xref: /openbmc/linux/net/smc/smc_core.h (revision addee42a)
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 
23 struct smc_lgr_list {			/* list of link group definition */
24 	struct list_head	list;
25 	spinlock_t		lock;	/* protects list of link groups */
26 };
27 
28 extern struct smc_lgr_list	smc_lgr_list; /* list of link groups */
29 
30 enum smc_lgr_role {		/* possible roles of a link group */
31 	SMC_CLNT,	/* client */
32 	SMC_SERV	/* server */
33 };
34 
35 enum smc_link_state {			/* possible states of a link */
36 	SMC_LNK_INACTIVE,	/* link is inactive */
37 	SMC_LNK_ACTIVATING,	/* link is being activated */
38 	SMC_LNK_ACTIVE		/* link is active */
39 };
40 
41 #define SMC_WR_BUF_SIZE		48	/* size of work request buffer */
42 
43 struct 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 
49 enum 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 
55 struct 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 
83 	struct ib_reg_wr	wr_reg;		/* WR register memory region */
84 	wait_queue_head_t	wr_reg_wait;	/* wait for wr_reg result */
85 	enum smc_wr_reg_state	wr_reg_state;	/* state of wr_reg request */
86 
87 	union ib_gid		gid;		/* gid matching used vlan id */
88 	u32			peer_qpn;	/* QP number of peer */
89 	enum ib_mtu		path_mtu;	/* used mtu */
90 	enum ib_mtu		peer_mtu;	/* mtu size of peer */
91 	u32			psn_initial;	/* QP tx initial packet seqno */
92 	u32			peer_psn;	/* QP rx initial packet seqno */
93 	u8			peer_mac[ETH_ALEN];	/* = gid[8:10||13:15] */
94 	u8			peer_gid[sizeof(union ib_gid)];	/* gid of peer*/
95 	u8			link_id;	/* unique # within link group */
96 
97 	enum smc_link_state	state;		/* state of link */
98 	struct completion	llc_confirm;	/* wait for rx of conf link */
99 	struct completion	llc_confirm_resp; /* wait 4 rx of cnf lnk rsp */
100 	int			llc_confirm_rc; /* rc from confirm link msg */
101 	int			llc_confirm_resp_rc; /* rc from conf_resp msg */
102 	struct completion	llc_add;	/* wait for rx of add link */
103 	struct completion	llc_add_resp;	/* wait for rx of add link rsp*/
104 };
105 
106 /* For now we just allow one parallel link per link group. The SMC protocol
107  * allows more (up to 8).
108  */
109 #define SMC_LINKS_PER_LGR_MAX	1
110 #define SMC_SINGLE_LINK		0
111 
112 #define SMC_FIRST_CONTACT	1		/* first contact to a peer */
113 #define SMC_REUSE_CONTACT	0		/* follow-on contact to a peer*/
114 
115 /* tx/rx buffer list element for sndbufs list and rmbs list of a lgr */
116 struct smc_buf_desc {
117 	struct list_head	list;
118 	void			*cpu_addr;	/* virtual address of buffer */
119 	struct sg_table		sgt[SMC_LINKS_PER_LGR_MAX];/* virtual buffer */
120 	struct ib_mr		*mr_rx[SMC_LINKS_PER_LGR_MAX];
121 						/* for rmb only: memory region
122 						 * incl. rkey provided to peer
123 						 */
124 	u32			order;		/* allocation order */
125 	u32			used;		/* currently used / unused */
126 	u8			reused	: 1;	/* new created / reused */
127 	u8			regerr	: 1;	/* err during registration */
128 };
129 
130 struct smc_rtoken {				/* address/key of remote RMB */
131 	u64			dma_addr;
132 	u32			rkey;
133 };
134 
135 #define SMC_LGR_ID_SIZE		4
136 
137 struct smc_link_group {
138 	struct list_head	list;
139 	enum smc_lgr_role	role;		/* client or server */
140 	struct smc_link		lnk[SMC_LINKS_PER_LGR_MAX];	/* smc link */
141 	char			peer_systemid[SMC_SYSTEMID_LEN];
142 						/* unique system_id of peer */
143 	struct rb_root		conns_all;	/* connection tree */
144 	rwlock_t		conns_lock;	/* protects conns_all */
145 	unsigned int		conns_num;	/* current # of connections */
146 	unsigned short		vlan_id;	/* vlan id of link group */
147 
148 	struct list_head	sndbufs[SMC_RMBE_SIZES];/* tx buffers */
149 	rwlock_t		sndbufs_lock;	/* protects tx buffers */
150 	struct list_head	rmbs[SMC_RMBE_SIZES];	/* rx buffers */
151 	rwlock_t		rmbs_lock;	/* protects rx buffers */
152 	struct smc_rtoken	rtokens[SMC_RMBS_PER_LGR_MAX]
153 				       [SMC_LINKS_PER_LGR_MAX];
154 						/* remote addr/key pairs */
155 	unsigned long		rtokens_used_mask[BITS_TO_LONGS(
156 							SMC_RMBS_PER_LGR_MAX)];
157 						/* used rtoken elements */
158 
159 	u8			id[SMC_LGR_ID_SIZE];	/* unique lgr id */
160 	struct delayed_work	free_work;	/* delayed freeing of an lgr */
161 	bool			sync_err;	/* lgr no longer fits to peer */
162 };
163 
164 /* Find the connection associated with the given alert token in the link group.
165  * To use rbtrees we have to implement our own search core.
166  * Requires @conns_lock
167  * @token	alert token to search for
168  * @lgr		 link group to search in
169  * Returns connection associated with token if found, NULL otherwise.
170  */
171 static inline struct smc_connection *smc_lgr_find_conn(
172 	u32 token, struct smc_link_group *lgr)
173 {
174 	struct smc_connection *res = NULL;
175 	struct rb_node *node;
176 
177 	node = lgr->conns_all.rb_node;
178 	while (node) {
179 		struct smc_connection *cur = rb_entry(node,
180 					struct smc_connection, alert_node);
181 
182 		if (cur->alert_token_local > token) {
183 			node = node->rb_left;
184 		} else {
185 			if (cur->alert_token_local < token) {
186 				node = node->rb_right;
187 			} else {
188 				res = cur;
189 				break;
190 			}
191 		}
192 	}
193 
194 	return res;
195 }
196 
197 struct smc_sock;
198 struct smc_clc_msg_accept_confirm;
199 
200 void smc_lgr_free(struct smc_link_group *lgr);
201 void smc_lgr_forget(struct smc_link_group *lgr);
202 void smc_lgr_terminate(struct smc_link_group *lgr);
203 int smc_buf_create(struct smc_sock *smc);
204 int smc_rmb_rtoken_handling(struct smc_connection *conn,
205 			    struct smc_clc_msg_accept_confirm *clc);
206 int smc_rtoken_add(struct smc_link_group *lgr, __be64 nw_vaddr, __be32 nw_rkey);
207 int smc_rtoken_delete(struct smc_link_group *lgr, __be32 nw_rkey);
208 void smc_sndbuf_sync_sg_for_cpu(struct smc_connection *conn);
209 void smc_sndbuf_sync_sg_for_device(struct smc_connection *conn);
210 void smc_rmb_sync_sg_for_cpu(struct smc_connection *conn);
211 void smc_rmb_sync_sg_for_device(struct smc_connection *conn);
212 #endif
213