2 * Copyright (C) 2013 Intel Corporation
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 #include <sys/cdefs.h>
28 __FBSDID("$FreeBSD$");
30 #include <sys/param.h>
31 #include <sys/kernel.h>
32 #include <sys/systm.h>
35 #include <sys/limits.h>
37 #include <sys/malloc.h>
38 #include <sys/module.h>
39 #include <sys/mutex.h>
40 #include <sys/queue.h>
41 #include <sys/socket.h>
42 #include <sys/sockio.h>
43 #include <sys/sysctl.h>
44 #include <sys/taskqueue.h>
46 #include <net/if_media.h>
47 #include <net/if_types.h>
48 #include <net/if_var.h>
50 #include <net/ethernet.h>
53 #include <machine/bus.h>
54 #include <machine/cpufunc.h>
55 #include <machine/pmap.h>
57 #include "../ntb_hw/ntb_hw.h"
60 * The Non-Transparent Bridge (NTB) is a device on some Intel processors that
61 * allows you to connect two systems using a PCI-e link.
63 * This module contains a protocol for sending and receiving messages, and
64 * exposes that protocol through a simulated ethernet device called ntb.
66 * NOTE: Much of the code in this module is shared with Linux. Any patches may
67 * be picked up and redistributed in Linux with a dual GPL/BSD license.
70 /* TODO: These functions should really be part of the kernel */
71 #define test_bit(pos, bitmap_addr) (*(bitmap_addr) & 1UL << (pos))
72 #define set_bit(pos, bitmap_addr) *(bitmap_addr) |= 1UL << (pos)
73 #define clear_bit(pos, bitmap_addr) *(bitmap_addr) &= ~(1UL << (pos))
75 #define KTR_NTB KTR_SPARE3
77 #define NTB_TRANSPORT_VERSION 3
78 #define NTB_RX_MAX_PKTS 64
79 #define NTB_RXQ_SIZE 300
81 static unsigned int transport_mtu = 0x4000 + ETHER_HDR_LEN + ETHER_CRC_LEN;
83 static unsigned int max_num_clients;
84 SYSCTL_UINT(_hw_ntb, OID_AUTO, max_num_clients, CTLFLAG_RDTUN,
85 &max_num_clients, 0, "Maximum number of NTB transport clients. "
86 "0 (default) - use all available NTB memory windows; "
87 "positive integer N - Limit to N memory windows.");
89 STAILQ_HEAD(ntb_queue_list, ntb_queue_entry);
91 struct ntb_queue_entry {
92 /* ntb_queue list reference */
93 STAILQ_ENTRY(ntb_queue_entry) entry;
95 /* info on data to be transfered */
106 struct ntb_transport_qp {
107 struct ntb_netdev *transport;
108 struct ntb_softc *ntb;
114 uint8_t qp_num; /* Only 64 QPs are allowed. 0-63 */
116 struct ntb_rx_info *rx_info;
117 struct ntb_rx_info *remote_rx_info;
119 void (*tx_handler)(struct ntb_transport_qp *qp, void *qp_data,
120 void *data, int len);
121 struct ntb_queue_list tx_free_q;
122 struct mtx ntb_tx_free_q_lock;
125 uint64_t tx_max_entry;
126 uint64_t tx_max_frame;
128 void (*rx_handler)(struct ntb_transport_qp *qp, void *qp_data,
129 void *data, int len);
130 struct ntb_queue_list rx_pend_q;
131 struct ntb_queue_list rx_free_q;
132 struct mtx ntb_rx_pend_q_lock;
133 struct mtx ntb_rx_free_q_lock;
134 struct task rx_completion_task;
137 uint64_t rx_max_entry;
138 uint64_t rx_max_frame;
140 void (*event_handler)(void *data, enum ntb_link_event status);
141 struct callout link_work;
142 struct callout queue_full;
143 struct callout rx_full;
145 uint64_t last_rx_no_buf;
150 uint64_t rx_ring_empty;
151 uint64_t rx_err_no_buf;
152 uint64_t rx_err_oflow;
156 uint64_t tx_ring_full;
159 struct ntb_queue_handlers {
160 void (*rx_handler)(struct ntb_transport_qp *qp, void *qp_data,
161 void *data, int len);
162 void (*tx_handler)(struct ntb_transport_qp *qp, void *qp_data,
163 void *data, int len);
164 void (*event_handler)(void *data, enum ntb_link_event status);
168 struct ntb_transport_mw {
175 struct ntb_softc *ntb;
177 struct ntb_transport_mw mw[NTB_MAX_NUM_MW];
178 struct ntb_transport_qp *qps;
182 struct callout link_work;
183 struct ntb_transport_qp *qp;
185 u_char eaddr[ETHER_ADDR_LEN];
190 static struct ntb_netdev net_softc;
193 IF_NTB_DESC_DONE_FLAG = 1 << 0,
194 IF_NTB_LINK_DOWN_FLAG = 1 << 1,
197 struct ntb_payload_header {
205 * The order of this enum is part of the if_ntb remote protocol. Do
206 * not reorder without bumping protocol version (and it's probably best
207 * to keep the protocol in lock-step with the Linux NTB driver.
214 * N.B.: transport_link_work assumes MW1 enums = MW0 + 2.
223 #define QP_TO_MW(ntb, qp) ((qp) % ntb_get_max_mw(ntb))
224 #define NTB_QP_DEF_NUM_ENTRIES 100
225 #define NTB_LINK_DOWN_TIMEOUT 10
227 static int ntb_handle_module_events(struct module *m, int what, void *arg);
228 static int ntb_setup_interface(void);
229 static int ntb_teardown_interface(void);
230 static void ntb_net_init(void *arg);
231 static int ntb_ioctl(struct ifnet *ifp, u_long command, caddr_t data);
232 static void ntb_start(struct ifnet *ifp);
233 static void ntb_net_tx_handler(struct ntb_transport_qp *qp, void *qp_data,
234 void *data, int len);
235 static void ntb_net_rx_handler(struct ntb_transport_qp *qp, void *qp_data,
236 void *data, int len);
237 static void ntb_net_event_handler(void *data, enum ntb_link_event status);
238 static int ntb_transport_init(struct ntb_softc *ntb);
239 static void ntb_transport_free(void *transport);
240 static void ntb_transport_init_queue(struct ntb_netdev *nt,
241 unsigned int qp_num);
242 static void ntb_transport_free_queue(struct ntb_transport_qp *qp);
243 static struct ntb_transport_qp * ntb_transport_create_queue(void *data,
244 struct ntb_softc *pdev, const struct ntb_queue_handlers *handlers);
245 static void ntb_transport_link_up(struct ntb_transport_qp *qp);
246 static int ntb_transport_tx_enqueue(struct ntb_transport_qp *qp, void *cb,
247 void *data, unsigned int len);
248 static int ntb_process_tx(struct ntb_transport_qp *qp,
249 struct ntb_queue_entry *entry);
250 static void ntb_tx_copy_task(struct ntb_transport_qp *qp,
251 struct ntb_queue_entry *entry, void *offset);
252 static void ntb_qp_full(void *arg);
253 static int ntb_transport_rxc_db(void *arg, int dummy);
254 static void ntb_rx_pendq_full(void *arg);
255 static int ntb_process_rxc(struct ntb_transport_qp *qp);
256 static void ntb_rx_copy_task(struct ntb_transport_qp *qp,
257 struct ntb_queue_entry *entry, void *offset);
258 static void ntb_rx_completion_task(void *arg, int pending);
259 static void ntb_transport_event_callback(void *data, enum ntb_hw_event event);
260 static void ntb_transport_link_work(void *arg);
261 static int ntb_set_mw(struct ntb_netdev *nt, int num_mw, unsigned int size);
262 static void ntb_free_mw(struct ntb_netdev *nt, int num_mw);
263 static void ntb_transport_setup_qp_mw(struct ntb_netdev *nt,
264 unsigned int qp_num);
265 static void ntb_qp_link_work(void *arg);
266 static void ntb_transport_link_cleanup(struct ntb_netdev *nt);
267 static void ntb_qp_link_down(struct ntb_transport_qp *qp);
268 static void ntb_qp_link_cleanup(struct ntb_transport_qp *qp);
269 static void ntb_transport_link_down(struct ntb_transport_qp *qp);
270 static void ntb_send_link_down(struct ntb_transport_qp *qp);
271 static void ntb_list_add(struct mtx *lock, struct ntb_queue_entry *entry,
272 struct ntb_queue_list *list);
273 static struct ntb_queue_entry *ntb_list_rm(struct mtx *lock,
274 struct ntb_queue_list *list);
275 static void create_random_local_eui48(u_char *eaddr);
276 static unsigned int ntb_transport_max_size(struct ntb_transport_qp *qp);
278 MALLOC_DEFINE(M_NTB_IF, "if_ntb", "ntb network driver");
280 /* Module setup and teardown */
282 ntb_handle_module_events(struct module *m, int what, void *arg)
288 err = ntb_setup_interface();
291 err = ntb_teardown_interface();
300 static moduledata_t if_ntb_mod = {
302 ntb_handle_module_events,
306 DECLARE_MODULE(if_ntb, if_ntb_mod, SI_SUB_KLD, SI_ORDER_ANY);
307 MODULE_DEPEND(if_ntb, ntb_hw, 1, 1, 1);
310 ntb_setup_interface(void)
313 struct ntb_queue_handlers handlers = { ntb_net_rx_handler,
314 ntb_net_tx_handler, ntb_net_event_handler };
316 net_softc.ntb = devclass_get_softc(devclass_find("ntb_hw"), 0);
317 if (net_softc.ntb == NULL) {
318 printf("ntb: Cannot find devclass\n");
322 ntb_transport_init(net_softc.ntb);
324 ifp = net_softc.ifp = if_alloc(IFT_ETHER);
326 printf("ntb: cannot allocate ifnet structure\n");
330 net_softc.qp = ntb_transport_create_queue(ifp, net_softc.ntb,
332 if_initname(ifp, "ntb", 0);
333 ifp->if_init = ntb_net_init;
334 ifp->if_softc = &net_softc;
335 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX;
336 ifp->if_ioctl = ntb_ioctl;
337 ifp->if_start = ntb_start;
338 IFQ_SET_MAXLEN(&ifp->if_snd, IFQ_MAXLEN);
339 ifp->if_snd.ifq_drv_maxlen = IFQ_MAXLEN;
340 IFQ_SET_READY(&ifp->if_snd);
341 create_random_local_eui48(net_softc.eaddr);
342 ether_ifattach(ifp, net_softc.eaddr);
343 ifp->if_capabilities = IFCAP_HWCSUM | IFCAP_JUMBO_MTU;
344 ifp->if_capenable = ifp->if_capabilities;
346 ntb_transport_link_up(net_softc.qp);
347 net_softc.bufsize = ntb_transport_max_size(net_softc.qp) +
348 sizeof(struct ether_header);
353 ntb_teardown_interface(void)
356 if (net_softc.qp != NULL)
357 ntb_transport_link_down(net_softc.qp);
359 if (net_softc.ifp != NULL) {
360 ether_ifdetach(net_softc.ifp);
361 if_free(net_softc.ifp);
364 if (net_softc.qp != NULL) {
365 ntb_transport_free_queue(net_softc.qp);
366 ntb_transport_free(&net_softc);
372 /* Network device interface */
375 ntb_net_init(void *arg)
377 struct ntb_netdev *ntb_softc = arg;
378 struct ifnet *ifp = ntb_softc->ifp;
380 ifp->if_drv_flags |= IFF_DRV_RUNNING;
381 ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
382 ifp->if_flags |= IFF_UP;
383 if_link_state_change(ifp, LINK_STATE_UP);
387 ntb_ioctl(struct ifnet *ifp, u_long command, caddr_t data)
389 struct ntb_netdev *nt = ifp->if_softc;
390 struct ifreq *ifr = (struct ifreq *)data;
396 if (ifr->ifr_mtu > ntb_transport_max_size(nt->qp) -
397 ETHER_HDR_LEN - ETHER_CRC_LEN) {
402 ifp->if_mtu = ifr->ifr_mtu;
406 error = ether_ioctl(ifp, command, data);
415 ntb_start(struct ifnet *ifp)
418 struct ntb_netdev *nt = ifp->if_softc;
421 mtx_lock(&nt->tx_lock);
422 ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
423 CTR0(KTR_NTB, "TX: ntb_start");
424 while (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) {
425 IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head);
426 CTR1(KTR_NTB, "TX: start mbuf %p", m_head);
427 rc = ntb_transport_tx_enqueue(nt->qp, m_head, m_head,
428 m_length(m_head, NULL));
431 "TX: could not tx mbuf %p. Returning to snd q",
434 ifp->if_drv_flags |= IFF_DRV_OACTIVE;
435 IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
436 callout_reset(&nt->qp->queue_full, hz / 1000,
443 mtx_unlock(&nt->tx_lock);
446 /* Network Device Callbacks */
448 ntb_net_tx_handler(struct ntb_transport_qp *qp, void *qp_data, void *data,
453 CTR1(KTR_NTB, "TX: tx_handler freeing mbuf %p", data);
457 ntb_net_rx_handler(struct ntb_transport_qp *qp, void *qp_data, void *data,
460 struct mbuf *m = data;
461 struct ifnet *ifp = qp_data;
463 CTR0(KTR_NTB, "RX: rx handler");
464 (*ifp->if_input)(ifp, m);
468 ntb_net_event_handler(void *data, enum ntb_link_event status)
475 /* XXX The Linux driver munges with the carrier status here. */
483 panic("Bogus ntb_link_event %u\n", status);
487 /* Transport Init and teardown */
490 ntb_transport_init(struct ntb_softc *ntb)
492 struct ntb_netdev *nt = &net_softc;
495 if (max_num_clients == 0)
496 nt->max_qps = MIN(ntb_get_max_cbs(ntb), ntb_get_max_mw(ntb));
498 nt->max_qps = MIN(ntb_get_max_cbs(ntb), max_num_clients);
500 ntb_register_transport(ntb, nt);
501 mtx_init(&nt->tx_lock, "ntb transport tx", NULL, MTX_DEF);
502 mtx_init(&nt->rx_lock, "ntb transport rx", NULL, MTX_DEF);
504 nt->qps = malloc(nt->max_qps * sizeof(struct ntb_transport_qp),
505 M_NTB_IF, M_WAITOK|M_ZERO);
507 nt->qp_bitmap = ((uint64_t) 1 << nt->max_qps) - 1;
509 for (i = 0; i < nt->max_qps; i++)
510 ntb_transport_init_queue(nt, i);
512 callout_init(&nt->link_work, 0);
514 rc = ntb_register_event_callback(ntb,
515 ntb_transport_event_callback);
519 if (ntb_query_link_status(ntb)) {
521 device_printf(ntb_get_device(ntb), "link up\n");
522 callout_reset(&nt->link_work, 0, ntb_transport_link_work, nt);
528 free(nt->qps, M_NTB_IF);
529 ntb_unregister_transport(ntb);
534 ntb_transport_free(void *transport)
536 struct ntb_netdev *nt = transport;
537 struct ntb_softc *ntb = nt->ntb;
540 ntb_transport_link_cleanup(nt);
542 callout_drain(&nt->link_work);
544 /* verify that all the qps are freed */
545 for (i = 0; i < nt->max_qps; i++)
546 if (!test_bit(i, &nt->qp_bitmap))
547 ntb_transport_free_queue(&nt->qps[i]);
549 ntb_unregister_event_callback(ntb);
551 for (i = 0; i < NTB_MAX_NUM_MW; i++)
554 free(nt->qps, M_NTB_IF);
555 ntb_unregister_transport(ntb);
559 ntb_transport_init_queue(struct ntb_netdev *nt, unsigned int qp_num)
561 struct ntb_transport_qp *qp;
562 unsigned int num_qps_mw, tx_size;
563 uint8_t mw_num, mw_max;
565 mw_max = ntb_get_max_mw(nt->ntb);
566 mw_num = QP_TO_MW(nt->ntb, qp_num);
568 qp = &nt->qps[qp_num];
572 qp->qp_link = NTB_LINK_DOWN;
573 qp->client_ready = NTB_LINK_DOWN;
574 qp->event_handler = NULL;
576 if (nt->max_qps % mw_max && mw_num + 1 < nt->max_qps / mw_max)
577 num_qps_mw = nt->max_qps / mw_max + 1;
579 num_qps_mw = nt->max_qps / mw_max;
581 tx_size = (unsigned int) ntb_get_mw_size(qp->ntb, mw_num) / num_qps_mw;
582 qp->rx_info = (struct ntb_rx_info *)
583 ((char *)ntb_get_mw_vbase(qp->ntb, mw_num) +
584 (qp_num / mw_max * tx_size));
585 tx_size -= sizeof(struct ntb_rx_info);
587 qp->tx_mw = qp->rx_info + 1;
588 /* Due to house-keeping, there must be at least 2 buffs */
589 qp->tx_max_frame = min(transport_mtu + sizeof(struct ntb_payload_header),
591 qp->tx_max_entry = tx_size / qp->tx_max_frame;
593 callout_init(&qp->link_work, 0);
594 callout_init(&qp->queue_full, 1);
595 callout_init(&qp->rx_full, 1);
597 mtx_init(&qp->ntb_rx_pend_q_lock, "ntb rx pend q", NULL, MTX_SPIN);
598 mtx_init(&qp->ntb_rx_free_q_lock, "ntb rx free q", NULL, MTX_SPIN);
599 mtx_init(&qp->ntb_tx_free_q_lock, "ntb tx free q", NULL, MTX_SPIN);
600 TASK_INIT(&qp->rx_completion_task, 0, ntb_rx_completion_task, qp);
602 STAILQ_INIT(&qp->rx_pend_q);
603 STAILQ_INIT(&qp->rx_free_q);
604 STAILQ_INIT(&qp->tx_free_q);
608 ntb_transport_free_queue(struct ntb_transport_qp *qp)
610 struct ntb_queue_entry *entry;
615 callout_drain(&qp->link_work);
617 ntb_unregister_db_callback(qp->ntb, qp->qp_num);
619 while ((entry = ntb_list_rm(&qp->ntb_rx_free_q_lock, &qp->rx_free_q)))
620 free(entry, M_NTB_IF);
622 while ((entry = ntb_list_rm(&qp->ntb_rx_pend_q_lock, &qp->rx_pend_q)))
623 free(entry, M_NTB_IF);
625 while ((entry = ntb_list_rm(&qp->ntb_tx_free_q_lock, &qp->tx_free_q)))
626 free(entry, M_NTB_IF);
628 set_bit(qp->qp_num, &qp->transport->qp_bitmap);
632 * ntb_transport_create_queue - Create a new NTB transport layer queue
633 * @rx_handler: receive callback function
634 * @tx_handler: transmit callback function
635 * @event_handler: event callback function
637 * Create a new NTB transport layer queue and provide the queue with a callback
638 * routine for both transmit and receive. The receive callback routine will be
639 * used to pass up data when the transport has received it on the queue. The
640 * transmit callback routine will be called when the transport has completed the
641 * transmission of the data on the queue and the data is ready to be freed.
643 * RETURNS: pointer to newly created ntb_queue, NULL on error.
645 static struct ntb_transport_qp *
646 ntb_transport_create_queue(void *data, struct ntb_softc *pdev,
647 const struct ntb_queue_handlers *handlers)
649 struct ntb_queue_entry *entry;
650 struct ntb_transport_qp *qp;
651 struct ntb_netdev *nt;
652 unsigned int free_queue;
655 nt = ntb_find_transport(pdev);
659 free_queue = ffs(nt->qp_bitmap);
663 /* decrement free_queue to make it zero based */
666 clear_bit(free_queue, &nt->qp_bitmap);
668 qp = &nt->qps[free_queue];
670 qp->rx_handler = handlers->rx_handler;
671 qp->tx_handler = handlers->tx_handler;
672 qp->event_handler = handlers->event_handler;
674 for (i = 0; i < NTB_QP_DEF_NUM_ENTRIES; i++) {
675 entry = malloc(sizeof(struct ntb_queue_entry), M_NTB_IF,
677 entry->cb_data = nt->ifp;
679 entry->len = transport_mtu;
680 ntb_list_add(&qp->ntb_rx_pend_q_lock, entry, &qp->rx_pend_q);
683 for (i = 0; i < NTB_QP_DEF_NUM_ENTRIES; i++) {
684 entry = malloc(sizeof(struct ntb_queue_entry), M_NTB_IF,
686 ntb_list_add(&qp->ntb_tx_free_q_lock, entry, &qp->tx_free_q);
689 rc = ntb_register_db_callback(qp->ntb, free_queue, qp,
690 ntb_transport_rxc_db);
697 while ((entry = ntb_list_rm(&qp->ntb_tx_free_q_lock, &qp->tx_free_q)))
698 free(entry, M_NTB_IF);
699 while ((entry = ntb_list_rm(&qp->ntb_rx_free_q_lock, &qp->rx_free_q)))
700 free(entry, M_NTB_IF);
701 set_bit(free_queue, &nt->qp_bitmap);
707 * ntb_transport_link_up - Notify NTB transport of client readiness to use queue
708 * @qp: NTB transport layer queue to be enabled
710 * Notify NTB transport layer of client readiness to use queue
713 ntb_transport_link_up(struct ntb_transport_qp *qp)
719 qp->client_ready = NTB_LINK_UP;
721 device_printf(ntb_get_device(qp->ntb), "qp client ready\n");
723 if (qp->transport->transport_link == NTB_LINK_UP)
724 callout_reset(&qp->link_work, 0, ntb_qp_link_work, qp);
732 * ntb_transport_tx_enqueue - Enqueue a new NTB queue entry
733 * @qp: NTB transport layer queue the entry is to be enqueued on
734 * @cb: per buffer pointer for callback function to use
735 * @data: pointer to data buffer that will be sent
736 * @len: length of the data buffer
738 * Enqueue a new transmit buffer onto the transport queue from which a NTB
739 * payload will be transmitted. This assumes that a lock is being held to
740 * serialize access to the qp.
742 * RETURNS: An appropriate ERRNO error value on error, or zero for success.
745 ntb_transport_tx_enqueue(struct ntb_transport_qp *qp, void *cb, void *data,
748 struct ntb_queue_entry *entry;
751 if (qp == NULL || qp->qp_link != NTB_LINK_UP || len == 0) {
752 CTR0(KTR_NTB, "TX: link not up");
756 entry = ntb_list_rm(&qp->ntb_tx_free_q_lock, &qp->tx_free_q);
758 CTR0(KTR_NTB, "TX: could not get entry from tx_free_q");
761 CTR1(KTR_NTB, "TX: got entry %p from tx_free_q", entry);
768 rc = ntb_process_tx(qp, entry);
770 ntb_list_add(&qp->ntb_tx_free_q_lock, entry, &qp->tx_free_q);
772 "TX: process_tx failed. Returning entry %p to tx_free_q",
779 ntb_process_tx(struct ntb_transport_qp *qp, struct ntb_queue_entry *entry)
783 offset = (char *)qp->tx_mw + qp->tx_max_frame * qp->tx_index;
785 "TX: process_tx: tx_pkts=%u, tx_index=%u, remote entry=%u",
786 qp->tx_pkts, qp->tx_index, qp->remote_rx_info->entry);
787 if (qp->tx_index == qp->remote_rx_info->entry) {
788 CTR0(KTR_NTB, "TX: ring full");
793 if (entry->len > qp->tx_max_frame - sizeof(struct ntb_payload_header)) {
794 if (qp->tx_handler != NULL)
795 qp->tx_handler(qp, qp->cb_data, entry->buf,
798 ntb_list_add(&qp->ntb_tx_free_q_lock, entry, &qp->tx_free_q);
800 "TX: frame too big. returning entry %p to tx_free_q",
804 CTR2(KTR_NTB, "TX: copying entry %p to offset %p", entry, offset);
805 ntb_tx_copy_task(qp, entry, offset);
808 qp->tx_index %= qp->tx_max_entry;
816 ntb_tx_copy_task(struct ntb_transport_qp *qp, struct ntb_queue_entry *entry,
819 struct ntb_payload_header *hdr;
821 CTR2(KTR_NTB, "TX: copying %d bytes to offset %p", entry->len, offset);
822 if (entry->buf != NULL)
823 m_copydata((struct mbuf *)entry->buf, 0, entry->len, offset);
825 hdr = (struct ntb_payload_header *)((char *)offset + qp->tx_max_frame -
826 sizeof(struct ntb_payload_header));
827 hdr->len = entry->len; /* TODO: replace with bus_space_write */
828 hdr->ver = qp->tx_pkts; /* TODO: replace with bus_space_write */
830 /* TODO: replace with bus_space_write */
831 hdr->flags = entry->flags | IF_NTB_DESC_DONE_FLAG;
833 ntb_ring_doorbell(qp->ntb, qp->qp_num);
836 * The entry length can only be zero if the packet is intended to be a
837 * "link down" or similar. Since no payload is being sent in these
838 * cases, there is nothing to add to the completion queue.
840 if (entry->len > 0) {
841 qp->tx_bytes += entry->len;
844 qp->tx_handler(qp, qp->cb_data, entry->cb_data,
849 "TX: entry %p sent. hdr->ver = %d, Returning to tx_free_q", entry,
851 ntb_list_add(&qp->ntb_tx_free_q_lock, entry, &qp->tx_free_q);
855 ntb_qp_full(void *arg)
858 CTR0(KTR_NTB, "TX: qp_full callout");
864 ntb_rx_pendq_full(void *arg)
867 CTR0(KTR_NTB, "RX: ntb_rx_pendq_full callout");
868 ntb_transport_rxc_db(arg, 0);
872 ntb_transport_rxc_db(void *arg, int dummy __unused)
874 struct ntb_transport_qp *qp = arg;
879 * Limit the number of packets processed in a single interrupt to
880 * provide fairness to others
882 mtx_lock(&qp->transport->rx_lock);
883 CTR0(KTR_NTB, "RX: transport_rx");
884 for (i = 0; i < MIN(qp->rx_max_entry, INT_MAX); i++) {
885 rc = ntb_process_rxc(qp);
887 CTR0(KTR_NTB, "RX: process_rxc failed");
891 mtx_unlock(&qp->transport->rx_lock);
897 ntb_process_rxc(struct ntb_transport_qp *qp)
899 struct ntb_payload_header *hdr;
900 struct ntb_queue_entry *entry;
904 ((char *)qp->rx_buff + qp->rx_max_frame * qp->rx_index);
906 ((char *)offset + qp->rx_max_frame -
907 sizeof(struct ntb_payload_header));
909 CTR1(KTR_NTB, "RX: process_rxc rx_index = %u", qp->rx_index);
910 entry = ntb_list_rm(&qp->ntb_rx_pend_q_lock, &qp->rx_pend_q);
913 CTR0(KTR_NTB, "RX: No entries in rx_pend_q");
916 callout_stop(&qp->rx_full);
917 CTR1(KTR_NTB, "RX: rx entry %p from rx_pend_q", entry);
919 if ((hdr->flags & IF_NTB_DESC_DONE_FLAG) == 0) {
921 "RX: hdr not done. Returning entry %p to rx_pend_q", entry);
922 ntb_list_add(&qp->ntb_rx_pend_q_lock, entry, &qp->rx_pend_q);
927 if (hdr->ver != (uint32_t) qp->rx_pkts) {
928 CTR3(KTR_NTB,"RX: ver != rx_pkts (%x != %lx). "
929 "Returning entry %p to rx_pend_q", hdr->ver, qp->rx_pkts,
931 ntb_list_add(&qp->ntb_rx_pend_q_lock, entry, &qp->rx_pend_q);
936 if ((hdr->flags & IF_NTB_LINK_DOWN_FLAG) != 0) {
937 ntb_qp_link_down(qp);
939 "RX: link down. adding entry %p back to rx_pend_q", entry);
940 ntb_list_add(&qp->ntb_rx_pend_q_lock, entry, &qp->rx_pend_q);
944 if (hdr->len <= entry->len) {
945 entry->len = hdr->len;
946 ntb_rx_copy_task(qp, entry, offset);
949 "RX: len too long. Returning entry %p to rx_pend_q", entry);
950 ntb_list_add(&qp->ntb_rx_pend_q_lock, entry, &qp->rx_pend_q);
955 qp->rx_bytes += hdr->len;
957 CTR1(KTR_NTB, "RX: received %ld rx_pkts", qp->rx_pkts);
961 /* Ensure that the data is globally visible before clearing the flag */
964 /* TODO: replace with bus_space_write */
965 qp->rx_info->entry = qp->rx_index;
968 qp->rx_index %= qp->rx_max_entry;
974 ntb_rx_copy_task(struct ntb_transport_qp *qp, struct ntb_queue_entry *entry,
977 struct ifnet *ifp = entry->cb_data;
978 unsigned int len = entry->len;
981 CTR2(KTR_NTB, "RX: copying %d bytes from offset %p", len, offset);
982 m = m_devget(offset, len, 0, ifp, NULL);
983 m->m_pkthdr.csum_flags = CSUM_IP_CHECKED | CSUM_IP_VALID;
985 entry->buf = (void *)m;
988 "RX: copied entry %p to mbuf %p. Adding entry to rx_free_q", entry,
990 ntb_list_add(&qp->ntb_rx_free_q_lock, entry, &qp->rx_free_q);
992 taskqueue_enqueue(taskqueue_swi, &qp->rx_completion_task);
996 ntb_rx_completion_task(void *arg, int pending)
998 struct ntb_transport_qp *qp = arg;
1000 struct ntb_queue_entry *entry;
1002 CTR0(KTR_NTB, "RX: rx_completion_task");
1004 while ((entry = ntb_list_rm(&qp->ntb_rx_free_q_lock, &qp->rx_free_q))) {
1006 CTR2(KTR_NTB, "RX: completing entry %p, mbuf %p", entry, m);
1007 if (qp->rx_handler && qp->client_ready == NTB_LINK_UP)
1008 qp->rx_handler(qp, qp->cb_data, m, entry->len);
1011 entry->len = qp->transport->bufsize;
1013 CTR1(KTR_NTB,"RX: entry %p removed from rx_free_q "
1014 "and added to rx_pend_q", entry);
1015 ntb_list_add(&qp->ntb_rx_pend_q_lock, entry, &qp->rx_pend_q);
1016 if (qp->rx_err_no_buf > qp->last_rx_no_buf) {
1017 qp->last_rx_no_buf = qp->rx_err_no_buf;
1018 CTR0(KTR_NTB, "RX: could spawn rx task");
1019 callout_reset(&qp->rx_full, hz / 1000, ntb_rx_pendq_full,
1025 /* Link Event handler */
1027 ntb_transport_event_callback(void *data, enum ntb_hw_event event)
1029 struct ntb_netdev *nt = data;
1032 case NTB_EVENT_HW_LINK_UP:
1034 device_printf(ntb_get_device(nt->ntb), "HW link up\n");
1035 callout_reset(&nt->link_work, 0, ntb_transport_link_work, nt);
1037 case NTB_EVENT_HW_LINK_DOWN:
1039 device_printf(ntb_get_device(nt->ntb), "HW link down\n");
1040 ntb_transport_link_cleanup(nt);
1043 panic("ntb: Unknown NTB event");
1049 ntb_transport_link_work(void *arg)
1051 struct ntb_netdev *nt = arg;
1052 struct ntb_softc *ntb = nt->ntb;
1053 struct ntb_transport_qp *qp;
1055 uint32_t val, i, num_mw;
1058 num_mw = ntb_get_max_mw(ntb);
1060 /* send the local info, in the opposite order of the way we read it */
1061 for (i = 0; i < num_mw; i++) {
1062 rc = ntb_write_remote_spad(ntb, IF_NTB_MW0_SZ_HIGH + (i * 2),
1063 (uint64_t)ntb_get_mw_size(ntb, i) >> 32);
1067 rc = ntb_write_remote_spad(ntb, IF_NTB_MW0_SZ_LOW + (i * 2),
1068 (uint32_t)ntb_get_mw_size(ntb, i));
1073 rc = ntb_write_remote_spad(ntb, IF_NTB_NUM_MWS, num_mw);
1077 rc = ntb_write_remote_spad(ntb, IF_NTB_NUM_QPS, nt->max_qps);
1081 rc = ntb_write_remote_spad(ntb, IF_NTB_VERSION, NTB_TRANSPORT_VERSION);
1085 /* Query the remote side for its info */
1086 rc = ntb_read_local_spad(ntb, IF_NTB_VERSION, &val);
1090 if (val != NTB_TRANSPORT_VERSION)
1093 rc = ntb_read_local_spad(ntb, IF_NTB_NUM_QPS, &val);
1097 if (val != nt->max_qps)
1100 rc = ntb_read_local_spad(ntb, IF_NTB_NUM_MWS, &val);
1107 for (i = 0; i < num_mw; i++) {
1108 rc = ntb_read_local_spad(ntb, IF_NTB_MW0_SZ_HIGH + (i * 2),
1113 val64 = (uint64_t)val << 32;
1115 rc = ntb_read_local_spad(ntb, IF_NTB_MW0_SZ_LOW + (i * 2),
1122 rc = ntb_set_mw(nt, i, val64);
1127 nt->transport_link = NTB_LINK_UP;
1129 device_printf(ntb_get_device(ntb), "transport link up\n");
1131 for (i = 0; i < nt->max_qps; i++) {
1134 ntb_transport_setup_qp_mw(nt, i);
1136 if (qp->client_ready == NTB_LINK_UP)
1137 callout_reset(&qp->link_work, 0, ntb_qp_link_work, qp);
1143 for (i = 0; i < NTB_MAX_NUM_MW; i++)
1146 if (ntb_query_link_status(ntb))
1147 callout_reset(&nt->link_work,
1148 NTB_LINK_DOWN_TIMEOUT * hz / 1000, ntb_transport_link_work, nt);
1152 ntb_set_mw(struct ntb_netdev *nt, int num_mw, unsigned int size)
1154 struct ntb_transport_mw *mw = &nt->mw[num_mw];
1156 /* No need to re-setup */
1157 if (mw->size == size)
1161 ntb_free_mw(nt, num_mw);
1163 /* Alloc memory for receiving data. Must be 4k aligned */
1166 mw->virt_addr = contigmalloc(mw->size, M_NTB_IF, M_ZERO, 0,
1167 BUS_SPACE_MAXADDR, mw->size, 0);
1168 if (mw->virt_addr == NULL) {
1170 printf("ntb: Unable to allocate MW buffer of size %zu\n",
1174 /* TODO: replace with bus_space_* functions */
1175 mw->dma_addr = vtophys(mw->virt_addr);
1178 * Ensure that the allocation from contigmalloc is aligned as
1179 * requested. XXX: This may not be needed -- brought in for parity
1180 * with the Linux driver.
1182 if (mw->dma_addr % size != 0) {
1183 device_printf(ntb_get_device(nt->ntb),
1184 "DMA memory 0x%jx not aligned to BAR size 0x%x\n",
1185 (uintmax_t)mw->dma_addr, size);
1186 ntb_free_mw(nt, num_mw);
1190 /* Notify HW the memory location of the receive buffer */
1191 ntb_set_mw_addr(nt->ntb, num_mw, mw->dma_addr);
1197 ntb_free_mw(struct ntb_netdev *nt, int num_mw)
1199 struct ntb_transport_mw *mw = &nt->mw[num_mw];
1201 if (mw->virt_addr == NULL)
1204 contigfree(mw->virt_addr, mw->size, M_NTB_IF);
1205 mw->virt_addr = NULL;
1209 ntb_transport_setup_qp_mw(struct ntb_netdev *nt, unsigned int qp_num)
1211 struct ntb_transport_qp *qp = &nt->qps[qp_num];
1213 unsigned int rx_size, num_qps_mw;
1214 uint8_t mw_num, mw_max;
1217 mw_max = ntb_get_max_mw(nt->ntb);
1218 mw_num = QP_TO_MW(nt->ntb, qp_num);
1220 if (nt->max_qps % mw_max && mw_num + 1 < nt->max_qps / mw_max)
1221 num_qps_mw = nt->max_qps / mw_max + 1;
1223 num_qps_mw = nt->max_qps / mw_max;
1225 rx_size = (unsigned int) nt->mw[mw_num].size / num_qps_mw;
1226 qp->remote_rx_info = (void *)((uint8_t *)nt->mw[mw_num].virt_addr +
1227 (qp_num / mw_max * rx_size));
1228 rx_size -= sizeof(struct ntb_rx_info);
1230 qp->rx_buff = qp->remote_rx_info + 1;
1231 /* Due to house-keeping, there must be at least 2 buffs */
1232 qp->rx_max_frame = min(transport_mtu + sizeof(struct ntb_payload_header),
1234 qp->rx_max_entry = rx_size / qp->rx_max_frame;
1237 qp->remote_rx_info->entry = qp->rx_max_entry - 1;
1239 /* setup the hdr offsets with 0's */
1240 for (i = 0; i < qp->rx_max_entry; i++) {
1241 offset = (void *)((uint8_t *)qp->rx_buff +
1242 qp->rx_max_frame * (i + 1) -
1243 sizeof(struct ntb_payload_header));
1244 memset(offset, 0, sizeof(struct ntb_payload_header));
1253 ntb_qp_link_work(void *arg)
1255 struct ntb_transport_qp *qp = arg;
1256 struct ntb_softc *ntb = qp->ntb;
1257 struct ntb_netdev *nt = qp->transport;
1261 rc = ntb_read_remote_spad(ntb, IF_NTB_QP_LINKS, &val);
1265 rc = ntb_write_remote_spad(ntb, IF_NTB_QP_LINKS, val | 1 << qp->qp_num);
1267 /* query remote spad for qp ready bits */
1268 rc = ntb_read_local_spad(ntb, IF_NTB_QP_LINKS, &val);
1270 /* See if the remote side is up */
1271 if ((1 << qp->qp_num & val) != 0) {
1272 qp->qp_link = NTB_LINK_UP;
1273 if (qp->event_handler != NULL)
1274 qp->event_handler(qp->cb_data, NTB_LINK_UP);
1276 device_printf(ntb_get_device(ntb), "qp link up\n");
1277 } else if (nt->transport_link == NTB_LINK_UP) {
1278 callout_reset(&qp->link_work,
1279 NTB_LINK_DOWN_TIMEOUT * hz / 1000, ntb_qp_link_work, qp);
1283 /* Link down event*/
1285 ntb_transport_link_cleanup(struct ntb_netdev *nt)
1289 /* Pass along the info to any clients */
1290 for (i = 0; i < nt->max_qps; i++)
1291 if (!test_bit(i, &nt->qp_bitmap))
1292 ntb_qp_link_down(&nt->qps[i]);
1294 if (nt->transport_link == NTB_LINK_DOWN)
1295 callout_drain(&nt->link_work);
1297 nt->transport_link = NTB_LINK_DOWN;
1300 * The scratchpad registers keep the values if the remote side
1301 * goes down, blast them now to give them a sane value the next
1302 * time they are accessed
1304 for (i = 0; i < IF_NTB_MAX_SPAD; i++)
1305 ntb_write_local_spad(nt->ntb, i, 0);
1310 ntb_qp_link_down(struct ntb_transport_qp *qp)
1313 ntb_qp_link_cleanup(qp);
1317 ntb_qp_link_cleanup(struct ntb_transport_qp *qp)
1319 struct ntb_netdev *nt = qp->transport;
1321 if (qp->qp_link == NTB_LINK_DOWN) {
1322 callout_drain(&qp->link_work);
1326 if (qp->event_handler != NULL)
1327 qp->event_handler(qp->cb_data, NTB_LINK_DOWN);
1329 qp->qp_link = NTB_LINK_DOWN;
1331 if (nt->transport_link == NTB_LINK_UP)
1332 callout_reset(&qp->link_work,
1333 NTB_LINK_DOWN_TIMEOUT * hz / 1000, ntb_qp_link_work, qp);
1336 /* Link commanded down */
1338 * ntb_transport_link_down - Notify NTB transport to no longer enqueue data
1339 * @qp: NTB transport layer queue to be disabled
1341 * Notify NTB transport layer of client's desire to no longer receive data on
1342 * transport queue specified. It is the client's responsibility to ensure all
1343 * entries on queue are purged or otherwise handled appropriately.
1346 ntb_transport_link_down(struct ntb_transport_qp *qp)
1353 qp->client_ready = NTB_LINK_DOWN;
1355 rc = ntb_read_remote_spad(qp->ntb, IF_NTB_QP_LINKS, &val);
1359 rc = ntb_write_remote_spad(qp->ntb, IF_NTB_QP_LINKS,
1360 val & ~(1 << qp->qp_num));
1362 if (qp->qp_link == NTB_LINK_UP)
1363 ntb_send_link_down(qp);
1365 callout_drain(&qp->link_work);
1370 ntb_send_link_down(struct ntb_transport_qp *qp)
1372 struct ntb_queue_entry *entry;
1375 if (qp->qp_link == NTB_LINK_DOWN)
1378 qp->qp_link = NTB_LINK_DOWN;
1380 for (i = 0; i < NTB_LINK_DOWN_TIMEOUT; i++) {
1381 entry = ntb_list_rm(&qp->ntb_tx_free_q_lock, &qp->tx_free_q);
1384 pause("NTB Wait for link down", hz / 10);
1390 entry->cb_data = NULL;
1393 entry->flags = IF_NTB_LINK_DOWN_FLAG;
1395 mtx_lock(&qp->transport->tx_lock);
1396 rc = ntb_process_tx(qp, entry);
1398 printf("ntb: Failed to send link down\n");
1399 mtx_unlock(&qp->transport->tx_lock);
1403 /* List Management */
1406 ntb_list_add(struct mtx *lock, struct ntb_queue_entry *entry,
1407 struct ntb_queue_list *list)
1410 mtx_lock_spin(lock);
1411 STAILQ_INSERT_TAIL(list, entry, entry);
1412 mtx_unlock_spin(lock);
1415 static struct ntb_queue_entry *
1416 ntb_list_rm(struct mtx *lock, struct ntb_queue_list *list)
1418 struct ntb_queue_entry *entry;
1420 mtx_lock_spin(lock);
1421 if (STAILQ_EMPTY(list)) {
1425 entry = STAILQ_FIRST(list);
1426 STAILQ_REMOVE_HEAD(list, entry);
1428 mtx_unlock_spin(lock);
1433 /* Helper functions */
1434 /* TODO: This too should really be part of the kernel */
1435 #define EUI48_MULTICAST 1 << 0
1436 #define EUI48_LOCALLY_ADMINISTERED 1 << 1
1438 create_random_local_eui48(u_char *eaddr)
1440 static uint8_t counter = 0;
1441 uint32_t seed = ticks;
1443 eaddr[0] = EUI48_LOCALLY_ADMINISTERED;
1444 memcpy(&eaddr[1], &seed, sizeof(uint32_t));
1445 eaddr[5] = counter++;
1449 * ntb_transport_max_size - Query the max payload size of a qp
1450 * @qp: NTB transport layer queue to be queried
1452 * Query the maximum payload size permissible on the given qp
1454 * RETURNS: the max payload size of a qp
1457 ntb_transport_max_size(struct ntb_transport_qp *qp)
1463 return (qp->tx_max_frame - sizeof(struct ntb_payload_header));