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