tjh.dev
Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
1/* SPDX-License-Identifier: GPL-2.0 */
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 <rdma/ib_verbs.h>
17
18#include "smc.h"
19#include "smc_ib.h"
20
21#define SMC_RMBS_PER_LGR_MAX 255 /* max. # of RMBs per link group */
22
23struct smc_lgr_list { /* list of link group definition */
24 struct list_head list;
25 spinlock_t lock; /* protects list of link groups */
26 u32 num; /* unique link group number */
27};
28
29enum smc_lgr_role { /* possible roles of a link group */
30 SMC_CLNT, /* client */
31 SMC_SERV /* server */
32};
33
34enum smc_link_state { /* possible states of a link */
35 SMC_LNK_INACTIVE, /* link is inactive */
36 SMC_LNK_ACTIVATING, /* link is being activated */
37 SMC_LNK_ACTIVE, /* link is active */
38 SMC_LNK_DELETING, /* link is being deleted */
39};
40
41#define SMC_WR_BUF_SIZE 48 /* size of work request buffer */
42
43struct smc_wr_buf {
44 u8 raw[SMC_WR_BUF_SIZE];
45};
46
47#define SMC_WR_REG_MR_WAIT_TIME (5 * HZ)/* wait time for ib_wr_reg_mr result */
48
49enum smc_wr_reg_state {
50 POSTED, /* ib_wr_reg_mr request posted */
51 CONFIRMED, /* ib_wr_reg_mr response: successful */
52 FAILED /* ib_wr_reg_mr response: failure */
53};
54
55struct smc_link {
56 struct smc_ib_device *smcibdev; /* ib-device */
57 u8 ibport; /* port - values 1 | 2 */
58 struct ib_pd *roce_pd; /* IB protection domain,
59 * unique for every RoCE QP
60 */
61 struct ib_qp *roce_qp; /* IB queue pair */
62 struct ib_qp_attr qp_attr; /* IB queue pair attributes */
63
64 struct smc_wr_buf *wr_tx_bufs; /* WR send payload buffers */
65 struct ib_send_wr *wr_tx_ibs; /* WR send meta data */
66 struct ib_sge *wr_tx_sges; /* WR send gather meta data */
67 struct smc_wr_tx_pend *wr_tx_pends; /* WR send waiting for CQE */
68 /* above four vectors have wr_tx_cnt elements and use the same index */
69 dma_addr_t wr_tx_dma_addr; /* DMA address of wr_tx_bufs */
70 atomic_long_t wr_tx_id; /* seq # of last sent WR */
71 unsigned long *wr_tx_mask; /* bit mask of used indexes */
72 u32 wr_tx_cnt; /* number of WR send buffers */
73 wait_queue_head_t wr_tx_wait; /* wait for free WR send buf */
74
75 struct smc_wr_buf *wr_rx_bufs; /* WR recv payload buffers */
76 struct ib_recv_wr *wr_rx_ibs; /* WR recv meta data */
77 struct ib_sge *wr_rx_sges; /* WR recv scatter meta data */
78 /* above three vectors have wr_rx_cnt elements and use the same index */
79 dma_addr_t wr_rx_dma_addr; /* DMA address of wr_rx_bufs */
80 u64 wr_rx_id; /* seq # of last recv WR */
81 u32 wr_rx_cnt; /* number of WR recv buffers */
82 unsigned long wr_rx_tstamp; /* jiffies when last buf rx */
83
84 struct ib_reg_wr wr_reg; /* WR register memory region */
85 wait_queue_head_t wr_reg_wait; /* wait for wr_reg result */
86 enum smc_wr_reg_state wr_reg_state; /* state of wr_reg request */
87
88 u8 gid[SMC_GID_SIZE];/* gid matching used vlan id*/
89 u8 sgid_index; /* gid index for vlan id */
90 u32 peer_qpn; /* QP number of peer */
91 enum ib_mtu path_mtu; /* used mtu */
92 enum ib_mtu peer_mtu; /* mtu size of peer */
93 u32 psn_initial; /* QP tx initial packet seqno */
94 u32 peer_psn; /* QP rx initial packet seqno */
95 u8 peer_mac[ETH_ALEN]; /* = gid[8:10||13:15] */
96 u8 peer_gid[SMC_GID_SIZE]; /* gid of peer*/
97 u8 link_id; /* unique # within link group */
98
99 enum smc_link_state state; /* state of link */
100 struct workqueue_struct *llc_wq; /* single thread work queue */
101 struct completion llc_confirm; /* wait for rx of conf link */
102 struct completion llc_confirm_resp; /* wait 4 rx of cnf lnk rsp */
103 int llc_confirm_rc; /* rc from confirm link msg */
104 int llc_confirm_resp_rc; /* rc from conf_resp msg */
105 struct completion llc_add; /* wait for rx of add link */
106 struct completion llc_add_resp; /* wait for rx of add link rsp*/
107 struct delayed_work llc_testlink_wrk; /* testlink worker */
108 struct completion llc_testlink_resp; /* wait for rx of testlink */
109 int llc_testlink_time; /* testlink interval */
110 struct completion llc_confirm_rkey; /* wait 4 rx of cnf rkey */
111 int llc_confirm_rkey_rc; /* rc from cnf rkey msg */
112};
113
114/* For now we just allow one parallel link per link group. The SMC protocol
115 * allows more (up to 8).
116 */
117#define SMC_LINKS_PER_LGR_MAX 1
118#define SMC_SINGLE_LINK 0
119
120#define SMC_FIRST_CONTACT 1 /* first contact to a peer */
121#define SMC_REUSE_CONTACT 0 /* follow-on contact to a peer*/
122
123/* tx/rx buffer list element for sndbufs list and rmbs list of a lgr */
124struct smc_buf_desc {
125 struct list_head list;
126 void *cpu_addr; /* virtual address of buffer */
127 struct page *pages;
128 int len; /* length of buffer */
129 u32 used; /* currently used / unused */
130 u8 reused : 1; /* new created / reused */
131 u8 regerr : 1; /* err during registration */
132 union {
133 struct { /* SMC-R */
134 struct sg_table sgt[SMC_LINKS_PER_LGR_MAX];
135 /* virtual buffer */
136 struct ib_mr *mr_rx[SMC_LINKS_PER_LGR_MAX];
137 /* for rmb only: memory region
138 * incl. rkey provided to peer
139 */
140 u32 order; /* allocation order */
141 };
142 struct { /* SMC-D */
143 unsigned short sba_idx;
144 /* SBA index number */
145 u64 token;
146 /* DMB token number */
147 dma_addr_t dma_addr;
148 /* DMA address */
149 };
150 };
151};
152
153struct smc_rtoken { /* address/key of remote RMB */
154 u64 dma_addr;
155 u32 rkey;
156};
157
158#define SMC_LGR_ID_SIZE 4
159#define SMC_BUF_MIN_SIZE 16384 /* minimum size of an RMB */
160#define SMC_RMBE_SIZES 16 /* number of distinct RMBE sizes */
161/* theoretically, the RFC states that largest size would be 512K,
162 * i.e. compressed 5 and thus 6 sizes (0..5), despite
163 * struct smc_clc_msg_accept_confirm.rmbe_size being a 4 bit value (0..15)
164 */
165
166struct smcd_dev;
167
168struct smc_link_group {
169 struct list_head list;
170 struct rb_root conns_all; /* connection tree */
171 rwlock_t conns_lock; /* protects conns_all */
172 unsigned int conns_num; /* current # of connections */
173 unsigned short vlan_id; /* vlan id of link group */
174
175 struct list_head sndbufs[SMC_RMBE_SIZES];/* tx buffers */
176 rwlock_t sndbufs_lock; /* protects tx buffers */
177 struct list_head rmbs[SMC_RMBE_SIZES]; /* rx buffers */
178 rwlock_t rmbs_lock; /* protects rx buffers */
179
180 u8 id[SMC_LGR_ID_SIZE]; /* unique lgr id */
181 struct delayed_work free_work; /* delayed freeing of an lgr */
182 u8 sync_err : 1; /* lgr no longer fits to peer */
183 u8 terminating : 1;/* lgr is terminating */
184
185 bool is_smcd; /* SMC-R or SMC-D */
186 union {
187 struct { /* SMC-R */
188 enum smc_lgr_role role;
189 /* client or server */
190 struct smc_link lnk[SMC_LINKS_PER_LGR_MAX];
191 /* smc link */
192 char peer_systemid[SMC_SYSTEMID_LEN];
193 /* unique system_id of peer */
194 struct smc_rtoken rtokens[SMC_RMBS_PER_LGR_MAX]
195 [SMC_LINKS_PER_LGR_MAX];
196 /* remote addr/key pairs */
197 DECLARE_BITMAP(rtokens_used_mask, SMC_RMBS_PER_LGR_MAX);
198 /* used rtoken elements */
199 };
200 struct { /* SMC-D */
201 u64 peer_gid;
202 /* Peer GID (remote) */
203 struct smcd_dev *smcd;
204 /* ISM device for VLAN reg. */
205 };
206 };
207};
208
209/* Find the connection associated with the given alert token in the link group.
210 * To use rbtrees we have to implement our own search core.
211 * Requires @conns_lock
212 * @token alert token to search for
213 * @lgr link group to search in
214 * Returns connection associated with token if found, NULL otherwise.
215 */
216static inline struct smc_connection *smc_lgr_find_conn(
217 u32 token, struct smc_link_group *lgr)
218{
219 struct smc_connection *res = NULL;
220 struct rb_node *node;
221
222 node = lgr->conns_all.rb_node;
223 while (node) {
224 struct smc_connection *cur = rb_entry(node,
225 struct smc_connection, alert_node);
226
227 if (cur->alert_token_local > token) {
228 node = node->rb_left;
229 } else {
230 if (cur->alert_token_local < token) {
231 node = node->rb_right;
232 } else {
233 res = cur;
234 break;
235 }
236 }
237 }
238
239 return res;
240}
241
242struct smc_sock;
243struct smc_clc_msg_accept_confirm;
244struct smc_clc_msg_local;
245
246void smc_lgr_free(struct smc_link_group *lgr);
247void smc_lgr_forget(struct smc_link_group *lgr);
248void smc_lgr_terminate(struct smc_link_group *lgr);
249void smc_port_terminate(struct smc_ib_device *smcibdev, u8 ibport);
250void smc_smcd_terminate(struct smcd_dev *dev, u64 peer_gid);
251int smc_buf_create(struct smc_sock *smc, bool is_smcd);
252int smc_uncompress_bufsize(u8 compressed);
253int smc_rmb_rtoken_handling(struct smc_connection *conn,
254 struct smc_clc_msg_accept_confirm *clc);
255int smc_rtoken_add(struct smc_link_group *lgr, __be64 nw_vaddr, __be32 nw_rkey);
256int smc_rtoken_delete(struct smc_link_group *lgr, __be32 nw_rkey);
257void smc_sndbuf_sync_sg_for_cpu(struct smc_connection *conn);
258void smc_sndbuf_sync_sg_for_device(struct smc_connection *conn);
259void smc_rmb_sync_sg_for_cpu(struct smc_connection *conn);
260void smc_rmb_sync_sg_for_device(struct smc_connection *conn);
261int smc_vlan_by_tcpsk(struct socket *clcsock, unsigned short *vlan_id);
262
263void smc_conn_free(struct smc_connection *conn);
264int smc_conn_create(struct smc_sock *smc, bool is_smcd, int srv_first_contact,
265 struct smc_ib_device *smcibdev, u8 ibport,
266 struct smc_clc_msg_local *lcl, struct smcd_dev *smcd,
267 u64 peer_gid);
268void smcd_conn_free(struct smc_connection *conn);
269void smc_lgr_schedule_free_work_fast(struct smc_link_group *lgr);
270void smc_core_exit(void);
271
272static inline struct smc_link_group *smc_get_lgr(struct smc_link *link)
273{
274 return container_of(link, struct smc_link_group, lnk[SMC_SINGLE_LINK]);
275}
276#endif