xref: /illumos-gate/usr/src/uts/common/io/comstar/port/srpt/srpt_impl.h (revision b6805bf78d2bbbeeaea8909a05623587b42d58b3)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
24  */
25 
26 #ifndef _SRPT_IMPL_H_
27 #define	_SRPT_IMPL_H_
28 
29 /*
30  * Prototypes and data structures for the SRP Target Port Provider.
31  */
32 
33 #include <sys/types.h>
34 #include <sys/ddi.h>
35 #include <sys/ib/ibtl/ibti.h>
36 #include <sys/modctl.h>
37 
38 #include <sys/stmf.h>
39 #include <sys/stmf_ioctl.h>
40 #include <sys/portif.h>
41 
42 #include <sys/ib/mgt/ibdma/ibdma.h>
43 
44 #ifdef __cplusplus
45 extern "C" {
46 #endif
47 
48 /* Format the session identifier */
49 #define	ALIAS_STR(s, a, b)						\
50 	((void) snprintf((s), sizeof ((s)), "%016llx:%016llx",		\
51 	    (u_longlong_t)(a), (u_longlong_t)(b)))
52 
53 /* Format the EUI name */
54 #define	EUI_STR(s, a)							\
55 	((void) snprintf((s), sizeof ((s)), "eui.%016llX", (u_longlong_t)(a)))
56 
57 /*
58  * We should/could consider making some of these values tunables.
59  * Specifically, SEND_MSG_SIZE and SEND_MSG_DEPTH.
60  */
61 enum {
62 	SRPT_DEFAULT_IOC_SRQ_SIZE = 4096,
63 	SRPT_DEFAULT_SEND_MSG_DEPTH = 128,
64 	/*
65 	 * SEND_MSG_SIZE must be a multiple of 64 as it is registered
66 	 * as memory regions with IB.  To support a scatter/gather table
67 	 * size of 32, the size must be at not less than 960.  To support
68 	 * the maximum scatter/gather table size of 255, the IU must
69 	 * be at least 4160 bytes.
70 	 */
71 	SRPT_DEFAULT_SEND_MSG_SIZE = 4160,
72 	SRPT_DEFAULT_MAX_RDMA_SIZE = 65536,
73 	SRPT_MIN_T_I_IU_LEN = 52,
74 	SRPT_EUI_ID_LEN = 20,
75 	SRPT_RECV_WC_POLL_SIZE = 16,
76 	SRPT_SEND_WC_POLL_SIZE = 16,
77 	SRPT_MAX_OUT_IO_PER_CMD = 16,
78 	SRPT_FENCE_SEND = 1,
79 	SRPT_NO_FENCE_SEND = 0
80 };
81 
82 struct srpt_target_port_s;
83 
84 #define	SRPT_ALIAS_LEN	(SRP_PORT_ID_LEN * 2 + 2)
85 
86 /*
87  * SRP Session - represents a SCSI I_T_Nexus.
88  *
89  * Sessions map 1 or more initiator logins to a specific I/O
90  * Controller SCSI Target Port.  Targets create sessions
91  * at initiator login and release when no longer referenced
92  * by a login.
93  */
94 typedef struct srpt_session_s {
95 	krwlock_t 			ss_rwlock;
96 	list_node_t			ss_node;
97 
98 	/*
99 	 * ADVANCED FEATURE, NOT YET SUPPORTED.
100 	 * In multi-channel mode, multiple RDMA communication
101 	 * channels may reference the same SCSI session.  When
102 	 * a channel releases its reference to the SCSI session,
103 	 * it should have no tasks associated with the session.
104 	 *
105 	 * If multi-channel is implemented, add a channel list
106 	 * to this object instead of tracking it on the target.
107 	 *
108 	 * Will also need a session state & mode.  Mode is to
109 	 * track if the session is MULTI or SINGLE channel.
110 	 */
111 
112 	stmf_scsi_session_t		*ss_ss;
113 	struct srpt_target_port_s	*ss_tgt;
114 	list_t				ss_task_list;
115 
116 	/*
117 	 * SRP Initiator and target identifiers are 128-bit.
118 	 *
119 	 * The specification defines the initiator to be 64-bits of
120 	 * ID extension and 64 bits of GUID, but these are really
121 	 * just a recommendation.  Generally the extension is used
122 	 * to create unique I_T_Nexus from the same initiator and
123 	 * target.  Initiators are inconsistent on the GUID they
124 	 * use, some use the HCA Node, some the HCA port.
125 	 *
126 	 * The specification defines the target to be 64-bits of
127 	 * service ID followed by 64-bits of I/O Controller GUID.
128 	 * In the case where there is a single default target
129 	 * service, they will be the same (our default).
130 	 */
131 	uint8_t				ss_i_id[SRP_PORT_ID_LEN];
132 	uint8_t				ss_t_id[SRP_PORT_ID_LEN];
133 
134 	/* So we can see the full 128-bit initiator login from stmfadm */
135 	char				ss_i_alias[SRPT_ALIAS_LEN];
136 	uint8_t				ss_hw_port;
137 
138 	char				ss_t_alias[SRPT_ALIAS_LEN];
139 	char				ss_i_name[SRPT_EUI_ID_LEN + 1];
140 	char				ss_t_name[SRPT_EUI_ID_LEN + 1];
141 	char				ss_i_gid[SRPT_ALIAS_LEN];
142 	char				ss_t_gid[SRPT_ALIAS_LEN];
143 } srpt_session_t;
144 
145 /*
146  * Send work request types.
147  */
148 typedef enum srpt_swqe_type_e {
149 	SRPT_SWQE_TYPE_DATA = 1,
150 	SRPT_SWQE_TYPE_RESP
151 } srpt_swqe_type_t;
152 
153 typedef struct srpt_swqe_s {
154 	srpt_swqe_type_t	sw_type;
155 	void			*sw_addr;
156 	ibt_wrid_t		sw_next;
157 } srpt_swqe_t;
158 
159 /*
160  * SRP Channel - the RDMA communications channel associated with
161  * a specific SRP login.
162  */
163 typedef enum srpt_channel_state_e {
164 	SRPT_CHANNEL_CONNECTING = 0,
165 	SRPT_CHANNEL_CONNECTED,
166 	SRPT_CHANNEL_DISCONNECTING
167 } srpt_channel_state_t;
168 
169 typedef struct srpt_channel_s {
170 	krwlock_t 			ch_rwlock;
171 
172 	kmutex_t			ch_reflock;
173 	uint_t				ch_refcnt;
174 	kcondvar_t			ch_cv_complete;
175 	uint_t				ch_cv_waiters;
176 
177 	list_node_t			ch_stp_node;
178 	srpt_channel_state_t		ch_state;
179 	ibt_cq_hdl_t			ch_scq_hdl;
180 	ibt_cq_hdl_t			ch_rcq_hdl;
181 	ibt_channel_hdl_t		ch_chan_hdl;
182 	ibt_chan_sizes_t		ch_sizes;
183 
184 	uint32_t			ch_req_lim_delta;
185 	uint32_t			ch_ti_iu_len;
186 	struct srpt_target_port_s	*ch_tgt;
187 	srpt_session_t			*ch_session;
188 
189 	/*
190 	 * Map IB send WQE request IDs to the
191 	 * apporpriate operation type (for errors).
192 	 */
193 	kmutex_t			ch_swqe_lock;
194 	srpt_swqe_t			*ch_swqe;
195 	uint32_t			ch_num_swqe;
196 	uint32_t			ch_head;
197 	uint32_t			ch_tail;
198 	uint32_t			ch_swqe_posted;
199 } srpt_channel_t;
200 
201 /*
202  * SRP Information Unit (IU).  Each IU structure contains
203  * the buffer for the IU itself (received over the RC
204  * channel), and all of the context required by the target
205  * to process this request represented by the IU.
206  * Available IU structures are managed on the I/O Controller
207  * shared receive queue.
208  */
209 enum {
210 	SRPT_IU_STMF_ABORTING	= 1 << 0,	/* STMF called abort */
211 	SRPT_IU_SRP_ABORTING	= 1 << 1,	/* SRP initiator aborting */
212 	SRPT_IU_ABORTED		= 1 << 2,	/* Task has been aborted */
213 	SRPT_IU_RESP_SENT	= 1 << 3	/* Response queued */
214 };
215 
216 typedef struct srpt_iu_s {
217 	/*
218 	 * The buffer for the IU itself.  When unused (a
219 	 * reference count of zero), this buffer is posted
220 	 * on the I/O Controllers SRPT SRQ.
221 	 */
222 	void			*iu_buf;
223 	ibt_wr_ds_t		iu_sge;
224 	struct srpt_ioc_s	*iu_ioc;
225 	uint_t			iu_pool_ndx;
226 	kmutex_t		iu_lock;
227 
228 	/*
229 	 * The following are reset for each IU request
230 	 * processed by this buffer.
231 	 */
232 	list_node_t		iu_ss_task_node;
233 	srpt_channel_t		*iu_ch;
234 
235 	uint_t			iu_num_rdescs;
236 	srp_direct_desc_t	*iu_rdescs;
237 	uint_t			iu_tot_xfer_len;
238 
239 	uint64_t		iu_tag;
240 	uint_t			iu_flags;
241 	uint32_t		iu_sq_posted_cnt;
242 	scsi_task_t		*iu_stmf_task;
243 } srpt_iu_t;
244 
245 /*
246  * SRP SCSI Target Port.  By default each HCA creates a single
247  * SCSI Target Port based on the associated I/O Controller
248  * (HCA) node GUID and made available through each physical
249  * hardware port of the I/O Controller.
250  */
251 typedef enum srpt_target_state_e {
252 	SRPT_TGT_STATE_OFFLINE = 0,
253 	SRPT_TGT_STATE_ONLINING,
254 	SRPT_TGT_STATE_ONLINE,
255 	SRPT_TGT_STATE_OFFLINING
256 } srpt_target_state_t;
257 
258 typedef struct srpt_hw_port_s {
259 	ibt_sbind_hdl_t		hwp_bind_hdl;
260 	ib_gid_t		hwp_gid;
261 } srpt_hw_port_t;
262 
263 typedef struct srpt_target_port_s {
264 	stmf_local_port_t	*tp_lport;
265 	struct srpt_ioc_s	*tp_ioc;
266 
267 	kmutex_t		tp_lock;
268 	srpt_target_state_t	tp_state;
269 	kcondvar_t		tp_offline_complete;
270 	uint_t			tp_drv_disabled;
271 
272 	/*
273 	 * We are using a simple list for channels right now, we
274 	 * probably should  switch this over to the AVL
275 	 * implementation eventually (but lookups are not done
276 	 * in the data path so this is not urgent).
277 	 */
278 	kmutex_t		tp_ch_list_lock;
279 	list_t			tp_ch_list;
280 
281 	/*
282 	 * A list of active sessions.  Session lifetime is
283 	 * determined by having active channels, but track
284 	 * them here for easier determination to when a
285 	 * target can truly be offlined, and as a step toward
286 	 * being session-focused rather than channel-focused.
287 	 * If we ever truly support multi-channel, move the
288 	 * channels to be part of the session object.
289 	 *
290 	 * Sessions should remain on this list until they
291 	 * are deregistered from STMF.  This allows the target
292 	 * to properly track when it can consider itself 'offline'.
293 	 */
294 	kmutex_t		tp_sess_list_lock;
295 	kcondvar_t		tp_sess_complete;
296 	list_t			tp_sess_list;
297 
298 	uint_t			tp_srp_enabled;
299 	ibt_srv_hdl_t		tp_ibt_svc_hdl;
300 	ibt_srv_desc_t		tp_ibt_svc_desc;
301 	ib_svc_id_t		tp_ibt_svc_id;
302 	scsi_devid_desc_t	*tp_scsi_devid;
303 	uint8_t			tp_srp_port_id[SRP_PORT_ID_LEN];
304 
305 	uint_t			tp_nports;
306 	srpt_hw_port_t		*tp_hw_port;
307 	/*
308 	 * track the number of active ports so we can offline the target if
309 	 * none
310 	 */
311 	uint32_t		tp_num_active_ports;
312 	/* state STMF wants the target in.  We may be offline due to no ports */
313 	srpt_target_state_t	tp_requested_state;
314 } srpt_target_port_t;
315 
316 /*
317  * SRP Target hardware device.  A SRP Target hardware device
318  * is an IB HCA.  All ports of the HCA comprise a single
319  * I/O Controller that is registered with the IB Device
320  * Managment Agent.
321  */
322 typedef struct srpt_ioc_s {
323 	list_node_t			ioc_node;
324 
325 	krwlock_t 			ioc_rwlock;
326 	ibt_hca_hdl_t			ioc_ibt_hdl;
327 	ibt_hca_attr_t			ioc_attr;
328 	ib_guid_t			ioc_guid;
329 
330 	/*
331 	 * By default each HCA is a single SRP.T10 service based on
332 	 * the HCA GUID.  We have implemented the target here as a
333 	 * pointer to facilitate moving to a list of targets if
334 	 * appropriate down the road.
335 	 */
336 	srpt_target_port_t		*ioc_tgt_port;
337 
338 
339 	/*
340 	 * Each HCA registers a single I/O Controller with the
341 	 * IB Device Management Agent.
342 	 */
343 	ibdma_hdl_t			ioc_ibdma_hdl;
344 	ib_dm_ioc_ctrl_profile_t	ioc_profile;
345 	ib_dm_srv_t			ioc_svc;
346 
347 	ibt_pd_hdl_t			ioc_pd_hdl;
348 	ibt_srq_sizes_t			ioc_srq_attr;
349 	ibt_srq_hdl_t			ioc_srq_hdl;
350 
351 	/*
352 	 * The I/O Controller pool of IU resources allocated
353 	 * at controller creation.
354 	 */
355 	uint32_t			ioc_num_iu_entries;
356 	srpt_iu_t			*ioc_iu_pool;
357 	ibt_mr_hdl_t			ioc_iu_mr_hdl;
358 	void				*ioc_iu_bufs;  /* iu buffer space */
359 
360 	/*
361 	 * Each I/O Controller has it's own data buffer
362 	 * vmem arena.  Pool is created at controller creation,
363 	 * and expanded as required.  This keeps IB memory
364 	 * registrations to a minimum in the data path.
365 	 */
366 	struct srpt_vmem_pool_s		*ioc_dbuf_pool;
367 	stmf_dbuf_store_t		*ioc_stmf_ds;
368 } srpt_ioc_t;
369 
370 /*
371  * Memory regions
372  */
373 typedef struct srpt_mr_s {
374 	ibt_mr_hdl_t			mr_hdl;
375 	ib_vaddr_t			mr_va;
376 	ib_memlen_t			mr_len;
377 	ibt_lkey_t			mr_lkey;
378 	ibt_rkey_t			mr_rkey;
379 	avl_node_t			mr_avl;
380 } srpt_mr_t;
381 
382 /*
383  * SRP Target vmem arena definition
384  */
385 typedef struct srpt_vmem_pool_s {
386 	srpt_ioc_t		*svp_ioc;
387 	ib_memlen_t		svp_chunksize;
388 	vmem_t			*svp_vmem;
389 	uint64_t		svp_total_size;
390 	uint64_t		svp_max_size;
391 	avl_tree_t		svp_mr_list;
392 	krwlock_t		svp_lock;
393 	ibt_mr_flags_t		svp_flags;
394 } srpt_vmem_pool_t;
395 
396 /*
397  * SRP port provider data buffer, allocated and freed
398  * via calls to the IOC datastore.
399  */
400 typedef struct srpt_ds_dbuf_s {
401 	stmf_data_buf_t			*db_stmf_buf;
402 	srpt_ioc_t			*db_ioc;
403 	ibt_mr_hdl_t			db_mr_hdl;
404 	ibt_wr_ds_t			db_sge;
405 	srpt_iu_t			*db_iu;
406 } srpt_ds_dbuf_t;
407 
408 /*
409  * SRP Target service state
410  */
411 typedef enum {
412 	SRPT_SVC_DISABLED,
413 	SRPT_SVC_ENABLED
414 } srpt_svc_state_t;
415 
416 typedef struct {
417 	ddi_modhandle_t		ibdmah;
418 	ibdma_hdl_t		(*ibdma_register)(ib_guid_t,
419 				    ib_dm_ioc_ctrl_profile_t *, ib_dm_srv_t *);
420 	ibdma_status_t		(*ibdma_unregister)(ibdma_hdl_t);
421 	ibdma_status_t		(*ibdma_update)(ibdma_hdl_t,
422 				    ib_dm_ioc_ctrl_profile_t *, ib_dm_srv_t *);
423 } srpt_ibdma_ops_t;
424 
425 /*
426  * SRP Target protocol driver context data structure, maintaining
427  * the global state of the protocol.
428  */
429 typedef struct srpt_ctxt_s {
430 	dev_info_t			*sc_dip;
431 	krwlock_t			sc_rwlock;
432 	srpt_svc_state_t		sc_svc_state;
433 
434 	ibt_clnt_hdl_t			sc_ibt_hdl;
435 
436 	/*
437 	 * SRP Target I/O Controllers. Each IBT HCA represents an
438 	 * I/O Controller.  Must hold rwlock as a writer to update.
439 	 */
440 	list_t				sc_ioc_list;
441 	uint_t				sc_num_iocs;
442 
443 	/* Back-end COMSTAR port provider interface. */
444 	stmf_port_provider_t		*sc_pp;
445 
446 	/* IBDMA entry points */
447 	srpt_ibdma_ops_t		sc_ibdma_ops;
448 
449 	/*
450 	 *  List of explicitly configured HCAs and their configurable
451 	 *  attributes.
452 	 */
453 	nvlist_t			*sc_cfg_hca_nv;
454 } srpt_ctxt_t;
455 
456 typedef struct srpt_iu_data_s {
457 	union {
458 		uint8_t			srp_op;
459 		srp_cmd_req_t		srp_cmd;
460 		srp_tsk_mgmt_t		srp_tsk_mgmt;
461 		srp_i_logout_t		srp_i_logout;
462 		srp_rsp_t		srp_rsp;
463 	} rx_iu;
464 } srpt_iu_data_t;
465 
466 extern srpt_ctxt_t *srpt_ctxt;
467 
468 /*
469  * For Non recoverable or Major Errors
470  */
471 #define	SRPT_LOG_L0	0
472 
473 /*
474  * For additional information on Non recoverable errors and
475  * warnings/informational message for sys-admin types.
476  */
477 #define	SRPT_LOG_L1	1
478 
479 /*
480  * debug only
481  * for more verbose trace than L1, for e.g. recoverable errors,
482  * or intersting trace
483  */
484 #define	SRPT_LOG_L2	2
485 
486 /*
487  * debug only
488  * for more verbose trace than L2, for e.g. printing function entries....
489  */
490 #define	SRPT_LOG_L3	3
491 
492 /*
493  * debug only
494  * for more verbose trace than L3, for e.g. printing minor function entries...
495  */
496 #define	SRPT_LOG_L4	4
497 
498 /*
499  * srpt_errlevel can be set in the debugger to enable additional logging.
500  * You can also add set srpt:srpt_errlevel={0,1,2,3,4} in /etc/system.
501  * The default log level is L1.
502  */
503 #define	SRPT_LOG_DEFAULT_LEVEL SRPT_LOG_L1
504 
505 extern uint_t srpt_errlevel;
506 
507 
508 #define	SRPT_DPRINTF_L0(...) cmn_err(CE_WARN, __VA_ARGS__)
509 #define	SRPT_DPRINTF_L1(...) cmn_err(CE_NOTE, __VA_ARGS__)
510 #define	SRPT_DPRINTF_L2(...)	if (srpt_errlevel >= SRPT_LOG_L2) { \
511 					cmn_err(CE_NOTE, __VA_ARGS__);\
512 				}
513 #ifdef	DEBUG
514 #define	SRPT_DPRINTF_L3(...)	if (srpt_errlevel >= SRPT_LOG_L3) { \
515 					cmn_err(CE_NOTE, __VA_ARGS__);\
516 				}
517 #define	SRPT_DPRINTF_L4(...)	if (srpt_errlevel >= SRPT_LOG_L4) { \
518 					cmn_err(CE_NOTE, __VA_ARGS__);\
519 				}
520 #else
521 #define	SRPT_DPRINTF_L3		0 &&
522 #define	SRPT_DPRINTF_L4		0 &&
523 #endif
524 
525 #ifdef __cplusplus
526 }
527 #endif
528 
529 #endif /* _SRPT_IMPL_H_ */
530