1 /* SPDX-License-Identifier: GPL-2.0 WITH Linux-syscall-note
3 * Copyright 2016-2022 HabanaLabs, Ltd.
11 #include <linux/types.h>
12 #include <linux/ioctl.h>
15 * Defines that are asic-specific but constitutes as ABI between kernel driver
18 #define GOYA_KMD_SRAM_RESERVED_SIZE_FROM_START 0x8000 /* 32KB */
19 #define GAUDI_DRIVER_SRAM_RESERVED_SIZE_FROM_START 0x80 /* 128 bytes */
22 * 128 SOBs reserved for collective wait
23 * 16 SOBs reserved for sync stream
25 #define GAUDI_FIRST_AVAILABLE_W_S_SYNC_OBJECT 144
28 * 64 monitors reserved for collective wait
29 * 8 monitors reserved for sync stream
31 #define GAUDI_FIRST_AVAILABLE_W_S_MONITOR 72
33 /* Max number of elements in timestamps registration buffers */
34 #define TS_MAX_ELEMENTS_NUM (1 << 20) /* 1MB */
37 * Goya queue Numbering
39 * The external queues (PCI DMA channels) MUST be before the internal queues
40 * and each group (PCI DMA channels and internal) must be contiguous inside
41 * itself but there can be a gap between the two groups (although not
46 GOYA_QUEUE_ID_DMA_0 = 0,
47 GOYA_QUEUE_ID_DMA_1 = 1,
48 GOYA_QUEUE_ID_DMA_2 = 2,
49 GOYA_QUEUE_ID_DMA_3 = 3,
50 GOYA_QUEUE_ID_DMA_4 = 4,
51 GOYA_QUEUE_ID_CPU_PQ = 5,
52 GOYA_QUEUE_ID_MME = 6, /* Internal queues start here */
53 GOYA_QUEUE_ID_TPC0 = 7,
54 GOYA_QUEUE_ID_TPC1 = 8,
55 GOYA_QUEUE_ID_TPC2 = 9,
56 GOYA_QUEUE_ID_TPC3 = 10,
57 GOYA_QUEUE_ID_TPC4 = 11,
58 GOYA_QUEUE_ID_TPC5 = 12,
59 GOYA_QUEUE_ID_TPC6 = 13,
60 GOYA_QUEUE_ID_TPC7 = 14,
65 * Gaudi queue Numbering
66 * External queues (PCI DMA channels) are DMA_0_*, DMA_1_* and DMA_5_*.
67 * Except one CPU queue, all the rest are internal queues.
71 GAUDI_QUEUE_ID_DMA_0_0 = 0, /* external */
72 GAUDI_QUEUE_ID_DMA_0_1 = 1, /* external */
73 GAUDI_QUEUE_ID_DMA_0_2 = 2, /* external */
74 GAUDI_QUEUE_ID_DMA_0_3 = 3, /* external */
75 GAUDI_QUEUE_ID_DMA_1_0 = 4, /* external */
76 GAUDI_QUEUE_ID_DMA_1_1 = 5, /* external */
77 GAUDI_QUEUE_ID_DMA_1_2 = 6, /* external */
78 GAUDI_QUEUE_ID_DMA_1_3 = 7, /* external */
79 GAUDI_QUEUE_ID_CPU_PQ = 8, /* CPU */
80 GAUDI_QUEUE_ID_DMA_2_0 = 9, /* internal */
81 GAUDI_QUEUE_ID_DMA_2_1 = 10, /* internal */
82 GAUDI_QUEUE_ID_DMA_2_2 = 11, /* internal */
83 GAUDI_QUEUE_ID_DMA_2_3 = 12, /* internal */
84 GAUDI_QUEUE_ID_DMA_3_0 = 13, /* internal */
85 GAUDI_QUEUE_ID_DMA_3_1 = 14, /* internal */
86 GAUDI_QUEUE_ID_DMA_3_2 = 15, /* internal */
87 GAUDI_QUEUE_ID_DMA_3_3 = 16, /* internal */
88 GAUDI_QUEUE_ID_DMA_4_0 = 17, /* internal */
89 GAUDI_QUEUE_ID_DMA_4_1 = 18, /* internal */
90 GAUDI_QUEUE_ID_DMA_4_2 = 19, /* internal */
91 GAUDI_QUEUE_ID_DMA_4_3 = 20, /* internal */
92 GAUDI_QUEUE_ID_DMA_5_0 = 21, /* internal */
93 GAUDI_QUEUE_ID_DMA_5_1 = 22, /* internal */
94 GAUDI_QUEUE_ID_DMA_5_2 = 23, /* internal */
95 GAUDI_QUEUE_ID_DMA_5_3 = 24, /* internal */
96 GAUDI_QUEUE_ID_DMA_6_0 = 25, /* internal */
97 GAUDI_QUEUE_ID_DMA_6_1 = 26, /* internal */
98 GAUDI_QUEUE_ID_DMA_6_2 = 27, /* internal */
99 GAUDI_QUEUE_ID_DMA_6_3 = 28, /* internal */
100 GAUDI_QUEUE_ID_DMA_7_0 = 29, /* internal */
101 GAUDI_QUEUE_ID_DMA_7_1 = 30, /* internal */
102 GAUDI_QUEUE_ID_DMA_7_2 = 31, /* internal */
103 GAUDI_QUEUE_ID_DMA_7_3 = 32, /* internal */
104 GAUDI_QUEUE_ID_MME_0_0 = 33, /* internal */
105 GAUDI_QUEUE_ID_MME_0_1 = 34, /* internal */
106 GAUDI_QUEUE_ID_MME_0_2 = 35, /* internal */
107 GAUDI_QUEUE_ID_MME_0_3 = 36, /* internal */
108 GAUDI_QUEUE_ID_MME_1_0 = 37, /* internal */
109 GAUDI_QUEUE_ID_MME_1_1 = 38, /* internal */
110 GAUDI_QUEUE_ID_MME_1_2 = 39, /* internal */
111 GAUDI_QUEUE_ID_MME_1_3 = 40, /* internal */
112 GAUDI_QUEUE_ID_TPC_0_0 = 41, /* internal */
113 GAUDI_QUEUE_ID_TPC_0_1 = 42, /* internal */
114 GAUDI_QUEUE_ID_TPC_0_2 = 43, /* internal */
115 GAUDI_QUEUE_ID_TPC_0_3 = 44, /* internal */
116 GAUDI_QUEUE_ID_TPC_1_0 = 45, /* internal */
117 GAUDI_QUEUE_ID_TPC_1_1 = 46, /* internal */
118 GAUDI_QUEUE_ID_TPC_1_2 = 47, /* internal */
119 GAUDI_QUEUE_ID_TPC_1_3 = 48, /* internal */
120 GAUDI_QUEUE_ID_TPC_2_0 = 49, /* internal */
121 GAUDI_QUEUE_ID_TPC_2_1 = 50, /* internal */
122 GAUDI_QUEUE_ID_TPC_2_2 = 51, /* internal */
123 GAUDI_QUEUE_ID_TPC_2_3 = 52, /* internal */
124 GAUDI_QUEUE_ID_TPC_3_0 = 53, /* internal */
125 GAUDI_QUEUE_ID_TPC_3_1 = 54, /* internal */
126 GAUDI_QUEUE_ID_TPC_3_2 = 55, /* internal */
127 GAUDI_QUEUE_ID_TPC_3_3 = 56, /* internal */
128 GAUDI_QUEUE_ID_TPC_4_0 = 57, /* internal */
129 GAUDI_QUEUE_ID_TPC_4_1 = 58, /* internal */
130 GAUDI_QUEUE_ID_TPC_4_2 = 59, /* internal */
131 GAUDI_QUEUE_ID_TPC_4_3 = 60, /* internal */
132 GAUDI_QUEUE_ID_TPC_5_0 = 61, /* internal */
133 GAUDI_QUEUE_ID_TPC_5_1 = 62, /* internal */
134 GAUDI_QUEUE_ID_TPC_5_2 = 63, /* internal */
135 GAUDI_QUEUE_ID_TPC_5_3 = 64, /* internal */
136 GAUDI_QUEUE_ID_TPC_6_0 = 65, /* internal */
137 GAUDI_QUEUE_ID_TPC_6_1 = 66, /* internal */
138 GAUDI_QUEUE_ID_TPC_6_2 = 67, /* internal */
139 GAUDI_QUEUE_ID_TPC_6_3 = 68, /* internal */
140 GAUDI_QUEUE_ID_TPC_7_0 = 69, /* internal */
141 GAUDI_QUEUE_ID_TPC_7_1 = 70, /* internal */
142 GAUDI_QUEUE_ID_TPC_7_2 = 71, /* internal */
143 GAUDI_QUEUE_ID_TPC_7_3 = 72, /* internal */
144 GAUDI_QUEUE_ID_NIC_0_0 = 73, /* internal */
145 GAUDI_QUEUE_ID_NIC_0_1 = 74, /* internal */
146 GAUDI_QUEUE_ID_NIC_0_2 = 75, /* internal */
147 GAUDI_QUEUE_ID_NIC_0_3 = 76, /* internal */
148 GAUDI_QUEUE_ID_NIC_1_0 = 77, /* internal */
149 GAUDI_QUEUE_ID_NIC_1_1 = 78, /* internal */
150 GAUDI_QUEUE_ID_NIC_1_2 = 79, /* internal */
151 GAUDI_QUEUE_ID_NIC_1_3 = 80, /* internal */
152 GAUDI_QUEUE_ID_NIC_2_0 = 81, /* internal */
153 GAUDI_QUEUE_ID_NIC_2_1 = 82, /* internal */
154 GAUDI_QUEUE_ID_NIC_2_2 = 83, /* internal */
155 GAUDI_QUEUE_ID_NIC_2_3 = 84, /* internal */
156 GAUDI_QUEUE_ID_NIC_3_0 = 85, /* internal */
157 GAUDI_QUEUE_ID_NIC_3_1 = 86, /* internal */
158 GAUDI_QUEUE_ID_NIC_3_2 = 87, /* internal */
159 GAUDI_QUEUE_ID_NIC_3_3 = 88, /* internal */
160 GAUDI_QUEUE_ID_NIC_4_0 = 89, /* internal */
161 GAUDI_QUEUE_ID_NIC_4_1 = 90, /* internal */
162 GAUDI_QUEUE_ID_NIC_4_2 = 91, /* internal */
163 GAUDI_QUEUE_ID_NIC_4_3 = 92, /* internal */
164 GAUDI_QUEUE_ID_NIC_5_0 = 93, /* internal */
165 GAUDI_QUEUE_ID_NIC_5_1 = 94, /* internal */
166 GAUDI_QUEUE_ID_NIC_5_2 = 95, /* internal */
167 GAUDI_QUEUE_ID_NIC_5_3 = 96, /* internal */
168 GAUDI_QUEUE_ID_NIC_6_0 = 97, /* internal */
169 GAUDI_QUEUE_ID_NIC_6_1 = 98, /* internal */
170 GAUDI_QUEUE_ID_NIC_6_2 = 99, /* internal */
171 GAUDI_QUEUE_ID_NIC_6_3 = 100, /* internal */
172 GAUDI_QUEUE_ID_NIC_7_0 = 101, /* internal */
173 GAUDI_QUEUE_ID_NIC_7_1 = 102, /* internal */
174 GAUDI_QUEUE_ID_NIC_7_2 = 103, /* internal */
175 GAUDI_QUEUE_ID_NIC_7_3 = 104, /* internal */
176 GAUDI_QUEUE_ID_NIC_8_0 = 105, /* internal */
177 GAUDI_QUEUE_ID_NIC_8_1 = 106, /* internal */
178 GAUDI_QUEUE_ID_NIC_8_2 = 107, /* internal */
179 GAUDI_QUEUE_ID_NIC_8_3 = 108, /* internal */
180 GAUDI_QUEUE_ID_NIC_9_0 = 109, /* internal */
181 GAUDI_QUEUE_ID_NIC_9_1 = 110, /* internal */
182 GAUDI_QUEUE_ID_NIC_9_2 = 111, /* internal */
183 GAUDI_QUEUE_ID_NIC_9_3 = 112, /* internal */
188 * In GAUDI2 we have two modes of operation in regard to queues:
189 * 1. Legacy mode, where each QMAN exposes 4 streams to the user
190 * 2. F/W mode, where we use F/W to schedule the JOBS to the different queues.
192 * When in legacy mode, the user sends the queue id per JOB according to
193 * enum gaudi2_queue_id below.
195 * When in F/W mode, the user sends a stream id per Command Submission. The
196 * stream id is a running number from 0 up to (N-1), where N is the number
197 * of streams the F/W exposes and is passed to the user in
198 * struct hl_info_hw_ip_info
201 enum gaudi2_queue_id {
202 GAUDI2_QUEUE_ID_PDMA_0_0 = 0,
203 GAUDI2_QUEUE_ID_PDMA_0_1 = 1,
204 GAUDI2_QUEUE_ID_PDMA_0_2 = 2,
205 GAUDI2_QUEUE_ID_PDMA_0_3 = 3,
206 GAUDI2_QUEUE_ID_PDMA_1_0 = 4,
207 GAUDI2_QUEUE_ID_PDMA_1_1 = 5,
208 GAUDI2_QUEUE_ID_PDMA_1_2 = 6,
209 GAUDI2_QUEUE_ID_PDMA_1_3 = 7,
210 GAUDI2_QUEUE_ID_DCORE0_EDMA_0_0 = 8,
211 GAUDI2_QUEUE_ID_DCORE0_EDMA_0_1 = 9,
212 GAUDI2_QUEUE_ID_DCORE0_EDMA_0_2 = 10,
213 GAUDI2_QUEUE_ID_DCORE0_EDMA_0_3 = 11,
214 GAUDI2_QUEUE_ID_DCORE0_EDMA_1_0 = 12,
215 GAUDI2_QUEUE_ID_DCORE0_EDMA_1_1 = 13,
216 GAUDI2_QUEUE_ID_DCORE0_EDMA_1_2 = 14,
217 GAUDI2_QUEUE_ID_DCORE0_EDMA_1_3 = 15,
218 GAUDI2_QUEUE_ID_DCORE0_MME_0_0 = 16,
219 GAUDI2_QUEUE_ID_DCORE0_MME_0_1 = 17,
220 GAUDI2_QUEUE_ID_DCORE0_MME_0_2 = 18,
221 GAUDI2_QUEUE_ID_DCORE0_MME_0_3 = 19,
222 GAUDI2_QUEUE_ID_DCORE0_TPC_0_0 = 20,
223 GAUDI2_QUEUE_ID_DCORE0_TPC_0_1 = 21,
224 GAUDI2_QUEUE_ID_DCORE0_TPC_0_2 = 22,
225 GAUDI2_QUEUE_ID_DCORE0_TPC_0_3 = 23,
226 GAUDI2_QUEUE_ID_DCORE0_TPC_1_0 = 24,
227 GAUDI2_QUEUE_ID_DCORE0_TPC_1_1 = 25,
228 GAUDI2_QUEUE_ID_DCORE0_TPC_1_2 = 26,
229 GAUDI2_QUEUE_ID_DCORE0_TPC_1_3 = 27,
230 GAUDI2_QUEUE_ID_DCORE0_TPC_2_0 = 28,
231 GAUDI2_QUEUE_ID_DCORE0_TPC_2_1 = 29,
232 GAUDI2_QUEUE_ID_DCORE0_TPC_2_2 = 30,
233 GAUDI2_QUEUE_ID_DCORE0_TPC_2_3 = 31,
234 GAUDI2_QUEUE_ID_DCORE0_TPC_3_0 = 32,
235 GAUDI2_QUEUE_ID_DCORE0_TPC_3_1 = 33,
236 GAUDI2_QUEUE_ID_DCORE0_TPC_3_2 = 34,
237 GAUDI2_QUEUE_ID_DCORE0_TPC_3_3 = 35,
238 GAUDI2_QUEUE_ID_DCORE0_TPC_4_0 = 36,
239 GAUDI2_QUEUE_ID_DCORE0_TPC_4_1 = 37,
240 GAUDI2_QUEUE_ID_DCORE0_TPC_4_2 = 38,
241 GAUDI2_QUEUE_ID_DCORE0_TPC_4_3 = 39,
242 GAUDI2_QUEUE_ID_DCORE0_TPC_5_0 = 40,
243 GAUDI2_QUEUE_ID_DCORE0_TPC_5_1 = 41,
244 GAUDI2_QUEUE_ID_DCORE0_TPC_5_2 = 42,
245 GAUDI2_QUEUE_ID_DCORE0_TPC_5_3 = 43,
246 GAUDI2_QUEUE_ID_DCORE0_TPC_6_0 = 44,
247 GAUDI2_QUEUE_ID_DCORE0_TPC_6_1 = 45,
248 GAUDI2_QUEUE_ID_DCORE0_TPC_6_2 = 46,
249 GAUDI2_QUEUE_ID_DCORE0_TPC_6_3 = 47,
250 GAUDI2_QUEUE_ID_DCORE1_EDMA_0_0 = 48,
251 GAUDI2_QUEUE_ID_DCORE1_EDMA_0_1 = 49,
252 GAUDI2_QUEUE_ID_DCORE1_EDMA_0_2 = 50,
253 GAUDI2_QUEUE_ID_DCORE1_EDMA_0_3 = 51,
254 GAUDI2_QUEUE_ID_DCORE1_EDMA_1_0 = 52,
255 GAUDI2_QUEUE_ID_DCORE1_EDMA_1_1 = 53,
256 GAUDI2_QUEUE_ID_DCORE1_EDMA_1_2 = 54,
257 GAUDI2_QUEUE_ID_DCORE1_EDMA_1_3 = 55,
258 GAUDI2_QUEUE_ID_DCORE1_MME_0_0 = 56,
259 GAUDI2_QUEUE_ID_DCORE1_MME_0_1 = 57,
260 GAUDI2_QUEUE_ID_DCORE1_MME_0_2 = 58,
261 GAUDI2_QUEUE_ID_DCORE1_MME_0_3 = 59,
262 GAUDI2_QUEUE_ID_DCORE1_TPC_0_0 = 60,
263 GAUDI2_QUEUE_ID_DCORE1_TPC_0_1 = 61,
264 GAUDI2_QUEUE_ID_DCORE1_TPC_0_2 = 62,
265 GAUDI2_QUEUE_ID_DCORE1_TPC_0_3 = 63,
266 GAUDI2_QUEUE_ID_DCORE1_TPC_1_0 = 64,
267 GAUDI2_QUEUE_ID_DCORE1_TPC_1_1 = 65,
268 GAUDI2_QUEUE_ID_DCORE1_TPC_1_2 = 66,
269 GAUDI2_QUEUE_ID_DCORE1_TPC_1_3 = 67,
270 GAUDI2_QUEUE_ID_DCORE1_TPC_2_0 = 68,
271 GAUDI2_QUEUE_ID_DCORE1_TPC_2_1 = 69,
272 GAUDI2_QUEUE_ID_DCORE1_TPC_2_2 = 70,
273 GAUDI2_QUEUE_ID_DCORE1_TPC_2_3 = 71,
274 GAUDI2_QUEUE_ID_DCORE1_TPC_3_0 = 72,
275 GAUDI2_QUEUE_ID_DCORE1_TPC_3_1 = 73,
276 GAUDI2_QUEUE_ID_DCORE1_TPC_3_2 = 74,
277 GAUDI2_QUEUE_ID_DCORE1_TPC_3_3 = 75,
278 GAUDI2_QUEUE_ID_DCORE1_TPC_4_0 = 76,
279 GAUDI2_QUEUE_ID_DCORE1_TPC_4_1 = 77,
280 GAUDI2_QUEUE_ID_DCORE1_TPC_4_2 = 78,
281 GAUDI2_QUEUE_ID_DCORE1_TPC_4_3 = 79,
282 GAUDI2_QUEUE_ID_DCORE1_TPC_5_0 = 80,
283 GAUDI2_QUEUE_ID_DCORE1_TPC_5_1 = 81,
284 GAUDI2_QUEUE_ID_DCORE1_TPC_5_2 = 82,
285 GAUDI2_QUEUE_ID_DCORE1_TPC_5_3 = 83,
286 GAUDI2_QUEUE_ID_DCORE2_EDMA_0_0 = 84,
287 GAUDI2_QUEUE_ID_DCORE2_EDMA_0_1 = 85,
288 GAUDI2_QUEUE_ID_DCORE2_EDMA_0_2 = 86,
289 GAUDI2_QUEUE_ID_DCORE2_EDMA_0_3 = 87,
290 GAUDI2_QUEUE_ID_DCORE2_EDMA_1_0 = 88,
291 GAUDI2_QUEUE_ID_DCORE2_EDMA_1_1 = 89,
292 GAUDI2_QUEUE_ID_DCORE2_EDMA_1_2 = 90,
293 GAUDI2_QUEUE_ID_DCORE2_EDMA_1_3 = 91,
294 GAUDI2_QUEUE_ID_DCORE2_MME_0_0 = 92,
295 GAUDI2_QUEUE_ID_DCORE2_MME_0_1 = 93,
296 GAUDI2_QUEUE_ID_DCORE2_MME_0_2 = 94,
297 GAUDI2_QUEUE_ID_DCORE2_MME_0_3 = 95,
298 GAUDI2_QUEUE_ID_DCORE2_TPC_0_0 = 96,
299 GAUDI2_QUEUE_ID_DCORE2_TPC_0_1 = 97,
300 GAUDI2_QUEUE_ID_DCORE2_TPC_0_2 = 98,
301 GAUDI2_QUEUE_ID_DCORE2_TPC_0_3 = 99,
302 GAUDI2_QUEUE_ID_DCORE2_TPC_1_0 = 100,
303 GAUDI2_QUEUE_ID_DCORE2_TPC_1_1 = 101,
304 GAUDI2_QUEUE_ID_DCORE2_TPC_1_2 = 102,
305 GAUDI2_QUEUE_ID_DCORE2_TPC_1_3 = 103,
306 GAUDI2_QUEUE_ID_DCORE2_TPC_2_0 = 104,
307 GAUDI2_QUEUE_ID_DCORE2_TPC_2_1 = 105,
308 GAUDI2_QUEUE_ID_DCORE2_TPC_2_2 = 106,
309 GAUDI2_QUEUE_ID_DCORE2_TPC_2_3 = 107,
310 GAUDI2_QUEUE_ID_DCORE2_TPC_3_0 = 108,
311 GAUDI2_QUEUE_ID_DCORE2_TPC_3_1 = 109,
312 GAUDI2_QUEUE_ID_DCORE2_TPC_3_2 = 110,
313 GAUDI2_QUEUE_ID_DCORE2_TPC_3_3 = 111,
314 GAUDI2_QUEUE_ID_DCORE2_TPC_4_0 = 112,
315 GAUDI2_QUEUE_ID_DCORE2_TPC_4_1 = 113,
316 GAUDI2_QUEUE_ID_DCORE2_TPC_4_2 = 114,
317 GAUDI2_QUEUE_ID_DCORE2_TPC_4_3 = 115,
318 GAUDI2_QUEUE_ID_DCORE2_TPC_5_0 = 116,
319 GAUDI2_QUEUE_ID_DCORE2_TPC_5_1 = 117,
320 GAUDI2_QUEUE_ID_DCORE2_TPC_5_2 = 118,
321 GAUDI2_QUEUE_ID_DCORE2_TPC_5_3 = 119,
322 GAUDI2_QUEUE_ID_DCORE3_EDMA_0_0 = 120,
323 GAUDI2_QUEUE_ID_DCORE3_EDMA_0_1 = 121,
324 GAUDI2_QUEUE_ID_DCORE3_EDMA_0_2 = 122,
325 GAUDI2_QUEUE_ID_DCORE3_EDMA_0_3 = 123,
326 GAUDI2_QUEUE_ID_DCORE3_EDMA_1_0 = 124,
327 GAUDI2_QUEUE_ID_DCORE3_EDMA_1_1 = 125,
328 GAUDI2_QUEUE_ID_DCORE3_EDMA_1_2 = 126,
329 GAUDI2_QUEUE_ID_DCORE3_EDMA_1_3 = 127,
330 GAUDI2_QUEUE_ID_DCORE3_MME_0_0 = 128,
331 GAUDI2_QUEUE_ID_DCORE3_MME_0_1 = 129,
332 GAUDI2_QUEUE_ID_DCORE3_MME_0_2 = 130,
333 GAUDI2_QUEUE_ID_DCORE3_MME_0_3 = 131,
334 GAUDI2_QUEUE_ID_DCORE3_TPC_0_0 = 132,
335 GAUDI2_QUEUE_ID_DCORE3_TPC_0_1 = 133,
336 GAUDI2_QUEUE_ID_DCORE3_TPC_0_2 = 134,
337 GAUDI2_QUEUE_ID_DCORE3_TPC_0_3 = 135,
338 GAUDI2_QUEUE_ID_DCORE3_TPC_1_0 = 136,
339 GAUDI2_QUEUE_ID_DCORE3_TPC_1_1 = 137,
340 GAUDI2_QUEUE_ID_DCORE3_TPC_1_2 = 138,
341 GAUDI2_QUEUE_ID_DCORE3_TPC_1_3 = 139,
342 GAUDI2_QUEUE_ID_DCORE3_TPC_2_0 = 140,
343 GAUDI2_QUEUE_ID_DCORE3_TPC_2_1 = 141,
344 GAUDI2_QUEUE_ID_DCORE3_TPC_2_2 = 142,
345 GAUDI2_QUEUE_ID_DCORE3_TPC_2_3 = 143,
346 GAUDI2_QUEUE_ID_DCORE3_TPC_3_0 = 144,
347 GAUDI2_QUEUE_ID_DCORE3_TPC_3_1 = 145,
348 GAUDI2_QUEUE_ID_DCORE3_TPC_3_2 = 146,
349 GAUDI2_QUEUE_ID_DCORE3_TPC_3_3 = 147,
350 GAUDI2_QUEUE_ID_DCORE3_TPC_4_0 = 148,
351 GAUDI2_QUEUE_ID_DCORE3_TPC_4_1 = 149,
352 GAUDI2_QUEUE_ID_DCORE3_TPC_4_2 = 150,
353 GAUDI2_QUEUE_ID_DCORE3_TPC_4_3 = 151,
354 GAUDI2_QUEUE_ID_DCORE3_TPC_5_0 = 152,
355 GAUDI2_QUEUE_ID_DCORE3_TPC_5_1 = 153,
356 GAUDI2_QUEUE_ID_DCORE3_TPC_5_2 = 154,
357 GAUDI2_QUEUE_ID_DCORE3_TPC_5_3 = 155,
358 GAUDI2_QUEUE_ID_NIC_0_0 = 156,
359 GAUDI2_QUEUE_ID_NIC_0_1 = 157,
360 GAUDI2_QUEUE_ID_NIC_0_2 = 158,
361 GAUDI2_QUEUE_ID_NIC_0_3 = 159,
362 GAUDI2_QUEUE_ID_NIC_1_0 = 160,
363 GAUDI2_QUEUE_ID_NIC_1_1 = 161,
364 GAUDI2_QUEUE_ID_NIC_1_2 = 162,
365 GAUDI2_QUEUE_ID_NIC_1_3 = 163,
366 GAUDI2_QUEUE_ID_NIC_2_0 = 164,
367 GAUDI2_QUEUE_ID_NIC_2_1 = 165,
368 GAUDI2_QUEUE_ID_NIC_2_2 = 166,
369 GAUDI2_QUEUE_ID_NIC_2_3 = 167,
370 GAUDI2_QUEUE_ID_NIC_3_0 = 168,
371 GAUDI2_QUEUE_ID_NIC_3_1 = 169,
372 GAUDI2_QUEUE_ID_NIC_3_2 = 170,
373 GAUDI2_QUEUE_ID_NIC_3_3 = 171,
374 GAUDI2_QUEUE_ID_NIC_4_0 = 172,
375 GAUDI2_QUEUE_ID_NIC_4_1 = 173,
376 GAUDI2_QUEUE_ID_NIC_4_2 = 174,
377 GAUDI2_QUEUE_ID_NIC_4_3 = 175,
378 GAUDI2_QUEUE_ID_NIC_5_0 = 176,
379 GAUDI2_QUEUE_ID_NIC_5_1 = 177,
380 GAUDI2_QUEUE_ID_NIC_5_2 = 178,
381 GAUDI2_QUEUE_ID_NIC_5_3 = 179,
382 GAUDI2_QUEUE_ID_NIC_6_0 = 180,
383 GAUDI2_QUEUE_ID_NIC_6_1 = 181,
384 GAUDI2_QUEUE_ID_NIC_6_2 = 182,
385 GAUDI2_QUEUE_ID_NIC_6_3 = 183,
386 GAUDI2_QUEUE_ID_NIC_7_0 = 184,
387 GAUDI2_QUEUE_ID_NIC_7_1 = 185,
388 GAUDI2_QUEUE_ID_NIC_7_2 = 186,
389 GAUDI2_QUEUE_ID_NIC_7_3 = 187,
390 GAUDI2_QUEUE_ID_NIC_8_0 = 188,
391 GAUDI2_QUEUE_ID_NIC_8_1 = 189,
392 GAUDI2_QUEUE_ID_NIC_8_2 = 190,
393 GAUDI2_QUEUE_ID_NIC_8_3 = 191,
394 GAUDI2_QUEUE_ID_NIC_9_0 = 192,
395 GAUDI2_QUEUE_ID_NIC_9_1 = 193,
396 GAUDI2_QUEUE_ID_NIC_9_2 = 194,
397 GAUDI2_QUEUE_ID_NIC_9_3 = 195,
398 GAUDI2_QUEUE_ID_NIC_10_0 = 196,
399 GAUDI2_QUEUE_ID_NIC_10_1 = 197,
400 GAUDI2_QUEUE_ID_NIC_10_2 = 198,
401 GAUDI2_QUEUE_ID_NIC_10_3 = 199,
402 GAUDI2_QUEUE_ID_NIC_11_0 = 200,
403 GAUDI2_QUEUE_ID_NIC_11_1 = 201,
404 GAUDI2_QUEUE_ID_NIC_11_2 = 202,
405 GAUDI2_QUEUE_ID_NIC_11_3 = 203,
406 GAUDI2_QUEUE_ID_NIC_12_0 = 204,
407 GAUDI2_QUEUE_ID_NIC_12_1 = 205,
408 GAUDI2_QUEUE_ID_NIC_12_2 = 206,
409 GAUDI2_QUEUE_ID_NIC_12_3 = 207,
410 GAUDI2_QUEUE_ID_NIC_13_0 = 208,
411 GAUDI2_QUEUE_ID_NIC_13_1 = 209,
412 GAUDI2_QUEUE_ID_NIC_13_2 = 210,
413 GAUDI2_QUEUE_ID_NIC_13_3 = 211,
414 GAUDI2_QUEUE_ID_NIC_14_0 = 212,
415 GAUDI2_QUEUE_ID_NIC_14_1 = 213,
416 GAUDI2_QUEUE_ID_NIC_14_2 = 214,
417 GAUDI2_QUEUE_ID_NIC_14_3 = 215,
418 GAUDI2_QUEUE_ID_NIC_15_0 = 216,
419 GAUDI2_QUEUE_ID_NIC_15_1 = 217,
420 GAUDI2_QUEUE_ID_NIC_15_2 = 218,
421 GAUDI2_QUEUE_ID_NIC_15_3 = 219,
422 GAUDI2_QUEUE_ID_NIC_16_0 = 220,
423 GAUDI2_QUEUE_ID_NIC_16_1 = 221,
424 GAUDI2_QUEUE_ID_NIC_16_2 = 222,
425 GAUDI2_QUEUE_ID_NIC_16_3 = 223,
426 GAUDI2_QUEUE_ID_NIC_17_0 = 224,
427 GAUDI2_QUEUE_ID_NIC_17_1 = 225,
428 GAUDI2_QUEUE_ID_NIC_17_2 = 226,
429 GAUDI2_QUEUE_ID_NIC_17_3 = 227,
430 GAUDI2_QUEUE_ID_NIC_18_0 = 228,
431 GAUDI2_QUEUE_ID_NIC_18_1 = 229,
432 GAUDI2_QUEUE_ID_NIC_18_2 = 230,
433 GAUDI2_QUEUE_ID_NIC_18_3 = 231,
434 GAUDI2_QUEUE_ID_NIC_19_0 = 232,
435 GAUDI2_QUEUE_ID_NIC_19_1 = 233,
436 GAUDI2_QUEUE_ID_NIC_19_2 = 234,
437 GAUDI2_QUEUE_ID_NIC_19_3 = 235,
438 GAUDI2_QUEUE_ID_NIC_20_0 = 236,
439 GAUDI2_QUEUE_ID_NIC_20_1 = 237,
440 GAUDI2_QUEUE_ID_NIC_20_2 = 238,
441 GAUDI2_QUEUE_ID_NIC_20_3 = 239,
442 GAUDI2_QUEUE_ID_NIC_21_0 = 240,
443 GAUDI2_QUEUE_ID_NIC_21_1 = 241,
444 GAUDI2_QUEUE_ID_NIC_21_2 = 242,
445 GAUDI2_QUEUE_ID_NIC_21_3 = 243,
446 GAUDI2_QUEUE_ID_NIC_22_0 = 244,
447 GAUDI2_QUEUE_ID_NIC_22_1 = 245,
448 GAUDI2_QUEUE_ID_NIC_22_2 = 246,
449 GAUDI2_QUEUE_ID_NIC_22_3 = 247,
450 GAUDI2_QUEUE_ID_NIC_23_0 = 248,
451 GAUDI2_QUEUE_ID_NIC_23_1 = 249,
452 GAUDI2_QUEUE_ID_NIC_23_2 = 250,
453 GAUDI2_QUEUE_ID_NIC_23_3 = 251,
454 GAUDI2_QUEUE_ID_ROT_0_0 = 252,
455 GAUDI2_QUEUE_ID_ROT_0_1 = 253,
456 GAUDI2_QUEUE_ID_ROT_0_2 = 254,
457 GAUDI2_QUEUE_ID_ROT_0_3 = 255,
458 GAUDI2_QUEUE_ID_ROT_1_0 = 256,
459 GAUDI2_QUEUE_ID_ROT_1_1 = 257,
460 GAUDI2_QUEUE_ID_ROT_1_2 = 258,
461 GAUDI2_QUEUE_ID_ROT_1_3 = 259,
462 GAUDI2_QUEUE_ID_CPU_PQ = 260,
469 * Used in the "busy_engines_mask" field in `struct hl_info_hw_idle'
472 enum goya_engine_id {
473 GOYA_ENGINE_ID_DMA_0 = 0,
474 GOYA_ENGINE_ID_DMA_1,
475 GOYA_ENGINE_ID_DMA_2,
476 GOYA_ENGINE_ID_DMA_3,
477 GOYA_ENGINE_ID_DMA_4,
478 GOYA_ENGINE_ID_MME_0,
479 GOYA_ENGINE_ID_TPC_0,
480 GOYA_ENGINE_ID_TPC_1,
481 GOYA_ENGINE_ID_TPC_2,
482 GOYA_ENGINE_ID_TPC_3,
483 GOYA_ENGINE_ID_TPC_4,
484 GOYA_ENGINE_ID_TPC_5,
485 GOYA_ENGINE_ID_TPC_6,
486 GOYA_ENGINE_ID_TPC_7,
490 enum gaudi_engine_id {
491 GAUDI_ENGINE_ID_DMA_0 = 0,
492 GAUDI_ENGINE_ID_DMA_1,
493 GAUDI_ENGINE_ID_DMA_2,
494 GAUDI_ENGINE_ID_DMA_3,
495 GAUDI_ENGINE_ID_DMA_4,
496 GAUDI_ENGINE_ID_DMA_5,
497 GAUDI_ENGINE_ID_DMA_6,
498 GAUDI_ENGINE_ID_DMA_7,
499 GAUDI_ENGINE_ID_MME_0,
500 GAUDI_ENGINE_ID_MME_1,
501 GAUDI_ENGINE_ID_MME_2,
502 GAUDI_ENGINE_ID_MME_3,
503 GAUDI_ENGINE_ID_TPC_0,
504 GAUDI_ENGINE_ID_TPC_1,
505 GAUDI_ENGINE_ID_TPC_2,
506 GAUDI_ENGINE_ID_TPC_3,
507 GAUDI_ENGINE_ID_TPC_4,
508 GAUDI_ENGINE_ID_TPC_5,
509 GAUDI_ENGINE_ID_TPC_6,
510 GAUDI_ENGINE_ID_TPC_7,
511 GAUDI_ENGINE_ID_NIC_0,
512 GAUDI_ENGINE_ID_NIC_1,
513 GAUDI_ENGINE_ID_NIC_2,
514 GAUDI_ENGINE_ID_NIC_3,
515 GAUDI_ENGINE_ID_NIC_4,
516 GAUDI_ENGINE_ID_NIC_5,
517 GAUDI_ENGINE_ID_NIC_6,
518 GAUDI_ENGINE_ID_NIC_7,
519 GAUDI_ENGINE_ID_NIC_8,
520 GAUDI_ENGINE_ID_NIC_9,
524 enum gaudi2_engine_id {
525 GAUDI2_DCORE0_ENGINE_ID_EDMA_0 = 0,
526 GAUDI2_DCORE0_ENGINE_ID_EDMA_1,
527 GAUDI2_DCORE0_ENGINE_ID_MME,
528 GAUDI2_DCORE0_ENGINE_ID_TPC_0,
529 GAUDI2_DCORE0_ENGINE_ID_TPC_1,
530 GAUDI2_DCORE0_ENGINE_ID_TPC_2,
531 GAUDI2_DCORE0_ENGINE_ID_TPC_3,
532 GAUDI2_DCORE0_ENGINE_ID_TPC_4,
533 GAUDI2_DCORE0_ENGINE_ID_TPC_5,
534 GAUDI2_DCORE0_ENGINE_ID_DEC_0,
535 GAUDI2_DCORE0_ENGINE_ID_DEC_1,
536 GAUDI2_DCORE1_ENGINE_ID_EDMA_0,
537 GAUDI2_DCORE1_ENGINE_ID_EDMA_1,
538 GAUDI2_DCORE1_ENGINE_ID_MME,
539 GAUDI2_DCORE1_ENGINE_ID_TPC_0,
540 GAUDI2_DCORE1_ENGINE_ID_TPC_1,
541 GAUDI2_DCORE1_ENGINE_ID_TPC_2,
542 GAUDI2_DCORE1_ENGINE_ID_TPC_3,
543 GAUDI2_DCORE1_ENGINE_ID_TPC_4,
544 GAUDI2_DCORE1_ENGINE_ID_TPC_5,
545 GAUDI2_DCORE1_ENGINE_ID_DEC_0,
546 GAUDI2_DCORE1_ENGINE_ID_DEC_1,
547 GAUDI2_DCORE2_ENGINE_ID_EDMA_0,
548 GAUDI2_DCORE2_ENGINE_ID_EDMA_1,
549 GAUDI2_DCORE2_ENGINE_ID_MME,
550 GAUDI2_DCORE2_ENGINE_ID_TPC_0,
551 GAUDI2_DCORE2_ENGINE_ID_TPC_1,
552 GAUDI2_DCORE2_ENGINE_ID_TPC_2,
553 GAUDI2_DCORE2_ENGINE_ID_TPC_3,
554 GAUDI2_DCORE2_ENGINE_ID_TPC_4,
555 GAUDI2_DCORE2_ENGINE_ID_TPC_5,
556 GAUDI2_DCORE2_ENGINE_ID_DEC_0,
557 GAUDI2_DCORE2_ENGINE_ID_DEC_1,
558 GAUDI2_DCORE3_ENGINE_ID_EDMA_0,
559 GAUDI2_DCORE3_ENGINE_ID_EDMA_1,
560 GAUDI2_DCORE3_ENGINE_ID_MME,
561 GAUDI2_DCORE3_ENGINE_ID_TPC_0,
562 GAUDI2_DCORE3_ENGINE_ID_TPC_1,
563 GAUDI2_DCORE3_ENGINE_ID_TPC_2,
564 GAUDI2_DCORE3_ENGINE_ID_TPC_3,
565 GAUDI2_DCORE3_ENGINE_ID_TPC_4,
566 GAUDI2_DCORE3_ENGINE_ID_TPC_5,
567 GAUDI2_DCORE3_ENGINE_ID_DEC_0,
568 GAUDI2_DCORE3_ENGINE_ID_DEC_1,
569 GAUDI2_DCORE0_ENGINE_ID_TPC_6,
570 GAUDI2_ENGINE_ID_PDMA_0,
571 GAUDI2_ENGINE_ID_PDMA_1,
572 GAUDI2_ENGINE_ID_ROT_0,
573 GAUDI2_ENGINE_ID_ROT_1,
574 GAUDI2_PCIE_ENGINE_ID_DEC_0,
575 GAUDI2_PCIE_ENGINE_ID_DEC_1,
576 GAUDI2_ENGINE_ID_NIC0_0,
577 GAUDI2_ENGINE_ID_NIC0_1,
578 GAUDI2_ENGINE_ID_NIC1_0,
579 GAUDI2_ENGINE_ID_NIC1_1,
580 GAUDI2_ENGINE_ID_NIC2_0,
581 GAUDI2_ENGINE_ID_NIC2_1,
582 GAUDI2_ENGINE_ID_NIC3_0,
583 GAUDI2_ENGINE_ID_NIC3_1,
584 GAUDI2_ENGINE_ID_NIC4_0,
585 GAUDI2_ENGINE_ID_NIC4_1,
586 GAUDI2_ENGINE_ID_NIC5_0,
587 GAUDI2_ENGINE_ID_NIC5_1,
588 GAUDI2_ENGINE_ID_NIC6_0,
589 GAUDI2_ENGINE_ID_NIC6_1,
590 GAUDI2_ENGINE_ID_NIC7_0,
591 GAUDI2_ENGINE_ID_NIC7_1,
592 GAUDI2_ENGINE_ID_NIC8_0,
593 GAUDI2_ENGINE_ID_NIC8_1,
594 GAUDI2_ENGINE_ID_NIC9_0,
595 GAUDI2_ENGINE_ID_NIC9_1,
596 GAUDI2_ENGINE_ID_NIC10_0,
597 GAUDI2_ENGINE_ID_NIC10_1,
598 GAUDI2_ENGINE_ID_NIC11_0,
599 GAUDI2_ENGINE_ID_NIC11_1,
600 GAUDI2_ENGINE_ID_SIZE
604 * ASIC specific PLL index
606 * Used to retrieve in frequency info of different IPs via
607 * HL_INFO_PLL_FREQUENCY under HL_IOCTL_INFO IOCTL. The enums need to be
608 * used as an index in struct hl_pll_frequency_info
611 enum hl_goya_pll_index {
622 enum hl_gaudi_pll_index {
623 HL_GAUDI_CPU_PLL = 0,
636 enum hl_gaudi2_pll_index {
637 HL_GAUDI2_CPU_PLL = 0,
653 * enum hl_goya_dma_direction - Direction of DMA operation inside a LIN_DMA packet that is
654 * submitted to the GOYA's DMA QMAN. This attribute is not relevant
655 * to the H/W but the kernel driver use it to parse the packet's
656 * addresses and patch/validate them.
657 * @HL_DMA_HOST_TO_DRAM: DMA operation from Host memory to GOYA's DDR.
658 * @HL_DMA_HOST_TO_SRAM: DMA operation from Host memory to GOYA's SRAM.
659 * @HL_DMA_DRAM_TO_SRAM: DMA operation from GOYA's DDR to GOYA's SRAM.
660 * @HL_DMA_SRAM_TO_DRAM: DMA operation from GOYA's SRAM to GOYA's DDR.
661 * @HL_DMA_SRAM_TO_HOST: DMA operation from GOYA's SRAM to Host memory.
662 * @HL_DMA_DRAM_TO_HOST: DMA operation from GOYA's DDR to Host memory.
663 * @HL_DMA_DRAM_TO_DRAM: DMA operation from GOYA's DDR to GOYA's DDR.
664 * @HL_DMA_SRAM_TO_SRAM: DMA operation from GOYA's SRAM to GOYA's SRAM.
665 * @HL_DMA_ENUM_MAX: number of values in enum
667 enum hl_goya_dma_direction {
680 * enum hl_device_status - Device status information.
681 * @HL_DEVICE_STATUS_OPERATIONAL: Device is operational.
682 * @HL_DEVICE_STATUS_IN_RESET: Device is currently during reset.
683 * @HL_DEVICE_STATUS_MALFUNCTION: Device is unusable.
684 * @HL_DEVICE_STATUS_NEEDS_RESET: Device needs reset because auto reset was disabled.
685 * @HL_DEVICE_STATUS_IN_DEVICE_CREATION: Device is operational but its creation is still in
687 * @HL_DEVICE_STATUS_IN_RESET_AFTER_DEVICE_RELEASE: Device is currently during reset that was
688 * triggered because the user released the device
689 * @HL_DEVICE_STATUS_LAST: Last status.
691 enum hl_device_status {
692 HL_DEVICE_STATUS_OPERATIONAL,
693 HL_DEVICE_STATUS_IN_RESET,
694 HL_DEVICE_STATUS_MALFUNCTION,
695 HL_DEVICE_STATUS_NEEDS_RESET,
696 HL_DEVICE_STATUS_IN_DEVICE_CREATION,
697 HL_DEVICE_STATUS_IN_RESET_AFTER_DEVICE_RELEASE,
698 HL_DEVICE_STATUS_LAST = HL_DEVICE_STATUS_IN_RESET_AFTER_DEVICE_RELEASE
701 enum hl_server_type {
702 HL_SERVER_TYPE_UNKNOWN = 0,
703 HL_SERVER_GAUDI_HLS1 = 1,
704 HL_SERVER_GAUDI_HLS1H = 2,
705 HL_SERVER_GAUDI_TYPE1 = 3,
706 HL_SERVER_GAUDI_TYPE2 = 4,
707 HL_SERVER_GAUDI2_HLS2 = 5
711 * Notifier event values - for the notification mechanism and the HL_INFO_GET_EVENTS command
713 * HL_NOTIFIER_EVENT_TPC_ASSERT - Indicates TPC assert event
714 * HL_NOTIFIER_EVENT_UNDEFINED_OPCODE - Indicates undefined operation code
715 * HL_NOTIFIER_EVENT_DEVICE_RESET - Indicates device requires a reset
716 * HL_NOTIFIER_EVENT_CS_TIMEOUT - Indicates CS timeout error
717 * HL_NOTIFIER_EVENT_DEVICE_UNAVAILABLE - Indicates device is unavailable
718 * HL_NOTIFIER_EVENT_USER_ENGINE_ERR - Indicates device engine in error state
719 * HL_NOTIFIER_EVENT_GENERAL_HW_ERR - Indicates device HW error
721 #define HL_NOTIFIER_EVENT_TPC_ASSERT (1ULL << 0)
722 #define HL_NOTIFIER_EVENT_UNDEFINED_OPCODE (1ULL << 1)
723 #define HL_NOTIFIER_EVENT_DEVICE_RESET (1ULL << 2)
724 #define HL_NOTIFIER_EVENT_CS_TIMEOUT (1ULL << 3)
725 #define HL_NOTIFIER_EVENT_DEVICE_UNAVAILABLE (1ULL << 4)
726 #define HL_NOTIFIER_EVENT_USER_ENGINE_ERR (1ULL << 5)
727 #define HL_NOTIFIER_EVENT_GENERAL_HW_ERR (1ULL << 6)
729 /* Opcode for management ioctl
731 * HW_IP_INFO - Receive information about different IP blocks in the
733 * HL_INFO_HW_EVENTS - Receive an array describing how many times each event
734 * occurred since the last hard reset.
735 * HL_INFO_DRAM_USAGE - Retrieve the dram usage inside the device and of the
736 * specific context. This is relevant only for devices
737 * where the dram is managed by the kernel driver
738 * HL_INFO_HW_IDLE - Retrieve information about the idle status of each
740 * HL_INFO_DEVICE_STATUS - Retrieve the device's status. This opcode doesn't
741 * require an open context.
742 * HL_INFO_DEVICE_UTILIZATION - Retrieve the total utilization of the device
743 * over the last period specified by the user.
744 * The period can be between 100ms to 1s, in
745 * resolution of 100ms. The return value is a
746 * percentage of the utilization rate.
747 * HL_INFO_HW_EVENTS_AGGREGATE - Receive an array describing how many times each
748 * event occurred since the driver was loaded.
749 * HL_INFO_CLK_RATE - Retrieve the current and maximum clock rate
750 * of the device in MHz. The maximum clock rate is
751 * configurable via sysfs parameter
752 * HL_INFO_RESET_COUNT - Retrieve the counts of the soft and hard reset
753 * operations performed on the device since the last
754 * time the driver was loaded.
755 * HL_INFO_TIME_SYNC - Retrieve the device's time alongside the host's time
756 * for synchronization.
757 * HL_INFO_CS_COUNTERS - Retrieve command submission counters
758 * HL_INFO_PCI_COUNTERS - Retrieve PCI counters
759 * HL_INFO_CLK_THROTTLE_REASON - Retrieve clock throttling reason
760 * HL_INFO_SYNC_MANAGER - Retrieve sync manager info per dcore
761 * HL_INFO_TOTAL_ENERGY - Retrieve total energy consumption
762 * HL_INFO_PLL_FREQUENCY - Retrieve PLL frequency
763 * HL_INFO_POWER - Retrieve power information
764 * HL_INFO_OPEN_STATS - Retrieve info regarding recent device open calls
765 * HL_INFO_DRAM_REPLACED_ROWS - Retrieve DRAM replaced rows info
766 * HL_INFO_DRAM_PENDING_ROWS - Retrieve DRAM pending rows num
767 * HL_INFO_LAST_ERR_OPEN_DEV_TIME - Retrieve timestamp of the last time the device was opened
768 * and CS timeout or razwi error occurred.
769 * HL_INFO_CS_TIMEOUT_EVENT - Retrieve CS timeout timestamp and its related CS sequence number.
770 * HL_INFO_RAZWI_EVENT - Retrieve parameters of razwi:
771 * Timestamp of razwi.
772 * The address which accessing it caused the razwi.
774 * Razwi cause, was it a page fault or MMU access error.
775 * HL_INFO_DEV_MEM_ALLOC_PAGE_SIZES - Retrieve valid page sizes for device memory allocation
776 * HL_INFO_SECURED_ATTESTATION - Retrieve attestation report of the boot.
777 * HL_INFO_REGISTER_EVENTFD - Register eventfd for event notifications.
778 * HL_INFO_UNREGISTER_EVENTFD - Unregister eventfd
779 * HL_INFO_GET_EVENTS - Retrieve the last occurred events
780 * HL_INFO_UNDEFINED_OPCODE_EVENT - Retrieve last undefined opcode error information.
782 #define HL_INFO_HW_IP_INFO 0
783 #define HL_INFO_HW_EVENTS 1
784 #define HL_INFO_DRAM_USAGE 2
785 #define HL_INFO_HW_IDLE 3
786 #define HL_INFO_DEVICE_STATUS 4
787 #define HL_INFO_DEVICE_UTILIZATION 6
788 #define HL_INFO_HW_EVENTS_AGGREGATE 7
789 #define HL_INFO_CLK_RATE 8
790 #define HL_INFO_RESET_COUNT 9
791 #define HL_INFO_TIME_SYNC 10
792 #define HL_INFO_CS_COUNTERS 11
793 #define HL_INFO_PCI_COUNTERS 12
794 #define HL_INFO_CLK_THROTTLE_REASON 13
795 #define HL_INFO_SYNC_MANAGER 14
796 #define HL_INFO_TOTAL_ENERGY 15
797 #define HL_INFO_PLL_FREQUENCY 16
798 #define HL_INFO_POWER 17
799 #define HL_INFO_OPEN_STATS 18
800 #define HL_INFO_DRAM_REPLACED_ROWS 21
801 #define HL_INFO_DRAM_PENDING_ROWS 22
802 #define HL_INFO_LAST_ERR_OPEN_DEV_TIME 23
803 #define HL_INFO_CS_TIMEOUT_EVENT 24
804 #define HL_INFO_RAZWI_EVENT 25
805 #define HL_INFO_DEV_MEM_ALLOC_PAGE_SIZES 26
806 #define HL_INFO_SECURED_ATTESTATION 27
807 #define HL_INFO_REGISTER_EVENTFD 28
808 #define HL_INFO_UNREGISTER_EVENTFD 29
809 #define HL_INFO_GET_EVENTS 30
810 #define HL_INFO_UNDEFINED_OPCODE_EVENT 31
811 #define HL_INFO_ENGINE_STATUS 32
813 #define HL_INFO_VERSION_MAX_LEN 128
814 #define HL_INFO_CARD_NAME_MAX_LEN 16
816 /* Maximum buffer size for retrieving engines status */
817 #define HL_ENGINES_DATA_MAX_SIZE SZ_1M
820 * struct hl_info_hw_ip_info - hardware information on various IPs in the ASIC
821 * @sram_base_address: The first SRAM physical base address that is free to be
823 * @dram_base_address: The first DRAM virtual or physical base address that is
824 * free to be used by the user.
825 * @dram_size: The DRAM size that is available to the user.
826 * @sram_size: The SRAM size that is available to the user.
827 * @num_of_events: The number of events that can be received from the f/w. This
828 * is needed so the user can what is the size of the h/w events
829 * array he needs to pass to the kernel when he wants to fetch
830 * the event counters.
831 * @device_id: PCI device ID of the ASIC.
832 * @module_id: Module ID of the ASIC for mezzanine cards in servers
834 * @decoder_enabled_mask: Bit-mask that represents which decoders are enabled.
835 * @first_available_interrupt_id: The first available interrupt ID for the user
836 * to be used when it works with user interrupts.
837 * Relevant for Gaudi2 and later.
838 * @server_type: Server type that the Gaudi ASIC is currently installed in.
839 * The value is according to enum hl_server_type
840 * @cpld_version: CPLD version on the board.
841 * @psoc_pci_pll_nr: PCI PLL NR value. Needed by the profiler in some ASICs.
842 * @psoc_pci_pll_nf: PCI PLL NF value. Needed by the profiler in some ASICs.
843 * @psoc_pci_pll_od: PCI PLL OD value. Needed by the profiler in some ASICs.
844 * @psoc_pci_pll_div_factor: PCI PLL DIV factor value. Needed by the profiler
846 * @tpc_enabled_mask: Bit-mask that represents which TPCs are enabled. Relevant
847 * for Goya/Gaudi only.
848 * @dram_enabled: Whether the DRAM is enabled.
849 * @security_enabled: Whether security is enabled on device.
850 * @mme_master_slave_mode: Indicate whether the MME is working in master/slave
851 * configuration. Relevant for Greco and later.
852 * @cpucp_version: The CPUCP f/w version.
853 * @card_name: The card name as passed by the f/w.
854 * @tpc_enabled_mask_ext: Bit-mask that represents which TPCs are enabled.
855 * Relevant for Greco and later.
856 * @dram_page_size: The DRAM physical page size.
857 * @edma_enabled_mask: Bit-mask that represents which EDMAs are enabled.
858 * Relevant for Gaudi2 and later.
859 * @number_of_user_interrupts: The number of interrupts that are available to the userspace
860 * application to use. Relevant for Gaudi2 and later.
861 * @device_mem_alloc_default_page_size: default page size used in device memory allocation.
863 struct hl_info_hw_ip_info {
864 __u64 sram_base_address;
865 __u64 dram_base_address;
871 __u32 decoder_enabled_mask;
872 __u16 first_available_interrupt_id;
875 __u32 psoc_pci_pll_nr;
876 __u32 psoc_pci_pll_nf;
877 __u32 psoc_pci_pll_od;
878 __u32 psoc_pci_pll_div_factor;
879 __u8 tpc_enabled_mask;
881 __u8 security_enabled;
882 __u8 mme_master_slave_mode;
883 __u8 cpucp_version[HL_INFO_VERSION_MAX_LEN];
884 __u8 card_name[HL_INFO_CARD_NAME_MAX_LEN];
885 __u64 tpc_enabled_mask_ext;
886 __u64 dram_page_size;
887 __u32 edma_enabled_mask;
888 __u16 number_of_user_interrupts;
891 __u64 device_mem_alloc_default_page_size;
894 struct hl_info_dram_usage {
899 #define HL_BUSY_ENGINES_MASK_EXT_SIZE 2
901 struct hl_info_hw_idle {
904 * Bitmask of busy engines.
905 * Bits definition is according to `enum <chip>_engine_id'.
907 __u32 busy_engines_mask;
910 * Extended Bitmask of busy engines.
911 * Bits definition is according to `enum <chip>_engine_id'.
913 __u64 busy_engines_mask_ext[HL_BUSY_ENGINES_MASK_EXT_SIZE];
916 struct hl_info_device_status {
921 struct hl_info_device_utilization {
926 struct hl_info_clk_rate {
927 __u32 cur_clk_rate_mhz;
928 __u32 max_clk_rate_mhz;
931 struct hl_info_reset_count {
932 __u32 hard_reset_cnt;
933 __u32 soft_reset_cnt;
936 struct hl_info_time_sync {
942 * struct hl_info_pci_counters - pci counters
943 * @rx_throughput: PCI rx throughput KBps
944 * @tx_throughput: PCI tx throughput KBps
945 * @replay_cnt: PCI replay counter
947 struct hl_info_pci_counters {
953 enum hl_clk_throttling_type {
954 HL_CLK_THROTTLE_TYPE_POWER,
955 HL_CLK_THROTTLE_TYPE_THERMAL,
956 HL_CLK_THROTTLE_TYPE_MAX
959 /* clk_throttling_reason masks */
960 #define HL_CLK_THROTTLE_POWER (1 << HL_CLK_THROTTLE_TYPE_POWER)
961 #define HL_CLK_THROTTLE_THERMAL (1 << HL_CLK_THROTTLE_TYPE_THERMAL)
964 * struct hl_info_clk_throttle - clock throttling reason
965 * @clk_throttling_reason: each bit represents a clk throttling reason
966 * @clk_throttling_timestamp_us: represents CPU timestamp in microseconds of the start-event
967 * @clk_throttling_duration_ns: the clock throttle time in nanosec
969 struct hl_info_clk_throttle {
970 __u32 clk_throttling_reason;
972 __u64 clk_throttling_timestamp_us[HL_CLK_THROTTLE_TYPE_MAX];
973 __u64 clk_throttling_duration_ns[HL_CLK_THROTTLE_TYPE_MAX];
977 * struct hl_info_energy - device energy information
978 * @total_energy_consumption: total device energy consumption
980 struct hl_info_energy {
981 __u64 total_energy_consumption;
984 #define HL_PLL_NUM_OUTPUTS 4
986 struct hl_pll_frequency_info {
987 __u16 output[HL_PLL_NUM_OUTPUTS];
991 * struct hl_open_stats_info - device open statistics information
992 * @open_counter: ever growing counter, increased on each successful dev open
993 * @last_open_period_ms: duration (ms) device was open last time
994 * @is_compute_ctx_active: Whether there is an active compute context executing
995 * @compute_ctx_in_release: true if the current compute context is being released
997 struct hl_open_stats_info {
999 __u64 last_open_period_ms;
1000 __u8 is_compute_ctx_active;
1001 __u8 compute_ctx_in_release;
1006 * struct hl_power_info - power information
1007 * @power: power consumption
1009 struct hl_power_info {
1014 * struct hl_info_sync_manager - sync manager information
1015 * @first_available_sync_object: first available sob
1016 * @first_available_monitor: first available monitor
1017 * @first_available_cq: first available cq
1019 struct hl_info_sync_manager {
1020 __u32 first_available_sync_object;
1021 __u32 first_available_monitor;
1022 __u32 first_available_cq;
1027 * struct hl_info_cs_counters - command submission counters
1028 * @total_out_of_mem_drop_cnt: total dropped due to memory allocation issue
1029 * @ctx_out_of_mem_drop_cnt: context dropped due to memory allocation issue
1030 * @total_parsing_drop_cnt: total dropped due to error in packet parsing
1031 * @ctx_parsing_drop_cnt: context dropped due to error in packet parsing
1032 * @total_queue_full_drop_cnt: total dropped due to queue full
1033 * @ctx_queue_full_drop_cnt: context dropped due to queue full
1034 * @total_device_in_reset_drop_cnt: total dropped due to device in reset
1035 * @ctx_device_in_reset_drop_cnt: context dropped due to device in reset
1036 * @total_max_cs_in_flight_drop_cnt: total dropped due to maximum CS in-flight
1037 * @ctx_max_cs_in_flight_drop_cnt: context dropped due to maximum CS in-flight
1038 * @total_validation_drop_cnt: total dropped due to validation error
1039 * @ctx_validation_drop_cnt: context dropped due to validation error
1041 struct hl_info_cs_counters {
1042 __u64 total_out_of_mem_drop_cnt;
1043 __u64 ctx_out_of_mem_drop_cnt;
1044 __u64 total_parsing_drop_cnt;
1045 __u64 ctx_parsing_drop_cnt;
1046 __u64 total_queue_full_drop_cnt;
1047 __u64 ctx_queue_full_drop_cnt;
1048 __u64 total_device_in_reset_drop_cnt;
1049 __u64 ctx_device_in_reset_drop_cnt;
1050 __u64 total_max_cs_in_flight_drop_cnt;
1051 __u64 ctx_max_cs_in_flight_drop_cnt;
1052 __u64 total_validation_drop_cnt;
1053 __u64 ctx_validation_drop_cnt;
1057 * struct hl_info_last_err_open_dev_time - last error boot information.
1058 * @timestamp: timestamp of last time the device was opened and error occurred.
1060 struct hl_info_last_err_open_dev_time {
1065 * struct hl_info_cs_timeout_event - last CS timeout information.
1066 * @timestamp: timestamp when last CS timeout event occurred.
1067 * @seq: sequence number of last CS timeout event.
1069 struct hl_info_cs_timeout_event {
1074 #define HL_RAZWI_PAGE_FAULT 0
1075 #define HL_RAZWI_MMU_ACCESS_ERROR 1
1078 * struct hl_info_razwi_event - razwi information.
1079 * @timestamp: timestamp of razwi.
1080 * @addr: address which accessing it caused razwi.
1081 * @engine_id_1: engine id of the razwi initiator, if it was initiated by engine that does not
1082 * have engine id it will be set to U16_MAX.
1083 * @engine_id_2: second engine id of razwi initiator. Might happen that razwi have 2 possible
1084 * engines which one them caused the razwi. In that case, it will contain the
1085 * second possible engine id, otherwise it will be set to U16_MAX.
1086 * @no_engine_id: if razwi initiator does not have engine id, this field will be set to 1,
1088 * @error_type: cause of razwi, page fault or access error, otherwise it will be set to U8_MAX.
1089 * @pad: padding to 64 bit.
1091 struct hl_info_razwi_event {
1101 #define MAX_QMAN_STREAMS_INFO 4
1102 #define OPCODE_INFO_MAX_ADDR_SIZE 8
1104 * struct hl_info_undefined_opcode_event - info about last undefined opcode error
1105 * @timestamp: timestamp of the undefined opcode error
1106 * @cb_addr_streams: CB addresses (per stream) that are currently exists in the PQ
1107 * entries. In case all streams array entries are
1108 * filled with values, it means the execution was in Lower-CP.
1109 * @cq_addr: the address of the current handled command buffer
1110 * @cq_size: the size of the current handled command buffer
1111 * @cb_addr_streams_len: num of streams - actual len of cb_addr_streams array.
1112 * should be equal to 1 in case of undefined opcode
1113 * in Upper-CP (specific stream) and equal to 4 incase
1114 * of undefined opcode in Lower-CP.
1115 * @engine_id: engine-id that the error occurred on
1116 * @stream_id: the stream id the error occurred on. In case the stream equals to
1117 * MAX_QMAN_STREAMS_INFO it means the error occurred on a Lower-CP.
1119 struct hl_info_undefined_opcode_event {
1121 __u64 cb_addr_streams[MAX_QMAN_STREAMS_INFO][OPCODE_INFO_MAX_ADDR_SIZE];
1124 __u32 cb_addr_streams_len;
1130 * struct hl_info_dev_memalloc_page_sizes - valid page sizes in device mem alloc information.
1131 * @page_order_bitmask: bitmap in which a set bit represents the order of the supported page size
1132 * (e.g. 0x2100000 means that 1MB and 32MB pages are supported).
1134 struct hl_info_dev_memalloc_page_sizes {
1135 __u64 page_order_bitmask;
1138 #define SEC_PCR_DATA_BUF_SZ 256
1139 #define SEC_PCR_QUOTE_BUF_SZ 510 /* (512 - 2) 2 bytes used for size */
1140 #define SEC_SIGNATURE_BUF_SZ 255 /* (256 - 1) 1 byte used for size */
1141 #define SEC_PUB_DATA_BUF_SZ 510 /* (512 - 2) 2 bytes used for size */
1142 #define SEC_CERTIFICATE_BUF_SZ 2046 /* (2048 - 2) 2 bytes used for size */
1145 * struct hl_info_sec_attest - attestation report of the boot
1146 * @nonce: number only used once. random number provided by host. this also passed to the quote
1147 * command as a qualifying data.
1148 * @pcr_quote_len: length of the attestation quote data (bytes)
1149 * @pub_data_len: length of the public data (bytes)
1150 * @certificate_len: length of the certificate (bytes)
1151 * @pcr_num_reg: number of PCR registers in the pcr_data array
1152 * @pcr_reg_len: length of each PCR register in the pcr_data array (bytes)
1153 * @quote_sig_len: length of the attestation report signature (bytes)
1154 * @pcr_data: raw values of the PCR registers
1155 * @pcr_quote: attestation report data structure
1156 * @quote_sig: signature structure of the attestation report
1157 * @public_data: public key for the signed attestation
1158 * (outPublic + name + qualifiedName)
1159 * @certificate: certificate for the attestation signing key
1161 struct hl_info_sec_attest {
1163 __u16 pcr_quote_len;
1165 __u16 certificate_len;
1169 __u8 pcr_data[SEC_PCR_DATA_BUF_SZ];
1170 __u8 pcr_quote[SEC_PCR_QUOTE_BUF_SZ];
1171 __u8 quote_sig[SEC_SIGNATURE_BUF_SZ];
1172 __u8 public_data[SEC_PUB_DATA_BUF_SZ];
1173 __u8 certificate[SEC_CERTIFICATE_BUF_SZ];
1185 * struct hl_info_args - Main structure to retrieve device related information.
1186 * @return_pointer: User space address of the relevant structure related to HL_INFO_* operation
1188 * @return_size: Size of the structure used in @return_pointer, just like "size" in "snprintf", it
1189 * limits how many bytes the kernel can write. For hw_events array, the size should be
1190 * hl_info_hw_ip_info.num_of_events * sizeof(__u32).
1191 * @op: Defines which type of information to be retrieved. Refer HL_INFO_* for details.
1192 * @dcore_id: DCORE id for which the information is relevant (for Gaudi refer to enum gaudi_dcores).
1193 * @ctx_id: Context ID of the user. Currently not in use.
1194 * @period_ms: Period value, in milliseconds, for utilization rate in range 100ms - 1000ms in 100 ms
1195 * resolution. Currently not in use.
1196 * @pll_index: Index as defined in hl_<asic type>_pll_index enumeration.
1197 * @eventfd: event file descriptor for event notifications.
1198 * @user_buffer_actual_size: Actual data size which was copied to user allocated buffer by the
1199 * driver. It is possible for the user to allocate buffer larger than
1200 * needed, hence updating this variable so user will know the exact amount
1201 * of bytes copied by the kernel to the buffer.
1202 * @sec_attest_nonce: Nonce number used for attestation report.
1203 * @pad: Padding to 64 bit.
1205 struct hl_info_args {
1206 __u64 return_pointer;
1216 __u32 user_buffer_actual_size;
1217 __u32 sec_attest_nonce;
1223 /* Opcode to create a new command buffer */
1224 #define HL_CB_OP_CREATE 0
1225 /* Opcode to destroy previously created command buffer */
1226 #define HL_CB_OP_DESTROY 1
1227 /* Opcode to retrieve information about a command buffer */
1228 #define HL_CB_OP_INFO 2
1230 /* 2MB minus 32 bytes for 2xMSG_PROT */
1231 #define HL_MAX_CB_SIZE (0x200000 - 32)
1233 /* Indicates whether the command buffer should be mapped to the device's MMU */
1234 #define HL_CB_FLAGS_MAP 0x1
1236 /* Used with HL_CB_OP_INFO opcode to get the device va address for kernel mapped CB */
1237 #define HL_CB_FLAGS_GET_DEVICE_VA 0x2
1240 /* Handle of CB or 0 if we want to create one */
1245 /* Size of CB. Maximum size is HL_MAX_CB_SIZE. The minimum size that
1246 * will be allocated, regardless of this parameter's value, is PAGE_SIZE
1250 /* Context ID - Currently not in use */
1262 /* Information about CB */
1264 /* Usage count of CB */
1269 /* CB mapped address to device MMU */
1277 struct hl_cb_out out;
1280 /* HL_CS_CHUNK_FLAGS_ values
1282 * HL_CS_CHUNK_FLAGS_USER_ALLOC_CB:
1283 * Indicates if the CB was allocated and mapped by userspace
1284 * (relevant to greco and above). User allocated CB is a command buffer,
1285 * allocated by the user, via malloc (or similar). After allocating the
1286 * CB, the user invokes - “memory ioctl” to map the user memory into a
1287 * device virtual address. The user provides this address via the
1288 * cb_handle field. The interface provides the ability to create a
1289 * large CBs, Which aren’t limited to “HL_MAX_CB_SIZE”. Therefore, it
1290 * increases the PCI-DMA queues throughput. This CB allocation method
1291 * also reduces the use of Linux DMA-able memory pool. Which are limited
1292 * and used by other Linux sub-systems.
1294 #define HL_CS_CHUNK_FLAGS_USER_ALLOC_CB 0x1
1297 * This structure size must always be fixed to 64-bytes for backward
1300 struct hl_cs_chunk {
1303 * For external queue, this represents a Handle of CB on the
1305 * For internal queue in Goya, this represents an SRAM or
1306 * a DRAM address of the internal CB. In Gaudi, this might also
1307 * represent a mapped host address of the CB.
1310 * For H/W queue, this represents either a Handle of CB on the
1311 * Host, or an SRAM, a DRAM, or a mapped host address of the CB.
1313 * A mapped host address is in the device address space, after
1314 * a host address was mapped by the device MMU.
1318 /* Relevant only when HL_CS_FLAGS_WAIT or
1319 * HL_CS_FLAGS_COLLECTIVE_WAIT is set
1320 * This holds address of array of u64 values that contain
1321 * signal CS sequence numbers. The wait described by
1322 * this job will listen on all those signals
1323 * (wait event per signal)
1325 __u64 signal_seq_arr;
1328 * Relevant only when HL_CS_FLAGS_WAIT or
1329 * HL_CS_FLAGS_COLLECTIVE_WAIT is set
1330 * along with HL_CS_FLAGS_ENCAP_SIGNALS.
1331 * This is the CS sequence which has the encapsulated signals.
1333 __u64 encaps_signal_seq;
1336 /* Index of queue to put the CB on */
1341 * Size of command buffer with valid packets
1342 * Can be smaller then actual CB size
1346 /* Relevant only when HL_CS_FLAGS_WAIT or
1347 * HL_CS_FLAGS_COLLECTIVE_WAIT is set.
1348 * Number of entries in signal_seq_arr
1350 __u32 num_signal_seq_arr;
1352 /* Relevant only when HL_CS_FLAGS_WAIT or
1353 * HL_CS_FLAGS_COLLECTIVE_WAIT is set along
1354 * with HL_CS_FLAGS_ENCAP_SIGNALS
1355 * This set the signals range that the user want to wait for
1356 * out of the whole reserved signals range.
1357 * e.g if the signals range is 20, and user don't want
1358 * to wait for signal 8, so he set this offset to 7, then
1359 * he call the API again with 9 and so on till 20.
1361 __u32 encaps_signal_offset;
1364 /* HL_CS_CHUNK_FLAGS_* */
1365 __u32 cs_chunk_flags;
1367 /* Relevant only when HL_CS_FLAGS_COLLECTIVE_WAIT is set.
1368 * This holds the collective engine ID. The wait described by this job
1369 * will sync with this engine and with all NICs before completion.
1371 __u32 collective_engine_id;
1373 /* Align structure to 64 bytes */
1377 /* SIGNAL/WAIT/COLLECTIVE_WAIT flags are mutually exclusive */
1378 #define HL_CS_FLAGS_FORCE_RESTORE 0x1
1379 #define HL_CS_FLAGS_SIGNAL 0x2
1380 #define HL_CS_FLAGS_WAIT 0x4
1381 #define HL_CS_FLAGS_COLLECTIVE_WAIT 0x8
1383 #define HL_CS_FLAGS_TIMESTAMP 0x20
1384 #define HL_CS_FLAGS_STAGED_SUBMISSION 0x40
1385 #define HL_CS_FLAGS_STAGED_SUBMISSION_FIRST 0x80
1386 #define HL_CS_FLAGS_STAGED_SUBMISSION_LAST 0x100
1387 #define HL_CS_FLAGS_CUSTOM_TIMEOUT 0x200
1388 #define HL_CS_FLAGS_SKIP_RESET_ON_TIMEOUT 0x400
1391 * The encapsulated signals CS is merged into the existing CS ioctls.
1392 * In order to use this feature need to follow the below procedure:
1393 * 1. Reserve signals, set the CS type to HL_CS_FLAGS_RESERVE_SIGNALS_ONLY
1394 * the output of this API will be the SOB offset from CFG_BASE.
1395 * this address will be used to patch CB cmds to do the signaling for this
1396 * SOB by incrementing it's value.
1397 * for reverting the reservation use HL_CS_FLAGS_UNRESERVE_SIGNALS_ONLY
1398 * CS type, note that this might fail if out-of-sync happened to the SOB
1399 * value, in case other signaling request to the same SOB occurred between
1400 * reserve-unreserve calls.
1401 * 2. Use the staged CS to do the encapsulated signaling jobs.
1402 * use HL_CS_FLAGS_STAGED_SUBMISSION and HL_CS_FLAGS_STAGED_SUBMISSION_FIRST
1403 * along with HL_CS_FLAGS_ENCAP_SIGNALS flag, and set encaps_signal_offset
1404 * field. This offset allows app to wait on part of the reserved signals.
1405 * 3. Use WAIT/COLLECTIVE WAIT CS along with HL_CS_FLAGS_ENCAP_SIGNALS flag
1406 * to wait for the encapsulated signals.
1408 #define HL_CS_FLAGS_ENCAP_SIGNALS 0x800
1409 #define HL_CS_FLAGS_RESERVE_SIGNALS_ONLY 0x1000
1410 #define HL_CS_FLAGS_UNRESERVE_SIGNALS_ONLY 0x2000
1413 * The engine cores CS is merged into the existing CS ioctls.
1414 * Use it to control the engine cores mode.
1416 #define HL_CS_FLAGS_ENGINE_CORE_COMMAND 0x4000
1418 #define HL_CS_STATUS_SUCCESS 0
1420 #define HL_MAX_JOBS_PER_CS 512
1422 /* HL_ENGINE_CORE_ values
1424 * HL_ENGINE_CORE_HALT: engine core halt
1425 * HL_ENGINE_CORE_RUN: engine core run
1427 #define HL_ENGINE_CORE_HALT (1 << 0)
1428 #define HL_ENGINE_CORE_RUN (1 << 1)
1434 /* this holds address of array of hl_cs_chunk for restore phase */
1435 __u64 chunks_restore;
1437 /* holds address of array of hl_cs_chunk for execution phase */
1438 __u64 chunks_execute;
1441 /* Valid only when HL_CS_FLAGS_ENGINE_CORE_COMMAND is set */
1443 /* this holds address of array of uint32 for engine_cores */
1446 /* number of engine cores in engine_cores array */
1447 __u32 num_engine_cores;
1449 /* the core command to be sent towards engine cores */
1456 * Sequence number of a staged submission CS
1457 * valid only if HL_CS_FLAGS_STAGED_SUBMISSION is set and
1458 * HL_CS_FLAGS_STAGED_SUBMISSION_FIRST is unset.
1463 * Encapsulated signals handle id
1464 * Valid for two flows:
1465 * 1. CS with encapsulated signals:
1466 * when HL_CS_FLAGS_STAGED_SUBMISSION and
1467 * HL_CS_FLAGS_STAGED_SUBMISSION_FIRST
1468 * and HL_CS_FLAGS_ENCAP_SIGNALS are set.
1469 * 2. unreserve signals:
1470 * valid when HL_CS_FLAGS_UNRESERVE_SIGNALS_ONLY is set.
1472 __u32 encaps_sig_handle_id;
1474 /* Valid only when HL_CS_FLAGS_RESERVE_SIGNALS_ONLY is set */
1476 /* Encapsulated signals number */
1477 __u32 encaps_signals_count;
1479 /* Encapsulated signals queue index (stream) */
1480 __u32 encaps_signals_q_idx;
1484 /* Number of chunks in restore phase array. Maximum number is
1485 * HL_MAX_JOBS_PER_CS
1487 __u32 num_chunks_restore;
1489 /* Number of chunks in execution array. Maximum number is
1490 * HL_MAX_JOBS_PER_CS
1492 __u32 num_chunks_execute;
1494 /* timeout in seconds - valid only if HL_CS_FLAGS_CUSTOM_TIMEOUT
1502 /* Context ID - Currently not in use */
1510 * seq holds the sequence number of the CS to pass to wait
1511 * ioctl. All values are valid except for 0 and ULLONG_MAX
1515 /* Valid only when HL_CS_FLAGS_RESERVE_SIGNALS_ONLY is set */
1517 /* This is the reserved signal handle id */
1520 /* This is the signals count */
1529 * SOB base address offset
1530 * Valid only when HL_CS_FLAGS_RESERVE_SIGNALS_ONLY or HL_CS_FLAGS_SIGNAL is set
1532 __u32 sob_base_addr_offset;
1535 * Count of completed signals in SOB before current signal submission.
1536 * Valid only when (HL_CS_FLAGS_ENCAP_SIGNALS & HL_CS_FLAGS_STAGED_SUBMISSION)
1537 * or HL_CS_FLAGS_SIGNAL is set
1539 __u16 sob_count_before_submission;
1545 struct hl_cs_out out;
1548 #define HL_WAIT_CS_FLAGS_INTERRUPT 0x2
1549 #define HL_WAIT_CS_FLAGS_INTERRUPT_MASK 0xFFF00000
1550 #define HL_WAIT_CS_FLAGS_ANY_CQ_INTERRUPT 0xFFF00000
1551 #define HL_WAIT_CS_FLAGS_ANY_DEC_INTERRUPT 0xFFE00000
1552 #define HL_WAIT_CS_FLAGS_MULTI_CS 0x4
1553 #define HL_WAIT_CS_FLAGS_INTERRUPT_KERNEL_CQ 0x10
1554 #define HL_WAIT_CS_FLAGS_REGISTER_INTERRUPT 0x20
1556 #define HL_WAIT_MULTI_CS_LIST_MAX_LEN 32
1558 struct hl_wait_cs_in {
1562 * In case of wait_cs holds the CS sequence number.
1563 * In case of wait for multi CS hold a user pointer to
1564 * an array of CS sequence numbers
1567 /* Absolute timeout to wait for command submission
1575 /* User address for completion comparison.
1576 * upon interrupt, driver will compare the value pointed
1577 * by this address with the supplied target value.
1578 * in order not to perform any comparison, set address
1580 * Relevant only when HL_WAIT_CS_FLAGS_INTERRUPT is set
1584 /* cq_counters_handle to a kernel mapped cb which contains
1586 * Relevant only when HL_WAIT_CS_FLAGS_INTERRUPT_KERNEL_CQ is set
1588 __u64 cq_counters_handle;
1591 /* Target value for completion comparison */
1596 /* Context ID - Currently not in use */
1599 /* HL_WAIT_CS_FLAGS_*
1600 * If HL_WAIT_CS_FLAGS_INTERRUPT is set, this field should include
1601 * interrupt id according to HL_WAIT_CS_FLAGS_INTERRUPT_MASK
1603 * in order to wait for any CQ interrupt, set interrupt value to
1604 * HL_WAIT_CS_FLAGS_ANY_CQ_INTERRUPT.
1606 * in order to wait for any decoder interrupt, set interrupt value to
1607 * HL_WAIT_CS_FLAGS_ANY_DEC_INTERRUPT.
1613 /* Multi CS API info- valid entries in multi-CS array */
1618 /* Absolute timeout to wait for an interrupt in microseconds.
1619 * Relevant only when HL_WAIT_CS_FLAGS_INTERRUPT is set
1621 __u64 interrupt_timeout_us;
1625 * cq counter offset inside the counters cb pointed by cq_counters_handle above.
1626 * upon interrupt, driver will compare the value pointed
1627 * by this address (cq_counters_handle + cq_counters_offset)
1628 * with the supplied target value.
1629 * relevant only when HL_WAIT_CS_FLAGS_INTERRUPT_KERNEL_CQ is set
1631 __u64 cq_counters_offset;
1634 * Timestamp_handle timestamps buffer handle.
1635 * relevant only when HL_WAIT_CS_FLAGS_REGISTER_INTERRUPT is set
1637 __u64 timestamp_handle;
1640 * Timestamp_offset is offset inside the timestamp buffer pointed by timestamp_handle above.
1641 * upon interrupt, if the cq reached the target value then driver will write
1642 * timestamp to this offset.
1643 * relevant only when HL_WAIT_CS_FLAGS_REGISTER_INTERRUPT is set
1645 __u64 timestamp_offset;
1648 #define HL_WAIT_CS_STATUS_COMPLETED 0
1649 #define HL_WAIT_CS_STATUS_BUSY 1
1650 #define HL_WAIT_CS_STATUS_TIMEDOUT 2
1651 #define HL_WAIT_CS_STATUS_ABORTED 3
1653 #define HL_WAIT_CS_STATUS_FLAG_GONE 0x1
1654 #define HL_WAIT_CS_STATUS_FLAG_TIMESTAMP_VLD 0x2
1656 struct hl_wait_cs_out {
1657 /* HL_WAIT_CS_STATUS_* */
1659 /* HL_WAIT_CS_STATUS_FLAG* */
1662 * valid only if HL_WAIT_CS_STATUS_FLAG_TIMESTAMP_VLD is set
1663 * for wait_cs: timestamp of CS completion
1664 * for wait_multi_cs: timestamp of FIRST CS completion
1666 __s64 timestamp_nsec;
1667 /* multi CS completion bitmap */
1668 __u32 cs_completion_map;
1672 union hl_wait_cs_args {
1673 struct hl_wait_cs_in in;
1674 struct hl_wait_cs_out out;
1677 /* Opcode to allocate device memory */
1678 #define HL_MEM_OP_ALLOC 0
1680 /* Opcode to free previously allocated device memory */
1681 #define HL_MEM_OP_FREE 1
1683 /* Opcode to map host and device memory */
1684 #define HL_MEM_OP_MAP 2
1686 /* Opcode to unmap previously mapped host and device memory */
1687 #define HL_MEM_OP_UNMAP 3
1689 /* Opcode to map a hw block */
1690 #define HL_MEM_OP_MAP_BLOCK 4
1692 /* Opcode to create DMA-BUF object for an existing device memory allocation
1693 * and to export an FD of that DMA-BUF back to the caller
1695 #define HL_MEM_OP_EXPORT_DMABUF_FD 5
1697 /* Opcode to create timestamps pool for user interrupts registration support
1698 * The memory will be allocated by the kernel driver, A timestamp buffer which the user
1699 * will get handle to it for mmap, and another internal buffer used by the
1700 * driver for registration management
1701 * The memory will be freed when the user closes the file descriptor(ctx close)
1703 #define HL_MEM_OP_TS_ALLOC 6
1706 #define HL_MEM_CONTIGUOUS 0x1
1707 #define HL_MEM_SHARED 0x2
1708 #define HL_MEM_USERPTR 0x4
1709 #define HL_MEM_FORCE_HINT 0x8
1710 #define HL_MEM_PREFETCH 0x40
1713 * structure hl_mem_in - structure that handle input args for memory IOCTL
1714 * @union arg: union of structures to be used based on the input operation
1715 * @op: specify the requested memory operation (one of the HL_MEM_OP_* definitions).
1716 * @flags: flags for the memory operation (one of the HL_MEM_* definitions).
1717 * For the HL_MEM_OP_EXPORT_DMABUF_FD opcode, this field holds the DMA-BUF file/FD flags.
1718 * @ctx_id: context ID - currently not in use.
1719 * @num_of_elements: number of timestamp elements used only with HL_MEM_OP_TS_ALLOC opcode.
1724 * structure for device memory allocation (used with the HL_MEM_OP_ALLOC op)
1725 * @mem_size: memory size to allocate
1726 * @page_size: page size to use on allocation. when the value is 0 the default page
1727 * size will be taken.
1735 * structure for free-ing device memory (used with the HL_MEM_OP_FREE op)
1736 * @handle: handle returned from HL_MEM_OP_ALLOC
1743 * structure for mapping device memory (used with the HL_MEM_OP_MAP op)
1744 * @hint_addr: requested virtual address of mapped memory.
1745 * the driver will try to map the requested region to this hint
1746 * address, as long as the address is valid and not already mapped.
1747 * the user should check the returned address of the IOCTL to make
1748 * sure he got the hint address.
1749 * passing 0 here means that the driver will choose the address itself.
1750 * @handle: handle returned from HL_MEM_OP_ALLOC.
1758 * structure for mapping host memory (used with the HL_MEM_OP_MAP op)
1759 * @host_virt_addr: address of allocated host memory.
1760 * @hint_addr: requested virtual address of mapped memory.
1761 * the driver will try to map the requested region to this hint
1762 * address, as long as the address is valid and not already mapped.
1763 * the user should check the returned address of the IOCTL to make
1764 * sure he got the hint address.
1765 * passing 0 here means that the driver will choose the address itself.
1766 * @size: size of allocated host memory.
1769 __u64 host_virt_addr;
1775 * structure for mapping hw block (used with the HL_MEM_OP_MAP_BLOCK op)
1776 * @block_addr:HW block address to map, a handle and size will be returned
1777 * to the user and will be used to mmap the relevant block.
1778 * only addresses from configuration space are allowed.
1785 * structure for unmapping host memory (used with the HL_MEM_OP_UNMAP op)
1786 * @device_virt_addr: virtual address returned from HL_MEM_OP_MAP
1789 __u64 device_virt_addr;
1793 * structure for exporting DMABUF object (used with
1794 * the HL_MEM_OP_EXPORT_DMABUF_FD op)
1795 * @handle: handle returned from HL_MEM_OP_ALLOC.
1796 * in Gaudi, where we don't have MMU for the device memory, the
1797 * driver expects a physical address (instead of a handle) in the
1798 * device memory space.
1799 * @mem_size: size of memory allocation. Relevant only for GAUDI
1810 __u32 num_of_elements;
1816 * Used for HL_MEM_OP_MAP as the virtual address that was
1817 * assigned in the device VA space.
1818 * A value of 0 means the requested operation failed.
1820 __u64 device_virt_addr;
1823 * Used in HL_MEM_OP_ALLOC
1824 * This is the assigned handle for the allocated memory
1830 * Used in HL_MEM_OP_MAP_BLOCK.
1831 * This is the assigned handle for the mapped block
1836 * Used in HL_MEM_OP_MAP_BLOCK
1837 * This is the size of the mapped block
1844 /* Returned in HL_MEM_OP_EXPORT_DMABUF_FD. Represents the
1845 * DMA-BUF object that was created to describe a memory
1846 * allocation on the device's memory space. The FD should be
1847 * passed to the importer driver
1854 struct hl_mem_in in;
1855 struct hl_mem_out out;
1858 #define HL_DEBUG_MAX_AUX_VALUES 10
1860 struct hl_debug_params_etr {
1861 /* Address in memory to allocate buffer */
1862 __u64 buffer_address;
1864 /* Size of buffer to allocate */
1867 /* Sink operation mode: SW fifo, HW fifo, Circular buffer */
1872 struct hl_debug_params_etf {
1873 /* Address in memory to allocate buffer */
1874 __u64 buffer_address;
1876 /* Size of buffer to allocate */
1879 /* Sink operation mode: SW fifo, HW fifo, Circular buffer */
1884 struct hl_debug_params_stm {
1885 /* Two bit masks for HW event and Stimulus Port */
1889 /* Trace source ID */
1892 /* Frequency for the timestamp register */
1896 struct hl_debug_params_bmon {
1897 /* Two address ranges that the user can request to filter */
1904 /* Capture window configuration */
1908 /* Trace source ID */
1911 /* Control register */
1914 /* Two more address ranges that the user can request to filter */
1922 struct hl_debug_params_spmu {
1923 /* Event types selection */
1924 __u64 event_types[HL_DEBUG_MAX_AUX_VALUES];
1926 /* Number of event types selection */
1927 __u32 event_types_num;
1929 /* TRC configuration register values */
1931 __u32 trc_ctrl_host_val;
1932 __u32 trc_en_host_val;
1935 /* Opcode for ETR component */
1936 #define HL_DEBUG_OP_ETR 0
1937 /* Opcode for ETF component */
1938 #define HL_DEBUG_OP_ETF 1
1939 /* Opcode for STM component */
1940 #define HL_DEBUG_OP_STM 2
1941 /* Opcode for FUNNEL component */
1942 #define HL_DEBUG_OP_FUNNEL 3
1943 /* Opcode for BMON component */
1944 #define HL_DEBUG_OP_BMON 4
1945 /* Opcode for SPMU component */
1946 #define HL_DEBUG_OP_SPMU 5
1947 /* Opcode for timestamp (deprecated) */
1948 #define HL_DEBUG_OP_TIMESTAMP 6
1949 /* Opcode for setting the device into or out of debug mode. The enable
1950 * variable should be 1 for enabling debug mode and 0 for disabling it
1952 #define HL_DEBUG_OP_SET_MODE 7
1954 struct hl_debug_args {
1956 * Pointer to user input structure.
1957 * This field is relevant to specific opcodes.
1960 /* Pointer to user output structure */
1962 /* Size of user input structure */
1964 /* Size of user output structure */
1969 * Register index in the component, taken from the debug_regs_index enum
1970 * in the various ASIC header files
1973 /* Enable/disable */
1975 /* Context ID - Currently not in use */
1980 * Various information operations such as:
1981 * - H/W IP information
1982 * - Current dram usage
1984 * The user calls this IOCTL with an opcode that describes the required
1985 * information. The user should supply a pointer to a user-allocated memory
1986 * chunk, which will be filled by the driver with the requested information.
1988 * The user supplies the maximum amount of size to copy into the user's memory,
1989 * in order to prevent data corruption in case of differences between the
1990 * definitions of structures in kernel and userspace, e.g. in case of old
1991 * userspace and new kernel driver
1993 #define HL_IOCTL_INFO \
1994 _IOWR('H', 0x01, struct hl_info_args)
1998 * - Request a Command Buffer
1999 * - Destroy a Command Buffer
2001 * The command buffers are memory blocks that reside in DMA-able address
2002 * space and are physically contiguous so they can be accessed by the device
2003 * directly. They are allocated using the coherent DMA API.
2005 * When creating a new CB, the IOCTL returns a handle of it, and the user-space
2006 * process needs to use that handle to mmap the buffer so it can access them.
2008 * In some instances, the device must access the command buffer through the
2009 * device's MMU, and thus its memory should be mapped. In these cases, user can
2010 * indicate the driver that such a mapping is required.
2011 * The resulting device virtual address will be used internally by the driver,
2012 * and won't be returned to user.
2015 #define HL_IOCTL_CB \
2016 _IOWR('H', 0x02, union hl_cb_args)
2019 * Command Submission
2021 * To submit work to the device, the user need to call this IOCTL with a set
2022 * of JOBS. That set of JOBS constitutes a CS object.
2023 * Each JOB will be enqueued on a specific queue, according to the user's input.
2024 * There can be more then one JOB per queue.
2026 * The CS IOCTL will receive two sets of JOBS. One set is for "restore" phase
2027 * and a second set is for "execution" phase.
2028 * The JOBS on the "restore" phase are enqueued only after context-switch
2029 * (or if its the first CS for this context). The user can also order the
2030 * driver to run the "restore" phase explicitly
2033 * There are two types of queues - external and internal. External queues
2034 * are DMA queues which transfer data from/to the Host. All other queues are
2035 * internal. The driver will get completion notifications from the device only
2036 * on JOBS which are enqueued in the external queues.
2039 * There is a single type of queue for all types of engines, either DMA engines
2040 * for transfers from/to the host or inside the device, or compute engines.
2041 * The driver will get completion notifications from the device for all queues.
2043 * For jobs on external queues, the user needs to create command buffers
2044 * through the CB ioctl and give the CB's handle to the CS ioctl. For jobs on
2045 * internal queues, the user needs to prepare a "command buffer" with packets
2046 * on either the device SRAM/DRAM or the host, and give the device address of
2047 * that buffer to the CS ioctl.
2048 * For jobs on H/W queues both options of command buffers are valid.
2050 * This IOCTL is asynchronous in regard to the actual execution of the CS. This
2051 * means it returns immediately after ALL the JOBS were enqueued on their
2052 * relevant queues. Therefore, the user mustn't assume the CS has been completed
2053 * or has even started to execute.
2055 * Upon successful enqueue, the IOCTL returns a sequence number which the user
2056 * can use with the "Wait for CS" IOCTL to check whether the handle's CS
2057 * non-internal JOBS have been completed. Note that if the CS has internal JOBS
2058 * which can execute AFTER the external JOBS have finished, the driver might
2059 * report that the CS has finished executing BEFORE the internal JOBS have
2060 * actually finished executing.
2062 * Even though the sequence number increments per CS, the user can NOT
2063 * automatically assume that if CS with sequence number N finished, then CS
2064 * with sequence number N-1 also finished. The user can make this assumption if
2065 * and only if CS N and CS N-1 are exactly the same (same CBs for the same
2068 #define HL_IOCTL_CS \
2069 _IOWR('H', 0x03, union hl_cs_args)
2072 * Wait for Command Submission
2074 * The user can call this IOCTL with a handle it received from the CS IOCTL
2075 * to wait until the handle's CS has finished executing. The user will wait
2076 * inside the kernel until the CS has finished or until the user-requested
2077 * timeout has expired.
2079 * If the timeout value is 0, the driver won't sleep at all. It will check
2080 * the status of the CS and return immediately
2082 * The return value of the IOCTL is a standard Linux error code. The possible
2085 * EINTR - Kernel waiting has been interrupted, e.g. due to OS signal
2086 * that the user process received
2087 * ETIMEDOUT - The CS has caused a timeout on the device
2088 * EIO - The CS was aborted (usually because the device was reset)
2089 * ENODEV - The device wants to do hard-reset (so user need to close FD)
2091 * The driver also returns a custom define in case the IOCTL call returned 0.
2092 * The define can be one of the following:
2094 * HL_WAIT_CS_STATUS_COMPLETED - The CS has been completed successfully (0)
2095 * HL_WAIT_CS_STATUS_BUSY - The CS is still executing (0)
2096 * HL_WAIT_CS_STATUS_TIMEDOUT - The CS has caused a timeout on the device
2098 * HL_WAIT_CS_STATUS_ABORTED - The CS was aborted, usually because the
2099 * device was reset (EIO)
2102 #define HL_IOCTL_WAIT_CS \
2103 _IOWR('H', 0x04, union hl_wait_cs_args)
2107 * - Map host memory to device MMU
2108 * - Unmap host memory from device MMU
2110 * This IOCTL allows the user to map host memory to the device MMU
2112 * For host memory, the IOCTL doesn't allocate memory. The user is supposed
2113 * to allocate the memory in user-space (malloc/new). The driver pins the
2114 * physical pages (up to the allowed limit by the OS), assigns a virtual
2115 * address in the device VA space and initializes the device MMU.
2117 * There is an option for the user to specify the requested virtual address.
2120 #define HL_IOCTL_MEMORY \
2121 _IOWR('H', 0x05, union hl_mem_args)
2125 * - Enable/disable the ETR/ETF/FUNNEL/STM/BMON/SPMU debug traces
2127 * This IOCTL allows the user to get debug traces from the chip.
2129 * Before the user can send configuration requests of the various
2130 * debug/profile engines, it needs to set the device into debug mode.
2131 * This is because the debug/profile infrastructure is shared component in the
2132 * device and we can't allow multiple users to access it at the same time.
2134 * Once a user set the device into debug mode, the driver won't allow other
2135 * users to "work" with the device, i.e. open a FD. If there are multiple users
2136 * opened on the device, the driver won't allow any user to debug the device.
2138 * For each configuration request, the user needs to provide the register index
2139 * and essential data such as buffer address and size.
2141 * Once the user has finished using the debug/profile engines, he should
2142 * set the device into non-debug mode, i.e. disable debug mode.
2144 * The driver can decide to "kick out" the user if he abuses this interface.
2147 #define HL_IOCTL_DEBUG \
2148 _IOWR('H', 0x06, struct hl_debug_args)
2150 #define HL_COMMAND_START 0x01
2151 #define HL_COMMAND_END 0x07
2153 #endif /* HABANALABS_H_ */