net/smc/smc_core.h

   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         SMC_LNK_DELETING,       /* link is being deleted */
  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_rdma_sge {                           /* sges for RDMA writes */
  56         struct ib_sge           wr_tx_rdma_sge[SMC_IB_MAX_SEND_SGE];
  57 };
  58
  59 #define SMC_MAX_RDMA_WRITES     2               /* max. # of RDMA writes per
  60                                                  * message send
  61                                                  */
  62
  63 struct smc_rdma_sges {                          /* sges per message send */
  64         struct smc_rdma_sge     tx_rdma_sge[SMC_MAX_RDMA_WRITES];
  65 };
  66
  67 struct smc_rdma_wr {                            /* work requests per message
  68                                                  * send
  69                                                  */
  70         struct ib_rdma_wr       wr_tx_rdma[SMC_MAX_RDMA_WRITES];
  71 };
  72
  73 struct smc_link {
  74         struct smc_ib_device    *smcibdev;      /* ib-device */
  75         u8                      ibport;         /* port - values 1 | 2 */
  76         struct ib_pd            *roce_pd;       /* IB protection domain,
  77                                                  * unique for every RoCE QP
  78                                                  */
  79         struct ib_qp            *roce_qp;       /* IB queue pair */
  80         struct ib_qp_attr       qp_attr;        /* IB queue pair attributes */
  81
  82         struct smc_wr_buf       *wr_tx_bufs;    /* WR send payload buffers */
  83         struct ib_send_wr       *wr_tx_ibs;     /* WR send meta data */
  84         struct ib_sge           *wr_tx_sges;    /* WR send gather meta data */
  85         struct smc_rdma_sges    *wr_tx_rdma_sges;/*RDMA WRITE gather meta data*/
  86         struct smc_rdma_wr      *wr_tx_rdmas;   /* WR RDMA WRITE */
  87         struct smc_wr_tx_pend   *wr_tx_pends;   /* WR send waiting for CQE */
  88         /* above four vectors have wr_tx_cnt elements and use the same index */
  89         dma_addr_t              wr_tx_dma_addr; /* DMA address of wr_tx_bufs */
  90         atomic_long_t           wr_tx_id;       /* seq # of last sent WR */
  91         unsigned long           *wr_tx_mask;    /* bit mask of used indexes */
  92         u32                     wr_tx_cnt;      /* number of WR send buffers */
  93         wait_queue_head_t       wr_tx_wait;     /* wait for free WR send buf */
  94
  95         struct smc_wr_buf       *wr_rx_bufs;    /* WR recv payload buffers */
  96         struct ib_recv_wr       *wr_rx_ibs;     /* WR recv meta data */
  97         struct ib_sge           *wr_rx_sges;    /* WR recv scatter meta data */
  98         /* above three vectors have wr_rx_cnt elements and use the same index */
  99         dma_addr_t              wr_rx_dma_addr; /* DMA address of wr_rx_bufs */
 100         u64                     wr_rx_id;       /* seq # of last recv WR */
 101         u32                     wr_rx_cnt;      /* number of WR recv buffers */
 102         unsigned long           wr_rx_tstamp;   /* jiffies when last buf rx */
 103
 104         struct ib_reg_wr        wr_reg;         /* WR register memory region */
 105         wait_queue_head_t       wr_reg_wait;    /* wait for wr_reg result */
 106         enum smc_wr_reg_state   wr_reg_state;   /* state of wr_reg request */
 107
 108         u8                      gid[SMC_GID_SIZE];/* gid matching used vlan id*/
 109         u8                      sgid_index;     /* gid index for vlan id      */
 110         u32                     peer_qpn;       /* QP number of peer */
 111         enum ib_mtu             path_mtu;       /* used mtu */
 112         enum ib_mtu             peer_mtu;       /* mtu size of peer */
 113         u32                     psn_initial;    /* QP tx initial packet seqno */
 114         u32                     peer_psn;       /* QP rx initial packet seqno */
 115         u8                      peer_mac[ETH_ALEN];     /* = gid[8:10||13:15] */
 116         u8                      peer_gid[SMC_GID_SIZE]; /* gid of peer*/
 117         u8                      link_id;        /* unique # within link group */
 118
 119         enum smc_link_state     state;          /* state of link */
 120         struct workqueue_struct *llc_wq;        /* single thread work queue */
 121         struct completion       llc_confirm;    /* wait for rx of conf link */
 122         struct completion       llc_confirm_resp; /* wait 4 rx of cnf lnk rsp */
 123         int                     llc_confirm_rc; /* rc from confirm link msg */
 124         int                     llc_confirm_resp_rc; /* rc from conf_resp msg */
 125         struct completion       llc_add;        /* wait for rx of add link */
 126         struct completion       llc_add_resp;   /* wait for rx of add link rsp*/
 127         struct delayed_work     llc_testlink_wrk; /* testlink worker */
 128         struct completion       llc_testlink_resp; /* wait for rx of testlink */
 129         int                     llc_testlink_time; /* testlink interval */
 130         struct completion       llc_confirm_rkey; /* wait 4 rx of cnf rkey */
 131         int                     llc_confirm_rkey_rc; /* rc from cnf rkey msg */
 132         struct completion       llc_delete_rkey; /* wait 4 rx of del rkey */
 133         int                     llc_delete_rkey_rc; /* rc from del rkey msg */
 134         struct mutex            llc_delete_rkey_mutex; /* serialize usage */
 135 };
 136
 137 /* For now we just allow one parallel link per link group. The SMC protocol
 138  * allows more (up to 8).
 139  */
 140 #define SMC_LINKS_PER_LGR_MAX   1
 141 #define SMC_SINGLE_LINK         0
 142
 143 #define SMC_FIRST_CONTACT       1               /* first contact to a peer */
 144 #define SMC_REUSE_CONTACT       0               /* follow-on contact to a peer*/
 145
 146 /* tx/rx buffer list element for sndbufs list and rmbs list of a lgr */
 147 struct smc_buf_desc {
 148         struct list_head        list;
 149         void                    *cpu_addr;      /* virtual address of buffer */
 150         struct page             *pages;
 151         int                     len;            /* length of buffer */
 152         u32                     used;           /* currently used / unused */
 153         u8                      wr_reg  : 1;    /* mem region registered */
 154         u8                      regerr  : 1;    /* err during registration */
 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                 struct { /* SMC-D */
 166                         unsigned short          sba_idx;
 167                                                 /* SBA index number */
 168                         u64                     token;
 169                                                 /* DMB token number */
 170                         dma_addr_t              dma_addr;
 171                                                 /* DMA address */
 172                 };
 173         };
 174 };
 175
 176 struct smc_rtoken {                             /* address/key of remote RMB */
 177         u64                     dma_addr;
 178         u32                     rkey;
 179 };
 180
 181 #define SMC_LGR_ID_SIZE         4
 182 #define SMC_BUF_MIN_SIZE        16384   /* minimum size of an RMB */
 183 #define SMC_RMBE_SIZES          16      /* number of distinct RMBE sizes */
 184 /* theoretically, the RFC states that largest size would be 512K,
 185  * i.e. compressed 5 and thus 6 sizes (0..5), despite
 186  * struct smc_clc_msg_accept_confirm.rmbe_size being a 4 bit value (0..15)
 187  */
 188
 189 struct smcd_dev;
 190
 191 struct smc_link_group {
 192         struct list_head        list;
 193         struct rb_root          conns_all;      /* connection tree */
 194         rwlock_t                conns_lock;     /* protects conns_all */
 195         unsigned int            conns_num;      /* current # of connections */
 196         unsigned short          vlan_id;        /* vlan id of link group */
 197
 198         struct list_head        sndbufs[SMC_RMBE_SIZES];/* tx buffers */
 199         rwlock_t                sndbufs_lock;   /* protects tx buffers */
 200         struct list_head        rmbs[SMC_RMBE_SIZES];   /* rx buffers */
 201         rwlock_t                rmbs_lock;      /* protects rx buffers */
 202
 203         u8                      id[SMC_LGR_ID_SIZE];    /* unique lgr id */
 204         struct delayed_work     free_work;      /* delayed freeing of an lgr */
 205         struct work_struct      terminate_work; /* abnormal lgr termination */
 206         u8                      sync_err : 1;   /* lgr no longer fits to peer */
 207         u8                      terminating : 1;/* lgr is terminating */
 208         u8                      freefast : 1;   /* free worker scheduled fast */
 209         u8                      freeing : 1;    /* lgr is being freed */
 210
 211         bool                    is_smcd;        /* SMC-R or SMC-D */
 212         union {
 213                 struct { /* SMC-R */
 214                         enum smc_lgr_role       role;
 215                                                 /* client or server */
 216                         struct smc_link         lnk[SMC_LINKS_PER_LGR_MAX];
 217                                                 /* smc link */
 218                         char                    peer_systemid[SMC_SYSTEMID_LEN];
 219                                                 /* unique system_id of peer */
 220                         struct smc_rtoken       rtokens[SMC_RMBS_PER_LGR_MAX]
 221                                                 [SMC_LINKS_PER_LGR_MAX];
 222                                                 /* remote addr/key pairs */
 223                         DECLARE_BITMAP(rtokens_used_mask, SMC_RMBS_PER_LGR_MAX);
 224                                                 /* used rtoken elements */
 225                 };
 226                 struct { /* SMC-D */
 227                         u64                     peer_gid;
 228                                                 /* Peer GID (remote) */
 229                         struct smcd_dev         *smcd;
 230                                                 /* ISM device for VLAN reg. */
 231                         u8                      peer_shutdown : 1;
 232                                                 /* peer triggered shutdownn */
 233                 };
 234         };
 235 };
 236
 237 struct smc_clc_msg_local;
 238
 239 struct smc_init_info {
 240         u8                      is_smcd;
 241         unsigned short          vlan_id;
 242         int                     srv_first_contact;
 243         int                     cln_first_contact;
 244         /* SMC-R */
 245         struct smc_clc_msg_local *ib_lcl;
 246         struct smc_ib_device    *ib_dev;
 247         u8                      ib_gid[SMC_GID_SIZE];
 248         u8                      ib_port;
 249         u32                     ib_clcqpn;
 250         /* SMC-D */
 251         u64                     ism_gid;
 252         struct smcd_dev         *ism_dev;
 253 };
 254
 255 /* Find the connection associated with the given alert token in the link group.
 256  * To use rbtrees we have to implement our own search core.
 257  * Requires @conns_lock
 258  * @token       alert token to search for
 259  * @lgr          link group to search in
 260  * Returns connection associated with token if found, NULL otherwise.
 261  */
 262 static inline struct smc_connection *smc_lgr_find_conn(
 263         u32 token, struct smc_link_group *lgr)
 264 {
 265         struct smc_connection *res = NULL;
 266         struct rb_node *node;
 267
 268         node = lgr->conns_all.rb_node;
 269         while (node) {
 270                 struct smc_connection *cur = rb_entry(node,
 271                                         struct smc_connection, alert_node);
 272
 273                 if (cur->alert_token_local > token) {
 274                         node = node->rb_left;
 275                 } else {
 276                         if (cur->alert_token_local < token) {
 277                                 node = node->rb_right;
 278                         } else {
 279                                 res = cur;
 280                                 break;
 281                         }
 282                 }
 283         }
 284
 285         return res;
 286 }
 287
 288 struct smc_sock;
 289 struct smc_clc_msg_accept_confirm;
 290 struct smc_clc_msg_local;
 291
 292 void smc_lgr_forget(struct smc_link_group *lgr);
 293 void smc_lgr_cleanup_early(struct smc_connection *conn);
 294 void smc_lgr_terminate_sched(struct smc_link_group *lgr);
 295 void smc_port_terminate(struct smc_ib_device *smcibdev, u8 ibport);
 296 void smc_smcd_terminate(struct smcd_dev *dev, u64 peer_gid,
 297                         unsigned short vlan);
 298 void smc_smcd_terminate_all(struct smcd_dev *dev);
 299 void smc_smcr_terminate_all(struct smc_ib_device *smcibdev);
 300 int smc_buf_create(struct smc_sock *smc, bool is_smcd);
 301 int smc_uncompress_bufsize(u8 compressed);
 302 int smc_rmb_rtoken_handling(struct smc_connection *conn,
 303                             struct smc_clc_msg_accept_confirm *clc);
 304 int smc_rtoken_add(struct smc_link_group *lgr, __be64 nw_vaddr, __be32 nw_rkey);
 305 int smc_rtoken_delete(struct smc_link_group *lgr, __be32 nw_rkey);
 306 void smc_sndbuf_sync_sg_for_cpu(struct smc_connection *conn);
 307 void smc_sndbuf_sync_sg_for_device(struct smc_connection *conn);
 308 void smc_rmb_sync_sg_for_cpu(struct smc_connection *conn);
 309 void smc_rmb_sync_sg_for_device(struct smc_connection *conn);
 310 int smc_vlan_by_tcpsk(struct socket *clcsock, struct smc_init_info *ini);
 311
 312 void smc_conn_free(struct smc_connection *conn);
 313 int smc_conn_create(struct smc_sock *smc, struct smc_init_info *ini);
 314 void smc_lgr_schedule_free_work_fast(struct smc_link_group *lgr);
 315 int smc_core_init(void);
 316 void smc_core_exit(void);
 317
 318 static inline struct smc_link_group *smc_get_lgr(struct smc_link *link)
 319 {
 320         return container_of(link, struct smc_link_group, lnk[SMC_SINGLE_LINK]);
 321 }
 322 #endif