Loading...
Note: File does not exist in v3.5.6.
1/* SPDX-License-Identifier: GPL-2.0 */
2/*
3 * Shared Memory Communications over RDMA (SMC-R) and RoCE
4 *
5 * Definitions for SMC Connections, Link Groups and Links
6 *
7 * Copyright IBM Corp. 2016
8 *
9 * Author(s): Ursula Braun <ubraun@linux.vnet.ibm.com>
10 */
11
12#ifndef _SMC_CORE_H
13#define _SMC_CORE_H
14
15#include <linux/atomic.h>
16#include <linux/smc.h>
17#include <linux/pci.h>
18#include <rdma/ib_verbs.h>
19#include <net/genetlink.h>
20
21#include "smc.h"
22#include "smc_ib.h"
23
24#define SMC_RMBS_PER_LGR_MAX 255 /* max. # of RMBs per link group */
25
26struct smc_lgr_list { /* list of link group definition */
27 struct list_head list;
28 spinlock_t lock; /* protects list of link groups */
29 u32 num; /* unique link group number */
30};
31
32enum smc_lgr_role { /* possible roles of a link group */
33 SMC_CLNT, /* client */
34 SMC_SERV /* server */
35};
36
37enum smc_link_state { /* possible states of a link */
38 SMC_LNK_UNUSED, /* link is unused */
39 SMC_LNK_INACTIVE, /* link is inactive */
40 SMC_LNK_ACTIVATING, /* link is being activated */
41 SMC_LNK_ACTIVE, /* link is active */
42};
43
44#define SMC_WR_BUF_SIZE 48 /* size of work request buffer */
45
46struct smc_wr_buf {
47 u8 raw[SMC_WR_BUF_SIZE];
48};
49
50#define SMC_WR_REG_MR_WAIT_TIME (5 * HZ)/* wait time for ib_wr_reg_mr result */
51
52enum smc_wr_reg_state {
53 POSTED, /* ib_wr_reg_mr request posted */
54 CONFIRMED, /* ib_wr_reg_mr response: successful */
55 FAILED /* ib_wr_reg_mr response: failure */
56};
57
58struct smc_rdma_sge { /* sges for RDMA writes */
59 struct ib_sge wr_tx_rdma_sge[SMC_IB_MAX_SEND_SGE];
60};
61
62#define SMC_MAX_RDMA_WRITES 2 /* max. # of RDMA writes per
63 * message send
64 */
65
66struct smc_rdma_sges { /* sges per message send */
67 struct smc_rdma_sge tx_rdma_sge[SMC_MAX_RDMA_WRITES];
68};
69
70struct smc_rdma_wr { /* work requests per message
71 * send
72 */
73 struct ib_rdma_wr wr_tx_rdma[SMC_MAX_RDMA_WRITES];
74};
75
76#define SMC_LGR_ID_SIZE 4
77
78struct smc_link {
79 struct smc_ib_device *smcibdev; /* ib-device */
80 u8 ibport; /* port - values 1 | 2 */
81 struct ib_pd *roce_pd; /* IB protection domain,
82 * unique for every RoCE QP
83 */
84 struct ib_qp *roce_qp; /* IB queue pair */
85 struct ib_qp_attr qp_attr; /* IB queue pair attributes */
86
87 struct smc_wr_buf *wr_tx_bufs; /* WR send payload buffers */
88 struct ib_send_wr *wr_tx_ibs; /* WR send meta data */
89 struct ib_sge *wr_tx_sges; /* WR send gather meta data */
90 struct smc_rdma_sges *wr_tx_rdma_sges;/*RDMA WRITE gather meta data*/
91 struct smc_rdma_wr *wr_tx_rdmas; /* WR RDMA WRITE */
92 struct smc_wr_tx_pend *wr_tx_pends; /* WR send waiting for CQE */
93 struct completion *wr_tx_compl; /* WR send CQE completion */
94 /* above four vectors have wr_tx_cnt elements and use the same index */
95 dma_addr_t wr_tx_dma_addr; /* DMA address of wr_tx_bufs */
96 atomic_long_t wr_tx_id; /* seq # of last sent WR */
97 unsigned long *wr_tx_mask; /* bit mask of used indexes */
98 u32 wr_tx_cnt; /* number of WR send buffers */
99 wait_queue_head_t wr_tx_wait; /* wait for free WR send buf */
100 atomic_t wr_tx_refcnt; /* tx refs to link */
101
102 struct smc_wr_buf *wr_rx_bufs; /* WR recv payload buffers */
103 struct ib_recv_wr *wr_rx_ibs; /* WR recv meta data */
104 struct ib_sge *wr_rx_sges; /* WR recv scatter meta data */
105 /* above three vectors have wr_rx_cnt elements and use the same index */
106 dma_addr_t wr_rx_dma_addr; /* DMA address of wr_rx_bufs */
107 u64 wr_rx_id; /* seq # of last recv WR */
108 u32 wr_rx_cnt; /* number of WR recv buffers */
109 unsigned long wr_rx_tstamp; /* jiffies when last buf rx */
110
111 struct ib_reg_wr wr_reg; /* WR register memory region */
112 wait_queue_head_t wr_reg_wait; /* wait for wr_reg result */
113 atomic_t wr_reg_refcnt; /* reg refs to link */
114 enum smc_wr_reg_state wr_reg_state; /* state of wr_reg request */
115
116 u8 gid[SMC_GID_SIZE];/* gid matching used vlan id*/
117 u8 sgid_index; /* gid index for vlan id */
118 u32 peer_qpn; /* QP number of peer */
119 enum ib_mtu path_mtu; /* used mtu */
120 enum ib_mtu peer_mtu; /* mtu size of peer */
121 u32 psn_initial; /* QP tx initial packet seqno */
122 u32 peer_psn; /* QP rx initial packet seqno */
123 u8 peer_mac[ETH_ALEN]; /* = gid[8:10||13:15] */
124 u8 peer_gid[SMC_GID_SIZE]; /* gid of peer*/
125 u8 link_id; /* unique # within link group */
126 u8 link_uid[SMC_LGR_ID_SIZE]; /* unique lnk id */
127 u8 peer_link_uid[SMC_LGR_ID_SIZE]; /* peer uid */
128 u8 link_idx; /* index in lgr link array */
129 u8 link_is_asym; /* is link asymmetric? */
130 struct smc_link_group *lgr; /* parent link group */
131 struct work_struct link_down_wrk; /* wrk to bring link down */
132 char ibname[IB_DEVICE_NAME_MAX]; /* ib device name */
133 int ndev_ifidx; /* network device ifindex */
134
135 enum smc_link_state state; /* state of link */
136 struct delayed_work llc_testlink_wrk; /* testlink worker */
137 struct completion llc_testlink_resp; /* wait for rx of testlink */
138 int llc_testlink_time; /* testlink interval */
139 atomic_t conn_cnt; /* connections on this link */
140};
141
142/* For now we just allow one parallel link per link group. The SMC protocol
143 * allows more (up to 8).
144 */
145#define SMC_LINKS_PER_LGR_MAX 3
146#define SMC_SINGLE_LINK 0
147
148/* tx/rx buffer list element for sndbufs list and rmbs list of a lgr */
149struct smc_buf_desc {
150 struct list_head list;
151 void *cpu_addr; /* virtual address of buffer */
152 struct page *pages;
153 int len; /* length of buffer */
154 u32 used; /* currently used / unused */
155 union {
156 struct { /* SMC-R */
157 struct sg_table sgt[SMC_LINKS_PER_LGR_MAX];
158 /* virtual buffer */
159 struct ib_mr *mr_rx[SMC_LINKS_PER_LGR_MAX];
160 /* for rmb only: memory region
161 * incl. rkey provided to peer
162 */
163 u32 order; /* allocation order */
164
165 u8 is_conf_rkey;
166 /* confirm_rkey done */
167 u8 is_reg_mr[SMC_LINKS_PER_LGR_MAX];
168 /* mem region registered */
169 u8 is_map_ib[SMC_LINKS_PER_LGR_MAX];
170 /* mem region mapped to lnk */
171 u8 is_reg_err;
172 /* buffer registration err */
173 };
174 struct { /* SMC-D */
175 unsigned short sba_idx;
176 /* SBA index number */
177 u64 token;
178 /* DMB token number */
179 dma_addr_t dma_addr;
180 /* DMA address */
181 };
182 };
183};
184
185struct smc_rtoken { /* address/key of remote RMB */
186 u64 dma_addr;
187 u32 rkey;
188};
189
190#define SMC_BUF_MIN_SIZE 16384 /* minimum size of an RMB */
191#define SMC_RMBE_SIZES 16 /* number of distinct RMBE sizes */
192/* theoretically, the RFC states that largest size would be 512K,
193 * i.e. compressed 5 and thus 6 sizes (0..5), despite
194 * struct smc_clc_msg_accept_confirm.rmbe_size being a 4 bit value (0..15)
195 */
196
197struct smcd_dev;
198
199enum smc_lgr_type { /* redundancy state of lgr */
200 SMC_LGR_NONE, /* no active links, lgr to be deleted */
201 SMC_LGR_SINGLE, /* 1 active RNIC on each peer */
202 SMC_LGR_SYMMETRIC, /* 2 active RNICs on each peer */
203 SMC_LGR_ASYMMETRIC_PEER, /* local has 2, peer 1 active RNICs */
204 SMC_LGR_ASYMMETRIC_LOCAL, /* local has 1, peer 2 active RNICs */
205};
206
207enum smc_llc_flowtype {
208 SMC_LLC_FLOW_NONE = 0,
209 SMC_LLC_FLOW_ADD_LINK = 2,
210 SMC_LLC_FLOW_DEL_LINK = 4,
211 SMC_LLC_FLOW_RKEY = 6,
212};
213
214struct smc_llc_qentry;
215
216struct smc_llc_flow {
217 enum smc_llc_flowtype type;
218 struct smc_llc_qentry *qentry;
219};
220
221struct smc_link_group {
222 struct list_head list;
223 struct rb_root conns_all; /* connection tree */
224 rwlock_t conns_lock; /* protects conns_all */
225 unsigned int conns_num; /* current # of connections */
226 unsigned short vlan_id; /* vlan id of link group */
227
228 struct list_head sndbufs[SMC_RMBE_SIZES];/* tx buffers */
229 struct mutex sndbufs_lock; /* protects tx buffers */
230 struct list_head rmbs[SMC_RMBE_SIZES]; /* rx buffers */
231 struct mutex rmbs_lock; /* protects rx buffers */
232
233 u8 id[SMC_LGR_ID_SIZE]; /* unique lgr id */
234 struct delayed_work free_work; /* delayed freeing of an lgr */
235 struct work_struct terminate_work; /* abnormal lgr termination */
236 struct workqueue_struct *tx_wq; /* wq for conn. tx workers */
237 u8 sync_err : 1; /* lgr no longer fits to peer */
238 u8 terminating : 1;/* lgr is terminating */
239 u8 freeing : 1; /* lgr is being freed */
240
241 bool is_smcd; /* SMC-R or SMC-D */
242 u8 smc_version;
243 u8 negotiated_eid[SMC_MAX_EID_LEN];
244 u8 peer_os; /* peer operating system */
245 u8 peer_smc_release;
246 u8 peer_hostname[SMC_MAX_HOSTNAME_LEN];
247 union {
248 struct { /* SMC-R */
249 enum smc_lgr_role role;
250 /* client or server */
251 struct smc_link lnk[SMC_LINKS_PER_LGR_MAX];
252 /* smc link */
253 char peer_systemid[SMC_SYSTEMID_LEN];
254 /* unique system_id of peer */
255 struct smc_rtoken rtokens[SMC_RMBS_PER_LGR_MAX]
256 [SMC_LINKS_PER_LGR_MAX];
257 /* remote addr/key pairs */
258 DECLARE_BITMAP(rtokens_used_mask, SMC_RMBS_PER_LGR_MAX);
259 /* used rtoken elements */
260 u8 next_link_id;
261 enum smc_lgr_type type;
262 /* redundancy state */
263 u8 pnet_id[SMC_MAX_PNETID_LEN + 1];
264 /* pnet id of this lgr */
265 struct list_head llc_event_q;
266 /* queue for llc events */
267 spinlock_t llc_event_q_lock;
268 /* protects llc_event_q */
269 struct mutex llc_conf_mutex;
270 /* protects lgr reconfig. */
271 struct work_struct llc_add_link_work;
272 struct work_struct llc_del_link_work;
273 struct work_struct llc_event_work;
274 /* llc event worker */
275 wait_queue_head_t llc_flow_waiter;
276 /* w4 next llc event */
277 wait_queue_head_t llc_msg_waiter;
278 /* w4 next llc msg */
279 struct smc_llc_flow llc_flow_lcl;
280 /* llc local control field */
281 struct smc_llc_flow llc_flow_rmt;
282 /* llc remote control field */
283 struct smc_llc_qentry *delayed_event;
284 /* arrived when flow active */
285 spinlock_t llc_flow_lock;
286 /* protects llc flow */
287 int llc_testlink_time;
288 /* link keep alive time */
289 u32 llc_termination_rsn;
290 /* rsn code for termination */
291 };
292 struct { /* SMC-D */
293 u64 peer_gid;
294 /* Peer GID (remote) */
295 struct smcd_dev *smcd;
296 /* ISM device for VLAN reg. */
297 u8 peer_shutdown : 1;
298 /* peer triggered shutdownn */
299 };
300 };
301};
302
303struct smc_clc_msg_local;
304
305struct smc_init_info {
306 u8 is_smcd;
307 u8 smc_type_v1;
308 u8 smc_type_v2;
309 u8 first_contact_peer;
310 u8 first_contact_local;
311 unsigned short vlan_id;
312 u32 rc;
313 /* SMC-R */
314 struct smc_clc_msg_local *ib_lcl;
315 struct smc_ib_device *ib_dev;
316 u8 ib_gid[SMC_GID_SIZE];
317 u8 ib_port;
318 u32 ib_clcqpn;
319 /* SMC-D */
320 u64 ism_peer_gid[SMC_MAX_ISM_DEVS + 1];
321 struct smcd_dev *ism_dev[SMC_MAX_ISM_DEVS + 1];
322 u16 ism_chid[SMC_MAX_ISM_DEVS + 1];
323 u8 ism_offered_cnt; /* # of ISM devices offered */
324 u8 ism_selected; /* index of selected ISM dev*/
325 u8 smcd_version;
326};
327
328/* Find the connection associated with the given alert token in the link group.
329 * To use rbtrees we have to implement our own search core.
330 * Requires @conns_lock
331 * @token alert token to search for
332 * @lgr link group to search in
333 * Returns connection associated with token if found, NULL otherwise.
334 */
335static inline struct smc_connection *smc_lgr_find_conn(
336 u32 token, struct smc_link_group *lgr)
337{
338 struct smc_connection *res = NULL;
339 struct rb_node *node;
340
341 node = lgr->conns_all.rb_node;
342 while (node) {
343 struct smc_connection *cur = rb_entry(node,
344 struct smc_connection, alert_node);
345
346 if (cur->alert_token_local > token) {
347 node = node->rb_left;
348 } else {
349 if (cur->alert_token_local < token) {
350 node = node->rb_right;
351 } else {
352 res = cur;
353 break;
354 }
355 }
356 }
357
358 return res;
359}
360
361/* returns true if the specified link is usable */
362static inline bool smc_link_usable(struct smc_link *lnk)
363{
364 if (lnk->state == SMC_LNK_UNUSED || lnk->state == SMC_LNK_INACTIVE)
365 return false;
366 return true;
367}
368
369static inline bool smc_link_active(struct smc_link *lnk)
370{
371 return lnk->state == SMC_LNK_ACTIVE;
372}
373
374static inline void smc_gid_be16_convert(__u8 *buf, u8 *gid_raw)
375{
376 sprintf(buf, "%04x:%04x:%04x:%04x:%04x:%04x:%04x:%04x",
377 be16_to_cpu(((__be16 *)gid_raw)[0]),
378 be16_to_cpu(((__be16 *)gid_raw)[1]),
379 be16_to_cpu(((__be16 *)gid_raw)[2]),
380 be16_to_cpu(((__be16 *)gid_raw)[3]),
381 be16_to_cpu(((__be16 *)gid_raw)[4]),
382 be16_to_cpu(((__be16 *)gid_raw)[5]),
383 be16_to_cpu(((__be16 *)gid_raw)[6]),
384 be16_to_cpu(((__be16 *)gid_raw)[7]));
385}
386
387struct smc_pci_dev {
388 __u32 pci_fid;
389 __u16 pci_pchid;
390 __u16 pci_vendor;
391 __u16 pci_device;
392 __u8 pci_id[SMC_PCI_ID_STR_LEN];
393};
394
395static inline void smc_set_pci_values(struct pci_dev *pci_dev,
396 struct smc_pci_dev *smc_dev)
397{
398 smc_dev->pci_vendor = pci_dev->vendor;
399 smc_dev->pci_device = pci_dev->device;
400 snprintf(smc_dev->pci_id, sizeof(smc_dev->pci_id), "%s",
401 pci_name(pci_dev));
402#if IS_ENABLED(CONFIG_S390)
403 { /* Set s390 specific PCI information */
404 struct zpci_dev *zdev;
405
406 zdev = to_zpci(pci_dev);
407 smc_dev->pci_fid = zdev->fid;
408 smc_dev->pci_pchid = zdev->pchid;
409 }
410#endif
411}
412
413struct smc_sock;
414struct smc_clc_msg_accept_confirm;
415
416void smc_lgr_cleanup_early(struct smc_connection *conn);
417void smc_lgr_terminate_sched(struct smc_link_group *lgr);
418void smcr_port_add(struct smc_ib_device *smcibdev, u8 ibport);
419void smcr_port_err(struct smc_ib_device *smcibdev, u8 ibport);
420void smc_smcd_terminate(struct smcd_dev *dev, u64 peer_gid,
421 unsigned short vlan);
422void smc_smcd_terminate_all(struct smcd_dev *dev);
423void smc_smcr_terminate_all(struct smc_ib_device *smcibdev);
424int smc_buf_create(struct smc_sock *smc, bool is_smcd);
425int smc_uncompress_bufsize(u8 compressed);
426int smc_rmb_rtoken_handling(struct smc_connection *conn, struct smc_link *link,
427 struct smc_clc_msg_accept_confirm *clc);
428int smc_rtoken_add(struct smc_link *lnk, __be64 nw_vaddr, __be32 nw_rkey);
429int smc_rtoken_delete(struct smc_link *lnk, __be32 nw_rkey);
430void smc_rtoken_set(struct smc_link_group *lgr, int link_idx, int link_idx_new,
431 __be32 nw_rkey_known, __be64 nw_vaddr, __be32 nw_rkey);
432void smc_rtoken_set2(struct smc_link_group *lgr, int rtok_idx, int link_id,
433 __be64 nw_vaddr, __be32 nw_rkey);
434void smc_sndbuf_sync_sg_for_cpu(struct smc_connection *conn);
435void smc_sndbuf_sync_sg_for_device(struct smc_connection *conn);
436void smc_rmb_sync_sg_for_cpu(struct smc_connection *conn);
437void smc_rmb_sync_sg_for_device(struct smc_connection *conn);
438int smc_vlan_by_tcpsk(struct socket *clcsock, struct smc_init_info *ini);
439
440void smc_conn_free(struct smc_connection *conn);
441int smc_conn_create(struct smc_sock *smc, struct smc_init_info *ini);
442void smc_lgr_schedule_free_work_fast(struct smc_link_group *lgr);
443int smc_core_init(void);
444void smc_core_exit(void);
445
446int smcr_link_init(struct smc_link_group *lgr, struct smc_link *lnk,
447 u8 link_idx, struct smc_init_info *ini);
448void smcr_link_clear(struct smc_link *lnk, bool log);
449void smc_switch_link_and_count(struct smc_connection *conn,
450 struct smc_link *to_lnk);
451int smcr_buf_map_lgr(struct smc_link *lnk);
452int smcr_buf_reg_lgr(struct smc_link *lnk);
453void smcr_lgr_set_type(struct smc_link_group *lgr, enum smc_lgr_type new_type);
454void smcr_lgr_set_type_asym(struct smc_link_group *lgr,
455 enum smc_lgr_type new_type, int asym_lnk_idx);
456int smcr_link_reg_rmb(struct smc_link *link, struct smc_buf_desc *rmb_desc);
457struct smc_link *smc_switch_conns(struct smc_link_group *lgr,
458 struct smc_link *from_lnk, bool is_dev_err);
459void smcr_link_down_cond(struct smc_link *lnk);
460void smcr_link_down_cond_sched(struct smc_link *lnk);
461int smc_nl_get_sys_info(struct sk_buff *skb, struct netlink_callback *cb);
462int smcr_nl_get_lgr(struct sk_buff *skb, struct netlink_callback *cb);
463int smcr_nl_get_link(struct sk_buff *skb, struct netlink_callback *cb);
464int smcd_nl_get_lgr(struct sk_buff *skb, struct netlink_callback *cb);
465
466static inline struct smc_link_group *smc_get_lgr(struct smc_link *link)
467{
468 return link->lgr;
469}
470#endif