#ifndef _SDP_H_ #define _SDP_H_ #include "opt_ddb.h" #include "opt_inet.h" #include "opt_ofed.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef DDB #include #endif #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef SDP_DEBUG #define CONFIG_INFINIBAND_SDP_DEBUG #endif #include "sdp_dbg.h" #undef LIST_HEAD /* From sys/queue.h */ #define LIST_HEAD(name, type) \ struct name { \ struct type *lh_first; /* first element */ \ } /* Interval between sucessive polls in the Tx routine when polling is used instead of interrupts (in per-core Tx rings) - should be power of 2 */ #define SDP_TX_POLL_MODER 16 #define SDP_TX_POLL_TIMEOUT (HZ / 20) #define SDP_NAGLE_TIMEOUT (HZ / 10) #define SDP_SRCAVAIL_CANCEL_TIMEOUT (HZ * 5) #define SDP_SRCAVAIL_ADV_TIMEOUT (1 * HZ) #define SDP_SRCAVAIL_PAYLOAD_LEN 1 #define SDP_RESOLVE_TIMEOUT 1000 #define SDP_ROUTE_TIMEOUT 1000 #define SDP_RETRY_COUNT 5 #define SDP_KEEPALIVE_TIME (120 * 60 * HZ) #define SDP_FIN_WAIT_TIMEOUT (60 * HZ) /* like TCP_FIN_TIMEOUT */ #define SDP_TX_SIZE 0x40 #define SDP_RX_SIZE 0x40 #define SDP_FMR_SIZE (MIN(0x1000, PAGE_SIZE) / sizeof(u64)) #define SDP_FMR_POOL_SIZE 1024 #define SDP_FMR_DIRTY_SIZE ( SDP_FMR_POOL_SIZE / 4 ) #define SDP_MAX_RDMA_READ_LEN (PAGE_SIZE * (SDP_FMR_SIZE - 2)) /* mb inlined data len - rest will be rx'ed into frags */ #define SDP_HEAD_SIZE (sizeof(struct sdp_bsdh)) /* limit tx payload len, if the sink supports bigger buffers than the source * can handle. * or rx fragment size (limited by sge->length size) */ #define SDP_MAX_PACKET (1 << 16) #define SDP_MAX_PAYLOAD (SDP_MAX_PACKET - SDP_HEAD_SIZE) #define SDP_MAX_RECV_SGES (SDP_MAX_PACKET / MCLBYTES) #define SDP_MAX_SEND_SGES (SDP_MAX_PACKET / MCLBYTES) + 2 #define SDP_NUM_WC 4 #define SDP_DEF_ZCOPY_THRESH 64*1024 #define SDP_MIN_ZCOPY_THRESH PAGE_SIZE #define SDP_MAX_ZCOPY_THRESH 1048576 #define SDP_OP_RECV 0x800000000LL #define SDP_OP_SEND 0x400000000LL #define SDP_OP_RDMA 0x200000000LL #define SDP_OP_NOP 0x100000000LL /* how long (in jiffies) to block sender till tx completion*/ #define SDP_BZCOPY_POLL_TIMEOUT (HZ / 10) #define SDP_AUTO_CONF 0xffff #define AUTO_MOD_DELAY (HZ / 4) struct sdp_mb_cb { __u32 seq; /* Starting sequence number */ struct bzcopy_state *bz; struct rx_srcavail_state *rx_sa; struct tx_srcavail_state *tx_sa; }; #define M_PUSH M_PROTO1 /* Do a 'push'. */ #define M_URG M_PROTO2 /* Mark as urgent (oob). */ #define SDP_SKB_CB(__mb) ((struct sdp_mb_cb *)&((__mb)->cb[0])) #define BZCOPY_STATE(mb) (SDP_SKB_CB(mb)->bz) #define RX_SRCAVAIL_STATE(mb) (SDP_SKB_CB(mb)->rx_sa) #define TX_SRCAVAIL_STATE(mb) (SDP_SKB_CB(mb)->tx_sa) #ifndef MIN #define MIN(a, b) (a < b ? a : b) #endif #define ring_head(ring) (atomic_read(&(ring).head)) #define ring_tail(ring) (atomic_read(&(ring).tail)) #define ring_posted(ring) (ring_head(ring) - ring_tail(ring)) #define rx_ring_posted(ssk) ring_posted(ssk->rx_ring) #ifdef SDP_ZCOPY #define tx_ring_posted(ssk) (ring_posted(ssk->tx_ring) + \ (ssk->tx_ring.rdma_inflight ? ssk->tx_ring.rdma_inflight->busy : 0)) #else #define tx_ring_posted(ssk) ring_posted(ssk->tx_ring) #endif extern int sdp_zcopy_thresh; extern int rcvbuf_initial_size; extern struct workqueue_struct *rx_comp_wq; extern struct ib_client sdp_client; enum sdp_mid { SDP_MID_HELLO = 0x0, SDP_MID_HELLO_ACK = 0x1, SDP_MID_DISCONN = 0x2, SDP_MID_ABORT = 0x3, SDP_MID_SENDSM = 0x4, SDP_MID_RDMARDCOMPL = 0x6, SDP_MID_SRCAVAIL_CANCEL = 0x8, SDP_MID_CHRCVBUF = 0xB, SDP_MID_CHRCVBUF_ACK = 0xC, SDP_MID_SINKAVAIL = 0xFD, SDP_MID_SRCAVAIL = 0xFE, SDP_MID_DATA = 0xFF, }; enum sdp_flags { SDP_OOB_PRES = 1 << 0, SDP_OOB_PEND = 1 << 1, }; enum { SDP_MIN_TX_CREDITS = 2 }; enum { SDP_ERR_ERROR = -4, SDP_ERR_FAULT = -3, SDP_NEW_SEG = -2, SDP_DO_WAIT_MEM = -1 }; struct sdp_bsdh { u8 mid; u8 flags; __u16 bufs; __u32 len; __u32 mseq; __u32 mseq_ack; } __attribute__((__packed__)); union cma_ip_addr { struct in6_addr ip6; struct { __u32 pad[3]; __u32 addr; } ip4; } __attribute__((__packed__)); /* TODO: too much? Can I avoid having the src/dst and port here? */ struct sdp_hh { struct sdp_bsdh bsdh; u8 majv_minv; u8 ipv_cap; u8 rsvd1; u8 max_adverts; __u32 desremrcvsz; __u32 localrcvsz; __u16 port; __u16 rsvd2; union cma_ip_addr src_addr; union cma_ip_addr dst_addr; u8 rsvd3[IB_CM_REQ_PRIVATE_DATA_SIZE - sizeof(struct sdp_bsdh) - 48]; } __attribute__((__packed__)); struct sdp_hah { struct sdp_bsdh bsdh; u8 majv_minv; u8 ipv_cap; u8 rsvd1; u8 ext_max_adverts; __u32 actrcvsz; u8 rsvd2[IB_CM_REP_PRIVATE_DATA_SIZE - sizeof(struct sdp_bsdh) - 8]; } __attribute__((__packed__)); struct sdp_rrch { __u32 len; } __attribute__((__packed__)); struct sdp_srcah { __u32 len; __u32 rkey; __u64 vaddr; } __attribute__((__packed__)); struct sdp_buf { struct mbuf *mb; u64 mapping[SDP_MAX_SEND_SGES]; } __attribute__((__packed__)); struct sdp_chrecvbuf { u32 size; } __attribute__((__packed__)); /* Context used for synchronous zero copy bcopy (BZCOPY) */ struct bzcopy_state { unsigned char __user *u_base; int u_len; int left; int page_cnt; int cur_page; int cur_offset; int busy; struct sdp_sock *ssk; struct page **pages; }; enum rx_sa_flag { RX_SA_ABORTED = 2, }; enum tx_sa_flag { TX_SA_SENDSM = 0x01, TX_SA_CROSS_SEND = 0x02, TX_SA_INTRRUPTED = 0x04, TX_SA_TIMEDOUT = 0x08, TX_SA_ERROR = 0x10, }; struct rx_srcavail_state { /* Advertised buffer stuff */ u32 mseq; u32 used; u32 reported; u32 len; u32 rkey; u64 vaddr; /* Dest buff info */ struct ib_umem *umem; struct ib_pool_fmr *fmr; /* Utility */ u8 busy; enum rx_sa_flag flags; }; struct tx_srcavail_state { /* Data below 'busy' will be reset */ u8 busy; struct ib_umem *umem; struct ib_pool_fmr *fmr; u32 bytes_sent; u32 bytes_acked; enum tx_sa_flag abort_flags; u8 posted; u32 mseq; }; struct sdp_tx_ring { #ifdef SDP_ZCOPY struct rx_srcavail_state *rdma_inflight; #endif struct sdp_buf *buffer; atomic_t head; atomic_t tail; struct ib_cq *cq; atomic_t credits; #define tx_credits(ssk) (atomic_read(&ssk->tx_ring.credits)) struct callout timer; u16 poll_cnt; }; struct sdp_rx_ring { struct sdp_buf *buffer; atomic_t head; atomic_t tail; struct ib_cq *cq; int destroyed; struct rwlock destroyed_lock; }; struct sdp_device { struct ib_pd *pd; struct ib_mr *mr; struct ib_fmr_pool *fmr_pool; }; struct sdp_moderation { unsigned long last_moder_packets; unsigned long last_moder_tx_packets; unsigned long last_moder_bytes; unsigned long last_moder_jiffies; int last_moder_time; u16 rx_usecs; u16 rx_frames; u16 tx_usecs; u32 pkt_rate_low; u16 rx_usecs_low; u32 pkt_rate_high; u16 rx_usecs_high; u16 sample_interval; u16 adaptive_rx_coal; u32 msg_enable; int moder_cnt; int moder_time; }; /* These are flags fields. */ #define SDP_TIMEWAIT 0x0001 /* In ssk timewait state. */ #define SDP_DROPPED 0x0002 /* Socket has been dropped. */ #define SDP_SOCKREF 0x0004 /* Holding a sockref for close. */ #define SDP_NODELAY 0x0008 /* Disble nagle. */ #define SDP_NEEDFIN 0x0010 /* Send a fin on the next tx. */ #define SDP_DREQWAIT 0x0020 /* Waiting on DREQ. */ #define SDP_DESTROY 0x0040 /* Being destroyed. */ #define SDP_DISCON 0x0080 /* rdma_disconnect is owed. */ /* These are oobflags */ #define SDP_HADOOB 0x0001 /* Had OOB data. */ #define SDP_HAVEOOB 0x0002 /* Have OOB data. */ struct sdp_sock { LIST_ENTRY(sdp_sock) list; struct socket *socket; struct rdma_cm_id *id; struct ib_device *ib_device; struct sdp_device *sdp_dev; struct ib_qp *qp; struct ucred *cred; struct callout keep2msl; /* 2msl and keepalive timer. */ struct callout nagle_timer; /* timeout waiting for ack */ struct ib_ucontext context; in_port_t lport; in_addr_t laddr; in_port_t fport; in_addr_t faddr; int flags; int oobflags; /* protected by rx lock. */ int state; int softerror; int recv_bytes; /* Bytes per recv. buf including header */ int xmit_size_goal; char iobc; struct sdp_rx_ring rx_ring; struct sdp_tx_ring tx_ring; struct rwlock lock; struct mbuf *rx_ctl_q; struct mbuf *rx_ctl_tail; int qp_active; /* XXX Flag. */ int max_sge; struct work_struct rx_comp_work; #define rcv_nxt(ssk) atomic_read(&(ssk->rcv_nxt)) atomic_t rcv_nxt; /* SDP specific */ atomic_t mseq_ack; #define mseq_ack(ssk) (atomic_read(&ssk->mseq_ack)) unsigned max_bufs; /* Initial buffers offered by other side */ unsigned min_bufs; /* Low water mark to wake senders */ unsigned long nagle_last_unacked; /* mseq of lastest unacked packet */ atomic_t remote_credits; #define remote_credits(ssk) (atomic_read(&ssk->remote_credits)) int poll_cq; /* SDP slow start */ int recv_request_head; /* mark the rx_head when the resize request was recieved */ int recv_request; /* XXX flag if request to resize was recieved */ unsigned long tx_packets; unsigned long rx_packets; unsigned long tx_bytes; unsigned long rx_bytes; struct sdp_moderation auto_mod; struct task shutdown_task; #ifdef SDP_ZCOPY struct tx_srcavail_state *tx_sa; struct rx_srcavail_state *rx_sa; spinlock_t tx_sa_lock; struct delayed_work srcavail_cancel_work; int srcavail_cancel_mseq; /* ZCOPY data: -1:use global; 0:disable zcopy; >0: zcopy threshold */ int zcopy_thresh; #endif }; #define sdp_sk(so) ((struct sdp_sock *)(so->so_pcb)) #define SDP_RLOCK(ssk) rw_rlock(&(ssk)->lock) #define SDP_WLOCK(ssk) rw_wlock(&(ssk)->lock) #define SDP_RUNLOCK(ssk) rw_runlock(&(ssk)->lock) #define SDP_WUNLOCK(ssk) rw_wunlock(&(ssk)->lock) #define SDP_WLOCK_ASSERT(ssk) rw_assert(&(ssk)->lock, RA_WLOCKED) #define SDP_RLOCK_ASSERT(ssk) rw_assert(&(ssk)->lock, RA_RLOCKED) #define SDP_LOCK_ASSERT(ssk) rw_assert(&(ssk)->lock, RA_LOCKED) static inline void tx_sa_reset(struct tx_srcavail_state *tx_sa) { memset((void *)&tx_sa->busy, 0, sizeof(*tx_sa) - offsetof(typeof(*tx_sa), busy)); } static inline void rx_ring_unlock(struct sdp_rx_ring *rx_ring) { rw_runlock(&rx_ring->destroyed_lock); } static inline int rx_ring_trylock(struct sdp_rx_ring *rx_ring) { rw_rlock(&rx_ring->destroyed_lock); if (rx_ring->destroyed) { rx_ring_unlock(rx_ring); return 0; } return 1; } static inline void rx_ring_destroy_lock(struct sdp_rx_ring *rx_ring) { rw_wlock(&rx_ring->destroyed_lock); rx_ring->destroyed = 1; rw_wunlock(&rx_ring->destroyed_lock); } static inline void sdp_arm_rx_cq(struct sdp_sock *ssk) { sdp_prf(ssk->socket, NULL, "Arming RX cq"); sdp_dbg_data(ssk->socket, "Arming RX cq\n"); ib_req_notify_cq(ssk->rx_ring.cq, IB_CQ_NEXT_COMP); } static inline void sdp_arm_tx_cq(struct sdp_sock *ssk) { sdp_prf(ssk->socket, NULL, "Arming TX cq"); sdp_dbg_data(ssk->socket, "Arming TX cq. credits: %d, posted: %d\n", tx_credits(ssk), tx_ring_posted(ssk)); ib_req_notify_cq(ssk->tx_ring.cq, IB_CQ_NEXT_COMP); } /* return the min of: * - tx credits * - free slots in tx_ring (not including SDP_MIN_TX_CREDITS */ static inline int tx_slots_free(struct sdp_sock *ssk) { int min_free; min_free = MIN(tx_credits(ssk), SDP_TX_SIZE - tx_ring_posted(ssk)); if (min_free < SDP_MIN_TX_CREDITS) return 0; return min_free - SDP_MIN_TX_CREDITS; }; /* utilities */ static inline char *mid2str(int mid) { #define ENUM2STR(e) [e] = #e static char *mid2str[] = { ENUM2STR(SDP_MID_HELLO), ENUM2STR(SDP_MID_HELLO_ACK), ENUM2STR(SDP_MID_ABORT), ENUM2STR(SDP_MID_DISCONN), ENUM2STR(SDP_MID_SENDSM), ENUM2STR(SDP_MID_RDMARDCOMPL), ENUM2STR(SDP_MID_SRCAVAIL_CANCEL), ENUM2STR(SDP_MID_CHRCVBUF), ENUM2STR(SDP_MID_CHRCVBUF_ACK), ENUM2STR(SDP_MID_DATA), ENUM2STR(SDP_MID_SRCAVAIL), ENUM2STR(SDP_MID_SINKAVAIL), }; if (mid >= ARRAY_SIZE(mid2str)) return NULL; return mid2str[mid]; } static inline struct mbuf * sdp_alloc_mb(struct socket *sk, u8 mid, int size, int wait) { struct sdp_bsdh *h; struct mbuf *mb; MGETHDR(mb, wait, MT_DATA); if (mb == NULL) return (NULL); mb->m_pkthdr.len = mb->m_len = sizeof(struct sdp_bsdh); h = mtod(mb, struct sdp_bsdh *); h->mid = mid; return mb; } static inline struct mbuf * sdp_alloc_mb_data(struct socket *sk, int wait) { return sdp_alloc_mb(sk, SDP_MID_DATA, 0, wait); } static inline struct mbuf * sdp_alloc_mb_disconnect(struct socket *sk, int wait) { return sdp_alloc_mb(sk, SDP_MID_DISCONN, 0, wait); } static inline void * mb_put(struct mbuf *mb, int len) { uint8_t *data; data = mb->m_data; data += mb->m_len; mb->m_len += len; return (void *)data; } static inline struct mbuf * sdp_alloc_mb_chrcvbuf_ack(struct socket *sk, int size, int wait) { struct mbuf *mb; struct sdp_chrecvbuf *resp_size; mb = sdp_alloc_mb(sk, SDP_MID_CHRCVBUF_ACK, sizeof(*resp_size), wait); if (mb == NULL) return (NULL); resp_size = (struct sdp_chrecvbuf *)mb_put(mb, sizeof *resp_size); resp_size->size = htonl(size); return mb; } static inline struct mbuf * sdp_alloc_mb_srcavail(struct socket *sk, u32 len, u32 rkey, u64 vaddr, int wait) { struct mbuf *mb; struct sdp_srcah *srcah; mb = sdp_alloc_mb(sk, SDP_MID_SRCAVAIL, sizeof(*srcah), wait); if (mb == NULL) return (NULL); srcah = (struct sdp_srcah *)mb_put(mb, sizeof(*srcah)); srcah->len = htonl(len); srcah->rkey = htonl(rkey); srcah->vaddr = cpu_to_be64(vaddr); return mb; } static inline struct mbuf * sdp_alloc_mb_srcavail_cancel(struct socket *sk, int wait) { return sdp_alloc_mb(sk, SDP_MID_SRCAVAIL_CANCEL, 0, wait); } static inline struct mbuf * sdp_alloc_mb_rdmardcompl(struct socket *sk, u32 len, int wait) { struct mbuf *mb; struct sdp_rrch *rrch; mb = sdp_alloc_mb(sk, SDP_MID_RDMARDCOMPL, sizeof(*rrch), wait); if (mb == NULL) return (NULL); rrch = (struct sdp_rrch *)mb_put(mb, sizeof(*rrch)); rrch->len = htonl(len); return mb; } static inline struct mbuf * sdp_alloc_mb_sendsm(struct socket *sk, int wait) { return sdp_alloc_mb(sk, SDP_MID_SENDSM, 0, wait); } static inline int sdp_tx_ring_slots_left(struct sdp_sock *ssk) { return SDP_TX_SIZE - tx_ring_posted(ssk); } static inline int credit_update_needed(struct sdp_sock *ssk) { int c; c = remote_credits(ssk); if (likely(c > SDP_MIN_TX_CREDITS)) c += c/2; return unlikely(c < rx_ring_posted(ssk)) && likely(tx_credits(ssk) > 0) && likely(sdp_tx_ring_slots_left(ssk)); } #define SDPSTATS_COUNTER_INC(stat) #define SDPSTATS_COUNTER_ADD(stat, val) #define SDPSTATS_COUNTER_MID_INC(stat, mid) #define SDPSTATS_HIST_LINEAR(stat, size) #define SDPSTATS_HIST(stat, size) static inline void sdp_cleanup_sdp_buf(struct sdp_sock *ssk, struct sdp_buf *sbuf, enum dma_data_direction dir) { struct ib_device *dev; struct mbuf *mb; int i; dev = ssk->ib_device; for (i = 0, mb = sbuf->mb; mb != NULL; mb = mb->m_next, i++) ib_dma_unmap_single(dev, sbuf->mapping[i], mb->m_len, dir); } /* sdp_main.c */ void sdp_set_default_moderation(struct sdp_sock *ssk); void sdp_start_keepalive_timer(struct socket *sk); void sdp_urg(struct sdp_sock *ssk, struct mbuf *mb); void sdp_cancel_dreq_wait_timeout(struct sdp_sock *ssk); void sdp_abort(struct socket *sk); struct sdp_sock *sdp_notify(struct sdp_sock *ssk, int error); /* sdp_cma.c */ int sdp_cma_handler(struct rdma_cm_id *, struct rdma_cm_event *); /* sdp_tx.c */ int sdp_tx_ring_create(struct sdp_sock *ssk, struct ib_device *device); void sdp_tx_ring_destroy(struct sdp_sock *ssk); int sdp_xmit_poll(struct sdp_sock *ssk, int force); void sdp_post_send(struct sdp_sock *ssk, struct mbuf *mb); void sdp_post_sends(struct sdp_sock *ssk, int wait); void sdp_post_keepalive(struct sdp_sock *ssk); /* sdp_rx.c */ void sdp_rx_ring_init(struct sdp_sock *ssk); int sdp_rx_ring_create(struct sdp_sock *ssk, struct ib_device *device); void sdp_rx_ring_destroy(struct sdp_sock *ssk); int sdp_resize_buffers(struct sdp_sock *ssk, u32 new_size); int sdp_init_buffers(struct sdp_sock *ssk, u32 new_size); void sdp_do_posts(struct sdp_sock *ssk); void sdp_rx_comp_full(struct sdp_sock *ssk); /* sdp_zcopy.c */ int sdp_sendmsg_zcopy(struct kiocb *iocb, struct socket *sk, struct iovec *iov); int sdp_handle_srcavail(struct sdp_sock *ssk, struct sdp_srcah *srcah); void sdp_handle_sendsm(struct sdp_sock *ssk, u32 mseq_ack); void sdp_handle_rdma_read_compl(struct sdp_sock *ssk, u32 mseq_ack, u32 bytes_completed); int sdp_handle_rdma_read_cqe(struct sdp_sock *ssk); int sdp_rdma_to_iovec(struct socket *sk, struct iovec *iov, struct mbuf *mb, unsigned long *used); int sdp_post_rdma_rd_compl(struct sdp_sock *ssk, struct rx_srcavail_state *rx_sa); int sdp_post_sendsm(struct socket *sk); void srcavail_cancel_timeout(struct work_struct *work); void sdp_abort_srcavail(struct socket *sk); void sdp_abort_rdma_read(struct socket *sk); int sdp_process_rx(struct sdp_sock *ssk); #endif