[linux-block.git] / net / smc / smc_cdc.c

// SPDX-License-Identifier: GPL-2.0
/*
 * Shared Memory Communications over RDMA (SMC-R) and RoCE
 *
 * Connection Data Control (CDC)
 * handles flow control
 *
 * Copyright IBM Corp. 2016
 *
 * Author(s):  Ursula Braun <ubraun@linux.vnet.ibm.com>
 */

#include <linux/spinlock.h>

#include "smc.h"
#include "smc_wr.h"
#include "smc_cdc.h"
#include "smc_tx.h"
#include "smc_rx.h"
#include "smc_close.h"

/********************************** send *************************************/

/* handler for send/transmission completion of a CDC msg */
static void smc_cdc_tx_handler(struct smc_wr_tx_pend_priv *pnd_snd,
			       struct smc_link *link,
			       enum ib_wc_status wc_status)
{
	struct smc_cdc_tx_pend *cdcpend = (struct smc_cdc_tx_pend *)pnd_snd;
	struct smc_connection *conn = cdcpend->conn;
	struct smc_sock *smc;
	int diff;

	smc = container_of(conn, struct smc_sock, conn);
	bh_lock_sock(&smc->sk);
	if (!wc_status) {
		diff = smc_curs_diff(cdcpend->conn->sndbuf_desc->len,
				     &cdcpend->conn->tx_curs_fin,
				     &cdcpend->cursor);
		/* sndbuf_space is decreased in smc_sendmsg */
		smp_mb__before_atomic();
		atomic_add(diff, &cdcpend->conn->sndbuf_space);
		/* guarantee 0 <= sndbuf_space <= sndbuf_desc->len */
		smp_mb__after_atomic();
		smc_curs_copy(&conn->tx_curs_fin, &cdcpend->cursor, conn);
		smc_curs_copy(&conn->local_tx_ctrl_fin, &cdcpend->p_cursor,
			      conn);
		conn->tx_cdc_seq_fin = cdcpend->ctrl_seq;
	}

	if (atomic_dec_and_test(&conn->cdc_pend_tx_wr)) {
		/* If user owns the sock_lock, mark the connection need sending.
		 * User context will later try to send when it release sock_lock
		 * in smc_release_cb()
		 */
		if (sock_owned_by_user(&smc->sk))
			conn->tx_in_release_sock = true;
		else
			smc_tx_pending(conn);

		if (unlikely(wq_has_sleeper(&conn->cdc_pend_tx_wq)))
			wake_up(&conn->cdc_pend_tx_wq);
	}
	WARN_ON(atomic_read(&conn->cdc_pend_tx_wr) < 0);

	smc_tx_sndbuf_nonfull(smc);
	bh_unlock_sock(&smc->sk);
}

int smc_cdc_get_free_slot(struct smc_connection *conn,
			  struct smc_link *link,
			  struct smc_wr_buf **wr_buf,
			  struct smc_rdma_wr **wr_rdma_buf,
			  struct smc_cdc_tx_pend **pend)
{
	int rc;

	rc = smc_wr_tx_get_free_slot(link, smc_cdc_tx_handler, wr_buf,
				     wr_rdma_buf,
				     (struct smc_wr_tx_pend_priv **)pend);
	if (conn->killed) {
		/* abnormal termination */
		if (!rc)
			smc_wr_tx_put_slot(link,
					   (struct smc_wr_tx_pend_priv *)(*pend));
		rc = -EPIPE;
	}
	return rc;
}

static inline void smc_cdc_add_pending_send(struct smc_connection *conn,
					    struct smc_cdc_tx_pend *pend)
{
	BUILD_BUG_ON_MSG(
		sizeof(struct smc_cdc_msg) > SMC_WR_BUF_SIZE,
		"must increase SMC_WR_BUF_SIZE to at least sizeof(struct smc_cdc_msg)");
	BUILD_BUG_ON_MSG(
		offsetofend(struct smc_cdc_msg, reserved) > SMC_WR_TX_SIZE,
		"must adapt SMC_WR_TX_SIZE to sizeof(struct smc_cdc_msg); if not all smc_wr upper layer protocols use the same message size any more, must start to set link->wr_tx_sges[i].length on each individual smc_wr_tx_send()");
	BUILD_BUG_ON_MSG(
		sizeof(struct smc_cdc_tx_pend) > SMC_WR_TX_PEND_PRIV_SIZE,
		"must increase SMC_WR_TX_PEND_PRIV_SIZE to at least sizeof(struct smc_cdc_tx_pend)");
	pend->conn = conn;
	pend->cursor = conn->tx_curs_sent;
	pend->p_cursor = conn->local_tx_ctrl.prod;
	pend->ctrl_seq = conn->tx_cdc_seq;
}

int smc_cdc_msg_send(struct smc_connection *conn,
		     struct smc_wr_buf *wr_buf,
		     struct smc_cdc_tx_pend *pend)
{
	struct smc_link *link = conn->lnk;
	union smc_host_cursor cfed;
	int rc;

	if (unlikely(!READ_ONCE(conn->sndbuf_desc)))
		return -ENOBUFS;

	smc_cdc_add_pending_send(conn, pend);

	conn->tx_cdc_seq++;
	conn->local_tx_ctrl.seqno = conn->tx_cdc_seq;
	smc_host_msg_to_cdc((struct smc_cdc_msg *)wr_buf, conn, &cfed);

	atomic_inc(&conn->cdc_pend_tx_wr);
	smp_mb__after_atomic(); /* Make sure cdc_pend_tx_wr added before post */

	rc = smc_wr_tx_send(link, (struct smc_wr_tx_pend_priv *)pend);
	if (!rc) {
		smc_curs_copy(&conn->rx_curs_confirmed, &cfed, conn);
		conn->local_rx_ctrl.prod_flags.cons_curs_upd_req = 0;
	} else {
		conn->tx_cdc_seq--;
		conn->local_tx_ctrl.seqno = conn->tx_cdc_seq;
		atomic_dec(&conn->cdc_pend_tx_wr);
	}

	return rc;
}

/* send a validation msg indicating the move of a conn to an other QP link */
int smcr_cdc_msg_send_validation(struct smc_connection *conn,
				 struct smc_cdc_tx_pend *pend,
				 struct smc_wr_buf *wr_buf)
{
	struct smc_host_cdc_msg *local = &conn->local_tx_ctrl;
	struct smc_link *link = conn->lnk;
	struct smc_cdc_msg *peer;
	int rc;

	peer = (struct smc_cdc_msg *)wr_buf;
	peer->common.type = local->common.type;
	peer->len = local->len;
	peer->seqno = htons(conn->tx_cdc_seq_fin); /* seqno last compl. tx */
	peer->token = htonl(local->token);
	peer->prod_flags.failover_validation = 1;

	/* We need to set pend->conn here to make sure smc_cdc_tx_handler()
	 * can handle properly
	 */
	smc_cdc_add_pending_send(conn, pend);

	atomic_inc(&conn->cdc_pend_tx_wr);
	smp_mb__after_atomic(); /* Make sure cdc_pend_tx_wr added before post */

	rc = smc_wr_tx_send(link, (struct smc_wr_tx_pend_priv *)pend);
	if (unlikely(rc))
		atomic_dec(&conn->cdc_pend_tx_wr);

	return rc;
}

static int smcr_cdc_get_slot_and_msg_send(struct smc_connection *conn)
{
	struct smc_cdc_tx_pend *pend;
	struct smc_wr_buf *wr_buf;
	struct smc_link *link;
	bool again = false;
	int rc;

again:
	link = conn->lnk;
	if (!smc_wr_tx_link_hold(link))
		return -ENOLINK;
	rc = smc_cdc_get_free_slot(conn, link, &wr_buf, NULL, &pend);
	if (rc)
		goto put_out;

	spin_lock_bh(&conn->send_lock);
	if (link != conn->lnk) {
		/* link of connection changed, try again one time*/
		spin_unlock_bh(&conn->send_lock);
		smc_wr_tx_put_slot(link,
				   (struct smc_wr_tx_pend_priv *)pend);
		smc_wr_tx_link_put(link);
		if (again)
			return -ENOLINK;
		again = true;
		goto again;
	}
	rc = smc_cdc_msg_send(conn, wr_buf, pend);
	spin_unlock_bh(&conn->send_lock);
put_out:
	smc_wr_tx_link_put(link);
	return rc;
}

int smc_cdc_get_slot_and_msg_send(struct smc_connection *conn)
{
	int rc;

	if (!smc_conn_lgr_valid(conn) ||
	    (conn->lgr->is_smcd && conn->lgr->peer_shutdown))
		return -EPIPE;

	if (conn->lgr->is_smcd) {
		spin_lock_bh(&conn->send_lock);
		rc = smcd_cdc_msg_send(conn);
		spin_unlock_bh(&conn->send_lock);
	} else {
		rc = smcr_cdc_get_slot_and_msg_send(conn);
	}

	return rc;
}

void smc_cdc_wait_pend_tx_wr(struct smc_connection *conn)
{
	wait_event(conn->cdc_pend_tx_wq, !atomic_read(&conn->cdc_pend_tx_wr));
}

/* Send a SMC-D CDC header.
 * This increments the free space available in our send buffer.
 * Also update the confirmed receive buffer with what was sent to the peer.
 */
int smcd_cdc_msg_send(struct smc_connection *conn)
{
	struct smc_sock *smc = container_of(conn, struct smc_sock, conn);
	union smc_host_cursor curs;
	struct smcd_cdc_msg cdc;
	int rc, diff;

	memset(&cdc, 0, sizeof(cdc));
	cdc.common.type = SMC_CDC_MSG_TYPE;
	curs.acurs.counter = atomic64_read(&conn->local_tx_ctrl.prod.acurs);
	cdc.prod.wrap = curs.wrap;
	cdc.prod.count = curs.count;
	curs.acurs.counter = atomic64_read(&conn->local_tx_ctrl.cons.acurs);
	cdc.cons.wrap = curs.wrap;
	cdc.cons.count = curs.count;
	cdc.cons.prod_flags = conn->local_tx_ctrl.prod_flags;
	cdc.cons.conn_state_flags = conn->local_tx_ctrl.conn_state_flags;
	rc = smcd_tx_ism_write(conn, &cdc, sizeof(cdc), 0, 1);
	if (rc)
		return rc;
	smc_curs_copy(&conn->rx_curs_confirmed, &curs, conn);
	conn->local_rx_ctrl.prod_flags.cons_curs_upd_req = 0;
	/* Calculate transmitted data and increment free send buffer space */
	diff = smc_curs_diff(conn->sndbuf_desc->len, &conn->tx_curs_fin,
			     &conn->tx_curs_sent);
	/* increased by confirmed number of bytes */
	smp_mb__before_atomic();
	atomic_add(diff, &conn->sndbuf_space);
	/* guarantee 0 <= sndbuf_space <= sndbuf_desc->len */
	smp_mb__after_atomic();
	smc_curs_copy(&conn->tx_curs_fin, &conn->tx_curs_sent, conn);

	smc_tx_sndbuf_nonfull(smc);
	return rc;
}

/********************************* receive ***********************************/

static inline bool smc_cdc_before(u16 seq1, u16 seq2)
{
	return (s16)(seq1 - seq2) < 0;
}

static void smc_cdc_handle_urg_data_arrival(struct smc_sock *smc,
					    int *diff_prod)
{
	struct smc_connection *conn = &smc->conn;
	char *base;

	/* new data included urgent business */
	smc_curs_copy(&conn->urg_curs, &conn->local_rx_ctrl.prod, conn);
	conn->urg_state = SMC_URG_VALID;
	if (!sock_flag(&smc->sk, SOCK_URGINLINE))
		/* we'll skip the urgent byte, so don't account for it */
		(*diff_prod)--;
	base = (char *)conn->rmb_desc->cpu_addr + conn->rx_off;
	if (conn->urg_curs.count)
		conn->urg_rx_byte = *(base + conn->urg_curs.count - 1);
	else
		conn->urg_rx_byte = *(base + conn->rmb_desc->len - 1);
	sk_send_sigurg(&smc->sk);
}

static void smc_cdc_msg_validate(struct smc_sock *smc, struct smc_cdc_msg *cdc,
				 struct smc_link *link)
{
	struct smc_connection *conn = &smc->conn;
	u16 recv_seq = ntohs(cdc->seqno);
	s16 diff;

	/* check that seqnum was seen before */
	diff = conn->local_rx_ctrl.seqno - recv_seq;
	if (diff < 0) { /* diff larger than 0x7fff */
		/* drop connection */
		conn->out_of_sync = 1;	/* prevent any further receives */
		spin_lock_bh(&conn->send_lock);
		conn->local_tx_ctrl.conn_state_flags.peer_conn_abort = 1;
		conn->lnk = link;
		spin_unlock_bh(&conn->send_lock);
		sock_hold(&smc->sk); /* sock_put in abort_work */
		if (!queue_work(smc_close_wq, &conn->abort_work))
			sock_put(&smc->sk);
	}
}

static void smc_cdc_msg_recv_action(struct smc_sock *smc,
				    struct smc_cdc_msg *cdc)
{
	union smc_host_cursor cons_old, prod_old;
	struct smc_connection *conn = &smc->conn;
	int diff_cons, diff_prod;

	smc_curs_copy(&prod_old, &conn->local_rx_ctrl.prod, conn);
	smc_curs_copy(&cons_old, &conn->local_rx_ctrl.cons, conn);
	smc_cdc_msg_to_host(&conn->local_rx_ctrl, cdc, conn);

	diff_cons = smc_curs_diff(conn->peer_rmbe_size, &cons_old,
				  &conn->local_rx_ctrl.cons);
	if (diff_cons) {
		/* peer_rmbe_space is decreased during data transfer with RDMA
		 * write
		 */
		smp_mb__before_atomic();
		atomic_add(diff_cons, &conn->peer_rmbe_space);
		/* guarantee 0 <= peer_rmbe_space <= peer_rmbe_size */
		smp_mb__after_atomic();
	}

	diff_prod = smc_curs_diff(conn->rmb_desc->len, &prod_old,
				  &conn->local_rx_ctrl.prod);
	if (diff_prod) {
		if (conn->local_rx_ctrl.prod_flags.urg_data_present)
			smc_cdc_handle_urg_data_arrival(smc, &diff_prod);
		/* bytes_to_rcv is decreased in smc_recvmsg */
		smp_mb__before_atomic();
		atomic_add(diff_prod, &conn->bytes_to_rcv);
		/* guarantee 0 <= bytes_to_rcv <= rmb_desc->len */
		smp_mb__after_atomic();
		smc->sk.sk_data_ready(&smc->sk);
	} else {
		if (conn->local_rx_ctrl.prod_flags.write_blocked)
			smc->sk.sk_data_ready(&smc->sk);
		if (conn->local_rx_ctrl.prod_flags.urg_data_pending)
			conn->urg_state = SMC_URG_NOTYET;
	}

	/* trigger sndbuf consumer: RDMA write into peer RMBE and CDC */
	if ((diff_cons && smc_tx_prepared_sends(conn)) ||
	    conn->local_rx_ctrl.prod_flags.cons_curs_upd_req ||
	    conn->local_rx_ctrl.prod_flags.urg_data_pending) {
		if (!sock_owned_by_user(&smc->sk))
			smc_tx_pending(conn);
		else
			conn->tx_in_release_sock = true;
	}

	if (diff_cons && conn->urg_tx_pend &&
	    atomic_read(&conn->peer_rmbe_space) == conn->peer_rmbe_size) {
		/* urg data confirmed by peer, indicate we're ready for more */
		conn->urg_tx_pend = false;
		smc->sk.sk_write_space(&smc->sk);
	}

	if (conn->local_rx_ctrl.conn_state_flags.peer_conn_abort) {
		smc->sk.sk_err = ECONNRESET;
		conn->local_tx_ctrl.conn_state_flags.peer_conn_abort = 1;
	}
	if (smc_cdc_rxed_any_close_or_senddone(conn)) {
		smc->sk.sk_shutdown |= RCV_SHUTDOWN;
		if (smc->clcsock && smc->clcsock->sk)
			smc->clcsock->sk->sk_shutdown |= RCV_SHUTDOWN;
		sock_set_flag(&smc->sk, SOCK_DONE);
		sock_hold(&smc->sk); /* sock_put in close_work */
		if (!queue_work(smc_close_wq, &conn->close_work))
			sock_put(&smc->sk);
	}
}

/* called under tasklet context */
static void smc_cdc_msg_recv(struct smc_sock *smc, struct smc_cdc_msg *cdc)
{
	sock_hold(&smc->sk);
	bh_lock_sock(&smc->sk);
	smc_cdc_msg_recv_action(smc, cdc);
	bh_unlock_sock(&smc->sk);
	sock_put(&smc->sk); /* no free sk in softirq-context */
}

/* Schedule a tasklet for this connection. Triggered from the ISM device IRQ
 * handler to indicate update in the DMBE.
 *
 * Context:
 * - tasklet context
 */
static void smcd_cdc_rx_tsklet(struct tasklet_struct *t)
{
	struct smc_connection *conn = from_tasklet(conn, t, rx_tsklet);
	struct smcd_cdc_msg *data_cdc;
	struct smcd_cdc_msg cdc;
	struct smc_sock *smc;

	if (!conn || conn->killed)
		return;

	data_cdc = (struct smcd_cdc_msg *)conn->rmb_desc->cpu_addr;
	smcd_curs_copy(&cdc.prod, &data_cdc->prod, conn);
	smcd_curs_copy(&cdc.cons, &data_cdc->cons, conn);
	smc = container_of(conn, struct smc_sock, conn);
	smc_cdc_msg_recv(smc, (struct smc_cdc_msg *)&cdc);
}

/* Initialize receive tasklet. Called from ISM device IRQ handler to start
 * receiver side.
 */
void smcd_cdc_rx_init(struct smc_connection *conn)
{
	tasklet_setup(&conn->rx_tsklet, smcd_cdc_rx_tsklet);
}

/***************************** init, exit, misc ******************************/

static void smc_cdc_rx_handler(struct ib_wc *wc, void *buf)
{
	struct smc_link *link = (struct smc_link *)wc->qp->qp_context;
	struct smc_cdc_msg *cdc = buf;
	struct smc_connection *conn;
	struct smc_link_group *lgr;
	struct smc_sock *smc;

	if (wc->byte_len < offsetof(struct smc_cdc_msg, reserved))
		return; /* short message */
	if (cdc->len != SMC_WR_TX_SIZE)
		return; /* invalid message */

	/* lookup connection */
	lgr = smc_get_lgr(link);
	read_lock_bh(&lgr->conns_lock);
	conn = smc_lgr_find_conn(ntohl(cdc->token), lgr);
	read_unlock_bh(&lgr->conns_lock);
	if (!conn || conn->out_of_sync)
		return;
	smc = container_of(conn, struct smc_sock, conn);

	if (cdc->prod_flags.failover_validation) {
		smc_cdc_msg_validate(smc, cdc, link);
		return;
	}
	if (smc_cdc_before(ntohs(cdc->seqno),
			   conn->local_rx_ctrl.seqno))
		/* received seqno is old */
		return;

	smc_cdc_msg_recv(smc, cdc);
}

static struct smc_wr_rx_handler smc_cdc_rx_handlers[] = {
	{
		.handler	= smc_cdc_rx_handler,
		.type		= SMC_CDC_MSG_TYPE
	},
	{
		.handler	= NULL,
	}
};

int __init smc_cdc_init(void)
{
	struct smc_wr_rx_handler *handler;
	int rc = 0;

	for (handler = smc_cdc_rx_handlers; handler->handler; handler++) {
		INIT_HLIST_NODE(&handler->list);
		rc = smc_wr_rx_register_handler(handler);
		if (rc)
			break;
	}
	return rc;
}
Commit	Line	Data
b2441318	1	// SPDX-License-Identifier: GPL-2.0
5f08318f UB	2	/*
	3	* Shared Memory Communications over RDMA (SMC-R) and RoCE
	4	*
	5	* Connection Data Control (CDC)
	6	* handles flow control
	7	*
	8	* Copyright IBM Corp. 2016
	9	*
	10	* Author(s): Ursula Braun <ubraun@linux.vnet.ibm.com>
	11	*/
	12
	13	#include <linux/spinlock.h>
	14
	15	#include "smc.h"
	16	#include "smc_wr.h"
	17	#include "smc_cdc.h"
e6727f39	18	#include "smc_tx.h"
952310cc	19	#include "smc_rx.h"
b38d7324	20	#include "smc_close.h"
5f08318f UB	21
	22	/******************************** send ***********************************/
	23
5f08318f UB	24	/* handler for send/transmission completion of a CDC msg */
	25	static void smc_cdc_tx_handler(struct smc_wr_tx_pend_priv *pnd_snd,
	26	struct smc_link *link,
	27	enum ib_wc_status wc_status)
	28	{
	29	struct smc_cdc_tx_pend cdcpend = (struct smc_cdc_tx_pend )pnd_snd;
bac6de7b	30	struct smc_connection *conn = cdcpend->conn;
5f08318f UB	31	struct smc_sock *smc;
	32	int diff;
	33
bac6de7b	34	smc = container_of(conn, struct smc_sock, conn);
5f08318f UB	35	bh_lock_sock(&smc->sk);
5f08318f UB	36	if (!wc_status) {
69cb7dc0	37	diff = smc_curs_diff(cdcpend->conn->sndbuf_desc->len,
5f08318f UB	38	&cdcpend->conn->tx_curs_fin,
	39	&cdcpend->cursor);
	40	/* sndbuf_space is decreased in smc_sendmsg */
	41	smp_mb__before_atomic();
	42	atomic_add(diff, &cdcpend->conn->sndbuf_space);
69cb7dc0	43	/* guarantee 0 <= sndbuf_space <= sndbuf_desc->len */
5f08318f	44	smp_mb__after_atomic();
bac6de7b	45	smc_curs_copy(&conn->tx_curs_fin, &cdcpend->cursor, conn);
f0ec4f1d KG	46	smc_curs_copy(&conn->local_tx_ctrl_fin, &cdcpend->p_cursor,
	47	conn);
	48	conn->tx_cdc_seq_fin = cdcpend->ctrl_seq;
5f08318f	49	}
349d4312	50
dcd2cf5f	51	if (atomic_dec_and_test(&conn->cdc_pend_tx_wr)) {
6b88af83 DL	52	/* If user owns the sock_lock, mark the connection need sending.
	53	* User context will later try to send when it release sock_lock
	54	* in smc_release_cb()
dcd2cf5f	55	*/
6b88af83 DL	56	if (sock_owned_by_user(&smc->sk))
	57	conn->tx_in_release_sock = true;
	58	else
	59	smc_tx_pending(conn);
	60
dcd2cf5f DL	61	if (unlikely(wq_has_sleeper(&conn->cdc_pend_tx_wq)))
	62	wake_up(&conn->cdc_pend_tx_wq);
	63	}
349d4312 DL	64	WARN_ON(atomic_read(&conn->cdc_pend_tx_wr) < 0);
349d4312 DL	65
e6727f39	66	smc_tx_sndbuf_nonfull(smc);
5f08318f UB	67	bh_unlock_sock(&smc->sk);
	68	}
	69
51957bc5	70	int smc_cdc_get_free_slot(struct smc_connection *conn,
c6f02ebe	71	struct smc_link *link,
5f08318f	72	struct smc_wr_buf **wr_buf,
ad6f317f	73	struct smc_rdma_wr **wr_rdma_buf,
5f08318f UB	74	struct smc_cdc_tx_pend **pend)
5f08318f UB	75	{
1a0a04c7	76	int rc;
51957bc5	77
1a0a04c7	78	rc = smc_wr_tx_get_free_slot(link, smc_cdc_tx_handler, wr_buf,
ad6f317f	79	wr_rdma_buf,
1a0a04c7	80	(struct smc_wr_tx_pend_priv **)pend);
2bced6ae	81	if (conn->killed) {
1a0a04c7	82	/* abnormal termination */
2bced6ae KG	83	if (!rc)
2bced6ae KG	84	smc_wr_tx_put_slot(link,
e225c9a5	85	(struct smc_wr_tx_pend_priv )(pend));
1a0a04c7	86	rc = -EPIPE;
2bced6ae	87	}
1a0a04c7	88	return rc;
5f08318f UB	89	}
	90
	91	static inline void smc_cdc_add_pending_send(struct smc_connection *conn,
	92	struct smc_cdc_tx_pend *pend)
	93	{
	94	BUILD_BUG_ON_MSG(
	95	sizeof(struct smc_cdc_msg) > SMC_WR_BUF_SIZE,
	96	"must increase SMC_WR_BUF_SIZE to at least sizeof(struct smc_cdc_msg)");
	97	BUILD_BUG_ON_MSG(
b9a22dd9	98	offsetofend(struct smc_cdc_msg, reserved) > SMC_WR_TX_SIZE,
5f08318f UB	99	"must adapt SMC_WR_TX_SIZE to sizeof(struct smc_cdc_msg); if not all smc_wr upper layer protocols use the same message size any more, must start to set link->wr_tx_sges[i].length on each individual smc_wr_tx_send()");
	100	BUILD_BUG_ON_MSG(
	101	sizeof(struct smc_cdc_tx_pend) > SMC_WR_TX_PEND_PRIV_SIZE,
	102	"must increase SMC_WR_TX_PEND_PRIV_SIZE to at least sizeof(struct smc_cdc_tx_pend)");
	103	pend->conn = conn;
	104	pend->cursor = conn->tx_curs_sent;
	105	pend->p_cursor = conn->local_tx_ctrl.prod;
	106	pend->ctrl_seq = conn->tx_cdc_seq;
	107	}
	108
	109	int smc_cdc_msg_send(struct smc_connection *conn,
	110	struct smc_wr_buf *wr_buf,
	111	struct smc_cdc_tx_pend *pend)
	112	{
387707fd	113	struct smc_link *link = conn->lnk;
b8649efa	114	union smc_host_cursor cfed;
5f08318f UB	115	int rc;
5f08318f UB	116
22a825c5 W	117	if (unlikely(!READ_ONCE(conn->sndbuf_desc)))
	118	return -ENOBUFS;
	119
5f08318f UB	120	smc_cdc_add_pending_send(conn, pend);
	121
	122	conn->tx_cdc_seq++;
	123	conn->local_tx_ctrl.seqno = conn->tx_cdc_seq;
ccc8ca9b	124	smc_host_msg_to_cdc((struct smc_cdc_msg *)wr_buf, conn, &cfed);
349d4312 DL	125
	126	atomic_inc(&conn->cdc_pend_tx_wr);
	127	smp_mb__after_atomic(); /* Make sure cdc_pend_tx_wr added before post */
	128
5f08318f	129	rc = smc_wr_tx_send(link, (struct smc_wr_tx_pend_priv *)pend);
4dff63c2	130	if (!rc) {
b8649efa	131	smc_curs_copy(&conn->rx_curs_confirmed, &cfed, conn);
4dff63c2	132	conn->local_rx_ctrl.prod_flags.cons_curs_upd_req = 0;
f0ec4f1d KG	133	} else {
	134	conn->tx_cdc_seq--;
	135	conn->local_tx_ctrl.seqno = conn->tx_cdc_seq;
349d4312	136	atomic_dec(&conn->cdc_pend_tx_wr);
4dff63c2	137	}
5f08318f UB	138
	139	return rc;
	140	}
	141
29bd73db	142	/* send a validation msg indicating the move of a conn to an other QP link */
b8ded9de KG	143	int smcr_cdc_msg_send_validation(struct smc_connection *conn,
	144	struct smc_cdc_tx_pend *pend,
	145	struct smc_wr_buf *wr_buf)
29bd73db KG	146	{
	147	struct smc_host_cdc_msg *local = &conn->local_tx_ctrl;
	148	struct smc_link *link = conn->lnk;
29bd73db KG	149	struct smc_cdc_msg *peer;
	150	int rc;
	151
29bd73db KG	152	peer = (struct smc_cdc_msg *)wr_buf;
	153	peer->common.type = local->common.type;
	154	peer->len = local->len;
	155	peer->seqno = htons(conn->tx_cdc_seq_fin); /* seqno last compl. tx */
	156	peer->token = htonl(local->token);
	157	peer->prod_flags.failover_validation = 1;
	158
349d4312 DL	159	/* We need to set pend->conn here to make sure smc_cdc_tx_handler()
	160	* can handle properly
	161	*/
	162	smc_cdc_add_pending_send(conn, pend);
	163
	164	atomic_inc(&conn->cdc_pend_tx_wr);
	165	smp_mb__after_atomic(); /* Make sure cdc_pend_tx_wr added before post */
	166
29bd73db	167	rc = smc_wr_tx_send(link, (struct smc_wr_tx_pend_priv *)pend);
349d4312 DL	168	if (unlikely(rc))
	169	atomic_dec(&conn->cdc_pend_tx_wr);
	170
29bd73db KG	171	return rc;
	172	}
	173
be244f28	174	static int smcr_cdc_get_slot_and_msg_send(struct smc_connection *conn)
5f08318f UB	175	{
	176	struct smc_cdc_tx_pend *pend;
	177	struct smc_wr_buf *wr_buf;
c6f02ebe KG	178	struct smc_link *link;
c6f02ebe KG	179	bool again = false;
5f08318f UB	180	int rc;
5f08318f UB	181
c6f02ebe KG	182	again:
c6f02ebe KG	183	link = conn->lnk;
95f7f3e7 KG	184	if (!smc_wr_tx_link_hold(link))
95f7f3e7 KG	185	return -ENOLINK;
c6f02ebe	186	rc = smc_cdc_get_free_slot(conn, link, &wr_buf, NULL, &pend);
5f08318f	187	if (rc)
95f7f3e7	188	goto put_out;
5f08318f	189
2dee25af	190	spin_lock_bh(&conn->send_lock);
c6f02ebe KG	191	if (link != conn->lnk) {
	192	/* link of connection changed, try again one time*/
	193	spin_unlock_bh(&conn->send_lock);
	194	smc_wr_tx_put_slot(link,
	195	(struct smc_wr_tx_pend_priv *)pend);
95f7f3e7	196	smc_wr_tx_link_put(link);
c6f02ebe KG	197	if (again)
	198	return -ENOLINK;
	199	again = true;
	200	goto again;
	201	}
2dee25af KG	202	rc = smc_cdc_msg_send(conn, wr_buf, pend);
2dee25af KG	203	spin_unlock_bh(&conn->send_lock);
95f7f3e7 KG	204	put_out:
95f7f3e7 KG	205	smc_wr_tx_link_put(link);
2dee25af	206	return rc;
5f08318f UB	207	}
5f08318f UB	208
be244f28 HW	209	int smc_cdc_get_slot_and_msg_send(struct smc_connection *conn)
	210	{
	211	int rc;
	212
ea89c6c0 WG	213	if (!smc_conn_lgr_valid(conn) \|\|
ea89c6c0 WG	214	(conn->lgr->is_smcd && conn->lgr->peer_shutdown))
50c6b20e UB	215	return -EPIPE;
50c6b20e UB	216
be244f28 HW	217	if (conn->lgr->is_smcd) {
	218	spin_lock_bh(&conn->send_lock);
	219	rc = smcd_cdc_msg_send(conn);
	220	spin_unlock_bh(&conn->send_lock);
	221	} else {
	222	rc = smcr_cdc_get_slot_and_msg_send(conn);
	223	}
	224
	225	return rc;
	226	}
	227
349d4312	228	void smc_cdc_wait_pend_tx_wr(struct smc_connection *conn)
5f08318f	229	{
349d4312	230	wait_event(conn->cdc_pend_tx_wq, !atomic_read(&conn->cdc_pend_tx_wr));
5f08318f UB	231	}
5f08318f UB	232
be244f28 HW	233	/* Send a SMC-D CDC header.
	234	* This increments the free space available in our send buffer.
	235	* Also update the confirmed receive buffer with what was sent to the peer.
	236	*/
	237	int smcd_cdc_msg_send(struct smc_connection *conn)
	238	{
	239	struct smc_sock *smc = container_of(conn, struct smc_sock, conn);
b9a22dd9	240	union smc_host_cursor curs;
be244f28 HW	241	struct smcd_cdc_msg cdc;
	242	int rc, diff;
	243
	244	memset(&cdc, 0, sizeof(cdc));
	245	cdc.common.type = SMC_CDC_MSG_TYPE;
b9a22dd9 UB	246	curs.acurs.counter = atomic64_read(&conn->local_tx_ctrl.prod.acurs);
	247	cdc.prod.wrap = curs.wrap;
	248	cdc.prod.count = curs.count;
	249	curs.acurs.counter = atomic64_read(&conn->local_tx_ctrl.cons.acurs);
	250	cdc.cons.wrap = curs.wrap;
	251	cdc.cons.count = curs.count;
	252	cdc.cons.prod_flags = conn->local_tx_ctrl.prod_flags;
	253	cdc.cons.conn_state_flags = conn->local_tx_ctrl.conn_state_flags;
be244f28 HW	254	rc = smcd_tx_ism_write(conn, &cdc, sizeof(cdc), 0, 1);
	255	if (rc)
	256	return rc;
b9a22dd9	257	smc_curs_copy(&conn->rx_curs_confirmed, &curs, conn);
4dff63c2	258	conn->local_rx_ctrl.prod_flags.cons_curs_upd_req = 0;
be244f28 HW	259	/* Calculate transmitted data and increment free send buffer space */
	260	diff = smc_curs_diff(conn->sndbuf_desc->len, &conn->tx_curs_fin,
	261	&conn->tx_curs_sent);
	262	/* increased by confirmed number of bytes */
	263	smp_mb__before_atomic();
	264	atomic_add(diff, &conn->sndbuf_space);
	265	/* guarantee 0 <= sndbuf_space <= sndbuf_desc->len */
	266	smp_mb__after_atomic();
bac6de7b	267	smc_curs_copy(&conn->tx_curs_fin, &conn->tx_curs_sent, conn);
be244f28 HW	268
	269	smc_tx_sndbuf_nonfull(smc);
	270	return rc;
	271	}
	272
5f08318f UB	273	/******************************* receive *********************************/
	274
	275	static inline bool smc_cdc_before(u16 seq1, u16 seq2)
	276	{
	277	return (s16)(seq1 - seq2) < 0;
	278	}
	279
de8474eb SR	280	static void smc_cdc_handle_urg_data_arrival(struct smc_sock *smc,
	281	int *diff_prod)
	282	{
	283	struct smc_connection *conn = &smc->conn;
	284	char *base;
	285
	286	/* new data included urgent business */
bac6de7b	287	smc_curs_copy(&conn->urg_curs, &conn->local_rx_ctrl.prod, conn);
de8474eb SR	288	conn->urg_state = SMC_URG_VALID;
	289	if (!sock_flag(&smc->sk, SOCK_URGINLINE))
	290	/* we'll skip the urgent byte, so don't account for it */
	291	(*diff_prod)--;
be244f28	292	base = (char *)conn->rmb_desc->cpu_addr + conn->rx_off;
de8474eb SR	293	if (conn->urg_curs.count)
	294	conn->urg_rx_byte = *(base + conn->urg_curs.count - 1);
	295	else
	296	conn->urg_rx_byte = *(base + conn->rmb_desc->len - 1);
	297	sk_send_sigurg(&smc->sk);
	298	}
	299
b286a065 KG	300	static void smc_cdc_msg_validate(struct smc_sock smc, struct smc_cdc_msg cdc,
	301	struct smc_link *link)
	302	{
	303	struct smc_connection *conn = &smc->conn;
	304	u16 recv_seq = ntohs(cdc->seqno);
	305	s16 diff;
	306
	307	/* check that seqnum was seen before */
	308	diff = conn->local_rx_ctrl.seqno - recv_seq;
	309	if (diff < 0) { /* diff larger than 0x7fff */
	310	/* drop connection */
	311	conn->out_of_sync = 1; /* prevent any further receives */
	312	spin_lock_bh(&conn->send_lock);
	313	conn->local_tx_ctrl.conn_state_flags.peer_conn_abort = 1;
	314	conn->lnk = link;
	315	spin_unlock_bh(&conn->send_lock);
	316	sock_hold(&smc->sk); /* sock_put in abort_work */
22ef473d	317	if (!queue_work(smc_close_wq, &conn->abort_work))
b286a065 KG	318	sock_put(&smc->sk);
	319	}
	320	}
	321
5f08318f	322	static void smc_cdc_msg_recv_action(struct smc_sock *smc,
5f08318f UB	323	struct smc_cdc_msg *cdc)
	324	{
	325	union smc_host_cursor cons_old, prod_old;
	326	struct smc_connection *conn = &smc->conn;
	327	int diff_cons, diff_prod;
	328
bac6de7b SR	329	smc_curs_copy(&prod_old, &conn->local_rx_ctrl.prod, conn);
bac6de7b SR	330	smc_curs_copy(&cons_old, &conn->local_rx_ctrl.cons, conn);
5f08318f UB	331	smc_cdc_msg_to_host(&conn->local_rx_ctrl, cdc, conn);
	332
	333	diff_cons = smc_curs_diff(conn->peer_rmbe_size, &cons_old,
	334	&conn->local_rx_ctrl.cons);
	335	if (diff_cons) {
	336	/* peer_rmbe_space is decreased during data transfer with RDMA
	337	* write
	338	*/
	339	smp_mb__before_atomic();
	340	atomic_add(diff_cons, &conn->peer_rmbe_space);
	341	/* guarantee 0 <= peer_rmbe_space <= peer_rmbe_size */
	342	smp_mb__after_atomic();
	343	}
	344
69cb7dc0	345	diff_prod = smc_curs_diff(conn->rmb_desc->len, &prod_old,
5f08318f UB	346	&conn->local_rx_ctrl.prod);
5f08318f UB	347	if (diff_prod) {
de8474eb SR	348	if (conn->local_rx_ctrl.prod_flags.urg_data_present)
de8474eb SR	349	smc_cdc_handle_urg_data_arrival(smc, &diff_prod);
5f08318f UB	350	/* bytes_to_rcv is decreased in smc_recvmsg */
	351	smp_mb__before_atomic();
	352	atomic_add(diff_prod, &conn->bytes_to_rcv);
69cb7dc0	353	/* guarantee 0 <= bytes_to_rcv <= rmb_desc->len */
5f08318f	354	smp_mb__after_atomic();
952310cc	355	smc->sk.sk_data_ready(&smc->sk);
de8474eb	356	} else {
cf0cfe53 KG	357	if (conn->local_rx_ctrl.prod_flags.write_blocked)
	358	smc->sk.sk_data_ready(&smc->sk);
	359	if (conn->local_rx_ctrl.prod_flags.urg_data_pending)
	360	conn->urg_state = SMC_URG_NOTYET;
5f08318f UB	361	}
5f08318f UB	362
51f1de79	363	/* trigger sndbuf consumer: RDMA write into peer RMBE and CDC */
cf0cfe53 KG	364	if ((diff_cons && smc_tx_prepared_sends(conn)) \|\|
cf0cfe53 KG	365	conn->local_rx_ctrl.prod_flags.cons_curs_upd_req \|\|
6b88af83 DL	366	conn->local_rx_ctrl.prod_flags.urg_data_pending) {
	367	if (!sock_owned_by_user(&smc->sk))
	368	smc_tx_pending(conn);
	369	else
	370	conn->tx_in_release_sock = true;
	371	}
cf0cfe53	372
de8474eb SR	373	if (diff_cons && conn->urg_tx_pend &&
	374	atomic_read(&conn->peer_rmbe_space) == conn->peer_rmbe_size) {
	375	/* urg data confirmed by peer, indicate we're ready for more */
	376	conn->urg_tx_pend = false;
	377	smc->sk.sk_write_space(&smc->sk);
	378	}
51f1de79	379
b38d7324	380	if (conn->local_rx_ctrl.conn_state_flags.peer_conn_abort) {
5f08318f	381	smc->sk.sk_err = ECONNRESET;
b38d7324 UB	382	conn->local_tx_ctrl.conn_state_flags.peer_conn_abort = 1;
b38d7324 UB	383	}
46c28dbd UB	384	if (smc_cdc_rxed_any_close_or_senddone(conn)) {
	385	smc->sk.sk_shutdown \|= RCV_SHUTDOWN;
	386	if (smc->clcsock && smc->clcsock->sk)
	387	smc->clcsock->sk->sk_shutdown \|= RCV_SHUTDOWN;
	388	sock_set_flag(&smc->sk, SOCK_DONE);
51f1de79	389	sock_hold(&smc->sk); /* sock_put in close_work */
22ef473d	390	if (!queue_work(smc_close_wq, &conn->close_work))
51f1de79	391	sock_put(&smc->sk);
b38d7324	392	}
5f08318f UB	393	}
	394
	395	/* called under tasklet context */
d7b0e37c	396	static void smc_cdc_msg_recv(struct smc_sock smc, struct smc_cdc_msg cdc)
5f08318f	397	{
5f08318f	398	sock_hold(&smc->sk);
5f08318f	399	bh_lock_sock(&smc->sk);
d7b0e37c	400	smc_cdc_msg_recv_action(smc, cdc);
5f08318f UB	401	bh_unlock_sock(&smc->sk);
	402	sock_put(&smc->sk); /* no free sk in softirq-context */
	403	}
	404
be244f28 HW	405	/* Schedule a tasklet for this connection. Triggered from the ISM device IRQ
	406	* handler to indicate update in the DMBE.
	407	*
	408	* Context:
	409	* - tasklet context
	410	*/
fcb8e3a3	411	static void smcd_cdc_rx_tsklet(struct tasklet_struct *t)
be244f28	412	{
fcb8e3a3	413	struct smc_connection *conn = from_tasklet(conn, t, rx_tsklet);
b9a22dd9	414	struct smcd_cdc_msg *data_cdc;
be244f28 HW	415	struct smcd_cdc_msg cdc;
	416	struct smc_sock *smc;
	417
b2900980	418	if (!conn \|\| conn->killed)
be244f28 HW	419	return;
be244f28 HW	420
b9a22dd9 UB	421	data_cdc = (struct smcd_cdc_msg *)conn->rmb_desc->cpu_addr;
	422	smcd_curs_copy(&cdc.prod, &data_cdc->prod, conn);
	423	smcd_curs_copy(&cdc.cons, &data_cdc->cons, conn);
be244f28 HW	424	smc = container_of(conn, struct smc_sock, conn);
	425	smc_cdc_msg_recv(smc, (struct smc_cdc_msg *)&cdc);
	426	}
	427
	428	/* Initialize receive tasklet. Called from ISM device IRQ handler to start
	429	* receiver side.
	430	*/
	431	void smcd_cdc_rx_init(struct smc_connection *conn)
	432	{
fcb8e3a3	433	tasklet_setup(&conn->rx_tsklet, smcd_cdc_rx_tsklet);
be244f28 HW	434	}
be244f28 HW	435
5f08318f UB	436	/*************************** init, exit, misc ****************************/
	437
	438	static void smc_cdc_rx_handler(struct ib_wc wc, void buf)
	439	{
	440	struct smc_link link = (struct smc_link )wc->qp->qp_context;
	441	struct smc_cdc_msg *cdc = buf;
d7b0e37c HW	442	struct smc_connection *conn;
	443	struct smc_link_group *lgr;
	444	struct smc_sock *smc;
5f08318f UB	445
	446	if (wc->byte_len < offsetof(struct smc_cdc_msg, reserved))
	447	return; /* short message */
cbba07a7	448	if (cdc->len != SMC_WR_TX_SIZE)
5f08318f	449	return; /* invalid message */
d7b0e37c HW	450
d7b0e37c HW	451	/* lookup connection */
00e5fb26	452	lgr = smc_get_lgr(link);
d7b0e37c HW	453	read_lock_bh(&lgr->conns_lock);
	454	conn = smc_lgr_find_conn(ntohl(cdc->token), lgr);
	455	read_unlock_bh(&lgr->conns_lock);
b286a065	456	if (!conn \|\| conn->out_of_sync)
d7b0e37c HW	457	return;
	458	smc = container_of(conn, struct smc_sock, conn);
	459
b286a065 KG	460	if (cdc->prod_flags.failover_validation) {
	461	smc_cdc_msg_validate(smc, cdc, link);
	462	return;
d7b0e37c	463	}
b286a065 KG	464	if (smc_cdc_before(ntohs(cdc->seqno),
	465	conn->local_rx_ctrl.seqno))
	466	/* received seqno is old */
	467	return;
	468
d7b0e37c	469	smc_cdc_msg_recv(smc, cdc);
5f08318f UB	470	}
	471
	472	static struct smc_wr_rx_handler smc_cdc_rx_handlers[] = {
	473	{
	474	.handler = smc_cdc_rx_handler,
	475	.type = SMC_CDC_MSG_TYPE
	476	},
	477	{
	478	.handler = NULL,
	479	}
	480	};
	481
	482	int __init smc_cdc_init(void)
	483	{
	484	struct smc_wr_rx_handler *handler;
	485	int rc = 0;
	486
	487	for (handler = smc_cdc_rx_handlers; handler->handler; handler++) {
	488	INIT_HLIST_NODE(&handler->list);
	489	rc = smc_wr_rx_register_handler(handler);
	490	if (rc)
	491	break;
	492	}
	493	return rc;
	494	}