ntb_transport.c revision 289273
1/*-
2 * Copyright (C) 2013 Intel Corporation
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
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.
13 *
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
24 * SUCH DAMAGE.
25 */
26
27#include <sys/cdefs.h>
28__FBSDID("$FreeBSD: head/sys/dev/ntb/if_ntb/if_ntb.c 289273 2015-10-13 23:42:13Z cem $");
29
30#include <sys/param.h>
31#include <sys/kernel.h>
32#include <sys/systm.h>
33#include <sys/bus.h>
34#include <sys/ktr.h>
35#include <sys/lock.h>
36#include <sys/malloc.h>
37#include <sys/module.h>
38#include <sys/mutex.h>
39#include <sys/queue.h>
40#include <sys/socket.h>
41#include <sys/sockio.h>
42#include <sys/sysctl.h>
43#include <sys/taskqueue.h>
44#include <net/if.h>
45#include <net/if_media.h>
46#include <net/if_types.h>
47#include <net/if_var.h>
48#include <net/bpf.h>
49#include <net/ethernet.h>
50#include <vm/vm.h>
51#include <vm/pmap.h>
52#include <machine/bus.h>
53#include <machine/cpufunc.h>
54#include <machine/pmap.h>
55
56#include "../ntb_hw/ntb_hw.h"
57
58/*
59 * The Non-Transparent Bridge (NTB) is a device on some Intel processors that
60 * allows you to connect two systems using a PCI-e link.
61 *
62 * This module contains a protocol for sending and receiving messages, and
63 * exposes that protocol through a simulated ethernet device called ntb.
64 *
65 * NOTE: Much of the code in this module is shared with Linux. Any patches may
66 * be picked up and redistributed in Linux with a dual GPL/BSD license.
67 */
68
69/* TODO: These functions should really be part of the kernel */
70#define test_bit(pos, bitmap_addr)  (*(bitmap_addr) & 1UL << (pos))
71#define set_bit(pos, bitmap_addr)   *(bitmap_addr) |= 1UL << (pos)
72#define clear_bit(pos, bitmap_addr) *(bitmap_addr) &= ~(1UL << (pos))
73
74#define KTR_NTB KTR_SPARE3
75
76#define NTB_TRANSPORT_VERSION	3
77#define NTB_RX_MAX_PKTS		64
78#define	NTB_RXQ_SIZE		300
79
80static unsigned int transport_mtu = 0x4000 + ETHER_HDR_LEN + ETHER_CRC_LEN;
81
82/*
83 * This is an oversimplification to work around Xeon Errata.  The second client
84 * may be usable for unidirectional traffic.
85 */
86static unsigned int max_num_clients = 1;
87
88STAILQ_HEAD(ntb_queue_list, ntb_queue_entry);
89
90struct ntb_queue_entry {
91	/* ntb_queue list reference */
92	STAILQ_ENTRY(ntb_queue_entry) entry;
93
94	/* info on data to be transfered */
95	void		*cb_data;
96	void		*buf;
97	uint64_t	len;
98	uint64_t	flags;
99};
100
101struct ntb_rx_info {
102	unsigned int entry;
103};
104
105struct ntb_transport_qp {
106	struct ntb_netdev	*transport;
107	struct ntb_softc	*ntb;
108
109	void			*cb_data;
110
111	bool			client_ready;
112	bool			qp_link;
113	uint8_t			qp_num;	/* Only 64 QPs are allowed.  0-63 */
114
115	struct ntb_rx_info	*rx_info;
116	struct ntb_rx_info	*remote_rx_info;
117
118	void (*tx_handler) (struct ntb_transport_qp *qp, void *qp_data,
119	    void *data, int len);
120	struct ntb_queue_list	tx_free_q;
121	struct mtx		ntb_tx_free_q_lock;
122	void			*tx_mw;
123	uint64_t		tx_index;
124	uint64_t		tx_max_entry;
125	uint64_t		tx_max_frame;
126
127	void (*rx_handler) (struct ntb_transport_qp *qp, void *qp_data,
128	    void *data, int len);
129	struct ntb_queue_list	rx_pend_q;
130	struct ntb_queue_list	rx_free_q;
131	struct mtx		ntb_rx_pend_q_lock;
132	struct mtx		ntb_rx_free_q_lock;
133	struct task		rx_completion_task;
134	void			*rx_buff;
135	uint64_t		rx_index;
136	uint64_t		rx_max_entry;
137	uint64_t		rx_max_frame;
138
139	void (*event_handler) (void *data, int status);
140	struct callout		link_work;
141	struct callout		queue_full;
142	struct callout		rx_full;
143
144	uint64_t		last_rx_no_buf;
145
146	/* Stats */
147	uint64_t		rx_bytes;
148	uint64_t		rx_pkts;
149	uint64_t		rx_ring_empty;
150	uint64_t		rx_err_no_buf;
151	uint64_t		rx_err_oflow;
152	uint64_t		rx_err_ver;
153	uint64_t		tx_bytes;
154	uint64_t		tx_pkts;
155	uint64_t		tx_ring_full;
156};
157
158struct ntb_queue_handlers {
159	void (*rx_handler) (struct ntb_transport_qp *qp, void *qp_data,
160	    void *data, int len);
161	void (*tx_handler) (struct ntb_transport_qp *qp, void *qp_data,
162	    void *data, int len);
163	void (*event_handler) (void *data, int status);
164};
165
166
167struct ntb_transport_mw {
168	size_t		size;
169	void		*virt_addr;
170	vm_paddr_t	dma_addr;
171};
172
173struct ntb_netdev {
174	struct ntb_softc	*ntb;
175	struct ifnet		*ifp;
176	struct ntb_transport_mw	mw[NTB_NUM_MW];
177	struct ntb_transport_qp	*qps;
178	uint64_t		max_qps;
179	uint64_t		qp_bitmap;
180	bool			transport_link;
181	struct callout		link_work;
182	struct ntb_transport_qp *qp;
183	uint64_t		bufsize;
184	u_char			eaddr[ETHER_ADDR_LEN];
185	struct mtx		tx_lock;
186	struct mtx		rx_lock;
187};
188
189static struct ntb_netdev net_softc;
190
191enum {
192	IF_NTB_DESC_DONE_FLAG = 1 << 0,
193	IF_NTB_LINK_DOWN_FLAG = 1 << 1,
194};
195
196struct ntb_payload_header {
197	uint64_t ver;
198	uint64_t len;
199	uint64_t flags;
200};
201
202enum {
203	/*
204	 * The order of this enum is part of the if_ntb remote protocol.  Do
205	 * not reorder without bumping protocol version (and it's probably best
206	 * to keep the protocol in lock-step with the Linux NTB driver.
207	 */
208	IF_NTB_VERSION = 0,
209	IF_NTB_QP_LINKS,
210	IF_NTB_NUM_QPS,
211	IF_NTB_NUM_MWS,
212	/*
213	 * N.B.: transport_link_work assumes MW1 enums = MW0 + 2.
214	 */
215	IF_NTB_MW0_SZ_HIGH,
216	IF_NTB_MW0_SZ_LOW,
217	IF_NTB_MW1_SZ_HIGH,
218	IF_NTB_MW1_SZ_LOW,
219	IF_NTB_MAX_SPAD,
220};
221
222#define QP_TO_MW(qp)		((qp) % NTB_NUM_MW)
223#define NTB_QP_DEF_NUM_ENTRIES	100
224#define NTB_LINK_DOWN_TIMEOUT	10
225
226static int ntb_handle_module_events(struct module *m, int what, void *arg);
227static int ntb_setup_interface(void);
228static int ntb_teardown_interface(void);
229static void ntb_net_init(void *arg);
230static int ntb_ioctl(struct ifnet *ifp, u_long command, caddr_t data);
231static void ntb_start(struct ifnet *ifp);
232static void ntb_net_tx_handler(struct ntb_transport_qp *qp, void *qp_data,
233    void *data, int len);
234static void ntb_net_rx_handler(struct ntb_transport_qp *qp, void *qp_data,
235    void *data, int len);
236static void ntb_net_event_handler(void *data, int status);
237static int ntb_transport_init(struct ntb_softc *ntb);
238static void ntb_transport_free(void *transport);
239static void ntb_transport_init_queue(struct ntb_netdev *nt,
240    unsigned int qp_num);
241static void ntb_transport_free_queue(struct ntb_transport_qp *qp);
242static struct ntb_transport_qp * ntb_transport_create_queue(void *data,
243    struct ntb_softc *pdev, const struct ntb_queue_handlers *handlers);
244static void ntb_transport_link_up(struct ntb_transport_qp *qp);
245static int ntb_transport_tx_enqueue(struct ntb_transport_qp *qp, void *cb,
246    void *data, unsigned int len);
247static int ntb_process_tx(struct ntb_transport_qp *qp,
248    struct ntb_queue_entry *entry);
249static void ntb_tx_copy_task(struct ntb_transport_qp *qp,
250    struct ntb_queue_entry *entry, void *offset);
251static void ntb_qp_full(void *arg);
252static void ntb_transport_rxc_db(void *data, int db_num);
253static void ntb_rx_pendq_full(void *arg);
254static void ntb_transport_rx(struct ntb_transport_qp *qp);
255static int ntb_process_rxc(struct ntb_transport_qp *qp);
256static void ntb_rx_copy_task(struct ntb_transport_qp *qp,
257    struct ntb_queue_entry *entry, void *offset);
258static void ntb_rx_completion_task(void *arg, int pending);
259static void ntb_transport_event_callback(void *data, enum ntb_hw_event event);
260static void ntb_transport_link_work(void *arg);
261static int ntb_set_mw(struct ntb_netdev *nt, int num_mw, unsigned int size);
262static void ntb_free_mw(struct ntb_netdev *nt, int num_mw);
263static void ntb_transport_setup_qp_mw(struct ntb_netdev *nt,
264    unsigned int qp_num);
265static void ntb_qp_link_work(void *arg);
266static void ntb_transport_link_cleanup(struct ntb_netdev *nt);
267static void ntb_qp_link_down(struct ntb_transport_qp *qp);
268static void ntb_qp_link_cleanup(struct ntb_transport_qp *qp);
269static void ntb_transport_link_down(struct ntb_transport_qp *qp);
270static void ntb_send_link_down(struct ntb_transport_qp *qp);
271static void ntb_list_add(struct mtx *lock, struct ntb_queue_entry *entry,
272    struct ntb_queue_list *list);
273static struct ntb_queue_entry *ntb_list_rm(struct mtx *lock,
274    struct ntb_queue_list *list);
275static void create_random_local_eui48(u_char *eaddr);
276static unsigned int ntb_transport_max_size(struct ntb_transport_qp *qp);
277
278MALLOC_DEFINE(M_NTB_IF, "if_ntb", "ntb network driver");
279
280/* Module setup and teardown */
281static int
282ntb_handle_module_events(struct module *m, int what, void *arg)
283{
284	int err = 0;
285
286	switch (what) {
287	case MOD_LOAD:
288		err = ntb_setup_interface();
289		break;
290	case MOD_UNLOAD:
291		err = ntb_teardown_interface();
292		break;
293	default:
294		err = EOPNOTSUPP;
295		break;
296	}
297	return (err);
298}
299
300static moduledata_t if_ntb_mod = {
301	"if_ntb",
302	ntb_handle_module_events,
303	NULL
304};
305
306DECLARE_MODULE(if_ntb, if_ntb_mod, SI_SUB_KLD, SI_ORDER_ANY);
307MODULE_DEPEND(if_ntb, ntb_hw, 1, 1, 1);
308
309static int
310ntb_setup_interface(void)
311{
312	struct ifnet *ifp;
313	struct ntb_queue_handlers handlers = { ntb_net_rx_handler,
314	    ntb_net_tx_handler, ntb_net_event_handler };
315
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");
319		return (ENXIO);
320	}
321
322	ntb_transport_init(net_softc.ntb);
323
324	ifp = net_softc.ifp = if_alloc(IFT_ETHER);
325	if (ifp == NULL) {
326		printf("ntb: cannot allocate ifnet structure\n");
327		return (ENOMEM);
328	}
329
330	net_softc.qp = ntb_transport_create_queue(ifp, net_softc.ntb,
331	    &handlers);
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;
345
346	ntb_transport_link_up(net_softc.qp);
347	net_softc.bufsize = ntb_transport_max_size(net_softc.qp) +
348	    sizeof(struct ether_header);
349	return (0);
350}
351
352static int
353ntb_teardown_interface(void)
354{
355
356	if (net_softc.qp != NULL)
357		ntb_transport_link_down(net_softc.qp);
358
359	if (net_softc.ifp != NULL) {
360		ether_ifdetach(net_softc.ifp);
361		if_free(net_softc.ifp);
362	}
363
364	if (net_softc.qp != NULL) {
365		ntb_transport_free_queue(net_softc.qp);
366		ntb_transport_free(&net_softc);
367	}
368
369	return (0);
370}
371
372/* Network device interface */
373
374static void
375ntb_net_init(void *arg)
376{
377	struct ntb_netdev *ntb_softc = arg;
378	struct ifnet *ifp = ntb_softc->ifp;
379
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);
384}
385
386static int
387ntb_ioctl(struct ifnet *ifp, u_long command, caddr_t data)
388{
389	struct ntb_netdev *nt = ifp->if_softc;
390	struct ifreq *ifr = (struct ifreq *)data;
391	int error = 0;
392
393	switch (command) {
394	case SIOCSIFMTU:
395	    {
396		if (ifr->ifr_mtu > ntb_transport_max_size(nt->qp) -
397		    ETHER_HDR_LEN - ETHER_CRC_LEN) {
398			error = EINVAL;
399			break;
400		}
401
402		ifp->if_mtu = ifr->ifr_mtu;
403		break;
404	    }
405	default:
406		error = ether_ioctl(ifp, command, data);
407		break;
408	}
409
410	return (error);
411}
412
413
414static void
415ntb_start(struct ifnet *ifp)
416{
417	struct mbuf *m_head;
418	struct ntb_netdev *nt = ifp->if_softc;
419	int rc;
420
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));
429		if (rc != 0) {
430			CTR1(KTR_NTB,
431			    "TX: could not tx mbuf %p. Returning to snd q",
432			    m_head);
433			if (rc == EAGAIN) {
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,
437				    ntb_qp_full, ifp);
438			}
439			break;
440		}
441
442	}
443	mtx_unlock(&nt->tx_lock);
444}
445
446/* Network Device Callbacks */
447static void
448ntb_net_tx_handler(struct ntb_transport_qp *qp, void *qp_data, void *data,
449    int len)
450{
451
452	m_freem(data);
453	CTR1(KTR_NTB, "TX: tx_handler freeing mbuf %p", data);
454}
455
456static void
457ntb_net_rx_handler(struct ntb_transport_qp *qp, void *qp_data, void *data,
458    int len)
459{
460	struct mbuf *m = data;
461	struct ifnet *ifp = qp_data;
462
463	CTR0(KTR_NTB, "RX: rx handler");
464	(*ifp->if_input)(ifp, m);
465}
466
467static void
468ntb_net_event_handler(void *data, int status)
469{
470
471}
472
473/* Transport Init and teardown */
474
475static int
476ntb_transport_init(struct ntb_softc *ntb)
477{
478	struct ntb_netdev *nt = &net_softc;
479	int rc, i;
480
481	nt->max_qps = max_num_clients;
482	ntb_register_transport(ntb, nt);
483	mtx_init(&nt->tx_lock, "ntb transport tx", NULL, MTX_DEF);
484	mtx_init(&nt->rx_lock, "ntb transport rx", NULL, MTX_DEF);
485
486	nt->qps = malloc(nt->max_qps * sizeof(struct ntb_transport_qp),
487			  M_NTB_IF, M_WAITOK|M_ZERO);
488
489	nt->qp_bitmap = ((uint64_t) 1 << nt->max_qps) - 1;
490
491	for (i = 0; i < nt->max_qps; i++)
492		ntb_transport_init_queue(nt, i);
493
494	callout_init(&nt->link_work, 0);
495
496	rc = ntb_register_event_callback(ntb,
497					 ntb_transport_event_callback);
498	if (rc != 0)
499		goto err;
500
501	if (ntb_query_link_status(ntb)) {
502		if (bootverbose)
503			device_printf(ntb_get_device(ntb), "link up\n");
504		callout_reset(&nt->link_work, 0, ntb_transport_link_work, nt);
505	}
506
507	return (0);
508
509err:
510	free(nt->qps, M_NTB_IF);
511	ntb_unregister_transport(ntb);
512	return (rc);
513}
514
515static void
516ntb_transport_free(void *transport)
517{
518	struct ntb_netdev *nt = transport;
519	struct ntb_softc *ntb = nt->ntb;
520	int i;
521
522	ntb_transport_link_cleanup(nt);
523
524	callout_drain(&nt->link_work);
525
526	/* verify that all the qps are freed */
527	for (i = 0; i < nt->max_qps; i++)
528		if (!test_bit(i, &nt->qp_bitmap))
529			ntb_transport_free_queue(&nt->qps[i]);
530
531	ntb_unregister_event_callback(ntb);
532
533	for (i = 0; i < NTB_NUM_MW; i++)
534		ntb_free_mw(nt, i);
535
536	free(nt->qps, M_NTB_IF);
537	ntb_unregister_transport(ntb);
538}
539
540static void
541ntb_transport_init_queue(struct ntb_netdev *nt, unsigned int qp_num)
542{
543	struct ntb_transport_qp *qp;
544	unsigned int num_qps_mw, tx_size;
545	uint8_t mw_num = QP_TO_MW(qp_num);
546
547	qp = &nt->qps[qp_num];
548	qp->qp_num = qp_num;
549	qp->transport = nt;
550	qp->ntb = nt->ntb;
551	qp->qp_link = NTB_LINK_DOWN;
552	qp->client_ready = NTB_LINK_DOWN;
553	qp->event_handler = NULL;
554
555	if (nt->max_qps % NTB_NUM_MW && mw_num < nt->max_qps % NTB_NUM_MW)
556		num_qps_mw = nt->max_qps / NTB_NUM_MW + 1;
557	else
558		num_qps_mw = nt->max_qps / NTB_NUM_MW;
559
560	tx_size = (unsigned int) ntb_get_mw_size(qp->ntb, mw_num) / num_qps_mw;
561	qp->rx_info = (struct ntb_rx_info *)
562	    ((char *)ntb_get_mw_vbase(qp->ntb, mw_num) +
563	    (qp_num / NTB_NUM_MW * tx_size));
564	tx_size -= sizeof(struct ntb_rx_info);
565
566	qp->tx_mw = qp->rx_info + 1;
567	/* Due to house-keeping, there must be at least 2 buffs */
568	qp->tx_max_frame = min(transport_mtu + sizeof(struct ntb_payload_header),
569	    tx_size / 2);
570	qp->tx_max_entry = tx_size / qp->tx_max_frame;
571
572	callout_init(&qp->link_work, 0);
573	callout_init(&qp->queue_full, 1);
574	callout_init(&qp->rx_full, 1);
575
576	mtx_init(&qp->ntb_rx_pend_q_lock, "ntb rx pend q", NULL, MTX_SPIN);
577	mtx_init(&qp->ntb_rx_free_q_lock, "ntb rx free q", NULL, MTX_SPIN);
578	mtx_init(&qp->ntb_tx_free_q_lock, "ntb tx free q", NULL, MTX_SPIN);
579	TASK_INIT(&qp->rx_completion_task, 0, ntb_rx_completion_task, qp);
580
581	STAILQ_INIT(&qp->rx_pend_q);
582	STAILQ_INIT(&qp->rx_free_q);
583	STAILQ_INIT(&qp->tx_free_q);
584}
585
586static void
587ntb_transport_free_queue(struct ntb_transport_qp *qp)
588{
589	struct ntb_queue_entry *entry;
590
591	if (qp == NULL)
592		return;
593
594	callout_drain(&qp->link_work);
595
596	ntb_unregister_db_callback(qp->ntb, qp->qp_num);
597
598	while ((entry = ntb_list_rm(&qp->ntb_rx_free_q_lock, &qp->rx_free_q)))
599		free(entry, M_NTB_IF);
600
601	while ((entry = ntb_list_rm(&qp->ntb_rx_pend_q_lock, &qp->rx_pend_q)))
602		free(entry, M_NTB_IF);
603
604	while ((entry = ntb_list_rm(&qp->ntb_tx_free_q_lock, &qp->tx_free_q)))
605		free(entry, M_NTB_IF);
606
607	set_bit(qp->qp_num, &qp->transport->qp_bitmap);
608}
609
610/**
611 * ntb_transport_create_queue - Create a new NTB transport layer queue
612 * @rx_handler: receive callback function
613 * @tx_handler: transmit callback function
614 * @event_handler: event callback function
615 *
616 * Create a new NTB transport layer queue and provide the queue with a callback
617 * routine for both transmit and receive.  The receive callback routine will be
618 * used to pass up data when the transport has received it on the queue.   The
619 * transmit callback routine will be called when the transport has completed the
620 * transmission of the data on the queue and the data is ready to be freed.
621 *
622 * RETURNS: pointer to newly created ntb_queue, NULL on error.
623 */
624static struct ntb_transport_qp *
625ntb_transport_create_queue(void *data, struct ntb_softc *pdev,
626    const struct ntb_queue_handlers *handlers)
627{
628	struct ntb_queue_entry *entry;
629	struct ntb_transport_qp *qp;
630	struct ntb_netdev *nt;
631	unsigned int free_queue;
632	int rc, i;
633
634	nt = ntb_find_transport(pdev);
635	if (nt == NULL)
636		goto err;
637
638	free_queue = ffs(nt->qp_bitmap);
639	if (free_queue == 0)
640		goto err;
641
642	/* decrement free_queue to make it zero based */
643	free_queue--;
644
645	clear_bit(free_queue, &nt->qp_bitmap);
646
647	qp = &nt->qps[free_queue];
648	qp->cb_data = data;
649	qp->rx_handler = handlers->rx_handler;
650	qp->tx_handler = handlers->tx_handler;
651	qp->event_handler = handlers->event_handler;
652
653	for (i = 0; i < NTB_QP_DEF_NUM_ENTRIES; i++) {
654		entry = malloc(sizeof(struct ntb_queue_entry), M_NTB_IF,
655		    M_WAITOK|M_ZERO);
656		entry->cb_data = nt->ifp;
657		entry->buf = NULL;
658		entry->len = transport_mtu;
659		ntb_list_add(&qp->ntb_rx_pend_q_lock, entry, &qp->rx_pend_q);
660	}
661
662	for (i = 0; i < NTB_QP_DEF_NUM_ENTRIES; i++) {
663		entry = malloc(sizeof(struct ntb_queue_entry), M_NTB_IF,
664		    M_WAITOK|M_ZERO);
665		ntb_list_add(&qp->ntb_tx_free_q_lock, entry, &qp->tx_free_q);
666	}
667
668	rc = ntb_register_db_callback(qp->ntb, free_queue, qp,
669				      ntb_transport_rxc_db);
670	if (rc != 0)
671		goto err1;
672
673	return (qp);
674
675err1:
676	while ((entry = ntb_list_rm(&qp->ntb_tx_free_q_lock, &qp->tx_free_q)))
677		free(entry, M_NTB_IF);
678	while ((entry = ntb_list_rm(&qp->ntb_rx_free_q_lock, &qp->rx_free_q)))
679		free(entry, M_NTB_IF);
680	set_bit(free_queue, &nt->qp_bitmap);
681err:
682	return (NULL);
683}
684
685/**
686 * ntb_transport_link_up - Notify NTB transport of client readiness to use queue
687 * @qp: NTB transport layer queue to be enabled
688 *
689 * Notify NTB transport layer of client readiness to use queue
690 */
691static void
692ntb_transport_link_up(struct ntb_transport_qp *qp)
693{
694
695	if (qp == NULL)
696		return;
697
698	qp->client_ready = NTB_LINK_UP;
699	if (bootverbose)
700		device_printf(ntb_get_device(qp->ntb), "qp client ready\n");
701
702	if (qp->transport->transport_link == NTB_LINK_UP)
703		callout_reset(&qp->link_work, 0, ntb_qp_link_work, qp);
704}
705
706
707
708/* Transport Tx */
709
710/**
711 * ntb_transport_tx_enqueue - Enqueue a new NTB queue entry
712 * @qp: NTB transport layer queue the entry is to be enqueued on
713 * @cb: per buffer pointer for callback function to use
714 * @data: pointer to data buffer that will be sent
715 * @len: length of the data buffer
716 *
717 * Enqueue a new transmit buffer onto the transport queue from which a NTB
718 * payload will be transmitted.  This assumes that a lock is being held to
719 * serialize access to the qp.
720 *
721 * RETURNS: An appropriate ERRNO error value on error, or zero for success.
722 */
723static int
724ntb_transport_tx_enqueue(struct ntb_transport_qp *qp, void *cb, void *data,
725    unsigned int len)
726{
727	struct ntb_queue_entry *entry;
728	int rc;
729
730	if (qp == NULL || qp->qp_link != NTB_LINK_UP || len == 0) {
731		CTR0(KTR_NTB, "TX: link not up");
732		return (EINVAL);
733	}
734
735	entry = ntb_list_rm(&qp->ntb_tx_free_q_lock, &qp->tx_free_q);
736	if (entry == NULL) {
737		CTR0(KTR_NTB, "TX: could not get entry from tx_free_q");
738		return (ENOMEM);
739	}
740	CTR1(KTR_NTB, "TX: got entry %p from tx_free_q", entry);
741
742	entry->cb_data = cb;
743	entry->buf = data;
744	entry->len = len;
745	entry->flags = 0;
746
747	rc = ntb_process_tx(qp, entry);
748	if (rc != 0) {
749		ntb_list_add(&qp->ntb_tx_free_q_lock, entry, &qp->tx_free_q);
750		CTR1(KTR_NTB,
751		    "TX: process_tx failed. Returning entry %p to tx_free_q",
752		    entry);
753	}
754	return (rc);
755}
756
757static int
758ntb_process_tx(struct ntb_transport_qp *qp, struct ntb_queue_entry *entry)
759{
760	void *offset;
761
762	offset = (char *)qp->tx_mw + qp->tx_max_frame * qp->tx_index;
763	CTR3(KTR_NTB,
764	    "TX: process_tx: tx_pkts=%u, tx_index=%u, remote entry=%u",
765	    qp->tx_pkts, qp->tx_index, qp->remote_rx_info->entry);
766	if (qp->tx_index == qp->remote_rx_info->entry) {
767		CTR0(KTR_NTB, "TX: ring full");
768		qp->tx_ring_full++;
769		return (EAGAIN);
770	}
771
772	if (entry->len > qp->tx_max_frame - sizeof(struct ntb_payload_header)) {
773		if (qp->tx_handler != NULL)
774			qp->tx_handler(qp, qp->cb_data, entry->buf,
775				       EIO);
776
777		ntb_list_add(&qp->ntb_tx_free_q_lock, entry, &qp->tx_free_q);
778		CTR1(KTR_NTB,
779		    "TX: frame too big. returning entry %p to tx_free_q",
780		    entry);
781		return (0);
782	}
783	CTR2(KTR_NTB, "TX: copying entry %p to offset %p", entry, offset);
784	ntb_tx_copy_task(qp, entry, offset);
785
786	qp->tx_index++;
787	qp->tx_index %= qp->tx_max_entry;
788
789	qp->tx_pkts++;
790
791	return (0);
792}
793
794static void
795ntb_tx_copy_task(struct ntb_transport_qp *qp, struct ntb_queue_entry *entry,
796    void *offset)
797{
798	struct ntb_payload_header *hdr;
799
800	CTR2(KTR_NTB, "TX: copying %d bytes to offset %p", entry->len, offset);
801	if (entry->buf != NULL)
802		m_copydata((struct mbuf *)entry->buf, 0, entry->len, offset);
803
804	hdr = (struct ntb_payload_header *)((char *)offset + qp->tx_max_frame -
805	    sizeof(struct ntb_payload_header));
806	hdr->len = entry->len; /* TODO: replace with bus_space_write */
807	hdr->ver = qp->tx_pkts; /* TODO: replace with bus_space_write */
808	wmb();
809	/* TODO: replace with bus_space_write */
810	hdr->flags = entry->flags | IF_NTB_DESC_DONE_FLAG;
811
812	ntb_ring_doorbell(qp->ntb, qp->qp_num);
813
814	/*
815	 * The entry length can only be zero if the packet is intended to be a
816	 * "link down" or similar.  Since no payload is being sent in these
817	 * cases, there is nothing to add to the completion queue.
818	 */
819	if (entry->len > 0) {
820		qp->tx_bytes += entry->len;
821
822		if (qp->tx_handler)
823			qp->tx_handler(qp, qp->cb_data, entry->cb_data,
824				       entry->len);
825	}
826
827	CTR2(KTR_NTB,
828	    "TX: entry %p sent. hdr->ver = %d, Returning to tx_free_q", entry,
829	    hdr->ver);
830	ntb_list_add(&qp->ntb_tx_free_q_lock, entry, &qp->tx_free_q);
831}
832
833static void
834ntb_qp_full(void *arg)
835{
836
837	CTR0(KTR_NTB, "TX: qp_full callout");
838	ntb_start(arg);
839}
840
841/* Transport Rx */
842static void
843ntb_transport_rxc_db(void *data, int db_num)
844{
845	struct ntb_transport_qp *qp = data;
846
847	ntb_transport_rx(qp);
848}
849
850static void
851ntb_rx_pendq_full(void *arg)
852{
853
854	CTR0(KTR_NTB, "RX: ntb_rx_pendq_full callout");
855	ntb_transport_rx(arg);
856}
857
858static void
859ntb_transport_rx(struct ntb_transport_qp *qp)
860{
861	uint64_t i;
862	int rc;
863
864	/*
865	 * Limit the number of packets processed in a single interrupt to
866	 * provide fairness to others
867	 */
868	mtx_lock(&qp->transport->rx_lock);
869	CTR0(KTR_NTB, "RX: transport_rx");
870	for (i = 0; i < qp->rx_max_entry; i++) {
871		rc = ntb_process_rxc(qp);
872		if (rc != 0) {
873			CTR0(KTR_NTB, "RX: process_rxc failed");
874			break;
875		}
876	}
877	mtx_unlock(&qp->transport->rx_lock);
878}
879
880static int
881ntb_process_rxc(struct ntb_transport_qp *qp)
882{
883	struct ntb_payload_header *hdr;
884	struct ntb_queue_entry *entry;
885	void *offset;
886
887	offset = (void *)
888	    ((char *)qp->rx_buff + qp->rx_max_frame * qp->rx_index);
889	hdr = (void *)
890	    ((char *)offset + qp->rx_max_frame -
891		sizeof(struct ntb_payload_header));
892
893	CTR1(KTR_NTB, "RX: process_rxc rx_index = %u", qp->rx_index);
894	entry = ntb_list_rm(&qp->ntb_rx_pend_q_lock, &qp->rx_pend_q);
895	if (entry == NULL) {
896		qp->rx_err_no_buf++;
897		CTR0(KTR_NTB, "RX: No entries in rx_pend_q");
898		return (ENOMEM);
899	}
900	callout_stop(&qp->rx_full);
901	CTR1(KTR_NTB, "RX: rx entry %p from rx_pend_q", entry);
902
903	if ((hdr->flags & IF_NTB_DESC_DONE_FLAG) == 0) {
904		CTR1(KTR_NTB,
905		    "RX: hdr not done. Returning entry %p to rx_pend_q", entry);
906		ntb_list_add(&qp->ntb_rx_pend_q_lock, entry, &qp->rx_pend_q);
907		qp->rx_ring_empty++;
908		return (EAGAIN);
909	}
910
911	if (hdr->ver != (uint32_t) qp->rx_pkts) {
912		CTR3(KTR_NTB,"RX: ver != rx_pkts (%x != %lx). "
913		    "Returning entry %p to rx_pend_q", hdr->ver, qp->rx_pkts,
914		    entry);
915		ntb_list_add(&qp->ntb_rx_pend_q_lock, entry, &qp->rx_pend_q);
916		qp->rx_err_ver++;
917		return (EIO);
918	}
919
920	if ((hdr->flags & IF_NTB_LINK_DOWN_FLAG) != 0) {
921		ntb_qp_link_down(qp);
922		CTR1(KTR_NTB,
923		    "RX: link down. adding entry %p back to rx_pend_q", entry);
924		ntb_list_add(&qp->ntb_rx_pend_q_lock, entry, &qp->rx_pend_q);
925		goto out;
926	}
927
928	if (hdr->len <= entry->len) {
929		entry->len = hdr->len;
930		ntb_rx_copy_task(qp, entry, offset);
931	} else {
932		CTR1(KTR_NTB,
933		    "RX: len too long. Returning entry %p to rx_pend_q", entry);
934		ntb_list_add(&qp->ntb_rx_pend_q_lock, entry, &qp->rx_pend_q);
935
936		qp->rx_err_oflow++;
937	}
938
939	qp->rx_bytes += hdr->len;
940	qp->rx_pkts++;
941	CTR1(KTR_NTB, "RX: received %ld rx_pkts", qp->rx_pkts);
942
943
944out:
945	/* Ensure that the data is globally visible before clearing the flag */
946	wmb();
947	hdr->flags = 0;
948	/* TODO: replace with bus_space_write */
949	qp->rx_info->entry = qp->rx_index;
950
951	qp->rx_index++;
952	qp->rx_index %= qp->rx_max_entry;
953
954	return (0);
955}
956
957static void
958ntb_rx_copy_task(struct ntb_transport_qp *qp, struct ntb_queue_entry *entry,
959    void *offset)
960{
961	struct ifnet *ifp = entry->cb_data;
962	unsigned int len = entry->len;
963	struct mbuf *m;
964
965	CTR2(KTR_NTB, "RX: copying %d bytes from offset %p", len, offset);
966	m = m_devget(offset, len, 0, ifp, NULL);
967	m->m_pkthdr.csum_flags = CSUM_IP_CHECKED | CSUM_IP_VALID;
968
969	entry->buf = (void *)m;
970
971	CTR2(KTR_NTB,
972	    "RX: copied entry %p to mbuf %p. Adding entry to rx_free_q", entry,
973	    m);
974	ntb_list_add(&qp->ntb_rx_free_q_lock, entry, &qp->rx_free_q);
975
976	taskqueue_enqueue(taskqueue_swi, &qp->rx_completion_task);
977}
978
979static void
980ntb_rx_completion_task(void *arg, int pending)
981{
982	struct ntb_transport_qp *qp = arg;
983	struct mbuf *m;
984	struct ntb_queue_entry *entry;
985
986	CTR0(KTR_NTB, "RX: rx_completion_task");
987
988	while ((entry = ntb_list_rm(&qp->ntb_rx_free_q_lock, &qp->rx_free_q))) {
989		m = entry->buf;
990		CTR2(KTR_NTB, "RX: completing entry %p, mbuf %p", entry, m);
991		if (qp->rx_handler && qp->client_ready == NTB_LINK_UP)
992			qp->rx_handler(qp, qp->cb_data, m, entry->len);
993
994		entry->buf = NULL;
995		entry->len = qp->transport->bufsize;
996
997		CTR1(KTR_NTB,"RX: entry %p removed from rx_free_q "
998		    "and added to rx_pend_q", entry);
999		ntb_list_add(&qp->ntb_rx_pend_q_lock, entry, &qp->rx_pend_q);
1000		if (qp->rx_err_no_buf > qp->last_rx_no_buf) {
1001			qp->last_rx_no_buf = qp->rx_err_no_buf;
1002			CTR0(KTR_NTB, "RX: could spawn rx task");
1003			callout_reset(&qp->rx_full, hz / 1000, ntb_rx_pendq_full,
1004			    qp);
1005		}
1006	}
1007}
1008
1009/* Link Event handler */
1010static void
1011ntb_transport_event_callback(void *data, enum ntb_hw_event event)
1012{
1013	struct ntb_netdev *nt = data;
1014
1015	switch (event) {
1016	case NTB_EVENT_HW_LINK_UP:
1017		if (bootverbose)
1018			device_printf(ntb_get_device(nt->ntb), "HW link up\n");
1019		callout_reset(&nt->link_work, 0, ntb_transport_link_work, nt);
1020		break;
1021	case NTB_EVENT_HW_LINK_DOWN:
1022		if (bootverbose)
1023			device_printf(ntb_get_device(nt->ntb), "HW link down\n");
1024		ntb_transport_link_cleanup(nt);
1025		break;
1026	default:
1027		panic("ntb: Unknown NTB event");
1028	}
1029}
1030
1031/* Link bring up */
1032static void
1033ntb_transport_link_work(void *arg)
1034{
1035	struct ntb_netdev *nt = arg;
1036	struct ntb_softc *ntb = nt->ntb;
1037	struct ntb_transport_qp *qp;
1038	uint64_t val64;
1039	uint32_t val, i, num_mw;
1040	int rc;
1041
1042	if (ntb_has_feature(ntb, NTB_REGS_THRU_MW))
1043		num_mw = NTB_NUM_MW - 1;
1044	else
1045		num_mw = NTB_NUM_MW;
1046
1047	/* send the local info, in the opposite order of the way we read it */
1048	for (i = 0; i < num_mw; i++) {
1049		rc = ntb_write_remote_spad(ntb, IF_NTB_MW0_SZ_HIGH + (i * 2),
1050		    (uint64_t)ntb_get_mw_size(ntb, i) >> 32);
1051		if (rc != 0)
1052			goto out;
1053
1054		rc = ntb_write_remote_spad(ntb, IF_NTB_MW0_SZ_LOW + (i * 2),
1055		    (uint32_t)ntb_get_mw_size(ntb, i));
1056		if (rc != 0)
1057			goto out;
1058	}
1059
1060	rc = ntb_write_remote_spad(ntb, IF_NTB_NUM_MWS, num_mw);
1061	if (rc != 0)
1062		goto out;
1063
1064	rc = ntb_write_remote_spad(ntb, IF_NTB_NUM_QPS, nt->max_qps);
1065	if (rc != 0)
1066		goto out;
1067
1068	rc = ntb_write_remote_spad(ntb, IF_NTB_VERSION, NTB_TRANSPORT_VERSION);
1069	if (rc != 0)
1070		goto out;
1071
1072	/* Query the remote side for its info */
1073	rc = ntb_read_local_spad(ntb, IF_NTB_VERSION, &val);
1074	if (rc != 0)
1075		goto out;
1076
1077	if (val != NTB_TRANSPORT_VERSION)
1078		goto out;
1079
1080	rc = ntb_read_local_spad(ntb, IF_NTB_NUM_QPS, &val);
1081	if (rc != 0)
1082		goto out;
1083
1084	if (val != nt->max_qps)
1085		goto out;
1086
1087	rc = ntb_read_local_spad(ntb, IF_NTB_NUM_MWS, &val);
1088	if (rc != 0)
1089		goto out;
1090
1091	if (val != num_mw)
1092		goto out;
1093
1094	for (i = 0; i < num_mw; i++) {
1095		rc = ntb_read_local_spad(ntb, IF_NTB_MW0_SZ_HIGH + (i * 2),
1096		    &val);
1097		if (rc != 0)
1098			goto free_mws;
1099
1100		val64 = (uint64_t)val << 32;
1101
1102		rc = ntb_read_local_spad(ntb, IF_NTB_MW0_SZ_LOW + (i * 2),
1103		    &val);
1104		if (rc != 0)
1105			goto free_mws;
1106
1107		val64 |= val;
1108
1109		rc = ntb_set_mw(nt, i, val64);
1110		if (rc != 0)
1111			goto free_mws;
1112	}
1113
1114	nt->transport_link = NTB_LINK_UP;
1115	if (bootverbose)
1116		device_printf(ntb_get_device(ntb), "transport link up\n");
1117
1118	for (i = 0; i < nt->max_qps; i++) {
1119		qp = &nt->qps[i];
1120
1121		ntb_transport_setup_qp_mw(nt, i);
1122
1123		if (qp->client_ready == NTB_LINK_UP)
1124			callout_reset(&qp->link_work, 0, ntb_qp_link_work, qp);
1125	}
1126
1127	return;
1128
1129free_mws:
1130	for (i = 0; i < NTB_NUM_MW; i++)
1131		ntb_free_mw(nt, i);
1132out:
1133	if (ntb_query_link_status(ntb))
1134		callout_reset(&nt->link_work,
1135		    NTB_LINK_DOWN_TIMEOUT * hz / 1000, ntb_transport_link_work, nt);
1136}
1137
1138static int
1139ntb_set_mw(struct ntb_netdev *nt, int num_mw, unsigned int size)
1140{
1141	struct ntb_transport_mw *mw = &nt->mw[num_mw];
1142
1143	/* No need to re-setup */
1144	if (mw->size == size)
1145		return (0);
1146
1147	if (mw->size != 0)
1148		ntb_free_mw(nt, num_mw);
1149
1150	/* Alloc memory for receiving data.  Must be 4k aligned */
1151	mw->size = size;
1152
1153	mw->virt_addr = contigmalloc(mw->size, M_NTB_IF, M_ZERO, 0,
1154	    BUS_SPACE_MAXADDR, mw->size, 0);
1155	if (mw->virt_addr == NULL) {
1156		mw->size = 0;
1157		printf("ntb: Unable to allocate MW buffer of size %d\n",
1158		    (int)mw->size);
1159		return (ENOMEM);
1160	}
1161	/* TODO: replace with bus_space_* functions */
1162	mw->dma_addr = vtophys(mw->virt_addr);
1163
1164	/* Notify HW the memory location of the receive buffer */
1165	ntb_set_mw_addr(nt->ntb, num_mw, mw->dma_addr);
1166
1167	return (0);
1168}
1169
1170static void
1171ntb_free_mw(struct ntb_netdev *nt, int num_mw)
1172{
1173	struct ntb_transport_mw *mw = &nt->mw[num_mw];
1174
1175	if (mw->virt_addr == NULL)
1176		return;
1177
1178	contigfree(mw->virt_addr, mw->size, M_NTB_IF);
1179	mw->virt_addr = NULL;
1180}
1181
1182static void
1183ntb_transport_setup_qp_mw(struct ntb_netdev *nt, unsigned int qp_num)
1184{
1185	struct ntb_transport_qp *qp = &nt->qps[qp_num];
1186	void *offset;
1187	unsigned int rx_size, num_qps_mw;
1188	uint8_t mw_num = QP_TO_MW(qp_num);
1189	unsigned int i;
1190
1191	if (nt->max_qps % NTB_NUM_MW && mw_num < nt->max_qps % NTB_NUM_MW)
1192		num_qps_mw = nt->max_qps / NTB_NUM_MW + 1;
1193	else
1194		num_qps_mw = nt->max_qps / NTB_NUM_MW;
1195
1196	rx_size = (unsigned int) nt->mw[mw_num].size / num_qps_mw;
1197	qp->remote_rx_info = (void *)((uint8_t *)nt->mw[mw_num].virt_addr +
1198			     (qp_num / NTB_NUM_MW * rx_size));
1199	rx_size -= sizeof(struct ntb_rx_info);
1200
1201	qp->rx_buff = qp->remote_rx_info + 1;
1202	/* Due to house-keeping, there must be at least 2 buffs */
1203	qp->rx_max_frame = min(transport_mtu + sizeof(struct ntb_payload_header),
1204	    rx_size / 2);
1205	qp->rx_max_entry = rx_size / qp->rx_max_frame;
1206	qp->rx_index = 0;
1207
1208	qp->remote_rx_info->entry = qp->rx_max_entry - 1;
1209
1210	/* setup the hdr offsets with 0's */
1211	for (i = 0; i < qp->rx_max_entry; i++) {
1212		offset = (void *)((uint8_t *)qp->rx_buff +
1213		    qp->rx_max_frame * (i + 1) -
1214		    sizeof(struct ntb_payload_header));
1215		memset(offset, 0, sizeof(struct ntb_payload_header));
1216	}
1217
1218	qp->rx_pkts = 0;
1219	qp->tx_pkts = 0;
1220	qp->tx_index = 0;
1221}
1222
1223static void
1224ntb_qp_link_work(void *arg)
1225{
1226	struct ntb_transport_qp *qp = arg;
1227	struct ntb_softc *ntb = qp->ntb;
1228	struct ntb_netdev *nt = qp->transport;
1229	int rc, val;
1230
1231
1232	rc = ntb_read_remote_spad(ntb, IF_NTB_QP_LINKS, &val);
1233	if (rc != 0)
1234		return;
1235
1236	rc = ntb_write_remote_spad(ntb, IF_NTB_QP_LINKS, val | 1 << qp->qp_num);
1237
1238	/* query remote spad for qp ready bits */
1239	rc = ntb_read_local_spad(ntb, IF_NTB_QP_LINKS, &val);
1240
1241	/* See if the remote side is up */
1242	if ((1 << qp->qp_num & val) != 0) {
1243		qp->qp_link = NTB_LINK_UP;
1244		if (qp->event_handler != NULL)
1245			qp->event_handler(qp->cb_data, NTB_LINK_UP);
1246		if (bootverbose)
1247			device_printf(ntb_get_device(ntb), "qp link up\n");
1248	} else if (nt->transport_link == NTB_LINK_UP) {
1249		callout_reset(&qp->link_work,
1250		    NTB_LINK_DOWN_TIMEOUT * hz / 1000, ntb_qp_link_work, qp);
1251	}
1252}
1253
1254/* Link down event*/
1255static void
1256ntb_transport_link_cleanup(struct ntb_netdev *nt)
1257{
1258	int i;
1259
1260	/* Pass along the info to any clients */
1261	for (i = 0; i < nt->max_qps; i++)
1262		if (!test_bit(i, &nt->qp_bitmap))
1263			ntb_qp_link_down(&nt->qps[i]);
1264
1265	if (nt->transport_link == NTB_LINK_DOWN)
1266		callout_drain(&nt->link_work);
1267	else
1268		nt->transport_link = NTB_LINK_DOWN;
1269
1270	/*
1271	 * The scratchpad registers keep the values if the remote side
1272	 * goes down, blast them now to give them a sane value the next
1273	 * time they are accessed
1274	 */
1275	for (i = 0; i < IF_NTB_MAX_SPAD; i++)
1276		ntb_write_local_spad(nt->ntb, i, 0);
1277}
1278
1279
1280static void
1281ntb_qp_link_down(struct ntb_transport_qp *qp)
1282{
1283
1284	ntb_qp_link_cleanup(qp);
1285}
1286
1287static void
1288ntb_qp_link_cleanup(struct ntb_transport_qp *qp)
1289{
1290	struct ntb_netdev *nt = qp->transport;
1291
1292	if (qp->qp_link == NTB_LINK_DOWN) {
1293		callout_drain(&qp->link_work);
1294		return;
1295	}
1296
1297	if (qp->event_handler != NULL)
1298		qp->event_handler(qp->cb_data, NTB_LINK_DOWN);
1299
1300	qp->qp_link = NTB_LINK_DOWN;
1301
1302	if (nt->transport_link == NTB_LINK_UP)
1303		callout_reset(&qp->link_work,
1304		    NTB_LINK_DOWN_TIMEOUT * hz / 1000, ntb_qp_link_work, qp);
1305}
1306
1307/* Link commanded down */
1308/**
1309 * ntb_transport_link_down - Notify NTB transport to no longer enqueue data
1310 * @qp: NTB transport layer queue to be disabled
1311 *
1312 * Notify NTB transport layer of client's desire to no longer receive data on
1313 * transport queue specified.  It is the client's responsibility to ensure all
1314 * entries on queue are purged or otherwise handled appropriately.
1315 */
1316static void
1317ntb_transport_link_down(struct ntb_transport_qp *qp)
1318{
1319	int rc, val;
1320
1321	if (qp == NULL)
1322		return;
1323
1324	qp->client_ready = NTB_LINK_DOWN;
1325
1326	rc = ntb_read_remote_spad(qp->ntb, IF_NTB_QP_LINKS, &val);
1327	if (rc != 0)
1328		return;
1329
1330	rc = ntb_write_remote_spad(qp->ntb, IF_NTB_QP_LINKS,
1331	   val & ~(1 << qp->qp_num));
1332
1333	if (qp->qp_link == NTB_LINK_UP)
1334		ntb_send_link_down(qp);
1335	else
1336		callout_drain(&qp->link_work);
1337
1338}
1339
1340static void
1341ntb_send_link_down(struct ntb_transport_qp *qp)
1342{
1343	struct ntb_queue_entry *entry;
1344	int i, rc;
1345
1346	if (qp->qp_link == NTB_LINK_DOWN)
1347		return;
1348
1349	qp->qp_link = NTB_LINK_DOWN;
1350
1351	for (i = 0; i < NTB_LINK_DOWN_TIMEOUT; i++) {
1352		entry = ntb_list_rm(&qp->ntb_tx_free_q_lock, &qp->tx_free_q);
1353		if (entry != NULL)
1354			break;
1355		pause("NTB Wait for link down", hz / 10);
1356	}
1357
1358	if (entry == NULL)
1359		return;
1360
1361	entry->cb_data = NULL;
1362	entry->buf = NULL;
1363	entry->len = 0;
1364	entry->flags = IF_NTB_LINK_DOWN_FLAG;
1365
1366	mtx_lock(&qp->transport->tx_lock);
1367	rc = ntb_process_tx(qp, entry);
1368	if (rc != 0)
1369		printf("ntb: Failed to send link down\n");
1370	mtx_unlock(&qp->transport->tx_lock);
1371}
1372
1373
1374/* List Management */
1375
1376static void
1377ntb_list_add(struct mtx *lock, struct ntb_queue_entry *entry,
1378    struct ntb_queue_list *list)
1379{
1380
1381	mtx_lock_spin(lock);
1382	STAILQ_INSERT_TAIL(list, entry, entry);
1383	mtx_unlock_spin(lock);
1384}
1385
1386static struct ntb_queue_entry *
1387ntb_list_rm(struct mtx *lock, struct ntb_queue_list *list)
1388{
1389	struct ntb_queue_entry *entry;
1390
1391	mtx_lock_spin(lock);
1392	if (STAILQ_EMPTY(list)) {
1393		entry = NULL;
1394		goto out;
1395	}
1396	entry = STAILQ_FIRST(list);
1397	STAILQ_REMOVE_HEAD(list, entry);
1398out:
1399	mtx_unlock_spin(lock);
1400
1401	return (entry);
1402}
1403
1404/* Helper functions */
1405/* TODO: This too should really be part of the kernel */
1406#define EUI48_MULTICAST			1 << 0
1407#define EUI48_LOCALLY_ADMINISTERED	1 << 1
1408static void
1409create_random_local_eui48(u_char *eaddr)
1410{
1411	static uint8_t counter = 0;
1412	uint32_t seed = ticks;
1413
1414	eaddr[0] = EUI48_LOCALLY_ADMINISTERED;
1415	memcpy(&eaddr[1], &seed, sizeof(uint32_t));
1416	eaddr[5] = counter++;
1417}
1418
1419/**
1420 * ntb_transport_max_size - Query the max payload size of a qp
1421 * @qp: NTB transport layer queue to be queried
1422 *
1423 * Query the maximum payload size permissible on the given qp
1424 *
1425 * RETURNS: the max payload size of a qp
1426 */
1427static unsigned int
1428ntb_transport_max_size(struct ntb_transport_qp *qp)
1429{
1430
1431	if (qp == NULL)
1432		return (0);
1433
1434	return (qp->tx_max_frame - sizeof(struct ntb_payload_header));
1435}
1436