2 * Copyright 2018 Advanced Micro Devices, Inc.
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:
11 * The above copyright notice and this permission notice shall be included in
12 * all copies or substantial portions of the Software.
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.
25 * GPU doorbell structures, functions & helpers
27 struct amdgpu_doorbell {
32 u32 num_doorbells; /* Number of doorbells actually reserved for amdgpu. */
35 /* Reserved doorbells for amdgpu (including multimedia).
36 * KFD can use all the rest in the 2M doorbell bar.
37 * For asic before vega10, doorbell is 32-bit, so the
38 * index/offset is in dword. For vega10 and after, doorbell
39 * can be 64-bit, so the index defined is in qword.
41 struct amdgpu_doorbell_index {
51 uint32_t userqueue_start;
52 uint32_t userqueue_end;
54 uint32_t sdma_engine0;
55 uint32_t sdma_engine1;
56 uint32_t sdma_engine2;
57 uint32_t sdma_engine3;
58 uint32_t sdma_engine4;
59 uint32_t sdma_engine5;
60 uint32_t sdma_engine6;
61 uint32_t sdma_engine7;
81 uint32_t max_assignment;
84 typedef enum _AMDGPU_DOORBELL_ASSIGNMENT
86 AMDGPU_DOORBELL_KIQ = 0x000,
87 AMDGPU_DOORBELL_HIQ = 0x001,
88 AMDGPU_DOORBELL_DIQ = 0x002,
89 AMDGPU_DOORBELL_MEC_RING0 = 0x010,
90 AMDGPU_DOORBELL_MEC_RING1 = 0x011,
91 AMDGPU_DOORBELL_MEC_RING2 = 0x012,
92 AMDGPU_DOORBELL_MEC_RING3 = 0x013,
93 AMDGPU_DOORBELL_MEC_RING4 = 0x014,
94 AMDGPU_DOORBELL_MEC_RING5 = 0x015,
95 AMDGPU_DOORBELL_MEC_RING6 = 0x016,
96 AMDGPU_DOORBELL_MEC_RING7 = 0x017,
97 AMDGPU_DOORBELL_GFX_RING0 = 0x020,
98 AMDGPU_DOORBELL_sDMA_ENGINE0 = 0x1E0,
99 AMDGPU_DOORBELL_sDMA_ENGINE1 = 0x1E1,
100 AMDGPU_DOORBELL_IH = 0x1E8,
101 AMDGPU_DOORBELL_MAX_ASSIGNMENT = 0x3FF,
102 AMDGPU_DOORBELL_INVALID = 0xFFFF
103 } AMDGPU_DOORBELL_ASSIGNMENT;
105 typedef enum _AMDGPU_VEGA20_DOORBELL_ASSIGNMENT
107 /* Compute + GFX: 0~255 */
108 AMDGPU_VEGA20_DOORBELL_KIQ = 0x000,
109 AMDGPU_VEGA20_DOORBELL_HIQ = 0x001,
110 AMDGPU_VEGA20_DOORBELL_DIQ = 0x002,
111 AMDGPU_VEGA20_DOORBELL_MEC_RING0 = 0x003,
112 AMDGPU_VEGA20_DOORBELL_MEC_RING1 = 0x004,
113 AMDGPU_VEGA20_DOORBELL_MEC_RING2 = 0x005,
114 AMDGPU_VEGA20_DOORBELL_MEC_RING3 = 0x006,
115 AMDGPU_VEGA20_DOORBELL_MEC_RING4 = 0x007,
116 AMDGPU_VEGA20_DOORBELL_MEC_RING5 = 0x008,
117 AMDGPU_VEGA20_DOORBELL_MEC_RING6 = 0x009,
118 AMDGPU_VEGA20_DOORBELL_MEC_RING7 = 0x00A,
119 AMDGPU_VEGA20_DOORBELL_USERQUEUE_START = 0x00B,
120 AMDGPU_VEGA20_DOORBELL_USERQUEUE_END = 0x08A,
121 AMDGPU_VEGA20_DOORBELL_GFX_RING0 = 0x08B,
123 AMDGPU_VEGA20_DOORBELL_sDMA_ENGINE0 = 0x100,
124 AMDGPU_VEGA20_DOORBELL_sDMA_ENGINE1 = 0x10A,
125 AMDGPU_VEGA20_DOORBELL_sDMA_ENGINE2 = 0x114,
126 AMDGPU_VEGA20_DOORBELL_sDMA_ENGINE3 = 0x11E,
127 AMDGPU_VEGA20_DOORBELL_sDMA_ENGINE4 = 0x128,
128 AMDGPU_VEGA20_DOORBELL_sDMA_ENGINE5 = 0x132,
129 AMDGPU_VEGA20_DOORBELL_sDMA_ENGINE6 = 0x13C,
130 AMDGPU_VEGA20_DOORBELL_sDMA_ENGINE7 = 0x146,
132 AMDGPU_VEGA20_DOORBELL_IH = 0x178,
134 * overlap the doorbell assignment with VCN as they are mutually exclusive
135 * VCE engine's doorbell is 32 bit and two VCE ring share one QWORD
137 AMDGPU_VEGA20_DOORBELL64_VCN0_1 = 0x188, /* lower 32 bits for VNC0 and upper 32 bits for VNC1 */
138 AMDGPU_VEGA20_DOORBELL64_VCN2_3 = 0x189,
139 AMDGPU_VEGA20_DOORBELL64_VCN4_5 = 0x18A,
140 AMDGPU_VEGA20_DOORBELL64_VCN6_7 = 0x18B,
142 AMDGPU_VEGA20_DOORBELL64_UVD_RING0_1 = 0x188,
143 AMDGPU_VEGA20_DOORBELL64_UVD_RING2_3 = 0x189,
144 AMDGPU_VEGA20_DOORBELL64_UVD_RING4_5 = 0x18A,
145 AMDGPU_VEGA20_DOORBELL64_UVD_RING6_7 = 0x18B,
147 AMDGPU_VEGA20_DOORBELL64_VCE_RING0_1 = 0x18C,
148 AMDGPU_VEGA20_DOORBELL64_VCE_RING2_3 = 0x18D,
149 AMDGPU_VEGA20_DOORBELL64_VCE_RING4_5 = 0x18E,
150 AMDGPU_VEGA20_DOORBELL64_VCE_RING6_7 = 0x18F,
151 AMDGPU_VEGA20_DOORBELL_MAX_ASSIGNMENT = 0x18F,
152 AMDGPU_VEGA20_DOORBELL_INVALID = 0xFFFF
153 } AMDGPU_VEGA20_DOORBELL_ASSIGNMENT;
156 * 64bit doorbell, offset are in QWORD, occupy 2KB doorbell space
158 typedef enum _AMDGPU_DOORBELL64_ASSIGNMENT
161 * All compute related doorbells: kiq, hiq, diq, traditional compute queue, user queue, should locate in
162 * a continues range so that programming CP_MEC_DOORBELL_RANGE_LOWER/UPPER can cover this range.
163 * Compute related doorbells are allocated from 0x00 to 0x8a
167 /* kernel scheduling */
168 AMDGPU_DOORBELL64_KIQ = 0x00,
170 /* HSA interface queue and debug queue */
171 AMDGPU_DOORBELL64_HIQ = 0x01,
172 AMDGPU_DOORBELL64_DIQ = 0x02,
174 /* Compute engines */
175 AMDGPU_DOORBELL64_MEC_RING0 = 0x03,
176 AMDGPU_DOORBELL64_MEC_RING1 = 0x04,
177 AMDGPU_DOORBELL64_MEC_RING2 = 0x05,
178 AMDGPU_DOORBELL64_MEC_RING3 = 0x06,
179 AMDGPU_DOORBELL64_MEC_RING4 = 0x07,
180 AMDGPU_DOORBELL64_MEC_RING5 = 0x08,
181 AMDGPU_DOORBELL64_MEC_RING6 = 0x09,
182 AMDGPU_DOORBELL64_MEC_RING7 = 0x0a,
184 /* User queue doorbell range (128 doorbells) */
185 AMDGPU_DOORBELL64_USERQUEUE_START = 0x0b,
186 AMDGPU_DOORBELL64_USERQUEUE_END = 0x8a,
188 /* Graphics engine */
189 AMDGPU_DOORBELL64_GFX_RING0 = 0x8b,
192 * Other graphics doorbells can be allocated here: from 0x8c to 0xdf
193 * Graphics voltage island aperture 1
194 * default non-graphics QWORD index is 0xe0 - 0xFF inclusive
197 /* For vega10 sriov, the sdma doorbell must be fixed as follow
198 * to keep the same setting with host driver, or it will
201 AMDGPU_DOORBELL64_sDMA_ENGINE0 = 0xF0,
202 AMDGPU_DOORBELL64_sDMA_HI_PRI_ENGINE0 = 0xF1,
203 AMDGPU_DOORBELL64_sDMA_ENGINE1 = 0xF2,
204 AMDGPU_DOORBELL64_sDMA_HI_PRI_ENGINE1 = 0xF3,
206 /* Interrupt handler */
207 AMDGPU_DOORBELL64_IH = 0xF4, /* For legacy interrupt ring buffer */
208 AMDGPU_DOORBELL64_IH_RING1 = 0xF5, /* For page migration request log */
209 AMDGPU_DOORBELL64_IH_RING2 = 0xF6, /* For page migration translation/invalidation log */
211 /* VCN engine use 32 bits doorbell */
212 AMDGPU_DOORBELL64_VCN0_1 = 0xF8, /* lower 32 bits for VNC0 and upper 32 bits for VNC1 */
213 AMDGPU_DOORBELL64_VCN2_3 = 0xF9,
214 AMDGPU_DOORBELL64_VCN4_5 = 0xFA,
215 AMDGPU_DOORBELL64_VCN6_7 = 0xFB,
217 /* overlap the doorbell assignment with VCN as they are mutually exclusive
218 * VCE engine's doorbell is 32 bit and two VCE ring share one QWORD
220 AMDGPU_DOORBELL64_UVD_RING0_1 = 0xF8,
221 AMDGPU_DOORBELL64_UVD_RING2_3 = 0xF9,
222 AMDGPU_DOORBELL64_UVD_RING4_5 = 0xFA,
223 AMDGPU_DOORBELL64_UVD_RING6_7 = 0xFB,
225 AMDGPU_DOORBELL64_VCE_RING0_1 = 0xFC,
226 AMDGPU_DOORBELL64_VCE_RING2_3 = 0xFD,
227 AMDGPU_DOORBELL64_VCE_RING4_5 = 0xFE,
228 AMDGPU_DOORBELL64_VCE_RING6_7 = 0xFF,
230 AMDGPU_DOORBELL64_MAX_ASSIGNMENT = 0xFF,
231 AMDGPU_DOORBELL64_INVALID = 0xFFFF
232 } AMDGPU_DOORBELL64_ASSIGNMENT;
234 u32 amdgpu_mm_rdoorbell(struct amdgpu_device *adev, u32 index);
235 void amdgpu_mm_wdoorbell(struct amdgpu_device *adev, u32 index, u32 v);
236 u64 amdgpu_mm_rdoorbell64(struct amdgpu_device *adev, u32 index);
237 void amdgpu_mm_wdoorbell64(struct amdgpu_device *adev, u32 index, u64 v);
239 #define RDOORBELL32(index) amdgpu_mm_rdoorbell(adev, (index))
240 #define WDOORBELL32(index, v) amdgpu_mm_wdoorbell(adev, (index), (v))
241 #define RDOORBELL64(index) amdgpu_mm_rdoorbell64(adev, (index))
242 #define WDOORBELL64(index, v) amdgpu_mm_wdoorbell64(adev, (index), (v))