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3169 				NUM_BANKS(ADDR_SURF_8_BANK));
3170 		mod2array[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3171 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3172 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3173 				NUM_BANKS(ADDR_SURF_8_BANK));
3174 		mod2array[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
3175 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
3176 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3177 				NUM_BANKS(ADDR_SURF_16_BANK));
3178 		mod2array[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
3179 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
3180 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3181 				NUM_BANKS(ADDR_SURF_16_BANK));
3182 		mod2array[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
3183 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
3184 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3185 				 NUM_BANKS(ADDR_SURF_16_BANK));
3186 		mod2array[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
3187 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
3188 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3189 				 NUM_BANKS(ADDR_SURF_16_BANK));
3190 		mod2array[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3191 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
3192 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3193 				 NUM_BANKS(ADDR_SURF_16_BANK));
3194 		mod2array[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3195 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3196 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3197 				 NUM_BANKS(ADDR_SURF_16_BANK));
3198 		mod2array[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3199 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3200 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3201 				 NUM_BANKS(ADDR_SURF_8_BANK));
3202 
3203 		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
3204 			if (reg_offset != 7 && reg_offset != 12 && reg_offset != 17 &&
3205 			    reg_offset != 23)
3206 				WREG32(mmGB_TILE_MODE0 + reg_offset, modearray[reg_offset]);
3207 
3208 		for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
3209 			if (reg_offset != 7)
3210 				WREG32(mmGB_MACROTILE_MODE0 + reg_offset, mod2array[reg_offset]);
3211 
3212 		break;
3213 	default:
3214 		dev_warn(adev->dev,
3215 			 "Unknown chip type (%d) in function gfx_v8_0_tiling_mode_table_init() falling through to CHIP_CARRIZO\n",
3216 			 adev->asic_type);
3217 		fallthrough;
3218 
3219 	case CHIP_CARRIZO:
3220 		modearray[0] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3221 				PIPE_CONFIG(ADDR_SURF_P2) |
3222 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) |
3223 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3224 		modearray[1] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3225 				PIPE_CONFIG(ADDR_SURF_P2) |
3226 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) |
3227 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3228 		modearray[2] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3229 				PIPE_CONFIG(ADDR_SURF_P2) |
3230 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) |
3231 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3232 		modearray[3] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3233 				PIPE_CONFIG(ADDR_SURF_P2) |
3234 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) |
3235 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3236 		modearray[4] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3237 				PIPE_CONFIG(ADDR_SURF_P2) |
3238 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
3239 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3240 		modearray[5] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
3241 				PIPE_CONFIG(ADDR_SURF_P2) |
3242 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
3243 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3244 		modearray[6] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
3245 				PIPE_CONFIG(ADDR_SURF_P2) |
3246 				TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) |
3247 				MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING));
3248 		modearray[8] = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) |
3249 				PIPE_CONFIG(ADDR_SURF_P2));
3250 		modearray[9] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
3251 				PIPE_CONFIG(ADDR_SURF_P2) |
3252 				MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
3253 				SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3254 		modearray[10] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3255 				 PIPE_CONFIG(ADDR_SURF_P2) |
3256 				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
3257 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3258 		modearray[11] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
3259 				 PIPE_CONFIG(ADDR_SURF_P2) |
3260 				 MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) |
3261 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
3262 		modearray[13] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
3263 				 PIPE_CONFIG(ADDR_SURF_P2) |
3264 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3265 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3266 		modearray[14] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3267 				 PIPE_CONFIG(ADDR_SURF_P2) |
3268 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3269 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3270 		modearray[15] = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) |
3271 				 PIPE_CONFIG(ADDR_SURF_P2) |
3272 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3273 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3274 		modearray[16] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
3275 				 PIPE_CONFIG(ADDR_SURF_P2) |
3276 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3277 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
3278 		modearray[18] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
3279 				 PIPE_CONFIG(ADDR_SURF_P2) |
3280 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3281 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3282 		modearray[19] = (ARRAY_MODE(ARRAY_1D_TILED_THICK) |
3283 				 PIPE_CONFIG(ADDR_SURF_P2) |
3284 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3285 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3286 		modearray[20] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
3287 				 PIPE_CONFIG(ADDR_SURF_P2) |
3288 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3289 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3290 		modearray[21] = (ARRAY_MODE(ARRAY_3D_TILED_THICK) |
3291 				 PIPE_CONFIG(ADDR_SURF_P2) |
3292 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3293 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3294 		modearray[22] = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) |
3295 				 PIPE_CONFIG(ADDR_SURF_P2) |
3296 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3297 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3298 		modearray[24] = (ARRAY_MODE(ARRAY_2D_TILED_THICK) |
3299 				 PIPE_CONFIG(ADDR_SURF_P2) |
3300 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) |
3301 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3302 		modearray[25] = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) |
3303 				 PIPE_CONFIG(ADDR_SURF_P2) |
3304 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3305 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3306 		modearray[26] = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) |
3307 				 PIPE_CONFIG(ADDR_SURF_P2) |
3308 				 MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) |
3309 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1));
3310 		modearray[27] = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) |
3311 				 PIPE_CONFIG(ADDR_SURF_P2) |
3312 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
3313 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3314 		modearray[28] = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) |
3315 				 PIPE_CONFIG(ADDR_SURF_P2) |
3316 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
3317 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2));
3318 		modearray[29] = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) |
3319 				 PIPE_CONFIG(ADDR_SURF_P2) |
3320 				 MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) |
3321 				 SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8));
3322 
3323 		mod2array[0] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3324 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
3325 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3326 				NUM_BANKS(ADDR_SURF_8_BANK));
3327 		mod2array[1] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3328 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
3329 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3330 				NUM_BANKS(ADDR_SURF_8_BANK));
3331 		mod2array[2] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3332 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3333 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3334 				NUM_BANKS(ADDR_SURF_8_BANK));
3335 		mod2array[3] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3336 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3337 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3338 				NUM_BANKS(ADDR_SURF_8_BANK));
3339 		mod2array[4] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3340 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3341 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3342 				NUM_BANKS(ADDR_SURF_8_BANK));
3343 		mod2array[5] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3344 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3345 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3346 				NUM_BANKS(ADDR_SURF_8_BANK));
3347 		mod2array[6] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3348 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3349 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3350 				NUM_BANKS(ADDR_SURF_8_BANK));
3351 		mod2array[8] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
3352 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) |
3353 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3354 				NUM_BANKS(ADDR_SURF_16_BANK));
3355 		mod2array[9] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_4) |
3356 				BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
3357 				MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3358 				NUM_BANKS(ADDR_SURF_16_BANK));
3359 		mod2array[10] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
3360 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) |
3361 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3362 				 NUM_BANKS(ADDR_SURF_16_BANK));
3363 		mod2array[11] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_2) |
3364 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
3365 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3366 				 NUM_BANKS(ADDR_SURF_16_BANK));
3367 		mod2array[12] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3368 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) |
3369 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3370 				 NUM_BANKS(ADDR_SURF_16_BANK));
3371 		mod2array[13] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3372 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3373 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_4) |
3374 				 NUM_BANKS(ADDR_SURF_16_BANK));
3375 		mod2array[14] = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) |
3376 				 BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) |
3377 				 MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) |
3378 				 NUM_BANKS(ADDR_SURF_8_BANK));
3379 
3380 		for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++)
3381 			if (reg_offset != 7 && reg_offset != 12 && reg_offset != 17 &&
3382 			    reg_offset != 23)
3383 				WREG32(mmGB_TILE_MODE0 + reg_offset, modearray[reg_offset]);
3384 
3385 		for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++)
3386 			if (reg_offset != 7)
3387 				WREG32(mmGB_MACROTILE_MODE0 + reg_offset, mod2array[reg_offset]);
3388 
3389 		break;
3390 	}
3391 }
3392 
3393 static void gfx_v8_0_select_se_sh(struct amdgpu_device *adev,
3394 				  u32 se_num, u32 sh_num, u32 instance,
3395 				  int xcc_id)
3396 {
3397 	u32 data;
3398 
3399 	if (instance == 0xffffffff)
3400 		data = REG_SET_FIELD(0, GRBM_GFX_INDEX, INSTANCE_BROADCAST_WRITES, 1);
3401 	else
3402 		data = REG_SET_FIELD(0, GRBM_GFX_INDEX, INSTANCE_INDEX, instance);
3403 
3404 	if (se_num == 0xffffffff)
3405 		data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SE_BROADCAST_WRITES, 1);
3406 	else
3407 		data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SE_INDEX, se_num);
3408 
3409 	if (sh_num == 0xffffffff)
3410 		data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SH_BROADCAST_WRITES, 1);
3411 	else
3412 		data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SH_INDEX, sh_num);
3413 
3414 	WREG32(mmGRBM_GFX_INDEX, data);
3415 }
3416 
3417 static void gfx_v8_0_select_me_pipe_q(struct amdgpu_device *adev,
3418 				  u32 me, u32 pipe, u32 q, u32 vm, u32 xcc_id)
3419 {
3420 	vi_srbm_select(adev, me, pipe, q, vm);
3421 }
3422 
3423 static u32 gfx_v8_0_get_rb_active_bitmap(struct amdgpu_device *adev)
3424 {
3425 	u32 data, mask;
3426 
3427 	data =  RREG32(mmCC_RB_BACKEND_DISABLE) |
3428 		RREG32(mmGC_USER_RB_BACKEND_DISABLE);
3429 
3430 	data = REG_GET_FIELD(data, GC_USER_RB_BACKEND_DISABLE, BACKEND_DISABLE);
3431 
3432 	mask = amdgpu_gfx_create_bitmask(adev->gfx.config.max_backends_per_se /
3433 					 adev->gfx.config.max_sh_per_se);
3434 
3435 	return (~data) & mask;
3436 }
3437 
3438 static void
3439 gfx_v8_0_raster_config(struct amdgpu_device *adev, u32 *rconf, u32 *rconf1)
3440 {
3441 	switch (adev->asic_type) {
3442 	case CHIP_FIJI:
3443 	case CHIP_VEGAM:
3444 		*rconf |= RB_MAP_PKR0(2) | RB_MAP_PKR1(2) |
3445 			  RB_XSEL2(1) | PKR_MAP(2) |
3446 			  PKR_XSEL(1) | PKR_YSEL(1) |
3447 			  SE_MAP(2) | SE_XSEL(2) | SE_YSEL(3);
3448 		*rconf1 |= SE_PAIR_MAP(2) | SE_PAIR_XSEL(3) |
3449 			   SE_PAIR_YSEL(2);
3450 		break;
3451 	case CHIP_TONGA:
3452 	case CHIP_POLARIS10:
3453 		*rconf |= RB_MAP_PKR0(2) | RB_XSEL2(1) | SE_MAP(2) |
3454 			  SE_XSEL(1) | SE_YSEL(1);
3455 		*rconf1 |= SE_PAIR_MAP(2) | SE_PAIR_XSEL(2) |
3456 			   SE_PAIR_YSEL(2);
3457 		break;
3458 	case CHIP_TOPAZ:
3459 	case CHIP_CARRIZO:
3460 		*rconf |= RB_MAP_PKR0(2);
3461 		*rconf1 |= 0x0;
3462 		break;
3463 	case CHIP_POLARIS11:
3464 	case CHIP_POLARIS12:
3465 		*rconf |= RB_MAP_PKR0(2) | RB_XSEL2(1) | SE_MAP(2) |
3466 			  SE_XSEL(1) | SE_YSEL(1);
3467 		*rconf1 |= 0x0;
3468 		break;
3469 	case CHIP_STONEY:
3470 		*rconf |= 0x0;
3471 		*rconf1 |= 0x0;
3472 		break;
3473 	default:
3474 		DRM_ERROR("unknown asic: 0x%x\n", adev->asic_type);
3475 		break;
3476 	}
3477 }
3478 
3479 static void
3480 gfx_v8_0_write_harvested_raster_configs(struct amdgpu_device *adev,
3481 					u32 raster_config, u32 raster_config_1,
3482 					unsigned rb_mask, unsigned num_rb)
3483 {
3484 	unsigned sh_per_se = max_t(unsigned, adev->gfx.config.max_sh_per_se, 1);
3485 	unsigned num_se = max_t(unsigned, adev->gfx.config.max_shader_engines, 1);
3486 	unsigned rb_per_pkr = min_t(unsigned, num_rb / num_se / sh_per_se, 2);
3487 	unsigned rb_per_se = num_rb / num_se;
3488 	unsigned se_mask[4];
3489 	unsigned se;
3490 
3491 	se_mask[0] = ((1 << rb_per_se) - 1) & rb_mask;
3492 	se_mask[1] = (se_mask[0] << rb_per_se) & rb_mask;
3493 	se_mask[2] = (se_mask[1] << rb_per_se) & rb_mask;
3494 	se_mask[3] = (se_mask[2] << rb_per_se) & rb_mask;
3495 
3496 	WARN_ON(!(num_se == 1 || num_se == 2 || num_se == 4));
3497 	WARN_ON(!(sh_per_se == 1 || sh_per_se == 2));
3498 	WARN_ON(!(rb_per_pkr == 1 || rb_per_pkr == 2));
3499 
3500 	if ((num_se > 2) && ((!se_mask[0] && !se_mask[1]) ||
3501 			     (!se_mask[2] && !se_mask[3]))) {
3502 		raster_config_1 &= ~SE_PAIR_MAP_MASK;
3503 
3504 		if (!se_mask[0] && !se_mask[1]) {
3505 			raster_config_1 |=
3506 				SE_PAIR_MAP(RASTER_CONFIG_SE_PAIR_MAP_3);
3507 		} else {
3508 			raster_config_1 |=
3509 				SE_PAIR_MAP(RASTER_CONFIG_SE_PAIR_MAP_0);
3510 		}
3511 	}
3512 
3513 	for (se = 0; se < num_se; se++) {
3514 		unsigned raster_config_se = raster_config;
3515 		unsigned pkr0_mask = ((1 << rb_per_pkr) - 1) << (se * rb_per_se);
3516 		unsigned pkr1_mask = pkr0_mask << rb_per_pkr;
3517 		int idx = (se / 2) * 2;
3518 
3519 		if ((num_se > 1) && (!se_mask[idx] || !se_mask[idx + 1])) {
3520 			raster_config_se &= ~SE_MAP_MASK;
3521 
3522 			if (!se_mask[idx]) {
3523 				raster_config_se |= SE_MAP(RASTER_CONFIG_SE_MAP_3);
3524 			} else {
3525 				raster_config_se |= SE_MAP(RASTER_CONFIG_SE_MAP_0);
3526 			}
3527 		}
3528 
3529 		pkr0_mask &= rb_mask;
3530 		pkr1_mask &= rb_mask;
3531 		if (rb_per_se > 2 && (!pkr0_mask || !pkr1_mask)) {
3532 			raster_config_se &= ~PKR_MAP_MASK;
3533 
3534 			if (!pkr0_mask) {
3535 				raster_config_se |= PKR_MAP(RASTER_CONFIG_PKR_MAP_3);
3536 			} else {
3537 				raster_config_se |= PKR_MAP(RASTER_CONFIG_PKR_MAP_0);
3538 			}
3539 		}
3540 
3541 		if (rb_per_se >= 2) {
3542 			unsigned rb0_mask = 1 << (se * rb_per_se);
3543 			unsigned rb1_mask = rb0_mask << 1;
3544 
3545 			rb0_mask &= rb_mask;
3546 			rb1_mask &= rb_mask;
3547 			if (!rb0_mask || !rb1_mask) {
3548 				raster_config_se &= ~RB_MAP_PKR0_MASK;
3549 
3550 				if (!rb0_mask) {
3551 					raster_config_se |=
3552 						RB_MAP_PKR0(RASTER_CONFIG_RB_MAP_3);
3553 				} else {
3554 					raster_config_se |=
3555 						RB_MAP_PKR0(RASTER_CONFIG_RB_MAP_0);
3556 				}
3557 			}
3558 
3559 			if (rb_per_se > 2) {
3560 				rb0_mask = 1 << (se * rb_per_se + rb_per_pkr);
3561 				rb1_mask = rb0_mask << 1;
3562 				rb0_mask &= rb_mask;
3563 				rb1_mask &= rb_mask;
3564 				if (!rb0_mask || !rb1_mask) {
3565 					raster_config_se &= ~RB_MAP_PKR1_MASK;
3566 
3567 					if (!rb0_mask) {
3568 						raster_config_se |=
3569 							RB_MAP_PKR1(RASTER_CONFIG_RB_MAP_3);
3570 					} else {
3571 						raster_config_se |=
3572 							RB_MAP_PKR1(RASTER_CONFIG_RB_MAP_0);
3573 					}
3574 				}
3575 			}
3576 		}
3577 
3578 		/* GRBM_GFX_INDEX has a different offset on VI */
3579 		gfx_v8_0_select_se_sh(adev, se, 0xffffffff, 0xffffffff, 0);
3580 		WREG32(mmPA_SC_RASTER_CONFIG, raster_config_se);
3581 		WREG32(mmPA_SC_RASTER_CONFIG_1, raster_config_1);
3582 	}
3583 
3584 	/* GRBM_GFX_INDEX has a different offset on VI */
3585 	gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 0);
3586 }
3587 
3588 static void gfx_v8_0_setup_rb(struct amdgpu_device *adev)
3589 {
3590 	int i, j;
3591 	u32 data;
3592 	u32 raster_config = 0, raster_config_1 = 0;
3593 	u32 active_rbs = 0;
3594 	u32 rb_bitmap_width_per_sh = adev->gfx.config.max_backends_per_se /
3595 					adev->gfx.config.max_sh_per_se;
3596 	unsigned num_rb_pipes;
3597 
3598 	mutex_lock(&adev->grbm_idx_mutex);
3599 	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
3600 		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
3601 			gfx_v8_0_select_se_sh(adev, i, j, 0xffffffff, 0);
3602 			data = gfx_v8_0_get_rb_active_bitmap(adev);
3603 			active_rbs |= data << ((i * adev->gfx.config.max_sh_per_se + j) *
3604 					       rb_bitmap_width_per_sh);
3605 		}
3606 	}
3607 	gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 0);
3608 
3609 	adev->gfx.config.backend_enable_mask = active_rbs;
3610 	adev->gfx.config.num_rbs = hweight32(active_rbs);
3611 
3612 	num_rb_pipes = min_t(unsigned, adev->gfx.config.max_backends_per_se *
3613 			     adev->gfx.config.max_shader_engines, 16);
3614 
3615 	gfx_v8_0_raster_config(adev, &raster_config, &raster_config_1);
3616 
3617 	if (!adev->gfx.config.backend_enable_mask ||
3618 			adev->gfx.config.num_rbs >= num_rb_pipes) {
3619 		WREG32(mmPA_SC_RASTER_CONFIG, raster_config);
3620 		WREG32(mmPA_SC_RASTER_CONFIG_1, raster_config_1);
3621 	} else {
3622 		gfx_v8_0_write_harvested_raster_configs(adev, raster_config, raster_config_1,
3623 							adev->gfx.config.backend_enable_mask,
3624 							num_rb_pipes);
3625 	}
3626 
3627 	/* cache the values for userspace */
3628 	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
3629 		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
3630 			gfx_v8_0_select_se_sh(adev, i, j, 0xffffffff, 0);
3631 			adev->gfx.config.rb_config[i][j].rb_backend_disable =
3632 				RREG32(mmCC_RB_BACKEND_DISABLE);
3633 			adev->gfx.config.rb_config[i][j].user_rb_backend_disable =
3634 				RREG32(mmGC_USER_RB_BACKEND_DISABLE);
3635 			adev->gfx.config.rb_config[i][j].raster_config =
3636 				RREG32(mmPA_SC_RASTER_CONFIG);
3637 			adev->gfx.config.rb_config[i][j].raster_config_1 =
3638 				RREG32(mmPA_SC_RASTER_CONFIG_1);
3639 		}
3640 	}
3641 	gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 0);
3642 	mutex_unlock(&adev->grbm_idx_mutex);
3643 }
3644 
3645 #define DEFAULT_SH_MEM_BASES	(0x6000)
3646 /**
3647  * gfx_v8_0_init_compute_vmid - gart enable
3648  *
3649  * @adev: amdgpu_device pointer
3650  *
3651  * Initialize compute vmid sh_mem registers
3652  *
3653  */
3654 static void gfx_v8_0_init_compute_vmid(struct amdgpu_device *adev)
3655 {
3656 	int i;
3657 	uint32_t sh_mem_config;
3658 	uint32_t sh_mem_bases;
3659 
3660 	/*
3661 	 * Configure apertures:
3662 	 * LDS:         0x60000000'00000000 - 0x60000001'00000000 (4GB)
3663 	 * Scratch:     0x60000001'00000000 - 0x60000002'00000000 (4GB)
3664 	 * GPUVM:       0x60010000'00000000 - 0x60020000'00000000 (1TB)
3665 	 */
3666 	sh_mem_bases = DEFAULT_SH_MEM_BASES | (DEFAULT_SH_MEM_BASES << 16);
3667 
3668 	sh_mem_config = SH_MEM_ADDRESS_MODE_HSA64 <<
3669 			SH_MEM_CONFIG__ADDRESS_MODE__SHIFT |
3670 			SH_MEM_ALIGNMENT_MODE_UNALIGNED <<
3671 			SH_MEM_CONFIG__ALIGNMENT_MODE__SHIFT |
3672 			MTYPE_CC << SH_MEM_CONFIG__DEFAULT_MTYPE__SHIFT |
3673 			SH_MEM_CONFIG__PRIVATE_ATC_MASK;
3674 
3675 	mutex_lock(&adev->srbm_mutex);
3676 	for (i = adev->vm_manager.first_kfd_vmid; i < AMDGPU_NUM_VMID; i++) {
3677 		vi_srbm_select(adev, 0, 0, 0, i);
3678 		/* CP and shaders */
3679 		WREG32(mmSH_MEM_CONFIG, sh_mem_config);
3680 		WREG32(mmSH_MEM_APE1_BASE, 1);
3681 		WREG32(mmSH_MEM_APE1_LIMIT, 0);
3682 		WREG32(mmSH_MEM_BASES, sh_mem_bases);
3683 	}
3684 	vi_srbm_select(adev, 0, 0, 0, 0);
3685 	mutex_unlock(&adev->srbm_mutex);
3686 
3687 	/* Initialize all compute VMIDs to have no GDS, GWS, or OA
3688 	   access. These should be enabled by FW for target VMIDs. */
3689 	for (i = adev->vm_manager.first_kfd_vmid; i < AMDGPU_NUM_VMID; i++) {
3690 		WREG32(amdgpu_gds_reg_offset[i].mem_base, 0);
3691 		WREG32(amdgpu_gds_reg_offset[i].mem_size, 0);
3692 		WREG32(amdgpu_gds_reg_offset[i].gws, 0);
3693 		WREG32(amdgpu_gds_reg_offset[i].oa, 0);
3694 	}
3695 }
3696 
3697 static void gfx_v8_0_init_gds_vmid(struct amdgpu_device *adev)
3698 {
3699 	int vmid;
3700 
3701 	/*
3702 	 * Initialize all compute and user-gfx VMIDs to have no GDS, GWS, or OA
3703 	 * access. Compute VMIDs should be enabled by FW for target VMIDs,
3704 	 * the driver can enable them for graphics. VMID0 should maintain
3705 	 * access so that HWS firmware can save/restore entries.
3706 	 */
3707 	for (vmid = 1; vmid < AMDGPU_NUM_VMID; vmid++) {
3708 		WREG32(amdgpu_gds_reg_offset[vmid].mem_base, 0);
3709 		WREG32(amdgpu_gds_reg_offset[vmid].mem_size, 0);
3710 		WREG32(amdgpu_gds_reg_offset[vmid].gws, 0);
3711 		WREG32(amdgpu_gds_reg_offset[vmid].oa, 0);
3712 	}
3713 }
3714 
3715 static void gfx_v8_0_config_init(struct amdgpu_device *adev)
3716 {
3717 	switch (adev->asic_type) {
3718 	default:
3719 		adev->gfx.config.double_offchip_lds_buf = 1;
3720 		break;
3721 	case CHIP_CARRIZO:
3722 	case CHIP_STONEY:
3723 		adev->gfx.config.double_offchip_lds_buf = 0;
3724 		break;
3725 	}
3726 }
3727 
3728 static void gfx_v8_0_constants_init(struct amdgpu_device *adev)
3729 {
3730 	u32 tmp, sh_static_mem_cfg;
3731 	int i;
3732 
3733 	WREG32_FIELD(GRBM_CNTL, READ_TIMEOUT, 0xFF);
3734 	WREG32(mmGB_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
3735 	WREG32(mmHDP_ADDR_CONFIG, adev->gfx.config.gb_addr_config);
3736 	WREG32(mmDMIF_ADDR_CALC, adev->gfx.config.gb_addr_config);
3737 
3738 	gfx_v8_0_tiling_mode_table_init(adev);
3739 	gfx_v8_0_setup_rb(adev);
3740 	gfx_v8_0_get_cu_info(adev);
3741 	gfx_v8_0_config_init(adev);
3742 
3743 	/* XXX SH_MEM regs */
3744 	/* where to put LDS, scratch, GPUVM in FSA64 space */
3745 	sh_static_mem_cfg = REG_SET_FIELD(0, SH_STATIC_MEM_CONFIG,
3746 				   SWIZZLE_ENABLE, 1);
3747 	sh_static_mem_cfg = REG_SET_FIELD(sh_static_mem_cfg, SH_STATIC_MEM_CONFIG,
3748 				   ELEMENT_SIZE, 1);
3749 	sh_static_mem_cfg = REG_SET_FIELD(sh_static_mem_cfg, SH_STATIC_MEM_CONFIG,
3750 				   INDEX_STRIDE, 3);
3751 	WREG32(mmSH_STATIC_MEM_CONFIG, sh_static_mem_cfg);
3752 
3753 	mutex_lock(&adev->srbm_mutex);
3754 	for (i = 0; i < adev->vm_manager.id_mgr[0].num_ids; i++) {
3755 		vi_srbm_select(adev, 0, 0, 0, i);
3756 		/* CP and shaders */
3757 		if (i == 0) {
3758 			tmp = REG_SET_FIELD(0, SH_MEM_CONFIG, DEFAULT_MTYPE, MTYPE_UC);
3759 			tmp = REG_SET_FIELD(tmp, SH_MEM_CONFIG, APE1_MTYPE, MTYPE_UC);
3760 			tmp = REG_SET_FIELD(tmp, SH_MEM_CONFIG, ALIGNMENT_MODE,
3761 					    SH_MEM_ALIGNMENT_MODE_UNALIGNED);
3762 			WREG32(mmSH_MEM_CONFIG, tmp);
3763 			WREG32(mmSH_MEM_BASES, 0);
3764 		} else {
3765 			tmp = REG_SET_FIELD(0, SH_MEM_CONFIG, DEFAULT_MTYPE, MTYPE_NC);
3766 			tmp = REG_SET_FIELD(tmp, SH_MEM_CONFIG, APE1_MTYPE, MTYPE_UC);
3767 			tmp = REG_SET_FIELD(tmp, SH_MEM_CONFIG, ALIGNMENT_MODE,
3768 					    SH_MEM_ALIGNMENT_MODE_UNALIGNED);
3769 			WREG32(mmSH_MEM_CONFIG, tmp);
3770 			tmp = adev->gmc.shared_aperture_start >> 48;
3771 			WREG32(mmSH_MEM_BASES, tmp);
3772 		}
3773 
3774 		WREG32(mmSH_MEM_APE1_BASE, 1);
3775 		WREG32(mmSH_MEM_APE1_LIMIT, 0);
3776 	}
3777 	vi_srbm_select(adev, 0, 0, 0, 0);
3778 	mutex_unlock(&adev->srbm_mutex);
3779 
3780 	gfx_v8_0_init_compute_vmid(adev);
3781 	gfx_v8_0_init_gds_vmid(adev);
3782 
3783 	mutex_lock(&adev->grbm_idx_mutex);
3784 	/*
3785 	 * making sure that the following register writes will be broadcasted
3786 	 * to all the shaders
3787 	 */
3788 	gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 0);
3789 
3790 	WREG32(mmPA_SC_FIFO_SIZE,
3791 		   (adev->gfx.config.sc_prim_fifo_size_frontend <<
3792 			PA_SC_FIFO_SIZE__SC_FRONTEND_PRIM_FIFO_SIZE__SHIFT) |
3793 		   (adev->gfx.config.sc_prim_fifo_size_backend <<
3794 			PA_SC_FIFO_SIZE__SC_BACKEND_PRIM_FIFO_SIZE__SHIFT) |
3795 		   (adev->gfx.config.sc_hiz_tile_fifo_size <<
3796 			PA_SC_FIFO_SIZE__SC_HIZ_TILE_FIFO_SIZE__SHIFT) |
3797 		   (adev->gfx.config.sc_earlyz_tile_fifo_size <<
3798 			PA_SC_FIFO_SIZE__SC_EARLYZ_TILE_FIFO_SIZE__SHIFT));
3799 
3800 	tmp = RREG32(mmSPI_ARB_PRIORITY);
3801 	tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS0, 2);
3802 	tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS1, 2);
3803 	tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS2, 2);
3804 	tmp = REG_SET_FIELD(tmp, SPI_ARB_PRIORITY, PIPE_ORDER_TS3, 2);
3805 	WREG32(mmSPI_ARB_PRIORITY, tmp);
3806 
3807 	mutex_unlock(&adev->grbm_idx_mutex);
3808 
3809 }
3810 
3811 static void gfx_v8_0_wait_for_rlc_serdes(struct amdgpu_device *adev)
3812 {
3813 	u32 i, j, k;
3814 	u32 mask;
3815 
3816 	mutex_lock(&adev->grbm_idx_mutex);
3817 	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
3818 		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
3819 			gfx_v8_0_select_se_sh(adev, i, j, 0xffffffff, 0);
3820 			for (k = 0; k < adev->usec_timeout; k++) {
3821 				if (RREG32(mmRLC_SERDES_CU_MASTER_BUSY) == 0)
3822 					break;
3823 				udelay(1);
3824 			}
3825 			if (k == adev->usec_timeout) {
3826 				gfx_v8_0_select_se_sh(adev, 0xffffffff,
3827 						      0xffffffff, 0xffffffff, 0);
3828 				mutex_unlock(&adev->grbm_idx_mutex);
3829 				DRM_INFO("Timeout wait for RLC serdes %u,%u\n",
3830 					 i, j);
3831 				return;
3832 			}
3833 		}
3834 	}
3835 	gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 0);
3836 	mutex_unlock(&adev->grbm_idx_mutex);
3837 
3838 	mask = RLC_SERDES_NONCU_MASTER_BUSY__SE_MASTER_BUSY_MASK |
3839 		RLC_SERDES_NONCU_MASTER_BUSY__GC_MASTER_BUSY_MASK |
3840 		RLC_SERDES_NONCU_MASTER_BUSY__TC0_MASTER_BUSY_MASK |
3841 		RLC_SERDES_NONCU_MASTER_BUSY__TC1_MASTER_BUSY_MASK;
3842 	for (k = 0; k < adev->usec_timeout; k++) {
3843 		if ((RREG32(mmRLC_SERDES_NONCU_MASTER_BUSY) & mask) == 0)
3844 			break;
3845 		udelay(1);
3846 	}
3847 }
3848 
3849 static void gfx_v8_0_enable_gui_idle_interrupt(struct amdgpu_device *adev,
3850 					       bool enable)
3851 {
3852 	u32 tmp = RREG32(mmCP_INT_CNTL_RING0);
3853 
3854 	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CNTX_BUSY_INT_ENABLE, enable ? 1 : 0);
3855 	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CNTX_EMPTY_INT_ENABLE, enable ? 1 : 0);
3856 	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CMP_BUSY_INT_ENABLE, enable ? 1 : 0);
3857 	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, GFX_IDLE_INT_ENABLE, enable ? 1 : 0);
3858 
3859 	WREG32(mmCP_INT_CNTL_RING0, tmp);
3860 }
3861 
3862 static void gfx_v8_0_init_csb(struct amdgpu_device *adev)
3863 {
3864 	adev->gfx.rlc.funcs->get_csb_buffer(adev, adev->gfx.rlc.cs_ptr);
3865 	/* csib */
3866 	WREG32(mmRLC_CSIB_ADDR_HI,
3867 			adev->gfx.rlc.clear_state_gpu_addr >> 32);
3868 	WREG32(mmRLC_CSIB_ADDR_LO,
3869 			adev->gfx.rlc.clear_state_gpu_addr & 0xfffffffc);
3870 	WREG32(mmRLC_CSIB_LENGTH,
3871 			adev->gfx.rlc.clear_state_size);
3872 }
3873 
3874 static void gfx_v8_0_parse_ind_reg_list(int *register_list_format,
3875 				int ind_offset,
3876 				int list_size,
3877 				int *unique_indices,
3878 				int *indices_count,
3879 				int max_indices,
3880 				int *ind_start_offsets,
3881 				int *offset_count,
3882 				int max_offset)
3883 {
3884 	int indices;
3885 	bool new_entry = true;
3886 
3887 	for (; ind_offset < list_size; ind_offset++) {
3888 
3889 		if (new_entry) {
3890 			new_entry = false;
3891 			ind_start_offsets[*offset_count] = ind_offset;
3892 			*offset_count = *offset_count + 1;
3893 			BUG_ON(*offset_count >= max_offset);
3894 		}
3895 
3896 		if (register_list_format[ind_offset] == 0xFFFFFFFF) {
3897 			new_entry = true;
3898 			continue;
3899 		}
3900 
3901 		ind_offset += 2;
3902 
3903 		/* look for the matching indice */
3904 		for (indices = 0;
3905 			indices < *indices_count;
3906 			indices++) {
3907 			if (unique_indices[indices] ==
3908 				register_list_format[ind_offset])
3909 				break;
3910 		}
3911 
3912 		if (indices >= *indices_count) {
3913 			unique_indices[*indices_count] =
3914 				register_list_format[ind_offset];
3915 			indices = *indices_count;
3916 			*indices_count = *indices_count + 1;
3917 			BUG_ON(*indices_count >= max_indices);
3918 		}
3919 
3920 		register_list_format[ind_offset] = indices;
3921 	}
3922 }
3923 
3924 static int gfx_v8_0_init_save_restore_list(struct amdgpu_device *adev)
3925 {
3926 	int i, temp, data;
3927 	int unique_indices[] = {0, 0, 0, 0, 0, 0, 0, 0};
3928 	int indices_count = 0;
3929 	int indirect_start_offsets[] = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0};
3930 	int offset_count = 0;
3931 
3932 	int list_size;
3933 	unsigned int *register_list_format =
3934 		kmemdup(adev->gfx.rlc.register_list_format,
3935 			adev->gfx.rlc.reg_list_format_size_bytes, GFP_KERNEL);
3936 	if (!register_list_format)
3937 		return -ENOMEM;
3938 
3939 	gfx_v8_0_parse_ind_reg_list(register_list_format,
3940 				RLC_FormatDirectRegListLength,
3941 				adev->gfx.rlc.reg_list_format_size_bytes >> 2,
3942 				unique_indices,
3943 				&indices_count,
3944 				ARRAY_SIZE(unique_indices),
3945 				indirect_start_offsets,
3946 				&offset_count,
3947 				ARRAY_SIZE(indirect_start_offsets));
3948 
3949 	/* save and restore list */
3950 	WREG32_FIELD(RLC_SRM_CNTL, AUTO_INCR_ADDR, 1);
3951 
3952 	WREG32(mmRLC_SRM_ARAM_ADDR, 0);
3953 	for (i = 0; i < adev->gfx.rlc.reg_list_size_bytes >> 2; i++)
3954 		WREG32(mmRLC_SRM_ARAM_DATA, adev->gfx.rlc.register_restore[i]);
3955 
3956 	/* indirect list */
3957 	WREG32(mmRLC_GPM_SCRATCH_ADDR, adev->gfx.rlc.reg_list_format_start);
3958 	for (i = 0; i < adev->gfx.rlc.reg_list_format_size_bytes >> 2; i++)
3959 		WREG32(mmRLC_GPM_SCRATCH_DATA, register_list_format[i]);
3960 
3961 	list_size = adev->gfx.rlc.reg_list_size_bytes >> 2;
3962 	list_size = list_size >> 1;
3963 	WREG32(mmRLC_GPM_SCRATCH_ADDR, adev->gfx.rlc.reg_restore_list_size);
3964 	WREG32(mmRLC_GPM_SCRATCH_DATA, list_size);
3965 
3966 	/* starting offsets starts */
3967 	WREG32(mmRLC_GPM_SCRATCH_ADDR,
3968 		adev->gfx.rlc.starting_offsets_start);
3969 	for (i = 0; i < ARRAY_SIZE(indirect_start_offsets); i++)
3970 		WREG32(mmRLC_GPM_SCRATCH_DATA,
3971 				indirect_start_offsets[i]);
3972 
3973 	/* unique indices */
3974 	temp = mmRLC_SRM_INDEX_CNTL_ADDR_0;
3975 	data = mmRLC_SRM_INDEX_CNTL_DATA_0;
3976 	for (i = 0; i < ARRAY_SIZE(unique_indices); i++) {
3977 		if (unique_indices[i] != 0) {
3978 			WREG32(temp + i, unique_indices[i] & 0x3FFFF);
3979 			WREG32(data + i, unique_indices[i] >> 20);
3980 		}
3981 	}
3982 	kfree(register_list_format);
3983 
3984 	return 0;
3985 }
3986 
3987 static void gfx_v8_0_enable_save_restore_machine(struct amdgpu_device *adev)
3988 {
3989 	WREG32_FIELD(RLC_SRM_CNTL, SRM_ENABLE, 1);
3990 }
3991 
3992 static void gfx_v8_0_init_power_gating(struct amdgpu_device *adev)
3993 {
3994 	uint32_t data;
3995 
3996 	WREG32_FIELD(CP_RB_WPTR_POLL_CNTL, IDLE_POLL_COUNT, 0x60);
3997 
3998 	data = REG_SET_FIELD(0, RLC_PG_DELAY, POWER_UP_DELAY, 0x10);
3999 	data = REG_SET_FIELD(data, RLC_PG_DELAY, POWER_DOWN_DELAY, 0x10);
4000 	data = REG_SET_FIELD(data, RLC_PG_DELAY, CMD_PROPAGATE_DELAY, 0x10);
4001 	data = REG_SET_FIELD(data, RLC_PG_DELAY, MEM_SLEEP_DELAY, 0x10);
4002 	WREG32(mmRLC_PG_DELAY, data);
4003 
4004 	WREG32_FIELD(RLC_PG_DELAY_2, SERDES_CMD_DELAY, 0x3);
4005 	WREG32_FIELD(RLC_AUTO_PG_CTRL, GRBM_REG_SAVE_GFX_IDLE_THRESHOLD, 0x55f0);
4006 
4007 }
4008 
4009 static void cz_enable_sck_slow_down_on_power_up(struct amdgpu_device *adev,
4010 						bool enable)
4011 {
4012 	WREG32_FIELD(RLC_PG_CNTL, SMU_CLK_SLOWDOWN_ON_PU_ENABLE, enable ? 1 : 0);
4013 }
4014 
4015 static void cz_enable_sck_slow_down_on_power_down(struct amdgpu_device *adev,
4016 						  bool enable)
4017 {
4018 	WREG32_FIELD(RLC_PG_CNTL, SMU_CLK_SLOWDOWN_ON_PD_ENABLE, enable ? 1 : 0);
4019 }
4020 
4021 static void cz_enable_cp_power_gating(struct amdgpu_device *adev, bool enable)
4022 {
4023 	WREG32_FIELD(RLC_PG_CNTL, CP_PG_DISABLE, enable ? 0 : 1);
4024 }
4025 
4026 static void gfx_v8_0_init_pg(struct amdgpu_device *adev)
4027 {
4028 	if ((adev->asic_type == CHIP_CARRIZO) ||
4029 	    (adev->asic_type == CHIP_STONEY)) {
4030 		gfx_v8_0_init_csb(adev);
4031 		gfx_v8_0_init_save_restore_list(adev);
4032 		gfx_v8_0_enable_save_restore_machine(adev);
4033 		WREG32(mmRLC_JUMP_TABLE_RESTORE, adev->gfx.rlc.cp_table_gpu_addr >> 8);
4034 		gfx_v8_0_init_power_gating(adev);
4035 		WREG32(mmRLC_PG_ALWAYS_ON_CU_MASK, adev->gfx.cu_info.ao_cu_mask);
4036 	} else if ((adev->asic_type == CHIP_POLARIS11) ||
4037 		   (adev->asic_type == CHIP_POLARIS12) ||
4038 		   (adev->asic_type == CHIP_VEGAM)) {
4039 		gfx_v8_0_init_csb(adev);
4040 		gfx_v8_0_init_save_restore_list(adev);
4041 		gfx_v8_0_enable_save_restore_machine(adev);
4042 		gfx_v8_0_init_power_gating(adev);
4043 	}
4044 
4045 }
4046 
4047 static void gfx_v8_0_rlc_stop(struct amdgpu_device *adev)
4048 {
4049 	WREG32_FIELD(RLC_CNTL, RLC_ENABLE_F32, 0);
4050 
4051 	gfx_v8_0_enable_gui_idle_interrupt(adev, false);
4052 	gfx_v8_0_wait_for_rlc_serdes(adev);
4053 }
4054 
4055 static void gfx_v8_0_rlc_reset(struct amdgpu_device *adev)
4056 {
4057 	WREG32_FIELD(GRBM_SOFT_RESET, SOFT_RESET_RLC, 1);
4058 	udelay(50);
4059 
4060 	WREG32_FIELD(GRBM_SOFT_RESET, SOFT_RESET_RLC, 0);
4061 	udelay(50);
4062 }
4063 
4064 static void gfx_v8_0_rlc_start(struct amdgpu_device *adev)
4065 {
4066 	WREG32_FIELD(RLC_CNTL, RLC_ENABLE_F32, 1);
4067 
4068 	/* carrizo do enable cp interrupt after cp inited */
4069 	if (!(adev->flags & AMD_IS_APU))
4070 		gfx_v8_0_enable_gui_idle_interrupt(adev, true);
4071 
4072 	udelay(50);
4073 }
4074 
4075 static int gfx_v8_0_rlc_resume(struct amdgpu_device *adev)
4076 {
4077 	if (amdgpu_sriov_vf(adev)) {
4078 		gfx_v8_0_init_csb(adev);
4079 		return 0;
4080 	}
4081 
4082 	adev->gfx.rlc.funcs->stop(adev);
4083 	adev->gfx.rlc.funcs->reset(adev);
4084 	gfx_v8_0_init_pg(adev);
4085 	adev->gfx.rlc.funcs->start(adev);
4086 
4087 	return 0;
4088 }
4089 
4090 static void gfx_v8_0_cp_gfx_enable(struct amdgpu_device *adev, bool enable)
4091 {
4092 	u32 tmp = RREG32(mmCP_ME_CNTL);
4093 
4094 	if (enable) {
4095 		tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, ME_HALT, 0);
4096 		tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, PFP_HALT, 0);
4097 		tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, CE_HALT, 0);
4098 	} else {
4099 		tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, ME_HALT, 1);
4100 		tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, PFP_HALT, 1);
4101 		tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, CE_HALT, 1);
4102 	}
4103 	WREG32(mmCP_ME_CNTL, tmp);
4104 	udelay(50);
4105 }
4106 
4107 static u32 gfx_v8_0_get_csb_size(struct amdgpu_device *adev)
4108 {
4109 	u32 count = 0;
4110 	const struct cs_section_def *sect = NULL;
4111 	const struct cs_extent_def *ext = NULL;
4112 
4113 	/* begin clear state */
4114 	count += 2;
4115 	/* context control state */
4116 	count += 3;
4117 
4118 	for (sect = vi_cs_data; sect->section != NULL; ++sect) {
4119 		for (ext = sect->section; ext->extent != NULL; ++ext) {
4120 			if (sect->id == SECT_CONTEXT)
4121 				count += 2 + ext->reg_count;
4122 			else
4123 				return 0;
4124 		}
4125 	}
4126 	/* pa_sc_raster_config/pa_sc_raster_config1 */
4127 	count += 4;
4128 	/* end clear state */
4129 	count += 2;
4130 	/* clear state */
4131 	count += 2;
4132 
4133 	return count;
4134 }
4135 
4136 static int gfx_v8_0_cp_gfx_start(struct amdgpu_device *adev)
4137 {
4138 	struct amdgpu_ring *ring = &adev->gfx.gfx_ring[0];
4139 	const struct cs_section_def *sect = NULL;
4140 	const struct cs_extent_def *ext = NULL;
4141 	int r, i;
4142 
4143 	/* init the CP */
4144 	WREG32(mmCP_MAX_CONTEXT, adev->gfx.config.max_hw_contexts - 1);
4145 	WREG32(mmCP_ENDIAN_SWAP, 0);
4146 	WREG32(mmCP_DEVICE_ID, 1);
4147 
4148 	gfx_v8_0_cp_gfx_enable(adev, true);
4149 
4150 	r = amdgpu_ring_alloc(ring, gfx_v8_0_get_csb_size(adev) + 4);
4151 	if (r) {
4152 		DRM_ERROR("amdgpu: cp failed to lock ring (%d).\n", r);
4153 		return r;
4154 	}
4155 
4156 	/* clear state buffer */
4157 	amdgpu_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
4158 	amdgpu_ring_write(ring, PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
4159 
4160 	amdgpu_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1));
4161 	amdgpu_ring_write(ring, 0x80000000);
4162 	amdgpu_ring_write(ring, 0x80000000);
4163 
4164 	for (sect = vi_cs_data; sect->section != NULL; ++sect) {
4165 		for (ext = sect->section; ext->extent != NULL; ++ext) {
4166 			if (sect->id == SECT_CONTEXT) {
4167 				amdgpu_ring_write(ring,
4168 				       PACKET3(PACKET3_SET_CONTEXT_REG,
4169 					       ext->reg_count));
4170 				amdgpu_ring_write(ring,
4171 				       ext->reg_index - PACKET3_SET_CONTEXT_REG_START);
4172 				for (i = 0; i < ext->reg_count; i++)
4173 					amdgpu_ring_write(ring, ext->extent[i]);
4174 			}
4175 		}
4176 	}
4177 
4178 	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONTEXT_REG, 2));
4179 	amdgpu_ring_write(ring, mmPA_SC_RASTER_CONFIG - PACKET3_SET_CONTEXT_REG_START);
4180 	amdgpu_ring_write(ring, adev->gfx.config.rb_config[0][0].raster_config);
4181 	amdgpu_ring_write(ring, adev->gfx.config.rb_config[0][0].raster_config_1);
4182 
4183 	amdgpu_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
4184 	amdgpu_ring_write(ring, PACKET3_PREAMBLE_END_CLEAR_STATE);
4185 
4186 	amdgpu_ring_write(ring, PACKET3(PACKET3_CLEAR_STATE, 0));
4187 	amdgpu_ring_write(ring, 0);
4188 
4189 	/* init the CE partitions */
4190 	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_BASE, 2));
4191 	amdgpu_ring_write(ring, PACKET3_BASE_INDEX(CE_PARTITION_BASE));
4192 	amdgpu_ring_write(ring, 0x8000);
4193 	amdgpu_ring_write(ring, 0x8000);
4194 
4195 	amdgpu_ring_commit(ring);
4196 
4197 	return 0;
4198 }
4199 static void gfx_v8_0_set_cpg_door_bell(struct amdgpu_device *adev, struct amdgpu_ring *ring)
4200 {
4201 	u32 tmp;
4202 	/* no gfx doorbells on iceland */
4203 	if (adev->asic_type == CHIP_TOPAZ)
4204 		return;
4205 
4206 	tmp = RREG32(mmCP_RB_DOORBELL_CONTROL);
4207 
4208 	if (ring->use_doorbell) {
4209 		tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL,
4210 				DOORBELL_OFFSET, ring->doorbell_index);
4211 		tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL,
4212 						DOORBELL_HIT, 0);
4213 		tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL,
4214 					    DOORBELL_EN, 1);
4215 	} else {
4216 		tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL, DOORBELL_EN, 0);
4217 	}
4218 
4219 	WREG32(mmCP_RB_DOORBELL_CONTROL, tmp);
4220 
4221 	if (adev->flags & AMD_IS_APU)
4222 		return;
4223 
4224 	tmp = REG_SET_FIELD(0, CP_RB_DOORBELL_RANGE_LOWER,
4225 					DOORBELL_RANGE_LOWER,
4226 					adev->doorbell_index.gfx_ring0);
4227 	WREG32(mmCP_RB_DOORBELL_RANGE_LOWER, tmp);
4228 
4229 	WREG32(mmCP_RB_DOORBELL_RANGE_UPPER,
4230 		CP_RB_DOORBELL_RANGE_UPPER__DOORBELL_RANGE_UPPER_MASK);
4231 }
4232 
4233 static int gfx_v8_0_cp_gfx_resume(struct amdgpu_device *adev)
4234 {
4235 	struct amdgpu_ring *ring;
4236 	u32 tmp;
4237 	u32 rb_bufsz;
4238 	u64 rb_addr, rptr_addr, wptr_gpu_addr;
4239 
4240 	/* Set the write pointer delay */
4241 	WREG32(mmCP_RB_WPTR_DELAY, 0);
4242 
4243 	/* set the RB to use vmid 0 */
4244 	WREG32(mmCP_RB_VMID, 0);
4245 
4246 	/* Set ring buffer size */
4247 	ring = &adev->gfx.gfx_ring[0];
4248 	rb_bufsz = order_base_2(ring->ring_size / 8);
4249 	tmp = REG_SET_FIELD(0, CP_RB0_CNTL, RB_BUFSZ, rb_bufsz);
4250 	tmp = REG_SET_FIELD(tmp, CP_RB0_CNTL, RB_BLKSZ, rb_bufsz - 2);
4251 	tmp = REG_SET_FIELD(tmp, CP_RB0_CNTL, MTYPE, 3);
4252 	tmp = REG_SET_FIELD(tmp, CP_RB0_CNTL, MIN_IB_AVAILSZ, 1);
4253 #ifdef __BIG_ENDIAN
4254 	tmp = REG_SET_FIELD(tmp, CP_RB0_CNTL, BUF_SWAP, 1);
4255 #endif
4256 	WREG32(mmCP_RB0_CNTL, tmp);
4257 
4258 	/* Initialize the ring buffer's read and write pointers */
4259 	WREG32(mmCP_RB0_CNTL, tmp | CP_RB0_CNTL__RB_RPTR_WR_ENA_MASK);
4260 	ring->wptr = 0;
4261 	WREG32(mmCP_RB0_WPTR, lower_32_bits(ring->wptr));
4262 
4263 	/* set the wb address wether it's enabled or not */
4264 	rptr_addr = ring->rptr_gpu_addr;
4265 	WREG32(mmCP_RB0_RPTR_ADDR, lower_32_bits(rptr_addr));
4266 	WREG32(mmCP_RB0_RPTR_ADDR_HI, upper_32_bits(rptr_addr) & 0xFF);
4267 
4268 	wptr_gpu_addr = ring->wptr_gpu_addr;
4269 	WREG32(mmCP_RB_WPTR_POLL_ADDR_LO, lower_32_bits(wptr_gpu_addr));
4270 	WREG32(mmCP_RB_WPTR_POLL_ADDR_HI, upper_32_bits(wptr_gpu_addr));
4271 	mdelay(1);
4272 	WREG32(mmCP_RB0_CNTL, tmp);
4273 
4274 	rb_addr = ring->gpu_addr >> 8;
4275 	WREG32(mmCP_RB0_BASE, rb_addr);
4276 	WREG32(mmCP_RB0_BASE_HI, upper_32_bits(rb_addr));
4277 
4278 	gfx_v8_0_set_cpg_door_bell(adev, ring);
4279 	/* start the ring */
4280 	amdgpu_ring_clear_ring(ring);
4281 	gfx_v8_0_cp_gfx_start(adev);
4282 
4283 	return 0;
4284 }
4285 
4286 static void gfx_v8_0_cp_compute_enable(struct amdgpu_device *adev, bool enable)
4287 {
4288 	if (enable) {
4289 		WREG32(mmCP_MEC_CNTL, 0);
4290 	} else {
4291 		WREG32(mmCP_MEC_CNTL, (CP_MEC_CNTL__MEC_ME1_HALT_MASK | CP_MEC_CNTL__MEC_ME2_HALT_MASK));
4292 		adev->gfx.kiq[0].ring.sched.ready = false;
4293 	}
4294 	udelay(50);
4295 }
4296 
4297 /* KIQ functions */
4298 static void gfx_v8_0_kiq_setting(struct amdgpu_ring *ring)
4299 {
4300 	uint32_t tmp;
4301 	struct amdgpu_device *adev = ring->adev;
4302 
4303 	/* tell RLC which is KIQ queue */
4304 	tmp = RREG32(mmRLC_CP_SCHEDULERS);
4305 	tmp &= 0xffffff00;
4306 	tmp |= (ring->me << 5) | (ring->pipe << 3) | (ring->queue);
4307 	WREG32(mmRLC_CP_SCHEDULERS, tmp);
4308 	tmp |= 0x80;
4309 	WREG32(mmRLC_CP_SCHEDULERS, tmp);
4310 }
4311 
4312 static int gfx_v8_0_kiq_kcq_enable(struct amdgpu_device *adev)
4313 {
4314 	struct amdgpu_ring *kiq_ring = &adev->gfx.kiq[0].ring;
4315 	uint64_t queue_mask = 0;
4316 	int r, i;
4317 
4318 	for (i = 0; i < AMDGPU_MAX_COMPUTE_QUEUES; ++i) {
4319 		if (!test_bit(i, adev->gfx.mec_bitmap[0].queue_bitmap))
4320 			continue;
4321 
4322 		/* This situation may be hit in the future if a new HW
4323 		 * generation exposes more than 64 queues. If so, the
4324 		 * definition of queue_mask needs updating */
4325 		if (WARN_ON(i >= (sizeof(queue_mask)*8))) {
4326 			DRM_ERROR("Invalid KCQ enabled: %d\n", i);
4327 			break;
4328 		}
4329 
4330 		queue_mask |= (1ull << i);
4331 	}
4332 
4333 	r = amdgpu_ring_alloc(kiq_ring, (8 * adev->gfx.num_compute_rings) + 8);
4334 	if (r) {
4335 		DRM_ERROR("Failed to lock KIQ (%d).\n", r);
4336 		return r;
4337 	}
4338 	/* set resources */
4339 	amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_SET_RESOURCES, 6));
4340 	amdgpu_ring_write(kiq_ring, 0);	/* vmid_mask:0 queue_type:0 (KIQ) */
4341 	amdgpu_ring_write(kiq_ring, lower_32_bits(queue_mask));	/* queue mask lo */
4342 	amdgpu_ring_write(kiq_ring, upper_32_bits(queue_mask));	/* queue mask hi */
4343 	amdgpu_ring_write(kiq_ring, 0);	/* gws mask lo */
4344 	amdgpu_ring_write(kiq_ring, 0);	/* gws mask hi */
4345 	amdgpu_ring_write(kiq_ring, 0);	/* oac mask */
4346 	amdgpu_ring_write(kiq_ring, 0);	/* gds heap base:0, gds heap size:0 */
4347 	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
4348 		struct amdgpu_ring *ring = &adev->gfx.compute_ring[i];
4349 		uint64_t mqd_addr = amdgpu_bo_gpu_offset(ring->mqd_obj);
4350 		uint64_t wptr_addr = ring->wptr_gpu_addr;
4351 
4352 		/* map queues */
4353 		amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_MAP_QUEUES, 5));
4354 		/* Q_sel:0, vmid:0, vidmem: 1, engine:0, num_Q:1*/
4355 		amdgpu_ring_write(kiq_ring,
4356 				  PACKET3_MAP_QUEUES_NUM_QUEUES(1));
4357 		amdgpu_ring_write(kiq_ring,
4358 				  PACKET3_MAP_QUEUES_DOORBELL_OFFSET(ring->doorbell_index) |
4359 				  PACKET3_MAP_QUEUES_QUEUE(ring->queue) |
4360 				  PACKET3_MAP_QUEUES_PIPE(ring->pipe) |
4361 				  PACKET3_MAP_QUEUES_ME(ring->me == 1 ? 0 : 1)); /* doorbell */
4362 		amdgpu_ring_write(kiq_ring, lower_32_bits(mqd_addr));
4363 		amdgpu_ring_write(kiq_ring, upper_32_bits(mqd_addr));
4364 		amdgpu_ring_write(kiq_ring, lower_32_bits(wptr_addr));
4365 		amdgpu_ring_write(kiq_ring, upper_32_bits(wptr_addr));
4366 	}
4367 
4368 	amdgpu_ring_commit(kiq_ring);
4369 
4370 	return 0;
4371 }
4372 
4373 static int gfx_v8_0_deactivate_hqd(struct amdgpu_device *adev, u32 req)
4374 {
4375 	int i, r = 0;
4376 
4377 	if (RREG32(mmCP_HQD_ACTIVE) & CP_HQD_ACTIVE__ACTIVE_MASK) {
4378 		WREG32_FIELD(CP_HQD_DEQUEUE_REQUEST, DEQUEUE_REQ, req);
4379 		for (i = 0; i < adev->usec_timeout; i++) {
4380 			if (!(RREG32(mmCP_HQD_ACTIVE) & CP_HQD_ACTIVE__ACTIVE_MASK))
4381 				break;
4382 			udelay(1);
4383 		}
4384 		if (i == adev->usec_timeout)
4385 			r = -ETIMEDOUT;
4386 	}
4387 	WREG32(mmCP_HQD_DEQUEUE_REQUEST, 0);
4388 	WREG32(mmCP_HQD_PQ_RPTR, 0);
4389 	WREG32(mmCP_HQD_PQ_WPTR, 0);
4390 
4391 	return r;
4392 }
4393 
4394 static void gfx_v8_0_mqd_set_priority(struct amdgpu_ring *ring, struct vi_mqd *mqd)
4395 {
4396 	struct amdgpu_device *adev = ring->adev;
4397 
4398 	if (ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE) {
4399 		if (amdgpu_gfx_is_high_priority_compute_queue(adev, ring)) {
4400 			mqd->cp_hqd_pipe_priority = AMDGPU_GFX_PIPE_PRIO_HIGH;
4401 			mqd->cp_hqd_queue_priority =
4402 				AMDGPU_GFX_QUEUE_PRIORITY_MAXIMUM;
4403 		}
4404 	}
4405 }
4406 
4407 static int gfx_v8_0_mqd_init(struct amdgpu_ring *ring)
4408 {
4409 	struct amdgpu_device *adev = ring->adev;
4410 	struct vi_mqd *mqd = ring->mqd_ptr;
4411 	uint64_t hqd_gpu_addr, wb_gpu_addr, eop_base_addr;
4412 	uint32_t tmp;
4413 
4414 	mqd->header = 0xC0310800;
4415 	mqd->compute_pipelinestat_enable = 0x00000001;
4416 	mqd->compute_static_thread_mgmt_se0 = 0xffffffff;
4417 	mqd->compute_static_thread_mgmt_se1 = 0xffffffff;
4418 	mqd->compute_static_thread_mgmt_se2 = 0xffffffff;
4419 	mqd->compute_static_thread_mgmt_se3 = 0xffffffff;
4420 	mqd->compute_misc_reserved = 0x00000003;
4421 	mqd->dynamic_cu_mask_addr_lo = lower_32_bits(ring->mqd_gpu_addr
4422 						     + offsetof(struct vi_mqd_allocation, dynamic_cu_mask));
4423 	mqd->dynamic_cu_mask_addr_hi = upper_32_bits(ring->mqd_gpu_addr
4424 						     + offsetof(struct vi_mqd_allocation, dynamic_cu_mask));
4425 	eop_base_addr = ring->eop_gpu_addr >> 8;
4426 	mqd->cp_hqd_eop_base_addr_lo = eop_base_addr;
4427 	mqd->cp_hqd_eop_base_addr_hi = upper_32_bits(eop_base_addr);
4428 
4429 	/* set the EOP size, register value is 2^(EOP_SIZE+1) dwords */
4430 	tmp = RREG32(mmCP_HQD_EOP_CONTROL);
4431 	tmp = REG_SET_FIELD(tmp, CP_HQD_EOP_CONTROL, EOP_SIZE,
4432 			(order_base_2(GFX8_MEC_HPD_SIZE / 4) - 1));
4433 
4434 	mqd->cp_hqd_eop_control = tmp;
4435 
4436 	/* enable doorbell? */
4437 	tmp = REG_SET_FIELD(RREG32(mmCP_HQD_PQ_DOORBELL_CONTROL),
4438 			    CP_HQD_PQ_DOORBELL_CONTROL,
4439 			    DOORBELL_EN,
4440 			    ring->use_doorbell ? 1 : 0);
4441 
4442 	mqd->cp_hqd_pq_doorbell_control = tmp;
4443 
4444 	/* set the pointer to the MQD */
4445 	mqd->cp_mqd_base_addr_lo = ring->mqd_gpu_addr & 0xfffffffc;
4446 	mqd->cp_mqd_base_addr_hi = upper_32_bits(ring->mqd_gpu_addr);
4447 
4448 	/* set MQD vmid to 0 */
4449 	tmp = RREG32(mmCP_MQD_CONTROL);
4450 	tmp = REG_SET_FIELD(tmp, CP_MQD_CONTROL, VMID, 0);
4451 	mqd->cp_mqd_control = tmp;
4452 
4453 	/* set the pointer to the HQD, this is similar CP_RB0_BASE/_HI */
4454 	hqd_gpu_addr = ring->gpu_addr >> 8;
4455 	mqd->cp_hqd_pq_base_lo = hqd_gpu_addr;
4456 	mqd->cp_hqd_pq_base_hi = upper_32_bits(hqd_gpu_addr);
4457 
4458 	/* set up the HQD, this is similar to CP_RB0_CNTL */
4459 	tmp = RREG32(mmCP_HQD_PQ_CONTROL);
4460 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, QUEUE_SIZE,
4461 			    (order_base_2(ring->ring_size / 4) - 1));
4462 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, RPTR_BLOCK_SIZE,
4463 			(order_base_2(AMDGPU_GPU_PAGE_SIZE / 4) - 1));
4464 #ifdef __BIG_ENDIAN
4465 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, ENDIAN_SWAP, 1);
4466 #endif
4467 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, UNORD_DISPATCH, 0);
4468 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, ROQ_PQ_IB_FLIP, 0);
4469 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, PRIV_STATE, 1);
4470 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, KMD_QUEUE, 1);
4471 	mqd->cp_hqd_pq_control = tmp;
4472 
4473 	/* set the wb address whether it's enabled or not */
4474 	wb_gpu_addr = ring->rptr_gpu_addr;
4475 	mqd->cp_hqd_pq_rptr_report_addr_lo = wb_gpu_addr & 0xfffffffc;
4476 	mqd->cp_hqd_pq_rptr_report_addr_hi =
4477 		upper_32_bits(wb_gpu_addr) & 0xffff;
4478 
4479 	/* only used if CP_PQ_WPTR_POLL_CNTL.CP_PQ_WPTR_POLL_CNTL__EN_MASK=1 */
4480 	wb_gpu_addr = ring->wptr_gpu_addr;
4481 	mqd->cp_hqd_pq_wptr_poll_addr_lo = wb_gpu_addr & 0xfffffffc;
4482 	mqd->cp_hqd_pq_wptr_poll_addr_hi = upper_32_bits(wb_gpu_addr) & 0xffff;
4483 
4484 	tmp = 0;
4485 	/* enable the doorbell if requested */
4486 	if (ring->use_doorbell) {
4487 		tmp = RREG32(mmCP_HQD_PQ_DOORBELL_CONTROL);
4488 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
4489 				DOORBELL_OFFSET, ring->doorbell_index);
4490 
4491 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
4492 					 DOORBELL_EN, 1);
4493 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
4494 					 DOORBELL_SOURCE, 0);
4495 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
4496 					 DOORBELL_HIT, 0);
4497 	}
4498 
4499 	mqd->cp_hqd_pq_doorbell_control = tmp;
4500 
4501 	/* reset read and write pointers, similar to CP_RB0_WPTR/_RPTR */
4502 	ring->wptr = 0;
4503 	mqd->cp_hqd_pq_wptr = ring->wptr;
4504 	mqd->cp_hqd_pq_rptr = RREG32(mmCP_HQD_PQ_RPTR);
4505 
4506 	/* set the vmid for the queue */
4507 	mqd->cp_hqd_vmid = 0;
4508 
4509 	tmp = RREG32(mmCP_HQD_PERSISTENT_STATE);
4510 	tmp = REG_SET_FIELD(tmp, CP_HQD_PERSISTENT_STATE, PRELOAD_SIZE, 0x53);
4511 	mqd->cp_hqd_persistent_state = tmp;
4512 
4513 	/* set MTYPE */
4514 	tmp = RREG32(mmCP_HQD_IB_CONTROL);
4515 	tmp = REG_SET_FIELD(tmp, CP_HQD_IB_CONTROL, MIN_IB_AVAIL_SIZE, 3);
4516 	tmp = REG_SET_FIELD(tmp, CP_HQD_IB_CONTROL, MTYPE, 3);
4517 	mqd->cp_hqd_ib_control = tmp;
4518 
4519 	tmp = RREG32(mmCP_HQD_IQ_TIMER);
4520 	tmp = REG_SET_FIELD(tmp, CP_HQD_IQ_TIMER, MTYPE, 3);
4521 	mqd->cp_hqd_iq_timer = tmp;
4522 
4523 	tmp = RREG32(mmCP_HQD_CTX_SAVE_CONTROL);
4524 	tmp = REG_SET_FIELD(tmp, CP_HQD_CTX_SAVE_CONTROL, MTYPE, 3);
4525 	mqd->cp_hqd_ctx_save_control = tmp;
4526 
4527 	/* defaults */
4528 	mqd->cp_hqd_eop_rptr = RREG32(mmCP_HQD_EOP_RPTR);
4529 	mqd->cp_hqd_eop_wptr = RREG32(mmCP_HQD_EOP_WPTR);
4530 	mqd->cp_hqd_ctx_save_base_addr_lo = RREG32(mmCP_HQD_CTX_SAVE_BASE_ADDR_LO);
4531 	mqd->cp_hqd_ctx_save_base_addr_hi = RREG32(mmCP_HQD_CTX_SAVE_BASE_ADDR_HI);
4532 	mqd->cp_hqd_cntl_stack_offset = RREG32(mmCP_HQD_CNTL_STACK_OFFSET);
4533 	mqd->cp_hqd_cntl_stack_size = RREG32(mmCP_HQD_CNTL_STACK_SIZE);
4534 	mqd->cp_hqd_wg_state_offset = RREG32(mmCP_HQD_WG_STATE_OFFSET);
4535 	mqd->cp_hqd_ctx_save_size = RREG32(mmCP_HQD_CTX_SAVE_SIZE);
4536 	mqd->cp_hqd_eop_done_events = RREG32(mmCP_HQD_EOP_EVENTS);
4537 	mqd->cp_hqd_error = RREG32(mmCP_HQD_ERROR);
4538 	mqd->cp_hqd_eop_wptr_mem = RREG32(mmCP_HQD_EOP_WPTR_MEM);
4539 	mqd->cp_hqd_eop_dones = RREG32(mmCP_HQD_EOP_DONES);
4540 
4541 	/* set static priority for a queue/ring */
4542 	gfx_v8_0_mqd_set_priority(ring, mqd);
4543 	mqd->cp_hqd_quantum = RREG32(mmCP_HQD_QUANTUM);
4544 
4545 	/* map_queues packet doesn't need activate the queue,
4546 	 * so only kiq need set this field.
4547 	 */
4548 	if (ring->funcs->type == AMDGPU_RING_TYPE_KIQ)
4549 		mqd->cp_hqd_active = 1;
4550 
4551 	return 0;
4552 }
4553 
4554 static int gfx_v8_0_mqd_commit(struct amdgpu_device *adev,
4555 			struct vi_mqd *mqd)
4556 {
4557 	uint32_t mqd_reg;
4558 	uint32_t *mqd_data;
4559 
4560 	/* HQD registers extend from mmCP_MQD_BASE_ADDR to mmCP_HQD_ERROR */
4561 	mqd_data = &mqd->cp_mqd_base_addr_lo;
4562 
4563 	/* disable wptr polling */
4564 	WREG32_FIELD(CP_PQ_WPTR_POLL_CNTL, EN, 0);
4565 
4566 	/* program all HQD registers */
4567 	for (mqd_reg = mmCP_HQD_VMID; mqd_reg <= mmCP_HQD_EOP_CONTROL; mqd_reg++)
4568 		WREG32(mqd_reg, mqd_data[mqd_reg - mmCP_MQD_BASE_ADDR]);
4569 
4570 	/* Tonga errata: EOP RPTR/WPTR should be left unmodified.
4571 	 * This is safe since EOP RPTR==WPTR for any inactive HQD
4572 	 * on ASICs that do not support context-save.
4573 	 * EOP writes/reads can start anywhere in the ring.
4574 	 */
4575 	if (adev->asic_type != CHIP_TONGA) {
4576 		WREG32(mmCP_HQD_EOP_RPTR, mqd->cp_hqd_eop_rptr);
4577 		WREG32(mmCP_HQD_EOP_WPTR, mqd->cp_hqd_eop_wptr);
4578 		WREG32(mmCP_HQD_EOP_WPTR_MEM, mqd->cp_hqd_eop_wptr_mem);
4579 	}
4580 
4581 	for (mqd_reg = mmCP_HQD_EOP_EVENTS; mqd_reg <= mmCP_HQD_ERROR; mqd_reg++)
4582 		WREG32(mqd_reg, mqd_data[mqd_reg - mmCP_MQD_BASE_ADDR]);
4583 
4584 	/* activate the HQD */
4585 	for (mqd_reg = mmCP_MQD_BASE_ADDR; mqd_reg <= mmCP_HQD_ACTIVE; mqd_reg++)
4586 		WREG32(mqd_reg, mqd_data[mqd_reg - mmCP_MQD_BASE_ADDR]);
4587 
4588 	return 0;
4589 }
4590 
4591 static int gfx_v8_0_kiq_init_queue(struct amdgpu_ring *ring)
4592 {
4593 	struct amdgpu_device *adev = ring->adev;
4594 	struct vi_mqd *mqd = ring->mqd_ptr;
4595 
4596 	gfx_v8_0_kiq_setting(ring);
4597 
4598 	if (amdgpu_in_reset(adev)) { /* for GPU_RESET case */
4599 		/* reset MQD to a clean status */
4600 		if (adev->gfx.kiq[0].mqd_backup)
4601 			memcpy(mqd, adev->gfx.kiq[0].mqd_backup, sizeof(struct vi_mqd_allocation));
4602 
4603 		/* reset ring buffer */
4604 		ring->wptr = 0;
4605 		amdgpu_ring_clear_ring(ring);
4606 		mutex_lock(&adev->srbm_mutex);
4607 		vi_srbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
4608 		gfx_v8_0_mqd_commit(adev, mqd);
4609 		vi_srbm_select(adev, 0, 0, 0, 0);
4610 		mutex_unlock(&adev->srbm_mutex);
4611 	} else {
4612 		memset((void *)mqd, 0, sizeof(struct vi_mqd_allocation));
4613 		((struct vi_mqd_allocation *)mqd)->dynamic_cu_mask = 0xFFFFFFFF;
4614 		((struct vi_mqd_allocation *)mqd)->dynamic_rb_mask = 0xFFFFFFFF;
4615 		if (amdgpu_sriov_vf(adev) && adev->in_suspend)
4616 			amdgpu_ring_clear_ring(ring);
4617 		mutex_lock(&adev->srbm_mutex);
4618 		vi_srbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
4619 		gfx_v8_0_mqd_init(ring);
4620 		gfx_v8_0_mqd_commit(adev, mqd);
4621 		vi_srbm_select(adev, 0, 0, 0, 0);
4622 		mutex_unlock(&adev->srbm_mutex);
4623 
4624 		if (adev->gfx.kiq[0].mqd_backup)
4625 			memcpy(adev->gfx.kiq[0].mqd_backup, mqd, sizeof(struct vi_mqd_allocation));
4626 	}
4627 
4628 	return 0;
4629 }
4630 
4631 static int gfx_v8_0_kcq_init_queue(struct amdgpu_ring *ring)
4632 {
4633 	struct amdgpu_device *adev = ring->adev;
4634 	struct vi_mqd *mqd = ring->mqd_ptr;
4635 	int mqd_idx = ring - &adev->gfx.compute_ring[0];
4636 
4637 	if (!amdgpu_in_reset(adev) && !adev->in_suspend) {
4638 		memset((void *)mqd, 0, sizeof(struct vi_mqd_allocation));
4639 		((struct vi_mqd_allocation *)mqd)->dynamic_cu_mask = 0xFFFFFFFF;
4640 		((struct vi_mqd_allocation *)mqd)->dynamic_rb_mask = 0xFFFFFFFF;
4641 		mutex_lock(&adev->srbm_mutex);
4642 		vi_srbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
4643 		gfx_v8_0_mqd_init(ring);
4644 		vi_srbm_select(adev, 0, 0, 0, 0);
4645 		mutex_unlock(&adev->srbm_mutex);
4646 
4647 		if (adev->gfx.mec.mqd_backup[mqd_idx])
4648 			memcpy(adev->gfx.mec.mqd_backup[mqd_idx], mqd, sizeof(struct vi_mqd_allocation));
4649 	} else {
4650 		/* restore MQD to a clean status */
4651 		if (adev->gfx.mec.mqd_backup[mqd_idx])
4652 			memcpy(mqd, adev->gfx.mec.mqd_backup[mqd_idx], sizeof(struct vi_mqd_allocation));
4653 		/* reset ring buffer */
4654 		ring->wptr = 0;
4655 		amdgpu_ring_clear_ring(ring);
4656 	}
4657 	return 0;
4658 }
4659 
4660 static void gfx_v8_0_set_mec_doorbell_range(struct amdgpu_device *adev)
4661 {
4662 	if (adev->asic_type > CHIP_TONGA) {
4663 		WREG32(mmCP_MEC_DOORBELL_RANGE_LOWER, adev->doorbell_index.kiq << 2);
4664 		WREG32(mmCP_MEC_DOORBELL_RANGE_UPPER, adev->doorbell_index.mec_ring7 << 2);
4665 	}
4666 	/* enable doorbells */
4667 	WREG32_FIELD(CP_PQ_STATUS, DOORBELL_ENABLE, 1);
4668 }
4669 
4670 static int gfx_v8_0_kiq_resume(struct amdgpu_device *adev)
4671 {
4672 	struct amdgpu_ring *ring;
4673 	int r;
4674 
4675 	ring = &adev->gfx.kiq[0].ring;
4676 
4677 	r = amdgpu_bo_reserve(ring->mqd_obj, false);
4678 	if (unlikely(r != 0))
4679 		return r;
4680 
4681 	r = amdgpu_bo_kmap(ring->mqd_obj, &ring->mqd_ptr);
4682 	if (unlikely(r != 0)) {
4683 		amdgpu_bo_unreserve(ring->mqd_obj);
4684 		return r;
4685 	}
4686 
4687 	gfx_v8_0_kiq_init_queue(ring);
4688 	amdgpu_bo_kunmap(ring->mqd_obj);
4689 	ring->mqd_ptr = NULL;
4690 	amdgpu_bo_unreserve(ring->mqd_obj);
4691 	return 0;
4692 }
4693 
4694 static int gfx_v8_0_kcq_resume(struct amdgpu_device *adev)
4695 {
4696 	struct amdgpu_ring *ring = NULL;
4697 	int r = 0, i;
4698 
4699 	gfx_v8_0_cp_compute_enable(adev, true);
4700 
4701 	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
4702 		ring = &adev->gfx.compute_ring[i];
4703 
4704 		r = amdgpu_bo_reserve(ring->mqd_obj, false);
4705 		if (unlikely(r != 0))
4706 			goto done;
4707 		r = amdgpu_bo_kmap(ring->mqd_obj, &ring->mqd_ptr);
4708 		if (!r) {
4709 			r = gfx_v8_0_kcq_init_queue(ring);
4710 			amdgpu_bo_kunmap(ring->mqd_obj);
4711 			ring->mqd_ptr = NULL;
4712 		}
4713 		amdgpu_bo_unreserve(ring->mqd_obj);
4714 		if (r)
4715 			goto done;
4716 	}
4717 
4718 	gfx_v8_0_set_mec_doorbell_range(adev);
4719 
4720 	r = gfx_v8_0_kiq_kcq_enable(adev);
4721 	if (r)
4722 		goto done;
4723 
4724 done:
4725 	return r;
4726 }
4727 
4728 static int gfx_v8_0_cp_test_all_rings(struct amdgpu_device *adev)
4729 {
4730 	int r, i;
4731 	struct amdgpu_ring *ring;
4732 
4733 	/* collect all the ring_tests here, gfx, kiq, compute */
4734 	ring = &adev->gfx.gfx_ring[0];
4735 	r = amdgpu_ring_test_helper(ring);
4736 	if (r)
4737 		return r;
4738 
4739 	ring = &adev->gfx.kiq[0].ring;
4740 	r = amdgpu_ring_test_helper(ring);
4741 	if (r)
4742 		return r;
4743 
4744 	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
4745 		ring = &adev->gfx.compute_ring[i];
4746 		amdgpu_ring_test_helper(ring);
4747 	}
4748 
4749 	return 0;
4750 }
4751 
4752 static int gfx_v8_0_cp_resume(struct amdgpu_device *adev)
4753 {
4754 	int r;
4755 
4756 	if (!(adev->flags & AMD_IS_APU))
4757 		gfx_v8_0_enable_gui_idle_interrupt(adev, false);
4758 
4759 	r = gfx_v8_0_kiq_resume(adev);
4760 	if (r)
4761 		return r;
4762 
4763 	r = gfx_v8_0_cp_gfx_resume(adev);
4764 	if (r)
4765 		return r;
4766 
4767 	r = gfx_v8_0_kcq_resume(adev);
4768 	if (r)
4769 		return r;
4770 
4771 	r = gfx_v8_0_cp_test_all_rings(adev);
4772 	if (r)
4773 		return r;
4774 
4775 	gfx_v8_0_enable_gui_idle_interrupt(adev, true);
4776 
4777 	return 0;
4778 }
4779 
4780 static void gfx_v8_0_cp_enable(struct amdgpu_device *adev, bool enable)
4781 {
4782 	gfx_v8_0_cp_gfx_enable(adev, enable);
4783 	gfx_v8_0_cp_compute_enable(adev, enable);
4784 }
4785 
4786 static int gfx_v8_0_hw_init(void *handle)
4787 {
4788 	int r;
4789 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4790 
4791 	gfx_v8_0_init_golden_registers(adev);
4792 	gfx_v8_0_constants_init(adev);
4793 
4794 	r = adev->gfx.rlc.funcs->resume(adev);
4795 	if (r)
4796 		return r;
4797 
4798 	r = gfx_v8_0_cp_resume(adev);
4799 
4800 	return r;
4801 }
4802 
4803 static int gfx_v8_0_kcq_disable(struct amdgpu_device *adev)
4804 {
4805 	int r, i;
4806 	struct amdgpu_ring *kiq_ring = &adev->gfx.kiq[0].ring;
4807 
4808 	r = amdgpu_ring_alloc(kiq_ring, 6 * adev->gfx.num_compute_rings);
4809 	if (r)
4810 		DRM_ERROR("Failed to lock KIQ (%d).\n", r);
4811 
4812 	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
4813 		struct amdgpu_ring *ring = &adev->gfx.compute_ring[i];
4814 
4815 		amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_UNMAP_QUEUES, 4));
4816 		amdgpu_ring_write(kiq_ring, /* Q_sel: 0, vmid: 0, engine: 0, num_Q: 1 */
4817 						PACKET3_UNMAP_QUEUES_ACTION(1) | /* RESET_QUEUES */
4818 						PACKET3_UNMAP_QUEUES_QUEUE_SEL(0) |
4819 						PACKET3_UNMAP_QUEUES_ENGINE_SEL(0) |
4820 						PACKET3_UNMAP_QUEUES_NUM_QUEUES(1));
4821 		amdgpu_ring_write(kiq_ring, PACKET3_UNMAP_QUEUES_DOORBELL_OFFSET0(ring->doorbell_index));
4822 		amdgpu_ring_write(kiq_ring, 0);
4823 		amdgpu_ring_write(kiq_ring, 0);
4824 		amdgpu_ring_write(kiq_ring, 0);
4825 	}
4826 	r = amdgpu_ring_test_helper(kiq_ring);
4827 	if (r)
4828 		DRM_ERROR("KCQ disable failed\n");
4829 
4830 	return r;
4831 }
4832 
4833 static bool gfx_v8_0_is_idle(void *handle)
4834 {
4835 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4836 
4837 	if (REG_GET_FIELD(RREG32(mmGRBM_STATUS), GRBM_STATUS, GUI_ACTIVE)
4838 		|| RREG32(mmGRBM_STATUS2) != 0x8)
4839 		return false;
4840 	else
4841 		return true;
4842 }
4843 
4844 static bool gfx_v8_0_rlc_is_idle(void *handle)
4845 {
4846 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4847 
4848 	if (RREG32(mmGRBM_STATUS2) != 0x8)
4849 		return false;
4850 	else
4851 		return true;
4852 }
4853 
4854 static int gfx_v8_0_wait_for_rlc_idle(void *handle)
4855 {
4856 	unsigned int i;
4857 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4858 
4859 	for (i = 0; i < adev->usec_timeout; i++) {
4860 		if (gfx_v8_0_rlc_is_idle(handle))
4861 			return 0;
4862 
4863 		udelay(1);
4864 	}
4865 	return -ETIMEDOUT;
4866 }
4867 
4868 static int gfx_v8_0_wait_for_idle(void *handle)
4869 {
4870 	unsigned int i;
4871 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4872 
4873 	for (i = 0; i < adev->usec_timeout; i++) {
4874 		if (gfx_v8_0_is_idle(handle))
4875 			return 0;
4876 
4877 		udelay(1);
4878 	}
4879 	return -ETIMEDOUT;
4880 }
4881 
4882 static int gfx_v8_0_hw_fini(void *handle)
4883 {
4884 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4885 
4886 	amdgpu_irq_put(adev, &adev->gfx.priv_reg_irq, 0);
4887 	amdgpu_irq_put(adev, &adev->gfx.priv_inst_irq, 0);
4888 
4889 	amdgpu_irq_put(adev, &adev->gfx.cp_ecc_error_irq, 0);
4890 
4891 	amdgpu_irq_put(adev, &adev->gfx.sq_irq, 0);
4892 
4893 	/* disable KCQ to avoid CPC touch memory not valid anymore */
4894 	gfx_v8_0_kcq_disable(adev);
4895 
4896 	if (amdgpu_sriov_vf(adev)) {
4897 		pr_debug("For SRIOV client, shouldn't do anything.\n");
4898 		return 0;
4899 	}
4900 	amdgpu_gfx_rlc_enter_safe_mode(adev, 0);
4901 	if (!gfx_v8_0_wait_for_idle(adev))
4902 		gfx_v8_0_cp_enable(adev, false);
4903 	else
4904 		pr_err("cp is busy, skip halt cp\n");
4905 	if (!gfx_v8_0_wait_for_rlc_idle(adev))
4906 		adev->gfx.rlc.funcs->stop(adev);
4907 	else
4908 		pr_err("rlc is busy, skip halt rlc\n");
4909 	amdgpu_gfx_rlc_exit_safe_mode(adev, 0);
4910 
4911 	return 0;
4912 }
4913 
4914 static int gfx_v8_0_suspend(void *handle)
4915 {
4916 	return gfx_v8_0_hw_fini(handle);
4917 }
4918 
4919 static int gfx_v8_0_resume(void *handle)
4920 {
4921 	return gfx_v8_0_hw_init(handle);
4922 }
4923 
4924 static bool gfx_v8_0_check_soft_reset(void *handle)
4925 {
4926 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4927 	u32 grbm_soft_reset = 0, srbm_soft_reset = 0;
4928 	u32 tmp;
4929 
4930 	/* GRBM_STATUS */
4931 	tmp = RREG32(mmGRBM_STATUS);
4932 	if (tmp & (GRBM_STATUS__PA_BUSY_MASK | GRBM_STATUS__SC_BUSY_MASK |
4933 		   GRBM_STATUS__BCI_BUSY_MASK | GRBM_STATUS__SX_BUSY_MASK |
4934 		   GRBM_STATUS__TA_BUSY_MASK | GRBM_STATUS__VGT_BUSY_MASK |
4935 		   GRBM_STATUS__DB_BUSY_MASK | GRBM_STATUS__CB_BUSY_MASK |
4936 		   GRBM_STATUS__GDS_BUSY_MASK | GRBM_STATUS__SPI_BUSY_MASK |
4937 		   GRBM_STATUS__IA_BUSY_MASK | GRBM_STATUS__IA_BUSY_NO_DMA_MASK |
4938 		   GRBM_STATUS__CP_BUSY_MASK | GRBM_STATUS__CP_COHERENCY_BUSY_MASK)) {
4939 		grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset,
4940 						GRBM_SOFT_RESET, SOFT_RESET_CP, 1);
4941 		grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset,
4942 						GRBM_SOFT_RESET, SOFT_RESET_GFX, 1);
4943 		srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset,
4944 						SRBM_SOFT_RESET, SOFT_RESET_GRBM, 1);
4945 	}
4946 
4947 	/* GRBM_STATUS2 */
4948 	tmp = RREG32(mmGRBM_STATUS2);
4949 	if (REG_GET_FIELD(tmp, GRBM_STATUS2, RLC_BUSY))
4950 		grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset,
4951 						GRBM_SOFT_RESET, SOFT_RESET_RLC, 1);
4952 
4953 	if (REG_GET_FIELD(tmp, GRBM_STATUS2, CPF_BUSY) ||
4954 	    REG_GET_FIELD(tmp, GRBM_STATUS2, CPC_BUSY) ||
4955 	    REG_GET_FIELD(tmp, GRBM_STATUS2, CPG_BUSY)) {
4956 		grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET,
4957 						SOFT_RESET_CPF, 1);
4958 		grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET,
4959 						SOFT_RESET_CPC, 1);
4960 		grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET,
4961 						SOFT_RESET_CPG, 1);
4962 		srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset, SRBM_SOFT_RESET,
4963 						SOFT_RESET_GRBM, 1);
4964 	}
4965 
4966 	/* SRBM_STATUS */
4967 	tmp = RREG32(mmSRBM_STATUS);
4968 	if (REG_GET_FIELD(tmp, SRBM_STATUS, GRBM_RQ_PENDING))
4969 		srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset,
4970 						SRBM_SOFT_RESET, SOFT_RESET_GRBM, 1);
4971 	if (REG_GET_FIELD(tmp, SRBM_STATUS, SEM_BUSY))
4972 		srbm_soft_reset = REG_SET_FIELD(srbm_soft_reset,
4973 						SRBM_SOFT_RESET, SOFT_RESET_SEM, 1);
4974 
4975 	if (grbm_soft_reset || srbm_soft_reset) {
4976 		adev->gfx.grbm_soft_reset = grbm_soft_reset;
4977 		adev->gfx.srbm_soft_reset = srbm_soft_reset;
4978 		return true;
4979 	} else {
4980 		adev->gfx.grbm_soft_reset = 0;
4981 		adev->gfx.srbm_soft_reset = 0;
4982 		return false;
4983 	}
4984 }
4985 
4986 static int gfx_v8_0_pre_soft_reset(void *handle)
4987 {
4988 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4989 	u32 grbm_soft_reset = 0;
4990 
4991 	if ((!adev->gfx.grbm_soft_reset) &&
4992 	    (!adev->gfx.srbm_soft_reset))
4993 		return 0;
4994 
4995 	grbm_soft_reset = adev->gfx.grbm_soft_reset;
4996 
4997 	/* stop the rlc */
4998 	adev->gfx.rlc.funcs->stop(adev);
4999 
5000 	if (REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CP) ||
5001 	    REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_GFX))
5002 		/* Disable GFX parsing/prefetching */
5003 		gfx_v8_0_cp_gfx_enable(adev, false);
5004 
5005 	if (REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CP) ||
5006 	    REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CPF) ||
5007 	    REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CPC) ||
5008 	    REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CPG)) {
5009 		int i;
5010 
5011 		for (i = 0; i < adev->gfx.num_compute_rings; i++) {
5012 			struct amdgpu_ring *ring = &adev->gfx.compute_ring[i];
5013 
5014 			mutex_lock(&adev->srbm_mutex);
5015 			vi_srbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
5016 			gfx_v8_0_deactivate_hqd(adev, 2);
5017 			vi_srbm_select(adev, 0, 0, 0, 0);
5018 			mutex_unlock(&adev->srbm_mutex);
5019 		}
5020 		/* Disable MEC parsing/prefetching */
5021 		gfx_v8_0_cp_compute_enable(adev, false);
5022 	}
5023 
5024 	return 0;
5025 }
5026 
5027 static int gfx_v8_0_soft_reset(void *handle)
5028 {
5029 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5030 	u32 grbm_soft_reset = 0, srbm_soft_reset = 0;
5031 	u32 tmp;
5032 
5033 	if ((!adev->gfx.grbm_soft_reset) &&
5034 	    (!adev->gfx.srbm_soft_reset))
5035 		return 0;
5036 
5037 	grbm_soft_reset = adev->gfx.grbm_soft_reset;
5038 	srbm_soft_reset = adev->gfx.srbm_soft_reset;
5039 
5040 	if (grbm_soft_reset || srbm_soft_reset) {
5041 		tmp = RREG32(mmGMCON_DEBUG);
5042 		tmp = REG_SET_FIELD(tmp, GMCON_DEBUG, GFX_STALL, 1);
5043 		tmp = REG_SET_FIELD(tmp, GMCON_DEBUG, GFX_CLEAR, 1);
5044 		WREG32(mmGMCON_DEBUG, tmp);
5045 		udelay(50);
5046 	}
5047 
5048 	if (grbm_soft_reset) {
5049 		tmp = RREG32(mmGRBM_SOFT_RESET);
5050 		tmp |= grbm_soft_reset;
5051 		dev_info(adev->dev, "GRBM_SOFT_RESET=0x%08X\n", tmp);
5052 		WREG32(mmGRBM_SOFT_RESET, tmp);
5053 		tmp = RREG32(mmGRBM_SOFT_RESET);
5054 
5055 		udelay(50);
5056 
5057 		tmp &= ~grbm_soft_reset;
5058 		WREG32(mmGRBM_SOFT_RESET, tmp);
5059 		tmp = RREG32(mmGRBM_SOFT_RESET);
5060 	}
5061 
5062 	if (srbm_soft_reset) {
5063 		tmp = RREG32(mmSRBM_SOFT_RESET);
5064 		tmp |= srbm_soft_reset;
5065 		dev_info(adev->dev, "SRBM_SOFT_RESET=0x%08X\n", tmp);
5066 		WREG32(mmSRBM_SOFT_RESET, tmp);
5067 		tmp = RREG32(mmSRBM_SOFT_RESET);
5068 
5069 		udelay(50);
5070 
5071 		tmp &= ~srbm_soft_reset;
5072 		WREG32(mmSRBM_SOFT_RESET, tmp);
5073 		tmp = RREG32(mmSRBM_SOFT_RESET);
5074 	}
5075 
5076 	if (grbm_soft_reset || srbm_soft_reset) {
5077 		tmp = RREG32(mmGMCON_DEBUG);
5078 		tmp = REG_SET_FIELD(tmp, GMCON_DEBUG, GFX_STALL, 0);
5079 		tmp = REG_SET_FIELD(tmp, GMCON_DEBUG, GFX_CLEAR, 0);
5080 		WREG32(mmGMCON_DEBUG, tmp);
5081 	}
5082 
5083 	/* Wait a little for things to settle down */
5084 	udelay(50);
5085 
5086 	return 0;
5087 }
5088 
5089 static int gfx_v8_0_post_soft_reset(void *handle)
5090 {
5091 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5092 	u32 grbm_soft_reset = 0;
5093 
5094 	if ((!adev->gfx.grbm_soft_reset) &&
5095 	    (!adev->gfx.srbm_soft_reset))
5096 		return 0;
5097 
5098 	grbm_soft_reset = adev->gfx.grbm_soft_reset;
5099 
5100 	if (REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CP) ||
5101 	    REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CPF) ||
5102 	    REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CPC) ||
5103 	    REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CPG)) {
5104 		int i;
5105 
5106 		for (i = 0; i < adev->gfx.num_compute_rings; i++) {
5107 			struct amdgpu_ring *ring = &adev->gfx.compute_ring[i];
5108 
5109 			mutex_lock(&adev->srbm_mutex);
5110 			vi_srbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
5111 			gfx_v8_0_deactivate_hqd(adev, 2);
5112 			vi_srbm_select(adev, 0, 0, 0, 0);
5113 			mutex_unlock(&adev->srbm_mutex);
5114 		}
5115 		gfx_v8_0_kiq_resume(adev);
5116 		gfx_v8_0_kcq_resume(adev);
5117 	}
5118 
5119 	if (REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_CP) ||
5120 	    REG_GET_FIELD(grbm_soft_reset, GRBM_SOFT_RESET, SOFT_RESET_GFX))
5121 		gfx_v8_0_cp_gfx_resume(adev);
5122 
5123 	gfx_v8_0_cp_test_all_rings(adev);
5124 
5125 	adev->gfx.rlc.funcs->start(adev);
5126 
5127 	return 0;
5128 }
5129 
5130 /**
5131  * gfx_v8_0_get_gpu_clock_counter - return GPU clock counter snapshot
5132  *
5133  * @adev: amdgpu_device pointer
5134  *
5135  * Fetches a GPU clock counter snapshot.
5136  * Returns the 64 bit clock counter snapshot.
5137  */
5138 static uint64_t gfx_v8_0_get_gpu_clock_counter(struct amdgpu_device *adev)
5139 {
5140 	uint64_t clock;
5141 
5142 	mutex_lock(&adev->gfx.gpu_clock_mutex);
5143 	WREG32(mmRLC_CAPTURE_GPU_CLOCK_COUNT, 1);
5144 	clock = (uint64_t)RREG32(mmRLC_GPU_CLOCK_COUNT_LSB) |
5145 		((uint64_t)RREG32(mmRLC_GPU_CLOCK_COUNT_MSB) << 32ULL);
5146 	mutex_unlock(&adev->gfx.gpu_clock_mutex);
5147 	return clock;
5148 }
5149 
5150 static void gfx_v8_0_ring_emit_gds_switch(struct amdgpu_ring *ring,
5151 					  uint32_t vmid,
5152 					  uint32_t gds_base, uint32_t gds_size,
5153 					  uint32_t gws_base, uint32_t gws_size,
5154 					  uint32_t oa_base, uint32_t oa_size)
5155 {
5156 	/* GDS Base */
5157 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
5158 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
5159 				WRITE_DATA_DST_SEL(0)));
5160 	amdgpu_ring_write(ring, amdgpu_gds_reg_offset[vmid].mem_base);
5161 	amdgpu_ring_write(ring, 0);
5162 	amdgpu_ring_write(ring, gds_base);
5163 
5164 	/* GDS Size */
5165 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
5166 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
5167 				WRITE_DATA_DST_SEL(0)));
5168 	amdgpu_ring_write(ring, amdgpu_gds_reg_offset[vmid].mem_size);
5169 	amdgpu_ring_write(ring, 0);
5170 	amdgpu_ring_write(ring, gds_size);
5171 
5172 	/* GWS */
5173 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
5174 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
5175 				WRITE_DATA_DST_SEL(0)));
5176 	amdgpu_ring_write(ring, amdgpu_gds_reg_offset[vmid].gws);
5177 	amdgpu_ring_write(ring, 0);
5178 	amdgpu_ring_write(ring, gws_size << GDS_GWS_VMID0__SIZE__SHIFT | gws_base);
5179 
5180 	/* OA */
5181 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
5182 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
5183 				WRITE_DATA_DST_SEL(0)));
5184 	amdgpu_ring_write(ring, amdgpu_gds_reg_offset[vmid].oa);
5185 	amdgpu_ring_write(ring, 0);
5186 	amdgpu_ring_write(ring, (1 << (oa_size + oa_base)) - (1 << oa_base));
5187 }
5188 
5189 static uint32_t wave_read_ind(struct amdgpu_device *adev, uint32_t simd, uint32_t wave, uint32_t address)
5190 {
5191 	WREG32(mmSQ_IND_INDEX,
5192 		(wave << SQ_IND_INDEX__WAVE_ID__SHIFT) |
5193 		(simd << SQ_IND_INDEX__SIMD_ID__SHIFT) |
5194 		(address << SQ_IND_INDEX__INDEX__SHIFT) |
5195 		(SQ_IND_INDEX__FORCE_READ_MASK));
5196 	return RREG32(mmSQ_IND_DATA);
5197 }
5198 
5199 static void wave_read_regs(struct amdgpu_device *adev, uint32_t simd,
5200 			   uint32_t wave, uint32_t thread,
5201 			   uint32_t regno, uint32_t num, uint32_t *out)
5202 {
5203 	WREG32(mmSQ_IND_INDEX,
5204 		(wave << SQ_IND_INDEX__WAVE_ID__SHIFT) |
5205 		(simd << SQ_IND_INDEX__SIMD_ID__SHIFT) |
5206 		(regno << SQ_IND_INDEX__INDEX__SHIFT) |
5207 		(thread << SQ_IND_INDEX__THREAD_ID__SHIFT) |
5208 		(SQ_IND_INDEX__FORCE_READ_MASK) |
5209 		(SQ_IND_INDEX__AUTO_INCR_MASK));
5210 	while (num--)
5211 		*(out++) = RREG32(mmSQ_IND_DATA);
5212 }
5213 
5214 static void gfx_v8_0_read_wave_data(struct amdgpu_device *adev, uint32_t xcc_id, uint32_t simd, uint32_t wave, uint32_t *dst, int *no_fields)
5215 {
5216 	/* type 0 wave data */
5217 	dst[(*no_fields)++] = 0;
5218 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_STATUS);
5219 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_PC_LO);
5220 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_PC_HI);
5221 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_EXEC_LO);
5222 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_EXEC_HI);
5223 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_HW_ID);
5224 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_INST_DW0);
5225 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_INST_DW1);
5226 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_GPR_ALLOC);
5227 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_LDS_ALLOC);
5228 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TRAPSTS);
5229 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_IB_STS);
5230 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TBA_LO);
5231 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TBA_HI);
5232 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TMA_LO);
5233 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_TMA_HI);
5234 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_IB_DBG0);
5235 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_M0);
5236 	dst[(*no_fields)++] = wave_read_ind(adev, simd, wave, ixSQ_WAVE_MODE);
5237 }
5238 
5239 static void gfx_v8_0_read_wave_sgprs(struct amdgpu_device *adev, uint32_t xcc_id, uint32_t simd,
5240 				     uint32_t wave, uint32_t start,
5241 				     uint32_t size, uint32_t *dst)
5242 {
5243 	wave_read_regs(
5244 		adev, simd, wave, 0,
5245 		start + SQIND_WAVE_SGPRS_OFFSET, size, dst);
5246 }
5247 
5248 
5249 static const struct amdgpu_gfx_funcs gfx_v8_0_gfx_funcs = {
5250 	.get_gpu_clock_counter = &gfx_v8_0_get_gpu_clock_counter,
5251 	.select_se_sh = &gfx_v8_0_select_se_sh,
5252 	.read_wave_data = &gfx_v8_0_read_wave_data,
5253 	.read_wave_sgprs = &gfx_v8_0_read_wave_sgprs,
5254 	.select_me_pipe_q = &gfx_v8_0_select_me_pipe_q
5255 };
5256 
5257 static int gfx_v8_0_early_init(void *handle)
5258 {
5259 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5260 
5261 	adev->gfx.xcc_mask = 1;
5262 	adev->gfx.num_gfx_rings = GFX8_NUM_GFX_RINGS;
5263 	adev->gfx.num_compute_rings = min(amdgpu_gfx_get_num_kcq(adev),
5264 					  AMDGPU_MAX_COMPUTE_RINGS);
5265 	adev->gfx.funcs = &gfx_v8_0_gfx_funcs;
5266 	gfx_v8_0_set_ring_funcs(adev);
5267 	gfx_v8_0_set_irq_funcs(adev);
5268 	gfx_v8_0_set_gds_init(adev);
5269 	gfx_v8_0_set_rlc_funcs(adev);
5270 
5271 	return 0;
5272 }
5273 
5274 static int gfx_v8_0_late_init(void *handle)
5275 {
5276 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5277 	int r;
5278 
5279 	r = amdgpu_irq_get(adev, &adev->gfx.priv_reg_irq, 0);
5280 	if (r)
5281 		return r;
5282 
5283 	r = amdgpu_irq_get(adev, &adev->gfx.priv_inst_irq, 0);
5284 	if (r)
5285 		return r;
5286 
5287 	/* requires IBs so do in late init after IB pool is initialized */
5288 	r = gfx_v8_0_do_edc_gpr_workarounds(adev);
5289 	if (r)
5290 		return r;
5291 
5292 	r = amdgpu_irq_get(adev, &adev->gfx.cp_ecc_error_irq, 0);
5293 	if (r) {
5294 		DRM_ERROR("amdgpu_irq_get() failed to get IRQ for EDC, r: %d.\n", r);
5295 		return r;
5296 	}
5297 
5298 	r = amdgpu_irq_get(adev, &adev->gfx.sq_irq, 0);
5299 	if (r) {
5300 		DRM_ERROR(
5301 			"amdgpu_irq_get() failed to get IRQ for SQ, r: %d.\n",
5302 			r);
5303 		return r;
5304 	}
5305 
5306 	return 0;
5307 }
5308 
5309 static void gfx_v8_0_enable_gfx_static_mg_power_gating(struct amdgpu_device *adev,
5310 						       bool enable)
5311 {
5312 	if ((adev->asic_type == CHIP_POLARIS11) ||
5313 	    (adev->asic_type == CHIP_POLARIS12) ||
5314 	    (adev->asic_type == CHIP_VEGAM))
5315 		/* Send msg to SMU via Powerplay */
5316 		amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_GFX, enable);
5317 
5318 	WREG32_FIELD(RLC_PG_CNTL, STATIC_PER_CU_PG_ENABLE, enable ? 1 : 0);
5319 }
5320 
5321 static void gfx_v8_0_enable_gfx_dynamic_mg_power_gating(struct amdgpu_device *adev,
5322 							bool enable)
5323 {
5324 	WREG32_FIELD(RLC_PG_CNTL, DYN_PER_CU_PG_ENABLE, enable ? 1 : 0);
5325 }
5326 
5327 static void polaris11_enable_gfx_quick_mg_power_gating(struct amdgpu_device *adev,
5328 		bool enable)
5329 {
5330 	WREG32_FIELD(RLC_PG_CNTL, QUICK_PG_ENABLE, enable ? 1 : 0);
5331 }
5332 
5333 static void cz_enable_gfx_cg_power_gating(struct amdgpu_device *adev,
5334 					  bool enable)
5335 {
5336 	WREG32_FIELD(RLC_PG_CNTL, GFX_POWER_GATING_ENABLE, enable ? 1 : 0);
5337 }
5338 
5339 static void cz_enable_gfx_pipeline_power_gating(struct amdgpu_device *adev,
5340 						bool enable)
5341 {
5342 	WREG32_FIELD(RLC_PG_CNTL, GFX_PIPELINE_PG_ENABLE, enable ? 1 : 0);
5343 
5344 	/* Read any GFX register to wake up GFX. */
5345 	if (!enable)
5346 		RREG32(mmDB_RENDER_CONTROL);
5347 }
5348 
5349 static void cz_update_gfx_cg_power_gating(struct amdgpu_device *adev,
5350 					  bool enable)
5351 {
5352 	if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_PG) && enable) {
5353 		cz_enable_gfx_cg_power_gating(adev, true);
5354 		if (adev->pg_flags & AMD_PG_SUPPORT_GFX_PIPELINE)
5355 			cz_enable_gfx_pipeline_power_gating(adev, true);
5356 	} else {
5357 		cz_enable_gfx_cg_power_gating(adev, false);
5358 		cz_enable_gfx_pipeline_power_gating(adev, false);
5359 	}
5360 }
5361 
5362 static int gfx_v8_0_set_powergating_state(void *handle,
5363 					  enum amd_powergating_state state)
5364 {
5365 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5366 	bool enable = (state == AMD_PG_STATE_GATE);
5367 
5368 	if (amdgpu_sriov_vf(adev))
5369 		return 0;
5370 
5371 	if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_SMG |
5372 				AMD_PG_SUPPORT_RLC_SMU_HS |
5373 				AMD_PG_SUPPORT_CP |
5374 				AMD_PG_SUPPORT_GFX_DMG))
5375 		amdgpu_gfx_rlc_enter_safe_mode(adev, 0);
5376 	switch (adev->asic_type) {
5377 	case CHIP_CARRIZO:
5378 	case CHIP_STONEY:
5379 
5380 		if (adev->pg_flags & AMD_PG_SUPPORT_RLC_SMU_HS) {
5381 			cz_enable_sck_slow_down_on_power_up(adev, true);
5382 			cz_enable_sck_slow_down_on_power_down(adev, true);
5383 		} else {
5384 			cz_enable_sck_slow_down_on_power_up(adev, false);
5385 			cz_enable_sck_slow_down_on_power_down(adev, false);
5386 		}
5387 		if (adev->pg_flags & AMD_PG_SUPPORT_CP)
5388 			cz_enable_cp_power_gating(adev, true);
5389 		else
5390 			cz_enable_cp_power_gating(adev, false);
5391 
5392 		cz_update_gfx_cg_power_gating(adev, enable);
5393 
5394 		if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_SMG) && enable)
5395 			gfx_v8_0_enable_gfx_static_mg_power_gating(adev, true);
5396 		else
5397 			gfx_v8_0_enable_gfx_static_mg_power_gating(adev, false);
5398 
5399 		if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_DMG) && enable)
5400 			gfx_v8_0_enable_gfx_dynamic_mg_power_gating(adev, true);
5401 		else
5402 			gfx_v8_0_enable_gfx_dynamic_mg_power_gating(adev, false);
5403 		break;
5404 	case CHIP_POLARIS11:
5405 	case CHIP_POLARIS12:
5406 	case CHIP_VEGAM:
5407 		if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_SMG) && enable)
5408 			gfx_v8_0_enable_gfx_static_mg_power_gating(adev, true);
5409 		else
5410 			gfx_v8_0_enable_gfx_static_mg_power_gating(adev, false);
5411 
5412 		if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_DMG) && enable)
5413 			gfx_v8_0_enable_gfx_dynamic_mg_power_gating(adev, true);
5414 		else
5415 			gfx_v8_0_enable_gfx_dynamic_mg_power_gating(adev, false);
5416 
5417 		if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_QUICK_MG) && enable)
5418 			polaris11_enable_gfx_quick_mg_power_gating(adev, true);
5419 		else
5420 			polaris11_enable_gfx_quick_mg_power_gating(adev, false);
5421 		break;
5422 	default:
5423 		break;
5424 	}
5425 	if (adev->pg_flags & (AMD_PG_SUPPORT_GFX_SMG |
5426 				AMD_PG_SUPPORT_RLC_SMU_HS |
5427 				AMD_PG_SUPPORT_CP |
5428 				AMD_PG_SUPPORT_GFX_DMG))
5429 		amdgpu_gfx_rlc_exit_safe_mode(adev, 0);
5430 	return 0;
5431 }
5432 
5433 static void gfx_v8_0_get_clockgating_state(void *handle, u64 *flags)
5434 {
5435 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5436 	int data;
5437 
5438 	if (amdgpu_sriov_vf(adev))
5439 		*flags = 0;
5440 
5441 	/* AMD_CG_SUPPORT_GFX_MGCG */
5442 	data = RREG32(mmRLC_CGTT_MGCG_OVERRIDE);
5443 	if (!(data & RLC_CGTT_MGCG_OVERRIDE__CPF_MASK))
5444 		*flags |= AMD_CG_SUPPORT_GFX_MGCG;
5445 
5446 	/* AMD_CG_SUPPORT_GFX_CGLG */
5447 	data = RREG32(mmRLC_CGCG_CGLS_CTRL);
5448 	if (data & RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK)
5449 		*flags |= AMD_CG_SUPPORT_GFX_CGCG;
5450 
5451 	/* AMD_CG_SUPPORT_GFX_CGLS */
5452 	if (data & RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK)
5453 		*flags |= AMD_CG_SUPPORT_GFX_CGLS;
5454 
5455 	/* AMD_CG_SUPPORT_GFX_CGTS */
5456 	data = RREG32(mmCGTS_SM_CTRL_REG);
5457 	if (!(data & CGTS_SM_CTRL_REG__OVERRIDE_MASK))
5458 		*flags |= AMD_CG_SUPPORT_GFX_CGTS;
5459 
5460 	/* AMD_CG_SUPPORT_GFX_CGTS_LS */
5461 	if (!(data & CGTS_SM_CTRL_REG__LS_OVERRIDE_MASK))
5462 		*flags |= AMD_CG_SUPPORT_GFX_CGTS_LS;
5463 
5464 	/* AMD_CG_SUPPORT_GFX_RLC_LS */
5465 	data = RREG32(mmRLC_MEM_SLP_CNTL);
5466 	if (data & RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK)
5467 		*flags |= AMD_CG_SUPPORT_GFX_RLC_LS | AMD_CG_SUPPORT_GFX_MGLS;
5468 
5469 	/* AMD_CG_SUPPORT_GFX_CP_LS */
5470 	data = RREG32(mmCP_MEM_SLP_CNTL);
5471 	if (data & CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK)
5472 		*flags |= AMD_CG_SUPPORT_GFX_CP_LS | AMD_CG_SUPPORT_GFX_MGLS;
5473 }
5474 
5475 static void gfx_v8_0_send_serdes_cmd(struct amdgpu_device *adev,
5476 				     uint32_t reg_addr, uint32_t cmd)
5477 {
5478 	uint32_t data;
5479 
5480 	gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 0);
5481 
5482 	WREG32(mmRLC_SERDES_WR_CU_MASTER_MASK, 0xffffffff);
5483 	WREG32(mmRLC_SERDES_WR_NONCU_MASTER_MASK, 0xffffffff);
5484 
5485 	data = RREG32(mmRLC_SERDES_WR_CTRL);
5486 	if (adev->asic_type == CHIP_STONEY)
5487 		data &= ~(RLC_SERDES_WR_CTRL__WRITE_COMMAND_MASK |
5488 			  RLC_SERDES_WR_CTRL__READ_COMMAND_MASK |
5489 			  RLC_SERDES_WR_CTRL__P1_SELECT_MASK |
5490 			  RLC_SERDES_WR_CTRL__P2_SELECT_MASK |
5491 			  RLC_SERDES_WR_CTRL__RDDATA_RESET_MASK |
5492 			  RLC_SERDES_WR_CTRL__POWER_DOWN_MASK |
5493 			  RLC_SERDES_WR_CTRL__POWER_UP_MASK |
5494 			  RLC_SERDES_WR_CTRL__SHORT_FORMAT_MASK |
5495 			  RLC_SERDES_WR_CTRL__SRBM_OVERRIDE_MASK);
5496 	else
5497 		data &= ~(RLC_SERDES_WR_CTRL__WRITE_COMMAND_MASK |
5498 			  RLC_SERDES_WR_CTRL__READ_COMMAND_MASK |
5499 			  RLC_SERDES_WR_CTRL__P1_SELECT_MASK |
5500 			  RLC_SERDES_WR_CTRL__P2_SELECT_MASK |
5501 			  RLC_SERDES_WR_CTRL__RDDATA_RESET_MASK |
5502 			  RLC_SERDES_WR_CTRL__POWER_DOWN_MASK |
5503 			  RLC_SERDES_WR_CTRL__POWER_UP_MASK |
5504 			  RLC_SERDES_WR_CTRL__SHORT_FORMAT_MASK |
5505 			  RLC_SERDES_WR_CTRL__BPM_DATA_MASK |
5506 			  RLC_SERDES_WR_CTRL__REG_ADDR_MASK |
5507 			  RLC_SERDES_WR_CTRL__SRBM_OVERRIDE_MASK);
5508 	data |= (RLC_SERDES_WR_CTRL__RSVD_BPM_ADDR_MASK |
5509 		 (cmd << RLC_SERDES_WR_CTRL__BPM_DATA__SHIFT) |
5510 		 (reg_addr << RLC_SERDES_WR_CTRL__REG_ADDR__SHIFT) |
5511 		 (0xff << RLC_SERDES_WR_CTRL__BPM_ADDR__SHIFT));
5512 
5513 	WREG32(mmRLC_SERDES_WR_CTRL, data);
5514 }
5515 
5516 #define MSG_ENTER_RLC_SAFE_MODE     1
5517 #define MSG_EXIT_RLC_SAFE_MODE      0
5518 #define RLC_GPR_REG2__REQ_MASK 0x00000001
5519 #define RLC_GPR_REG2__REQ__SHIFT 0
5520 #define RLC_GPR_REG2__MESSAGE__SHIFT 0x00000001
5521 #define RLC_GPR_REG2__MESSAGE_MASK 0x0000001e
5522 
5523 static bool gfx_v8_0_is_rlc_enabled(struct amdgpu_device *adev)
5524 {
5525 	uint32_t rlc_setting;
5526 
5527 	rlc_setting = RREG32(mmRLC_CNTL);
5528 	if (!(rlc_setting & RLC_CNTL__RLC_ENABLE_F32_MASK))
5529 		return false;
5530 
5531 	return true;
5532 }
5533 
5534 static void gfx_v8_0_set_safe_mode(struct amdgpu_device *adev, int xcc_id)
5535 {
5536 	uint32_t data;
5537 	unsigned i;
5538 	data = RREG32(mmRLC_CNTL);
5539 	data |= RLC_SAFE_MODE__CMD_MASK;
5540 	data &= ~RLC_SAFE_MODE__MESSAGE_MASK;
5541 	data |= (1 << RLC_SAFE_MODE__MESSAGE__SHIFT);
5542 	WREG32(mmRLC_SAFE_MODE, data);
5543 
5544 	/* wait for RLC_SAFE_MODE */
5545 	for (i = 0; i < adev->usec_timeout; i++) {
5546 		if ((RREG32(mmRLC_GPM_STAT) &
5547 		     (RLC_GPM_STAT__GFX_CLOCK_STATUS_MASK |
5548 		      RLC_GPM_STAT__GFX_POWER_STATUS_MASK)) ==
5549 		    (RLC_GPM_STAT__GFX_CLOCK_STATUS_MASK |
5550 		     RLC_GPM_STAT__GFX_POWER_STATUS_MASK))
5551 			break;
5552 		udelay(1);
5553 	}
5554 	for (i = 0; i < adev->usec_timeout; i++) {
5555 		if (!REG_GET_FIELD(RREG32(mmRLC_SAFE_MODE), RLC_SAFE_MODE, CMD))
5556 			break;
5557 		udelay(1);
5558 	}
5559 }
5560 
5561 static void gfx_v8_0_unset_safe_mode(struct amdgpu_device *adev, int xcc_id)
5562 {
5563 	uint32_t data;
5564 	unsigned i;
5565 
5566 	data = RREG32(mmRLC_CNTL);
5567 	data |= RLC_SAFE_MODE__CMD_MASK;
5568 	data &= ~RLC_SAFE_MODE__MESSAGE_MASK;
5569 	WREG32(mmRLC_SAFE_MODE, data);
5570 
5571 	for (i = 0; i < adev->usec_timeout; i++) {
5572 		if (!REG_GET_FIELD(RREG32(mmRLC_SAFE_MODE), RLC_SAFE_MODE, CMD))
5573 			break;
5574 		udelay(1);
5575 	}
5576 }
5577 
5578 static void gfx_v8_0_update_spm_vmid(struct amdgpu_device *adev, struct amdgpu_ring *ring, unsigned vmid)
5579 {
5580 	u32 data;
5581 
5582 	amdgpu_gfx_off_ctrl(adev, false);
5583 
5584 	if (amdgpu_sriov_is_pp_one_vf(adev))
5585 		data = RREG32_NO_KIQ(mmRLC_SPM_VMID);
5586 	else
5587 		data = RREG32(mmRLC_SPM_VMID);
5588 
5589 	data &= ~RLC_SPM_VMID__RLC_SPM_VMID_MASK;
5590 	data |= (vmid & RLC_SPM_VMID__RLC_SPM_VMID_MASK) << RLC_SPM_VMID__RLC_SPM_VMID__SHIFT;
5591 
5592 	if (amdgpu_sriov_is_pp_one_vf(adev))
5593 		WREG32_NO_KIQ(mmRLC_SPM_VMID, data);
5594 	else
5595 		WREG32(mmRLC_SPM_VMID, data);
5596 
5597 	amdgpu_gfx_off_ctrl(adev, true);
5598 }
5599 
5600 static const struct amdgpu_rlc_funcs iceland_rlc_funcs = {
5601 	.is_rlc_enabled = gfx_v8_0_is_rlc_enabled,
5602 	.set_safe_mode = gfx_v8_0_set_safe_mode,
5603 	.unset_safe_mode = gfx_v8_0_unset_safe_mode,
5604 	.init = gfx_v8_0_rlc_init,
5605 	.get_csb_size = gfx_v8_0_get_csb_size,
5606 	.get_csb_buffer = gfx_v8_0_get_csb_buffer,
5607 	.get_cp_table_num = gfx_v8_0_cp_jump_table_num,
5608 	.resume = gfx_v8_0_rlc_resume,
5609 	.stop = gfx_v8_0_rlc_stop,
5610 	.reset = gfx_v8_0_rlc_reset,
5611 	.start = gfx_v8_0_rlc_start,
5612 	.update_spm_vmid = gfx_v8_0_update_spm_vmid
5613 };
5614 
5615 static void gfx_v8_0_update_medium_grain_clock_gating(struct amdgpu_device *adev,
5616 						      bool enable)
5617 {
5618 	uint32_t temp, data;
5619 
5620 	amdgpu_gfx_rlc_enter_safe_mode(adev, 0);
5621 
5622 	/* It is disabled by HW by default */
5623 	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGCG)) {
5624 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS) {
5625 			if (adev->cg_flags & AMD_CG_SUPPORT_GFX_RLC_LS)
5626 				/* 1 - RLC memory Light sleep */
5627 				WREG32_FIELD(RLC_MEM_SLP_CNTL, RLC_MEM_LS_EN, 1);
5628 
5629 			if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CP_LS)
5630 				WREG32_FIELD(CP_MEM_SLP_CNTL, CP_MEM_LS_EN, 1);
5631 		}
5632 
5633 		/* 3 - RLC_CGTT_MGCG_OVERRIDE */
5634 		temp = data = RREG32(mmRLC_CGTT_MGCG_OVERRIDE);
5635 		if (adev->flags & AMD_IS_APU)
5636 			data &= ~(RLC_CGTT_MGCG_OVERRIDE__CPF_MASK |
5637 				  RLC_CGTT_MGCG_OVERRIDE__RLC_MASK |
5638 				  RLC_CGTT_MGCG_OVERRIDE__MGCG_MASK);
5639 		else
5640 			data &= ~(RLC_CGTT_MGCG_OVERRIDE__CPF_MASK |
5641 				  RLC_CGTT_MGCG_OVERRIDE__RLC_MASK |
5642 				  RLC_CGTT_MGCG_OVERRIDE__MGCG_MASK |
5643 				  RLC_CGTT_MGCG_OVERRIDE__GRBM_MASK);
5644 
5645 		if (temp != data)
5646 			WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data);
5647 
5648 		/* 4 - wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */
5649 		gfx_v8_0_wait_for_rlc_serdes(adev);
5650 
5651 		/* 5 - clear mgcg override */
5652 		gfx_v8_0_send_serdes_cmd(adev, BPM_REG_MGCG_OVERRIDE, CLE_BPM_SERDES_CMD);
5653 
5654 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGTS) {
5655 			/* 6 - Enable CGTS(Tree Shade) MGCG /MGLS */
5656 			temp = data = RREG32(mmCGTS_SM_CTRL_REG);
5657 			data &= ~(CGTS_SM_CTRL_REG__SM_MODE_MASK);
5658 			data |= (0x2 << CGTS_SM_CTRL_REG__SM_MODE__SHIFT);
5659 			data |= CGTS_SM_CTRL_REG__SM_MODE_ENABLE_MASK;
5660 			data &= ~CGTS_SM_CTRL_REG__OVERRIDE_MASK;
5661 			if ((adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS) &&
5662 			    (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGTS_LS))
5663 				data &= ~CGTS_SM_CTRL_REG__LS_OVERRIDE_MASK;
5664 			data |= CGTS_SM_CTRL_REG__ON_MONITOR_ADD_EN_MASK;
5665 			data |= (0x96 << CGTS_SM_CTRL_REG__ON_MONITOR_ADD__SHIFT);
5666 			if (temp != data)
5667 				WREG32(mmCGTS_SM_CTRL_REG, data);
5668 		}
5669 		udelay(50);
5670 
5671 		/* 7 - wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */
5672 		gfx_v8_0_wait_for_rlc_serdes(adev);
5673 	} else {
5674 		/* 1 - MGCG_OVERRIDE[0] for CP and MGCG_OVERRIDE[1] for RLC */
5675 		temp = data = RREG32(mmRLC_CGTT_MGCG_OVERRIDE);
5676 		data |= (RLC_CGTT_MGCG_OVERRIDE__CPF_MASK |
5677 				RLC_CGTT_MGCG_OVERRIDE__RLC_MASK |
5678 				RLC_CGTT_MGCG_OVERRIDE__MGCG_MASK |
5679 				RLC_CGTT_MGCG_OVERRIDE__GRBM_MASK);
5680 		if (temp != data)
5681 			WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data);
5682 
5683 		/* 2 - disable MGLS in RLC */
5684 		data = RREG32(mmRLC_MEM_SLP_CNTL);
5685 		if (data & RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK) {
5686 			data &= ~RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK;
5687 			WREG32(mmRLC_MEM_SLP_CNTL, data);
5688 		}
5689 
5690 		/* 3 - disable MGLS in CP */
5691 		data = RREG32(mmCP_MEM_SLP_CNTL);
5692 		if (data & CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK) {
5693 			data &= ~CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK;
5694 			WREG32(mmCP_MEM_SLP_CNTL, data);
5695 		}
5696 
5697 		/* 4 - Disable CGTS(Tree Shade) MGCG and MGLS */
5698 		temp = data = RREG32(mmCGTS_SM_CTRL_REG);
5699 		data |= (CGTS_SM_CTRL_REG__OVERRIDE_MASK |
5700 				CGTS_SM_CTRL_REG__LS_OVERRIDE_MASK);
5701 		if (temp != data)
5702 			WREG32(mmCGTS_SM_CTRL_REG, data);
5703 
5704 		/* 5 - wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */
5705 		gfx_v8_0_wait_for_rlc_serdes(adev);
5706 
5707 		/* 6 - set mgcg override */
5708 		gfx_v8_0_send_serdes_cmd(adev, BPM_REG_MGCG_OVERRIDE, SET_BPM_SERDES_CMD);
5709 
5710 		udelay(50);
5711 
5712 		/* 7- wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */
5713 		gfx_v8_0_wait_for_rlc_serdes(adev);
5714 	}
5715 
5716 	amdgpu_gfx_rlc_exit_safe_mode(adev, 0);
5717 }
5718 
5719 static void gfx_v8_0_update_coarse_grain_clock_gating(struct amdgpu_device *adev,
5720 						      bool enable)
5721 {
5722 	uint32_t temp, temp1, data, data1;
5723 
5724 	temp = data = RREG32(mmRLC_CGCG_CGLS_CTRL);
5725 
5726 	amdgpu_gfx_rlc_enter_safe_mode(adev, 0);
5727 
5728 	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGCG)) {
5729 		temp1 = data1 =	RREG32(mmRLC_CGTT_MGCG_OVERRIDE);
5730 		data1 &= ~RLC_CGTT_MGCG_OVERRIDE__CGCG_MASK;
5731 		if (temp1 != data1)
5732 			WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data1);
5733 
5734 		/* : wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */
5735 		gfx_v8_0_wait_for_rlc_serdes(adev);
5736 
5737 		/* 2 - clear cgcg override */
5738 		gfx_v8_0_send_serdes_cmd(adev, BPM_REG_CGCG_OVERRIDE, CLE_BPM_SERDES_CMD);
5739 
5740 		/* wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */
5741 		gfx_v8_0_wait_for_rlc_serdes(adev);
5742 
5743 		/* 3 - write cmd to set CGLS */
5744 		gfx_v8_0_send_serdes_cmd(adev, BPM_REG_CGLS_EN, SET_BPM_SERDES_CMD);
5745 
5746 		/* 4 - enable cgcg */
5747 		data |= RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK;
5748 
5749 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGLS) {
5750 			/* enable cgls*/
5751 			data |= RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK;
5752 
5753 			temp1 = data1 =	RREG32(mmRLC_CGTT_MGCG_OVERRIDE);
5754 			data1 &= ~RLC_CGTT_MGCG_OVERRIDE__CGLS_MASK;
5755 
5756 			if (temp1 != data1)
5757 				WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data1);
5758 		} else {
5759 			data &= ~RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK;
5760 		}
5761 
5762 		if (temp != data)
5763 			WREG32(mmRLC_CGCG_CGLS_CTRL, data);
5764 
5765 		/* 5 enable cntx_empty_int_enable/cntx_busy_int_enable/
5766 		 * Cmp_busy/GFX_Idle interrupts
5767 		 */
5768 		gfx_v8_0_enable_gui_idle_interrupt(adev, true);
5769 	} else {
5770 		/* disable cntx_empty_int_enable & GFX Idle interrupt */
5771 		gfx_v8_0_enable_gui_idle_interrupt(adev, false);
5772 
5773 		/* TEST CGCG */
5774 		temp1 = data1 =	RREG32(mmRLC_CGTT_MGCG_OVERRIDE);
5775 		data1 |= (RLC_CGTT_MGCG_OVERRIDE__CGCG_MASK |
5776 				RLC_CGTT_MGCG_OVERRIDE__CGLS_MASK);
5777 		if (temp1 != data1)
5778 			WREG32(mmRLC_CGTT_MGCG_OVERRIDE, data1);
5779 
5780 		/* read gfx register to wake up cgcg */
5781 		RREG32(mmCB_CGTT_SCLK_CTRL);
5782 		RREG32(mmCB_CGTT_SCLK_CTRL);
5783 		RREG32(mmCB_CGTT_SCLK_CTRL);
5784 		RREG32(mmCB_CGTT_SCLK_CTRL);
5785 
5786 		/* wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */
5787 		gfx_v8_0_wait_for_rlc_serdes(adev);
5788 
5789 		/* write cmd to Set CGCG Override */
5790 		gfx_v8_0_send_serdes_cmd(adev, BPM_REG_CGCG_OVERRIDE, SET_BPM_SERDES_CMD);
5791 
5792 		/* wait for RLC_SERDES_CU_MASTER & RLC_SERDES_NONCU_MASTER idle */
5793 		gfx_v8_0_wait_for_rlc_serdes(adev);
5794 
5795 		/* write cmd to Clear CGLS */
5796 		gfx_v8_0_send_serdes_cmd(adev, BPM_REG_CGLS_EN, CLE_BPM_SERDES_CMD);
5797 
5798 		/* disable cgcg, cgls should be disabled too. */
5799 		data &= ~(RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK |
5800 			  RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK);
5801 		if (temp != data)
5802 			WREG32(mmRLC_CGCG_CGLS_CTRL, data);
5803 		/* enable interrupts again for PG */
5804 		gfx_v8_0_enable_gui_idle_interrupt(adev, true);
5805 	}
5806 
5807 	gfx_v8_0_wait_for_rlc_serdes(adev);
5808 
5809 	amdgpu_gfx_rlc_exit_safe_mode(adev, 0);
5810 }
5811 static int gfx_v8_0_update_gfx_clock_gating(struct amdgpu_device *adev,
5812 					    bool enable)
5813 {
5814 	if (enable) {
5815 		/* CGCG/CGLS should be enabled after MGCG/MGLS/TS(CG/LS)
5816 		 * ===  MGCG + MGLS + TS(CG/LS) ===
5817 		 */
5818 		gfx_v8_0_update_medium_grain_clock_gating(adev, enable);
5819 		gfx_v8_0_update_coarse_grain_clock_gating(adev, enable);
5820 	} else {
5821 		/* CGCG/CGLS should be disabled before MGCG/MGLS/TS(CG/LS)
5822 		 * ===  CGCG + CGLS ===
5823 		 */
5824 		gfx_v8_0_update_coarse_grain_clock_gating(adev, enable);
5825 		gfx_v8_0_update_medium_grain_clock_gating(adev, enable);
5826 	}
5827 	return 0;
5828 }
5829 
5830 static int gfx_v8_0_tonga_update_gfx_clock_gating(struct amdgpu_device *adev,
5831 					  enum amd_clockgating_state state)
5832 {
5833 	uint32_t msg_id, pp_state = 0;
5834 	uint32_t pp_support_state = 0;
5835 
5836 	if (adev->cg_flags & (AMD_CG_SUPPORT_GFX_CGCG | AMD_CG_SUPPORT_GFX_CGLS)) {
5837 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGLS) {
5838 			pp_support_state = PP_STATE_SUPPORT_LS;
5839 			pp_state = PP_STATE_LS;
5840 		}
5841 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGCG) {
5842 			pp_support_state |= PP_STATE_SUPPORT_CG;
5843 			pp_state |= PP_STATE_CG;
5844 		}
5845 		if (state == AMD_CG_STATE_UNGATE)
5846 			pp_state = 0;
5847 
5848 		msg_id = PP_CG_MSG_ID(PP_GROUP_GFX,
5849 				PP_BLOCK_GFX_CG,
5850 				pp_support_state,
5851 				pp_state);
5852 		amdgpu_dpm_set_clockgating_by_smu(adev, msg_id);
5853 	}
5854 
5855 	if (adev->cg_flags & (AMD_CG_SUPPORT_GFX_MGCG | AMD_CG_SUPPORT_GFX_MGLS)) {
5856 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS) {
5857 			pp_support_state = PP_STATE_SUPPORT_LS;
5858 			pp_state = PP_STATE_LS;
5859 		}
5860 
5861 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGCG) {
5862 			pp_support_state |= PP_STATE_SUPPORT_CG;
5863 			pp_state |= PP_STATE_CG;
5864 		}
5865 
5866 		if (state == AMD_CG_STATE_UNGATE)
5867 			pp_state = 0;
5868 
5869 		msg_id = PP_CG_MSG_ID(PP_GROUP_GFX,
5870 				PP_BLOCK_GFX_MG,
5871 				pp_support_state,
5872 				pp_state);
5873 		amdgpu_dpm_set_clockgating_by_smu(adev, msg_id);
5874 	}
5875 
5876 	return 0;
5877 }
5878 
5879 static int gfx_v8_0_polaris_update_gfx_clock_gating(struct amdgpu_device *adev,
5880 					  enum amd_clockgating_state state)
5881 {
5882 
5883 	uint32_t msg_id, pp_state = 0;
5884 	uint32_t pp_support_state = 0;
5885 
5886 	if (adev->cg_flags & (AMD_CG_SUPPORT_GFX_CGCG | AMD_CG_SUPPORT_GFX_CGLS)) {
5887 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGLS) {
5888 			pp_support_state = PP_STATE_SUPPORT_LS;
5889 			pp_state = PP_STATE_LS;
5890 		}
5891 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGCG) {
5892 			pp_support_state |= PP_STATE_SUPPORT_CG;
5893 			pp_state |= PP_STATE_CG;
5894 		}
5895 		if (state == AMD_CG_STATE_UNGATE)
5896 			pp_state = 0;
5897 
5898 		msg_id = PP_CG_MSG_ID(PP_GROUP_GFX,
5899 				PP_BLOCK_GFX_CG,
5900 				pp_support_state,
5901 				pp_state);
5902 		amdgpu_dpm_set_clockgating_by_smu(adev, msg_id);
5903 	}
5904 
5905 	if (adev->cg_flags & (AMD_CG_SUPPORT_GFX_3D_CGCG | AMD_CG_SUPPORT_GFX_3D_CGLS)) {
5906 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_3D_CGLS) {
5907 			pp_support_state = PP_STATE_SUPPORT_LS;
5908 			pp_state = PP_STATE_LS;
5909 		}
5910 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_3D_CGCG) {
5911 			pp_support_state |= PP_STATE_SUPPORT_CG;
5912 			pp_state |= PP_STATE_CG;
5913 		}
5914 		if (state == AMD_CG_STATE_UNGATE)
5915 			pp_state = 0;
5916 
5917 		msg_id = PP_CG_MSG_ID(PP_GROUP_GFX,
5918 				PP_BLOCK_GFX_3D,
5919 				pp_support_state,
5920 				pp_state);
5921 		amdgpu_dpm_set_clockgating_by_smu(adev, msg_id);
5922 	}
5923 
5924 	if (adev->cg_flags & (AMD_CG_SUPPORT_GFX_MGCG | AMD_CG_SUPPORT_GFX_MGLS)) {
5925 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS) {
5926 			pp_support_state = PP_STATE_SUPPORT_LS;
5927 			pp_state = PP_STATE_LS;
5928 		}
5929 
5930 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGCG) {
5931 			pp_support_state |= PP_STATE_SUPPORT_CG;
5932 			pp_state |= PP_STATE_CG;
5933 		}
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_MG,
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_RLC_LS) {
5946 		pp_support_state = PP_STATE_SUPPORT_LS;
5947 
5948 		if (state == AMD_CG_STATE_UNGATE)
5949 			pp_state = 0;
5950 		else
5951 			pp_state = PP_STATE_LS;
5952 
5953 		msg_id = PP_CG_MSG_ID(PP_GROUP_GFX,
5954 				PP_BLOCK_GFX_RLC,
5955 				pp_support_state,
5956 				pp_state);
5957 		amdgpu_dpm_set_clockgating_by_smu(adev, msg_id);
5958 	}
5959 
5960 	if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CP_LS) {
5961 		pp_support_state = PP_STATE_SUPPORT_LS;
5962 
5963 		if (state == AMD_CG_STATE_UNGATE)
5964 			pp_state = 0;
5965 		else
5966 			pp_state = PP_STATE_LS;
5967 		msg_id = PP_CG_MSG_ID(PP_GROUP_GFX,
5968 			PP_BLOCK_GFX_CP,
5969 			pp_support_state,
5970 			pp_state);
5971 		amdgpu_dpm_set_clockgating_by_smu(adev, msg_id);
5972 	}
5973 
5974 	return 0;
5975 }
5976 
5977 static int gfx_v8_0_set_clockgating_state(void *handle,
5978 					  enum amd_clockgating_state state)
5979 {
5980 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
5981 
5982 	if (amdgpu_sriov_vf(adev))
5983 		return 0;
5984 
5985 	switch (adev->asic_type) {
5986 	case CHIP_FIJI:
5987 	case CHIP_CARRIZO:
5988 	case CHIP_STONEY:
5989 		gfx_v8_0_update_gfx_clock_gating(adev,
5990 						 state == AMD_CG_STATE_GATE);
5991 		break;
5992 	case CHIP_TONGA:
5993 		gfx_v8_0_tonga_update_gfx_clock_gating(adev, state);
5994 		break;
5995 	case CHIP_POLARIS10:
5996 	case CHIP_POLARIS11:
5997 	case CHIP_POLARIS12:
5998 	case CHIP_VEGAM:
5999 		gfx_v8_0_polaris_update_gfx_clock_gating(adev, state);
6000 		break;
6001 	default:
6002 		break;
6003 	}
6004 	return 0;
6005 }
6006 
6007 static u64 gfx_v8_0_ring_get_rptr(struct amdgpu_ring *ring)
6008 {
6009 	return *ring->rptr_cpu_addr;
6010 }
6011 
6012 static u64 gfx_v8_0_ring_get_wptr_gfx(struct amdgpu_ring *ring)
6013 {
6014 	struct amdgpu_device *adev = ring->adev;
6015 
6016 	if (ring->use_doorbell)
6017 		/* XXX check if swapping is necessary on BE */
6018 		return *ring->wptr_cpu_addr;
6019 	else
6020 		return RREG32(mmCP_RB0_WPTR);
6021 }
6022 
6023 static void gfx_v8_0_ring_set_wptr_gfx(struct amdgpu_ring *ring)
6024 {
6025 	struct amdgpu_device *adev = ring->adev;
6026 
6027 	if (ring->use_doorbell) {
6028 		/* XXX check if swapping is necessary on BE */
6029 		*ring->wptr_cpu_addr = lower_32_bits(ring->wptr);
6030 		WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
6031 	} else {
6032 		WREG32(mmCP_RB0_WPTR, lower_32_bits(ring->wptr));
6033 		(void)RREG32(mmCP_RB0_WPTR);
6034 	}
6035 }
6036 
6037 static void gfx_v8_0_ring_emit_hdp_flush(struct amdgpu_ring *ring)
6038 {
6039 	u32 ref_and_mask, reg_mem_engine;
6040 
6041 	if ((ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE) ||
6042 	    (ring->funcs->type == AMDGPU_RING_TYPE_KIQ)) {
6043 		switch (ring->me) {
6044 		case 1:
6045 			ref_and_mask = GPU_HDP_FLUSH_DONE__CP2_MASK << ring->pipe;
6046 			break;
6047 		case 2:
6048 			ref_and_mask = GPU_HDP_FLUSH_DONE__CP6_MASK << ring->pipe;
6049 			break;
6050 		default:
6051 			return;
6052 		}
6053 		reg_mem_engine = 0;
6054 	} else {
6055 		ref_and_mask = GPU_HDP_FLUSH_DONE__CP0_MASK;
6056 		reg_mem_engine = WAIT_REG_MEM_ENGINE(1); /* pfp */
6057 	}
6058 
6059 	amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
6060 	amdgpu_ring_write(ring, (WAIT_REG_MEM_OPERATION(1) | /* write, wait, write */
6061 				 WAIT_REG_MEM_FUNCTION(3) |  /* == */
6062 				 reg_mem_engine));
6063 	amdgpu_ring_write(ring, mmGPU_HDP_FLUSH_REQ);
6064 	amdgpu_ring_write(ring, mmGPU_HDP_FLUSH_DONE);
6065 	amdgpu_ring_write(ring, ref_and_mask);
6066 	amdgpu_ring_write(ring, ref_and_mask);
6067 	amdgpu_ring_write(ring, 0x20); /* poll interval */
6068 }
6069 
6070 static void gfx_v8_0_ring_emit_vgt_flush(struct amdgpu_ring *ring)
6071 {
6072 	amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE, 0));
6073 	amdgpu_ring_write(ring, EVENT_TYPE(VS_PARTIAL_FLUSH) |
6074 		EVENT_INDEX(4));
6075 
6076 	amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE, 0));
6077 	amdgpu_ring_write(ring, EVENT_TYPE(VGT_FLUSH) |
6078 		EVENT_INDEX(0));
6079 }
6080 
6081 static void gfx_v8_0_ring_emit_ib_gfx(struct amdgpu_ring *ring,
6082 					struct amdgpu_job *job,
6083 					struct amdgpu_ib *ib,
6084 					uint32_t flags)
6085 {
6086 	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
6087 	u32 header, control = 0;
6088 
6089 	if (ib->flags & AMDGPU_IB_FLAG_CE)
6090 		header = PACKET3(PACKET3_INDIRECT_BUFFER_CONST, 2);
6091 	else
6092 		header = PACKET3(PACKET3_INDIRECT_BUFFER, 2);
6093 
6094 	control |= ib->length_dw | (vmid << 24);
6095 
6096 	if (amdgpu_sriov_vf(ring->adev) && (ib->flags & AMDGPU_IB_FLAG_PREEMPT)) {
6097 		control |= INDIRECT_BUFFER_PRE_ENB(1);
6098 
6099 		if (!(ib->flags & AMDGPU_IB_FLAG_CE) && vmid)
6100 			gfx_v8_0_ring_emit_de_meta(ring);
6101 	}
6102 
6103 	amdgpu_ring_write(ring, header);
6104 	amdgpu_ring_write(ring,
6105 #ifdef __BIG_ENDIAN
6106 			  (2 << 0) |
6107 #endif
6108 			  (ib->gpu_addr & 0xFFFFFFFC));
6109 	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr) & 0xFFFF);
6110 	amdgpu_ring_write(ring, control);
6111 }
6112 
6113 static void gfx_v8_0_ring_emit_ib_compute(struct amdgpu_ring *ring,
6114 					  struct amdgpu_job *job,
6115 					  struct amdgpu_ib *ib,
6116 					  uint32_t flags)
6117 {
6118 	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
6119 	u32 control = INDIRECT_BUFFER_VALID | ib->length_dw | (vmid << 24);
6120 
6121 	/* Currently, there is a high possibility to get wave ID mismatch
6122 	 * between ME and GDS, leading to a hw deadlock, because ME generates
6123 	 * different wave IDs than the GDS expects. This situation happens
6124 	 * randomly when at least 5 compute pipes use GDS ordered append.
6125 	 * The wave IDs generated by ME are also wrong after suspend/resume.
6126 	 * Those are probably bugs somewhere else in the kernel driver.
6127 	 *
6128 	 * Writing GDS_COMPUTE_MAX_WAVE_ID resets wave ID counters in ME and
6129 	 * GDS to 0 for this ring (me/pipe).
6130 	 */
6131 	if (ib->flags & AMDGPU_IB_FLAG_RESET_GDS_MAX_WAVE_ID) {
6132 		amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1));
6133 		amdgpu_ring_write(ring, mmGDS_COMPUTE_MAX_WAVE_ID - PACKET3_SET_CONFIG_REG_START);
6134 		amdgpu_ring_write(ring, ring->adev->gds.gds_compute_max_wave_id);
6135 	}
6136 
6137 	amdgpu_ring_write(ring, PACKET3(PACKET3_INDIRECT_BUFFER, 2));
6138 	amdgpu_ring_write(ring,
6139 #ifdef __BIG_ENDIAN
6140 				(2 << 0) |
6141 #endif
6142 				(ib->gpu_addr & 0xFFFFFFFC));
6143 	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr) & 0xFFFF);
6144 	amdgpu_ring_write(ring, control);
6145 }
6146 
6147 static void gfx_v8_0_ring_emit_fence_gfx(struct amdgpu_ring *ring, u64 addr,
6148 					 u64 seq, unsigned flags)
6149 {
6150 	bool write64bit = flags & AMDGPU_FENCE_FLAG_64BIT;
6151 	bool int_sel = flags & AMDGPU_FENCE_FLAG_INT;
6152 	bool exec = flags & AMDGPU_FENCE_FLAG_EXEC;
6153 
6154 	/* Workaround for cache flush problems. First send a dummy EOP
6155 	 * event down the pipe with seq one below.
6156 	 */
6157 	amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE_EOP, 4));
6158 	amdgpu_ring_write(ring, (EOP_TCL1_ACTION_EN |
6159 				 EOP_TC_ACTION_EN |
6160 				 EOP_TC_WB_ACTION_EN |
6161 				 EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
6162 				 EVENT_INDEX(5)));
6163 	amdgpu_ring_write(ring, addr & 0xfffffffc);
6164 	amdgpu_ring_write(ring, (upper_32_bits(addr) & 0xffff) |
6165 				DATA_SEL(1) | INT_SEL(0));
6166 	amdgpu_ring_write(ring, lower_32_bits(seq - 1));
6167 	amdgpu_ring_write(ring, upper_32_bits(seq - 1));
6168 
6169 	/* Then send the real EOP event down the pipe:
6170 	 * EVENT_WRITE_EOP - flush caches, send int */
6171 	amdgpu_ring_write(ring, PACKET3(PACKET3_EVENT_WRITE_EOP, 4));
6172 	amdgpu_ring_write(ring, (EOP_TCL1_ACTION_EN |
6173 				 EOP_TC_ACTION_EN |
6174 				 EOP_TC_WB_ACTION_EN |
6175 				 EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
6176 				 EVENT_INDEX(5) |
6177 				 (exec ? EOP_EXEC : 0)));
6178 	amdgpu_ring_write(ring, addr & 0xfffffffc);
6179 	amdgpu_ring_write(ring, (upper_32_bits(addr) & 0xffff) |
6180 			  DATA_SEL(write64bit ? 2 : 1) | INT_SEL(int_sel ? 2 : 0));
6181 	amdgpu_ring_write(ring, lower_32_bits(seq));
6182 	amdgpu_ring_write(ring, upper_32_bits(seq));
6183 
6184 }
6185 
6186 static void gfx_v8_0_ring_emit_pipeline_sync(struct amdgpu_ring *ring)
6187 {
6188 	int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX);
6189 	uint32_t seq = ring->fence_drv.sync_seq;
6190 	uint64_t addr = ring->fence_drv.gpu_addr;
6191 
6192 	amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
6193 	amdgpu_ring_write(ring, (WAIT_REG_MEM_MEM_SPACE(1) | /* memory */
6194 				 WAIT_REG_MEM_FUNCTION(3) | /* equal */
6195 				 WAIT_REG_MEM_ENGINE(usepfp))); /* pfp or me */
6196 	amdgpu_ring_write(ring, addr & 0xfffffffc);
6197 	amdgpu_ring_write(ring, upper_32_bits(addr) & 0xffffffff);
6198 	amdgpu_ring_write(ring, seq);
6199 	amdgpu_ring_write(ring, 0xffffffff);
6200 	amdgpu_ring_write(ring, 4); /* poll interval */
6201 }
6202 
6203 static void gfx_v8_0_ring_emit_vm_flush(struct amdgpu_ring *ring,
6204 					unsigned vmid, uint64_t pd_addr)
6205 {
6206 	int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX);
6207 
6208 	amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr);
6209 
6210 	/* wait for the invalidate to complete */
6211 	amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
6212 	amdgpu_ring_write(ring, (WAIT_REG_MEM_OPERATION(0) | /* wait */
6213 				 WAIT_REG_MEM_FUNCTION(0) |  /* always */
6214 				 WAIT_REG_MEM_ENGINE(0))); /* me */
6215 	amdgpu_ring_write(ring, mmVM_INVALIDATE_REQUEST);
6216 	amdgpu_ring_write(ring, 0);
6217 	amdgpu_ring_write(ring, 0); /* ref */
6218 	amdgpu_ring_write(ring, 0); /* mask */
6219 	amdgpu_ring_write(ring, 0x20); /* poll interval */
6220 
6221 	/* compute doesn't have PFP */
6222 	if (usepfp) {
6223 		/* sync PFP to ME, otherwise we might get invalid PFP reads */
6224 		amdgpu_ring_write(ring, PACKET3(PACKET3_PFP_SYNC_ME, 0));
6225 		amdgpu_ring_write(ring, 0x0);
6226 	}
6227 }
6228 
6229 static u64 gfx_v8_0_ring_get_wptr_compute(struct amdgpu_ring *ring)
6230 {
6231 	return *ring->wptr_cpu_addr;
6232 }
6233 
6234 static void gfx_v8_0_ring_set_wptr_compute(struct amdgpu_ring *ring)
6235 {
6236 	struct amdgpu_device *adev = ring->adev;
6237 
6238 	/* XXX check if swapping is necessary on BE */
6239 	*ring->wptr_cpu_addr = lower_32_bits(ring->wptr);
6240 	WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr));
6241 }
6242 
6243 static void gfx_v8_0_ring_emit_fence_compute(struct amdgpu_ring *ring,
6244 					     u64 addr, u64 seq,
6245 					     unsigned flags)
6246 {
6247 	bool write64bit = flags & AMDGPU_FENCE_FLAG_64BIT;
6248 	bool int_sel = flags & AMDGPU_FENCE_FLAG_INT;
6249 
6250 	/* RELEASE_MEM - flush caches, send int */
6251 	amdgpu_ring_write(ring, PACKET3(PACKET3_RELEASE_MEM, 5));
6252 	amdgpu_ring_write(ring, (EOP_TCL1_ACTION_EN |
6253 				 EOP_TC_ACTION_EN |
6254 				 EOP_TC_WB_ACTION_EN |
6255 				 EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
6256 				 EVENT_INDEX(5)));
6257 	amdgpu_ring_write(ring, DATA_SEL(write64bit ? 2 : 1) | INT_SEL(int_sel ? 2 : 0));
6258 	amdgpu_ring_write(ring, addr & 0xfffffffc);
6259 	amdgpu_ring_write(ring, upper_32_bits(addr));
6260 	amdgpu_ring_write(ring, lower_32_bits(seq));
6261 	amdgpu_ring_write(ring, upper_32_bits(seq));
6262 }
6263 
6264 static void gfx_v8_0_ring_emit_fence_kiq(struct amdgpu_ring *ring, u64 addr,
6265 					 u64 seq, unsigned int flags)
6266 {
6267 	/* we only allocate 32bit for each seq wb address */
6268 	BUG_ON(flags & AMDGPU_FENCE_FLAG_64BIT);
6269 
6270 	/* write fence seq to the "addr" */
6271 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
6272 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
6273 				 WRITE_DATA_DST_SEL(5) | WR_CONFIRM));
6274 	amdgpu_ring_write(ring, lower_32_bits(addr));
6275 	amdgpu_ring_write(ring, upper_32_bits(addr));
6276 	amdgpu_ring_write(ring, lower_32_bits(seq));
6277 
6278 	if (flags & AMDGPU_FENCE_FLAG_INT) {
6279 		/* set register to trigger INT */
6280 		amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
6281 		amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
6282 					 WRITE_DATA_DST_SEL(0) | WR_CONFIRM));
6283 		amdgpu_ring_write(ring, mmCPC_INT_STATUS);
6284 		amdgpu_ring_write(ring, 0);
6285 		amdgpu_ring_write(ring, 0x20000000); /* src_id is 178 */
6286 	}
6287 }
6288 
6289 static void gfx_v8_ring_emit_sb(struct amdgpu_ring *ring)
6290 {
6291 	amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
6292 	amdgpu_ring_write(ring, 0);
6293 }
6294 
6295 static void gfx_v8_ring_emit_cntxcntl(struct amdgpu_ring *ring, uint32_t flags)
6296 {
6297 	uint32_t dw2 = 0;
6298 
6299 	if (amdgpu_sriov_vf(ring->adev))
6300 		gfx_v8_0_ring_emit_ce_meta(ring);
6301 
6302 	dw2 |= 0x80000000; /* set load_enable otherwise this package is just NOPs */
6303 	if (flags & AMDGPU_HAVE_CTX_SWITCH) {
6304 		gfx_v8_0_ring_emit_vgt_flush(ring);
6305 		/* set load_global_config & load_global_uconfig */
6306 		dw2 |= 0x8001;
6307 		/* set load_cs_sh_regs */
6308 		dw2 |= 0x01000000;
6309 		/* set load_per_context_state & load_gfx_sh_regs for GFX */
6310 		dw2 |= 0x10002;
6311 
6312 		/* set load_ce_ram if preamble presented */
6313 		if (AMDGPU_PREAMBLE_IB_PRESENT & flags)
6314 			dw2 |= 0x10000000;
6315 	} else {
6316 		/* still load_ce_ram if this is the first time preamble presented
6317 		 * although there is no context switch happens.
6318 		 */
6319 		if (AMDGPU_PREAMBLE_IB_PRESENT_FIRST & flags)
6320 			dw2 |= 0x10000000;
6321 	}
6322 
6323 	amdgpu_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1));
6324 	amdgpu_ring_write(ring, dw2);
6325 	amdgpu_ring_write(ring, 0);
6326 }
6327 
6328 static unsigned gfx_v8_0_ring_emit_init_cond_exec(struct amdgpu_ring *ring,
6329 						  uint64_t addr)
6330 {
6331 	unsigned ret;
6332 
6333 	amdgpu_ring_write(ring, PACKET3(PACKET3_COND_EXEC, 3));
6334 	amdgpu_ring_write(ring, lower_32_bits(addr));
6335 	amdgpu_ring_write(ring, upper_32_bits(addr));
6336 	/* discard following DWs if *cond_exec_gpu_addr==0 */
6337 	amdgpu_ring_write(ring, 0);
6338 	ret = ring->wptr & ring->buf_mask;
6339 	/* patch dummy value later */
6340 	amdgpu_ring_write(ring, 0);
6341 	return ret;
6342 }
6343 
6344 static void gfx_v8_0_ring_emit_rreg(struct amdgpu_ring *ring, uint32_t reg,
6345 				    uint32_t reg_val_offs)
6346 {
6347 	struct amdgpu_device *adev = ring->adev;
6348 
6349 	amdgpu_ring_write(ring, PACKET3(PACKET3_COPY_DATA, 4));
6350 	amdgpu_ring_write(ring, 0 |	/* src: register*/
6351 				(5 << 8) |	/* dst: memory */
6352 				(1 << 20));	/* write confirm */
6353 	amdgpu_ring_write(ring, reg);
6354 	amdgpu_ring_write(ring, 0);
6355 	amdgpu_ring_write(ring, lower_32_bits(adev->wb.gpu_addr +
6356 				reg_val_offs * 4));
6357 	amdgpu_ring_write(ring, upper_32_bits(adev->wb.gpu_addr +
6358 				reg_val_offs * 4));
6359 }
6360 
6361 static void gfx_v8_0_ring_emit_wreg(struct amdgpu_ring *ring, uint32_t reg,
6362 				  uint32_t val)
6363 {
6364 	uint32_t cmd;
6365 
6366 	switch (ring->funcs->type) {
6367 	case AMDGPU_RING_TYPE_GFX:
6368 		cmd = WRITE_DATA_ENGINE_SEL(1) | WR_CONFIRM;
6369 		break;
6370 	case AMDGPU_RING_TYPE_KIQ:
6371 		cmd = 1 << 16; /* no inc addr */
6372 		break;
6373 	default:
6374 		cmd = WR_CONFIRM;
6375 		break;
6376 	}
6377 
6378 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
6379 	amdgpu_ring_write(ring, cmd);
6380 	amdgpu_ring_write(ring, reg);
6381 	amdgpu_ring_write(ring, 0);
6382 	amdgpu_ring_write(ring, val);
6383 }
6384 
6385 static void gfx_v8_0_wait_reg_mem(struct amdgpu_ring *ring, int eng_sel,
6386 				  int mem_space, int opt, uint32_t addr0,
6387 				  uint32_t addr1, uint32_t ref, uint32_t mask,
6388 				  uint32_t inv)
6389 {
6390 	amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
6391 	amdgpu_ring_write(ring,
6392 			  /* memory (1) or register (0) */
6393 			  (WAIT_REG_MEM_MEM_SPACE(mem_space) |
6394 			   WAIT_REG_MEM_OPERATION(opt) | /* wait */
6395 			   WAIT_REG_MEM_FUNCTION(3) |  /* equal */
6396 			   WAIT_REG_MEM_ENGINE(eng_sel)));
6397 
6398 	if (mem_space)
6399 		BUG_ON(addr0 & 0x3); /* Dword align */
6400 	amdgpu_ring_write(ring, addr0);
6401 	amdgpu_ring_write(ring, addr1);
6402 	amdgpu_ring_write(ring, ref);
6403 	amdgpu_ring_write(ring, mask);
6404 	amdgpu_ring_write(ring, inv); /* poll interval */
6405 }
6406 
6407 static void gfx_v8_0_ring_emit_reg_wait(struct amdgpu_ring *ring, uint32_t reg,
6408 					uint32_t val, uint32_t mask)
6409 {
6410 	gfx_v8_0_wait_reg_mem(ring, 0, 0, 0, reg, 0, val, mask, 0x20);
6411 }
6412 
6413 static void gfx_v8_0_ring_soft_recovery(struct amdgpu_ring *ring, unsigned vmid)
6414 {
6415 	struct amdgpu_device *adev = ring->adev;
6416 	uint32_t value = 0;
6417 
6418 	value = REG_SET_FIELD(value, SQ_CMD, CMD, 0x03);
6419 	value = REG_SET_FIELD(value, SQ_CMD, MODE, 0x01);
6420 	value = REG_SET_FIELD(value, SQ_CMD, CHECK_VMID, 1);
6421 	value = REG_SET_FIELD(value, SQ_CMD, VM_ID, vmid);
6422 	WREG32(mmSQ_CMD, value);
6423 }
6424 
6425 static void gfx_v8_0_set_gfx_eop_interrupt_state(struct amdgpu_device *adev,
6426 						 enum amdgpu_interrupt_state state)
6427 {
6428 	WREG32_FIELD(CP_INT_CNTL_RING0, TIME_STAMP_INT_ENABLE,
6429 		     state == AMDGPU_IRQ_STATE_DISABLE ? 0 : 1);
6430 }
6431 
6432 static void gfx_v8_0_set_compute_eop_interrupt_state(struct amdgpu_device *adev,
6433 						     int me, int pipe,
6434 						     enum amdgpu_interrupt_state state)
6435 {
6436 	u32 mec_int_cntl, mec_int_cntl_reg;
6437 
6438 	/*
6439 	 * amdgpu controls only the first MEC. That's why this function only
6440 	 * handles the setting of interrupts for this specific MEC. All other
6441 	 * pipes' interrupts are set by amdkfd.
6442 	 */
6443 
6444 	if (me == 1) {
6445 		switch (pipe) {
6446 		case 0:
6447 			mec_int_cntl_reg = mmCP_ME1_PIPE0_INT_CNTL;
6448 			break;
6449 		case 1:
6450 			mec_int_cntl_reg = mmCP_ME1_PIPE1_INT_CNTL;
6451 			break;
6452 		case 2:
6453 			mec_int_cntl_reg = mmCP_ME1_PIPE2_INT_CNTL;
6454 			break;
6455 		case 3:
6456 			mec_int_cntl_reg = mmCP_ME1_PIPE3_INT_CNTL;
6457 			break;
6458 		default:
6459 			DRM_DEBUG("invalid pipe %d\n", pipe);
6460 			return;
6461 		}
6462 	} else {
6463 		DRM_DEBUG("invalid me %d\n", me);
6464 		return;
6465 	}
6466 
6467 	switch (state) {
6468 	case AMDGPU_IRQ_STATE_DISABLE:
6469 		mec_int_cntl = RREG32(mec_int_cntl_reg);
6470 		mec_int_cntl &= ~CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK;
6471 		WREG32(mec_int_cntl_reg, mec_int_cntl);
6472 		break;
6473 	case AMDGPU_IRQ_STATE_ENABLE:
6474 		mec_int_cntl = RREG32(mec_int_cntl_reg);
6475 		mec_int_cntl |= CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK;
6476 		WREG32(mec_int_cntl_reg, mec_int_cntl);
6477 		break;
6478 	default:
6479 		break;
6480 	}
6481 }
6482 
6483 static int gfx_v8_0_set_priv_reg_fault_state(struct amdgpu_device *adev,
6484 					     struct amdgpu_irq_src *source,
6485 					     unsigned type,
6486 					     enum amdgpu_interrupt_state state)
6487 {
6488 	WREG32_FIELD(CP_INT_CNTL_RING0, PRIV_REG_INT_ENABLE,
6489 		     state == AMDGPU_IRQ_STATE_DISABLE ? 0 : 1);
6490 
6491 	return 0;
6492 }
6493 
6494 static int gfx_v8_0_set_priv_inst_fault_state(struct amdgpu_device *adev,
6495 					      struct amdgpu_irq_src *source,
6496 					      unsigned type,
6497 					      enum amdgpu_interrupt_state state)
6498 {
6499 	WREG32_FIELD(CP_INT_CNTL_RING0, PRIV_INSTR_INT_ENABLE,
6500 		     state == AMDGPU_IRQ_STATE_DISABLE ? 0 : 1);
6501 
6502 	return 0;
6503 }
6504 
6505 static int gfx_v8_0_set_eop_interrupt_state(struct amdgpu_device *adev,
6506 					    struct amdgpu_irq_src *src,
6507 					    unsigned type,
6508 					    enum amdgpu_interrupt_state state)
6509 {
6510 	switch (type) {
6511 	case AMDGPU_CP_IRQ_GFX_ME0_PIPE0_EOP:
6512 		gfx_v8_0_set_gfx_eop_interrupt_state(adev, state);
6513 		break;
6514 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP:
6515 		gfx_v8_0_set_compute_eop_interrupt_state(adev, 1, 0, state);
6516 		break;
6517 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE1_EOP:
6518 		gfx_v8_0_set_compute_eop_interrupt_state(adev, 1, 1, state);
6519 		break;
6520 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE2_EOP:
6521 		gfx_v8_0_set_compute_eop_interrupt_state(adev, 1, 2, state);
6522 		break;
6523 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE3_EOP:
6524 		gfx_v8_0_set_compute_eop_interrupt_state(adev, 1, 3, state);
6525 		break;
6526 	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE0_EOP:
6527 		gfx_v8_0_set_compute_eop_interrupt_state(adev, 2, 0, state);
6528 		break;
6529 	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE1_EOP:
6530 		gfx_v8_0_set_compute_eop_interrupt_state(adev, 2, 1, state);
6531 		break;
6532 	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE2_EOP:
6533 		gfx_v8_0_set_compute_eop_interrupt_state(adev, 2, 2, state);
6534 		break;
6535 	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE3_EOP:
6536 		gfx_v8_0_set_compute_eop_interrupt_state(adev, 2, 3, state);
6537 		break;
6538 	default:
6539 		break;
6540 	}
6541 	return 0;
6542 }
6543 
6544 static int gfx_v8_0_set_cp_ecc_int_state(struct amdgpu_device *adev,
6545 					 struct amdgpu_irq_src *source,
6546 					 unsigned int type,
6547 					 enum amdgpu_interrupt_state state)
6548 {
6549 	int enable_flag;
6550 
6551 	switch (state) {
6552 	case AMDGPU_IRQ_STATE_DISABLE:
6553 		enable_flag = 0;
6554 		break;
6555 
6556 	case AMDGPU_IRQ_STATE_ENABLE:
6557 		enable_flag = 1;
6558 		break;
6559 
6560 	default:
6561 		return -EINVAL;
6562 	}
6563 
6564 	WREG32_FIELD(CP_INT_CNTL, CP_ECC_ERROR_INT_ENABLE, enable_flag);
6565 	WREG32_FIELD(CP_INT_CNTL_RING0, CP_ECC_ERROR_INT_ENABLE, enable_flag);
6566 	WREG32_FIELD(CP_INT_CNTL_RING1, CP_ECC_ERROR_INT_ENABLE, enable_flag);
6567 	WREG32_FIELD(CP_INT_CNTL_RING2, CP_ECC_ERROR_INT_ENABLE, enable_flag);
6568 	WREG32_FIELD(CPC_INT_CNTL, CP_ECC_ERROR_INT_ENABLE, enable_flag);
6569 	WREG32_FIELD(CP_ME1_PIPE0_INT_CNTL, CP_ECC_ERROR_INT_ENABLE,
6570 		     enable_flag);
6571 	WREG32_FIELD(CP_ME1_PIPE1_INT_CNTL, CP_ECC_ERROR_INT_ENABLE,
6572 		     enable_flag);
6573 	WREG32_FIELD(CP_ME1_PIPE2_INT_CNTL, CP_ECC_ERROR_INT_ENABLE,
6574 		     enable_flag);
6575 	WREG32_FIELD(CP_ME1_PIPE3_INT_CNTL, CP_ECC_ERROR_INT_ENABLE,
6576 		     enable_flag);
6577 	WREG32_FIELD(CP_ME2_PIPE0_INT_CNTL, CP_ECC_ERROR_INT_ENABLE,
6578 		     enable_flag);
6579 	WREG32_FIELD(CP_ME2_PIPE1_INT_CNTL, CP_ECC_ERROR_INT_ENABLE,
6580 		     enable_flag);
6581 	WREG32_FIELD(CP_ME2_PIPE2_INT_CNTL, CP_ECC_ERROR_INT_ENABLE,
6582 		     enable_flag);
6583 	WREG32_FIELD(CP_ME2_PIPE3_INT_CNTL, CP_ECC_ERROR_INT_ENABLE,
6584 		     enable_flag);
6585 
6586 	return 0;
6587 }
6588 
6589 static int gfx_v8_0_set_sq_int_state(struct amdgpu_device *adev,
6590 				     struct amdgpu_irq_src *source,
6591 				     unsigned int type,
6592 				     enum amdgpu_interrupt_state state)
6593 {
6594 	int enable_flag;
6595 
6596 	switch (state) {
6597 	case AMDGPU_IRQ_STATE_DISABLE:
6598 		enable_flag = 1;
6599 		break;
6600 
6601 	case AMDGPU_IRQ_STATE_ENABLE:
6602 		enable_flag = 0;
6603 		break;
6604 
6605 	default:
6606 		return -EINVAL;
6607 	}
6608 
6609 	WREG32_FIELD(SQ_INTERRUPT_MSG_CTRL, STALL,
6610 		     enable_flag);
6611 
6612 	return 0;
6613 }
6614 
6615 static int gfx_v8_0_eop_irq(struct amdgpu_device *adev,
6616 			    struct amdgpu_irq_src *source,
6617 			    struct amdgpu_iv_entry *entry)
6618 {
6619 	int i;
6620 	u8 me_id, pipe_id, queue_id;
6621 	struct amdgpu_ring *ring;
6622 
6623 	DRM_DEBUG("IH: CP EOP\n");
6624 	me_id = (entry->ring_id & 0x0c) >> 2;
6625 	pipe_id = (entry->ring_id & 0x03) >> 0;
6626 	queue_id = (entry->ring_id & 0x70) >> 4;
6627 
6628 	switch (me_id) {
6629 	case 0:
6630 		amdgpu_fence_process(&adev->gfx.gfx_ring[0]);
6631 		break;
6632 	case 1:
6633 	case 2:
6634 		for (i = 0; i < adev->gfx.num_compute_rings; i++) {
6635 			ring = &adev->gfx.compute_ring[i];
6636 			/* Per-queue interrupt is supported for MEC starting from VI.
6637 			  * The interrupt can only be enabled/disabled per pipe instead of per queue.
6638 			  */
6639 			if ((ring->me == me_id) && (ring->pipe == pipe_id) && (ring->queue == queue_id))
6640 				amdgpu_fence_process(ring);
6641 		}
6642 		break;
6643 	}
6644 	return 0;
6645 }
6646 
6647 static void gfx_v8_0_fault(struct amdgpu_device *adev,
6648 			   struct amdgpu_iv_entry *entry)
6649 {
6650 	u8 me_id, pipe_id, queue_id;
6651 	struct amdgpu_ring *ring;
6652 	int i;
6653 
6654 	me_id = (entry->ring_id & 0x0c) >> 2;
6655 	pipe_id = (entry->ring_id & 0x03) >> 0;
6656 	queue_id = (entry->ring_id & 0x70) >> 4;
6657 
6658 	switch (me_id) {
6659 	case 0:
6660 		drm_sched_fault(&adev->gfx.gfx_ring[0].sched);
6661 		break;
6662 	case 1:
6663 	case 2:
6664 		for (i = 0; i < adev->gfx.num_compute_rings; i++) {
6665 			ring = &adev->gfx.compute_ring[i];
6666 			if (ring->me == me_id && ring->pipe == pipe_id &&
6667 			    ring->queue == queue_id)
6668 				drm_sched_fault(&ring->sched);
6669 		}
6670 		break;
6671 	}
6672 }
6673 
6674 static int gfx_v8_0_priv_reg_irq(struct amdgpu_device *adev,
6675 				 struct amdgpu_irq_src *source,
6676 				 struct amdgpu_iv_entry *entry)
6677 {
6678 	DRM_ERROR("Illegal register access in command stream\n");
6679 	gfx_v8_0_fault(adev, entry);
6680 	return 0;
6681 }
6682 
6683 static int gfx_v8_0_priv_inst_irq(struct amdgpu_device *adev,
6684 				  struct amdgpu_irq_src *source,
6685 				  struct amdgpu_iv_entry *entry)
6686 {
6687 	DRM_ERROR("Illegal instruction in command stream\n");
6688 	gfx_v8_0_fault(adev, entry);
6689 	return 0;
6690 }
6691 
6692 static int gfx_v8_0_cp_ecc_error_irq(struct amdgpu_device *adev,
6693 				     struct amdgpu_irq_src *source,
6694 				     struct amdgpu_iv_entry *entry)
6695 {
6696 	DRM_ERROR("CP EDC/ECC error detected.");
6697 	return 0;
6698 }
6699 
6700 static void gfx_v8_0_parse_sq_irq(struct amdgpu_device *adev, unsigned ih_data,
6701 				  bool from_wq)
6702 {
6703 	u32 enc, se_id, sh_id, cu_id;
6704 	char type[20];
6705 	int sq_edc_source = -1;
6706 
6707 	enc = REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_CMN, ENCODING);
6708 	se_id = REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_CMN, SE_ID);
6709 
6710 	switch (enc) {
6711 		case 0:
6712 			DRM_INFO("SQ general purpose intr detected:"
6713 					"se_id %d, immed_overflow %d, host_reg_overflow %d,"
6714 					"host_cmd_overflow %d, cmd_timestamp %d,"
6715 					"reg_timestamp %d, thread_trace_buff_full %d,"
6716 					"wlt %d, thread_trace %d.\n",
6717 					se_id,
6718 					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_AUTO, IMMED_OVERFLOW),
6719 					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_AUTO, HOST_REG_OVERFLOW),
6720 					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_AUTO, HOST_CMD_OVERFLOW),
6721 					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_AUTO, CMD_TIMESTAMP),
6722 					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_AUTO, REG_TIMESTAMP),
6723 					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_AUTO, THREAD_TRACE_BUF_FULL),
6724 					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_AUTO, WLT),
6725 					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_AUTO, THREAD_TRACE)
6726 					);
6727 			break;
6728 		case 1:
6729 		case 2:
6730 
6731 			cu_id = REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_WAVE, CU_ID);
6732 			sh_id = REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_WAVE, SH_ID);
6733 
6734 			/*
6735 			 * This function can be called either directly from ISR
6736 			 * or from BH in which case we can access SQ_EDC_INFO
6737 			 * instance
6738 			 */
6739 			if (from_wq) {
6740 				mutex_lock(&adev->grbm_idx_mutex);
6741 				gfx_v8_0_select_se_sh(adev, se_id, sh_id, cu_id, 0);
6742 
6743 				sq_edc_source = REG_GET_FIELD(RREG32(mmSQ_EDC_INFO), SQ_EDC_INFO, SOURCE);
6744 
6745 				gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 0);
6746 				mutex_unlock(&adev->grbm_idx_mutex);
6747 			}
6748 
6749 			if (enc == 1)
6750 				sprintf(type, "instruction intr");
6751 			else
6752 				sprintf(type, "EDC/ECC error");
6753 
6754 			DRM_INFO(
6755 				"SQ %s detected: "
6756 					"se_id %d, sh_id %d, cu_id %d, simd_id %d, wave_id %d, vm_id %d "
6757 					"trap %s, sq_ed_info.source %s.\n",
6758 					type, se_id, sh_id, cu_id,
6759 					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_WAVE, SIMD_ID),
6760 					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_WAVE, WAVE_ID),
6761 					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_WAVE, VM_ID),
6762 					REG_GET_FIELD(ih_data, SQ_INTERRUPT_WORD_WAVE, PRIV) ? "true" : "false",
6763 					(sq_edc_source != -1) ? sq_edc_source_names[sq_edc_source] : "unavailable"
6764 				);
6765 			break;
6766 		default:
6767 			DRM_ERROR("SQ invalid encoding type\n.");
6768 	}
6769 }
6770 
6771 static void gfx_v8_0_sq_irq_work_func(struct work_struct *work)
6772 {
6773 
6774 	struct amdgpu_device *adev = container_of(work, struct amdgpu_device, gfx.sq_work.work);
6775 	struct sq_work *sq_work = container_of(work, struct sq_work, work);
6776 
6777 	gfx_v8_0_parse_sq_irq(adev, sq_work->ih_data, true);
6778 }
6779 
6780 static int gfx_v8_0_sq_irq(struct amdgpu_device *adev,
6781 			   struct amdgpu_irq_src *source,
6782 			   struct amdgpu_iv_entry *entry)
6783 {
6784 	unsigned ih_data = entry->src_data[0];
6785 
6786 	/*
6787 	 * Try to submit work so SQ_EDC_INFO can be accessed from
6788 	 * BH. If previous work submission hasn't finished yet
6789 	 * just print whatever info is possible directly from the ISR.
6790 	 */
6791 	if (work_pending(&adev->gfx.sq_work.work)) {
6792 		gfx_v8_0_parse_sq_irq(adev, ih_data, false);
6793 	} else {
6794 		adev->gfx.sq_work.ih_data = ih_data;
6795 		schedule_work(&adev->gfx.sq_work.work);
6796 	}
6797 
6798 	return 0;
6799 }
6800 
6801 static void gfx_v8_0_emit_mem_sync(struct amdgpu_ring *ring)
6802 {
6803 	amdgpu_ring_write(ring, PACKET3(PACKET3_SURFACE_SYNC, 3));
6804 	amdgpu_ring_write(ring, PACKET3_TCL1_ACTION_ENA |
6805 			  PACKET3_TC_ACTION_ENA |
6806 			  PACKET3_SH_KCACHE_ACTION_ENA |
6807 			  PACKET3_SH_ICACHE_ACTION_ENA |
6808 			  PACKET3_TC_WB_ACTION_ENA);  /* CP_COHER_CNTL */
6809 	amdgpu_ring_write(ring, 0xffffffff);  /* CP_COHER_SIZE */
6810 	amdgpu_ring_write(ring, 0);  /* CP_COHER_BASE */
6811 	amdgpu_ring_write(ring, 0x0000000A); /* poll interval */
6812 }
6813 
6814 static void gfx_v8_0_emit_mem_sync_compute(struct amdgpu_ring *ring)
6815 {
6816 	amdgpu_ring_write(ring, PACKET3(PACKET3_ACQUIRE_MEM, 5));
6817 	amdgpu_ring_write(ring, PACKET3_TCL1_ACTION_ENA |
6818 			  PACKET3_TC_ACTION_ENA |
6819 			  PACKET3_SH_KCACHE_ACTION_ENA |
6820 			  PACKET3_SH_ICACHE_ACTION_ENA |
6821 			  PACKET3_TC_WB_ACTION_ENA);  /* CP_COHER_CNTL */
6822 	amdgpu_ring_write(ring, 0xffffffff);	/* CP_COHER_SIZE */
6823 	amdgpu_ring_write(ring, 0xff);		/* CP_COHER_SIZE_HI */
6824 	amdgpu_ring_write(ring, 0);		/* CP_COHER_BASE */
6825 	amdgpu_ring_write(ring, 0);		/* CP_COHER_BASE_HI */
6826 	amdgpu_ring_write(ring, 0x0000000A);	/* poll interval */
6827 }
6828 
6829 
6830 /* mmSPI_WCL_PIPE_PERCENT_CS[0-7]_DEFAULT values are same */
6831 #define mmSPI_WCL_PIPE_PERCENT_CS_DEFAULT	0x0000007f
6832 static void gfx_v8_0_emit_wave_limit_cs(struct amdgpu_ring *ring,
6833 					uint32_t pipe, bool enable)
6834 {
6835 	uint32_t val;
6836 	uint32_t wcl_cs_reg;
6837 
6838 	val = enable ? 0x1 : mmSPI_WCL_PIPE_PERCENT_CS_DEFAULT;
6839 
6840 	switch (pipe) {
6841 	case 0:
6842 		wcl_cs_reg = mmSPI_WCL_PIPE_PERCENT_CS0;
6843 		break;
6844 	case 1:
6845 		wcl_cs_reg = mmSPI_WCL_PIPE_PERCENT_CS1;
6846 		break;
6847 	case 2:
6848 		wcl_cs_reg = mmSPI_WCL_PIPE_PERCENT_CS2;
6849 		break;
6850 	case 3:
6851 		wcl_cs_reg = mmSPI_WCL_PIPE_PERCENT_CS3;
6852 		break;
6853 	default:
6854 		DRM_DEBUG("invalid pipe %d\n", pipe);
6855 		return;
6856 	}
6857 
6858 	amdgpu_ring_emit_wreg(ring, wcl_cs_reg, val);
6859 
6860 }
6861 
6862 #define mmSPI_WCL_PIPE_PERCENT_GFX_DEFAULT	0x07ffffff
6863 static void gfx_v8_0_emit_wave_limit(struct amdgpu_ring *ring, bool enable)
6864 {
6865 	struct amdgpu_device *adev = ring->adev;
6866 	uint32_t val;
6867 	int i;
6868 
6869 	/* mmSPI_WCL_PIPE_PERCENT_GFX is 7 bit multiplier register to limit
6870 	 * number of gfx waves. Setting 5 bit will make sure gfx only gets
6871 	 * around 25% of gpu resources.
6872 	 */
6873 	val = enable ? 0x1f : mmSPI_WCL_PIPE_PERCENT_GFX_DEFAULT;
6874 	amdgpu_ring_emit_wreg(ring, mmSPI_WCL_PIPE_PERCENT_GFX, val);
6875 
6876 	/* Restrict waves for normal/low priority compute queues as well
6877 	 * to get best QoS for high priority compute jobs.
6878 	 *
6879 	 * amdgpu controls only 1st ME(0-3 CS pipes).
6880 	 */
6881 	for (i = 0; i < adev->gfx.mec.num_pipe_per_mec; i++) {
6882 		if (i != ring->pipe)
6883 			gfx_v8_0_emit_wave_limit_cs(ring, i, enable);
6884 
6885 	}
6886 
6887 }
6888 
6889 static int gfx_v8_0_reset_kgq(struct amdgpu_ring *ring, unsigned int vmid)
6890 {
6891 	struct amdgpu_device *adev = ring->adev;
6892 	struct amdgpu_kiq *kiq = &adev->gfx.kiq[0];
6893 	struct amdgpu_ring *kiq_ring = &kiq->ring;
6894 	unsigned long flags;
6895 	u32 tmp;
6896 	int r;
6897 
6898 	if (amdgpu_sriov_vf(adev))
6899 		return -EINVAL;
6900 
6901 	if (!kiq->pmf || !kiq->pmf->kiq_unmap_queues)
6902 		return -EINVAL;
6903 
6904 	spin_lock_irqsave(&kiq->ring_lock, flags);
6905 
6906 	if (amdgpu_ring_alloc(kiq_ring, 5)) {
6907 		spin_unlock_irqrestore(&kiq->ring_lock, flags);
6908 		return -ENOMEM;
6909 	}
6910 
6911 	tmp = REG_SET_FIELD(0, CP_VMID_RESET, RESET_REQUEST, 1 << vmid);
6912 	gfx_v8_0_ring_emit_wreg(kiq_ring, mmCP_VMID_RESET, tmp);
6913 	amdgpu_ring_commit(kiq_ring);
6914 
6915 	spin_unlock_irqrestore(&kiq->ring_lock, flags);
6916 
6917 	r = amdgpu_ring_test_ring(kiq_ring);
6918 	if (r)
6919 		return r;
6920 
6921 	if (amdgpu_ring_alloc(ring, 7 + 12 + 5))
6922 		return -ENOMEM;
6923 	gfx_v8_0_ring_emit_fence_gfx(ring, ring->fence_drv.gpu_addr,
6924 				     ring->fence_drv.sync_seq, AMDGPU_FENCE_FLAG_EXEC);
6925 	gfx_v8_0_ring_emit_reg_wait(ring, mmCP_VMID_RESET, 0, 0xffff);
6926 	gfx_v8_0_ring_emit_wreg(ring, mmCP_VMID_RESET, 0);
6927 
6928 	return amdgpu_ring_test_ring(ring);
6929 }
6930 
6931 static const struct amd_ip_funcs gfx_v8_0_ip_funcs = {
6932 	.name = "gfx_v8_0",
6933 	.early_init = gfx_v8_0_early_init,
6934 	.late_init = gfx_v8_0_late_init,
6935 	.sw_init = gfx_v8_0_sw_init,
6936 	.sw_fini = gfx_v8_0_sw_fini,
6937 	.hw_init = gfx_v8_0_hw_init,
6938 	.hw_fini = gfx_v8_0_hw_fini,
6939 	.suspend = gfx_v8_0_suspend,
6940 	.resume = gfx_v8_0_resume,
6941 	.is_idle = gfx_v8_0_is_idle,
6942 	.wait_for_idle = gfx_v8_0_wait_for_idle,
6943 	.check_soft_reset = gfx_v8_0_check_soft_reset,
6944 	.pre_soft_reset = gfx_v8_0_pre_soft_reset,
6945 	.soft_reset = gfx_v8_0_soft_reset,
6946 	.post_soft_reset = gfx_v8_0_post_soft_reset,
6947 	.set_clockgating_state = gfx_v8_0_set_clockgating_state,
6948 	.set_powergating_state = gfx_v8_0_set_powergating_state,
6949 	.get_clockgating_state = gfx_v8_0_get_clockgating_state,
6950 	.dump_ip_state = NULL,
6951 	.print_ip_state = NULL,
6952 };
6953 
6954 static const struct amdgpu_ring_funcs gfx_v8_0_ring_funcs_gfx = {
6955 	.type = AMDGPU_RING_TYPE_GFX,
6956 	.align_mask = 0xff,
6957 	.nop = PACKET3(PACKET3_NOP, 0x3FFF),
6958 	.support_64bit_ptrs = false,
6959 	.get_rptr = gfx_v8_0_ring_get_rptr,
6960 	.get_wptr = gfx_v8_0_ring_get_wptr_gfx,
6961 	.set_wptr = gfx_v8_0_ring_set_wptr_gfx,
6962 	.emit_frame_size = /* maximum 215dw if count 16 IBs in */
6963 		5 +  /* COND_EXEC */
6964 		7 +  /* PIPELINE_SYNC */
6965 		VI_FLUSH_GPU_TLB_NUM_WREG * 5 + 9 + /* VM_FLUSH */
6966 		12 +  /* FENCE for VM_FLUSH */
6967 		20 + /* GDS switch */
6968 		4 + /* double SWITCH_BUFFER,
6969 		       the first COND_EXEC jump to the place just
6970 			   prior to this double SWITCH_BUFFER  */
6971 		5 + /* COND_EXEC */
6972 		7 +	 /*	HDP_flush */
6973 		4 +	 /*	VGT_flush */
6974 		14 + /*	CE_META */
6975 		31 + /*	DE_META */
6976 		3 + /* CNTX_CTRL */
6977 		5 + /* HDP_INVL */
6978 		12 + 12 + /* FENCE x2 */
6979 		2 + /* SWITCH_BUFFER */
6980 		5, /* SURFACE_SYNC */
6981 	.emit_ib_size =	4, /* gfx_v8_0_ring_emit_ib_gfx */
6982 	.emit_ib = gfx_v8_0_ring_emit_ib_gfx,
6983 	.emit_fence = gfx_v8_0_ring_emit_fence_gfx,
6984 	.emit_pipeline_sync = gfx_v8_0_ring_emit_pipeline_sync,
6985 	.emit_vm_flush = gfx_v8_0_ring_emit_vm_flush,
6986 	.emit_gds_switch = gfx_v8_0_ring_emit_gds_switch,
6987 	.emit_hdp_flush = gfx_v8_0_ring_emit_hdp_flush,
6988 	.test_ring = gfx_v8_0_ring_test_ring,
6989 	.test_ib = gfx_v8_0_ring_test_ib,
6990 	.insert_nop = amdgpu_ring_insert_nop,
6991 	.pad_ib = amdgpu_ring_generic_pad_ib,
6992 	.emit_switch_buffer = gfx_v8_ring_emit_sb,
6993 	.emit_cntxcntl = gfx_v8_ring_emit_cntxcntl,
6994 	.init_cond_exec = gfx_v8_0_ring_emit_init_cond_exec,
6995 	.emit_wreg = gfx_v8_0_ring_emit_wreg,
6996 	.soft_recovery = gfx_v8_0_ring_soft_recovery,
6997 	.emit_mem_sync = gfx_v8_0_emit_mem_sync,
6998 	.reset = gfx_v8_0_reset_kgq,
6999 };
7000 
7001 static const struct amdgpu_ring_funcs gfx_v8_0_ring_funcs_compute = {
7002 	.type = AMDGPU_RING_TYPE_COMPUTE,
7003 	.align_mask = 0xff,
7004 	.nop = PACKET3(PACKET3_NOP, 0x3FFF),
7005 	.support_64bit_ptrs = false,
7006 	.get_rptr = gfx_v8_0_ring_get_rptr,
7007 	.get_wptr = gfx_v8_0_ring_get_wptr_compute,
7008 	.set_wptr = gfx_v8_0_ring_set_wptr_compute,
7009 	.emit_frame_size =
7010 		20 + /* gfx_v8_0_ring_emit_gds_switch */
7011 		7 + /* gfx_v8_0_ring_emit_hdp_flush */
7012 		5 + /* hdp_invalidate */
7013 		7 + /* gfx_v8_0_ring_emit_pipeline_sync */
7014 		VI_FLUSH_GPU_TLB_NUM_WREG * 5 + 7 + /* gfx_v8_0_ring_emit_vm_flush */
7015 		7 + 7 + 7 + /* gfx_v8_0_ring_emit_fence_compute x3 for user fence, vm fence */
7016 		7 + /* gfx_v8_0_emit_mem_sync_compute */
7017 		5 + /* gfx_v8_0_emit_wave_limit for updating mmSPI_WCL_PIPE_PERCENT_GFX register */
7018 		15, /* for updating 3 mmSPI_WCL_PIPE_PERCENT_CS registers */
7019 	.emit_ib_size =	7, /* gfx_v8_0_ring_emit_ib_compute */
7020 	.emit_ib = gfx_v8_0_ring_emit_ib_compute,
7021 	.emit_fence = gfx_v8_0_ring_emit_fence_compute,
7022 	.emit_pipeline_sync = gfx_v8_0_ring_emit_pipeline_sync,
7023 	.emit_vm_flush = gfx_v8_0_ring_emit_vm_flush,
7024 	.emit_gds_switch = gfx_v8_0_ring_emit_gds_switch,
7025 	.emit_hdp_flush = gfx_v8_0_ring_emit_hdp_flush,
7026 	.test_ring = gfx_v8_0_ring_test_ring,
7027 	.test_ib = gfx_v8_0_ring_test_ib,
7028 	.insert_nop = amdgpu_ring_insert_nop,
7029 	.pad_ib = amdgpu_ring_generic_pad_ib,
7030 	.emit_wreg = gfx_v8_0_ring_emit_wreg,
7031 	.soft_recovery = gfx_v8_0_ring_soft_recovery,
7032 	.emit_mem_sync = gfx_v8_0_emit_mem_sync_compute,
7033 	.emit_wave_limit = gfx_v8_0_emit_wave_limit,
7034 };
7035 
7036 static const struct amdgpu_ring_funcs gfx_v8_0_ring_funcs_kiq = {
7037 	.type = AMDGPU_RING_TYPE_KIQ,
7038 	.align_mask = 0xff,
7039 	.nop = PACKET3(PACKET3_NOP, 0x3FFF),
7040 	.support_64bit_ptrs = false,
7041 	.get_rptr = gfx_v8_0_ring_get_rptr,
7042 	.get_wptr = gfx_v8_0_ring_get_wptr_compute,
7043 	.set_wptr = gfx_v8_0_ring_set_wptr_compute,
7044 	.emit_frame_size =
7045 		20 + /* gfx_v8_0_ring_emit_gds_switch */
7046 		7 + /* gfx_v8_0_ring_emit_hdp_flush */
7047 		5 + /* hdp_invalidate */
7048 		7 + /* gfx_v8_0_ring_emit_pipeline_sync */
7049 		17 + /* gfx_v8_0_ring_emit_vm_flush */
7050 		7 + 7 + 7, /* gfx_v8_0_ring_emit_fence_kiq x3 for user fence, vm fence */
7051 	.emit_ib_size =	7, /* gfx_v8_0_ring_emit_ib_compute */
7052 	.emit_fence = gfx_v8_0_ring_emit_fence_kiq,
7053 	.test_ring = gfx_v8_0_ring_test_ring,
7054 	.insert_nop = amdgpu_ring_insert_nop,
7055 	.pad_ib = amdgpu_ring_generic_pad_ib,
7056 	.emit_rreg = gfx_v8_0_ring_emit_rreg,
7057 	.emit_wreg = gfx_v8_0_ring_emit_wreg,
7058 };
7059 
7060 static void gfx_v8_0_set_ring_funcs(struct amdgpu_device *adev)
7061 {
7062 	int i;
7063 
7064 	adev->gfx.kiq[0].ring.funcs = &gfx_v8_0_ring_funcs_kiq;
7065 
7066 	for (i = 0; i < adev->gfx.num_gfx_rings; i++)
7067 		adev->gfx.gfx_ring[i].funcs = &gfx_v8_0_ring_funcs_gfx;
7068 
7069 	for (i = 0; i < adev->gfx.num_compute_rings; i++)
7070 		adev->gfx.compute_ring[i].funcs = &gfx_v8_0_ring_funcs_compute;
7071 }
7072 
7073 static const struct amdgpu_irq_src_funcs gfx_v8_0_eop_irq_funcs = {
7074 	.set = gfx_v8_0_set_eop_interrupt_state,
7075 	.process = gfx_v8_0_eop_irq,
7076 };
7077 
7078 static const struct amdgpu_irq_src_funcs gfx_v8_0_priv_reg_irq_funcs = {
7079 	.set = gfx_v8_0_set_priv_reg_fault_state,
7080 	.process = gfx_v8_0_priv_reg_irq,
7081 };
7082 
7083 static const struct amdgpu_irq_src_funcs gfx_v8_0_priv_inst_irq_funcs = {
7084 	.set = gfx_v8_0_set_priv_inst_fault_state,
7085 	.process = gfx_v8_0_priv_inst_irq,
7086 };
7087 
7088 static const struct amdgpu_irq_src_funcs gfx_v8_0_cp_ecc_error_irq_funcs = {
7089 	.set = gfx_v8_0_set_cp_ecc_int_state,
7090 	.process = gfx_v8_0_cp_ecc_error_irq,
7091 };
7092 
7093 static const struct amdgpu_irq_src_funcs gfx_v8_0_sq_irq_funcs = {
7094 	.set = gfx_v8_0_set_sq_int_state,
7095 	.process = gfx_v8_0_sq_irq,
7096 };
7097 
7098 static void gfx_v8_0_set_irq_funcs(struct amdgpu_device *adev)
7099 {
7100 	adev->gfx.eop_irq.num_types = AMDGPU_CP_IRQ_LAST;
7101 	adev->gfx.eop_irq.funcs = &gfx_v8_0_eop_irq_funcs;
7102 
7103 	adev->gfx.priv_reg_irq.num_types = 1;
7104 	adev->gfx.priv_reg_irq.funcs = &gfx_v8_0_priv_reg_irq_funcs;
7105 
7106 	adev->gfx.priv_inst_irq.num_types = 1;
7107 	adev->gfx.priv_inst_irq.funcs = &gfx_v8_0_priv_inst_irq_funcs;
7108 
7109 	adev->gfx.cp_ecc_error_irq.num_types = 1;
7110 	adev->gfx.cp_ecc_error_irq.funcs = &gfx_v8_0_cp_ecc_error_irq_funcs;
7111 
7112 	adev->gfx.sq_irq.num_types = 1;
7113 	adev->gfx.sq_irq.funcs = &gfx_v8_0_sq_irq_funcs;
7114 }
7115 
7116 static void gfx_v8_0_set_rlc_funcs(struct amdgpu_device *adev)
7117 {
7118 	adev->gfx.rlc.funcs = &iceland_rlc_funcs;
7119 }
7120 
7121 static void gfx_v8_0_set_gds_init(struct amdgpu_device *adev)
7122 {
7123 	/* init asci gds info */
7124 	adev->gds.gds_size = RREG32(mmGDS_VMID0_SIZE);
7125 	adev->gds.gws_size = 64;
7126 	adev->gds.oa_size = 16;
7127 	adev->gds.gds_compute_max_wave_id = RREG32(mmGDS_COMPUTE_MAX_WAVE_ID);
7128 }
7129 
7130 static void gfx_v8_0_set_user_cu_inactive_bitmap(struct amdgpu_device *adev,
7131 						 u32 bitmap)
7132 {
7133 	u32 data;
7134 
7135 	if (!bitmap)
7136 		return;
7137 
7138 	data = bitmap << GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_CUS__SHIFT;
7139 	data &= GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_CUS_MASK;
7140 
7141 	WREG32(mmGC_USER_SHADER_ARRAY_CONFIG, data);
7142 }
7143 
7144 static u32 gfx_v8_0_get_cu_active_bitmap(struct amdgpu_device *adev)
7145 {
7146 	u32 data, mask;
7147 
7148 	data =  RREG32(mmCC_GC_SHADER_ARRAY_CONFIG) |
7149 		RREG32(mmGC_USER_SHADER_ARRAY_CONFIG);
7150 
7151 	mask = amdgpu_gfx_create_bitmask(adev->gfx.config.max_cu_per_sh);
7152 
7153 	return ~REG_GET_FIELD(data, CC_GC_SHADER_ARRAY_CONFIG, INACTIVE_CUS) & mask;
7154 }
7155 
7156 static void gfx_v8_0_get_cu_info(struct amdgpu_device *adev)
7157 {
7158 	int i, j, k, counter, active_cu_number = 0;
7159 	u32 mask, bitmap, ao_bitmap, ao_cu_mask = 0;
7160 	struct amdgpu_cu_info *cu_info = &adev->gfx.cu_info;
7161 	unsigned disable_masks[4 * 2];
7162 	u32 ao_cu_num;
7163 
7164 	memset(cu_info, 0, sizeof(*cu_info));
7165 
7166 	if (adev->flags & AMD_IS_APU)
7167 		ao_cu_num = 2;
7168 	else
7169 		ao_cu_num = adev->gfx.config.max_cu_per_sh;
7170 
7171 	amdgpu_gfx_parse_disable_cu(disable_masks, 4, 2);
7172 
7173 	mutex_lock(&adev->grbm_idx_mutex);
7174 	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
7175 		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
7176 			mask = 1;
7177 			ao_bitmap = 0;
7178 			counter = 0;
7179 			gfx_v8_0_select_se_sh(adev, i, j, 0xffffffff, 0);
7180 			if (i < 4 && j < 2)
7181 				gfx_v8_0_set_user_cu_inactive_bitmap(
7182 					adev, disable_masks[i * 2 + j]);
7183 			bitmap = gfx_v8_0_get_cu_active_bitmap(adev);
7184 			cu_info->bitmap[0][i][j] = bitmap;
7185 
7186 			for (k = 0; k < adev->gfx.config.max_cu_per_sh; k ++) {
7187 				if (bitmap & mask) {
7188 					if (counter < ao_cu_num)
7189 						ao_bitmap |= mask;
7190 					counter ++;
7191 				}
7192 				mask <<= 1;
7193 			}
7194 			active_cu_number += counter;
7195 			if (i < 2 && j < 2)
7196 				ao_cu_mask |= (ao_bitmap << (i * 16 + j * 8));
7197 			cu_info->ao_cu_bitmap[i][j] = ao_bitmap;
7198 		}
7199 	}
7200 	gfx_v8_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff, 0);
7201 	mutex_unlock(&adev->grbm_idx_mutex);
7202 
7203 	cu_info->number = active_cu_number;
7204 	cu_info->ao_cu_mask = ao_cu_mask;
7205 	cu_info->simd_per_cu = NUM_SIMD_PER_CU;
7206 	cu_info->max_waves_per_simd = 10;
7207 	cu_info->max_scratch_slots_per_cu = 32;
7208 	cu_info->wave_front_size = 64;
7209 	cu_info->lds_size = 64;
7210 }
7211 
7212 const struct amdgpu_ip_block_version gfx_v8_0_ip_block =
7213 {
7214 	.type = AMD_IP_BLOCK_TYPE_GFX,
7215 	.major = 8,
7216 	.minor = 0,
7217 	.rev = 0,
7218 	.funcs = &gfx_v8_0_ip_funcs,
7219 };
7220 
7221 const struct amdgpu_ip_block_version gfx_v8_1_ip_block =
7222 {
7223 	.type = AMD_IP_BLOCK_TYPE_GFX,
7224 	.major = 8,
7225 	.minor = 1,
7226 	.rev = 0,
7227 	.funcs = &gfx_v8_0_ip_funcs,
7228 };
7229 
7230 static void gfx_v8_0_ring_emit_ce_meta(struct amdgpu_ring *ring)
7231 {
7232 	uint64_t ce_payload_addr;
7233 	int cnt_ce;
7234 	union {
7235 		struct vi_ce_ib_state regular;
7236 		struct vi_ce_ib_state_chained_ib chained;
7237 	} ce_payload = {};
7238 
7239 	if (ring->adev->virt.chained_ib_support) {
7240 		ce_payload_addr = amdgpu_csa_vaddr(ring->adev) +
7241 			offsetof(struct vi_gfx_meta_data_chained_ib, ce_payload);
7242 		cnt_ce = (sizeof(ce_payload.chained) >> 2) + 4 - 2;
7243 	} else {
7244 		ce_payload_addr = amdgpu_csa_vaddr(ring->adev) +
7245 			offsetof(struct vi_gfx_meta_data, ce_payload);
7246 		cnt_ce = (sizeof(ce_payload.regular) >> 2) + 4 - 2;
7247 	}
7248 
7249 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, cnt_ce));
7250 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(2) |
7251 				WRITE_DATA_DST_SEL(8) |
7252 				WR_CONFIRM) |
7253 				WRITE_DATA_CACHE_POLICY(0));
7254 	amdgpu_ring_write(ring, lower_32_bits(ce_payload_addr));
7255 	amdgpu_ring_write(ring, upper_32_bits(ce_payload_addr));
7256 	amdgpu_ring_write_multiple(ring, (void *)&ce_payload, cnt_ce - 2);
7257 }
7258 
7259 static void gfx_v8_0_ring_emit_de_meta(struct amdgpu_ring *ring)
7260 {
7261 	uint64_t de_payload_addr, gds_addr, csa_addr;
7262 	int cnt_de;
7263 	union {
7264 		struct vi_de_ib_state regular;
7265 		struct vi_de_ib_state_chained_ib chained;
7266 	} de_payload = {};
7267 
7268 	csa_addr = amdgpu_csa_vaddr(ring->adev);
7269 	gds_addr = csa_addr + 4096;
7270 	if (ring->adev->virt.chained_ib_support) {
7271 		de_payload.chained.gds_backup_addrlo = lower_32_bits(gds_addr);
7272 		de_payload.chained.gds_backup_addrhi = upper_32_bits(gds_addr);
7273 		de_payload_addr = csa_addr + offsetof(struct vi_gfx_meta_data_chained_ib, de_payload);
7274 		cnt_de = (sizeof(de_payload.chained) >> 2) + 4 - 2;
7275 	} else {
7276 		de_payload.regular.gds_backup_addrlo = lower_32_bits(gds_addr);
7277 		de_payload.regular.gds_backup_addrhi = upper_32_bits(gds_addr);
7278 		de_payload_addr = csa_addr + offsetof(struct vi_gfx_meta_data, de_payload);
7279 		cnt_de = (sizeof(de_payload.regular) >> 2) + 4 - 2;
7280 	}
7281 
7282 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, cnt_de));
7283 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(1) |
7284 				WRITE_DATA_DST_SEL(8) |
7285 				WR_CONFIRM) |
7286 				WRITE_DATA_CACHE_POLICY(0));
7287 	amdgpu_ring_write(ring, lower_32_bits(de_payload_addr));
7288 	amdgpu_ring_write(ring, upper_32_bits(de_payload_addr));
7289 	amdgpu_ring_write_multiple(ring, (void *)&de_payload, cnt_de - 2);
7290 }
7291