xref: /freebsd/sys/net/iflib.c (revision c203bd70b5957f85616424b6fa374479372d06e3)
1 /*-
2  * Copyright (c) 2014-2018, Matthew Macy <mmacy@mattmacy.io>
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 are met:
7  *
8  *  1. Redistributions of source code must retain the above copyright notice,
9  *     this list of conditions and the following disclaimer.
10  *
11  *  2. Neither the name of Matthew Macy nor the names of its
12  *     contributors may be used to endorse or promote products derived from
13  *     this software without specific prior written permission.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
16  * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18  * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
19  * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
20  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
21  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
22  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
23  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
24  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
25  * POSSIBILITY OF SUCH DAMAGE.
26  */
27 
28 #include <sys/cdefs.h>
29 __FBSDID("$FreeBSD$");
30 
31 #include "opt_inet.h"
32 #include "opt_inet6.h"
33 #include "opt_acpi.h"
34 #include "opt_sched.h"
35 
36 #include <sys/param.h>
37 #include <sys/types.h>
38 #include <sys/bus.h>
39 #include <sys/eventhandler.h>
40 #include <sys/kernel.h>
41 #include <sys/lock.h>
42 #include <sys/mutex.h>
43 #include <sys/module.h>
44 #include <sys/kobj.h>
45 #include <sys/rman.h>
46 #include <sys/sbuf.h>
47 #include <sys/smp.h>
48 #include <sys/socket.h>
49 #include <sys/sockio.h>
50 #include <sys/sysctl.h>
51 #include <sys/syslog.h>
52 #include <sys/taskqueue.h>
53 #include <sys/limits.h>
54 
55 #include <net/if.h>
56 #include <net/if_var.h>
57 #include <net/if_types.h>
58 #include <net/if_media.h>
59 #include <net/bpf.h>
60 #include <net/ethernet.h>
61 #include <net/mp_ring.h>
62 #include <net/debugnet.h>
63 #include <net/pfil.h>
64 #include <net/vnet.h>
65 
66 #include <netinet/in.h>
67 #include <netinet/in_pcb.h>
68 #include <netinet/tcp_lro.h>
69 #include <netinet/in_systm.h>
70 #include <netinet/if_ether.h>
71 #include <netinet/ip.h>
72 #include <netinet/ip6.h>
73 #include <netinet/tcp.h>
74 #include <netinet/ip_var.h>
75 #include <netinet6/ip6_var.h>
76 
77 #include <machine/bus.h>
78 #include <machine/in_cksum.h>
79 
80 #include <vm/vm.h>
81 #include <vm/pmap.h>
82 
83 #include <dev/led/led.h>
84 #include <dev/pci/pcireg.h>
85 #include <dev/pci/pcivar.h>
86 #include <dev/pci/pci_private.h>
87 
88 #include <net/iflib.h>
89 #include <net/iflib_private.h>
90 
91 #include "ifdi_if.h"
92 
93 #ifdef PCI_IOV
94 #include <dev/pci/pci_iov.h>
95 #endif
96 
97 #include <sys/bitstring.h>
98 /*
99  * enable accounting of every mbuf as it comes in to and goes out of
100  * iflib's software descriptor references
101  */
102 #define MEMORY_LOGGING 0
103 /*
104  * Enable mbuf vectors for compressing long mbuf chains
105  */
106 
107 /*
108  * NB:
109  * - Prefetching in tx cleaning should perhaps be a tunable. The distance ahead
110  *   we prefetch needs to be determined by the time spent in m_free vis a vis
111  *   the cost of a prefetch. This will of course vary based on the workload:
112  *      - NFLX's m_free path is dominated by vm-based M_EXT manipulation which
113  *        is quite expensive, thus suggesting very little prefetch.
114  *      - small packet forwarding which is just returning a single mbuf to
115  *        UMA will typically be very fast vis a vis the cost of a memory
116  *        access.
117  */
118 
119 /*
120  * File organization:
121  *  - private structures
122  *  - iflib private utility functions
123  *  - ifnet functions
124  *  - vlan registry and other exported functions
125  *  - iflib public core functions
126  *
127  *
128  */
129 MALLOC_DEFINE(M_IFLIB, "iflib", "ifnet library");
130 
131 #define	IFLIB_RXEOF_MORE (1U << 0)
132 #define	IFLIB_RXEOF_EMPTY (2U << 0)
133 
134 struct iflib_txq;
135 typedef struct iflib_txq *iflib_txq_t;
136 struct iflib_rxq;
137 typedef struct iflib_rxq *iflib_rxq_t;
138 struct iflib_fl;
139 typedef struct iflib_fl *iflib_fl_t;
140 
141 struct iflib_ctx;
142 
143 static void iru_init(if_rxd_update_t iru, iflib_rxq_t rxq, uint8_t flid);
144 static void iflib_timer(void *arg);
145 static void iflib_tqg_detach(if_ctx_t ctx);
146 
147 typedef struct iflib_filter_info {
148 	driver_filter_t *ifi_filter;
149 	void *ifi_filter_arg;
150 	struct grouptask *ifi_task;
151 	void *ifi_ctx;
152 } *iflib_filter_info_t;
153 
154 struct iflib_ctx {
155 	KOBJ_FIELDS;
156 	/*
157 	 * Pointer to hardware driver's softc
158 	 */
159 	void *ifc_softc;
160 	device_t ifc_dev;
161 	if_t ifc_ifp;
162 
163 	cpuset_t ifc_cpus;
164 	if_shared_ctx_t ifc_sctx;
165 	struct if_softc_ctx ifc_softc_ctx;
166 
167 	struct sx ifc_ctx_sx;
168 	struct mtx ifc_state_mtx;
169 
170 	iflib_txq_t ifc_txqs;
171 	iflib_rxq_t ifc_rxqs;
172 	uint32_t ifc_if_flags;
173 	uint32_t ifc_flags;
174 	uint32_t ifc_max_fl_buf_size;
175 	uint32_t ifc_rx_mbuf_sz;
176 
177 	int ifc_link_state;
178 	int ifc_watchdog_events;
179 	struct cdev *ifc_led_dev;
180 	struct resource *ifc_msix_mem;
181 
182 	struct if_irq ifc_legacy_irq;
183 	struct grouptask ifc_admin_task;
184 	struct grouptask ifc_vflr_task;
185 	struct iflib_filter_info ifc_filter_info;
186 	struct ifmedia	ifc_media;
187 	struct ifmedia	*ifc_mediap;
188 
189 	struct sysctl_oid *ifc_sysctl_node;
190 	uint16_t ifc_sysctl_ntxqs;
191 	uint16_t ifc_sysctl_nrxqs;
192 	uint16_t ifc_sysctl_qs_eq_override;
193 	uint16_t ifc_sysctl_rx_budget;
194 	uint16_t ifc_sysctl_tx_abdicate;
195 	uint16_t ifc_sysctl_core_offset;
196 #define	CORE_OFFSET_UNSPECIFIED	0xffff
197 	uint8_t  ifc_sysctl_separate_txrx;
198 
199 	qidx_t ifc_sysctl_ntxds[8];
200 	qidx_t ifc_sysctl_nrxds[8];
201 	struct if_txrx ifc_txrx;
202 #define isc_txd_encap  ifc_txrx.ift_txd_encap
203 #define isc_txd_flush  ifc_txrx.ift_txd_flush
204 #define isc_txd_credits_update  ifc_txrx.ift_txd_credits_update
205 #define isc_rxd_available ifc_txrx.ift_rxd_available
206 #define isc_rxd_pkt_get ifc_txrx.ift_rxd_pkt_get
207 #define isc_rxd_refill ifc_txrx.ift_rxd_refill
208 #define isc_rxd_flush ifc_txrx.ift_rxd_flush
209 #define isc_legacy_intr ifc_txrx.ift_legacy_intr
210 	eventhandler_tag ifc_vlan_attach_event;
211 	eventhandler_tag ifc_vlan_detach_event;
212 	struct ether_addr ifc_mac;
213 };
214 
215 void *
216 iflib_get_softc(if_ctx_t ctx)
217 {
218 
219 	return (ctx->ifc_softc);
220 }
221 
222 device_t
223 iflib_get_dev(if_ctx_t ctx)
224 {
225 
226 	return (ctx->ifc_dev);
227 }
228 
229 if_t
230 iflib_get_ifp(if_ctx_t ctx)
231 {
232 
233 	return (ctx->ifc_ifp);
234 }
235 
236 struct ifmedia *
237 iflib_get_media(if_ctx_t ctx)
238 {
239 
240 	return (ctx->ifc_mediap);
241 }
242 
243 uint32_t
244 iflib_get_flags(if_ctx_t ctx)
245 {
246 	return (ctx->ifc_flags);
247 }
248 
249 void
250 iflib_set_mac(if_ctx_t ctx, uint8_t mac[ETHER_ADDR_LEN])
251 {
252 
253 	bcopy(mac, ctx->ifc_mac.octet, ETHER_ADDR_LEN);
254 }
255 
256 if_softc_ctx_t
257 iflib_get_softc_ctx(if_ctx_t ctx)
258 {
259 
260 	return (&ctx->ifc_softc_ctx);
261 }
262 
263 if_shared_ctx_t
264 iflib_get_sctx(if_ctx_t ctx)
265 {
266 
267 	return (ctx->ifc_sctx);
268 }
269 
270 #define IP_ALIGNED(m) ((((uintptr_t)(m)->m_data) & 0x3) == 0x2)
271 #define CACHE_PTR_INCREMENT (CACHE_LINE_SIZE/sizeof(void*))
272 #define CACHE_PTR_NEXT(ptr) ((void *)(((uintptr_t)(ptr)+CACHE_LINE_SIZE-1) & (CACHE_LINE_SIZE-1)))
273 
274 #define LINK_ACTIVE(ctx) ((ctx)->ifc_link_state == LINK_STATE_UP)
275 #define CTX_IS_VF(ctx) ((ctx)->ifc_sctx->isc_flags & IFLIB_IS_VF)
276 
277 typedef struct iflib_sw_rx_desc_array {
278 	bus_dmamap_t	*ifsd_map;         /* bus_dma maps for packet */
279 	struct mbuf	**ifsd_m;           /* pkthdr mbufs */
280 	caddr_t		*ifsd_cl;          /* direct cluster pointer for rx */
281 	bus_addr_t	*ifsd_ba;          /* bus addr of cluster for rx */
282 } iflib_rxsd_array_t;
283 
284 typedef struct iflib_sw_tx_desc_array {
285 	bus_dmamap_t    *ifsd_map;         /* bus_dma maps for packet */
286 	bus_dmamap_t	*ifsd_tso_map;     /* bus_dma maps for TSO packet */
287 	struct mbuf    **ifsd_m;           /* pkthdr mbufs */
288 } if_txsd_vec_t;
289 
290 /* magic number that should be high enough for any hardware */
291 #define IFLIB_MAX_TX_SEGS		128
292 #define IFLIB_RX_COPY_THRESH		128
293 #define IFLIB_MAX_RX_REFRESH		32
294 /* The minimum descriptors per second before we start coalescing */
295 #define IFLIB_MIN_DESC_SEC		16384
296 #define IFLIB_DEFAULT_TX_UPDATE_FREQ	16
297 #define IFLIB_QUEUE_IDLE		0
298 #define IFLIB_QUEUE_HUNG		1
299 #define IFLIB_QUEUE_WORKING		2
300 /* maximum number of txqs that can share an rx interrupt */
301 #define IFLIB_MAX_TX_SHARED_INTR	4
302 
303 /* this should really scale with ring size - this is a fairly arbitrary value */
304 #define TX_BATCH_SIZE			32
305 
306 #define IFLIB_RESTART_BUDGET		8
307 
308 #define CSUM_OFFLOAD		(CSUM_IP_TSO|CSUM_IP6_TSO|CSUM_IP| \
309 				 CSUM_IP_UDP|CSUM_IP_TCP|CSUM_IP_SCTP| \
310 				 CSUM_IP6_UDP|CSUM_IP6_TCP|CSUM_IP6_SCTP)
311 
312 struct iflib_txq {
313 	qidx_t		ift_in_use;
314 	qidx_t		ift_cidx;
315 	qidx_t		ift_cidx_processed;
316 	qidx_t		ift_pidx;
317 	uint8_t		ift_gen;
318 	uint8_t		ift_br_offset;
319 	uint16_t	ift_npending;
320 	uint16_t	ift_db_pending;
321 	uint16_t	ift_rs_pending;
322 	/* implicit pad */
323 	uint8_t		ift_txd_size[8];
324 	uint64_t	ift_processed;
325 	uint64_t	ift_cleaned;
326 	uint64_t	ift_cleaned_prev;
327 #if MEMORY_LOGGING
328 	uint64_t	ift_enqueued;
329 	uint64_t	ift_dequeued;
330 #endif
331 	uint64_t	ift_no_tx_dma_setup;
332 	uint64_t	ift_no_desc_avail;
333 	uint64_t	ift_mbuf_defrag_failed;
334 	uint64_t	ift_mbuf_defrag;
335 	uint64_t	ift_map_failed;
336 	uint64_t	ift_txd_encap_efbig;
337 	uint64_t	ift_pullups;
338 	uint64_t	ift_last_timer_tick;
339 
340 	struct mtx	ift_mtx;
341 	struct mtx	ift_db_mtx;
342 
343 	/* constant values */
344 	if_ctx_t	ift_ctx;
345 	struct ifmp_ring        *ift_br;
346 	struct grouptask	ift_task;
347 	qidx_t		ift_size;
348 	uint16_t	ift_id;
349 	struct callout	ift_timer;
350 #ifdef DEV_NETMAP
351 	struct callout	ift_netmap_timer;
352 #endif /* DEV_NETMAP */
353 
354 	if_txsd_vec_t	ift_sds;
355 	uint8_t		ift_qstatus;
356 	uint8_t		ift_closed;
357 	uint8_t		ift_update_freq;
358 	struct iflib_filter_info ift_filter_info;
359 	bus_dma_tag_t	ift_buf_tag;
360 	bus_dma_tag_t	ift_tso_buf_tag;
361 	iflib_dma_info_t	ift_ifdi;
362 #define	MTX_NAME_LEN	32
363 	char                    ift_mtx_name[MTX_NAME_LEN];
364 	bus_dma_segment_t	ift_segs[IFLIB_MAX_TX_SEGS]  __aligned(CACHE_LINE_SIZE);
365 #ifdef IFLIB_DIAGNOSTICS
366 	uint64_t ift_cpu_exec_count[256];
367 #endif
368 } __aligned(CACHE_LINE_SIZE);
369 
370 struct iflib_fl {
371 	qidx_t		ifl_cidx;
372 	qidx_t		ifl_pidx;
373 	qidx_t		ifl_credits;
374 	uint8_t		ifl_gen;
375 	uint8_t		ifl_rxd_size;
376 #if MEMORY_LOGGING
377 	uint64_t	ifl_m_enqueued;
378 	uint64_t	ifl_m_dequeued;
379 	uint64_t	ifl_cl_enqueued;
380 	uint64_t	ifl_cl_dequeued;
381 #endif
382 	/* implicit pad */
383 	bitstr_t 	*ifl_rx_bitmap;
384 	qidx_t		ifl_fragidx;
385 	/* constant */
386 	qidx_t		ifl_size;
387 	uint16_t	ifl_buf_size;
388 	uint16_t	ifl_cltype;
389 	uma_zone_t	ifl_zone;
390 	iflib_rxsd_array_t	ifl_sds;
391 	iflib_rxq_t	ifl_rxq;
392 	uint8_t		ifl_id;
393 	bus_dma_tag_t	ifl_buf_tag;
394 	iflib_dma_info_t	ifl_ifdi;
395 	uint64_t	ifl_bus_addrs[IFLIB_MAX_RX_REFRESH] __aligned(CACHE_LINE_SIZE);
396 	qidx_t		ifl_rxd_idxs[IFLIB_MAX_RX_REFRESH];
397 }  __aligned(CACHE_LINE_SIZE);
398 
399 static inline qidx_t
400 get_inuse(int size, qidx_t cidx, qidx_t pidx, uint8_t gen)
401 {
402 	qidx_t used;
403 
404 	if (pidx > cidx)
405 		used = pidx - cidx;
406 	else if (pidx < cidx)
407 		used = size - cidx + pidx;
408 	else if (gen == 0 && pidx == cidx)
409 		used = 0;
410 	else if (gen == 1 && pidx == cidx)
411 		used = size;
412 	else
413 		panic("bad state");
414 
415 	return (used);
416 }
417 
418 #define TXQ_AVAIL(txq) (txq->ift_size - get_inuse(txq->ift_size, txq->ift_cidx, txq->ift_pidx, txq->ift_gen))
419 
420 #define IDXDIFF(head, tail, wrap) \
421 	((head) >= (tail) ? (head) - (tail) : (wrap) - (tail) + (head))
422 
423 struct iflib_rxq {
424 	if_ctx_t	ifr_ctx;
425 	iflib_fl_t	ifr_fl;
426 	uint64_t	ifr_rx_irq;
427 	struct pfil_head	*pfil;
428 	/*
429 	 * If there is a separate completion queue (IFLIB_HAS_RXCQ), this is
430 	 * the completion queue consumer index.  Otherwise it's unused.
431 	 */
432 	qidx_t		ifr_cq_cidx;
433 	uint16_t	ifr_id;
434 	uint8_t		ifr_nfl;
435 	uint8_t		ifr_ntxqirq;
436 	uint8_t		ifr_txqid[IFLIB_MAX_TX_SHARED_INTR];
437 	uint8_t		ifr_fl_offset;
438 	struct lro_ctrl			ifr_lc;
439 	struct grouptask        ifr_task;
440 	struct callout		ifr_watchdog;
441 	struct iflib_filter_info ifr_filter_info;
442 	iflib_dma_info_t		ifr_ifdi;
443 
444 	/* dynamically allocate if any drivers need a value substantially larger than this */
445 	struct if_rxd_frag	ifr_frags[IFLIB_MAX_RX_SEGS] __aligned(CACHE_LINE_SIZE);
446 #ifdef IFLIB_DIAGNOSTICS
447 	uint64_t ifr_cpu_exec_count[256];
448 #endif
449 }  __aligned(CACHE_LINE_SIZE);
450 
451 typedef struct if_rxsd {
452 	caddr_t *ifsd_cl;
453 	iflib_fl_t ifsd_fl;
454 } *if_rxsd_t;
455 
456 /* multiple of word size */
457 #ifdef __LP64__
458 #define PKT_INFO_SIZE	6
459 #define RXD_INFO_SIZE	5
460 #define PKT_TYPE uint64_t
461 #else
462 #define PKT_INFO_SIZE	11
463 #define RXD_INFO_SIZE	8
464 #define PKT_TYPE uint32_t
465 #endif
466 #define PKT_LOOP_BOUND  ((PKT_INFO_SIZE/3)*3)
467 #define RXD_LOOP_BOUND  ((RXD_INFO_SIZE/4)*4)
468 
469 typedef struct if_pkt_info_pad {
470 	PKT_TYPE pkt_val[PKT_INFO_SIZE];
471 } *if_pkt_info_pad_t;
472 typedef struct if_rxd_info_pad {
473 	PKT_TYPE rxd_val[RXD_INFO_SIZE];
474 } *if_rxd_info_pad_t;
475 
476 CTASSERT(sizeof(struct if_pkt_info_pad) == sizeof(struct if_pkt_info));
477 CTASSERT(sizeof(struct if_rxd_info_pad) == sizeof(struct if_rxd_info));
478 
479 static inline void
480 pkt_info_zero(if_pkt_info_t pi)
481 {
482 	if_pkt_info_pad_t pi_pad;
483 
484 	pi_pad = (if_pkt_info_pad_t)pi;
485 	pi_pad->pkt_val[0] = 0; pi_pad->pkt_val[1] = 0; pi_pad->pkt_val[2] = 0;
486 	pi_pad->pkt_val[3] = 0; pi_pad->pkt_val[4] = 0; pi_pad->pkt_val[5] = 0;
487 #ifndef __LP64__
488 	pi_pad->pkt_val[6] = 0; pi_pad->pkt_val[7] = 0; pi_pad->pkt_val[8] = 0;
489 	pi_pad->pkt_val[9] = 0; pi_pad->pkt_val[10] = 0;
490 #endif
491 }
492 
493 static device_method_t iflib_pseudo_methods[] = {
494 	DEVMETHOD(device_attach, noop_attach),
495 	DEVMETHOD(device_detach, iflib_pseudo_detach),
496 	DEVMETHOD_END
497 };
498 
499 driver_t iflib_pseudodriver = {
500 	"iflib_pseudo", iflib_pseudo_methods, sizeof(struct iflib_ctx),
501 };
502 
503 static inline void
504 rxd_info_zero(if_rxd_info_t ri)
505 {
506 	if_rxd_info_pad_t ri_pad;
507 	int i;
508 
509 	ri_pad = (if_rxd_info_pad_t)ri;
510 	for (i = 0; i < RXD_LOOP_BOUND; i += 4) {
511 		ri_pad->rxd_val[i] = 0;
512 		ri_pad->rxd_val[i+1] = 0;
513 		ri_pad->rxd_val[i+2] = 0;
514 		ri_pad->rxd_val[i+3] = 0;
515 	}
516 #ifdef __LP64__
517 	ri_pad->rxd_val[RXD_INFO_SIZE-1] = 0;
518 #endif
519 }
520 
521 /*
522  * Only allow a single packet to take up most 1/nth of the tx ring
523  */
524 #define MAX_SINGLE_PACKET_FRACTION 12
525 #define IF_BAD_DMA (bus_addr_t)-1
526 
527 #define CTX_ACTIVE(ctx) ((if_getdrvflags((ctx)->ifc_ifp) & IFF_DRV_RUNNING))
528 
529 #define CTX_LOCK_INIT(_sc)  sx_init(&(_sc)->ifc_ctx_sx, "iflib ctx lock")
530 #define CTX_LOCK(ctx) sx_xlock(&(ctx)->ifc_ctx_sx)
531 #define CTX_UNLOCK(ctx) sx_xunlock(&(ctx)->ifc_ctx_sx)
532 #define CTX_LOCK_DESTROY(ctx) sx_destroy(&(ctx)->ifc_ctx_sx)
533 
534 #define STATE_LOCK_INIT(_sc, _name)  mtx_init(&(_sc)->ifc_state_mtx, _name, "iflib state lock", MTX_DEF)
535 #define STATE_LOCK(ctx) mtx_lock(&(ctx)->ifc_state_mtx)
536 #define STATE_UNLOCK(ctx) mtx_unlock(&(ctx)->ifc_state_mtx)
537 #define STATE_LOCK_DESTROY(ctx) mtx_destroy(&(ctx)->ifc_state_mtx)
538 
539 #define CALLOUT_LOCK(txq)	mtx_lock(&txq->ift_mtx)
540 #define CALLOUT_UNLOCK(txq) 	mtx_unlock(&txq->ift_mtx)
541 
542 void
543 iflib_set_detach(if_ctx_t ctx)
544 {
545 	STATE_LOCK(ctx);
546 	ctx->ifc_flags |= IFC_IN_DETACH;
547 	STATE_UNLOCK(ctx);
548 }
549 
550 /* Our boot-time initialization hook */
551 static int	iflib_module_event_handler(module_t, int, void *);
552 
553 static moduledata_t iflib_moduledata = {
554 	"iflib",
555 	iflib_module_event_handler,
556 	NULL
557 };
558 
559 DECLARE_MODULE(iflib, iflib_moduledata, SI_SUB_INIT_IF, SI_ORDER_ANY);
560 MODULE_VERSION(iflib, 1);
561 
562 MODULE_DEPEND(iflib, pci, 1, 1, 1);
563 MODULE_DEPEND(iflib, ether, 1, 1, 1);
564 
565 TASKQGROUP_DEFINE(if_io_tqg, mp_ncpus, 1);
566 TASKQGROUP_DEFINE(if_config_tqg, 1, 1);
567 
568 #ifndef IFLIB_DEBUG_COUNTERS
569 #ifdef INVARIANTS
570 #define IFLIB_DEBUG_COUNTERS 1
571 #else
572 #define IFLIB_DEBUG_COUNTERS 0
573 #endif /* !INVARIANTS */
574 #endif
575 
576 static SYSCTL_NODE(_net, OID_AUTO, iflib, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
577     "iflib driver parameters");
578 
579 /*
580  * XXX need to ensure that this can't accidentally cause the head to be moved backwards
581  */
582 static int iflib_min_tx_latency = 0;
583 SYSCTL_INT(_net_iflib, OID_AUTO, min_tx_latency, CTLFLAG_RW,
584 		   &iflib_min_tx_latency, 0, "minimize transmit latency at the possible expense of throughput");
585 static int iflib_no_tx_batch = 0;
586 SYSCTL_INT(_net_iflib, OID_AUTO, no_tx_batch, CTLFLAG_RW,
587 		   &iflib_no_tx_batch, 0, "minimize transmit latency at the possible expense of throughput");
588 static int iflib_timer_default = 1000;
589 SYSCTL_INT(_net_iflib, OID_AUTO, timer_default, CTLFLAG_RW,
590 		   &iflib_timer_default, 0, "number of ticks between iflib_timer calls");
591 
592 
593 #if IFLIB_DEBUG_COUNTERS
594 
595 static int iflib_tx_seen;
596 static int iflib_tx_sent;
597 static int iflib_tx_encap;
598 static int iflib_rx_allocs;
599 static int iflib_fl_refills;
600 static int iflib_fl_refills_large;
601 static int iflib_tx_frees;
602 
603 SYSCTL_INT(_net_iflib, OID_AUTO, tx_seen, CTLFLAG_RD,
604 		   &iflib_tx_seen, 0, "# TX mbufs seen");
605 SYSCTL_INT(_net_iflib, OID_AUTO, tx_sent, CTLFLAG_RD,
606 		   &iflib_tx_sent, 0, "# TX mbufs sent");
607 SYSCTL_INT(_net_iflib, OID_AUTO, tx_encap, CTLFLAG_RD,
608 		   &iflib_tx_encap, 0, "# TX mbufs encapped");
609 SYSCTL_INT(_net_iflib, OID_AUTO, tx_frees, CTLFLAG_RD,
610 		   &iflib_tx_frees, 0, "# TX frees");
611 SYSCTL_INT(_net_iflib, OID_AUTO, rx_allocs, CTLFLAG_RD,
612 		   &iflib_rx_allocs, 0, "# RX allocations");
613 SYSCTL_INT(_net_iflib, OID_AUTO, fl_refills, CTLFLAG_RD,
614 		   &iflib_fl_refills, 0, "# refills");
615 SYSCTL_INT(_net_iflib, OID_AUTO, fl_refills_large, CTLFLAG_RD,
616 		   &iflib_fl_refills_large, 0, "# large refills");
617 
618 static int iflib_txq_drain_flushing;
619 static int iflib_txq_drain_oactive;
620 static int iflib_txq_drain_notready;
621 
622 SYSCTL_INT(_net_iflib, OID_AUTO, txq_drain_flushing, CTLFLAG_RD,
623 		   &iflib_txq_drain_flushing, 0, "# drain flushes");
624 SYSCTL_INT(_net_iflib, OID_AUTO, txq_drain_oactive, CTLFLAG_RD,
625 		   &iflib_txq_drain_oactive, 0, "# drain oactives");
626 SYSCTL_INT(_net_iflib, OID_AUTO, txq_drain_notready, CTLFLAG_RD,
627 		   &iflib_txq_drain_notready, 0, "# drain notready");
628 
629 static int iflib_encap_load_mbuf_fail;
630 static int iflib_encap_pad_mbuf_fail;
631 static int iflib_encap_txq_avail_fail;
632 static int iflib_encap_txd_encap_fail;
633 
634 SYSCTL_INT(_net_iflib, OID_AUTO, encap_load_mbuf_fail, CTLFLAG_RD,
635 		   &iflib_encap_load_mbuf_fail, 0, "# busdma load failures");
636 SYSCTL_INT(_net_iflib, OID_AUTO, encap_pad_mbuf_fail, CTLFLAG_RD,
637 		   &iflib_encap_pad_mbuf_fail, 0, "# runt frame pad failures");
638 SYSCTL_INT(_net_iflib, OID_AUTO, encap_txq_avail_fail, CTLFLAG_RD,
639 		   &iflib_encap_txq_avail_fail, 0, "# txq avail failures");
640 SYSCTL_INT(_net_iflib, OID_AUTO, encap_txd_encap_fail, CTLFLAG_RD,
641 		   &iflib_encap_txd_encap_fail, 0, "# driver encap failures");
642 
643 static int iflib_task_fn_rxs;
644 static int iflib_rx_intr_enables;
645 static int iflib_fast_intrs;
646 static int iflib_rx_unavail;
647 static int iflib_rx_ctx_inactive;
648 static int iflib_rx_if_input;
649 static int iflib_rxd_flush;
650 
651 static int iflib_verbose_debug;
652 
653 SYSCTL_INT(_net_iflib, OID_AUTO, task_fn_rx, CTLFLAG_RD,
654 		   &iflib_task_fn_rxs, 0, "# task_fn_rx calls");
655 SYSCTL_INT(_net_iflib, OID_AUTO, rx_intr_enables, CTLFLAG_RD,
656 		   &iflib_rx_intr_enables, 0, "# RX intr enables");
657 SYSCTL_INT(_net_iflib, OID_AUTO, fast_intrs, CTLFLAG_RD,
658 		   &iflib_fast_intrs, 0, "# fast_intr calls");
659 SYSCTL_INT(_net_iflib, OID_AUTO, rx_unavail, CTLFLAG_RD,
660 		   &iflib_rx_unavail, 0, "# times rxeof called with no available data");
661 SYSCTL_INT(_net_iflib, OID_AUTO, rx_ctx_inactive, CTLFLAG_RD,
662 		   &iflib_rx_ctx_inactive, 0, "# times rxeof called with inactive context");
663 SYSCTL_INT(_net_iflib, OID_AUTO, rx_if_input, CTLFLAG_RD,
664 		   &iflib_rx_if_input, 0, "# times rxeof called if_input");
665 SYSCTL_INT(_net_iflib, OID_AUTO, rxd_flush, CTLFLAG_RD,
666 	         &iflib_rxd_flush, 0, "# times rxd_flush called");
667 SYSCTL_INT(_net_iflib, OID_AUTO, verbose_debug, CTLFLAG_RW,
668 		   &iflib_verbose_debug, 0, "enable verbose debugging");
669 
670 #define DBG_COUNTER_INC(name) atomic_add_int(&(iflib_ ## name), 1)
671 static void
672 iflib_debug_reset(void)
673 {
674 	iflib_tx_seen = iflib_tx_sent = iflib_tx_encap = iflib_rx_allocs =
675 		iflib_fl_refills = iflib_fl_refills_large = iflib_tx_frees =
676 		iflib_txq_drain_flushing = iflib_txq_drain_oactive =
677 		iflib_txq_drain_notready =
678 		iflib_encap_load_mbuf_fail = iflib_encap_pad_mbuf_fail =
679 		iflib_encap_txq_avail_fail = iflib_encap_txd_encap_fail =
680 		iflib_task_fn_rxs = iflib_rx_intr_enables = iflib_fast_intrs =
681 		iflib_rx_unavail =
682 		iflib_rx_ctx_inactive = iflib_rx_if_input =
683 		iflib_rxd_flush = 0;
684 }
685 
686 #else
687 #define DBG_COUNTER_INC(name)
688 static void iflib_debug_reset(void) {}
689 #endif
690 
691 #define IFLIB_DEBUG 0
692 
693 static void iflib_tx_structures_free(if_ctx_t ctx);
694 static void iflib_rx_structures_free(if_ctx_t ctx);
695 static int iflib_queues_alloc(if_ctx_t ctx);
696 static int iflib_tx_credits_update(if_ctx_t ctx, iflib_txq_t txq);
697 static int iflib_rxd_avail(if_ctx_t ctx, iflib_rxq_t rxq, qidx_t cidx, qidx_t budget);
698 static int iflib_qset_structures_setup(if_ctx_t ctx);
699 static int iflib_msix_init(if_ctx_t ctx);
700 static int iflib_legacy_setup(if_ctx_t ctx, driver_filter_t filter, void *filterarg, int *rid, const char *str);
701 static void iflib_txq_check_drain(iflib_txq_t txq, int budget);
702 static uint32_t iflib_txq_can_drain(struct ifmp_ring *);
703 #ifdef ALTQ
704 static void iflib_altq_if_start(if_t ifp);
705 static int iflib_altq_if_transmit(if_t ifp, struct mbuf *m);
706 #endif
707 static int iflib_register(if_ctx_t);
708 static void iflib_deregister(if_ctx_t);
709 static void iflib_unregister_vlan_handlers(if_ctx_t ctx);
710 static uint16_t iflib_get_mbuf_size_for(unsigned int size);
711 static void iflib_init_locked(if_ctx_t ctx);
712 static void iflib_add_device_sysctl_pre(if_ctx_t ctx);
713 static void iflib_add_device_sysctl_post(if_ctx_t ctx);
714 static void iflib_ifmp_purge(iflib_txq_t txq);
715 static void _iflib_pre_assert(if_softc_ctx_t scctx);
716 static void iflib_if_init_locked(if_ctx_t ctx);
717 static void iflib_free_intr_mem(if_ctx_t ctx);
718 #ifndef __NO_STRICT_ALIGNMENT
719 static struct mbuf * iflib_fixup_rx(struct mbuf *m);
720 #endif
721 
722 static SLIST_HEAD(cpu_offset_list, cpu_offset) cpu_offsets =
723     SLIST_HEAD_INITIALIZER(cpu_offsets);
724 struct cpu_offset {
725 	SLIST_ENTRY(cpu_offset) entries;
726 	cpuset_t	set;
727 	unsigned int	refcount;
728 	uint16_t	offset;
729 };
730 static struct mtx cpu_offset_mtx;
731 MTX_SYSINIT(iflib_cpu_offset, &cpu_offset_mtx, "iflib_cpu_offset lock",
732     MTX_DEF);
733 
734 DEBUGNET_DEFINE(iflib);
735 
736 static int
737 iflib_num_rx_descs(if_ctx_t ctx)
738 {
739 	if_softc_ctx_t scctx = &ctx->ifc_softc_ctx;
740 	if_shared_ctx_t sctx = ctx->ifc_sctx;
741 	uint16_t first_rxq = (sctx->isc_flags & IFLIB_HAS_RXCQ) ? 1 : 0;
742 
743 	return scctx->isc_nrxd[first_rxq];
744 }
745 
746 static int
747 iflib_num_tx_descs(if_ctx_t ctx)
748 {
749 	if_softc_ctx_t scctx = &ctx->ifc_softc_ctx;
750 	if_shared_ctx_t sctx = ctx->ifc_sctx;
751 	uint16_t first_txq = (sctx->isc_flags & IFLIB_HAS_TXCQ) ? 1 : 0;
752 
753 	return scctx->isc_ntxd[first_txq];
754 }
755 
756 #ifdef DEV_NETMAP
757 #include <sys/selinfo.h>
758 #include <net/netmap.h>
759 #include <dev/netmap/netmap_kern.h>
760 
761 MODULE_DEPEND(iflib, netmap, 1, 1, 1);
762 
763 static int netmap_fl_refill(iflib_rxq_t rxq, struct netmap_kring *kring, bool init);
764 static void iflib_netmap_timer(void *arg);
765 
766 /*
767  * device-specific sysctl variables:
768  *
769  * iflib_crcstrip: 0: keep CRC in rx frames (default), 1: strip it.
770  *	During regular operations the CRC is stripped, but on some
771  *	hardware reception of frames not multiple of 64 is slower,
772  *	so using crcstrip=0 helps in benchmarks.
773  *
774  * iflib_rx_miss, iflib_rx_miss_bufs:
775  *	count packets that might be missed due to lost interrupts.
776  */
777 SYSCTL_DECL(_dev_netmap);
778 /*
779  * The xl driver by default strips CRCs and we do not override it.
780  */
781 
782 int iflib_crcstrip = 1;
783 SYSCTL_INT(_dev_netmap, OID_AUTO, iflib_crcstrip,
784     CTLFLAG_RW, &iflib_crcstrip, 1, "strip CRC on RX frames");
785 
786 int iflib_rx_miss, iflib_rx_miss_bufs;
787 SYSCTL_INT(_dev_netmap, OID_AUTO, iflib_rx_miss,
788     CTLFLAG_RW, &iflib_rx_miss, 0, "potentially missed RX intr");
789 SYSCTL_INT(_dev_netmap, OID_AUTO, iflib_rx_miss_bufs,
790     CTLFLAG_RW, &iflib_rx_miss_bufs, 0, "potentially missed RX intr bufs");
791 
792 /*
793  * Register/unregister. We are already under netmap lock.
794  * Only called on the first register or the last unregister.
795  */
796 static int
797 iflib_netmap_register(struct netmap_adapter *na, int onoff)
798 {
799 	if_t ifp = na->ifp;
800 	if_ctx_t ctx = ifp->if_softc;
801 	int status;
802 
803 	CTX_LOCK(ctx);
804 	if (!CTX_IS_VF(ctx))
805 		IFDI_CRCSTRIP_SET(ctx, onoff, iflib_crcstrip);
806 
807 	iflib_stop(ctx);
808 
809 	/*
810 	 * Enable (or disable) netmap flags, and intercept (or restore)
811 	 * ifp->if_transmit. This is done once the device has been stopped
812 	 * to prevent race conditions. Also, this must be done after
813 	 * calling netmap_disable_all_rings() and before calling
814 	 * netmap_enable_all_rings(), so that these two functions see the
815 	 * updated state of the NAF_NETMAP_ON bit.
816 	 */
817 	if (onoff) {
818 		nm_set_native_flags(na);
819 	} else {
820 		nm_clear_native_flags(na);
821 	}
822 
823 	iflib_init_locked(ctx);
824 	IFDI_CRCSTRIP_SET(ctx, onoff, iflib_crcstrip); // XXX why twice ?
825 	status = ifp->if_drv_flags & IFF_DRV_RUNNING ? 0 : 1;
826 	if (status)
827 		nm_clear_native_flags(na);
828 	CTX_UNLOCK(ctx);
829 	return (status);
830 }
831 
832 static int
833 netmap_fl_refill(iflib_rxq_t rxq, struct netmap_kring *kring, bool init)
834 {
835 	struct netmap_adapter *na = kring->na;
836 	u_int const lim = kring->nkr_num_slots - 1;
837 	struct netmap_ring *ring = kring->ring;
838 	bus_dmamap_t *map;
839 	struct if_rxd_update iru;
840 	if_ctx_t ctx = rxq->ifr_ctx;
841 	iflib_fl_t fl = &rxq->ifr_fl[0];
842 	u_int nic_i_first, nic_i;
843 	u_int nm_i;
844 	int i, n;
845 #if IFLIB_DEBUG_COUNTERS
846 	int rf_count = 0;
847 #endif
848 
849 	/*
850 	 * This function is used both at initialization and in rxsync.
851 	 * At initialization we need to prepare (with isc_rxd_refill())
852 	 * all the netmap buffers currently owned by the kernel, in
853 	 * such a way to keep fl->ifl_pidx and kring->nr_hwcur in sync
854 	 * (except for kring->nkr_hwofs). These may be less than
855 	 * kring->nkr_num_slots if netmap_reset() was called while
856 	 * an application using the kring that still owned some
857 	 * buffers.
858 	 * At rxsync time, both indexes point to the next buffer to be
859 	 * refilled.
860 	 * In any case we publish (with isc_rxd_flush()) up to
861 	 * (fl->ifl_pidx - 1) % N (included), to avoid the NIC tail/prod
862 	 * pointer to overrun the head/cons pointer, although this is
863 	 * not necessary for some NICs (e.g. vmx).
864 	 */
865 	if (__predict_false(init)) {
866 		n = kring->nkr_num_slots - nm_kr_rxspace(kring);
867 	} else {
868 		n = kring->rhead - kring->nr_hwcur;
869 		if (n == 0)
870 			return (0); /* Nothing to do. */
871 		if (n < 0)
872 			n += kring->nkr_num_slots;
873 	}
874 
875 	iru_init(&iru, rxq, 0 /* flid */);
876 	map = fl->ifl_sds.ifsd_map;
877 	nic_i = fl->ifl_pidx;
878 	nm_i = netmap_idx_n2k(kring, nic_i);
879 	if (__predict_false(init)) {
880 		/*
881 		 * On init/reset, nic_i must be 0, and we must
882 		 * start to refill from hwtail (see netmap_reset()).
883 		 */
884 		MPASS(nic_i == 0);
885 		MPASS(nm_i == kring->nr_hwtail);
886 	} else
887 		MPASS(nm_i == kring->nr_hwcur);
888 	DBG_COUNTER_INC(fl_refills);
889 	while (n > 0) {
890 #if IFLIB_DEBUG_COUNTERS
891 		if (++rf_count == 9)
892 			DBG_COUNTER_INC(fl_refills_large);
893 #endif
894 		nic_i_first = nic_i;
895 		for (i = 0; n > 0 && i < IFLIB_MAX_RX_REFRESH; n--, i++) {
896 			struct netmap_slot *slot = &ring->slot[nm_i];
897 			void *addr = PNMB(na, slot, &fl->ifl_bus_addrs[i]);
898 
899 			MPASS(i < IFLIB_MAX_RX_REFRESH);
900 
901 			if (addr == NETMAP_BUF_BASE(na)) /* bad buf */
902 			        return netmap_ring_reinit(kring);
903 
904 			fl->ifl_rxd_idxs[i] = nic_i;
905 
906 			if (__predict_false(init)) {
907 				netmap_load_map(na, fl->ifl_buf_tag,
908 				    map[nic_i], addr);
909 			} else if (slot->flags & NS_BUF_CHANGED) {
910 				/* buffer has changed, reload map */
911 				netmap_reload_map(na, fl->ifl_buf_tag,
912 				    map[nic_i], addr);
913 			}
914 			bus_dmamap_sync(fl->ifl_buf_tag, map[nic_i],
915 			    BUS_DMASYNC_PREREAD);
916 			slot->flags &= ~NS_BUF_CHANGED;
917 
918 			nm_i = nm_next(nm_i, lim);
919 			nic_i = nm_next(nic_i, lim);
920 		}
921 
922 		iru.iru_pidx = nic_i_first;
923 		iru.iru_count = i;
924 		ctx->isc_rxd_refill(ctx->ifc_softc, &iru);
925 	}
926 	fl->ifl_pidx = nic_i;
927 	/*
928 	 * At the end of the loop we must have refilled everything
929 	 * we could possibly refill.
930 	 */
931 	MPASS(nm_i == kring->rhead);
932 	kring->nr_hwcur = nm_i;
933 
934 	bus_dmamap_sync(fl->ifl_ifdi->idi_tag, fl->ifl_ifdi->idi_map,
935 	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
936 	ctx->isc_rxd_flush(ctx->ifc_softc, rxq->ifr_id, fl->ifl_id,
937 	    nm_prev(nic_i, lim));
938 	DBG_COUNTER_INC(rxd_flush);
939 
940 	return (0);
941 }
942 
943 #define NETMAP_TX_TIMER_US	90
944 
945 /*
946  * Reconcile kernel and user view of the transmit ring.
947  *
948  * All information is in the kring.
949  * Userspace wants to send packets up to the one before kring->rhead,
950  * kernel knows kring->nr_hwcur is the first unsent packet.
951  *
952  * Here we push packets out (as many as possible), and possibly
953  * reclaim buffers from previously completed transmission.
954  *
955  * The caller (netmap) guarantees that there is only one instance
956  * running at any time. Any interference with other driver
957  * methods should be handled by the individual drivers.
958  */
959 static int
960 iflib_netmap_txsync(struct netmap_kring *kring, int flags)
961 {
962 	struct netmap_adapter *na = kring->na;
963 	if_t ifp = na->ifp;
964 	struct netmap_ring *ring = kring->ring;
965 	u_int nm_i;	/* index into the netmap kring */
966 	u_int nic_i;	/* index into the NIC ring */
967 	u_int n;
968 	u_int const lim = kring->nkr_num_slots - 1;
969 	u_int const head = kring->rhead;
970 	struct if_pkt_info pi;
971 
972 	/*
973 	 * interrupts on every tx packet are expensive so request
974 	 * them every half ring, or where NS_REPORT is set
975 	 */
976 	u_int report_frequency = kring->nkr_num_slots >> 1;
977 	/* device-specific */
978 	if_ctx_t ctx = ifp->if_softc;
979 	iflib_txq_t txq = &ctx->ifc_txqs[kring->ring_id];
980 
981 	bus_dmamap_sync(txq->ift_ifdi->idi_tag, txq->ift_ifdi->idi_map,
982 	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
983 
984 	/*
985 	 * First part: process new packets to send.
986 	 * nm_i is the current index in the netmap kring,
987 	 * nic_i is the corresponding index in the NIC ring.
988 	 *
989 	 * If we have packets to send (nm_i != head)
990 	 * iterate over the netmap ring, fetch length and update
991 	 * the corresponding slot in the NIC ring. Some drivers also
992 	 * need to update the buffer's physical address in the NIC slot
993 	 * even NS_BUF_CHANGED is not set (PNMB computes the addresses).
994 	 *
995 	 * The netmap_reload_map() calls is especially expensive,
996 	 * even when (as in this case) the tag is 0, so do only
997 	 * when the buffer has actually changed.
998 	 *
999 	 * If possible do not set the report/intr bit on all slots,
1000 	 * but only a few times per ring or when NS_REPORT is set.
1001 	 *
1002 	 * Finally, on 10G and faster drivers, it might be useful
1003 	 * to prefetch the next slot and txr entry.
1004 	 */
1005 
1006 	nm_i = kring->nr_hwcur;
1007 	if (nm_i != head) {	/* we have new packets to send */
1008 		uint32_t pkt_len = 0, seg_idx = 0;
1009 		int nic_i_start = -1, flags = 0;
1010 		pkt_info_zero(&pi);
1011 		pi.ipi_segs = txq->ift_segs;
1012 		pi.ipi_qsidx = kring->ring_id;
1013 		nic_i = netmap_idx_k2n(kring, nm_i);
1014 
1015 		__builtin_prefetch(&ring->slot[nm_i]);
1016 		__builtin_prefetch(&txq->ift_sds.ifsd_m[nic_i]);
1017 		__builtin_prefetch(&txq->ift_sds.ifsd_map[nic_i]);
1018 
1019 		for (n = 0; nm_i != head; n++) {
1020 			struct netmap_slot *slot = &ring->slot[nm_i];
1021 			u_int len = slot->len;
1022 			uint64_t paddr;
1023 			void *addr = PNMB(na, slot, &paddr);
1024 
1025 			flags |= (slot->flags & NS_REPORT ||
1026 				nic_i == 0 || nic_i == report_frequency) ?
1027 				IPI_TX_INTR : 0;
1028 
1029 			/*
1030 			 * If this is the first packet fragment, save the
1031 			 * index of the first NIC slot for later.
1032 			 */
1033 			if (nic_i_start < 0)
1034 				nic_i_start = nic_i;
1035 
1036 			pi.ipi_segs[seg_idx].ds_addr = paddr;
1037 			pi.ipi_segs[seg_idx].ds_len = len;
1038 			if (len) {
1039 				pkt_len += len;
1040 				seg_idx++;
1041 			}
1042 
1043 			if (!(slot->flags & NS_MOREFRAG)) {
1044 				pi.ipi_len = pkt_len;
1045 				pi.ipi_nsegs = seg_idx;
1046 				pi.ipi_pidx = nic_i_start;
1047 				pi.ipi_ndescs = 0;
1048 				pi.ipi_flags = flags;
1049 
1050 				/* Prepare the NIC TX ring. */
1051 				ctx->isc_txd_encap(ctx->ifc_softc, &pi);
1052 				DBG_COUNTER_INC(tx_encap);
1053 
1054 				/* Reinit per-packet info for the next one. */
1055 				flags = seg_idx = pkt_len = 0;
1056 				nic_i_start = -1;
1057 			}
1058 
1059 			/* prefetch for next round */
1060 			__builtin_prefetch(&ring->slot[nm_i + 1]);
1061 			__builtin_prefetch(&txq->ift_sds.ifsd_m[nic_i + 1]);
1062 			__builtin_prefetch(&txq->ift_sds.ifsd_map[nic_i + 1]);
1063 
1064 			NM_CHECK_ADDR_LEN(na, addr, len);
1065 
1066 			if (slot->flags & NS_BUF_CHANGED) {
1067 				/* buffer has changed, reload map */
1068 				netmap_reload_map(na, txq->ift_buf_tag,
1069 				    txq->ift_sds.ifsd_map[nic_i], addr);
1070 			}
1071 			/* make sure changes to the buffer are synced */
1072 			bus_dmamap_sync(txq->ift_buf_tag,
1073 			    txq->ift_sds.ifsd_map[nic_i],
1074 			    BUS_DMASYNC_PREWRITE);
1075 
1076 			slot->flags &= ~(NS_REPORT | NS_BUF_CHANGED | NS_MOREFRAG);
1077 			nm_i = nm_next(nm_i, lim);
1078 			nic_i = nm_next(nic_i, lim);
1079 		}
1080 		kring->nr_hwcur = nm_i;
1081 
1082 		/* synchronize the NIC ring */
1083 		bus_dmamap_sync(txq->ift_ifdi->idi_tag, txq->ift_ifdi->idi_map,
1084 		    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
1085 
1086 		/* (re)start the tx unit up to slot nic_i (excluded) */
1087 		ctx->isc_txd_flush(ctx->ifc_softc, txq->ift_id, nic_i);
1088 	}
1089 
1090 	/*
1091 	 * Second part: reclaim buffers for completed transmissions.
1092 	 *
1093 	 * If there are unclaimed buffers, attempt to reclaim them.
1094 	 * If we don't manage to reclaim them all, and TX IRQs are not in use,
1095 	 * trigger a per-tx-queue timer to try again later.
1096 	 */
1097 	if (kring->nr_hwtail != nm_prev(kring->nr_hwcur, lim)) {
1098 		if (iflib_tx_credits_update(ctx, txq)) {
1099 			/* some tx completed, increment avail */
1100 			nic_i = txq->ift_cidx_processed;
1101 			kring->nr_hwtail = nm_prev(netmap_idx_n2k(kring, nic_i), lim);
1102 		}
1103 	}
1104 
1105 	if (!(ctx->ifc_flags & IFC_NETMAP_TX_IRQ))
1106 		if (kring->nr_hwtail != nm_prev(kring->nr_hwcur, lim)) {
1107 			callout_reset_sbt_on(&txq->ift_netmap_timer,
1108 			    NETMAP_TX_TIMER_US * SBT_1US, SBT_1US,
1109 			    iflib_netmap_timer, txq,
1110 			    txq->ift_netmap_timer.c_cpu, 0);
1111 		}
1112 	return (0);
1113 }
1114 
1115 /*
1116  * Reconcile kernel and user view of the receive ring.
1117  * Same as for the txsync, this routine must be efficient.
1118  * The caller guarantees a single invocations, but races against
1119  * the rest of the driver should be handled here.
1120  *
1121  * On call, kring->rhead is the first packet that userspace wants
1122  * to keep, and kring->rcur is the wakeup point.
1123  * The kernel has previously reported packets up to kring->rtail.
1124  *
1125  * If (flags & NAF_FORCE_READ) also check for incoming packets irrespective
1126  * of whether or not we received an interrupt.
1127  */
1128 static int
1129 iflib_netmap_rxsync(struct netmap_kring *kring, int flags)
1130 {
1131 	struct netmap_adapter *na = kring->na;
1132 	struct netmap_ring *ring = kring->ring;
1133 	if_t ifp = na->ifp;
1134 	uint32_t nm_i;	/* index into the netmap ring */
1135 	uint32_t nic_i;	/* index into the NIC ring */
1136 	u_int n;
1137 	u_int const lim = kring->nkr_num_slots - 1;
1138 	int force_update = (flags & NAF_FORCE_READ) || kring->nr_kflags & NKR_PENDINTR;
1139 	int i = 0;
1140 
1141 	if_ctx_t ctx = ifp->if_softc;
1142 	if_shared_ctx_t sctx = ctx->ifc_sctx;
1143 	if_softc_ctx_t scctx = &ctx->ifc_softc_ctx;
1144 	iflib_rxq_t rxq = &ctx->ifc_rxqs[kring->ring_id];
1145 	iflib_fl_t fl = &rxq->ifr_fl[0];
1146 	struct if_rxd_info ri;
1147 	qidx_t *cidxp;
1148 
1149 	/*
1150 	 * netmap only uses free list 0, to avoid out of order consumption
1151 	 * of receive buffers
1152 	 */
1153 
1154 	bus_dmamap_sync(fl->ifl_ifdi->idi_tag, fl->ifl_ifdi->idi_map,
1155 	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
1156 
1157 	/*
1158 	 * First part: import newly received packets.
1159 	 *
1160 	 * nm_i is the index of the next free slot in the netmap ring,
1161 	 * nic_i is the index of the next received packet in the NIC ring
1162 	 * (or in the free list 0 if IFLIB_HAS_RXCQ is set), and they may
1163 	 * differ in case if_init() has been called while
1164 	 * in netmap mode. For the receive ring we have
1165 	 *
1166 	 *	nic_i = fl->ifl_cidx;
1167 	 *	nm_i = kring->nr_hwtail (previous)
1168 	 * and
1169 	 *	nm_i == (nic_i + kring->nkr_hwofs) % ring_size
1170 	 *
1171 	 * fl->ifl_cidx is set to 0 on a ring reinit
1172 	 */
1173 	if (netmap_no_pendintr || force_update) {
1174 		uint32_t hwtail_lim = nm_prev(kring->nr_hwcur, lim);
1175 		bool have_rxcq = sctx->isc_flags & IFLIB_HAS_RXCQ;
1176 		int crclen = iflib_crcstrip ? 0 : 4;
1177 		int error, avail;
1178 
1179 		/*
1180 		 * For the free list consumer index, we use the same
1181 		 * logic as in iflib_rxeof().
1182 		 */
1183 		if (have_rxcq)
1184 			cidxp = &rxq->ifr_cq_cidx;
1185 		else
1186 			cidxp = &fl->ifl_cidx;
1187 		avail = ctx->isc_rxd_available(ctx->ifc_softc,
1188 		    rxq->ifr_id, *cidxp, USHRT_MAX);
1189 
1190 		nic_i = fl->ifl_cidx;
1191 		nm_i = netmap_idx_n2k(kring, nic_i);
1192 		MPASS(nm_i == kring->nr_hwtail);
1193 		for (n = 0; avail > 0 && nm_i != hwtail_lim; n++, avail--) {
1194 			rxd_info_zero(&ri);
1195 			ri.iri_frags = rxq->ifr_frags;
1196 			ri.iri_qsidx = kring->ring_id;
1197 			ri.iri_ifp = ctx->ifc_ifp;
1198 			ri.iri_cidx = *cidxp;
1199 
1200 			error = ctx->isc_rxd_pkt_get(ctx->ifc_softc, &ri);
1201 			for (i = 0; i < ri.iri_nfrags; i++) {
1202 				if (error) {
1203 					ring->slot[nm_i].len = 0;
1204 					ring->slot[nm_i].flags = 0;
1205 				} else {
1206 					ring->slot[nm_i].len = ri.iri_frags[i].irf_len;
1207 					if (i == (ri.iri_nfrags - 1)) {
1208 						ring->slot[nm_i].len -= crclen;
1209 						ring->slot[nm_i].flags = 0;
1210 					} else
1211 						ring->slot[nm_i].flags = NS_MOREFRAG;
1212 				}
1213 
1214 				bus_dmamap_sync(fl->ifl_buf_tag,
1215 				    fl->ifl_sds.ifsd_map[nic_i], BUS_DMASYNC_POSTREAD);
1216 				nm_i = nm_next(nm_i, lim);
1217 				fl->ifl_cidx = nic_i = nm_next(nic_i, lim);
1218 			}
1219 
1220 			if (have_rxcq) {
1221 				*cidxp = ri.iri_cidx;
1222 				while (*cidxp >= scctx->isc_nrxd[0])
1223 					*cidxp -= scctx->isc_nrxd[0];
1224 			}
1225 
1226 		}
1227 		if (n) { /* update the state variables */
1228 			if (netmap_no_pendintr && !force_update) {
1229 				/* diagnostics */
1230 				iflib_rx_miss ++;
1231 				iflib_rx_miss_bufs += n;
1232 			}
1233 			kring->nr_hwtail = nm_i;
1234 		}
1235 		kring->nr_kflags &= ~NKR_PENDINTR;
1236 	}
1237 	/*
1238 	 * Second part: skip past packets that userspace has released.
1239 	 * (kring->nr_hwcur to head excluded),
1240 	 * and make the buffers available for reception.
1241 	 * As usual nm_i is the index in the netmap ring,
1242 	 * nic_i is the index in the NIC ring, and
1243 	 * nm_i == (nic_i + kring->nkr_hwofs) % ring_size
1244 	 */
1245 	netmap_fl_refill(rxq, kring, false);
1246 
1247 	return (0);
1248 }
1249 
1250 static void
1251 iflib_netmap_intr(struct netmap_adapter *na, int onoff)
1252 {
1253 	if_ctx_t ctx = na->ifp->if_softc;
1254 
1255 	CTX_LOCK(ctx);
1256 	if (onoff) {
1257 		IFDI_INTR_ENABLE(ctx);
1258 	} else {
1259 		IFDI_INTR_DISABLE(ctx);
1260 	}
1261 	CTX_UNLOCK(ctx);
1262 }
1263 
1264 static int
1265 iflib_netmap_attach(if_ctx_t ctx)
1266 {
1267 	struct netmap_adapter na;
1268 
1269 	bzero(&na, sizeof(na));
1270 
1271 	na.ifp = ctx->ifc_ifp;
1272 	na.na_flags = NAF_BDG_MAYSLEEP | NAF_MOREFRAG;
1273 	MPASS(ctx->ifc_softc_ctx.isc_ntxqsets);
1274 	MPASS(ctx->ifc_softc_ctx.isc_nrxqsets);
1275 
1276 	na.num_tx_desc = iflib_num_tx_descs(ctx);
1277 	na.num_rx_desc = iflib_num_rx_descs(ctx);
1278 	na.nm_txsync = iflib_netmap_txsync;
1279 	na.nm_rxsync = iflib_netmap_rxsync;
1280 	na.nm_register = iflib_netmap_register;
1281 	na.nm_intr = iflib_netmap_intr;
1282 	na.num_tx_rings = ctx->ifc_softc_ctx.isc_ntxqsets;
1283 	na.num_rx_rings = ctx->ifc_softc_ctx.isc_nrxqsets;
1284 	return (netmap_attach(&na));
1285 }
1286 
1287 static int
1288 iflib_netmap_txq_init(if_ctx_t ctx, iflib_txq_t txq)
1289 {
1290 	struct netmap_adapter *na = NA(ctx->ifc_ifp);
1291 	struct netmap_slot *slot;
1292 
1293 	slot = netmap_reset(na, NR_TX, txq->ift_id, 0);
1294 	if (slot == NULL)
1295 		return (0);
1296 	for (int i = 0; i < ctx->ifc_softc_ctx.isc_ntxd[0]; i++) {
1297 		/*
1298 		 * In netmap mode, set the map for the packet buffer.
1299 		 * NOTE: Some drivers (not this one) also need to set
1300 		 * the physical buffer address in the NIC ring.
1301 		 * netmap_idx_n2k() maps a nic index, i, into the corresponding
1302 		 * netmap slot index, si
1303 		 */
1304 		int si = netmap_idx_n2k(na->tx_rings[txq->ift_id], i);
1305 		netmap_load_map(na, txq->ift_buf_tag, txq->ift_sds.ifsd_map[i],
1306 		    NMB(na, slot + si));
1307 	}
1308 	return (1);
1309 }
1310 
1311 static int
1312 iflib_netmap_rxq_init(if_ctx_t ctx, iflib_rxq_t rxq)
1313 {
1314 	struct netmap_adapter *na = NA(ctx->ifc_ifp);
1315 	struct netmap_kring *kring;
1316 	struct netmap_slot *slot;
1317 
1318 	slot = netmap_reset(na, NR_RX, rxq->ifr_id, 0);
1319 	if (slot == NULL)
1320 		return (0);
1321 	kring = na->rx_rings[rxq->ifr_id];
1322 	netmap_fl_refill(rxq, kring, true);
1323 	return (1);
1324 }
1325 
1326 static void
1327 iflib_netmap_timer(void *arg)
1328 {
1329 	iflib_txq_t txq = arg;
1330 	if_ctx_t ctx = txq->ift_ctx;
1331 
1332 	/*
1333 	 * Wake up the netmap application, to give it a chance to
1334 	 * call txsync and reclaim more completed TX buffers.
1335 	 */
1336 	netmap_tx_irq(ctx->ifc_ifp, txq->ift_id);
1337 }
1338 
1339 #define iflib_netmap_detach(ifp) netmap_detach(ifp)
1340 
1341 #else
1342 #define iflib_netmap_txq_init(ctx, txq) (0)
1343 #define iflib_netmap_rxq_init(ctx, rxq) (0)
1344 #define iflib_netmap_detach(ifp)
1345 #define netmap_enable_all_rings(ifp)
1346 #define netmap_disable_all_rings(ifp)
1347 
1348 #define iflib_netmap_attach(ctx) (0)
1349 #define netmap_rx_irq(ifp, qid, budget) (0)
1350 #endif
1351 
1352 #if defined(__i386__) || defined(__amd64__)
1353 static __inline void
1354 prefetch(void *x)
1355 {
1356 	__asm volatile("prefetcht0 %0" :: "m" (*(unsigned long *)x));
1357 }
1358 static __inline void
1359 prefetch2cachelines(void *x)
1360 {
1361 	__asm volatile("prefetcht0 %0" :: "m" (*(unsigned long *)x));
1362 #if (CACHE_LINE_SIZE < 128)
1363 	__asm volatile("prefetcht0 %0" :: "m" (*(((unsigned long *)x)+CACHE_LINE_SIZE/(sizeof(unsigned long)))));
1364 #endif
1365 }
1366 #else
1367 #define prefetch(x)
1368 #define prefetch2cachelines(x)
1369 #endif
1370 
1371 static void
1372 iru_init(if_rxd_update_t iru, iflib_rxq_t rxq, uint8_t flid)
1373 {
1374 	iflib_fl_t fl;
1375 
1376 	fl = &rxq->ifr_fl[flid];
1377 	iru->iru_paddrs = fl->ifl_bus_addrs;
1378 	iru->iru_idxs = fl->ifl_rxd_idxs;
1379 	iru->iru_qsidx = rxq->ifr_id;
1380 	iru->iru_buf_size = fl->ifl_buf_size;
1381 	iru->iru_flidx = fl->ifl_id;
1382 }
1383 
1384 static void
1385 _iflib_dmamap_cb(void *arg, bus_dma_segment_t *segs, int nseg, int err)
1386 {
1387 	if (err)
1388 		return;
1389 	*(bus_addr_t *) arg = segs[0].ds_addr;
1390 }
1391 
1392 int
1393 iflib_dma_alloc_align(if_ctx_t ctx, int size, int align, iflib_dma_info_t dma, int mapflags)
1394 {
1395 	int err;
1396 	device_t dev = ctx->ifc_dev;
1397 
1398 	err = bus_dma_tag_create(bus_get_dma_tag(dev),	/* parent */
1399 				align, 0,		/* alignment, bounds */
1400 				BUS_SPACE_MAXADDR,	/* lowaddr */
1401 				BUS_SPACE_MAXADDR,	/* highaddr */
1402 				NULL, NULL,		/* filter, filterarg */
1403 				size,			/* maxsize */
1404 				1,			/* nsegments */
1405 				size,			/* maxsegsize */
1406 				BUS_DMA_ALLOCNOW,	/* flags */
1407 				NULL,			/* lockfunc */
1408 				NULL,			/* lockarg */
1409 				&dma->idi_tag);
1410 	if (err) {
1411 		device_printf(dev,
1412 		    "%s: bus_dma_tag_create failed: %d\n",
1413 		    __func__, err);
1414 		goto fail_0;
1415 	}
1416 
1417 	err = bus_dmamem_alloc(dma->idi_tag, (void**) &dma->idi_vaddr,
1418 	    BUS_DMA_NOWAIT | BUS_DMA_COHERENT | BUS_DMA_ZERO, &dma->idi_map);
1419 	if (err) {
1420 		device_printf(dev,
1421 		    "%s: bus_dmamem_alloc(%ju) failed: %d\n",
1422 		    __func__, (uintmax_t)size, err);
1423 		goto fail_1;
1424 	}
1425 
1426 	dma->idi_paddr = IF_BAD_DMA;
1427 	err = bus_dmamap_load(dma->idi_tag, dma->idi_map, dma->idi_vaddr,
1428 	    size, _iflib_dmamap_cb, &dma->idi_paddr, mapflags | BUS_DMA_NOWAIT);
1429 	if (err || dma->idi_paddr == IF_BAD_DMA) {
1430 		device_printf(dev,
1431 		    "%s: bus_dmamap_load failed: %d\n",
1432 		    __func__, err);
1433 		goto fail_2;
1434 	}
1435 
1436 	dma->idi_size = size;
1437 	return (0);
1438 
1439 fail_2:
1440 	bus_dmamem_free(dma->idi_tag, dma->idi_vaddr, dma->idi_map);
1441 fail_1:
1442 	bus_dma_tag_destroy(dma->idi_tag);
1443 fail_0:
1444 	dma->idi_tag = NULL;
1445 
1446 	return (err);
1447 }
1448 
1449 int
1450 iflib_dma_alloc(if_ctx_t ctx, int size, iflib_dma_info_t dma, int mapflags)
1451 {
1452 	if_shared_ctx_t sctx = ctx->ifc_sctx;
1453 
1454 	KASSERT(sctx->isc_q_align != 0, ("alignment value not initialized"));
1455 
1456 	return (iflib_dma_alloc_align(ctx, size, sctx->isc_q_align, dma, mapflags));
1457 }
1458 
1459 int
1460 iflib_dma_alloc_multi(if_ctx_t ctx, int *sizes, iflib_dma_info_t *dmalist, int mapflags, int count)
1461 {
1462 	int i, err;
1463 	iflib_dma_info_t *dmaiter;
1464 
1465 	dmaiter = dmalist;
1466 	for (i = 0; i < count; i++, dmaiter++) {
1467 		if ((err = iflib_dma_alloc(ctx, sizes[i], *dmaiter, mapflags)) != 0)
1468 			break;
1469 	}
1470 	if (err)
1471 		iflib_dma_free_multi(dmalist, i);
1472 	return (err);
1473 }
1474 
1475 void
1476 iflib_dma_free(iflib_dma_info_t dma)
1477 {
1478 	if (dma->idi_tag == NULL)
1479 		return;
1480 	if (dma->idi_paddr != IF_BAD_DMA) {
1481 		bus_dmamap_sync(dma->idi_tag, dma->idi_map,
1482 		    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
1483 		bus_dmamap_unload(dma->idi_tag, dma->idi_map);
1484 		dma->idi_paddr = IF_BAD_DMA;
1485 	}
1486 	if (dma->idi_vaddr != NULL) {
1487 		bus_dmamem_free(dma->idi_tag, dma->idi_vaddr, dma->idi_map);
1488 		dma->idi_vaddr = NULL;
1489 	}
1490 	bus_dma_tag_destroy(dma->idi_tag);
1491 	dma->idi_tag = NULL;
1492 }
1493 
1494 void
1495 iflib_dma_free_multi(iflib_dma_info_t *dmalist, int count)
1496 {
1497 	int i;
1498 	iflib_dma_info_t *dmaiter = dmalist;
1499 
1500 	for (i = 0; i < count; i++, dmaiter++)
1501 		iflib_dma_free(*dmaiter);
1502 }
1503 
1504 static int
1505 iflib_fast_intr(void *arg)
1506 {
1507 	iflib_filter_info_t info = arg;
1508 	struct grouptask *gtask = info->ifi_task;
1509 	int result;
1510 
1511 	DBG_COUNTER_INC(fast_intrs);
1512 	if (info->ifi_filter != NULL) {
1513 		result = info->ifi_filter(info->ifi_filter_arg);
1514 		if ((result & FILTER_SCHEDULE_THREAD) == 0)
1515 			return (result);
1516 	}
1517 
1518 	GROUPTASK_ENQUEUE(gtask);
1519 	return (FILTER_HANDLED);
1520 }
1521 
1522 static int
1523 iflib_fast_intr_rxtx(void *arg)
1524 {
1525 	iflib_filter_info_t info = arg;
1526 	struct grouptask *gtask = info->ifi_task;
1527 	if_ctx_t ctx;
1528 	iflib_rxq_t rxq = (iflib_rxq_t)info->ifi_ctx;
1529 	iflib_txq_t txq;
1530 	void *sc;
1531 	int i, cidx, result;
1532 	qidx_t txqid;
1533 	bool intr_enable, intr_legacy;
1534 
1535 	DBG_COUNTER_INC(fast_intrs);
1536 	if (info->ifi_filter != NULL) {
1537 		result = info->ifi_filter(info->ifi_filter_arg);
1538 		if ((result & FILTER_SCHEDULE_THREAD) == 0)
1539 			return (result);
1540 	}
1541 
1542 	ctx = rxq->ifr_ctx;
1543 	sc = ctx->ifc_softc;
1544 	intr_enable = false;
1545 	intr_legacy = !!(ctx->ifc_flags & IFC_LEGACY);
1546 	MPASS(rxq->ifr_ntxqirq);
1547 	for (i = 0; i < rxq->ifr_ntxqirq; i++) {
1548 		txqid = rxq->ifr_txqid[i];
1549 		txq = &ctx->ifc_txqs[txqid];
1550 		bus_dmamap_sync(txq->ift_ifdi->idi_tag, txq->ift_ifdi->idi_map,
1551 		    BUS_DMASYNC_POSTREAD);
1552 		if (!ctx->isc_txd_credits_update(sc, txqid, false)) {
1553 			if (intr_legacy)
1554 				intr_enable = true;
1555 			else
1556 				IFDI_TX_QUEUE_INTR_ENABLE(ctx, txqid);
1557 			continue;
1558 		}
1559 		GROUPTASK_ENQUEUE(&txq->ift_task);
1560 	}
1561 	if (ctx->ifc_sctx->isc_flags & IFLIB_HAS_RXCQ)
1562 		cidx = rxq->ifr_cq_cidx;
1563 	else
1564 		cidx = rxq->ifr_fl[0].ifl_cidx;
1565 	if (iflib_rxd_avail(ctx, rxq, cidx, 1))
1566 		GROUPTASK_ENQUEUE(gtask);
1567 	else {
1568 		if (intr_legacy)
1569 			intr_enable = true;
1570 		else
1571 			IFDI_RX_QUEUE_INTR_ENABLE(ctx, rxq->ifr_id);
1572 		DBG_COUNTER_INC(rx_intr_enables);
1573 	}
1574 	if (intr_enable)
1575 		IFDI_INTR_ENABLE(ctx);
1576 	return (FILTER_HANDLED);
1577 }
1578 
1579 static int
1580 iflib_fast_intr_ctx(void *arg)
1581 {
1582 	iflib_filter_info_t info = arg;
1583 	struct grouptask *gtask = info->ifi_task;
1584 	int result;
1585 
1586 	DBG_COUNTER_INC(fast_intrs);
1587 	if (info->ifi_filter != NULL) {
1588 		result = info->ifi_filter(info->ifi_filter_arg);
1589 		if ((result & FILTER_SCHEDULE_THREAD) == 0)
1590 			return (result);
1591 	}
1592 
1593 	GROUPTASK_ENQUEUE(gtask);
1594 	return (FILTER_HANDLED);
1595 }
1596 
1597 static int
1598 _iflib_irq_alloc(if_ctx_t ctx, if_irq_t irq, int rid,
1599 		 driver_filter_t filter, driver_intr_t handler, void *arg,
1600 		 const char *name)
1601 {
1602 	struct resource *res;
1603 	void *tag = NULL;
1604 	device_t dev = ctx->ifc_dev;
1605 	int flags, i, rc;
1606 
1607 	flags = RF_ACTIVE;
1608 	if (ctx->ifc_flags & IFC_LEGACY)
1609 		flags |= RF_SHAREABLE;
1610 	MPASS(rid < 512);
1611 	i = rid;
1612 	res = bus_alloc_resource_any(dev, SYS_RES_IRQ, &i, flags);
1613 	if (res == NULL) {
1614 		device_printf(dev,
1615 		    "failed to allocate IRQ for rid %d, name %s.\n", rid, name);
1616 		return (ENOMEM);
1617 	}
1618 	irq->ii_res = res;
1619 	KASSERT(filter == NULL || handler == NULL, ("filter and handler can't both be non-NULL"));
1620 	rc = bus_setup_intr(dev, res, INTR_MPSAFE | INTR_TYPE_NET,
1621 						filter, handler, arg, &tag);
1622 	if (rc != 0) {
1623 		device_printf(dev,
1624 		    "failed to setup interrupt for rid %d, name %s: %d\n",
1625 					  rid, name ? name : "unknown", rc);
1626 		return (rc);
1627 	} else if (name)
1628 		bus_describe_intr(dev, res, tag, "%s", name);
1629 
1630 	irq->ii_tag = tag;
1631 	return (0);
1632 }
1633 
1634 /*********************************************************************
1635  *
1636  *  Allocate DMA resources for TX buffers as well as memory for the TX
1637  *  mbuf map.  TX DMA maps (non-TSO/TSO) and TX mbuf map are kept in a
1638  *  iflib_sw_tx_desc_array structure, storing all the information that
1639  *  is needed to transmit a packet on the wire.  This is called only
1640  *  once at attach, setup is done every reset.
1641  *
1642  **********************************************************************/
1643 static int
1644 iflib_txsd_alloc(iflib_txq_t txq)
1645 {
1646 	if_ctx_t ctx = txq->ift_ctx;
1647 	if_shared_ctx_t sctx = ctx->ifc_sctx;
1648 	if_softc_ctx_t scctx = &ctx->ifc_softc_ctx;
1649 	device_t dev = ctx->ifc_dev;
1650 	bus_size_t tsomaxsize;
1651 	int err, nsegments, ntsosegments;
1652 	bool tso;
1653 
1654 	nsegments = scctx->isc_tx_nsegments;
1655 	ntsosegments = scctx->isc_tx_tso_segments_max;
1656 	tsomaxsize = scctx->isc_tx_tso_size_max;
1657 	if (if_getcapabilities(ctx->ifc_ifp) & IFCAP_VLAN_MTU)
1658 		tsomaxsize += sizeof(struct ether_vlan_header);
1659 	MPASS(scctx->isc_ntxd[0] > 0);
1660 	MPASS(scctx->isc_ntxd[txq->ift_br_offset] > 0);
1661 	MPASS(nsegments > 0);
1662 	if (if_getcapabilities(ctx->ifc_ifp) & IFCAP_TSO) {
1663 		MPASS(ntsosegments > 0);
1664 		MPASS(sctx->isc_tso_maxsize >= tsomaxsize);
1665 	}
1666 
1667 	/*
1668 	 * Set up DMA tags for TX buffers.
1669 	 */
1670 	if ((err = bus_dma_tag_create(bus_get_dma_tag(dev),
1671 			       1, 0,			/* alignment, bounds */
1672 			       BUS_SPACE_MAXADDR,	/* lowaddr */
1673 			       BUS_SPACE_MAXADDR,	/* highaddr */
1674 			       NULL, NULL,		/* filter, filterarg */
1675 			       sctx->isc_tx_maxsize,		/* maxsize */
1676 			       nsegments,	/* nsegments */
1677 			       sctx->isc_tx_maxsegsize,	/* maxsegsize */
1678 			       0,			/* flags */
1679 			       NULL,			/* lockfunc */
1680 			       NULL,			/* lockfuncarg */
1681 			       &txq->ift_buf_tag))) {
1682 		device_printf(dev,"Unable to allocate TX DMA tag: %d\n", err);
1683 		device_printf(dev,"maxsize: %ju nsegments: %d maxsegsize: %ju\n",
1684 		    (uintmax_t)sctx->isc_tx_maxsize, nsegments, (uintmax_t)sctx->isc_tx_maxsegsize);
1685 		goto fail;
1686 	}
1687 	tso = (if_getcapabilities(ctx->ifc_ifp) & IFCAP_TSO) != 0;
1688 	if (tso && (err = bus_dma_tag_create(bus_get_dma_tag(dev),
1689 			       1, 0,			/* alignment, bounds */
1690 			       BUS_SPACE_MAXADDR,	/* lowaddr */
1691 			       BUS_SPACE_MAXADDR,	/* highaddr */
1692 			       NULL, NULL,		/* filter, filterarg */
1693 			       tsomaxsize,		/* maxsize */
1694 			       ntsosegments,	/* nsegments */
1695 			       sctx->isc_tso_maxsegsize,/* maxsegsize */
1696 			       0,			/* flags */
1697 			       NULL,			/* lockfunc */
1698 			       NULL,			/* lockfuncarg */
1699 			       &txq->ift_tso_buf_tag))) {
1700 		device_printf(dev, "Unable to allocate TSO TX DMA tag: %d\n",
1701 		    err);
1702 		goto fail;
1703 	}
1704 
1705 	/* Allocate memory for the TX mbuf map. */
1706 	if (!(txq->ift_sds.ifsd_m =
1707 	    (struct mbuf **) malloc(sizeof(struct mbuf *) *
1708 	    scctx->isc_ntxd[txq->ift_br_offset], M_IFLIB, M_NOWAIT | M_ZERO))) {
1709 		device_printf(dev, "Unable to allocate TX mbuf map memory\n");
1710 		err = ENOMEM;
1711 		goto fail;
1712 	}
1713 
1714 	/*
1715 	 * Create the DMA maps for TX buffers.
1716 	 */
1717 	if ((txq->ift_sds.ifsd_map = (bus_dmamap_t *)malloc(
1718 	    sizeof(bus_dmamap_t) * scctx->isc_ntxd[txq->ift_br_offset],
1719 	    M_IFLIB, M_NOWAIT | M_ZERO)) == NULL) {
1720 		device_printf(dev,
1721 		    "Unable to allocate TX buffer DMA map memory\n");
1722 		err = ENOMEM;
1723 		goto fail;
1724 	}
1725 	if (tso && (txq->ift_sds.ifsd_tso_map = (bus_dmamap_t *)malloc(
1726 	    sizeof(bus_dmamap_t) * scctx->isc_ntxd[txq->ift_br_offset],
1727 	    M_IFLIB, M_NOWAIT | M_ZERO)) == NULL) {
1728 		device_printf(dev,
1729 		    "Unable to allocate TSO TX buffer map memory\n");
1730 		err = ENOMEM;
1731 		goto fail;
1732 	}
1733 	for (int i = 0; i < scctx->isc_ntxd[txq->ift_br_offset]; i++) {
1734 		err = bus_dmamap_create(txq->ift_buf_tag, 0,
1735 		    &txq->ift_sds.ifsd_map[i]);
1736 		if (err != 0) {
1737 			device_printf(dev, "Unable to create TX DMA map\n");
1738 			goto fail;
1739 		}
1740 		if (!tso)
1741 			continue;
1742 		err = bus_dmamap_create(txq->ift_tso_buf_tag, 0,
1743 		    &txq->ift_sds.ifsd_tso_map[i]);
1744 		if (err != 0) {
1745 			device_printf(dev, "Unable to create TSO TX DMA map\n");
1746 			goto fail;
1747 		}
1748 	}
1749 	return (0);
1750 fail:
1751 	/* We free all, it handles case where we are in the middle */
1752 	iflib_tx_structures_free(ctx);
1753 	return (err);
1754 }
1755 
1756 static void
1757 iflib_txsd_destroy(if_ctx_t ctx, iflib_txq_t txq, int i)
1758 {
1759 	bus_dmamap_t map;
1760 
1761 	if (txq->ift_sds.ifsd_map != NULL) {
1762 		map = txq->ift_sds.ifsd_map[i];
1763 		bus_dmamap_sync(txq->ift_buf_tag, map, BUS_DMASYNC_POSTWRITE);
1764 		bus_dmamap_unload(txq->ift_buf_tag, map);
1765 		bus_dmamap_destroy(txq->ift_buf_tag, map);
1766 		txq->ift_sds.ifsd_map[i] = NULL;
1767 	}
1768 
1769 	if (txq->ift_sds.ifsd_tso_map != NULL) {
1770 		map = txq->ift_sds.ifsd_tso_map[i];
1771 		bus_dmamap_sync(txq->ift_tso_buf_tag, map,
1772 		    BUS_DMASYNC_POSTWRITE);
1773 		bus_dmamap_unload(txq->ift_tso_buf_tag, map);
1774 		bus_dmamap_destroy(txq->ift_tso_buf_tag, map);
1775 		txq->ift_sds.ifsd_tso_map[i] = NULL;
1776 	}
1777 }
1778 
1779 static void
1780 iflib_txq_destroy(iflib_txq_t txq)
1781 {
1782 	if_ctx_t ctx = txq->ift_ctx;
1783 
1784 	for (int i = 0; i < txq->ift_size; i++)
1785 		iflib_txsd_destroy(ctx, txq, i);
1786 
1787 	if (txq->ift_br != NULL) {
1788 		ifmp_ring_free(txq->ift_br);
1789 		txq->ift_br = NULL;
1790 	}
1791 
1792 	mtx_destroy(&txq->ift_mtx);
1793 
1794 	if (txq->ift_sds.ifsd_map != NULL) {
1795 		free(txq->ift_sds.ifsd_map, M_IFLIB);
1796 		txq->ift_sds.ifsd_map = NULL;
1797 	}
1798 	if (txq->ift_sds.ifsd_tso_map != NULL) {
1799 		free(txq->ift_sds.ifsd_tso_map, M_IFLIB);
1800 		txq->ift_sds.ifsd_tso_map = NULL;
1801 	}
1802 	if (txq->ift_sds.ifsd_m != NULL) {
1803 		free(txq->ift_sds.ifsd_m, M_IFLIB);
1804 		txq->ift_sds.ifsd_m = NULL;
1805 	}
1806 	if (txq->ift_buf_tag != NULL) {
1807 		bus_dma_tag_destroy(txq->ift_buf_tag);
1808 		txq->ift_buf_tag = NULL;
1809 	}
1810 	if (txq->ift_tso_buf_tag != NULL) {
1811 		bus_dma_tag_destroy(txq->ift_tso_buf_tag);
1812 		txq->ift_tso_buf_tag = NULL;
1813 	}
1814 	if (txq->ift_ifdi != NULL) {
1815 		free(txq->ift_ifdi, M_IFLIB);
1816 	}
1817 }
1818 
1819 static void
1820 iflib_txsd_free(if_ctx_t ctx, iflib_txq_t txq, int i)
1821 {
1822 	struct mbuf **mp;
1823 
1824 	mp = &txq->ift_sds.ifsd_m[i];
1825 	if (*mp == NULL)
1826 		return;
1827 
1828 	if (txq->ift_sds.ifsd_map != NULL) {
1829 		bus_dmamap_sync(txq->ift_buf_tag,
1830 		    txq->ift_sds.ifsd_map[i], BUS_DMASYNC_POSTWRITE);
1831 		bus_dmamap_unload(txq->ift_buf_tag, txq->ift_sds.ifsd_map[i]);
1832 	}
1833 	if (txq->ift_sds.ifsd_tso_map != NULL) {
1834 		bus_dmamap_sync(txq->ift_tso_buf_tag,
1835 		    txq->ift_sds.ifsd_tso_map[i], BUS_DMASYNC_POSTWRITE);
1836 		bus_dmamap_unload(txq->ift_tso_buf_tag,
1837 		    txq->ift_sds.ifsd_tso_map[i]);
1838 	}
1839 	m_freem(*mp);
1840 	DBG_COUNTER_INC(tx_frees);
1841 	*mp = NULL;
1842 }
1843 
1844 static int
1845 iflib_txq_setup(iflib_txq_t txq)
1846 {
1847 	if_ctx_t ctx = txq->ift_ctx;
1848 	if_softc_ctx_t scctx = &ctx->ifc_softc_ctx;
1849 	if_shared_ctx_t sctx = ctx->ifc_sctx;
1850 	iflib_dma_info_t di;
1851 	int i;
1852 
1853 	/* Set number of descriptors available */
1854 	txq->ift_qstatus = IFLIB_QUEUE_IDLE;
1855 	/* XXX make configurable */
1856 	txq->ift_update_freq = IFLIB_DEFAULT_TX_UPDATE_FREQ;
1857 
1858 	/* Reset indices */
1859 	txq->ift_cidx_processed = 0;
1860 	txq->ift_pidx = txq->ift_cidx = txq->ift_npending = 0;
1861 	txq->ift_size = scctx->isc_ntxd[txq->ift_br_offset];
1862 
1863 	for (i = 0, di = txq->ift_ifdi; i < sctx->isc_ntxqs; i++, di++)
1864 		bzero((void *)di->idi_vaddr, di->idi_size);
1865 
1866 	IFDI_TXQ_SETUP(ctx, txq->ift_id);
1867 	for (i = 0, di = txq->ift_ifdi; i < sctx->isc_ntxqs; i++, di++)
1868 		bus_dmamap_sync(di->idi_tag, di->idi_map,
1869 		    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
1870 	return (0);
1871 }
1872 
1873 /*********************************************************************
1874  *
1875  *  Allocate DMA resources for RX buffers as well as memory for the RX
1876  *  mbuf map, direct RX cluster pointer map and RX cluster bus address
1877  *  map.  RX DMA map, RX mbuf map, direct RX cluster pointer map and
1878  *  RX cluster map are kept in a iflib_sw_rx_desc_array structure.
1879  *  Since we use use one entry in iflib_sw_rx_desc_array per received
1880  *  packet, the maximum number of entries we'll need is equal to the
1881  *  number of hardware receive descriptors that we've allocated.
1882  *
1883  **********************************************************************/
1884 static int
1885 iflib_rxsd_alloc(iflib_rxq_t rxq)
1886 {
1887 	if_ctx_t ctx = rxq->ifr_ctx;
1888 	if_shared_ctx_t sctx = ctx->ifc_sctx;
1889 	if_softc_ctx_t scctx = &ctx->ifc_softc_ctx;
1890 	device_t dev = ctx->ifc_dev;
1891 	iflib_fl_t fl;
1892 	int			err;
1893 
1894 	MPASS(scctx->isc_nrxd[0] > 0);
1895 	MPASS(scctx->isc_nrxd[rxq->ifr_fl_offset] > 0);
1896 
1897 	fl = rxq->ifr_fl;
1898 	for (int i = 0; i <  rxq->ifr_nfl; i++, fl++) {
1899 		fl->ifl_size = scctx->isc_nrxd[rxq->ifr_fl_offset]; /* this isn't necessarily the same */
1900 		/* Set up DMA tag for RX buffers. */
1901 		err = bus_dma_tag_create(bus_get_dma_tag(dev), /* parent */
1902 					 1, 0,			/* alignment, bounds */
1903 					 BUS_SPACE_MAXADDR,	/* lowaddr */
1904 					 BUS_SPACE_MAXADDR,	/* highaddr */
1905 					 NULL, NULL,		/* filter, filterarg */
1906 					 sctx->isc_rx_maxsize,	/* maxsize */
1907 					 sctx->isc_rx_nsegments,	/* nsegments */
1908 					 sctx->isc_rx_maxsegsize,	/* maxsegsize */
1909 					 0,			/* flags */
1910 					 NULL,			/* lockfunc */
1911 					 NULL,			/* lockarg */
1912 					 &fl->ifl_buf_tag);
1913 		if (err) {
1914 			device_printf(dev,
1915 			    "Unable to allocate RX DMA tag: %d\n", err);
1916 			goto fail;
1917 		}
1918 
1919 		/* Allocate memory for the RX mbuf map. */
1920 		if (!(fl->ifl_sds.ifsd_m =
1921 		      (struct mbuf **) malloc(sizeof(struct mbuf *) *
1922 					      scctx->isc_nrxd[rxq->ifr_fl_offset], M_IFLIB, M_NOWAIT | M_ZERO))) {
1923 			device_printf(dev,
1924 			    "Unable to allocate RX mbuf map memory\n");
1925 			err = ENOMEM;
1926 			goto fail;
1927 		}
1928 
1929 		/* Allocate memory for the direct RX cluster pointer map. */
1930 		if (!(fl->ifl_sds.ifsd_cl =
1931 		      (caddr_t *) malloc(sizeof(caddr_t) *
1932 					      scctx->isc_nrxd[rxq->ifr_fl_offset], M_IFLIB, M_NOWAIT | M_ZERO))) {
1933 			device_printf(dev,
1934 			    "Unable to allocate RX cluster map memory\n");
1935 			err = ENOMEM;
1936 			goto fail;
1937 		}
1938 
1939 		/* Allocate memory for the RX cluster bus address map. */
1940 		if (!(fl->ifl_sds.ifsd_ba =
1941 		      (bus_addr_t *) malloc(sizeof(bus_addr_t) *
1942 					      scctx->isc_nrxd[rxq->ifr_fl_offset], M_IFLIB, M_NOWAIT | M_ZERO))) {
1943 			device_printf(dev,
1944 			    "Unable to allocate RX bus address map memory\n");
1945 			err = ENOMEM;
1946 			goto fail;
1947 		}
1948 
1949 		/*
1950 		 * Create the DMA maps for RX buffers.
1951 		 */
1952 		if (!(fl->ifl_sds.ifsd_map =
1953 		      (bus_dmamap_t *) malloc(sizeof(bus_dmamap_t) * scctx->isc_nrxd[rxq->ifr_fl_offset], M_IFLIB, M_NOWAIT | M_ZERO))) {
1954 			device_printf(dev,
1955 			    "Unable to allocate RX buffer DMA map memory\n");
1956 			err = ENOMEM;
1957 			goto fail;
1958 		}
1959 		for (int i = 0; i < scctx->isc_nrxd[rxq->ifr_fl_offset]; i++) {
1960 			err = bus_dmamap_create(fl->ifl_buf_tag, 0,
1961 			    &fl->ifl_sds.ifsd_map[i]);
1962 			if (err != 0) {
1963 				device_printf(dev, "Unable to create RX buffer DMA map\n");
1964 				goto fail;
1965 			}
1966 		}
1967 	}
1968 	return (0);
1969 
1970 fail:
1971 	iflib_rx_structures_free(ctx);
1972 	return (err);
1973 }
1974 
1975 /*
1976  * Internal service routines
1977  */
1978 
1979 struct rxq_refill_cb_arg {
1980 	int               error;
1981 	bus_dma_segment_t seg;
1982 	int               nseg;
1983 };
1984 
1985 static void
1986 _rxq_refill_cb(void *arg, bus_dma_segment_t *segs, int nseg, int error)
1987 {
1988 	struct rxq_refill_cb_arg *cb_arg = arg;
1989 
1990 	cb_arg->error = error;
1991 	cb_arg->seg = segs[0];
1992 	cb_arg->nseg = nseg;
1993 }
1994 
1995 /**
1996  * iflib_fl_refill - refill an rxq free-buffer list
1997  * @ctx: the iflib context
1998  * @fl: the free list to refill
1999  * @count: the number of new buffers to allocate
2000  *
2001  * (Re)populate an rxq free-buffer list with up to @count new packet buffers.
2002  * The caller must assure that @count does not exceed the queue's capacity
2003  * minus one (since we always leave a descriptor unavailable).
2004  */
2005 static uint8_t
2006 iflib_fl_refill(if_ctx_t ctx, iflib_fl_t fl, int count)
2007 {
2008 	struct if_rxd_update iru;
2009 	struct rxq_refill_cb_arg cb_arg;
2010 	struct mbuf *m;
2011 	caddr_t cl, *sd_cl;
2012 	struct mbuf **sd_m;
2013 	bus_dmamap_t *sd_map;
2014 	bus_addr_t bus_addr, *sd_ba;
2015 	int err, frag_idx, i, idx, n, pidx;
2016 	qidx_t credits;
2017 
2018 	MPASS(count <= fl->ifl_size - fl->ifl_credits - 1);
2019 
2020 	sd_m = fl->ifl_sds.ifsd_m;
2021 	sd_map = fl->ifl_sds.ifsd_map;
2022 	sd_cl = fl->ifl_sds.ifsd_cl;
2023 	sd_ba = fl->ifl_sds.ifsd_ba;
2024 	pidx = fl->ifl_pidx;
2025 	idx = pidx;
2026 	frag_idx = fl->ifl_fragidx;
2027 	credits = fl->ifl_credits;
2028 
2029 	i = 0;
2030 	n = count;
2031 	MPASS(n > 0);
2032 	MPASS(credits + n <= fl->ifl_size);
2033 
2034 	if (pidx < fl->ifl_cidx)
2035 		MPASS(pidx + n <= fl->ifl_cidx);
2036 	if (pidx == fl->ifl_cidx && (credits < fl->ifl_size))
2037 		MPASS(fl->ifl_gen == 0);
2038 	if (pidx > fl->ifl_cidx)
2039 		MPASS(n <= fl->ifl_size - pidx + fl->ifl_cidx);
2040 
2041 	DBG_COUNTER_INC(fl_refills);
2042 	if (n > 8)
2043 		DBG_COUNTER_INC(fl_refills_large);
2044 	iru_init(&iru, fl->ifl_rxq, fl->ifl_id);
2045 	while (n-- > 0) {
2046 		/*
2047 		 * We allocate an uninitialized mbuf + cluster, mbuf is
2048 		 * initialized after rx.
2049 		 *
2050 		 * If the cluster is still set then we know a minimum sized
2051 		 * packet was received
2052 		 */
2053 		bit_ffc_at(fl->ifl_rx_bitmap, frag_idx, fl->ifl_size,
2054 		    &frag_idx);
2055 		if (frag_idx < 0)
2056 			bit_ffc(fl->ifl_rx_bitmap, fl->ifl_size, &frag_idx);
2057 		MPASS(frag_idx >= 0);
2058 		if ((cl = sd_cl[frag_idx]) == NULL) {
2059 			cl = uma_zalloc(fl->ifl_zone, M_NOWAIT);
2060 			if (__predict_false(cl == NULL))
2061 				break;
2062 
2063 			cb_arg.error = 0;
2064 			MPASS(sd_map != NULL);
2065 			err = bus_dmamap_load(fl->ifl_buf_tag, sd_map[frag_idx],
2066 			    cl, fl->ifl_buf_size, _rxq_refill_cb, &cb_arg,
2067 			    BUS_DMA_NOWAIT);
2068 			if (__predict_false(err != 0 || cb_arg.error)) {
2069 				uma_zfree(fl->ifl_zone, cl);
2070 				break;
2071 			}
2072 
2073 			sd_ba[frag_idx] = bus_addr = cb_arg.seg.ds_addr;
2074 			sd_cl[frag_idx] = cl;
2075 #if MEMORY_LOGGING
2076 			fl->ifl_cl_enqueued++;
2077 #endif
2078 		} else {
2079 			bus_addr = sd_ba[frag_idx];
2080 		}
2081 		bus_dmamap_sync(fl->ifl_buf_tag, sd_map[frag_idx],
2082 		    BUS_DMASYNC_PREREAD);
2083 
2084 		if (sd_m[frag_idx] == NULL) {
2085 			m = m_gethdr(M_NOWAIT, MT_NOINIT);
2086 			if (__predict_false(m == NULL))
2087 				break;
2088 			sd_m[frag_idx] = m;
2089 		}
2090 		bit_set(fl->ifl_rx_bitmap, frag_idx);
2091 #if MEMORY_LOGGING
2092 		fl->ifl_m_enqueued++;
2093 #endif
2094 
2095 		DBG_COUNTER_INC(rx_allocs);
2096 		fl->ifl_rxd_idxs[i] = frag_idx;
2097 		fl->ifl_bus_addrs[i] = bus_addr;
2098 		credits++;
2099 		i++;
2100 		MPASS(credits <= fl->ifl_size);
2101 		if (++idx == fl->ifl_size) {
2102 #ifdef INVARIANTS
2103 			fl->ifl_gen = 1;
2104 #endif
2105 			idx = 0;
2106 		}
2107 		if (n == 0 || i == IFLIB_MAX_RX_REFRESH) {
2108 			iru.iru_pidx = pidx;
2109 			iru.iru_count = i;
2110 			ctx->isc_rxd_refill(ctx->ifc_softc, &iru);
2111 			fl->ifl_pidx = idx;
2112 			fl->ifl_credits = credits;
2113 			pidx = idx;
2114 			i = 0;
2115 		}
2116 	}
2117 
2118 	if (n < count - 1) {
2119 		if (i != 0) {
2120 			iru.iru_pidx = pidx;
2121 			iru.iru_count = i;
2122 			ctx->isc_rxd_refill(ctx->ifc_softc, &iru);
2123 			fl->ifl_pidx = idx;
2124 			fl->ifl_credits = credits;
2125 		}
2126 		DBG_COUNTER_INC(rxd_flush);
2127 		bus_dmamap_sync(fl->ifl_ifdi->idi_tag, fl->ifl_ifdi->idi_map,
2128 		    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
2129 		ctx->isc_rxd_flush(ctx->ifc_softc, fl->ifl_rxq->ifr_id,
2130 		    fl->ifl_id, fl->ifl_pidx);
2131 		if (__predict_true(bit_test(fl->ifl_rx_bitmap, frag_idx))) {
2132 			fl->ifl_fragidx = frag_idx + 1;
2133 			if (fl->ifl_fragidx == fl->ifl_size)
2134 				fl->ifl_fragidx = 0;
2135 		} else {
2136 			fl->ifl_fragidx = frag_idx;
2137 		}
2138 	}
2139 
2140 	return (n == -1 ? 0 : IFLIB_RXEOF_EMPTY);
2141 }
2142 
2143 static inline uint8_t
2144 iflib_fl_refill_all(if_ctx_t ctx, iflib_fl_t fl)
2145 {
2146 	/*
2147 	 * We leave an unused descriptor to avoid pidx to catch up with cidx.
2148 	 * This is important as it confuses most NICs. For instance,
2149 	 * Intel NICs have (per receive ring) RDH and RDT registers, where
2150 	 * RDH points to the next receive descriptor to be used by the NIC,
2151 	 * and RDT for the next receive descriptor to be published by the
2152 	 * driver to the NIC (RDT - 1 is thus the last valid one).
2153 	 * The condition RDH == RDT means no descriptors are available to
2154 	 * the NIC, and thus it would be ambiguous if it also meant that
2155 	 * all the descriptors are available to the NIC.
2156 	 */
2157 	int32_t reclaimable = fl->ifl_size - fl->ifl_credits - 1;
2158 #ifdef INVARIANTS
2159 	int32_t delta = fl->ifl_size - get_inuse(fl->ifl_size, fl->ifl_cidx, fl->ifl_pidx, fl->ifl_gen) - 1;
2160 #endif
2161 
2162 	MPASS(fl->ifl_credits <= fl->ifl_size);
2163 	MPASS(reclaimable == delta);
2164 
2165 	if (reclaimable > 0)
2166 		return (iflib_fl_refill(ctx, fl, reclaimable));
2167 	return (0);
2168 }
2169 
2170 uint8_t
2171 iflib_in_detach(if_ctx_t ctx)
2172 {
2173 	bool in_detach;
2174 
2175 	STATE_LOCK(ctx);
2176 	in_detach = !!(ctx->ifc_flags & IFC_IN_DETACH);
2177 	STATE_UNLOCK(ctx);
2178 	return (in_detach);
2179 }
2180 
2181 static void
2182 iflib_fl_bufs_free(iflib_fl_t fl)
2183 {
2184 	iflib_dma_info_t idi = fl->ifl_ifdi;
2185 	bus_dmamap_t sd_map;
2186 	uint32_t i;
2187 
2188 	for (i = 0; i < fl->ifl_size; i++) {
2189 		struct mbuf **sd_m = &fl->ifl_sds.ifsd_m[i];
2190 		caddr_t *sd_cl = &fl->ifl_sds.ifsd_cl[i];
2191 
2192 		if (*sd_cl != NULL) {
2193 			sd_map = fl->ifl_sds.ifsd_map[i];
2194 			bus_dmamap_sync(fl->ifl_buf_tag, sd_map,
2195 			    BUS_DMASYNC_POSTREAD);
2196 			bus_dmamap_unload(fl->ifl_buf_tag, sd_map);
2197 			uma_zfree(fl->ifl_zone, *sd_cl);
2198 			*sd_cl = NULL;
2199 			if (*sd_m != NULL) {
2200 				m_init(*sd_m, M_NOWAIT, MT_DATA, 0);
2201 				uma_zfree(zone_mbuf, *sd_m);
2202 				*sd_m = NULL;
2203 			}
2204 		} else {
2205 			MPASS(*sd_m == NULL);
2206 		}
2207 #if MEMORY_LOGGING
2208 		fl->ifl_m_dequeued++;
2209 		fl->ifl_cl_dequeued++;
2210 #endif
2211 	}
2212 #ifdef INVARIANTS
2213 	for (i = 0; i < fl->ifl_size; i++) {
2214 		MPASS(fl->ifl_sds.ifsd_cl[i] == NULL);
2215 		MPASS(fl->ifl_sds.ifsd_m[i] == NULL);
2216 	}
2217 #endif
2218 	/*
2219 	 * Reset free list values
2220 	 */
2221 	fl->ifl_credits = fl->ifl_cidx = fl->ifl_pidx = fl->ifl_gen = fl->ifl_fragidx = 0;
2222 	bzero(idi->idi_vaddr, idi->idi_size);
2223 }
2224 
2225 /*********************************************************************
2226  *
2227  *  Initialize a free list and its buffers.
2228  *
2229  **********************************************************************/
2230 static int
2231 iflib_fl_setup(iflib_fl_t fl)
2232 {
2233 	iflib_rxq_t rxq = fl->ifl_rxq;
2234 	if_ctx_t ctx = rxq->ifr_ctx;
2235 	if_softc_ctx_t scctx = &ctx->ifc_softc_ctx;
2236 	int qidx;
2237 
2238 	bit_nclear(fl->ifl_rx_bitmap, 0, fl->ifl_size - 1);
2239 	/*
2240 	** Free current RX buffer structs and their mbufs
2241 	*/
2242 	iflib_fl_bufs_free(fl);
2243 	/* Now replenish the mbufs */
2244 	MPASS(fl->ifl_credits == 0);
2245 	qidx = rxq->ifr_fl_offset + fl->ifl_id;
2246 	if (scctx->isc_rxd_buf_size[qidx] != 0)
2247 		fl->ifl_buf_size = scctx->isc_rxd_buf_size[qidx];
2248 	else
2249 		fl->ifl_buf_size = ctx->ifc_rx_mbuf_sz;
2250 	/*
2251 	 * ifl_buf_size may be a driver-supplied value, so pull it up
2252 	 * to the selected mbuf size.
2253 	 */
2254 	fl->ifl_buf_size = iflib_get_mbuf_size_for(fl->ifl_buf_size);
2255 	if (fl->ifl_buf_size > ctx->ifc_max_fl_buf_size)
2256 		ctx->ifc_max_fl_buf_size = fl->ifl_buf_size;
2257 	fl->ifl_cltype = m_gettype(fl->ifl_buf_size);
2258 	fl->ifl_zone = m_getzone(fl->ifl_buf_size);
2259 
2260 	/*
2261 	 * Avoid pre-allocating zillions of clusters to an idle card
2262 	 * potentially speeding up attach. In any case make sure
2263 	 * to leave a descriptor unavailable. See the comment in
2264 	 * iflib_fl_refill_all().
2265 	 */
2266 	MPASS(fl->ifl_size > 0);
2267 	(void)iflib_fl_refill(ctx, fl, min(128, fl->ifl_size - 1));
2268 	if (min(128, fl->ifl_size - 1) != fl->ifl_credits)
2269 		return (ENOBUFS);
2270 	/*
2271 	 * handle failure
2272 	 */
2273 	MPASS(rxq != NULL);
2274 	MPASS(fl->ifl_ifdi != NULL);
2275 	bus_dmamap_sync(fl->ifl_ifdi->idi_tag, fl->ifl_ifdi->idi_map,
2276 	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
2277 	return (0);
2278 }
2279 
2280 /*********************************************************************
2281  *
2282  *  Free receive ring data structures
2283  *
2284  **********************************************************************/
2285 static void
2286 iflib_rx_sds_free(iflib_rxq_t rxq)
2287 {
2288 	iflib_fl_t fl;
2289 	int i, j;
2290 
2291 	if (rxq->ifr_fl != NULL) {
2292 		for (i = 0; i < rxq->ifr_nfl; i++) {
2293 			fl = &rxq->ifr_fl[i];
2294 			if (fl->ifl_buf_tag != NULL) {
2295 				if (fl->ifl_sds.ifsd_map != NULL) {
2296 					for (j = 0; j < fl->ifl_size; j++) {
2297 						bus_dmamap_sync(
2298 						    fl->ifl_buf_tag,
2299 						    fl->ifl_sds.ifsd_map[j],
2300 						    BUS_DMASYNC_POSTREAD);
2301 						bus_dmamap_unload(
2302 						    fl->ifl_buf_tag,
2303 						    fl->ifl_sds.ifsd_map[j]);
2304 						bus_dmamap_destroy(
2305 						    fl->ifl_buf_tag,
2306 						    fl->ifl_sds.ifsd_map[j]);
2307 					}
2308 				}
2309 				bus_dma_tag_destroy(fl->ifl_buf_tag);
2310 				fl->ifl_buf_tag = NULL;
2311 			}
2312 			free(fl->ifl_sds.ifsd_m, M_IFLIB);
2313 			free(fl->ifl_sds.ifsd_cl, M_IFLIB);
2314 			free(fl->ifl_sds.ifsd_ba, M_IFLIB);
2315 			free(fl->ifl_sds.ifsd_map, M_IFLIB);
2316 			free(fl->ifl_rx_bitmap, M_IFLIB);
2317 			fl->ifl_sds.ifsd_m = NULL;
2318 			fl->ifl_sds.ifsd_cl = NULL;
2319 			fl->ifl_sds.ifsd_ba = NULL;
2320 			fl->ifl_sds.ifsd_map = NULL;
2321 			fl->ifl_rx_bitmap = NULL;
2322 		}
2323 		free(rxq->ifr_fl, M_IFLIB);
2324 		rxq->ifr_fl = NULL;
2325 		free(rxq->ifr_ifdi, M_IFLIB);
2326 		rxq->ifr_ifdi = NULL;
2327 		rxq->ifr_cq_cidx = 0;
2328 	}
2329 }
2330 
2331 /*
2332  * Timer routine
2333  */
2334 static void
2335 iflib_timer(void *arg)
2336 {
2337 	iflib_txq_t txq = arg;
2338 	if_ctx_t ctx = txq->ift_ctx;
2339 	if_softc_ctx_t sctx = &ctx->ifc_softc_ctx;
2340 	uint64_t this_tick = ticks;
2341 
2342 	if (!(if_getdrvflags(ctx->ifc_ifp) & IFF_DRV_RUNNING))
2343 		return;
2344 
2345 	/*
2346 	** Check on the state of the TX queue(s), this
2347 	** can be done without the lock because its RO
2348 	** and the HUNG state will be static if set.
2349 	*/
2350 	if (this_tick - txq->ift_last_timer_tick >= iflib_timer_default) {
2351 		txq->ift_last_timer_tick = this_tick;
2352 		IFDI_TIMER(ctx, txq->ift_id);
2353 		if ((txq->ift_qstatus == IFLIB_QUEUE_HUNG) &&
2354 		    ((txq->ift_cleaned_prev == txq->ift_cleaned) ||
2355 		     (sctx->isc_pause_frames == 0)))
2356 			goto hung;
2357 
2358 		if (txq->ift_qstatus != IFLIB_QUEUE_IDLE &&
2359 		    ifmp_ring_is_stalled(txq->ift_br)) {
2360 			KASSERT(ctx->ifc_link_state == LINK_STATE_UP,
2361 			    ("queue can't be marked as hung if interface is down"));
2362 			txq->ift_qstatus = IFLIB_QUEUE_HUNG;
2363 		}
2364 		txq->ift_cleaned_prev = txq->ift_cleaned;
2365 	}
2366 	/* handle any laggards */
2367 	if (txq->ift_db_pending)
2368 		GROUPTASK_ENQUEUE(&txq->ift_task);
2369 
2370 	sctx->isc_pause_frames = 0;
2371 	if (if_getdrvflags(ctx->ifc_ifp) & IFF_DRV_RUNNING)
2372 		callout_reset_on(&txq->ift_timer, iflib_timer_default, iflib_timer,
2373 		    txq, txq->ift_timer.c_cpu);
2374 	return;
2375 
2376  hung:
2377 	device_printf(ctx->ifc_dev,
2378 	    "Watchdog timeout (TX: %d desc avail: %d pidx: %d) -- resetting\n",
2379 	    txq->ift_id, TXQ_AVAIL(txq), txq->ift_pidx);
2380 	STATE_LOCK(ctx);
2381 	if_setdrvflagbits(ctx->ifc_ifp, IFF_DRV_OACTIVE, IFF_DRV_RUNNING);
2382 	ctx->ifc_flags |= (IFC_DO_WATCHDOG|IFC_DO_RESET);
2383 	iflib_admin_intr_deferred(ctx);
2384 	STATE_UNLOCK(ctx);
2385 }
2386 
2387 static uint16_t
2388 iflib_get_mbuf_size_for(unsigned int size)
2389 {
2390 
2391 	if (size <= MCLBYTES)
2392 		return (MCLBYTES);
2393 	else
2394 		return (MJUMPAGESIZE);
2395 }
2396 
2397 static void
2398 iflib_calc_rx_mbuf_sz(if_ctx_t ctx)
2399 {
2400 	if_softc_ctx_t sctx = &ctx->ifc_softc_ctx;
2401 
2402 	/*
2403 	 * XXX don't set the max_frame_size to larger
2404 	 * than the hardware can handle
2405 	 */
2406 	ctx->ifc_rx_mbuf_sz =
2407 	    iflib_get_mbuf_size_for(sctx->isc_max_frame_size);
2408 }
2409 
2410 uint32_t
2411 iflib_get_rx_mbuf_sz(if_ctx_t ctx)
2412 {
2413 
2414 	return (ctx->ifc_rx_mbuf_sz);
2415 }
2416 
2417 static void
2418 iflib_init_locked(if_ctx_t ctx)
2419 {
2420 	if_softc_ctx_t sctx = &ctx->ifc_softc_ctx;
2421 	if_softc_ctx_t scctx = &ctx->ifc_softc_ctx;
2422 	if_t ifp = ctx->ifc_ifp;
2423 	iflib_fl_t fl;
2424 	iflib_txq_t txq;
2425 	iflib_rxq_t rxq;
2426 	int i, j, tx_ip_csum_flags, tx_ip6_csum_flags;
2427 
2428 	if_setdrvflagbits(ifp, IFF_DRV_OACTIVE, IFF_DRV_RUNNING);
2429 	IFDI_INTR_DISABLE(ctx);
2430 
2431 	/*
2432 	 * See iflib_stop(). Useful in case iflib_init_locked() is
2433 	 * called without first calling iflib_stop().
2434 	 */
2435 	netmap_disable_all_rings(ifp);
2436 
2437 	tx_ip_csum_flags = scctx->isc_tx_csum_flags & (CSUM_IP | CSUM_TCP | CSUM_UDP | CSUM_SCTP);
2438 	tx_ip6_csum_flags = scctx->isc_tx_csum_flags & (CSUM_IP6_TCP | CSUM_IP6_UDP | CSUM_IP6_SCTP);
2439 	/* Set hardware offload abilities */
2440 	if_clearhwassist(ifp);
2441 	if (if_getcapenable(ifp) & IFCAP_TXCSUM)
2442 		if_sethwassistbits(ifp, tx_ip_csum_flags, 0);
2443 	if (if_getcapenable(ifp) & IFCAP_TXCSUM_IPV6)
2444 		if_sethwassistbits(ifp,  tx_ip6_csum_flags, 0);
2445 	if (if_getcapenable(ifp) & IFCAP_TSO4)
2446 		if_sethwassistbits(ifp, CSUM_IP_TSO, 0);
2447 	if (if_getcapenable(ifp) & IFCAP_TSO6)
2448 		if_sethwassistbits(ifp, CSUM_IP6_TSO, 0);
2449 
2450 	for (i = 0, txq = ctx->ifc_txqs; i < sctx->isc_ntxqsets; i++, txq++) {
2451 		CALLOUT_LOCK(txq);
2452 		callout_stop(&txq->ift_timer);
2453 #ifdef DEV_NETMAP
2454 		callout_stop(&txq->ift_netmap_timer);
2455 #endif /* DEV_NETMAP */
2456 		CALLOUT_UNLOCK(txq);
2457 		iflib_netmap_txq_init(ctx, txq);
2458 	}
2459 
2460 	/*
2461 	 * Calculate a suitable Rx mbuf size prior to calling IFDI_INIT, so
2462 	 * that drivers can use the value when setting up the hardware receive
2463 	 * buffers.
2464 	 */
2465 	iflib_calc_rx_mbuf_sz(ctx);
2466 
2467 #ifdef INVARIANTS
2468 	i = if_getdrvflags(ifp);
2469 #endif
2470 	IFDI_INIT(ctx);
2471 	MPASS(if_getdrvflags(ifp) == i);
2472 	for (i = 0, rxq = ctx->ifc_rxqs; i < sctx->isc_nrxqsets; i++, rxq++) {
2473 		if (iflib_netmap_rxq_init(ctx, rxq) > 0) {
2474 			/* This rxq is in netmap mode. Skip normal init. */
2475 			continue;
2476 		}
2477 		for (j = 0, fl = rxq->ifr_fl; j < rxq->ifr_nfl; j++, fl++) {
2478 			if (iflib_fl_setup(fl)) {
2479 				device_printf(ctx->ifc_dev,
2480 				    "setting up free list %d failed - "
2481 				    "check cluster settings\n", j);
2482 				goto done;
2483 			}
2484 		}
2485 	}
2486 done:
2487 	if_setdrvflagbits(ctx->ifc_ifp, IFF_DRV_RUNNING, IFF_DRV_OACTIVE);
2488 	IFDI_INTR_ENABLE(ctx);
2489 	txq = ctx->ifc_txqs;
2490 	for (i = 0; i < sctx->isc_ntxqsets; i++, txq++)
2491 		callout_reset_on(&txq->ift_timer, iflib_timer_default, iflib_timer, txq,
2492 			txq->ift_timer.c_cpu);
2493 
2494         /* Re-enable txsync/rxsync. */
2495 	netmap_enable_all_rings(ifp);
2496 }
2497 
2498 static int
2499 iflib_media_change(if_t ifp)
2500 {
2501 	if_ctx_t ctx = if_getsoftc(ifp);
2502 	int err;
2503 
2504 	CTX_LOCK(ctx);
2505 	if ((err = IFDI_MEDIA_CHANGE(ctx)) == 0)
2506 		iflib_init_locked(ctx);
2507 	CTX_UNLOCK(ctx);
2508 	return (err);
2509 }
2510 
2511 static void
2512 iflib_media_status(if_t ifp, struct ifmediareq *ifmr)
2513 {
2514 	if_ctx_t ctx = if_getsoftc(ifp);
2515 
2516 	CTX_LOCK(ctx);
2517 	IFDI_UPDATE_ADMIN_STATUS(ctx);
2518 	IFDI_MEDIA_STATUS(ctx, ifmr);
2519 	CTX_UNLOCK(ctx);
2520 }
2521 
2522 void
2523 iflib_stop(if_ctx_t ctx)
2524 {
2525 	iflib_txq_t txq = ctx->ifc_txqs;
2526 	iflib_rxq_t rxq = ctx->ifc_rxqs;
2527 	if_softc_ctx_t scctx = &ctx->ifc_softc_ctx;
2528 	if_shared_ctx_t sctx = ctx->ifc_sctx;
2529 	iflib_dma_info_t di;
2530 	iflib_fl_t fl;
2531 	int i, j;
2532 
2533 	/* Tell the stack that the interface is no longer active */
2534 	if_setdrvflagbits(ctx->ifc_ifp, IFF_DRV_OACTIVE, IFF_DRV_RUNNING);
2535 
2536 	IFDI_INTR_DISABLE(ctx);
2537 	DELAY(1000);
2538 	IFDI_STOP(ctx);
2539 	DELAY(1000);
2540 
2541 	/*
2542 	 * Stop any pending txsync/rxsync and prevent new ones
2543 	 * form starting. Processes blocked in poll() will get
2544 	 * POLLERR.
2545 	 */
2546 	netmap_disable_all_rings(ctx->ifc_ifp);
2547 
2548 	iflib_debug_reset();
2549 	/* Wait for current tx queue users to exit to disarm watchdog timer. */
2550 	for (i = 0; i < scctx->isc_ntxqsets; i++, txq++) {
2551 		/* make sure all transmitters have completed before proceeding XXX */
2552 
2553 		CALLOUT_LOCK(txq);
2554 		callout_stop(&txq->ift_timer);
2555 #ifdef DEV_NETMAP
2556 		callout_stop(&txq->ift_netmap_timer);
2557 #endif /* DEV_NETMAP */
2558 		CALLOUT_UNLOCK(txq);
2559 
2560 		/* clean any enqueued buffers */
2561 		iflib_ifmp_purge(txq);
2562 		/* Free any existing tx buffers. */
2563 		for (j = 0; j < txq->ift_size; j++) {
2564 			iflib_txsd_free(ctx, txq, j);
2565 		}
2566 		txq->ift_processed = txq->ift_cleaned = txq->ift_cidx_processed = 0;
2567 		txq->ift_in_use = txq->ift_gen = txq->ift_cidx = txq->ift_pidx = txq->ift_no_desc_avail = 0;
2568 		txq->ift_closed = txq->ift_mbuf_defrag = txq->ift_mbuf_defrag_failed = 0;
2569 		txq->ift_no_tx_dma_setup = txq->ift_txd_encap_efbig = txq->ift_map_failed = 0;
2570 		txq->ift_pullups = 0;
2571 		ifmp_ring_reset_stats(txq->ift_br);
2572 		for (j = 0, di = txq->ift_ifdi; j < sctx->isc_ntxqs; j++, di++)
2573 			bzero((void *)di->idi_vaddr, di->idi_size);
2574 	}
2575 	for (i = 0; i < scctx->isc_nrxqsets; i++, rxq++) {
2576 		/* make sure all transmitters have completed before proceeding XXX */
2577 
2578 		rxq->ifr_cq_cidx = 0;
2579 		for (j = 0, di = rxq->ifr_ifdi; j < sctx->isc_nrxqs; j++, di++)
2580 			bzero((void *)di->idi_vaddr, di->idi_size);
2581 		/* also resets the free lists pidx/cidx */
2582 		for (j = 0, fl = rxq->ifr_fl; j < rxq->ifr_nfl; j++, fl++)
2583 			iflib_fl_bufs_free(fl);
2584 	}
2585 }
2586 
2587 static inline caddr_t
2588 calc_next_rxd(iflib_fl_t fl, int cidx)
2589 {
2590 	qidx_t size;
2591 	int nrxd;
2592 	caddr_t start, end, cur, next;
2593 
2594 	nrxd = fl->ifl_size;
2595 	size = fl->ifl_rxd_size;
2596 	start = fl->ifl_ifdi->idi_vaddr;
2597 
2598 	if (__predict_false(size == 0))
2599 		return (start);
2600 	cur = start + size*cidx;
2601 	end = start + size*nrxd;
2602 	next = CACHE_PTR_NEXT(cur);
2603 	return (next < end ? next : start);
2604 }
2605 
2606 static inline void
2607 prefetch_pkts(iflib_fl_t fl, int cidx)
2608 {
2609 	int nextptr;
2610 	int nrxd = fl->ifl_size;
2611 	caddr_t next_rxd;
2612 
2613 	nextptr = (cidx + CACHE_PTR_INCREMENT) & (nrxd-1);
2614 	prefetch(&fl->ifl_sds.ifsd_m[nextptr]);
2615 	prefetch(&fl->ifl_sds.ifsd_cl[nextptr]);
2616 	next_rxd = calc_next_rxd(fl, cidx);
2617 	prefetch(next_rxd);
2618 	prefetch(fl->ifl_sds.ifsd_m[(cidx + 1) & (nrxd-1)]);
2619 	prefetch(fl->ifl_sds.ifsd_m[(cidx + 2) & (nrxd-1)]);
2620 	prefetch(fl->ifl_sds.ifsd_m[(cidx + 3) & (nrxd-1)]);
2621 	prefetch(fl->ifl_sds.ifsd_m[(cidx + 4) & (nrxd-1)]);
2622 	prefetch(fl->ifl_sds.ifsd_cl[(cidx + 1) & (nrxd-1)]);
2623 	prefetch(fl->ifl_sds.ifsd_cl[(cidx + 2) & (nrxd-1)]);
2624 	prefetch(fl->ifl_sds.ifsd_cl[(cidx + 3) & (nrxd-1)]);
2625 	prefetch(fl->ifl_sds.ifsd_cl[(cidx + 4) & (nrxd-1)]);
2626 }
2627 
2628 static struct mbuf *
2629 rxd_frag_to_sd(iflib_rxq_t rxq, if_rxd_frag_t irf, bool unload, if_rxsd_t sd,
2630     int *pf_rv, if_rxd_info_t ri)
2631 {
2632 	bus_dmamap_t map;
2633 	iflib_fl_t fl;
2634 	caddr_t payload;
2635 	struct mbuf *m;
2636 	int flid, cidx, len, next;
2637 
2638 	map = NULL;
2639 	flid = irf->irf_flid;
2640 	cidx = irf->irf_idx;
2641 	fl = &rxq->ifr_fl[flid];
2642 	sd->ifsd_fl = fl;
2643 	m = fl->ifl_sds.ifsd_m[cidx];
2644 	sd->ifsd_cl = &fl->ifl_sds.ifsd_cl[cidx];
2645 	fl->ifl_credits--;
2646 #if MEMORY_LOGGING
2647 	fl->ifl_m_dequeued++;
2648 #endif
2649 	if (rxq->ifr_ctx->ifc_flags & IFC_PREFETCH)
2650 		prefetch_pkts(fl, cidx);
2651 	next = (cidx + CACHE_PTR_INCREMENT) & (fl->ifl_size-1);
2652 	prefetch(&fl->ifl_sds.ifsd_map[next]);
2653 	map = fl->ifl_sds.ifsd_map[cidx];
2654 
2655 	bus_dmamap_sync(fl->ifl_buf_tag, map, BUS_DMASYNC_POSTREAD);
2656 
2657 	if (rxq->pfil != NULL && PFIL_HOOKED_IN(rxq->pfil) && pf_rv != NULL &&
2658 	    irf->irf_len != 0) {
2659 		payload  = *sd->ifsd_cl;
2660 		payload +=  ri->iri_pad;
2661 		len = ri->iri_len - ri->iri_pad;
2662 		*pf_rv = pfil_run_hooks(rxq->pfil, payload, ri->iri_ifp,
2663 		    len | PFIL_MEMPTR | PFIL_IN, NULL);
2664 		switch (*pf_rv) {
2665 		case PFIL_DROPPED:
2666 		case PFIL_CONSUMED:
2667 			/*
2668 			 * The filter ate it.  Everything is recycled.
2669 			 */
2670 			m = NULL;
2671 			unload = 0;
2672 			break;
2673 		case PFIL_REALLOCED:
2674 			/*
2675 			 * The filter copied it.  Everything is recycled.
2676 			 */
2677 			m = pfil_mem2mbuf(payload);
2678 			unload = 0;
2679 			break;
2680 		case PFIL_PASS:
2681 			/*
2682 			 * Filter said it was OK, so receive like
2683 			 * normal
2684 			 */
2685 			fl->ifl_sds.ifsd_m[cidx] = NULL;
2686 			break;
2687 		default:
2688 			MPASS(0);
2689 		}
2690 	} else {
2691 		fl->ifl_sds.ifsd_m[cidx] = NULL;
2692 		if (pf_rv != NULL)
2693 			*pf_rv = PFIL_PASS;
2694 	}
2695 
2696 	if (unload && irf->irf_len != 0)
2697 		bus_dmamap_unload(fl->ifl_buf_tag, map);
2698 	fl->ifl_cidx = (fl->ifl_cidx + 1) & (fl->ifl_size-1);
2699 	if (__predict_false(fl->ifl_cidx == 0))
2700 		fl->ifl_gen = 0;
2701 	bit_clear(fl->ifl_rx_bitmap, cidx);
2702 	return (m);
2703 }
2704 
2705 static struct mbuf *
2706 assemble_segments(iflib_rxq_t rxq, if_rxd_info_t ri, if_rxsd_t sd, int *pf_rv)
2707 {
2708 	struct mbuf *m, *mh, *mt;
2709 	caddr_t cl;
2710 	int  *pf_rv_ptr, flags, i, padlen;
2711 	bool consumed;
2712 
2713 	i = 0;
2714 	mh = NULL;
2715 	consumed = false;
2716 	*pf_rv = PFIL_PASS;
2717 	pf_rv_ptr = pf_rv;
2718 	do {
2719 		m = rxd_frag_to_sd(rxq, &ri->iri_frags[i], !consumed, sd,
2720 		    pf_rv_ptr, ri);
2721 
2722 		MPASS(*sd->ifsd_cl != NULL);
2723 
2724 		/*
2725 		 * Exclude zero-length frags & frags from
2726 		 * packets the filter has consumed or dropped
2727 		 */
2728 		if (ri->iri_frags[i].irf_len == 0 || consumed ||
2729 		    *pf_rv == PFIL_CONSUMED || *pf_rv == PFIL_DROPPED) {
2730 			if (mh == NULL) {
2731 				/* everything saved here */
2732 				consumed = true;
2733 				pf_rv_ptr = NULL;
2734 				continue;
2735 			}
2736 			/* XXX we can save the cluster here, but not the mbuf */
2737 			m_init(m, M_NOWAIT, MT_DATA, 0);
2738 			m_free(m);
2739 			continue;
2740 		}
2741 		if (mh == NULL) {
2742 			flags = M_PKTHDR|M_EXT;
2743 			mh = mt = m;
2744 			padlen = ri->iri_pad;
2745 		} else {
2746 			flags = M_EXT;
2747 			mt->m_next = m;
2748 			mt = m;
2749 			/* assuming padding is only on the first fragment */
2750 			padlen = 0;
2751 		}
2752 		cl = *sd->ifsd_cl;
2753 		*sd->ifsd_cl = NULL;
2754 
2755 		/* Can these two be made one ? */
2756 		m_init(m, M_NOWAIT, MT_DATA, flags);
2757 		m_cljset(m, cl, sd->ifsd_fl->ifl_cltype);
2758 		/*
2759 		 * These must follow m_init and m_cljset
2760 		 */
2761 		m->m_data += padlen;
2762 		ri->iri_len -= padlen;
2763 		m->m_len = ri->iri_frags[i].irf_len;
2764 	} while (++i < ri->iri_nfrags);
2765 
2766 	return (mh);
2767 }
2768 
2769 /*
2770  * Process one software descriptor
2771  */
2772 static struct mbuf *
2773 iflib_rxd_pkt_get(iflib_rxq_t rxq, if_rxd_info_t ri)
2774 {
2775 	struct if_rxsd sd;
2776 	struct mbuf *m;
2777 	int pf_rv;
2778 
2779 	/* should I merge this back in now that the two paths are basically duplicated? */
2780 	if (ri->iri_nfrags == 1 &&
2781 	    ri->iri_frags[0].irf_len != 0 &&
2782 	    ri->iri_frags[0].irf_len <= MIN(IFLIB_RX_COPY_THRESH, MHLEN)) {
2783 		m = rxd_frag_to_sd(rxq, &ri->iri_frags[0], false, &sd,
2784 		    &pf_rv, ri);
2785 		if (pf_rv != PFIL_PASS && pf_rv != PFIL_REALLOCED)
2786 			return (m);
2787 		if (pf_rv == PFIL_PASS) {
2788 			m_init(m, M_NOWAIT, MT_DATA, M_PKTHDR);
2789 #ifndef __NO_STRICT_ALIGNMENT
2790 			if (!IP_ALIGNED(m))
2791 				m->m_data += 2;
2792 #endif
2793 			memcpy(m->m_data, *sd.ifsd_cl, ri->iri_len);
2794 			m->m_len = ri->iri_frags[0].irf_len;
2795 		}
2796 	} else {
2797 		m = assemble_segments(rxq, ri, &sd, &pf_rv);
2798 		if (m == NULL)
2799 			return (NULL);
2800 		if (pf_rv != PFIL_PASS && pf_rv != PFIL_REALLOCED)
2801 			return (m);
2802 	}
2803 	m->m_pkthdr.len = ri->iri_len;
2804 	m->m_pkthdr.rcvif = ri->iri_ifp;
2805 	m->m_flags |= ri->iri_flags;
2806 	m->m_pkthdr.ether_vtag = ri->iri_vtag;
2807 	m->m_pkthdr.flowid = ri->iri_flowid;
2808 	M_HASHTYPE_SET(m, ri->iri_rsstype);
2809 	m->m_pkthdr.csum_flags = ri->iri_csum_flags;
2810 	m->m_pkthdr.csum_data = ri->iri_csum_data;
2811 	return (m);
2812 }
2813 
2814 #if defined(INET6) || defined(INET)
2815 static void
2816 iflib_get_ip_forwarding(struct lro_ctrl *lc, bool *v4, bool *v6)
2817 {
2818 	CURVNET_SET(lc->ifp->if_vnet);
2819 #if defined(INET6)
2820 	*v6 = V_ip6_forwarding;
2821 #endif
2822 #if defined(INET)
2823 	*v4 = V_ipforwarding;
2824 #endif
2825 	CURVNET_RESTORE();
2826 }
2827 
2828 /*
2829  * Returns true if it's possible this packet could be LROed.
2830  * if it returns false, it is guaranteed that tcp_lro_rx()
2831  * would not return zero.
2832  */
2833 static bool
2834 iflib_check_lro_possible(struct mbuf *m, bool v4_forwarding, bool v6_forwarding)
2835 {
2836 	struct ether_header *eh;
2837 
2838 	eh = mtod(m, struct ether_header *);
2839 	switch (eh->ether_type) {
2840 #if defined(INET6)
2841 		case htons(ETHERTYPE_IPV6):
2842 			return (!v6_forwarding);
2843 #endif
2844 #if defined (INET)
2845 		case htons(ETHERTYPE_IP):
2846 			return (!v4_forwarding);
2847 #endif
2848 	}
2849 
2850 	return false;
2851 }
2852 #else
2853 static void
2854 iflib_get_ip_forwarding(struct lro_ctrl *lc __unused, bool *v4 __unused, bool *v6 __unused)
2855 {
2856 }
2857 #endif
2858 
2859 static void
2860 _task_fn_rx_watchdog(void *context)
2861 {
2862 	iflib_rxq_t rxq = context;
2863 
2864 	GROUPTASK_ENQUEUE(&rxq->ifr_task);
2865 }
2866 
2867 static uint8_t
2868 iflib_rxeof(iflib_rxq_t rxq, qidx_t budget)
2869 {
2870 	if_t ifp;
2871 	if_ctx_t ctx = rxq->ifr_ctx;
2872 	if_shared_ctx_t sctx = ctx->ifc_sctx;
2873 	if_softc_ctx_t scctx = &ctx->ifc_softc_ctx;
2874 	int avail, i;
2875 	qidx_t *cidxp;
2876 	struct if_rxd_info ri;
2877 	int err, budget_left, rx_bytes, rx_pkts;
2878 	iflib_fl_t fl;
2879 	int lro_enabled;
2880 	bool v4_forwarding, v6_forwarding, lro_possible;
2881 	uint8_t retval = 0;
2882 
2883 	/*
2884 	 * XXX early demux data packets so that if_input processing only handles
2885 	 * acks in interrupt context
2886 	 */
2887 	struct mbuf *m, *mh, *mt, *mf;
2888 
2889 	NET_EPOCH_ASSERT();
2890 
2891 	lro_possible = v4_forwarding = v6_forwarding = false;
2892 	ifp = ctx->ifc_ifp;
2893 	mh = mt = NULL;
2894 	MPASS(budget > 0);
2895 	rx_pkts	= rx_bytes = 0;
2896 	if (sctx->isc_flags & IFLIB_HAS_RXCQ)
2897 		cidxp = &rxq->ifr_cq_cidx;
2898 	else
2899 		cidxp = &rxq->ifr_fl[0].ifl_cidx;
2900 	if ((avail = iflib_rxd_avail(ctx, rxq, *cidxp, budget)) == 0) {
2901 		for (i = 0, fl = &rxq->ifr_fl[0]; i < sctx->isc_nfl; i++, fl++)
2902 			retval |= iflib_fl_refill_all(ctx, fl);
2903 		DBG_COUNTER_INC(rx_unavail);
2904 		return (retval);
2905 	}
2906 
2907 	/* pfil needs the vnet to be set */
2908 	CURVNET_SET_QUIET(ifp->if_vnet);
2909 	for (budget_left = budget; budget_left > 0 && avail > 0;) {
2910 		if (__predict_false(!CTX_ACTIVE(ctx))) {
2911 			DBG_COUNTER_INC(rx_ctx_inactive);
2912 			break;
2913 		}
2914 		/*
2915 		 * Reset client set fields to their default values
2916 		 */
2917 		rxd_info_zero(&ri);
2918 		ri.iri_qsidx = rxq->ifr_id;
2919 		ri.iri_cidx = *cidxp;
2920 		ri.iri_ifp = ifp;
2921 		ri.iri_frags = rxq->ifr_frags;
2922 		err = ctx->isc_rxd_pkt_get(ctx->ifc_softc, &ri);
2923 
2924 		if (err)
2925 			goto err;
2926 		rx_pkts += 1;
2927 		rx_bytes += ri.iri_len;
2928 		if (sctx->isc_flags & IFLIB_HAS_RXCQ) {
2929 			*cidxp = ri.iri_cidx;
2930 			/* Update our consumer index */
2931 			/* XXX NB: shurd - check if this is still safe */
2932 			while (rxq->ifr_cq_cidx >= scctx->isc_nrxd[0])
2933 				rxq->ifr_cq_cidx -= scctx->isc_nrxd[0];
2934 			/* was this only a completion queue message? */
2935 			if (__predict_false(ri.iri_nfrags == 0))
2936 				continue;
2937 		}
2938 		MPASS(ri.iri_nfrags != 0);
2939 		MPASS(ri.iri_len != 0);
2940 
2941 		/* will advance the cidx on the corresponding free lists */
2942 		m = iflib_rxd_pkt_get(rxq, &ri);
2943 		avail--;
2944 		budget_left--;
2945 		if (avail == 0 && budget_left)
2946 			avail = iflib_rxd_avail(ctx, rxq, *cidxp, budget_left);
2947 
2948 		if (__predict_false(m == NULL))
2949 			continue;
2950 
2951 		/* imm_pkt: -- cxgb */
2952 		if (mh == NULL)
2953 			mh = mt = m;
2954 		else {
2955 			mt->m_nextpkt = m;
2956 			mt = m;
2957 		}
2958 	}
2959 	CURVNET_RESTORE();
2960 	/* make sure that we can refill faster than drain */
2961 	for (i = 0, fl = &rxq->ifr_fl[0]; i < sctx->isc_nfl; i++, fl++)
2962 		retval |= iflib_fl_refill_all(ctx, fl);
2963 
2964 	lro_enabled = (if_getcapenable(ifp) & IFCAP_LRO);
2965 	if (lro_enabled)
2966 		iflib_get_ip_forwarding(&rxq->ifr_lc, &v4_forwarding, &v6_forwarding);
2967 	mt = mf = NULL;
2968 	while (mh != NULL) {
2969 		m = mh;
2970 		mh = mh->m_nextpkt;
2971 		m->m_nextpkt = NULL;
2972 #ifndef __NO_STRICT_ALIGNMENT
2973 		if (!IP_ALIGNED(m) && (m = iflib_fixup_rx(m)) == NULL)
2974 			continue;
2975 #endif
2976 		rx_bytes += m->m_pkthdr.len;
2977 		rx_pkts++;
2978 #if defined(INET6) || defined(INET)
2979 		if (lro_enabled) {
2980 			if (!lro_possible) {
2981 				lro_possible = iflib_check_lro_possible(m, v4_forwarding, v6_forwarding);
2982 				if (lro_possible && mf != NULL) {
2983 					ifp->if_input(ifp, mf);
2984 					DBG_COUNTER_INC(rx_if_input);
2985 					mt = mf = NULL;
2986 				}
2987 			}
2988 			if ((m->m_pkthdr.csum_flags & (CSUM_L4_CALC|CSUM_L4_VALID)) ==
2989 			    (CSUM_L4_CALC|CSUM_L4_VALID)) {
2990 				if (lro_possible && tcp_lro_rx(&rxq->ifr_lc, m, 0) == 0)
2991 					continue;
2992 			}
2993 		}
2994 #endif
2995 		if (lro_possible) {
2996 			ifp->if_input(ifp, m);
2997 			DBG_COUNTER_INC(rx_if_input);
2998 			continue;
2999 		}
3000 
3001 		if (mf == NULL)
3002 			mf = m;
3003 		if (mt != NULL)
3004 			mt->m_nextpkt = m;
3005 		mt = m;
3006 	}
3007 	if (mf != NULL) {
3008 		ifp->if_input(ifp, mf);
3009 		DBG_COUNTER_INC(rx_if_input);
3010 	}
3011 
3012 	if_inc_counter(ifp, IFCOUNTER_IBYTES, rx_bytes);
3013 	if_inc_counter(ifp, IFCOUNTER_IPACKETS, rx_pkts);
3014 
3015 	/*
3016 	 * Flush any outstanding LRO work
3017 	 */
3018 #if defined(INET6) || defined(INET)
3019 	tcp_lro_flush_all(&rxq->ifr_lc);
3020 #endif
3021 	if (avail != 0 || iflib_rxd_avail(ctx, rxq, *cidxp, 1) != 0)
3022 		retval |= IFLIB_RXEOF_MORE;
3023 	return (retval);
3024 err:
3025 	STATE_LOCK(ctx);
3026 	ctx->ifc_flags |= IFC_DO_RESET;
3027 	iflib_admin_intr_deferred(ctx);
3028 	STATE_UNLOCK(ctx);
3029 	return (0);
3030 }
3031 
3032 #define TXD_NOTIFY_COUNT(txq) (((txq)->ift_size / (txq)->ift_update_freq)-1)
3033 static inline qidx_t
3034 txq_max_db_deferred(iflib_txq_t txq, qidx_t in_use)
3035 {
3036 	qidx_t notify_count = TXD_NOTIFY_COUNT(txq);
3037 	qidx_t minthresh = txq->ift_size / 8;
3038 	if (in_use > 4*minthresh)
3039 		return (notify_count);
3040 	if (in_use > 2*minthresh)
3041 		return (notify_count >> 1);
3042 	if (in_use > minthresh)
3043 		return (notify_count >> 3);
3044 	return (0);
3045 }
3046 
3047 static inline qidx_t
3048 txq_max_rs_deferred(iflib_txq_t txq)
3049 {
3050 	qidx_t notify_count = TXD_NOTIFY_COUNT(txq);
3051 	qidx_t minthresh = txq->ift_size / 8;
3052 	if (txq->ift_in_use > 4*minthresh)
3053 		return (notify_count);
3054 	if (txq->ift_in_use > 2*minthresh)
3055 		return (notify_count >> 1);
3056 	if (txq->ift_in_use > minthresh)
3057 		return (notify_count >> 2);
3058 	return (2);
3059 }
3060 
3061 #define M_CSUM_FLAGS(m) ((m)->m_pkthdr.csum_flags)
3062 #define M_HAS_VLANTAG(m) (m->m_flags & M_VLANTAG)
3063 
3064 #define TXQ_MAX_DB_DEFERRED(txq, in_use) txq_max_db_deferred((txq), (in_use))
3065 #define TXQ_MAX_RS_DEFERRED(txq) txq_max_rs_deferred(txq)
3066 #define TXQ_MAX_DB_CONSUMED(size) (size >> 4)
3067 
3068 /* forward compatibility for cxgb */
3069 #define FIRST_QSET(ctx) 0
3070 #define NTXQSETS(ctx) ((ctx)->ifc_softc_ctx.isc_ntxqsets)
3071 #define NRXQSETS(ctx) ((ctx)->ifc_softc_ctx.isc_nrxqsets)
3072 #define QIDX(ctx, m) ((((m)->m_pkthdr.flowid & ctx->ifc_softc_ctx.isc_rss_table_mask) % NTXQSETS(ctx)) + FIRST_QSET(ctx))
3073 #define DESC_RECLAIMABLE(q) ((int)((q)->ift_processed - (q)->ift_cleaned - (q)->ift_ctx->ifc_softc_ctx.isc_tx_nsegments))
3074 
3075 /* XXX we should be setting this to something other than zero */
3076 #define RECLAIM_THRESH(ctx) ((ctx)->ifc_sctx->isc_tx_reclaim_thresh)
3077 #define	MAX_TX_DESC(ctx) MAX((ctx)->ifc_softc_ctx.isc_tx_tso_segments_max, \
3078     (ctx)->ifc_softc_ctx.isc_tx_nsegments)
3079 
3080 static inline bool
3081 iflib_txd_db_check(iflib_txq_t txq, int ring)
3082 {
3083 	if_ctx_t ctx = txq->ift_ctx;
3084 	qidx_t dbval, max;
3085 
3086 	max = TXQ_MAX_DB_DEFERRED(txq, txq->ift_in_use);
3087 
3088 	/* force || threshold exceeded || at the edge of the ring */
3089 	if (ring || (txq->ift_db_pending >= max) || (TXQ_AVAIL(txq) <= MAX_TX_DESC(ctx) + 2)) {
3090 
3091 		/*
3092 		 * 'npending' is used if the card's doorbell is in terms of the number of descriptors
3093 		 * pending flush (BRCM). 'pidx' is used in cases where the card's doorbeel uses the
3094 		 * producer index explicitly (INTC).
3095 		 */
3096 		dbval = txq->ift_npending ? txq->ift_npending : txq->ift_pidx;
3097 		bus_dmamap_sync(txq->ift_ifdi->idi_tag, txq->ift_ifdi->idi_map,
3098 		    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
3099 		ctx->isc_txd_flush(ctx->ifc_softc, txq->ift_id, dbval);
3100 
3101 		/*
3102 		 * Absent bugs there are zero packets pending so reset pending counts to zero.
3103 		 */
3104 		txq->ift_db_pending = txq->ift_npending = 0;
3105 		return (true);
3106 	}
3107 	return (false);
3108 }
3109 
3110 #ifdef PKT_DEBUG
3111 static void
3112 print_pkt(if_pkt_info_t pi)
3113 {
3114 	printf("pi len:  %d qsidx: %d nsegs: %d ndescs: %d flags: %x pidx: %d\n",
3115 	       pi->ipi_len, pi->ipi_qsidx, pi->ipi_nsegs, pi->ipi_ndescs, pi->ipi_flags, pi->ipi_pidx);
3116 	printf("pi new_pidx: %d csum_flags: %lx tso_segsz: %d mflags: %x vtag: %d\n",
3117 	       pi->ipi_new_pidx, pi->ipi_csum_flags, pi->ipi_tso_segsz, pi->ipi_mflags, pi->ipi_vtag);
3118 	printf("pi etype: %d ehdrlen: %d ip_hlen: %d ipproto: %d\n",
3119 	       pi->ipi_etype, pi->ipi_ehdrlen, pi->ipi_ip_hlen, pi->ipi_ipproto);
3120 }
3121 #endif
3122 
3123 #define IS_TSO4(pi) ((pi)->ipi_csum_flags & CSUM_IP_TSO)
3124 #define IS_TX_OFFLOAD4(pi) ((pi)->ipi_csum_flags & (CSUM_IP_TCP | CSUM_IP_TSO))
3125 #define IS_TSO6(pi) ((pi)->ipi_csum_flags & CSUM_IP6_TSO)
3126 #define IS_TX_OFFLOAD6(pi) ((pi)->ipi_csum_flags & (CSUM_IP6_TCP | CSUM_IP6_TSO))
3127 
3128 static int
3129 iflib_parse_header(iflib_txq_t txq, if_pkt_info_t pi, struct mbuf **mp)
3130 {
3131 	if_shared_ctx_t sctx = txq->ift_ctx->ifc_sctx;
3132 	struct ether_vlan_header *eh;
3133 	struct mbuf *m;
3134 
3135 	m = *mp;
3136 	if ((sctx->isc_flags & IFLIB_NEED_SCRATCH) &&
3137 	    M_WRITABLE(m) == 0) {
3138 		if ((m = m_dup(m, M_NOWAIT)) == NULL) {
3139 			return (ENOMEM);
3140 		} else {
3141 			m_freem(*mp);
3142 			DBG_COUNTER_INC(tx_frees);
3143 			*mp = m;
3144 		}
3145 	}
3146 
3147 	/*
3148 	 * Determine where frame payload starts.
3149 	 * Jump over vlan headers if already present,
3150 	 * helpful for QinQ too.
3151 	 */
3152 	if (__predict_false(m->m_len < sizeof(*eh))) {
3153 		txq->ift_pullups++;
3154 		if (__predict_false((m = m_pullup(m, sizeof(*eh))) == NULL))
3155 			return (ENOMEM);
3156 	}
3157 	eh = mtod(m, struct ether_vlan_header *);
3158 	if (eh->evl_encap_proto == htons(ETHERTYPE_VLAN)) {
3159 		pi->ipi_etype = ntohs(eh->evl_proto);
3160 		pi->ipi_ehdrlen = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN;
3161 	} else {
3162 		pi->ipi_etype = ntohs(eh->evl_encap_proto);
3163 		pi->ipi_ehdrlen = ETHER_HDR_LEN;
3164 	}
3165 
3166 	switch (pi->ipi_etype) {
3167 #ifdef INET
3168 	case ETHERTYPE_IP:
3169 	{
3170 		struct mbuf *n;
3171 		struct ip *ip = NULL;
3172 		struct tcphdr *th = NULL;
3173 		int minthlen;
3174 
3175 		minthlen = min(m->m_pkthdr.len, pi->ipi_ehdrlen + sizeof(*ip) + sizeof(*th));
3176 		if (__predict_false(m->m_len < minthlen)) {
3177 			/*
3178 			 * if this code bloat is causing too much of a hit
3179 			 * move it to a separate function and mark it noinline
3180 			 */
3181 			if (m->m_len == pi->ipi_ehdrlen) {
3182 				n = m->m_next;
3183 				MPASS(n);
3184 				if (n->m_len >= sizeof(*ip))  {
3185 					ip = (struct ip *)n->m_data;
3186 					if (n->m_len >= (ip->ip_hl << 2) + sizeof(*th))
3187 						th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
3188 				} else {
3189 					txq->ift_pullups++;
3190 					if (__predict_false((m = m_pullup(m, minthlen)) == NULL))
3191 						return (ENOMEM);
3192 					ip = (struct ip *)(m->m_data + pi->ipi_ehdrlen);
3193 				}
3194 			} else {
3195 				txq->ift_pullups++;
3196 				if (__predict_false((m = m_pullup(m, minthlen)) == NULL))
3197 					return (ENOMEM);
3198 				ip = (struct ip *)(m->m_data + pi->ipi_ehdrlen);
3199 				if (m->m_len >= (ip->ip_hl << 2) + sizeof(*th))
3200 					th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
3201 			}
3202 		} else {
3203 			ip = (struct ip *)(m->m_data + pi->ipi_ehdrlen);
3204 			if (m->m_len >= (ip->ip_hl << 2) + sizeof(*th))
3205 				th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
3206 		}
3207 		pi->ipi_ip_hlen = ip->ip_hl << 2;
3208 		pi->ipi_ipproto = ip->ip_p;
3209 		pi->ipi_flags |= IPI_TX_IPV4;
3210 
3211 		/* TCP checksum offload may require TCP header length */
3212 		if (IS_TX_OFFLOAD4(pi)) {
3213 			if (__predict_true(pi->ipi_ipproto == IPPROTO_TCP)) {
3214 				if (__predict_false(th == NULL)) {
3215 					txq->ift_pullups++;
3216 					if (__predict_false((m = m_pullup(m, (ip->ip_hl << 2) + sizeof(*th))) == NULL))
3217 						return (ENOMEM);
3218 					th = (struct tcphdr *)((caddr_t)ip + pi->ipi_ip_hlen);
3219 				}
3220 				pi->ipi_tcp_hflags = th->th_flags;
3221 				pi->ipi_tcp_hlen = th->th_off << 2;
3222 				pi->ipi_tcp_seq = th->th_seq;
3223 			}
3224 			if (IS_TSO4(pi)) {
3225 				if (__predict_false(ip->ip_p != IPPROTO_TCP))
3226 					return (ENXIO);
3227 				/*
3228 				 * TSO always requires hardware checksum offload.
3229 				 */
3230 				pi->ipi_csum_flags |= (CSUM_IP_TCP | CSUM_IP);
3231 				th->th_sum = in_pseudo(ip->ip_src.s_addr,
3232 						       ip->ip_dst.s_addr, htons(IPPROTO_TCP));
3233 				pi->ipi_tso_segsz = m->m_pkthdr.tso_segsz;
3234 				if (sctx->isc_flags & IFLIB_TSO_INIT_IP) {
3235 					ip->ip_sum = 0;
3236 					ip->ip_len = htons(pi->ipi_ip_hlen + pi->ipi_tcp_hlen + pi->ipi_tso_segsz);
3237 				}
3238 			}
3239 		}
3240 		if ((sctx->isc_flags & IFLIB_NEED_ZERO_CSUM) && (pi->ipi_csum_flags & CSUM_IP))
3241                        ip->ip_sum = 0;
3242 
3243 		break;
3244 	}
3245 #endif
3246 #ifdef INET6
3247 	case ETHERTYPE_IPV6:
3248 	{
3249 		struct ip6_hdr *ip6 = (struct ip6_hdr *)(m->m_data + pi->ipi_ehdrlen);
3250 		struct tcphdr *th;
3251 		pi->ipi_ip_hlen = sizeof(struct ip6_hdr);
3252 
3253 		if (__predict_false(m->m_len < pi->ipi_ehdrlen + sizeof(struct ip6_hdr))) {
3254 			txq->ift_pullups++;
3255 			if (__predict_false((m = m_pullup(m, pi->ipi_ehdrlen + sizeof(struct ip6_hdr))) == NULL))
3256 				return (ENOMEM);
3257 		}
3258 		th = (struct tcphdr *)((caddr_t)ip6 + pi->ipi_ip_hlen);
3259 
3260 		/* XXX-BZ this will go badly in case of ext hdrs. */
3261 		pi->ipi_ipproto = ip6->ip6_nxt;
3262 		pi->ipi_flags |= IPI_TX_IPV6;
3263 
3264 		/* TCP checksum offload may require TCP header length */
3265 		if (IS_TX_OFFLOAD6(pi)) {
3266 			if (pi->ipi_ipproto == IPPROTO_TCP) {
3267 				if (__predict_false(m->m_len < pi->ipi_ehdrlen + sizeof(struct ip6_hdr) + sizeof(struct tcphdr))) {
3268 					txq->ift_pullups++;
3269 					if (__predict_false((m = m_pullup(m, pi->ipi_ehdrlen + sizeof(struct ip6_hdr) + sizeof(struct tcphdr))) == NULL))
3270 						return (ENOMEM);
3271 				}
3272 				pi->ipi_tcp_hflags = th->th_flags;
3273 				pi->ipi_tcp_hlen = th->th_off << 2;
3274 				pi->ipi_tcp_seq = th->th_seq;
3275 			}
3276 			if (IS_TSO6(pi)) {
3277 				if (__predict_false(ip6->ip6_nxt != IPPROTO_TCP))
3278 					return (ENXIO);
3279 				/*
3280 				 * TSO always requires hardware checksum offload.
3281 				 */
3282 				pi->ipi_csum_flags |= CSUM_IP6_TCP;
3283 				th->th_sum = in6_cksum_pseudo(ip6, 0, IPPROTO_TCP, 0);
3284 				pi->ipi_tso_segsz = m->m_pkthdr.tso_segsz;
3285 			}
3286 		}
3287 		break;
3288 	}
3289 #endif
3290 	default:
3291 		pi->ipi_csum_flags &= ~CSUM_OFFLOAD;
3292 		pi->ipi_ip_hlen = 0;
3293 		break;
3294 	}
3295 	*mp = m;
3296 
3297 	return (0);
3298 }
3299 
3300 /*
3301  * If dodgy hardware rejects the scatter gather chain we've handed it
3302  * we'll need to remove the mbuf chain from ifsg_m[] before we can add the
3303  * m_defrag'd mbufs
3304  */
3305 static __noinline struct mbuf *
3306 iflib_remove_mbuf(iflib_txq_t txq)
3307 {
3308 	int ntxd, pidx;
3309 	struct mbuf *m, **ifsd_m;
3310 
3311 	ifsd_m = txq->ift_sds.ifsd_m;
3312 	ntxd = txq->ift_size;
3313 	pidx = txq->ift_pidx & (ntxd - 1);
3314 	ifsd_m = txq->ift_sds.ifsd_m;
3315 	m = ifsd_m[pidx];
3316 	ifsd_m[pidx] = NULL;
3317 	bus_dmamap_unload(txq->ift_buf_tag, txq->ift_sds.ifsd_map[pidx]);
3318 	if (txq->ift_sds.ifsd_tso_map != NULL)
3319 		bus_dmamap_unload(txq->ift_tso_buf_tag,
3320 		    txq->ift_sds.ifsd_tso_map[pidx]);
3321 #if MEMORY_LOGGING
3322 	txq->ift_dequeued++;
3323 #endif
3324 	return (m);
3325 }
3326 
3327 static inline caddr_t
3328 calc_next_txd(iflib_txq_t txq, int cidx, uint8_t qid)
3329 {
3330 	qidx_t size;
3331 	int ntxd;
3332 	caddr_t start, end, cur, next;
3333 
3334 	ntxd = txq->ift_size;
3335 	size = txq->ift_txd_size[qid];
3336 	start = txq->ift_ifdi[qid].idi_vaddr;
3337 
3338 	if (__predict_false(size == 0))
3339 		return (start);
3340 	cur = start + size*cidx;
3341 	end = start + size*ntxd;
3342 	next = CACHE_PTR_NEXT(cur);
3343 	return (next < end ? next : start);
3344 }
3345 
3346 /*
3347  * Pad an mbuf to ensure a minimum ethernet frame size.
3348  * min_frame_size is the frame size (less CRC) to pad the mbuf to
3349  */
3350 static __noinline int
3351 iflib_ether_pad(device_t dev, struct mbuf **m_head, uint16_t min_frame_size)
3352 {
3353 	/*
3354 	 * 18 is enough bytes to pad an ARP packet to 46 bytes, and
3355 	 * and ARP message is the smallest common payload I can think of
3356 	 */
3357 	static char pad[18];	/* just zeros */
3358 	int n;
3359 	struct mbuf *new_head;
3360 
3361 	if (!M_WRITABLE(*m_head)) {
3362 		new_head = m_dup(*m_head, M_NOWAIT);
3363 		if (new_head == NULL) {
3364 			m_freem(*m_head);
3365 			device_printf(dev, "cannot pad short frame, m_dup() failed");
3366 			DBG_COUNTER_INC(encap_pad_mbuf_fail);
3367 			DBG_COUNTER_INC(tx_frees);
3368 			return ENOMEM;
3369 		}
3370 		m_freem(*m_head);
3371 		*m_head = new_head;
3372 	}
3373 
3374 	for (n = min_frame_size - (*m_head)->m_pkthdr.len;
3375 	     n > 0; n -= sizeof(pad))
3376 		if (!m_append(*m_head, min(n, sizeof(pad)), pad))
3377 			break;
3378 
3379 	if (n > 0) {
3380 		m_freem(*m_head);
3381 		device_printf(dev, "cannot pad short frame\n");
3382 		DBG_COUNTER_INC(encap_pad_mbuf_fail);
3383 		DBG_COUNTER_INC(tx_frees);
3384 		return (ENOBUFS);
3385 	}
3386 
3387 	return 0;
3388 }
3389 
3390 static int
3391 iflib_encap(iflib_txq_t txq, struct mbuf **m_headp)
3392 {
3393 	if_ctx_t		ctx;
3394 	if_shared_ctx_t		sctx;
3395 	if_softc_ctx_t		scctx;
3396 	bus_dma_tag_t		buf_tag;
3397 	bus_dma_segment_t	*segs;
3398 	struct mbuf		*m_head, **ifsd_m;
3399 	void			*next_txd;
3400 	bus_dmamap_t		map;
3401 	struct if_pkt_info	pi;
3402 	int remap = 0;
3403 	int err, nsegs, ndesc, max_segs, pidx, cidx, next, ntxd;
3404 
3405 	ctx = txq->ift_ctx;
3406 	sctx = ctx->ifc_sctx;
3407 	scctx = &ctx->ifc_softc_ctx;
3408 	segs = txq->ift_segs;
3409 	ntxd = txq->ift_size;
3410 	m_head = *m_headp;
3411 	map = NULL;
3412 
3413 	/*
3414 	 * If we're doing TSO the next descriptor to clean may be quite far ahead
3415 	 */
3416 	cidx = txq->ift_cidx;
3417 	pidx = txq->ift_pidx;
3418 	if (ctx->ifc_flags & IFC_PREFETCH) {
3419 		next = (cidx + CACHE_PTR_INCREMENT) & (ntxd-1);
3420 		if (!(ctx->ifc_flags & IFLIB_HAS_TXCQ)) {
3421 			next_txd = calc_next_txd(txq, cidx, 0);
3422 			prefetch(next_txd);
3423 		}
3424 
3425 		/* prefetch the next cache line of mbuf pointers and flags */
3426 		prefetch(&txq->ift_sds.ifsd_m[next]);
3427 		prefetch(&txq->ift_sds.ifsd_map[next]);
3428 		next = (cidx + CACHE_LINE_SIZE) & (ntxd-1);
3429 	}
3430 	map = txq->ift_sds.ifsd_map[pidx];
3431 	ifsd_m = txq->ift_sds.ifsd_m;
3432 
3433 	if (m_head->m_pkthdr.csum_flags & CSUM_TSO) {
3434 		buf_tag = txq->ift_tso_buf_tag;
3435 		max_segs = scctx->isc_tx_tso_segments_max;
3436 		map = txq->ift_sds.ifsd_tso_map[pidx];
3437 		MPASS(buf_tag != NULL);
3438 		MPASS(max_segs > 0);
3439 	} else {
3440 		buf_tag = txq->ift_buf_tag;
3441 		max_segs = scctx->isc_tx_nsegments;
3442 		map = txq->ift_sds.ifsd_map[pidx];
3443 	}
3444 	if ((sctx->isc_flags & IFLIB_NEED_ETHER_PAD) &&
3445 	    __predict_false(m_head->m_pkthdr.len < scctx->isc_min_frame_size)) {
3446 		err = iflib_ether_pad(ctx->ifc_dev, m_headp, scctx->isc_min_frame_size);
3447 		if (err) {
3448 			DBG_COUNTER_INC(encap_txd_encap_fail);
3449 			return err;
3450 		}
3451 	}
3452 	m_head = *m_headp;
3453 
3454 	pkt_info_zero(&pi);
3455 	pi.ipi_mflags = (m_head->m_flags & (M_VLANTAG|M_BCAST|M_MCAST));
3456 	pi.ipi_pidx = pidx;
3457 	pi.ipi_qsidx = txq->ift_id;
3458 	pi.ipi_len = m_head->m_pkthdr.len;
3459 	pi.ipi_csum_flags = m_head->m_pkthdr.csum_flags;
3460 	pi.ipi_vtag = M_HAS_VLANTAG(m_head) ? m_head->m_pkthdr.ether_vtag : 0;
3461 
3462 	/* deliberate bitwise OR to make one condition */
3463 	if (__predict_true((pi.ipi_csum_flags | pi.ipi_vtag))) {
3464 		if (__predict_false((err = iflib_parse_header(txq, &pi, m_headp)) != 0)) {
3465 			DBG_COUNTER_INC(encap_txd_encap_fail);
3466 			return (err);
3467 		}
3468 		m_head = *m_headp;
3469 	}
3470 
3471 retry:
3472 	err = bus_dmamap_load_mbuf_sg(buf_tag, map, m_head, segs, &nsegs,
3473 	    BUS_DMA_NOWAIT);
3474 defrag:
3475 	if (__predict_false(err)) {
3476 		switch (err) {
3477 		case EFBIG:
3478 			/* try collapse once and defrag once */
3479 			if (remap == 0) {
3480 				m_head = m_collapse(*m_headp, M_NOWAIT, max_segs);
3481 				/* try defrag if collapsing fails */
3482 				if (m_head == NULL)
3483 					remap++;
3484 			}
3485 			if (remap == 1) {
3486 				txq->ift_mbuf_defrag++;
3487 				m_head = m_defrag(*m_headp, M_NOWAIT);
3488 			}
3489 			/*
3490 			 * remap should never be >1 unless bus_dmamap_load_mbuf_sg
3491 			 * failed to map an mbuf that was run through m_defrag
3492 			 */
3493 			MPASS(remap <= 1);
3494 			if (__predict_false(m_head == NULL || remap > 1))
3495 				goto defrag_failed;
3496 			remap++;
3497 			*m_headp = m_head;
3498 			goto retry;
3499 			break;
3500 		case ENOMEM:
3501 			txq->ift_no_tx_dma_setup++;
3502 			break;
3503 		default:
3504 			txq->ift_no_tx_dma_setup++;
3505 			m_freem(*m_headp);
3506 			DBG_COUNTER_INC(tx_frees);
3507 			*m_headp = NULL;
3508 			break;
3509 		}
3510 		txq->ift_map_failed++;
3511 		DBG_COUNTER_INC(encap_load_mbuf_fail);
3512 		DBG_COUNTER_INC(encap_txd_encap_fail);
3513 		return (err);
3514 	}
3515 	ifsd_m[pidx] = m_head;
3516 	/*
3517 	 * XXX assumes a 1 to 1 relationship between segments and
3518 	 *        descriptors - this does not hold true on all drivers, e.g.
3519 	 *        cxgb
3520 	 */
3521 	if (__predict_false(nsegs + 2 > TXQ_AVAIL(txq))) {
3522 		txq->ift_no_desc_avail++;
3523 		bus_dmamap_unload(buf_tag, map);
3524 		DBG_COUNTER_INC(encap_txq_avail_fail);
3525 		DBG_COUNTER_INC(encap_txd_encap_fail);
3526 		if ((txq->ift_task.gt_task.ta_flags & TASK_ENQUEUED) == 0)
3527 			GROUPTASK_ENQUEUE(&txq->ift_task);
3528 		return (ENOBUFS);
3529 	}
3530 	/*
3531 	 * On Intel cards we can greatly reduce the number of TX interrupts
3532 	 * we see by only setting report status on every Nth descriptor.
3533 	 * However, this also means that the driver will need to keep track
3534 	 * of the descriptors that RS was set on to check them for the DD bit.
3535 	 */
3536 	txq->ift_rs_pending += nsegs + 1;
3537 	if (txq->ift_rs_pending > TXQ_MAX_RS_DEFERRED(txq) ||
3538 	     iflib_no_tx_batch || (TXQ_AVAIL(txq) - nsegs) <= MAX_TX_DESC(ctx) + 2) {
3539 		pi.ipi_flags |= IPI_TX_INTR;
3540 		txq->ift_rs_pending = 0;
3541 	}
3542 
3543 	pi.ipi_segs = segs;
3544 	pi.ipi_nsegs = nsegs;
3545 
3546 	MPASS(pidx >= 0 && pidx < txq->ift_size);
3547 #ifdef PKT_DEBUG
3548 	print_pkt(&pi);
3549 #endif
3550 	if ((err = ctx->isc_txd_encap(ctx->ifc_softc, &pi)) == 0) {
3551 		bus_dmamap_sync(buf_tag, map, BUS_DMASYNC_PREWRITE);
3552 		DBG_COUNTER_INC(tx_encap);
3553 		MPASS(pi.ipi_new_pidx < txq->ift_size);
3554 
3555 		ndesc = pi.ipi_new_pidx - pi.ipi_pidx;
3556 		if (pi.ipi_new_pidx < pi.ipi_pidx) {
3557 			ndesc += txq->ift_size;
3558 			txq->ift_gen = 1;
3559 		}
3560 		/*
3561 		 * drivers can need as many as
3562 		 * two sentinels
3563 		 */
3564 		MPASS(ndesc <= pi.ipi_nsegs + 2);
3565 		MPASS(pi.ipi_new_pidx != pidx);
3566 		MPASS(ndesc > 0);
3567 		txq->ift_in_use += ndesc;
3568 		txq->ift_db_pending += ndesc;
3569 
3570 		/*
3571 		 * We update the last software descriptor again here because there may
3572 		 * be a sentinel and/or there may be more mbufs than segments
3573 		 */
3574 		txq->ift_pidx = pi.ipi_new_pidx;
3575 		txq->ift_npending += pi.ipi_ndescs;
3576 	} else {
3577 		*m_headp = m_head = iflib_remove_mbuf(txq);
3578 		if (err == EFBIG) {
3579 			txq->ift_txd_encap_efbig++;
3580 			if (remap < 2) {
3581 				remap = 1;
3582 				goto defrag;
3583 			}
3584 		}
3585 		goto defrag_failed;
3586 	}
3587 	/*
3588 	 * err can't possibly be non-zero here, so we don't neet to test it
3589 	 * to see if we need to DBG_COUNTER_INC(encap_txd_encap_fail).
3590 	 */
3591 	return (err);
3592 
3593 defrag_failed:
3594 	txq->ift_mbuf_defrag_failed++;
3595 	txq->ift_map_failed++;
3596 	m_freem(*m_headp);
3597 	DBG_COUNTER_INC(tx_frees);
3598 	*m_headp = NULL;
3599 	DBG_COUNTER_INC(encap_txd_encap_fail);
3600 	return (ENOMEM);
3601 }
3602 
3603 static void
3604 iflib_tx_desc_free(iflib_txq_t txq, int n)
3605 {
3606 	uint32_t qsize, cidx, mask, gen;
3607 	struct mbuf *m, **ifsd_m;
3608 	bool do_prefetch;
3609 
3610 	cidx = txq->ift_cidx;
3611 	gen = txq->ift_gen;
3612 	qsize = txq->ift_size;
3613 	mask = qsize-1;
3614 	ifsd_m = txq->ift_sds.ifsd_m;
3615 	do_prefetch = (txq->ift_ctx->ifc_flags & IFC_PREFETCH);
3616 
3617 	while (n-- > 0) {
3618 		if (do_prefetch) {
3619 			prefetch(ifsd_m[(cidx + 3) & mask]);
3620 			prefetch(ifsd_m[(cidx + 4) & mask]);
3621 		}
3622 		if ((m = ifsd_m[cidx]) != NULL) {
3623 			prefetch(&ifsd_m[(cidx + CACHE_PTR_INCREMENT) & mask]);
3624 			if (m->m_pkthdr.csum_flags & CSUM_TSO) {
3625 				bus_dmamap_sync(txq->ift_tso_buf_tag,
3626 				    txq->ift_sds.ifsd_tso_map[cidx],
3627 				    BUS_DMASYNC_POSTWRITE);
3628 				bus_dmamap_unload(txq->ift_tso_buf_tag,
3629 				    txq->ift_sds.ifsd_tso_map[cidx]);
3630 			} else {
3631 				bus_dmamap_sync(txq->ift_buf_tag,
3632 				    txq->ift_sds.ifsd_map[cidx],
3633 				    BUS_DMASYNC_POSTWRITE);
3634 				bus_dmamap_unload(txq->ift_buf_tag,
3635 				    txq->ift_sds.ifsd_map[cidx]);
3636 			}
3637 			/* XXX we don't support any drivers that batch packets yet */
3638 			MPASS(m->m_nextpkt == NULL);
3639 			m_freem(m);
3640 			ifsd_m[cidx] = NULL;
3641 #if MEMORY_LOGGING
3642 			txq->ift_dequeued++;
3643 #endif
3644 			DBG_COUNTER_INC(tx_frees);
3645 		}
3646 		if (__predict_false(++cidx == qsize)) {
3647 			cidx = 0;
3648 			gen = 0;
3649 		}
3650 	}
3651 	txq->ift_cidx = cidx;
3652 	txq->ift_gen = gen;
3653 }
3654 
3655 static __inline int
3656 iflib_completed_tx_reclaim(iflib_txq_t txq, int thresh)
3657 {
3658 	int reclaim;
3659 	if_ctx_t ctx = txq->ift_ctx;
3660 
3661 	KASSERT(thresh >= 0, ("invalid threshold to reclaim"));
3662 	MPASS(thresh /*+ MAX_TX_DESC(txq->ift_ctx) */ < txq->ift_size);
3663 
3664 	/*
3665 	 * Need a rate-limiting check so that this isn't called every time
3666 	 */
3667 	iflib_tx_credits_update(ctx, txq);
3668 	reclaim = DESC_RECLAIMABLE(txq);
3669 
3670 	if (reclaim <= thresh /* + MAX_TX_DESC(txq->ift_ctx) */) {
3671 #ifdef INVARIANTS
3672 		if (iflib_verbose_debug) {
3673 			printf("%s processed=%ju cleaned=%ju tx_nsegments=%d reclaim=%d thresh=%d\n", __FUNCTION__,
3674 			       txq->ift_processed, txq->ift_cleaned, txq->ift_ctx->ifc_softc_ctx.isc_tx_nsegments,
3675 			       reclaim, thresh);
3676 		}
3677 #endif
3678 		return (0);
3679 	}
3680 	iflib_tx_desc_free(txq, reclaim);
3681 	txq->ift_cleaned += reclaim;
3682 	txq->ift_in_use -= reclaim;
3683 
3684 	return (reclaim);
3685 }
3686 
3687 static struct mbuf **
3688 _ring_peek_one(struct ifmp_ring *r, int cidx, int offset, int remaining)
3689 {
3690 	int next, size;
3691 	struct mbuf **items;
3692 
3693 	size = r->size;
3694 	next = (cidx + CACHE_PTR_INCREMENT) & (size-1);
3695 	items = __DEVOLATILE(struct mbuf **, &r->items[0]);
3696 
3697 	prefetch(items[(cidx + offset) & (size-1)]);
3698 	if (remaining > 1) {
3699 		prefetch2cachelines(&items[next]);
3700 		prefetch2cachelines(items[(cidx + offset + 1) & (size-1)]);
3701 		prefetch2cachelines(items[(cidx + offset + 2) & (size-1)]);
3702 		prefetch2cachelines(items[(cidx + offset + 3) & (size-1)]);
3703 	}
3704 	return (__DEVOLATILE(struct mbuf **, &r->items[(cidx + offset) & (size-1)]));
3705 }
3706 
3707 static void
3708 iflib_txq_check_drain(iflib_txq_t txq, int budget)
3709 {
3710 
3711 	ifmp_ring_check_drainage(txq->ift_br, budget);
3712 }
3713 
3714 static uint32_t
3715 iflib_txq_can_drain(struct ifmp_ring *r)
3716 {
3717 	iflib_txq_t txq = r->cookie;
3718 	if_ctx_t ctx = txq->ift_ctx;
3719 
3720 	if (TXQ_AVAIL(txq) > MAX_TX_DESC(ctx) + 2)
3721 		return (1);
3722 	bus_dmamap_sync(txq->ift_ifdi->idi_tag, txq->ift_ifdi->idi_map,
3723 	    BUS_DMASYNC_POSTREAD);
3724 	return (ctx->isc_txd_credits_update(ctx->ifc_softc, txq->ift_id,
3725 	    false));
3726 }
3727 
3728 static uint32_t
3729 iflib_txq_drain(struct ifmp_ring *r, uint32_t cidx, uint32_t pidx)
3730 {
3731 	iflib_txq_t txq = r->cookie;
3732 	if_ctx_t ctx = txq->ift_ctx;
3733 	if_t ifp = ctx->ifc_ifp;
3734 	struct mbuf *m, **mp;
3735 	int avail, bytes_sent, skipped, count, err, i;
3736 	int mcast_sent, pkt_sent, reclaimed;
3737 	bool do_prefetch, rang, ring;
3738 
3739 	if (__predict_false(!(if_getdrvflags(ifp) & IFF_DRV_RUNNING) ||
3740 			    !LINK_ACTIVE(ctx))) {
3741 		DBG_COUNTER_INC(txq_drain_notready);
3742 		return (0);
3743 	}
3744 	reclaimed = iflib_completed_tx_reclaim(txq, RECLAIM_THRESH(ctx));
3745 	rang = iflib_txd_db_check(txq, reclaimed && txq->ift_db_pending);
3746 	avail = IDXDIFF(pidx, cidx, r->size);
3747 
3748 	if (__predict_false(ctx->ifc_flags & IFC_QFLUSH)) {
3749 		/*
3750 		 * The driver is unloading so we need to free all pending packets.
3751 		 */
3752 		DBG_COUNTER_INC(txq_drain_flushing);
3753 		for (i = 0; i < avail; i++) {
3754 			if (__predict_true(r->items[(cidx + i) & (r->size-1)] != (void *)txq))
3755 				m_freem(r->items[(cidx + i) & (r->size-1)]);
3756 			r->items[(cidx + i) & (r->size-1)] = NULL;
3757 		}
3758 		return (avail);
3759 	}
3760 
3761 	if (__predict_false(if_getdrvflags(ctx->ifc_ifp) & IFF_DRV_OACTIVE)) {
3762 		txq->ift_qstatus = IFLIB_QUEUE_IDLE;
3763 		CALLOUT_LOCK(txq);
3764 		callout_stop(&txq->ift_timer);
3765 		CALLOUT_UNLOCK(txq);
3766 		DBG_COUNTER_INC(txq_drain_oactive);
3767 		return (0);
3768 	}
3769 
3770 	/*
3771 	 * If we've reclaimed any packets this queue cannot be hung.
3772 	 */
3773 	if (reclaimed)
3774 		txq->ift_qstatus = IFLIB_QUEUE_IDLE;
3775 	skipped = mcast_sent = bytes_sent = pkt_sent = 0;
3776 	count = MIN(avail, TX_BATCH_SIZE);
3777 #ifdef INVARIANTS
3778 	if (iflib_verbose_debug)
3779 		printf("%s avail=%d ifc_flags=%x txq_avail=%d ", __FUNCTION__,
3780 		       avail, ctx->ifc_flags, TXQ_AVAIL(txq));
3781 #endif
3782 	do_prefetch = (ctx->ifc_flags & IFC_PREFETCH);
3783 	err = 0;
3784 	for (i = 0; i < count && TXQ_AVAIL(txq) >= MAX_TX_DESC(ctx) + 2; i++) {
3785 		int rem = do_prefetch ? count - i : 0;
3786 
3787 		mp = _ring_peek_one(r, cidx, i, rem);
3788 		MPASS(mp != NULL && *mp != NULL);
3789 
3790 		/*
3791 		 * Completion interrupts will use the address of the txq
3792 		 * as a sentinel to enqueue _something_ in order to acquire
3793 		 * the lock on the mp_ring (there's no direct lock call).
3794 		 * We obviously whave to check for these sentinel cases
3795 		 * and skip them.
3796 		 */
3797 		if (__predict_false(*mp == (struct mbuf *)txq)) {
3798 			skipped++;
3799 			continue;
3800 		}
3801 		err = iflib_encap(txq, mp);
3802 		if (__predict_false(err)) {
3803 			/* no room - bail out */
3804 			if (err == ENOBUFS)
3805 				break;
3806 			skipped++;
3807 			/* we can't send this packet - skip it */
3808 			continue;
3809 		}
3810 		pkt_sent++;
3811 		m = *mp;
3812 		DBG_COUNTER_INC(tx_sent);
3813 		bytes_sent += m->m_pkthdr.len;
3814 		mcast_sent += !!(m->m_flags & M_MCAST);
3815 
3816 		if (__predict_false(!(ifp->if_drv_flags & IFF_DRV_RUNNING)))
3817 			break;
3818 		ETHER_BPF_MTAP(ifp, m);
3819 		rang = iflib_txd_db_check(txq, false);
3820 	}
3821 
3822 	/* deliberate use of bitwise or to avoid gratuitous short-circuit */
3823 	ring = rang ? false  : (iflib_min_tx_latency | err);
3824 	iflib_txd_db_check(txq, ring);
3825 	if_inc_counter(ifp, IFCOUNTER_OBYTES, bytes_sent);
3826 	if_inc_counter(ifp, IFCOUNTER_OPACKETS, pkt_sent);
3827 	if (mcast_sent)
3828 		if_inc_counter(ifp, IFCOUNTER_OMCASTS, mcast_sent);
3829 #ifdef INVARIANTS
3830 	if (iflib_verbose_debug)
3831 		printf("consumed=%d\n", skipped + pkt_sent);
3832 #endif
3833 	return (skipped + pkt_sent);
3834 }
3835 
3836 static uint32_t
3837 iflib_txq_drain_always(struct ifmp_ring *r)
3838 {
3839 	return (1);
3840 }
3841 
3842 static uint32_t
3843 iflib_txq_drain_free(struct ifmp_ring *r, uint32_t cidx, uint32_t pidx)
3844 {
3845 	int i, avail;
3846 	struct mbuf **mp;
3847 	iflib_txq_t txq;
3848 
3849 	txq = r->cookie;
3850 
3851 	txq->ift_qstatus = IFLIB_QUEUE_IDLE;
3852 	CALLOUT_LOCK(txq);
3853 	callout_stop(&txq->ift_timer);
3854 	CALLOUT_UNLOCK(txq);
3855 
3856 	avail = IDXDIFF(pidx, cidx, r->size);
3857 	for (i = 0; i < avail; i++) {
3858 		mp = _ring_peek_one(r, cidx, i, avail - i);
3859 		if (__predict_false(*mp == (struct mbuf *)txq))
3860 			continue;
3861 		m_freem(*mp);
3862 		DBG_COUNTER_INC(tx_frees);
3863 	}
3864 	MPASS(ifmp_ring_is_stalled(r) == 0);
3865 	return (avail);
3866 }
3867 
3868 static void
3869 iflib_ifmp_purge(iflib_txq_t txq)
3870 {
3871 	struct ifmp_ring *r;
3872 
3873 	r = txq->ift_br;
3874 	r->drain = iflib_txq_drain_free;
3875 	r->can_drain = iflib_txq_drain_always;
3876 
3877 	ifmp_ring_check_drainage(r, r->size);
3878 
3879 	r->drain = iflib_txq_drain;
3880 	r->can_drain = iflib_txq_can_drain;
3881 }
3882 
3883 static void
3884 _task_fn_tx(void *context)
3885 {
3886 	iflib_txq_t txq = context;
3887 	if_ctx_t ctx = txq->ift_ctx;
3888 	if_t ifp = ctx->ifc_ifp;
3889 	int abdicate = ctx->ifc_sysctl_tx_abdicate;
3890 
3891 #ifdef IFLIB_DIAGNOSTICS
3892 	txq->ift_cpu_exec_count[curcpu]++;
3893 #endif
3894 	if (!(if_getdrvflags(ifp) & IFF_DRV_RUNNING))
3895 		return;
3896 #ifdef DEV_NETMAP
3897 	if ((if_getcapenable(ifp) & IFCAP_NETMAP) &&
3898 	    netmap_tx_irq(ifp, txq->ift_id))
3899 		goto skip_ifmp;
3900 #endif
3901 #ifdef ALTQ
3902 	if (ALTQ_IS_ENABLED(&ifp->if_snd))
3903 		iflib_altq_if_start(ifp);
3904 #endif
3905 	if (txq->ift_db_pending)
3906 		ifmp_ring_enqueue(txq->ift_br, (void **)&txq, 1, TX_BATCH_SIZE, abdicate);
3907 	else if (!abdicate)
3908 		ifmp_ring_check_drainage(txq->ift_br, TX_BATCH_SIZE);
3909 	/*
3910 	 * When abdicating, we always need to check drainage, not just when we don't enqueue
3911 	 */
3912 	if (abdicate)
3913 		ifmp_ring_check_drainage(txq->ift_br, TX_BATCH_SIZE);
3914 #ifdef DEV_NETMAP
3915 skip_ifmp:
3916 #endif
3917 	if (ctx->ifc_flags & IFC_LEGACY)
3918 		IFDI_INTR_ENABLE(ctx);
3919 	else
3920 		IFDI_TX_QUEUE_INTR_ENABLE(ctx, txq->ift_id);
3921 }
3922 
3923 static void
3924 _task_fn_rx(void *context)
3925 {
3926 	iflib_rxq_t rxq = context;
3927 	if_ctx_t ctx = rxq->ifr_ctx;
3928 	uint8_t more;
3929 	uint16_t budget;
3930 #ifdef DEV_NETMAP
3931 	u_int work = 0;
3932 	int nmirq;
3933 #endif
3934 
3935 #ifdef IFLIB_DIAGNOSTICS
3936 	rxq->ifr_cpu_exec_count[curcpu]++;
3937 #endif
3938 	DBG_COUNTER_INC(task_fn_rxs);
3939 	if (__predict_false(!(if_getdrvflags(ctx->ifc_ifp) & IFF_DRV_RUNNING)))
3940 		return;
3941 #ifdef DEV_NETMAP
3942 	nmirq = netmap_rx_irq(ctx->ifc_ifp, rxq->ifr_id, &work);
3943 	if (nmirq != NM_IRQ_PASS) {
3944 		more = (nmirq == NM_IRQ_RESCHED) ? IFLIB_RXEOF_MORE : 0;
3945 		goto skip_rxeof;
3946 	}
3947 #endif
3948 	budget = ctx->ifc_sysctl_rx_budget;
3949 	if (budget == 0)
3950 		budget = 16;	/* XXX */
3951 	more = iflib_rxeof(rxq, budget);
3952 #ifdef DEV_NETMAP
3953 skip_rxeof:
3954 #endif
3955 	if ((more & IFLIB_RXEOF_MORE) == 0) {
3956 		if (ctx->ifc_flags & IFC_LEGACY)
3957 			IFDI_INTR_ENABLE(ctx);
3958 		else
3959 			IFDI_RX_QUEUE_INTR_ENABLE(ctx, rxq->ifr_id);
3960 		DBG_COUNTER_INC(rx_intr_enables);
3961 	}
3962 	if (__predict_false(!(if_getdrvflags(ctx->ifc_ifp) & IFF_DRV_RUNNING)))
3963 		return;
3964 
3965 	if (more & IFLIB_RXEOF_MORE)
3966 		GROUPTASK_ENQUEUE(&rxq->ifr_task);
3967 	else if (more & IFLIB_RXEOF_EMPTY)
3968 		callout_reset_curcpu(&rxq->ifr_watchdog, 1, &_task_fn_rx_watchdog, rxq);
3969 }
3970 
3971 static void
3972 _task_fn_admin(void *context)
3973 {
3974 	if_ctx_t ctx = context;
3975 	if_softc_ctx_t sctx = &ctx->ifc_softc_ctx;
3976 	iflib_txq_t txq;
3977 	int i;
3978 	bool oactive, running, do_reset, do_watchdog, in_detach;
3979 
3980 	STATE_LOCK(ctx);
3981 	running = (if_getdrvflags(ctx->ifc_ifp) & IFF_DRV_RUNNING);
3982 	oactive = (if_getdrvflags(ctx->ifc_ifp) & IFF_DRV_OACTIVE);
3983 	do_reset = (ctx->ifc_flags & IFC_DO_RESET);
3984 	do_watchdog = (ctx->ifc_flags & IFC_DO_WATCHDOG);
3985 	in_detach = (ctx->ifc_flags & IFC_IN_DETACH);
3986 	ctx->ifc_flags &= ~(IFC_DO_RESET|IFC_DO_WATCHDOG);
3987 	STATE_UNLOCK(ctx);
3988 
3989 	if ((!running && !oactive) && !(ctx->ifc_sctx->isc_flags & IFLIB_ADMIN_ALWAYS_RUN))
3990 		return;
3991 	if (in_detach)
3992 		return;
3993 
3994 	CTX_LOCK(ctx);
3995 	for (txq = ctx->ifc_txqs, i = 0; i < sctx->isc_ntxqsets; i++, txq++) {
3996 		CALLOUT_LOCK(txq);
3997 		callout_stop(&txq->ift_timer);
3998 		CALLOUT_UNLOCK(txq);
3999 	}
4000 	if (do_watchdog) {
4001 		ctx->ifc_watchdog_events++;
4002 		IFDI_WATCHDOG_RESET(ctx);
4003 	}
4004 	IFDI_UPDATE_ADMIN_STATUS(ctx);
4005 	for (txq = ctx->ifc_txqs, i = 0; i < sctx->isc_ntxqsets; i++, txq++) {
4006 		callout_reset_on(&txq->ift_timer, iflib_timer_default, iflib_timer, txq,
4007 		    txq->ift_timer.c_cpu);
4008 	}
4009 	IFDI_LINK_INTR_ENABLE(ctx);
4010 	if (do_reset)
4011 		iflib_if_init_locked(ctx);
4012 	CTX_UNLOCK(ctx);
4013 
4014 	if (LINK_ACTIVE(ctx) == 0)
4015 		return;
4016 	for (txq = ctx->ifc_txqs, i = 0; i < sctx->isc_ntxqsets; i++, txq++)
4017 		iflib_txq_check_drain(txq, IFLIB_RESTART_BUDGET);
4018 }
4019 
4020 static void
4021 _task_fn_iov(void *context)
4022 {
4023 	if_ctx_t ctx = context;
4024 
4025 	if (!(if_getdrvflags(ctx->ifc_ifp) & IFF_DRV_RUNNING) &&
4026 	    !(ctx->ifc_sctx->isc_flags & IFLIB_ADMIN_ALWAYS_RUN))
4027 		return;
4028 
4029 	CTX_LOCK(ctx);
4030 	IFDI_VFLR_HANDLE(ctx);
4031 	CTX_UNLOCK(ctx);
4032 }
4033 
4034 static int
4035 iflib_sysctl_int_delay(SYSCTL_HANDLER_ARGS)
4036 {
4037 	int err;
4038 	if_int_delay_info_t info;
4039 	if_ctx_t ctx;
4040 
4041 	info = (if_int_delay_info_t)arg1;
4042 	ctx = info->iidi_ctx;
4043 	info->iidi_req = req;
4044 	info->iidi_oidp = oidp;
4045 	CTX_LOCK(ctx);
4046 	err = IFDI_SYSCTL_INT_DELAY(ctx, info);
4047 	CTX_UNLOCK(ctx);
4048 	return (err);
4049 }
4050 
4051 /*********************************************************************
4052  *
4053  *  IFNET FUNCTIONS
4054  *
4055  **********************************************************************/
4056 
4057 static void
4058 iflib_if_init_locked(if_ctx_t ctx)
4059 {
4060 	iflib_stop(ctx);
4061 	iflib_init_locked(ctx);
4062 }
4063 
4064 static void
4065 iflib_if_init(void *arg)
4066 {
4067 	if_ctx_t ctx = arg;
4068 
4069 	CTX_LOCK(ctx);
4070 	iflib_if_init_locked(ctx);
4071 	CTX_UNLOCK(ctx);
4072 }
4073 
4074 static int
4075 iflib_if_transmit(if_t ifp, struct mbuf *m)
4076 {
4077 	if_ctx_t	ctx = if_getsoftc(ifp);
4078 
4079 	iflib_txq_t txq;
4080 	int err, qidx;
4081 	int abdicate = ctx->ifc_sysctl_tx_abdicate;
4082 
4083 	if (__predict_false((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 || !LINK_ACTIVE(ctx))) {
4084 		DBG_COUNTER_INC(tx_frees);
4085 		m_freem(m);
4086 		return (ENETDOWN);
4087 	}
4088 
4089 	MPASS(m->m_nextpkt == NULL);
4090 	/* ALTQ-enabled interfaces always use queue 0. */
4091 	qidx = 0;
4092 	if ((NTXQSETS(ctx) > 1) && M_HASHTYPE_GET(m) && !ALTQ_IS_ENABLED(&ifp->if_snd))
4093 		qidx = QIDX(ctx, m);
4094 	/*
4095 	 * XXX calculate buf_ring based on flowid (divvy up bits?)
4096 	 */
4097 	txq = &ctx->ifc_txqs[qidx];
4098 
4099 #ifdef DRIVER_BACKPRESSURE
4100 	if (txq->ift_closed) {
4101 		while (m != NULL) {
4102 			next = m->m_nextpkt;
4103 			m->m_nextpkt = NULL;
4104 			m_freem(m);
4105 			DBG_COUNTER_INC(tx_frees);
4106 			m = next;
4107 		}
4108 		return (ENOBUFS);
4109 	}
4110 #endif
4111 #ifdef notyet
4112 	qidx = count = 0;
4113 	mp = marr;
4114 	next = m;
4115 	do {
4116 		count++;
4117 		next = next->m_nextpkt;
4118 	} while (next != NULL);
4119 
4120 	if (count > nitems(marr))
4121 		if ((mp = malloc(count*sizeof(struct mbuf *), M_IFLIB, M_NOWAIT)) == NULL) {
4122 			/* XXX check nextpkt */
4123 			m_freem(m);
4124 			/* XXX simplify for now */
4125 			DBG_COUNTER_INC(tx_frees);
4126 			return (ENOBUFS);
4127 		}
4128 	for (next = m, i = 0; next != NULL; i++) {
4129 		mp[i] = next;
4130 		next = next->m_nextpkt;
4131 		mp[i]->m_nextpkt = NULL;
4132 	}
4133 #endif
4134 	DBG_COUNTER_INC(tx_seen);
4135 	err = ifmp_ring_enqueue(txq->ift_br, (void **)&m, 1, TX_BATCH_SIZE, abdicate);
4136 
4137 	if (abdicate)
4138 		GROUPTASK_ENQUEUE(&txq->ift_task);
4139  	if (err) {
4140 		if (!abdicate)
4141 			GROUPTASK_ENQUEUE(&txq->ift_task);
4142 		/* support forthcoming later */
4143 #ifdef DRIVER_BACKPRESSURE
4144 		txq->ift_closed = TRUE;
4145 #endif
4146 		ifmp_ring_check_drainage(txq->ift_br, TX_BATCH_SIZE);
4147 		m_freem(m);
4148 		DBG_COUNTER_INC(tx_frees);
4149 	}
4150 
4151 	return (err);
4152 }
4153 
4154 #ifdef ALTQ
4155 /*
4156  * The overall approach to integrating iflib with ALTQ is to continue to use
4157  * the iflib mp_ring machinery between the ALTQ queue(s) and the hardware
4158  * ring.  Technically, when using ALTQ, queueing to an intermediate mp_ring
4159  * is redundant/unnecessary, but doing so minimizes the amount of
4160  * ALTQ-specific code required in iflib.  It is assumed that the overhead of
4161  * redundantly queueing to an intermediate mp_ring is swamped by the
4162  * performance limitations inherent in using ALTQ.
4163  *
4164  * When ALTQ support is compiled in, all iflib drivers will use a transmit
4165  * routine, iflib_altq_if_transmit(), that checks if ALTQ is enabled for the
4166  * given interface.  If ALTQ is enabled for an interface, then all
4167  * transmitted packets for that interface will be submitted to the ALTQ
4168  * subsystem via IFQ_ENQUEUE().  We don't use the legacy if_transmit()
4169  * implementation because it uses IFQ_HANDOFF(), which will duplicatively
4170  * update stats that the iflib machinery handles, and which is sensitve to
4171  * the disused IFF_DRV_OACTIVE flag.  Additionally, iflib_altq_if_start()
4172  * will be installed as the start routine for use by ALTQ facilities that
4173  * need to trigger queue drains on a scheduled basis.
4174  *
4175  */
4176 static void
4177 iflib_altq_if_start(if_t ifp)
4178 {
4179 	struct ifaltq *ifq = &ifp->if_snd;
4180 	struct mbuf *m;
4181 
4182 	IFQ_LOCK(ifq);
4183 	IFQ_DEQUEUE_NOLOCK(ifq, m);
4184 	while (m != NULL) {
4185 		iflib_if_transmit(ifp, m);
4186 		IFQ_DEQUEUE_NOLOCK(ifq, m);
4187 	}
4188 	IFQ_UNLOCK(ifq);
4189 }
4190 
4191 static int
4192 iflib_altq_if_transmit(if_t ifp, struct mbuf *m)
4193 {
4194 	int err;
4195 
4196 	if (ALTQ_IS_ENABLED(&ifp->if_snd)) {
4197 		IFQ_ENQUEUE(&ifp->if_snd, m, err);
4198 		if (err == 0)
4199 			iflib_altq_if_start(ifp);
4200 	} else
4201 		err = iflib_if_transmit(ifp, m);
4202 
4203 	return (err);
4204 }
4205 #endif /* ALTQ */
4206 
4207 static void
4208 iflib_if_qflush(if_t ifp)
4209 {
4210 	if_ctx_t ctx = if_getsoftc(ifp);
4211 	iflib_txq_t txq = ctx->ifc_txqs;
4212 	int i;
4213 
4214 	STATE_LOCK(ctx);
4215 	ctx->ifc_flags |= IFC_QFLUSH;
4216 	STATE_UNLOCK(ctx);
4217 	for (i = 0; i < NTXQSETS(ctx); i++, txq++)
4218 		while (!(ifmp_ring_is_idle(txq->ift_br) || ifmp_ring_is_stalled(txq->ift_br)))
4219 			iflib_txq_check_drain(txq, 0);
4220 	STATE_LOCK(ctx);
4221 	ctx->ifc_flags &= ~IFC_QFLUSH;
4222 	STATE_UNLOCK(ctx);
4223 
4224 	/*
4225 	 * When ALTQ is enabled, this will also take care of purging the
4226 	 * ALTQ queue(s).
4227 	 */
4228 	if_qflush(ifp);
4229 }
4230 
4231 #define IFCAP_FLAGS (IFCAP_HWCSUM_IPV6 | IFCAP_HWCSUM | IFCAP_LRO | \
4232 		     IFCAP_TSO | IFCAP_VLAN_HWTAGGING | IFCAP_HWSTATS | \
4233 		     IFCAP_VLAN_MTU | IFCAP_VLAN_HWFILTER | \
4234 		     IFCAP_VLAN_HWTSO | IFCAP_VLAN_HWCSUM | IFCAP_MEXTPG)
4235 
4236 static int
4237 iflib_if_ioctl(if_t ifp, u_long command, caddr_t data)
4238 {
4239 	if_ctx_t ctx = if_getsoftc(ifp);
4240 	struct ifreq	*ifr = (struct ifreq *)data;
4241 #if defined(INET) || defined(INET6)
4242 	struct ifaddr	*ifa = (struct ifaddr *)data;
4243 #endif
4244 	bool		avoid_reset = false;
4245 	int		err = 0, reinit = 0, bits;
4246 
4247 	switch (command) {
4248 	case SIOCSIFADDR:
4249 #ifdef INET
4250 		if (ifa->ifa_addr->sa_family == AF_INET)
4251 			avoid_reset = true;
4252 #endif
4253 #ifdef INET6
4254 		if (ifa->ifa_addr->sa_family == AF_INET6)
4255 			avoid_reset = true;
4256 #endif
4257 		/*
4258 		** Calling init results in link renegotiation,
4259 		** so we avoid doing it when possible.
4260 		*/
4261 		if (avoid_reset) {
4262 			if_setflagbits(ifp, IFF_UP,0);
4263 			if (!(if_getdrvflags(ifp) & IFF_DRV_RUNNING))
4264 				reinit = 1;
4265 #ifdef INET
4266 			if (!(if_getflags(ifp) & IFF_NOARP))
4267 				arp_ifinit(ifp, ifa);
4268 #endif
4269 		} else
4270 			err = ether_ioctl(ifp, command, data);
4271 		break;
4272 	case SIOCSIFMTU:
4273 		CTX_LOCK(ctx);
4274 		if (ifr->ifr_mtu == if_getmtu(ifp)) {
4275 			CTX_UNLOCK(ctx);
4276 			break;
4277 		}
4278 		bits = if_getdrvflags(ifp);
4279 		/* stop the driver and free any clusters before proceeding */
4280 		iflib_stop(ctx);
4281 
4282 		if ((err = IFDI_MTU_SET(ctx, ifr->ifr_mtu)) == 0) {
4283 			STATE_LOCK(ctx);
4284 			if (ifr->ifr_mtu > ctx->ifc_max_fl_buf_size)
4285 				ctx->ifc_flags |= IFC_MULTISEG;
4286 			else
4287 				ctx->ifc_flags &= ~IFC_MULTISEG;
4288 			STATE_UNLOCK(ctx);
4289 			err = if_setmtu(ifp, ifr->ifr_mtu);
4290 		}
4291 		iflib_init_locked(ctx);
4292 		STATE_LOCK(ctx);
4293 		if_setdrvflags(ifp, bits);
4294 		STATE_UNLOCK(ctx);
4295 		CTX_UNLOCK(ctx);
4296 		break;
4297 	case SIOCSIFFLAGS:
4298 		CTX_LOCK(ctx);
4299 		if (if_getflags(ifp) & IFF_UP) {
4300 			if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) {
4301 				if ((if_getflags(ifp) ^ ctx->ifc_if_flags) &
4302 				    (IFF_PROMISC | IFF_ALLMULTI)) {
4303 					CTX_UNLOCK(ctx);
4304 					err = IFDI_PROMISC_SET(ctx, if_getflags(ifp));
4305 					CTX_LOCK(ctx);
4306 				}
4307 			} else
4308 				reinit = 1;
4309 		} else if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) {
4310 			iflib_stop(ctx);
4311 		}
4312 		ctx->ifc_if_flags = if_getflags(ifp);
4313 		CTX_UNLOCK(ctx);
4314 		break;
4315 	case SIOCADDMULTI:
4316 	case SIOCDELMULTI:
4317 		if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) {
4318 			CTX_LOCK(ctx);
4319 			IFDI_INTR_DISABLE(ctx);
4320 			IFDI_MULTI_SET(ctx);
4321 			IFDI_INTR_ENABLE(ctx);
4322 			CTX_UNLOCK(ctx);
4323 		}
4324 		break;
4325 	case SIOCSIFMEDIA:
4326 		CTX_LOCK(ctx);
4327 		IFDI_MEDIA_SET(ctx);
4328 		CTX_UNLOCK(ctx);
4329 		/* FALLTHROUGH */
4330 	case SIOCGIFMEDIA:
4331 	case SIOCGIFXMEDIA:
4332 		err = ifmedia_ioctl(ifp, ifr, ctx->ifc_mediap, command);
4333 		break;
4334 	case SIOCGI2C:
4335 	{
4336 		struct ifi2creq i2c;
4337 
4338 		err = copyin(ifr_data_get_ptr(ifr), &i2c, sizeof(i2c));
4339 		if (err != 0)
4340 			break;
4341 		if (i2c.dev_addr != 0xA0 && i2c.dev_addr != 0xA2) {
4342 			err = EINVAL;
4343 			break;
4344 		}
4345 		if (i2c.len > sizeof(i2c.data)) {
4346 			err = EINVAL;
4347 			break;
4348 		}
4349 
4350 		if ((err = IFDI_I2C_REQ(ctx, &i2c)) == 0)
4351 			err = copyout(&i2c, ifr_data_get_ptr(ifr),
4352 			    sizeof(i2c));
4353 		break;
4354 	}
4355 	case SIOCSIFCAP:
4356 	{
4357 		int mask, setmask, oldmask;
4358 
4359 		oldmask = if_getcapenable(ifp);
4360 		mask = ifr->ifr_reqcap ^ oldmask;
4361 		mask &= ctx->ifc_softc_ctx.isc_capabilities | IFCAP_MEXTPG;
4362 		setmask = 0;
4363 #ifdef TCP_OFFLOAD
4364 		setmask |= mask & (IFCAP_TOE4|IFCAP_TOE6);
4365 #endif
4366 		setmask |= (mask & IFCAP_FLAGS);
4367 		setmask |= (mask & IFCAP_WOL);
4368 
4369 		/*
4370 		 * If any RX csum has changed, change all the ones that
4371 		 * are supported by the driver.
4372 		 */
4373 		if (setmask & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6)) {
4374 			setmask |= ctx->ifc_softc_ctx.isc_capabilities &
4375 			    (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6);
4376 		}
4377 
4378 		/*
4379 		 * want to ensure that traffic has stopped before we change any of the flags
4380 		 */
4381 		if (setmask) {
4382 			CTX_LOCK(ctx);
4383 			bits = if_getdrvflags(ifp);
4384 			if (bits & IFF_DRV_RUNNING && setmask & ~IFCAP_WOL)
4385 				iflib_stop(ctx);
4386 			STATE_LOCK(ctx);
4387 			if_togglecapenable(ifp, setmask);
4388 			STATE_UNLOCK(ctx);
4389 			if (bits & IFF_DRV_RUNNING && setmask & ~IFCAP_WOL)
4390 				iflib_init_locked(ctx);
4391 			STATE_LOCK(ctx);
4392 			if_setdrvflags(ifp, bits);
4393 			STATE_UNLOCK(ctx);
4394 			CTX_UNLOCK(ctx);
4395 		}
4396 		if_vlancap(ifp);
4397 		break;
4398 	}
4399 	case SIOCGPRIVATE_0:
4400 	case SIOCSDRVSPEC:
4401 	case SIOCGDRVSPEC:
4402 		CTX_LOCK(ctx);
4403 		err = IFDI_PRIV_IOCTL(ctx, command, data);
4404 		CTX_UNLOCK(ctx);
4405 		break;
4406 	default:
4407 		err = ether_ioctl(ifp, command, data);
4408 		break;
4409 	}
4410 	if (reinit)
4411 		iflib_if_init(ctx);
4412 	return (err);
4413 }
4414 
4415 static uint64_t
4416 iflib_if_get_counter(if_t ifp, ift_counter cnt)
4417 {
4418 	if_ctx_t ctx = if_getsoftc(ifp);
4419 
4420 	return (IFDI_GET_COUNTER(ctx, cnt));
4421 }
4422 
4423 /*********************************************************************
4424  *
4425  *  OTHER FUNCTIONS EXPORTED TO THE STACK
4426  *
4427  **********************************************************************/
4428 
4429 static void
4430 iflib_vlan_register(void *arg, if_t ifp, uint16_t vtag)
4431 {
4432 	if_ctx_t ctx = if_getsoftc(ifp);
4433 
4434 	if ((void *)ctx != arg)
4435 		return;
4436 
4437 	if ((vtag == 0) || (vtag > 4095))
4438 		return;
4439 
4440 	if (iflib_in_detach(ctx))
4441 		return;
4442 
4443 	CTX_LOCK(ctx);
4444 	/* Driver may need all untagged packets to be flushed */
4445 	if (IFDI_NEEDS_RESTART(ctx, IFLIB_RESTART_VLAN_CONFIG))
4446 		iflib_stop(ctx);
4447 	IFDI_VLAN_REGISTER(ctx, vtag);
4448 	/* Re-init to load the changes, if required */
4449 	if (IFDI_NEEDS_RESTART(ctx, IFLIB_RESTART_VLAN_CONFIG))
4450 		iflib_init_locked(ctx);
4451 	CTX_UNLOCK(ctx);
4452 }
4453 
4454 static void
4455 iflib_vlan_unregister(void *arg, if_t ifp, uint16_t vtag)
4456 {
4457 	if_ctx_t ctx = if_getsoftc(ifp);
4458 
4459 	if ((void *)ctx != arg)
4460 		return;
4461 
4462 	if ((vtag == 0) || (vtag > 4095))
4463 		return;
4464 
4465 	CTX_LOCK(ctx);
4466 	/* Driver may need all tagged packets to be flushed */
4467 	if (IFDI_NEEDS_RESTART(ctx, IFLIB_RESTART_VLAN_CONFIG))
4468 		iflib_stop(ctx);
4469 	IFDI_VLAN_UNREGISTER(ctx, vtag);
4470 	/* Re-init to load the changes, if required */
4471 	if (IFDI_NEEDS_RESTART(ctx, IFLIB_RESTART_VLAN_CONFIG))
4472 		iflib_init_locked(ctx);
4473 	CTX_UNLOCK(ctx);
4474 }
4475 
4476 static void
4477 iflib_led_func(void *arg, int onoff)
4478 {
4479 	if_ctx_t ctx = arg;
4480 
4481 	CTX_LOCK(ctx);
4482 	IFDI_LED_FUNC(ctx, onoff);
4483 	CTX_UNLOCK(ctx);
4484 }
4485 
4486 /*********************************************************************
4487  *
4488  *  BUS FUNCTION DEFINITIONS
4489  *
4490  **********************************************************************/
4491 
4492 int
4493 iflib_device_probe(device_t dev)
4494 {
4495 	const pci_vendor_info_t *ent;
4496 	if_shared_ctx_t sctx;
4497 	uint16_t pci_device_id, pci_rev_id, pci_subdevice_id, pci_subvendor_id;
4498 	uint16_t pci_vendor_id;
4499 
4500 	if ((sctx = DEVICE_REGISTER(dev)) == NULL || sctx->isc_magic != IFLIB_MAGIC)
4501 		return (ENOTSUP);
4502 
4503 	pci_vendor_id = pci_get_vendor(dev);
4504 	pci_device_id = pci_get_device(dev);
4505 	pci_subvendor_id = pci_get_subvendor(dev);
4506 	pci_subdevice_id = pci_get_subdevice(dev);
4507 	pci_rev_id = pci_get_revid(dev);
4508 	if (sctx->isc_parse_devinfo != NULL)
4509 		sctx->isc_parse_devinfo(&pci_device_id, &pci_subvendor_id, &pci_subdevice_id, &pci_rev_id);
4510 
4511 	ent = sctx->isc_vendor_info;
4512 	while (ent->pvi_vendor_id != 0) {
4513 		if (pci_vendor_id != ent->pvi_vendor_id) {
4514 			ent++;
4515 			continue;
4516 		}
4517 		if ((pci_device_id == ent->pvi_device_id) &&
4518 		    ((pci_subvendor_id == ent->pvi_subvendor_id) ||
4519 		     (ent->pvi_subvendor_id == 0)) &&
4520 		    ((pci_subdevice_id == ent->pvi_subdevice_id) ||
4521 		     (ent->pvi_subdevice_id == 0)) &&
4522 		    ((pci_rev_id == ent->pvi_rev_id) ||
4523 		     (ent->pvi_rev_id == 0))) {
4524 			device_set_desc_copy(dev, ent->pvi_name);
4525 			/* this needs to be changed to zero if the bus probing code
4526 			 * ever stops re-probing on best match because the sctx
4527 			 * may have its values over written by register calls
4528 			 * in subsequent probes
4529 			 */
4530 			return (BUS_PROBE_DEFAULT);
4531 		}
4532 		ent++;
4533 	}
4534 	return (ENXIO);
4535 }
4536 
4537 int
4538 iflib_device_probe_vendor(device_t dev)
4539 {
4540 	int probe;
4541 
4542 	probe = iflib_device_probe(dev);
4543 	if (probe == BUS_PROBE_DEFAULT)
4544 		return (BUS_PROBE_VENDOR);
4545 	else
4546 		return (probe);
4547 }
4548 
4549 static void
4550 iflib_reset_qvalues(if_ctx_t ctx)
4551 {
4552 	if_softc_ctx_t scctx = &ctx->ifc_softc_ctx;
4553 	if_shared_ctx_t sctx = ctx->ifc_sctx;
4554 	device_t dev = ctx->ifc_dev;
4555 	int i;
4556 
4557 	if (ctx->ifc_sysctl_ntxqs != 0)
4558 		scctx->isc_ntxqsets = ctx->ifc_sysctl_ntxqs;
4559 	if (ctx->ifc_sysctl_nrxqs != 0)
4560 		scctx->isc_nrxqsets = ctx->ifc_sysctl_nrxqs;
4561 
4562 	for (i = 0; i < sctx->isc_ntxqs; i++) {
4563 		if (ctx->ifc_sysctl_ntxds[i] != 0)
4564 			scctx->isc_ntxd[i] = ctx->ifc_sysctl_ntxds[i];
4565 		else
4566 			scctx->isc_ntxd[i] = sctx->isc_ntxd_default[i];
4567 	}
4568 
4569 	for (i = 0; i < sctx->isc_nrxqs; i++) {
4570 		if (ctx->ifc_sysctl_nrxds[i] != 0)
4571 			scctx->isc_nrxd[i] = ctx->ifc_sysctl_nrxds[i];
4572 		else
4573 			scctx->isc_nrxd[i] = sctx->isc_nrxd_default[i];
4574 	}
4575 
4576 	for (i = 0; i < sctx->isc_nrxqs; i++) {
4577 		if (scctx->isc_nrxd[i] < sctx->isc_nrxd_min[i]) {
4578 			device_printf(dev, "nrxd%d: %d less than nrxd_min %d - resetting to min\n",
4579 				      i, scctx->isc_nrxd[i], sctx->isc_nrxd_min[i]);
4580 			scctx->isc_nrxd[i] = sctx->isc_nrxd_min[i];
4581 		}
4582 		if (scctx->isc_nrxd[i] > sctx->isc_nrxd_max[i]) {
4583 			device_printf(dev, "nrxd%d: %d greater than nrxd_max %d - resetting to max\n",
4584 				      i, scctx->isc_nrxd[i], sctx->isc_nrxd_max[i]);
4585 			scctx->isc_nrxd[i] = sctx->isc_nrxd_max[i];
4586 		}
4587 		if (!powerof2(scctx->isc_nrxd[i])) {
4588 			device_printf(dev, "nrxd%d: %d is not a power of 2 - using default value of %d\n",
4589 				      i, scctx->isc_nrxd[i], sctx->isc_nrxd_default[i]);
4590 			scctx->isc_nrxd[i] = sctx->isc_nrxd_default[i];
4591 		}
4592 	}
4593 
4594 	for (i = 0; i < sctx->isc_ntxqs; i++) {
4595 		if (scctx->isc_ntxd[i] < sctx->isc_ntxd_min[i]) {
4596 			device_printf(dev, "ntxd%d: %d less than ntxd_min %d - resetting to min\n",
4597 				      i, scctx->isc_ntxd[i], sctx->isc_ntxd_min[i]);
4598 			scctx->isc_ntxd[i] = sctx->isc_ntxd_min[i];
4599 		}
4600 		if (scctx->isc_ntxd[i] > sctx->isc_ntxd_max[i]) {
4601 			device_printf(dev, "ntxd%d: %d greater than ntxd_max %d - resetting to max\n",
4602 				      i, scctx->isc_ntxd[i], sctx->isc_ntxd_max[i]);
4603 			scctx->isc_ntxd[i] = sctx->isc_ntxd_max[i];
4604 		}
4605 		if (!powerof2(scctx->isc_ntxd[i])) {
4606 			device_printf(dev, "ntxd%d: %d is not a power of 2 - using default value of %d\n",
4607 				      i, scctx->isc_ntxd[i], sctx->isc_ntxd_default[i]);
4608 			scctx->isc_ntxd[i] = sctx->isc_ntxd_default[i];
4609 		}
4610 	}
4611 }
4612 
4613 static void
4614 iflib_add_pfil(if_ctx_t ctx)
4615 {
4616 	struct pfil_head *pfil;
4617 	struct pfil_head_args pa;
4618 	iflib_rxq_t rxq;
4619 	int i;
4620 
4621 	pa.pa_version = PFIL_VERSION;
4622 	pa.pa_flags = PFIL_IN;
4623 	pa.pa_type = PFIL_TYPE_ETHERNET;
4624 	pa.pa_headname = ctx->ifc_ifp->if_xname;
4625 	pfil = pfil_head_register(&pa);
4626 
4627 	for (i = 0, rxq = ctx->ifc_rxqs; i < NRXQSETS(ctx); i++, rxq++) {
4628 		rxq->pfil = pfil;
4629 	}
4630 }
4631 
4632 static void
4633 iflib_rem_pfil(if_ctx_t ctx)
4634 {
4635 	struct pfil_head *pfil;
4636 	iflib_rxq_t rxq;
4637 	int i;
4638 
4639 	rxq = ctx->ifc_rxqs;
4640 	pfil = rxq->pfil;
4641 	for (i = 0; i < NRXQSETS(ctx); i++, rxq++) {
4642 		rxq->pfil = NULL;
4643 	}
4644 	pfil_head_unregister(pfil);
4645 }
4646 
4647 static uint16_t
4648 get_ctx_core_offset(if_ctx_t ctx)
4649 {
4650 	if_softc_ctx_t scctx = &ctx->ifc_softc_ctx;
4651 	struct cpu_offset *op;
4652 	uint16_t qc;
4653 	uint16_t ret = ctx->ifc_sysctl_core_offset;
4654 
4655 	if (ret != CORE_OFFSET_UNSPECIFIED)
4656 		return (ret);
4657 
4658 	if (ctx->ifc_sysctl_separate_txrx)
4659 		qc = scctx->isc_ntxqsets + scctx->isc_nrxqsets;
4660 	else
4661 		qc = max(scctx->isc_ntxqsets, scctx->isc_nrxqsets);
4662 
4663 	mtx_lock(&cpu_offset_mtx);
4664 	SLIST_FOREACH(op, &cpu_offsets, entries) {
4665 		if (CPU_CMP(&ctx->ifc_cpus, &op->set) == 0) {
4666 			ret = op->offset;
4667 			op->offset += qc;
4668 			MPASS(op->refcount < UINT_MAX);
4669 			op->refcount++;
4670 			break;
4671 		}
4672 	}
4673 	if (ret == CORE_OFFSET_UNSPECIFIED) {
4674 		ret = 0;
4675 		op = malloc(sizeof(struct cpu_offset), M_IFLIB,
4676 		    M_NOWAIT | M_ZERO);
4677 		if (op == NULL) {
4678 			device_printf(ctx->ifc_dev,
4679 			    "allocation for cpu offset failed.\n");
4680 		} else {
4681 			op->offset = qc;
4682 			op->refcount = 1;
4683 			CPU_COPY(&ctx->ifc_cpus, &op->set);
4684 			SLIST_INSERT_HEAD(&cpu_offsets, op, entries);
4685 		}
4686 	}
4687 	mtx_unlock(&cpu_offset_mtx);
4688 
4689 	return (ret);
4690 }
4691 
4692 static void
4693 unref_ctx_core_offset(if_ctx_t ctx)
4694 {
4695 	struct cpu_offset *op, *top;
4696 
4697 	mtx_lock(&cpu_offset_mtx);
4698 	SLIST_FOREACH_SAFE(op, &cpu_offsets, entries, top) {
4699 		if (CPU_CMP(&ctx->ifc_cpus, &op->set) == 0) {
4700 			MPASS(op->refcount > 0);
4701 			op->refcount--;
4702 			if (op->refcount == 0) {
4703 				SLIST_REMOVE(&cpu_offsets, op, cpu_offset, entries);
4704 				free(op, M_IFLIB);
4705 			}
4706 			break;
4707 		}
4708 	}
4709 	mtx_unlock(&cpu_offset_mtx);
4710 }
4711 
4712 int
4713 iflib_device_register(device_t dev, void *sc, if_shared_ctx_t sctx, if_ctx_t *ctxp)
4714 {
4715 	if_ctx_t ctx;
4716 	if_t ifp;
4717 	if_softc_ctx_t scctx;
4718 	kobjop_desc_t kobj_desc;
4719 	kobj_method_t *kobj_method;
4720 	int err, msix, rid;
4721 	int num_txd, num_rxd;
4722 
4723 	ctx = malloc(sizeof(* ctx), M_IFLIB, M_WAITOK|M_ZERO);
4724 
4725 	if (sc == NULL) {
4726 		sc = malloc(sctx->isc_driver->size, M_IFLIB, M_WAITOK|M_ZERO);
4727 		device_set_softc(dev, ctx);
4728 		ctx->ifc_flags |= IFC_SC_ALLOCATED;
4729 	}
4730 
4731 	ctx->ifc_sctx = sctx;
4732 	ctx->ifc_dev = dev;
4733 	ctx->ifc_softc = sc;
4734 
4735 	if ((err = iflib_register(ctx)) != 0) {
4736 		device_printf(dev, "iflib_register failed %d\n", err);
4737 		goto fail_ctx_free;
4738 	}
4739 	iflib_add_device_sysctl_pre(ctx);
4740 
4741 	scctx = &ctx->ifc_softc_ctx;
4742 	ifp = ctx->ifc_ifp;
4743 
4744 	iflib_reset_qvalues(ctx);
4745 	CTX_LOCK(ctx);
4746 	if ((err = IFDI_ATTACH_PRE(ctx)) != 0) {
4747 		device_printf(dev, "IFDI_ATTACH_PRE failed %d\n", err);
4748 		goto fail_unlock;
4749 	}
4750 	_iflib_pre_assert(scctx);
4751 	ctx->ifc_txrx = *scctx->isc_txrx;
4752 
4753 	if (sctx->isc_flags & IFLIB_DRIVER_MEDIA)
4754 		ctx->ifc_mediap = scctx->isc_media;
4755 
4756 #ifdef INVARIANTS
4757 	if (scctx->isc_capabilities & IFCAP_TXCSUM)
4758 		MPASS(scctx->isc_tx_csum_flags);
4759 #endif
4760 
4761 	if_setcapabilities(ifp,
4762 	    scctx->isc_capabilities | IFCAP_HWSTATS | IFCAP_MEXTPG);
4763 	if_setcapenable(ifp,
4764 	    scctx->isc_capenable | IFCAP_HWSTATS | IFCAP_MEXTPG);
4765 
4766 	if (scctx->isc_ntxqsets == 0 || (scctx->isc_ntxqsets_max && scctx->isc_ntxqsets_max < scctx->isc_ntxqsets))
4767 		scctx->isc_ntxqsets = scctx->isc_ntxqsets_max;
4768 	if (scctx->isc_nrxqsets == 0 || (scctx->isc_nrxqsets_max && scctx->isc_nrxqsets_max < scctx->isc_nrxqsets))
4769 		scctx->isc_nrxqsets = scctx->isc_nrxqsets_max;
4770 
4771 	num_txd = iflib_num_tx_descs(ctx);
4772 	num_rxd = iflib_num_rx_descs(ctx);
4773 
4774 	/* XXX change for per-queue sizes */
4775 	device_printf(dev, "Using %d TX descriptors and %d RX descriptors\n",
4776 	    num_txd, num_rxd);
4777 
4778 	if (scctx->isc_tx_nsegments > num_txd / MAX_SINGLE_PACKET_FRACTION)
4779 		scctx->isc_tx_nsegments = max(1, num_txd /
4780 		    MAX_SINGLE_PACKET_FRACTION);
4781 	if (scctx->isc_tx_tso_segments_max > num_txd /
4782 	    MAX_SINGLE_PACKET_FRACTION)
4783 		scctx->isc_tx_tso_segments_max = max(1,
4784 		    num_txd / MAX_SINGLE_PACKET_FRACTION);
4785 
4786 	/* TSO parameters - dig these out of the data sheet - simply correspond to tag setup */
4787 	if (if_getcapabilities(ifp) & IFCAP_TSO) {
4788 		/*
4789 		 * The stack can't handle a TSO size larger than IP_MAXPACKET,
4790 		 * but some MACs do.
4791 		 */
4792 		if_sethwtsomax(ifp, min(scctx->isc_tx_tso_size_max,
4793 		    IP_MAXPACKET));
4794 		/*
4795 		 * Take maximum number of m_pullup(9)'s in iflib_parse_header()
4796 		 * into account.  In the worst case, each of these calls will
4797 		 * add another mbuf and, thus, the requirement for another DMA
4798 		 * segment.  So for best performance, it doesn't make sense to
4799 		 * advertize a maximum of TSO segments that typically will
4800 		 * require defragmentation in iflib_encap().
4801 		 */
4802 		if_sethwtsomaxsegcount(ifp, scctx->isc_tx_tso_segments_max - 3);
4803 		if_sethwtsomaxsegsize(ifp, scctx->isc_tx_tso_segsize_max);
4804 	}
4805 	if (scctx->isc_rss_table_size == 0)
4806 		scctx->isc_rss_table_size = 64;
4807 	scctx->isc_rss_table_mask = scctx->isc_rss_table_size-1;
4808 
4809 	GROUPTASK_INIT(&ctx->ifc_admin_task, 0, _task_fn_admin, ctx);
4810 	/* XXX format name */
4811 	taskqgroup_attach(qgroup_if_config_tqg, &ctx->ifc_admin_task, ctx,
4812 	    NULL, NULL, "admin");
4813 
4814 	/* Set up cpu set.  If it fails, use the set of all CPUs. */
4815 	if (bus_get_cpus(dev, INTR_CPUS, sizeof(ctx->ifc_cpus), &ctx->ifc_cpus) != 0) {
4816 		device_printf(dev, "Unable to fetch CPU list\n");
4817 		CPU_COPY(&all_cpus, &ctx->ifc_cpus);
4818 	}
4819 	MPASS(CPU_COUNT(&ctx->ifc_cpus) > 0);
4820 
4821 	/*
4822 	** Now set up MSI or MSI-X, should return us the number of supported
4823 	** vectors (will be 1 for a legacy interrupt and MSI).
4824 	*/
4825 	if (sctx->isc_flags & IFLIB_SKIP_MSIX) {
4826 		msix = scctx->isc_vectors;
4827 	} else if (scctx->isc_msix_bar != 0)
4828 	       /*
4829 		* The simple fact that isc_msix_bar is not 0 does not mean we
4830 		* we have a good value there that is known to work.
4831 		*/
4832 		msix = iflib_msix_init(ctx);
4833 	else {
4834 		scctx->isc_vectors = 1;
4835 		scctx->isc_ntxqsets = 1;
4836 		scctx->isc_nrxqsets = 1;
4837 		scctx->isc_intr = IFLIB_INTR_LEGACY;
4838 		msix = 0;
4839 	}
4840 	/* Get memory for the station queues */
4841 	if ((err = iflib_queues_alloc(ctx))) {
4842 		device_printf(dev, "Unable to allocate queue memory\n");
4843 		goto fail_intr_free;
4844 	}
4845 
4846 	if ((err = iflib_qset_structures_setup(ctx)))
4847 		goto fail_queues;
4848 
4849 	/*
4850 	 * Now that we know how many queues there are, get the core offset.
4851 	 */
4852 	ctx->ifc_sysctl_core_offset = get_ctx_core_offset(ctx);
4853 
4854 	if (msix > 1) {
4855 		/*
4856 		 * When using MSI-X, ensure that ifdi_{r,t}x_queue_intr_enable
4857 		 * aren't the default NULL implementation.
4858 		 */
4859 		kobj_desc = &ifdi_rx_queue_intr_enable_desc;
4860 		kobj_method = kobj_lookup_method(((kobj_t)ctx)->ops->cls, NULL,
4861 		    kobj_desc);
4862 		if (kobj_method == &kobj_desc->deflt) {
4863 			device_printf(dev,
4864 			    "MSI-X requires ifdi_rx_queue_intr_enable method");
4865 			err = EOPNOTSUPP;
4866 			goto fail_queues;
4867 		}
4868 		kobj_desc = &ifdi_tx_queue_intr_enable_desc;
4869 		kobj_method = kobj_lookup_method(((kobj_t)ctx)->ops->cls, NULL,
4870 		    kobj_desc);
4871 		if (kobj_method == &kobj_desc->deflt) {
4872 			device_printf(dev,
4873 			    "MSI-X requires ifdi_tx_queue_intr_enable method");
4874 			err = EOPNOTSUPP;
4875 			goto fail_queues;
4876 		}
4877 
4878 		/*
4879 		 * Assign the MSI-X vectors.
4880 		 * Note that the default NULL ifdi_msix_intr_assign method will
4881 		 * fail here, too.
4882 		 */
4883 		err = IFDI_MSIX_INTR_ASSIGN(ctx, msix);
4884 		if (err != 0) {
4885 			device_printf(dev, "IFDI_MSIX_INTR_ASSIGN failed %d\n",
4886 			    err);
4887 			goto fail_queues;
4888 		}
4889 	} else if (scctx->isc_intr != IFLIB_INTR_MSIX) {
4890 		rid = 0;
4891 		if (scctx->isc_intr == IFLIB_INTR_MSI) {
4892 			MPASS(msix == 1);
4893 			rid = 1;
4894 		}
4895 		if ((err = iflib_legacy_setup(ctx, ctx->isc_legacy_intr, ctx->ifc_softc, &rid, "irq0")) != 0) {
4896 			device_printf(dev, "iflib_legacy_setup failed %d\n", err);
4897 			goto fail_queues;
4898 		}
4899 	} else {
4900 		device_printf(dev,
4901 		    "Cannot use iflib with only 1 MSI-X interrupt!\n");
4902 		err = ENODEV;
4903 		goto fail_queues;
4904 	}
4905 
4906 	ether_ifattach(ctx->ifc_ifp, ctx->ifc_mac.octet);
4907 
4908 	if ((err = IFDI_ATTACH_POST(ctx)) != 0) {
4909 		device_printf(dev, "IFDI_ATTACH_POST failed %d\n", err);
4910 		goto fail_detach;
4911 	}
4912 
4913 	/*
4914 	 * Tell the upper layer(s) if IFCAP_VLAN_MTU is supported.
4915 	 * This must appear after the call to ether_ifattach() because
4916 	 * ether_ifattach() sets if_hdrlen to the default value.
4917 	 */
4918 	if (if_getcapabilities(ifp) & IFCAP_VLAN_MTU)
4919 		if_setifheaderlen(ifp, sizeof(struct ether_vlan_header));
4920 
4921 	if ((err = iflib_netmap_attach(ctx))) {
4922 		device_printf(ctx->ifc_dev, "netmap attach failed: %d\n", err);
4923 		goto fail_detach;
4924 	}
4925 	*ctxp = ctx;
4926 
4927 	DEBUGNET_SET(ctx->ifc_ifp, iflib);
4928 
4929 	if_setgetcounterfn(ctx->ifc_ifp, iflib_if_get_counter);
4930 	iflib_add_device_sysctl_post(ctx);
4931 	iflib_add_pfil(ctx);
4932 	ctx->ifc_flags |= IFC_INIT_DONE;
4933 	CTX_UNLOCK(ctx);
4934 
4935 	return (0);
4936 
4937 fail_detach:
4938 	ether_ifdetach(ctx->ifc_ifp);
4939 fail_queues:
4940 	iflib_tqg_detach(ctx);
4941 	iflib_tx_structures_free(ctx);
4942 	iflib_rx_structures_free(ctx);
4943 	IFDI_DETACH(ctx);
4944 	IFDI_QUEUES_FREE(ctx);
4945 fail_intr_free:
4946 	iflib_free_intr_mem(ctx);
4947 fail_unlock:
4948 	CTX_UNLOCK(ctx);
4949 	iflib_deregister(ctx);
4950 fail_ctx_free:
4951 	device_set_softc(ctx->ifc_dev, NULL);
4952         if (ctx->ifc_flags & IFC_SC_ALLOCATED)
4953                 free(ctx->ifc_softc, M_IFLIB);
4954         free(ctx, M_IFLIB);
4955 	return (err);
4956 }
4957 
4958 int
4959 iflib_pseudo_register(device_t dev, if_shared_ctx_t sctx, if_ctx_t *ctxp,
4960 					  struct iflib_cloneattach_ctx *clctx)
4961 {
4962 	int num_txd, num_rxd;
4963 	int err;
4964 	if_ctx_t ctx;
4965 	if_t ifp;
4966 	if_softc_ctx_t scctx;
4967 	int i;
4968 	void *sc;
4969 
4970 	ctx = malloc(sizeof(*ctx), M_IFLIB, M_WAITOK|M_ZERO);
4971 	sc = malloc(sctx->isc_driver->size, M_IFLIB, M_WAITOK|M_ZERO);
4972 	ctx->ifc_flags |= IFC_SC_ALLOCATED;
4973 	if (sctx->isc_flags & (IFLIB_PSEUDO|IFLIB_VIRTUAL))
4974 		ctx->ifc_flags |= IFC_PSEUDO;
4975 
4976 	ctx->ifc_sctx = sctx;
4977 	ctx->ifc_softc = sc;
4978 	ctx->ifc_dev = dev;
4979 
4980 	if ((err = iflib_register(ctx)) != 0) {
4981 		device_printf(dev, "%s: iflib_register failed %d\n", __func__, err);
4982 		goto fail_ctx_free;
4983 	}
4984 	iflib_add_device_sysctl_pre(ctx);
4985 
4986 	scctx = &ctx->ifc_softc_ctx;
4987 	ifp = ctx->ifc_ifp;
4988 
4989 	iflib_reset_qvalues(ctx);
4990 	CTX_LOCK(ctx);
4991 	if ((err = IFDI_ATTACH_PRE(ctx)) != 0) {
4992 		device_printf(dev, "IFDI_ATTACH_PRE failed %d\n", err);
4993 		goto fail_unlock;
4994 	}
4995 	if (sctx->isc_flags & IFLIB_GEN_MAC)
4996 		ether_gen_addr(ifp, &ctx->ifc_mac);
4997 	if ((err = IFDI_CLONEATTACH(ctx, clctx->cc_ifc, clctx->cc_name,
4998 								clctx->cc_params)) != 0) {
4999 		device_printf(dev, "IFDI_CLONEATTACH failed %d\n", err);
5000 		goto fail_unlock;
5001 	}
5002 #ifdef INVARIANTS
5003 	if (scctx->isc_capabilities & IFCAP_TXCSUM)
5004 		MPASS(scctx->isc_tx_csum_flags);
5005 #endif
5006 
5007 	if_setcapabilities(ifp, scctx->isc_capabilities | IFCAP_HWSTATS | IFCAP_LINKSTATE);
5008 	if_setcapenable(ifp, scctx->isc_capenable | IFCAP_HWSTATS | IFCAP_LINKSTATE);
5009 
5010 	ifp->if_flags |= IFF_NOGROUP;
5011 	if (sctx->isc_flags & IFLIB_PSEUDO) {
5012 		ifmedia_add(ctx->ifc_mediap, IFM_ETHER | IFM_AUTO, 0, NULL);
5013 		ifmedia_set(ctx->ifc_mediap, IFM_ETHER | IFM_AUTO);
5014 		if (sctx->isc_flags & IFLIB_PSEUDO_ETHER) {
5015 			ether_ifattach(ctx->ifc_ifp, ctx->ifc_mac.octet);
5016 		} else {
5017 			if_attach(ctx->ifc_ifp);
5018 			bpfattach(ctx->ifc_ifp, DLT_NULL, sizeof(u_int32_t));
5019 		}
5020 
5021 		if ((err = IFDI_ATTACH_POST(ctx)) != 0) {
5022 			device_printf(dev, "IFDI_ATTACH_POST failed %d\n", err);
5023 			goto fail_detach;
5024 		}
5025 		*ctxp = ctx;
5026 
5027 		/*
5028 		 * Tell the upper layer(s) if IFCAP_VLAN_MTU is supported.
5029 		 * This must appear after the call to ether_ifattach() because
5030 		 * ether_ifattach() sets if_hdrlen to the default value.
5031 		 */
5032 		if (if_getcapabilities(ifp) & IFCAP_VLAN_MTU)
5033 			if_setifheaderlen(ifp,
5034 			    sizeof(struct ether_vlan_header));
5035 
5036 		if_setgetcounterfn(ctx->ifc_ifp, iflib_if_get_counter);
5037 		iflib_add_device_sysctl_post(ctx);
5038 		ctx->ifc_flags |= IFC_INIT_DONE;
5039 		CTX_UNLOCK(ctx);
5040 		return (0);
5041 	}
5042 	ifmedia_add(ctx->ifc_mediap, IFM_ETHER | IFM_1000_T | IFM_FDX, 0, NULL);
5043 	ifmedia_add(ctx->ifc_mediap, IFM_ETHER | IFM_AUTO, 0, NULL);
5044 	ifmedia_set(ctx->ifc_mediap, IFM_ETHER | IFM_AUTO);
5045 
5046 	_iflib_pre_assert(scctx);
5047 	ctx->ifc_txrx = *scctx->isc_txrx;
5048 
5049 	if (scctx->isc_ntxqsets == 0 || (scctx->isc_ntxqsets_max && scctx->isc_ntxqsets_max < scctx->isc_ntxqsets))
5050 		scctx->isc_ntxqsets = scctx->isc_ntxqsets_max;
5051 	if (scctx->isc_nrxqsets == 0 || (scctx->isc_nrxqsets_max && scctx->isc_nrxqsets_max < scctx->isc_nrxqsets))
5052 		scctx->isc_nrxqsets = scctx->isc_nrxqsets_max;
5053 
5054 	num_txd = iflib_num_tx_descs(ctx);
5055 	num_rxd = iflib_num_rx_descs(ctx);
5056 
5057 	/* XXX change for per-queue sizes */
5058 	device_printf(dev, "Using %d TX descriptors and %d RX descriptors\n",
5059 	    num_txd, num_rxd);
5060 
5061 	if (scctx->isc_tx_nsegments > num_txd / MAX_SINGLE_PACKET_FRACTION)
5062 		scctx->isc_tx_nsegments = max(1, num_txd /
5063 		    MAX_SINGLE_PACKET_FRACTION);
5064 	if (scctx->isc_tx_tso_segments_max > num_txd /
5065 	    MAX_SINGLE_PACKET_FRACTION)
5066 		scctx->isc_tx_tso_segments_max = max(1,
5067 		    num_txd / MAX_SINGLE_PACKET_FRACTION);
5068 
5069 	/* TSO parameters - dig these out of the data sheet - simply correspond to tag setup */
5070 	if (if_getcapabilities(ifp) & IFCAP_TSO) {
5071 		/*
5072 		 * The stack can't handle a TSO size larger than IP_MAXPACKET,
5073 		 * but some MACs do.
5074 		 */
5075 		if_sethwtsomax(ifp, min(scctx->isc_tx_tso_size_max,
5076 		    IP_MAXPACKET));
5077 		/*
5078 		 * Take maximum number of m_pullup(9)'s in iflib_parse_header()
5079 		 * into account.  In the worst case, each of these calls will
5080 		 * add another mbuf and, thus, the requirement for another DMA
5081 		 * segment.  So for best performance, it doesn't make sense to
5082 		 * advertize a maximum of TSO segments that typically will
5083 		 * require defragmentation in iflib_encap().
5084 		 */
5085 		if_sethwtsomaxsegcount(ifp, scctx->isc_tx_tso_segments_max - 3);
5086 		if_sethwtsomaxsegsize(ifp, scctx->isc_tx_tso_segsize_max);
5087 	}
5088 	if (scctx->isc_rss_table_size == 0)
5089 		scctx->isc_rss_table_size = 64;
5090 	scctx->isc_rss_table_mask = scctx->isc_rss_table_size-1;
5091 
5092 	GROUPTASK_INIT(&ctx->ifc_admin_task, 0, _task_fn_admin, ctx);
5093 	/* XXX format name */
5094 	taskqgroup_attach(qgroup_if_config_tqg, &ctx->ifc_admin_task, ctx,
5095 	    NULL, NULL, "admin");
5096 
5097 	/* XXX --- can support > 1 -- but keep it simple for now */
5098 	scctx->isc_intr = IFLIB_INTR_LEGACY;
5099 
5100 	/* Get memory for the station queues */
5101 	if ((err = iflib_queues_alloc(ctx))) {
5102 		device_printf(dev, "Unable to allocate queue memory\n");
5103 		goto fail_iflib_detach;
5104 	}
5105 
5106 	if ((err = iflib_qset_structures_setup(ctx))) {
5107 		device_printf(dev, "qset structure setup failed %d\n", err);
5108 		goto fail_queues;
5109 	}
5110 
5111 	/*
5112 	 * XXX What if anything do we want to do about interrupts?
5113 	 */
5114 	ether_ifattach(ctx->ifc_ifp, ctx->ifc_mac.octet);
5115 	if ((err = IFDI_ATTACH_POST(ctx)) != 0) {
5116 		device_printf(dev, "IFDI_ATTACH_POST failed %d\n", err);
5117 		goto fail_detach;
5118 	}
5119 
5120 	/*
5121 	 * Tell the upper layer(s) if IFCAP_VLAN_MTU is supported.
5122 	 * This must appear after the call to ether_ifattach() because
5123 	 * ether_ifattach() sets if_hdrlen to the default value.
5124 	 */
5125 	if (if_getcapabilities(ifp) & IFCAP_VLAN_MTU)
5126 		if_setifheaderlen(ifp, sizeof(struct ether_vlan_header));
5127 
5128 	/* XXX handle more than one queue */
5129 	for (i = 0; i < scctx->isc_nrxqsets; i++)
5130 		IFDI_RX_CLSET(ctx, 0, i, ctx->ifc_rxqs[i].ifr_fl[0].ifl_sds.ifsd_cl);
5131 
5132 	*ctxp = ctx;
5133 
5134 	if_setgetcounterfn(ctx->ifc_ifp, iflib_if_get_counter);
5135 	iflib_add_device_sysctl_post(ctx);
5136 	ctx->ifc_flags |= IFC_INIT_DONE;
5137 	CTX_UNLOCK(ctx);
5138 
5139 	return (0);
5140 fail_detach:
5141 	ether_ifdetach(ctx->ifc_ifp);
5142 fail_queues:
5143 	iflib_tqg_detach(ctx);
5144 	iflib_tx_structures_free(ctx);
5145 	iflib_rx_structures_free(ctx);
5146 fail_iflib_detach:
5147 	IFDI_DETACH(ctx);
5148 	IFDI_QUEUES_FREE(ctx);
5149 fail_unlock:
5150 	CTX_UNLOCK(ctx);
5151 	iflib_deregister(ctx);
5152 fail_ctx_free:
5153 	free(ctx->ifc_softc, M_IFLIB);
5154 	free(ctx, M_IFLIB);
5155 	return (err);
5156 }
5157 
5158 int
5159 iflib_pseudo_deregister(if_ctx_t ctx)
5160 {
5161 	if_t ifp = ctx->ifc_ifp;
5162 	if_shared_ctx_t sctx = ctx->ifc_sctx;
5163 
5164 	/* Unregister VLAN event handlers early */
5165 	iflib_unregister_vlan_handlers(ctx);
5166 
5167 	if ((sctx->isc_flags & IFLIB_PSEUDO)  &&
5168 		(sctx->isc_flags & IFLIB_PSEUDO_ETHER) == 0) {
5169 		bpfdetach(ifp);
5170 		if_detach(ifp);
5171 	} else {
5172 		ether_ifdetach(ifp);
5173 	}
5174 
5175 	iflib_tqg_detach(ctx);
5176 	iflib_tx_structures_free(ctx);
5177 	iflib_rx_structures_free(ctx);
5178 	IFDI_DETACH(ctx);
5179 	IFDI_QUEUES_FREE(ctx);
5180 
5181 	iflib_deregister(ctx);
5182 
5183 	if (ctx->ifc_flags & IFC_SC_ALLOCATED)
5184 		free(ctx->ifc_softc, M_IFLIB);
5185 	free(ctx, M_IFLIB);
5186 	return (0);
5187 }
5188 
5189 int
5190 iflib_device_attach(device_t dev)
5191 {
5192 	if_ctx_t ctx;
5193 	if_shared_ctx_t sctx;
5194 
5195 	if ((sctx = DEVICE_REGISTER(dev)) == NULL || sctx->isc_magic != IFLIB_MAGIC)
5196 		return (ENOTSUP);
5197 
5198 	pci_enable_busmaster(dev);
5199 
5200 	return (iflib_device_register(dev, NULL, sctx, &ctx));
5201 }
5202 
5203 int
5204 iflib_device_deregister(if_ctx_t ctx)
5205 {
5206 	if_t ifp = ctx->ifc_ifp;
5207 	device_t dev = ctx->ifc_dev;
5208 
5209 	/* Make sure VLANS are not using driver */
5210 	if (if_vlantrunkinuse(ifp)) {
5211 		device_printf(dev, "Vlan in use, detach first\n");
5212 		return (EBUSY);
5213 	}
5214 #ifdef PCI_IOV
5215 	if (!CTX_IS_VF(ctx) && pci_iov_detach(dev) != 0) {
5216 		device_printf(dev, "SR-IOV in use; detach first.\n");
5217 		return (EBUSY);
5218 	}
5219 #endif
5220 
5221 	STATE_LOCK(ctx);
5222 	ctx->ifc_flags |= IFC_IN_DETACH;
5223 	STATE_UNLOCK(ctx);
5224 
5225 	/* Unregister VLAN handlers before calling iflib_stop() */
5226 	iflib_unregister_vlan_handlers(ctx);
5227 
5228 	iflib_netmap_detach(ifp);
5229 	ether_ifdetach(ifp);
5230 
5231 	CTX_LOCK(ctx);
5232 	iflib_stop(ctx);
5233 	CTX_UNLOCK(ctx);
5234 
5235 	iflib_rem_pfil(ctx);
5236 	if (ctx->ifc_led_dev != NULL)
5237 		led_destroy(ctx->ifc_led_dev);
5238 
5239 	iflib_tqg_detach(ctx);
5240 	iflib_tx_structures_free(ctx);
5241 	iflib_rx_structures_free(ctx);
5242 
5243 	CTX_LOCK(ctx);
5244 	IFDI_DETACH(ctx);
5245 	IFDI_QUEUES_FREE(ctx);
5246 	CTX_UNLOCK(ctx);
5247 
5248 	/* ether_ifdetach calls if_qflush - lock must be destroy afterwards*/
5249 	iflib_free_intr_mem(ctx);
5250 
5251 	bus_generic_detach(dev);
5252 
5253 	iflib_deregister(ctx);
5254 
5255 	device_set_softc(ctx->ifc_dev, NULL);
5256 	if (ctx->ifc_flags & IFC_SC_ALLOCATED)
5257 		free(ctx->ifc_softc, M_IFLIB);
5258 	unref_ctx_core_offset(ctx);
5259 	free(ctx, M_IFLIB);
5260 	return (0);
5261 }
5262 
5263 static void
5264 iflib_tqg_detach(if_ctx_t ctx)
5265 {
5266 	iflib_txq_t txq;
5267 	iflib_rxq_t rxq;
5268 	int i;
5269 	struct taskqgroup *tqg;
5270 
5271 	/* XXX drain any dependent tasks */
5272 	tqg = qgroup_if_io_tqg;
5273 	for (txq = ctx->ifc_txqs, i = 0; i < NTXQSETS(ctx); i++, txq++) {
5274 		callout_drain(&txq->ift_timer);
5275 #ifdef DEV_NETMAP
5276 		callout_drain(&txq->ift_netmap_timer);
5277 #endif /* DEV_NETMAP */
5278 		if (txq->ift_task.gt_uniq != NULL)
5279 			taskqgroup_detach(tqg, &txq->ift_task);
5280 	}
5281 	for (i = 0, rxq = ctx->ifc_rxqs; i < NRXQSETS(ctx); i++, rxq++) {
5282 		if (rxq->ifr_task.gt_uniq != NULL)
5283 			taskqgroup_detach(tqg, &rxq->ifr_task);
5284 	}
5285 	tqg = qgroup_if_config_tqg;
5286 	if (ctx->ifc_admin_task.gt_uniq != NULL)
5287 		taskqgroup_detach(tqg, &ctx->ifc_admin_task);
5288 	if (ctx->ifc_vflr_task.gt_uniq != NULL)
5289 		taskqgroup_detach(tqg, &ctx->ifc_vflr_task);
5290 }
5291 
5292 static void
5293 iflib_free_intr_mem(if_ctx_t ctx)
5294 {
5295 
5296 	if (ctx->ifc_softc_ctx.isc_intr != IFLIB_INTR_MSIX) {
5297 		iflib_irq_free(ctx, &ctx->ifc_legacy_irq);
5298 	}
5299 	if (ctx->ifc_softc_ctx.isc_intr != IFLIB_INTR_LEGACY) {
5300 		pci_release_msi(ctx->ifc_dev);
5301 	}
5302 	if (ctx->ifc_msix_mem != NULL) {
5303 		bus_release_resource(ctx->ifc_dev, SYS_RES_MEMORY,
5304 		    rman_get_rid(ctx->ifc_msix_mem), ctx->ifc_msix_mem);
5305 		ctx->ifc_msix_mem = NULL;
5306 	}
5307 }
5308 
5309 int
5310 iflib_device_detach(device_t dev)
5311 {
5312 	if_ctx_t ctx = device_get_softc(dev);
5313 
5314 	return (iflib_device_deregister(ctx));
5315 }
5316 
5317 int
5318 iflib_device_suspend(device_t dev)
5319 {
5320 	if_ctx_t ctx = device_get_softc(dev);
5321 
5322 	CTX_LOCK(ctx);
5323 	IFDI_SUSPEND(ctx);
5324 	CTX_UNLOCK(ctx);
5325 
5326 	return bus_generic_suspend(dev);
5327 }
5328 int
5329 iflib_device_shutdown(device_t dev)
5330 {
5331 	if_ctx_t ctx = device_get_softc(dev);
5332 
5333 	CTX_LOCK(ctx);
5334 	IFDI_SHUTDOWN(ctx);
5335 	CTX_UNLOCK(ctx);
5336 
5337 	return bus_generic_suspend(dev);
5338 }
5339 
5340 int
5341 iflib_device_resume(device_t dev)
5342 {
5343 	if_ctx_t ctx = device_get_softc(dev);
5344 	iflib_txq_t txq = ctx->ifc_txqs;
5345 
5346 	CTX_LOCK(ctx);
5347 	IFDI_RESUME(ctx);
5348 	iflib_if_init_locked(ctx);
5349 	CTX_UNLOCK(ctx);
5350 	for (int i = 0; i < NTXQSETS(ctx); i++, txq++)
5351 		iflib_txq_check_drain(txq, IFLIB_RESTART_BUDGET);
5352 
5353 	return (bus_generic_resume(dev));
5354 }
5355 
5356 int
5357 iflib_device_iov_init(device_t dev, uint16_t num_vfs, const nvlist_t *params)
5358 {
5359 	int error;
5360 	if_ctx_t ctx = device_get_softc(dev);
5361 
5362 	CTX_LOCK(ctx);
5363 	error = IFDI_IOV_INIT(ctx, num_vfs, params);
5364 	CTX_UNLOCK(ctx);
5365 
5366 	return (error);
5367 }
5368 
5369 void
5370 iflib_device_iov_uninit(device_t dev)
5371 {
5372 	if_ctx_t ctx = device_get_softc(dev);
5373 
5374 	CTX_LOCK(ctx);
5375 	IFDI_IOV_UNINIT(ctx);
5376 	CTX_UNLOCK(ctx);
5377 }
5378 
5379 int
5380 iflib_device_iov_add_vf(device_t dev, uint16_t vfnum, const nvlist_t *params)
5381 {
5382 	int error;
5383 	if_ctx_t ctx = device_get_softc(dev);
5384 
5385 	CTX_LOCK(ctx);
5386 	error = IFDI_IOV_VF_ADD(ctx, vfnum, params);
5387 	CTX_UNLOCK(ctx);
5388 
5389 	return (error);
5390 }
5391 
5392 /*********************************************************************
5393  *
5394  *  MODULE FUNCTION DEFINITIONS
5395  *
5396  **********************************************************************/
5397 
5398 /*
5399  * - Start a fast taskqueue thread for each core
5400  * - Start a taskqueue for control operations
5401  */
5402 static int
5403 iflib_module_init(void)
5404 {
5405 	iflib_timer_default = hz / 2;
5406 	return (0);
5407 }
5408 
5409 static int
5410 iflib_module_event_handler(module_t mod, int what, void *arg)
5411 {
5412 	int err;
5413 
5414 	switch (what) {
5415 	case MOD_LOAD:
5416 		if ((err = iflib_module_init()) != 0)
5417 			return (err);
5418 		break;
5419 	case MOD_UNLOAD:
5420 		return (EBUSY);
5421 	default:
5422 		return (EOPNOTSUPP);
5423 	}
5424 
5425 	return (0);
5426 }
5427 
5428 /*********************************************************************
5429  *
5430  *  PUBLIC FUNCTION DEFINITIONS
5431  *     ordered as in iflib.h
5432  *
5433  **********************************************************************/
5434 
5435 static void
5436 _iflib_assert(if_shared_ctx_t sctx)
5437 {
5438 	int i;
5439 
5440 	MPASS(sctx->isc_tx_maxsize);
5441 	MPASS(sctx->isc_tx_maxsegsize);
5442 
5443 	MPASS(sctx->isc_rx_maxsize);
5444 	MPASS(sctx->isc_rx_nsegments);
5445 	MPASS(sctx->isc_rx_maxsegsize);
5446 
5447 	MPASS(sctx->isc_nrxqs >= 1 && sctx->isc_nrxqs <= 8);
5448 	for (i = 0; i < sctx->isc_nrxqs; i++) {
5449 		MPASS(sctx->isc_nrxd_min[i]);
5450 		MPASS(powerof2(sctx->isc_nrxd_min[i]));
5451 		MPASS(sctx->isc_nrxd_max[i]);
5452 		MPASS(powerof2(sctx->isc_nrxd_max[i]));
5453 		MPASS(sctx->isc_nrxd_default[i]);
5454 		MPASS(powerof2(sctx->isc_nrxd_default[i]));
5455 	}
5456 
5457 	MPASS(sctx->isc_ntxqs >= 1 && sctx->isc_ntxqs <= 8);
5458 	for (i = 0; i < sctx->isc_ntxqs; i++) {
5459 		MPASS(sctx->isc_ntxd_min[i]);
5460 		MPASS(powerof2(sctx->isc_ntxd_min[i]));
5461 		MPASS(sctx->isc_ntxd_max[i]);
5462 		MPASS(powerof2(sctx->isc_ntxd_max[i]));
5463 		MPASS(sctx->isc_ntxd_default[i]);
5464 		MPASS(powerof2(sctx->isc_ntxd_default[i]));
5465 	}
5466 }
5467 
5468 static void
5469 _iflib_pre_assert(if_softc_ctx_t scctx)
5470 {
5471 
5472 	MPASS(scctx->isc_txrx->ift_txd_encap);
5473 	MPASS(scctx->isc_txrx->ift_txd_flush);
5474 	MPASS(scctx->isc_txrx->ift_txd_credits_update);
5475 	MPASS(scctx->isc_txrx->ift_rxd_available);
5476 	MPASS(scctx->isc_txrx->ift_rxd_pkt_get);
5477 	MPASS(scctx->isc_txrx->ift_rxd_refill);
5478 	MPASS(scctx->isc_txrx->ift_rxd_flush);
5479 }
5480 
5481 static int
5482 iflib_register(if_ctx_t ctx)
5483 {
5484 	if_shared_ctx_t sctx = ctx->ifc_sctx;
5485 	driver_t *driver = sctx->isc_driver;
5486 	device_t dev = ctx->ifc_dev;
5487 	if_t ifp;
5488 	u_char type;
5489 	int iflags;
5490 
5491 	if ((sctx->isc_flags & IFLIB_PSEUDO) == 0)
5492 		_iflib_assert(sctx);
5493 
5494 	CTX_LOCK_INIT(ctx);
5495 	STATE_LOCK_INIT(ctx, device_get_nameunit(ctx->ifc_dev));
5496 	if (sctx->isc_flags & IFLIB_PSEUDO) {
5497 		if (sctx->isc_flags & IFLIB_PSEUDO_ETHER)
5498 			type = IFT_ETHER;
5499 		else
5500 			type = IFT_PPP;
5501 	} else
5502 		type = IFT_ETHER;
5503 	ifp = ctx->ifc_ifp = if_alloc(type);
5504 	if (ifp == NULL) {
5505 		device_printf(dev, "can not allocate ifnet structure\n");
5506 		return (ENOMEM);
5507 	}
5508 
5509 	/*
5510 	 * Initialize our context's device specific methods
5511 	 */
5512 	kobj_init((kobj_t) ctx, (kobj_class_t) driver);
5513 	kobj_class_compile((kobj_class_t) driver);
5514 
5515 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
5516 	if_setsoftc(ifp, ctx);
5517 	if_setdev(ifp, dev);
5518 	if_setinitfn(ifp, iflib_if_init);
5519 	if_setioctlfn(ifp, iflib_if_ioctl);
5520 #ifdef ALTQ
5521 	if_setstartfn(ifp, iflib_altq_if_start);
5522 	if_settransmitfn(ifp, iflib_altq_if_transmit);
5523 	if_setsendqready(ifp);
5524 #else
5525 	if_settransmitfn(ifp, iflib_if_transmit);
5526 #endif
5527 	if_setqflushfn(ifp, iflib_if_qflush);
5528 	iflags = IFF_MULTICAST | IFF_KNOWSEPOCH;
5529 
5530 	if ((sctx->isc_flags & IFLIB_PSEUDO) &&
5531 		(sctx->isc_flags & IFLIB_PSEUDO_ETHER) == 0)
5532 		iflags |= IFF_POINTOPOINT;
5533 	else
5534 		iflags |= IFF_BROADCAST | IFF_SIMPLEX;
5535 	if_setflags(ifp, iflags);
5536 	ctx->ifc_vlan_attach_event =
5537 		EVENTHANDLER_REGISTER(vlan_config, iflib_vlan_register, ctx,
5538 							  EVENTHANDLER_PRI_FIRST);
5539 	ctx->ifc_vlan_detach_event =
5540 		EVENTHANDLER_REGISTER(vlan_unconfig, iflib_vlan_unregister, ctx,
5541 							  EVENTHANDLER_PRI_FIRST);
5542 
5543 	if ((sctx->isc_flags & IFLIB_DRIVER_MEDIA) == 0) {
5544 		ctx->ifc_mediap = &ctx->ifc_media;
5545 		ifmedia_init(ctx->ifc_mediap, IFM_IMASK,
5546 		    iflib_media_change, iflib_media_status);
5547 	}
5548 	return (0);
5549 }
5550 
5551 static void
5552 iflib_unregister_vlan_handlers(if_ctx_t ctx)
5553 {
5554 	/* Unregister VLAN events */
5555 	if (ctx->ifc_vlan_attach_event != NULL) {
5556 		EVENTHANDLER_DEREGISTER(vlan_config, ctx->ifc_vlan_attach_event);
5557 		ctx->ifc_vlan_attach_event = NULL;
5558 	}
5559 	if (ctx->ifc_vlan_detach_event != NULL) {
5560 		EVENTHANDLER_DEREGISTER(vlan_unconfig, ctx->ifc_vlan_detach_event);
5561 		ctx->ifc_vlan_detach_event = NULL;
5562 	}
5563 
5564 }
5565 
5566 static void
5567 iflib_deregister(if_ctx_t ctx)
5568 {
5569 	if_t ifp = ctx->ifc_ifp;
5570 
5571 	/* Remove all media */
5572 	ifmedia_removeall(&ctx->ifc_media);
5573 
5574 	/* Ensure that VLAN event handlers are unregistered */
5575 	iflib_unregister_vlan_handlers(ctx);
5576 
5577 	/* Release kobject reference */
5578 	kobj_delete((kobj_t) ctx, NULL);
5579 
5580 	/* Free the ifnet structure */
5581 	if_free(ifp);
5582 
5583 	STATE_LOCK_DESTROY(ctx);
5584 
5585 	/* ether_ifdetach calls if_qflush - lock must be destroy afterwards*/
5586 	CTX_LOCK_DESTROY(ctx);
5587 }
5588 
5589 static int
5590 iflib_queues_alloc(if_ctx_t ctx)
5591 {
5592 	if_shared_ctx_t sctx = ctx->ifc_sctx;
5593 	if_softc_ctx_t scctx = &ctx->ifc_softc_ctx;
5594 	device_t dev = ctx->ifc_dev;
5595 	int nrxqsets = scctx->isc_nrxqsets;
5596 	int ntxqsets = scctx->isc_ntxqsets;
5597 	iflib_txq_t txq;
5598 	iflib_rxq_t rxq;
5599 	iflib_fl_t fl = NULL;
5600 	int i, j, cpu, err, txconf, rxconf;
5601 	iflib_dma_info_t ifdip;
5602 	uint32_t *rxqsizes = scctx->isc_rxqsizes;
5603 	uint32_t *txqsizes = scctx->isc_txqsizes;
5604 	uint8_t nrxqs = sctx->isc_nrxqs;
5605 	uint8_t ntxqs = sctx->isc_ntxqs;
5606 	int nfree_lists = sctx->isc_nfl ? sctx->isc_nfl : 1;
5607 	int fl_offset = (sctx->isc_flags & IFLIB_HAS_RXCQ ? 1 : 0);
5608 	caddr_t *vaddrs;
5609 	uint64_t *paddrs;
5610 
5611 	KASSERT(ntxqs > 0, ("number of queues per qset must be at least 1"));
5612 	KASSERT(nrxqs > 0, ("number of queues per qset must be at least 1"));
5613 	KASSERT(nrxqs >= fl_offset + nfree_lists,
5614            ("there must be at least a rxq for each free list"));
5615 
5616 	/* Allocate the TX ring struct memory */
5617 	if (!(ctx->ifc_txqs =
5618 	    (iflib_txq_t) malloc(sizeof(struct iflib_txq) *
5619 	    ntxqsets, M_IFLIB, M_NOWAIT | M_ZERO))) {
5620 		device_printf(dev, "Unable to allocate TX ring memory\n");
5621 		err = ENOMEM;
5622 		goto fail;
5623 	}
5624 
5625 	/* Now allocate the RX */
5626 	if (!(ctx->ifc_rxqs =
5627 	    (iflib_rxq_t) malloc(sizeof(struct iflib_rxq) *
5628 	    nrxqsets, M_IFLIB, M_NOWAIT | M_ZERO))) {
5629 		device_printf(dev, "Unable to allocate RX ring memory\n");
5630 		err = ENOMEM;
5631 		goto rx_fail;
5632 	}
5633 
5634 	txq = ctx->ifc_txqs;
5635 	rxq = ctx->ifc_rxqs;
5636 
5637 	/*
5638 	 * XXX handle allocation failure
5639 	 */
5640 	for (txconf = i = 0, cpu = CPU_FIRST(); i < ntxqsets; i++, txconf++, txq++, cpu = CPU_NEXT(cpu)) {
5641 		/* Set up some basics */
5642 
5643 		if ((ifdip = malloc(sizeof(struct iflib_dma_info) * ntxqs,
5644 		    M_IFLIB, M_NOWAIT | M_ZERO)) == NULL) {
5645 			device_printf(dev,
5646 			    "Unable to allocate TX DMA info memory\n");
5647 			err = ENOMEM;
5648 			goto err_tx_desc;
5649 		}
5650 		txq->ift_ifdi = ifdip;
5651 		for (j = 0; j < ntxqs; j++, ifdip++) {
5652 			if (iflib_dma_alloc(ctx, txqsizes[j], ifdip, 0)) {
5653 				device_printf(dev,
5654 				    "Unable to allocate TX descriptors\n");
5655 				err = ENOMEM;
5656 				goto err_tx_desc;
5657 			}
5658 			txq->ift_txd_size[j] = scctx->isc_txd_size[j];
5659 			bzero((void *)ifdip->idi_vaddr, txqsizes[j]);
5660 		}
5661 		txq->ift_ctx = ctx;
5662 		txq->ift_id = i;
5663 		if (sctx->isc_flags & IFLIB_HAS_TXCQ) {
5664 			txq->ift_br_offset = 1;
5665 		} else {
5666 			txq->ift_br_offset = 0;
5667 		}
5668 
5669 		if (iflib_txsd_alloc(txq)) {
5670 			device_printf(dev, "Critical Failure setting up TX buffers\n");
5671 			err = ENOMEM;
5672 			goto err_tx_desc;
5673 		}
5674 
5675 		/* Initialize the TX lock */
5676 		snprintf(txq->ift_mtx_name, MTX_NAME_LEN, "%s:TX(%d):callout",
5677 		    device_get_nameunit(dev), txq->ift_id);
5678 		mtx_init(&txq->ift_mtx, txq->ift_mtx_name, NULL, MTX_DEF);
5679 		callout_init_mtx(&txq->ift_timer, &txq->ift_mtx, 0);
5680 		txq->ift_timer.c_cpu = cpu;
5681 #ifdef DEV_NETMAP
5682 		callout_init_mtx(&txq->ift_netmap_timer, &txq->ift_mtx, 0);
5683 		txq->ift_netmap_timer.c_cpu = cpu;
5684 #endif /* DEV_NETMAP */
5685 
5686 		err = ifmp_ring_alloc(&txq->ift_br, 2048, txq, iflib_txq_drain,
5687 				      iflib_txq_can_drain, M_IFLIB, M_WAITOK);
5688 		if (err) {
5689 			/* XXX free any allocated rings */
5690 			device_printf(dev, "Unable to allocate buf_ring\n");
5691 			goto err_tx_desc;
5692 		}
5693 	}
5694 
5695 	for (rxconf = i = 0; i < nrxqsets; i++, rxconf++, rxq++) {
5696 		/* Set up some basics */
5697 		callout_init(&rxq->ifr_watchdog, 1);
5698 
5699 		if ((ifdip = malloc(sizeof(struct iflib_dma_info) * nrxqs,
5700 		   M_IFLIB, M_NOWAIT | M_ZERO)) == NULL) {
5701 			device_printf(dev,
5702 			    "Unable to allocate RX DMA info memory\n");
5703 			err = ENOMEM;
5704 			goto err_tx_desc;
5705 		}
5706 
5707 		rxq->ifr_ifdi = ifdip;
5708 		/* XXX this needs to be changed if #rx queues != #tx queues */
5709 		rxq->ifr_ntxqirq = 1;
5710 		rxq->ifr_txqid[0] = i;
5711 		for (j = 0; j < nrxqs; j++, ifdip++) {
5712 			if (iflib_dma_alloc(ctx, rxqsizes[j], ifdip, 0)) {
5713 				device_printf(dev,
5714 				    "Unable to allocate RX descriptors\n");
5715 				err = ENOMEM;
5716 				goto err_tx_desc;
5717 			}
5718 			bzero((void *)ifdip->idi_vaddr, rxqsizes[j]);
5719 		}
5720 		rxq->ifr_ctx = ctx;
5721 		rxq->ifr_id = i;
5722 		rxq->ifr_fl_offset = fl_offset;
5723 		rxq->ifr_nfl = nfree_lists;
5724 		if (!(fl =
5725 			  (iflib_fl_t) malloc(sizeof(struct iflib_fl) * nfree_lists, M_IFLIB, M_NOWAIT | M_ZERO))) {
5726 			device_printf(dev, "Unable to allocate free list memory\n");
5727 			err = ENOMEM;
5728 			goto err_tx_desc;
5729 		}
5730 		rxq->ifr_fl = fl;
5731 		for (j = 0; j < nfree_lists; j++) {
5732 			fl[j].ifl_rxq = rxq;
5733 			fl[j].ifl_id = j;
5734 			fl[j].ifl_ifdi = &rxq->ifr_ifdi[j + rxq->ifr_fl_offset];
5735 			fl[j].ifl_rxd_size = scctx->isc_rxd_size[j];
5736 		}
5737 		/* Allocate receive buffers for the ring */
5738 		if (iflib_rxsd_alloc(rxq)) {
5739 			device_printf(dev,
5740 			    "Critical Failure setting up receive buffers\n");
5741 			err = ENOMEM;
5742 			goto err_rx_desc;
5743 		}
5744 
5745 		for (j = 0, fl = rxq->ifr_fl; j < rxq->ifr_nfl; j++, fl++)
5746 			fl->ifl_rx_bitmap = bit_alloc(fl->ifl_size, M_IFLIB,
5747 			    M_WAITOK);
5748 	}
5749 
5750 	/* TXQs */
5751 	vaddrs = malloc(sizeof(caddr_t)*ntxqsets*ntxqs, M_IFLIB, M_WAITOK);
5752 	paddrs = malloc(sizeof(uint64_t)*ntxqsets*ntxqs, M_IFLIB, M_WAITOK);
5753 	for (i = 0; i < ntxqsets; i++) {
5754 		iflib_dma_info_t di = ctx->ifc_txqs[i].ift_ifdi;
5755 
5756 		for (j = 0; j < ntxqs; j++, di++) {
5757 			vaddrs[i*ntxqs + j] = di->idi_vaddr;
5758 			paddrs[i*ntxqs + j] = di->idi_paddr;
5759 		}
5760 	}
5761 	if ((err = IFDI_TX_QUEUES_ALLOC(ctx, vaddrs, paddrs, ntxqs, ntxqsets)) != 0) {
5762 		device_printf(ctx->ifc_dev,
5763 		    "Unable to allocate device TX queue\n");
5764 		iflib_tx_structures_free(ctx);
5765 		free(vaddrs, M_IFLIB);
5766 		free(paddrs, M_IFLIB);
5767 		goto err_rx_desc;
5768 	}
5769 	free(vaddrs, M_IFLIB);
5770 	free(paddrs, M_IFLIB);
5771 
5772 	/* RXQs */
5773 	vaddrs = malloc(sizeof(caddr_t)*nrxqsets*nrxqs, M_IFLIB, M_WAITOK);
5774 	paddrs = malloc(sizeof(uint64_t)*nrxqsets*nrxqs, M_IFLIB, M_WAITOK);
5775 	for (i = 0; i < nrxqsets; i++) {
5776 		iflib_dma_info_t di = ctx->ifc_rxqs[i].ifr_ifdi;
5777 
5778 		for (j = 0; j < nrxqs; j++, di++) {
5779 			vaddrs[i*nrxqs + j] = di->idi_vaddr;
5780 			paddrs[i*nrxqs + j] = di->idi_paddr;
5781 		}
5782 	}
5783 	if ((err = IFDI_RX_QUEUES_ALLOC(ctx, vaddrs, paddrs, nrxqs, nrxqsets)) != 0) {
5784 		device_printf(ctx->ifc_dev,
5785 		    "Unable to allocate device RX queue\n");
5786 		iflib_tx_structures_free(ctx);
5787 		free(vaddrs, M_IFLIB);
5788 		free(paddrs, M_IFLIB);
5789 		goto err_rx_desc;
5790 	}
5791 	free(vaddrs, M_IFLIB);
5792 	free(paddrs, M_IFLIB);
5793 
5794 	return (0);
5795 
5796 /* XXX handle allocation failure changes */
5797 err_rx_desc:
5798 err_tx_desc:
5799 rx_fail:
5800 	if (ctx->ifc_rxqs != NULL)
5801 		free(ctx->ifc_rxqs, M_IFLIB);
5802 	ctx->ifc_rxqs = NULL;
5803 	if (ctx->ifc_txqs != NULL)
5804 		free(ctx->ifc_txqs, M_IFLIB);
5805 	ctx->ifc_txqs = NULL;
5806 fail:
5807 	return (err);
5808 }
5809 
5810 static int
5811 iflib_tx_structures_setup(if_ctx_t ctx)
5812 {
5813 	iflib_txq_t txq = ctx->ifc_txqs;
5814 	int i;
5815 
5816 	for (i = 0; i < NTXQSETS(ctx); i++, txq++)
5817 		iflib_txq_setup(txq);
5818 
5819 	return (0);
5820 }
5821 
5822 static void
5823 iflib_tx_structures_free(if_ctx_t ctx)
5824 {
5825 	iflib_txq_t txq = ctx->ifc_txqs;
5826 	if_shared_ctx_t sctx = ctx->ifc_sctx;
5827 	int i, j;
5828 
5829 	for (i = 0; i < NTXQSETS(ctx); i++, txq++) {
5830 		for (j = 0; j < sctx->isc_ntxqs; j++)
5831 			iflib_dma_free(&txq->ift_ifdi[j]);
5832 		iflib_txq_destroy(txq);
5833 	}
5834 	free(ctx->ifc_txqs, M_IFLIB);
5835 	ctx->ifc_txqs = NULL;
5836 }
5837 
5838 /*********************************************************************
5839  *
5840  *  Initialize all receive rings.
5841  *
5842  **********************************************************************/
5843 static int
5844 iflib_rx_structures_setup(if_ctx_t ctx)
5845 {
5846 	iflib_rxq_t rxq = ctx->ifc_rxqs;
5847 	int q;
5848 #if defined(INET6) || defined(INET)
5849 	int err, i;
5850 #endif
5851 
5852 	for (q = 0; q < ctx->ifc_softc_ctx.isc_nrxqsets; q++, rxq++) {
5853 #if defined(INET6) || defined(INET)
5854 		if (if_getcapabilities(ctx->ifc_ifp) & IFCAP_LRO) {
5855 			err = tcp_lro_init_args(&rxq->ifr_lc, ctx->ifc_ifp,
5856 			    TCP_LRO_ENTRIES, min(1024,
5857 			    ctx->ifc_softc_ctx.isc_nrxd[rxq->ifr_fl_offset]));
5858 			if (err != 0) {
5859 				device_printf(ctx->ifc_dev,
5860 				    "LRO Initialization failed!\n");
5861 				goto fail;
5862 			}
5863 		}
5864 #endif
5865 		IFDI_RXQ_SETUP(ctx, rxq->ifr_id);
5866 	}
5867 	return (0);
5868 #if defined(INET6) || defined(INET)
5869 fail:
5870 	/*
5871 	 * Free LRO resources allocated so far, we will only handle
5872 	 * the rings that completed, the failing case will have
5873 	 * cleaned up for itself.  'q' failed, so its the terminus.
5874 	 */
5875 	rxq = ctx->ifc_rxqs;
5876 	for (i = 0; i < q; ++i, rxq++) {
5877 		if (if_getcapabilities(ctx->ifc_ifp) & IFCAP_LRO)
5878 			tcp_lro_free(&rxq->ifr_lc);
5879 	}
5880 	return (err);
5881 #endif
5882 }
5883 
5884 /*********************************************************************
5885  *
5886  *  Free all receive rings.
5887  *
5888  **********************************************************************/
5889 static void
5890 iflib_rx_structures_free(if_ctx_t ctx)
5891 {
5892 	iflib_rxq_t rxq = ctx->ifc_rxqs;
5893 	if_shared_ctx_t sctx = ctx->ifc_sctx;
5894 	int i, j;
5895 
5896 	for (i = 0; i < ctx->ifc_softc_ctx.isc_nrxqsets; i++, rxq++) {
5897 		for (j = 0; j < sctx->isc_nrxqs; j++)
5898 			iflib_dma_free(&rxq->ifr_ifdi[j]);
5899 		iflib_rx_sds_free(rxq);
5900 #if defined(INET6) || defined(INET)
5901 		if (if_getcapabilities(ctx->ifc_ifp) & IFCAP_LRO)
5902 			tcp_lro_free(&rxq->ifr_lc);
5903 #endif
5904 	}
5905 	free(ctx->ifc_rxqs, M_IFLIB);
5906 	ctx->ifc_rxqs = NULL;
5907 }
5908 
5909 static int
5910 iflib_qset_structures_setup(if_ctx_t ctx)
5911 {
5912 	int err;
5913 
5914 	/*
5915 	 * It is expected that the caller takes care of freeing queues if this
5916 	 * fails.
5917 	 */
5918 	if ((err = iflib_tx_structures_setup(ctx)) != 0) {
5919 		device_printf(ctx->ifc_dev, "iflib_tx_structures_setup failed: %d\n", err);
5920 		return (err);
5921 	}
5922 
5923 	if ((err = iflib_rx_structures_setup(ctx)) != 0)
5924 		device_printf(ctx->ifc_dev, "iflib_rx_structures_setup failed: %d\n", err);
5925 
5926 	return (err);
5927 }
5928 
5929 int
5930 iflib_irq_alloc(if_ctx_t ctx, if_irq_t irq, int rid,
5931 		driver_filter_t filter, void *filter_arg, driver_intr_t handler, void *arg, const char *name)
5932 {
5933 
5934 	return (_iflib_irq_alloc(ctx, irq, rid, filter, handler, arg, name));
5935 }
5936 
5937 #ifdef SMP
5938 static int
5939 find_nth(if_ctx_t ctx, int qid)
5940 {
5941 	cpuset_t cpus;
5942 	int i, cpuid, eqid, count;
5943 
5944 	CPU_COPY(&ctx->ifc_cpus, &cpus);
5945 	count = CPU_COUNT(&cpus);
5946 	eqid = qid % count;
5947 	/* clear up to the qid'th bit */
5948 	for (i = 0; i < eqid; i++) {
5949 		cpuid = CPU_FFS(&cpus);
5950 		MPASS(cpuid != 0);
5951 		CPU_CLR(cpuid-1, &cpus);
5952 	}
5953 	cpuid = CPU_FFS(&cpus);
5954 	MPASS(cpuid != 0);
5955 	return (cpuid-1);
5956 }
5957 
5958 #ifdef SCHED_ULE
5959 extern struct cpu_group *cpu_top;              /* CPU topology */
5960 
5961 static int
5962 find_child_with_core(int cpu, struct cpu_group *grp)
5963 {
5964 	int i;
5965 
5966 	if (grp->cg_children == 0)
5967 		return -1;
5968 
5969 	MPASS(grp->cg_child);
5970 	for (i = 0; i < grp->cg_children; i++) {
5971 		if (CPU_ISSET(cpu, &grp->cg_child[i].cg_mask))
5972 			return i;
5973 	}
5974 
5975 	return -1;
5976 }
5977 
5978 /*
5979  * Find the nth "close" core to the specified core
5980  * "close" is defined as the deepest level that shares
5981  * at least an L2 cache.  With threads, this will be
5982  * threads on the same core.  If the shared cache is L3
5983  * or higher, simply returns the same core.
5984  */
5985 static int
5986 find_close_core(int cpu, int core_offset)
5987 {
5988 	struct cpu_group *grp;
5989 	int i;
5990 	int fcpu;
5991 	cpuset_t cs;
5992 
5993 	grp = cpu_top;
5994 	if (grp == NULL)
5995 		return cpu;
5996 	i = 0;
5997 	while ((i = find_child_with_core(cpu, grp)) != -1) {
5998 		/* If the child only has one cpu, don't descend */
5999 		if (grp->cg_child[i].cg_count <= 1)
6000 			break;
6001 		grp = &grp->cg_child[i];
6002 	}
6003 
6004 	/* If they don't share at least an L2 cache, use the same CPU */
6005 	if (grp->cg_level > CG_SHARE_L2 || grp->cg_level == CG_SHARE_NONE)
6006 		return cpu;
6007 
6008 	/* Now pick one */
6009 	CPU_COPY(&grp->cg_mask, &cs);
6010 
6011 	/* Add the selected CPU offset to core offset. */
6012 	for (i = 0; (fcpu = CPU_FFS(&cs)) != 0; i++) {
6013 		if (fcpu - 1 == cpu)
6014 			break;
6015 		CPU_CLR(fcpu - 1, &cs);
6016 	}
6017 	MPASS(fcpu);
6018 
6019 	core_offset += i;
6020 
6021 	CPU_COPY(&grp->cg_mask, &cs);
6022 	for (i = core_offset % grp->cg_count; i > 0; i--) {
6023 		MPASS(CPU_FFS(&cs));
6024 		CPU_CLR(CPU_FFS(&cs) - 1, &cs);
6025 	}
6026 	MPASS(CPU_FFS(&cs));
6027 	return CPU_FFS(&cs) - 1;
6028 }
6029 #else
6030 static int
6031 find_close_core(int cpu, int core_offset __unused)
6032 {
6033 	return cpu;
6034 }
6035 #endif
6036 
6037 static int
6038 get_core_offset(if_ctx_t ctx, iflib_intr_type_t type, int qid)
6039 {
6040 	switch (type) {
6041 	case IFLIB_INTR_TX:
6042 		/* TX queues get cores which share at least an L2 cache with the corresponding RX queue */
6043 		/* XXX handle multiple RX threads per core and more than two core per L2 group */
6044 		return qid / CPU_COUNT(&ctx->ifc_cpus) + 1;
6045 	case IFLIB_INTR_RX:
6046 	case IFLIB_INTR_RXTX:
6047 		/* RX queues get the specified core */
6048 		return qid / CPU_COUNT(&ctx->ifc_cpus);
6049 	default:
6050 		return -1;
6051 	}
6052 }
6053 #else
6054 #define get_core_offset(ctx, type, qid)	CPU_FIRST()
6055 #define find_close_core(cpuid, tid)	CPU_FIRST()
6056 #define find_nth(ctx, gid)		CPU_FIRST()
6057 #endif
6058 
6059 /* Just to avoid copy/paste */
6060 static inline int
6061 iflib_irq_set_affinity(if_ctx_t ctx, if_irq_t irq, iflib_intr_type_t type,
6062     int qid, struct grouptask *gtask, struct taskqgroup *tqg, void *uniq,
6063     const char *name)
6064 {
6065 	device_t dev;
6066 	int co, cpuid, err, tid;
6067 
6068 	dev = ctx->ifc_dev;
6069 	co = ctx->ifc_sysctl_core_offset;
6070 	if (ctx->ifc_sysctl_separate_txrx && type == IFLIB_INTR_TX)
6071 		co += ctx->ifc_softc_ctx.isc_nrxqsets;
6072 	cpuid = find_nth(ctx, qid + co);
6073 	tid = get_core_offset(ctx, type, qid);
6074 	if (tid < 0) {
6075 		device_printf(dev, "get_core_offset failed\n");
6076 		return (EOPNOTSUPP);
6077 	}
6078 	cpuid = find_close_core(cpuid, tid);
6079 	err = taskqgroup_attach_cpu(tqg, gtask, uniq, cpuid, dev, irq->ii_res,
6080 	    name);
6081 	if (err) {
6082 		device_printf(dev, "taskqgroup_attach_cpu failed %d\n", err);
6083 		return (err);
6084 	}
6085 #ifdef notyet
6086 	if (cpuid > ctx->ifc_cpuid_highest)
6087 		ctx->ifc_cpuid_highest = cpuid;
6088 #endif
6089 	return (0);
6090 }
6091 
6092 int
6093 iflib_irq_alloc_generic(if_ctx_t ctx, if_irq_t irq, int rid,
6094 			iflib_intr_type_t type, driver_filter_t *filter,
6095 			void *filter_arg, int qid, const char *name)
6096 {
6097 	device_t dev;
6098 	struct grouptask *gtask;
6099 	struct taskqgroup *tqg;
6100 	iflib_filter_info_t info;
6101 	gtask_fn_t *fn;
6102 	int tqrid, err;
6103 	driver_filter_t *intr_fast;
6104 	void *q;
6105 
6106 	info = &ctx->ifc_filter_info;
6107 	tqrid = rid;
6108 
6109 	switch (type) {
6110 	/* XXX merge tx/rx for netmap? */
6111 	case IFLIB_INTR_TX:
6112 		q = &ctx->ifc_txqs[qid];
6113 		info = &ctx->ifc_txqs[qid].ift_filter_info;
6114 		gtask = &ctx->ifc_txqs[qid].ift_task;
6115 		tqg = qgroup_if_io_tqg;
6116 		fn = _task_fn_tx;
6117 		intr_fast = iflib_fast_intr;
6118 		GROUPTASK_INIT(gtask, 0, fn, q);
6119 		ctx->ifc_flags |= IFC_NETMAP_TX_IRQ;
6120 		break;
6121 	case IFLIB_INTR_RX:
6122 		q = &ctx->ifc_rxqs[qid];
6123 		info = &ctx->ifc_rxqs[qid].ifr_filter_info;
6124 		gtask = &ctx->ifc_rxqs[qid].ifr_task;
6125 		tqg = qgroup_if_io_tqg;
6126 		fn = _task_fn_rx;
6127 		intr_fast = iflib_fast_intr;
6128 		NET_GROUPTASK_INIT(gtask, 0, fn, q);
6129 		break;
6130 	case IFLIB_INTR_RXTX:
6131 		q = &ctx->ifc_rxqs[qid];
6132 		info = &ctx->ifc_rxqs[qid].ifr_filter_info;
6133 		gtask = &ctx->ifc_rxqs[qid].ifr_task;
6134 		tqg = qgroup_if_io_tqg;
6135 		fn = _task_fn_rx;
6136 		intr_fast = iflib_fast_intr_rxtx;
6137 		NET_GROUPTASK_INIT(gtask, 0, fn, q);
6138 		break;
6139 	case IFLIB_INTR_ADMIN:
6140 		q = ctx;
6141 		tqrid = -1;
6142 		info = &ctx->ifc_filter_info;
6143 		gtask = &ctx->ifc_admin_task;
6144 		tqg = qgroup_if_config_tqg;
6145 		fn = _task_fn_admin;
6146 		intr_fast = iflib_fast_intr_ctx;
6147 		break;
6148 	default:
6149 		device_printf(ctx->ifc_dev, "%s: unknown net intr type\n",
6150 		    __func__);
6151 		return (EINVAL);
6152 	}
6153 
6154 	info->ifi_filter = filter;
6155 	info->ifi_filter_arg = filter_arg;
6156 	info->ifi_task = gtask;
6157 	info->ifi_ctx = q;
6158 
6159 	dev = ctx->ifc_dev;
6160 	err = _iflib_irq_alloc(ctx, irq, rid, intr_fast, NULL, info,  name);
6161 	if (err != 0) {
6162 		device_printf(dev, "_iflib_irq_alloc failed %d\n", err);
6163 		return (err);
6164 	}
6165 	if (type == IFLIB_INTR_ADMIN)
6166 		return (0);
6167 
6168 	if (tqrid != -1) {
6169 		err = iflib_irq_set_affinity(ctx, irq, type, qid, gtask, tqg,
6170 		    q, name);
6171 		if (err)
6172 			return (err);
6173 	} else {
6174 		taskqgroup_attach(tqg, gtask, q, dev, irq->ii_res, name);
6175 	}
6176 
6177 	return (0);
6178 }
6179 
6180 void
6181 iflib_softirq_alloc_generic(if_ctx_t ctx, if_irq_t irq, iflib_intr_type_t type, void *arg, int qid, const char *name)
6182 {
6183 	struct grouptask *gtask;
6184 	struct taskqgroup *tqg;
6185 	gtask_fn_t *fn;
6186 	void *q;
6187 	int err;
6188 
6189 	switch (type) {
6190 	case IFLIB_INTR_TX:
6191 		q = &ctx->ifc_txqs[qid];
6192 		gtask = &ctx->ifc_txqs[qid].ift_task;
6193 		tqg = qgroup_if_io_tqg;
6194 		fn = _task_fn_tx;
6195 		GROUPTASK_INIT(gtask, 0, fn, q);
6196 		break;
6197 	case IFLIB_INTR_RX:
6198 		q = &ctx->ifc_rxqs[qid];
6199 		gtask = &ctx->ifc_rxqs[qid].ifr_task;
6200 		tqg = qgroup_if_io_tqg;
6201 		fn = _task_fn_rx;
6202 		NET_GROUPTASK_INIT(gtask, 0, fn, q);
6203 		break;
6204 	case IFLIB_INTR_IOV:
6205 		q = ctx;
6206 		gtask = &ctx->ifc_vflr_task;
6207 		tqg = qgroup_if_config_tqg;
6208 		fn = _task_fn_iov;
6209 		GROUPTASK_INIT(gtask, 0, fn, q);
6210 		break;
6211 	default:
6212 		panic("unknown net intr type");
6213 	}
6214 	if (irq != NULL) {
6215 		err = iflib_irq_set_affinity(ctx, irq, type, qid, gtask, tqg,
6216 		    q, name);
6217 		if (err)
6218 			taskqgroup_attach(tqg, gtask, q, ctx->ifc_dev,
6219 			    irq->ii_res, name);
6220 	} else {
6221 		taskqgroup_attach(tqg, gtask, q, NULL, NULL, name);
6222 	}
6223 }
6224 
6225 void
6226 iflib_irq_free(if_ctx_t ctx, if_irq_t irq)
6227 {
6228 
6229 	if (irq->ii_tag)
6230 		bus_teardown_intr(ctx->ifc_dev, irq->ii_res, irq->ii_tag);
6231 
6232 	if (irq->ii_res)
6233 		bus_release_resource(ctx->ifc_dev, SYS_RES_IRQ,
6234 		    rman_get_rid(irq->ii_res), irq->ii_res);
6235 }
6236 
6237 static int
6238 iflib_legacy_setup(if_ctx_t ctx, driver_filter_t filter, void *filter_arg, int *rid, const char *name)
6239 {
6240 	iflib_txq_t txq = ctx->ifc_txqs;
6241 	iflib_rxq_t rxq = ctx->ifc_rxqs;
6242 	if_irq_t irq = &ctx->ifc_legacy_irq;
6243 	iflib_filter_info_t info;
6244 	device_t dev;
6245 	struct grouptask *gtask;
6246 	struct resource *res;
6247 	struct taskqgroup *tqg;
6248 	void *q;
6249 	int err, tqrid;
6250 	bool rx_only;
6251 
6252 	q = &ctx->ifc_rxqs[0];
6253 	info = &rxq[0].ifr_filter_info;
6254 	gtask = &rxq[0].ifr_task;
6255 	tqg = qgroup_if_io_tqg;
6256 	tqrid = *rid;
6257 	rx_only = (ctx->ifc_sctx->isc_flags & IFLIB_SINGLE_IRQ_RX_ONLY) != 0;
6258 
6259 	ctx->ifc_flags |= IFC_LEGACY;
6260 	info->ifi_filter = filter;
6261 	info->ifi_filter_arg = filter_arg;
6262 	info->ifi_task = gtask;
6263 	info->ifi_ctx = rx_only ? ctx : q;
6264 
6265 	dev = ctx->ifc_dev;
6266 	/* We allocate a single interrupt resource */
6267 	err = _iflib_irq_alloc(ctx, irq, tqrid, rx_only ? iflib_fast_intr_ctx :
6268 	    iflib_fast_intr_rxtx, NULL, info, name);
6269 	if (err != 0)
6270 		return (err);
6271 	NET_GROUPTASK_INIT(gtask, 0, _task_fn_rx, q);
6272 	res = irq->ii_res;
6273 	taskqgroup_attach(tqg, gtask, q, dev, res, name);
6274 
6275 	GROUPTASK_INIT(&txq->ift_task, 0, _task_fn_tx, txq);
6276 	taskqgroup_attach(qgroup_if_io_tqg, &txq->ift_task, txq, dev, res,
6277 	    "tx");
6278 	return (0);
6279 }
6280 
6281 void
6282 iflib_led_create(if_ctx_t ctx)
6283 {
6284 
6285 	ctx->ifc_led_dev = led_create(iflib_led_func, ctx,
6286 	    device_get_nameunit(ctx->ifc_dev));
6287 }
6288 
6289 void
6290 iflib_tx_intr_deferred(if_ctx_t ctx, int txqid)
6291 {
6292 
6293 	GROUPTASK_ENQUEUE(&ctx->ifc_txqs[txqid].ift_task);
6294 }
6295 
6296 void
6297 iflib_rx_intr_deferred(if_ctx_t ctx, int rxqid)
6298 {
6299 
6300 	GROUPTASK_ENQUEUE(&ctx->ifc_rxqs[rxqid].ifr_task);
6301 }
6302 
6303 void
6304 iflib_admin_intr_deferred(if_ctx_t ctx)
6305 {
6306 
6307 	MPASS(ctx->ifc_admin_task.gt_taskqueue != NULL);
6308 	GROUPTASK_ENQUEUE(&ctx->ifc_admin_task);
6309 }
6310 
6311 void
6312 iflib_iov_intr_deferred(if_ctx_t ctx)
6313 {
6314 
6315 	GROUPTASK_ENQUEUE(&ctx->ifc_vflr_task);
6316 }
6317 
6318 void
6319 iflib_io_tqg_attach(struct grouptask *gt, void *uniq, int cpu, const char *name)
6320 {
6321 
6322 	taskqgroup_attach_cpu(qgroup_if_io_tqg, gt, uniq, cpu, NULL, NULL,
6323 	    name);
6324 }
6325 
6326 void
6327 iflib_config_gtask_init(void *ctx, struct grouptask *gtask, gtask_fn_t *fn,
6328 	const char *name)
6329 {
6330 
6331 	GROUPTASK_INIT(gtask, 0, fn, ctx);
6332 	taskqgroup_attach(qgroup_if_config_tqg, gtask, gtask, NULL, NULL,
6333 	    name);
6334 }
6335 
6336 void
6337 iflib_config_gtask_deinit(struct grouptask *gtask)
6338 {
6339 
6340 	taskqgroup_detach(qgroup_if_config_tqg, gtask);
6341 }
6342 
6343 void
6344 iflib_link_state_change(if_ctx_t ctx, int link_state, uint64_t baudrate)
6345 {
6346 	if_t ifp = ctx->ifc_ifp;
6347 	iflib_txq_t txq = ctx->ifc_txqs;
6348 
6349 	if_setbaudrate(ifp, baudrate);
6350 	if (baudrate >= IF_Gbps(10)) {
6351 		STATE_LOCK(ctx);
6352 		ctx->ifc_flags |= IFC_PREFETCH;
6353 		STATE_UNLOCK(ctx);
6354 	}
6355 	/* If link down, disable watchdog */
6356 	if ((ctx->ifc_link_state == LINK_STATE_UP) && (link_state == LINK_STATE_DOWN)) {
6357 		for (int i = 0; i < ctx->ifc_softc_ctx.isc_ntxqsets; i++, txq++)
6358 			txq->ift_qstatus = IFLIB_QUEUE_IDLE;
6359 	}
6360 	ctx->ifc_link_state = link_state;
6361 	if_link_state_change(ifp, link_state);
6362 }
6363 
6364 static int
6365 iflib_tx_credits_update(if_ctx_t ctx, iflib_txq_t txq)
6366 {
6367 	int credits;
6368 #ifdef INVARIANTS
6369 	int credits_pre = txq->ift_cidx_processed;
6370 #endif
6371 
6372 	bus_dmamap_sync(txq->ift_ifdi->idi_tag, txq->ift_ifdi->idi_map,
6373 	    BUS_DMASYNC_POSTREAD);
6374 	if ((credits = ctx->isc_txd_credits_update(ctx->ifc_softc, txq->ift_id, true)) == 0)
6375 		return (0);
6376 
6377 	txq->ift_processed += credits;
6378 	txq->ift_cidx_processed += credits;
6379 
6380 	MPASS(credits_pre + credits == txq->ift_cidx_processed);
6381 	if (txq->ift_cidx_processed >= txq->ift_size)
6382 		txq->ift_cidx_processed -= txq->ift_size;
6383 	return (credits);
6384 }
6385 
6386 static int
6387 iflib_rxd_avail(if_ctx_t ctx, iflib_rxq_t rxq, qidx_t cidx, qidx_t budget)
6388 {
6389 	iflib_fl_t fl;
6390 	u_int i;
6391 
6392 	for (i = 0, fl = &rxq->ifr_fl[0]; i < rxq->ifr_nfl; i++, fl++)
6393 		bus_dmamap_sync(fl->ifl_ifdi->idi_tag, fl->ifl_ifdi->idi_map,
6394 		    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
6395 	return (ctx->isc_rxd_available(ctx->ifc_softc, rxq->ifr_id, cidx,
6396 	    budget));
6397 }
6398 
6399 void
6400 iflib_add_int_delay_sysctl(if_ctx_t ctx, const char *name,
6401 	const char *description, if_int_delay_info_t info,
6402 	int offset, int value)
6403 {
6404 	info->iidi_ctx = ctx;
6405 	info->iidi_offset = offset;
6406 	info->iidi_value = value;
6407 	SYSCTL_ADD_PROC(device_get_sysctl_ctx(ctx->ifc_dev),
6408 	    SYSCTL_CHILDREN(device_get_sysctl_tree(ctx->ifc_dev)),
6409 	    OID_AUTO, name, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
6410 	    info, 0, iflib_sysctl_int_delay, "I", description);
6411 }
6412 
6413 struct sx *
6414 iflib_ctx_lock_get(if_ctx_t ctx)
6415 {
6416 
6417 	return (&ctx->ifc_ctx_sx);
6418 }
6419 
6420 static int
6421 iflib_msix_init(if_ctx_t ctx)
6422 {
6423 	device_t dev = ctx->ifc_dev;
6424 	if_shared_ctx_t sctx = ctx->ifc_sctx;
6425 	if_softc_ctx_t scctx = &ctx->ifc_softc_ctx;
6426 	int admincnt, bar, err, iflib_num_rx_queues, iflib_num_tx_queues;
6427 	int msgs, queuemsgs, queues, rx_queues, tx_queues, vectors;
6428 
6429 	iflib_num_tx_queues = ctx->ifc_sysctl_ntxqs;
6430 	iflib_num_rx_queues = ctx->ifc_sysctl_nrxqs;
6431 
6432 	if (bootverbose)
6433 		device_printf(dev, "msix_init qsets capped at %d\n",
6434 		    imax(scctx->isc_ntxqsets, scctx->isc_nrxqsets));
6435 
6436 	/* Override by tuneable */
6437 	if (scctx->isc_disable_msix)
6438 		goto msi;
6439 
6440 	/* First try MSI-X */
6441 	if ((msgs = pci_msix_count(dev)) == 0) {
6442 		if (bootverbose)
6443 			device_printf(dev, "MSI-X not supported or disabled\n");
6444 		goto msi;
6445 	}
6446 
6447 	bar = ctx->ifc_softc_ctx.isc_msix_bar;
6448 	/*
6449 	 * bar == -1 => "trust me I know what I'm doing"
6450 	 * Some drivers are for hardware that is so shoddily
6451 	 * documented that no one knows which bars are which
6452 	 * so the developer has to map all bars. This hack
6453 	 * allows shoddy garbage to use MSI-X in this framework.
6454 	 */
6455 	if (bar != -1) {
6456 		ctx->ifc_msix_mem = bus_alloc_resource_any(dev,
6457 	            SYS_RES_MEMORY, &bar, RF_ACTIVE);
6458 		if (ctx->ifc_msix_mem == NULL) {
6459 			device_printf(dev, "Unable to map MSI-X table\n");
6460 			goto msi;
6461 		}
6462 	}
6463 
6464 	admincnt = sctx->isc_admin_intrcnt;
6465 #if IFLIB_DEBUG
6466 	/* use only 1 qset in debug mode */
6467 	queuemsgs = min(msgs - admincnt, 1);
6468 #else
6469 	queuemsgs = msgs - admincnt;
6470 #endif
6471 #ifdef RSS
6472 	queues = imin(queuemsgs, rss_getnumbuckets());
6473 #else
6474 	queues = queuemsgs;
6475 #endif
6476 	queues = imin(CPU_COUNT(&ctx->ifc_cpus), queues);
6477 	if (bootverbose)
6478 		device_printf(dev,
6479 		    "intr CPUs: %d queue msgs: %d admincnt: %d\n",
6480 		    CPU_COUNT(&ctx->ifc_cpus), queuemsgs, admincnt);
6481 #ifdef  RSS
6482 	/* If we're doing RSS, clamp at the number of RSS buckets */
6483 	if (queues > rss_getnumbuckets())
6484 		queues = rss_getnumbuckets();
6485 #endif
6486 	if (iflib_num_rx_queues > 0 && iflib_num_rx_queues < queuemsgs - admincnt)
6487 		rx_queues = iflib_num_rx_queues;
6488 	else
6489 		rx_queues = queues;
6490 
6491 	if (rx_queues > scctx->isc_nrxqsets)
6492 		rx_queues = scctx->isc_nrxqsets;
6493 
6494 	/*
6495 	 * We want this to be all logical CPUs by default
6496 	 */
6497 	if (iflib_num_tx_queues > 0 && iflib_num_tx_queues < queues)
6498 		tx_queues = iflib_num_tx_queues;
6499 	else
6500 		tx_queues = mp_ncpus;
6501 
6502 	if (tx_queues > scctx->isc_ntxqsets)
6503 		tx_queues = scctx->isc_ntxqsets;
6504 
6505 	if (ctx->ifc_sysctl_qs_eq_override == 0) {
6506 #ifdef INVARIANTS
6507 		if (tx_queues != rx_queues)
6508 			device_printf(dev,
6509 			    "queue equality override not set, capping rx_queues at %d and tx_queues at %d\n",
6510 			    min(rx_queues, tx_queues), min(rx_queues, tx_queues));
6511 #endif
6512 		tx_queues = min(rx_queues, tx_queues);
6513 		rx_queues = min(rx_queues, tx_queues);
6514 	}
6515 
6516 	vectors = rx_queues + admincnt;
6517 	if (msgs < vectors) {
6518 		device_printf(dev,
6519 		    "insufficient number of MSI-X vectors "
6520 		    "(supported %d, need %d)\n", msgs, vectors);
6521 		goto msi;
6522 	}
6523 
6524 	device_printf(dev, "Using %d RX queues %d TX queues\n", rx_queues,
6525 	    tx_queues);
6526 	msgs = vectors;
6527 	if ((err = pci_alloc_msix(dev, &vectors)) == 0) {
6528 		if (vectors != msgs) {
6529 			device_printf(dev,
6530 			    "Unable to allocate sufficient MSI-X vectors "
6531 			    "(got %d, need %d)\n", vectors, msgs);
6532 			pci_release_msi(dev);
6533 			if (bar != -1) {
6534 				bus_release_resource(dev, SYS_RES_MEMORY, bar,
6535 				    ctx->ifc_msix_mem);
6536 				ctx->ifc_msix_mem = NULL;
6537 			}
6538 			goto msi;
6539 		}
6540 		device_printf(dev, "Using MSI-X interrupts with %d vectors\n",
6541 		    vectors);
6542 		scctx->isc_vectors = vectors;
6543 		scctx->isc_nrxqsets = rx_queues;
6544 		scctx->isc_ntxqsets = tx_queues;
6545 		scctx->isc_intr = IFLIB_INTR_MSIX;
6546 
6547 		return (vectors);
6548 	} else {
6549 		device_printf(dev,
6550 		    "failed to allocate %d MSI-X vectors, err: %d\n", vectors,
6551 		    err);
6552 		if (bar != -1) {
6553 			bus_release_resource(dev, SYS_RES_MEMORY, bar,
6554 			    ctx->ifc_msix_mem);
6555 			ctx->ifc_msix_mem = NULL;
6556 		}
6557 	}
6558 
6559 msi:
6560 	vectors = pci_msi_count(dev);
6561 	scctx->isc_nrxqsets = 1;
6562 	scctx->isc_ntxqsets = 1;
6563 	scctx->isc_vectors = vectors;
6564 	if (vectors == 1 && pci_alloc_msi(dev, &vectors) == 0) {
6565 		device_printf(dev,"Using an MSI interrupt\n");
6566 		scctx->isc_intr = IFLIB_INTR_MSI;
6567 	} else {
6568 		scctx->isc_vectors = 1;
6569 		device_printf(dev,"Using a Legacy interrupt\n");
6570 		scctx->isc_intr = IFLIB_INTR_LEGACY;
6571 	}
6572 
6573 	return (vectors);
6574 }
6575 
6576 static const char *ring_states[] = { "IDLE", "BUSY", "STALLED", "ABDICATED" };
6577 
6578 static int
6579 mp_ring_state_handler(SYSCTL_HANDLER_ARGS)
6580 {
6581 	int rc;
6582 	uint16_t *state = ((uint16_t *)oidp->oid_arg1);
6583 	struct sbuf *sb;
6584 	const char *ring_state = "UNKNOWN";
6585 
6586 	/* XXX needed ? */
6587 	rc = sysctl_wire_old_buffer(req, 0);
6588 	MPASS(rc == 0);
6589 	if (rc != 0)
6590 		return (rc);
6591 	sb = sbuf_new_for_sysctl(NULL, NULL, 80, req);
6592 	MPASS(sb != NULL);
6593 	if (sb == NULL)
6594 		return (ENOMEM);
6595 	if (state[3] <= 3)
6596 		ring_state = ring_states[state[3]];
6597 
6598 	sbuf_printf(sb, "pidx_head: %04hd pidx_tail: %04hd cidx: %04hd state: %s",
6599 		    state[0], state[1], state[2], ring_state);
6600 	rc = sbuf_finish(sb);
6601 	sbuf_delete(sb);
6602         return(rc);
6603 }
6604 
6605 enum iflib_ndesc_handler {
6606 	IFLIB_NTXD_HANDLER,
6607 	IFLIB_NRXD_HANDLER,
6608 };
6609 
6610 static int
6611 mp_ndesc_handler(SYSCTL_HANDLER_ARGS)
6612 {
6613 	if_ctx_t ctx = (void *)arg1;
6614 	enum iflib_ndesc_handler type = arg2;
6615 	char buf[256] = {0};
6616 	qidx_t *ndesc;
6617 	char *p, *next;
6618 	int nqs, rc, i;
6619 
6620 	nqs = 8;
6621 	switch(type) {
6622 	case IFLIB_NTXD_HANDLER:
6623 		ndesc = ctx->ifc_sysctl_ntxds;
6624 		if (ctx->ifc_sctx)
6625 			nqs = ctx->ifc_sctx->isc_ntxqs;
6626 		break;
6627 	case IFLIB_NRXD_HANDLER:
6628 		ndesc = ctx->ifc_sysctl_nrxds;
6629 		if (ctx->ifc_sctx)
6630 			nqs = ctx->ifc_sctx->isc_nrxqs;
6631 		break;
6632 	default:
6633 		printf("%s: unhandled type\n", __func__);
6634 		return (EINVAL);
6635 	}
6636 	if (nqs == 0)
6637 		nqs = 8;
6638 
6639 	for (i=0; i<8; i++) {
6640 		if (i >= nqs)
6641 			break;
6642 		if (i)
6643 			strcat(buf, ",");
6644 		sprintf(strchr(buf, 0), "%d", ndesc[i]);
6645 	}
6646 
6647 	rc = sysctl_handle_string(oidp, buf, sizeof(buf), req);
6648 	if (rc || req->newptr == NULL)
6649 		return rc;
6650 
6651 	for (i = 0, next = buf, p = strsep(&next, " ,"); i < 8 && p;
6652 	    i++, p = strsep(&next, " ,")) {
6653 		ndesc[i] = strtoul(p, NULL, 10);
6654 	}
6655 
6656 	return(rc);
6657 }
6658 
6659 #define NAME_BUFLEN 32
6660 static void
6661 iflib_add_device_sysctl_pre(if_ctx_t ctx)
6662 {
6663         device_t dev = iflib_get_dev(ctx);
6664 	struct sysctl_oid_list *child, *oid_list;
6665 	struct sysctl_ctx_list *ctx_list;
6666 	struct sysctl_oid *node;
6667 
6668 	ctx_list = device_get_sysctl_ctx(dev);
6669 	child = SYSCTL_CHILDREN(device_get_sysctl_tree(dev));
6670 	ctx->ifc_sysctl_node = node = SYSCTL_ADD_NODE(ctx_list, child, OID_AUTO, "iflib",
6671 	    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "IFLIB fields");
6672 	oid_list = SYSCTL_CHILDREN(node);
6673 
6674 	SYSCTL_ADD_CONST_STRING(ctx_list, oid_list, OID_AUTO, "driver_version",
6675 		       CTLFLAG_RD, ctx->ifc_sctx->isc_driver_version,
6676 		       "driver version");
6677 
6678 	SYSCTL_ADD_U16(ctx_list, oid_list, OID_AUTO, "override_ntxqs",
6679 		       CTLFLAG_RWTUN, &ctx->ifc_sysctl_ntxqs, 0,
6680 			"# of txqs to use, 0 => use default #");
6681 	SYSCTL_ADD_U16(ctx_list, oid_list, OID_AUTO, "override_nrxqs",
6682 		       CTLFLAG_RWTUN, &ctx->ifc_sysctl_nrxqs, 0,
6683 			"# of rxqs to use, 0 => use default #");
6684 	SYSCTL_ADD_U16(ctx_list, oid_list, OID_AUTO, "override_qs_enable",
6685 		       CTLFLAG_RWTUN, &ctx->ifc_sysctl_qs_eq_override, 0,
6686                        "permit #txq != #rxq");
6687 	SYSCTL_ADD_INT(ctx_list, oid_list, OID_AUTO, "disable_msix",
6688                       CTLFLAG_RWTUN, &ctx->ifc_softc_ctx.isc_disable_msix, 0,
6689                       "disable MSI-X (default 0)");
6690 	SYSCTL_ADD_U16(ctx_list, oid_list, OID_AUTO, "rx_budget",
6691 		       CTLFLAG_RWTUN, &ctx->ifc_sysctl_rx_budget, 0,
6692 		       "set the RX budget");
6693 	SYSCTL_ADD_U16(ctx_list, oid_list, OID_AUTO, "tx_abdicate",
6694 		       CTLFLAG_RWTUN, &ctx->ifc_sysctl_tx_abdicate, 0,
6695 		       "cause TX to abdicate instead of running to completion");
6696 	ctx->ifc_sysctl_core_offset = CORE_OFFSET_UNSPECIFIED;
6697 	SYSCTL_ADD_U16(ctx_list, oid_list, OID_AUTO, "core_offset",
6698 		       CTLFLAG_RDTUN, &ctx->ifc_sysctl_core_offset, 0,
6699 		       "offset to start using cores at");
6700 	SYSCTL_ADD_U8(ctx_list, oid_list, OID_AUTO, "separate_txrx",
6701 		       CTLFLAG_RDTUN, &ctx->ifc_sysctl_separate_txrx, 0,
6702 		       "use separate cores for TX and RX");
6703 
6704 	/* XXX change for per-queue sizes */
6705 	SYSCTL_ADD_PROC(ctx_list, oid_list, OID_AUTO, "override_ntxds",
6706 	    CTLTYPE_STRING | CTLFLAG_RWTUN | CTLFLAG_NEEDGIANT, ctx,
6707 	    IFLIB_NTXD_HANDLER, mp_ndesc_handler, "A",
6708 	    "list of # of TX descriptors to use, 0 = use default #");
6709 	SYSCTL_ADD_PROC(ctx_list, oid_list, OID_AUTO, "override_nrxds",
6710 	    CTLTYPE_STRING | CTLFLAG_RWTUN | CTLFLAG_NEEDGIANT, ctx,
6711 	    IFLIB_NRXD_HANDLER, mp_ndesc_handler, "A",
6712 	    "list of # of RX descriptors to use, 0 = use default #");
6713 }
6714 
6715 static void
6716 iflib_add_device_sysctl_post(if_ctx_t ctx)
6717 {
6718 	if_shared_ctx_t sctx = ctx->ifc_sctx;
6719 	if_softc_ctx_t scctx = &ctx->ifc_softc_ctx;
6720         device_t dev = iflib_get_dev(ctx);
6721 	struct sysctl_oid_list *child;
6722 	struct sysctl_ctx_list *ctx_list;
6723 	iflib_fl_t fl;
6724 	iflib_txq_t txq;
6725 	iflib_rxq_t rxq;
6726 	int i, j;
6727 	char namebuf[NAME_BUFLEN];
6728 	char *qfmt;
6729 	struct sysctl_oid *queue_node, *fl_node, *node;
6730 	struct sysctl_oid_list *queue_list, *fl_list;
6731 	ctx_list = device_get_sysctl_ctx(dev);
6732 
6733 	node = ctx->ifc_sysctl_node;
6734 	child = SYSCTL_CHILDREN(node);
6735 
6736 	if (scctx->isc_ntxqsets > 100)
6737 		qfmt = "txq%03d";
6738 	else if (scctx->isc_ntxqsets > 10)
6739 		qfmt = "txq%02d";
6740 	else
6741 		qfmt = "txq%d";
6742 	for (i = 0, txq = ctx->ifc_txqs; i < scctx->isc_ntxqsets; i++, txq++) {
6743 		snprintf(namebuf, NAME_BUFLEN, qfmt, i);
6744 		queue_node = SYSCTL_ADD_NODE(ctx_list, child, OID_AUTO, namebuf,
6745 		    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "Queue Name");
6746 		queue_list = SYSCTL_CHILDREN(queue_node);
6747 #if MEMORY_LOGGING
6748 		SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "txq_dequeued",
6749 				CTLFLAG_RD,
6750 				&txq->ift_dequeued, "total mbufs freed");
6751 		SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "txq_enqueued",
6752 				CTLFLAG_RD,
6753 				&txq->ift_enqueued, "total mbufs enqueued");
6754 #endif
6755 		SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "mbuf_defrag",
6756 				   CTLFLAG_RD,
6757 				   &txq->ift_mbuf_defrag, "# of times m_defrag was called");
6758 		SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "m_pullups",
6759 				   CTLFLAG_RD,
6760 				   &txq->ift_pullups, "# of times m_pullup was called");
6761 		SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "mbuf_defrag_failed",
6762 				   CTLFLAG_RD,
6763 				   &txq->ift_mbuf_defrag_failed, "# of times m_defrag failed");
6764 		SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "no_desc_avail",
6765 				   CTLFLAG_RD,
6766 				   &txq->ift_no_desc_avail, "# of times no descriptors were available");
6767 		SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "tx_map_failed",
6768 				   CTLFLAG_RD,
6769 				   &txq->ift_map_failed, "# of times DMA map failed");
6770 		SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "txd_encap_efbig",
6771 				   CTLFLAG_RD,
6772 				   &txq->ift_txd_encap_efbig, "# of times txd_encap returned EFBIG");
6773 		SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "no_tx_dma_setup",
6774 				   CTLFLAG_RD,
6775 				   &txq->ift_no_tx_dma_setup, "# of times map failed for other than EFBIG");
6776 		SYSCTL_ADD_U16(ctx_list, queue_list, OID_AUTO, "txq_pidx",
6777 				   CTLFLAG_RD,
6778 				   &txq->ift_pidx, 1, "Producer Index");
6779 		SYSCTL_ADD_U16(ctx_list, queue_list, OID_AUTO, "txq_cidx",
6780 				   CTLFLAG_RD,
6781 				   &txq->ift_cidx, 1, "Consumer Index");
6782 		SYSCTL_ADD_U16(ctx_list, queue_list, OID_AUTO, "txq_cidx_processed",
6783 				   CTLFLAG_RD,
6784 				   &txq->ift_cidx_processed, 1, "Consumer Index seen by credit update");
6785 		SYSCTL_ADD_U16(ctx_list, queue_list, OID_AUTO, "txq_in_use",
6786 				   CTLFLAG_RD,
6787 				   &txq->ift_in_use, 1, "descriptors in use");
6788 		SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "txq_processed",
6789 				   CTLFLAG_RD,
6790 				   &txq->ift_processed, "descriptors procesed for clean");
6791 		SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "txq_cleaned",
6792 				   CTLFLAG_RD,
6793 				   &txq->ift_cleaned, "total cleaned");
6794 		SYSCTL_ADD_PROC(ctx_list, queue_list, OID_AUTO, "ring_state",
6795 		    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT,
6796 		    __DEVOLATILE(uint64_t *, &txq->ift_br->state), 0,
6797 		    mp_ring_state_handler, "A", "soft ring state");
6798 		SYSCTL_ADD_COUNTER_U64(ctx_list, queue_list, OID_AUTO, "r_enqueues",
6799 				       CTLFLAG_RD, &txq->ift_br->enqueues,
6800 				       "# of enqueues to the mp_ring for this queue");
6801 		SYSCTL_ADD_COUNTER_U64(ctx_list, queue_list, OID_AUTO, "r_drops",
6802 				       CTLFLAG_RD, &txq->ift_br->drops,
6803 				       "# of drops in the mp_ring for this queue");
6804 		SYSCTL_ADD_COUNTER_U64(ctx_list, queue_list, OID_AUTO, "r_starts",
6805 				       CTLFLAG_RD, &txq->ift_br->starts,
6806 				       "# of normal consumer starts in the mp_ring for this queue");
6807 		SYSCTL_ADD_COUNTER_U64(ctx_list, queue_list, OID_AUTO, "r_stalls",
6808 				       CTLFLAG_RD, &txq->ift_br->stalls,
6809 					       "# of consumer stalls in the mp_ring for this queue");
6810 		SYSCTL_ADD_COUNTER_U64(ctx_list, queue_list, OID_AUTO, "r_restarts",
6811 			       CTLFLAG_RD, &txq->ift_br->restarts,
6812 				       "# of consumer restarts in the mp_ring for this queue");
6813 		SYSCTL_ADD_COUNTER_U64(ctx_list, queue_list, OID_AUTO, "r_abdications",
6814 				       CTLFLAG_RD, &txq->ift_br->abdications,
6815 				       "# of consumer abdications in the mp_ring for this queue");
6816 	}
6817 
6818 	if (scctx->isc_nrxqsets > 100)
6819 		qfmt = "rxq%03d";
6820 	else if (scctx->isc_nrxqsets > 10)
6821 		qfmt = "rxq%02d";
6822 	else
6823 		qfmt = "rxq%d";
6824 	for (i = 0, rxq = ctx->ifc_rxqs; i < scctx->isc_nrxqsets; i++, rxq++) {
6825 		snprintf(namebuf, NAME_BUFLEN, qfmt, i);
6826 		queue_node = SYSCTL_ADD_NODE(ctx_list, child, OID_AUTO, namebuf,
6827 		    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "Queue Name");
6828 		queue_list = SYSCTL_CHILDREN(queue_node);
6829 		if (sctx->isc_flags & IFLIB_HAS_RXCQ) {
6830 			SYSCTL_ADD_U16(ctx_list, queue_list, OID_AUTO, "rxq_cq_cidx",
6831 				       CTLFLAG_RD,
6832 				       &rxq->ifr_cq_cidx, 1, "Consumer Index");
6833 		}
6834 
6835 		for (j = 0, fl = rxq->ifr_fl; j < rxq->ifr_nfl; j++, fl++) {
6836 			snprintf(namebuf, NAME_BUFLEN, "rxq_fl%d", j);
6837 			fl_node = SYSCTL_ADD_NODE(ctx_list, queue_list, OID_AUTO, namebuf,
6838 			    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "freelist Name");
6839 			fl_list = SYSCTL_CHILDREN(fl_node);
6840 			SYSCTL_ADD_U16(ctx_list, fl_list, OID_AUTO, "pidx",
6841 				       CTLFLAG_RD,
6842 				       &fl->ifl_pidx, 1, "Producer Index");
6843 			SYSCTL_ADD_U16(ctx_list, fl_list, OID_AUTO, "cidx",
6844 				       CTLFLAG_RD,
6845 				       &fl->ifl_cidx, 1, "Consumer Index");
6846 			SYSCTL_ADD_U16(ctx_list, fl_list, OID_AUTO, "credits",
6847 				       CTLFLAG_RD,
6848 				       &fl->ifl_credits, 1, "credits available");
6849 			SYSCTL_ADD_U16(ctx_list, fl_list, OID_AUTO, "buf_size",
6850 				       CTLFLAG_RD,
6851 				       &fl->ifl_buf_size, 1, "buffer size");
6852 #if MEMORY_LOGGING
6853 			SYSCTL_ADD_QUAD(ctx_list, fl_list, OID_AUTO, "fl_m_enqueued",
6854 					CTLFLAG_RD,
6855 					&fl->ifl_m_enqueued, "mbufs allocated");
6856 			SYSCTL_ADD_QUAD(ctx_list, fl_list, OID_AUTO, "fl_m_dequeued",
6857 					CTLFLAG_RD,
6858 					&fl->ifl_m_dequeued, "mbufs freed");
6859 			SYSCTL_ADD_QUAD(ctx_list, fl_list, OID_AUTO, "fl_cl_enqueued",
6860 					CTLFLAG_RD,
6861 					&fl->ifl_cl_enqueued, "clusters allocated");
6862 			SYSCTL_ADD_QUAD(ctx_list, fl_list, OID_AUTO, "fl_cl_dequeued",
6863 					CTLFLAG_RD,
6864 					&fl->ifl_cl_dequeued, "clusters freed");
6865 #endif
6866 		}
6867 	}
6868 
6869 }
6870 
6871 void
6872 iflib_request_reset(if_ctx_t ctx)
6873 {
6874 
6875 	STATE_LOCK(ctx);
6876 	ctx->ifc_flags |= IFC_DO_RESET;
6877 	STATE_UNLOCK(ctx);
6878 }
6879 
6880 #ifndef __NO_STRICT_ALIGNMENT
6881 static struct mbuf *
6882 iflib_fixup_rx(struct mbuf *m)
6883 {
6884 	struct mbuf *n;
6885 
6886 	if (m->m_len <= (MCLBYTES - ETHER_HDR_LEN)) {
6887 		bcopy(m->m_data, m->m_data + ETHER_HDR_LEN, m->m_len);
6888 		m->m_data += ETHER_HDR_LEN;
6889 		n = m;
6890 	} else {
6891 		MGETHDR(n, M_NOWAIT, MT_DATA);
6892 		if (n == NULL) {
6893 			m_freem(m);
6894 			return (NULL);
6895 		}
6896 		bcopy(m->m_data, n->m_data, ETHER_HDR_LEN);
6897 		m->m_data += ETHER_HDR_LEN;
6898 		m->m_len -= ETHER_HDR_LEN;
6899 		n->m_len = ETHER_HDR_LEN;
6900 		M_MOVE_PKTHDR(n, m);
6901 		n->m_next = m;
6902 	}
6903 	return (n);
6904 }
6905 #endif
6906 
6907 #ifdef DEBUGNET
6908 static void
6909 iflib_debugnet_init(if_t ifp, int *nrxr, int *ncl, int *clsize)
6910 {
6911 	if_ctx_t ctx;
6912 
6913 	ctx = if_getsoftc(ifp);
6914 	CTX_LOCK(ctx);
6915 	*nrxr = NRXQSETS(ctx);
6916 	*ncl = ctx->ifc_rxqs[0].ifr_fl->ifl_size;
6917 	*clsize = ctx->ifc_rxqs[0].ifr_fl->ifl_buf_size;
6918 	CTX_UNLOCK(ctx);
6919 }
6920 
6921 static void
6922 iflib_debugnet_event(if_t ifp, enum debugnet_ev event)
6923 {
6924 	if_ctx_t ctx;
6925 	if_softc_ctx_t scctx;
6926 	iflib_fl_t fl;
6927 	iflib_rxq_t rxq;
6928 	int i, j;
6929 
6930 	ctx = if_getsoftc(ifp);
6931 	scctx = &ctx->ifc_softc_ctx;
6932 
6933 	switch (event) {
6934 	case DEBUGNET_START:
6935 		for (i = 0; i < scctx->isc_nrxqsets; i++) {
6936 			rxq = &ctx->ifc_rxqs[i];
6937 			for (j = 0; j < rxq->ifr_nfl; j++) {
6938 				fl = rxq->ifr_fl;
6939 				fl->ifl_zone = m_getzone(fl->ifl_buf_size);
6940 			}
6941 		}
6942 		iflib_no_tx_batch = 1;
6943 		break;
6944 	default:
6945 		break;
6946 	}
6947 }
6948 
6949 static int
6950 iflib_debugnet_transmit(if_t ifp, struct mbuf *m)
6951 {
6952 	if_ctx_t ctx;
6953 	iflib_txq_t txq;
6954 	int error;
6955 
6956 	ctx = if_getsoftc(ifp);
6957 	if ((if_getdrvflags(ifp) & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
6958 	    IFF_DRV_RUNNING)
6959 		return (EBUSY);
6960 
6961 	txq = &ctx->ifc_txqs[0];
6962 	error = iflib_encap(txq, &m);
6963 	if (error == 0)
6964 		(void)iflib_txd_db_check(txq, true);
6965 	return (error);
6966 }
6967 
6968 static int
6969 iflib_debugnet_poll(if_t ifp, int count)
6970 {
6971 	struct epoch_tracker et;
6972 	if_ctx_t ctx;
6973 	if_softc_ctx_t scctx;
6974 	iflib_txq_t txq;
6975 	int i;
6976 
6977 	ctx = if_getsoftc(ifp);
6978 	scctx = &ctx->ifc_softc_ctx;
6979 
6980 	if ((if_getdrvflags(ifp) & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
6981 	    IFF_DRV_RUNNING)
6982 		return (EBUSY);
6983 
6984 	txq = &ctx->ifc_txqs[0];
6985 	(void)iflib_completed_tx_reclaim(txq, RECLAIM_THRESH(ctx));
6986 
6987 	NET_EPOCH_ENTER(et);
6988 	for (i = 0; i < scctx->isc_nrxqsets; i++)
6989 		(void)iflib_rxeof(&ctx->ifc_rxqs[i], 16 /* XXX */);
6990 	NET_EPOCH_EXIT(et);
6991 	return (0);
6992 }
6993 #endif /* DEBUGNET */
6994