xref: /linux/drivers/infiniband/hw/mlx5/devx.c (revision ae22a94997b8a03dcb3c922857c203246711f9d4)
1 // SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB
2 /*
3  * Copyright (c) 2018, Mellanox Technologies inc.  All rights reserved.
4  */
5 
6 #include <rdma/ib_user_verbs.h>
7 #include <rdma/ib_verbs.h>
8 #include <rdma/uverbs_types.h>
9 #include <rdma/uverbs_ioctl.h>
10 #include <rdma/mlx5_user_ioctl_cmds.h>
11 #include <rdma/mlx5_user_ioctl_verbs.h>
12 #include <rdma/ib_umem.h>
13 #include <rdma/uverbs_std_types.h>
14 #include <linux/mlx5/driver.h>
15 #include <linux/mlx5/fs.h>
16 #include "mlx5_ib.h"
17 #include "devx.h"
18 #include "qp.h"
19 #include <linux/xarray.h>
20 
21 #define UVERBS_MODULE_NAME mlx5_ib
22 #include <rdma/uverbs_named_ioctl.h>
23 
24 static void dispatch_event_fd(struct list_head *fd_list, const void *data);
25 
26 enum devx_obj_flags {
27 	DEVX_OBJ_FLAGS_INDIRECT_MKEY = 1 << 0,
28 	DEVX_OBJ_FLAGS_DCT = 1 << 1,
29 	DEVX_OBJ_FLAGS_CQ = 1 << 2,
30 };
31 
32 struct devx_async_data {
33 	struct mlx5_ib_dev *mdev;
34 	struct list_head list;
35 	struct devx_async_cmd_event_file *ev_file;
36 	struct mlx5_async_work cb_work;
37 	u16 cmd_out_len;
38 	/* must be last field in this structure */
39 	struct mlx5_ib_uapi_devx_async_cmd_hdr hdr;
40 };
41 
42 struct devx_async_event_data {
43 	struct list_head list; /* headed in ev_file->event_list */
44 	struct mlx5_ib_uapi_devx_async_event_hdr hdr;
45 };
46 
47 /* first level XA value data structure */
48 struct devx_event {
49 	struct xarray object_ids; /* second XA level, Key = object id */
50 	struct list_head unaffiliated_list;
51 };
52 
53 /* second level XA value data structure */
54 struct devx_obj_event {
55 	struct rcu_head rcu;
56 	struct list_head obj_sub_list;
57 };
58 
59 struct devx_event_subscription {
60 	struct list_head file_list; /* headed in ev_file->
61 				     * subscribed_events_list
62 				     */
63 	struct list_head xa_list; /* headed in devx_event->unaffiliated_list or
64 				   * devx_obj_event->obj_sub_list
65 				   */
66 	struct list_head obj_list; /* headed in devx_object */
67 	struct list_head event_list; /* headed in ev_file->event_list or in
68 				      * temp list via subscription
69 				      */
70 
71 	u8 is_cleaned:1;
72 	u32 xa_key_level1;
73 	u32 xa_key_level2;
74 	struct rcu_head	rcu;
75 	u64 cookie;
76 	struct devx_async_event_file *ev_file;
77 	struct eventfd_ctx *eventfd;
78 };
79 
80 struct devx_async_event_file {
81 	struct ib_uobject uobj;
82 	/* Head of events that are subscribed to this FD */
83 	struct list_head subscribed_events_list;
84 	spinlock_t lock;
85 	wait_queue_head_t poll_wait;
86 	struct list_head event_list;
87 	struct mlx5_ib_dev *dev;
88 	u8 omit_data:1;
89 	u8 is_overflow_err:1;
90 	u8 is_destroyed:1;
91 };
92 
93 struct devx_umem {
94 	struct mlx5_core_dev		*mdev;
95 	struct ib_umem			*umem;
96 	u32				dinlen;
97 	u32				dinbox[MLX5_ST_SZ_DW(destroy_umem_in)];
98 };
99 
100 struct devx_umem_reg_cmd {
101 	void				*in;
102 	u32				inlen;
103 	u32				out[MLX5_ST_SZ_DW(create_umem_out)];
104 };
105 
106 static struct mlx5_ib_ucontext *
107 devx_ufile2uctx(const struct uverbs_attr_bundle *attrs)
108 {
109 	return to_mucontext(ib_uverbs_get_ucontext(attrs));
110 }
111 
112 int mlx5_ib_devx_create(struct mlx5_ib_dev *dev, bool is_user)
113 {
114 	u32 in[MLX5_ST_SZ_DW(create_uctx_in)] = {};
115 	u32 out[MLX5_ST_SZ_DW(create_uctx_out)] = {};
116 	void *uctx;
117 	int err;
118 	u16 uid;
119 	u32 cap = 0;
120 
121 	/* 0 means not supported */
122 	if (!MLX5_CAP_GEN(dev->mdev, log_max_uctx))
123 		return -EINVAL;
124 
125 	uctx = MLX5_ADDR_OF(create_uctx_in, in, uctx);
126 	if (is_user && capable(CAP_NET_RAW) &&
127 	    (MLX5_CAP_GEN(dev->mdev, uctx_cap) & MLX5_UCTX_CAP_RAW_TX))
128 		cap |= MLX5_UCTX_CAP_RAW_TX;
129 	if (is_user && capable(CAP_SYS_RAWIO) &&
130 	    (MLX5_CAP_GEN(dev->mdev, uctx_cap) &
131 	     MLX5_UCTX_CAP_INTERNAL_DEV_RES))
132 		cap |= MLX5_UCTX_CAP_INTERNAL_DEV_RES;
133 
134 	MLX5_SET(create_uctx_in, in, opcode, MLX5_CMD_OP_CREATE_UCTX);
135 	MLX5_SET(uctx, uctx, cap, cap);
136 
137 	err = mlx5_cmd_exec(dev->mdev, in, sizeof(in), out, sizeof(out));
138 	if (err)
139 		return err;
140 
141 	uid = MLX5_GET(create_uctx_out, out, uid);
142 	return uid;
143 }
144 
145 void mlx5_ib_devx_destroy(struct mlx5_ib_dev *dev, u16 uid)
146 {
147 	u32 in[MLX5_ST_SZ_DW(destroy_uctx_in)] = {};
148 	u32 out[MLX5_ST_SZ_DW(destroy_uctx_out)] = {};
149 
150 	MLX5_SET(destroy_uctx_in, in, opcode, MLX5_CMD_OP_DESTROY_UCTX);
151 	MLX5_SET(destroy_uctx_in, in, uid, uid);
152 
153 	mlx5_cmd_exec(dev->mdev, in, sizeof(in), out, sizeof(out));
154 }
155 
156 static bool is_legacy_unaffiliated_event_num(u16 event_num)
157 {
158 	switch (event_num) {
159 	case MLX5_EVENT_TYPE_PORT_CHANGE:
160 		return true;
161 	default:
162 		return false;
163 	}
164 }
165 
166 static bool is_legacy_obj_event_num(u16 event_num)
167 {
168 	switch (event_num) {
169 	case MLX5_EVENT_TYPE_PATH_MIG:
170 	case MLX5_EVENT_TYPE_COMM_EST:
171 	case MLX5_EVENT_TYPE_SQ_DRAINED:
172 	case MLX5_EVENT_TYPE_SRQ_LAST_WQE:
173 	case MLX5_EVENT_TYPE_SRQ_RQ_LIMIT:
174 	case MLX5_EVENT_TYPE_CQ_ERROR:
175 	case MLX5_EVENT_TYPE_WQ_CATAS_ERROR:
176 	case MLX5_EVENT_TYPE_PATH_MIG_FAILED:
177 	case MLX5_EVENT_TYPE_WQ_INVAL_REQ_ERROR:
178 	case MLX5_EVENT_TYPE_WQ_ACCESS_ERROR:
179 	case MLX5_EVENT_TYPE_SRQ_CATAS_ERROR:
180 	case MLX5_EVENT_TYPE_DCT_DRAINED:
181 	case MLX5_EVENT_TYPE_COMP:
182 	case MLX5_EVENT_TYPE_DCT_KEY_VIOLATION:
183 	case MLX5_EVENT_TYPE_XRQ_ERROR:
184 		return true;
185 	default:
186 		return false;
187 	}
188 }
189 
190 static u16 get_legacy_obj_type(u16 opcode)
191 {
192 	switch (opcode) {
193 	case MLX5_CMD_OP_CREATE_RQ:
194 		return MLX5_EVENT_QUEUE_TYPE_RQ;
195 	case MLX5_CMD_OP_CREATE_QP:
196 		return MLX5_EVENT_QUEUE_TYPE_QP;
197 	case MLX5_CMD_OP_CREATE_SQ:
198 		return MLX5_EVENT_QUEUE_TYPE_SQ;
199 	case MLX5_CMD_OP_CREATE_DCT:
200 		return MLX5_EVENT_QUEUE_TYPE_DCT;
201 	default:
202 		return 0;
203 	}
204 }
205 
206 static u16 get_dec_obj_type(struct devx_obj *obj, u16 event_num)
207 {
208 	u16 opcode;
209 
210 	opcode = (obj->obj_id >> 32) & 0xffff;
211 
212 	if (is_legacy_obj_event_num(event_num))
213 		return get_legacy_obj_type(opcode);
214 
215 	switch (opcode) {
216 	case MLX5_CMD_OP_CREATE_GENERAL_OBJECT:
217 		return (obj->obj_id >> 48);
218 	case MLX5_CMD_OP_CREATE_RQ:
219 		return MLX5_OBJ_TYPE_RQ;
220 	case MLX5_CMD_OP_CREATE_QP:
221 		return MLX5_OBJ_TYPE_QP;
222 	case MLX5_CMD_OP_CREATE_SQ:
223 		return MLX5_OBJ_TYPE_SQ;
224 	case MLX5_CMD_OP_CREATE_DCT:
225 		return MLX5_OBJ_TYPE_DCT;
226 	case MLX5_CMD_OP_CREATE_TIR:
227 		return MLX5_OBJ_TYPE_TIR;
228 	case MLX5_CMD_OP_CREATE_TIS:
229 		return MLX5_OBJ_TYPE_TIS;
230 	case MLX5_CMD_OP_CREATE_PSV:
231 		return MLX5_OBJ_TYPE_PSV;
232 	case MLX5_OBJ_TYPE_MKEY:
233 		return MLX5_OBJ_TYPE_MKEY;
234 	case MLX5_CMD_OP_CREATE_RMP:
235 		return MLX5_OBJ_TYPE_RMP;
236 	case MLX5_CMD_OP_CREATE_XRC_SRQ:
237 		return MLX5_OBJ_TYPE_XRC_SRQ;
238 	case MLX5_CMD_OP_CREATE_XRQ:
239 		return MLX5_OBJ_TYPE_XRQ;
240 	case MLX5_CMD_OP_CREATE_RQT:
241 		return MLX5_OBJ_TYPE_RQT;
242 	case MLX5_CMD_OP_ALLOC_FLOW_COUNTER:
243 		return MLX5_OBJ_TYPE_FLOW_COUNTER;
244 	case MLX5_CMD_OP_CREATE_CQ:
245 		return MLX5_OBJ_TYPE_CQ;
246 	default:
247 		return 0;
248 	}
249 }
250 
251 static u16 get_event_obj_type(unsigned long event_type, struct mlx5_eqe *eqe)
252 {
253 	switch (event_type) {
254 	case MLX5_EVENT_TYPE_WQ_CATAS_ERROR:
255 	case MLX5_EVENT_TYPE_WQ_ACCESS_ERROR:
256 	case MLX5_EVENT_TYPE_WQ_INVAL_REQ_ERROR:
257 	case MLX5_EVENT_TYPE_SRQ_LAST_WQE:
258 	case MLX5_EVENT_TYPE_PATH_MIG:
259 	case MLX5_EVENT_TYPE_PATH_MIG_FAILED:
260 	case MLX5_EVENT_TYPE_COMM_EST:
261 	case MLX5_EVENT_TYPE_SQ_DRAINED:
262 	case MLX5_EVENT_TYPE_SRQ_RQ_LIMIT:
263 	case MLX5_EVENT_TYPE_SRQ_CATAS_ERROR:
264 		return eqe->data.qp_srq.type;
265 	case MLX5_EVENT_TYPE_CQ_ERROR:
266 	case MLX5_EVENT_TYPE_XRQ_ERROR:
267 		return 0;
268 	case MLX5_EVENT_TYPE_DCT_DRAINED:
269 	case MLX5_EVENT_TYPE_DCT_KEY_VIOLATION:
270 		return MLX5_EVENT_QUEUE_TYPE_DCT;
271 	default:
272 		return MLX5_GET(affiliated_event_header, &eqe->data, obj_type);
273 	}
274 }
275 
276 static u32 get_dec_obj_id(u64 obj_id)
277 {
278 	return (obj_id & 0xffffffff);
279 }
280 
281 /*
282  * As the obj_id in the firmware is not globally unique the object type
283  * must be considered upon checking for a valid object id.
284  * For that the opcode of the creator command is encoded as part of the obj_id.
285  */
286 static u64 get_enc_obj_id(u32 opcode, u32 obj_id)
287 {
288 	return ((u64)opcode << 32) | obj_id;
289 }
290 
291 static u32 devx_get_created_obj_id(const void *in, const void *out, u16 opcode)
292 {
293 	switch (opcode) {
294 	case MLX5_CMD_OP_CREATE_GENERAL_OBJECT:
295 		return MLX5_GET(general_obj_out_cmd_hdr, out, obj_id);
296 	case MLX5_CMD_OP_CREATE_UMEM:
297 		return MLX5_GET(create_umem_out, out, umem_id);
298 	case MLX5_CMD_OP_CREATE_MKEY:
299 		return MLX5_GET(create_mkey_out, out, mkey_index);
300 	case MLX5_CMD_OP_CREATE_CQ:
301 		return MLX5_GET(create_cq_out, out, cqn);
302 	case MLX5_CMD_OP_ALLOC_PD:
303 		return MLX5_GET(alloc_pd_out, out, pd);
304 	case MLX5_CMD_OP_ALLOC_TRANSPORT_DOMAIN:
305 		return MLX5_GET(alloc_transport_domain_out, out,
306 				transport_domain);
307 	case MLX5_CMD_OP_CREATE_RMP:
308 		return MLX5_GET(create_rmp_out, out, rmpn);
309 	case MLX5_CMD_OP_CREATE_SQ:
310 		return MLX5_GET(create_sq_out, out, sqn);
311 	case MLX5_CMD_OP_CREATE_RQ:
312 		return MLX5_GET(create_rq_out, out, rqn);
313 	case MLX5_CMD_OP_CREATE_RQT:
314 		return MLX5_GET(create_rqt_out, out, rqtn);
315 	case MLX5_CMD_OP_CREATE_TIR:
316 		return MLX5_GET(create_tir_out, out, tirn);
317 	case MLX5_CMD_OP_CREATE_TIS:
318 		return MLX5_GET(create_tis_out, out, tisn);
319 	case MLX5_CMD_OP_ALLOC_Q_COUNTER:
320 		return MLX5_GET(alloc_q_counter_out, out, counter_set_id);
321 	case MLX5_CMD_OP_CREATE_FLOW_TABLE:
322 		return MLX5_GET(create_flow_table_out, out, table_id);
323 	case MLX5_CMD_OP_CREATE_FLOW_GROUP:
324 		return MLX5_GET(create_flow_group_out, out, group_id);
325 	case MLX5_CMD_OP_SET_FLOW_TABLE_ENTRY:
326 		return MLX5_GET(set_fte_in, in, flow_index);
327 	case MLX5_CMD_OP_ALLOC_FLOW_COUNTER:
328 		return MLX5_GET(alloc_flow_counter_out, out, flow_counter_id);
329 	case MLX5_CMD_OP_ALLOC_PACKET_REFORMAT_CONTEXT:
330 		return MLX5_GET(alloc_packet_reformat_context_out, out,
331 				packet_reformat_id);
332 	case MLX5_CMD_OP_ALLOC_MODIFY_HEADER_CONTEXT:
333 		return MLX5_GET(alloc_modify_header_context_out, out,
334 				modify_header_id);
335 	case MLX5_CMD_OP_CREATE_SCHEDULING_ELEMENT:
336 		return MLX5_GET(create_scheduling_element_out, out,
337 				scheduling_element_id);
338 	case MLX5_CMD_OP_ADD_VXLAN_UDP_DPORT:
339 		return MLX5_GET(add_vxlan_udp_dport_in, in, vxlan_udp_port);
340 	case MLX5_CMD_OP_SET_L2_TABLE_ENTRY:
341 		return MLX5_GET(set_l2_table_entry_in, in, table_index);
342 	case MLX5_CMD_OP_CREATE_QP:
343 		return MLX5_GET(create_qp_out, out, qpn);
344 	case MLX5_CMD_OP_CREATE_SRQ:
345 		return MLX5_GET(create_srq_out, out, srqn);
346 	case MLX5_CMD_OP_CREATE_XRC_SRQ:
347 		return MLX5_GET(create_xrc_srq_out, out, xrc_srqn);
348 	case MLX5_CMD_OP_CREATE_DCT:
349 		return MLX5_GET(create_dct_out, out, dctn);
350 	case MLX5_CMD_OP_CREATE_XRQ:
351 		return MLX5_GET(create_xrq_out, out, xrqn);
352 	case MLX5_CMD_OP_ATTACH_TO_MCG:
353 		return MLX5_GET(attach_to_mcg_in, in, qpn);
354 	case MLX5_CMD_OP_ALLOC_XRCD:
355 		return MLX5_GET(alloc_xrcd_out, out, xrcd);
356 	case MLX5_CMD_OP_CREATE_PSV:
357 		return MLX5_GET(create_psv_out, out, psv0_index);
358 	default:
359 		/* The entry must match to one of the devx_is_obj_create_cmd */
360 		WARN_ON(true);
361 		return 0;
362 	}
363 }
364 
365 static u64 devx_get_obj_id(const void *in)
366 {
367 	u16 opcode = MLX5_GET(general_obj_in_cmd_hdr, in, opcode);
368 	u64 obj_id;
369 
370 	switch (opcode) {
371 	case MLX5_CMD_OP_MODIFY_GENERAL_OBJECT:
372 	case MLX5_CMD_OP_QUERY_GENERAL_OBJECT:
373 		obj_id = get_enc_obj_id(MLX5_CMD_OP_CREATE_GENERAL_OBJECT |
374 					MLX5_GET(general_obj_in_cmd_hdr, in,
375 						 obj_type) << 16,
376 					MLX5_GET(general_obj_in_cmd_hdr, in,
377 						 obj_id));
378 		break;
379 	case MLX5_CMD_OP_QUERY_MKEY:
380 		obj_id = get_enc_obj_id(MLX5_CMD_OP_CREATE_MKEY,
381 					MLX5_GET(query_mkey_in, in,
382 						 mkey_index));
383 		break;
384 	case MLX5_CMD_OP_QUERY_CQ:
385 		obj_id = get_enc_obj_id(MLX5_CMD_OP_CREATE_CQ,
386 					MLX5_GET(query_cq_in, in, cqn));
387 		break;
388 	case MLX5_CMD_OP_MODIFY_CQ:
389 		obj_id = get_enc_obj_id(MLX5_CMD_OP_CREATE_CQ,
390 					MLX5_GET(modify_cq_in, in, cqn));
391 		break;
392 	case MLX5_CMD_OP_QUERY_SQ:
393 		obj_id = get_enc_obj_id(MLX5_CMD_OP_CREATE_SQ,
394 					MLX5_GET(query_sq_in, in, sqn));
395 		break;
396 	case MLX5_CMD_OP_MODIFY_SQ:
397 		obj_id = get_enc_obj_id(MLX5_CMD_OP_CREATE_SQ,
398 					MLX5_GET(modify_sq_in, in, sqn));
399 		break;
400 	case MLX5_CMD_OP_QUERY_RQ:
401 		obj_id = get_enc_obj_id(MLX5_CMD_OP_CREATE_RQ,
402 					MLX5_GET(query_rq_in, in, rqn));
403 		break;
404 	case MLX5_CMD_OP_MODIFY_RQ:
405 		obj_id = get_enc_obj_id(MLX5_CMD_OP_CREATE_RQ,
406 					MLX5_GET(modify_rq_in, in, rqn));
407 		break;
408 	case MLX5_CMD_OP_QUERY_RMP:
409 		obj_id = get_enc_obj_id(MLX5_CMD_OP_CREATE_RMP,
410 					MLX5_GET(query_rmp_in, in, rmpn));
411 		break;
412 	case MLX5_CMD_OP_MODIFY_RMP:
413 		obj_id = get_enc_obj_id(MLX5_CMD_OP_CREATE_RMP,
414 					MLX5_GET(modify_rmp_in, in, rmpn));
415 		break;
416 	case MLX5_CMD_OP_QUERY_RQT:
417 		obj_id = get_enc_obj_id(MLX5_CMD_OP_CREATE_RQT,
418 					MLX5_GET(query_rqt_in, in, rqtn));
419 		break;
420 	case MLX5_CMD_OP_MODIFY_RQT:
421 		obj_id = get_enc_obj_id(MLX5_CMD_OP_CREATE_RQT,
422 					MLX5_GET(modify_rqt_in, in, rqtn));
423 		break;
424 	case MLX5_CMD_OP_QUERY_TIR:
425 		obj_id = get_enc_obj_id(MLX5_CMD_OP_CREATE_TIR,
426 					MLX5_GET(query_tir_in, in, tirn));
427 		break;
428 	case MLX5_CMD_OP_MODIFY_TIR:
429 		obj_id = get_enc_obj_id(MLX5_CMD_OP_CREATE_TIR,
430 					MLX5_GET(modify_tir_in, in, tirn));
431 		break;
432 	case MLX5_CMD_OP_QUERY_TIS:
433 		obj_id = get_enc_obj_id(MLX5_CMD_OP_CREATE_TIS,
434 					MLX5_GET(query_tis_in, in, tisn));
435 		break;
436 	case MLX5_CMD_OP_MODIFY_TIS:
437 		obj_id = get_enc_obj_id(MLX5_CMD_OP_CREATE_TIS,
438 					MLX5_GET(modify_tis_in, in, tisn));
439 		break;
440 	case MLX5_CMD_OP_QUERY_FLOW_TABLE:
441 		obj_id = get_enc_obj_id(MLX5_CMD_OP_CREATE_FLOW_TABLE,
442 					MLX5_GET(query_flow_table_in, in,
443 						 table_id));
444 		break;
445 	case MLX5_CMD_OP_MODIFY_FLOW_TABLE:
446 		obj_id = get_enc_obj_id(MLX5_CMD_OP_CREATE_FLOW_TABLE,
447 					MLX5_GET(modify_flow_table_in, in,
448 						 table_id));
449 		break;
450 	case MLX5_CMD_OP_QUERY_FLOW_GROUP:
451 		obj_id = get_enc_obj_id(MLX5_CMD_OP_CREATE_FLOW_GROUP,
452 					MLX5_GET(query_flow_group_in, in,
453 						 group_id));
454 		break;
455 	case MLX5_CMD_OP_QUERY_FLOW_TABLE_ENTRY:
456 		obj_id = get_enc_obj_id(MLX5_CMD_OP_SET_FLOW_TABLE_ENTRY,
457 					MLX5_GET(query_fte_in, in,
458 						 flow_index));
459 		break;
460 	case MLX5_CMD_OP_SET_FLOW_TABLE_ENTRY:
461 		obj_id = get_enc_obj_id(MLX5_CMD_OP_SET_FLOW_TABLE_ENTRY,
462 					MLX5_GET(set_fte_in, in, flow_index));
463 		break;
464 	case MLX5_CMD_OP_QUERY_Q_COUNTER:
465 		obj_id = get_enc_obj_id(MLX5_CMD_OP_ALLOC_Q_COUNTER,
466 					MLX5_GET(query_q_counter_in, in,
467 						 counter_set_id));
468 		break;
469 	case MLX5_CMD_OP_QUERY_FLOW_COUNTER:
470 		obj_id = get_enc_obj_id(MLX5_CMD_OP_ALLOC_FLOW_COUNTER,
471 					MLX5_GET(query_flow_counter_in, in,
472 						 flow_counter_id));
473 		break;
474 	case MLX5_CMD_OP_QUERY_MODIFY_HEADER_CONTEXT:
475 		obj_id = get_enc_obj_id(MLX5_CMD_OP_ALLOC_MODIFY_HEADER_CONTEXT,
476 					MLX5_GET(query_modify_header_context_in,
477 						 in, modify_header_id));
478 		break;
479 	case MLX5_CMD_OP_QUERY_SCHEDULING_ELEMENT:
480 		obj_id = get_enc_obj_id(MLX5_CMD_OP_CREATE_SCHEDULING_ELEMENT,
481 					MLX5_GET(query_scheduling_element_in,
482 						 in, scheduling_element_id));
483 		break;
484 	case MLX5_CMD_OP_MODIFY_SCHEDULING_ELEMENT:
485 		obj_id = get_enc_obj_id(MLX5_CMD_OP_CREATE_SCHEDULING_ELEMENT,
486 					MLX5_GET(modify_scheduling_element_in,
487 						 in, scheduling_element_id));
488 		break;
489 	case MLX5_CMD_OP_ADD_VXLAN_UDP_DPORT:
490 		obj_id = get_enc_obj_id(MLX5_CMD_OP_ADD_VXLAN_UDP_DPORT,
491 					MLX5_GET(add_vxlan_udp_dport_in, in,
492 						 vxlan_udp_port));
493 		break;
494 	case MLX5_CMD_OP_QUERY_L2_TABLE_ENTRY:
495 		obj_id = get_enc_obj_id(MLX5_CMD_OP_SET_L2_TABLE_ENTRY,
496 					MLX5_GET(query_l2_table_entry_in, in,
497 						 table_index));
498 		break;
499 	case MLX5_CMD_OP_SET_L2_TABLE_ENTRY:
500 		obj_id = get_enc_obj_id(MLX5_CMD_OP_SET_L2_TABLE_ENTRY,
501 					MLX5_GET(set_l2_table_entry_in, in,
502 						 table_index));
503 		break;
504 	case MLX5_CMD_OP_QUERY_QP:
505 		obj_id = get_enc_obj_id(MLX5_CMD_OP_CREATE_QP,
506 					MLX5_GET(query_qp_in, in, qpn));
507 		break;
508 	case MLX5_CMD_OP_RST2INIT_QP:
509 		obj_id = get_enc_obj_id(MLX5_CMD_OP_CREATE_QP,
510 					MLX5_GET(rst2init_qp_in, in, qpn));
511 		break;
512 	case MLX5_CMD_OP_INIT2INIT_QP:
513 		obj_id = get_enc_obj_id(MLX5_CMD_OP_CREATE_QP,
514 					MLX5_GET(init2init_qp_in, in, qpn));
515 		break;
516 	case MLX5_CMD_OP_INIT2RTR_QP:
517 		obj_id = get_enc_obj_id(MLX5_CMD_OP_CREATE_QP,
518 					MLX5_GET(init2rtr_qp_in, in, qpn));
519 		break;
520 	case MLX5_CMD_OP_RTR2RTS_QP:
521 		obj_id = get_enc_obj_id(MLX5_CMD_OP_CREATE_QP,
522 					MLX5_GET(rtr2rts_qp_in, in, qpn));
523 		break;
524 	case MLX5_CMD_OP_RTS2RTS_QP:
525 		obj_id = get_enc_obj_id(MLX5_CMD_OP_CREATE_QP,
526 					MLX5_GET(rts2rts_qp_in, in, qpn));
527 		break;
528 	case MLX5_CMD_OP_SQERR2RTS_QP:
529 		obj_id = get_enc_obj_id(MLX5_CMD_OP_CREATE_QP,
530 					MLX5_GET(sqerr2rts_qp_in, in, qpn));
531 		break;
532 	case MLX5_CMD_OP_2ERR_QP:
533 		obj_id = get_enc_obj_id(MLX5_CMD_OP_CREATE_QP,
534 					MLX5_GET(qp_2err_in, in, qpn));
535 		break;
536 	case MLX5_CMD_OP_2RST_QP:
537 		obj_id = get_enc_obj_id(MLX5_CMD_OP_CREATE_QP,
538 					MLX5_GET(qp_2rst_in, in, qpn));
539 		break;
540 	case MLX5_CMD_OP_QUERY_DCT:
541 		obj_id = get_enc_obj_id(MLX5_CMD_OP_CREATE_DCT,
542 					MLX5_GET(query_dct_in, in, dctn));
543 		break;
544 	case MLX5_CMD_OP_QUERY_XRQ:
545 	case MLX5_CMD_OP_QUERY_XRQ_DC_PARAMS_ENTRY:
546 	case MLX5_CMD_OP_QUERY_XRQ_ERROR_PARAMS:
547 		obj_id = get_enc_obj_id(MLX5_CMD_OP_CREATE_XRQ,
548 					MLX5_GET(query_xrq_in, in, xrqn));
549 		break;
550 	case MLX5_CMD_OP_QUERY_XRC_SRQ:
551 		obj_id = get_enc_obj_id(MLX5_CMD_OP_CREATE_XRC_SRQ,
552 					MLX5_GET(query_xrc_srq_in, in,
553 						 xrc_srqn));
554 		break;
555 	case MLX5_CMD_OP_ARM_XRC_SRQ:
556 		obj_id = get_enc_obj_id(MLX5_CMD_OP_CREATE_XRC_SRQ,
557 					MLX5_GET(arm_xrc_srq_in, in, xrc_srqn));
558 		break;
559 	case MLX5_CMD_OP_QUERY_SRQ:
560 		obj_id = get_enc_obj_id(MLX5_CMD_OP_CREATE_SRQ,
561 					MLX5_GET(query_srq_in, in, srqn));
562 		break;
563 	case MLX5_CMD_OP_ARM_RQ:
564 		obj_id = get_enc_obj_id(MLX5_CMD_OP_CREATE_RQ,
565 					MLX5_GET(arm_rq_in, in, srq_number));
566 		break;
567 	case MLX5_CMD_OP_ARM_DCT_FOR_KEY_VIOLATION:
568 		obj_id = get_enc_obj_id(MLX5_CMD_OP_CREATE_DCT,
569 					MLX5_GET(drain_dct_in, in, dctn));
570 		break;
571 	case MLX5_CMD_OP_ARM_XRQ:
572 	case MLX5_CMD_OP_SET_XRQ_DC_PARAMS_ENTRY:
573 	case MLX5_CMD_OP_RELEASE_XRQ_ERROR:
574 	case MLX5_CMD_OP_MODIFY_XRQ:
575 		obj_id = get_enc_obj_id(MLX5_CMD_OP_CREATE_XRQ,
576 					MLX5_GET(arm_xrq_in, in, xrqn));
577 		break;
578 	case MLX5_CMD_OP_QUERY_PACKET_REFORMAT_CONTEXT:
579 		obj_id = get_enc_obj_id
580 				(MLX5_CMD_OP_ALLOC_PACKET_REFORMAT_CONTEXT,
581 				 MLX5_GET(query_packet_reformat_context_in,
582 					  in, packet_reformat_id));
583 		break;
584 	default:
585 		obj_id = 0;
586 	}
587 
588 	return obj_id;
589 }
590 
591 static bool devx_is_valid_obj_id(struct uverbs_attr_bundle *attrs,
592 				 struct ib_uobject *uobj, const void *in)
593 {
594 	struct mlx5_ib_dev *dev = mlx5_udata_to_mdev(&attrs->driver_udata);
595 	u64 obj_id = devx_get_obj_id(in);
596 
597 	if (!obj_id)
598 		return false;
599 
600 	switch (uobj_get_object_id(uobj)) {
601 	case UVERBS_OBJECT_CQ:
602 		return get_enc_obj_id(MLX5_CMD_OP_CREATE_CQ,
603 				      to_mcq(uobj->object)->mcq.cqn) ==
604 				      obj_id;
605 
606 	case UVERBS_OBJECT_SRQ:
607 	{
608 		struct mlx5_core_srq *srq = &(to_msrq(uobj->object)->msrq);
609 		u16 opcode;
610 
611 		switch (srq->common.res) {
612 		case MLX5_RES_XSRQ:
613 			opcode = MLX5_CMD_OP_CREATE_XRC_SRQ;
614 			break;
615 		case MLX5_RES_XRQ:
616 			opcode = MLX5_CMD_OP_CREATE_XRQ;
617 			break;
618 		default:
619 			if (!dev->mdev->issi)
620 				opcode = MLX5_CMD_OP_CREATE_SRQ;
621 			else
622 				opcode = MLX5_CMD_OP_CREATE_RMP;
623 		}
624 
625 		return get_enc_obj_id(opcode,
626 				      to_msrq(uobj->object)->msrq.srqn) ==
627 				      obj_id;
628 	}
629 
630 	case UVERBS_OBJECT_QP:
631 	{
632 		struct mlx5_ib_qp *qp = to_mqp(uobj->object);
633 
634 		if (qp->type == IB_QPT_RAW_PACKET ||
635 		    (qp->flags & IB_QP_CREATE_SOURCE_QPN)) {
636 			struct mlx5_ib_raw_packet_qp *raw_packet_qp =
637 							 &qp->raw_packet_qp;
638 			struct mlx5_ib_rq *rq = &raw_packet_qp->rq;
639 			struct mlx5_ib_sq *sq = &raw_packet_qp->sq;
640 
641 			return (get_enc_obj_id(MLX5_CMD_OP_CREATE_RQ,
642 					       rq->base.mqp.qpn) == obj_id ||
643 				get_enc_obj_id(MLX5_CMD_OP_CREATE_SQ,
644 					       sq->base.mqp.qpn) == obj_id ||
645 				get_enc_obj_id(MLX5_CMD_OP_CREATE_TIR,
646 					       rq->tirn) == obj_id ||
647 				get_enc_obj_id(MLX5_CMD_OP_CREATE_TIS,
648 					       sq->tisn) == obj_id);
649 		}
650 
651 		if (qp->type == MLX5_IB_QPT_DCT)
652 			return get_enc_obj_id(MLX5_CMD_OP_CREATE_DCT,
653 					      qp->dct.mdct.mqp.qpn) == obj_id;
654 		return get_enc_obj_id(MLX5_CMD_OP_CREATE_QP,
655 				      qp->ibqp.qp_num) == obj_id;
656 	}
657 
658 	case UVERBS_OBJECT_WQ:
659 		return get_enc_obj_id(MLX5_CMD_OP_CREATE_RQ,
660 				      to_mrwq(uobj->object)->core_qp.qpn) ==
661 				      obj_id;
662 
663 	case UVERBS_OBJECT_RWQ_IND_TBL:
664 		return get_enc_obj_id(MLX5_CMD_OP_CREATE_RQT,
665 				      to_mrwq_ind_table(uobj->object)->rqtn) ==
666 				      obj_id;
667 
668 	case MLX5_IB_OBJECT_DEVX_OBJ:
669 	{
670 		u16 opcode = MLX5_GET(general_obj_in_cmd_hdr, in, opcode);
671 		struct devx_obj *devx_uobj = uobj->object;
672 
673 		if (opcode == MLX5_CMD_OP_QUERY_FLOW_COUNTER &&
674 		    devx_uobj->flow_counter_bulk_size) {
675 			u64 end;
676 
677 			end = devx_uobj->obj_id +
678 				devx_uobj->flow_counter_bulk_size;
679 			return devx_uobj->obj_id <= obj_id && end > obj_id;
680 		}
681 
682 		return devx_uobj->obj_id == obj_id;
683 	}
684 
685 	default:
686 		return false;
687 	}
688 }
689 
690 static void devx_set_umem_valid(const void *in)
691 {
692 	u16 opcode = MLX5_GET(general_obj_in_cmd_hdr, in, opcode);
693 
694 	switch (opcode) {
695 	case MLX5_CMD_OP_CREATE_MKEY:
696 		MLX5_SET(create_mkey_in, in, mkey_umem_valid, 1);
697 		break;
698 	case MLX5_CMD_OP_CREATE_CQ:
699 	{
700 		void *cqc;
701 
702 		MLX5_SET(create_cq_in, in, cq_umem_valid, 1);
703 		cqc = MLX5_ADDR_OF(create_cq_in, in, cq_context);
704 		MLX5_SET(cqc, cqc, dbr_umem_valid, 1);
705 		break;
706 	}
707 	case MLX5_CMD_OP_CREATE_QP:
708 	{
709 		void *qpc;
710 
711 		qpc = MLX5_ADDR_OF(create_qp_in, in, qpc);
712 		MLX5_SET(qpc, qpc, dbr_umem_valid, 1);
713 		MLX5_SET(create_qp_in, in, wq_umem_valid, 1);
714 		break;
715 	}
716 
717 	case MLX5_CMD_OP_CREATE_RQ:
718 	{
719 		void *rqc, *wq;
720 
721 		rqc = MLX5_ADDR_OF(create_rq_in, in, ctx);
722 		wq  = MLX5_ADDR_OF(rqc, rqc, wq);
723 		MLX5_SET(wq, wq, dbr_umem_valid, 1);
724 		MLX5_SET(wq, wq, wq_umem_valid, 1);
725 		break;
726 	}
727 
728 	case MLX5_CMD_OP_CREATE_SQ:
729 	{
730 		void *sqc, *wq;
731 
732 		sqc = MLX5_ADDR_OF(create_sq_in, in, ctx);
733 		wq = MLX5_ADDR_OF(sqc, sqc, wq);
734 		MLX5_SET(wq, wq, dbr_umem_valid, 1);
735 		MLX5_SET(wq, wq, wq_umem_valid, 1);
736 		break;
737 	}
738 
739 	case MLX5_CMD_OP_MODIFY_CQ:
740 		MLX5_SET(modify_cq_in, in, cq_umem_valid, 1);
741 		break;
742 
743 	case MLX5_CMD_OP_CREATE_RMP:
744 	{
745 		void *rmpc, *wq;
746 
747 		rmpc = MLX5_ADDR_OF(create_rmp_in, in, ctx);
748 		wq = MLX5_ADDR_OF(rmpc, rmpc, wq);
749 		MLX5_SET(wq, wq, dbr_umem_valid, 1);
750 		MLX5_SET(wq, wq, wq_umem_valid, 1);
751 		break;
752 	}
753 
754 	case MLX5_CMD_OP_CREATE_XRQ:
755 	{
756 		void *xrqc, *wq;
757 
758 		xrqc = MLX5_ADDR_OF(create_xrq_in, in, xrq_context);
759 		wq = MLX5_ADDR_OF(xrqc, xrqc, wq);
760 		MLX5_SET(wq, wq, dbr_umem_valid, 1);
761 		MLX5_SET(wq, wq, wq_umem_valid, 1);
762 		break;
763 	}
764 
765 	case MLX5_CMD_OP_CREATE_XRC_SRQ:
766 	{
767 		void *xrc_srqc;
768 
769 		MLX5_SET(create_xrc_srq_in, in, xrc_srq_umem_valid, 1);
770 		xrc_srqc = MLX5_ADDR_OF(create_xrc_srq_in, in,
771 					xrc_srq_context_entry);
772 		MLX5_SET(xrc_srqc, xrc_srqc, dbr_umem_valid, 1);
773 		break;
774 	}
775 
776 	default:
777 		return;
778 	}
779 }
780 
781 static bool devx_is_obj_create_cmd(const void *in, u16 *opcode)
782 {
783 	*opcode = MLX5_GET(general_obj_in_cmd_hdr, in, opcode);
784 
785 	switch (*opcode) {
786 	case MLX5_CMD_OP_CREATE_GENERAL_OBJECT:
787 	case MLX5_CMD_OP_CREATE_MKEY:
788 	case MLX5_CMD_OP_CREATE_CQ:
789 	case MLX5_CMD_OP_ALLOC_PD:
790 	case MLX5_CMD_OP_ALLOC_TRANSPORT_DOMAIN:
791 	case MLX5_CMD_OP_CREATE_RMP:
792 	case MLX5_CMD_OP_CREATE_SQ:
793 	case MLX5_CMD_OP_CREATE_RQ:
794 	case MLX5_CMD_OP_CREATE_RQT:
795 	case MLX5_CMD_OP_CREATE_TIR:
796 	case MLX5_CMD_OP_CREATE_TIS:
797 	case MLX5_CMD_OP_ALLOC_Q_COUNTER:
798 	case MLX5_CMD_OP_CREATE_FLOW_TABLE:
799 	case MLX5_CMD_OP_CREATE_FLOW_GROUP:
800 	case MLX5_CMD_OP_ALLOC_FLOW_COUNTER:
801 	case MLX5_CMD_OP_ALLOC_PACKET_REFORMAT_CONTEXT:
802 	case MLX5_CMD_OP_ALLOC_MODIFY_HEADER_CONTEXT:
803 	case MLX5_CMD_OP_CREATE_SCHEDULING_ELEMENT:
804 	case MLX5_CMD_OP_ADD_VXLAN_UDP_DPORT:
805 	case MLX5_CMD_OP_SET_L2_TABLE_ENTRY:
806 	case MLX5_CMD_OP_CREATE_QP:
807 	case MLX5_CMD_OP_CREATE_SRQ:
808 	case MLX5_CMD_OP_CREATE_XRC_SRQ:
809 	case MLX5_CMD_OP_CREATE_DCT:
810 	case MLX5_CMD_OP_CREATE_XRQ:
811 	case MLX5_CMD_OP_ATTACH_TO_MCG:
812 	case MLX5_CMD_OP_ALLOC_XRCD:
813 		return true;
814 	case MLX5_CMD_OP_SET_FLOW_TABLE_ENTRY:
815 	{
816 		u16 op_mod = MLX5_GET(set_fte_in, in, op_mod);
817 		if (op_mod == 0)
818 			return true;
819 		return false;
820 	}
821 	case MLX5_CMD_OP_CREATE_PSV:
822 	{
823 		u8 num_psv = MLX5_GET(create_psv_in, in, num_psv);
824 
825 		if (num_psv == 1)
826 			return true;
827 		return false;
828 	}
829 	default:
830 		return false;
831 	}
832 }
833 
834 static bool devx_is_obj_modify_cmd(const void *in)
835 {
836 	u16 opcode = MLX5_GET(general_obj_in_cmd_hdr, in, opcode);
837 
838 	switch (opcode) {
839 	case MLX5_CMD_OP_MODIFY_GENERAL_OBJECT:
840 	case MLX5_CMD_OP_MODIFY_CQ:
841 	case MLX5_CMD_OP_MODIFY_RMP:
842 	case MLX5_CMD_OP_MODIFY_SQ:
843 	case MLX5_CMD_OP_MODIFY_RQ:
844 	case MLX5_CMD_OP_MODIFY_RQT:
845 	case MLX5_CMD_OP_MODIFY_TIR:
846 	case MLX5_CMD_OP_MODIFY_TIS:
847 	case MLX5_CMD_OP_MODIFY_FLOW_TABLE:
848 	case MLX5_CMD_OP_MODIFY_SCHEDULING_ELEMENT:
849 	case MLX5_CMD_OP_ADD_VXLAN_UDP_DPORT:
850 	case MLX5_CMD_OP_SET_L2_TABLE_ENTRY:
851 	case MLX5_CMD_OP_RST2INIT_QP:
852 	case MLX5_CMD_OP_INIT2RTR_QP:
853 	case MLX5_CMD_OP_INIT2INIT_QP:
854 	case MLX5_CMD_OP_RTR2RTS_QP:
855 	case MLX5_CMD_OP_RTS2RTS_QP:
856 	case MLX5_CMD_OP_SQERR2RTS_QP:
857 	case MLX5_CMD_OP_2ERR_QP:
858 	case MLX5_CMD_OP_2RST_QP:
859 	case MLX5_CMD_OP_ARM_XRC_SRQ:
860 	case MLX5_CMD_OP_ARM_RQ:
861 	case MLX5_CMD_OP_ARM_DCT_FOR_KEY_VIOLATION:
862 	case MLX5_CMD_OP_ARM_XRQ:
863 	case MLX5_CMD_OP_SET_XRQ_DC_PARAMS_ENTRY:
864 	case MLX5_CMD_OP_RELEASE_XRQ_ERROR:
865 	case MLX5_CMD_OP_MODIFY_XRQ:
866 		return true;
867 	case MLX5_CMD_OP_SET_FLOW_TABLE_ENTRY:
868 	{
869 		u16 op_mod = MLX5_GET(set_fte_in, in, op_mod);
870 
871 		if (op_mod == 1)
872 			return true;
873 		return false;
874 	}
875 	default:
876 		return false;
877 	}
878 }
879 
880 static bool devx_is_obj_query_cmd(const void *in)
881 {
882 	u16 opcode = MLX5_GET(general_obj_in_cmd_hdr, in, opcode);
883 
884 	switch (opcode) {
885 	case MLX5_CMD_OP_QUERY_GENERAL_OBJECT:
886 	case MLX5_CMD_OP_QUERY_MKEY:
887 	case MLX5_CMD_OP_QUERY_CQ:
888 	case MLX5_CMD_OP_QUERY_RMP:
889 	case MLX5_CMD_OP_QUERY_SQ:
890 	case MLX5_CMD_OP_QUERY_RQ:
891 	case MLX5_CMD_OP_QUERY_RQT:
892 	case MLX5_CMD_OP_QUERY_TIR:
893 	case MLX5_CMD_OP_QUERY_TIS:
894 	case MLX5_CMD_OP_QUERY_Q_COUNTER:
895 	case MLX5_CMD_OP_QUERY_FLOW_TABLE:
896 	case MLX5_CMD_OP_QUERY_FLOW_GROUP:
897 	case MLX5_CMD_OP_QUERY_FLOW_TABLE_ENTRY:
898 	case MLX5_CMD_OP_QUERY_FLOW_COUNTER:
899 	case MLX5_CMD_OP_QUERY_MODIFY_HEADER_CONTEXT:
900 	case MLX5_CMD_OP_QUERY_SCHEDULING_ELEMENT:
901 	case MLX5_CMD_OP_QUERY_L2_TABLE_ENTRY:
902 	case MLX5_CMD_OP_QUERY_QP:
903 	case MLX5_CMD_OP_QUERY_SRQ:
904 	case MLX5_CMD_OP_QUERY_XRC_SRQ:
905 	case MLX5_CMD_OP_QUERY_DCT:
906 	case MLX5_CMD_OP_QUERY_XRQ:
907 	case MLX5_CMD_OP_QUERY_XRQ_DC_PARAMS_ENTRY:
908 	case MLX5_CMD_OP_QUERY_XRQ_ERROR_PARAMS:
909 	case MLX5_CMD_OP_QUERY_PACKET_REFORMAT_CONTEXT:
910 		return true;
911 	default:
912 		return false;
913 	}
914 }
915 
916 static bool devx_is_whitelist_cmd(void *in)
917 {
918 	u16 opcode = MLX5_GET(general_obj_in_cmd_hdr, in, opcode);
919 
920 	switch (opcode) {
921 	case MLX5_CMD_OP_QUERY_HCA_CAP:
922 	case MLX5_CMD_OP_QUERY_HCA_VPORT_CONTEXT:
923 	case MLX5_CMD_OP_QUERY_ESW_VPORT_CONTEXT:
924 	case MLX5_CMD_OP_QUERY_ESW_FUNCTIONS:
925 		return true;
926 	default:
927 		return false;
928 	}
929 }
930 
931 static int devx_get_uid(struct mlx5_ib_ucontext *c, void *cmd_in)
932 {
933 	if (devx_is_whitelist_cmd(cmd_in)) {
934 		struct mlx5_ib_dev *dev;
935 
936 		if (c->devx_uid)
937 			return c->devx_uid;
938 
939 		dev = to_mdev(c->ibucontext.device);
940 		if (dev->devx_whitelist_uid)
941 			return dev->devx_whitelist_uid;
942 
943 		return -EOPNOTSUPP;
944 	}
945 
946 	if (!c->devx_uid)
947 		return -EINVAL;
948 
949 	return c->devx_uid;
950 }
951 
952 static bool devx_is_general_cmd(void *in, struct mlx5_ib_dev *dev)
953 {
954 	u16 opcode = MLX5_GET(general_obj_in_cmd_hdr, in, opcode);
955 
956 	/* Pass all cmds for vhca_tunnel as general, tracking is done in FW */
957 	if ((MLX5_CAP_GEN_64(dev->mdev, vhca_tunnel_commands) &&
958 	     MLX5_GET(general_obj_in_cmd_hdr, in, vhca_tunnel_id)) ||
959 	    (opcode >= MLX5_CMD_OP_GENERAL_START &&
960 	     opcode < MLX5_CMD_OP_GENERAL_END))
961 		return true;
962 
963 	switch (opcode) {
964 	case MLX5_CMD_OP_QUERY_HCA_CAP:
965 	case MLX5_CMD_OP_QUERY_HCA_VPORT_CONTEXT:
966 	case MLX5_CMD_OP_QUERY_ESW_VPORT_CONTEXT:
967 	case MLX5_CMD_OP_QUERY_VPORT_STATE:
968 	case MLX5_CMD_OP_QUERY_ADAPTER:
969 	case MLX5_CMD_OP_QUERY_ISSI:
970 	case MLX5_CMD_OP_QUERY_NIC_VPORT_CONTEXT:
971 	case MLX5_CMD_OP_QUERY_ROCE_ADDRESS:
972 	case MLX5_CMD_OP_QUERY_VNIC_ENV:
973 	case MLX5_CMD_OP_QUERY_VPORT_COUNTER:
974 	case MLX5_CMD_OP_GET_DROPPED_PACKET_LOG:
975 	case MLX5_CMD_OP_NOP:
976 	case MLX5_CMD_OP_QUERY_CONG_STATUS:
977 	case MLX5_CMD_OP_QUERY_CONG_PARAMS:
978 	case MLX5_CMD_OP_QUERY_CONG_STATISTICS:
979 	case MLX5_CMD_OP_QUERY_LAG:
980 	case MLX5_CMD_OP_QUERY_ESW_FUNCTIONS:
981 		return true;
982 	default:
983 		return false;
984 	}
985 }
986 
987 static int UVERBS_HANDLER(MLX5_IB_METHOD_DEVX_QUERY_EQN)(
988 	struct uverbs_attr_bundle *attrs)
989 {
990 	struct mlx5_ib_ucontext *c;
991 	struct mlx5_ib_dev *dev;
992 	int user_vector;
993 	int dev_eqn;
994 	int err;
995 
996 	if (uverbs_copy_from(&user_vector, attrs,
997 			     MLX5_IB_ATTR_DEVX_QUERY_EQN_USER_VEC))
998 		return -EFAULT;
999 
1000 	c = devx_ufile2uctx(attrs);
1001 	if (IS_ERR(c))
1002 		return PTR_ERR(c);
1003 	dev = to_mdev(c->ibucontext.device);
1004 
1005 	err = mlx5_comp_eqn_get(dev->mdev, user_vector, &dev_eqn);
1006 	if (err < 0)
1007 		return err;
1008 
1009 	if (uverbs_copy_to(attrs, MLX5_IB_ATTR_DEVX_QUERY_EQN_DEV_EQN,
1010 			   &dev_eqn, sizeof(dev_eqn)))
1011 		return -EFAULT;
1012 
1013 	return 0;
1014 }
1015 
1016 /*
1017  *Security note:
1018  * The hardware protection mechanism works like this: Each device object that
1019  * is subject to UAR doorbells (QP/SQ/CQ) gets a UAR ID (called uar_page in
1020  * the device specification manual) upon its creation. Then upon doorbell,
1021  * hardware fetches the object context for which the doorbell was rang, and
1022  * validates that the UAR through which the DB was rang matches the UAR ID
1023  * of the object.
1024  * If no match the doorbell is silently ignored by the hardware. Of course,
1025  * the user cannot ring a doorbell on a UAR that was not mapped to it.
1026  * Now in devx, as the devx kernel does not manipulate the QP/SQ/CQ command
1027  * mailboxes (except tagging them with UID), we expose to the user its UAR
1028  * ID, so it can embed it in these objects in the expected specification
1029  * format. So the only thing the user can do is hurt itself by creating a
1030  * QP/SQ/CQ with a UAR ID other than his, and then in this case other users
1031  * may ring a doorbell on its objects.
1032  * The consequence of that will be that another user can schedule a QP/SQ
1033  * of the buggy user for execution (just insert it to the hardware schedule
1034  * queue or arm its CQ for event generation), no further harm is expected.
1035  */
1036 static int UVERBS_HANDLER(MLX5_IB_METHOD_DEVX_QUERY_UAR)(
1037 	struct uverbs_attr_bundle *attrs)
1038 {
1039 	struct mlx5_ib_ucontext *c;
1040 	struct mlx5_ib_dev *dev;
1041 	u32 user_idx;
1042 	s32 dev_idx;
1043 
1044 	c = devx_ufile2uctx(attrs);
1045 	if (IS_ERR(c))
1046 		return PTR_ERR(c);
1047 	dev = to_mdev(c->ibucontext.device);
1048 
1049 	if (uverbs_copy_from(&user_idx, attrs,
1050 			     MLX5_IB_ATTR_DEVX_QUERY_UAR_USER_IDX))
1051 		return -EFAULT;
1052 
1053 	dev_idx = bfregn_to_uar_index(dev, &c->bfregi, user_idx, true);
1054 	if (dev_idx < 0)
1055 		return dev_idx;
1056 
1057 	if (uverbs_copy_to(attrs, MLX5_IB_ATTR_DEVX_QUERY_UAR_DEV_IDX,
1058 			   &dev_idx, sizeof(dev_idx)))
1059 		return -EFAULT;
1060 
1061 	return 0;
1062 }
1063 
1064 static int UVERBS_HANDLER(MLX5_IB_METHOD_DEVX_OTHER)(
1065 	struct uverbs_attr_bundle *attrs)
1066 {
1067 	struct mlx5_ib_ucontext *c;
1068 	struct mlx5_ib_dev *dev;
1069 	void *cmd_in = uverbs_attr_get_alloced_ptr(
1070 		attrs, MLX5_IB_ATTR_DEVX_OTHER_CMD_IN);
1071 	int cmd_out_len = uverbs_attr_get_len(attrs,
1072 					MLX5_IB_ATTR_DEVX_OTHER_CMD_OUT);
1073 	void *cmd_out;
1074 	int err, err2;
1075 	int uid;
1076 
1077 	c = devx_ufile2uctx(attrs);
1078 	if (IS_ERR(c))
1079 		return PTR_ERR(c);
1080 	dev = to_mdev(c->ibucontext.device);
1081 
1082 	uid = devx_get_uid(c, cmd_in);
1083 	if (uid < 0)
1084 		return uid;
1085 
1086 	/* Only white list of some general HCA commands are allowed for this method. */
1087 	if (!devx_is_general_cmd(cmd_in, dev))
1088 		return -EINVAL;
1089 
1090 	cmd_out = uverbs_zalloc(attrs, cmd_out_len);
1091 	if (IS_ERR(cmd_out))
1092 		return PTR_ERR(cmd_out);
1093 
1094 	MLX5_SET(general_obj_in_cmd_hdr, cmd_in, uid, uid);
1095 	err = mlx5_cmd_do(dev->mdev, cmd_in,
1096 			  uverbs_attr_get_len(attrs, MLX5_IB_ATTR_DEVX_OTHER_CMD_IN),
1097 			  cmd_out, cmd_out_len);
1098 	if (err && err != -EREMOTEIO)
1099 		return err;
1100 
1101 	err2 = uverbs_copy_to(attrs, MLX5_IB_ATTR_DEVX_OTHER_CMD_OUT, cmd_out,
1102 			      cmd_out_len);
1103 
1104 	return err2 ?: err;
1105 }
1106 
1107 static void devx_obj_build_destroy_cmd(void *in, void *out, void *din,
1108 				       u32 *dinlen,
1109 				       u32 *obj_id)
1110 {
1111 	u16 opcode = MLX5_GET(general_obj_in_cmd_hdr, in, opcode);
1112 	u16 uid = MLX5_GET(general_obj_in_cmd_hdr, in, uid);
1113 
1114 	*obj_id = devx_get_created_obj_id(in, out, opcode);
1115 	*dinlen = MLX5_ST_SZ_BYTES(general_obj_in_cmd_hdr);
1116 	MLX5_SET(general_obj_in_cmd_hdr, din, uid, uid);
1117 
1118 	switch (opcode) {
1119 	case MLX5_CMD_OP_CREATE_GENERAL_OBJECT:
1120 		MLX5_SET(general_obj_in_cmd_hdr, din, opcode, MLX5_CMD_OP_DESTROY_GENERAL_OBJECT);
1121 		MLX5_SET(general_obj_in_cmd_hdr, din, obj_id, *obj_id);
1122 		MLX5_SET(general_obj_in_cmd_hdr, din, obj_type,
1123 			 MLX5_GET(general_obj_in_cmd_hdr, in, obj_type));
1124 		break;
1125 
1126 	case MLX5_CMD_OP_CREATE_UMEM:
1127 		MLX5_SET(destroy_umem_in, din, opcode,
1128 			 MLX5_CMD_OP_DESTROY_UMEM);
1129 		MLX5_SET(destroy_umem_in, din, umem_id, *obj_id);
1130 		break;
1131 	case MLX5_CMD_OP_CREATE_MKEY:
1132 		MLX5_SET(destroy_mkey_in, din, opcode,
1133 			 MLX5_CMD_OP_DESTROY_MKEY);
1134 		MLX5_SET(destroy_mkey_in, din, mkey_index, *obj_id);
1135 		break;
1136 	case MLX5_CMD_OP_CREATE_CQ:
1137 		MLX5_SET(destroy_cq_in, din, opcode, MLX5_CMD_OP_DESTROY_CQ);
1138 		MLX5_SET(destroy_cq_in, din, cqn, *obj_id);
1139 		break;
1140 	case MLX5_CMD_OP_ALLOC_PD:
1141 		MLX5_SET(dealloc_pd_in, din, opcode, MLX5_CMD_OP_DEALLOC_PD);
1142 		MLX5_SET(dealloc_pd_in, din, pd, *obj_id);
1143 		break;
1144 	case MLX5_CMD_OP_ALLOC_TRANSPORT_DOMAIN:
1145 		MLX5_SET(dealloc_transport_domain_in, din, opcode,
1146 			 MLX5_CMD_OP_DEALLOC_TRANSPORT_DOMAIN);
1147 		MLX5_SET(dealloc_transport_domain_in, din, transport_domain,
1148 			 *obj_id);
1149 		break;
1150 	case MLX5_CMD_OP_CREATE_RMP:
1151 		MLX5_SET(destroy_rmp_in, din, opcode, MLX5_CMD_OP_DESTROY_RMP);
1152 		MLX5_SET(destroy_rmp_in, din, rmpn, *obj_id);
1153 		break;
1154 	case MLX5_CMD_OP_CREATE_SQ:
1155 		MLX5_SET(destroy_sq_in, din, opcode, MLX5_CMD_OP_DESTROY_SQ);
1156 		MLX5_SET(destroy_sq_in, din, sqn, *obj_id);
1157 		break;
1158 	case MLX5_CMD_OP_CREATE_RQ:
1159 		MLX5_SET(destroy_rq_in, din, opcode, MLX5_CMD_OP_DESTROY_RQ);
1160 		MLX5_SET(destroy_rq_in, din, rqn, *obj_id);
1161 		break;
1162 	case MLX5_CMD_OP_CREATE_RQT:
1163 		MLX5_SET(destroy_rqt_in, din, opcode, MLX5_CMD_OP_DESTROY_RQT);
1164 		MLX5_SET(destroy_rqt_in, din, rqtn, *obj_id);
1165 		break;
1166 	case MLX5_CMD_OP_CREATE_TIR:
1167 		MLX5_SET(destroy_tir_in, din, opcode, MLX5_CMD_OP_DESTROY_TIR);
1168 		MLX5_SET(destroy_tir_in, din, tirn, *obj_id);
1169 		break;
1170 	case MLX5_CMD_OP_CREATE_TIS:
1171 		MLX5_SET(destroy_tis_in, din, opcode, MLX5_CMD_OP_DESTROY_TIS);
1172 		MLX5_SET(destroy_tis_in, din, tisn, *obj_id);
1173 		break;
1174 	case MLX5_CMD_OP_ALLOC_Q_COUNTER:
1175 		MLX5_SET(dealloc_q_counter_in, din, opcode,
1176 			 MLX5_CMD_OP_DEALLOC_Q_COUNTER);
1177 		MLX5_SET(dealloc_q_counter_in, din, counter_set_id, *obj_id);
1178 		break;
1179 	case MLX5_CMD_OP_CREATE_FLOW_TABLE:
1180 		*dinlen = MLX5_ST_SZ_BYTES(destroy_flow_table_in);
1181 		MLX5_SET(destroy_flow_table_in, din, other_vport,
1182 			 MLX5_GET(create_flow_table_in,  in, other_vport));
1183 		MLX5_SET(destroy_flow_table_in, din, vport_number,
1184 			 MLX5_GET(create_flow_table_in,  in, vport_number));
1185 		MLX5_SET(destroy_flow_table_in, din, table_type,
1186 			 MLX5_GET(create_flow_table_in,  in, table_type));
1187 		MLX5_SET(destroy_flow_table_in, din, table_id, *obj_id);
1188 		MLX5_SET(destroy_flow_table_in, din, opcode,
1189 			 MLX5_CMD_OP_DESTROY_FLOW_TABLE);
1190 		break;
1191 	case MLX5_CMD_OP_CREATE_FLOW_GROUP:
1192 		*dinlen = MLX5_ST_SZ_BYTES(destroy_flow_group_in);
1193 		MLX5_SET(destroy_flow_group_in, din, other_vport,
1194 			 MLX5_GET(create_flow_group_in, in, other_vport));
1195 		MLX5_SET(destroy_flow_group_in, din, vport_number,
1196 			 MLX5_GET(create_flow_group_in, in, vport_number));
1197 		MLX5_SET(destroy_flow_group_in, din, table_type,
1198 			 MLX5_GET(create_flow_group_in, in, table_type));
1199 		MLX5_SET(destroy_flow_group_in, din, table_id,
1200 			 MLX5_GET(create_flow_group_in, in, table_id));
1201 		MLX5_SET(destroy_flow_group_in, din, group_id, *obj_id);
1202 		MLX5_SET(destroy_flow_group_in, din, opcode,
1203 			 MLX5_CMD_OP_DESTROY_FLOW_GROUP);
1204 		break;
1205 	case MLX5_CMD_OP_SET_FLOW_TABLE_ENTRY:
1206 		*dinlen = MLX5_ST_SZ_BYTES(delete_fte_in);
1207 		MLX5_SET(delete_fte_in, din, other_vport,
1208 			 MLX5_GET(set_fte_in,  in, other_vport));
1209 		MLX5_SET(delete_fte_in, din, vport_number,
1210 			 MLX5_GET(set_fte_in, in, vport_number));
1211 		MLX5_SET(delete_fte_in, din, table_type,
1212 			 MLX5_GET(set_fte_in, in, table_type));
1213 		MLX5_SET(delete_fte_in, din, table_id,
1214 			 MLX5_GET(set_fte_in, in, table_id));
1215 		MLX5_SET(delete_fte_in, din, flow_index, *obj_id);
1216 		MLX5_SET(delete_fte_in, din, opcode,
1217 			 MLX5_CMD_OP_DELETE_FLOW_TABLE_ENTRY);
1218 		break;
1219 	case MLX5_CMD_OP_ALLOC_FLOW_COUNTER:
1220 		MLX5_SET(dealloc_flow_counter_in, din, opcode,
1221 			 MLX5_CMD_OP_DEALLOC_FLOW_COUNTER);
1222 		MLX5_SET(dealloc_flow_counter_in, din, flow_counter_id,
1223 			 *obj_id);
1224 		break;
1225 	case MLX5_CMD_OP_ALLOC_PACKET_REFORMAT_CONTEXT:
1226 		MLX5_SET(dealloc_packet_reformat_context_in, din, opcode,
1227 			 MLX5_CMD_OP_DEALLOC_PACKET_REFORMAT_CONTEXT);
1228 		MLX5_SET(dealloc_packet_reformat_context_in, din,
1229 			 packet_reformat_id, *obj_id);
1230 		break;
1231 	case MLX5_CMD_OP_ALLOC_MODIFY_HEADER_CONTEXT:
1232 		MLX5_SET(dealloc_modify_header_context_in, din, opcode,
1233 			 MLX5_CMD_OP_DEALLOC_MODIFY_HEADER_CONTEXT);
1234 		MLX5_SET(dealloc_modify_header_context_in, din,
1235 			 modify_header_id, *obj_id);
1236 		break;
1237 	case MLX5_CMD_OP_CREATE_SCHEDULING_ELEMENT:
1238 		*dinlen = MLX5_ST_SZ_BYTES(destroy_scheduling_element_in);
1239 		MLX5_SET(destroy_scheduling_element_in, din,
1240 			 scheduling_hierarchy,
1241 			 MLX5_GET(create_scheduling_element_in, in,
1242 				  scheduling_hierarchy));
1243 		MLX5_SET(destroy_scheduling_element_in, din,
1244 			 scheduling_element_id, *obj_id);
1245 		MLX5_SET(destroy_scheduling_element_in, din, opcode,
1246 			 MLX5_CMD_OP_DESTROY_SCHEDULING_ELEMENT);
1247 		break;
1248 	case MLX5_CMD_OP_ADD_VXLAN_UDP_DPORT:
1249 		*dinlen = MLX5_ST_SZ_BYTES(delete_vxlan_udp_dport_in);
1250 		MLX5_SET(delete_vxlan_udp_dport_in, din, vxlan_udp_port, *obj_id);
1251 		MLX5_SET(delete_vxlan_udp_dport_in, din, opcode,
1252 			 MLX5_CMD_OP_DELETE_VXLAN_UDP_DPORT);
1253 		break;
1254 	case MLX5_CMD_OP_SET_L2_TABLE_ENTRY:
1255 		*dinlen = MLX5_ST_SZ_BYTES(delete_l2_table_entry_in);
1256 		MLX5_SET(delete_l2_table_entry_in, din, table_index, *obj_id);
1257 		MLX5_SET(delete_l2_table_entry_in, din, opcode,
1258 			 MLX5_CMD_OP_DELETE_L2_TABLE_ENTRY);
1259 		break;
1260 	case MLX5_CMD_OP_CREATE_QP:
1261 		MLX5_SET(destroy_qp_in, din, opcode, MLX5_CMD_OP_DESTROY_QP);
1262 		MLX5_SET(destroy_qp_in, din, qpn, *obj_id);
1263 		break;
1264 	case MLX5_CMD_OP_CREATE_SRQ:
1265 		MLX5_SET(destroy_srq_in, din, opcode, MLX5_CMD_OP_DESTROY_SRQ);
1266 		MLX5_SET(destroy_srq_in, din, srqn, *obj_id);
1267 		break;
1268 	case MLX5_CMD_OP_CREATE_XRC_SRQ:
1269 		MLX5_SET(destroy_xrc_srq_in, din, opcode,
1270 			 MLX5_CMD_OP_DESTROY_XRC_SRQ);
1271 		MLX5_SET(destroy_xrc_srq_in, din, xrc_srqn, *obj_id);
1272 		break;
1273 	case MLX5_CMD_OP_CREATE_DCT:
1274 		MLX5_SET(destroy_dct_in, din, opcode, MLX5_CMD_OP_DESTROY_DCT);
1275 		MLX5_SET(destroy_dct_in, din, dctn, *obj_id);
1276 		break;
1277 	case MLX5_CMD_OP_CREATE_XRQ:
1278 		MLX5_SET(destroy_xrq_in, din, opcode, MLX5_CMD_OP_DESTROY_XRQ);
1279 		MLX5_SET(destroy_xrq_in, din, xrqn, *obj_id);
1280 		break;
1281 	case MLX5_CMD_OP_ATTACH_TO_MCG:
1282 		*dinlen = MLX5_ST_SZ_BYTES(detach_from_mcg_in);
1283 		MLX5_SET(detach_from_mcg_in, din, qpn,
1284 			 MLX5_GET(attach_to_mcg_in, in, qpn));
1285 		memcpy(MLX5_ADDR_OF(detach_from_mcg_in, din, multicast_gid),
1286 		       MLX5_ADDR_OF(attach_to_mcg_in, in, multicast_gid),
1287 		       MLX5_FLD_SZ_BYTES(attach_to_mcg_in, multicast_gid));
1288 		MLX5_SET(detach_from_mcg_in, din, opcode,
1289 			 MLX5_CMD_OP_DETACH_FROM_MCG);
1290 		MLX5_SET(detach_from_mcg_in, din, qpn, *obj_id);
1291 		break;
1292 	case MLX5_CMD_OP_ALLOC_XRCD:
1293 		MLX5_SET(dealloc_xrcd_in, din, opcode,
1294 			 MLX5_CMD_OP_DEALLOC_XRCD);
1295 		MLX5_SET(dealloc_xrcd_in, din, xrcd, *obj_id);
1296 		break;
1297 	case MLX5_CMD_OP_CREATE_PSV:
1298 		MLX5_SET(destroy_psv_in, din, opcode,
1299 			 MLX5_CMD_OP_DESTROY_PSV);
1300 		MLX5_SET(destroy_psv_in, din, psvn, *obj_id);
1301 		break;
1302 	default:
1303 		/* The entry must match to one of the devx_is_obj_create_cmd */
1304 		WARN_ON(true);
1305 		break;
1306 	}
1307 }
1308 
1309 static int devx_handle_mkey_indirect(struct devx_obj *obj,
1310 				     struct mlx5_ib_dev *dev,
1311 				     void *in, void *out)
1312 {
1313 	struct mlx5_ib_mkey *mkey = &obj->mkey;
1314 	void *mkc;
1315 	u8 key;
1316 
1317 	mkc = MLX5_ADDR_OF(create_mkey_in, in, memory_key_mkey_entry);
1318 	key = MLX5_GET(mkc, mkc, mkey_7_0);
1319 	mkey->key = mlx5_idx_to_mkey(
1320 			MLX5_GET(create_mkey_out, out, mkey_index)) | key;
1321 	mkey->type = MLX5_MKEY_INDIRECT_DEVX;
1322 	mkey->ndescs = MLX5_GET(mkc, mkc, translations_octword_size);
1323 	init_waitqueue_head(&mkey->wait);
1324 
1325 	return mlx5r_store_odp_mkey(dev, mkey);
1326 }
1327 
1328 static int devx_handle_mkey_create(struct mlx5_ib_dev *dev,
1329 				   struct devx_obj *obj,
1330 				   void *in, int in_len)
1331 {
1332 	int min_len = MLX5_BYTE_OFF(create_mkey_in, memory_key_mkey_entry) +
1333 			MLX5_FLD_SZ_BYTES(create_mkey_in,
1334 			memory_key_mkey_entry);
1335 	void *mkc;
1336 	u8 access_mode;
1337 
1338 	if (in_len < min_len)
1339 		return -EINVAL;
1340 
1341 	mkc = MLX5_ADDR_OF(create_mkey_in, in, memory_key_mkey_entry);
1342 
1343 	access_mode = MLX5_GET(mkc, mkc, access_mode_1_0);
1344 	access_mode |= MLX5_GET(mkc, mkc, access_mode_4_2) << 2;
1345 
1346 	if (access_mode == MLX5_MKC_ACCESS_MODE_KLMS ||
1347 		access_mode == MLX5_MKC_ACCESS_MODE_KSM) {
1348 		if (IS_ENABLED(CONFIG_INFINIBAND_ON_DEMAND_PAGING))
1349 			obj->flags |= DEVX_OBJ_FLAGS_INDIRECT_MKEY;
1350 		return 0;
1351 	}
1352 
1353 	MLX5_SET(create_mkey_in, in, mkey_umem_valid, 1);
1354 	return 0;
1355 }
1356 
1357 static void devx_cleanup_subscription(struct mlx5_ib_dev *dev,
1358 				      struct devx_event_subscription *sub)
1359 {
1360 	struct devx_event *event;
1361 	struct devx_obj_event *xa_val_level2;
1362 
1363 	if (sub->is_cleaned)
1364 		return;
1365 
1366 	sub->is_cleaned = 1;
1367 	list_del_rcu(&sub->xa_list);
1368 
1369 	if (list_empty(&sub->obj_list))
1370 		return;
1371 
1372 	list_del_rcu(&sub->obj_list);
1373 	/* check whether key level 1 for this obj_sub_list is empty */
1374 	event = xa_load(&dev->devx_event_table.event_xa,
1375 			sub->xa_key_level1);
1376 	WARN_ON(!event);
1377 
1378 	xa_val_level2 = xa_load(&event->object_ids, sub->xa_key_level2);
1379 	if (list_empty(&xa_val_level2->obj_sub_list)) {
1380 		xa_erase(&event->object_ids,
1381 			 sub->xa_key_level2);
1382 		kfree_rcu(xa_val_level2, rcu);
1383 	}
1384 }
1385 
1386 static int devx_obj_cleanup(struct ib_uobject *uobject,
1387 			    enum rdma_remove_reason why,
1388 			    struct uverbs_attr_bundle *attrs)
1389 {
1390 	u32 out[MLX5_ST_SZ_DW(general_obj_out_cmd_hdr)];
1391 	struct mlx5_devx_event_table *devx_event_table;
1392 	struct devx_obj *obj = uobject->object;
1393 	struct devx_event_subscription *sub_entry, *tmp;
1394 	struct mlx5_ib_dev *dev;
1395 	int ret;
1396 
1397 	dev = mlx5_udata_to_mdev(&attrs->driver_udata);
1398 	if (obj->flags & DEVX_OBJ_FLAGS_INDIRECT_MKEY &&
1399 	    xa_erase(&obj->ib_dev->odp_mkeys,
1400 		     mlx5_base_mkey(obj->mkey.key)))
1401 		/*
1402 		 * The pagefault_single_data_segment() does commands against
1403 		 * the mmkey, we must wait for that to stop before freeing the
1404 		 * mkey, as another allocation could get the same mkey #.
1405 		 */
1406 		mlx5r_deref_wait_odp_mkey(&obj->mkey);
1407 
1408 	if (obj->flags & DEVX_OBJ_FLAGS_DCT)
1409 		ret = mlx5_core_destroy_dct(obj->ib_dev, &obj->core_dct);
1410 	else if (obj->flags & DEVX_OBJ_FLAGS_CQ)
1411 		ret = mlx5_core_destroy_cq(obj->ib_dev->mdev, &obj->core_cq);
1412 	else
1413 		ret = mlx5_cmd_exec(obj->ib_dev->mdev, obj->dinbox,
1414 				    obj->dinlen, out, sizeof(out));
1415 	if (ret)
1416 		return ret;
1417 
1418 	devx_event_table = &dev->devx_event_table;
1419 
1420 	mutex_lock(&devx_event_table->event_xa_lock);
1421 	list_for_each_entry_safe(sub_entry, tmp, &obj->event_sub, obj_list)
1422 		devx_cleanup_subscription(dev, sub_entry);
1423 	mutex_unlock(&devx_event_table->event_xa_lock);
1424 
1425 	kfree(obj);
1426 	return ret;
1427 }
1428 
1429 static void devx_cq_comp(struct mlx5_core_cq *mcq, struct mlx5_eqe *eqe)
1430 {
1431 	struct devx_obj *obj = container_of(mcq, struct devx_obj, core_cq);
1432 	struct mlx5_devx_event_table *table;
1433 	struct devx_event *event;
1434 	struct devx_obj_event *obj_event;
1435 	u32 obj_id = mcq->cqn;
1436 
1437 	table = &obj->ib_dev->devx_event_table;
1438 	rcu_read_lock();
1439 	event = xa_load(&table->event_xa, MLX5_EVENT_TYPE_COMP);
1440 	if (!event)
1441 		goto out;
1442 
1443 	obj_event = xa_load(&event->object_ids, obj_id);
1444 	if (!obj_event)
1445 		goto out;
1446 
1447 	dispatch_event_fd(&obj_event->obj_sub_list, eqe);
1448 out:
1449 	rcu_read_unlock();
1450 }
1451 
1452 static bool is_apu_cq(struct mlx5_ib_dev *dev, const void *in)
1453 {
1454 	if (!MLX5_CAP_GEN(dev->mdev, apu) ||
1455 	    !MLX5_GET(cqc, MLX5_ADDR_OF(create_cq_in, in, cq_context), apu_cq))
1456 		return false;
1457 
1458 	return true;
1459 }
1460 
1461 static int UVERBS_HANDLER(MLX5_IB_METHOD_DEVX_OBJ_CREATE)(
1462 	struct uverbs_attr_bundle *attrs)
1463 {
1464 	void *cmd_in = uverbs_attr_get_alloced_ptr(attrs, MLX5_IB_ATTR_DEVX_OBJ_CREATE_CMD_IN);
1465 	int cmd_out_len =  uverbs_attr_get_len(attrs,
1466 					MLX5_IB_ATTR_DEVX_OBJ_CREATE_CMD_OUT);
1467 	int cmd_in_len = uverbs_attr_get_len(attrs,
1468 					MLX5_IB_ATTR_DEVX_OBJ_CREATE_CMD_IN);
1469 	void *cmd_out;
1470 	struct ib_uobject *uobj = uverbs_attr_get_uobject(
1471 		attrs, MLX5_IB_ATTR_DEVX_OBJ_CREATE_HANDLE);
1472 	struct mlx5_ib_ucontext *c = rdma_udata_to_drv_context(
1473 		&attrs->driver_udata, struct mlx5_ib_ucontext, ibucontext);
1474 	struct mlx5_ib_dev *dev = to_mdev(c->ibucontext.device);
1475 	u32 out[MLX5_ST_SZ_DW(general_obj_out_cmd_hdr)];
1476 	struct devx_obj *obj;
1477 	u16 obj_type = 0;
1478 	int err, err2 = 0;
1479 	int uid;
1480 	u32 obj_id;
1481 	u16 opcode;
1482 
1483 	if (MLX5_GET(general_obj_in_cmd_hdr, cmd_in, vhca_tunnel_id))
1484 		return -EINVAL;
1485 
1486 	uid = devx_get_uid(c, cmd_in);
1487 	if (uid < 0)
1488 		return uid;
1489 
1490 	if (!devx_is_obj_create_cmd(cmd_in, &opcode))
1491 		return -EINVAL;
1492 
1493 	cmd_out = uverbs_zalloc(attrs, cmd_out_len);
1494 	if (IS_ERR(cmd_out))
1495 		return PTR_ERR(cmd_out);
1496 
1497 	obj = kzalloc(sizeof(struct devx_obj), GFP_KERNEL);
1498 	if (!obj)
1499 		return -ENOMEM;
1500 
1501 	MLX5_SET(general_obj_in_cmd_hdr, cmd_in, uid, uid);
1502 	if (opcode == MLX5_CMD_OP_CREATE_MKEY) {
1503 		err = devx_handle_mkey_create(dev, obj, cmd_in, cmd_in_len);
1504 		if (err)
1505 			goto obj_free;
1506 	} else {
1507 		devx_set_umem_valid(cmd_in);
1508 	}
1509 
1510 	if (opcode == MLX5_CMD_OP_CREATE_DCT) {
1511 		obj->flags |= DEVX_OBJ_FLAGS_DCT;
1512 		err = mlx5_core_create_dct(dev, &obj->core_dct, cmd_in,
1513 					   cmd_in_len, cmd_out, cmd_out_len);
1514 	} else if (opcode == MLX5_CMD_OP_CREATE_CQ &&
1515 		   !is_apu_cq(dev, cmd_in)) {
1516 		obj->flags |= DEVX_OBJ_FLAGS_CQ;
1517 		obj->core_cq.comp = devx_cq_comp;
1518 		err = mlx5_create_cq(dev->mdev, &obj->core_cq,
1519 				     cmd_in, cmd_in_len, cmd_out,
1520 				     cmd_out_len);
1521 	} else {
1522 		err = mlx5_cmd_do(dev->mdev, cmd_in, cmd_in_len,
1523 				  cmd_out, cmd_out_len);
1524 	}
1525 
1526 	if (err == -EREMOTEIO)
1527 		err2 = uverbs_copy_to(attrs,
1528 				      MLX5_IB_ATTR_DEVX_OBJ_CREATE_CMD_OUT,
1529 				      cmd_out, cmd_out_len);
1530 	if (err)
1531 		goto obj_free;
1532 
1533 	if (opcode == MLX5_CMD_OP_ALLOC_FLOW_COUNTER) {
1534 		u32 bulk = MLX5_GET(alloc_flow_counter_in,
1535 				    cmd_in,
1536 				    flow_counter_bulk_log_size);
1537 
1538 		if (bulk)
1539 			bulk = 1 << bulk;
1540 		else
1541 			bulk = 128UL * MLX5_GET(alloc_flow_counter_in,
1542 						cmd_in,
1543 						flow_counter_bulk);
1544 		obj->flow_counter_bulk_size = bulk;
1545 	}
1546 
1547 	uobj->object = obj;
1548 	INIT_LIST_HEAD(&obj->event_sub);
1549 	obj->ib_dev = dev;
1550 	devx_obj_build_destroy_cmd(cmd_in, cmd_out, obj->dinbox, &obj->dinlen,
1551 				   &obj_id);
1552 	WARN_ON(obj->dinlen > MLX5_MAX_DESTROY_INBOX_SIZE_DW * sizeof(u32));
1553 
1554 	err = uverbs_copy_to(attrs, MLX5_IB_ATTR_DEVX_OBJ_CREATE_CMD_OUT, cmd_out, cmd_out_len);
1555 	if (err)
1556 		goto obj_destroy;
1557 
1558 	if (opcode == MLX5_CMD_OP_CREATE_GENERAL_OBJECT)
1559 		obj_type = MLX5_GET(general_obj_in_cmd_hdr, cmd_in, obj_type);
1560 	obj->obj_id = get_enc_obj_id(opcode | obj_type << 16, obj_id);
1561 
1562 	if (obj->flags & DEVX_OBJ_FLAGS_INDIRECT_MKEY) {
1563 		err = devx_handle_mkey_indirect(obj, dev, cmd_in, cmd_out);
1564 		if (err)
1565 			goto obj_destroy;
1566 	}
1567 	return 0;
1568 
1569 obj_destroy:
1570 	if (obj->flags & DEVX_OBJ_FLAGS_DCT)
1571 		mlx5_core_destroy_dct(obj->ib_dev, &obj->core_dct);
1572 	else if (obj->flags & DEVX_OBJ_FLAGS_CQ)
1573 		mlx5_core_destroy_cq(obj->ib_dev->mdev, &obj->core_cq);
1574 	else
1575 		mlx5_cmd_exec(obj->ib_dev->mdev, obj->dinbox, obj->dinlen, out,
1576 			      sizeof(out));
1577 obj_free:
1578 	kfree(obj);
1579 	return err2 ?: err;
1580 }
1581 
1582 static int UVERBS_HANDLER(MLX5_IB_METHOD_DEVX_OBJ_MODIFY)(
1583 	struct uverbs_attr_bundle *attrs)
1584 {
1585 	void *cmd_in = uverbs_attr_get_alloced_ptr(attrs, MLX5_IB_ATTR_DEVX_OBJ_MODIFY_CMD_IN);
1586 	int cmd_out_len = uverbs_attr_get_len(attrs,
1587 					MLX5_IB_ATTR_DEVX_OBJ_MODIFY_CMD_OUT);
1588 	struct ib_uobject *uobj = uverbs_attr_get_uobject(attrs,
1589 							  MLX5_IB_ATTR_DEVX_OBJ_MODIFY_HANDLE);
1590 	struct mlx5_ib_ucontext *c = rdma_udata_to_drv_context(
1591 		&attrs->driver_udata, struct mlx5_ib_ucontext, ibucontext);
1592 	struct mlx5_ib_dev *mdev = to_mdev(c->ibucontext.device);
1593 	void *cmd_out;
1594 	int err, err2;
1595 	int uid;
1596 
1597 	if (MLX5_GET(general_obj_in_cmd_hdr, cmd_in, vhca_tunnel_id))
1598 		return -EINVAL;
1599 
1600 	uid = devx_get_uid(c, cmd_in);
1601 	if (uid < 0)
1602 		return uid;
1603 
1604 	if (!devx_is_obj_modify_cmd(cmd_in))
1605 		return -EINVAL;
1606 
1607 	if (!devx_is_valid_obj_id(attrs, uobj, cmd_in))
1608 		return -EINVAL;
1609 
1610 	cmd_out = uverbs_zalloc(attrs, cmd_out_len);
1611 	if (IS_ERR(cmd_out))
1612 		return PTR_ERR(cmd_out);
1613 
1614 	MLX5_SET(general_obj_in_cmd_hdr, cmd_in, uid, uid);
1615 	devx_set_umem_valid(cmd_in);
1616 
1617 	err = mlx5_cmd_do(mdev->mdev, cmd_in,
1618 			  uverbs_attr_get_len(attrs, MLX5_IB_ATTR_DEVX_OBJ_MODIFY_CMD_IN),
1619 			  cmd_out, cmd_out_len);
1620 	if (err && err != -EREMOTEIO)
1621 		return err;
1622 
1623 	err2 = uverbs_copy_to(attrs, MLX5_IB_ATTR_DEVX_OBJ_MODIFY_CMD_OUT,
1624 			      cmd_out, cmd_out_len);
1625 
1626 	return err2 ?: err;
1627 }
1628 
1629 static int UVERBS_HANDLER(MLX5_IB_METHOD_DEVX_OBJ_QUERY)(
1630 	struct uverbs_attr_bundle *attrs)
1631 {
1632 	void *cmd_in = uverbs_attr_get_alloced_ptr(attrs, MLX5_IB_ATTR_DEVX_OBJ_QUERY_CMD_IN);
1633 	int cmd_out_len = uverbs_attr_get_len(attrs,
1634 					      MLX5_IB_ATTR_DEVX_OBJ_QUERY_CMD_OUT);
1635 	struct ib_uobject *uobj = uverbs_attr_get_uobject(attrs,
1636 							  MLX5_IB_ATTR_DEVX_OBJ_QUERY_HANDLE);
1637 	struct mlx5_ib_ucontext *c = rdma_udata_to_drv_context(
1638 		&attrs->driver_udata, struct mlx5_ib_ucontext, ibucontext);
1639 	void *cmd_out;
1640 	int err, err2;
1641 	int uid;
1642 	struct mlx5_ib_dev *mdev = to_mdev(c->ibucontext.device);
1643 
1644 	if (MLX5_GET(general_obj_in_cmd_hdr, cmd_in, vhca_tunnel_id))
1645 		return -EINVAL;
1646 
1647 	uid = devx_get_uid(c, cmd_in);
1648 	if (uid < 0)
1649 		return uid;
1650 
1651 	if (!devx_is_obj_query_cmd(cmd_in))
1652 		return -EINVAL;
1653 
1654 	if (!devx_is_valid_obj_id(attrs, uobj, cmd_in))
1655 		return -EINVAL;
1656 
1657 	cmd_out = uverbs_zalloc(attrs, cmd_out_len);
1658 	if (IS_ERR(cmd_out))
1659 		return PTR_ERR(cmd_out);
1660 
1661 	MLX5_SET(general_obj_in_cmd_hdr, cmd_in, uid, uid);
1662 	err = mlx5_cmd_do(mdev->mdev, cmd_in,
1663 			  uverbs_attr_get_len(attrs, MLX5_IB_ATTR_DEVX_OBJ_QUERY_CMD_IN),
1664 			  cmd_out, cmd_out_len);
1665 	if (err && err != -EREMOTEIO)
1666 		return err;
1667 
1668 	err2 = uverbs_copy_to(attrs, MLX5_IB_ATTR_DEVX_OBJ_QUERY_CMD_OUT,
1669 			      cmd_out, cmd_out_len);
1670 
1671 	return err2 ?: err;
1672 }
1673 
1674 struct devx_async_event_queue {
1675 	spinlock_t		lock;
1676 	wait_queue_head_t	poll_wait;
1677 	struct list_head	event_list;
1678 	atomic_t		bytes_in_use;
1679 	u8			is_destroyed:1;
1680 };
1681 
1682 struct devx_async_cmd_event_file {
1683 	struct ib_uobject		uobj;
1684 	struct devx_async_event_queue	ev_queue;
1685 	struct mlx5_async_ctx		async_ctx;
1686 };
1687 
1688 static void devx_init_event_queue(struct devx_async_event_queue *ev_queue)
1689 {
1690 	spin_lock_init(&ev_queue->lock);
1691 	INIT_LIST_HEAD(&ev_queue->event_list);
1692 	init_waitqueue_head(&ev_queue->poll_wait);
1693 	atomic_set(&ev_queue->bytes_in_use, 0);
1694 	ev_queue->is_destroyed = 0;
1695 }
1696 
1697 static int UVERBS_HANDLER(MLX5_IB_METHOD_DEVX_ASYNC_CMD_FD_ALLOC)(
1698 	struct uverbs_attr_bundle *attrs)
1699 {
1700 	struct devx_async_cmd_event_file *ev_file;
1701 
1702 	struct ib_uobject *uobj = uverbs_attr_get_uobject(
1703 		attrs, MLX5_IB_ATTR_DEVX_ASYNC_CMD_FD_ALLOC_HANDLE);
1704 	struct mlx5_ib_dev *mdev = mlx5_udata_to_mdev(&attrs->driver_udata);
1705 
1706 	ev_file = container_of(uobj, struct devx_async_cmd_event_file,
1707 			       uobj);
1708 	devx_init_event_queue(&ev_file->ev_queue);
1709 	mlx5_cmd_init_async_ctx(mdev->mdev, &ev_file->async_ctx);
1710 	return 0;
1711 }
1712 
1713 static int UVERBS_HANDLER(MLX5_IB_METHOD_DEVX_ASYNC_EVENT_FD_ALLOC)(
1714 	struct uverbs_attr_bundle *attrs)
1715 {
1716 	struct ib_uobject *uobj = uverbs_attr_get_uobject(
1717 		attrs, MLX5_IB_ATTR_DEVX_ASYNC_EVENT_FD_ALLOC_HANDLE);
1718 	struct devx_async_event_file *ev_file;
1719 	struct mlx5_ib_ucontext *c = rdma_udata_to_drv_context(
1720 		&attrs->driver_udata, struct mlx5_ib_ucontext, ibucontext);
1721 	struct mlx5_ib_dev *dev = to_mdev(c->ibucontext.device);
1722 	u32 flags;
1723 	int err;
1724 
1725 	err = uverbs_get_flags32(&flags, attrs,
1726 		MLX5_IB_ATTR_DEVX_ASYNC_EVENT_FD_ALLOC_FLAGS,
1727 		MLX5_IB_UAPI_DEVX_CR_EV_CH_FLAGS_OMIT_DATA);
1728 
1729 	if (err)
1730 		return err;
1731 
1732 	ev_file = container_of(uobj, struct devx_async_event_file,
1733 			       uobj);
1734 	spin_lock_init(&ev_file->lock);
1735 	INIT_LIST_HEAD(&ev_file->event_list);
1736 	init_waitqueue_head(&ev_file->poll_wait);
1737 	if (flags & MLX5_IB_UAPI_DEVX_CR_EV_CH_FLAGS_OMIT_DATA)
1738 		ev_file->omit_data = 1;
1739 	INIT_LIST_HEAD(&ev_file->subscribed_events_list);
1740 	ev_file->dev = dev;
1741 	get_device(&dev->ib_dev.dev);
1742 	return 0;
1743 }
1744 
1745 static void devx_query_callback(int status, struct mlx5_async_work *context)
1746 {
1747 	struct devx_async_data *async_data =
1748 		container_of(context, struct devx_async_data, cb_work);
1749 	struct devx_async_cmd_event_file *ev_file = async_data->ev_file;
1750 	struct devx_async_event_queue *ev_queue = &ev_file->ev_queue;
1751 	unsigned long flags;
1752 
1753 	/*
1754 	 * Note that if the struct devx_async_cmd_event_file uobj begins to be
1755 	 * destroyed it will block at mlx5_cmd_cleanup_async_ctx() until this
1756 	 * routine returns, ensuring that it always remains valid here.
1757 	 */
1758 	spin_lock_irqsave(&ev_queue->lock, flags);
1759 	list_add_tail(&async_data->list, &ev_queue->event_list);
1760 	spin_unlock_irqrestore(&ev_queue->lock, flags);
1761 
1762 	wake_up_interruptible(&ev_queue->poll_wait);
1763 }
1764 
1765 #define MAX_ASYNC_BYTES_IN_USE (1024 * 1024) /* 1MB */
1766 
1767 static int UVERBS_HANDLER(MLX5_IB_METHOD_DEVX_OBJ_ASYNC_QUERY)(
1768 	struct uverbs_attr_bundle *attrs)
1769 {
1770 	void *cmd_in = uverbs_attr_get_alloced_ptr(attrs,
1771 				MLX5_IB_ATTR_DEVX_OBJ_QUERY_ASYNC_CMD_IN);
1772 	struct ib_uobject *uobj = uverbs_attr_get_uobject(
1773 				attrs,
1774 				MLX5_IB_ATTR_DEVX_OBJ_QUERY_ASYNC_HANDLE);
1775 	u16 cmd_out_len;
1776 	struct mlx5_ib_ucontext *c = rdma_udata_to_drv_context(
1777 		&attrs->driver_udata, struct mlx5_ib_ucontext, ibucontext);
1778 	struct ib_uobject *fd_uobj;
1779 	int err;
1780 	int uid;
1781 	struct mlx5_ib_dev *mdev = to_mdev(c->ibucontext.device);
1782 	struct devx_async_cmd_event_file *ev_file;
1783 	struct devx_async_data *async_data;
1784 
1785 	if (MLX5_GET(general_obj_in_cmd_hdr, cmd_in, vhca_tunnel_id))
1786 		return -EINVAL;
1787 
1788 	uid = devx_get_uid(c, cmd_in);
1789 	if (uid < 0)
1790 		return uid;
1791 
1792 	if (!devx_is_obj_query_cmd(cmd_in))
1793 		return -EINVAL;
1794 
1795 	err = uverbs_get_const(&cmd_out_len, attrs,
1796 			       MLX5_IB_ATTR_DEVX_OBJ_QUERY_ASYNC_OUT_LEN);
1797 	if (err)
1798 		return err;
1799 
1800 	if (!devx_is_valid_obj_id(attrs, uobj, cmd_in))
1801 		return -EINVAL;
1802 
1803 	fd_uobj = uverbs_attr_get_uobject(attrs,
1804 				MLX5_IB_ATTR_DEVX_OBJ_QUERY_ASYNC_FD);
1805 	if (IS_ERR(fd_uobj))
1806 		return PTR_ERR(fd_uobj);
1807 
1808 	ev_file = container_of(fd_uobj, struct devx_async_cmd_event_file,
1809 			       uobj);
1810 
1811 	if (atomic_add_return(cmd_out_len, &ev_file->ev_queue.bytes_in_use) >
1812 			MAX_ASYNC_BYTES_IN_USE) {
1813 		atomic_sub(cmd_out_len, &ev_file->ev_queue.bytes_in_use);
1814 		return -EAGAIN;
1815 	}
1816 
1817 	async_data = kvzalloc(struct_size(async_data, hdr.out_data,
1818 					  cmd_out_len), GFP_KERNEL);
1819 	if (!async_data) {
1820 		err = -ENOMEM;
1821 		goto sub_bytes;
1822 	}
1823 
1824 	err = uverbs_copy_from(&async_data->hdr.wr_id, attrs,
1825 			       MLX5_IB_ATTR_DEVX_OBJ_QUERY_ASYNC_WR_ID);
1826 	if (err)
1827 		goto free_async;
1828 
1829 	async_data->cmd_out_len = cmd_out_len;
1830 	async_data->mdev = mdev;
1831 	async_data->ev_file = ev_file;
1832 
1833 	MLX5_SET(general_obj_in_cmd_hdr, cmd_in, uid, uid);
1834 	err = mlx5_cmd_exec_cb(&ev_file->async_ctx, cmd_in,
1835 		    uverbs_attr_get_len(attrs,
1836 				MLX5_IB_ATTR_DEVX_OBJ_QUERY_ASYNC_CMD_IN),
1837 		    async_data->hdr.out_data,
1838 		    async_data->cmd_out_len,
1839 		    devx_query_callback, &async_data->cb_work);
1840 
1841 	if (err)
1842 		goto free_async;
1843 
1844 	return 0;
1845 
1846 free_async:
1847 	kvfree(async_data);
1848 sub_bytes:
1849 	atomic_sub(cmd_out_len, &ev_file->ev_queue.bytes_in_use);
1850 	return err;
1851 }
1852 
1853 static void
1854 subscribe_event_xa_dealloc(struct mlx5_devx_event_table *devx_event_table,
1855 			   u32 key_level1,
1856 			   bool is_level2,
1857 			   u32 key_level2)
1858 {
1859 	struct devx_event *event;
1860 	struct devx_obj_event *xa_val_level2;
1861 
1862 	/* Level 1 is valid for future use, no need to free */
1863 	if (!is_level2)
1864 		return;
1865 
1866 	event = xa_load(&devx_event_table->event_xa, key_level1);
1867 	WARN_ON(!event);
1868 
1869 	xa_val_level2 = xa_load(&event->object_ids,
1870 				key_level2);
1871 	if (list_empty(&xa_val_level2->obj_sub_list)) {
1872 		xa_erase(&event->object_ids,
1873 			 key_level2);
1874 		kfree_rcu(xa_val_level2, rcu);
1875 	}
1876 }
1877 
1878 static int
1879 subscribe_event_xa_alloc(struct mlx5_devx_event_table *devx_event_table,
1880 			 u32 key_level1,
1881 			 bool is_level2,
1882 			 u32 key_level2)
1883 {
1884 	struct devx_obj_event *obj_event;
1885 	struct devx_event *event;
1886 	int err;
1887 
1888 	event = xa_load(&devx_event_table->event_xa, key_level1);
1889 	if (!event) {
1890 		event = kzalloc(sizeof(*event), GFP_KERNEL);
1891 		if (!event)
1892 			return -ENOMEM;
1893 
1894 		INIT_LIST_HEAD(&event->unaffiliated_list);
1895 		xa_init(&event->object_ids);
1896 
1897 		err = xa_insert(&devx_event_table->event_xa,
1898 				key_level1,
1899 				event,
1900 				GFP_KERNEL);
1901 		if (err) {
1902 			kfree(event);
1903 			return err;
1904 		}
1905 	}
1906 
1907 	if (!is_level2)
1908 		return 0;
1909 
1910 	obj_event = xa_load(&event->object_ids, key_level2);
1911 	if (!obj_event) {
1912 		obj_event = kzalloc(sizeof(*obj_event), GFP_KERNEL);
1913 		if (!obj_event)
1914 			/* Level1 is valid for future use, no need to free */
1915 			return -ENOMEM;
1916 
1917 		err = xa_insert(&event->object_ids,
1918 				key_level2,
1919 				obj_event,
1920 				GFP_KERNEL);
1921 		if (err) {
1922 			kfree(obj_event);
1923 			return err;
1924 		}
1925 		INIT_LIST_HEAD(&obj_event->obj_sub_list);
1926 	}
1927 
1928 	return 0;
1929 }
1930 
1931 static bool is_valid_events_legacy(int num_events, u16 *event_type_num_list,
1932 				   struct devx_obj *obj)
1933 {
1934 	int i;
1935 
1936 	for (i = 0; i < num_events; i++) {
1937 		if (obj) {
1938 			if (!is_legacy_obj_event_num(event_type_num_list[i]))
1939 				return false;
1940 		} else if (!is_legacy_unaffiliated_event_num(
1941 				event_type_num_list[i])) {
1942 			return false;
1943 		}
1944 	}
1945 
1946 	return true;
1947 }
1948 
1949 #define MAX_SUPP_EVENT_NUM 255
1950 static bool is_valid_events(struct mlx5_core_dev *dev,
1951 			    int num_events, u16 *event_type_num_list,
1952 			    struct devx_obj *obj)
1953 {
1954 	__be64 *aff_events;
1955 	__be64 *unaff_events;
1956 	int mask_entry;
1957 	int mask_bit;
1958 	int i;
1959 
1960 	if (MLX5_CAP_GEN(dev, event_cap)) {
1961 		aff_events = MLX5_CAP_DEV_EVENT(dev,
1962 						user_affiliated_events);
1963 		unaff_events = MLX5_CAP_DEV_EVENT(dev,
1964 						  user_unaffiliated_events);
1965 	} else {
1966 		return is_valid_events_legacy(num_events, event_type_num_list,
1967 					      obj);
1968 	}
1969 
1970 	for (i = 0; i < num_events; i++) {
1971 		if (event_type_num_list[i] > MAX_SUPP_EVENT_NUM)
1972 			return false;
1973 
1974 		mask_entry = event_type_num_list[i] / 64;
1975 		mask_bit = event_type_num_list[i] % 64;
1976 
1977 		if (obj) {
1978 			/* CQ completion */
1979 			if (event_type_num_list[i] == 0)
1980 				continue;
1981 
1982 			if (!(be64_to_cpu(aff_events[mask_entry]) &
1983 					(1ull << mask_bit)))
1984 				return false;
1985 
1986 			continue;
1987 		}
1988 
1989 		if (!(be64_to_cpu(unaff_events[mask_entry]) &
1990 				(1ull << mask_bit)))
1991 			return false;
1992 	}
1993 
1994 	return true;
1995 }
1996 
1997 #define MAX_NUM_EVENTS 16
1998 static int UVERBS_HANDLER(MLX5_IB_METHOD_DEVX_SUBSCRIBE_EVENT)(
1999 	struct uverbs_attr_bundle *attrs)
2000 {
2001 	struct ib_uobject *devx_uobj = uverbs_attr_get_uobject(
2002 				attrs,
2003 				MLX5_IB_ATTR_DEVX_SUBSCRIBE_EVENT_OBJ_HANDLE);
2004 	struct mlx5_ib_ucontext *c = rdma_udata_to_drv_context(
2005 		&attrs->driver_udata, struct mlx5_ib_ucontext, ibucontext);
2006 	struct mlx5_ib_dev *dev = to_mdev(c->ibucontext.device);
2007 	struct ib_uobject *fd_uobj;
2008 	struct devx_obj *obj = NULL;
2009 	struct devx_async_event_file *ev_file;
2010 	struct mlx5_devx_event_table *devx_event_table = &dev->devx_event_table;
2011 	u16 *event_type_num_list;
2012 	struct devx_event_subscription *event_sub, *tmp_sub;
2013 	struct list_head sub_list;
2014 	int redirect_fd;
2015 	bool use_eventfd = false;
2016 	int num_events;
2017 	u16 obj_type = 0;
2018 	u64 cookie = 0;
2019 	u32 obj_id = 0;
2020 	int err;
2021 	int i;
2022 
2023 	if (!c->devx_uid)
2024 		return -EINVAL;
2025 
2026 	if (!IS_ERR(devx_uobj)) {
2027 		obj = (struct devx_obj *)devx_uobj->object;
2028 		if (obj)
2029 			obj_id = get_dec_obj_id(obj->obj_id);
2030 	}
2031 
2032 	fd_uobj = uverbs_attr_get_uobject(attrs,
2033 				MLX5_IB_ATTR_DEVX_SUBSCRIBE_EVENT_FD_HANDLE);
2034 	if (IS_ERR(fd_uobj))
2035 		return PTR_ERR(fd_uobj);
2036 
2037 	ev_file = container_of(fd_uobj, struct devx_async_event_file,
2038 			       uobj);
2039 
2040 	if (uverbs_attr_is_valid(attrs,
2041 				 MLX5_IB_ATTR_DEVX_SUBSCRIBE_EVENT_FD_NUM)) {
2042 		err = uverbs_copy_from(&redirect_fd, attrs,
2043 			       MLX5_IB_ATTR_DEVX_SUBSCRIBE_EVENT_FD_NUM);
2044 		if (err)
2045 			return err;
2046 
2047 		use_eventfd = true;
2048 	}
2049 
2050 	if (uverbs_attr_is_valid(attrs,
2051 				 MLX5_IB_ATTR_DEVX_SUBSCRIBE_EVENT_COOKIE)) {
2052 		if (use_eventfd)
2053 			return -EINVAL;
2054 
2055 		err = uverbs_copy_from(&cookie, attrs,
2056 				MLX5_IB_ATTR_DEVX_SUBSCRIBE_EVENT_COOKIE);
2057 		if (err)
2058 			return err;
2059 	}
2060 
2061 	num_events = uverbs_attr_ptr_get_array_size(
2062 		attrs, MLX5_IB_ATTR_DEVX_SUBSCRIBE_EVENT_TYPE_NUM_LIST,
2063 		sizeof(u16));
2064 
2065 	if (num_events < 0)
2066 		return num_events;
2067 
2068 	if (num_events > MAX_NUM_EVENTS)
2069 		return -EINVAL;
2070 
2071 	event_type_num_list = uverbs_attr_get_alloced_ptr(attrs,
2072 			MLX5_IB_ATTR_DEVX_SUBSCRIBE_EVENT_TYPE_NUM_LIST);
2073 
2074 	if (!is_valid_events(dev->mdev, num_events, event_type_num_list, obj))
2075 		return -EINVAL;
2076 
2077 	INIT_LIST_HEAD(&sub_list);
2078 
2079 	/* Protect from concurrent subscriptions to same XA entries to allow
2080 	 * both to succeed
2081 	 */
2082 	mutex_lock(&devx_event_table->event_xa_lock);
2083 	for (i = 0; i < num_events; i++) {
2084 		u32 key_level1;
2085 
2086 		if (obj)
2087 			obj_type = get_dec_obj_type(obj,
2088 						    event_type_num_list[i]);
2089 		key_level1 = event_type_num_list[i] | obj_type << 16;
2090 
2091 		err = subscribe_event_xa_alloc(devx_event_table,
2092 					       key_level1,
2093 					       obj,
2094 					       obj_id);
2095 		if (err)
2096 			goto err;
2097 
2098 		event_sub = kzalloc(sizeof(*event_sub), GFP_KERNEL);
2099 		if (!event_sub) {
2100 			err = -ENOMEM;
2101 			goto err;
2102 		}
2103 
2104 		list_add_tail(&event_sub->event_list, &sub_list);
2105 		uverbs_uobject_get(&ev_file->uobj);
2106 		if (use_eventfd) {
2107 			event_sub->eventfd =
2108 				eventfd_ctx_fdget(redirect_fd);
2109 
2110 			if (IS_ERR(event_sub->eventfd)) {
2111 				err = PTR_ERR(event_sub->eventfd);
2112 				event_sub->eventfd = NULL;
2113 				goto err;
2114 			}
2115 		}
2116 
2117 		event_sub->cookie = cookie;
2118 		event_sub->ev_file = ev_file;
2119 		/* May be needed upon cleanup the devx object/subscription */
2120 		event_sub->xa_key_level1 = key_level1;
2121 		event_sub->xa_key_level2 = obj_id;
2122 		INIT_LIST_HEAD(&event_sub->obj_list);
2123 	}
2124 
2125 	/* Once all the allocations and the XA data insertions were done we
2126 	 * can go ahead and add all the subscriptions to the relevant lists
2127 	 * without concern of a failure.
2128 	 */
2129 	list_for_each_entry_safe(event_sub, tmp_sub, &sub_list, event_list) {
2130 		struct devx_event *event;
2131 		struct devx_obj_event *obj_event;
2132 
2133 		list_del_init(&event_sub->event_list);
2134 
2135 		spin_lock_irq(&ev_file->lock);
2136 		list_add_tail_rcu(&event_sub->file_list,
2137 				  &ev_file->subscribed_events_list);
2138 		spin_unlock_irq(&ev_file->lock);
2139 
2140 		event = xa_load(&devx_event_table->event_xa,
2141 				event_sub->xa_key_level1);
2142 		WARN_ON(!event);
2143 
2144 		if (!obj) {
2145 			list_add_tail_rcu(&event_sub->xa_list,
2146 					  &event->unaffiliated_list);
2147 			continue;
2148 		}
2149 
2150 		obj_event = xa_load(&event->object_ids, obj_id);
2151 		WARN_ON(!obj_event);
2152 		list_add_tail_rcu(&event_sub->xa_list,
2153 				  &obj_event->obj_sub_list);
2154 		list_add_tail_rcu(&event_sub->obj_list,
2155 				  &obj->event_sub);
2156 	}
2157 
2158 	mutex_unlock(&devx_event_table->event_xa_lock);
2159 	return 0;
2160 
2161 err:
2162 	list_for_each_entry_safe(event_sub, tmp_sub, &sub_list, event_list) {
2163 		list_del(&event_sub->event_list);
2164 
2165 		subscribe_event_xa_dealloc(devx_event_table,
2166 					   event_sub->xa_key_level1,
2167 					   obj,
2168 					   obj_id);
2169 
2170 		if (event_sub->eventfd)
2171 			eventfd_ctx_put(event_sub->eventfd);
2172 		uverbs_uobject_put(&event_sub->ev_file->uobj);
2173 		kfree(event_sub);
2174 	}
2175 
2176 	mutex_unlock(&devx_event_table->event_xa_lock);
2177 	return err;
2178 }
2179 
2180 static int devx_umem_get(struct mlx5_ib_dev *dev, struct ib_ucontext *ucontext,
2181 			 struct uverbs_attr_bundle *attrs,
2182 			 struct devx_umem *obj, u32 access_flags)
2183 {
2184 	u64 addr;
2185 	size_t size;
2186 	int err;
2187 
2188 	if (uverbs_copy_from(&addr, attrs, MLX5_IB_ATTR_DEVX_UMEM_REG_ADDR) ||
2189 	    uverbs_copy_from(&size, attrs, MLX5_IB_ATTR_DEVX_UMEM_REG_LEN))
2190 		return -EFAULT;
2191 
2192 	err = ib_check_mr_access(&dev->ib_dev, access_flags);
2193 	if (err)
2194 		return err;
2195 
2196 	if (uverbs_attr_is_valid(attrs, MLX5_IB_ATTR_DEVX_UMEM_REG_DMABUF_FD)) {
2197 		struct ib_umem_dmabuf *umem_dmabuf;
2198 		int dmabuf_fd;
2199 
2200 		err = uverbs_get_raw_fd(&dmabuf_fd, attrs,
2201 					MLX5_IB_ATTR_DEVX_UMEM_REG_DMABUF_FD);
2202 		if (err)
2203 			return -EFAULT;
2204 
2205 		umem_dmabuf = ib_umem_dmabuf_get_pinned(
2206 			&dev->ib_dev, addr, size, dmabuf_fd, access_flags);
2207 		if (IS_ERR(umem_dmabuf))
2208 			return PTR_ERR(umem_dmabuf);
2209 		obj->umem = &umem_dmabuf->umem;
2210 	} else {
2211 		obj->umem = ib_umem_get(&dev->ib_dev, addr, size, access_flags);
2212 		if (IS_ERR(obj->umem))
2213 			return PTR_ERR(obj->umem);
2214 	}
2215 	return 0;
2216 }
2217 
2218 static unsigned int devx_umem_find_best_pgsize(struct ib_umem *umem,
2219 					       unsigned long pgsz_bitmap)
2220 {
2221 	unsigned long page_size;
2222 
2223 	/* Don't bother checking larger page sizes as offset must be zero and
2224 	 * total DEVX umem length must be equal to total umem length.
2225 	 */
2226 	pgsz_bitmap &= GENMASK_ULL(max_t(u64, order_base_2(umem->length),
2227 					 PAGE_SHIFT),
2228 				   MLX5_ADAPTER_PAGE_SHIFT);
2229 	if (!pgsz_bitmap)
2230 		return 0;
2231 
2232 	page_size = ib_umem_find_best_pgoff(umem, pgsz_bitmap, U64_MAX);
2233 	if (!page_size)
2234 		return 0;
2235 
2236 	/* If the page_size is less than the CPU page size then we can use the
2237 	 * offset and create a umem which is a subset of the page list.
2238 	 * For larger page sizes we can't be sure the DMA  list reflects the
2239 	 * VA so we must ensure that the umem extent is exactly equal to the
2240 	 * page list. Reduce the page size until one of these cases is true.
2241 	 */
2242 	while ((ib_umem_dma_offset(umem, page_size) != 0 ||
2243 		(umem->length % page_size) != 0) &&
2244 		page_size > PAGE_SIZE)
2245 		page_size /= 2;
2246 
2247 	return page_size;
2248 }
2249 
2250 static int devx_umem_reg_cmd_alloc(struct mlx5_ib_dev *dev,
2251 				   struct uverbs_attr_bundle *attrs,
2252 				   struct devx_umem *obj,
2253 				   struct devx_umem_reg_cmd *cmd,
2254 				   int access)
2255 {
2256 	unsigned long pgsz_bitmap;
2257 	unsigned int page_size;
2258 	__be64 *mtt;
2259 	void *umem;
2260 	int ret;
2261 
2262 	/*
2263 	 * If the user does not pass in pgsz_bitmap then the user promises not
2264 	 * to use umem_offset!=0 in any commands that allocate on top of the
2265 	 * umem.
2266 	 *
2267 	 * If the user wants to use a umem_offset then it must pass in
2268 	 * pgsz_bitmap which guides the maximum page size and thus maximum
2269 	 * object alignment inside the umem. See the PRM.
2270 	 *
2271 	 * Users are not allowed to use IOVA here, mkeys are not supported on
2272 	 * umem.
2273 	 */
2274 	ret = uverbs_get_const_default(&pgsz_bitmap, attrs,
2275 			MLX5_IB_ATTR_DEVX_UMEM_REG_PGSZ_BITMAP,
2276 			GENMASK_ULL(63,
2277 				    min(PAGE_SHIFT, MLX5_ADAPTER_PAGE_SHIFT)));
2278 	if (ret)
2279 		return ret;
2280 
2281 	page_size = devx_umem_find_best_pgsize(obj->umem, pgsz_bitmap);
2282 	if (!page_size)
2283 		return -EINVAL;
2284 
2285 	cmd->inlen = MLX5_ST_SZ_BYTES(create_umem_in) +
2286 		     (MLX5_ST_SZ_BYTES(mtt) *
2287 		      ib_umem_num_dma_blocks(obj->umem, page_size));
2288 	cmd->in = uverbs_zalloc(attrs, cmd->inlen);
2289 	if (IS_ERR(cmd->in))
2290 		return PTR_ERR(cmd->in);
2291 
2292 	umem = MLX5_ADDR_OF(create_umem_in, cmd->in, umem);
2293 	mtt = (__be64 *)MLX5_ADDR_OF(umem, umem, mtt);
2294 
2295 	MLX5_SET(create_umem_in, cmd->in, opcode, MLX5_CMD_OP_CREATE_UMEM);
2296 	MLX5_SET64(umem, umem, num_of_mtt,
2297 		   ib_umem_num_dma_blocks(obj->umem, page_size));
2298 	MLX5_SET(umem, umem, log_page_size,
2299 		 order_base_2(page_size) - MLX5_ADAPTER_PAGE_SHIFT);
2300 	MLX5_SET(umem, umem, page_offset,
2301 		 ib_umem_dma_offset(obj->umem, page_size));
2302 
2303 	if (mlx5_umem_needs_ats(dev, obj->umem, access))
2304 		MLX5_SET(umem, umem, ats, 1);
2305 
2306 	mlx5_ib_populate_pas(obj->umem, page_size, mtt,
2307 			     (obj->umem->writable ? MLX5_IB_MTT_WRITE : 0) |
2308 				     MLX5_IB_MTT_READ);
2309 	return 0;
2310 }
2311 
2312 static int UVERBS_HANDLER(MLX5_IB_METHOD_DEVX_UMEM_REG)(
2313 	struct uverbs_attr_bundle *attrs)
2314 {
2315 	struct devx_umem_reg_cmd cmd;
2316 	struct devx_umem *obj;
2317 	struct ib_uobject *uobj = uverbs_attr_get_uobject(
2318 		attrs, MLX5_IB_ATTR_DEVX_UMEM_REG_HANDLE);
2319 	u32 obj_id;
2320 	struct mlx5_ib_ucontext *c = rdma_udata_to_drv_context(
2321 		&attrs->driver_udata, struct mlx5_ib_ucontext, ibucontext);
2322 	struct mlx5_ib_dev *dev = to_mdev(c->ibucontext.device);
2323 	int access_flags;
2324 	int err;
2325 
2326 	if (!c->devx_uid)
2327 		return -EINVAL;
2328 
2329 	err = uverbs_get_flags32(&access_flags, attrs,
2330 				 MLX5_IB_ATTR_DEVX_UMEM_REG_ACCESS,
2331 				 IB_ACCESS_LOCAL_WRITE |
2332 				 IB_ACCESS_REMOTE_WRITE |
2333 				 IB_ACCESS_REMOTE_READ |
2334 				 IB_ACCESS_RELAXED_ORDERING);
2335 	if (err)
2336 		return err;
2337 
2338 	obj = kzalloc(sizeof(struct devx_umem), GFP_KERNEL);
2339 	if (!obj)
2340 		return -ENOMEM;
2341 
2342 	err = devx_umem_get(dev, &c->ibucontext, attrs, obj, access_flags);
2343 	if (err)
2344 		goto err_obj_free;
2345 
2346 	err = devx_umem_reg_cmd_alloc(dev, attrs, obj, &cmd, access_flags);
2347 	if (err)
2348 		goto err_umem_release;
2349 
2350 	MLX5_SET(create_umem_in, cmd.in, uid, c->devx_uid);
2351 	err = mlx5_cmd_exec(dev->mdev, cmd.in, cmd.inlen, cmd.out,
2352 			    sizeof(cmd.out));
2353 	if (err)
2354 		goto err_umem_release;
2355 
2356 	obj->mdev = dev->mdev;
2357 	uobj->object = obj;
2358 	devx_obj_build_destroy_cmd(cmd.in, cmd.out, obj->dinbox, &obj->dinlen, &obj_id);
2359 	uverbs_finalize_uobj_create(attrs, MLX5_IB_ATTR_DEVX_UMEM_REG_HANDLE);
2360 
2361 	err = uverbs_copy_to(attrs, MLX5_IB_ATTR_DEVX_UMEM_REG_OUT_ID, &obj_id,
2362 			     sizeof(obj_id));
2363 	return err;
2364 
2365 err_umem_release:
2366 	ib_umem_release(obj->umem);
2367 err_obj_free:
2368 	kfree(obj);
2369 	return err;
2370 }
2371 
2372 static int devx_umem_cleanup(struct ib_uobject *uobject,
2373 			     enum rdma_remove_reason why,
2374 			     struct uverbs_attr_bundle *attrs)
2375 {
2376 	struct devx_umem *obj = uobject->object;
2377 	u32 out[MLX5_ST_SZ_DW(general_obj_out_cmd_hdr)];
2378 	int err;
2379 
2380 	err = mlx5_cmd_exec(obj->mdev, obj->dinbox, obj->dinlen, out, sizeof(out));
2381 	if (err)
2382 		return err;
2383 
2384 	ib_umem_release(obj->umem);
2385 	kfree(obj);
2386 	return 0;
2387 }
2388 
2389 static bool is_unaffiliated_event(struct mlx5_core_dev *dev,
2390 				  unsigned long event_type)
2391 {
2392 	__be64 *unaff_events;
2393 	int mask_entry;
2394 	int mask_bit;
2395 
2396 	if (!MLX5_CAP_GEN(dev, event_cap))
2397 		return is_legacy_unaffiliated_event_num(event_type);
2398 
2399 	unaff_events = MLX5_CAP_DEV_EVENT(dev,
2400 					  user_unaffiliated_events);
2401 	WARN_ON(event_type > MAX_SUPP_EVENT_NUM);
2402 
2403 	mask_entry = event_type / 64;
2404 	mask_bit = event_type % 64;
2405 
2406 	if (!(be64_to_cpu(unaff_events[mask_entry]) & (1ull << mask_bit)))
2407 		return false;
2408 
2409 	return true;
2410 }
2411 
2412 static u32 devx_get_obj_id_from_event(unsigned long event_type, void *data)
2413 {
2414 	struct mlx5_eqe *eqe = data;
2415 	u32 obj_id = 0;
2416 
2417 	switch (event_type) {
2418 	case MLX5_EVENT_TYPE_SRQ_CATAS_ERROR:
2419 	case MLX5_EVENT_TYPE_SRQ_RQ_LIMIT:
2420 	case MLX5_EVENT_TYPE_PATH_MIG:
2421 	case MLX5_EVENT_TYPE_COMM_EST:
2422 	case MLX5_EVENT_TYPE_SQ_DRAINED:
2423 	case MLX5_EVENT_TYPE_SRQ_LAST_WQE:
2424 	case MLX5_EVENT_TYPE_WQ_CATAS_ERROR:
2425 	case MLX5_EVENT_TYPE_PATH_MIG_FAILED:
2426 	case MLX5_EVENT_TYPE_WQ_INVAL_REQ_ERROR:
2427 	case MLX5_EVENT_TYPE_WQ_ACCESS_ERROR:
2428 		obj_id = be32_to_cpu(eqe->data.qp_srq.qp_srq_n) & 0xffffff;
2429 		break;
2430 	case MLX5_EVENT_TYPE_XRQ_ERROR:
2431 		obj_id = be32_to_cpu(eqe->data.xrq_err.type_xrqn) & 0xffffff;
2432 		break;
2433 	case MLX5_EVENT_TYPE_DCT_DRAINED:
2434 	case MLX5_EVENT_TYPE_DCT_KEY_VIOLATION:
2435 		obj_id = be32_to_cpu(eqe->data.dct.dctn) & 0xffffff;
2436 		break;
2437 	case MLX5_EVENT_TYPE_CQ_ERROR:
2438 		obj_id = be32_to_cpu(eqe->data.cq_err.cqn) & 0xffffff;
2439 		break;
2440 	default:
2441 		obj_id = MLX5_GET(affiliated_event_header, &eqe->data, obj_id);
2442 		break;
2443 	}
2444 
2445 	return obj_id;
2446 }
2447 
2448 static int deliver_event(struct devx_event_subscription *event_sub,
2449 			 const void *data)
2450 {
2451 	struct devx_async_event_file *ev_file;
2452 	struct devx_async_event_data *event_data;
2453 	unsigned long flags;
2454 
2455 	ev_file = event_sub->ev_file;
2456 
2457 	if (ev_file->omit_data) {
2458 		spin_lock_irqsave(&ev_file->lock, flags);
2459 		if (!list_empty(&event_sub->event_list) ||
2460 		    ev_file->is_destroyed) {
2461 			spin_unlock_irqrestore(&ev_file->lock, flags);
2462 			return 0;
2463 		}
2464 
2465 		list_add_tail(&event_sub->event_list, &ev_file->event_list);
2466 		spin_unlock_irqrestore(&ev_file->lock, flags);
2467 		wake_up_interruptible(&ev_file->poll_wait);
2468 		return 0;
2469 	}
2470 
2471 	event_data = kzalloc(sizeof(*event_data) + sizeof(struct mlx5_eqe),
2472 			     GFP_ATOMIC);
2473 	if (!event_data) {
2474 		spin_lock_irqsave(&ev_file->lock, flags);
2475 		ev_file->is_overflow_err = 1;
2476 		spin_unlock_irqrestore(&ev_file->lock, flags);
2477 		return -ENOMEM;
2478 	}
2479 
2480 	event_data->hdr.cookie = event_sub->cookie;
2481 	memcpy(event_data->hdr.out_data, data, sizeof(struct mlx5_eqe));
2482 
2483 	spin_lock_irqsave(&ev_file->lock, flags);
2484 	if (!ev_file->is_destroyed)
2485 		list_add_tail(&event_data->list, &ev_file->event_list);
2486 	else
2487 		kfree(event_data);
2488 	spin_unlock_irqrestore(&ev_file->lock, flags);
2489 	wake_up_interruptible(&ev_file->poll_wait);
2490 
2491 	return 0;
2492 }
2493 
2494 static void dispatch_event_fd(struct list_head *fd_list,
2495 			      const void *data)
2496 {
2497 	struct devx_event_subscription *item;
2498 
2499 	list_for_each_entry_rcu(item, fd_list, xa_list) {
2500 		if (item->eventfd)
2501 			eventfd_signal(item->eventfd);
2502 		else
2503 			deliver_event(item, data);
2504 	}
2505 }
2506 
2507 static int devx_event_notifier(struct notifier_block *nb,
2508 			       unsigned long event_type, void *data)
2509 {
2510 	struct mlx5_devx_event_table *table;
2511 	struct mlx5_ib_dev *dev;
2512 	struct devx_event *event;
2513 	struct devx_obj_event *obj_event;
2514 	u16 obj_type = 0;
2515 	bool is_unaffiliated;
2516 	u32 obj_id;
2517 
2518 	/* Explicit filtering to kernel events which may occur frequently */
2519 	if (event_type == MLX5_EVENT_TYPE_CMD ||
2520 	    event_type == MLX5_EVENT_TYPE_PAGE_REQUEST)
2521 		return NOTIFY_OK;
2522 
2523 	table = container_of(nb, struct mlx5_devx_event_table, devx_nb.nb);
2524 	dev = container_of(table, struct mlx5_ib_dev, devx_event_table);
2525 	is_unaffiliated = is_unaffiliated_event(dev->mdev, event_type);
2526 
2527 	if (!is_unaffiliated)
2528 		obj_type = get_event_obj_type(event_type, data);
2529 
2530 	rcu_read_lock();
2531 	event = xa_load(&table->event_xa, event_type | (obj_type << 16));
2532 	if (!event) {
2533 		rcu_read_unlock();
2534 		return NOTIFY_DONE;
2535 	}
2536 
2537 	if (is_unaffiliated) {
2538 		dispatch_event_fd(&event->unaffiliated_list, data);
2539 		rcu_read_unlock();
2540 		return NOTIFY_OK;
2541 	}
2542 
2543 	obj_id = devx_get_obj_id_from_event(event_type, data);
2544 	obj_event = xa_load(&event->object_ids, obj_id);
2545 	if (!obj_event) {
2546 		rcu_read_unlock();
2547 		return NOTIFY_DONE;
2548 	}
2549 
2550 	dispatch_event_fd(&obj_event->obj_sub_list, data);
2551 
2552 	rcu_read_unlock();
2553 	return NOTIFY_OK;
2554 }
2555 
2556 int mlx5_ib_devx_init(struct mlx5_ib_dev *dev)
2557 {
2558 	struct mlx5_devx_event_table *table = &dev->devx_event_table;
2559 	int uid;
2560 
2561 	uid = mlx5_ib_devx_create(dev, false);
2562 	if (uid > 0) {
2563 		dev->devx_whitelist_uid = uid;
2564 		xa_init(&table->event_xa);
2565 		mutex_init(&table->event_xa_lock);
2566 		MLX5_NB_INIT(&table->devx_nb, devx_event_notifier, NOTIFY_ANY);
2567 		mlx5_eq_notifier_register(dev->mdev, &table->devx_nb);
2568 	}
2569 
2570 	return 0;
2571 }
2572 
2573 void mlx5_ib_devx_cleanup(struct mlx5_ib_dev *dev)
2574 {
2575 	struct mlx5_devx_event_table *table = &dev->devx_event_table;
2576 	struct devx_event_subscription *sub, *tmp;
2577 	struct devx_event *event;
2578 	void *entry;
2579 	unsigned long id;
2580 
2581 	if (dev->devx_whitelist_uid) {
2582 		mlx5_eq_notifier_unregister(dev->mdev, &table->devx_nb);
2583 		mutex_lock(&dev->devx_event_table.event_xa_lock);
2584 		xa_for_each(&table->event_xa, id, entry) {
2585 			event = entry;
2586 			list_for_each_entry_safe(
2587 				sub, tmp, &event->unaffiliated_list, xa_list)
2588 				devx_cleanup_subscription(dev, sub);
2589 			kfree(entry);
2590 		}
2591 		mutex_unlock(&dev->devx_event_table.event_xa_lock);
2592 		xa_destroy(&table->event_xa);
2593 
2594 		mlx5_ib_devx_destroy(dev, dev->devx_whitelist_uid);
2595 	}
2596 }
2597 
2598 static ssize_t devx_async_cmd_event_read(struct file *filp, char __user *buf,
2599 					 size_t count, loff_t *pos)
2600 {
2601 	struct devx_async_cmd_event_file *comp_ev_file = filp->private_data;
2602 	struct devx_async_event_queue *ev_queue = &comp_ev_file->ev_queue;
2603 	struct devx_async_data *event;
2604 	int ret = 0;
2605 	size_t eventsz;
2606 
2607 	spin_lock_irq(&ev_queue->lock);
2608 
2609 	while (list_empty(&ev_queue->event_list)) {
2610 		spin_unlock_irq(&ev_queue->lock);
2611 
2612 		if (filp->f_flags & O_NONBLOCK)
2613 			return -EAGAIN;
2614 
2615 		if (wait_event_interruptible(
2616 			    ev_queue->poll_wait,
2617 			    (!list_empty(&ev_queue->event_list) ||
2618 			     ev_queue->is_destroyed))) {
2619 			return -ERESTARTSYS;
2620 		}
2621 
2622 		spin_lock_irq(&ev_queue->lock);
2623 		if (ev_queue->is_destroyed) {
2624 			spin_unlock_irq(&ev_queue->lock);
2625 			return -EIO;
2626 		}
2627 	}
2628 
2629 	event = list_entry(ev_queue->event_list.next,
2630 			   struct devx_async_data, list);
2631 	eventsz = event->cmd_out_len +
2632 			sizeof(struct mlx5_ib_uapi_devx_async_cmd_hdr);
2633 
2634 	if (eventsz > count) {
2635 		spin_unlock_irq(&ev_queue->lock);
2636 		return -ENOSPC;
2637 	}
2638 
2639 	list_del(ev_queue->event_list.next);
2640 	spin_unlock_irq(&ev_queue->lock);
2641 
2642 	if (copy_to_user(buf, &event->hdr, eventsz))
2643 		ret = -EFAULT;
2644 	else
2645 		ret = eventsz;
2646 
2647 	atomic_sub(event->cmd_out_len, &ev_queue->bytes_in_use);
2648 	kvfree(event);
2649 	return ret;
2650 }
2651 
2652 static __poll_t devx_async_cmd_event_poll(struct file *filp,
2653 					      struct poll_table_struct *wait)
2654 {
2655 	struct devx_async_cmd_event_file *comp_ev_file = filp->private_data;
2656 	struct devx_async_event_queue *ev_queue = &comp_ev_file->ev_queue;
2657 	__poll_t pollflags = 0;
2658 
2659 	poll_wait(filp, &ev_queue->poll_wait, wait);
2660 
2661 	spin_lock_irq(&ev_queue->lock);
2662 	if (ev_queue->is_destroyed)
2663 		pollflags = EPOLLIN | EPOLLRDNORM | EPOLLRDHUP;
2664 	else if (!list_empty(&ev_queue->event_list))
2665 		pollflags = EPOLLIN | EPOLLRDNORM;
2666 	spin_unlock_irq(&ev_queue->lock);
2667 
2668 	return pollflags;
2669 }
2670 
2671 static const struct file_operations devx_async_cmd_event_fops = {
2672 	.owner	 = THIS_MODULE,
2673 	.read	 = devx_async_cmd_event_read,
2674 	.poll    = devx_async_cmd_event_poll,
2675 	.release = uverbs_uobject_fd_release,
2676 	.llseek	 = no_llseek,
2677 };
2678 
2679 static ssize_t devx_async_event_read(struct file *filp, char __user *buf,
2680 				     size_t count, loff_t *pos)
2681 {
2682 	struct devx_async_event_file *ev_file = filp->private_data;
2683 	struct devx_event_subscription *event_sub;
2684 	struct devx_async_event_data *event;
2685 	int ret = 0;
2686 	size_t eventsz;
2687 	bool omit_data;
2688 	void *event_data;
2689 
2690 	omit_data = ev_file->omit_data;
2691 
2692 	spin_lock_irq(&ev_file->lock);
2693 
2694 	if (ev_file->is_overflow_err) {
2695 		ev_file->is_overflow_err = 0;
2696 		spin_unlock_irq(&ev_file->lock);
2697 		return -EOVERFLOW;
2698 	}
2699 
2700 
2701 	while (list_empty(&ev_file->event_list)) {
2702 		spin_unlock_irq(&ev_file->lock);
2703 
2704 		if (filp->f_flags & O_NONBLOCK)
2705 			return -EAGAIN;
2706 
2707 		if (wait_event_interruptible(ev_file->poll_wait,
2708 			    (!list_empty(&ev_file->event_list) ||
2709 			     ev_file->is_destroyed))) {
2710 			return -ERESTARTSYS;
2711 		}
2712 
2713 		spin_lock_irq(&ev_file->lock);
2714 		if (ev_file->is_destroyed) {
2715 			spin_unlock_irq(&ev_file->lock);
2716 			return -EIO;
2717 		}
2718 	}
2719 
2720 	if (omit_data) {
2721 		event_sub = list_first_entry(&ev_file->event_list,
2722 					struct devx_event_subscription,
2723 					event_list);
2724 		eventsz = sizeof(event_sub->cookie);
2725 		event_data = &event_sub->cookie;
2726 	} else {
2727 		event = list_first_entry(&ev_file->event_list,
2728 				      struct devx_async_event_data, list);
2729 		eventsz = sizeof(struct mlx5_eqe) +
2730 			sizeof(struct mlx5_ib_uapi_devx_async_event_hdr);
2731 		event_data = &event->hdr;
2732 	}
2733 
2734 	if (eventsz > count) {
2735 		spin_unlock_irq(&ev_file->lock);
2736 		return -EINVAL;
2737 	}
2738 
2739 	if (omit_data)
2740 		list_del_init(&event_sub->event_list);
2741 	else
2742 		list_del(&event->list);
2743 
2744 	spin_unlock_irq(&ev_file->lock);
2745 
2746 	if (copy_to_user(buf, event_data, eventsz))
2747 		/* This points to an application issue, not a kernel concern */
2748 		ret = -EFAULT;
2749 	else
2750 		ret = eventsz;
2751 
2752 	if (!omit_data)
2753 		kfree(event);
2754 	return ret;
2755 }
2756 
2757 static __poll_t devx_async_event_poll(struct file *filp,
2758 				      struct poll_table_struct *wait)
2759 {
2760 	struct devx_async_event_file *ev_file = filp->private_data;
2761 	__poll_t pollflags = 0;
2762 
2763 	poll_wait(filp, &ev_file->poll_wait, wait);
2764 
2765 	spin_lock_irq(&ev_file->lock);
2766 	if (ev_file->is_destroyed)
2767 		pollflags = EPOLLIN | EPOLLRDNORM | EPOLLRDHUP;
2768 	else if (!list_empty(&ev_file->event_list))
2769 		pollflags = EPOLLIN | EPOLLRDNORM;
2770 	spin_unlock_irq(&ev_file->lock);
2771 
2772 	return pollflags;
2773 }
2774 
2775 static void devx_free_subscription(struct rcu_head *rcu)
2776 {
2777 	struct devx_event_subscription *event_sub =
2778 		container_of(rcu, struct devx_event_subscription, rcu);
2779 
2780 	if (event_sub->eventfd)
2781 		eventfd_ctx_put(event_sub->eventfd);
2782 	uverbs_uobject_put(&event_sub->ev_file->uobj);
2783 	kfree(event_sub);
2784 }
2785 
2786 static const struct file_operations devx_async_event_fops = {
2787 	.owner	 = THIS_MODULE,
2788 	.read	 = devx_async_event_read,
2789 	.poll    = devx_async_event_poll,
2790 	.release = uverbs_uobject_fd_release,
2791 	.llseek	 = no_llseek,
2792 };
2793 
2794 static void devx_async_cmd_event_destroy_uobj(struct ib_uobject *uobj,
2795 					      enum rdma_remove_reason why)
2796 {
2797 	struct devx_async_cmd_event_file *comp_ev_file =
2798 		container_of(uobj, struct devx_async_cmd_event_file,
2799 			     uobj);
2800 	struct devx_async_event_queue *ev_queue = &comp_ev_file->ev_queue;
2801 	struct devx_async_data *entry, *tmp;
2802 
2803 	spin_lock_irq(&ev_queue->lock);
2804 	ev_queue->is_destroyed = 1;
2805 	spin_unlock_irq(&ev_queue->lock);
2806 	wake_up_interruptible(&ev_queue->poll_wait);
2807 
2808 	mlx5_cmd_cleanup_async_ctx(&comp_ev_file->async_ctx);
2809 
2810 	spin_lock_irq(&comp_ev_file->ev_queue.lock);
2811 	list_for_each_entry_safe(entry, tmp,
2812 				 &comp_ev_file->ev_queue.event_list, list) {
2813 		list_del(&entry->list);
2814 		kvfree(entry);
2815 	}
2816 	spin_unlock_irq(&comp_ev_file->ev_queue.lock);
2817 };
2818 
2819 static void devx_async_event_destroy_uobj(struct ib_uobject *uobj,
2820 					  enum rdma_remove_reason why)
2821 {
2822 	struct devx_async_event_file *ev_file =
2823 		container_of(uobj, struct devx_async_event_file,
2824 			     uobj);
2825 	struct devx_event_subscription *event_sub, *event_sub_tmp;
2826 	struct mlx5_ib_dev *dev = ev_file->dev;
2827 
2828 	spin_lock_irq(&ev_file->lock);
2829 	ev_file->is_destroyed = 1;
2830 
2831 	/* free the pending events allocation */
2832 	if (ev_file->omit_data) {
2833 		struct devx_event_subscription *event_sub, *tmp;
2834 
2835 		list_for_each_entry_safe(event_sub, tmp, &ev_file->event_list,
2836 					 event_list)
2837 			list_del_init(&event_sub->event_list);
2838 
2839 	} else {
2840 		struct devx_async_event_data *entry, *tmp;
2841 
2842 		list_for_each_entry_safe(entry, tmp, &ev_file->event_list,
2843 					 list) {
2844 			list_del(&entry->list);
2845 			kfree(entry);
2846 		}
2847 	}
2848 
2849 	spin_unlock_irq(&ev_file->lock);
2850 	wake_up_interruptible(&ev_file->poll_wait);
2851 
2852 	mutex_lock(&dev->devx_event_table.event_xa_lock);
2853 	/* delete the subscriptions which are related to this FD */
2854 	list_for_each_entry_safe(event_sub, event_sub_tmp,
2855 				 &ev_file->subscribed_events_list, file_list) {
2856 		devx_cleanup_subscription(dev, event_sub);
2857 		list_del_rcu(&event_sub->file_list);
2858 		/* subscription may not be used by the read API any more */
2859 		call_rcu(&event_sub->rcu, devx_free_subscription);
2860 	}
2861 	mutex_unlock(&dev->devx_event_table.event_xa_lock);
2862 
2863 	put_device(&dev->ib_dev.dev);
2864 };
2865 
2866 DECLARE_UVERBS_NAMED_METHOD(
2867 	MLX5_IB_METHOD_DEVX_UMEM_REG,
2868 	UVERBS_ATTR_IDR(MLX5_IB_ATTR_DEVX_UMEM_REG_HANDLE,
2869 			MLX5_IB_OBJECT_DEVX_UMEM,
2870 			UVERBS_ACCESS_NEW,
2871 			UA_MANDATORY),
2872 	UVERBS_ATTR_PTR_IN(MLX5_IB_ATTR_DEVX_UMEM_REG_ADDR,
2873 			   UVERBS_ATTR_TYPE(u64),
2874 			   UA_MANDATORY),
2875 	UVERBS_ATTR_PTR_IN(MLX5_IB_ATTR_DEVX_UMEM_REG_LEN,
2876 			   UVERBS_ATTR_TYPE(u64),
2877 			   UA_MANDATORY),
2878 	UVERBS_ATTR_RAW_FD(MLX5_IB_ATTR_DEVX_UMEM_REG_DMABUF_FD,
2879 			   UA_OPTIONAL),
2880 	UVERBS_ATTR_FLAGS_IN(MLX5_IB_ATTR_DEVX_UMEM_REG_ACCESS,
2881 			     enum ib_access_flags),
2882 	UVERBS_ATTR_CONST_IN(MLX5_IB_ATTR_DEVX_UMEM_REG_PGSZ_BITMAP,
2883 			     u64),
2884 	UVERBS_ATTR_PTR_OUT(MLX5_IB_ATTR_DEVX_UMEM_REG_OUT_ID,
2885 			    UVERBS_ATTR_TYPE(u32),
2886 			    UA_MANDATORY));
2887 
2888 DECLARE_UVERBS_NAMED_METHOD_DESTROY(
2889 	MLX5_IB_METHOD_DEVX_UMEM_DEREG,
2890 	UVERBS_ATTR_IDR(MLX5_IB_ATTR_DEVX_UMEM_DEREG_HANDLE,
2891 			MLX5_IB_OBJECT_DEVX_UMEM,
2892 			UVERBS_ACCESS_DESTROY,
2893 			UA_MANDATORY));
2894 
2895 DECLARE_UVERBS_NAMED_METHOD(
2896 	MLX5_IB_METHOD_DEVX_QUERY_EQN,
2897 	UVERBS_ATTR_PTR_IN(MLX5_IB_ATTR_DEVX_QUERY_EQN_USER_VEC,
2898 			   UVERBS_ATTR_TYPE(u32),
2899 			   UA_MANDATORY),
2900 	UVERBS_ATTR_PTR_OUT(MLX5_IB_ATTR_DEVX_QUERY_EQN_DEV_EQN,
2901 			    UVERBS_ATTR_TYPE(u32),
2902 			    UA_MANDATORY));
2903 
2904 DECLARE_UVERBS_NAMED_METHOD(
2905 	MLX5_IB_METHOD_DEVX_QUERY_UAR,
2906 	UVERBS_ATTR_PTR_IN(MLX5_IB_ATTR_DEVX_QUERY_UAR_USER_IDX,
2907 			   UVERBS_ATTR_TYPE(u32),
2908 			   UA_MANDATORY),
2909 	UVERBS_ATTR_PTR_OUT(MLX5_IB_ATTR_DEVX_QUERY_UAR_DEV_IDX,
2910 			    UVERBS_ATTR_TYPE(u32),
2911 			    UA_MANDATORY));
2912 
2913 DECLARE_UVERBS_NAMED_METHOD(
2914 	MLX5_IB_METHOD_DEVX_OTHER,
2915 	UVERBS_ATTR_PTR_IN(
2916 		MLX5_IB_ATTR_DEVX_OTHER_CMD_IN,
2917 		UVERBS_ATTR_MIN_SIZE(MLX5_ST_SZ_BYTES(general_obj_in_cmd_hdr)),
2918 		UA_MANDATORY,
2919 		UA_ALLOC_AND_COPY),
2920 	UVERBS_ATTR_PTR_OUT(
2921 		MLX5_IB_ATTR_DEVX_OTHER_CMD_OUT,
2922 		UVERBS_ATTR_MIN_SIZE(MLX5_ST_SZ_BYTES(general_obj_out_cmd_hdr)),
2923 		UA_MANDATORY));
2924 
2925 DECLARE_UVERBS_NAMED_METHOD(
2926 	MLX5_IB_METHOD_DEVX_OBJ_CREATE,
2927 	UVERBS_ATTR_IDR(MLX5_IB_ATTR_DEVX_OBJ_CREATE_HANDLE,
2928 			MLX5_IB_OBJECT_DEVX_OBJ,
2929 			UVERBS_ACCESS_NEW,
2930 			UA_MANDATORY),
2931 	UVERBS_ATTR_PTR_IN(
2932 		MLX5_IB_ATTR_DEVX_OBJ_CREATE_CMD_IN,
2933 		UVERBS_ATTR_MIN_SIZE(MLX5_ST_SZ_BYTES(general_obj_in_cmd_hdr)),
2934 		UA_MANDATORY,
2935 		UA_ALLOC_AND_COPY),
2936 	UVERBS_ATTR_PTR_OUT(
2937 		MLX5_IB_ATTR_DEVX_OBJ_CREATE_CMD_OUT,
2938 		UVERBS_ATTR_MIN_SIZE(MLX5_ST_SZ_BYTES(general_obj_out_cmd_hdr)),
2939 		UA_MANDATORY));
2940 
2941 DECLARE_UVERBS_NAMED_METHOD_DESTROY(
2942 	MLX5_IB_METHOD_DEVX_OBJ_DESTROY,
2943 	UVERBS_ATTR_IDR(MLX5_IB_ATTR_DEVX_OBJ_DESTROY_HANDLE,
2944 			MLX5_IB_OBJECT_DEVX_OBJ,
2945 			UVERBS_ACCESS_DESTROY,
2946 			UA_MANDATORY));
2947 
2948 DECLARE_UVERBS_NAMED_METHOD(
2949 	MLX5_IB_METHOD_DEVX_OBJ_MODIFY,
2950 	UVERBS_ATTR_IDR(MLX5_IB_ATTR_DEVX_OBJ_MODIFY_HANDLE,
2951 			UVERBS_IDR_ANY_OBJECT,
2952 			UVERBS_ACCESS_READ,
2953 			UA_MANDATORY),
2954 	UVERBS_ATTR_PTR_IN(
2955 		MLX5_IB_ATTR_DEVX_OBJ_MODIFY_CMD_IN,
2956 		UVERBS_ATTR_MIN_SIZE(MLX5_ST_SZ_BYTES(general_obj_in_cmd_hdr)),
2957 		UA_MANDATORY,
2958 		UA_ALLOC_AND_COPY),
2959 	UVERBS_ATTR_PTR_OUT(
2960 		MLX5_IB_ATTR_DEVX_OBJ_MODIFY_CMD_OUT,
2961 		UVERBS_ATTR_MIN_SIZE(MLX5_ST_SZ_BYTES(general_obj_out_cmd_hdr)),
2962 		UA_MANDATORY));
2963 
2964 DECLARE_UVERBS_NAMED_METHOD(
2965 	MLX5_IB_METHOD_DEVX_OBJ_QUERY,
2966 	UVERBS_ATTR_IDR(MLX5_IB_ATTR_DEVX_OBJ_QUERY_HANDLE,
2967 			UVERBS_IDR_ANY_OBJECT,
2968 			UVERBS_ACCESS_READ,
2969 			UA_MANDATORY),
2970 	UVERBS_ATTR_PTR_IN(
2971 		MLX5_IB_ATTR_DEVX_OBJ_QUERY_CMD_IN,
2972 		UVERBS_ATTR_MIN_SIZE(MLX5_ST_SZ_BYTES(general_obj_in_cmd_hdr)),
2973 		UA_MANDATORY,
2974 		UA_ALLOC_AND_COPY),
2975 	UVERBS_ATTR_PTR_OUT(
2976 		MLX5_IB_ATTR_DEVX_OBJ_QUERY_CMD_OUT,
2977 		UVERBS_ATTR_MIN_SIZE(MLX5_ST_SZ_BYTES(general_obj_out_cmd_hdr)),
2978 		UA_MANDATORY));
2979 
2980 DECLARE_UVERBS_NAMED_METHOD(
2981 	MLX5_IB_METHOD_DEVX_OBJ_ASYNC_QUERY,
2982 	UVERBS_ATTR_IDR(MLX5_IB_ATTR_DEVX_OBJ_QUERY_HANDLE,
2983 			UVERBS_IDR_ANY_OBJECT,
2984 			UVERBS_ACCESS_READ,
2985 			UA_MANDATORY),
2986 	UVERBS_ATTR_PTR_IN(
2987 		MLX5_IB_ATTR_DEVX_OBJ_QUERY_CMD_IN,
2988 		UVERBS_ATTR_MIN_SIZE(MLX5_ST_SZ_BYTES(general_obj_in_cmd_hdr)),
2989 		UA_MANDATORY,
2990 		UA_ALLOC_AND_COPY),
2991 	UVERBS_ATTR_CONST_IN(MLX5_IB_ATTR_DEVX_OBJ_QUERY_ASYNC_OUT_LEN,
2992 		u16, UA_MANDATORY),
2993 	UVERBS_ATTR_FD(MLX5_IB_ATTR_DEVX_OBJ_QUERY_ASYNC_FD,
2994 		MLX5_IB_OBJECT_DEVX_ASYNC_CMD_FD,
2995 		UVERBS_ACCESS_READ,
2996 		UA_MANDATORY),
2997 	UVERBS_ATTR_PTR_IN(MLX5_IB_ATTR_DEVX_OBJ_QUERY_ASYNC_WR_ID,
2998 		UVERBS_ATTR_TYPE(u64),
2999 		UA_MANDATORY));
3000 
3001 DECLARE_UVERBS_NAMED_METHOD(
3002 	MLX5_IB_METHOD_DEVX_SUBSCRIBE_EVENT,
3003 	UVERBS_ATTR_FD(MLX5_IB_ATTR_DEVX_SUBSCRIBE_EVENT_FD_HANDLE,
3004 		MLX5_IB_OBJECT_DEVX_ASYNC_EVENT_FD,
3005 		UVERBS_ACCESS_READ,
3006 		UA_MANDATORY),
3007 	UVERBS_ATTR_IDR(MLX5_IB_ATTR_DEVX_SUBSCRIBE_EVENT_OBJ_HANDLE,
3008 		MLX5_IB_OBJECT_DEVX_OBJ,
3009 		UVERBS_ACCESS_READ,
3010 		UA_OPTIONAL),
3011 	UVERBS_ATTR_PTR_IN(MLX5_IB_ATTR_DEVX_SUBSCRIBE_EVENT_TYPE_NUM_LIST,
3012 		UVERBS_ATTR_MIN_SIZE(sizeof(u16)),
3013 		UA_MANDATORY,
3014 		UA_ALLOC_AND_COPY),
3015 	UVERBS_ATTR_PTR_IN(MLX5_IB_ATTR_DEVX_SUBSCRIBE_EVENT_COOKIE,
3016 		UVERBS_ATTR_TYPE(u64),
3017 		UA_OPTIONAL),
3018 	UVERBS_ATTR_PTR_IN(MLX5_IB_ATTR_DEVX_SUBSCRIBE_EVENT_FD_NUM,
3019 		UVERBS_ATTR_TYPE(u32),
3020 		UA_OPTIONAL));
3021 
3022 DECLARE_UVERBS_GLOBAL_METHODS(MLX5_IB_OBJECT_DEVX,
3023 			      &UVERBS_METHOD(MLX5_IB_METHOD_DEVX_OTHER),
3024 			      &UVERBS_METHOD(MLX5_IB_METHOD_DEVX_QUERY_UAR),
3025 			      &UVERBS_METHOD(MLX5_IB_METHOD_DEVX_QUERY_EQN),
3026 			      &UVERBS_METHOD(MLX5_IB_METHOD_DEVX_SUBSCRIBE_EVENT));
3027 
3028 DECLARE_UVERBS_NAMED_OBJECT(MLX5_IB_OBJECT_DEVX_OBJ,
3029 			    UVERBS_TYPE_ALLOC_IDR(devx_obj_cleanup),
3030 			    &UVERBS_METHOD(MLX5_IB_METHOD_DEVX_OBJ_CREATE),
3031 			    &UVERBS_METHOD(MLX5_IB_METHOD_DEVX_OBJ_DESTROY),
3032 			    &UVERBS_METHOD(MLX5_IB_METHOD_DEVX_OBJ_MODIFY),
3033 			    &UVERBS_METHOD(MLX5_IB_METHOD_DEVX_OBJ_QUERY),
3034 			    &UVERBS_METHOD(MLX5_IB_METHOD_DEVX_OBJ_ASYNC_QUERY));
3035 
3036 DECLARE_UVERBS_NAMED_OBJECT(MLX5_IB_OBJECT_DEVX_UMEM,
3037 			    UVERBS_TYPE_ALLOC_IDR(devx_umem_cleanup),
3038 			    &UVERBS_METHOD(MLX5_IB_METHOD_DEVX_UMEM_REG),
3039 			    &UVERBS_METHOD(MLX5_IB_METHOD_DEVX_UMEM_DEREG));
3040 
3041 
3042 DECLARE_UVERBS_NAMED_METHOD(
3043 	MLX5_IB_METHOD_DEVX_ASYNC_CMD_FD_ALLOC,
3044 	UVERBS_ATTR_FD(MLX5_IB_ATTR_DEVX_ASYNC_CMD_FD_ALLOC_HANDLE,
3045 			MLX5_IB_OBJECT_DEVX_ASYNC_CMD_FD,
3046 			UVERBS_ACCESS_NEW,
3047 			UA_MANDATORY));
3048 
3049 DECLARE_UVERBS_NAMED_OBJECT(
3050 	MLX5_IB_OBJECT_DEVX_ASYNC_CMD_FD,
3051 	UVERBS_TYPE_ALLOC_FD(sizeof(struct devx_async_cmd_event_file),
3052 			     devx_async_cmd_event_destroy_uobj,
3053 			     &devx_async_cmd_event_fops, "[devx_async_cmd]",
3054 			     O_RDONLY),
3055 	&UVERBS_METHOD(MLX5_IB_METHOD_DEVX_ASYNC_CMD_FD_ALLOC));
3056 
3057 DECLARE_UVERBS_NAMED_METHOD(
3058 	MLX5_IB_METHOD_DEVX_ASYNC_EVENT_FD_ALLOC,
3059 	UVERBS_ATTR_FD(MLX5_IB_ATTR_DEVX_ASYNC_EVENT_FD_ALLOC_HANDLE,
3060 			MLX5_IB_OBJECT_DEVX_ASYNC_EVENT_FD,
3061 			UVERBS_ACCESS_NEW,
3062 			UA_MANDATORY),
3063 	UVERBS_ATTR_FLAGS_IN(MLX5_IB_ATTR_DEVX_ASYNC_EVENT_FD_ALLOC_FLAGS,
3064 			enum mlx5_ib_uapi_devx_create_event_channel_flags,
3065 			UA_MANDATORY));
3066 
3067 DECLARE_UVERBS_NAMED_OBJECT(
3068 	MLX5_IB_OBJECT_DEVX_ASYNC_EVENT_FD,
3069 	UVERBS_TYPE_ALLOC_FD(sizeof(struct devx_async_event_file),
3070 			     devx_async_event_destroy_uobj,
3071 			     &devx_async_event_fops, "[devx_async_event]",
3072 			     O_RDONLY),
3073 	&UVERBS_METHOD(MLX5_IB_METHOD_DEVX_ASYNC_EVENT_FD_ALLOC));
3074 
3075 static bool devx_is_supported(struct ib_device *device)
3076 {
3077 	struct mlx5_ib_dev *dev = to_mdev(device);
3078 
3079 	return MLX5_CAP_GEN(dev->mdev, log_max_uctx);
3080 }
3081 
3082 const struct uapi_definition mlx5_ib_devx_defs[] = {
3083 	UAPI_DEF_CHAIN_OBJ_TREE_NAMED(
3084 		MLX5_IB_OBJECT_DEVX,
3085 		UAPI_DEF_IS_OBJ_SUPPORTED(devx_is_supported)),
3086 	UAPI_DEF_CHAIN_OBJ_TREE_NAMED(
3087 		MLX5_IB_OBJECT_DEVX_OBJ,
3088 		UAPI_DEF_IS_OBJ_SUPPORTED(devx_is_supported)),
3089 	UAPI_DEF_CHAIN_OBJ_TREE_NAMED(
3090 		MLX5_IB_OBJECT_DEVX_UMEM,
3091 		UAPI_DEF_IS_OBJ_SUPPORTED(devx_is_supported)),
3092 	UAPI_DEF_CHAIN_OBJ_TREE_NAMED(
3093 		MLX5_IB_OBJECT_DEVX_ASYNC_CMD_FD,
3094 		UAPI_DEF_IS_OBJ_SUPPORTED(devx_is_supported)),
3095 	UAPI_DEF_CHAIN_OBJ_TREE_NAMED(
3096 		MLX5_IB_OBJECT_DEVX_ASYNC_EVENT_FD,
3097 		UAPI_DEF_IS_OBJ_SUPPORTED(devx_is_supported)),
3098 	{},
3099 };
3100