1 /*
2  * Copyright 2019 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 #include "amdgpu.h"
30 #include "amdgpu_gfx.h"
31 #include "amdgpu_psp.h"
32 #include "amdgpu_smu.h"
33 #include "nv.h"
34 #include "nvd.h"
35 
36 #include "gc/gc_10_1_0_offset.h"
37 #include "gc/gc_10_1_0_sh_mask.h"
38 #include "navi10_enum.h"
39 #include "hdp/hdp_5_0_0_offset.h"
40 #include "ivsrcid/gfx/irqsrcs_gfx_10_1.h"
41 
42 #include "soc15.h"
43 #include "soc15_common.h"
44 #include "clearstate_gfx10.h"
45 #include "v10_structs.h"
46 #include "gfx_v10_0.h"
47 #include "nbio_v2_3.h"
48 
49 /**
50  * Navi10 has two graphic rings to share each graphic pipe.
51  * 1. Primary ring
52  * 2. Async ring
53  *
54  * In bring-up phase, it just used primary ring so set gfx ring number as 1 at
55  * first.
56  */
57 #define GFX10_NUM_GFX_RINGS	2
58 #define GFX10_MEC_HPD_SIZE	2048
59 
60 #define F32_CE_PROGRAM_RAM_SIZE		65536
61 #define RLCG_UCODE_LOADING_START_ADDRESS	0x00002000L
62 
63 #define mmCGTT_GS_NGG_CLK_CTRL	0x5087
64 #define mmCGTT_GS_NGG_CLK_CTRL_BASE_IDX	1
65 
66 MODULE_FIRMWARE("amdgpu/navi10_ce.bin");
67 MODULE_FIRMWARE("amdgpu/navi10_pfp.bin");
68 MODULE_FIRMWARE("amdgpu/navi10_me.bin");
69 MODULE_FIRMWARE("amdgpu/navi10_mec.bin");
70 MODULE_FIRMWARE("amdgpu/navi10_mec2.bin");
71 MODULE_FIRMWARE("amdgpu/navi10_rlc.bin");
72 
73 MODULE_FIRMWARE("amdgpu/navi14_ce_wks.bin");
74 MODULE_FIRMWARE("amdgpu/navi14_pfp_wks.bin");
75 MODULE_FIRMWARE("amdgpu/navi14_me_wks.bin");
76 MODULE_FIRMWARE("amdgpu/navi14_mec_wks.bin");
77 MODULE_FIRMWARE("amdgpu/navi14_mec2_wks.bin");
78 MODULE_FIRMWARE("amdgpu/navi14_ce.bin");
79 MODULE_FIRMWARE("amdgpu/navi14_pfp.bin");
80 MODULE_FIRMWARE("amdgpu/navi14_me.bin");
81 MODULE_FIRMWARE("amdgpu/navi14_mec.bin");
82 MODULE_FIRMWARE("amdgpu/navi14_mec2.bin");
83 MODULE_FIRMWARE("amdgpu/navi14_rlc.bin");
84 
85 MODULE_FIRMWARE("amdgpu/navi12_ce.bin");
86 MODULE_FIRMWARE("amdgpu/navi12_pfp.bin");
87 MODULE_FIRMWARE("amdgpu/navi12_me.bin");
88 MODULE_FIRMWARE("amdgpu/navi12_mec.bin");
89 MODULE_FIRMWARE("amdgpu/navi12_mec2.bin");
90 MODULE_FIRMWARE("amdgpu/navi12_rlc.bin");
91 
92 static const struct soc15_reg_golden golden_settings_gc_10_1[] =
93 {
94 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL_4, 0xffffffff, 0x00400014),
95 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCGTT_CPF_CLK_CTRL, 0xfcff8fff, 0xf8000100),
96 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCGTT_SPI_CLK_CTRL, 0xcd000000, 0x0d000100),
97 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCGTT_SQ_CLK_CTRL, 0x60000ff0, 0x60000100),
98 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCGTT_SQG_CLK_CTRL, 0x40000000, 0x40000100),
99 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCGTT_VGT_CLK_CTRL, 0xffff8fff, 0xffff8100),
100 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCGTT_WD_CLK_CTRL, 0xfeff8fff, 0xfeff8100),
101 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCH_PIPE_STEER, 0xffffffff, 0xe4e4e4e4),
102 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCH_VC5_ENABLE, 0x00000002, 0x00000000),
103 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_SD_CNTL, 0x000007ff, 0x000005ff),
104 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DEBUG, 0x20000000, 0x20000000),
105 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DEBUG2, 0xffffffff, 0x00000420),
106 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DEBUG3, 0x00000200, 0x00000200),
107 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DEBUG4, 0x07900000, 0x04900000),
108 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DFSM_TILES_IN_FLIGHT, 0x0000ffff, 0x0000003f),
109 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_LAST_OF_BURST_CONFIG, 0xffffffff, 0x03860204),
110 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGCR_GENERAL_CNTL, 0x1ff0ffff, 0x00000500),
111 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGE_PRIV_CONTROL, 0x000007ff, 0x000001fe),
112 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGL1_PIPE_STEER, 0xffffffff, 0xe4e4e4e4),
113 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGL2_PIPE_STEER_0, 0x77777777, 0x10321032),
114 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGL2_PIPE_STEER_1, 0x77777777, 0x02310231),
115 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGL2A_ADDR_MATCH_MASK, 0xffffffff, 0xffffffcf),
116 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGL2C_ADDR_MATCH_MASK, 0xffffffff, 0xffffffcf),
117 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGL2C_CGTT_SCLK_CTRL, 0x10000000, 0x10000100),
118 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGL2C_CTRL2, 0xffffffff, 0x1402002f),
119 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGL2C_CTRL3, 0xffff9fff, 0x00001188),
120 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE, 0x3fffffff, 0x08000009),
121 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE_1, 0x00400000, 0x04440000),
122 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000),
123 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmRMI_SPARE, 0xffffffff, 0xffff3101),
124 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmSQ_ALU_CLK_CTRL, 0xffffffff, 0xffffffff),
125 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmSQ_ARB_CONFIG, 0x00000100, 0x00000130),
126 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmSQ_LDS_CLK_CTRL, 0xffffffff, 0xffffffff),
127 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTA_CNTL_AUX, 0xfff7ffff, 0x01030000),
128 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CNTL, 0x60000010, 0x479c0010),
129 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmUTCL1_CGTT_CLK_CTRL, 0xfeff0fff, 0x40000100),
130 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmUTCL1_CTRL, 0x00c00000, 0x00c00000)
131 };
132 
133 static const struct soc15_reg_golden golden_settings_gc_10_0_nv10[] =
134 {
135 	/* Pending on emulation bring up */
136 };
137 
138 static const struct soc15_reg_golden golden_settings_gc_10_1_1[] =
139 {
140 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL_4, 0xffffffff, 0x003c0014),
141 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCGTT_GS_NGG_CLK_CTRL, 0xffff8fff, 0xffff8100),
142 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCGTT_IA_CLK_CTRL, 0xffff0fff, 0xffff0100),
143 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCGTT_SPI_CLK_CTRL, 0xcd000000, 0x0d000100),
144 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCGTT_SQ_CLK_CTRL, 0xf8ff0fff, 0x60000100),
145 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCGTT_SQG_CLK_CTRL, 0x40000ff0, 0x40000100),
146 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCGTT_VGT_CLK_CTRL, 0xffff8fff, 0xffff8100),
147 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCGTT_WD_CLK_CTRL, 0xffff8fff, 0xffff8100),
148 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCH_PIPE_STEER, 0xffffffff, 0xe4e4e4e4),
149 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCH_VC5_ENABLE, 0x00000002, 0x00000000),
150 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_SD_CNTL, 0x800007ff, 0x000005ff),
151 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DEBUG, 0xffffffff, 0x20000000),
152 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DEBUG2, 0xffffffff, 0x00000420),
153 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DEBUG3, 0x00000200, 0x00000200),
154 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DEBUG4, 0xffffffff, 0x04900000),
155 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DFSM_TILES_IN_FLIGHT, 0x0000ffff, 0x0000003f),
156 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_LAST_OF_BURST_CONFIG, 0xffffffff, 0x03860204),
157 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGCR_GENERAL_CNTL, 0x1ff0ffff, 0x00000500),
158 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGE_PRIV_CONTROL, 0x000007ff, 0x000001fe),
159 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGL1_PIPE_STEER, 0xffffffff, 0xe4e4e4e4),
160 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGL2A_ADDR_MATCH_MASK, 0xffffffff, 0xffffffe7),
161 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGL2C_ADDR_MATCH_MASK, 0xffffffff, 0xffffffe7),
162 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGL2C_CGTT_SCLK_CTRL, 0xffff0fff, 0x10000100),
163 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGL2C_CTRL2, 0xffffffff, 0x1402002f),
164 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGL2C_CTRL3, 0xffffbfff, 0x00000188),
165 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE, 0x3fffffff, 0x08000009),
166 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE_1, 0x00400000, 0x04440000),
167 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000),
168 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmRMI_SPARE, 0xffffffff, 0xffff3101),
169 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmSQ_ALU_CLK_CTRL, 0xffffffff, 0xffffffff),
170 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmSQ_ARB_CONFIG, 0x00000133, 0x00000130),
171 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmSQ_LDS_CLK_CTRL, 0xffffffff, 0xffffffff),
172 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTA_CNTL_AUX, 0xfff7ffff, 0x01030000),
173 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CNTL, 0x60000010, 0x479c0010),
174 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmUTCL1_CTRL, 0x00c00000, 0x00c00000),
175 };
176 
177 static const struct soc15_reg_golden golden_settings_gc_10_1_2[] =
178 {
179 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCB_HW_CONTROL_4, 0x003e001f, 0x003c0014),
180 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCGTT_GS_NGG_CLK_CTRL, 0xffff8fff, 0xffff8100),
181 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCGTT_IA_CLK_CTRL, 0xffff0fff, 0xffff0100),
182 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCGTT_SPI_CLK_CTRL, 0xff7f0fff, 0x0d000100),
183 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCGTT_SQ_CLK_CTRL, 0xffffcfff, 0x60000100),
184 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCGTT_SQG_CLK_CTRL, 0xffff0fff, 0x40000100),
185 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCGTT_VGT_CLK_CTRL, 0xffff8fff, 0xffff8100),
186 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCGTT_WD_CLK_CTRL, 0xffff8fff, 0xffff8100),
187 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCH_PIPE_STEER, 0xffffffff, 0xe4e4e4e4),
188 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCH_VC5_ENABLE, 0x00000003, 0x00000000),
189 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmCP_SD_CNTL, 0x800007ff, 0x000005ff),
190 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DEBUG, 0xffffffff, 0x20000000),
191 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DEBUG2, 0xffffffff, 0x00000420),
192 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DEBUG3, 0xffffffff, 0x00000200),
193 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DEBUG4, 0xffffffff, 0x04800000),
194 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_DFSM_TILES_IN_FLIGHT, 0x0000ffff, 0x0000003f),
195 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmDB_LAST_OF_BURST_CONFIG, 0xffffffff, 0x03860204),
196 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGCR_GENERAL_CNTL, 0x1ff0ffff, 0x00000500),
197 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGE_PRIV_CONTROL, 0x00007fff, 0x000001fe),
198 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGL1_PIPE_STEER, 0xffffffff, 0xe4e4e4e4),
199 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGL2_PIPE_STEER_0, 0x77777777, 0x10321032),
200 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGL2_PIPE_STEER_1, 0x77777777, 0x02310231),
201 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGL2A_ADDR_MATCH_MASK, 0xffffffff, 0xffffffcf),
202 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGL2C_ADDR_MATCH_MASK, 0xffffffff, 0xffffffcf),
203 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGL2C_CGTT_SCLK_CTRL, 0xffff0fff, 0x10000100),
204 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGL2C_CTRL2, 0xffffffff, 0x1402002f),
205 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmGL2C_CTRL3, 0xffffbfff, 0x00000188),
206 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_BINNER_EVENT_CNTL_0, 0xffffffff, 0x842a4c02),
207 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_BINNER_TIMEOUT_COUNTER, 0xffffffff, 0x00000800),
208 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE, 0x3fffffff, 0x08000009),
209 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE_1, 0xffffffff, 0x04440000),
210 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_ENHANCE_2, 0x00000820, 0x00000820),
211 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmPA_SC_LINE_STIPPLE_STATE, 0x0000ff0f, 0x00000000),
212 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmRMI_SPARE, 0xffffffff, 0xffff3101),
213 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmSQ_ALU_CLK_CTRL, 0xffffffff, 0xffffffff),
214 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmSQ_ARB_CONFIG, 0x00000133, 0x00000130),
215 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmSQ_LDS_CLK_CTRL, 0xffffffff, 0xffffffff),
216 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTA_CNTL_AUX, 0xfff7ffff, 0x01030000),
217 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmTCP_CNTL, 0xffdf80ff, 0x479c0010),
218 	SOC15_REG_GOLDEN_VALUE(GC, 0, mmUTCL1_CTRL, 0xffffffff, 0x00800000)
219 };
220 
221 static const struct soc15_reg_golden golden_settings_gc_10_1_nv14[] =
222 {
223 	/* Pending on emulation bring up */
224 };
225 
226 static const struct soc15_reg_golden golden_settings_gc_10_1_2_nv12[] =
227 {
228 	/* Pending on emulation bring up */
229 };
230 
231 #define DEFAULT_SH_MEM_CONFIG \
232 	((SH_MEM_ADDRESS_MODE_64 << SH_MEM_CONFIG__ADDRESS_MODE__SHIFT) | \
233 	 (SH_MEM_ALIGNMENT_MODE_UNALIGNED << SH_MEM_CONFIG__ALIGNMENT_MODE__SHIFT) | \
234 	 (SH_MEM_RETRY_MODE_ALL << SH_MEM_CONFIG__RETRY_MODE__SHIFT) | \
235 	 (3 << SH_MEM_CONFIG__INITIAL_INST_PREFETCH__SHIFT))
236 
237 
238 static void gfx_v10_0_set_ring_funcs(struct amdgpu_device *adev);
239 static void gfx_v10_0_set_irq_funcs(struct amdgpu_device *adev);
240 static void gfx_v10_0_set_gds_init(struct amdgpu_device *adev);
241 static void gfx_v10_0_set_rlc_funcs(struct amdgpu_device *adev);
242 static int gfx_v10_0_get_cu_info(struct amdgpu_device *adev,
243                                  struct amdgpu_cu_info *cu_info);
244 static uint64_t gfx_v10_0_get_gpu_clock_counter(struct amdgpu_device *adev);
245 static void gfx_v10_0_select_se_sh(struct amdgpu_device *adev, u32 se_num,
246 				   u32 sh_num, u32 instance);
247 static u32 gfx_v10_0_get_wgp_active_bitmap_per_sh(struct amdgpu_device *adev);
248 
249 static int gfx_v10_0_rlc_backdoor_autoload_buffer_init(struct amdgpu_device *adev);
250 static void gfx_v10_0_rlc_backdoor_autoload_buffer_fini(struct amdgpu_device *adev);
251 static int gfx_v10_0_rlc_backdoor_autoload_enable(struct amdgpu_device *adev);
252 static int gfx_v10_0_wait_for_rlc_autoload_complete(struct amdgpu_device *adev);
253 static void gfx_v10_0_ring_emit_ce_meta(struct amdgpu_ring *ring, bool resume);
254 static void gfx_v10_0_ring_emit_de_meta(struct amdgpu_ring *ring, bool resume);
255 static void gfx_v10_0_ring_emit_tmz(struct amdgpu_ring *ring, bool start);
256 
257 static void gfx10_kiq_set_resources(struct amdgpu_ring *kiq_ring, uint64_t queue_mask)
258 {
259 	amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_SET_RESOURCES, 6));
260 	amdgpu_ring_write(kiq_ring, PACKET3_SET_RESOURCES_VMID_MASK(0) |
261 			  PACKET3_SET_RESOURCES_QUEUE_TYPE(0));	/* vmid_mask:0 queue_type:0 (KIQ) */
262 	amdgpu_ring_write(kiq_ring, lower_32_bits(queue_mask));	/* queue mask lo */
263 	amdgpu_ring_write(kiq_ring, upper_32_bits(queue_mask));	/* queue mask hi */
264 	amdgpu_ring_write(kiq_ring, 0);	/* gws mask lo */
265 	amdgpu_ring_write(kiq_ring, 0);	/* gws mask hi */
266 	amdgpu_ring_write(kiq_ring, 0);	/* oac mask */
267 	amdgpu_ring_write(kiq_ring, 0);	/* gds heap base:0, gds heap size:0 */
268 }
269 
270 static void gfx10_kiq_map_queues(struct amdgpu_ring *kiq_ring,
271 				 struct amdgpu_ring *ring)
272 {
273 	struct amdgpu_device *adev = kiq_ring->adev;
274 	uint64_t mqd_addr = amdgpu_bo_gpu_offset(ring->mqd_obj);
275 	uint64_t wptr_addr = adev->wb.gpu_addr + (ring->wptr_offs * 4);
276 	uint32_t eng_sel = ring->funcs->type == AMDGPU_RING_TYPE_GFX ? 4 : 0;
277 
278 	amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_MAP_QUEUES, 5));
279 	/* Q_sel:0, vmid:0, vidmem: 1, engine:0, num_Q:1*/
280 	amdgpu_ring_write(kiq_ring, /* Q_sel: 0, vmid: 0, engine: 0, num_Q: 1 */
281 			  PACKET3_MAP_QUEUES_QUEUE_SEL(0) | /* Queue_Sel */
282 			  PACKET3_MAP_QUEUES_VMID(0) | /* VMID */
283 			  PACKET3_MAP_QUEUES_QUEUE(ring->queue) |
284 			  PACKET3_MAP_QUEUES_PIPE(ring->pipe) |
285 			  PACKET3_MAP_QUEUES_ME((ring->me == 1 ? 0 : 1)) |
286 			  PACKET3_MAP_QUEUES_QUEUE_TYPE(0) | /*queue_type: normal compute queue */
287 			  PACKET3_MAP_QUEUES_ALLOC_FORMAT(0) | /* alloc format: all_on_one_pipe */
288 			  PACKET3_MAP_QUEUES_ENGINE_SEL(eng_sel) |
289 			  PACKET3_MAP_QUEUES_NUM_QUEUES(1)); /* num_queues: must be 1 */
290 	amdgpu_ring_write(kiq_ring, PACKET3_MAP_QUEUES_DOORBELL_OFFSET(ring->doorbell_index));
291 	amdgpu_ring_write(kiq_ring, lower_32_bits(mqd_addr));
292 	amdgpu_ring_write(kiq_ring, upper_32_bits(mqd_addr));
293 	amdgpu_ring_write(kiq_ring, lower_32_bits(wptr_addr));
294 	amdgpu_ring_write(kiq_ring, upper_32_bits(wptr_addr));
295 }
296 
297 static void gfx10_kiq_unmap_queues(struct amdgpu_ring *kiq_ring,
298 				   struct amdgpu_ring *ring,
299 				   enum amdgpu_unmap_queues_action action,
300 				   u64 gpu_addr, u64 seq)
301 {
302 	uint32_t eng_sel = ring->funcs->type == AMDGPU_RING_TYPE_GFX ? 4 : 0;
303 
304 	amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_UNMAP_QUEUES, 4));
305 	amdgpu_ring_write(kiq_ring, /* Q_sel: 0, vmid: 0, engine: 0, num_Q: 1 */
306 			  PACKET3_UNMAP_QUEUES_ACTION(action) |
307 			  PACKET3_UNMAP_QUEUES_QUEUE_SEL(0) |
308 			  PACKET3_UNMAP_QUEUES_ENGINE_SEL(eng_sel) |
309 			  PACKET3_UNMAP_QUEUES_NUM_QUEUES(1));
310 	amdgpu_ring_write(kiq_ring,
311 		  PACKET3_UNMAP_QUEUES_DOORBELL_OFFSET0(ring->doorbell_index));
312 
313 	if (action == PREEMPT_QUEUES_NO_UNMAP) {
314 		amdgpu_ring_write(kiq_ring, lower_32_bits(gpu_addr));
315 		amdgpu_ring_write(kiq_ring, upper_32_bits(gpu_addr));
316 		amdgpu_ring_write(kiq_ring, seq);
317 	} else {
318 		amdgpu_ring_write(kiq_ring, 0);
319 		amdgpu_ring_write(kiq_ring, 0);
320 		amdgpu_ring_write(kiq_ring, 0);
321 	}
322 }
323 
324 static void gfx10_kiq_query_status(struct amdgpu_ring *kiq_ring,
325 				   struct amdgpu_ring *ring,
326 				   u64 addr,
327 				   u64 seq)
328 {
329 	uint32_t eng_sel = ring->funcs->type == AMDGPU_RING_TYPE_GFX ? 4 : 0;
330 
331 	amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_QUERY_STATUS, 5));
332 	amdgpu_ring_write(kiq_ring,
333 			  PACKET3_QUERY_STATUS_CONTEXT_ID(0) |
334 			  PACKET3_QUERY_STATUS_INTERRUPT_SEL(0) |
335 			  PACKET3_QUERY_STATUS_COMMAND(2));
336 	amdgpu_ring_write(kiq_ring, /* Q_sel: 0, vmid: 0, engine: 0, num_Q: 1 */
337 			  PACKET3_QUERY_STATUS_DOORBELL_OFFSET(ring->doorbell_index) |
338 			  PACKET3_QUERY_STATUS_ENG_SEL(eng_sel));
339 	amdgpu_ring_write(kiq_ring, lower_32_bits(addr));
340 	amdgpu_ring_write(kiq_ring, upper_32_bits(addr));
341 	amdgpu_ring_write(kiq_ring, lower_32_bits(seq));
342 	amdgpu_ring_write(kiq_ring, upper_32_bits(seq));
343 }
344 
345 static const struct kiq_pm4_funcs gfx_v10_0_kiq_pm4_funcs = {
346 	.kiq_set_resources = gfx10_kiq_set_resources,
347 	.kiq_map_queues = gfx10_kiq_map_queues,
348 	.kiq_unmap_queues = gfx10_kiq_unmap_queues,
349 	.kiq_query_status = gfx10_kiq_query_status,
350 	.set_resources_size = 8,
351 	.map_queues_size = 7,
352 	.unmap_queues_size = 6,
353 	.query_status_size = 7,
354 };
355 
356 static void gfx_v10_0_set_kiq_pm4_funcs(struct amdgpu_device *adev)
357 {
358 	adev->gfx.kiq.pmf = &gfx_v10_0_kiq_pm4_funcs;
359 }
360 
361 static void gfx_v10_0_init_golden_registers(struct amdgpu_device *adev)
362 {
363 	switch (adev->asic_type) {
364 	case CHIP_NAVI10:
365 		soc15_program_register_sequence(adev,
366 						golden_settings_gc_10_1,
367 						(const u32)ARRAY_SIZE(golden_settings_gc_10_1));
368 		soc15_program_register_sequence(adev,
369 						golden_settings_gc_10_0_nv10,
370 						(const u32)ARRAY_SIZE(golden_settings_gc_10_0_nv10));
371 		break;
372 	case CHIP_NAVI14:
373 		soc15_program_register_sequence(adev,
374 						golden_settings_gc_10_1_1,
375 						(const u32)ARRAY_SIZE(golden_settings_gc_10_1_1));
376 		soc15_program_register_sequence(adev,
377 						golden_settings_gc_10_1_nv14,
378 						(const u32)ARRAY_SIZE(golden_settings_gc_10_1_nv14));
379 		break;
380 	case CHIP_NAVI12:
381 		soc15_program_register_sequence(adev,
382 						golden_settings_gc_10_1_2,
383 						(const u32)ARRAY_SIZE(golden_settings_gc_10_1_2));
384 		soc15_program_register_sequence(adev,
385 						golden_settings_gc_10_1_2_nv12,
386 						(const u32)ARRAY_SIZE(golden_settings_gc_10_1_2_nv12));
387 		break;
388 	default:
389 		break;
390 	}
391 }
392 
393 static void gfx_v10_0_scratch_init(struct amdgpu_device *adev)
394 {
395 	adev->gfx.scratch.num_reg = 8;
396 	adev->gfx.scratch.reg_base = SOC15_REG_OFFSET(GC, 0, mmSCRATCH_REG0);
397 	adev->gfx.scratch.free_mask = (1u << adev->gfx.scratch.num_reg) - 1;
398 }
399 
400 static void gfx_v10_0_write_data_to_reg(struct amdgpu_ring *ring, int eng_sel,
401 				       bool wc, uint32_t reg, uint32_t val)
402 {
403 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
404 	amdgpu_ring_write(ring, WRITE_DATA_ENGINE_SEL(eng_sel) |
405 			  WRITE_DATA_DST_SEL(0) | (wc ? WR_CONFIRM : 0));
406 	amdgpu_ring_write(ring, reg);
407 	amdgpu_ring_write(ring, 0);
408 	amdgpu_ring_write(ring, val);
409 }
410 
411 static void gfx_v10_0_wait_reg_mem(struct amdgpu_ring *ring, int eng_sel,
412 				  int mem_space, int opt, uint32_t addr0,
413 				  uint32_t addr1, uint32_t ref, uint32_t mask,
414 				  uint32_t inv)
415 {
416 	amdgpu_ring_write(ring, PACKET3(PACKET3_WAIT_REG_MEM, 5));
417 	amdgpu_ring_write(ring,
418 			  /* memory (1) or register (0) */
419 			  (WAIT_REG_MEM_MEM_SPACE(mem_space) |
420 			   WAIT_REG_MEM_OPERATION(opt) | /* wait */
421 			   WAIT_REG_MEM_FUNCTION(3) |  /* equal */
422 			   WAIT_REG_MEM_ENGINE(eng_sel)));
423 
424 	if (mem_space)
425 		BUG_ON(addr0 & 0x3); /* Dword align */
426 	amdgpu_ring_write(ring, addr0);
427 	amdgpu_ring_write(ring, addr1);
428 	amdgpu_ring_write(ring, ref);
429 	amdgpu_ring_write(ring, mask);
430 	amdgpu_ring_write(ring, inv); /* poll interval */
431 }
432 
433 static int gfx_v10_0_ring_test_ring(struct amdgpu_ring *ring)
434 {
435 	struct amdgpu_device *adev = ring->adev;
436 	uint32_t scratch;
437 	uint32_t tmp = 0;
438 	unsigned i;
439 	int r;
440 
441 	r = amdgpu_gfx_scratch_get(adev, &scratch);
442 	if (r) {
443 		DRM_ERROR("amdgpu: cp failed to get scratch reg (%d).\n", r);
444 		return r;
445 	}
446 
447 	WREG32(scratch, 0xCAFEDEAD);
448 
449 	r = amdgpu_ring_alloc(ring, 3);
450 	if (r) {
451 		DRM_ERROR("amdgpu: cp failed to lock ring %d (%d).\n",
452 			  ring->idx, r);
453 		amdgpu_gfx_scratch_free(adev, scratch);
454 		return r;
455 	}
456 
457 	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_UCONFIG_REG, 1));
458 	amdgpu_ring_write(ring, (scratch - PACKET3_SET_UCONFIG_REG_START));
459 	amdgpu_ring_write(ring, 0xDEADBEEF);
460 	amdgpu_ring_commit(ring);
461 
462 	for (i = 0; i < adev->usec_timeout; i++) {
463 		tmp = RREG32(scratch);
464 		if (tmp == 0xDEADBEEF)
465 			break;
466 		if (amdgpu_emu_mode == 1)
467 			msleep(1);
468 		else
469 			udelay(1);
470 	}
471 	if (i < adev->usec_timeout) {
472 		if (amdgpu_emu_mode == 1)
473 			DRM_INFO("ring test on %d succeeded in %d msecs\n",
474 				 ring->idx, i);
475 		else
476 			DRM_INFO("ring test on %d succeeded in %d usecs\n",
477 				 ring->idx, i);
478 	} else {
479 		DRM_ERROR("amdgpu: ring %d test failed (scratch(0x%04X)=0x%08X)\n",
480 			  ring->idx, scratch, tmp);
481 		r = -EINVAL;
482 	}
483 	amdgpu_gfx_scratch_free(adev, scratch);
484 
485 	return r;
486 }
487 
488 static int gfx_v10_0_ring_test_ib(struct amdgpu_ring *ring, long timeout)
489 {
490 	struct amdgpu_device *adev = ring->adev;
491 	struct amdgpu_ib ib;
492 	struct dma_fence *f = NULL;
493 	uint32_t scratch;
494 	uint32_t tmp = 0;
495 	long r;
496 
497 	r = amdgpu_gfx_scratch_get(adev, &scratch);
498 	if (r) {
499 		DRM_ERROR("amdgpu: failed to get scratch reg (%ld).\n", r);
500 		return r;
501 	}
502 
503 	WREG32(scratch, 0xCAFEDEAD);
504 
505 	memset(&ib, 0, sizeof(ib));
506 	r = amdgpu_ib_get(adev, NULL, 256, &ib);
507 	if (r) {
508 		DRM_ERROR("amdgpu: failed to get ib (%ld).\n", r);
509 		goto err1;
510 	}
511 
512 	ib.ptr[0] = PACKET3(PACKET3_SET_UCONFIG_REG, 1);
513 	ib.ptr[1] = ((scratch - PACKET3_SET_UCONFIG_REG_START));
514 	ib.ptr[2] = 0xDEADBEEF;
515 	ib.length_dw = 3;
516 
517 	r = amdgpu_ib_schedule(ring, 1, &ib, NULL, &f);
518 	if (r)
519 		goto err2;
520 
521 	r = dma_fence_wait_timeout(f, false, timeout);
522 	if (r == 0) {
523 		DRM_ERROR("amdgpu: IB test timed out.\n");
524 		r = -ETIMEDOUT;
525 		goto err2;
526 	} else if (r < 0) {
527 		DRM_ERROR("amdgpu: fence wait failed (%ld).\n", r);
528 		goto err2;
529 	}
530 
531 	tmp = RREG32(scratch);
532 	if (tmp == 0xDEADBEEF) {
533 		DRM_INFO("ib test on ring %d succeeded\n", ring->idx);
534 		r = 0;
535 	} else {
536 		DRM_ERROR("amdgpu: ib test failed (scratch(0x%04X)=0x%08X)\n",
537 			  scratch, tmp);
538 		r = -EINVAL;
539 	}
540 err2:
541 	amdgpu_ib_free(adev, &ib, NULL);
542 	dma_fence_put(f);
543 err1:
544 	amdgpu_gfx_scratch_free(adev, scratch);
545 
546 	return r;
547 }
548 
549 static void gfx_v10_0_free_microcode(struct amdgpu_device *adev)
550 {
551 	release_firmware(adev->gfx.pfp_fw);
552 	adev->gfx.pfp_fw = NULL;
553 	release_firmware(adev->gfx.me_fw);
554 	adev->gfx.me_fw = NULL;
555 	release_firmware(adev->gfx.ce_fw);
556 	adev->gfx.ce_fw = NULL;
557 	release_firmware(adev->gfx.rlc_fw);
558 	adev->gfx.rlc_fw = NULL;
559 	release_firmware(adev->gfx.mec_fw);
560 	adev->gfx.mec_fw = NULL;
561 	release_firmware(adev->gfx.mec2_fw);
562 	adev->gfx.mec2_fw = NULL;
563 
564 	kfree(adev->gfx.rlc.register_list_format);
565 }
566 
567 static void gfx_v10_0_check_fw_write_wait(struct amdgpu_device *adev)
568 {
569 	adev->gfx.cp_fw_write_wait = false;
570 
571 	switch (adev->asic_type) {
572 	case CHIP_NAVI10:
573 	case CHIP_NAVI12:
574 	case CHIP_NAVI14:
575 		if ((adev->gfx.me_fw_version >= 0x00000046) &&
576 		    (adev->gfx.me_feature_version >= 27) &&
577 		    (adev->gfx.pfp_fw_version >= 0x00000068) &&
578 		    (adev->gfx.pfp_feature_version >= 27) &&
579 		    (adev->gfx.mec_fw_version >= 0x0000005b) &&
580 		    (adev->gfx.mec_feature_version >= 27))
581 			adev->gfx.cp_fw_write_wait = true;
582 		break;
583 	default:
584 		break;
585 	}
586 
587 	if (adev->gfx.cp_fw_write_wait == false)
588 		DRM_WARN_ONCE("Warning: check cp_fw_version and update it to realize \
589 			      GRBM requires 1-cycle delay in cp firmware\n");
590 }
591 
592 
593 static void gfx_v10_0_init_rlc_ext_microcode(struct amdgpu_device *adev)
594 {
595 	const struct rlc_firmware_header_v2_1 *rlc_hdr;
596 
597 	rlc_hdr = (const struct rlc_firmware_header_v2_1 *)adev->gfx.rlc_fw->data;
598 	adev->gfx.rlc_srlc_fw_version = le32_to_cpu(rlc_hdr->save_restore_list_cntl_ucode_ver);
599 	adev->gfx.rlc_srlc_feature_version = le32_to_cpu(rlc_hdr->save_restore_list_cntl_feature_ver);
600 	adev->gfx.rlc.save_restore_list_cntl_size_bytes = le32_to_cpu(rlc_hdr->save_restore_list_cntl_size_bytes);
601 	adev->gfx.rlc.save_restore_list_cntl = (u8 *)rlc_hdr + le32_to_cpu(rlc_hdr->save_restore_list_cntl_offset_bytes);
602 	adev->gfx.rlc_srlg_fw_version = le32_to_cpu(rlc_hdr->save_restore_list_gpm_ucode_ver);
603 	adev->gfx.rlc_srlg_feature_version = le32_to_cpu(rlc_hdr->save_restore_list_gpm_feature_ver);
604 	adev->gfx.rlc.save_restore_list_gpm_size_bytes = le32_to_cpu(rlc_hdr->save_restore_list_gpm_size_bytes);
605 	adev->gfx.rlc.save_restore_list_gpm = (u8 *)rlc_hdr + le32_to_cpu(rlc_hdr->save_restore_list_gpm_offset_bytes);
606 	adev->gfx.rlc_srls_fw_version = le32_to_cpu(rlc_hdr->save_restore_list_srm_ucode_ver);
607 	adev->gfx.rlc_srls_feature_version = le32_to_cpu(rlc_hdr->save_restore_list_srm_feature_ver);
608 	adev->gfx.rlc.save_restore_list_srm_size_bytes = le32_to_cpu(rlc_hdr->save_restore_list_srm_size_bytes);
609 	adev->gfx.rlc.save_restore_list_srm = (u8 *)rlc_hdr + le32_to_cpu(rlc_hdr->save_restore_list_srm_offset_bytes);
610 	adev->gfx.rlc.reg_list_format_direct_reg_list_length =
611 			le32_to_cpu(rlc_hdr->reg_list_format_direct_reg_list_length);
612 }
613 
614 static void gfx_v10_0_check_gfxoff_flag(struct amdgpu_device *adev)
615 {
616 	switch (adev->asic_type) {
617 	case CHIP_NAVI10:
618 		adev->pm.pp_feature &= ~PP_GFXOFF_MASK;
619 		break;
620 	default:
621 		break;
622 	}
623 }
624 
625 static int gfx_v10_0_init_microcode(struct amdgpu_device *adev)
626 {
627 	const char *chip_name;
628 	char fw_name[40];
629 	char wks[10];
630 	int err;
631 	struct amdgpu_firmware_info *info = NULL;
632 	const struct common_firmware_header *header = NULL;
633 	const struct gfx_firmware_header_v1_0 *cp_hdr;
634 	const struct rlc_firmware_header_v2_0 *rlc_hdr;
635 	unsigned int *tmp = NULL;
636 	unsigned int i = 0;
637 	uint16_t version_major;
638 	uint16_t version_minor;
639 
640 	DRM_DEBUG("\n");
641 
642 	memset(wks, 0, sizeof(wks));
643 	switch (adev->asic_type) {
644 	case CHIP_NAVI10:
645 		chip_name = "navi10";
646 		break;
647 	case CHIP_NAVI14:
648 		chip_name = "navi14";
649 		if (!(adev->pdev->device == 0x7340 &&
650 		      adev->pdev->revision != 0x00))
651 			snprintf(wks, sizeof(wks), "_wks");
652 		break;
653 	case CHIP_NAVI12:
654 		chip_name = "navi12";
655 		break;
656 	default:
657 		BUG();
658 	}
659 
660 	snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_pfp%s.bin", chip_name, wks);
661 	err = request_firmware(&adev->gfx.pfp_fw, fw_name, adev->dev);
662 	if (err)
663 		goto out;
664 	err = amdgpu_ucode_validate(adev->gfx.pfp_fw);
665 	if (err)
666 		goto out;
667 	cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.pfp_fw->data;
668 	adev->gfx.pfp_fw_version = le32_to_cpu(cp_hdr->header.ucode_version);
669 	adev->gfx.pfp_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version);
670 
671 	snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_me%s.bin", chip_name, wks);
672 	err = request_firmware(&adev->gfx.me_fw, fw_name, adev->dev);
673 	if (err)
674 		goto out;
675 	err = amdgpu_ucode_validate(adev->gfx.me_fw);
676 	if (err)
677 		goto out;
678 	cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.me_fw->data;
679 	adev->gfx.me_fw_version = le32_to_cpu(cp_hdr->header.ucode_version);
680 	adev->gfx.me_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version);
681 
682 	snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_ce%s.bin", chip_name, wks);
683 	err = request_firmware(&adev->gfx.ce_fw, fw_name, adev->dev);
684 	if (err)
685 		goto out;
686 	err = amdgpu_ucode_validate(adev->gfx.ce_fw);
687 	if (err)
688 		goto out;
689 	cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.ce_fw->data;
690 	adev->gfx.ce_fw_version = le32_to_cpu(cp_hdr->header.ucode_version);
691 	adev->gfx.ce_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version);
692 
693 	if (!amdgpu_sriov_vf(adev)) {
694 		snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_rlc.bin", chip_name);
695 		err = request_firmware(&adev->gfx.rlc_fw, fw_name, adev->dev);
696 		if (err)
697 			goto out;
698 		err = amdgpu_ucode_validate(adev->gfx.rlc_fw);
699 		rlc_hdr = (const struct rlc_firmware_header_v2_0 *)adev->gfx.rlc_fw->data;
700 		version_major = le16_to_cpu(rlc_hdr->header.header_version_major);
701 		version_minor = le16_to_cpu(rlc_hdr->header.header_version_minor);
702 		if (version_major == 2 && version_minor == 1)
703 			adev->gfx.rlc.is_rlc_v2_1 = true;
704 
705 		adev->gfx.rlc_fw_version = le32_to_cpu(rlc_hdr->header.ucode_version);
706 		adev->gfx.rlc_feature_version = le32_to_cpu(rlc_hdr->ucode_feature_version);
707 		adev->gfx.rlc.save_and_restore_offset =
708 			le32_to_cpu(rlc_hdr->save_and_restore_offset);
709 		adev->gfx.rlc.clear_state_descriptor_offset =
710 			le32_to_cpu(rlc_hdr->clear_state_descriptor_offset);
711 		adev->gfx.rlc.avail_scratch_ram_locations =
712 			le32_to_cpu(rlc_hdr->avail_scratch_ram_locations);
713 		adev->gfx.rlc.reg_restore_list_size =
714 			le32_to_cpu(rlc_hdr->reg_restore_list_size);
715 		adev->gfx.rlc.reg_list_format_start =
716 			le32_to_cpu(rlc_hdr->reg_list_format_start);
717 		adev->gfx.rlc.reg_list_format_separate_start =
718 			le32_to_cpu(rlc_hdr->reg_list_format_separate_start);
719 		adev->gfx.rlc.starting_offsets_start =
720 			le32_to_cpu(rlc_hdr->starting_offsets_start);
721 		adev->gfx.rlc.reg_list_format_size_bytes =
722 			le32_to_cpu(rlc_hdr->reg_list_format_size_bytes);
723 		adev->gfx.rlc.reg_list_size_bytes =
724 			le32_to_cpu(rlc_hdr->reg_list_size_bytes);
725 		adev->gfx.rlc.register_list_format =
726 			kmalloc(adev->gfx.rlc.reg_list_format_size_bytes +
727 					adev->gfx.rlc.reg_list_size_bytes, GFP_KERNEL);
728 		if (!adev->gfx.rlc.register_list_format) {
729 			err = -ENOMEM;
730 			goto out;
731 		}
732 
733 		tmp = (unsigned int *)((uintptr_t)rlc_hdr +
734 							   le32_to_cpu(rlc_hdr->reg_list_format_array_offset_bytes));
735 		for (i = 0 ; i < (rlc_hdr->reg_list_format_size_bytes >> 2); i++)
736 			adev->gfx.rlc.register_list_format[i] =	le32_to_cpu(tmp[i]);
737 
738 		adev->gfx.rlc.register_restore = adev->gfx.rlc.register_list_format + i;
739 
740 		tmp = (unsigned int *)((uintptr_t)rlc_hdr +
741 							   le32_to_cpu(rlc_hdr->reg_list_array_offset_bytes));
742 		for (i = 0 ; i < (rlc_hdr->reg_list_size_bytes >> 2); i++)
743 			adev->gfx.rlc.register_restore[i] = le32_to_cpu(tmp[i]);
744 
745 		if (adev->gfx.rlc.is_rlc_v2_1)
746 			gfx_v10_0_init_rlc_ext_microcode(adev);
747 	}
748 
749 	snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec%s.bin", chip_name, wks);
750 	err = request_firmware(&adev->gfx.mec_fw, fw_name, adev->dev);
751 	if (err)
752 		goto out;
753 	err = amdgpu_ucode_validate(adev->gfx.mec_fw);
754 	if (err)
755 		goto out;
756 	cp_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data;
757 	adev->gfx.mec_fw_version = le32_to_cpu(cp_hdr->header.ucode_version);
758 	adev->gfx.mec_feature_version = le32_to_cpu(cp_hdr->ucode_feature_version);
759 
760 	snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_mec2%s.bin", chip_name, wks);
761 	err = request_firmware(&adev->gfx.mec2_fw, fw_name, adev->dev);
762 	if (!err) {
763 		err = amdgpu_ucode_validate(adev->gfx.mec2_fw);
764 		if (err)
765 			goto out;
766 		cp_hdr = (const struct gfx_firmware_header_v1_0 *)
767 		adev->gfx.mec2_fw->data;
768 		adev->gfx.mec2_fw_version =
769 		le32_to_cpu(cp_hdr->header.ucode_version);
770 		adev->gfx.mec2_feature_version =
771 		le32_to_cpu(cp_hdr->ucode_feature_version);
772 	} else {
773 		err = 0;
774 		adev->gfx.mec2_fw = NULL;
775 	}
776 
777 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
778 		info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_PFP];
779 		info->ucode_id = AMDGPU_UCODE_ID_CP_PFP;
780 		info->fw = adev->gfx.pfp_fw;
781 		header = (const struct common_firmware_header *)info->fw->data;
782 		adev->firmware.fw_size +=
783 			ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
784 
785 		info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_ME];
786 		info->ucode_id = AMDGPU_UCODE_ID_CP_ME;
787 		info->fw = adev->gfx.me_fw;
788 		header = (const struct common_firmware_header *)info->fw->data;
789 		adev->firmware.fw_size +=
790 			ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
791 
792 		info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_CE];
793 		info->ucode_id = AMDGPU_UCODE_ID_CP_CE;
794 		info->fw = adev->gfx.ce_fw;
795 		header = (const struct common_firmware_header *)info->fw->data;
796 		adev->firmware.fw_size +=
797 			ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
798 
799 		info = &adev->firmware.ucode[AMDGPU_UCODE_ID_RLC_G];
800 		info->ucode_id = AMDGPU_UCODE_ID_RLC_G;
801 		info->fw = adev->gfx.rlc_fw;
802 		header = (const struct common_firmware_header *)info->fw->data;
803 		adev->firmware.fw_size +=
804 			ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
805 
806 		if (adev->gfx.rlc.is_rlc_v2_1 &&
807 		    adev->gfx.rlc.save_restore_list_cntl_size_bytes &&
808 		    adev->gfx.rlc.save_restore_list_gpm_size_bytes &&
809 		    adev->gfx.rlc.save_restore_list_srm_size_bytes) {
810 			info = &adev->firmware.ucode[AMDGPU_UCODE_ID_RLC_RESTORE_LIST_CNTL];
811 			info->ucode_id = AMDGPU_UCODE_ID_RLC_RESTORE_LIST_CNTL;
812 			info->fw = adev->gfx.rlc_fw;
813 			adev->firmware.fw_size +=
814 				ALIGN(adev->gfx.rlc.save_restore_list_cntl_size_bytes, PAGE_SIZE);
815 
816 			info = &adev->firmware.ucode[AMDGPU_UCODE_ID_RLC_RESTORE_LIST_GPM_MEM];
817 			info->ucode_id = AMDGPU_UCODE_ID_RLC_RESTORE_LIST_GPM_MEM;
818 			info->fw = adev->gfx.rlc_fw;
819 			adev->firmware.fw_size +=
820 				ALIGN(adev->gfx.rlc.save_restore_list_gpm_size_bytes, PAGE_SIZE);
821 
822 			info = &adev->firmware.ucode[AMDGPU_UCODE_ID_RLC_RESTORE_LIST_SRM_MEM];
823 			info->ucode_id = AMDGPU_UCODE_ID_RLC_RESTORE_LIST_SRM_MEM;
824 			info->fw = adev->gfx.rlc_fw;
825 			adev->firmware.fw_size +=
826 				ALIGN(adev->gfx.rlc.save_restore_list_srm_size_bytes, PAGE_SIZE);
827 		}
828 
829 		info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_MEC1];
830 		info->ucode_id = AMDGPU_UCODE_ID_CP_MEC1;
831 		info->fw = adev->gfx.mec_fw;
832 		header = (const struct common_firmware_header *)info->fw->data;
833 		cp_hdr = (const struct gfx_firmware_header_v1_0 *)info->fw->data;
834 		adev->firmware.fw_size +=
835 			ALIGN(le32_to_cpu(header->ucode_size_bytes) -
836 			      le32_to_cpu(cp_hdr->jt_size) * 4, PAGE_SIZE);
837 
838 		info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_MEC1_JT];
839 		info->ucode_id = AMDGPU_UCODE_ID_CP_MEC1_JT;
840 		info->fw = adev->gfx.mec_fw;
841 		adev->firmware.fw_size +=
842 			ALIGN(le32_to_cpu(cp_hdr->jt_size) * 4, PAGE_SIZE);
843 
844 		if (adev->gfx.mec2_fw) {
845 			info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_MEC2];
846 			info->ucode_id = AMDGPU_UCODE_ID_CP_MEC2;
847 			info->fw = adev->gfx.mec2_fw;
848 			header = (const struct common_firmware_header *)info->fw->data;
849 			cp_hdr = (const struct gfx_firmware_header_v1_0 *)info->fw->data;
850 			adev->firmware.fw_size +=
851 				ALIGN(le32_to_cpu(header->ucode_size_bytes) -
852 				      le32_to_cpu(cp_hdr->jt_size) * 4,
853 				      PAGE_SIZE);
854 			info = &adev->firmware.ucode[AMDGPU_UCODE_ID_CP_MEC2_JT];
855 			info->ucode_id = AMDGPU_UCODE_ID_CP_MEC2_JT;
856 			info->fw = adev->gfx.mec2_fw;
857 			adev->firmware.fw_size +=
858 				ALIGN(le32_to_cpu(cp_hdr->jt_size) * 4,
859 				      PAGE_SIZE);
860 		}
861 	}
862 
863 	gfx_v10_0_check_fw_write_wait(adev);
864 out:
865 	if (err) {
866 		dev_err(adev->dev,
867 			"gfx10: Failed to load firmware \"%s\"\n",
868 			fw_name);
869 		release_firmware(adev->gfx.pfp_fw);
870 		adev->gfx.pfp_fw = NULL;
871 		release_firmware(adev->gfx.me_fw);
872 		adev->gfx.me_fw = NULL;
873 		release_firmware(adev->gfx.ce_fw);
874 		adev->gfx.ce_fw = NULL;
875 		release_firmware(adev->gfx.rlc_fw);
876 		adev->gfx.rlc_fw = NULL;
877 		release_firmware(adev->gfx.mec_fw);
878 		adev->gfx.mec_fw = NULL;
879 		release_firmware(adev->gfx.mec2_fw);
880 		adev->gfx.mec2_fw = NULL;
881 	}
882 
883 	gfx_v10_0_check_gfxoff_flag(adev);
884 
885 	return err;
886 }
887 
888 static u32 gfx_v10_0_get_csb_size(struct amdgpu_device *adev)
889 {
890 	u32 count = 0;
891 	const struct cs_section_def *sect = NULL;
892 	const struct cs_extent_def *ext = NULL;
893 
894 	/* begin clear state */
895 	count += 2;
896 	/* context control state */
897 	count += 3;
898 
899 	for (sect = gfx10_cs_data; sect->section != NULL; ++sect) {
900 		for (ext = sect->section; ext->extent != NULL; ++ext) {
901 			if (sect->id == SECT_CONTEXT)
902 				count += 2 + ext->reg_count;
903 			else
904 				return 0;
905 		}
906 	}
907 
908 	/* set PA_SC_TILE_STEERING_OVERRIDE */
909 	count += 3;
910 	/* end clear state */
911 	count += 2;
912 	/* clear state */
913 	count += 2;
914 
915 	return count;
916 }
917 
918 static void gfx_v10_0_get_csb_buffer(struct amdgpu_device *adev,
919 				    volatile u32 *buffer)
920 {
921 	u32 count = 0, i;
922 	const struct cs_section_def *sect = NULL;
923 	const struct cs_extent_def *ext = NULL;
924 	int ctx_reg_offset;
925 
926 	if (adev->gfx.rlc.cs_data == NULL)
927 		return;
928 	if (buffer == NULL)
929 		return;
930 
931 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0));
932 	buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
933 
934 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CONTEXT_CONTROL, 1));
935 	buffer[count++] = cpu_to_le32(0x80000000);
936 	buffer[count++] = cpu_to_le32(0x80000000);
937 
938 	for (sect = adev->gfx.rlc.cs_data; sect->section != NULL; ++sect) {
939 		for (ext = sect->section; ext->extent != NULL; ++ext) {
940 			if (sect->id == SECT_CONTEXT) {
941 				buffer[count++] =
942 					cpu_to_le32(PACKET3(PACKET3_SET_CONTEXT_REG, ext->reg_count));
943 				buffer[count++] = cpu_to_le32(ext->reg_index -
944 						PACKET3_SET_CONTEXT_REG_START);
945 				for (i = 0; i < ext->reg_count; i++)
946 					buffer[count++] = cpu_to_le32(ext->extent[i]);
947 			} else {
948 				return;
949 			}
950 		}
951 	}
952 
953 	ctx_reg_offset =
954 		SOC15_REG_OFFSET(GC, 0, mmPA_SC_TILE_STEERING_OVERRIDE) - PACKET3_SET_CONTEXT_REG_START;
955 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_SET_CONTEXT_REG, 1));
956 	buffer[count++] = cpu_to_le32(ctx_reg_offset);
957 	buffer[count++] = cpu_to_le32(adev->gfx.config.pa_sc_tile_steering_override);
958 
959 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_PREAMBLE_CNTL, 0));
960 	buffer[count++] = cpu_to_le32(PACKET3_PREAMBLE_END_CLEAR_STATE);
961 
962 	buffer[count++] = cpu_to_le32(PACKET3(PACKET3_CLEAR_STATE, 0));
963 	buffer[count++] = cpu_to_le32(0);
964 }
965 
966 static void gfx_v10_0_rlc_fini(struct amdgpu_device *adev)
967 {
968 	/* clear state block */
969 	amdgpu_bo_free_kernel(&adev->gfx.rlc.clear_state_obj,
970 			&adev->gfx.rlc.clear_state_gpu_addr,
971 			(void **)&adev->gfx.rlc.cs_ptr);
972 
973 	/* jump table block */
974 	amdgpu_bo_free_kernel(&adev->gfx.rlc.cp_table_obj,
975 			&adev->gfx.rlc.cp_table_gpu_addr,
976 			(void **)&adev->gfx.rlc.cp_table_ptr);
977 }
978 
979 static int gfx_v10_0_rlc_init(struct amdgpu_device *adev)
980 {
981 	const struct cs_section_def *cs_data;
982 	int r;
983 
984 	adev->gfx.rlc.cs_data = gfx10_cs_data;
985 
986 	cs_data = adev->gfx.rlc.cs_data;
987 
988 	if (cs_data) {
989 		/* init clear state block */
990 		r = amdgpu_gfx_rlc_init_csb(adev);
991 		if (r)
992 			return r;
993 	}
994 
995 	return 0;
996 }
997 
998 static void gfx_v10_0_mec_fini(struct amdgpu_device *adev)
999 {
1000 	amdgpu_bo_free_kernel(&adev->gfx.mec.hpd_eop_obj, NULL, NULL);
1001 	amdgpu_bo_free_kernel(&adev->gfx.mec.mec_fw_obj, NULL, NULL);
1002 }
1003 
1004 static int gfx_v10_0_me_init(struct amdgpu_device *adev)
1005 {
1006 	int r;
1007 
1008 	bitmap_zero(adev->gfx.me.queue_bitmap, AMDGPU_MAX_GFX_QUEUES);
1009 
1010 	amdgpu_gfx_graphics_queue_acquire(adev);
1011 
1012 	r = gfx_v10_0_init_microcode(adev);
1013 	if (r)
1014 		DRM_ERROR("Failed to load gfx firmware!\n");
1015 
1016 	return r;
1017 }
1018 
1019 static int gfx_v10_0_mec_init(struct amdgpu_device *adev)
1020 {
1021 	int r;
1022 	u32 *hpd;
1023 	const __le32 *fw_data = NULL;
1024 	unsigned fw_size;
1025 	u32 *fw = NULL;
1026 	size_t mec_hpd_size;
1027 
1028 	const struct gfx_firmware_header_v1_0 *mec_hdr = NULL;
1029 
1030 	bitmap_zero(adev->gfx.mec.queue_bitmap, AMDGPU_MAX_COMPUTE_QUEUES);
1031 
1032 	/* take ownership of the relevant compute queues */
1033 	amdgpu_gfx_compute_queue_acquire(adev);
1034 	mec_hpd_size = adev->gfx.num_compute_rings * GFX10_MEC_HPD_SIZE;
1035 
1036 	r = amdgpu_bo_create_reserved(adev, mec_hpd_size, PAGE_SIZE,
1037 				      AMDGPU_GEM_DOMAIN_GTT,
1038 				      &adev->gfx.mec.hpd_eop_obj,
1039 				      &adev->gfx.mec.hpd_eop_gpu_addr,
1040 				      (void **)&hpd);
1041 	if (r) {
1042 		dev_warn(adev->dev, "(%d) create HDP EOP bo failed\n", r);
1043 		gfx_v10_0_mec_fini(adev);
1044 		return r;
1045 	}
1046 
1047 	memset(hpd, 0, adev->gfx.mec.hpd_eop_obj->tbo.mem.size);
1048 
1049 	amdgpu_bo_kunmap(adev->gfx.mec.hpd_eop_obj);
1050 	amdgpu_bo_unreserve(adev->gfx.mec.hpd_eop_obj);
1051 
1052 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_DIRECT) {
1053 		mec_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data;
1054 
1055 		fw_data = (const __le32 *) (adev->gfx.mec_fw->data +
1056 			 le32_to_cpu(mec_hdr->header.ucode_array_offset_bytes));
1057 		fw_size = le32_to_cpu(mec_hdr->header.ucode_size_bytes);
1058 
1059 		r = amdgpu_bo_create_reserved(adev, mec_hdr->header.ucode_size_bytes,
1060 					      PAGE_SIZE, AMDGPU_GEM_DOMAIN_GTT,
1061 					      &adev->gfx.mec.mec_fw_obj,
1062 					      &adev->gfx.mec.mec_fw_gpu_addr,
1063 					      (void **)&fw);
1064 		if (r) {
1065 			dev_err(adev->dev, "(%d) failed to create mec fw bo\n", r);
1066 			gfx_v10_0_mec_fini(adev);
1067 			return r;
1068 		}
1069 
1070 		memcpy(fw, fw_data, fw_size);
1071 
1072 		amdgpu_bo_kunmap(adev->gfx.mec.mec_fw_obj);
1073 		amdgpu_bo_unreserve(adev->gfx.mec.mec_fw_obj);
1074 	}
1075 
1076 	return 0;
1077 }
1078 
1079 static uint32_t wave_read_ind(struct amdgpu_device *adev, uint32_t wave, uint32_t address)
1080 {
1081 	WREG32_SOC15(GC, 0, mmSQ_IND_INDEX,
1082 		(wave << SQ_IND_INDEX__WAVE_ID__SHIFT) |
1083 		(address << SQ_IND_INDEX__INDEX__SHIFT));
1084 	return RREG32_SOC15(GC, 0, mmSQ_IND_DATA);
1085 }
1086 
1087 static void wave_read_regs(struct amdgpu_device *adev, uint32_t wave,
1088 			   uint32_t thread, uint32_t regno,
1089 			   uint32_t num, uint32_t *out)
1090 {
1091 	WREG32_SOC15(GC, 0, mmSQ_IND_INDEX,
1092 		(wave << SQ_IND_INDEX__WAVE_ID__SHIFT) |
1093 		(regno << SQ_IND_INDEX__INDEX__SHIFT) |
1094 		(thread << SQ_IND_INDEX__WORKITEM_ID__SHIFT) |
1095 		(SQ_IND_INDEX__AUTO_INCR_MASK));
1096 	while (num--)
1097 		*(out++) = RREG32_SOC15(GC, 0, mmSQ_IND_DATA);
1098 }
1099 
1100 static void gfx_v10_0_read_wave_data(struct amdgpu_device *adev, uint32_t simd, uint32_t wave, uint32_t *dst, int *no_fields)
1101 {
1102 	/* in gfx10 the SIMD_ID is specified as part of the INSTANCE
1103 	 * field when performing a select_se_sh so it should be
1104 	 * zero here */
1105 	WARN_ON(simd != 0);
1106 
1107 	/* type 2 wave data */
1108 	dst[(*no_fields)++] = 2;
1109 	dst[(*no_fields)++] = wave_read_ind(adev, wave, ixSQ_WAVE_STATUS);
1110 	dst[(*no_fields)++] = wave_read_ind(adev, wave, ixSQ_WAVE_PC_LO);
1111 	dst[(*no_fields)++] = wave_read_ind(adev, wave, ixSQ_WAVE_PC_HI);
1112 	dst[(*no_fields)++] = wave_read_ind(adev, wave, ixSQ_WAVE_EXEC_LO);
1113 	dst[(*no_fields)++] = wave_read_ind(adev, wave, ixSQ_WAVE_EXEC_HI);
1114 	dst[(*no_fields)++] = wave_read_ind(adev, wave, ixSQ_WAVE_HW_ID1);
1115 	dst[(*no_fields)++] = wave_read_ind(adev, wave, ixSQ_WAVE_HW_ID2);
1116 	dst[(*no_fields)++] = wave_read_ind(adev, wave, ixSQ_WAVE_INST_DW0);
1117 	dst[(*no_fields)++] = wave_read_ind(adev, wave, ixSQ_WAVE_GPR_ALLOC);
1118 	dst[(*no_fields)++] = wave_read_ind(adev, wave, ixSQ_WAVE_LDS_ALLOC);
1119 	dst[(*no_fields)++] = wave_read_ind(adev, wave, ixSQ_WAVE_TRAPSTS);
1120 	dst[(*no_fields)++] = wave_read_ind(adev, wave, ixSQ_WAVE_IB_STS);
1121 	dst[(*no_fields)++] = wave_read_ind(adev, wave, ixSQ_WAVE_IB_STS2);
1122 	dst[(*no_fields)++] = wave_read_ind(adev, wave, ixSQ_WAVE_IB_DBG1);
1123 	dst[(*no_fields)++] = wave_read_ind(adev, wave, ixSQ_WAVE_M0);
1124 }
1125 
1126 static void gfx_v10_0_read_wave_sgprs(struct amdgpu_device *adev, uint32_t simd,
1127 				     uint32_t wave, uint32_t start,
1128 				     uint32_t size, uint32_t *dst)
1129 {
1130 	WARN_ON(simd != 0);
1131 
1132 	wave_read_regs(
1133 		adev, wave, 0, start + SQIND_WAVE_SGPRS_OFFSET, size,
1134 		dst);
1135 }
1136 
1137 static void gfx_v10_0_read_wave_vgprs(struct amdgpu_device *adev, uint32_t simd,
1138 				      uint32_t wave, uint32_t thread,
1139 				      uint32_t start, uint32_t size,
1140 				      uint32_t *dst)
1141 {
1142 	wave_read_regs(
1143 		adev, wave, thread,
1144 		start + SQIND_WAVE_VGPRS_OFFSET, size, dst);
1145 }
1146 
1147 static void gfx_v10_0_select_me_pipe_q(struct amdgpu_device *adev,
1148 									  u32 me, u32 pipe, u32 q, u32 vm)
1149  {
1150        nv_grbm_select(adev, me, pipe, q, vm);
1151  }
1152 
1153 
1154 static const struct amdgpu_gfx_funcs gfx_v10_0_gfx_funcs = {
1155 	.get_gpu_clock_counter = &gfx_v10_0_get_gpu_clock_counter,
1156 	.select_se_sh = &gfx_v10_0_select_se_sh,
1157 	.read_wave_data = &gfx_v10_0_read_wave_data,
1158 	.read_wave_sgprs = &gfx_v10_0_read_wave_sgprs,
1159 	.read_wave_vgprs = &gfx_v10_0_read_wave_vgprs,
1160 	.select_me_pipe_q = &gfx_v10_0_select_me_pipe_q,
1161 };
1162 
1163 static void gfx_v10_0_gpu_early_init(struct amdgpu_device *adev)
1164 {
1165 	u32 gb_addr_config;
1166 
1167 	adev->gfx.funcs = &gfx_v10_0_gfx_funcs;
1168 
1169 	switch (adev->asic_type) {
1170 	case CHIP_NAVI10:
1171 	case CHIP_NAVI14:
1172 	case CHIP_NAVI12:
1173 		adev->gfx.config.max_hw_contexts = 8;
1174 		adev->gfx.config.sc_prim_fifo_size_frontend = 0x20;
1175 		adev->gfx.config.sc_prim_fifo_size_backend = 0x100;
1176 		adev->gfx.config.sc_hiz_tile_fifo_size = 0;
1177 		adev->gfx.config.sc_earlyz_tile_fifo_size = 0x4C0;
1178 		gb_addr_config = RREG32_SOC15(GC, 0, mmGB_ADDR_CONFIG);
1179 		break;
1180 	default:
1181 		BUG();
1182 		break;
1183 	}
1184 
1185 	adev->gfx.config.gb_addr_config = gb_addr_config;
1186 
1187 	adev->gfx.config.gb_addr_config_fields.num_pipes = 1 <<
1188 			REG_GET_FIELD(adev->gfx.config.gb_addr_config,
1189 				      GB_ADDR_CONFIG, NUM_PIPES);
1190 
1191 	adev->gfx.config.max_tile_pipes =
1192 		adev->gfx.config.gb_addr_config_fields.num_pipes;
1193 
1194 	adev->gfx.config.gb_addr_config_fields.max_compress_frags = 1 <<
1195 			REG_GET_FIELD(adev->gfx.config.gb_addr_config,
1196 				      GB_ADDR_CONFIG, MAX_COMPRESSED_FRAGS);
1197 	adev->gfx.config.gb_addr_config_fields.num_rb_per_se = 1 <<
1198 			REG_GET_FIELD(adev->gfx.config.gb_addr_config,
1199 				      GB_ADDR_CONFIG, NUM_RB_PER_SE);
1200 	adev->gfx.config.gb_addr_config_fields.num_se = 1 <<
1201 			REG_GET_FIELD(adev->gfx.config.gb_addr_config,
1202 				      GB_ADDR_CONFIG, NUM_SHADER_ENGINES);
1203 	adev->gfx.config.gb_addr_config_fields.pipe_interleave_size = 1 << (8 +
1204 			REG_GET_FIELD(adev->gfx.config.gb_addr_config,
1205 				      GB_ADDR_CONFIG, PIPE_INTERLEAVE_SIZE));
1206 }
1207 
1208 static int gfx_v10_0_gfx_ring_init(struct amdgpu_device *adev, int ring_id,
1209 				   int me, int pipe, int queue)
1210 {
1211 	int r;
1212 	struct amdgpu_ring *ring;
1213 	unsigned int irq_type;
1214 
1215 	ring = &adev->gfx.gfx_ring[ring_id];
1216 
1217 	ring->me = me;
1218 	ring->pipe = pipe;
1219 	ring->queue = queue;
1220 
1221 	ring->ring_obj = NULL;
1222 	ring->use_doorbell = true;
1223 
1224 	if (!ring_id)
1225 		ring->doorbell_index = adev->doorbell_index.gfx_ring0 << 1;
1226 	else
1227 		ring->doorbell_index = adev->doorbell_index.gfx_ring1 << 1;
1228 	sprintf(ring->name, "gfx_%d.%d.%d", ring->me, ring->pipe, ring->queue);
1229 
1230 	irq_type = AMDGPU_CP_IRQ_GFX_ME0_PIPE0_EOP + ring->pipe;
1231 	r = amdgpu_ring_init(adev, ring, 1024,
1232 			     &adev->gfx.eop_irq, irq_type);
1233 	if (r)
1234 		return r;
1235 	return 0;
1236 }
1237 
1238 static int gfx_v10_0_compute_ring_init(struct amdgpu_device *adev, int ring_id,
1239 				       int mec, int pipe, int queue)
1240 {
1241 	int r;
1242 	unsigned irq_type;
1243 	struct amdgpu_ring *ring = &adev->gfx.compute_ring[ring_id];
1244 
1245 	ring = &adev->gfx.compute_ring[ring_id];
1246 
1247 	/* mec0 is me1 */
1248 	ring->me = mec + 1;
1249 	ring->pipe = pipe;
1250 	ring->queue = queue;
1251 
1252 	ring->ring_obj = NULL;
1253 	ring->use_doorbell = true;
1254 	ring->doorbell_index = (adev->doorbell_index.mec_ring0 + ring_id) << 1;
1255 	ring->eop_gpu_addr = adev->gfx.mec.hpd_eop_gpu_addr
1256 				+ (ring_id * GFX10_MEC_HPD_SIZE);
1257 	sprintf(ring->name, "comp_%d.%d.%d", ring->me, ring->pipe, ring->queue);
1258 
1259 	irq_type = AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP
1260 		+ ((ring->me - 1) * adev->gfx.mec.num_pipe_per_mec)
1261 		+ ring->pipe;
1262 
1263 	/* type-2 packets are deprecated on MEC, use type-3 instead */
1264 	r = amdgpu_ring_init(adev, ring, 1024,
1265 			     &adev->gfx.eop_irq, irq_type);
1266 	if (r)
1267 		return r;
1268 
1269 	return 0;
1270 }
1271 
1272 static int gfx_v10_0_sw_init(void *handle)
1273 {
1274 	int i, j, k, r, ring_id = 0;
1275 	struct amdgpu_kiq *kiq;
1276 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1277 
1278 	switch (adev->asic_type) {
1279 	case CHIP_NAVI10:
1280 	case CHIP_NAVI14:
1281 	case CHIP_NAVI12:
1282 		adev->gfx.me.num_me = 1;
1283 		adev->gfx.me.num_pipe_per_me = 2;
1284 		adev->gfx.me.num_queue_per_pipe = 1;
1285 		adev->gfx.mec.num_mec = 2;
1286 		adev->gfx.mec.num_pipe_per_mec = 4;
1287 		adev->gfx.mec.num_queue_per_pipe = 8;
1288 		break;
1289 	default:
1290 		adev->gfx.me.num_me = 1;
1291 		adev->gfx.me.num_pipe_per_me = 1;
1292 		adev->gfx.me.num_queue_per_pipe = 1;
1293 		adev->gfx.mec.num_mec = 1;
1294 		adev->gfx.mec.num_pipe_per_mec = 4;
1295 		adev->gfx.mec.num_queue_per_pipe = 8;
1296 		break;
1297 	}
1298 
1299 	/* KIQ event */
1300 	r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_GRBM_CP,
1301 			      GFX_10_1__SRCID__CP_IB2_INTERRUPT_PKT,
1302 			      &adev->gfx.kiq.irq);
1303 	if (r)
1304 		return r;
1305 
1306 	/* EOP Event */
1307 	r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_GRBM_CP,
1308 			      GFX_10_1__SRCID__CP_EOP_INTERRUPT,
1309 			      &adev->gfx.eop_irq);
1310 	if (r)
1311 		return r;
1312 
1313 	/* Privileged reg */
1314 	r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_GRBM_CP, GFX_10_1__SRCID__CP_PRIV_REG_FAULT,
1315 			      &adev->gfx.priv_reg_irq);
1316 	if (r)
1317 		return r;
1318 
1319 	/* Privileged inst */
1320 	r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_GRBM_CP, GFX_10_1__SRCID__CP_PRIV_INSTR_FAULT,
1321 			      &adev->gfx.priv_inst_irq);
1322 	if (r)
1323 		return r;
1324 
1325 	adev->gfx.gfx_current_status = AMDGPU_GFX_NORMAL_MODE;
1326 
1327 	gfx_v10_0_scratch_init(adev);
1328 
1329 	r = gfx_v10_0_me_init(adev);
1330 	if (r)
1331 		return r;
1332 
1333 	r = gfx_v10_0_rlc_init(adev);
1334 	if (r) {
1335 		DRM_ERROR("Failed to init rlc BOs!\n");
1336 		return r;
1337 	}
1338 
1339 	r = gfx_v10_0_mec_init(adev);
1340 	if (r) {
1341 		DRM_ERROR("Failed to init MEC BOs!\n");
1342 		return r;
1343 	}
1344 
1345 	/* set up the gfx ring */
1346 	for (i = 0; i < adev->gfx.me.num_me; i++) {
1347 		for (j = 0; j < adev->gfx.me.num_queue_per_pipe; j++) {
1348 			for (k = 0; k < adev->gfx.me.num_pipe_per_me; k++) {
1349 				if (!amdgpu_gfx_is_me_queue_enabled(adev, i, k, j))
1350 					continue;
1351 
1352 				r = gfx_v10_0_gfx_ring_init(adev, ring_id,
1353 							    i, k, j);
1354 				if (r)
1355 					return r;
1356 				ring_id++;
1357 			}
1358 		}
1359 	}
1360 
1361 	ring_id = 0;
1362 	/* set up the compute queues - allocate horizontally across pipes */
1363 	for (i = 0; i < adev->gfx.mec.num_mec; ++i) {
1364 		for (j = 0; j < adev->gfx.mec.num_queue_per_pipe; j++) {
1365 			for (k = 0; k < adev->gfx.mec.num_pipe_per_mec; k++) {
1366 				if (!amdgpu_gfx_is_mec_queue_enabled(adev, i, k,
1367 								     j))
1368 					continue;
1369 
1370 				r = gfx_v10_0_compute_ring_init(adev, ring_id,
1371 								i, k, j);
1372 				if (r)
1373 					return r;
1374 
1375 				ring_id++;
1376 			}
1377 		}
1378 	}
1379 
1380 	r = amdgpu_gfx_kiq_init(adev, GFX10_MEC_HPD_SIZE);
1381 	if (r) {
1382 		DRM_ERROR("Failed to init KIQ BOs!\n");
1383 		return r;
1384 	}
1385 
1386 	kiq = &adev->gfx.kiq;
1387 	r = amdgpu_gfx_kiq_init_ring(adev, &kiq->ring, &kiq->irq);
1388 	if (r)
1389 		return r;
1390 
1391 	r = amdgpu_gfx_mqd_sw_init(adev, sizeof(struct v10_compute_mqd));
1392 	if (r)
1393 		return r;
1394 
1395 	/* allocate visible FB for rlc auto-loading fw */
1396 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_RLC_BACKDOOR_AUTO) {
1397 		r = gfx_v10_0_rlc_backdoor_autoload_buffer_init(adev);
1398 		if (r)
1399 			return r;
1400 	}
1401 
1402 	adev->gfx.ce_ram_size = F32_CE_PROGRAM_RAM_SIZE;
1403 
1404 	gfx_v10_0_gpu_early_init(adev);
1405 
1406 	return 0;
1407 }
1408 
1409 static void gfx_v10_0_pfp_fini(struct amdgpu_device *adev)
1410 {
1411 	amdgpu_bo_free_kernel(&adev->gfx.pfp.pfp_fw_obj,
1412 			      &adev->gfx.pfp.pfp_fw_gpu_addr,
1413 			      (void **)&adev->gfx.pfp.pfp_fw_ptr);
1414 }
1415 
1416 static void gfx_v10_0_ce_fini(struct amdgpu_device *adev)
1417 {
1418 	amdgpu_bo_free_kernel(&adev->gfx.ce.ce_fw_obj,
1419 			      &adev->gfx.ce.ce_fw_gpu_addr,
1420 			      (void **)&adev->gfx.ce.ce_fw_ptr);
1421 }
1422 
1423 static void gfx_v10_0_me_fini(struct amdgpu_device *adev)
1424 {
1425 	amdgpu_bo_free_kernel(&adev->gfx.me.me_fw_obj,
1426 			      &adev->gfx.me.me_fw_gpu_addr,
1427 			      (void **)&adev->gfx.me.me_fw_ptr);
1428 }
1429 
1430 static int gfx_v10_0_sw_fini(void *handle)
1431 {
1432 	int i;
1433 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
1434 
1435 	for (i = 0; i < adev->gfx.num_gfx_rings; i++)
1436 		amdgpu_ring_fini(&adev->gfx.gfx_ring[i]);
1437 	for (i = 0; i < adev->gfx.num_compute_rings; i++)
1438 		amdgpu_ring_fini(&adev->gfx.compute_ring[i]);
1439 
1440 	amdgpu_gfx_mqd_sw_fini(adev);
1441 	amdgpu_gfx_kiq_free_ring(&adev->gfx.kiq.ring);
1442 	amdgpu_gfx_kiq_fini(adev);
1443 
1444 	gfx_v10_0_pfp_fini(adev);
1445 	gfx_v10_0_ce_fini(adev);
1446 	gfx_v10_0_me_fini(adev);
1447 	gfx_v10_0_rlc_fini(adev);
1448 	gfx_v10_0_mec_fini(adev);
1449 
1450 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_RLC_BACKDOOR_AUTO)
1451 		gfx_v10_0_rlc_backdoor_autoload_buffer_fini(adev);
1452 
1453 	gfx_v10_0_free_microcode(adev);
1454 
1455 	return 0;
1456 }
1457 
1458 
1459 static void gfx_v10_0_tiling_mode_table_init(struct amdgpu_device *adev)
1460 {
1461 	/* TODO */
1462 }
1463 
1464 static void gfx_v10_0_select_se_sh(struct amdgpu_device *adev, u32 se_num,
1465 				   u32 sh_num, u32 instance)
1466 {
1467 	u32 data;
1468 
1469 	if (instance == 0xffffffff)
1470 		data = REG_SET_FIELD(0, GRBM_GFX_INDEX,
1471 				     INSTANCE_BROADCAST_WRITES, 1);
1472 	else
1473 		data = REG_SET_FIELD(0, GRBM_GFX_INDEX, INSTANCE_INDEX,
1474 				     instance);
1475 
1476 	if (se_num == 0xffffffff)
1477 		data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SE_BROADCAST_WRITES,
1478 				     1);
1479 	else
1480 		data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SE_INDEX, se_num);
1481 
1482 	if (sh_num == 0xffffffff)
1483 		data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SA_BROADCAST_WRITES,
1484 				     1);
1485 	else
1486 		data = REG_SET_FIELD(data, GRBM_GFX_INDEX, SA_INDEX, sh_num);
1487 
1488 	WREG32_SOC15(GC, 0, mmGRBM_GFX_INDEX, data);
1489 }
1490 
1491 static u32 gfx_v10_0_get_rb_active_bitmap(struct amdgpu_device *adev)
1492 {
1493 	u32 data, mask;
1494 
1495 	data = RREG32_SOC15(GC, 0, mmCC_RB_BACKEND_DISABLE);
1496 	data |= RREG32_SOC15(GC, 0, mmGC_USER_RB_BACKEND_DISABLE);
1497 
1498 	data &= CC_RB_BACKEND_DISABLE__BACKEND_DISABLE_MASK;
1499 	data >>= GC_USER_RB_BACKEND_DISABLE__BACKEND_DISABLE__SHIFT;
1500 
1501 	mask = amdgpu_gfx_create_bitmask(adev->gfx.config.max_backends_per_se /
1502 					 adev->gfx.config.max_sh_per_se);
1503 
1504 	return (~data) & mask;
1505 }
1506 
1507 static void gfx_v10_0_setup_rb(struct amdgpu_device *adev)
1508 {
1509 	int i, j;
1510 	u32 data;
1511 	u32 active_rbs = 0;
1512 	u32 rb_bitmap_width_per_sh = adev->gfx.config.max_backends_per_se /
1513 					adev->gfx.config.max_sh_per_se;
1514 
1515 	mutex_lock(&adev->grbm_idx_mutex);
1516 	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
1517 		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
1518 			gfx_v10_0_select_se_sh(adev, i, j, 0xffffffff);
1519 			data = gfx_v10_0_get_rb_active_bitmap(adev);
1520 			active_rbs |= data << ((i * adev->gfx.config.max_sh_per_se + j) *
1521 					       rb_bitmap_width_per_sh);
1522 		}
1523 	}
1524 	gfx_v10_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
1525 	mutex_unlock(&adev->grbm_idx_mutex);
1526 
1527 	adev->gfx.config.backend_enable_mask = active_rbs;
1528 	adev->gfx.config.num_rbs = hweight32(active_rbs);
1529 }
1530 
1531 static u32 gfx_v10_0_init_pa_sc_tile_steering_override(struct amdgpu_device *adev)
1532 {
1533 	uint32_t num_sc;
1534 	uint32_t enabled_rb_per_sh;
1535 	uint32_t active_rb_bitmap;
1536 	uint32_t num_rb_per_sc;
1537 	uint32_t num_packer_per_sc;
1538 	uint32_t pa_sc_tile_steering_override;
1539 
1540 	/* init num_sc */
1541 	num_sc = adev->gfx.config.max_shader_engines * adev->gfx.config.max_sh_per_se *
1542 			adev->gfx.config.num_sc_per_sh;
1543 	/* init num_rb_per_sc */
1544 	active_rb_bitmap = gfx_v10_0_get_rb_active_bitmap(adev);
1545 	enabled_rb_per_sh = hweight32(active_rb_bitmap);
1546 	num_rb_per_sc = enabled_rb_per_sh / adev->gfx.config.num_sc_per_sh;
1547 	/* init num_packer_per_sc */
1548 	num_packer_per_sc = adev->gfx.config.num_packer_per_sc;
1549 
1550 	pa_sc_tile_steering_override = 0;
1551 	pa_sc_tile_steering_override |=
1552 		(order_base_2(num_sc) << PA_SC_TILE_STEERING_OVERRIDE__NUM_SC__SHIFT) &
1553 		PA_SC_TILE_STEERING_OVERRIDE__NUM_SC_MASK;
1554 	pa_sc_tile_steering_override |=
1555 		(order_base_2(num_rb_per_sc) << PA_SC_TILE_STEERING_OVERRIDE__NUM_RB_PER_SC__SHIFT) &
1556 		PA_SC_TILE_STEERING_OVERRIDE__NUM_RB_PER_SC_MASK;
1557 	pa_sc_tile_steering_override |=
1558 		(order_base_2(num_packer_per_sc) << PA_SC_TILE_STEERING_OVERRIDE__NUM_PACKER_PER_SC__SHIFT) &
1559 		PA_SC_TILE_STEERING_OVERRIDE__NUM_PACKER_PER_SC_MASK;
1560 
1561 	return pa_sc_tile_steering_override;
1562 }
1563 
1564 #define DEFAULT_SH_MEM_BASES	(0x6000)
1565 #define FIRST_COMPUTE_VMID	(8)
1566 #define LAST_COMPUTE_VMID	(16)
1567 
1568 static void gfx_v10_0_init_compute_vmid(struct amdgpu_device *adev)
1569 {
1570 	int i;
1571 	uint32_t sh_mem_bases;
1572 
1573 	/*
1574 	 * Configure apertures:
1575 	 * LDS:         0x60000000'00000000 - 0x60000001'00000000 (4GB)
1576 	 * Scratch:     0x60000001'00000000 - 0x60000002'00000000 (4GB)
1577 	 * GPUVM:       0x60010000'00000000 - 0x60020000'00000000 (1TB)
1578 	 */
1579 	sh_mem_bases = DEFAULT_SH_MEM_BASES | (DEFAULT_SH_MEM_BASES << 16);
1580 
1581 	mutex_lock(&adev->srbm_mutex);
1582 	for (i = FIRST_COMPUTE_VMID; i < LAST_COMPUTE_VMID; i++) {
1583 		nv_grbm_select(adev, 0, 0, 0, i);
1584 		/* CP and shaders */
1585 		WREG32_SOC15(GC, 0, mmSH_MEM_CONFIG, DEFAULT_SH_MEM_CONFIG);
1586 		WREG32_SOC15(GC, 0, mmSH_MEM_BASES, sh_mem_bases);
1587 	}
1588 	nv_grbm_select(adev, 0, 0, 0, 0);
1589 	mutex_unlock(&adev->srbm_mutex);
1590 
1591 	/* Initialize all compute VMIDs to have no GDS, GWS, or OA
1592 	   acccess. These should be enabled by FW for target VMIDs. */
1593 	for (i = FIRST_COMPUTE_VMID; i < LAST_COMPUTE_VMID; i++) {
1594 		WREG32_SOC15_OFFSET(GC, 0, mmGDS_VMID0_BASE, 2 * i, 0);
1595 		WREG32_SOC15_OFFSET(GC, 0, mmGDS_VMID0_SIZE, 2 * i, 0);
1596 		WREG32_SOC15_OFFSET(GC, 0, mmGDS_GWS_VMID0, i, 0);
1597 		WREG32_SOC15_OFFSET(GC, 0, mmGDS_OA_VMID0, i, 0);
1598 	}
1599 }
1600 
1601 static void gfx_v10_0_init_gds_vmid(struct amdgpu_device *adev)
1602 {
1603 	int vmid;
1604 
1605 	/*
1606 	 * Initialize all compute and user-gfx VMIDs to have no GDS, GWS, or OA
1607 	 * access. Compute VMIDs should be enabled by FW for target VMIDs,
1608 	 * the driver can enable them for graphics. VMID0 should maintain
1609 	 * access so that HWS firmware can save/restore entries.
1610 	 */
1611 	for (vmid = 1; vmid < 16; vmid++) {
1612 		WREG32_SOC15_OFFSET(GC, 0, mmGDS_VMID0_BASE, 2 * vmid, 0);
1613 		WREG32_SOC15_OFFSET(GC, 0, mmGDS_VMID0_SIZE, 2 * vmid, 0);
1614 		WREG32_SOC15_OFFSET(GC, 0, mmGDS_GWS_VMID0, vmid, 0);
1615 		WREG32_SOC15_OFFSET(GC, 0, mmGDS_OA_VMID0, vmid, 0);
1616 	}
1617 }
1618 
1619 
1620 static void gfx_v10_0_tcp_harvest(struct amdgpu_device *adev)
1621 {
1622 	int i, j, k;
1623 	int max_wgp_per_sh = adev->gfx.config.max_cu_per_sh >> 1;
1624 	u32 tmp, wgp_active_bitmap = 0;
1625 	u32 gcrd_targets_disable_tcp = 0;
1626 	u32 utcl_invreq_disable = 0;
1627 	/*
1628 	 * GCRD_TARGETS_DISABLE field contains
1629 	 * for Navi10/Navi12: GL1C=[18:15], SQC=[14:10], TCP=[9:0]
1630 	 * for Navi14: GL1C=[21:18], SQC=[17:12], TCP=[11:0]
1631 	 */
1632 	u32 gcrd_targets_disable_mask = amdgpu_gfx_create_bitmask(
1633 		2 * max_wgp_per_sh + /* TCP */
1634 		max_wgp_per_sh + /* SQC */
1635 		4); /* GL1C */
1636 	/*
1637 	 * UTCL1_UTCL0_INVREQ_DISABLE field contains
1638 	 * for Navi10Navi12: SQG=[24], RMI=[23:20], SQC=[19:10], TCP=[9:0]
1639 	 * for Navi14: SQG=[28], RMI=[27:24], SQC=[23:12], TCP=[11:0]
1640 	 */
1641 	u32 utcl_invreq_disable_mask = amdgpu_gfx_create_bitmask(
1642 		2 * max_wgp_per_sh + /* TCP */
1643 		2 * max_wgp_per_sh + /* SQC */
1644 		4 + /* RMI */
1645 		1); /* SQG */
1646 
1647 	if (adev->asic_type == CHIP_NAVI10 ||
1648 	    adev->asic_type == CHIP_NAVI14 ||
1649 	    adev->asic_type == CHIP_NAVI12) {
1650 		mutex_lock(&adev->grbm_idx_mutex);
1651 		for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
1652 			for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
1653 				gfx_v10_0_select_se_sh(adev, i, j, 0xffffffff);
1654 				wgp_active_bitmap = gfx_v10_0_get_wgp_active_bitmap_per_sh(adev);
1655 				/*
1656 				 * Set corresponding TCP bits for the inactive WGPs in
1657 				 * GCRD_SA_TARGETS_DISABLE
1658 				 */
1659 				gcrd_targets_disable_tcp = 0;
1660 				/* Set TCP & SQC bits in UTCL1_UTCL0_INVREQ_DISABLE */
1661 				utcl_invreq_disable = 0;
1662 
1663 				for (k = 0; k < max_wgp_per_sh; k++) {
1664 					if (!(wgp_active_bitmap & (1 << k))) {
1665 						gcrd_targets_disable_tcp |= 3 << (2 * k);
1666 						utcl_invreq_disable |= (3 << (2 * k)) |
1667 							(3 << (2 * (max_wgp_per_sh + k)));
1668 					}
1669 				}
1670 
1671 				tmp = RREG32_SOC15(GC, 0, mmUTCL1_UTCL0_INVREQ_DISABLE);
1672 				/* only override TCP & SQC bits */
1673 				tmp &= 0xffffffff << (4 * max_wgp_per_sh);
1674 				tmp |= (utcl_invreq_disable & utcl_invreq_disable_mask);
1675 				WREG32_SOC15(GC, 0, mmUTCL1_UTCL0_INVREQ_DISABLE, tmp);
1676 
1677 				tmp = RREG32_SOC15(GC, 0, mmGCRD_SA_TARGETS_DISABLE);
1678 				/* only override TCP bits */
1679 				tmp &= 0xffffffff << (2 * max_wgp_per_sh);
1680 				tmp |= (gcrd_targets_disable_tcp & gcrd_targets_disable_mask);
1681 				WREG32_SOC15(GC, 0, mmGCRD_SA_TARGETS_DISABLE, tmp);
1682 			}
1683 		}
1684 
1685 		gfx_v10_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
1686 		mutex_unlock(&adev->grbm_idx_mutex);
1687 	}
1688 }
1689 
1690 static void gfx_v10_0_get_tcc_info(struct amdgpu_device *adev)
1691 {
1692 	/* TCCs are global (not instanced). */
1693 	uint32_t tcc_disable = RREG32_SOC15(GC, 0, mmCGTS_TCC_DISABLE) |
1694 			       RREG32_SOC15(GC, 0, mmCGTS_USER_TCC_DISABLE);
1695 
1696 	adev->gfx.config.tcc_disabled_mask =
1697 		REG_GET_FIELD(tcc_disable, CGTS_TCC_DISABLE, TCC_DISABLE) |
1698 		(REG_GET_FIELD(tcc_disable, CGTS_TCC_DISABLE, HI_TCC_DISABLE) << 16);
1699 }
1700 
1701 static void gfx_v10_0_constants_init(struct amdgpu_device *adev)
1702 {
1703 	u32 tmp;
1704 	int i;
1705 
1706 	WREG32_FIELD15(GC, 0, GRBM_CNTL, READ_TIMEOUT, 0xff);
1707 
1708 	gfx_v10_0_tiling_mode_table_init(adev);
1709 
1710 	gfx_v10_0_setup_rb(adev);
1711 	gfx_v10_0_get_cu_info(adev, &adev->gfx.cu_info);
1712 	gfx_v10_0_get_tcc_info(adev);
1713 	adev->gfx.config.pa_sc_tile_steering_override =
1714 		gfx_v10_0_init_pa_sc_tile_steering_override(adev);
1715 
1716 	/* XXX SH_MEM regs */
1717 	/* where to put LDS, scratch, GPUVM in FSA64 space */
1718 	mutex_lock(&adev->srbm_mutex);
1719 	for (i = 0; i < adev->vm_manager.id_mgr[AMDGPU_GFXHUB_0].num_ids; i++) {
1720 		nv_grbm_select(adev, 0, 0, 0, i);
1721 		/* CP and shaders */
1722 		WREG32_SOC15(GC, 0, mmSH_MEM_CONFIG, DEFAULT_SH_MEM_CONFIG);
1723 		if (i != 0) {
1724 			tmp = REG_SET_FIELD(0, SH_MEM_BASES, PRIVATE_BASE,
1725 				(adev->gmc.private_aperture_start >> 48));
1726 			tmp = REG_SET_FIELD(tmp, SH_MEM_BASES, SHARED_BASE,
1727 				(adev->gmc.shared_aperture_start >> 48));
1728 			WREG32_SOC15(GC, 0, mmSH_MEM_BASES, tmp);
1729 		}
1730 	}
1731 	nv_grbm_select(adev, 0, 0, 0, 0);
1732 
1733 	mutex_unlock(&adev->srbm_mutex);
1734 
1735 	gfx_v10_0_init_compute_vmid(adev);
1736 	gfx_v10_0_init_gds_vmid(adev);
1737 
1738 }
1739 
1740 static void gfx_v10_0_enable_gui_idle_interrupt(struct amdgpu_device *adev,
1741 					       bool enable)
1742 {
1743 	u32 tmp = RREG32_SOC15(GC, 0, mmCP_INT_CNTL_RING0);
1744 
1745 	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CNTX_BUSY_INT_ENABLE,
1746 			    enable ? 1 : 0);
1747 	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CNTX_EMPTY_INT_ENABLE,
1748 			    enable ? 1 : 0);
1749 	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CMP_BUSY_INT_ENABLE,
1750 			    enable ? 1 : 0);
1751 	tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, GFX_IDLE_INT_ENABLE,
1752 			    enable ? 1 : 0);
1753 
1754 	WREG32_SOC15(GC, 0, mmCP_INT_CNTL_RING0, tmp);
1755 }
1756 
1757 static int gfx_v10_0_init_csb(struct amdgpu_device *adev)
1758 {
1759 	adev->gfx.rlc.funcs->get_csb_buffer(adev, adev->gfx.rlc.cs_ptr);
1760 
1761 	/* csib */
1762 	WREG32_SOC15(GC, 0, mmRLC_CSIB_ADDR_HI,
1763 		     adev->gfx.rlc.clear_state_gpu_addr >> 32);
1764 	WREG32_SOC15(GC, 0, mmRLC_CSIB_ADDR_LO,
1765 		     adev->gfx.rlc.clear_state_gpu_addr & 0xfffffffc);
1766 	WREG32_SOC15(GC, 0, mmRLC_CSIB_LENGTH, adev->gfx.rlc.clear_state_size);
1767 
1768 	return 0;
1769 }
1770 
1771 void gfx_v10_0_rlc_stop(struct amdgpu_device *adev)
1772 {
1773 	u32 tmp = RREG32_SOC15(GC, 0, mmRLC_CNTL);
1774 
1775 	tmp = REG_SET_FIELD(tmp, RLC_CNTL, RLC_ENABLE_F32, 0);
1776 	WREG32_SOC15(GC, 0, mmRLC_CNTL, tmp);
1777 }
1778 
1779 static void gfx_v10_0_rlc_reset(struct amdgpu_device *adev)
1780 {
1781 	WREG32_FIELD15(GC, 0, GRBM_SOFT_RESET, SOFT_RESET_RLC, 1);
1782 	udelay(50);
1783 	WREG32_FIELD15(GC, 0, GRBM_SOFT_RESET, SOFT_RESET_RLC, 0);
1784 	udelay(50);
1785 }
1786 
1787 static void gfx_v10_0_rlc_smu_handshake_cntl(struct amdgpu_device *adev,
1788 					     bool enable)
1789 {
1790 	uint32_t rlc_pg_cntl;
1791 
1792 	rlc_pg_cntl = RREG32_SOC15(GC, 0, mmRLC_PG_CNTL);
1793 
1794 	if (!enable) {
1795 		/* RLC_PG_CNTL[23] = 0 (default)
1796 		 * RLC will wait for handshake acks with SMU
1797 		 * GFXOFF will be enabled
1798 		 * RLC_PG_CNTL[23] = 1
1799 		 * RLC will not issue any message to SMU
1800 		 * hence no handshake between SMU & RLC
1801 		 * GFXOFF will be disabled
1802 		 */
1803 		rlc_pg_cntl |= 0x800000;
1804 	} else
1805 		rlc_pg_cntl &= ~0x800000;
1806 	WREG32_SOC15(GC, 0, mmRLC_PG_CNTL, rlc_pg_cntl);
1807 }
1808 
1809 static void gfx_v10_0_rlc_start(struct amdgpu_device *adev)
1810 {
1811 	/* TODO: enable rlc & smu handshake until smu
1812 	 * and gfxoff feature works as expected */
1813 	if (!(amdgpu_pp_feature_mask & PP_GFXOFF_MASK))
1814 		gfx_v10_0_rlc_smu_handshake_cntl(adev, false);
1815 
1816 	WREG32_FIELD15(GC, 0, RLC_CNTL, RLC_ENABLE_F32, 1);
1817 	udelay(50);
1818 }
1819 
1820 static void gfx_v10_0_rlc_enable_srm(struct amdgpu_device *adev)
1821 {
1822 	uint32_t tmp;
1823 
1824 	/* enable Save Restore Machine */
1825 	tmp = RREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_SRM_CNTL));
1826 	tmp |= RLC_SRM_CNTL__AUTO_INCR_ADDR_MASK;
1827 	tmp |= RLC_SRM_CNTL__SRM_ENABLE_MASK;
1828 	WREG32(SOC15_REG_OFFSET(GC, 0, mmRLC_SRM_CNTL), tmp);
1829 }
1830 
1831 static int gfx_v10_0_rlc_load_microcode(struct amdgpu_device *adev)
1832 {
1833 	const struct rlc_firmware_header_v2_0 *hdr;
1834 	const __le32 *fw_data;
1835 	unsigned i, fw_size;
1836 
1837 	if (!adev->gfx.rlc_fw)
1838 		return -EINVAL;
1839 
1840 	hdr = (const struct rlc_firmware_header_v2_0 *)adev->gfx.rlc_fw->data;
1841 	amdgpu_ucode_print_rlc_hdr(&hdr->header);
1842 
1843 	fw_data = (const __le32 *)(adev->gfx.rlc_fw->data +
1844 			   le32_to_cpu(hdr->header.ucode_array_offset_bytes));
1845 	fw_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4;
1846 
1847 	WREG32_SOC15(GC, 0, mmRLC_GPM_UCODE_ADDR,
1848 		     RLCG_UCODE_LOADING_START_ADDRESS);
1849 
1850 	for (i = 0; i < fw_size; i++)
1851 		WREG32_SOC15(GC, 0, mmRLC_GPM_UCODE_DATA,
1852 			     le32_to_cpup(fw_data++));
1853 
1854 	WREG32_SOC15(GC, 0, mmRLC_GPM_UCODE_ADDR, adev->gfx.rlc_fw_version);
1855 
1856 	return 0;
1857 }
1858 
1859 static int gfx_v10_0_rlc_resume(struct amdgpu_device *adev)
1860 {
1861 	int r;
1862 
1863 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_PSP) {
1864 
1865 		r = gfx_v10_0_wait_for_rlc_autoload_complete(adev);
1866 		if (r)
1867 			return r;
1868 
1869 		gfx_v10_0_init_csb(adev);
1870 
1871 		if (!amdgpu_sriov_vf(adev)) /* enable RLC SRM */
1872 			gfx_v10_0_rlc_enable_srm(adev);
1873 	} else {
1874 		adev->gfx.rlc.funcs->stop(adev);
1875 
1876 		/* disable CG */
1877 		WREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL, 0);
1878 
1879 		/* disable PG */
1880 		WREG32_SOC15(GC, 0, mmRLC_PG_CNTL, 0);
1881 
1882 		if (adev->firmware.load_type == AMDGPU_FW_LOAD_DIRECT) {
1883 			/* legacy rlc firmware loading */
1884 			r = gfx_v10_0_rlc_load_microcode(adev);
1885 			if (r)
1886 				return r;
1887 		} else if (adev->firmware.load_type == AMDGPU_FW_LOAD_RLC_BACKDOOR_AUTO) {
1888 			/* rlc backdoor autoload firmware */
1889 			r = gfx_v10_0_rlc_backdoor_autoload_enable(adev);
1890 			if (r)
1891 				return r;
1892 		}
1893 
1894 		gfx_v10_0_init_csb(adev);
1895 
1896 		adev->gfx.rlc.funcs->start(adev);
1897 
1898 		if (adev->firmware.load_type == AMDGPU_FW_LOAD_RLC_BACKDOOR_AUTO) {
1899 			r = gfx_v10_0_wait_for_rlc_autoload_complete(adev);
1900 			if (r)
1901 				return r;
1902 		}
1903 	}
1904 	return 0;
1905 }
1906 
1907 static struct {
1908 	FIRMWARE_ID	id;
1909 	unsigned int	offset;
1910 	unsigned int	size;
1911 } rlc_autoload_info[FIRMWARE_ID_MAX];
1912 
1913 static int gfx_v10_0_parse_rlc_toc(struct amdgpu_device *adev)
1914 {
1915 	int ret;
1916 	RLC_TABLE_OF_CONTENT *rlc_toc;
1917 
1918 	ret = amdgpu_bo_create_reserved(adev, adev->psp.toc_bin_size, PAGE_SIZE,
1919 					AMDGPU_GEM_DOMAIN_GTT,
1920 					&adev->gfx.rlc.rlc_toc_bo,
1921 					&adev->gfx.rlc.rlc_toc_gpu_addr,
1922 					(void **)&adev->gfx.rlc.rlc_toc_buf);
1923 	if (ret) {
1924 		dev_err(adev->dev, "(%d) failed to create rlc toc bo\n", ret);
1925 		return ret;
1926 	}
1927 
1928 	/* Copy toc from psp sos fw to rlc toc buffer */
1929 	memcpy(adev->gfx.rlc.rlc_toc_buf, adev->psp.toc_start_addr, adev->psp.toc_bin_size);
1930 
1931 	rlc_toc = (RLC_TABLE_OF_CONTENT *)adev->gfx.rlc.rlc_toc_buf;
1932 	while (rlc_toc && (rlc_toc->id > FIRMWARE_ID_INVALID) &&
1933 		(rlc_toc->id < FIRMWARE_ID_MAX)) {
1934 		if ((rlc_toc->id >= FIRMWARE_ID_CP_CE) &&
1935 		    (rlc_toc->id <= FIRMWARE_ID_CP_MES)) {
1936 			/* Offset needs 4KB alignment */
1937 			rlc_toc->offset = ALIGN(rlc_toc->offset * 4, PAGE_SIZE);
1938 		}
1939 
1940 		rlc_autoload_info[rlc_toc->id].id = rlc_toc->id;
1941 		rlc_autoload_info[rlc_toc->id].offset = rlc_toc->offset * 4;
1942 		rlc_autoload_info[rlc_toc->id].size = rlc_toc->size * 4;
1943 
1944 		rlc_toc++;
1945 	};
1946 
1947 	return 0;
1948 }
1949 
1950 static uint32_t gfx_v10_0_calc_toc_total_size(struct amdgpu_device *adev)
1951 {
1952 	uint32_t total_size = 0;
1953 	FIRMWARE_ID id;
1954 	int ret;
1955 
1956 	ret = gfx_v10_0_parse_rlc_toc(adev);
1957 	if (ret) {
1958 		dev_err(adev->dev, "failed to parse rlc toc\n");
1959 		return 0;
1960 	}
1961 
1962 	for (id = FIRMWARE_ID_RLC_G_UCODE; id < FIRMWARE_ID_MAX; id++)
1963 		total_size += rlc_autoload_info[id].size;
1964 
1965 	/* In case the offset in rlc toc ucode is aligned */
1966 	if (total_size < rlc_autoload_info[FIRMWARE_ID_MAX-1].offset)
1967 		total_size = rlc_autoload_info[FIRMWARE_ID_MAX-1].offset +
1968 				rlc_autoload_info[FIRMWARE_ID_MAX-1].size;
1969 
1970 	return total_size;
1971 }
1972 
1973 static int gfx_v10_0_rlc_backdoor_autoload_buffer_init(struct amdgpu_device *adev)
1974 {
1975 	int r;
1976 	uint32_t total_size;
1977 
1978 	total_size = gfx_v10_0_calc_toc_total_size(adev);
1979 
1980 	r = amdgpu_bo_create_reserved(adev, total_size, PAGE_SIZE,
1981 				      AMDGPU_GEM_DOMAIN_GTT,
1982 				      &adev->gfx.rlc.rlc_autoload_bo,
1983 				      &adev->gfx.rlc.rlc_autoload_gpu_addr,
1984 				      (void **)&adev->gfx.rlc.rlc_autoload_ptr);
1985 	if (r) {
1986 		dev_err(adev->dev, "(%d) failed to create fw autoload bo\n", r);
1987 		return r;
1988 	}
1989 
1990 	return 0;
1991 }
1992 
1993 static void gfx_v10_0_rlc_backdoor_autoload_buffer_fini(struct amdgpu_device *adev)
1994 {
1995 	amdgpu_bo_free_kernel(&adev->gfx.rlc.rlc_toc_bo,
1996 			      &adev->gfx.rlc.rlc_toc_gpu_addr,
1997 			      (void **)&adev->gfx.rlc.rlc_toc_buf);
1998 	amdgpu_bo_free_kernel(&adev->gfx.rlc.rlc_autoload_bo,
1999 			      &adev->gfx.rlc.rlc_autoload_gpu_addr,
2000 			      (void **)&adev->gfx.rlc.rlc_autoload_ptr);
2001 }
2002 
2003 static void gfx_v10_0_rlc_backdoor_autoload_copy_ucode(struct amdgpu_device *adev,
2004 						       FIRMWARE_ID id,
2005 						       const void *fw_data,
2006 						       uint32_t fw_size)
2007 {
2008 	uint32_t toc_offset;
2009 	uint32_t toc_fw_size;
2010 	char *ptr = adev->gfx.rlc.rlc_autoload_ptr;
2011 
2012 	if (id <= FIRMWARE_ID_INVALID || id >= FIRMWARE_ID_MAX)
2013 		return;
2014 
2015 	toc_offset = rlc_autoload_info[id].offset;
2016 	toc_fw_size = rlc_autoload_info[id].size;
2017 
2018 	if (fw_size == 0)
2019 		fw_size = toc_fw_size;
2020 
2021 	if (fw_size > toc_fw_size)
2022 		fw_size = toc_fw_size;
2023 
2024 	memcpy(ptr + toc_offset, fw_data, fw_size);
2025 
2026 	if (fw_size < toc_fw_size)
2027 		memset(ptr + toc_offset + fw_size, 0, toc_fw_size - fw_size);
2028 }
2029 
2030 static void gfx_v10_0_rlc_backdoor_autoload_copy_toc_ucode(struct amdgpu_device *adev)
2031 {
2032 	void *data;
2033 	uint32_t size;
2034 
2035 	data = adev->gfx.rlc.rlc_toc_buf;
2036 	size = rlc_autoload_info[FIRMWARE_ID_RLC_TOC].size;
2037 
2038 	gfx_v10_0_rlc_backdoor_autoload_copy_ucode(adev,
2039 						   FIRMWARE_ID_RLC_TOC,
2040 						   data, size);
2041 }
2042 
2043 static void gfx_v10_0_rlc_backdoor_autoload_copy_gfx_ucode(struct amdgpu_device *adev)
2044 {
2045 	const __le32 *fw_data;
2046 	uint32_t fw_size;
2047 	const struct gfx_firmware_header_v1_0 *cp_hdr;
2048 	const struct rlc_firmware_header_v2_0 *rlc_hdr;
2049 
2050 	/* pfp ucode */
2051 	cp_hdr = (const struct gfx_firmware_header_v1_0 *)
2052 		adev->gfx.pfp_fw->data;
2053 	fw_data = (const __le32 *)(adev->gfx.pfp_fw->data +
2054 		le32_to_cpu(cp_hdr->header.ucode_array_offset_bytes));
2055 	fw_size = le32_to_cpu(cp_hdr->header.ucode_size_bytes);
2056 	gfx_v10_0_rlc_backdoor_autoload_copy_ucode(adev,
2057 						   FIRMWARE_ID_CP_PFP,
2058 						   fw_data, fw_size);
2059 
2060 	/* ce ucode */
2061 	cp_hdr = (const struct gfx_firmware_header_v1_0 *)
2062 		adev->gfx.ce_fw->data;
2063 	fw_data = (const __le32 *)(adev->gfx.ce_fw->data +
2064 		le32_to_cpu(cp_hdr->header.ucode_array_offset_bytes));
2065 	fw_size = le32_to_cpu(cp_hdr->header.ucode_size_bytes);
2066 	gfx_v10_0_rlc_backdoor_autoload_copy_ucode(adev,
2067 						   FIRMWARE_ID_CP_CE,
2068 						   fw_data, fw_size);
2069 
2070 	/* me ucode */
2071 	cp_hdr = (const struct gfx_firmware_header_v1_0 *)
2072 		adev->gfx.me_fw->data;
2073 	fw_data = (const __le32 *)(adev->gfx.me_fw->data +
2074 		le32_to_cpu(cp_hdr->header.ucode_array_offset_bytes));
2075 	fw_size = le32_to_cpu(cp_hdr->header.ucode_size_bytes);
2076 	gfx_v10_0_rlc_backdoor_autoload_copy_ucode(adev,
2077 						   FIRMWARE_ID_CP_ME,
2078 						   fw_data, fw_size);
2079 
2080 	/* rlc ucode */
2081 	rlc_hdr = (const struct rlc_firmware_header_v2_0 *)
2082 		adev->gfx.rlc_fw->data;
2083 	fw_data = (const __le32 *)(adev->gfx.rlc_fw->data +
2084 		le32_to_cpu(rlc_hdr->header.ucode_array_offset_bytes));
2085 	fw_size = le32_to_cpu(rlc_hdr->header.ucode_size_bytes);
2086 	gfx_v10_0_rlc_backdoor_autoload_copy_ucode(adev,
2087 						   FIRMWARE_ID_RLC_G_UCODE,
2088 						   fw_data, fw_size);
2089 
2090 	/* mec1 ucode */
2091 	cp_hdr = (const struct gfx_firmware_header_v1_0 *)
2092 		adev->gfx.mec_fw->data;
2093 	fw_data = (const __le32 *) (adev->gfx.mec_fw->data +
2094 		le32_to_cpu(cp_hdr->header.ucode_array_offset_bytes));
2095 	fw_size = le32_to_cpu(cp_hdr->header.ucode_size_bytes) -
2096 		cp_hdr->jt_size * 4;
2097 	gfx_v10_0_rlc_backdoor_autoload_copy_ucode(adev,
2098 						   FIRMWARE_ID_CP_MEC,
2099 						   fw_data, fw_size);
2100 	/* mec2 ucode is not necessary if mec2 ucode is same as mec1 */
2101 }
2102 
2103 /* Temporarily put sdma part here */
2104 static void gfx_v10_0_rlc_backdoor_autoload_copy_sdma_ucode(struct amdgpu_device *adev)
2105 {
2106 	const __le32 *fw_data;
2107 	uint32_t fw_size;
2108 	const struct sdma_firmware_header_v1_0 *sdma_hdr;
2109 	int i;
2110 
2111 	for (i = 0; i < adev->sdma.num_instances; i++) {
2112 		sdma_hdr = (const struct sdma_firmware_header_v1_0 *)
2113 			adev->sdma.instance[i].fw->data;
2114 		fw_data = (const __le32 *) (adev->sdma.instance[i].fw->data +
2115 			le32_to_cpu(sdma_hdr->header.ucode_array_offset_bytes));
2116 		fw_size = le32_to_cpu(sdma_hdr->header.ucode_size_bytes);
2117 
2118 		if (i == 0) {
2119 			gfx_v10_0_rlc_backdoor_autoload_copy_ucode(adev,
2120 				FIRMWARE_ID_SDMA0_UCODE, fw_data, fw_size);
2121 			gfx_v10_0_rlc_backdoor_autoload_copy_ucode(adev,
2122 				FIRMWARE_ID_SDMA0_JT,
2123 				(uint32_t *)fw_data +
2124 				sdma_hdr->jt_offset,
2125 				sdma_hdr->jt_size * 4);
2126 		} else if (i == 1) {
2127 			gfx_v10_0_rlc_backdoor_autoload_copy_ucode(adev,
2128 				FIRMWARE_ID_SDMA1_UCODE, fw_data, fw_size);
2129 			gfx_v10_0_rlc_backdoor_autoload_copy_ucode(adev,
2130 				FIRMWARE_ID_SDMA1_JT,
2131 				(uint32_t *)fw_data +
2132 				sdma_hdr->jt_offset,
2133 				sdma_hdr->jt_size * 4);
2134 		}
2135 	}
2136 }
2137 
2138 static int gfx_v10_0_rlc_backdoor_autoload_enable(struct amdgpu_device *adev)
2139 {
2140 	uint32_t rlc_g_offset, rlc_g_size, tmp;
2141 	uint64_t gpu_addr;
2142 
2143 	gfx_v10_0_rlc_backdoor_autoload_copy_toc_ucode(adev);
2144 	gfx_v10_0_rlc_backdoor_autoload_copy_sdma_ucode(adev);
2145 	gfx_v10_0_rlc_backdoor_autoload_copy_gfx_ucode(adev);
2146 
2147 	rlc_g_offset = rlc_autoload_info[FIRMWARE_ID_RLC_G_UCODE].offset;
2148 	rlc_g_size = rlc_autoload_info[FIRMWARE_ID_RLC_G_UCODE].size;
2149 	gpu_addr = adev->gfx.rlc.rlc_autoload_gpu_addr + rlc_g_offset;
2150 
2151 	WREG32_SOC15(GC, 0, mmRLC_HYP_BOOTLOAD_ADDR_HI, upper_32_bits(gpu_addr));
2152 	WREG32_SOC15(GC, 0, mmRLC_HYP_BOOTLOAD_ADDR_LO, lower_32_bits(gpu_addr));
2153 	WREG32_SOC15(GC, 0, mmRLC_HYP_BOOTLOAD_SIZE, rlc_g_size);
2154 
2155 	tmp = RREG32_SOC15(GC, 0, mmRLC_HYP_RESET_VECTOR);
2156 	if (!(tmp & (RLC_HYP_RESET_VECTOR__COLD_BOOT_EXIT_MASK |
2157 		   RLC_HYP_RESET_VECTOR__VDDGFX_EXIT_MASK))) {
2158 		DRM_ERROR("Neither COLD_BOOT_EXIT nor VDDGFX_EXIT is set\n");
2159 		return -EINVAL;
2160 	}
2161 
2162 	tmp = RREG32_SOC15(GC, 0, mmRLC_CNTL);
2163 	if (tmp & RLC_CNTL__RLC_ENABLE_F32_MASK) {
2164 		DRM_ERROR("RLC ROM should halt itself\n");
2165 		return -EINVAL;
2166 	}
2167 
2168 	return 0;
2169 }
2170 
2171 static int gfx_v10_0_rlc_backdoor_autoload_config_me_cache(struct amdgpu_device *adev)
2172 {
2173 	uint32_t usec_timeout = 50000;  /* wait for 50ms */
2174 	uint32_t tmp;
2175 	int i;
2176 	uint64_t addr;
2177 
2178 	/* Trigger an invalidation of the L1 instruction caches */
2179 	tmp = RREG32_SOC15(GC, 0, mmCP_ME_IC_OP_CNTL);
2180 	tmp = REG_SET_FIELD(tmp, CP_ME_IC_OP_CNTL, INVALIDATE_CACHE, 1);
2181 	WREG32_SOC15(GC, 0, mmCP_ME_IC_OP_CNTL, tmp);
2182 
2183 	/* Wait for invalidation complete */
2184 	for (i = 0; i < usec_timeout; i++) {
2185 		tmp = RREG32_SOC15(GC, 0, mmCP_ME_IC_OP_CNTL);
2186 		if (1 == REG_GET_FIELD(tmp, CP_ME_IC_OP_CNTL,
2187 			INVALIDATE_CACHE_COMPLETE))
2188 			break;
2189 		udelay(1);
2190 	}
2191 
2192 	if (i >= usec_timeout) {
2193 		dev_err(adev->dev, "failed to invalidate instruction cache\n");
2194 		return -EINVAL;
2195 	}
2196 
2197 	/* Program me ucode address into intruction cache address register */
2198 	addr = adev->gfx.rlc.rlc_autoload_gpu_addr +
2199 		rlc_autoload_info[FIRMWARE_ID_CP_ME].offset;
2200 	WREG32_SOC15(GC, 0, mmCP_ME_IC_BASE_LO,
2201 			lower_32_bits(addr) & 0xFFFFF000);
2202 	WREG32_SOC15(GC, 0, mmCP_ME_IC_BASE_HI,
2203 			upper_32_bits(addr));
2204 
2205 	return 0;
2206 }
2207 
2208 static int gfx_v10_0_rlc_backdoor_autoload_config_ce_cache(struct amdgpu_device *adev)
2209 {
2210 	uint32_t usec_timeout = 50000;  /* wait for 50ms */
2211 	uint32_t tmp;
2212 	int i;
2213 	uint64_t addr;
2214 
2215 	/* Trigger an invalidation of the L1 instruction caches */
2216 	tmp = RREG32_SOC15(GC, 0, mmCP_CE_IC_OP_CNTL);
2217 	tmp = REG_SET_FIELD(tmp, CP_CE_IC_OP_CNTL, INVALIDATE_CACHE, 1);
2218 	WREG32_SOC15(GC, 0, mmCP_CE_IC_OP_CNTL, tmp);
2219 
2220 	/* Wait for invalidation complete */
2221 	for (i = 0; i < usec_timeout; i++) {
2222 		tmp = RREG32_SOC15(GC, 0, mmCP_CE_IC_OP_CNTL);
2223 		if (1 == REG_GET_FIELD(tmp, CP_CE_IC_OP_CNTL,
2224 			INVALIDATE_CACHE_COMPLETE))
2225 			break;
2226 		udelay(1);
2227 	}
2228 
2229 	if (i >= usec_timeout) {
2230 		dev_err(adev->dev, "failed to invalidate instruction cache\n");
2231 		return -EINVAL;
2232 	}
2233 
2234 	/* Program ce ucode address into intruction cache address register */
2235 	addr = adev->gfx.rlc.rlc_autoload_gpu_addr +
2236 		rlc_autoload_info[FIRMWARE_ID_CP_CE].offset;
2237 	WREG32_SOC15(GC, 0, mmCP_CE_IC_BASE_LO,
2238 			lower_32_bits(addr) & 0xFFFFF000);
2239 	WREG32_SOC15(GC, 0, mmCP_CE_IC_BASE_HI,
2240 			upper_32_bits(addr));
2241 
2242 	return 0;
2243 }
2244 
2245 static int gfx_v10_0_rlc_backdoor_autoload_config_pfp_cache(struct amdgpu_device *adev)
2246 {
2247 	uint32_t usec_timeout = 50000;  /* wait for 50ms */
2248 	uint32_t tmp;
2249 	int i;
2250 	uint64_t addr;
2251 
2252 	/* Trigger an invalidation of the L1 instruction caches */
2253 	tmp = RREG32_SOC15(GC, 0, mmCP_PFP_IC_OP_CNTL);
2254 	tmp = REG_SET_FIELD(tmp, CP_PFP_IC_OP_CNTL, INVALIDATE_CACHE, 1);
2255 	WREG32_SOC15(GC, 0, mmCP_PFP_IC_OP_CNTL, tmp);
2256 
2257 	/* Wait for invalidation complete */
2258 	for (i = 0; i < usec_timeout; i++) {
2259 		tmp = RREG32_SOC15(GC, 0, mmCP_PFP_IC_OP_CNTL);
2260 		if (1 == REG_GET_FIELD(tmp, CP_PFP_IC_OP_CNTL,
2261 			INVALIDATE_CACHE_COMPLETE))
2262 			break;
2263 		udelay(1);
2264 	}
2265 
2266 	if (i >= usec_timeout) {
2267 		dev_err(adev->dev, "failed to invalidate instruction cache\n");
2268 		return -EINVAL;
2269 	}
2270 
2271 	/* Program pfp ucode address into intruction cache address register */
2272 	addr = adev->gfx.rlc.rlc_autoload_gpu_addr +
2273 		rlc_autoload_info[FIRMWARE_ID_CP_PFP].offset;
2274 	WREG32_SOC15(GC, 0, mmCP_PFP_IC_BASE_LO,
2275 			lower_32_bits(addr) & 0xFFFFF000);
2276 	WREG32_SOC15(GC, 0, mmCP_PFP_IC_BASE_HI,
2277 			upper_32_bits(addr));
2278 
2279 	return 0;
2280 }
2281 
2282 static int gfx_v10_0_rlc_backdoor_autoload_config_mec_cache(struct amdgpu_device *adev)
2283 {
2284 	uint32_t usec_timeout = 50000;  /* wait for 50ms */
2285 	uint32_t tmp;
2286 	int i;
2287 	uint64_t addr;
2288 
2289 	/* Trigger an invalidation of the L1 instruction caches */
2290 	tmp = RREG32_SOC15(GC, 0, mmCP_CPC_IC_OP_CNTL);
2291 	tmp = REG_SET_FIELD(tmp, CP_CPC_IC_OP_CNTL, INVALIDATE_CACHE, 1);
2292 	WREG32_SOC15(GC, 0, mmCP_CPC_IC_OP_CNTL, tmp);
2293 
2294 	/* Wait for invalidation complete */
2295 	for (i = 0; i < usec_timeout; i++) {
2296 		tmp = RREG32_SOC15(GC, 0, mmCP_CPC_IC_OP_CNTL);
2297 		if (1 == REG_GET_FIELD(tmp, CP_CPC_IC_OP_CNTL,
2298 			INVALIDATE_CACHE_COMPLETE))
2299 			break;
2300 		udelay(1);
2301 	}
2302 
2303 	if (i >= usec_timeout) {
2304 		dev_err(adev->dev, "failed to invalidate instruction cache\n");
2305 		return -EINVAL;
2306 	}
2307 
2308 	/* Program mec1 ucode address into intruction cache address register */
2309 	addr = adev->gfx.rlc.rlc_autoload_gpu_addr +
2310 		rlc_autoload_info[FIRMWARE_ID_CP_MEC].offset;
2311 	WREG32_SOC15(GC, 0, mmCP_CPC_IC_BASE_LO,
2312 			lower_32_bits(addr) & 0xFFFFF000);
2313 	WREG32_SOC15(GC, 0, mmCP_CPC_IC_BASE_HI,
2314 			upper_32_bits(addr));
2315 
2316 	return 0;
2317 }
2318 
2319 static int gfx_v10_0_wait_for_rlc_autoload_complete(struct amdgpu_device *adev)
2320 {
2321 	uint32_t cp_status;
2322 	uint32_t bootload_status;
2323 	int i, r;
2324 
2325 	for (i = 0; i < adev->usec_timeout; i++) {
2326 		cp_status = RREG32_SOC15(GC, 0, mmCP_STAT);
2327 		bootload_status = RREG32_SOC15(GC, 0, mmRLC_RLCS_BOOTLOAD_STATUS);
2328 		if ((cp_status == 0) &&
2329 		    (REG_GET_FIELD(bootload_status,
2330 			RLC_RLCS_BOOTLOAD_STATUS, BOOTLOAD_COMPLETE) == 1)) {
2331 			break;
2332 		}
2333 		udelay(1);
2334 	}
2335 
2336 	if (i >= adev->usec_timeout) {
2337 		dev_err(adev->dev, "rlc autoload: gc ucode autoload timeout\n");
2338 		return -ETIMEDOUT;
2339 	}
2340 
2341 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_RLC_BACKDOOR_AUTO) {
2342 		r = gfx_v10_0_rlc_backdoor_autoload_config_me_cache(adev);
2343 		if (r)
2344 			return r;
2345 
2346 		r = gfx_v10_0_rlc_backdoor_autoload_config_ce_cache(adev);
2347 		if (r)
2348 			return r;
2349 
2350 		r = gfx_v10_0_rlc_backdoor_autoload_config_pfp_cache(adev);
2351 		if (r)
2352 			return r;
2353 
2354 		r = gfx_v10_0_rlc_backdoor_autoload_config_mec_cache(adev);
2355 		if (r)
2356 			return r;
2357 	}
2358 
2359 	return 0;
2360 }
2361 
2362 static int gfx_v10_0_cp_gfx_enable(struct amdgpu_device *adev, bool enable)
2363 {
2364 	int i;
2365 	u32 tmp = RREG32_SOC15(GC, 0, mmCP_ME_CNTL);
2366 
2367 	tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, ME_HALT, enable ? 0 : 1);
2368 	tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, PFP_HALT, enable ? 0 : 1);
2369 	tmp = REG_SET_FIELD(tmp, CP_ME_CNTL, CE_HALT, enable ? 0 : 1);
2370 	if (!enable) {
2371 		for (i = 0; i < adev->gfx.num_gfx_rings; i++)
2372 			adev->gfx.gfx_ring[i].sched.ready = false;
2373 	}
2374 	WREG32_SOC15(GC, 0, mmCP_ME_CNTL, tmp);
2375 
2376 	for (i = 0; i < adev->usec_timeout; i++) {
2377 		if (RREG32_SOC15(GC, 0, mmCP_STAT) == 0)
2378 			break;
2379 		udelay(1);
2380 	}
2381 
2382 	if (i >= adev->usec_timeout)
2383 		DRM_ERROR("failed to %s cp gfx\n", enable ? "unhalt" : "halt");
2384 
2385 	return 0;
2386 }
2387 
2388 static int gfx_v10_0_cp_gfx_load_pfp_microcode(struct amdgpu_device *adev)
2389 {
2390 	int r;
2391 	const struct gfx_firmware_header_v1_0 *pfp_hdr;
2392 	const __le32 *fw_data;
2393 	unsigned i, fw_size;
2394 	uint32_t tmp;
2395 	uint32_t usec_timeout = 50000;  /* wait for 50ms */
2396 
2397 	pfp_hdr = (const struct gfx_firmware_header_v1_0 *)
2398 		adev->gfx.pfp_fw->data;
2399 
2400 	amdgpu_ucode_print_gfx_hdr(&pfp_hdr->header);
2401 
2402 	fw_data = (const __le32 *)(adev->gfx.pfp_fw->data +
2403 		le32_to_cpu(pfp_hdr->header.ucode_array_offset_bytes));
2404 	fw_size = le32_to_cpu(pfp_hdr->header.ucode_size_bytes);
2405 
2406 	r = amdgpu_bo_create_reserved(adev, pfp_hdr->header.ucode_size_bytes,
2407 				      PAGE_SIZE, AMDGPU_GEM_DOMAIN_GTT,
2408 				      &adev->gfx.pfp.pfp_fw_obj,
2409 				      &adev->gfx.pfp.pfp_fw_gpu_addr,
2410 				      (void **)&adev->gfx.pfp.pfp_fw_ptr);
2411 	if (r) {
2412 		dev_err(adev->dev, "(%d) failed to create pfp fw bo\n", r);
2413 		gfx_v10_0_pfp_fini(adev);
2414 		return r;
2415 	}
2416 
2417 	memcpy(adev->gfx.pfp.pfp_fw_ptr, fw_data, fw_size);
2418 
2419 	amdgpu_bo_kunmap(adev->gfx.pfp.pfp_fw_obj);
2420 	amdgpu_bo_unreserve(adev->gfx.pfp.pfp_fw_obj);
2421 
2422 	/* Trigger an invalidation of the L1 instruction caches */
2423 	tmp = RREG32_SOC15(GC, 0, mmCP_PFP_IC_OP_CNTL);
2424 	tmp = REG_SET_FIELD(tmp, CP_PFP_IC_OP_CNTL, INVALIDATE_CACHE, 1);
2425 	WREG32_SOC15(GC, 0, mmCP_PFP_IC_OP_CNTL, tmp);
2426 
2427 	/* Wait for invalidation complete */
2428 	for (i = 0; i < usec_timeout; i++) {
2429 		tmp = RREG32_SOC15(GC, 0, mmCP_PFP_IC_OP_CNTL);
2430 		if (1 == REG_GET_FIELD(tmp, CP_PFP_IC_OP_CNTL,
2431 			INVALIDATE_CACHE_COMPLETE))
2432 			break;
2433 		udelay(1);
2434 	}
2435 
2436 	if (i >= usec_timeout) {
2437 		dev_err(adev->dev, "failed to invalidate instruction cache\n");
2438 		return -EINVAL;
2439 	}
2440 
2441 	if (amdgpu_emu_mode == 1)
2442 		adev->nbio.funcs->hdp_flush(adev, NULL);
2443 
2444 	tmp = RREG32_SOC15(GC, 0, mmCP_PFP_IC_BASE_CNTL);
2445 	tmp = REG_SET_FIELD(tmp, CP_PFP_IC_BASE_CNTL, VMID, 0);
2446 	tmp = REG_SET_FIELD(tmp, CP_PFP_IC_BASE_CNTL, CACHE_POLICY, 0);
2447 	tmp = REG_SET_FIELD(tmp, CP_PFP_IC_BASE_CNTL, EXE_DISABLE, 0);
2448 	tmp = REG_SET_FIELD(tmp, CP_PFP_IC_BASE_CNTL, ADDRESS_CLAMP, 1);
2449 	WREG32_SOC15(GC, 0, mmCP_PFP_IC_BASE_CNTL, tmp);
2450 	WREG32_SOC15(GC, 0, mmCP_PFP_IC_BASE_LO,
2451 		adev->gfx.pfp.pfp_fw_gpu_addr & 0xFFFFF000);
2452 	WREG32_SOC15(GC, 0, mmCP_PFP_IC_BASE_HI,
2453 		upper_32_bits(adev->gfx.pfp.pfp_fw_gpu_addr));
2454 
2455 	return 0;
2456 }
2457 
2458 static int gfx_v10_0_cp_gfx_load_ce_microcode(struct amdgpu_device *adev)
2459 {
2460 	int r;
2461 	const struct gfx_firmware_header_v1_0 *ce_hdr;
2462 	const __le32 *fw_data;
2463 	unsigned i, fw_size;
2464 	uint32_t tmp;
2465 	uint32_t usec_timeout = 50000;  /* wait for 50ms */
2466 
2467 	ce_hdr = (const struct gfx_firmware_header_v1_0 *)
2468 		adev->gfx.ce_fw->data;
2469 
2470 	amdgpu_ucode_print_gfx_hdr(&ce_hdr->header);
2471 
2472 	fw_data = (const __le32 *)(adev->gfx.ce_fw->data +
2473 		le32_to_cpu(ce_hdr->header.ucode_array_offset_bytes));
2474 	fw_size = le32_to_cpu(ce_hdr->header.ucode_size_bytes);
2475 
2476 	r = amdgpu_bo_create_reserved(adev, ce_hdr->header.ucode_size_bytes,
2477 				      PAGE_SIZE, AMDGPU_GEM_DOMAIN_GTT,
2478 				      &adev->gfx.ce.ce_fw_obj,
2479 				      &adev->gfx.ce.ce_fw_gpu_addr,
2480 				      (void **)&adev->gfx.ce.ce_fw_ptr);
2481 	if (r) {
2482 		dev_err(adev->dev, "(%d) failed to create ce fw bo\n", r);
2483 		gfx_v10_0_ce_fini(adev);
2484 		return r;
2485 	}
2486 
2487 	memcpy(adev->gfx.ce.ce_fw_ptr, fw_data, fw_size);
2488 
2489 	amdgpu_bo_kunmap(adev->gfx.ce.ce_fw_obj);
2490 	amdgpu_bo_unreserve(adev->gfx.ce.ce_fw_obj);
2491 
2492 	/* Trigger an invalidation of the L1 instruction caches */
2493 	tmp = RREG32_SOC15(GC, 0, mmCP_CE_IC_OP_CNTL);
2494 	tmp = REG_SET_FIELD(tmp, CP_CE_IC_OP_CNTL, INVALIDATE_CACHE, 1);
2495 	WREG32_SOC15(GC, 0, mmCP_CE_IC_OP_CNTL, tmp);
2496 
2497 	/* Wait for invalidation complete */
2498 	for (i = 0; i < usec_timeout; i++) {
2499 		tmp = RREG32_SOC15(GC, 0, mmCP_CE_IC_OP_CNTL);
2500 		if (1 == REG_GET_FIELD(tmp, CP_CE_IC_OP_CNTL,
2501 			INVALIDATE_CACHE_COMPLETE))
2502 			break;
2503 		udelay(1);
2504 	}
2505 
2506 	if (i >= usec_timeout) {
2507 		dev_err(adev->dev, "failed to invalidate instruction cache\n");
2508 		return -EINVAL;
2509 	}
2510 
2511 	if (amdgpu_emu_mode == 1)
2512 		adev->nbio.funcs->hdp_flush(adev, NULL);
2513 
2514 	tmp = RREG32_SOC15(GC, 0, mmCP_CE_IC_BASE_CNTL);
2515 	tmp = REG_SET_FIELD(tmp, CP_CE_IC_BASE_CNTL, VMID, 0);
2516 	tmp = REG_SET_FIELD(tmp, CP_CE_IC_BASE_CNTL, CACHE_POLICY, 0);
2517 	tmp = REG_SET_FIELD(tmp, CP_CE_IC_BASE_CNTL, EXE_DISABLE, 0);
2518 	tmp = REG_SET_FIELD(tmp, CP_CE_IC_BASE_CNTL, ADDRESS_CLAMP, 1);
2519 	WREG32_SOC15(GC, 0, mmCP_CE_IC_BASE_LO,
2520 		adev->gfx.ce.ce_fw_gpu_addr & 0xFFFFF000);
2521 	WREG32_SOC15(GC, 0, mmCP_CE_IC_BASE_HI,
2522 		upper_32_bits(adev->gfx.ce.ce_fw_gpu_addr));
2523 
2524 	return 0;
2525 }
2526 
2527 static int gfx_v10_0_cp_gfx_load_me_microcode(struct amdgpu_device *adev)
2528 {
2529 	int r;
2530 	const struct gfx_firmware_header_v1_0 *me_hdr;
2531 	const __le32 *fw_data;
2532 	unsigned i, fw_size;
2533 	uint32_t tmp;
2534 	uint32_t usec_timeout = 50000;  /* wait for 50ms */
2535 
2536 	me_hdr = (const struct gfx_firmware_header_v1_0 *)
2537 		adev->gfx.me_fw->data;
2538 
2539 	amdgpu_ucode_print_gfx_hdr(&me_hdr->header);
2540 
2541 	fw_data = (const __le32 *)(adev->gfx.me_fw->data +
2542 		le32_to_cpu(me_hdr->header.ucode_array_offset_bytes));
2543 	fw_size = le32_to_cpu(me_hdr->header.ucode_size_bytes);
2544 
2545 	r = amdgpu_bo_create_reserved(adev, me_hdr->header.ucode_size_bytes,
2546 				      PAGE_SIZE, AMDGPU_GEM_DOMAIN_GTT,
2547 				      &adev->gfx.me.me_fw_obj,
2548 				      &adev->gfx.me.me_fw_gpu_addr,
2549 				      (void **)&adev->gfx.me.me_fw_ptr);
2550 	if (r) {
2551 		dev_err(adev->dev, "(%d) failed to create me fw bo\n", r);
2552 		gfx_v10_0_me_fini(adev);
2553 		return r;
2554 	}
2555 
2556 	memcpy(adev->gfx.me.me_fw_ptr, fw_data, fw_size);
2557 
2558 	amdgpu_bo_kunmap(adev->gfx.me.me_fw_obj);
2559 	amdgpu_bo_unreserve(adev->gfx.me.me_fw_obj);
2560 
2561 	/* Trigger an invalidation of the L1 instruction caches */
2562 	tmp = RREG32_SOC15(GC, 0, mmCP_ME_IC_OP_CNTL);
2563 	tmp = REG_SET_FIELD(tmp, CP_ME_IC_OP_CNTL, INVALIDATE_CACHE, 1);
2564 	WREG32_SOC15(GC, 0, mmCP_ME_IC_OP_CNTL, tmp);
2565 
2566 	/* Wait for invalidation complete */
2567 	for (i = 0; i < usec_timeout; i++) {
2568 		tmp = RREG32_SOC15(GC, 0, mmCP_ME_IC_OP_CNTL);
2569 		if (1 == REG_GET_FIELD(tmp, CP_ME_IC_OP_CNTL,
2570 			INVALIDATE_CACHE_COMPLETE))
2571 			break;
2572 		udelay(1);
2573 	}
2574 
2575 	if (i >= usec_timeout) {
2576 		dev_err(adev->dev, "failed to invalidate instruction cache\n");
2577 		return -EINVAL;
2578 	}
2579 
2580 	if (amdgpu_emu_mode == 1)
2581 		adev->nbio.funcs->hdp_flush(adev, NULL);
2582 
2583 	tmp = RREG32_SOC15(GC, 0, mmCP_ME_IC_BASE_CNTL);
2584 	tmp = REG_SET_FIELD(tmp, CP_ME_IC_BASE_CNTL, VMID, 0);
2585 	tmp = REG_SET_FIELD(tmp, CP_ME_IC_BASE_CNTL, CACHE_POLICY, 0);
2586 	tmp = REG_SET_FIELD(tmp, CP_ME_IC_BASE_CNTL, EXE_DISABLE, 0);
2587 	tmp = REG_SET_FIELD(tmp, CP_ME_IC_BASE_CNTL, ADDRESS_CLAMP, 1);
2588 	WREG32_SOC15(GC, 0, mmCP_ME_IC_BASE_LO,
2589 		adev->gfx.me.me_fw_gpu_addr & 0xFFFFF000);
2590 	WREG32_SOC15(GC, 0, mmCP_ME_IC_BASE_HI,
2591 		upper_32_bits(adev->gfx.me.me_fw_gpu_addr));
2592 
2593 	return 0;
2594 }
2595 
2596 static int gfx_v10_0_cp_gfx_load_microcode(struct amdgpu_device *adev)
2597 {
2598 	int r;
2599 
2600 	if (!adev->gfx.me_fw || !adev->gfx.pfp_fw || !adev->gfx.ce_fw)
2601 		return -EINVAL;
2602 
2603 	gfx_v10_0_cp_gfx_enable(adev, false);
2604 
2605 	r = gfx_v10_0_cp_gfx_load_pfp_microcode(adev);
2606 	if (r) {
2607 		dev_err(adev->dev, "(%d) failed to load pfp fw\n", r);
2608 		return r;
2609 	}
2610 
2611 	r = gfx_v10_0_cp_gfx_load_ce_microcode(adev);
2612 	if (r) {
2613 		dev_err(adev->dev, "(%d) failed to load ce fw\n", r);
2614 		return r;
2615 	}
2616 
2617 	r = gfx_v10_0_cp_gfx_load_me_microcode(adev);
2618 	if (r) {
2619 		dev_err(adev->dev, "(%d) failed to load me fw\n", r);
2620 		return r;
2621 	}
2622 
2623 	return 0;
2624 }
2625 
2626 static int gfx_v10_0_cp_gfx_start(struct amdgpu_device *adev)
2627 {
2628 	struct amdgpu_ring *ring;
2629 	const struct cs_section_def *sect = NULL;
2630 	const struct cs_extent_def *ext = NULL;
2631 	int r, i;
2632 	int ctx_reg_offset;
2633 
2634 	/* init the CP */
2635 	WREG32_SOC15(GC, 0, mmCP_MAX_CONTEXT,
2636 		     adev->gfx.config.max_hw_contexts - 1);
2637 	WREG32_SOC15(GC, 0, mmCP_DEVICE_ID, 1);
2638 
2639 	gfx_v10_0_cp_gfx_enable(adev, true);
2640 
2641 	ring = &adev->gfx.gfx_ring[0];
2642 	r = amdgpu_ring_alloc(ring, gfx_v10_0_get_csb_size(adev) + 4);
2643 	if (r) {
2644 		DRM_ERROR("amdgpu: cp failed to lock ring (%d).\n", r);
2645 		return r;
2646 	}
2647 
2648 	amdgpu_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
2649 	amdgpu_ring_write(ring, PACKET3_PREAMBLE_BEGIN_CLEAR_STATE);
2650 
2651 	amdgpu_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1));
2652 	amdgpu_ring_write(ring, 0x80000000);
2653 	amdgpu_ring_write(ring, 0x80000000);
2654 
2655 	for (sect = gfx10_cs_data; sect->section != NULL; ++sect) {
2656 		for (ext = sect->section; ext->extent != NULL; ++ext) {
2657 			if (sect->id == SECT_CONTEXT) {
2658 				amdgpu_ring_write(ring,
2659 						  PACKET3(PACKET3_SET_CONTEXT_REG,
2660 							  ext->reg_count));
2661 				amdgpu_ring_write(ring, ext->reg_index -
2662 						  PACKET3_SET_CONTEXT_REG_START);
2663 				for (i = 0; i < ext->reg_count; i++)
2664 					amdgpu_ring_write(ring, ext->extent[i]);
2665 			}
2666 		}
2667 	}
2668 
2669 	ctx_reg_offset =
2670 		SOC15_REG_OFFSET(GC, 0, mmPA_SC_TILE_STEERING_OVERRIDE) - PACKET3_SET_CONTEXT_REG_START;
2671 	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONTEXT_REG, 1));
2672 	amdgpu_ring_write(ring, ctx_reg_offset);
2673 	amdgpu_ring_write(ring, adev->gfx.config.pa_sc_tile_steering_override);
2674 
2675 	amdgpu_ring_write(ring, PACKET3(PACKET3_PREAMBLE_CNTL, 0));
2676 	amdgpu_ring_write(ring, PACKET3_PREAMBLE_END_CLEAR_STATE);
2677 
2678 	amdgpu_ring_write(ring, PACKET3(PACKET3_CLEAR_STATE, 0));
2679 	amdgpu_ring_write(ring, 0);
2680 
2681 	amdgpu_ring_write(ring, PACKET3(PACKET3_SET_BASE, 2));
2682 	amdgpu_ring_write(ring, PACKET3_BASE_INDEX(CE_PARTITION_BASE));
2683 	amdgpu_ring_write(ring, 0x8000);
2684 	amdgpu_ring_write(ring, 0x8000);
2685 
2686 	amdgpu_ring_commit(ring);
2687 
2688 	/* submit cs packet to copy state 0 to next available state */
2689 	ring = &adev->gfx.gfx_ring[1];
2690 	r = amdgpu_ring_alloc(ring, 2);
2691 	if (r) {
2692 		DRM_ERROR("amdgpu: cp failed to lock ring (%d).\n", r);
2693 		return r;
2694 	}
2695 
2696 	amdgpu_ring_write(ring, PACKET3(PACKET3_CLEAR_STATE, 0));
2697 	amdgpu_ring_write(ring, 0);
2698 
2699 	amdgpu_ring_commit(ring);
2700 
2701 	return 0;
2702 }
2703 
2704 static void gfx_v10_0_cp_gfx_switch_pipe(struct amdgpu_device *adev,
2705 					 CP_PIPE_ID pipe)
2706 {
2707 	u32 tmp;
2708 
2709 	tmp = RREG32_SOC15(GC, 0, mmGRBM_GFX_CNTL);
2710 	tmp = REG_SET_FIELD(tmp, GRBM_GFX_CNTL, PIPEID, pipe);
2711 
2712 	WREG32_SOC15(GC, 0, mmGRBM_GFX_CNTL, tmp);
2713 }
2714 
2715 static void gfx_v10_0_cp_gfx_set_doorbell(struct amdgpu_device *adev,
2716 					  struct amdgpu_ring *ring)
2717 {
2718 	u32 tmp;
2719 
2720 	tmp = RREG32_SOC15(GC, 0, mmCP_RB_DOORBELL_CONTROL);
2721 	if (ring->use_doorbell) {
2722 		tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL,
2723 				    DOORBELL_OFFSET, ring->doorbell_index);
2724 		tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL,
2725 				    DOORBELL_EN, 1);
2726 	} else {
2727 		tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL,
2728 				    DOORBELL_EN, 0);
2729 	}
2730 	WREG32_SOC15(GC, 0, mmCP_RB_DOORBELL_CONTROL, tmp);
2731 	tmp = REG_SET_FIELD(0, CP_RB_DOORBELL_RANGE_LOWER,
2732 			    DOORBELL_RANGE_LOWER, ring->doorbell_index);
2733 	WREG32_SOC15(GC, 0, mmCP_RB_DOORBELL_RANGE_LOWER, tmp);
2734 
2735 	WREG32_SOC15(GC, 0, mmCP_RB_DOORBELL_RANGE_UPPER,
2736 		     CP_RB_DOORBELL_RANGE_UPPER__DOORBELL_RANGE_UPPER_MASK);
2737 }
2738 
2739 static int gfx_v10_0_cp_gfx_resume(struct amdgpu_device *adev)
2740 {
2741 	struct amdgpu_ring *ring;
2742 	u32 tmp;
2743 	u32 rb_bufsz;
2744 	u64 rb_addr, rptr_addr, wptr_gpu_addr;
2745 	u32 i;
2746 
2747 	/* Set the write pointer delay */
2748 	WREG32_SOC15(GC, 0, mmCP_RB_WPTR_DELAY, 0);
2749 
2750 	/* set the RB to use vmid 0 */
2751 	WREG32_SOC15(GC, 0, mmCP_RB_VMID, 0);
2752 
2753 	/* Init gfx ring 0 for pipe 0 */
2754 	mutex_lock(&adev->srbm_mutex);
2755 	gfx_v10_0_cp_gfx_switch_pipe(adev, PIPE_ID0);
2756 
2757 	/* Set ring buffer size */
2758 	ring = &adev->gfx.gfx_ring[0];
2759 	rb_bufsz = order_base_2(ring->ring_size / 8);
2760 	tmp = REG_SET_FIELD(0, CP_RB0_CNTL, RB_BUFSZ, rb_bufsz);
2761 	tmp = REG_SET_FIELD(tmp, CP_RB0_CNTL, RB_BLKSZ, rb_bufsz - 2);
2762 #ifdef __BIG_ENDIAN
2763 	tmp = REG_SET_FIELD(tmp, CP_RB0_CNTL, BUF_SWAP, 1);
2764 #endif
2765 	WREG32_SOC15(GC, 0, mmCP_RB0_CNTL, tmp);
2766 
2767 	/* Initialize the ring buffer's write pointers */
2768 	ring->wptr = 0;
2769 	WREG32_SOC15(GC, 0, mmCP_RB0_WPTR, lower_32_bits(ring->wptr));
2770 	WREG32_SOC15(GC, 0, mmCP_RB0_WPTR_HI, upper_32_bits(ring->wptr));
2771 
2772 	/* set the wb address wether it's enabled or not */
2773 	rptr_addr = adev->wb.gpu_addr + (ring->rptr_offs * 4);
2774 	WREG32_SOC15(GC, 0, mmCP_RB0_RPTR_ADDR, lower_32_bits(rptr_addr));
2775 	WREG32_SOC15(GC, 0, mmCP_RB0_RPTR_ADDR_HI, upper_32_bits(rptr_addr) &
2776 		     CP_RB_RPTR_ADDR_HI__RB_RPTR_ADDR_HI_MASK);
2777 
2778 	wptr_gpu_addr = adev->wb.gpu_addr + (ring->wptr_offs * 4);
2779 	WREG32_SOC15(GC, 0, mmCP_RB_WPTR_POLL_ADDR_LO,
2780 		     lower_32_bits(wptr_gpu_addr));
2781 	WREG32_SOC15(GC, 0, mmCP_RB_WPTR_POLL_ADDR_HI,
2782 		     upper_32_bits(wptr_gpu_addr));
2783 
2784 	mdelay(1);
2785 	WREG32_SOC15(GC, 0, mmCP_RB0_CNTL, tmp);
2786 
2787 	rb_addr = ring->gpu_addr >> 8;
2788 	WREG32_SOC15(GC, 0, mmCP_RB0_BASE, rb_addr);
2789 	WREG32_SOC15(GC, 0, mmCP_RB0_BASE_HI, upper_32_bits(rb_addr));
2790 
2791 	WREG32_SOC15(GC, 0, mmCP_RB_ACTIVE, 1);
2792 
2793 	gfx_v10_0_cp_gfx_set_doorbell(adev, ring);
2794 	mutex_unlock(&adev->srbm_mutex);
2795 
2796 	/* Init gfx ring 1 for pipe 1 */
2797 	mutex_lock(&adev->srbm_mutex);
2798 	gfx_v10_0_cp_gfx_switch_pipe(adev, PIPE_ID1);
2799 	ring = &adev->gfx.gfx_ring[1];
2800 	rb_bufsz = order_base_2(ring->ring_size / 8);
2801 	tmp = REG_SET_FIELD(0, CP_RB1_CNTL, RB_BUFSZ, rb_bufsz);
2802 	tmp = REG_SET_FIELD(tmp, CP_RB1_CNTL, RB_BLKSZ, rb_bufsz - 2);
2803 	WREG32_SOC15(GC, 0, mmCP_RB1_CNTL, tmp);
2804 	/* Initialize the ring buffer's write pointers */
2805 	ring->wptr = 0;
2806 	WREG32_SOC15(GC, 0, mmCP_RB1_WPTR, lower_32_bits(ring->wptr));
2807 	WREG32_SOC15(GC, 0, mmCP_RB1_WPTR_HI, upper_32_bits(ring->wptr));
2808 	/* Set the wb address wether it's enabled or not */
2809 	rptr_addr = adev->wb.gpu_addr + (ring->rptr_offs * 4);
2810 	WREG32_SOC15(GC, 0, mmCP_RB1_RPTR_ADDR, lower_32_bits(rptr_addr));
2811 	WREG32_SOC15(GC, 0, mmCP_RB1_RPTR_ADDR_HI, upper_32_bits(rptr_addr) &
2812 		CP_RB1_RPTR_ADDR_HI__RB_RPTR_ADDR_HI_MASK);
2813 	wptr_gpu_addr = adev->wb.gpu_addr + (ring->wptr_offs * 4);
2814 	WREG32_SOC15(GC, 0, mmCP_RB_WPTR_POLL_ADDR_LO,
2815 		lower_32_bits(wptr_gpu_addr));
2816 	WREG32_SOC15(GC, 0, mmCP_RB_WPTR_POLL_ADDR_HI,
2817 		upper_32_bits(wptr_gpu_addr));
2818 
2819 	mdelay(1);
2820 	WREG32_SOC15(GC, 0, mmCP_RB1_CNTL, tmp);
2821 
2822 	rb_addr = ring->gpu_addr >> 8;
2823 	WREG32_SOC15(GC, 0, mmCP_RB1_BASE, rb_addr);
2824 	WREG32_SOC15(GC, 0, mmCP_RB1_BASE_HI, upper_32_bits(rb_addr));
2825 	WREG32_SOC15(GC, 0, mmCP_RB1_ACTIVE, 1);
2826 
2827 	gfx_v10_0_cp_gfx_set_doorbell(adev, ring);
2828 	mutex_unlock(&adev->srbm_mutex);
2829 
2830 	/* Switch to pipe 0 */
2831 	mutex_lock(&adev->srbm_mutex);
2832 	gfx_v10_0_cp_gfx_switch_pipe(adev, PIPE_ID0);
2833 	mutex_unlock(&adev->srbm_mutex);
2834 
2835 	/* start the ring */
2836 	gfx_v10_0_cp_gfx_start(adev);
2837 
2838 	for (i = 0; i < adev->gfx.num_gfx_rings; i++) {
2839 		ring = &adev->gfx.gfx_ring[i];
2840 		ring->sched.ready = true;
2841 	}
2842 
2843 	return 0;
2844 }
2845 
2846 static void gfx_v10_0_cp_compute_enable(struct amdgpu_device *adev, bool enable)
2847 {
2848 	int i;
2849 
2850 	if (enable) {
2851 		WREG32_SOC15(GC, 0, mmCP_MEC_CNTL, 0);
2852 	} else {
2853 		WREG32_SOC15(GC, 0, mmCP_MEC_CNTL,
2854 			     (CP_MEC_CNTL__MEC_ME1_HALT_MASK |
2855 			      CP_MEC_CNTL__MEC_ME2_HALT_MASK));
2856 		for (i = 0; i < adev->gfx.num_compute_rings; i++)
2857 			adev->gfx.compute_ring[i].sched.ready = false;
2858 		adev->gfx.kiq.ring.sched.ready = false;
2859 	}
2860 	udelay(50);
2861 }
2862 
2863 static int gfx_v10_0_cp_compute_load_microcode(struct amdgpu_device *adev)
2864 {
2865 	const struct gfx_firmware_header_v1_0 *mec_hdr;
2866 	const __le32 *fw_data;
2867 	unsigned i;
2868 	u32 tmp;
2869 	u32 usec_timeout = 50000; /* Wait for 50 ms */
2870 
2871 	if (!adev->gfx.mec_fw)
2872 		return -EINVAL;
2873 
2874 	gfx_v10_0_cp_compute_enable(adev, false);
2875 
2876 	mec_hdr = (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data;
2877 	amdgpu_ucode_print_gfx_hdr(&mec_hdr->header);
2878 
2879 	fw_data = (const __le32 *)
2880 		(adev->gfx.mec_fw->data +
2881 		 le32_to_cpu(mec_hdr->header.ucode_array_offset_bytes));
2882 
2883 	/* Trigger an invalidation of the L1 instruction caches */
2884 	tmp = RREG32_SOC15(GC, 0, mmCP_CPC_IC_OP_CNTL);
2885 	tmp = REG_SET_FIELD(tmp, CP_CPC_IC_OP_CNTL, INVALIDATE_CACHE, 1);
2886 	WREG32_SOC15(GC, 0, mmCP_CPC_IC_OP_CNTL, tmp);
2887 
2888 	/* Wait for invalidation complete */
2889 	for (i = 0; i < usec_timeout; i++) {
2890 		tmp = RREG32_SOC15(GC, 0, mmCP_CPC_IC_OP_CNTL);
2891 		if (1 == REG_GET_FIELD(tmp, CP_CPC_IC_OP_CNTL,
2892 				       INVALIDATE_CACHE_COMPLETE))
2893 			break;
2894 		udelay(1);
2895 	}
2896 
2897 	if (i >= usec_timeout) {
2898 		dev_err(adev->dev, "failed to invalidate instruction cache\n");
2899 		return -EINVAL;
2900 	}
2901 
2902 	if (amdgpu_emu_mode == 1)
2903 		adev->nbio.funcs->hdp_flush(adev, NULL);
2904 
2905 	tmp = RREG32_SOC15(GC, 0, mmCP_CPC_IC_BASE_CNTL);
2906 	tmp = REG_SET_FIELD(tmp, CP_CPC_IC_BASE_CNTL, CACHE_POLICY, 0);
2907 	tmp = REG_SET_FIELD(tmp, CP_CPC_IC_BASE_CNTL, EXE_DISABLE, 0);
2908 	tmp = REG_SET_FIELD(tmp, CP_CPC_IC_BASE_CNTL, ADDRESS_CLAMP, 1);
2909 	WREG32_SOC15(GC, 0, mmCP_CPC_IC_BASE_CNTL, tmp);
2910 
2911 	WREG32_SOC15(GC, 0, mmCP_CPC_IC_BASE_LO, adev->gfx.mec.mec_fw_gpu_addr &
2912 		     0xFFFFF000);
2913 	WREG32_SOC15(GC, 0, mmCP_CPC_IC_BASE_HI,
2914 		     upper_32_bits(adev->gfx.mec.mec_fw_gpu_addr));
2915 
2916 	/* MEC1 */
2917 	WREG32_SOC15(GC, 0, mmCP_MEC_ME1_UCODE_ADDR, 0);
2918 
2919 	for (i = 0; i < mec_hdr->jt_size; i++)
2920 		WREG32_SOC15(GC, 0, mmCP_MEC_ME1_UCODE_DATA,
2921 			     le32_to_cpup(fw_data + mec_hdr->jt_offset + i));
2922 
2923 	WREG32_SOC15(GC, 0, mmCP_MEC_ME1_UCODE_ADDR, adev->gfx.mec_fw_version);
2924 
2925 	/*
2926 	 * TODO: Loading MEC2 firmware is only necessary if MEC2 should run
2927 	 * different microcode than MEC1.
2928 	 */
2929 
2930 	return 0;
2931 }
2932 
2933 static void gfx_v10_0_kiq_setting(struct amdgpu_ring *ring)
2934 {
2935 	uint32_t tmp;
2936 	struct amdgpu_device *adev = ring->adev;
2937 
2938 	/* tell RLC which is KIQ queue */
2939 	tmp = RREG32_SOC15(GC, 0, mmRLC_CP_SCHEDULERS);
2940 	tmp &= 0xffffff00;
2941 	tmp |= (ring->me << 5) | (ring->pipe << 3) | (ring->queue);
2942 	WREG32_SOC15(GC, 0, mmRLC_CP_SCHEDULERS, tmp);
2943 	tmp |= 0x80;
2944 	WREG32_SOC15(GC, 0, mmRLC_CP_SCHEDULERS, tmp);
2945 }
2946 
2947 static int gfx_v10_0_gfx_mqd_init(struct amdgpu_ring *ring)
2948 {
2949 	struct amdgpu_device *adev = ring->adev;
2950 	struct v10_gfx_mqd *mqd = ring->mqd_ptr;
2951 	uint64_t hqd_gpu_addr, wb_gpu_addr;
2952 	uint32_t tmp;
2953 	uint32_t rb_bufsz;
2954 
2955 	/* set up gfx hqd wptr */
2956 	mqd->cp_gfx_hqd_wptr = 0;
2957 	mqd->cp_gfx_hqd_wptr_hi = 0;
2958 
2959 	/* set the pointer to the MQD */
2960 	mqd->cp_mqd_base_addr = ring->mqd_gpu_addr & 0xfffffffc;
2961 	mqd->cp_mqd_base_addr_hi = upper_32_bits(ring->mqd_gpu_addr);
2962 
2963 	/* set up mqd control */
2964 	tmp = RREG32_SOC15(GC, 0, mmCP_GFX_MQD_CONTROL);
2965 	tmp = REG_SET_FIELD(tmp, CP_GFX_MQD_CONTROL, VMID, 0);
2966 	tmp = REG_SET_FIELD(tmp, CP_GFX_MQD_CONTROL, PRIV_STATE, 1);
2967 	tmp = REG_SET_FIELD(tmp, CP_GFX_MQD_CONTROL, CACHE_POLICY, 0);
2968 	mqd->cp_gfx_mqd_control = tmp;
2969 
2970 	/* set up gfx_hqd_vimd with 0x0 to indicate the ring buffer's vmid */
2971 	tmp = RREG32_SOC15(GC, 0, mmCP_GFX_HQD_VMID);
2972 	tmp = REG_SET_FIELD(tmp, CP_GFX_HQD_VMID, VMID, 0);
2973 	mqd->cp_gfx_hqd_vmid = 0;
2974 
2975 	/* set up default queue priority level
2976 	 * 0x0 = low priority, 0x1 = high priority */
2977 	tmp = RREG32_SOC15(GC, 0, mmCP_GFX_HQD_QUEUE_PRIORITY);
2978 	tmp = REG_SET_FIELD(tmp, CP_GFX_HQD_QUEUE_PRIORITY, PRIORITY_LEVEL, 0);
2979 	mqd->cp_gfx_hqd_queue_priority = tmp;
2980 
2981 	/* set up time quantum */
2982 	tmp = RREG32_SOC15(GC, 0, mmCP_GFX_HQD_QUANTUM);
2983 	tmp = REG_SET_FIELD(tmp, CP_GFX_HQD_QUANTUM, QUANTUM_EN, 1);
2984 	mqd->cp_gfx_hqd_quantum = tmp;
2985 
2986 	/* set up gfx hqd base. this is similar as CP_RB_BASE */
2987 	hqd_gpu_addr = ring->gpu_addr >> 8;
2988 	mqd->cp_gfx_hqd_base = hqd_gpu_addr;
2989 	mqd->cp_gfx_hqd_base_hi = upper_32_bits(hqd_gpu_addr);
2990 
2991 	/* set up hqd_rptr_addr/_hi, similar as CP_RB_RPTR */
2992 	wb_gpu_addr = adev->wb.gpu_addr + (ring->rptr_offs * 4);
2993 	mqd->cp_gfx_hqd_rptr_addr = wb_gpu_addr & 0xfffffffc;
2994 	mqd->cp_gfx_hqd_rptr_addr_hi =
2995 		upper_32_bits(wb_gpu_addr) & 0xffff;
2996 
2997 	/* set up rb_wptr_poll addr */
2998 	wb_gpu_addr = adev->wb.gpu_addr + (ring->wptr_offs * 4);
2999 	mqd->cp_rb_wptr_poll_addr_lo = wb_gpu_addr & 0xfffffffc;
3000 	mqd->cp_rb_wptr_poll_addr_hi = upper_32_bits(wb_gpu_addr) & 0xffff;
3001 
3002 	/* set up the gfx_hqd_control, similar as CP_RB0_CNTL */
3003 	rb_bufsz = order_base_2(ring->ring_size / 4) - 1;
3004 	tmp = RREG32_SOC15(GC, 0, mmCP_GFX_HQD_CNTL);
3005 	tmp = REG_SET_FIELD(tmp, CP_GFX_HQD_CNTL, RB_BUFSZ, rb_bufsz);
3006 	tmp = REG_SET_FIELD(tmp, CP_GFX_HQD_CNTL, RB_BLKSZ, rb_bufsz - 2);
3007 #ifdef __BIG_ENDIAN
3008 	tmp = REG_SET_FIELD(tmp, CP_GFX_HQD_CNTL, BUF_SWAP, 1);
3009 #endif
3010 	mqd->cp_gfx_hqd_cntl = tmp;
3011 
3012 	/* set up cp_doorbell_control */
3013 	tmp = RREG32_SOC15(GC, 0, mmCP_RB_DOORBELL_CONTROL);
3014 	if (ring->use_doorbell) {
3015 		tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL,
3016 				    DOORBELL_OFFSET, ring->doorbell_index);
3017 		tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL,
3018 				    DOORBELL_EN, 1);
3019 	} else
3020 		tmp = REG_SET_FIELD(tmp, CP_RB_DOORBELL_CONTROL,
3021 				    DOORBELL_EN, 0);
3022 	mqd->cp_rb_doorbell_control = tmp;
3023 
3024 	/* reset read and write pointers, similar to CP_RB0_WPTR/_RPTR */
3025 	ring->wptr = 0;
3026 	mqd->cp_gfx_hqd_rptr = RREG32_SOC15(GC, 0, mmCP_GFX_HQD_RPTR);
3027 
3028 	/* active the queue */
3029 	mqd->cp_gfx_hqd_active = 1;
3030 
3031 	return 0;
3032 }
3033 
3034 #ifdef BRING_UP_DEBUG
3035 static int gfx_v10_0_gfx_queue_init_register(struct amdgpu_ring *ring)
3036 {
3037 	struct amdgpu_device *adev = ring->adev;
3038 	struct v10_gfx_mqd *mqd = ring->mqd_ptr;
3039 
3040 	/* set mmCP_GFX_HQD_WPTR/_HI to 0 */
3041 	WREG32_SOC15(GC, 0, mmCP_GFX_HQD_WPTR, mqd->cp_gfx_hqd_wptr);
3042 	WREG32_SOC15(GC, 0, mmCP_GFX_HQD_WPTR_HI, mqd->cp_gfx_hqd_wptr_hi);
3043 
3044 	/* set GFX_MQD_BASE */
3045 	WREG32_SOC15(GC, 0, mmCP_MQD_BASE_ADDR, mqd->cp_mqd_base_addr);
3046 	WREG32_SOC15(GC, 0, mmCP_MQD_BASE_ADDR_HI, mqd->cp_mqd_base_addr_hi);
3047 
3048 	/* set GFX_MQD_CONTROL */
3049 	WREG32_SOC15(GC, 0, mmCP_GFX_MQD_CONTROL, mqd->cp_gfx_mqd_control);
3050 
3051 	/* set GFX_HQD_VMID to 0 */
3052 	WREG32_SOC15(GC, 0, mmCP_GFX_HQD_VMID, mqd->cp_gfx_hqd_vmid);
3053 
3054 	WREG32_SOC15(GC, 0, mmCP_GFX_HQD_QUEUE_PRIORITY,
3055 			mqd->cp_gfx_hqd_queue_priority);
3056 	WREG32_SOC15(GC, 0, mmCP_GFX_HQD_QUANTUM, mqd->cp_gfx_hqd_quantum);
3057 
3058 	/* set GFX_HQD_BASE, similar as CP_RB_BASE */
3059 	WREG32_SOC15(GC, 0, mmCP_GFX_HQD_BASE, mqd->cp_gfx_hqd_base);
3060 	WREG32_SOC15(GC, 0, mmCP_GFX_HQD_BASE_HI, mqd->cp_gfx_hqd_base_hi);
3061 
3062 	/* set GFX_HQD_RPTR_ADDR, similar as CP_RB_RPTR */
3063 	WREG32_SOC15(GC, 0, mmCP_GFX_HQD_RPTR_ADDR, mqd->cp_gfx_hqd_rptr_addr);
3064 	WREG32_SOC15(GC, 0, mmCP_GFX_HQD_RPTR_ADDR_HI, mqd->cp_gfx_hqd_rptr_addr_hi);
3065 
3066 	/* set GFX_HQD_CNTL, similar as CP_RB_CNTL */
3067 	WREG32_SOC15(GC, 0, mmCP_GFX_HQD_CNTL, mqd->cp_gfx_hqd_cntl);
3068 
3069 	/* set RB_WPTR_POLL_ADDR */
3070 	WREG32_SOC15(GC, 0, mmCP_RB_WPTR_POLL_ADDR_LO, mqd->cp_rb_wptr_poll_addr_lo);
3071 	WREG32_SOC15(GC, 0, mmCP_RB_WPTR_POLL_ADDR_HI, mqd->cp_rb_wptr_poll_addr_hi);
3072 
3073 	/* set RB_DOORBELL_CONTROL */
3074 	WREG32_SOC15(GC, 0, mmCP_RB_DOORBELL_CONTROL, mqd->cp_rb_doorbell_control);
3075 
3076 	/* active the queue */
3077 	WREG32_SOC15(GC, 0, mmCP_GFX_HQD_ACTIVE, mqd->cp_gfx_hqd_active);
3078 
3079 	return 0;
3080 }
3081 #endif
3082 
3083 static int gfx_v10_0_gfx_init_queue(struct amdgpu_ring *ring)
3084 {
3085 	struct amdgpu_device *adev = ring->adev;
3086 	struct v10_gfx_mqd *mqd = ring->mqd_ptr;
3087 	int mqd_idx = ring - &adev->gfx.gfx_ring[0];
3088 
3089 	if (!adev->in_gpu_reset && !adev->in_suspend) {
3090 		memset((void *)mqd, 0, sizeof(*mqd));
3091 		mutex_lock(&adev->srbm_mutex);
3092 		nv_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
3093 		gfx_v10_0_gfx_mqd_init(ring);
3094 #ifdef BRING_UP_DEBUG
3095 		gfx_v10_0_gfx_queue_init_register(ring);
3096 #endif
3097 		nv_grbm_select(adev, 0, 0, 0, 0);
3098 		mutex_unlock(&adev->srbm_mutex);
3099 		if (adev->gfx.me.mqd_backup[mqd_idx])
3100 			memcpy(adev->gfx.me.mqd_backup[mqd_idx], mqd, sizeof(*mqd));
3101 	} else if (adev->in_gpu_reset) {
3102 		/* reset mqd with the backup copy */
3103 		if (adev->gfx.me.mqd_backup[mqd_idx])
3104 			memcpy(mqd, adev->gfx.me.mqd_backup[mqd_idx], sizeof(*mqd));
3105 		/* reset the ring */
3106 		ring->wptr = 0;
3107 		adev->wb.wb[ring->wptr_offs] = 0;
3108 		amdgpu_ring_clear_ring(ring);
3109 #ifdef BRING_UP_DEBUG
3110 		mutex_lock(&adev->srbm_mutex);
3111 		nv_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
3112 		gfx_v10_0_gfx_queue_init_register(ring);
3113 		nv_grbm_select(adev, 0, 0, 0, 0);
3114 		mutex_unlock(&adev->srbm_mutex);
3115 #endif
3116 	} else {
3117 		amdgpu_ring_clear_ring(ring);
3118 	}
3119 
3120 	return 0;
3121 }
3122 
3123 #ifndef BRING_UP_DEBUG
3124 static int gfx_v10_0_kiq_enable_kgq(struct amdgpu_device *adev)
3125 {
3126 	struct amdgpu_kiq *kiq = &adev->gfx.kiq;
3127 	struct amdgpu_ring *kiq_ring = &adev->gfx.kiq.ring;
3128 	int r, i;
3129 
3130 	if (!kiq->pmf || !kiq->pmf->kiq_map_queues)
3131 		return -EINVAL;
3132 
3133 	r = amdgpu_ring_alloc(kiq_ring, kiq->pmf->map_queues_size *
3134 					adev->gfx.num_gfx_rings);
3135 	if (r) {
3136 		DRM_ERROR("Failed to lock KIQ (%d).\n", r);
3137 		return r;
3138 	}
3139 
3140 	for (i = 0; i < adev->gfx.num_gfx_rings; i++)
3141 		kiq->pmf->kiq_map_queues(kiq_ring, &adev->gfx.gfx_ring[i]);
3142 
3143 	r = amdgpu_ring_test_ring(kiq_ring);
3144 	if (r) {
3145 		DRM_ERROR("kfq enable failed\n");
3146 		kiq_ring->sched.ready = false;
3147 	}
3148 	return r;
3149 }
3150 #endif
3151 
3152 static int gfx_v10_0_cp_async_gfx_ring_resume(struct amdgpu_device *adev)
3153 {
3154 	int r, i;
3155 	struct amdgpu_ring *ring;
3156 
3157 	for (i = 0; i < adev->gfx.num_gfx_rings; i++) {
3158 		ring = &adev->gfx.gfx_ring[i];
3159 
3160 		r = amdgpu_bo_reserve(ring->mqd_obj, false);
3161 		if (unlikely(r != 0))
3162 			goto done;
3163 
3164 		r = amdgpu_bo_kmap(ring->mqd_obj, (void **)&ring->mqd_ptr);
3165 		if (!r) {
3166 			r = gfx_v10_0_gfx_init_queue(ring);
3167 			amdgpu_bo_kunmap(ring->mqd_obj);
3168 			ring->mqd_ptr = NULL;
3169 		}
3170 		amdgpu_bo_unreserve(ring->mqd_obj);
3171 		if (r)
3172 			goto done;
3173 	}
3174 #ifndef BRING_UP_DEBUG
3175 	r = gfx_v10_0_kiq_enable_kgq(adev);
3176 	if (r)
3177 		goto done;
3178 #endif
3179 	r = gfx_v10_0_cp_gfx_start(adev);
3180 	if (r)
3181 		goto done;
3182 
3183 	for (i = 0; i < adev->gfx.num_gfx_rings; i++) {
3184 		ring = &adev->gfx.gfx_ring[i];
3185 		ring->sched.ready = true;
3186 	}
3187 done:
3188 	return r;
3189 }
3190 
3191 static int gfx_v10_0_compute_mqd_init(struct amdgpu_ring *ring)
3192 {
3193 	struct amdgpu_device *adev = ring->adev;
3194 	struct v10_compute_mqd *mqd = ring->mqd_ptr;
3195 	uint64_t hqd_gpu_addr, wb_gpu_addr, eop_base_addr;
3196 	uint32_t tmp;
3197 
3198 	mqd->header = 0xC0310800;
3199 	mqd->compute_pipelinestat_enable = 0x00000001;
3200 	mqd->compute_static_thread_mgmt_se0 = 0xffffffff;
3201 	mqd->compute_static_thread_mgmt_se1 = 0xffffffff;
3202 	mqd->compute_static_thread_mgmt_se2 = 0xffffffff;
3203 	mqd->compute_static_thread_mgmt_se3 = 0xffffffff;
3204 	mqd->compute_misc_reserved = 0x00000003;
3205 
3206 	eop_base_addr = ring->eop_gpu_addr >> 8;
3207 	mqd->cp_hqd_eop_base_addr_lo = eop_base_addr;
3208 	mqd->cp_hqd_eop_base_addr_hi = upper_32_bits(eop_base_addr);
3209 
3210 	/* set the EOP size, register value is 2^(EOP_SIZE+1) dwords */
3211 	tmp = RREG32_SOC15(GC, 0, mmCP_HQD_EOP_CONTROL);
3212 	tmp = REG_SET_FIELD(tmp, CP_HQD_EOP_CONTROL, EOP_SIZE,
3213 			(order_base_2(GFX10_MEC_HPD_SIZE / 4) - 1));
3214 
3215 	mqd->cp_hqd_eop_control = tmp;
3216 
3217 	/* enable doorbell? */
3218 	tmp = RREG32_SOC15(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL);
3219 
3220 	if (ring->use_doorbell) {
3221 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3222 				    DOORBELL_OFFSET, ring->doorbell_index);
3223 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3224 				    DOORBELL_EN, 1);
3225 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3226 				    DOORBELL_SOURCE, 0);
3227 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3228 				    DOORBELL_HIT, 0);
3229 	} else {
3230 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3231 				    DOORBELL_EN, 0);
3232 	}
3233 
3234 	mqd->cp_hqd_pq_doorbell_control = tmp;
3235 
3236 	/* disable the queue if it's active */
3237 	ring->wptr = 0;
3238 	mqd->cp_hqd_dequeue_request = 0;
3239 	mqd->cp_hqd_pq_rptr = 0;
3240 	mqd->cp_hqd_pq_wptr_lo = 0;
3241 	mqd->cp_hqd_pq_wptr_hi = 0;
3242 
3243 	/* set the pointer to the MQD */
3244 	mqd->cp_mqd_base_addr_lo = ring->mqd_gpu_addr & 0xfffffffc;
3245 	mqd->cp_mqd_base_addr_hi = upper_32_bits(ring->mqd_gpu_addr);
3246 
3247 	/* set MQD vmid to 0 */
3248 	tmp = RREG32_SOC15(GC, 0, mmCP_MQD_CONTROL);
3249 	tmp = REG_SET_FIELD(tmp, CP_MQD_CONTROL, VMID, 0);
3250 	mqd->cp_mqd_control = tmp;
3251 
3252 	/* set the pointer to the HQD, this is similar CP_RB0_BASE/_HI */
3253 	hqd_gpu_addr = ring->gpu_addr >> 8;
3254 	mqd->cp_hqd_pq_base_lo = hqd_gpu_addr;
3255 	mqd->cp_hqd_pq_base_hi = upper_32_bits(hqd_gpu_addr);
3256 
3257 	/* set up the HQD, this is similar to CP_RB0_CNTL */
3258 	tmp = RREG32_SOC15(GC, 0, mmCP_HQD_PQ_CONTROL);
3259 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, QUEUE_SIZE,
3260 			    (order_base_2(ring->ring_size / 4) - 1));
3261 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, RPTR_BLOCK_SIZE,
3262 			    ((order_base_2(AMDGPU_GPU_PAGE_SIZE / 4) - 1) << 8));
3263 #ifdef __BIG_ENDIAN
3264 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, ENDIAN_SWAP, 1);
3265 #endif
3266 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, UNORD_DISPATCH, 0);
3267 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, TUNNEL_DISPATCH, 0);
3268 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, PRIV_STATE, 1);
3269 	tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_CONTROL, KMD_QUEUE, 1);
3270 	mqd->cp_hqd_pq_control = tmp;
3271 
3272 	/* set the wb address whether it's enabled or not */
3273 	wb_gpu_addr = adev->wb.gpu_addr + (ring->rptr_offs * 4);
3274 	mqd->cp_hqd_pq_rptr_report_addr_lo = wb_gpu_addr & 0xfffffffc;
3275 	mqd->cp_hqd_pq_rptr_report_addr_hi =
3276 		upper_32_bits(wb_gpu_addr) & 0xffff;
3277 
3278 	/* only used if CP_PQ_WPTR_POLL_CNTL.CP_PQ_WPTR_POLL_CNTL__EN_MASK=1 */
3279 	wb_gpu_addr = adev->wb.gpu_addr + (ring->wptr_offs * 4);
3280 	mqd->cp_hqd_pq_wptr_poll_addr_lo = wb_gpu_addr & 0xfffffffc;
3281 	mqd->cp_hqd_pq_wptr_poll_addr_hi = upper_32_bits(wb_gpu_addr) & 0xffff;
3282 
3283 	tmp = 0;
3284 	/* enable the doorbell if requested */
3285 	if (ring->use_doorbell) {
3286 		tmp = RREG32_SOC15(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL);
3287 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3288 				DOORBELL_OFFSET, ring->doorbell_index);
3289 
3290 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3291 				    DOORBELL_EN, 1);
3292 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3293 				    DOORBELL_SOURCE, 0);
3294 		tmp = REG_SET_FIELD(tmp, CP_HQD_PQ_DOORBELL_CONTROL,
3295 				    DOORBELL_HIT, 0);
3296 	}
3297 
3298 	mqd->cp_hqd_pq_doorbell_control = tmp;
3299 
3300 	/* reset read and write pointers, similar to CP_RB0_WPTR/_RPTR */
3301 	ring->wptr = 0;
3302 	mqd->cp_hqd_pq_rptr = RREG32_SOC15(GC, 0, mmCP_HQD_PQ_RPTR);
3303 
3304 	/* set the vmid for the queue */
3305 	mqd->cp_hqd_vmid = 0;
3306 
3307 	tmp = RREG32_SOC15(GC, 0, mmCP_HQD_PERSISTENT_STATE);
3308 	tmp = REG_SET_FIELD(tmp, CP_HQD_PERSISTENT_STATE, PRELOAD_SIZE, 0x53);
3309 	mqd->cp_hqd_persistent_state = tmp;
3310 
3311 	/* set MIN_IB_AVAIL_SIZE */
3312 	tmp = RREG32_SOC15(GC, 0, mmCP_HQD_IB_CONTROL);
3313 	tmp = REG_SET_FIELD(tmp, CP_HQD_IB_CONTROL, MIN_IB_AVAIL_SIZE, 3);
3314 	mqd->cp_hqd_ib_control = tmp;
3315 
3316 	/* activate the queue */
3317 	mqd->cp_hqd_active = 1;
3318 
3319 	return 0;
3320 }
3321 
3322 static int gfx_v10_0_kiq_init_register(struct amdgpu_ring *ring)
3323 {
3324 	struct amdgpu_device *adev = ring->adev;
3325 	struct v10_compute_mqd *mqd = ring->mqd_ptr;
3326 	int j;
3327 
3328 	/* disable wptr polling */
3329 	WREG32_FIELD15(GC, 0, CP_PQ_WPTR_POLL_CNTL, EN, 0);
3330 
3331 	/* write the EOP addr */
3332 	WREG32_SOC15(GC, 0, mmCP_HQD_EOP_BASE_ADDR,
3333 	       mqd->cp_hqd_eop_base_addr_lo);
3334 	WREG32_SOC15(GC, 0, mmCP_HQD_EOP_BASE_ADDR_HI,
3335 	       mqd->cp_hqd_eop_base_addr_hi);
3336 
3337 	/* set the EOP size, register value is 2^(EOP_SIZE+1) dwords */
3338 	WREG32_SOC15(GC, 0, mmCP_HQD_EOP_CONTROL,
3339 	       mqd->cp_hqd_eop_control);
3340 
3341 	/* enable doorbell? */
3342 	WREG32_SOC15(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL,
3343 	       mqd->cp_hqd_pq_doorbell_control);
3344 
3345 	/* disable the queue if it's active */
3346 	if (RREG32_SOC15(GC, 0, mmCP_HQD_ACTIVE) & 1) {
3347 		WREG32_SOC15(GC, 0, mmCP_HQD_DEQUEUE_REQUEST, 1);
3348 		for (j = 0; j < adev->usec_timeout; j++) {
3349 			if (!(RREG32_SOC15(GC, 0, mmCP_HQD_ACTIVE) & 1))
3350 				break;
3351 			udelay(1);
3352 		}
3353 		WREG32_SOC15(GC, 0, mmCP_HQD_DEQUEUE_REQUEST,
3354 		       mqd->cp_hqd_dequeue_request);
3355 		WREG32_SOC15(GC, 0, mmCP_HQD_PQ_RPTR,
3356 		       mqd->cp_hqd_pq_rptr);
3357 		WREG32_SOC15(GC, 0, mmCP_HQD_PQ_WPTR_LO,
3358 		       mqd->cp_hqd_pq_wptr_lo);
3359 		WREG32_SOC15(GC, 0, mmCP_HQD_PQ_WPTR_HI,
3360 		       mqd->cp_hqd_pq_wptr_hi);
3361 	}
3362 
3363 	/* set the pointer to the MQD */
3364 	WREG32_SOC15(GC, 0, mmCP_MQD_BASE_ADDR,
3365 	       mqd->cp_mqd_base_addr_lo);
3366 	WREG32_SOC15(GC, 0, mmCP_MQD_BASE_ADDR_HI,
3367 	       mqd->cp_mqd_base_addr_hi);
3368 
3369 	/* set MQD vmid to 0 */
3370 	WREG32_SOC15(GC, 0, mmCP_MQD_CONTROL,
3371 	       mqd->cp_mqd_control);
3372 
3373 	/* set the pointer to the HQD, this is similar CP_RB0_BASE/_HI */
3374 	WREG32_SOC15(GC, 0, mmCP_HQD_PQ_BASE,
3375 	       mqd->cp_hqd_pq_base_lo);
3376 	WREG32_SOC15(GC, 0, mmCP_HQD_PQ_BASE_HI,
3377 	       mqd->cp_hqd_pq_base_hi);
3378 
3379 	/* set up the HQD, this is similar to CP_RB0_CNTL */
3380 	WREG32_SOC15(GC, 0, mmCP_HQD_PQ_CONTROL,
3381 	       mqd->cp_hqd_pq_control);
3382 
3383 	/* set the wb address whether it's enabled or not */
3384 	WREG32_SOC15(GC, 0, mmCP_HQD_PQ_RPTR_REPORT_ADDR,
3385 		mqd->cp_hqd_pq_rptr_report_addr_lo);
3386 	WREG32_SOC15(GC, 0, mmCP_HQD_PQ_RPTR_REPORT_ADDR_HI,
3387 		mqd->cp_hqd_pq_rptr_report_addr_hi);
3388 
3389 	/* only used if CP_PQ_WPTR_POLL_CNTL.CP_PQ_WPTR_POLL_CNTL__EN_MASK=1 */
3390 	WREG32_SOC15(GC, 0, mmCP_HQD_PQ_WPTR_POLL_ADDR,
3391 	       mqd->cp_hqd_pq_wptr_poll_addr_lo);
3392 	WREG32_SOC15(GC, 0, mmCP_HQD_PQ_WPTR_POLL_ADDR_HI,
3393 	       mqd->cp_hqd_pq_wptr_poll_addr_hi);
3394 
3395 	/* enable the doorbell if requested */
3396 	if (ring->use_doorbell) {
3397 		WREG32_SOC15(GC, 0, mmCP_MEC_DOORBELL_RANGE_LOWER,
3398 			(adev->doorbell_index.kiq * 2) << 2);
3399 		WREG32_SOC15(GC, 0, mmCP_MEC_DOORBELL_RANGE_UPPER,
3400 			(adev->doorbell_index.userqueue_end * 2) << 2);
3401 	}
3402 
3403 	WREG32_SOC15(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL,
3404 	       mqd->cp_hqd_pq_doorbell_control);
3405 
3406 	/* reset read and write pointers, similar to CP_RB0_WPTR/_RPTR */
3407 	WREG32_SOC15(GC, 0, mmCP_HQD_PQ_WPTR_LO,
3408 	       mqd->cp_hqd_pq_wptr_lo);
3409 	WREG32_SOC15(GC, 0, mmCP_HQD_PQ_WPTR_HI,
3410 	       mqd->cp_hqd_pq_wptr_hi);
3411 
3412 	/* set the vmid for the queue */
3413 	WREG32_SOC15(GC, 0, mmCP_HQD_VMID, mqd->cp_hqd_vmid);
3414 
3415 	WREG32_SOC15(GC, 0, mmCP_HQD_PERSISTENT_STATE,
3416 	       mqd->cp_hqd_persistent_state);
3417 
3418 	/* activate the queue */
3419 	WREG32_SOC15(GC, 0, mmCP_HQD_ACTIVE,
3420 	       mqd->cp_hqd_active);
3421 
3422 	if (ring->use_doorbell)
3423 		WREG32_FIELD15(GC, 0, CP_PQ_STATUS, DOORBELL_ENABLE, 1);
3424 
3425 	return 0;
3426 }
3427 
3428 static int gfx_v10_0_kiq_init_queue(struct amdgpu_ring *ring)
3429 {
3430 	struct amdgpu_device *adev = ring->adev;
3431 	struct v10_compute_mqd *mqd = ring->mqd_ptr;
3432 	int mqd_idx = AMDGPU_MAX_COMPUTE_RINGS;
3433 
3434 	gfx_v10_0_kiq_setting(ring);
3435 
3436 	if (adev->in_gpu_reset) { /* for GPU_RESET case */
3437 		/* reset MQD to a clean status */
3438 		if (adev->gfx.mec.mqd_backup[mqd_idx])
3439 			memcpy(mqd, adev->gfx.mec.mqd_backup[mqd_idx], sizeof(*mqd));
3440 
3441 		/* reset ring buffer */
3442 		ring->wptr = 0;
3443 		amdgpu_ring_clear_ring(ring);
3444 
3445 		mutex_lock(&adev->srbm_mutex);
3446 		nv_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
3447 		gfx_v10_0_kiq_init_register(ring);
3448 		nv_grbm_select(adev, 0, 0, 0, 0);
3449 		mutex_unlock(&adev->srbm_mutex);
3450 	} else {
3451 		memset((void *)mqd, 0, sizeof(*mqd));
3452 		mutex_lock(&adev->srbm_mutex);
3453 		nv_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
3454 		gfx_v10_0_compute_mqd_init(ring);
3455 		gfx_v10_0_kiq_init_register(ring);
3456 		nv_grbm_select(adev, 0, 0, 0, 0);
3457 		mutex_unlock(&adev->srbm_mutex);
3458 
3459 		if (adev->gfx.mec.mqd_backup[mqd_idx])
3460 			memcpy(adev->gfx.mec.mqd_backup[mqd_idx], mqd, sizeof(*mqd));
3461 	}
3462 
3463 	return 0;
3464 }
3465 
3466 static int gfx_v10_0_kcq_init_queue(struct amdgpu_ring *ring)
3467 {
3468 	struct amdgpu_device *adev = ring->adev;
3469 	struct v10_compute_mqd *mqd = ring->mqd_ptr;
3470 	int mqd_idx = ring - &adev->gfx.compute_ring[0];
3471 
3472 	if (!adev->in_gpu_reset && !adev->in_suspend) {
3473 		memset((void *)mqd, 0, sizeof(*mqd));
3474 		mutex_lock(&adev->srbm_mutex);
3475 		nv_grbm_select(adev, ring->me, ring->pipe, ring->queue, 0);
3476 		gfx_v10_0_compute_mqd_init(ring);
3477 		nv_grbm_select(adev, 0, 0, 0, 0);
3478 		mutex_unlock(&adev->srbm_mutex);
3479 
3480 		if (adev->gfx.mec.mqd_backup[mqd_idx])
3481 			memcpy(adev->gfx.mec.mqd_backup[mqd_idx], mqd, sizeof(*mqd));
3482 	} else if (adev->in_gpu_reset) { /* for GPU_RESET case */
3483 		/* reset MQD to a clean status */
3484 		if (adev->gfx.mec.mqd_backup[mqd_idx])
3485 			memcpy(mqd, adev->gfx.mec.mqd_backup[mqd_idx], sizeof(*mqd));
3486 
3487 		/* reset ring buffer */
3488 		ring->wptr = 0;
3489 		amdgpu_ring_clear_ring(ring);
3490 	} else {
3491 		amdgpu_ring_clear_ring(ring);
3492 	}
3493 
3494 	return 0;
3495 }
3496 
3497 static int gfx_v10_0_kiq_resume(struct amdgpu_device *adev)
3498 {
3499 	struct amdgpu_ring *ring;
3500 	int r;
3501 
3502 	ring = &adev->gfx.kiq.ring;
3503 
3504 	r = amdgpu_bo_reserve(ring->mqd_obj, false);
3505 	if (unlikely(r != 0))
3506 		return r;
3507 
3508 	r = amdgpu_bo_kmap(ring->mqd_obj, (void **)&ring->mqd_ptr);
3509 	if (unlikely(r != 0))
3510 		return r;
3511 
3512 	gfx_v10_0_kiq_init_queue(ring);
3513 	amdgpu_bo_kunmap(ring->mqd_obj);
3514 	ring->mqd_ptr = NULL;
3515 	amdgpu_bo_unreserve(ring->mqd_obj);
3516 	ring->sched.ready = true;
3517 	return 0;
3518 }
3519 
3520 static int gfx_v10_0_kcq_resume(struct amdgpu_device *adev)
3521 {
3522 	struct amdgpu_ring *ring = NULL;
3523 	int r = 0, i;
3524 
3525 	gfx_v10_0_cp_compute_enable(adev, true);
3526 
3527 	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
3528 		ring = &adev->gfx.compute_ring[i];
3529 
3530 		r = amdgpu_bo_reserve(ring->mqd_obj, false);
3531 		if (unlikely(r != 0))
3532 			goto done;
3533 		r = amdgpu_bo_kmap(ring->mqd_obj, (void **)&ring->mqd_ptr);
3534 		if (!r) {
3535 			r = gfx_v10_0_kcq_init_queue(ring);
3536 			amdgpu_bo_kunmap(ring->mqd_obj);
3537 			ring->mqd_ptr = NULL;
3538 		}
3539 		amdgpu_bo_unreserve(ring->mqd_obj);
3540 		if (r)
3541 			goto done;
3542 	}
3543 
3544 	r = amdgpu_gfx_enable_kcq(adev);
3545 done:
3546 	return r;
3547 }
3548 
3549 static int gfx_v10_0_cp_resume(struct amdgpu_device *adev)
3550 {
3551 	int r, i;
3552 	struct amdgpu_ring *ring;
3553 
3554 	if (!(adev->flags & AMD_IS_APU))
3555 		gfx_v10_0_enable_gui_idle_interrupt(adev, false);
3556 
3557 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_DIRECT) {
3558 		/* legacy firmware loading */
3559 		r = gfx_v10_0_cp_gfx_load_microcode(adev);
3560 		if (r)
3561 			return r;
3562 
3563 		r = gfx_v10_0_cp_compute_load_microcode(adev);
3564 		if (r)
3565 			return r;
3566 	}
3567 
3568 	r = gfx_v10_0_kiq_resume(adev);
3569 	if (r)
3570 		return r;
3571 
3572 	r = gfx_v10_0_kcq_resume(adev);
3573 	if (r)
3574 		return r;
3575 
3576 	if (!amdgpu_async_gfx_ring) {
3577 		r = gfx_v10_0_cp_gfx_resume(adev);
3578 		if (r)
3579 			return r;
3580 	} else {
3581 		r = gfx_v10_0_cp_async_gfx_ring_resume(adev);
3582 		if (r)
3583 			return r;
3584 	}
3585 
3586 	for (i = 0; i < adev->gfx.num_gfx_rings; i++) {
3587 		ring = &adev->gfx.gfx_ring[i];
3588 		DRM_INFO("gfx %d ring me %d pipe %d q %d\n",
3589 			 i, ring->me, ring->pipe, ring->queue);
3590 		r = amdgpu_ring_test_ring(ring);
3591 		if (r) {
3592 			ring->sched.ready = false;
3593 			return r;
3594 		}
3595 	}
3596 
3597 	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
3598 		ring = &adev->gfx.compute_ring[i];
3599 		ring->sched.ready = true;
3600 		DRM_INFO("compute ring %d mec %d pipe %d q %d\n",
3601 			 i, ring->me, ring->pipe, ring->queue);
3602 		r = amdgpu_ring_test_ring(ring);
3603 		if (r)
3604 			ring->sched.ready = false;
3605 	}
3606 
3607 	return 0;
3608 }
3609 
3610 static void gfx_v10_0_cp_enable(struct amdgpu_device *adev, bool enable)
3611 {
3612 	gfx_v10_0_cp_gfx_enable(adev, enable);
3613 	gfx_v10_0_cp_compute_enable(adev, enable);
3614 }
3615 
3616 static bool gfx_v10_0_check_grbm_cam_remapping(struct amdgpu_device *adev)
3617 {
3618 	uint32_t data, pattern = 0xDEADBEEF;
3619 
3620 	/* check if mmVGT_ESGS_RING_SIZE_UMD
3621 	 * has been remapped to mmVGT_ESGS_RING_SIZE */
3622 	data = RREG32_SOC15(GC, 0, mmVGT_ESGS_RING_SIZE);
3623 
3624 	WREG32_SOC15(GC, 0, mmVGT_ESGS_RING_SIZE, 0);
3625 
3626 	WREG32_SOC15(GC, 0, mmVGT_ESGS_RING_SIZE_UMD, pattern);
3627 
3628 	if (RREG32_SOC15(GC, 0, mmVGT_ESGS_RING_SIZE) == pattern) {
3629 		WREG32_SOC15(GC, 0, mmVGT_ESGS_RING_SIZE_UMD, data);
3630 		return true;
3631 	} else {
3632 		WREG32_SOC15(GC, 0, mmVGT_ESGS_RING_SIZE, data);
3633 		return false;
3634 	}
3635 }
3636 
3637 static void gfx_v10_0_setup_grbm_cam_remapping(struct amdgpu_device *adev)
3638 {
3639 	uint32_t data;
3640 
3641 	/* initialize cam_index to 0
3642 	 * index will auto-inc after each data writting */
3643 	WREG32_SOC15(GC, 0, mmGRBM_CAM_INDEX, 0);
3644 
3645 	/* mmVGT_TF_RING_SIZE_UMD -> mmVGT_TF_RING_SIZE */
3646 	data = (SOC15_REG_OFFSET(GC, 0, mmVGT_TF_RING_SIZE_UMD) <<
3647 		GRBM_CAM_DATA__CAM_ADDR__SHIFT) |
3648 	       (SOC15_REG_OFFSET(GC, 0, mmVGT_TF_RING_SIZE) <<
3649 		GRBM_CAM_DATA__CAM_REMAPADDR__SHIFT);
3650 	WREG32_SOC15(GC, 0, mmGRBM_CAM_DATA_UPPER, 0);
3651 	WREG32_SOC15(GC, 0, mmGRBM_CAM_DATA, data);
3652 
3653 	/* mmVGT_TF_MEMORY_BASE_UMD -> mmVGT_TF_MEMORY_BASE */
3654 	data = (SOC15_REG_OFFSET(GC, 0, mmVGT_TF_MEMORY_BASE_UMD) <<
3655 		GRBM_CAM_DATA__CAM_ADDR__SHIFT) |
3656 	       (SOC15_REG_OFFSET(GC, 0, mmVGT_TF_MEMORY_BASE) <<
3657 		GRBM_CAM_DATA__CAM_REMAPADDR__SHIFT);
3658 	WREG32_SOC15(GC, 0, mmGRBM_CAM_DATA_UPPER, 0);
3659 	WREG32_SOC15(GC, 0, mmGRBM_CAM_DATA, data);
3660 
3661 	/* mmVGT_TF_MEMORY_BASE_HI_UMD -> mmVGT_TF_MEMORY_BASE_HI */
3662 	data = (SOC15_REG_OFFSET(GC, 0, mmVGT_TF_MEMORY_BASE_HI_UMD) <<
3663 		GRBM_CAM_DATA__CAM_ADDR__SHIFT) |
3664 	       (SOC15_REG_OFFSET(GC, 0, mmVGT_TF_MEMORY_BASE_HI) <<
3665 		GRBM_CAM_DATA__CAM_REMAPADDR__SHIFT);
3666 	WREG32_SOC15(GC, 0, mmGRBM_CAM_DATA_UPPER, 0);
3667 	WREG32_SOC15(GC, 0, mmGRBM_CAM_DATA, data);
3668 
3669 	/* mmVGT_HS_OFFCHIP_PARAM_UMD -> mmVGT_HS_OFFCHIP_PARAM */
3670 	data = (SOC15_REG_OFFSET(GC, 0, mmVGT_HS_OFFCHIP_PARAM_UMD) <<
3671 		GRBM_CAM_DATA__CAM_ADDR__SHIFT) |
3672 	       (SOC15_REG_OFFSET(GC, 0, mmVGT_HS_OFFCHIP_PARAM) <<
3673 		GRBM_CAM_DATA__CAM_REMAPADDR__SHIFT);
3674 	WREG32_SOC15(GC, 0, mmGRBM_CAM_DATA_UPPER, 0);
3675 	WREG32_SOC15(GC, 0, mmGRBM_CAM_DATA, data);
3676 
3677 	/* mmVGT_ESGS_RING_SIZE_UMD -> mmVGT_ESGS_RING_SIZE */
3678 	data = (SOC15_REG_OFFSET(GC, 0, mmVGT_ESGS_RING_SIZE_UMD) <<
3679 		GRBM_CAM_DATA__CAM_ADDR__SHIFT) |
3680 	       (SOC15_REG_OFFSET(GC, 0, mmVGT_ESGS_RING_SIZE) <<
3681 		GRBM_CAM_DATA__CAM_REMAPADDR__SHIFT);
3682 	WREG32_SOC15(GC, 0, mmGRBM_CAM_DATA_UPPER, 0);
3683 	WREG32_SOC15(GC, 0, mmGRBM_CAM_DATA, data);
3684 
3685 	/* mmVGT_GSVS_RING_SIZE_UMD -> mmVGT_GSVS_RING_SIZE */
3686 	data = (SOC15_REG_OFFSET(GC, 0, mmVGT_GSVS_RING_SIZE_UMD) <<
3687 		GRBM_CAM_DATA__CAM_ADDR__SHIFT) |
3688 	       (SOC15_REG_OFFSET(GC, 0, mmVGT_GSVS_RING_SIZE) <<
3689 		GRBM_CAM_DATA__CAM_REMAPADDR__SHIFT);
3690 	WREG32_SOC15(GC, 0, mmGRBM_CAM_DATA_UPPER, 0);
3691 	WREG32_SOC15(GC, 0, mmGRBM_CAM_DATA, data);
3692 
3693 	/* mmSPI_CONFIG_CNTL_REMAP -> mmSPI_CONFIG_CNTL */
3694 	data = (SOC15_REG_OFFSET(GC, 0, mmSPI_CONFIG_CNTL_REMAP) <<
3695 		GRBM_CAM_DATA__CAM_ADDR__SHIFT) |
3696 	       (SOC15_REG_OFFSET(GC, 0, mmSPI_CONFIG_CNTL) <<
3697 		GRBM_CAM_DATA__CAM_REMAPADDR__SHIFT);
3698 	WREG32_SOC15(GC, 0, mmGRBM_CAM_DATA_UPPER, 0);
3699 	WREG32_SOC15(GC, 0, mmGRBM_CAM_DATA, data);
3700 }
3701 
3702 static int gfx_v10_0_hw_init(void *handle)
3703 {
3704 	int r;
3705 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
3706 
3707 	if (!amdgpu_emu_mode)
3708 		gfx_v10_0_init_golden_registers(adev);
3709 
3710 	if (adev->firmware.load_type == AMDGPU_FW_LOAD_DIRECT) {
3711 		/**
3712 		 * For gfx 10, rlc firmware loading relies on smu firmware is
3713 		 * loaded firstly, so in direct type, it has to load smc ucode
3714 		 * here before rlc.
3715 		 */
3716 		r = smu_load_microcode(&adev->smu);
3717 		if (r)
3718 			return r;
3719 
3720 		r = smu_check_fw_status(&adev->smu);
3721 		if (r) {
3722 			pr_err("SMC firmware status is not correct\n");
3723 			return r;
3724 		}
3725 	}
3726 
3727 	/* if GRBM CAM not remapped, set up the remapping */
3728 	if (!gfx_v10_0_check_grbm_cam_remapping(adev))
3729 		gfx_v10_0_setup_grbm_cam_remapping(adev);
3730 
3731 	gfx_v10_0_constants_init(adev);
3732 
3733 	r = gfx_v10_0_rlc_resume(adev);
3734 	if (r)
3735 		return r;
3736 
3737 	/*
3738 	 * init golden registers and rlc resume may override some registers,
3739 	 * reconfig them here
3740 	 */
3741 	gfx_v10_0_tcp_harvest(adev);
3742 
3743 	r = gfx_v10_0_cp_resume(adev);
3744 	if (r)
3745 		return r;
3746 
3747 	return r;
3748 }
3749 
3750 #ifndef BRING_UP_DEBUG
3751 static int gfx_v10_0_kiq_disable_kgq(struct amdgpu_device *adev)
3752 {
3753 	struct amdgpu_kiq *kiq = &adev->gfx.kiq;
3754 	struct amdgpu_ring *kiq_ring = &kiq->ring;
3755 	int i;
3756 
3757 	if (!kiq->pmf || !kiq->pmf->kiq_unmap_queues)
3758 		return -EINVAL;
3759 
3760 	if (amdgpu_ring_alloc(kiq_ring, kiq->pmf->unmap_queues_size *
3761 					adev->gfx.num_gfx_rings))
3762 		return -ENOMEM;
3763 
3764 	for (i = 0; i < adev->gfx.num_gfx_rings; i++)
3765 		kiq->pmf->kiq_unmap_queues(kiq_ring, &adev->gfx.gfx_ring[i],
3766 					   PREEMPT_QUEUES, 0, 0);
3767 
3768 	return amdgpu_ring_test_ring(kiq_ring);
3769 }
3770 #endif
3771 
3772 static int gfx_v10_0_hw_fini(void *handle)
3773 {
3774 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
3775 	int r;
3776 
3777 	amdgpu_irq_put(adev, &adev->gfx.priv_reg_irq, 0);
3778 	amdgpu_irq_put(adev, &adev->gfx.priv_inst_irq, 0);
3779 #ifndef BRING_UP_DEBUG
3780 	if (amdgpu_async_gfx_ring) {
3781 		r = gfx_v10_0_kiq_disable_kgq(adev);
3782 		if (r)
3783 			DRM_ERROR("KGQ disable failed\n");
3784 	}
3785 #endif
3786 	if (amdgpu_gfx_disable_kcq(adev))
3787 		DRM_ERROR("KCQ disable failed\n");
3788 	if (amdgpu_sriov_vf(adev)) {
3789 		gfx_v10_0_cp_gfx_enable(adev, false);
3790 		return 0;
3791 	}
3792 	gfx_v10_0_cp_enable(adev, false);
3793 	gfx_v10_0_enable_gui_idle_interrupt(adev, false);
3794 
3795 	return 0;
3796 }
3797 
3798 static int gfx_v10_0_suspend(void *handle)
3799 {
3800 	return gfx_v10_0_hw_fini(handle);
3801 }
3802 
3803 static int gfx_v10_0_resume(void *handle)
3804 {
3805 	return gfx_v10_0_hw_init(handle);
3806 }
3807 
3808 static bool gfx_v10_0_is_idle(void *handle)
3809 {
3810 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
3811 
3812 	if (REG_GET_FIELD(RREG32_SOC15(GC, 0, mmGRBM_STATUS),
3813 				GRBM_STATUS, GUI_ACTIVE))
3814 		return false;
3815 	else
3816 		return true;
3817 }
3818 
3819 static int gfx_v10_0_wait_for_idle(void *handle)
3820 {
3821 	unsigned i;
3822 	u32 tmp;
3823 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
3824 
3825 	for (i = 0; i < adev->usec_timeout; i++) {
3826 		/* read MC_STATUS */
3827 		tmp = RREG32_SOC15(GC, 0, mmGRBM_STATUS) &
3828 			GRBM_STATUS__GUI_ACTIVE_MASK;
3829 
3830 		if (!REG_GET_FIELD(tmp, GRBM_STATUS, GUI_ACTIVE))
3831 			return 0;
3832 		udelay(1);
3833 	}
3834 	return -ETIMEDOUT;
3835 }
3836 
3837 static int gfx_v10_0_soft_reset(void *handle)
3838 {
3839 	u32 grbm_soft_reset = 0;
3840 	u32 tmp;
3841 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
3842 
3843 	/* GRBM_STATUS */
3844 	tmp = RREG32_SOC15(GC, 0, mmGRBM_STATUS);
3845 	if (tmp & (GRBM_STATUS__PA_BUSY_MASK | GRBM_STATUS__SC_BUSY_MASK |
3846 		   GRBM_STATUS__BCI_BUSY_MASK | GRBM_STATUS__SX_BUSY_MASK |
3847 		   GRBM_STATUS__TA_BUSY_MASK | GRBM_STATUS__DB_BUSY_MASK |
3848 		   GRBM_STATUS__CB_BUSY_MASK | GRBM_STATUS__GDS_BUSY_MASK |
3849 		   GRBM_STATUS__SPI_BUSY_MASK | GRBM_STATUS__GE_BUSY_NO_DMA_MASK
3850 		   | GRBM_STATUS__BCI_BUSY_MASK)) {
3851 		grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset,
3852 						GRBM_SOFT_RESET, SOFT_RESET_CP,
3853 						1);
3854 		grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset,
3855 						GRBM_SOFT_RESET, SOFT_RESET_GFX,
3856 						1);
3857 	}
3858 
3859 	if (tmp & (GRBM_STATUS__CP_BUSY_MASK | GRBM_STATUS__CP_COHERENCY_BUSY_MASK)) {
3860 		grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset,
3861 						GRBM_SOFT_RESET, SOFT_RESET_CP,
3862 						1);
3863 	}
3864 
3865 	/* GRBM_STATUS2 */
3866 	tmp = RREG32_SOC15(GC, 0, mmGRBM_STATUS2);
3867 	if (REG_GET_FIELD(tmp, GRBM_STATUS2, RLC_BUSY))
3868 		grbm_soft_reset = REG_SET_FIELD(grbm_soft_reset,
3869 						GRBM_SOFT_RESET, SOFT_RESET_RLC,
3870 						1);
3871 
3872 	if (grbm_soft_reset) {
3873 		/* stop the rlc */
3874 		gfx_v10_0_rlc_stop(adev);
3875 
3876 		/* Disable GFX parsing/prefetching */
3877 		gfx_v10_0_cp_gfx_enable(adev, false);
3878 
3879 		/* Disable MEC parsing/prefetching */
3880 		gfx_v10_0_cp_compute_enable(adev, false);
3881 
3882 		if (grbm_soft_reset) {
3883 			tmp = RREG32_SOC15(GC, 0, mmGRBM_SOFT_RESET);
3884 			tmp |= grbm_soft_reset;
3885 			dev_info(adev->dev, "GRBM_SOFT_RESET=0x%08X\n", tmp);
3886 			WREG32_SOC15(GC, 0, mmGRBM_SOFT_RESET, tmp);
3887 			tmp = RREG32_SOC15(GC, 0, mmGRBM_SOFT_RESET);
3888 
3889 			udelay(50);
3890 
3891 			tmp &= ~grbm_soft_reset;
3892 			WREG32_SOC15(GC, 0, mmGRBM_SOFT_RESET, tmp);
3893 			tmp = RREG32_SOC15(GC, 0, mmGRBM_SOFT_RESET);
3894 		}
3895 
3896 		/* Wait a little for things to settle down */
3897 		udelay(50);
3898 	}
3899 	return 0;
3900 }
3901 
3902 static uint64_t gfx_v10_0_get_gpu_clock_counter(struct amdgpu_device *adev)
3903 {
3904 	uint64_t clock;
3905 
3906 	mutex_lock(&adev->gfx.gpu_clock_mutex);
3907 	WREG32_SOC15(GC, 0, mmRLC_CAPTURE_GPU_CLOCK_COUNT, 1);
3908 	clock = (uint64_t)RREG32_SOC15(GC, 0, mmRLC_GPU_CLOCK_COUNT_LSB) |
3909 		((uint64_t)RREG32_SOC15(GC, 0, mmRLC_GPU_CLOCK_COUNT_MSB) << 32ULL);
3910 	mutex_unlock(&adev->gfx.gpu_clock_mutex);
3911 	return clock;
3912 }
3913 
3914 static void gfx_v10_0_ring_emit_gds_switch(struct amdgpu_ring *ring,
3915 					   uint32_t vmid,
3916 					   uint32_t gds_base, uint32_t gds_size,
3917 					   uint32_t gws_base, uint32_t gws_size,
3918 					   uint32_t oa_base, uint32_t oa_size)
3919 {
3920 	struct amdgpu_device *adev = ring->adev;
3921 
3922 	/* GDS Base */
3923 	gfx_v10_0_write_data_to_reg(ring, 0, false,
3924 				    SOC15_REG_OFFSET(GC, 0, mmGDS_VMID0_BASE) + 2 * vmid,
3925 				    gds_base);
3926 
3927 	/* GDS Size */
3928 	gfx_v10_0_write_data_to_reg(ring, 0, false,
3929 				    SOC15_REG_OFFSET(GC, 0, mmGDS_VMID0_SIZE) + 2 * vmid,
3930 				    gds_size);
3931 
3932 	/* GWS */
3933 	gfx_v10_0_write_data_to_reg(ring, 0, false,
3934 				    SOC15_REG_OFFSET(GC, 0, mmGDS_GWS_VMID0) + vmid,
3935 				    gws_size << GDS_GWS_VMID0__SIZE__SHIFT | gws_base);
3936 
3937 	/* OA */
3938 	gfx_v10_0_write_data_to_reg(ring, 0, false,
3939 				    SOC15_REG_OFFSET(GC, 0, mmGDS_OA_VMID0) + vmid,
3940 				    (1 << (oa_size + oa_base)) - (1 << oa_base));
3941 }
3942 
3943 static int gfx_v10_0_early_init(void *handle)
3944 {
3945 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
3946 
3947 	adev->gfx.num_gfx_rings = GFX10_NUM_GFX_RINGS;
3948 	adev->gfx.num_compute_rings = AMDGPU_MAX_COMPUTE_RINGS;
3949 
3950 	gfx_v10_0_set_kiq_pm4_funcs(adev);
3951 	gfx_v10_0_set_ring_funcs(adev);
3952 	gfx_v10_0_set_irq_funcs(adev);
3953 	gfx_v10_0_set_gds_init(adev);
3954 	gfx_v10_0_set_rlc_funcs(adev);
3955 
3956 	return 0;
3957 }
3958 
3959 static int gfx_v10_0_late_init(void *handle)
3960 {
3961 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
3962 	int r;
3963 
3964 	r = amdgpu_irq_get(adev, &adev->gfx.priv_reg_irq, 0);
3965 	if (r)
3966 		return r;
3967 
3968 	r = amdgpu_irq_get(adev, &adev->gfx.priv_inst_irq, 0);
3969 	if (r)
3970 		return r;
3971 
3972 	return 0;
3973 }
3974 
3975 static bool gfx_v10_0_is_rlc_enabled(struct amdgpu_device *adev)
3976 {
3977 	uint32_t rlc_cntl;
3978 
3979 	/* if RLC is not enabled, do nothing */
3980 	rlc_cntl = RREG32_SOC15(GC, 0, mmRLC_CNTL);
3981 	return (REG_GET_FIELD(rlc_cntl, RLC_CNTL, RLC_ENABLE_F32)) ? true : false;
3982 }
3983 
3984 static void gfx_v10_0_set_safe_mode(struct amdgpu_device *adev)
3985 {
3986 	uint32_t data;
3987 	unsigned i;
3988 
3989 	data = RLC_SAFE_MODE__CMD_MASK;
3990 	data |= (1 << RLC_SAFE_MODE__MESSAGE__SHIFT);
3991 	WREG32_SOC15(GC, 0, mmRLC_SAFE_MODE, data);
3992 
3993 	/* wait for RLC_SAFE_MODE */
3994 	for (i = 0; i < adev->usec_timeout; i++) {
3995 		if (!REG_GET_FIELD(RREG32_SOC15(GC, 0, mmRLC_SAFE_MODE), RLC_SAFE_MODE, CMD))
3996 			break;
3997 		udelay(1);
3998 	}
3999 }
4000 
4001 static void gfx_v10_0_unset_safe_mode(struct amdgpu_device *adev)
4002 {
4003 	uint32_t data;
4004 
4005 	data = RLC_SAFE_MODE__CMD_MASK;
4006 	WREG32_SOC15(GC, 0, mmRLC_SAFE_MODE, data);
4007 }
4008 
4009 static void gfx_v10_0_update_medium_grain_clock_gating(struct amdgpu_device *adev,
4010 						      bool enable)
4011 {
4012 	uint32_t data, def;
4013 
4014 	/* It is disabled by HW by default */
4015 	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGCG)) {
4016 		/* 1 - RLC_CGTT_MGCG_OVERRIDE */
4017 		def = data = RREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE);
4018 		data &= ~(RLC_CGTT_MGCG_OVERRIDE__GRBM_CGTT_SCLK_OVERRIDE_MASK |
4019 			  RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGCG_OVERRIDE_MASK |
4020 			  RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGLS_OVERRIDE_MASK);
4021 
4022 		/* only for Vega10 & Raven1 */
4023 		data |= RLC_CGTT_MGCG_OVERRIDE__RLC_CGTT_SCLK_OVERRIDE_MASK;
4024 
4025 		if (def != data)
4026 			WREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE, data);
4027 
4028 		/* MGLS is a global flag to control all MGLS in GFX */
4029 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGLS) {
4030 			/* 2 - RLC memory Light sleep */
4031 			if (adev->cg_flags & AMD_CG_SUPPORT_GFX_RLC_LS) {
4032 				def = data = RREG32_SOC15(GC, 0, mmRLC_MEM_SLP_CNTL);
4033 				data |= RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK;
4034 				if (def != data)
4035 					WREG32_SOC15(GC, 0, mmRLC_MEM_SLP_CNTL, data);
4036 			}
4037 			/* 3 - CP memory Light sleep */
4038 			if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CP_LS) {
4039 				def = data = RREG32_SOC15(GC, 0, mmCP_MEM_SLP_CNTL);
4040 				data |= CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK;
4041 				if (def != data)
4042 					WREG32_SOC15(GC, 0, mmCP_MEM_SLP_CNTL, data);
4043 			}
4044 		}
4045 	} else {
4046 		/* 1 - MGCG_OVERRIDE */
4047 		def = data = RREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE);
4048 		data |= (RLC_CGTT_MGCG_OVERRIDE__RLC_CGTT_SCLK_OVERRIDE_MASK |
4049 			 RLC_CGTT_MGCG_OVERRIDE__GRBM_CGTT_SCLK_OVERRIDE_MASK |
4050 			 RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGCG_OVERRIDE_MASK |
4051 			 RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGLS_OVERRIDE_MASK);
4052 		if (def != data)
4053 			WREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE, data);
4054 
4055 		/* 2 - disable MGLS in RLC */
4056 		data = RREG32_SOC15(GC, 0, mmRLC_MEM_SLP_CNTL);
4057 		if (data & RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK) {
4058 			data &= ~RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK;
4059 			WREG32_SOC15(GC, 0, mmRLC_MEM_SLP_CNTL, data);
4060 		}
4061 
4062 		/* 3 - disable MGLS in CP */
4063 		data = RREG32_SOC15(GC, 0, mmCP_MEM_SLP_CNTL);
4064 		if (data & CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK) {
4065 			data &= ~CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK;
4066 			WREG32_SOC15(GC, 0, mmCP_MEM_SLP_CNTL, data);
4067 		}
4068 	}
4069 }
4070 
4071 static void gfx_v10_0_update_3d_clock_gating(struct amdgpu_device *adev,
4072 					   bool enable)
4073 {
4074 	uint32_t data, def;
4075 
4076 	/* Enable 3D CGCG/CGLS */
4077 	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_3D_CGCG)) {
4078 		/* write cmd to clear cgcg/cgls ov */
4079 		def = data = RREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE);
4080 		/* unset CGCG override */
4081 		data &= ~RLC_CGTT_MGCG_OVERRIDE__GFXIP_GFX3D_CG_OVERRIDE_MASK;
4082 		/* update CGCG and CGLS override bits */
4083 		if (def != data)
4084 			WREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE, data);
4085 		/* enable 3Dcgcg FSM(0x0000363f) */
4086 		def = RREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL_3D);
4087 		data = (0x36 << RLC_CGCG_CGLS_CTRL_3D__CGCG_GFX_IDLE_THRESHOLD__SHIFT) |
4088 			RLC_CGCG_CGLS_CTRL_3D__CGCG_EN_MASK;
4089 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_3D_CGLS)
4090 			data |= (0x000F << RLC_CGCG_CGLS_CTRL_3D__CGLS_REP_COMPANSAT_DELAY__SHIFT) |
4091 				RLC_CGCG_CGLS_CTRL_3D__CGLS_EN_MASK;
4092 		if (def != data)
4093 			WREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL_3D, data);
4094 
4095 		/* set IDLE_POLL_COUNT(0x00900100) */
4096 		def = RREG32_SOC15(GC, 0, mmCP_RB_WPTR_POLL_CNTL);
4097 		data = (0x0100 << CP_RB_WPTR_POLL_CNTL__POLL_FREQUENCY__SHIFT) |
4098 			(0x0090 << CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT__SHIFT);
4099 		if (def != data)
4100 			WREG32_SOC15(GC, 0, mmCP_RB_WPTR_POLL_CNTL, data);
4101 	} else {
4102 		/* Disable CGCG/CGLS */
4103 		def = data = RREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL_3D);
4104 		/* disable cgcg, cgls should be disabled */
4105 		data &= ~(RLC_CGCG_CGLS_CTRL_3D__CGCG_EN_MASK |
4106 			  RLC_CGCG_CGLS_CTRL_3D__CGLS_EN_MASK);
4107 		/* disable cgcg and cgls in FSM */
4108 		if (def != data)
4109 			WREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL_3D, data);
4110 	}
4111 }
4112 
4113 static void gfx_v10_0_update_coarse_grain_clock_gating(struct amdgpu_device *adev,
4114 						      bool enable)
4115 {
4116 	uint32_t def, data;
4117 
4118 	if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGCG)) {
4119 		def = data = RREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE);
4120 		/* unset CGCG override */
4121 		data &= ~RLC_CGTT_MGCG_OVERRIDE__GFXIP_CGCG_OVERRIDE_MASK;
4122 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGLS)
4123 			data &= ~RLC_CGTT_MGCG_OVERRIDE__GFXIP_CGLS_OVERRIDE_MASK;
4124 		else
4125 			data |= RLC_CGTT_MGCG_OVERRIDE__GFXIP_CGLS_OVERRIDE_MASK;
4126 		/* update CGCG and CGLS override bits */
4127 		if (def != data)
4128 			WREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE, data);
4129 
4130 		/* enable cgcg FSM(0x0000363F) */
4131 		def = RREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL);
4132 		data = (0x36 << RLC_CGCG_CGLS_CTRL__CGCG_GFX_IDLE_THRESHOLD__SHIFT) |
4133 			RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK;
4134 		if (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGLS)
4135 			data |= (0x000F << RLC_CGCG_CGLS_CTRL__CGLS_REP_COMPANSAT_DELAY__SHIFT) |
4136 				RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK;
4137 		if (def != data)
4138 			WREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL, data);
4139 
4140 		/* set IDLE_POLL_COUNT(0x00900100) */
4141 		def = RREG32_SOC15(GC, 0, mmCP_RB_WPTR_POLL_CNTL);
4142 		data = (0x0100 << CP_RB_WPTR_POLL_CNTL__POLL_FREQUENCY__SHIFT) |
4143 			(0x0090 << CP_RB_WPTR_POLL_CNTL__IDLE_POLL_COUNT__SHIFT);
4144 		if (def != data)
4145 			WREG32_SOC15(GC, 0, mmCP_RB_WPTR_POLL_CNTL, data);
4146 	} else {
4147 		def = data = RREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL);
4148 		/* reset CGCG/CGLS bits */
4149 		data &= ~(RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK | RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK);
4150 		/* disable cgcg and cgls in FSM */
4151 		if (def != data)
4152 			WREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL, data);
4153 	}
4154 }
4155 
4156 static int gfx_v10_0_update_gfx_clock_gating(struct amdgpu_device *adev,
4157 					    bool enable)
4158 {
4159 	amdgpu_gfx_rlc_enter_safe_mode(adev);
4160 
4161 	if (enable) {
4162 		/* CGCG/CGLS should be enabled after MGCG/MGLS
4163 		 * ===  MGCG + MGLS ===
4164 		 */
4165 		gfx_v10_0_update_medium_grain_clock_gating(adev, enable);
4166 		/* ===  CGCG /CGLS for GFX 3D Only === */
4167 		gfx_v10_0_update_3d_clock_gating(adev, enable);
4168 		/* ===  CGCG + CGLS === */
4169 		gfx_v10_0_update_coarse_grain_clock_gating(adev, enable);
4170 	} else {
4171 		/* CGCG/CGLS should be disabled before MGCG/MGLS
4172 		 * ===  CGCG + CGLS ===
4173 		 */
4174 		gfx_v10_0_update_coarse_grain_clock_gating(adev, enable);
4175 		/* ===  CGCG /CGLS for GFX 3D Only === */
4176 		gfx_v10_0_update_3d_clock_gating(adev, enable);
4177 		/* ===  MGCG + MGLS === */
4178 		gfx_v10_0_update_medium_grain_clock_gating(adev, enable);
4179 	}
4180 
4181 	if (adev->cg_flags &
4182 	    (AMD_CG_SUPPORT_GFX_MGCG |
4183 	     AMD_CG_SUPPORT_GFX_CGLS |
4184 	     AMD_CG_SUPPORT_GFX_CGCG |
4185 	     AMD_CG_SUPPORT_GFX_CGLS |
4186 	     AMD_CG_SUPPORT_GFX_3D_CGCG |
4187 	     AMD_CG_SUPPORT_GFX_3D_CGLS))
4188 		gfx_v10_0_enable_gui_idle_interrupt(adev, enable);
4189 
4190 	amdgpu_gfx_rlc_exit_safe_mode(adev);
4191 
4192 	return 0;
4193 }
4194 
4195 static const struct amdgpu_rlc_funcs gfx_v10_0_rlc_funcs = {
4196 	.is_rlc_enabled = gfx_v10_0_is_rlc_enabled,
4197 	.set_safe_mode = gfx_v10_0_set_safe_mode,
4198 	.unset_safe_mode = gfx_v10_0_unset_safe_mode,
4199 	.init = gfx_v10_0_rlc_init,
4200 	.get_csb_size = gfx_v10_0_get_csb_size,
4201 	.get_csb_buffer = gfx_v10_0_get_csb_buffer,
4202 	.resume = gfx_v10_0_rlc_resume,
4203 	.stop = gfx_v10_0_rlc_stop,
4204 	.reset = gfx_v10_0_rlc_reset,
4205 	.start = gfx_v10_0_rlc_start
4206 };
4207 
4208 static int gfx_v10_0_set_powergating_state(void *handle,
4209 					  enum amd_powergating_state state)
4210 {
4211 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4212 	bool enable = (state == AMD_PG_STATE_GATE) ? true : false;
4213 	switch (adev->asic_type) {
4214 	case CHIP_NAVI10:
4215 	case CHIP_NAVI14:
4216 		if (!enable) {
4217 			amdgpu_gfx_off_ctrl(adev, false);
4218 			cancel_delayed_work_sync(&adev->gfx.gfx_off_delay_work);
4219 		} else
4220 			amdgpu_gfx_off_ctrl(adev, true);
4221 		break;
4222 	default:
4223 		break;
4224 	}
4225 	return 0;
4226 }
4227 
4228 static int gfx_v10_0_set_clockgating_state(void *handle,
4229 					  enum amd_clockgating_state state)
4230 {
4231 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4232 
4233 	switch (adev->asic_type) {
4234 	case CHIP_NAVI10:
4235 	case CHIP_NAVI14:
4236 	case CHIP_NAVI12:
4237 		gfx_v10_0_update_gfx_clock_gating(adev,
4238 						 state == AMD_CG_STATE_GATE ? true : false);
4239 		break;
4240 	default:
4241 		break;
4242 	}
4243 	return 0;
4244 }
4245 
4246 static void gfx_v10_0_get_clockgating_state(void *handle, u32 *flags)
4247 {
4248 	struct amdgpu_device *adev = (struct amdgpu_device *)handle;
4249 	int data;
4250 
4251 	/* AMD_CG_SUPPORT_GFX_MGCG */
4252 	data = RREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE);
4253 	if (!(data & RLC_CGTT_MGCG_OVERRIDE__GFXIP_MGCG_OVERRIDE_MASK))
4254 		*flags |= AMD_CG_SUPPORT_GFX_MGCG;
4255 
4256 	/* AMD_CG_SUPPORT_GFX_CGCG */
4257 	data = RREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL);
4258 	if (data & RLC_CGCG_CGLS_CTRL__CGCG_EN_MASK)
4259 		*flags |= AMD_CG_SUPPORT_GFX_CGCG;
4260 
4261 	/* AMD_CG_SUPPORT_GFX_CGLS */
4262 	if (data & RLC_CGCG_CGLS_CTRL__CGLS_EN_MASK)
4263 		*flags |= AMD_CG_SUPPORT_GFX_CGLS;
4264 
4265 	/* AMD_CG_SUPPORT_GFX_RLC_LS */
4266 	data = RREG32_SOC15(GC, 0, mmRLC_MEM_SLP_CNTL);
4267 	if (data & RLC_MEM_SLP_CNTL__RLC_MEM_LS_EN_MASK)
4268 		*flags |= AMD_CG_SUPPORT_GFX_RLC_LS | AMD_CG_SUPPORT_GFX_MGLS;
4269 
4270 	/* AMD_CG_SUPPORT_GFX_CP_LS */
4271 	data = RREG32_SOC15(GC, 0, mmCP_MEM_SLP_CNTL);
4272 	if (data & CP_MEM_SLP_CNTL__CP_MEM_LS_EN_MASK)
4273 		*flags |= AMD_CG_SUPPORT_GFX_CP_LS | AMD_CG_SUPPORT_GFX_MGLS;
4274 
4275 	/* AMD_CG_SUPPORT_GFX_3D_CGCG */
4276 	data = RREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL_3D);
4277 	if (data & RLC_CGCG_CGLS_CTRL_3D__CGCG_EN_MASK)
4278 		*flags |= AMD_CG_SUPPORT_GFX_3D_CGCG;
4279 
4280 	/* AMD_CG_SUPPORT_GFX_3D_CGLS */
4281 	if (data & RLC_CGCG_CGLS_CTRL_3D__CGLS_EN_MASK)
4282 		*flags |= AMD_CG_SUPPORT_GFX_3D_CGLS;
4283 }
4284 
4285 static u64 gfx_v10_0_ring_get_rptr_gfx(struct amdgpu_ring *ring)
4286 {
4287 	return ring->adev->wb.wb[ring->rptr_offs]; /* gfx10 is 32bit rptr*/
4288 }
4289 
4290 static u64 gfx_v10_0_ring_get_wptr_gfx(struct amdgpu_ring *ring)
4291 {
4292 	struct amdgpu_device *adev = ring->adev;
4293 	u64 wptr;
4294 
4295 	/* XXX check if swapping is necessary on BE */
4296 	if (ring->use_doorbell) {
4297 		wptr = atomic64_read((atomic64_t *)&adev->wb.wb[ring->wptr_offs]);
4298 	} else {
4299 		wptr = RREG32_SOC15(GC, 0, mmCP_RB0_WPTR);
4300 		wptr += (u64)RREG32_SOC15(GC, 0, mmCP_RB0_WPTR_HI) << 32;
4301 	}
4302 
4303 	return wptr;
4304 }
4305 
4306 static void gfx_v10_0_ring_set_wptr_gfx(struct amdgpu_ring *ring)
4307 {
4308 	struct amdgpu_device *adev = ring->adev;
4309 
4310 	if (ring->use_doorbell) {
4311 		/* XXX check if swapping is necessary on BE */
4312 		atomic64_set((atomic64_t *)&adev->wb.wb[ring->wptr_offs], ring->wptr);
4313 		WDOORBELL64(ring->doorbell_index, ring->wptr);
4314 	} else {
4315 		WREG32_SOC15(GC, 0, mmCP_RB0_WPTR, lower_32_bits(ring->wptr));
4316 		WREG32_SOC15(GC, 0, mmCP_RB0_WPTR_HI, upper_32_bits(ring->wptr));
4317 	}
4318 }
4319 
4320 static u64 gfx_v10_0_ring_get_rptr_compute(struct amdgpu_ring *ring)
4321 {
4322 	return ring->adev->wb.wb[ring->rptr_offs]; /* gfx10 hardware is 32bit rptr */
4323 }
4324 
4325 static u64 gfx_v10_0_ring_get_wptr_compute(struct amdgpu_ring *ring)
4326 {
4327 	u64 wptr;
4328 
4329 	/* XXX check if swapping is necessary on BE */
4330 	if (ring->use_doorbell)
4331 		wptr = atomic64_read((atomic64_t *)&ring->adev->wb.wb[ring->wptr_offs]);
4332 	else
4333 		BUG();
4334 	return wptr;
4335 }
4336 
4337 static void gfx_v10_0_ring_set_wptr_compute(struct amdgpu_ring *ring)
4338 {
4339 	struct amdgpu_device *adev = ring->adev;
4340 
4341 	/* XXX check if swapping is necessary on BE */
4342 	if (ring->use_doorbell) {
4343 		atomic64_set((atomic64_t *)&adev->wb.wb[ring->wptr_offs], ring->wptr);
4344 		WDOORBELL64(ring->doorbell_index, ring->wptr);
4345 	} else {
4346 		BUG(); /* only DOORBELL method supported on gfx10 now */
4347 	}
4348 }
4349 
4350 static void gfx_v10_0_ring_emit_hdp_flush(struct amdgpu_ring *ring)
4351 {
4352 	struct amdgpu_device *adev = ring->adev;
4353 	u32 ref_and_mask, reg_mem_engine;
4354 	const struct nbio_hdp_flush_reg *nbio_hf_reg = adev->nbio.hdp_flush_reg;
4355 
4356 	if (ring->funcs->type == AMDGPU_RING_TYPE_COMPUTE) {
4357 		switch (ring->me) {
4358 		case 1:
4359 			ref_and_mask = nbio_hf_reg->ref_and_mask_cp2 << ring->pipe;
4360 			break;
4361 		case 2:
4362 			ref_and_mask = nbio_hf_reg->ref_and_mask_cp6 << ring->pipe;
4363 			break;
4364 		default:
4365 			return;
4366 		}
4367 		reg_mem_engine = 0;
4368 	} else {
4369 		ref_and_mask = nbio_hf_reg->ref_and_mask_cp0;
4370 		reg_mem_engine = 1; /* pfp */
4371 	}
4372 
4373 	gfx_v10_0_wait_reg_mem(ring, reg_mem_engine, 0, 1,
4374 			       adev->nbio.funcs->get_hdp_flush_req_offset(adev),
4375 			       adev->nbio.funcs->get_hdp_flush_done_offset(adev),
4376 			       ref_and_mask, ref_and_mask, 0x20);
4377 }
4378 
4379 static void gfx_v10_0_ring_emit_ib_gfx(struct amdgpu_ring *ring,
4380 				       struct amdgpu_job *job,
4381 				       struct amdgpu_ib *ib,
4382 				       uint32_t flags)
4383 {
4384 	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
4385 	u32 header, control = 0;
4386 
4387 	if (ib->flags & AMDGPU_IB_FLAG_CE)
4388 		header = PACKET3(PACKET3_INDIRECT_BUFFER_CNST, 2);
4389 	else
4390 		header = PACKET3(PACKET3_INDIRECT_BUFFER, 2);
4391 
4392 	control |= ib->length_dw | (vmid << 24);
4393 
4394 	if (amdgpu_mcbp && (ib->flags & AMDGPU_IB_FLAG_PREEMPT)) {
4395 		control |= INDIRECT_BUFFER_PRE_ENB(1);
4396 
4397 		if (flags & AMDGPU_IB_PREEMPTED)
4398 			control |= INDIRECT_BUFFER_PRE_RESUME(1);
4399 
4400 		if (!(ib->flags & AMDGPU_IB_FLAG_CE))
4401 			gfx_v10_0_ring_emit_de_meta(ring,
4402 				    flags & AMDGPU_IB_PREEMPTED ? true : false);
4403 	}
4404 
4405 	amdgpu_ring_write(ring, header);
4406 	BUG_ON(ib->gpu_addr & 0x3); /* Dword align */
4407 	amdgpu_ring_write(ring,
4408 #ifdef __BIG_ENDIAN
4409 		(2 << 0) |
4410 #endif
4411 		lower_32_bits(ib->gpu_addr));
4412 	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr));
4413 	amdgpu_ring_write(ring, control);
4414 }
4415 
4416 static void gfx_v10_0_ring_emit_ib_compute(struct amdgpu_ring *ring,
4417 					   struct amdgpu_job *job,
4418 					   struct amdgpu_ib *ib,
4419 					   uint32_t flags)
4420 {
4421 	unsigned vmid = AMDGPU_JOB_GET_VMID(job);
4422 	u32 control = INDIRECT_BUFFER_VALID | ib->length_dw | (vmid << 24);
4423 
4424 	/* Currently, there is a high possibility to get wave ID mismatch
4425 	 * between ME and GDS, leading to a hw deadlock, because ME generates
4426 	 * different wave IDs than the GDS expects. This situation happens
4427 	 * randomly when at least 5 compute pipes use GDS ordered append.
4428 	 * The wave IDs generated by ME are also wrong after suspend/resume.
4429 	 * Those are probably bugs somewhere else in the kernel driver.
4430 	 *
4431 	 * Writing GDS_COMPUTE_MAX_WAVE_ID resets wave ID counters in ME and
4432 	 * GDS to 0 for this ring (me/pipe).
4433 	 */
4434 	if (ib->flags & AMDGPU_IB_FLAG_RESET_GDS_MAX_WAVE_ID) {
4435 		amdgpu_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1));
4436 		amdgpu_ring_write(ring, mmGDS_COMPUTE_MAX_WAVE_ID);
4437 		amdgpu_ring_write(ring, ring->adev->gds.gds_compute_max_wave_id);
4438 	}
4439 
4440 	amdgpu_ring_write(ring, PACKET3(PACKET3_INDIRECT_BUFFER, 2));
4441 	BUG_ON(ib->gpu_addr & 0x3); /* Dword align */
4442 	amdgpu_ring_write(ring,
4443 #ifdef __BIG_ENDIAN
4444 				(2 << 0) |
4445 #endif
4446 				lower_32_bits(ib->gpu_addr));
4447 	amdgpu_ring_write(ring, upper_32_bits(ib->gpu_addr));
4448 	amdgpu_ring_write(ring, control);
4449 }
4450 
4451 static void gfx_v10_0_ring_emit_fence(struct amdgpu_ring *ring, u64 addr,
4452 				     u64 seq, unsigned flags)
4453 {
4454 	struct amdgpu_device *adev = ring->adev;
4455 	bool write64bit = flags & AMDGPU_FENCE_FLAG_64BIT;
4456 	bool int_sel = flags & AMDGPU_FENCE_FLAG_INT;
4457 
4458 	/* Interrupt not work fine on GFX10.1 model yet. Use fallback instead */
4459 	if (adev->pdev->device == 0x50)
4460 		int_sel = false;
4461 
4462 	/* RELEASE_MEM - flush caches, send int */
4463 	amdgpu_ring_write(ring, PACKET3(PACKET3_RELEASE_MEM, 6));
4464 	amdgpu_ring_write(ring, (PACKET3_RELEASE_MEM_GCR_SEQ |
4465 				 PACKET3_RELEASE_MEM_GCR_GL2_WB |
4466 				 PACKET3_RELEASE_MEM_GCR_GLM_INV | /* must be set with GLM_WB */
4467 				 PACKET3_RELEASE_MEM_GCR_GLM_WB |
4468 				 PACKET3_RELEASE_MEM_CACHE_POLICY(3) |
4469 				 PACKET3_RELEASE_MEM_EVENT_TYPE(CACHE_FLUSH_AND_INV_TS_EVENT) |
4470 				 PACKET3_RELEASE_MEM_EVENT_INDEX(5)));
4471 	amdgpu_ring_write(ring, (PACKET3_RELEASE_MEM_DATA_SEL(write64bit ? 2 : 1) |
4472 				 PACKET3_RELEASE_MEM_INT_SEL(int_sel ? 2 : 0)));
4473 
4474 	/*
4475 	 * the address should be Qword aligned if 64bit write, Dword
4476 	 * aligned if only send 32bit data low (discard data high)
4477 	 */
4478 	if (write64bit)
4479 		BUG_ON(addr & 0x7);
4480 	else
4481 		BUG_ON(addr & 0x3);
4482 	amdgpu_ring_write(ring, lower_32_bits(addr));
4483 	amdgpu_ring_write(ring, upper_32_bits(addr));
4484 	amdgpu_ring_write(ring, lower_32_bits(seq));
4485 	amdgpu_ring_write(ring, upper_32_bits(seq));
4486 	amdgpu_ring_write(ring, 0);
4487 }
4488 
4489 static void gfx_v10_0_ring_emit_pipeline_sync(struct amdgpu_ring *ring)
4490 {
4491 	int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX);
4492 	uint32_t seq = ring->fence_drv.sync_seq;
4493 	uint64_t addr = ring->fence_drv.gpu_addr;
4494 
4495 	gfx_v10_0_wait_reg_mem(ring, usepfp, 1, 0, lower_32_bits(addr),
4496 			       upper_32_bits(addr), seq, 0xffffffff, 4);
4497 }
4498 
4499 static void gfx_v10_0_ring_emit_vm_flush(struct amdgpu_ring *ring,
4500 					 unsigned vmid, uint64_t pd_addr)
4501 {
4502 	amdgpu_gmc_emit_flush_gpu_tlb(ring, vmid, pd_addr);
4503 
4504 	/* compute doesn't have PFP */
4505 	if (ring->funcs->type == AMDGPU_RING_TYPE_GFX) {
4506 		/* sync PFP to ME, otherwise we might get invalid PFP reads */
4507 		amdgpu_ring_write(ring, PACKET3(PACKET3_PFP_SYNC_ME, 0));
4508 		amdgpu_ring_write(ring, 0x0);
4509 	}
4510 }
4511 
4512 static void gfx_v10_0_ring_emit_fence_kiq(struct amdgpu_ring *ring, u64 addr,
4513 					  u64 seq, unsigned int flags)
4514 {
4515 	struct amdgpu_device *adev = ring->adev;
4516 
4517 	/* we only allocate 32bit for each seq wb address */
4518 	BUG_ON(flags & AMDGPU_FENCE_FLAG_64BIT);
4519 
4520 	/* write fence seq to the "addr" */
4521 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
4522 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
4523 				 WRITE_DATA_DST_SEL(5) | WR_CONFIRM));
4524 	amdgpu_ring_write(ring, lower_32_bits(addr));
4525 	amdgpu_ring_write(ring, upper_32_bits(addr));
4526 	amdgpu_ring_write(ring, lower_32_bits(seq));
4527 
4528 	if (flags & AMDGPU_FENCE_FLAG_INT) {
4529 		/* set register to trigger INT */
4530 		amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
4531 		amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(0) |
4532 					 WRITE_DATA_DST_SEL(0) | WR_CONFIRM));
4533 		amdgpu_ring_write(ring, SOC15_REG_OFFSET(GC, 0, mmCPC_INT_STATUS));
4534 		amdgpu_ring_write(ring, 0);
4535 		amdgpu_ring_write(ring, 0x20000000); /* src_id is 178 */
4536 	}
4537 }
4538 
4539 static void gfx_v10_0_ring_emit_sb(struct amdgpu_ring *ring)
4540 {
4541 	amdgpu_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
4542 	amdgpu_ring_write(ring, 0);
4543 }
4544 
4545 static void gfx_v10_0_ring_emit_cntxcntl(struct amdgpu_ring *ring, uint32_t flags)
4546 {
4547 	uint32_t dw2 = 0;
4548 
4549 	if (amdgpu_mcbp)
4550 		gfx_v10_0_ring_emit_ce_meta(ring,
4551 				    flags & AMDGPU_IB_PREEMPTED ? true : false);
4552 
4553 	gfx_v10_0_ring_emit_tmz(ring, true);
4554 
4555 	dw2 |= 0x80000000; /* set load_enable otherwise this package is just NOPs */
4556 	if (flags & AMDGPU_HAVE_CTX_SWITCH) {
4557 		/* set load_global_config & load_global_uconfig */
4558 		dw2 |= 0x8001;
4559 		/* set load_cs_sh_regs */
4560 		dw2 |= 0x01000000;
4561 		/* set load_per_context_state & load_gfx_sh_regs for GFX */
4562 		dw2 |= 0x10002;
4563 
4564 		/* set load_ce_ram if preamble presented */
4565 		if (AMDGPU_PREAMBLE_IB_PRESENT & flags)
4566 			dw2 |= 0x10000000;
4567 	} else {
4568 		/* still load_ce_ram if this is the first time preamble presented
4569 		 * although there is no context switch happens.
4570 		 */
4571 		if (AMDGPU_PREAMBLE_IB_PRESENT_FIRST & flags)
4572 			dw2 |= 0x10000000;
4573 	}
4574 
4575 	amdgpu_ring_write(ring, PACKET3(PACKET3_CONTEXT_CONTROL, 1));
4576 	amdgpu_ring_write(ring, dw2);
4577 	amdgpu_ring_write(ring, 0);
4578 }
4579 
4580 static unsigned gfx_v10_0_ring_emit_init_cond_exec(struct amdgpu_ring *ring)
4581 {
4582 	unsigned ret;
4583 
4584 	amdgpu_ring_write(ring, PACKET3(PACKET3_COND_EXEC, 3));
4585 	amdgpu_ring_write(ring, lower_32_bits(ring->cond_exe_gpu_addr));
4586 	amdgpu_ring_write(ring, upper_32_bits(ring->cond_exe_gpu_addr));
4587 	amdgpu_ring_write(ring, 0); /* discard following DWs if *cond_exec_gpu_addr==0 */
4588 	ret = ring->wptr & ring->buf_mask;
4589 	amdgpu_ring_write(ring, 0x55aa55aa); /* patch dummy value later */
4590 
4591 	return ret;
4592 }
4593 
4594 static void gfx_v10_0_ring_emit_patch_cond_exec(struct amdgpu_ring *ring, unsigned offset)
4595 {
4596 	unsigned cur;
4597 	BUG_ON(offset > ring->buf_mask);
4598 	BUG_ON(ring->ring[offset] != 0x55aa55aa);
4599 
4600 	cur = (ring->wptr - 1) & ring->buf_mask;
4601 	if (likely(cur > offset))
4602 		ring->ring[offset] = cur - offset;
4603 	else
4604 		ring->ring[offset] = (ring->buf_mask + 1) - offset + cur;
4605 }
4606 
4607 static int gfx_v10_0_ring_preempt_ib(struct amdgpu_ring *ring)
4608 {
4609 	int i, r = 0;
4610 	struct amdgpu_device *adev = ring->adev;
4611 	struct amdgpu_kiq *kiq = &adev->gfx.kiq;
4612 	struct amdgpu_ring *kiq_ring = &kiq->ring;
4613 
4614 	if (!kiq->pmf || !kiq->pmf->kiq_unmap_queues)
4615 		return -EINVAL;
4616 
4617 	if (amdgpu_ring_alloc(kiq_ring, kiq->pmf->unmap_queues_size))
4618 		return -ENOMEM;
4619 
4620 	/* assert preemption condition */
4621 	amdgpu_ring_set_preempt_cond_exec(ring, false);
4622 
4623 	/* assert IB preemption, emit the trailing fence */
4624 	kiq->pmf->kiq_unmap_queues(kiq_ring, ring, PREEMPT_QUEUES_NO_UNMAP,
4625 				   ring->trail_fence_gpu_addr,
4626 				   ++ring->trail_seq);
4627 	amdgpu_ring_commit(kiq_ring);
4628 
4629 	/* poll the trailing fence */
4630 	for (i = 0; i < adev->usec_timeout; i++) {
4631 		if (ring->trail_seq ==
4632 		    le32_to_cpu(*(ring->trail_fence_cpu_addr)))
4633 			break;
4634 		udelay(1);
4635 	}
4636 
4637 	if (i >= adev->usec_timeout) {
4638 		r = -EINVAL;
4639 		DRM_ERROR("ring %d failed to preempt ib\n", ring->idx);
4640 	}
4641 
4642 	/* deassert preemption condition */
4643 	amdgpu_ring_set_preempt_cond_exec(ring, true);
4644 	return r;
4645 }
4646 
4647 static void gfx_v10_0_ring_emit_ce_meta(struct amdgpu_ring *ring, bool resume)
4648 {
4649 	struct amdgpu_device *adev = ring->adev;
4650 	struct v10_ce_ib_state ce_payload = {0};
4651 	uint64_t csa_addr;
4652 	int cnt;
4653 
4654 	cnt = (sizeof(ce_payload) >> 2) + 4 - 2;
4655 	csa_addr = amdgpu_csa_vaddr(ring->adev);
4656 
4657 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, cnt));
4658 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(2) |
4659 				 WRITE_DATA_DST_SEL(8) |
4660 				 WR_CONFIRM) |
4661 				 WRITE_DATA_CACHE_POLICY(0));
4662 	amdgpu_ring_write(ring, lower_32_bits(csa_addr +
4663 			      offsetof(struct v10_gfx_meta_data, ce_payload)));
4664 	amdgpu_ring_write(ring, upper_32_bits(csa_addr +
4665 			      offsetof(struct v10_gfx_meta_data, ce_payload)));
4666 
4667 	if (resume)
4668 		amdgpu_ring_write_multiple(ring, adev->virt.csa_cpu_addr +
4669 					   offsetof(struct v10_gfx_meta_data,
4670 						    ce_payload),
4671 					   sizeof(ce_payload) >> 2);
4672 	else
4673 		amdgpu_ring_write_multiple(ring, (void *)&ce_payload,
4674 					   sizeof(ce_payload) >> 2);
4675 }
4676 
4677 static void gfx_v10_0_ring_emit_de_meta(struct amdgpu_ring *ring, bool resume)
4678 {
4679 	struct amdgpu_device *adev = ring->adev;
4680 	struct v10_de_ib_state de_payload = {0};
4681 	uint64_t csa_addr, gds_addr;
4682 	int cnt;
4683 
4684 	csa_addr = amdgpu_csa_vaddr(ring->adev);
4685 	gds_addr = ALIGN(csa_addr + AMDGPU_CSA_SIZE - adev->gds.gds_size,
4686 			 PAGE_SIZE);
4687 	de_payload.gds_backup_addrlo = lower_32_bits(gds_addr);
4688 	de_payload.gds_backup_addrhi = upper_32_bits(gds_addr);
4689 
4690 	cnt = (sizeof(de_payload) >> 2) + 4 - 2;
4691 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, cnt));
4692 	amdgpu_ring_write(ring, (WRITE_DATA_ENGINE_SEL(1) |
4693 				 WRITE_DATA_DST_SEL(8) |
4694 				 WR_CONFIRM) |
4695 				 WRITE_DATA_CACHE_POLICY(0));
4696 	amdgpu_ring_write(ring, lower_32_bits(csa_addr +
4697 			      offsetof(struct v10_gfx_meta_data, de_payload)));
4698 	amdgpu_ring_write(ring, upper_32_bits(csa_addr +
4699 			      offsetof(struct v10_gfx_meta_data, de_payload)));
4700 
4701 	if (resume)
4702 		amdgpu_ring_write_multiple(ring, adev->virt.csa_cpu_addr +
4703 					   offsetof(struct v10_gfx_meta_data,
4704 						    de_payload),
4705 					   sizeof(de_payload) >> 2);
4706 	else
4707 		amdgpu_ring_write_multiple(ring, (void *)&de_payload,
4708 					   sizeof(de_payload) >> 2);
4709 }
4710 
4711 static void gfx_v10_0_ring_emit_tmz(struct amdgpu_ring *ring, bool start)
4712 {
4713 	amdgpu_ring_write(ring, PACKET3(PACKET3_FRAME_CONTROL, 0));
4714 	amdgpu_ring_write(ring, FRAME_CMD(start ? 0 : 1)); /* frame_end */
4715 }
4716 
4717 static void gfx_v10_0_ring_emit_rreg(struct amdgpu_ring *ring, uint32_t reg)
4718 {
4719 	struct amdgpu_device *adev = ring->adev;
4720 
4721 	amdgpu_ring_write(ring, PACKET3(PACKET3_COPY_DATA, 4));
4722 	amdgpu_ring_write(ring, 0 |	/* src: register*/
4723 				(5 << 8) |	/* dst: memory */
4724 				(1 << 20));	/* write confirm */
4725 	amdgpu_ring_write(ring, reg);
4726 	amdgpu_ring_write(ring, 0);
4727 	amdgpu_ring_write(ring, lower_32_bits(adev->wb.gpu_addr +
4728 				adev->virt.reg_val_offs * 4));
4729 	amdgpu_ring_write(ring, upper_32_bits(adev->wb.gpu_addr +
4730 				adev->virt.reg_val_offs * 4));
4731 }
4732 
4733 static void gfx_v10_0_ring_emit_wreg(struct amdgpu_ring *ring, uint32_t reg,
4734 				   uint32_t val)
4735 {
4736 	uint32_t cmd = 0;
4737 
4738 	switch (ring->funcs->type) {
4739 	case AMDGPU_RING_TYPE_GFX:
4740 		cmd = WRITE_DATA_ENGINE_SEL(1) | WR_CONFIRM;
4741 		break;
4742 	case AMDGPU_RING_TYPE_KIQ:
4743 		cmd = (1 << 16); /* no inc addr */
4744 		break;
4745 	default:
4746 		cmd = WR_CONFIRM;
4747 		break;
4748 	}
4749 	amdgpu_ring_write(ring, PACKET3(PACKET3_WRITE_DATA, 3));
4750 	amdgpu_ring_write(ring, cmd);
4751 	amdgpu_ring_write(ring, reg);
4752 	amdgpu_ring_write(ring, 0);
4753 	amdgpu_ring_write(ring, val);
4754 }
4755 
4756 static void gfx_v10_0_ring_emit_reg_wait(struct amdgpu_ring *ring, uint32_t reg,
4757 					uint32_t val, uint32_t mask)
4758 {
4759 	gfx_v10_0_wait_reg_mem(ring, 0, 0, 0, reg, 0, val, mask, 0x20);
4760 }
4761 
4762 static void gfx_v10_0_ring_emit_reg_write_reg_wait(struct amdgpu_ring *ring,
4763 						   uint32_t reg0, uint32_t reg1,
4764 						   uint32_t ref, uint32_t mask)
4765 {
4766 	int usepfp = (ring->funcs->type == AMDGPU_RING_TYPE_GFX);
4767 	struct amdgpu_device *adev = ring->adev;
4768 	bool fw_version_ok = false;
4769 
4770 	fw_version_ok = adev->gfx.cp_fw_write_wait;
4771 
4772 	if (fw_version_ok)
4773 		gfx_v10_0_wait_reg_mem(ring, usepfp, 0, 1, reg0, reg1,
4774 				       ref, mask, 0x20);
4775 	else
4776 		amdgpu_ring_emit_reg_write_reg_wait_helper(ring, reg0, reg1,
4777 							   ref, mask);
4778 }
4779 
4780 static void
4781 gfx_v10_0_set_gfx_eop_interrupt_state(struct amdgpu_device *adev,
4782 				      uint32_t me, uint32_t pipe,
4783 				      enum amdgpu_interrupt_state state)
4784 {
4785 	uint32_t cp_int_cntl, cp_int_cntl_reg;
4786 
4787 	if (!me) {
4788 		switch (pipe) {
4789 		case 0:
4790 			cp_int_cntl_reg = SOC15_REG_OFFSET(GC, 0, mmCP_INT_CNTL_RING0);
4791 			break;
4792 		case 1:
4793 			cp_int_cntl_reg = SOC15_REG_OFFSET(GC, 0, mmCP_INT_CNTL_RING1);
4794 			break;
4795 		default:
4796 			DRM_DEBUG("invalid pipe %d\n", pipe);
4797 			return;
4798 		}
4799 	} else {
4800 		DRM_DEBUG("invalid me %d\n", me);
4801 		return;
4802 	}
4803 
4804 	switch (state) {
4805 	case AMDGPU_IRQ_STATE_DISABLE:
4806 		cp_int_cntl = RREG32(cp_int_cntl_reg);
4807 		cp_int_cntl = REG_SET_FIELD(cp_int_cntl, CP_INT_CNTL_RING0,
4808 					    TIME_STAMP_INT_ENABLE, 0);
4809 		WREG32(cp_int_cntl_reg, cp_int_cntl);
4810 		break;
4811 	case AMDGPU_IRQ_STATE_ENABLE:
4812 		cp_int_cntl = RREG32(cp_int_cntl_reg);
4813 		cp_int_cntl = REG_SET_FIELD(cp_int_cntl, CP_INT_CNTL_RING0,
4814 					    TIME_STAMP_INT_ENABLE, 1);
4815 		WREG32(cp_int_cntl_reg, cp_int_cntl);
4816 		break;
4817 	default:
4818 		break;
4819 	}
4820 }
4821 
4822 static void gfx_v10_0_set_compute_eop_interrupt_state(struct amdgpu_device *adev,
4823 						     int me, int pipe,
4824 						     enum amdgpu_interrupt_state state)
4825 {
4826 	u32 mec_int_cntl, mec_int_cntl_reg;
4827 
4828 	/*
4829 	 * amdgpu controls only the first MEC. That's why this function only
4830 	 * handles the setting of interrupts for this specific MEC. All other
4831 	 * pipes' interrupts are set by amdkfd.
4832 	 */
4833 
4834 	if (me == 1) {
4835 		switch (pipe) {
4836 		case 0:
4837 			mec_int_cntl_reg = SOC15_REG_OFFSET(GC, 0, mmCP_ME1_PIPE0_INT_CNTL);
4838 			break;
4839 		case 1:
4840 			mec_int_cntl_reg = SOC15_REG_OFFSET(GC, 0, mmCP_ME1_PIPE1_INT_CNTL);
4841 			break;
4842 		case 2:
4843 			mec_int_cntl_reg = SOC15_REG_OFFSET(GC, 0, mmCP_ME1_PIPE2_INT_CNTL);
4844 			break;
4845 		case 3:
4846 			mec_int_cntl_reg = SOC15_REG_OFFSET(GC, 0, mmCP_ME1_PIPE3_INT_CNTL);
4847 			break;
4848 		default:
4849 			DRM_DEBUG("invalid pipe %d\n", pipe);
4850 			return;
4851 		}
4852 	} else {
4853 		DRM_DEBUG("invalid me %d\n", me);
4854 		return;
4855 	}
4856 
4857 	switch (state) {
4858 	case AMDGPU_IRQ_STATE_DISABLE:
4859 		mec_int_cntl = RREG32(mec_int_cntl_reg);
4860 		mec_int_cntl = REG_SET_FIELD(mec_int_cntl, CP_ME1_PIPE0_INT_CNTL,
4861 					     TIME_STAMP_INT_ENABLE, 0);
4862 		WREG32(mec_int_cntl_reg, mec_int_cntl);
4863 		break;
4864 	case AMDGPU_IRQ_STATE_ENABLE:
4865 		mec_int_cntl = RREG32(mec_int_cntl_reg);
4866 		mec_int_cntl = REG_SET_FIELD(mec_int_cntl, CP_ME1_PIPE0_INT_CNTL,
4867 					     TIME_STAMP_INT_ENABLE, 1);
4868 		WREG32(mec_int_cntl_reg, mec_int_cntl);
4869 		break;
4870 	default:
4871 		break;
4872 	}
4873 }
4874 
4875 static int gfx_v10_0_set_eop_interrupt_state(struct amdgpu_device *adev,
4876 					    struct amdgpu_irq_src *src,
4877 					    unsigned type,
4878 					    enum amdgpu_interrupt_state state)
4879 {
4880 	switch (type) {
4881 	case AMDGPU_CP_IRQ_GFX_ME0_PIPE0_EOP:
4882 		gfx_v10_0_set_gfx_eop_interrupt_state(adev, 0, 0, state);
4883 		break;
4884 	case AMDGPU_CP_IRQ_GFX_ME0_PIPE1_EOP:
4885 		gfx_v10_0_set_gfx_eop_interrupt_state(adev, 0, 1, state);
4886 		break;
4887 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE0_EOP:
4888 		gfx_v10_0_set_compute_eop_interrupt_state(adev, 1, 0, state);
4889 		break;
4890 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE1_EOP:
4891 		gfx_v10_0_set_compute_eop_interrupt_state(adev, 1, 1, state);
4892 		break;
4893 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE2_EOP:
4894 		gfx_v10_0_set_compute_eop_interrupt_state(adev, 1, 2, state);
4895 		break;
4896 	case AMDGPU_CP_IRQ_COMPUTE_MEC1_PIPE3_EOP:
4897 		gfx_v10_0_set_compute_eop_interrupt_state(adev, 1, 3, state);
4898 		break;
4899 	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE0_EOP:
4900 		gfx_v10_0_set_compute_eop_interrupt_state(adev, 2, 0, state);
4901 		break;
4902 	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE1_EOP:
4903 		gfx_v10_0_set_compute_eop_interrupt_state(adev, 2, 1, state);
4904 		break;
4905 	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE2_EOP:
4906 		gfx_v10_0_set_compute_eop_interrupt_state(adev, 2, 2, state);
4907 		break;
4908 	case AMDGPU_CP_IRQ_COMPUTE_MEC2_PIPE3_EOP:
4909 		gfx_v10_0_set_compute_eop_interrupt_state(adev, 2, 3, state);
4910 		break;
4911 	default:
4912 		break;
4913 	}
4914 	return 0;
4915 }
4916 
4917 static int gfx_v10_0_eop_irq(struct amdgpu_device *adev,
4918 			     struct amdgpu_irq_src *source,
4919 			     struct amdgpu_iv_entry *entry)
4920 {
4921 	int i;
4922 	u8 me_id, pipe_id, queue_id;
4923 	struct amdgpu_ring *ring;
4924 
4925 	DRM_DEBUG("IH: CP EOP\n");
4926 	me_id = (entry->ring_id & 0x0c) >> 2;
4927 	pipe_id = (entry->ring_id & 0x03) >> 0;
4928 	queue_id = (entry->ring_id & 0x70) >> 4;
4929 
4930 	switch (me_id) {
4931 	case 0:
4932 		if (pipe_id == 0)
4933 			amdgpu_fence_process(&adev->gfx.gfx_ring[0]);
4934 		else
4935 			amdgpu_fence_process(&adev->gfx.gfx_ring[1]);
4936 		break;
4937 	case 1:
4938 	case 2:
4939 		for (i = 0; i < adev->gfx.num_compute_rings; i++) {
4940 			ring = &adev->gfx.compute_ring[i];
4941 			/* Per-queue interrupt is supported for MEC starting from VI.
4942 			  * The interrupt can only be enabled/disabled per pipe instead of per queue.
4943 			  */
4944 			if ((ring->me == me_id) && (ring->pipe == pipe_id) && (ring->queue == queue_id))
4945 				amdgpu_fence_process(ring);
4946 		}
4947 		break;
4948 	}
4949 	return 0;
4950 }
4951 
4952 static int gfx_v10_0_set_priv_reg_fault_state(struct amdgpu_device *adev,
4953 					      struct amdgpu_irq_src *source,
4954 					      unsigned type,
4955 					      enum amdgpu_interrupt_state state)
4956 {
4957 	switch (state) {
4958 	case AMDGPU_IRQ_STATE_DISABLE:
4959 	case AMDGPU_IRQ_STATE_ENABLE:
4960 		WREG32_FIELD15(GC, 0, CP_INT_CNTL_RING0,
4961 			       PRIV_REG_INT_ENABLE,
4962 			       state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0);
4963 		break;
4964 	default:
4965 		break;
4966 	}
4967 
4968 	return 0;
4969 }
4970 
4971 static int gfx_v10_0_set_priv_inst_fault_state(struct amdgpu_device *adev,
4972 					       struct amdgpu_irq_src *source,
4973 					       unsigned type,
4974 					       enum amdgpu_interrupt_state state)
4975 {
4976 	switch (state) {
4977 	case AMDGPU_IRQ_STATE_DISABLE:
4978 	case AMDGPU_IRQ_STATE_ENABLE:
4979 		WREG32_FIELD15(GC, 0, CP_INT_CNTL_RING0,
4980 			       PRIV_INSTR_INT_ENABLE,
4981 			       state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0);
4982 	default:
4983 		break;
4984 	}
4985 
4986 	return 0;
4987 }
4988 
4989 static void gfx_v10_0_handle_priv_fault(struct amdgpu_device *adev,
4990 					struct amdgpu_iv_entry *entry)
4991 {
4992 	u8 me_id, pipe_id, queue_id;
4993 	struct amdgpu_ring *ring;
4994 	int i;
4995 
4996 	me_id = (entry->ring_id & 0x0c) >> 2;
4997 	pipe_id = (entry->ring_id & 0x03) >> 0;
4998 	queue_id = (entry->ring_id & 0x70) >> 4;
4999 
5000 	switch (me_id) {
5001 	case 0:
5002 		for (i = 0; i < adev->gfx.num_gfx_rings; i++) {
5003 			ring = &adev->gfx.gfx_ring[i];
5004 			/* we only enabled 1 gfx queue per pipe for now */
5005 			if (ring->me == me_id && ring->pipe == pipe_id)
5006 				drm_sched_fault(&ring->sched);
5007 		}
5008 		break;
5009 	case 1:
5010 	case 2:
5011 		for (i = 0; i < adev->gfx.num_compute_rings; i++) {
5012 			ring = &adev->gfx.compute_ring[i];
5013 			if (ring->me == me_id && ring->pipe == pipe_id &&
5014 			    ring->queue == queue_id)
5015 				drm_sched_fault(&ring->sched);
5016 		}
5017 		break;
5018 	default:
5019 		BUG();
5020 	}
5021 }
5022 
5023 static int gfx_v10_0_priv_reg_irq(struct amdgpu_device *adev,
5024 				  struct amdgpu_irq_src *source,
5025 				  struct amdgpu_iv_entry *entry)
5026 {
5027 	DRM_ERROR("Illegal register access in command stream\n");
5028 	gfx_v10_0_handle_priv_fault(adev, entry);
5029 	return 0;
5030 }
5031 
5032 static int gfx_v10_0_priv_inst_irq(struct amdgpu_device *adev,
5033 				   struct amdgpu_irq_src *source,
5034 				   struct amdgpu_iv_entry *entry)
5035 {
5036 	DRM_ERROR("Illegal instruction in command stream\n");
5037 	gfx_v10_0_handle_priv_fault(adev, entry);
5038 	return 0;
5039 }
5040 
5041 static int gfx_v10_0_kiq_set_interrupt_state(struct amdgpu_device *adev,
5042 					     struct amdgpu_irq_src *src,
5043 					     unsigned int type,
5044 					     enum amdgpu_interrupt_state state)
5045 {
5046 	uint32_t tmp, target;
5047 	struct amdgpu_ring *ring = &(adev->gfx.kiq.ring);
5048 
5049 	if (ring->me == 1)
5050 		target = SOC15_REG_OFFSET(GC, 0, mmCP_ME1_PIPE0_INT_CNTL);
5051 	else
5052 		target = SOC15_REG_OFFSET(GC, 0, mmCP_ME2_PIPE0_INT_CNTL);
5053 	target += ring->pipe;
5054 
5055 	switch (type) {
5056 	case AMDGPU_CP_KIQ_IRQ_DRIVER0:
5057 		if (state == AMDGPU_IRQ_STATE_DISABLE) {
5058 			tmp = RREG32_SOC15(GC, 0, mmCPC_INT_CNTL);
5059 			tmp = REG_SET_FIELD(tmp, CPC_INT_CNTL,
5060 					    GENERIC2_INT_ENABLE, 0);
5061 			WREG32_SOC15(GC, 0, mmCPC_INT_CNTL, tmp);
5062 
5063 			tmp = RREG32(target);
5064 			tmp = REG_SET_FIELD(tmp, CP_ME2_PIPE0_INT_CNTL,
5065 					    GENERIC2_INT_ENABLE, 0);
5066 			WREG32(target, tmp);
5067 		} else {
5068 			tmp = RREG32_SOC15(GC, 0, mmCPC_INT_CNTL);
5069 			tmp = REG_SET_FIELD(tmp, CPC_INT_CNTL,
5070 					    GENERIC2_INT_ENABLE, 1);
5071 			WREG32_SOC15(GC, 0, mmCPC_INT_CNTL, tmp);
5072 
5073 			tmp = RREG32(target);
5074 			tmp = REG_SET_FIELD(tmp, CP_ME2_PIPE0_INT_CNTL,
5075 					    GENERIC2_INT_ENABLE, 1);
5076 			WREG32(target, tmp);
5077 		}
5078 		break;
5079 	default:
5080 		BUG(); /* kiq only support GENERIC2_INT now */
5081 		break;
5082 	}
5083 	return 0;
5084 }
5085 
5086 static int gfx_v10_0_kiq_irq(struct amdgpu_device *adev,
5087 			     struct amdgpu_irq_src *source,
5088 			     struct amdgpu_iv_entry *entry)
5089 {
5090 	u8 me_id, pipe_id, queue_id;
5091 	struct amdgpu_ring *ring = &(adev->gfx.kiq.ring);
5092 
5093 	me_id = (entry->ring_id & 0x0c) >> 2;
5094 	pipe_id = (entry->ring_id & 0x03) >> 0;
5095 	queue_id = (entry->ring_id & 0x70) >> 4;
5096 	DRM_DEBUG("IH: CPC GENERIC2_INT, me:%d, pipe:%d, queue:%d\n",
5097 		   me_id, pipe_id, queue_id);
5098 
5099 	amdgpu_fence_process(ring);
5100 	return 0;
5101 }
5102 
5103 static const struct amd_ip_funcs gfx_v10_0_ip_funcs = {
5104 	.name = "gfx_v10_0",
5105 	.early_init = gfx_v10_0_early_init,
5106 	.late_init = gfx_v10_0_late_init,
5107 	.sw_init = gfx_v10_0_sw_init,
5108 	.sw_fini = gfx_v10_0_sw_fini,
5109 	.hw_init = gfx_v10_0_hw_init,
5110 	.hw_fini = gfx_v10_0_hw_fini,
5111 	.suspend = gfx_v10_0_suspend,
5112 	.resume = gfx_v10_0_resume,
5113 	.is_idle = gfx_v10_0_is_idle,
5114 	.wait_for_idle = gfx_v10_0_wait_for_idle,
5115 	.soft_reset = gfx_v10_0_soft_reset,
5116 	.set_clockgating_state = gfx_v10_0_set_clockgating_state,
5117 	.set_powergating_state = gfx_v10_0_set_powergating_state,
5118 	.get_clockgating_state = gfx_v10_0_get_clockgating_state,
5119 };
5120 
5121 static const struct amdgpu_ring_funcs gfx_v10_0_ring_funcs_gfx = {
5122 	.type = AMDGPU_RING_TYPE_GFX,
5123 	.align_mask = 0xff,
5124 	.nop = PACKET3(PACKET3_NOP, 0x3FFF),
5125 	.support_64bit_ptrs = true,
5126 	.vmhub = AMDGPU_GFXHUB_0,
5127 	.get_rptr = gfx_v10_0_ring_get_rptr_gfx,
5128 	.get_wptr = gfx_v10_0_ring_get_wptr_gfx,
5129 	.set_wptr = gfx_v10_0_ring_set_wptr_gfx,
5130 	.emit_frame_size = /* totally 242 maximum if 16 IBs */
5131 		5 + /* COND_EXEC */
5132 		7 + /* PIPELINE_SYNC */
5133 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 5 +
5134 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 7 +
5135 		2 + /* VM_FLUSH */
5136 		8 + /* FENCE for VM_FLUSH */
5137 		20 + /* GDS switch */
5138 		4 + /* double SWITCH_BUFFER,
5139 		     * the first COND_EXEC jump to the place
5140 		     * just prior to this double SWITCH_BUFFER
5141 		     */
5142 		5 + /* COND_EXEC */
5143 		7 + /* HDP_flush */
5144 		4 + /* VGT_flush */
5145 		14 + /*	CE_META */
5146 		31 + /*	DE_META */
5147 		3 + /* CNTX_CTRL */
5148 		5 + /* HDP_INVL */
5149 		8 + 8 + /* FENCE x2 */
5150 		2, /* SWITCH_BUFFER */
5151 	.emit_ib_size =	4, /* gfx_v10_0_ring_emit_ib_gfx */
5152 	.emit_ib = gfx_v10_0_ring_emit_ib_gfx,
5153 	.emit_fence = gfx_v10_0_ring_emit_fence,
5154 	.emit_pipeline_sync = gfx_v10_0_ring_emit_pipeline_sync,
5155 	.emit_vm_flush = gfx_v10_0_ring_emit_vm_flush,
5156 	.emit_gds_switch = gfx_v10_0_ring_emit_gds_switch,
5157 	.emit_hdp_flush = gfx_v10_0_ring_emit_hdp_flush,
5158 	.test_ring = gfx_v10_0_ring_test_ring,
5159 	.test_ib = gfx_v10_0_ring_test_ib,
5160 	.insert_nop = amdgpu_ring_insert_nop,
5161 	.pad_ib = amdgpu_ring_generic_pad_ib,
5162 	.emit_switch_buffer = gfx_v10_0_ring_emit_sb,
5163 	.emit_cntxcntl = gfx_v10_0_ring_emit_cntxcntl,
5164 	.init_cond_exec = gfx_v10_0_ring_emit_init_cond_exec,
5165 	.patch_cond_exec = gfx_v10_0_ring_emit_patch_cond_exec,
5166 	.preempt_ib = gfx_v10_0_ring_preempt_ib,
5167 	.emit_tmz = gfx_v10_0_ring_emit_tmz,
5168 	.emit_wreg = gfx_v10_0_ring_emit_wreg,
5169 	.emit_reg_wait = gfx_v10_0_ring_emit_reg_wait,
5170 	.emit_reg_write_reg_wait = gfx_v10_0_ring_emit_reg_write_reg_wait,
5171 };
5172 
5173 static const struct amdgpu_ring_funcs gfx_v10_0_ring_funcs_compute = {
5174 	.type = AMDGPU_RING_TYPE_COMPUTE,
5175 	.align_mask = 0xff,
5176 	.nop = PACKET3(PACKET3_NOP, 0x3FFF),
5177 	.support_64bit_ptrs = true,
5178 	.vmhub = AMDGPU_GFXHUB_0,
5179 	.get_rptr = gfx_v10_0_ring_get_rptr_compute,
5180 	.get_wptr = gfx_v10_0_ring_get_wptr_compute,
5181 	.set_wptr = gfx_v10_0_ring_set_wptr_compute,
5182 	.emit_frame_size =
5183 		20 + /* gfx_v10_0_ring_emit_gds_switch */
5184 		7 + /* gfx_v10_0_ring_emit_hdp_flush */
5185 		5 + /* hdp invalidate */
5186 		7 + /* gfx_v10_0_ring_emit_pipeline_sync */
5187 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 5 +
5188 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 7 +
5189 		2 + /* gfx_v10_0_ring_emit_vm_flush */
5190 		8 + 8 + 8, /* gfx_v10_0_ring_emit_fence x3 for user fence, vm fence */
5191 	.emit_ib_size =	7, /* gfx_v10_0_ring_emit_ib_compute */
5192 	.emit_ib = gfx_v10_0_ring_emit_ib_compute,
5193 	.emit_fence = gfx_v10_0_ring_emit_fence,
5194 	.emit_pipeline_sync = gfx_v10_0_ring_emit_pipeline_sync,
5195 	.emit_vm_flush = gfx_v10_0_ring_emit_vm_flush,
5196 	.emit_gds_switch = gfx_v10_0_ring_emit_gds_switch,
5197 	.emit_hdp_flush = gfx_v10_0_ring_emit_hdp_flush,
5198 	.test_ring = gfx_v10_0_ring_test_ring,
5199 	.test_ib = gfx_v10_0_ring_test_ib,
5200 	.insert_nop = amdgpu_ring_insert_nop,
5201 	.pad_ib = amdgpu_ring_generic_pad_ib,
5202 	.emit_wreg = gfx_v10_0_ring_emit_wreg,
5203 	.emit_reg_wait = gfx_v10_0_ring_emit_reg_wait,
5204 	.emit_reg_write_reg_wait = gfx_v10_0_ring_emit_reg_write_reg_wait,
5205 };
5206 
5207 static const struct amdgpu_ring_funcs gfx_v10_0_ring_funcs_kiq = {
5208 	.type = AMDGPU_RING_TYPE_KIQ,
5209 	.align_mask = 0xff,
5210 	.nop = PACKET3(PACKET3_NOP, 0x3FFF),
5211 	.support_64bit_ptrs = true,
5212 	.vmhub = AMDGPU_GFXHUB_0,
5213 	.get_rptr = gfx_v10_0_ring_get_rptr_compute,
5214 	.get_wptr = gfx_v10_0_ring_get_wptr_compute,
5215 	.set_wptr = gfx_v10_0_ring_set_wptr_compute,
5216 	.emit_frame_size =
5217 		20 + /* gfx_v10_0_ring_emit_gds_switch */
5218 		7 + /* gfx_v10_0_ring_emit_hdp_flush */
5219 		5 + /*hdp invalidate */
5220 		7 + /* gfx_v10_0_ring_emit_pipeline_sync */
5221 		SOC15_FLUSH_GPU_TLB_NUM_WREG * 5 +
5222 		SOC15_FLUSH_GPU_TLB_NUM_REG_WAIT * 7 +
5223 		2 + /* gfx_v10_0_ring_emit_vm_flush */
5224 		8 + 8 + 8, /* gfx_v10_0_ring_emit_fence_kiq x3 for user fence, vm fence */
5225 	.emit_ib_size =	7, /* gfx_v10_0_ring_emit_ib_compute */
5226 	.emit_ib = gfx_v10_0_ring_emit_ib_compute,
5227 	.emit_fence = gfx_v10_0_ring_emit_fence_kiq,
5228 	.test_ring = gfx_v10_0_ring_test_ring,
5229 	.test_ib = gfx_v10_0_ring_test_ib,
5230 	.insert_nop = amdgpu_ring_insert_nop,
5231 	.pad_ib = amdgpu_ring_generic_pad_ib,
5232 	.emit_rreg = gfx_v10_0_ring_emit_rreg,
5233 	.emit_wreg = gfx_v10_0_ring_emit_wreg,
5234 	.emit_reg_wait = gfx_v10_0_ring_emit_reg_wait,
5235 	.emit_reg_write_reg_wait = gfx_v10_0_ring_emit_reg_write_reg_wait,
5236 };
5237 
5238 static void gfx_v10_0_set_ring_funcs(struct amdgpu_device *adev)
5239 {
5240 	int i;
5241 
5242 	adev->gfx.kiq.ring.funcs = &gfx_v10_0_ring_funcs_kiq;
5243 
5244 	for (i = 0; i < adev->gfx.num_gfx_rings; i++)
5245 		adev->gfx.gfx_ring[i].funcs = &gfx_v10_0_ring_funcs_gfx;
5246 
5247 	for (i = 0; i < adev->gfx.num_compute_rings; i++)
5248 		adev->gfx.compute_ring[i].funcs = &gfx_v10_0_ring_funcs_compute;
5249 }
5250 
5251 static const struct amdgpu_irq_src_funcs gfx_v10_0_eop_irq_funcs = {
5252 	.set = gfx_v10_0_set_eop_interrupt_state,
5253 	.process = gfx_v10_0_eop_irq,
5254 };
5255 
5256 static const struct amdgpu_irq_src_funcs gfx_v10_0_priv_reg_irq_funcs = {
5257 	.set = gfx_v10_0_set_priv_reg_fault_state,
5258 	.process = gfx_v10_0_priv_reg_irq,
5259 };
5260 
5261 static const struct amdgpu_irq_src_funcs gfx_v10_0_priv_inst_irq_funcs = {
5262 	.set = gfx_v10_0_set_priv_inst_fault_state,
5263 	.process = gfx_v10_0_priv_inst_irq,
5264 };
5265 
5266 static const struct amdgpu_irq_src_funcs gfx_v10_0_kiq_irq_funcs = {
5267 	.set = gfx_v10_0_kiq_set_interrupt_state,
5268 	.process = gfx_v10_0_kiq_irq,
5269 };
5270 
5271 static void gfx_v10_0_set_irq_funcs(struct amdgpu_device *adev)
5272 {
5273 	adev->gfx.eop_irq.num_types = AMDGPU_CP_IRQ_LAST;
5274 	adev->gfx.eop_irq.funcs = &gfx_v10_0_eop_irq_funcs;
5275 
5276 	adev->gfx.kiq.irq.num_types = AMDGPU_CP_KIQ_IRQ_LAST;
5277 	adev->gfx.kiq.irq.funcs = &gfx_v10_0_kiq_irq_funcs;
5278 
5279 	adev->gfx.priv_reg_irq.num_types = 1;
5280 	adev->gfx.priv_reg_irq.funcs = &gfx_v10_0_priv_reg_irq_funcs;
5281 
5282 	adev->gfx.priv_inst_irq.num_types = 1;
5283 	adev->gfx.priv_inst_irq.funcs = &gfx_v10_0_priv_inst_irq_funcs;
5284 }
5285 
5286 static void gfx_v10_0_set_rlc_funcs(struct amdgpu_device *adev)
5287 {
5288 	switch (adev->asic_type) {
5289 	case CHIP_NAVI10:
5290 	case CHIP_NAVI14:
5291 	case CHIP_NAVI12:
5292 		adev->gfx.rlc.funcs = &gfx_v10_0_rlc_funcs;
5293 		break;
5294 	default:
5295 		break;
5296 	}
5297 }
5298 
5299 static void gfx_v10_0_set_gds_init(struct amdgpu_device *adev)
5300 {
5301 	unsigned total_cu = adev->gfx.config.max_cu_per_sh *
5302 			    adev->gfx.config.max_sh_per_se *
5303 			    adev->gfx.config.max_shader_engines;
5304 
5305 	adev->gds.gds_size = 0x10000;
5306 	adev->gds.gds_compute_max_wave_id = total_cu * 32 - 1;
5307 	adev->gds.gws_size = 64;
5308 	adev->gds.oa_size = 16;
5309 }
5310 
5311 static void gfx_v10_0_set_user_wgp_inactive_bitmap_per_sh(struct amdgpu_device *adev,
5312 							  u32 bitmap)
5313 {
5314 	u32 data;
5315 
5316 	if (!bitmap)
5317 		return;
5318 
5319 	data = bitmap << GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_WGPS__SHIFT;
5320 	data &= GC_USER_SHADER_ARRAY_CONFIG__INACTIVE_WGPS_MASK;
5321 
5322 	WREG32_SOC15(GC, 0, mmGC_USER_SHADER_ARRAY_CONFIG, data);
5323 }
5324 
5325 static u32 gfx_v10_0_get_wgp_active_bitmap_per_sh(struct amdgpu_device *adev)
5326 {
5327 	u32 data, wgp_bitmask;
5328 	data = RREG32_SOC15(GC, 0, mmCC_GC_SHADER_ARRAY_CONFIG);
5329 	data |= RREG32_SOC15(GC, 0, mmGC_USER_SHADER_ARRAY_CONFIG);
5330 
5331 	data &= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_WGPS_MASK;
5332 	data >>= CC_GC_SHADER_ARRAY_CONFIG__INACTIVE_WGPS__SHIFT;
5333 
5334 	wgp_bitmask =
5335 		amdgpu_gfx_create_bitmask(adev->gfx.config.max_cu_per_sh >> 1);
5336 
5337 	return (~data) & wgp_bitmask;
5338 }
5339 
5340 static u32 gfx_v10_0_get_cu_active_bitmap_per_sh(struct amdgpu_device *adev)
5341 {
5342 	u32 wgp_idx, wgp_active_bitmap;
5343 	u32 cu_bitmap_per_wgp, cu_active_bitmap;
5344 
5345 	wgp_active_bitmap = gfx_v10_0_get_wgp_active_bitmap_per_sh(adev);
5346 	cu_active_bitmap = 0;
5347 
5348 	for (wgp_idx = 0; wgp_idx < 16; wgp_idx++) {
5349 		/* if there is one WGP enabled, it means 2 CUs will be enabled */
5350 		cu_bitmap_per_wgp = 3 << (2 * wgp_idx);
5351 		if (wgp_active_bitmap & (1 << wgp_idx))
5352 			cu_active_bitmap |= cu_bitmap_per_wgp;
5353 	}
5354 
5355 	return cu_active_bitmap;
5356 }
5357 
5358 static int gfx_v10_0_get_cu_info(struct amdgpu_device *adev,
5359 				 struct amdgpu_cu_info *cu_info)
5360 {
5361 	int i, j, k, counter, active_cu_number = 0;
5362 	u32 mask, bitmap, ao_bitmap, ao_cu_mask = 0;
5363 	unsigned disable_masks[4 * 2];
5364 
5365 	if (!adev || !cu_info)
5366 		return -EINVAL;
5367 
5368 	amdgpu_gfx_parse_disable_cu(disable_masks, 4, 2);
5369 
5370 	mutex_lock(&adev->grbm_idx_mutex);
5371 	for (i = 0; i < adev->gfx.config.max_shader_engines; i++) {
5372 		for (j = 0; j < adev->gfx.config.max_sh_per_se; j++) {
5373 			mask = 1;
5374 			ao_bitmap = 0;
5375 			counter = 0;
5376 			gfx_v10_0_select_se_sh(adev, i, j, 0xffffffff);
5377 			if (i < 4 && j < 2)
5378 				gfx_v10_0_set_user_wgp_inactive_bitmap_per_sh(
5379 					adev, disable_masks[i * 2 + j]);
5380 			bitmap = gfx_v10_0_get_cu_active_bitmap_per_sh(adev);
5381 			cu_info->bitmap[i][j] = bitmap;
5382 
5383 			for (k = 0; k < adev->gfx.config.max_cu_per_sh; k++) {
5384 				if (bitmap & mask) {
5385 					if (counter < adev->gfx.config.max_cu_per_sh)
5386 						ao_bitmap |= mask;
5387 					counter++;
5388 				}
5389 				mask <<= 1;
5390 			}
5391 			active_cu_number += counter;
5392 			if (i < 2 && j < 2)
5393 				ao_cu_mask |= (ao_bitmap << (i * 16 + j * 8));
5394 			cu_info->ao_cu_bitmap[i][j] = ao_bitmap;
5395 		}
5396 	}
5397 	gfx_v10_0_select_se_sh(adev, 0xffffffff, 0xffffffff, 0xffffffff);
5398 	mutex_unlock(&adev->grbm_idx_mutex);
5399 
5400 	cu_info->number = active_cu_number;
5401 	cu_info->ao_cu_mask = ao_cu_mask;
5402 	cu_info->simd_per_cu = NUM_SIMD_PER_CU;
5403 
5404 	return 0;
5405 }
5406 
5407 const struct amdgpu_ip_block_version gfx_v10_0_ip_block =
5408 {
5409 	.type = AMD_IP_BLOCK_TYPE_GFX,
5410 	.major = 10,
5411 	.minor = 0,
5412 	.rev = 0,
5413 	.funcs = &gfx_v10_0_ip_funcs,
5414 };
5415