Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
1/* SPDX-License-Identifier: GPL-2.0 WITH Linux-syscall-note
2 *
3 * Copyright 2016-2022 HabanaLabs, Ltd.
4 * All Rights Reserved.
5 *
6 */
7
8#ifndef HABANALABS_H_
9#define HABANALABS_H_
10
11#include <linux/types.h>
12#include <linux/ioctl.h>
13
14/*
15 * Defines that are asic-specific but constitutes as ABI between kernel driver
16 * and userspace
17 */
18#define GOYA_KMD_SRAM_RESERVED_SIZE_FROM_START 0x8000 /* 32KB */
19#define GAUDI_DRIVER_SRAM_RESERVED_SIZE_FROM_START 0x80 /* 128 bytes */
20
21/*
22 * 128 SOBs reserved for collective wait
23 * 16 SOBs reserved for sync stream
24 */
25#define GAUDI_FIRST_AVAILABLE_W_S_SYNC_OBJECT 144
26
27/*
28 * 64 monitors reserved for collective wait
29 * 8 monitors reserved for sync stream
30 */
31#define GAUDI_FIRST_AVAILABLE_W_S_MONITOR 72
32
33/* Max number of elements in timestamps registration buffers */
34#define TS_MAX_ELEMENTS_NUM (1 << 20) /* 1MB */
35
36/*
37 * Goya queue Numbering
38 *
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
42 * recommended)
43 */
44
45enum goya_queue_id {
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,
61 GOYA_QUEUE_ID_SIZE
62};
63
64/*
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.
68 */
69
70enum gaudi_queue_id {
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 */
184 GAUDI_QUEUE_ID_SIZE
185};
186
187/*
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.
191 *
192 * When in legacy mode, the user sends the queue id per JOB according to
193 * enum gaudi2_queue_id below.
194 *
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
199 */
200
201enum 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,
463 GAUDI2_QUEUE_ID_SIZE
464};
465
466/*
467 * Engine Numbering
468 *
469 * Used in the "busy_engines_mask" field in `struct hl_info_hw_idle'
470 */
471
472enum 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,
487 GOYA_ENGINE_ID_SIZE
488};
489
490enum 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,
521 GAUDI_ENGINE_ID_SIZE
522};
523
524enum 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
601};
602
603/*
604 * ASIC specific PLL index
605 *
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
609 */
610
611enum hl_goya_pll_index {
612 HL_GOYA_CPU_PLL = 0,
613 HL_GOYA_IC_PLL,
614 HL_GOYA_MC_PLL,
615 HL_GOYA_MME_PLL,
616 HL_GOYA_PCI_PLL,
617 HL_GOYA_EMMC_PLL,
618 HL_GOYA_TPC_PLL,
619 HL_GOYA_PLL_MAX
620};
621
622enum hl_gaudi_pll_index {
623 HL_GAUDI_CPU_PLL = 0,
624 HL_GAUDI_PCI_PLL,
625 HL_GAUDI_SRAM_PLL,
626 HL_GAUDI_HBM_PLL,
627 HL_GAUDI_NIC_PLL,
628 HL_GAUDI_DMA_PLL,
629 HL_GAUDI_MESH_PLL,
630 HL_GAUDI_MME_PLL,
631 HL_GAUDI_TPC_PLL,
632 HL_GAUDI_IF_PLL,
633 HL_GAUDI_PLL_MAX
634};
635
636enum hl_gaudi2_pll_index {
637 HL_GAUDI2_CPU_PLL = 0,
638 HL_GAUDI2_PCI_PLL,
639 HL_GAUDI2_SRAM_PLL,
640 HL_GAUDI2_HBM_PLL,
641 HL_GAUDI2_NIC_PLL,
642 HL_GAUDI2_DMA_PLL,
643 HL_GAUDI2_MESH_PLL,
644 HL_GAUDI2_MME_PLL,
645 HL_GAUDI2_TPC_PLL,
646 HL_GAUDI2_IF_PLL,
647 HL_GAUDI2_VID_PLL,
648 HL_GAUDI2_MSS_PLL,
649 HL_GAUDI2_PLL_MAX
650};
651
652/**
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
666 */
667enum hl_goya_dma_direction {
668 HL_DMA_HOST_TO_DRAM,
669 HL_DMA_HOST_TO_SRAM,
670 HL_DMA_DRAM_TO_SRAM,
671 HL_DMA_SRAM_TO_DRAM,
672 HL_DMA_SRAM_TO_HOST,
673 HL_DMA_DRAM_TO_HOST,
674 HL_DMA_DRAM_TO_DRAM,
675 HL_DMA_SRAM_TO_SRAM,
676 HL_DMA_ENUM_MAX
677};
678
679/**
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
686 * progress.
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.
690 */
691enum 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
699};
700
701enum 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
708};
709
710/* Opcode for management ioctl
711 *
712 * HW_IP_INFO - Receive information about different IP blocks in the
713 * device.
714 * HL_INFO_HW_EVENTS - Receive an array describing how many times each event
715 * occurred since the last hard reset.
716 * HL_INFO_DRAM_USAGE - Retrieve the dram usage inside the device and of the
717 * specific context. This is relevant only for devices
718 * where the dram is managed by the kernel driver
719 * HL_INFO_HW_IDLE - Retrieve information about the idle status of each
720 * internal engine.
721 * HL_INFO_DEVICE_STATUS - Retrieve the device's status. This opcode doesn't
722 * require an open context.
723 * HL_INFO_DEVICE_UTILIZATION - Retrieve the total utilization of the device
724 * over the last period specified by the user.
725 * The period can be between 100ms to 1s, in
726 * resolution of 100ms. The return value is a
727 * percentage of the utilization rate.
728 * HL_INFO_HW_EVENTS_AGGREGATE - Receive an array describing how many times each
729 * event occurred since the driver was loaded.
730 * HL_INFO_CLK_RATE - Retrieve the current and maximum clock rate
731 * of the device in MHz. The maximum clock rate is
732 * configurable via sysfs parameter
733 * HL_INFO_RESET_COUNT - Retrieve the counts of the soft and hard reset
734 * operations performed on the device since the last
735 * time the driver was loaded.
736 * HL_INFO_TIME_SYNC - Retrieve the device's time alongside the host's time
737 * for synchronization.
738 * HL_INFO_CS_COUNTERS - Retrieve command submission counters
739 * HL_INFO_PCI_COUNTERS - Retrieve PCI counters
740 * HL_INFO_CLK_THROTTLE_REASON - Retrieve clock throttling reason
741 * HL_INFO_SYNC_MANAGER - Retrieve sync manager info per dcore
742 * HL_INFO_TOTAL_ENERGY - Retrieve total energy consumption
743 * HL_INFO_PLL_FREQUENCY - Retrieve PLL frequency
744 * HL_INFO_POWER - Retrieve power information
745 * HL_INFO_OPEN_STATS - Retrieve info regarding recent device open calls
746 * HL_INFO_DRAM_REPLACED_ROWS - Retrieve DRAM replaced rows info
747 * HL_INFO_DRAM_PENDING_ROWS - Retrieve DRAM pending rows num
748 * HL_INFO_LAST_ERR_OPEN_DEV_TIME - Retrieve timestamp of the last time the device was opened
749 * and CS timeout or razwi error occurred.
750 * HL_INFO_CS_TIMEOUT_EVENT - Retrieve CS timeout timestamp and its related CS sequence number.
751 * HL_INFO_RAZWI_EVENT - Retrieve parameters of razwi:
752 * Timestamp of razwi.
753 * The address which accessing it caused the razwi.
754 * Razwi initiator.
755 * Razwi cause, was it a page fault or MMU access error.
756 * HL_INFO_DEV_MEM_ALLOC_PAGE_SIZES - Retrieve valid page sizes for device memory allocation
757 * HL_INFO_REGISTER_EVENTFD - Register eventfd for event notifications.
758 * HL_INFO_UNREGISTER_EVENTFD - Unregister eventfd
759 * HL_INFO_GET_EVENTS - Retrieve the last occurred events
760 * HL_INFO_UNDEFINED_OPCODE_EVENT - Retrieve last undefined opcode error information.
761 */
762#define HL_INFO_HW_IP_INFO 0
763#define HL_INFO_HW_EVENTS 1
764#define HL_INFO_DRAM_USAGE 2
765#define HL_INFO_HW_IDLE 3
766#define HL_INFO_DEVICE_STATUS 4
767#define HL_INFO_DEVICE_UTILIZATION 6
768#define HL_INFO_HW_EVENTS_AGGREGATE 7
769#define HL_INFO_CLK_RATE 8
770#define HL_INFO_RESET_COUNT 9
771#define HL_INFO_TIME_SYNC 10
772#define HL_INFO_CS_COUNTERS 11
773#define HL_INFO_PCI_COUNTERS 12
774#define HL_INFO_CLK_THROTTLE_REASON 13
775#define HL_INFO_SYNC_MANAGER 14
776#define HL_INFO_TOTAL_ENERGY 15
777#define HL_INFO_PLL_FREQUENCY 16
778#define HL_INFO_POWER 17
779#define HL_INFO_OPEN_STATS 18
780#define HL_INFO_DRAM_REPLACED_ROWS 21
781#define HL_INFO_DRAM_PENDING_ROWS 22
782#define HL_INFO_LAST_ERR_OPEN_DEV_TIME 23
783#define HL_INFO_CS_TIMEOUT_EVENT 24
784#define HL_INFO_RAZWI_EVENT 25
785#define HL_INFO_DEV_MEM_ALLOC_PAGE_SIZES 26
786#define HL_INFO_REGISTER_EVENTFD 28
787#define HL_INFO_UNREGISTER_EVENTFD 29
788#define HL_INFO_GET_EVENTS 30
789#define HL_INFO_UNDEFINED_OPCODE_EVENT 31
790
791#define HL_INFO_VERSION_MAX_LEN 128
792#define HL_INFO_CARD_NAME_MAX_LEN 16
793
794/**
795 * struct hl_info_hw_ip_info - hardware information on various IPs in the ASIC
796 * @sram_base_address: The first SRAM physical base address that is free to be
797 * used by the user.
798 * @dram_base_address: The first DRAM virtual or physical base address that is
799 * free to be used by the user.
800 * @dram_size: The DRAM size that is available to the user.
801 * @sram_size: The SRAM size that is available to the user.
802 * @num_of_events: The number of events that can be received from the f/w. This
803 * is needed so the user can what is the size of the h/w events
804 * array he needs to pass to the kernel when he wants to fetch
805 * the event counters.
806 * @device_id: PCI device ID of the ASIC.
807 * @module_id: Module ID of the ASIC for mezzanine cards in servers
808 * (From OCP spec).
809 * @decoder_enabled_mask: Bit-mask that represents which decoders are enabled.
810 * @first_available_interrupt_id: The first available interrupt ID for the user
811 * to be used when it works with user interrupts.
812 * Relevant for Gaudi2 and later.
813 * @server_type: Server type that the Gaudi ASIC is currently installed in.
814 * The value is according to enum hl_server_type
815 * @cpld_version: CPLD version on the board.
816 * @psoc_pci_pll_nr: PCI PLL NR value. Needed by the profiler in some ASICs.
817 * @psoc_pci_pll_nf: PCI PLL NF value. Needed by the profiler in some ASICs.
818 * @psoc_pci_pll_od: PCI PLL OD value. Needed by the profiler in some ASICs.
819 * @psoc_pci_pll_div_factor: PCI PLL DIV factor value. Needed by the profiler
820 * in some ASICs.
821 * @tpc_enabled_mask: Bit-mask that represents which TPCs are enabled. Relevant
822 * for Goya/Gaudi only.
823 * @dram_enabled: Whether the DRAM is enabled.
824 * @mme_master_slave_mode: Indicate whether the MME is working in master/slave
825 * configuration. Relevant for Greco and later.
826 * @cpucp_version: The CPUCP f/w version.
827 * @card_name: The card name as passed by the f/w.
828 * @tpc_enabled_mask_ext: Bit-mask that represents which TPCs are enabled.
829 * Relevant for Greco and later.
830 * @dram_page_size: The DRAM physical page size.
831 * @edma_enabled_mask: Bit-mask that represents which EDMAs are enabled.
832 * Relevant for Gaudi2 and later.
833 * @number_of_user_interrupts: The number of interrupts that are available to the userspace
834 * application to use. Relevant for Gaudi2 and later.
835 * @device_mem_alloc_default_page_size: default page size used in device memory allocation.
836 */
837struct hl_info_hw_ip_info {
838 __u64 sram_base_address;
839 __u64 dram_base_address;
840 __u64 dram_size;
841 __u32 sram_size;
842 __u32 num_of_events;
843 __u32 device_id;
844 __u32 module_id;
845 __u32 decoder_enabled_mask;
846 __u16 first_available_interrupt_id;
847 __u16 server_type;
848 __u32 cpld_version;
849 __u32 psoc_pci_pll_nr;
850 __u32 psoc_pci_pll_nf;
851 __u32 psoc_pci_pll_od;
852 __u32 psoc_pci_pll_div_factor;
853 __u8 tpc_enabled_mask;
854 __u8 dram_enabled;
855 __u8 reserved;
856 __u8 mme_master_slave_mode;
857 __u8 cpucp_version[HL_INFO_VERSION_MAX_LEN];
858 __u8 card_name[HL_INFO_CARD_NAME_MAX_LEN];
859 __u64 tpc_enabled_mask_ext;
860 __u64 dram_page_size;
861 __u32 edma_enabled_mask;
862 __u16 number_of_user_interrupts;
863 __u16 pad2;
864 __u64 reserved4;
865 __u64 device_mem_alloc_default_page_size;
866};
867
868struct hl_info_dram_usage {
869 __u64 dram_free_mem;
870 __u64 ctx_dram_mem;
871};
872
873#define HL_BUSY_ENGINES_MASK_EXT_SIZE 2
874
875struct hl_info_hw_idle {
876 __u32 is_idle;
877 /*
878 * Bitmask of busy engines.
879 * Bits definition is according to `enum <chip>_enging_id'.
880 */
881 __u32 busy_engines_mask;
882
883 /*
884 * Extended Bitmask of busy engines.
885 * Bits definition is according to `enum <chip>_enging_id'.
886 */
887 __u64 busy_engines_mask_ext[HL_BUSY_ENGINES_MASK_EXT_SIZE];
888};
889
890struct hl_info_device_status {
891 __u32 status;
892 __u32 pad;
893};
894
895struct hl_info_device_utilization {
896 __u32 utilization;
897 __u32 pad;
898};
899
900struct hl_info_clk_rate {
901 __u32 cur_clk_rate_mhz;
902 __u32 max_clk_rate_mhz;
903};
904
905struct hl_info_reset_count {
906 __u32 hard_reset_cnt;
907 __u32 soft_reset_cnt;
908};
909
910struct hl_info_time_sync {
911 __u64 device_time;
912 __u64 host_time;
913};
914
915/**
916 * struct hl_info_pci_counters - pci counters
917 * @rx_throughput: PCI rx throughput KBps
918 * @tx_throughput: PCI tx throughput KBps
919 * @replay_cnt: PCI replay counter
920 */
921struct hl_info_pci_counters {
922 __u64 rx_throughput;
923 __u64 tx_throughput;
924 __u64 replay_cnt;
925};
926
927enum hl_clk_throttling_type {
928 HL_CLK_THROTTLE_TYPE_POWER,
929 HL_CLK_THROTTLE_TYPE_THERMAL,
930 HL_CLK_THROTTLE_TYPE_MAX
931};
932
933/* clk_throttling_reason masks */
934#define HL_CLK_THROTTLE_POWER (1 << HL_CLK_THROTTLE_TYPE_POWER)
935#define HL_CLK_THROTTLE_THERMAL (1 << HL_CLK_THROTTLE_TYPE_THERMAL)
936
937/**
938 * struct hl_info_clk_throttle - clock throttling reason
939 * @clk_throttling_reason: each bit represents a clk throttling reason
940 * @clk_throttling_timestamp_us: represents CPU timestamp in microseconds of the start-event
941 * @clk_throttling_duration_ns: the clock throttle time in nanosec
942 */
943struct hl_info_clk_throttle {
944 __u32 clk_throttling_reason;
945 __u32 pad;
946 __u64 clk_throttling_timestamp_us[HL_CLK_THROTTLE_TYPE_MAX];
947 __u64 clk_throttling_duration_ns[HL_CLK_THROTTLE_TYPE_MAX];
948};
949
950/**
951 * struct hl_info_energy - device energy information
952 * @total_energy_consumption: total device energy consumption
953 */
954struct hl_info_energy {
955 __u64 total_energy_consumption;
956};
957
958#define HL_PLL_NUM_OUTPUTS 4
959
960struct hl_pll_frequency_info {
961 __u16 output[HL_PLL_NUM_OUTPUTS];
962};
963
964/**
965 * struct hl_open_stats_info - device open statistics information
966 * @open_counter: ever growing counter, increased on each successful dev open
967 * @last_open_period_ms: duration (ms) device was open last time
968 * @is_compute_ctx_active: Whether there is an active compute context executing
969 * @compute_ctx_in_release: true if the current compute context is being released
970 */
971struct hl_open_stats_info {
972 __u64 open_counter;
973 __u64 last_open_period_ms;
974 __u8 is_compute_ctx_active;
975 __u8 compute_ctx_in_release;
976 __u8 pad[6];
977};
978
979/**
980 * struct hl_power_info - power information
981 * @power: power consumption
982 */
983struct hl_power_info {
984 __u64 power;
985};
986
987/**
988 * struct hl_info_sync_manager - sync manager information
989 * @first_available_sync_object: first available sob
990 * @first_available_monitor: first available monitor
991 * @first_available_cq: first available cq
992 */
993struct hl_info_sync_manager {
994 __u32 first_available_sync_object;
995 __u32 first_available_monitor;
996 __u32 first_available_cq;
997 __u32 reserved;
998};
999
1000/**
1001 * struct hl_info_cs_counters - command submission counters
1002 * @total_out_of_mem_drop_cnt: total dropped due to memory allocation issue
1003 * @ctx_out_of_mem_drop_cnt: context dropped due to memory allocation issue
1004 * @total_parsing_drop_cnt: total dropped due to error in packet parsing
1005 * @ctx_parsing_drop_cnt: context dropped due to error in packet parsing
1006 * @total_queue_full_drop_cnt: total dropped due to queue full
1007 * @ctx_queue_full_drop_cnt: context dropped due to queue full
1008 * @total_device_in_reset_drop_cnt: total dropped due to device in reset
1009 * @ctx_device_in_reset_drop_cnt: context dropped due to device in reset
1010 * @total_max_cs_in_flight_drop_cnt: total dropped due to maximum CS in-flight
1011 * @ctx_max_cs_in_flight_drop_cnt: context dropped due to maximum CS in-flight
1012 * @total_validation_drop_cnt: total dropped due to validation error
1013 * @ctx_validation_drop_cnt: context dropped due to validation error
1014 */
1015struct hl_info_cs_counters {
1016 __u64 total_out_of_mem_drop_cnt;
1017 __u64 ctx_out_of_mem_drop_cnt;
1018 __u64 total_parsing_drop_cnt;
1019 __u64 ctx_parsing_drop_cnt;
1020 __u64 total_queue_full_drop_cnt;
1021 __u64 ctx_queue_full_drop_cnt;
1022 __u64 total_device_in_reset_drop_cnt;
1023 __u64 ctx_device_in_reset_drop_cnt;
1024 __u64 total_max_cs_in_flight_drop_cnt;
1025 __u64 ctx_max_cs_in_flight_drop_cnt;
1026 __u64 total_validation_drop_cnt;
1027 __u64 ctx_validation_drop_cnt;
1028};
1029
1030/**
1031 * struct hl_info_last_err_open_dev_time - last error boot information.
1032 * @timestamp: timestamp of last time the device was opened and error occurred.
1033 */
1034struct hl_info_last_err_open_dev_time {
1035 __s64 timestamp;
1036};
1037
1038/**
1039 * struct hl_info_cs_timeout_event - last CS timeout information.
1040 * @timestamp: timestamp when last CS timeout event occurred.
1041 * @seq: sequence number of last CS timeout event.
1042 */
1043struct hl_info_cs_timeout_event {
1044 __s64 timestamp;
1045 __u64 seq;
1046};
1047
1048#define HL_RAZWI_PAGE_FAULT 0
1049#define HL_RAZWI_MMU_ACCESS_ERROR 1
1050
1051/**
1052 * struct hl_info_razwi_event - razwi information.
1053 * @timestamp: timestamp of razwi.
1054 * @addr: address which accessing it caused razwi.
1055 * @engine_id_1: engine id of the razwi initiator, if it was initiated by engine that does not
1056 * have engine id it will be set to U16_MAX.
1057 * @engine_id_2: second engine id of razwi initiator. Might happen that razwi have 2 possible
1058 * engines which one them caused the razwi. In that case, it will contain the
1059 * second possible engine id, otherwise it will be set to U16_MAX.
1060 * @no_engine_id: if razwi initiator does not have engine id, this field will be set to 1,
1061 * otherwise 0.
1062 * @error_type: cause of razwi, page fault or access error, otherwise it will be set to U8_MAX.
1063 * @pad: padding to 64 bit.
1064 */
1065struct hl_info_razwi_event {
1066 __s64 timestamp;
1067 __u64 addr;
1068 __u16 engine_id_1;
1069 __u16 engine_id_2;
1070 __u8 no_engine_id;
1071 __u8 error_type;
1072 __u8 pad[2];
1073};
1074
1075#define MAX_QMAN_STREAMS_INFO 4
1076#define OPCODE_INFO_MAX_ADDR_SIZE 8
1077/**
1078 * struct hl_info_undefined_opcode_event - info about last undefined opcode error
1079 * @timestamp: timestamp of the undefined opcode error
1080 * @cb_addr_streams: CB addresses (per stream) that are currently exists in the PQ
1081 * entiers. In case all streams array entries are
1082 * filled with values, it means the execution was in Lower-CP.
1083 * @cq_addr: the address of the current handled command buffer
1084 * @cq_size: the size of the current handled command buffer
1085 * @cb_addr_streams_len: num of streams - actual len of cb_addr_streams array.
1086 * should be equal to 1 incase of undefined opcode
1087 * in Upper-CP (specific stream) and equal to 4 incase
1088 * of undefined opcode in Lower-CP.
1089 * @engine_id: engine-id that the error occurred on
1090 * @stream_id: the stream id the error occurred on. In case the stream equals to
1091 * MAX_QMAN_STREAMS_INFO it means the error occurred on a Lower-CP.
1092 */
1093struct hl_info_undefined_opcode_event {
1094 __s64 timestamp;
1095 __u64 cb_addr_streams[MAX_QMAN_STREAMS_INFO][OPCODE_INFO_MAX_ADDR_SIZE];
1096 __u64 cq_addr;
1097 __u32 cq_size;
1098 __u32 cb_addr_streams_len;
1099 __u32 engine_id;
1100 __u32 stream_id;
1101};
1102
1103/**
1104 * struct hl_info_dev_memalloc_page_sizes - valid page sizes in device mem alloc information.
1105 * @page_order_bitmask: bitmap in which a set bit represents the order of the supported page size
1106 * (e.g. 0x2100000 means that 1MB and 32MB pages are supported).
1107 */
1108struct hl_info_dev_memalloc_page_sizes {
1109 __u64 page_order_bitmask;
1110};
1111
1112enum gaudi_dcores {
1113 HL_GAUDI_WS_DCORE,
1114 HL_GAUDI_WN_DCORE,
1115 HL_GAUDI_EN_DCORE,
1116 HL_GAUDI_ES_DCORE
1117};
1118
1119/**
1120 * struct hl_info_args - Main structure to retrieve device related information.
1121 * @return_pointer: User space address of the relevant structure related to HL_INFO_* operation
1122 * mentioned in @op.
1123 * @return_size: Size of the structure used in @return_pointer, just like "size" in "snprintf", it
1124 * limits how many bytes the kernel can write. For hw_events array, the size should be
1125 * hl_info_hw_ip_info.num_of_events * sizeof(__u32).
1126 * @op: Defines which type of information to be retrieved. Refer HL_INFO_* for details.
1127 * @dcore_id: DCORE id for which the information is relevant (for Gaudi refer to enum gaudi_dcores).
1128 * @ctx_id: Context ID of the user. Currently not in use.
1129 * @period_ms: Period value, in milliseconds, for utilization rate in range 100ms - 1000ms in 100 ms
1130 * resolution. Currently not in use.
1131 * @pll_index: Index as defined in hl_<asic type>_pll_index enumeration.
1132 * @eventfd: event file descriptor for event notifications.
1133 * @pad: Padding to 64 bit.
1134 */
1135struct hl_info_args {
1136 __u64 return_pointer;
1137 __u32 return_size;
1138 __u32 op;
1139
1140 union {
1141 __u32 dcore_id;
1142 __u32 ctx_id;
1143 __u32 period_ms;
1144 __u32 pll_index;
1145 __u32 eventfd;
1146 };
1147
1148 __u32 pad;
1149};
1150
1151/* Opcode to create a new command buffer */
1152#define HL_CB_OP_CREATE 0
1153/* Opcode to destroy previously created command buffer */
1154#define HL_CB_OP_DESTROY 1
1155/* Opcode to retrieve information about a command buffer */
1156#define HL_CB_OP_INFO 2
1157
1158/* 2MB minus 32 bytes for 2xMSG_PROT */
1159#define HL_MAX_CB_SIZE (0x200000 - 32)
1160
1161/* Indicates whether the command buffer should be mapped to the device's MMU */
1162#define HL_CB_FLAGS_MAP 0x1
1163
1164/* Used with HL_CB_OP_INFO opcode to get the device va address for kernel mapped CB */
1165#define HL_CB_FLAGS_GET_DEVICE_VA 0x2
1166
1167struct hl_cb_in {
1168 /* Handle of CB or 0 if we want to create one */
1169 __u64 cb_handle;
1170 /* HL_CB_OP_* */
1171 __u32 op;
1172
1173 /* Size of CB. Maximum size is HL_MAX_CB_SIZE. The minimum size that
1174 * will be allocated, regardless of this parameter's value, is PAGE_SIZE
1175 */
1176 __u32 cb_size;
1177
1178 /* Context ID - Currently not in use */
1179 __u32 ctx_id;
1180 /* HL_CB_FLAGS_* */
1181 __u32 flags;
1182};
1183
1184struct hl_cb_out {
1185 union {
1186 /* Handle of CB */
1187 __u64 cb_handle;
1188
1189 union {
1190 /* Information about CB */
1191 struct {
1192 /* Usage count of CB */
1193 __u32 usage_cnt;
1194 __u32 pad;
1195 };
1196
1197 /* CB mapped address to device MMU */
1198 __u64 device_va;
1199 };
1200 };
1201};
1202
1203union hl_cb_args {
1204 struct hl_cb_in in;
1205 struct hl_cb_out out;
1206};
1207
1208/* HL_CS_CHUNK_FLAGS_ values
1209 *
1210 * HL_CS_CHUNK_FLAGS_USER_ALLOC_CB:
1211 * Indicates if the CB was allocated and mapped by userspace
1212 * (relevant to greco and above). User allocated CB is a command buffer,
1213 * allocated by the user, via malloc (or similar). After allocating the
1214 * CB, the user invokes - “memory ioctl” to map the user memory into a
1215 * device virtual address. The user provides this address via the
1216 * cb_handle field. The interface provides the ability to create a
1217 * large CBs, Which aren’t limited to “HL_MAX_CB_SIZE”. Therefore, it
1218 * increases the PCI-DMA queues throughput. This CB allocation method
1219 * also reduces the use of Linux DMA-able memory pool. Which are limited
1220 * and used by other Linux sub-systems.
1221 */
1222#define HL_CS_CHUNK_FLAGS_USER_ALLOC_CB 0x1
1223
1224/*
1225 * This structure size must always be fixed to 64-bytes for backward
1226 * compatibility
1227 */
1228struct hl_cs_chunk {
1229 union {
1230 /* Goya/Gaudi:
1231 * For external queue, this represents a Handle of CB on the
1232 * Host.
1233 * For internal queue in Goya, this represents an SRAM or
1234 * a DRAM address of the internal CB. In Gaudi, this might also
1235 * represent a mapped host address of the CB.
1236 *
1237 * Greco onwards:
1238 * For H/W queue, this represents either a Handle of CB on the
1239 * Host, or an SRAM, a DRAM, or a mapped host address of the CB.
1240 *
1241 * A mapped host address is in the device address space, after
1242 * a host address was mapped by the device MMU.
1243 */
1244 __u64 cb_handle;
1245
1246 /* Relevant only when HL_CS_FLAGS_WAIT or
1247 * HL_CS_FLAGS_COLLECTIVE_WAIT is set
1248 * This holds address of array of u64 values that contain
1249 * signal CS sequence numbers. The wait described by
1250 * this job will listen on all those signals
1251 * (wait event per signal)
1252 */
1253 __u64 signal_seq_arr;
1254
1255 /*
1256 * Relevant only when HL_CS_FLAGS_WAIT or
1257 * HL_CS_FLAGS_COLLECTIVE_WAIT is set
1258 * along with HL_CS_FLAGS_ENCAP_SIGNALS.
1259 * This is the CS sequence which has the encapsulated signals.
1260 */
1261 __u64 encaps_signal_seq;
1262 };
1263
1264 /* Index of queue to put the CB on */
1265 __u32 queue_index;
1266
1267 union {
1268 /*
1269 * Size of command buffer with valid packets
1270 * Can be smaller then actual CB size
1271 */
1272 __u32 cb_size;
1273
1274 /* Relevant only when HL_CS_FLAGS_WAIT or
1275 * HL_CS_FLAGS_COLLECTIVE_WAIT is set.
1276 * Number of entries in signal_seq_arr
1277 */
1278 __u32 num_signal_seq_arr;
1279
1280 /* Relevant only when HL_CS_FLAGS_WAIT or
1281 * HL_CS_FLAGS_COLLECTIVE_WAIT is set along
1282 * with HL_CS_FLAGS_ENCAP_SIGNALS
1283 * This set the signals range that the user want to wait for
1284 * out of the whole reserved signals range.
1285 * e.g if the signals range is 20, and user don't want
1286 * to wait for signal 8, so he set this offset to 7, then
1287 * he call the API again with 9 and so on till 20.
1288 */
1289 __u32 encaps_signal_offset;
1290 };
1291
1292 /* HL_CS_CHUNK_FLAGS_* */
1293 __u32 cs_chunk_flags;
1294
1295 /* Relevant only when HL_CS_FLAGS_COLLECTIVE_WAIT is set.
1296 * This holds the collective engine ID. The wait described by this job
1297 * will sync with this engine and with all NICs before completion.
1298 */
1299 __u32 collective_engine_id;
1300
1301 /* Align structure to 64 bytes */
1302 __u32 pad[10];
1303};
1304
1305/* SIGNAL/WAIT/COLLECTIVE_WAIT flags are mutually exclusive */
1306#define HL_CS_FLAGS_FORCE_RESTORE 0x1
1307#define HL_CS_FLAGS_SIGNAL 0x2
1308#define HL_CS_FLAGS_WAIT 0x4
1309#define HL_CS_FLAGS_COLLECTIVE_WAIT 0x8
1310
1311#define HL_CS_FLAGS_TIMESTAMP 0x20
1312#define HL_CS_FLAGS_STAGED_SUBMISSION 0x40
1313#define HL_CS_FLAGS_STAGED_SUBMISSION_FIRST 0x80
1314#define HL_CS_FLAGS_STAGED_SUBMISSION_LAST 0x100
1315#define HL_CS_FLAGS_CUSTOM_TIMEOUT 0x200
1316#define HL_CS_FLAGS_SKIP_RESET_ON_TIMEOUT 0x400
1317
1318/*
1319 * The encapsulated signals CS is merged into the existing CS ioctls.
1320 * In order to use this feature need to follow the below procedure:
1321 * 1. Reserve signals, set the CS type to HL_CS_FLAGS_RESERVE_SIGNALS_ONLY
1322 * the output of this API will be the SOB offset from CFG_BASE.
1323 * this address will be used to patch CB cmds to do the signaling for this
1324 * SOB by incrementing it's value.
1325 * for reverting the reservation use HL_CS_FLAGS_UNRESERVE_SIGNALS_ONLY
1326 * CS type, note that this might fail if out-of-sync happened to the SOB
1327 * value, in case other signaling request to the same SOB occurred between
1328 * reserve-unreserve calls.
1329 * 2. Use the staged CS to do the encapsulated signaling jobs.
1330 * use HL_CS_FLAGS_STAGED_SUBMISSION and HL_CS_FLAGS_STAGED_SUBMISSION_FIRST
1331 * along with HL_CS_FLAGS_ENCAP_SIGNALS flag, and set encaps_signal_offset
1332 * field. This offset allows app to wait on part of the reserved signals.
1333 * 3. Use WAIT/COLLECTIVE WAIT CS along with HL_CS_FLAGS_ENCAP_SIGNALS flag
1334 * to wait for the encapsulated signals.
1335 */
1336#define HL_CS_FLAGS_ENCAP_SIGNALS 0x800
1337#define HL_CS_FLAGS_RESERVE_SIGNALS_ONLY 0x1000
1338#define HL_CS_FLAGS_UNRESERVE_SIGNALS_ONLY 0x2000
1339
1340#define HL_CS_STATUS_SUCCESS 0
1341
1342#define HL_MAX_JOBS_PER_CS 512
1343
1344struct hl_cs_in {
1345
1346 /* this holds address of array of hl_cs_chunk for restore phase */
1347 __u64 chunks_restore;
1348
1349 /* holds address of array of hl_cs_chunk for execution phase */
1350 __u64 chunks_execute;
1351
1352 union {
1353 /*
1354 * Sequence number of a staged submission CS
1355 * valid only if HL_CS_FLAGS_STAGED_SUBMISSION is set and
1356 * HL_CS_FLAGS_STAGED_SUBMISSION_FIRST is unset.
1357 */
1358 __u64 seq;
1359
1360 /*
1361 * Encapsulated signals handle id
1362 * Valid for two flows:
1363 * 1. CS with encapsulated signals:
1364 * when HL_CS_FLAGS_STAGED_SUBMISSION and
1365 * HL_CS_FLAGS_STAGED_SUBMISSION_FIRST
1366 * and HL_CS_FLAGS_ENCAP_SIGNALS are set.
1367 * 2. unreserve signals:
1368 * valid when HL_CS_FLAGS_UNRESERVE_SIGNALS_ONLY is set.
1369 */
1370 __u32 encaps_sig_handle_id;
1371
1372 /* Valid only when HL_CS_FLAGS_RESERVE_SIGNALS_ONLY is set */
1373 struct {
1374 /* Encapsulated signals number */
1375 __u32 encaps_signals_count;
1376
1377 /* Encapsulated signals queue index (stream) */
1378 __u32 encaps_signals_q_idx;
1379 };
1380 };
1381
1382 /* Number of chunks in restore phase array. Maximum number is
1383 * HL_MAX_JOBS_PER_CS
1384 */
1385 __u32 num_chunks_restore;
1386
1387 /* Number of chunks in execution array. Maximum number is
1388 * HL_MAX_JOBS_PER_CS
1389 */
1390 __u32 num_chunks_execute;
1391
1392 /* timeout in seconds - valid only if HL_CS_FLAGS_CUSTOM_TIMEOUT
1393 * is set
1394 */
1395 __u32 timeout;
1396
1397 /* HL_CS_FLAGS_* */
1398 __u32 cs_flags;
1399
1400 /* Context ID - Currently not in use */
1401 __u32 ctx_id;
1402 __u8 pad[4];
1403};
1404
1405struct hl_cs_out {
1406 union {
1407 /*
1408 * seq holds the sequence number of the CS to pass to wait
1409 * ioctl. All values are valid except for 0 and ULLONG_MAX
1410 */
1411 __u64 seq;
1412
1413 /* Valid only when HL_CS_FLAGS_RESERVE_SIGNALS_ONLY is set */
1414 struct {
1415 /* This is the resereved signal handle id */
1416 __u32 handle_id;
1417
1418 /* This is the signals count */
1419 __u32 count;
1420 };
1421 };
1422
1423 /* HL_CS_STATUS */
1424 __u32 status;
1425
1426 /*
1427 * SOB base address offset
1428 * Valid only when HL_CS_FLAGS_RESERVE_SIGNALS_ONLY or HL_CS_FLAGS_SIGNAL is set
1429 */
1430 __u32 sob_base_addr_offset;
1431
1432 /*
1433 * Count of completed signals in SOB before current signal submission.
1434 * Valid only when (HL_CS_FLAGS_ENCAP_SIGNALS & HL_CS_FLAGS_STAGED_SUBMISSION)
1435 * or HL_CS_FLAGS_SIGNAL is set
1436 */
1437 __u16 sob_count_before_submission;
1438 __u16 pad[3];
1439};
1440
1441union hl_cs_args {
1442 struct hl_cs_in in;
1443 struct hl_cs_out out;
1444};
1445
1446#define HL_WAIT_CS_FLAGS_INTERRUPT 0x2
1447#define HL_WAIT_CS_FLAGS_INTERRUPT_MASK 0xFFF00000
1448#define HL_WAIT_CS_FLAGS_ANY_CQ_INTERRUPT 0xFFF00000
1449#define HL_WAIT_CS_FLAGS_ANY_DEC_INTERRUPT 0xFFE00000
1450#define HL_WAIT_CS_FLAGS_MULTI_CS 0x4
1451#define HL_WAIT_CS_FLAGS_INTERRUPT_KERNEL_CQ 0x10
1452#define HL_WAIT_CS_FLAGS_REGISTER_INTERRUPT 0x20
1453
1454#define HL_WAIT_MULTI_CS_LIST_MAX_LEN 32
1455
1456struct hl_wait_cs_in {
1457 union {
1458 struct {
1459 /*
1460 * In case of wait_cs holds the CS sequence number.
1461 * In case of wait for multi CS hold a user pointer to
1462 * an array of CS sequence numbers
1463 */
1464 __u64 seq;
1465 /* Absolute timeout to wait for command submission
1466 * in microseconds
1467 */
1468 __u64 timeout_us;
1469 };
1470
1471 struct {
1472 union {
1473 /* User address for completion comparison.
1474 * upon interrupt, driver will compare the value pointed
1475 * by this address with the supplied target value.
1476 * in order not to perform any comparison, set address
1477 * to all 1s.
1478 * Relevant only when HL_WAIT_CS_FLAGS_INTERRUPT is set
1479 */
1480 __u64 addr;
1481
1482 /* cq_counters_handle to a kernel mapped cb which contains
1483 * cq counters.
1484 * Relevant only when HL_WAIT_CS_FLAGS_INTERRUPT_KERNEL_CQ is set
1485 */
1486 __u64 cq_counters_handle;
1487 };
1488
1489 /* Target value for completion comparison */
1490 __u64 target;
1491 };
1492 };
1493
1494 /* Context ID - Currently not in use */
1495 __u32 ctx_id;
1496
1497 /* HL_WAIT_CS_FLAGS_*
1498 * If HL_WAIT_CS_FLAGS_INTERRUPT is set, this field should include
1499 * interrupt id according to HL_WAIT_CS_FLAGS_INTERRUPT_MASK
1500 *
1501 * in order to wait for any CQ interrupt, set interrupt value to
1502 * HL_WAIT_CS_FLAGS_ANY_CQ_INTERRUPT.
1503 *
1504 * in order to wait for any decoder interrupt, set interrupt value to
1505 * HL_WAIT_CS_FLAGS_ANY_DEC_INTERRUPT.
1506 */
1507 __u32 flags;
1508
1509 union {
1510 struct {
1511 /* Multi CS API info- valid entries in multi-CS array */
1512 __u8 seq_arr_len;
1513 __u8 pad[7];
1514 };
1515
1516 /* Absolute timeout to wait for an interrupt in microseconds.
1517 * Relevant only when HL_WAIT_CS_FLAGS_INTERRUPT is set
1518 */
1519 __u64 interrupt_timeout_us;
1520 };
1521
1522 /*
1523 * cq counter offset inside the counters cb pointed by cq_counters_handle above.
1524 * upon interrupt, driver will compare the value pointed
1525 * by this address (cq_counters_handle + cq_counters_offset)
1526 * with the supplied target value.
1527 * relevant only when HL_WAIT_CS_FLAGS_INTERRUPT_KERNEL_CQ is set
1528 */
1529 __u64 cq_counters_offset;
1530
1531 /*
1532 * Timestamp_handle timestamps buffer handle.
1533 * relevant only when HL_WAIT_CS_FLAGS_REGISTER_INTERRUPT is set
1534 */
1535 __u64 timestamp_handle;
1536
1537 /*
1538 * Timestamp_offset is offset inside the timestamp buffer pointed by timestamp_handle above.
1539 * upon interrupt, if the cq reached the target value then driver will write
1540 * timestamp to this offset.
1541 * relevant only when HL_WAIT_CS_FLAGS_REGISTER_INTERRUPT is set
1542 */
1543 __u64 timestamp_offset;
1544};
1545
1546#define HL_WAIT_CS_STATUS_COMPLETED 0
1547#define HL_WAIT_CS_STATUS_BUSY 1
1548#define HL_WAIT_CS_STATUS_TIMEDOUT 2
1549#define HL_WAIT_CS_STATUS_ABORTED 3
1550
1551#define HL_WAIT_CS_STATUS_FLAG_GONE 0x1
1552#define HL_WAIT_CS_STATUS_FLAG_TIMESTAMP_VLD 0x2
1553
1554struct hl_wait_cs_out {
1555 /* HL_WAIT_CS_STATUS_* */
1556 __u32 status;
1557 /* HL_WAIT_CS_STATUS_FLAG* */
1558 __u32 flags;
1559 /*
1560 * valid only if HL_WAIT_CS_STATUS_FLAG_TIMESTAMP_VLD is set
1561 * for wait_cs: timestamp of CS completion
1562 * for wait_multi_cs: timestamp of FIRST CS completion
1563 */
1564 __s64 timestamp_nsec;
1565 /* multi CS completion bitmap */
1566 __u32 cs_completion_map;
1567 __u32 pad;
1568};
1569
1570union hl_wait_cs_args {
1571 struct hl_wait_cs_in in;
1572 struct hl_wait_cs_out out;
1573};
1574
1575/* Opcode to allocate device memory */
1576#define HL_MEM_OP_ALLOC 0
1577
1578/* Opcode to free previously allocated device memory */
1579#define HL_MEM_OP_FREE 1
1580
1581/* Opcode to map host and device memory */
1582#define HL_MEM_OP_MAP 2
1583
1584/* Opcode to unmap previously mapped host and device memory */
1585#define HL_MEM_OP_UNMAP 3
1586
1587/* Opcode to map a hw block */
1588#define HL_MEM_OP_MAP_BLOCK 4
1589
1590/* Opcode to create DMA-BUF object for an existing device memory allocation
1591 * and to export an FD of that DMA-BUF back to the caller
1592 */
1593#define HL_MEM_OP_EXPORT_DMABUF_FD 5
1594
1595/* Opcode to create timestamps pool for user interrupts registration support
1596 * The memory will be allocated by the kernel driver, A timestamp buffer which the user
1597 * will get handle to it for mmap, and another internal buffer used by the
1598 * driver for registration management
1599 * The memory will be freed when the user closes the file descriptor(ctx close)
1600 */
1601#define HL_MEM_OP_TS_ALLOC 6
1602
1603/* Memory flags */
1604#define HL_MEM_CONTIGUOUS 0x1
1605#define HL_MEM_SHARED 0x2
1606#define HL_MEM_USERPTR 0x4
1607#define HL_MEM_FORCE_HINT 0x8
1608#define HL_MEM_PREFETCH 0x40
1609
1610/**
1611 * structure hl_mem_in - structure that handle input args for memory IOCTL
1612 * @union arg: union of structures to be used based on the input operation
1613 * @op: specify the requested memory operation (one of the HL_MEM_OP_* definitions).
1614 * @flags: flags for the memory operation (one of the HL_MEM_* definitions).
1615 * For the HL_MEM_OP_EXPORT_DMABUF_FD opcode, this field holds the DMA-BUF file/FD flags.
1616 * @ctx_id: context ID - currently not in use.
1617 * @num_of_elements: number of timestamp elements used only with HL_MEM_OP_TS_ALLOC opcode.
1618 */
1619struct hl_mem_in {
1620 union {
1621 /**
1622 * structure for device memory allocation (used with the HL_MEM_OP_ALLOC op)
1623 * @mem_size: memory size to allocate
1624 * @page_size: page size to use on allocation. when the value is 0 the default page
1625 * size will be taken.
1626 */
1627 struct {
1628 __u64 mem_size;
1629 __u64 page_size;
1630 } alloc;
1631
1632 /**
1633 * structure for free-ing device memory (used with the HL_MEM_OP_FREE op)
1634 * @handle: handle returned from HL_MEM_OP_ALLOC
1635 */
1636 struct {
1637 __u64 handle;
1638 } free;
1639
1640 /**
1641 * structure for mapping device memory (used with the HL_MEM_OP_MAP op)
1642 * @hint_addr: requested virtual address of mapped memory.
1643 * the driver will try to map the requested region to this hint
1644 * address, as long as the address is valid and not already mapped.
1645 * the user should check the returned address of the IOCTL to make
1646 * sure he got the hint address.
1647 * passing 0 here means that the driver will choose the address itself.
1648 * @handle: handle returned from HL_MEM_OP_ALLOC.
1649 */
1650 struct {
1651 __u64 hint_addr;
1652 __u64 handle;
1653 } map_device;
1654
1655 /**
1656 * structure for mapping host memory (used with the HL_MEM_OP_MAP op)
1657 * @host_virt_addr: address of allocated host memory.
1658 * @hint_addr: requested virtual address of mapped memory.
1659 * the driver will try to map the requested region to this hint
1660 * address, as long as the address is valid and not already mapped.
1661 * the user should check the returned address of the IOCTL to make
1662 * sure he got the hint address.
1663 * passing 0 here means that the driver will choose the address itself.
1664 * @size: size of allocated host memory.
1665 */
1666 struct {
1667 __u64 host_virt_addr;
1668 __u64 hint_addr;
1669 __u64 mem_size;
1670 } map_host;
1671
1672 /**
1673 * structure for mapping hw block (used with the HL_MEM_OP_MAP_BLOCK op)
1674 * @block_addr:HW block address to map, a handle and size will be returned
1675 * to the user and will be used to mmap the relevant block.
1676 * only addresses from configuration space are allowed.
1677 */
1678 struct {
1679 __u64 block_addr;
1680 } map_block;
1681
1682 /**
1683 * structure for unmapping host memory (used with the HL_MEM_OP_UNMAP op)
1684 * @device_virt_addr: virtual address returned from HL_MEM_OP_MAP
1685 */
1686 struct {
1687 __u64 device_virt_addr;
1688 } unmap;
1689
1690 /**
1691 * structure for exporting DMABUF object (used with
1692 * the HL_MEM_OP_EXPORT_DMABUF_FD op)
1693 * @handle: handle returned from HL_MEM_OP_ALLOC.
1694 * in Gaudi, where we don't have MMU for the device memory, the
1695 * driver expects a physical address (instead of a handle) in the
1696 * device memory space.
1697 * @mem_size: size of memory allocation. Relevant only for GAUDI
1698 */
1699 struct {
1700 __u64 handle;
1701 __u64 mem_size;
1702 } export_dmabuf_fd;
1703 };
1704
1705 __u32 op;
1706 __u32 flags;
1707 __u32 ctx_id;
1708 __u32 num_of_elements;
1709};
1710
1711struct hl_mem_out {
1712 union {
1713 /*
1714 * Used for HL_MEM_OP_MAP as the virtual address that was
1715 * assigned in the device VA space.
1716 * A value of 0 means the requested operation failed.
1717 */
1718 __u64 device_virt_addr;
1719
1720 /*
1721 * Used in HL_MEM_OP_ALLOC
1722 * This is the assigned handle for the allocated memory
1723 */
1724 __u64 handle;
1725
1726 struct {
1727 /*
1728 * Used in HL_MEM_OP_MAP_BLOCK.
1729 * This is the assigned handle for the mapped block
1730 */
1731 __u64 block_handle;
1732
1733 /*
1734 * Used in HL_MEM_OP_MAP_BLOCK
1735 * This is the size of the mapped block
1736 */
1737 __u32 block_size;
1738
1739 __u32 pad;
1740 };
1741
1742 /* Returned in HL_MEM_OP_EXPORT_DMABUF_FD. Represents the
1743 * DMA-BUF object that was created to describe a memory
1744 * allocation on the device's memory space. The FD should be
1745 * passed to the importer driver
1746 */
1747 __s32 fd;
1748 };
1749};
1750
1751union hl_mem_args {
1752 struct hl_mem_in in;
1753 struct hl_mem_out out;
1754};
1755
1756#define HL_DEBUG_MAX_AUX_VALUES 10
1757
1758struct hl_debug_params_etr {
1759 /* Address in memory to allocate buffer */
1760 __u64 buffer_address;
1761
1762 /* Size of buffer to allocate */
1763 __u64 buffer_size;
1764
1765 /* Sink operation mode: SW fifo, HW fifo, Circular buffer */
1766 __u32 sink_mode;
1767 __u32 pad;
1768};
1769
1770struct hl_debug_params_etf {
1771 /* Address in memory to allocate buffer */
1772 __u64 buffer_address;
1773
1774 /* Size of buffer to allocate */
1775 __u64 buffer_size;
1776
1777 /* Sink operation mode: SW fifo, HW fifo, Circular buffer */
1778 __u32 sink_mode;
1779 __u32 pad;
1780};
1781
1782struct hl_debug_params_stm {
1783 /* Two bit masks for HW event and Stimulus Port */
1784 __u64 he_mask;
1785 __u64 sp_mask;
1786
1787 /* Trace source ID */
1788 __u32 id;
1789
1790 /* Frequency for the timestamp register */
1791 __u32 frequency;
1792};
1793
1794struct hl_debug_params_bmon {
1795 /* Two address ranges that the user can request to filter */
1796 __u64 start_addr0;
1797 __u64 addr_mask0;
1798
1799 __u64 start_addr1;
1800 __u64 addr_mask1;
1801
1802 /* Capture window configuration */
1803 __u32 bw_win;
1804 __u32 win_capture;
1805
1806 /* Trace source ID */
1807 __u32 id;
1808
1809 /* Control register */
1810 __u32 control;
1811
1812 /* Two more address ranges that the user can request to filter */
1813 __u64 start_addr2;
1814 __u64 end_addr2;
1815
1816 __u64 start_addr3;
1817 __u64 end_addr3;
1818};
1819
1820struct hl_debug_params_spmu {
1821 /* Event types selection */
1822 __u64 event_types[HL_DEBUG_MAX_AUX_VALUES];
1823
1824 /* Number of event types selection */
1825 __u32 event_types_num;
1826
1827 /* TRC configuration register values */
1828 __u32 pmtrc_val;
1829 __u32 trc_ctrl_host_val;
1830 __u32 trc_en_host_val;
1831};
1832
1833/* Opcode for ETR component */
1834#define HL_DEBUG_OP_ETR 0
1835/* Opcode for ETF component */
1836#define HL_DEBUG_OP_ETF 1
1837/* Opcode for STM component */
1838#define HL_DEBUG_OP_STM 2
1839/* Opcode for FUNNEL component */
1840#define HL_DEBUG_OP_FUNNEL 3
1841/* Opcode for BMON component */
1842#define HL_DEBUG_OP_BMON 4
1843/* Opcode for SPMU component */
1844#define HL_DEBUG_OP_SPMU 5
1845/* Opcode for timestamp (deprecated) */
1846#define HL_DEBUG_OP_TIMESTAMP 6
1847/* Opcode for setting the device into or out of debug mode. The enable
1848 * variable should be 1 for enabling debug mode and 0 for disabling it
1849 */
1850#define HL_DEBUG_OP_SET_MODE 7
1851
1852struct hl_debug_args {
1853 /*
1854 * Pointer to user input structure.
1855 * This field is relevant to specific opcodes.
1856 */
1857 __u64 input_ptr;
1858 /* Pointer to user output structure */
1859 __u64 output_ptr;
1860 /* Size of user input structure */
1861 __u32 input_size;
1862 /* Size of user output structure */
1863 __u32 output_size;
1864 /* HL_DEBUG_OP_* */
1865 __u32 op;
1866 /*
1867 * Register index in the component, taken from the debug_regs_index enum
1868 * in the various ASIC header files
1869 */
1870 __u32 reg_idx;
1871 /* Enable/disable */
1872 __u32 enable;
1873 /* Context ID - Currently not in use */
1874 __u32 ctx_id;
1875};
1876
1877/*
1878 * Notifier event values - for the notification mechanism and the HL_INFO_GET_EVENTS command
1879 *
1880 * HL_NOTIFIER_EVENT_TPC_ASSERT - Indicates TPC assert event
1881 * HL_NOTIFIER_EVENT_UNDEFINED_OPCODE - Indicates undefined operation code
1882 * HL_NOTIFIER_EVENT_DEVICE_RESET - Indicates device requires a reset
1883 * HL_NOTIFIER_EVENT_CS_TIMEOUT - Indicates CS timeout error
1884 * HL_NOTIFIER_EVENT_DEVICE_UNAVAILABLE - Indicates device is unavailable
1885 */
1886#define HL_NOTIFIER_EVENT_TPC_ASSERT (1ULL << 0)
1887#define HL_NOTIFIER_EVENT_UNDEFINED_OPCODE (1ULL << 1)
1888#define HL_NOTIFIER_EVENT_DEVICE_RESET (1ULL << 2)
1889#define HL_NOTIFIER_EVENT_CS_TIMEOUT (1ULL << 3)
1890#define HL_NOTIFIER_EVENT_DEVICE_UNAVAILABLE (1ULL << 4)
1891
1892/*
1893 * Various information operations such as:
1894 * - H/W IP information
1895 * - Current dram usage
1896 *
1897 * The user calls this IOCTL with an opcode that describes the required
1898 * information. The user should supply a pointer to a user-allocated memory
1899 * chunk, which will be filled by the driver with the requested information.
1900 *
1901 * The user supplies the maximum amount of size to copy into the user's memory,
1902 * in order to prevent data corruption in case of differences between the
1903 * definitions of structures in kernel and userspace, e.g. in case of old
1904 * userspace and new kernel driver
1905 */
1906#define HL_IOCTL_INFO \
1907 _IOWR('H', 0x01, struct hl_info_args)
1908
1909/*
1910 * Command Buffer
1911 * - Request a Command Buffer
1912 * - Destroy a Command Buffer
1913 *
1914 * The command buffers are memory blocks that reside in DMA-able address
1915 * space and are physically contiguous so they can be accessed by the device
1916 * directly. They are allocated using the coherent DMA API.
1917 *
1918 * When creating a new CB, the IOCTL returns a handle of it, and the user-space
1919 * process needs to use that handle to mmap the buffer so it can access them.
1920 *
1921 * In some instances, the device must access the command buffer through the
1922 * device's MMU, and thus its memory should be mapped. In these cases, user can
1923 * indicate the driver that such a mapping is required.
1924 * The resulting device virtual address will be used internally by the driver,
1925 * and won't be returned to user.
1926 *
1927 */
1928#define HL_IOCTL_CB \
1929 _IOWR('H', 0x02, union hl_cb_args)
1930
1931/*
1932 * Command Submission
1933 *
1934 * To submit work to the device, the user need to call this IOCTL with a set
1935 * of JOBS. That set of JOBS constitutes a CS object.
1936 * Each JOB will be enqueued on a specific queue, according to the user's input.
1937 * There can be more then one JOB per queue.
1938 *
1939 * The CS IOCTL will receive two sets of JOBS. One set is for "restore" phase
1940 * and a second set is for "execution" phase.
1941 * The JOBS on the "restore" phase are enqueued only after context-switch
1942 * (or if its the first CS for this context). The user can also order the
1943 * driver to run the "restore" phase explicitly
1944 *
1945 * Goya/Gaudi:
1946 * There are two types of queues - external and internal. External queues
1947 * are DMA queues which transfer data from/to the Host. All other queues are
1948 * internal. The driver will get completion notifications from the device only
1949 * on JOBS which are enqueued in the external queues.
1950 *
1951 * Greco onwards:
1952 * There is a single type of queue for all types of engines, either DMA engines
1953 * for transfers from/to the host or inside the device, or compute engines.
1954 * The driver will get completion notifications from the device for all queues.
1955 *
1956 * For jobs on external queues, the user needs to create command buffers
1957 * through the CB ioctl and give the CB's handle to the CS ioctl. For jobs on
1958 * internal queues, the user needs to prepare a "command buffer" with packets
1959 * on either the device SRAM/DRAM or the host, and give the device address of
1960 * that buffer to the CS ioctl.
1961 * For jobs on H/W queues both options of command buffers are valid.
1962 *
1963 * This IOCTL is asynchronous in regard to the actual execution of the CS. This
1964 * means it returns immediately after ALL the JOBS were enqueued on their
1965 * relevant queues. Therefore, the user mustn't assume the CS has been completed
1966 * or has even started to execute.
1967 *
1968 * Upon successful enqueue, the IOCTL returns a sequence number which the user
1969 * can use with the "Wait for CS" IOCTL to check whether the handle's CS
1970 * non-internal JOBS have been completed. Note that if the CS has internal JOBS
1971 * which can execute AFTER the external JOBS have finished, the driver might
1972 * report that the CS has finished executing BEFORE the internal JOBS have
1973 * actually finished executing.
1974 *
1975 * Even though the sequence number increments per CS, the user can NOT
1976 * automatically assume that if CS with sequence number N finished, then CS
1977 * with sequence number N-1 also finished. The user can make this assumption if
1978 * and only if CS N and CS N-1 are exactly the same (same CBs for the same
1979 * queues).
1980 */
1981#define HL_IOCTL_CS \
1982 _IOWR('H', 0x03, union hl_cs_args)
1983
1984/*
1985 * Wait for Command Submission
1986 *
1987 * The user can call this IOCTL with a handle it received from the CS IOCTL
1988 * to wait until the handle's CS has finished executing. The user will wait
1989 * inside the kernel until the CS has finished or until the user-requested
1990 * timeout has expired.
1991 *
1992 * If the timeout value is 0, the driver won't sleep at all. It will check
1993 * the status of the CS and return immediately
1994 *
1995 * The return value of the IOCTL is a standard Linux error code. The possible
1996 * values are:
1997 *
1998 * EINTR - Kernel waiting has been interrupted, e.g. due to OS signal
1999 * that the user process received
2000 * ETIMEDOUT - The CS has caused a timeout on the device
2001 * EIO - The CS was aborted (usually because the device was reset)
2002 * ENODEV - The device wants to do hard-reset (so user need to close FD)
2003 *
2004 * The driver also returns a custom define in case the IOCTL call returned 0.
2005 * The define can be one of the following:
2006 *
2007 * HL_WAIT_CS_STATUS_COMPLETED - The CS has been completed successfully (0)
2008 * HL_WAIT_CS_STATUS_BUSY - The CS is still executing (0)
2009 * HL_WAIT_CS_STATUS_TIMEDOUT - The CS has caused a timeout on the device
2010 * (ETIMEDOUT)
2011 * HL_WAIT_CS_STATUS_ABORTED - The CS was aborted, usually because the
2012 * device was reset (EIO)
2013 */
2014
2015#define HL_IOCTL_WAIT_CS \
2016 _IOWR('H', 0x04, union hl_wait_cs_args)
2017
2018/*
2019 * Memory
2020 * - Map host memory to device MMU
2021 * - Unmap host memory from device MMU
2022 *
2023 * This IOCTL allows the user to map host memory to the device MMU
2024 *
2025 * For host memory, the IOCTL doesn't allocate memory. The user is supposed
2026 * to allocate the memory in user-space (malloc/new). The driver pins the
2027 * physical pages (up to the allowed limit by the OS), assigns a virtual
2028 * address in the device VA space and initializes the device MMU.
2029 *
2030 * There is an option for the user to specify the requested virtual address.
2031 *
2032 */
2033#define HL_IOCTL_MEMORY \
2034 _IOWR('H', 0x05, union hl_mem_args)
2035
2036/*
2037 * Debug
2038 * - Enable/disable the ETR/ETF/FUNNEL/STM/BMON/SPMU debug traces
2039 *
2040 * This IOCTL allows the user to get debug traces from the chip.
2041 *
2042 * Before the user can send configuration requests of the various
2043 * debug/profile engines, it needs to set the device into debug mode.
2044 * This is because the debug/profile infrastructure is shared component in the
2045 * device and we can't allow multiple users to access it at the same time.
2046 *
2047 * Once a user set the device into debug mode, the driver won't allow other
2048 * users to "work" with the device, i.e. open a FD. If there are multiple users
2049 * opened on the device, the driver won't allow any user to debug the device.
2050 *
2051 * For each configuration request, the user needs to provide the register index
2052 * and essential data such as buffer address and size.
2053 *
2054 * Once the user has finished using the debug/profile engines, he should
2055 * set the device into non-debug mode, i.e. disable debug mode.
2056 *
2057 * The driver can decide to "kick out" the user if he abuses this interface.
2058 *
2059 */
2060#define HL_IOCTL_DEBUG \
2061 _IOWR('H', 0x06, struct hl_debug_args)
2062
2063#define HL_COMMAND_START 0x01
2064#define HL_COMMAND_END 0x07
2065
2066#endif /* HABANALABS_H_ */