xref: /freebsd/sys/net/iflib.c (revision 094fc1ed0f2627525c7b0342efcbad5be7a8546a)
1 /*-
2  * Copyright (c) 2014-2017, Matthew Macy <mmacy@nextbsd.org>
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 
35 #include <sys/param.h>
36 #include <sys/types.h>
37 #include <sys/bus.h>
38 #include <sys/eventhandler.h>
39 #include <sys/sockio.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/sysctl.h>
50 #include <sys/syslog.h>
51 #include <sys/taskqueue.h>
52 #include <sys/limits.h>
53 
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 
63 #include <netinet/in.h>
64 #include <netinet/in_pcb.h>
65 #include <netinet/tcp_lro.h>
66 #include <netinet/in_systm.h>
67 #include <netinet/if_ether.h>
68 #include <netinet/ip.h>
69 #include <netinet/ip6.h>
70 #include <netinet/tcp.h>
71 
72 #include <machine/bus.h>
73 #include <machine/in_cksum.h>
74 
75 #include <vm/vm.h>
76 #include <vm/pmap.h>
77 
78 #include <dev/led/led.h>
79 #include <dev/pci/pcireg.h>
80 #include <dev/pci/pcivar.h>
81 #include <dev/pci/pci_private.h>
82 
83 #include <net/iflib.h>
84 
85 #include "ifdi_if.h"
86 
87 #if defined(__i386__) || defined(__amd64__)
88 #include <sys/memdesc.h>
89 #include <machine/bus.h>
90 #include <machine/md_var.h>
91 #include <machine/specialreg.h>
92 #include <x86/include/busdma_impl.h>
93 #include <x86/iommu/busdma_dmar.h>
94 #endif
95 
96 #include <sys/bitstring.h>
97 /*
98  * enable accounting of every mbuf as it comes in to and goes out of
99  * iflib's software descriptor references
100  */
101 #define MEMORY_LOGGING 0
102 /*
103  * Enable mbuf vectors for compressing long mbuf chains
104  */
105 
106 /*
107  * NB:
108  * - Prefetching in tx cleaning should perhaps be a tunable. The distance ahead
109  *   we prefetch needs to be determined by the time spent in m_free vis a vis
110  *   the cost of a prefetch. This will of course vary based on the workload:
111  *      - NFLX's m_free path is dominated by vm-based M_EXT manipulation which
112  *        is quite expensive, thus suggesting very little prefetch.
113  *      - small packet forwarding which is just returning a single mbuf to
114  *        UMA will typically be very fast vis a vis the cost of a memory
115  *        access.
116  */
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 static MALLOC_DEFINE(M_IFLIB, "iflib", "ifnet library");
130 
131 struct iflib_txq;
132 typedef struct iflib_txq *iflib_txq_t;
133 struct iflib_rxq;
134 typedef struct iflib_rxq *iflib_rxq_t;
135 struct iflib_fl;
136 typedef struct iflib_fl *iflib_fl_t;
137 
138 struct iflib_ctx;
139 
140 typedef struct iflib_filter_info {
141 	driver_filter_t *ifi_filter;
142 	void *ifi_filter_arg;
143 	struct grouptask *ifi_task;
144 	void *ifi_ctx;
145 } *iflib_filter_info_t;
146 
147 struct iflib_ctx {
148 	KOBJ_FIELDS;
149    /*
150    * Pointer to hardware driver's softc
151    */
152 	void *ifc_softc;
153 	device_t ifc_dev;
154 	if_t ifc_ifp;
155 
156 	cpuset_t ifc_cpus;
157 	if_shared_ctx_t ifc_sctx;
158 	struct if_softc_ctx ifc_softc_ctx;
159 
160 	struct mtx ifc_mtx;
161 
162 	uint16_t ifc_nhwtxqs;
163 	uint16_t ifc_nhwrxqs;
164 
165 	iflib_txq_t ifc_txqs;
166 	iflib_rxq_t ifc_rxqs;
167 	uint32_t ifc_if_flags;
168 	uint32_t ifc_flags;
169 	uint32_t ifc_max_fl_buf_size;
170 	int ifc_in_detach;
171 
172 	int ifc_link_state;
173 	int ifc_link_irq;
174 	int ifc_watchdog_events;
175 	struct cdev *ifc_led_dev;
176 	struct resource *ifc_msix_mem;
177 
178 	struct if_irq ifc_legacy_irq;
179 	struct grouptask ifc_admin_task;
180 	struct grouptask ifc_vflr_task;
181 	struct iflib_filter_info ifc_filter_info;
182 	struct ifmedia	ifc_media;
183 
184 	struct sysctl_oid *ifc_sysctl_node;
185 	uint16_t ifc_sysctl_ntxqs;
186 	uint16_t ifc_sysctl_nrxqs;
187 	uint16_t ifc_sysctl_qs_eq_override;
188 	uint16_t ifc_sysctl_rx_budget;
189 
190 	qidx_t ifc_sysctl_ntxds[8];
191 	qidx_t ifc_sysctl_nrxds[8];
192 	struct if_txrx ifc_txrx;
193 #define isc_txd_encap  ifc_txrx.ift_txd_encap
194 #define isc_txd_flush  ifc_txrx.ift_txd_flush
195 #define isc_txd_credits_update  ifc_txrx.ift_txd_credits_update
196 #define isc_rxd_available ifc_txrx.ift_rxd_available
197 #define isc_rxd_pkt_get ifc_txrx.ift_rxd_pkt_get
198 #define isc_rxd_refill ifc_txrx.ift_rxd_refill
199 #define isc_rxd_flush ifc_txrx.ift_rxd_flush
200 #define isc_rxd_refill ifc_txrx.ift_rxd_refill
201 #define isc_rxd_refill ifc_txrx.ift_rxd_refill
202 #define isc_legacy_intr ifc_txrx.ift_legacy_intr
203 	eventhandler_tag ifc_vlan_attach_event;
204 	eventhandler_tag ifc_vlan_detach_event;
205 	uint8_t ifc_mac[ETHER_ADDR_LEN];
206 	char ifc_mtx_name[16];
207 };
208 
209 
210 void *
211 iflib_get_softc(if_ctx_t ctx)
212 {
213 
214 	return (ctx->ifc_softc);
215 }
216 
217 device_t
218 iflib_get_dev(if_ctx_t ctx)
219 {
220 
221 	return (ctx->ifc_dev);
222 }
223 
224 if_t
225 iflib_get_ifp(if_ctx_t ctx)
226 {
227 
228 	return (ctx->ifc_ifp);
229 }
230 
231 struct ifmedia *
232 iflib_get_media(if_ctx_t ctx)
233 {
234 
235 	return (&ctx->ifc_media);
236 }
237 
238 void
239 iflib_set_mac(if_ctx_t ctx, uint8_t mac[ETHER_ADDR_LEN])
240 {
241 
242 	bcopy(mac, ctx->ifc_mac, ETHER_ADDR_LEN);
243 }
244 
245 if_softc_ctx_t
246 iflib_get_softc_ctx(if_ctx_t ctx)
247 {
248 
249 	return (&ctx->ifc_softc_ctx);
250 }
251 
252 if_shared_ctx_t
253 iflib_get_sctx(if_ctx_t ctx)
254 {
255 
256 	return (ctx->ifc_sctx);
257 }
258 
259 #define IP_ALIGNED(m) ((((uintptr_t)(m)->m_data) & 0x3) == 0x2)
260 #define CACHE_PTR_INCREMENT (CACHE_LINE_SIZE/sizeof(void*))
261 #define CACHE_PTR_NEXT(ptr) ((void *)(((uintptr_t)(ptr)+CACHE_LINE_SIZE-1) & (CACHE_LINE_SIZE-1)))
262 
263 #define LINK_ACTIVE(ctx) ((ctx)->ifc_link_state == LINK_STATE_UP)
264 #define CTX_IS_VF(ctx) ((ctx)->ifc_sctx->isc_flags & IFLIB_IS_VF)
265 
266 #define RX_SW_DESC_MAP_CREATED	(1 << 0)
267 #define TX_SW_DESC_MAP_CREATED	(1 << 1)
268 #define RX_SW_DESC_INUSE        (1 << 3)
269 #define TX_SW_DESC_MAPPED       (1 << 4)
270 
271 #define	M_TOOBIG		M_PROTO1
272 
273 typedef struct iflib_sw_rx_desc_array {
274 	bus_dmamap_t	*ifsd_map;         /* bus_dma maps for packet */
275 	struct mbuf	**ifsd_m;           /* pkthdr mbufs */
276 	caddr_t		*ifsd_cl;          /* direct cluster pointer for rx */
277 	uint8_t		*ifsd_flags;
278 } iflib_rxsd_array_t;
279 
280 typedef struct iflib_sw_tx_desc_array {
281 	bus_dmamap_t    *ifsd_map;         /* bus_dma maps for packet */
282 	struct mbuf    **ifsd_m;           /* pkthdr mbufs */
283 	uint8_t		*ifsd_flags;
284 } if_txsd_vec_t;
285 
286 
287 /* magic number that should be high enough for any hardware */
288 #define IFLIB_MAX_TX_SEGS		128
289 #define IFLIB_MAX_RX_SEGS		32
290 #define IFLIB_RX_COPY_THRESH		128
291 #define IFLIB_MAX_RX_REFRESH		32
292 /* The minimum descriptors per second before we start coalescing */
293 #define IFLIB_MIN_DESC_SEC		16384
294 #define IFLIB_DEFAULT_TX_UPDATE_FREQ	16
295 #define IFLIB_QUEUE_IDLE		0
296 #define IFLIB_QUEUE_HUNG		1
297 #define IFLIB_QUEUE_WORKING		2
298 /* maximum number of txqs that can share an rx interrupt */
299 #define IFLIB_MAX_TX_SHARED_INTR	4
300 
301 /* this should really scale with ring size - this is a fairly arbitrary value */
302 #define TX_BATCH_SIZE			32
303 
304 #define IFLIB_RESTART_BUDGET		8
305 
306 #define	IFC_LEGACY		0x001
307 #define	IFC_QFLUSH		0x002
308 #define	IFC_MULTISEG		0x004
309 #define	IFC_DMAR		0x008
310 #define	IFC_SC_ALLOCATED	0x010
311 #define	IFC_INIT_DONE		0x020
312 #define	IFC_PREFETCH		0x040
313 #define	IFC_DO_RESET		0x080
314 #define	IFC_CHECK_HUNG		0x100
315 
316 #define CSUM_OFFLOAD		(CSUM_IP_TSO|CSUM_IP6_TSO|CSUM_IP| \
317 				 CSUM_IP_UDP|CSUM_IP_TCP|CSUM_IP_SCTP| \
318 				 CSUM_IP6_UDP|CSUM_IP6_TCP|CSUM_IP6_SCTP)
319 struct iflib_txq {
320 	qidx_t		ift_in_use;
321 	qidx_t		ift_cidx;
322 	qidx_t		ift_cidx_processed;
323 	qidx_t		ift_pidx;
324 	uint8_t		ift_gen;
325 	uint8_t		ift_br_offset;
326 	uint16_t	ift_npending;
327 	uint16_t	ift_db_pending;
328 	uint16_t	ift_rs_pending;
329 	/* implicit pad */
330 	uint8_t		ift_txd_size[8];
331 	uint64_t	ift_processed;
332 	uint64_t	ift_cleaned;
333 	uint64_t	ift_cleaned_prev;
334 #if MEMORY_LOGGING
335 	uint64_t	ift_enqueued;
336 	uint64_t	ift_dequeued;
337 #endif
338 	uint64_t	ift_no_tx_dma_setup;
339 	uint64_t	ift_no_desc_avail;
340 	uint64_t	ift_mbuf_defrag_failed;
341 	uint64_t	ift_mbuf_defrag;
342 	uint64_t	ift_map_failed;
343 	uint64_t	ift_txd_encap_efbig;
344 	uint64_t	ift_pullups;
345 
346 	struct mtx	ift_mtx;
347 	struct mtx	ift_db_mtx;
348 
349 	/* constant values */
350 	if_ctx_t	ift_ctx;
351 	struct ifmp_ring        *ift_br;
352 	struct grouptask	ift_task;
353 	qidx_t		ift_size;
354 	uint16_t	ift_id;
355 	struct callout	ift_timer;
356 
357 	if_txsd_vec_t	ift_sds;
358 	uint8_t		ift_qstatus;
359 	uint8_t		ift_closed;
360 	uint8_t		ift_update_freq;
361 	struct iflib_filter_info ift_filter_info;
362 	bus_dma_tag_t		ift_desc_tag;
363 	bus_dma_tag_t		ift_tso_desc_tag;
364 	iflib_dma_info_t	ift_ifdi;
365 #define MTX_NAME_LEN 16
366 	char                    ift_mtx_name[MTX_NAME_LEN];
367 	char                    ift_db_mtx_name[MTX_NAME_LEN];
368 	bus_dma_segment_t	ift_segs[IFLIB_MAX_TX_SEGS]  __aligned(CACHE_LINE_SIZE);
369 #ifdef IFLIB_DIAGNOSTICS
370 	uint64_t ift_cpu_exec_count[256];
371 #endif
372 } __aligned(CACHE_LINE_SIZE);
373 
374 struct iflib_fl {
375 	qidx_t		ifl_cidx;
376 	qidx_t		ifl_pidx;
377 	qidx_t		ifl_credits;
378 	uint8_t		ifl_gen;
379 	uint8_t		ifl_rxd_size;
380 #if MEMORY_LOGGING
381 	uint64_t	ifl_m_enqueued;
382 	uint64_t	ifl_m_dequeued;
383 	uint64_t	ifl_cl_enqueued;
384 	uint64_t	ifl_cl_dequeued;
385 #endif
386 	/* implicit pad */
387 
388 	bitstr_t 	*ifl_rx_bitmap;
389 	qidx_t		ifl_fragidx;
390 	/* constant */
391 	qidx_t		ifl_size;
392 	uint16_t	ifl_buf_size;
393 	uint16_t	ifl_cltype;
394 	uma_zone_t	ifl_zone;
395 	iflib_rxsd_array_t	ifl_sds;
396 	iflib_rxq_t	ifl_rxq;
397 	uint8_t		ifl_id;
398 	bus_dma_tag_t           ifl_desc_tag;
399 	iflib_dma_info_t	ifl_ifdi;
400 	uint64_t	ifl_bus_addrs[IFLIB_MAX_RX_REFRESH] __aligned(CACHE_LINE_SIZE);
401 	caddr_t		ifl_vm_addrs[IFLIB_MAX_RX_REFRESH];
402 	qidx_t	ifl_rxd_idxs[IFLIB_MAX_RX_REFRESH];
403 }  __aligned(CACHE_LINE_SIZE);
404 
405 static inline qidx_t
406 get_inuse(int size, qidx_t cidx, qidx_t pidx, uint8_t gen)
407 {
408 	qidx_t used;
409 
410 	if (pidx > cidx)
411 		used = pidx - cidx;
412 	else if (pidx < cidx)
413 		used = size - cidx + pidx;
414 	else if (gen == 0 && pidx == cidx)
415 		used = 0;
416 	else if (gen == 1 && pidx == cidx)
417 		used = size;
418 	else
419 		panic("bad state");
420 
421 	return (used);
422 }
423 
424 #define TXQ_AVAIL(txq) (txq->ift_size - get_inuse(txq->ift_size, txq->ift_cidx, txq->ift_pidx, txq->ift_gen))
425 
426 #define IDXDIFF(head, tail, wrap) \
427 	((head) >= (tail) ? (head) - (tail) : (wrap) - (tail) + (head))
428 
429 struct iflib_rxq {
430 	/* If there is a separate completion queue -
431 	 * these are the cq cidx and pidx. Otherwise
432 	 * these are unused.
433 	 */
434 	qidx_t		ifr_size;
435 	qidx_t		ifr_cq_cidx;
436 	qidx_t		ifr_cq_pidx;
437 	uint8_t		ifr_cq_gen;
438 	uint8_t		ifr_fl_offset;
439 
440 	if_ctx_t	ifr_ctx;
441 	iflib_fl_t	ifr_fl;
442 	uint64_t	ifr_rx_irq;
443 	uint16_t	ifr_id;
444 	uint8_t		ifr_lro_enabled;
445 	uint8_t		ifr_nfl;
446 	uint8_t		ifr_ntxqirq;
447 	uint8_t		ifr_txqid[IFLIB_MAX_TX_SHARED_INTR];
448 	struct lro_ctrl			ifr_lc;
449 	struct grouptask        ifr_task;
450 	struct iflib_filter_info ifr_filter_info;
451 	iflib_dma_info_t		ifr_ifdi;
452 
453 	/* dynamically allocate if any drivers need a value substantially larger than this */
454 	struct if_rxd_frag	ifr_frags[IFLIB_MAX_RX_SEGS] __aligned(CACHE_LINE_SIZE);
455 #ifdef IFLIB_DIAGNOSTICS
456 	uint64_t ifr_cpu_exec_count[256];
457 #endif
458 }  __aligned(CACHE_LINE_SIZE);
459 
460 typedef struct if_rxsd {
461 	caddr_t *ifsd_cl;
462 	struct mbuf **ifsd_m;
463 	iflib_fl_t ifsd_fl;
464 	qidx_t ifsd_cidx;
465 } *if_rxsd_t;
466 
467 /* multiple of word size */
468 #ifdef __LP64__
469 #define PKT_INFO_SIZE	6
470 #define RXD_INFO_SIZE	5
471 #define PKT_TYPE uint64_t
472 #else
473 #define PKT_INFO_SIZE	11
474 #define RXD_INFO_SIZE	8
475 #define PKT_TYPE uint32_t
476 #endif
477 #define PKT_LOOP_BOUND  ((PKT_INFO_SIZE/3)*3)
478 #define RXD_LOOP_BOUND  ((RXD_INFO_SIZE/4)*4)
479 
480 typedef struct if_pkt_info_pad {
481 	PKT_TYPE pkt_val[PKT_INFO_SIZE];
482 } *if_pkt_info_pad_t;
483 typedef struct if_rxd_info_pad {
484 	PKT_TYPE rxd_val[RXD_INFO_SIZE];
485 } *if_rxd_info_pad_t;
486 
487 CTASSERT(sizeof(struct if_pkt_info_pad) == sizeof(struct if_pkt_info));
488 CTASSERT(sizeof(struct if_rxd_info_pad) == sizeof(struct if_rxd_info));
489 
490 
491 static inline void
492 pkt_info_zero(if_pkt_info_t pi)
493 {
494 	if_pkt_info_pad_t pi_pad;
495 
496 	pi_pad = (if_pkt_info_pad_t)pi;
497 	pi_pad->pkt_val[0] = 0; pi_pad->pkt_val[1] = 0; pi_pad->pkt_val[2] = 0;
498 	pi_pad->pkt_val[3] = 0; pi_pad->pkt_val[4] = 0; pi_pad->pkt_val[5] = 0;
499 #ifndef __LP64__
500 	pi_pad->pkt_val[6] = 0; pi_pad->pkt_val[7] = 0; pi_pad->pkt_val[8] = 0;
501 	pi_pad->pkt_val[9] = 0; pi_pad->pkt_val[10] = 0;
502 #endif
503 }
504 
505 static inline void
506 rxd_info_zero(if_rxd_info_t ri)
507 {
508 	if_rxd_info_pad_t ri_pad;
509 	int i;
510 
511 	ri_pad = (if_rxd_info_pad_t)ri;
512 	for (i = 0; i < RXD_LOOP_BOUND; i += 4) {
513 		ri_pad->rxd_val[i] = 0;
514 		ri_pad->rxd_val[i+1] = 0;
515 		ri_pad->rxd_val[i+2] = 0;
516 		ri_pad->rxd_val[i+3] = 0;
517 	}
518 #ifdef __LP64__
519 	ri_pad->rxd_val[RXD_INFO_SIZE-1] = 0;
520 #endif
521 }
522 
523 /*
524  * Only allow a single packet to take up most 1/nth of the tx ring
525  */
526 #define MAX_SINGLE_PACKET_FRACTION 12
527 #define IF_BAD_DMA (bus_addr_t)-1
528 
529 #define CTX_ACTIVE(ctx) ((if_getdrvflags((ctx)->ifc_ifp) & IFF_DRV_RUNNING))
530 
531 #define CTX_LOCK_INIT(_sc, _name)  mtx_init(&(_sc)->ifc_mtx, _name, "iflib ctx lock", MTX_DEF)
532 
533 #define CTX_LOCK(ctx) mtx_lock(&(ctx)->ifc_mtx)
534 #define CTX_UNLOCK(ctx) mtx_unlock(&(ctx)->ifc_mtx)
535 #define CTX_LOCK_DESTROY(ctx) mtx_destroy(&(ctx)->ifc_mtx)
536 
537 
538 #define CALLOUT_LOCK(txq)	mtx_lock(&txq->ift_mtx)
539 #define CALLOUT_UNLOCK(txq) 	mtx_unlock(&txq->ift_mtx)
540 
541 
542 /* Our boot-time initialization hook */
543 static int	iflib_module_event_handler(module_t, int, void *);
544 
545 static moduledata_t iflib_moduledata = {
546 	"iflib",
547 	iflib_module_event_handler,
548 	NULL
549 };
550 
551 DECLARE_MODULE(iflib, iflib_moduledata, SI_SUB_INIT_IF, SI_ORDER_ANY);
552 MODULE_VERSION(iflib, 1);
553 
554 MODULE_DEPEND(iflib, pci, 1, 1, 1);
555 MODULE_DEPEND(iflib, ether, 1, 1, 1);
556 
557 TASKQGROUP_DEFINE(if_io_tqg, mp_ncpus, 1);
558 TASKQGROUP_DEFINE(if_config_tqg, 1, 1);
559 
560 #ifndef IFLIB_DEBUG_COUNTERS
561 #ifdef INVARIANTS
562 #define IFLIB_DEBUG_COUNTERS 1
563 #else
564 #define IFLIB_DEBUG_COUNTERS 0
565 #endif /* !INVARIANTS */
566 #endif
567 
568 static SYSCTL_NODE(_net, OID_AUTO, iflib, CTLFLAG_RD, 0,
569                    "iflib driver parameters");
570 
571 /*
572  * XXX need to ensure that this can't accidentally cause the head to be moved backwards
573  */
574 static int iflib_min_tx_latency = 0;
575 SYSCTL_INT(_net_iflib, OID_AUTO, min_tx_latency, CTLFLAG_RW,
576 		   &iflib_min_tx_latency, 0, "minimize transmit latency at the possible expense of throughput");
577 static int iflib_no_tx_batch = 0;
578 SYSCTL_INT(_net_iflib, OID_AUTO, no_tx_batch, CTLFLAG_RW,
579 		   &iflib_no_tx_batch, 0, "minimize transmit latency at the possible expense of throughput");
580 
581 
582 #if IFLIB_DEBUG_COUNTERS
583 
584 static int iflib_tx_seen;
585 static int iflib_tx_sent;
586 static int iflib_tx_encap;
587 static int iflib_rx_allocs;
588 static int iflib_fl_refills;
589 static int iflib_fl_refills_large;
590 static int iflib_tx_frees;
591 
592 SYSCTL_INT(_net_iflib, OID_AUTO, tx_seen, CTLFLAG_RD,
593 		   &iflib_tx_seen, 0, "# tx mbufs seen");
594 SYSCTL_INT(_net_iflib, OID_AUTO, tx_sent, CTLFLAG_RD,
595 		   &iflib_tx_sent, 0, "# tx mbufs sent");
596 SYSCTL_INT(_net_iflib, OID_AUTO, tx_encap, CTLFLAG_RD,
597 		   &iflib_tx_encap, 0, "# tx mbufs encapped");
598 SYSCTL_INT(_net_iflib, OID_AUTO, tx_frees, CTLFLAG_RD,
599 		   &iflib_tx_frees, 0, "# tx frees");
600 SYSCTL_INT(_net_iflib, OID_AUTO, rx_allocs, CTLFLAG_RD,
601 		   &iflib_rx_allocs, 0, "# rx allocations");
602 SYSCTL_INT(_net_iflib, OID_AUTO, fl_refills, CTLFLAG_RD,
603 		   &iflib_fl_refills, 0, "# refills");
604 SYSCTL_INT(_net_iflib, OID_AUTO, fl_refills_large, CTLFLAG_RD,
605 		   &iflib_fl_refills_large, 0, "# large refills");
606 
607 
608 static int iflib_txq_drain_flushing;
609 static int iflib_txq_drain_oactive;
610 static int iflib_txq_drain_notready;
611 static int iflib_txq_drain_encapfail;
612 
613 SYSCTL_INT(_net_iflib, OID_AUTO, txq_drain_flushing, CTLFLAG_RD,
614 		   &iflib_txq_drain_flushing, 0, "# drain flushes");
615 SYSCTL_INT(_net_iflib, OID_AUTO, txq_drain_oactive, CTLFLAG_RD,
616 		   &iflib_txq_drain_oactive, 0, "# drain oactives");
617 SYSCTL_INT(_net_iflib, OID_AUTO, txq_drain_notready, CTLFLAG_RD,
618 		   &iflib_txq_drain_notready, 0, "# drain notready");
619 SYSCTL_INT(_net_iflib, OID_AUTO, txq_drain_encapfail, CTLFLAG_RD,
620 		   &iflib_txq_drain_encapfail, 0, "# drain encap fails");
621 
622 
623 static int iflib_encap_load_mbuf_fail;
624 static int iflib_encap_txq_avail_fail;
625 static int iflib_encap_txd_encap_fail;
626 
627 SYSCTL_INT(_net_iflib, OID_AUTO, encap_load_mbuf_fail, CTLFLAG_RD,
628 		   &iflib_encap_load_mbuf_fail, 0, "# busdma load failures");
629 SYSCTL_INT(_net_iflib, OID_AUTO, encap_txq_avail_fail, CTLFLAG_RD,
630 		   &iflib_encap_txq_avail_fail, 0, "# txq avail failures");
631 SYSCTL_INT(_net_iflib, OID_AUTO, encap_txd_encap_fail, CTLFLAG_RD,
632 		   &iflib_encap_txd_encap_fail, 0, "# driver encap failures");
633 
634 static int iflib_task_fn_rxs;
635 static int iflib_rx_intr_enables;
636 static int iflib_fast_intrs;
637 static int iflib_intr_link;
638 static int iflib_intr_msix;
639 static int iflib_rx_unavail;
640 static int iflib_rx_ctx_inactive;
641 static int iflib_rx_zero_len;
642 static int iflib_rx_if_input;
643 static int iflib_rx_mbuf_null;
644 static int iflib_rxd_flush;
645 
646 static int iflib_verbose_debug;
647 
648 SYSCTL_INT(_net_iflib, OID_AUTO, intr_link, CTLFLAG_RD,
649 		   &iflib_intr_link, 0, "# intr link calls");
650 SYSCTL_INT(_net_iflib, OID_AUTO, intr_msix, CTLFLAG_RD,
651 		   &iflib_intr_msix, 0, "# intr msix calls");
652 SYSCTL_INT(_net_iflib, OID_AUTO, task_fn_rx, CTLFLAG_RD,
653 		   &iflib_task_fn_rxs, 0, "# task_fn_rx calls");
654 SYSCTL_INT(_net_iflib, OID_AUTO, rx_intr_enables, CTLFLAG_RD,
655 		   &iflib_rx_intr_enables, 0, "# rx intr enables");
656 SYSCTL_INT(_net_iflib, OID_AUTO, fast_intrs, CTLFLAG_RD,
657 		   &iflib_fast_intrs, 0, "# fast_intr calls");
658 SYSCTL_INT(_net_iflib, OID_AUTO, rx_unavail, CTLFLAG_RD,
659 		   &iflib_rx_unavail, 0, "# times rxeof called with no available data");
660 SYSCTL_INT(_net_iflib, OID_AUTO, rx_ctx_inactive, CTLFLAG_RD,
661 		   &iflib_rx_ctx_inactive, 0, "# times rxeof called with inactive context");
662 SYSCTL_INT(_net_iflib, OID_AUTO, rx_zero_len, CTLFLAG_RD,
663 		   &iflib_rx_zero_len, 0, "# times rxeof saw zero len mbuf");
664 SYSCTL_INT(_net_iflib, OID_AUTO, rx_if_input, CTLFLAG_RD,
665 		   &iflib_rx_if_input, 0, "# times rxeof called if_input");
666 SYSCTL_INT(_net_iflib, OID_AUTO, rx_mbuf_null, CTLFLAG_RD,
667 		   &iflib_rx_mbuf_null, 0, "# times rxeof got null mbuf");
668 SYSCTL_INT(_net_iflib, OID_AUTO, rxd_flush, CTLFLAG_RD,
669 	         &iflib_rxd_flush, 0, "# times rxd_flush called");
670 SYSCTL_INT(_net_iflib, OID_AUTO, verbose_debug, CTLFLAG_RW,
671 		   &iflib_verbose_debug, 0, "enable verbose debugging");
672 
673 #define DBG_COUNTER_INC(name) atomic_add_int(&(iflib_ ## name), 1)
674 static void
675 iflib_debug_reset(void)
676 {
677 	iflib_tx_seen = iflib_tx_sent = iflib_tx_encap = iflib_rx_allocs =
678 		iflib_fl_refills = iflib_fl_refills_large = iflib_tx_frees =
679 		iflib_txq_drain_flushing = iflib_txq_drain_oactive =
680 		iflib_txq_drain_notready = iflib_txq_drain_encapfail =
681 		iflib_encap_load_mbuf_fail = iflib_encap_txq_avail_fail =
682 		iflib_encap_txd_encap_fail = iflib_task_fn_rxs = iflib_rx_intr_enables =
683 		iflib_fast_intrs = iflib_intr_link = iflib_intr_msix = iflib_rx_unavail =
684 		iflib_rx_ctx_inactive = iflib_rx_zero_len = iflib_rx_if_input =
685 		iflib_rx_mbuf_null = iflib_rxd_flush = 0;
686 }
687 
688 #else
689 #define DBG_COUNTER_INC(name)
690 static void iflib_debug_reset(void) {}
691 #endif
692 
693 
694 
695 #define IFLIB_DEBUG 0
696 
697 static void iflib_tx_structures_free(if_ctx_t ctx);
698 static void iflib_rx_structures_free(if_ctx_t ctx);
699 static int iflib_queues_alloc(if_ctx_t ctx);
700 static int iflib_tx_credits_update(if_ctx_t ctx, iflib_txq_t txq);
701 static int iflib_rxd_avail(if_ctx_t ctx, iflib_rxq_t rxq, qidx_t cidx, qidx_t budget);
702 static int iflib_qset_structures_setup(if_ctx_t ctx);
703 static int iflib_msix_init(if_ctx_t ctx);
704 static int iflib_legacy_setup(if_ctx_t ctx, driver_filter_t filter, void *filterarg, int *rid, char *str);
705 static void iflib_txq_check_drain(iflib_txq_t txq, int budget);
706 static uint32_t iflib_txq_can_drain(struct ifmp_ring *);
707 static int iflib_register(if_ctx_t);
708 static void iflib_init_locked(if_ctx_t ctx);
709 static void iflib_add_device_sysctl_pre(if_ctx_t ctx);
710 static void iflib_add_device_sysctl_post(if_ctx_t ctx);
711 static void iflib_ifmp_purge(iflib_txq_t txq);
712 static void _iflib_pre_assert(if_softc_ctx_t scctx);
713 static void iflib_stop(if_ctx_t ctx);
714 static void iflib_if_init_locked(if_ctx_t ctx);
715 #ifndef __NO_STRICT_ALIGNMENT
716 static struct mbuf * iflib_fixup_rx(struct mbuf *m);
717 #endif
718 
719 #ifdef DEV_NETMAP
720 #include <sys/selinfo.h>
721 #include <net/netmap.h>
722 #include <dev/netmap/netmap_kern.h>
723 
724 MODULE_DEPEND(iflib, netmap, 1, 1, 1);
725 
726 /*
727  * device-specific sysctl variables:
728  *
729  * iflib_crcstrip: 0: keep CRC in rx frames (default), 1: strip it.
730  *	During regular operations the CRC is stripped, but on some
731  *	hardware reception of frames not multiple of 64 is slower,
732  *	so using crcstrip=0 helps in benchmarks.
733  *
734  * iflib_rx_miss, iflib_rx_miss_bufs:
735  *	count packets that might be missed due to lost interrupts.
736  */
737 SYSCTL_DECL(_dev_netmap);
738 /*
739  * The xl driver by default strips CRCs and we do not override it.
740  */
741 
742 int iflib_crcstrip = 1;
743 SYSCTL_INT(_dev_netmap, OID_AUTO, iflib_crcstrip,
744     CTLFLAG_RW, &iflib_crcstrip, 1, "strip CRC on rx frames");
745 
746 int iflib_rx_miss, iflib_rx_miss_bufs;
747 SYSCTL_INT(_dev_netmap, OID_AUTO, iflib_rx_miss,
748     CTLFLAG_RW, &iflib_rx_miss, 0, "potentially missed rx intr");
749 SYSCTL_INT(_dev_netmap, OID_AUTO, iflib_rx_miss_bufs,
750     CTLFLAG_RW, &iflib_rx_miss_bufs, 0, "potentially missed rx intr bufs");
751 
752 /*
753  * Register/unregister. We are already under netmap lock.
754  * Only called on the first register or the last unregister.
755  */
756 static int
757 iflib_netmap_register(struct netmap_adapter *na, int onoff)
758 {
759 	struct ifnet *ifp = na->ifp;
760 	if_ctx_t ctx = ifp->if_softc;
761 	int status;
762 
763 	CTX_LOCK(ctx);
764 	IFDI_INTR_DISABLE(ctx);
765 
766 	/* Tell the stack that the interface is no longer active */
767 	ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
768 
769 	if (!CTX_IS_VF(ctx))
770 		IFDI_CRCSTRIP_SET(ctx, onoff, iflib_crcstrip);
771 
772 	/* enable or disable flags and callbacks in na and ifp */
773 	if (onoff) {
774 		nm_set_native_flags(na);
775 	} else {
776 		nm_clear_native_flags(na);
777 	}
778 	iflib_stop(ctx);
779 	iflib_init_locked(ctx);
780 	IFDI_CRCSTRIP_SET(ctx, onoff, iflib_crcstrip); // XXX why twice ?
781 	status = ifp->if_drv_flags & IFF_DRV_RUNNING ? 0 : 1;
782 	if (status)
783 		nm_clear_native_flags(na);
784 	CTX_UNLOCK(ctx);
785 	return (status);
786 }
787 
788 /*
789  * Reconcile kernel and user view of the transmit ring.
790  *
791  * All information is in the kring.
792  * Userspace wants to send packets up to the one before kring->rhead,
793  * kernel knows kring->nr_hwcur is the first unsent packet.
794  *
795  * Here we push packets out (as many as possible), and possibly
796  * reclaim buffers from previously completed transmission.
797  *
798  * The caller (netmap) guarantees that there is only one instance
799  * running at any time. Any interference with other driver
800  * methods should be handled by the individual drivers.
801  */
802 static int
803 iflib_netmap_txsync(struct netmap_kring *kring, int flags)
804 {
805 	struct netmap_adapter *na = kring->na;
806 	struct ifnet *ifp = na->ifp;
807 	struct netmap_ring *ring = kring->ring;
808 	u_int nm_i;	/* index into the netmap ring */
809 	u_int nic_i;	/* index into the NIC ring */
810 	u_int n;
811 	u_int const lim = kring->nkr_num_slots - 1;
812 	u_int const head = kring->rhead;
813 	struct if_pkt_info pi;
814 
815 	/*
816 	 * interrupts on every tx packet are expensive so request
817 	 * them every half ring, or where NS_REPORT is set
818 	 */
819 	u_int report_frequency = kring->nkr_num_slots >> 1;
820 	/* device-specific */
821 	if_ctx_t ctx = ifp->if_softc;
822 	iflib_txq_t txq = &ctx->ifc_txqs[kring->ring_id];
823 
824 	if (txq->ift_sds.ifsd_map)
825 		bus_dmamap_sync(txq->ift_desc_tag, txq->ift_ifdi->idi_map,
826 				BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
827 
828 
829 	/*
830 	 * First part: process new packets to send.
831 	 * nm_i is the current index in the netmap ring,
832 	 * nic_i is the corresponding index in the NIC ring.
833 	 *
834 	 * If we have packets to send (nm_i != head)
835 	 * iterate over the netmap ring, fetch length and update
836 	 * the corresponding slot in the NIC ring. Some drivers also
837 	 * need to update the buffer's physical address in the NIC slot
838 	 * even NS_BUF_CHANGED is not set (PNMB computes the addresses).
839 	 *
840 	 * The netmap_reload_map() calls is especially expensive,
841 	 * even when (as in this case) the tag is 0, so do only
842 	 * when the buffer has actually changed.
843 	 *
844 	 * If possible do not set the report/intr bit on all slots,
845 	 * but only a few times per ring or when NS_REPORT is set.
846 	 *
847 	 * Finally, on 10G and faster drivers, it might be useful
848 	 * to prefetch the next slot and txr entry.
849 	 */
850 
851 	nm_i = kring->nr_hwcur;
852 	pkt_info_zero(&pi);
853 	pi.ipi_segs = txq->ift_segs;
854 	pi.ipi_qsidx = kring->ring_id;
855 	if (nm_i != head) {	/* we have new packets to send */
856 		nic_i = netmap_idx_k2n(kring, nm_i);
857 
858 		__builtin_prefetch(&ring->slot[nm_i]);
859 		__builtin_prefetch(&txq->ift_sds.ifsd_m[nic_i]);
860 		if (txq->ift_sds.ifsd_map)
861 			__builtin_prefetch(&txq->ift_sds.ifsd_map[nic_i]);
862 
863 		for (n = 0; nm_i != head; n++) {
864 			struct netmap_slot *slot = &ring->slot[nm_i];
865 			u_int len = slot->len;
866 			uint64_t paddr;
867 			void *addr = PNMB(na, slot, &paddr);
868 			int flags = (slot->flags & NS_REPORT ||
869 				nic_i == 0 || nic_i == report_frequency) ?
870 				IPI_TX_INTR : 0;
871 
872 			/* device-specific */
873 			pi.ipi_len = len;
874 			pi.ipi_segs[0].ds_addr = paddr;
875 			pi.ipi_segs[0].ds_len = len;
876 			pi.ipi_nsegs = 1;
877 			pi.ipi_ndescs = 0;
878 			pi.ipi_pidx = nic_i;
879 			pi.ipi_flags = flags;
880 
881 			/* Fill the slot in the NIC ring. */
882 			ctx->isc_txd_encap(ctx->ifc_softc, &pi);
883 
884 			/* prefetch for next round */
885 			__builtin_prefetch(&ring->slot[nm_i + 1]);
886 			__builtin_prefetch(&txq->ift_sds.ifsd_m[nic_i + 1]);
887 			if (txq->ift_sds.ifsd_map) {
888 				__builtin_prefetch(&txq->ift_sds.ifsd_map[nic_i + 1]);
889 
890 				NM_CHECK_ADDR_LEN(na, addr, len);
891 
892 				if (slot->flags & NS_BUF_CHANGED) {
893 					/* buffer has changed, reload map */
894 					netmap_reload_map(na, txq->ift_desc_tag, txq->ift_sds.ifsd_map[nic_i], addr);
895 				}
896 				/* make sure changes to the buffer are synced */
897 				bus_dmamap_sync(txq->ift_ifdi->idi_tag, txq->ift_sds.ifsd_map[nic_i],
898 						BUS_DMASYNC_PREWRITE);
899 			}
900 			slot->flags &= ~(NS_REPORT | NS_BUF_CHANGED);
901 			nm_i = nm_next(nm_i, lim);
902 			nic_i = nm_next(nic_i, lim);
903 		}
904 		kring->nr_hwcur = head;
905 
906 		/* synchronize the NIC ring */
907 		if (txq->ift_sds.ifsd_map)
908 			bus_dmamap_sync(txq->ift_desc_tag, txq->ift_ifdi->idi_map,
909 						BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
910 
911 		/* (re)start the tx unit up to slot nic_i (excluded) */
912 		ctx->isc_txd_flush(ctx->ifc_softc, txq->ift_id, nic_i);
913 	}
914 
915 	/*
916 	 * Second part: reclaim buffers for completed transmissions.
917 	 */
918 	if (iflib_tx_credits_update(ctx, txq)) {
919 		/* some tx completed, increment avail */
920 		nic_i = txq->ift_cidx_processed;
921 		kring->nr_hwtail = nm_prev(netmap_idx_n2k(kring, nic_i), lim);
922 	}
923 	return (0);
924 }
925 
926 /*
927  * Reconcile kernel and user view of the receive ring.
928  * Same as for the txsync, this routine must be efficient.
929  * The caller guarantees a single invocations, but races against
930  * the rest of the driver should be handled here.
931  *
932  * On call, kring->rhead is the first packet that userspace wants
933  * to keep, and kring->rcur is the wakeup point.
934  * The kernel has previously reported packets up to kring->rtail.
935  *
936  * If (flags & NAF_FORCE_READ) also check for incoming packets irrespective
937  * of whether or not we received an interrupt.
938  */
939 static int
940 iflib_netmap_rxsync(struct netmap_kring *kring, int flags)
941 {
942 	struct netmap_adapter *na = kring->na;
943 	struct netmap_ring *ring = kring->ring;
944 	uint32_t nm_i;	/* index into the netmap ring */
945 	uint32_t nic_i, nic_i_start;	/* index into the NIC ring */
946 	u_int i, n;
947 	u_int const lim = kring->nkr_num_slots - 1;
948 	u_int const head = kring->rhead;
949 	int force_update = (flags & NAF_FORCE_READ) || kring->nr_kflags & NKR_PENDINTR;
950 	struct if_rxd_info ri;
951 	struct if_rxd_update iru;
952 
953 	struct ifnet *ifp = na->ifp;
954 	if_ctx_t ctx = ifp->if_softc;
955 	iflib_rxq_t rxq = &ctx->ifc_rxqs[kring->ring_id];
956 	iflib_fl_t fl = rxq->ifr_fl;
957 	if (head > lim)
958 		return netmap_ring_reinit(kring);
959 
960 	/* XXX check sync modes */
961 	for (i = 0, fl = rxq->ifr_fl; i < rxq->ifr_nfl; i++, fl++) {
962 		if (fl->ifl_sds.ifsd_map == NULL)
963 			continue;
964 		bus_dmamap_sync(rxq->ifr_fl[i].ifl_desc_tag, fl->ifl_ifdi->idi_map,
965 				BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
966 	}
967 	/*
968 	 * First part: import newly received packets.
969 	 *
970 	 * nm_i is the index of the next free slot in the netmap ring,
971 	 * nic_i is the index of the next received packet in the NIC ring,
972 	 * and they may differ in case if_init() has been called while
973 	 * in netmap mode. For the receive ring we have
974 	 *
975 	 *	nic_i = rxr->next_check;
976 	 *	nm_i = kring->nr_hwtail (previous)
977 	 * and
978 	 *	nm_i == (nic_i + kring->nkr_hwofs) % ring_size
979 	 *
980 	 * rxr->next_check is set to 0 on a ring reinit
981 	 */
982 	if (netmap_no_pendintr || force_update) {
983 		int crclen = iflib_crcstrip ? 0 : 4;
984 		int error, avail;
985 		uint16_t slot_flags = kring->nkr_slot_flags;
986 
987 		for (fl = rxq->ifr_fl, i = 0; i < rxq->ifr_nfl; i++, fl++) {
988 			nic_i = fl->ifl_cidx;
989 			nm_i = netmap_idx_n2k(kring, nic_i);
990 			avail = iflib_rxd_avail(ctx, rxq, nic_i, USHRT_MAX);
991 			for (n = 0; avail > 0; n++, avail--) {
992 				rxd_info_zero(&ri);
993 				ri.iri_frags = rxq->ifr_frags;
994 				ri.iri_qsidx = kring->ring_id;
995 				ri.iri_ifp = ctx->ifc_ifp;
996 				ri.iri_cidx = nic_i;
997 
998 				error = ctx->isc_rxd_pkt_get(ctx->ifc_softc, &ri);
999 				ring->slot[nm_i].len = error ? 0 : ri.iri_len - crclen;
1000 				ring->slot[nm_i].flags = slot_flags;
1001 				if (fl->ifl_sds.ifsd_map)
1002 					bus_dmamap_sync(fl->ifl_ifdi->idi_tag,
1003 							fl->ifl_sds.ifsd_map[nic_i], BUS_DMASYNC_POSTREAD);
1004 				nm_i = nm_next(nm_i, lim);
1005 				nic_i = nm_next(nic_i, lim);
1006 			}
1007 			if (n) { /* update the state variables */
1008 				if (netmap_no_pendintr && !force_update) {
1009 					/* diagnostics */
1010 					iflib_rx_miss ++;
1011 					iflib_rx_miss_bufs += n;
1012 				}
1013 				fl->ifl_cidx = nic_i;
1014 				kring->nr_hwtail = nm_i;
1015 			}
1016 			kring->nr_kflags &= ~NKR_PENDINTR;
1017 		}
1018 	}
1019 	/*
1020 	 * Second part: skip past packets that userspace has released.
1021 	 * (kring->nr_hwcur to head excluded),
1022 	 * and make the buffers available for reception.
1023 	 * As usual nm_i is the index in the netmap ring,
1024 	 * nic_i is the index in the NIC ring, and
1025 	 * nm_i == (nic_i + kring->nkr_hwofs) % ring_size
1026 	 */
1027 	/* XXX not sure how this will work with multiple free lists */
1028 	nm_i = kring->nr_hwcur;
1029 	if (nm_i == head)
1030 		return (0);
1031 
1032 	iru.iru_paddrs = fl->ifl_bus_addrs;
1033 	iru.iru_vaddrs = &fl->ifl_vm_addrs[0];
1034 	iru.iru_idxs = fl->ifl_rxd_idxs;
1035 	iru.iru_qsidx = rxq->ifr_id;
1036 	iru.iru_buf_size = fl->ifl_buf_size;
1037 	iru.iru_flidx = fl->ifl_id;
1038 	nic_i_start = nic_i = netmap_idx_k2n(kring, nm_i);
1039 	for (i = 0; nm_i != head; i++) {
1040 		struct netmap_slot *slot = &ring->slot[nm_i];
1041 		void *addr = PNMB(na, slot, &fl->ifl_bus_addrs[i]);
1042 
1043 		if (addr == NETMAP_BUF_BASE(na)) /* bad buf */
1044 			goto ring_reset;
1045 
1046 		fl->ifl_vm_addrs[i] = addr;
1047 		if (fl->ifl_sds.ifsd_map && (slot->flags & NS_BUF_CHANGED)) {
1048 			/* buffer has changed, reload map */
1049 			netmap_reload_map(na, fl->ifl_ifdi->idi_tag, fl->ifl_sds.ifsd_map[nic_i], addr);
1050 		}
1051 		slot->flags &= ~NS_BUF_CHANGED;
1052 
1053 		nm_i = nm_next(nm_i, lim);
1054 		fl->ifl_rxd_idxs[i] = nic_i = nm_next(nic_i, lim);
1055 		if (nm_i != head && i < IFLIB_MAX_RX_REFRESH)
1056 			continue;
1057 
1058 		iru.iru_pidx = nic_i_start;
1059 		iru.iru_count = i;
1060 		i = 0;
1061 		ctx->isc_rxd_refill(ctx->ifc_softc, &iru);
1062 		if (fl->ifl_sds.ifsd_map == NULL) {
1063 			nic_i_start = nic_i;
1064 			continue;
1065 		}
1066 		nic_i = nic_i_start;
1067 		for (n = 0; n < iru.iru_count; n++) {
1068 			bus_dmamap_sync(fl->ifl_ifdi->idi_tag, fl->ifl_sds.ifsd_map[nic_i],
1069 					BUS_DMASYNC_PREREAD);
1070 			nic_i = nm_next(nic_i, lim);
1071 		}
1072 		nic_i_start = nic_i;
1073 	}
1074 	kring->nr_hwcur = head;
1075 
1076 	if (fl->ifl_sds.ifsd_map)
1077 		bus_dmamap_sync(fl->ifl_ifdi->idi_tag, fl->ifl_ifdi->idi_map,
1078 				BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
1079 	/*
1080 	 * IMPORTANT: we must leave one free slot in the ring,
1081 	 * so move nic_i back by one unit
1082 	 */
1083 	nic_i = nm_prev(nic_i, lim);
1084 	ctx->isc_rxd_flush(ctx->ifc_softc, rxq->ifr_id, fl->ifl_id, nic_i);
1085 	return 0;
1086 
1087 ring_reset:
1088 	return netmap_ring_reinit(kring);
1089 }
1090 
1091 static void
1092 iflib_netmap_intr(struct netmap_adapter *na, int onoff)
1093 {
1094 	struct ifnet *ifp = na->ifp;
1095 	if_ctx_t ctx = ifp->if_softc;
1096 
1097 	CTX_LOCK(ctx);
1098 	if (onoff) {
1099 		IFDI_INTR_ENABLE(ctx);
1100 	} else {
1101 		IFDI_INTR_DISABLE(ctx);
1102 	}
1103 	CTX_UNLOCK(ctx);
1104 }
1105 
1106 
1107 static int
1108 iflib_netmap_attach(if_ctx_t ctx)
1109 {
1110 	struct netmap_adapter na;
1111 	if_softc_ctx_t scctx = &ctx->ifc_softc_ctx;
1112 
1113 	bzero(&na, sizeof(na));
1114 
1115 	na.ifp = ctx->ifc_ifp;
1116 	na.na_flags = NAF_BDG_MAYSLEEP;
1117 	MPASS(ctx->ifc_softc_ctx.isc_ntxqsets);
1118 	MPASS(ctx->ifc_softc_ctx.isc_nrxqsets);
1119 
1120 	na.num_tx_desc = scctx->isc_ntxd[0];
1121 	na.num_rx_desc = scctx->isc_nrxd[0];
1122 	na.nm_txsync = iflib_netmap_txsync;
1123 	na.nm_rxsync = iflib_netmap_rxsync;
1124 	na.nm_register = iflib_netmap_register;
1125 	na.nm_intr = iflib_netmap_intr;
1126 	na.num_tx_rings = ctx->ifc_softc_ctx.isc_ntxqsets;
1127 	na.num_rx_rings = ctx->ifc_softc_ctx.isc_nrxqsets;
1128 	return (netmap_attach(&na));
1129 }
1130 
1131 static void
1132 iflib_netmap_txq_init(if_ctx_t ctx, iflib_txq_t txq)
1133 {
1134 	struct netmap_adapter *na = NA(ctx->ifc_ifp);
1135 	struct netmap_slot *slot;
1136 
1137 	slot = netmap_reset(na, NR_TX, txq->ift_id, 0);
1138 	if (slot == NULL)
1139 		return;
1140 	if (txq->ift_sds.ifsd_map == NULL)
1141 		return;
1142 
1143 	for (int i = 0; i < ctx->ifc_softc_ctx.isc_ntxd[0]; i++) {
1144 
1145 		/*
1146 		 * In netmap mode, set the map for the packet buffer.
1147 		 * NOTE: Some drivers (not this one) also need to set
1148 		 * the physical buffer address in the NIC ring.
1149 		 * netmap_idx_n2k() maps a nic index, i, into the corresponding
1150 		 * netmap slot index, si
1151 		 */
1152 		int si = netmap_idx_n2k(&na->tx_rings[txq->ift_id], i);
1153 		netmap_load_map(na, txq->ift_desc_tag, txq->ift_sds.ifsd_map[i], NMB(na, slot + si));
1154 	}
1155 }
1156 static void
1157 iflib_netmap_rxq_init(if_ctx_t ctx, iflib_rxq_t rxq)
1158 {
1159 	struct netmap_adapter *na = NA(ctx->ifc_ifp);
1160 	struct netmap_slot *slot;
1161 	struct if_rxd_update iru;
1162 	iflib_fl_t fl;
1163 	bus_dmamap_t *map;
1164 	int nrxd;
1165 	uint32_t i, j, pidx_start;
1166 
1167 	slot = netmap_reset(na, NR_RX, rxq->ifr_id, 0);
1168 	if (slot == NULL)
1169 		return;
1170 	fl = &rxq->ifr_fl[0];
1171 	map = fl->ifl_sds.ifsd_map;
1172 	nrxd = ctx->ifc_softc_ctx.isc_nrxd[0];
1173 	iru.iru_paddrs = fl->ifl_bus_addrs;
1174 	iru.iru_vaddrs = &fl->ifl_vm_addrs[0];
1175 	iru.iru_idxs = fl->ifl_rxd_idxs;
1176 	iru.iru_qsidx = rxq->ifr_id;
1177 	iru.iru_buf_size = rxq->ifr_fl[0].ifl_buf_size;
1178 	iru.iru_flidx = 0;
1179 
1180 	for (pidx_start = i = j = 0; i < nrxd; i++, j++) {
1181 		int sj = netmap_idx_n2k(&na->rx_rings[rxq->ifr_id], i);
1182 		void *addr;
1183 
1184 		fl->ifl_rxd_idxs[j] = i;
1185 		addr = fl->ifl_vm_addrs[j] = PNMB(na, slot + sj, &fl->ifl_bus_addrs[j]);
1186 		if (map) {
1187 			netmap_load_map(na, rxq->ifr_fl[0].ifl_ifdi->idi_tag, *map, addr);
1188 			map++;
1189 		}
1190 
1191 		if (j < IFLIB_MAX_RX_REFRESH && i < nrxd - 1)
1192 			continue;
1193 
1194 		iru.iru_pidx = pidx_start;
1195 		pidx_start = i;
1196 		iru.iru_count = j;
1197 		j = 0;
1198 		MPASS(pidx_start + j <= nrxd);
1199 		/* Update descriptors and the cached value */
1200 		ctx->isc_rxd_refill(ctx->ifc_softc, &iru);
1201 	}
1202 	/* preserve queue */
1203 	if (ctx->ifc_ifp->if_capenable & IFCAP_NETMAP) {
1204 		struct netmap_kring *kring = &na->rx_rings[rxq->ifr_id];
1205 		int t = na->num_rx_desc - 1 - nm_kr_rxspace(kring);
1206 		ctx->isc_rxd_flush(ctx->ifc_softc, rxq->ifr_id, 0 /* fl_id */, t);
1207 	} else
1208 		ctx->isc_rxd_flush(ctx->ifc_softc, rxq->ifr_id, 0 /* fl_id */, nrxd-1);
1209 }
1210 
1211 #define iflib_netmap_detach(ifp) netmap_detach(ifp)
1212 
1213 #else
1214 #define iflib_netmap_txq_init(ctx, txq)
1215 #define iflib_netmap_rxq_init(ctx, rxq)
1216 #define iflib_netmap_detach(ifp)
1217 
1218 #define iflib_netmap_attach(ctx) (0)
1219 #define netmap_rx_irq(ifp, qid, budget) (0)
1220 #define netmap_tx_irq(ifp, qid) do {} while (0)
1221 
1222 #endif
1223 
1224 #if defined(__i386__) || defined(__amd64__)
1225 static __inline void
1226 prefetch(void *x)
1227 {
1228 	__asm volatile("prefetcht0 %0" :: "m" (*(unsigned long *)x));
1229 }
1230 #else
1231 #define prefetch(x)
1232 #endif
1233 
1234 static void
1235 _iflib_dmamap_cb(void *arg, bus_dma_segment_t *segs, int nseg, int err)
1236 {
1237 	if (err)
1238 		return;
1239 	*(bus_addr_t *) arg = segs[0].ds_addr;
1240 }
1241 
1242 int
1243 iflib_dma_alloc(if_ctx_t ctx, int size, iflib_dma_info_t dma, int mapflags)
1244 {
1245 	int err;
1246 	if_shared_ctx_t sctx = ctx->ifc_sctx;
1247 	device_t dev = ctx->ifc_dev;
1248 
1249 	KASSERT(sctx->isc_q_align != 0, ("alignment value not initialized"));
1250 
1251 	err = bus_dma_tag_create(bus_get_dma_tag(dev), /* parent */
1252 				sctx->isc_q_align, 0,	/* alignment, bounds */
1253 				BUS_SPACE_MAXADDR,	/* lowaddr */
1254 				BUS_SPACE_MAXADDR,	/* highaddr */
1255 				NULL, NULL,		/* filter, filterarg */
1256 				size,			/* maxsize */
1257 				1,			/* nsegments */
1258 				size,			/* maxsegsize */
1259 				BUS_DMA_ALLOCNOW,	/* flags */
1260 				NULL,			/* lockfunc */
1261 				NULL,			/* lockarg */
1262 				&dma->idi_tag);
1263 	if (err) {
1264 		device_printf(dev,
1265 		    "%s: bus_dma_tag_create failed: %d\n",
1266 		    __func__, err);
1267 		goto fail_0;
1268 	}
1269 
1270 	err = bus_dmamem_alloc(dma->idi_tag, (void**) &dma->idi_vaddr,
1271 	    BUS_DMA_NOWAIT | BUS_DMA_COHERENT | BUS_DMA_ZERO, &dma->idi_map);
1272 	if (err) {
1273 		device_printf(dev,
1274 		    "%s: bus_dmamem_alloc(%ju) failed: %d\n",
1275 		    __func__, (uintmax_t)size, err);
1276 		goto fail_1;
1277 	}
1278 
1279 	dma->idi_paddr = IF_BAD_DMA;
1280 	err = bus_dmamap_load(dma->idi_tag, dma->idi_map, dma->idi_vaddr,
1281 	    size, _iflib_dmamap_cb, &dma->idi_paddr, mapflags | BUS_DMA_NOWAIT);
1282 	if (err || dma->idi_paddr == IF_BAD_DMA) {
1283 		device_printf(dev,
1284 		    "%s: bus_dmamap_load failed: %d\n",
1285 		    __func__, err);
1286 		goto fail_2;
1287 	}
1288 
1289 	dma->idi_size = size;
1290 	return (0);
1291 
1292 fail_2:
1293 	bus_dmamem_free(dma->idi_tag, dma->idi_vaddr, dma->idi_map);
1294 fail_1:
1295 	bus_dma_tag_destroy(dma->idi_tag);
1296 fail_0:
1297 	dma->idi_tag = NULL;
1298 
1299 	return (err);
1300 }
1301 
1302 int
1303 iflib_dma_alloc_multi(if_ctx_t ctx, int *sizes, iflib_dma_info_t *dmalist, int mapflags, int count)
1304 {
1305 	int i, err;
1306 	iflib_dma_info_t *dmaiter;
1307 
1308 	dmaiter = dmalist;
1309 	for (i = 0; i < count; i++, dmaiter++) {
1310 		if ((err = iflib_dma_alloc(ctx, sizes[i], *dmaiter, mapflags)) != 0)
1311 			break;
1312 	}
1313 	if (err)
1314 		iflib_dma_free_multi(dmalist, i);
1315 	return (err);
1316 }
1317 
1318 void
1319 iflib_dma_free(iflib_dma_info_t dma)
1320 {
1321 	if (dma->idi_tag == NULL)
1322 		return;
1323 	if (dma->idi_paddr != IF_BAD_DMA) {
1324 		bus_dmamap_sync(dma->idi_tag, dma->idi_map,
1325 		    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
1326 		bus_dmamap_unload(dma->idi_tag, dma->idi_map);
1327 		dma->idi_paddr = IF_BAD_DMA;
1328 	}
1329 	if (dma->idi_vaddr != NULL) {
1330 		bus_dmamem_free(dma->idi_tag, dma->idi_vaddr, dma->idi_map);
1331 		dma->idi_vaddr = NULL;
1332 	}
1333 	bus_dma_tag_destroy(dma->idi_tag);
1334 	dma->idi_tag = NULL;
1335 }
1336 
1337 void
1338 iflib_dma_free_multi(iflib_dma_info_t *dmalist, int count)
1339 {
1340 	int i;
1341 	iflib_dma_info_t *dmaiter = dmalist;
1342 
1343 	for (i = 0; i < count; i++, dmaiter++)
1344 		iflib_dma_free(*dmaiter);
1345 }
1346 
1347 #ifdef EARLY_AP_STARTUP
1348 static const int iflib_started = 1;
1349 #else
1350 /*
1351  * We used to abuse the smp_started flag to decide if the queues have been
1352  * fully initialized (by late taskqgroup_adjust() calls in a SYSINIT()).
1353  * That gave bad races, since the SYSINIT() runs strictly after smp_started
1354  * is set.  Run a SYSINIT() strictly after that to just set a usable
1355  * completion flag.
1356  */
1357 
1358 static int iflib_started;
1359 
1360 static void
1361 iflib_record_started(void *arg)
1362 {
1363 	iflib_started = 1;
1364 }
1365 
1366 SYSINIT(iflib_record_started, SI_SUB_SMP + 1, SI_ORDER_FIRST,
1367 	iflib_record_started, NULL);
1368 #endif
1369 
1370 static int
1371 iflib_fast_intr(void *arg)
1372 {
1373 	iflib_filter_info_t info = arg;
1374 	struct grouptask *gtask = info->ifi_task;
1375 	if (!iflib_started)
1376 		return (FILTER_HANDLED);
1377 
1378 	DBG_COUNTER_INC(fast_intrs);
1379 	if (info->ifi_filter != NULL && info->ifi_filter(info->ifi_filter_arg) == FILTER_HANDLED)
1380 		return (FILTER_HANDLED);
1381 
1382 	GROUPTASK_ENQUEUE(gtask);
1383 	return (FILTER_HANDLED);
1384 }
1385 
1386 static int
1387 iflib_fast_intr_rxtx(void *arg)
1388 {
1389 	iflib_filter_info_t info = arg;
1390 	struct grouptask *gtask = info->ifi_task;
1391 	iflib_rxq_t rxq = (iflib_rxq_t)info->ifi_ctx;
1392 	if_ctx_t ctx;
1393 	int i, cidx;
1394 
1395 	if (!iflib_started)
1396 		return (FILTER_HANDLED);
1397 
1398 	DBG_COUNTER_INC(fast_intrs);
1399 	if (info->ifi_filter != NULL && info->ifi_filter(info->ifi_filter_arg) == FILTER_HANDLED)
1400 		return (FILTER_HANDLED);
1401 
1402 	for (i = 0; i < rxq->ifr_ntxqirq; i++) {
1403 		qidx_t txqid = rxq->ifr_txqid[i];
1404 
1405 		ctx = rxq->ifr_ctx;
1406 
1407 		if (!ctx->isc_txd_credits_update(ctx->ifc_softc, txqid, false)) {
1408 			IFDI_TX_QUEUE_INTR_ENABLE(ctx, txqid);
1409 			continue;
1410 		}
1411 		GROUPTASK_ENQUEUE(&ctx->ifc_txqs[txqid].ift_task);
1412 	}
1413 	if (ctx->ifc_sctx->isc_flags & IFLIB_HAS_RXCQ)
1414 		cidx = rxq->ifr_cq_cidx;
1415 	else
1416 		cidx = rxq->ifr_fl[0].ifl_cidx;
1417 	if (iflib_rxd_avail(ctx, rxq, cidx, 1))
1418 		GROUPTASK_ENQUEUE(gtask);
1419 	else
1420 		IFDI_RX_QUEUE_INTR_ENABLE(ctx, rxq->ifr_id);
1421 	return (FILTER_HANDLED);
1422 }
1423 
1424 
1425 static int
1426 iflib_fast_intr_ctx(void *arg)
1427 {
1428 	iflib_filter_info_t info = arg;
1429 	struct grouptask *gtask = info->ifi_task;
1430 
1431 	if (!iflib_started)
1432 		return (FILTER_HANDLED);
1433 
1434 	DBG_COUNTER_INC(fast_intrs);
1435 	if (info->ifi_filter != NULL && info->ifi_filter(info->ifi_filter_arg) == FILTER_HANDLED)
1436 		return (FILTER_HANDLED);
1437 
1438 	GROUPTASK_ENQUEUE(gtask);
1439 	return (FILTER_HANDLED);
1440 }
1441 
1442 static int
1443 _iflib_irq_alloc(if_ctx_t ctx, if_irq_t irq, int rid,
1444 	driver_filter_t filter, driver_intr_t handler, void *arg,
1445 				 char *name)
1446 {
1447 	int rc, flags;
1448 	struct resource *res;
1449 	void *tag = NULL;
1450 	device_t dev = ctx->ifc_dev;
1451 
1452 	flags = RF_ACTIVE;
1453 	if (ctx->ifc_flags & IFC_LEGACY)
1454 		flags |= RF_SHAREABLE;
1455 	MPASS(rid < 512);
1456 	irq->ii_rid = rid;
1457 	res = bus_alloc_resource_any(dev, SYS_RES_IRQ, &irq->ii_rid, flags);
1458 	if (res == NULL) {
1459 		device_printf(dev,
1460 		    "failed to allocate IRQ for rid %d, name %s.\n", rid, name);
1461 		return (ENOMEM);
1462 	}
1463 	irq->ii_res = res;
1464 	KASSERT(filter == NULL || handler == NULL, ("filter and handler can't both be non-NULL"));
1465 	rc = bus_setup_intr(dev, res, INTR_MPSAFE | INTR_TYPE_NET,
1466 						filter, handler, arg, &tag);
1467 	if (rc != 0) {
1468 		device_printf(dev,
1469 		    "failed to setup interrupt for rid %d, name %s: %d\n",
1470 					  rid, name ? name : "unknown", rc);
1471 		return (rc);
1472 	} else if (name)
1473 		bus_describe_intr(dev, res, tag, "%s", name);
1474 
1475 	irq->ii_tag = tag;
1476 	return (0);
1477 }
1478 
1479 
1480 /*********************************************************************
1481  *
1482  *  Allocate memory for tx_buffer structures. The tx_buffer stores all
1483  *  the information needed to transmit a packet on the wire. This is
1484  *  called only once at attach, setup is done every reset.
1485  *
1486  **********************************************************************/
1487 
1488 static int
1489 iflib_txsd_alloc(iflib_txq_t txq)
1490 {
1491 	if_ctx_t ctx = txq->ift_ctx;
1492 	if_shared_ctx_t sctx = ctx->ifc_sctx;
1493 	if_softc_ctx_t scctx = &ctx->ifc_softc_ctx;
1494 	device_t dev = ctx->ifc_dev;
1495 	int err, nsegments, ntsosegments;
1496 
1497 	nsegments = scctx->isc_tx_nsegments;
1498 	ntsosegments = scctx->isc_tx_tso_segments_max;
1499 	MPASS(scctx->isc_ntxd[0] > 0);
1500 	MPASS(scctx->isc_ntxd[txq->ift_br_offset] > 0);
1501 	MPASS(nsegments > 0);
1502 	MPASS(ntsosegments > 0);
1503 	/*
1504 	 * Setup DMA descriptor areas.
1505 	 */
1506 	if ((err = bus_dma_tag_create(bus_get_dma_tag(dev),
1507 			       1, 0,			/* alignment, bounds */
1508 			       BUS_SPACE_MAXADDR,	/* lowaddr */
1509 			       BUS_SPACE_MAXADDR,	/* highaddr */
1510 			       NULL, NULL,		/* filter, filterarg */
1511 			       sctx->isc_tx_maxsize,		/* maxsize */
1512 			       nsegments,	/* nsegments */
1513 			       sctx->isc_tx_maxsegsize,	/* maxsegsize */
1514 			       0,			/* flags */
1515 			       NULL,			/* lockfunc */
1516 			       NULL,			/* lockfuncarg */
1517 			       &txq->ift_desc_tag))) {
1518 		device_printf(dev,"Unable to allocate TX DMA tag: %d\n", err);
1519 		device_printf(dev,"maxsize: %ju nsegments: %d maxsegsize: %ju\n",
1520 		    (uintmax_t)sctx->isc_tx_maxsize, nsegments, (uintmax_t)sctx->isc_tx_maxsegsize);
1521 		goto fail;
1522 	}
1523 	if ((err = bus_dma_tag_create(bus_get_dma_tag(dev),
1524 			       1, 0,			/* alignment, bounds */
1525 			       BUS_SPACE_MAXADDR,	/* lowaddr */
1526 			       BUS_SPACE_MAXADDR,	/* highaddr */
1527 			       NULL, NULL,		/* filter, filterarg */
1528 			       scctx->isc_tx_tso_size_max,		/* maxsize */
1529 			       ntsosegments,	/* nsegments */
1530 			       scctx->isc_tx_tso_segsize_max,	/* maxsegsize */
1531 			       0,			/* flags */
1532 			       NULL,			/* lockfunc */
1533 			       NULL,			/* lockfuncarg */
1534 			       &txq->ift_tso_desc_tag))) {
1535 		device_printf(dev,"Unable to allocate TX TSO DMA tag: %d\n", err);
1536 
1537 		goto fail;
1538 	}
1539 	if (!(txq->ift_sds.ifsd_flags =
1540 	    (uint8_t *) malloc(sizeof(uint8_t) *
1541 	    scctx->isc_ntxd[txq->ift_br_offset], M_IFLIB, M_NOWAIT | M_ZERO))) {
1542 		device_printf(dev, "Unable to allocate tx_buffer memory\n");
1543 		err = ENOMEM;
1544 		goto fail;
1545 	}
1546 	if (!(txq->ift_sds.ifsd_m =
1547 	    (struct mbuf **) malloc(sizeof(struct mbuf *) *
1548 	    scctx->isc_ntxd[txq->ift_br_offset], M_IFLIB, M_NOWAIT | M_ZERO))) {
1549 		device_printf(dev, "Unable to allocate tx_buffer memory\n");
1550 		err = ENOMEM;
1551 		goto fail;
1552 	}
1553 
1554         /* Create the descriptor buffer dma maps */
1555 #if defined(ACPI_DMAR) || (! (defined(__i386__) || defined(__amd64__)))
1556 	if ((ctx->ifc_flags & IFC_DMAR) == 0)
1557 		return (0);
1558 
1559 	if (!(txq->ift_sds.ifsd_map =
1560 	    (bus_dmamap_t *) malloc(sizeof(bus_dmamap_t) * scctx->isc_ntxd[txq->ift_br_offset], M_IFLIB, M_NOWAIT | M_ZERO))) {
1561 		device_printf(dev, "Unable to allocate tx_buffer map memory\n");
1562 		err = ENOMEM;
1563 		goto fail;
1564 	}
1565 
1566 	for (int i = 0; i < scctx->isc_ntxd[txq->ift_br_offset]; i++) {
1567 		err = bus_dmamap_create(txq->ift_desc_tag, 0, &txq->ift_sds.ifsd_map[i]);
1568 		if (err != 0) {
1569 			device_printf(dev, "Unable to create TX DMA map\n");
1570 			goto fail;
1571 		}
1572 	}
1573 #endif
1574 	return (0);
1575 fail:
1576 	/* We free all, it handles case where we are in the middle */
1577 	iflib_tx_structures_free(ctx);
1578 	return (err);
1579 }
1580 
1581 static void
1582 iflib_txsd_destroy(if_ctx_t ctx, iflib_txq_t txq, int i)
1583 {
1584 	bus_dmamap_t map;
1585 
1586 	map = NULL;
1587 	if (txq->ift_sds.ifsd_map != NULL)
1588 		map = txq->ift_sds.ifsd_map[i];
1589 	if (map != NULL) {
1590 		bus_dmamap_unload(txq->ift_desc_tag, map);
1591 		bus_dmamap_destroy(txq->ift_desc_tag, map);
1592 		txq->ift_sds.ifsd_map[i] = NULL;
1593 	}
1594 }
1595 
1596 static void
1597 iflib_txq_destroy(iflib_txq_t txq)
1598 {
1599 	if_ctx_t ctx = txq->ift_ctx;
1600 
1601 	for (int i = 0; i < txq->ift_size; i++)
1602 		iflib_txsd_destroy(ctx, txq, i);
1603 	if (txq->ift_sds.ifsd_map != NULL) {
1604 		free(txq->ift_sds.ifsd_map, M_IFLIB);
1605 		txq->ift_sds.ifsd_map = NULL;
1606 	}
1607 	if (txq->ift_sds.ifsd_m != NULL) {
1608 		free(txq->ift_sds.ifsd_m, M_IFLIB);
1609 		txq->ift_sds.ifsd_m = NULL;
1610 	}
1611 	if (txq->ift_sds.ifsd_flags != NULL) {
1612 		free(txq->ift_sds.ifsd_flags, M_IFLIB);
1613 		txq->ift_sds.ifsd_flags = NULL;
1614 	}
1615 	if (txq->ift_desc_tag != NULL) {
1616 		bus_dma_tag_destroy(txq->ift_desc_tag);
1617 		txq->ift_desc_tag = NULL;
1618 	}
1619 	if (txq->ift_tso_desc_tag != NULL) {
1620 		bus_dma_tag_destroy(txq->ift_tso_desc_tag);
1621 		txq->ift_tso_desc_tag = NULL;
1622 	}
1623 }
1624 
1625 static void
1626 iflib_txsd_free(if_ctx_t ctx, iflib_txq_t txq, int i)
1627 {
1628 	struct mbuf **mp;
1629 
1630 	mp = &txq->ift_sds.ifsd_m[i];
1631 	if (*mp == NULL)
1632 		return;
1633 
1634 	if (txq->ift_sds.ifsd_map != NULL) {
1635 		bus_dmamap_sync(txq->ift_desc_tag,
1636 				txq->ift_sds.ifsd_map[i],
1637 				BUS_DMASYNC_POSTWRITE);
1638 		bus_dmamap_unload(txq->ift_desc_tag,
1639 				  txq->ift_sds.ifsd_map[i]);
1640 	}
1641 	m_free(*mp);
1642 	DBG_COUNTER_INC(tx_frees);
1643 	*mp = NULL;
1644 }
1645 
1646 static int
1647 iflib_txq_setup(iflib_txq_t txq)
1648 {
1649 	if_ctx_t ctx = txq->ift_ctx;
1650 	if_softc_ctx_t scctx = &ctx->ifc_softc_ctx;
1651 	iflib_dma_info_t di;
1652 	int i;
1653 
1654 	/* Set number of descriptors available */
1655 	txq->ift_qstatus = IFLIB_QUEUE_IDLE;
1656 	/* XXX make configurable */
1657 	txq->ift_update_freq = IFLIB_DEFAULT_TX_UPDATE_FREQ;
1658 
1659 	/* Reset indices */
1660 	txq->ift_cidx_processed = 0;
1661 	txq->ift_pidx = txq->ift_cidx = txq->ift_npending = 0;
1662 	txq->ift_size = scctx->isc_ntxd[txq->ift_br_offset];
1663 
1664 	for (i = 0, di = txq->ift_ifdi; i < ctx->ifc_nhwtxqs; i++, di++)
1665 		bzero((void *)di->idi_vaddr, di->idi_size);
1666 
1667 	IFDI_TXQ_SETUP(ctx, txq->ift_id);
1668 	for (i = 0, di = txq->ift_ifdi; i < ctx->ifc_nhwtxqs; i++, di++)
1669 		bus_dmamap_sync(di->idi_tag, di->idi_map,
1670 						BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
1671 	return (0);
1672 }
1673 
1674 /*********************************************************************
1675  *
1676  *  Allocate memory for rx_buffer structures. Since we use one
1677  *  rx_buffer per received packet, the maximum number of rx_buffer's
1678  *  that we'll need is equal to the number of receive descriptors
1679  *  that we've allocated.
1680  *
1681  **********************************************************************/
1682 static int
1683 iflib_rxsd_alloc(iflib_rxq_t rxq)
1684 {
1685 	if_ctx_t ctx = rxq->ifr_ctx;
1686 	if_shared_ctx_t sctx = ctx->ifc_sctx;
1687 	if_softc_ctx_t scctx = &ctx->ifc_softc_ctx;
1688 	device_t dev = ctx->ifc_dev;
1689 	iflib_fl_t fl;
1690 	int			err;
1691 
1692 	MPASS(scctx->isc_nrxd[0] > 0);
1693 	MPASS(scctx->isc_nrxd[rxq->ifr_fl_offset] > 0);
1694 
1695 	fl = rxq->ifr_fl;
1696 	for (int i = 0; i <  rxq->ifr_nfl; i++, fl++) {
1697 		fl->ifl_size = scctx->isc_nrxd[rxq->ifr_fl_offset]; /* this isn't necessarily the same */
1698 		err = bus_dma_tag_create(bus_get_dma_tag(dev), /* parent */
1699 					 1, 0,			/* alignment, bounds */
1700 					 BUS_SPACE_MAXADDR,	/* lowaddr */
1701 					 BUS_SPACE_MAXADDR,	/* highaddr */
1702 					 NULL, NULL,		/* filter, filterarg */
1703 					 sctx->isc_rx_maxsize,	/* maxsize */
1704 					 sctx->isc_rx_nsegments,	/* nsegments */
1705 					 sctx->isc_rx_maxsegsize,	/* maxsegsize */
1706 					 0,			/* flags */
1707 					 NULL,			/* lockfunc */
1708 					 NULL,			/* lockarg */
1709 					 &fl->ifl_desc_tag);
1710 		if (err) {
1711 			device_printf(dev, "%s: bus_dma_tag_create failed %d\n",
1712 				__func__, err);
1713 			goto fail;
1714 		}
1715 		if (!(fl->ifl_sds.ifsd_flags =
1716 		      (uint8_t *) malloc(sizeof(uint8_t) *
1717 					 scctx->isc_nrxd[rxq->ifr_fl_offset], M_IFLIB, M_NOWAIT | M_ZERO))) {
1718 			device_printf(dev, "Unable to allocate tx_buffer memory\n");
1719 			err = ENOMEM;
1720 			goto fail;
1721 		}
1722 		if (!(fl->ifl_sds.ifsd_m =
1723 		      (struct mbuf **) malloc(sizeof(struct mbuf *) *
1724 					      scctx->isc_nrxd[rxq->ifr_fl_offset], M_IFLIB, M_NOWAIT | M_ZERO))) {
1725 			device_printf(dev, "Unable to allocate tx_buffer memory\n");
1726 			err = ENOMEM;
1727 			goto fail;
1728 		}
1729 		if (!(fl->ifl_sds.ifsd_cl =
1730 		      (caddr_t *) malloc(sizeof(caddr_t) *
1731 					      scctx->isc_nrxd[rxq->ifr_fl_offset], M_IFLIB, M_NOWAIT | M_ZERO))) {
1732 			device_printf(dev, "Unable to allocate tx_buffer memory\n");
1733 			err = ENOMEM;
1734 			goto fail;
1735 		}
1736 
1737 		/* Create the descriptor buffer dma maps */
1738 #if defined(ACPI_DMAR) || (! (defined(__i386__) || defined(__amd64__)))
1739 		if ((ctx->ifc_flags & IFC_DMAR) == 0)
1740 			continue;
1741 
1742 		if (!(fl->ifl_sds.ifsd_map =
1743 		      (bus_dmamap_t *) malloc(sizeof(bus_dmamap_t) * scctx->isc_nrxd[rxq->ifr_fl_offset], M_IFLIB, M_NOWAIT | M_ZERO))) {
1744 			device_printf(dev, "Unable to allocate tx_buffer map memory\n");
1745 			err = ENOMEM;
1746 			goto fail;
1747 		}
1748 
1749 		for (int i = 0; i < scctx->isc_nrxd[rxq->ifr_fl_offset]; i++) {
1750 			err = bus_dmamap_create(fl->ifl_desc_tag, 0, &fl->ifl_sds.ifsd_map[i]);
1751 			if (err != 0) {
1752 				device_printf(dev, "Unable to create RX buffer DMA map\n");
1753 				goto fail;
1754 			}
1755 		}
1756 #endif
1757 	}
1758 	return (0);
1759 
1760 fail:
1761 	iflib_rx_structures_free(ctx);
1762 	return (err);
1763 }
1764 
1765 
1766 /*
1767  * Internal service routines
1768  */
1769 
1770 struct rxq_refill_cb_arg {
1771 	int               error;
1772 	bus_dma_segment_t seg;
1773 	int               nseg;
1774 };
1775 
1776 static void
1777 _rxq_refill_cb(void *arg, bus_dma_segment_t *segs, int nseg, int error)
1778 {
1779 	struct rxq_refill_cb_arg *cb_arg = arg;
1780 
1781 	cb_arg->error = error;
1782 	cb_arg->seg = segs[0];
1783 	cb_arg->nseg = nseg;
1784 }
1785 
1786 
1787 #ifdef ACPI_DMAR
1788 #define IS_DMAR(ctx) (ctx->ifc_flags & IFC_DMAR)
1789 #else
1790 #define IS_DMAR(ctx) (0)
1791 #endif
1792 
1793 /**
1794  *	rxq_refill - refill an rxq  free-buffer list
1795  *	@ctx: the iflib context
1796  *	@rxq: the free-list to refill
1797  *	@n: the number of new buffers to allocate
1798  *
1799  *	(Re)populate an rxq free-buffer list with up to @n new packet buffers.
1800  *	The caller must assure that @n does not exceed the queue's capacity.
1801  */
1802 static void
1803 _iflib_fl_refill(if_ctx_t ctx, iflib_fl_t fl, int count)
1804 {
1805 	struct mbuf *m;
1806 	int idx, frag_idx = fl->ifl_fragidx;
1807         int pidx = fl->ifl_pidx;
1808 	caddr_t cl, *sd_cl;
1809 	struct mbuf **sd_m;
1810 	uint8_t *sd_flags;
1811 	struct if_rxd_update iru;
1812 	bus_dmamap_t *sd_map;
1813 	int n, i = 0;
1814 	uint64_t bus_addr;
1815 	int err;
1816 
1817 	sd_m = fl->ifl_sds.ifsd_m;
1818 	sd_map = fl->ifl_sds.ifsd_map;
1819 	sd_cl = fl->ifl_sds.ifsd_cl;
1820 	sd_flags = fl->ifl_sds.ifsd_flags;
1821 	idx = pidx;
1822 
1823 	n  = count;
1824 	MPASS(n > 0);
1825 	MPASS(fl->ifl_credits + n <= fl->ifl_size);
1826 
1827 	if (pidx < fl->ifl_cidx)
1828 		MPASS(pidx + n <= fl->ifl_cidx);
1829 	if (pidx == fl->ifl_cidx && (fl->ifl_credits < fl->ifl_size))
1830 		MPASS(fl->ifl_gen == 0);
1831 	if (pidx > fl->ifl_cidx)
1832 		MPASS(n <= fl->ifl_size - pidx + fl->ifl_cidx);
1833 
1834 	DBG_COUNTER_INC(fl_refills);
1835 	if (n > 8)
1836 		DBG_COUNTER_INC(fl_refills_large);
1837 	iru.iru_paddrs = fl->ifl_bus_addrs;
1838 	iru.iru_vaddrs = &fl->ifl_vm_addrs[0];
1839 	iru.iru_idxs = fl->ifl_rxd_idxs;
1840 	iru.iru_qsidx = fl->ifl_rxq->ifr_id;
1841 	iru.iru_buf_size = fl->ifl_buf_size;
1842 	iru.iru_flidx = fl->ifl_id;
1843 	while (n--) {
1844 		/*
1845 		 * We allocate an uninitialized mbuf + cluster, mbuf is
1846 		 * initialized after rx.
1847 		 *
1848 		 * If the cluster is still set then we know a minimum sized packet was received
1849 		 */
1850 		bit_ffc_at(fl->ifl_rx_bitmap, frag_idx, fl->ifl_size,  &frag_idx);
1851 		if ((frag_idx < 0) || (frag_idx >= fl->ifl_size))
1852                 	bit_ffc(fl->ifl_rx_bitmap, fl->ifl_size, &frag_idx);
1853 		if ((cl = sd_cl[frag_idx]) == NULL) {
1854                        if ((cl = sd_cl[frag_idx] = m_cljget(NULL, M_NOWAIT, fl->ifl_buf_size)) == NULL)
1855 				break;
1856 #if MEMORY_LOGGING
1857 			fl->ifl_cl_enqueued++;
1858 #endif
1859 		}
1860 		if ((m = m_gethdr(M_NOWAIT, MT_NOINIT)) == NULL) {
1861 			break;
1862 		}
1863 #if MEMORY_LOGGING
1864 		fl->ifl_m_enqueued++;
1865 #endif
1866 
1867 		DBG_COUNTER_INC(rx_allocs);
1868 #if defined(__i386__) || defined(__amd64__)
1869 		if (!IS_DMAR(ctx)) {
1870 			bus_addr = pmap_kextract((vm_offset_t)cl);
1871 		} else
1872 #endif
1873 		{
1874 			struct rxq_refill_cb_arg cb_arg;
1875 			iflib_rxq_t q;
1876 
1877 			cb_arg.error = 0;
1878 			q = fl->ifl_rxq;
1879 			MPASS(sd_map != NULL);
1880 			MPASS(sd_map[frag_idx] != NULL);
1881 			err = bus_dmamap_load(fl->ifl_desc_tag, sd_map[frag_idx],
1882 		         cl, fl->ifl_buf_size, _rxq_refill_cb, &cb_arg, 0);
1883 			bus_dmamap_sync(fl->ifl_desc_tag, sd_map[frag_idx],
1884 					BUS_DMASYNC_PREREAD);
1885 
1886 			if (err != 0 || cb_arg.error) {
1887 				/*
1888 				 * !zone_pack ?
1889 				 */
1890 				if (fl->ifl_zone == zone_pack)
1891 					uma_zfree(fl->ifl_zone, cl);
1892 				m_free(m);
1893 				n = 0;
1894 				goto done;
1895 			}
1896 			bus_addr = cb_arg.seg.ds_addr;
1897 		}
1898                 bit_set(fl->ifl_rx_bitmap, frag_idx);
1899 		sd_flags[frag_idx] |= RX_SW_DESC_INUSE;
1900 
1901 		MPASS(sd_m[frag_idx] == NULL);
1902 		sd_cl[frag_idx] = cl;
1903 		sd_m[frag_idx] = m;
1904 		fl->ifl_rxd_idxs[i] = frag_idx;
1905 		fl->ifl_bus_addrs[i] = bus_addr;
1906 		fl->ifl_vm_addrs[i] = cl;
1907 		fl->ifl_credits++;
1908 		i++;
1909 		MPASS(fl->ifl_credits <= fl->ifl_size);
1910 		if (++idx == fl->ifl_size) {
1911 			fl->ifl_gen = 1;
1912 			idx = 0;
1913 		}
1914 		if (n == 0 || i == IFLIB_MAX_RX_REFRESH) {
1915 			iru.iru_pidx = pidx;
1916 			iru.iru_count = i;
1917 			ctx->isc_rxd_refill(ctx->ifc_softc, &iru);
1918 			i = 0;
1919 			pidx = idx;
1920 			fl->ifl_pidx = idx;
1921 		}
1922 
1923 	}
1924 done:
1925 	DBG_COUNTER_INC(rxd_flush);
1926 	if (fl->ifl_pidx == 0)
1927 		pidx = fl->ifl_size - 1;
1928 	else
1929 		pidx = fl->ifl_pidx - 1;
1930 
1931 	if (sd_map)
1932 		bus_dmamap_sync(fl->ifl_ifdi->idi_tag, fl->ifl_ifdi->idi_map,
1933 				BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
1934 	ctx->isc_rxd_flush(ctx->ifc_softc, fl->ifl_rxq->ifr_id, fl->ifl_id, pidx);
1935 	fl->ifl_fragidx = frag_idx;
1936 }
1937 
1938 static __inline void
1939 __iflib_fl_refill_lt(if_ctx_t ctx, iflib_fl_t fl, int max)
1940 {
1941 	/* we avoid allowing pidx to catch up with cidx as it confuses ixl */
1942 	int32_t reclaimable = fl->ifl_size - fl->ifl_credits - 1;
1943 #ifdef INVARIANTS
1944 	int32_t delta = fl->ifl_size - get_inuse(fl->ifl_size, fl->ifl_cidx, fl->ifl_pidx, fl->ifl_gen) - 1;
1945 #endif
1946 
1947 	MPASS(fl->ifl_credits <= fl->ifl_size);
1948 	MPASS(reclaimable == delta);
1949 
1950 	if (reclaimable > 0)
1951 		_iflib_fl_refill(ctx, fl, min(max, reclaimable));
1952 }
1953 
1954 static void
1955 iflib_fl_bufs_free(iflib_fl_t fl)
1956 {
1957 	iflib_dma_info_t idi = fl->ifl_ifdi;
1958 	uint32_t i;
1959 
1960 	for (i = 0; i < fl->ifl_size; i++) {
1961 		struct mbuf **sd_m = &fl->ifl_sds.ifsd_m[i];
1962 		uint8_t *sd_flags = &fl->ifl_sds.ifsd_flags[i];
1963 		caddr_t *sd_cl = &fl->ifl_sds.ifsd_cl[i];
1964 
1965 		if (*sd_flags & RX_SW_DESC_INUSE) {
1966 			if (fl->ifl_sds.ifsd_map != NULL) {
1967 				bus_dmamap_t sd_map = fl->ifl_sds.ifsd_map[i];
1968 				bus_dmamap_unload(fl->ifl_desc_tag, sd_map);
1969 				bus_dmamap_destroy(fl->ifl_desc_tag, sd_map);
1970 			}
1971 			if (*sd_m != NULL) {
1972 				m_init(*sd_m, M_NOWAIT, MT_DATA, 0);
1973 				uma_zfree(zone_mbuf, *sd_m);
1974 			}
1975 			if (*sd_cl != NULL)
1976 				uma_zfree(fl->ifl_zone, *sd_cl);
1977 			*sd_flags = 0;
1978 		} else {
1979 			MPASS(*sd_cl == NULL);
1980 			MPASS(*sd_m == NULL);
1981 		}
1982 #if MEMORY_LOGGING
1983 		fl->ifl_m_dequeued++;
1984 		fl->ifl_cl_dequeued++;
1985 #endif
1986 		*sd_cl = NULL;
1987 		*sd_m = NULL;
1988 	}
1989 #ifdef INVARIANTS
1990 	for (i = 0; i < fl->ifl_size; i++) {
1991 		MPASS(fl->ifl_sds.ifsd_flags[i] == 0);
1992 		MPASS(fl->ifl_sds.ifsd_cl[i] == NULL);
1993 		MPASS(fl->ifl_sds.ifsd_m[i] == NULL);
1994 	}
1995 #endif
1996 	/*
1997 	 * Reset free list values
1998 	 */
1999 	fl->ifl_credits = fl->ifl_cidx = fl->ifl_pidx = fl->ifl_gen = fl->ifl_fragidx = 0;
2000 	bzero(idi->idi_vaddr, idi->idi_size);
2001 }
2002 
2003 /*********************************************************************
2004  *
2005  *  Initialize a receive ring and its buffers.
2006  *
2007  **********************************************************************/
2008 static int
2009 iflib_fl_setup(iflib_fl_t fl)
2010 {
2011 	iflib_rxq_t rxq = fl->ifl_rxq;
2012 	if_ctx_t ctx = rxq->ifr_ctx;
2013 	if_softc_ctx_t sctx = &ctx->ifc_softc_ctx;
2014 
2015 	bit_nclear(fl->ifl_rx_bitmap, 0, fl->ifl_size);
2016 	/*
2017 	** Free current RX buffer structs and their mbufs
2018 	*/
2019 	iflib_fl_bufs_free(fl);
2020 	/* Now replenish the mbufs */
2021 	MPASS(fl->ifl_credits == 0);
2022 	/*
2023 	 * XXX don't set the max_frame_size to larger
2024 	 * than the hardware can handle
2025 	 */
2026 	if (sctx->isc_max_frame_size <= 2048)
2027 		fl->ifl_buf_size = MCLBYTES;
2028 #ifndef CONTIGMALLOC_WORKS
2029 	else
2030 		fl->ifl_buf_size = MJUMPAGESIZE;
2031 #else
2032 	else if (sctx->isc_max_frame_size <= 4096)
2033 		fl->ifl_buf_size = MJUMPAGESIZE;
2034 	else if (sctx->isc_max_frame_size <= 9216)
2035 		fl->ifl_buf_size = MJUM9BYTES;
2036 	else
2037 		fl->ifl_buf_size = MJUM16BYTES;
2038 #endif
2039 	if (fl->ifl_buf_size > ctx->ifc_max_fl_buf_size)
2040 		ctx->ifc_max_fl_buf_size = fl->ifl_buf_size;
2041 	fl->ifl_cltype = m_gettype(fl->ifl_buf_size);
2042 	fl->ifl_zone = m_getzone(fl->ifl_buf_size);
2043 
2044 
2045 	/* avoid pre-allocating zillions of clusters to an idle card
2046 	 * potentially speeding up attach
2047 	 */
2048 	_iflib_fl_refill(ctx, fl, min(128, fl->ifl_size));
2049 	MPASS(min(128, fl->ifl_size) == fl->ifl_credits);
2050 	if (min(128, fl->ifl_size) != fl->ifl_credits)
2051 		return (ENOBUFS);
2052 	/*
2053 	 * handle failure
2054 	 */
2055 	MPASS(rxq != NULL);
2056 	MPASS(fl->ifl_ifdi != NULL);
2057 	bus_dmamap_sync(fl->ifl_ifdi->idi_tag, fl->ifl_ifdi->idi_map,
2058 	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
2059 	return (0);
2060 }
2061 
2062 /*********************************************************************
2063  *
2064  *  Free receive ring data structures
2065  *
2066  **********************************************************************/
2067 static void
2068 iflib_rx_sds_free(iflib_rxq_t rxq)
2069 {
2070 	iflib_fl_t fl;
2071 	int i;
2072 
2073 	if (rxq->ifr_fl != NULL) {
2074 		for (i = 0; i < rxq->ifr_nfl; i++) {
2075 			fl = &rxq->ifr_fl[i];
2076 			if (fl->ifl_desc_tag != NULL) {
2077 				bus_dma_tag_destroy(fl->ifl_desc_tag);
2078 				fl->ifl_desc_tag = NULL;
2079 			}
2080 			free(fl->ifl_sds.ifsd_m, M_IFLIB);
2081 			free(fl->ifl_sds.ifsd_cl, M_IFLIB);
2082 			/* XXX destroy maps first */
2083 			free(fl->ifl_sds.ifsd_map, M_IFLIB);
2084 			fl->ifl_sds.ifsd_m = NULL;
2085 			fl->ifl_sds.ifsd_cl = NULL;
2086 			fl->ifl_sds.ifsd_map = NULL;
2087 		}
2088 		free(rxq->ifr_fl, M_IFLIB);
2089 		rxq->ifr_fl = NULL;
2090 		rxq->ifr_cq_gen = rxq->ifr_cq_cidx = rxq->ifr_cq_pidx = 0;
2091 	}
2092 }
2093 
2094 /*
2095  * MI independent logic
2096  *
2097  */
2098 static void
2099 iflib_timer(void *arg)
2100 {
2101 	iflib_txq_t txq = arg;
2102 	if_ctx_t ctx = txq->ift_ctx;
2103 	if_softc_ctx_t sctx = &ctx->ifc_softc_ctx;
2104 
2105 	if (!(if_getdrvflags(ctx->ifc_ifp) & IFF_DRV_RUNNING))
2106 		return;
2107 	/*
2108 	** Check on the state of the TX queue(s), this
2109 	** can be done without the lock because its RO
2110 	** and the HUNG state will be static if set.
2111 	*/
2112 	IFDI_TIMER(ctx, txq->ift_id);
2113 	if ((txq->ift_qstatus == IFLIB_QUEUE_HUNG) &&
2114 	    ((txq->ift_cleaned_prev == txq->ift_cleaned) ||
2115 	     (sctx->isc_pause_frames == 0)))
2116 		goto hung;
2117 
2118 	if (ifmp_ring_is_stalled(txq->ift_br))
2119 		txq->ift_qstatus = IFLIB_QUEUE_HUNG;
2120 	txq->ift_cleaned_prev = txq->ift_cleaned;
2121 	/* handle any laggards */
2122 	if (txq->ift_db_pending)
2123 		GROUPTASK_ENQUEUE(&txq->ift_task);
2124 
2125 	sctx->isc_pause_frames = 0;
2126 	if (if_getdrvflags(ctx->ifc_ifp) & IFF_DRV_RUNNING)
2127 		callout_reset_on(&txq->ift_timer, hz/2, iflib_timer, txq, txq->ift_timer.c_cpu);
2128 	return;
2129 hung:
2130 	CTX_LOCK(ctx);
2131 	if_setdrvflagbits(ctx->ifc_ifp, IFF_DRV_OACTIVE, IFF_DRV_RUNNING);
2132 	device_printf(ctx->ifc_dev,  "TX(%d) desc avail = %d, pidx = %d\n",
2133 				  txq->ift_id, TXQ_AVAIL(txq), txq->ift_pidx);
2134 
2135 	IFDI_WATCHDOG_RESET(ctx);
2136 	ctx->ifc_watchdog_events++;
2137 
2138 	ctx->ifc_flags |= IFC_DO_RESET;
2139 	iflib_admin_intr_deferred(ctx);
2140 	CTX_UNLOCK(ctx);
2141 }
2142 
2143 static void
2144 iflib_init_locked(if_ctx_t ctx)
2145 {
2146 	if_softc_ctx_t sctx = &ctx->ifc_softc_ctx;
2147 	if_softc_ctx_t scctx = &ctx->ifc_softc_ctx;
2148 	if_t ifp = ctx->ifc_ifp;
2149 	iflib_fl_t fl;
2150 	iflib_txq_t txq;
2151 	iflib_rxq_t rxq;
2152 	int i, j, tx_ip_csum_flags, tx_ip6_csum_flags;
2153 
2154 
2155 	if_setdrvflagbits(ifp, IFF_DRV_OACTIVE, IFF_DRV_RUNNING);
2156 	IFDI_INTR_DISABLE(ctx);
2157 
2158 	tx_ip_csum_flags = scctx->isc_tx_csum_flags & (CSUM_IP | CSUM_TCP | CSUM_UDP | CSUM_SCTP);
2159 	tx_ip6_csum_flags = scctx->isc_tx_csum_flags & (CSUM_IP6_TCP | CSUM_IP6_UDP | CSUM_IP6_SCTP);
2160 	/* Set hardware offload abilities */
2161 	if_clearhwassist(ifp);
2162 	if (if_getcapenable(ifp) & IFCAP_TXCSUM)
2163 		if_sethwassistbits(ifp, tx_ip_csum_flags, 0);
2164 	if (if_getcapenable(ifp) & IFCAP_TXCSUM_IPV6)
2165 		if_sethwassistbits(ifp,  tx_ip6_csum_flags, 0);
2166 	if (if_getcapenable(ifp) & IFCAP_TSO4)
2167 		if_sethwassistbits(ifp, CSUM_IP_TSO, 0);
2168 	if (if_getcapenable(ifp) & IFCAP_TSO6)
2169 		if_sethwassistbits(ifp, CSUM_IP6_TSO, 0);
2170 
2171 	for (i = 0, txq = ctx->ifc_txqs; i < sctx->isc_ntxqsets; i++, txq++) {
2172 		CALLOUT_LOCK(txq);
2173 		callout_stop(&txq->ift_timer);
2174 		CALLOUT_UNLOCK(txq);
2175 		iflib_netmap_txq_init(ctx, txq);
2176 	}
2177 #ifdef INVARIANTS
2178 	i = if_getdrvflags(ifp);
2179 #endif
2180 	IFDI_INIT(ctx);
2181 	MPASS(if_getdrvflags(ifp) == i);
2182 	for (i = 0, rxq = ctx->ifc_rxqs; i < sctx->isc_nrxqsets; i++, rxq++) {
2183 		/* XXX this should really be done on a per-queue basis */
2184 		if (if_getcapenable(ifp) & IFCAP_NETMAP) {
2185 			MPASS(rxq->ifr_id == i);
2186 			iflib_netmap_rxq_init(ctx, rxq);
2187 			continue;
2188 		}
2189 		for (j = 0, fl = rxq->ifr_fl; j < rxq->ifr_nfl; j++, fl++) {
2190 			if (iflib_fl_setup(fl)) {
2191 				device_printf(ctx->ifc_dev, "freelist setup failed - check cluster settings\n");
2192 				goto done;
2193 			}
2194 		}
2195 	}
2196 	done:
2197 	if_setdrvflagbits(ctx->ifc_ifp, IFF_DRV_RUNNING, IFF_DRV_OACTIVE);
2198 	IFDI_INTR_ENABLE(ctx);
2199 	txq = ctx->ifc_txqs;
2200 	for (i = 0; i < sctx->isc_ntxqsets; i++, txq++)
2201 		callout_reset_on(&txq->ift_timer, hz/2, iflib_timer, txq,
2202 			txq->ift_timer.c_cpu);
2203 }
2204 
2205 static int
2206 iflib_media_change(if_t ifp)
2207 {
2208 	if_ctx_t ctx = if_getsoftc(ifp);
2209 	int err;
2210 
2211 	CTX_LOCK(ctx);
2212 	if ((err = IFDI_MEDIA_CHANGE(ctx)) == 0)
2213 		iflib_init_locked(ctx);
2214 	CTX_UNLOCK(ctx);
2215 	return (err);
2216 }
2217 
2218 static void
2219 iflib_media_status(if_t ifp, struct ifmediareq *ifmr)
2220 {
2221 	if_ctx_t ctx = if_getsoftc(ifp);
2222 
2223 	CTX_LOCK(ctx);
2224 	IFDI_UPDATE_ADMIN_STATUS(ctx);
2225 	IFDI_MEDIA_STATUS(ctx, ifmr);
2226 	CTX_UNLOCK(ctx);
2227 }
2228 
2229 static void
2230 iflib_stop(if_ctx_t ctx)
2231 {
2232 	iflib_txq_t txq = ctx->ifc_txqs;
2233 	iflib_rxq_t rxq = ctx->ifc_rxqs;
2234 	if_softc_ctx_t scctx = &ctx->ifc_softc_ctx;
2235 	iflib_dma_info_t di;
2236 	iflib_fl_t fl;
2237 	int i, j;
2238 
2239 	/* Tell the stack that the interface is no longer active */
2240 	if_setdrvflagbits(ctx->ifc_ifp, IFF_DRV_OACTIVE, IFF_DRV_RUNNING);
2241 
2242 	IFDI_INTR_DISABLE(ctx);
2243 	DELAY(1000);
2244 	IFDI_STOP(ctx);
2245 	DELAY(1000);
2246 
2247 	iflib_debug_reset();
2248 	/* Wait for current tx queue users to exit to disarm watchdog timer. */
2249 	for (i = 0; i < scctx->isc_ntxqsets; i++, txq++) {
2250 		/* make sure all transmitters have completed before proceeding XXX */
2251 
2252 		/* clean any enqueued buffers */
2253 		iflib_ifmp_purge(txq);
2254 		/* Free any existing tx buffers. */
2255 		for (j = 0; j < txq->ift_size; j++) {
2256 			iflib_txsd_free(ctx, txq, j);
2257 		}
2258 		txq->ift_processed = txq->ift_cleaned = txq->ift_cidx_processed = 0;
2259 		txq->ift_in_use = txq->ift_gen = txq->ift_cidx = txq->ift_pidx = txq->ift_no_desc_avail = 0;
2260 		txq->ift_closed = txq->ift_mbuf_defrag = txq->ift_mbuf_defrag_failed = 0;
2261 		txq->ift_no_tx_dma_setup = txq->ift_txd_encap_efbig = txq->ift_map_failed = 0;
2262 		txq->ift_pullups = 0;
2263 		ifmp_ring_reset_stats(txq->ift_br);
2264 		for (j = 0, di = txq->ift_ifdi; j < ctx->ifc_nhwtxqs; j++, di++)
2265 			bzero((void *)di->idi_vaddr, di->idi_size);
2266 	}
2267 	for (i = 0; i < scctx->isc_nrxqsets; i++, rxq++) {
2268 		/* make sure all transmitters have completed before proceeding XXX */
2269 
2270 		for (j = 0, di = txq->ift_ifdi; j < ctx->ifc_nhwrxqs; j++, di++)
2271 			bzero((void *)di->idi_vaddr, di->idi_size);
2272 		/* also resets the free lists pidx/cidx */
2273 		for (j = 0, fl = rxq->ifr_fl; j < rxq->ifr_nfl; j++, fl++)
2274 			iflib_fl_bufs_free(fl);
2275 	}
2276 }
2277 
2278 static inline caddr_t
2279 calc_next_rxd(iflib_fl_t fl, int cidx)
2280 {
2281 	qidx_t size;
2282 	int nrxd;
2283 	caddr_t start, end, cur, next;
2284 
2285 	nrxd = fl->ifl_size;
2286 	size = fl->ifl_rxd_size;
2287 	start = fl->ifl_ifdi->idi_vaddr;
2288 
2289 	if (__predict_false(size == 0))
2290 		return (start);
2291 	cur = start + size*cidx;
2292 	end = start + size*nrxd;
2293 	next = CACHE_PTR_NEXT(cur);
2294 	return (next < end ? next : start);
2295 }
2296 
2297 static inline void
2298 prefetch_pkts(iflib_fl_t fl, int cidx)
2299 {
2300 	int nextptr;
2301 	int nrxd = fl->ifl_size;
2302 	caddr_t next_rxd;
2303 
2304 
2305 	nextptr = (cidx + CACHE_PTR_INCREMENT) & (nrxd-1);
2306 	prefetch(&fl->ifl_sds.ifsd_m[nextptr]);
2307 	prefetch(&fl->ifl_sds.ifsd_cl[nextptr]);
2308 	next_rxd = calc_next_rxd(fl, cidx);
2309 	prefetch(next_rxd);
2310 	prefetch(fl->ifl_sds.ifsd_m[(cidx + 1) & (nrxd-1)]);
2311 	prefetch(fl->ifl_sds.ifsd_m[(cidx + 2) & (nrxd-1)]);
2312 	prefetch(fl->ifl_sds.ifsd_m[(cidx + 3) & (nrxd-1)]);
2313 	prefetch(fl->ifl_sds.ifsd_m[(cidx + 4) & (nrxd-1)]);
2314 	prefetch(fl->ifl_sds.ifsd_cl[(cidx + 1) & (nrxd-1)]);
2315 	prefetch(fl->ifl_sds.ifsd_cl[(cidx + 2) & (nrxd-1)]);
2316 	prefetch(fl->ifl_sds.ifsd_cl[(cidx + 3) & (nrxd-1)]);
2317 	prefetch(fl->ifl_sds.ifsd_cl[(cidx + 4) & (nrxd-1)]);
2318 }
2319 
2320 static void
2321 rxd_frag_to_sd(iflib_rxq_t rxq, if_rxd_frag_t irf, int unload, if_rxsd_t sd)
2322 {
2323 	int flid, cidx;
2324 	bus_dmamap_t map;
2325 	iflib_fl_t fl;
2326 	iflib_dma_info_t di;
2327 	int next;
2328 
2329 	map = NULL;
2330 	flid = irf->irf_flid;
2331 	cidx = irf->irf_idx;
2332 	fl = &rxq->ifr_fl[flid];
2333 	sd->ifsd_fl = fl;
2334 	sd->ifsd_cidx = cidx;
2335 	sd->ifsd_m = &fl->ifl_sds.ifsd_m[cidx];
2336 	sd->ifsd_cl = &fl->ifl_sds.ifsd_cl[cidx];
2337 	fl->ifl_credits--;
2338 #if MEMORY_LOGGING
2339 	fl->ifl_m_dequeued++;
2340 #endif
2341 	if (rxq->ifr_ctx->ifc_flags & IFC_PREFETCH)
2342 		prefetch_pkts(fl, cidx);
2343 	if (fl->ifl_sds.ifsd_map != NULL) {
2344 		next = (cidx + CACHE_PTR_INCREMENT) & (fl->ifl_size-1);
2345 		prefetch(&fl->ifl_sds.ifsd_map[next]);
2346 		map = fl->ifl_sds.ifsd_map[cidx];
2347 		di = fl->ifl_ifdi;
2348 		next = (cidx + CACHE_LINE_SIZE) & (fl->ifl_size-1);
2349 		prefetch(&fl->ifl_sds.ifsd_flags[next]);
2350 		bus_dmamap_sync(di->idi_tag, di->idi_map,
2351 				BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2352 
2353 	/* not valid assert if bxe really does SGE from non-contiguous elements */
2354 		MPASS(fl->ifl_cidx == cidx);
2355 		if (unload)
2356 			bus_dmamap_unload(fl->ifl_desc_tag, map);
2357 	}
2358 	fl->ifl_cidx = (fl->ifl_cidx + 1) & (fl->ifl_size-1);
2359 	if (__predict_false(fl->ifl_cidx == 0))
2360 		fl->ifl_gen = 0;
2361 	if (map != NULL)
2362 		bus_dmamap_sync(fl->ifl_ifdi->idi_tag, fl->ifl_ifdi->idi_map,
2363 			BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
2364         bit_clear(fl->ifl_rx_bitmap, cidx);
2365 }
2366 
2367 static struct mbuf *
2368 assemble_segments(iflib_rxq_t rxq, if_rxd_info_t ri, if_rxsd_t sd)
2369 {
2370 	int i, padlen , flags;
2371 	struct mbuf *m, *mh, *mt;
2372 	caddr_t cl;
2373 
2374 	i = 0;
2375 	mh = NULL;
2376 	do {
2377 		rxd_frag_to_sd(rxq, &ri->iri_frags[i], TRUE, sd);
2378 
2379 		MPASS(*sd->ifsd_cl != NULL);
2380 		MPASS(*sd->ifsd_m != NULL);
2381 
2382 		/* Don't include zero-length frags */
2383 		if (ri->iri_frags[i].irf_len == 0) {
2384 			/* XXX we can save the cluster here, but not the mbuf */
2385 			m_init(*sd->ifsd_m, M_NOWAIT, MT_DATA, 0);
2386 			m_free(*sd->ifsd_m);
2387 			*sd->ifsd_m = NULL;
2388 			continue;
2389 		}
2390 		m = *sd->ifsd_m;
2391 		*sd->ifsd_m = NULL;
2392 		if (mh == NULL) {
2393 			flags = M_PKTHDR|M_EXT;
2394 			mh = mt = m;
2395 			padlen = ri->iri_pad;
2396 		} else {
2397 			flags = M_EXT;
2398 			mt->m_next = m;
2399 			mt = m;
2400 			/* assuming padding is only on the first fragment */
2401 			padlen = 0;
2402 		}
2403 		cl = *sd->ifsd_cl;
2404 		*sd->ifsd_cl = NULL;
2405 
2406 		/* Can these two be made one ? */
2407 		m_init(m, M_NOWAIT, MT_DATA, flags);
2408 		m_cljset(m, cl, sd->ifsd_fl->ifl_cltype);
2409 		/*
2410 		 * These must follow m_init and m_cljset
2411 		 */
2412 		m->m_data += padlen;
2413 		ri->iri_len -= padlen;
2414 		m->m_len = ri->iri_frags[i].irf_len;
2415 	} while (++i < ri->iri_nfrags);
2416 
2417 	return (mh);
2418 }
2419 
2420 /*
2421  * Process one software descriptor
2422  */
2423 static struct mbuf *
2424 iflib_rxd_pkt_get(iflib_rxq_t rxq, if_rxd_info_t ri)
2425 {
2426 	struct if_rxsd sd;
2427 	struct mbuf *m;
2428 
2429 	/* should I merge this back in now that the two paths are basically duplicated? */
2430 	if (ri->iri_nfrags == 1 &&
2431 	    ri->iri_frags[0].irf_len <= IFLIB_RX_COPY_THRESH) {
2432 		rxd_frag_to_sd(rxq, &ri->iri_frags[0], FALSE, &sd);
2433 		m = *sd.ifsd_m;
2434 		*sd.ifsd_m = NULL;
2435 		m_init(m, M_NOWAIT, MT_DATA, M_PKTHDR);
2436 #ifndef __NO_STRICT_ALIGNMENT
2437 		if (!IP_ALIGNED(m))
2438 			m->m_data += 2;
2439 #endif
2440 		memcpy(m->m_data, *sd.ifsd_cl, ri->iri_len);
2441 		m->m_len = ri->iri_frags[0].irf_len;
2442        } else {
2443 		m = assemble_segments(rxq, ri, &sd);
2444 	}
2445 	m->m_pkthdr.len = ri->iri_len;
2446 	m->m_pkthdr.rcvif = ri->iri_ifp;
2447 	m->m_flags |= ri->iri_flags;
2448 	m->m_pkthdr.ether_vtag = ri->iri_vtag;
2449 	m->m_pkthdr.flowid = ri->iri_flowid;
2450 	M_HASHTYPE_SET(m, ri->iri_rsstype);
2451 	m->m_pkthdr.csum_flags = ri->iri_csum_flags;
2452 	m->m_pkthdr.csum_data = ri->iri_csum_data;
2453 	return (m);
2454 }
2455 
2456 static bool
2457 iflib_rxeof(iflib_rxq_t rxq, qidx_t budget)
2458 {
2459 	if_ctx_t ctx = rxq->ifr_ctx;
2460 	if_shared_ctx_t sctx = ctx->ifc_sctx;
2461 	if_softc_ctx_t scctx = &ctx->ifc_softc_ctx;
2462 	int avail, i;
2463 	qidx_t *cidxp;
2464 	struct if_rxd_info ri;
2465 	int err, budget_left, rx_bytes, rx_pkts;
2466 	iflib_fl_t fl;
2467 	struct ifnet *ifp;
2468 	int lro_enabled;
2469 
2470 	/*
2471 	 * XXX early demux data packets so that if_input processing only handles
2472 	 * acks in interrupt context
2473 	 */
2474 	struct mbuf *m, *mh, *mt, *mf;
2475 
2476 	ifp = ctx->ifc_ifp;
2477 	mh = mt = NULL;
2478 	MPASS(budget > 0);
2479 	rx_pkts	= rx_bytes = 0;
2480 	if (sctx->isc_flags & IFLIB_HAS_RXCQ)
2481 		cidxp = &rxq->ifr_cq_cidx;
2482 	else
2483 		cidxp = &rxq->ifr_fl[0].ifl_cidx;
2484 	if ((avail = iflib_rxd_avail(ctx, rxq, *cidxp, budget)) == 0) {
2485 		for (i = 0, fl = &rxq->ifr_fl[0]; i < sctx->isc_nfl; i++, fl++)
2486 			__iflib_fl_refill_lt(ctx, fl, budget + 8);
2487 		DBG_COUNTER_INC(rx_unavail);
2488 		return (false);
2489 	}
2490 
2491 	for (budget_left = budget; (budget_left > 0) && (avail > 0); budget_left--, avail--) {
2492 		if (__predict_false(!CTX_ACTIVE(ctx))) {
2493 			DBG_COUNTER_INC(rx_ctx_inactive);
2494 			break;
2495 		}
2496 		/*
2497 		 * Reset client set fields to their default values
2498 		 */
2499 		rxd_info_zero(&ri);
2500 		ri.iri_qsidx = rxq->ifr_id;
2501 		ri.iri_cidx = *cidxp;
2502 		ri.iri_ifp = ifp;
2503 		ri.iri_frags = rxq->ifr_frags;
2504 		err = ctx->isc_rxd_pkt_get(ctx->ifc_softc, &ri);
2505 
2506 		if (err)
2507 			goto err;
2508 		if (sctx->isc_flags & IFLIB_HAS_RXCQ) {
2509 			*cidxp = ri.iri_cidx;
2510 			/* Update our consumer index */
2511 			/* XXX NB: shurd - check if this is still safe */
2512 			while (rxq->ifr_cq_cidx >= scctx->isc_nrxd[0]) {
2513 				rxq->ifr_cq_cidx -= scctx->isc_nrxd[0];
2514 				rxq->ifr_cq_gen = 0;
2515 			}
2516 			/* was this only a completion queue message? */
2517 			if (__predict_false(ri.iri_nfrags == 0))
2518 				continue;
2519 		}
2520 		MPASS(ri.iri_nfrags != 0);
2521 		MPASS(ri.iri_len != 0);
2522 
2523 		/* will advance the cidx on the corresponding free lists */
2524 		m = iflib_rxd_pkt_get(rxq, &ri);
2525 		if (avail == 0 && budget_left)
2526 			avail = iflib_rxd_avail(ctx, rxq, *cidxp, budget_left);
2527 
2528 		if (__predict_false(m == NULL)) {
2529 			DBG_COUNTER_INC(rx_mbuf_null);
2530 			continue;
2531 		}
2532 		/* imm_pkt: -- cxgb */
2533 		if (mh == NULL)
2534 			mh = mt = m;
2535 		else {
2536 			mt->m_nextpkt = m;
2537 			mt = m;
2538 		}
2539 	}
2540 	/* make sure that we can refill faster than drain */
2541 	for (i = 0, fl = &rxq->ifr_fl[0]; i < sctx->isc_nfl; i++, fl++)
2542 		__iflib_fl_refill_lt(ctx, fl, budget + 8);
2543 
2544 	lro_enabled = (if_getcapenable(ifp) & IFCAP_LRO);
2545 	mt = mf = NULL;
2546 	while (mh != NULL) {
2547 		m = mh;
2548 		if (mf == NULL)
2549 			mf = m;
2550 		mh = mh->m_nextpkt;
2551 		m->m_nextpkt = NULL;
2552 #ifndef __NO_STRICT_ALIGNMENT
2553 		if (!IP_ALIGNED(m) && (m = iflib_fixup_rx(m)) == NULL)
2554 			continue;
2555 #endif
2556 		rx_bytes += m->m_pkthdr.len;
2557 		rx_pkts++;
2558 #if defined(INET6) || defined(INET)
2559 		if (lro_enabled && tcp_lro_rx(&rxq->ifr_lc, m, 0) == 0) {
2560 			if (mf == m)
2561 				mf = NULL;
2562 			continue;
2563 		}
2564 #endif
2565 		if (mt != NULL)
2566 			mt->m_nextpkt = m;
2567 		mt = m;
2568 	}
2569 	if (mf != NULL) {
2570 		ifp->if_input(ifp, mf);
2571 		DBG_COUNTER_INC(rx_if_input);
2572 	}
2573 
2574 	if_inc_counter(ifp, IFCOUNTER_IBYTES, rx_bytes);
2575 	if_inc_counter(ifp, IFCOUNTER_IPACKETS, rx_pkts);
2576 
2577 	/*
2578 	 * Flush any outstanding LRO work
2579 	 */
2580 #if defined(INET6) || defined(INET)
2581 	tcp_lro_flush_all(&rxq->ifr_lc);
2582 #endif
2583 	if (avail)
2584 		return true;
2585 	return (iflib_rxd_avail(ctx, rxq, *cidxp, 1));
2586 err:
2587 	CTX_LOCK(ctx);
2588 	ctx->ifc_flags |= IFC_DO_RESET;
2589 	iflib_admin_intr_deferred(ctx);
2590 	CTX_UNLOCK(ctx);
2591 	return (false);
2592 }
2593 
2594 #define TXD_NOTIFY_COUNT(txq) (((txq)->ift_size / (txq)->ift_update_freq)-1)
2595 static inline qidx_t
2596 txq_max_db_deferred(iflib_txq_t txq, qidx_t in_use)
2597 {
2598 	qidx_t notify_count = TXD_NOTIFY_COUNT(txq);
2599 	qidx_t minthresh = txq->ift_size / 8;
2600 	if (in_use > 4*minthresh)
2601 		return (notify_count);
2602 	if (in_use > 2*minthresh)
2603 		return (notify_count >> 1);
2604 	if (in_use > minthresh)
2605 		return (notify_count >> 3);
2606 	return (0);
2607 }
2608 
2609 static inline qidx_t
2610 txq_max_rs_deferred(iflib_txq_t txq)
2611 {
2612 	qidx_t notify_count = TXD_NOTIFY_COUNT(txq);
2613 	qidx_t minthresh = txq->ift_size / 8;
2614 	if (txq->ift_in_use > 4*minthresh)
2615 		return (notify_count);
2616 	if (txq->ift_in_use > 2*minthresh)
2617 		return (notify_count >> 1);
2618 	if (txq->ift_in_use > minthresh)
2619 		return (notify_count >> 2);
2620 	return (2);
2621 }
2622 
2623 #define M_CSUM_FLAGS(m) ((m)->m_pkthdr.csum_flags)
2624 #define M_HAS_VLANTAG(m) (m->m_flags & M_VLANTAG)
2625 
2626 #define TXQ_MAX_DB_DEFERRED(txq, in_use) txq_max_db_deferred((txq), (in_use))
2627 #define TXQ_MAX_RS_DEFERRED(txq) txq_max_rs_deferred(txq)
2628 #define TXQ_MAX_DB_CONSUMED(size) (size >> 4)
2629 
2630 /* forward compatibility for cxgb */
2631 #define FIRST_QSET(ctx) 0
2632 #define NTXQSETS(ctx) ((ctx)->ifc_softc_ctx.isc_ntxqsets)
2633 #define NRXQSETS(ctx) ((ctx)->ifc_softc_ctx.isc_nrxqsets)
2634 #define QIDX(ctx, m) ((((m)->m_pkthdr.flowid & ctx->ifc_softc_ctx.isc_rss_table_mask) % NTXQSETS(ctx)) + FIRST_QSET(ctx))
2635 #define DESC_RECLAIMABLE(q) ((int)((q)->ift_processed - (q)->ift_cleaned - (q)->ift_ctx->ifc_softc_ctx.isc_tx_nsegments))
2636 
2637 /* XXX we should be setting this to something other than zero */
2638 #define RECLAIM_THRESH(ctx) ((ctx)->ifc_sctx->isc_tx_reclaim_thresh)
2639 #define MAX_TX_DESC(ctx) ((ctx)->ifc_softc_ctx.isc_tx_tso_segments_max)
2640 
2641 static inline bool
2642 iflib_txd_db_check(if_ctx_t ctx, iflib_txq_t txq, int ring, qidx_t in_use)
2643 {
2644 	qidx_t dbval, max;
2645 	bool rang;
2646 
2647 	rang = false;
2648 	max = TXQ_MAX_DB_DEFERRED(txq, in_use);
2649 	if (ring || txq->ift_db_pending >= max) {
2650 		dbval = txq->ift_npending ? txq->ift_npending : txq->ift_pidx;
2651 		ctx->isc_txd_flush(ctx->ifc_softc, txq->ift_id, dbval);
2652 		txq->ift_db_pending = txq->ift_npending = 0;
2653 		rang = true;
2654 	}
2655 	return (rang);
2656 }
2657 
2658 #ifdef PKT_DEBUG
2659 static void
2660 print_pkt(if_pkt_info_t pi)
2661 {
2662 	printf("pi len:  %d qsidx: %d nsegs: %d ndescs: %d flags: %x pidx: %d\n",
2663 	       pi->ipi_len, pi->ipi_qsidx, pi->ipi_nsegs, pi->ipi_ndescs, pi->ipi_flags, pi->ipi_pidx);
2664 	printf("pi new_pidx: %d csum_flags: %lx tso_segsz: %d mflags: %x vtag: %d\n",
2665 	       pi->ipi_new_pidx, pi->ipi_csum_flags, pi->ipi_tso_segsz, pi->ipi_mflags, pi->ipi_vtag);
2666 	printf("pi etype: %d ehdrlen: %d ip_hlen: %d ipproto: %d\n",
2667 	       pi->ipi_etype, pi->ipi_ehdrlen, pi->ipi_ip_hlen, pi->ipi_ipproto);
2668 }
2669 #endif
2670 
2671 #define IS_TSO4(pi) ((pi)->ipi_csum_flags & CSUM_IP_TSO)
2672 #define IS_TSO6(pi) ((pi)->ipi_csum_flags & CSUM_IP6_TSO)
2673 
2674 static int
2675 iflib_parse_header(iflib_txq_t txq, if_pkt_info_t pi, struct mbuf **mp)
2676 {
2677 	if_shared_ctx_t sctx = txq->ift_ctx->ifc_sctx;
2678 	struct ether_vlan_header *eh;
2679 	struct mbuf *m, *n;
2680 
2681 	n = m = *mp;
2682 	if ((sctx->isc_flags & IFLIB_NEED_SCRATCH) &&
2683 	    M_WRITABLE(m) == 0) {
2684 		if ((m = m_dup(m, M_NOWAIT)) == NULL) {
2685 			return (ENOMEM);
2686 		} else {
2687 			m_freem(*mp);
2688 			n = *mp = m;
2689 		}
2690 	}
2691 
2692 	/*
2693 	 * Determine where frame payload starts.
2694 	 * Jump over vlan headers if already present,
2695 	 * helpful for QinQ too.
2696 	 */
2697 	if (__predict_false(m->m_len < sizeof(*eh))) {
2698 		txq->ift_pullups++;
2699 		if (__predict_false((m = m_pullup(m, sizeof(*eh))) == NULL))
2700 			return (ENOMEM);
2701 	}
2702 	eh = mtod(m, struct ether_vlan_header *);
2703 	if (eh->evl_encap_proto == htons(ETHERTYPE_VLAN)) {
2704 		pi->ipi_etype = ntohs(eh->evl_proto);
2705 		pi->ipi_ehdrlen = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN;
2706 	} else {
2707 		pi->ipi_etype = ntohs(eh->evl_encap_proto);
2708 		pi->ipi_ehdrlen = ETHER_HDR_LEN;
2709 	}
2710 
2711 	if (if_getmtu(txq->ift_ctx->ifc_ifp) >= pi->ipi_len) {
2712 		pi->ipi_csum_flags &= ~(CSUM_IP_TSO|CSUM_IP6_TSO);
2713 	}
2714 
2715 	switch (pi->ipi_etype) {
2716 #ifdef INET
2717 	case ETHERTYPE_IP:
2718 	{
2719 		struct ip *ip = NULL;
2720 		struct tcphdr *th = NULL;
2721 		int minthlen;
2722 
2723 		minthlen = min(m->m_pkthdr.len, pi->ipi_ehdrlen + sizeof(*ip) + sizeof(*th));
2724 		if (__predict_false(m->m_len < minthlen)) {
2725 			/*
2726 			 * if this code bloat is causing too much of a hit
2727 			 * move it to a separate function and mark it noinline
2728 			 */
2729 			if (m->m_len == pi->ipi_ehdrlen) {
2730 				n = m->m_next;
2731 				MPASS(n);
2732 				if (n->m_len >= sizeof(*ip))  {
2733 					ip = (struct ip *)n->m_data;
2734 					if (n->m_len >= (ip->ip_hl << 2) + sizeof(*th))
2735 						th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
2736 				} else {
2737 					txq->ift_pullups++;
2738 					if (__predict_false((m = m_pullup(m, minthlen)) == NULL))
2739 						return (ENOMEM);
2740 					ip = (struct ip *)(m->m_data + pi->ipi_ehdrlen);
2741 				}
2742 			} else {
2743 				txq->ift_pullups++;
2744 				if (__predict_false((m = m_pullup(m, minthlen)) == NULL))
2745 					return (ENOMEM);
2746 				ip = (struct ip *)(m->m_data + pi->ipi_ehdrlen);
2747 				if (m->m_len >= (ip->ip_hl << 2) + sizeof(*th))
2748 					th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
2749 			}
2750 		} else {
2751 			ip = (struct ip *)(m->m_data + pi->ipi_ehdrlen);
2752 			if (m->m_len >= (ip->ip_hl << 2) + sizeof(*th))
2753 				th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
2754 		}
2755 		pi->ipi_ip_hlen = ip->ip_hl << 2;
2756 		pi->ipi_ipproto = ip->ip_p;
2757 		pi->ipi_flags |= IPI_TX_IPV4;
2758 
2759 		if ((sctx->isc_flags & IFLIB_NEED_ZERO_CSUM) && (pi->ipi_csum_flags & CSUM_IP))
2760                        ip->ip_sum = 0;
2761 
2762 		if (IS_TSO4(pi)) {
2763 			if (pi->ipi_ipproto == IPPROTO_TCP) {
2764 				if (__predict_false(th == NULL)) {
2765 					txq->ift_pullups++;
2766 					if (__predict_false((m = m_pullup(m, (ip->ip_hl << 2) + sizeof(*th))) == NULL))
2767 						return (ENOMEM);
2768 					th = (struct tcphdr *)((caddr_t)ip + pi->ipi_ip_hlen);
2769 				}
2770 				pi->ipi_tcp_hflags = th->th_flags;
2771 				pi->ipi_tcp_hlen = th->th_off << 2;
2772 				pi->ipi_tcp_seq = th->th_seq;
2773 			}
2774 			if (__predict_false(ip->ip_p != IPPROTO_TCP))
2775 				return (ENXIO);
2776 			th->th_sum = in_pseudo(ip->ip_src.s_addr,
2777 					       ip->ip_dst.s_addr, htons(IPPROTO_TCP));
2778 			pi->ipi_tso_segsz = m->m_pkthdr.tso_segsz;
2779 			if (sctx->isc_flags & IFLIB_TSO_INIT_IP) {
2780 				ip->ip_sum = 0;
2781 				ip->ip_len = htons(pi->ipi_ip_hlen + pi->ipi_tcp_hlen + pi->ipi_tso_segsz);
2782 			}
2783 		}
2784 		break;
2785 	}
2786 #endif
2787 #ifdef INET6
2788 	case ETHERTYPE_IPV6:
2789 	{
2790 		struct ip6_hdr *ip6 = (struct ip6_hdr *)(m->m_data + pi->ipi_ehdrlen);
2791 		struct tcphdr *th;
2792 		pi->ipi_ip_hlen = sizeof(struct ip6_hdr);
2793 
2794 		if (__predict_false(m->m_len < pi->ipi_ehdrlen + sizeof(struct ip6_hdr))) {
2795 			if (__predict_false((m = m_pullup(m, pi->ipi_ehdrlen + sizeof(struct ip6_hdr))) == NULL))
2796 				return (ENOMEM);
2797 		}
2798 		th = (struct tcphdr *)((caddr_t)ip6 + pi->ipi_ip_hlen);
2799 
2800 		/* XXX-BZ this will go badly in case of ext hdrs. */
2801 		pi->ipi_ipproto = ip6->ip6_nxt;
2802 		pi->ipi_flags |= IPI_TX_IPV6;
2803 
2804 		if (IS_TSO6(pi)) {
2805 			if (pi->ipi_ipproto == IPPROTO_TCP) {
2806 				if (__predict_false(m->m_len < pi->ipi_ehdrlen + sizeof(struct ip6_hdr) + sizeof(struct tcphdr))) {
2807 					if (__predict_false((m = m_pullup(m, pi->ipi_ehdrlen + sizeof(struct ip6_hdr) + sizeof(struct tcphdr))) == NULL))
2808 						return (ENOMEM);
2809 				}
2810 				pi->ipi_tcp_hflags = th->th_flags;
2811 				pi->ipi_tcp_hlen = th->th_off << 2;
2812 			}
2813 
2814 			if (__predict_false(ip6->ip6_nxt != IPPROTO_TCP))
2815 				return (ENXIO);
2816 			/*
2817 			 * The corresponding flag is set by the stack in the IPv4
2818 			 * TSO case, but not in IPv6 (at least in FreeBSD 10.2).
2819 			 * So, set it here because the rest of the flow requires it.
2820 			 */
2821 			pi->ipi_csum_flags |= CSUM_TCP_IPV6;
2822 			th->th_sum = in6_cksum_pseudo(ip6, 0, IPPROTO_TCP, 0);
2823 			pi->ipi_tso_segsz = m->m_pkthdr.tso_segsz;
2824 		}
2825 		break;
2826 	}
2827 #endif
2828 	default:
2829 		pi->ipi_csum_flags &= ~CSUM_OFFLOAD;
2830 		pi->ipi_ip_hlen = 0;
2831 		break;
2832 	}
2833 	*mp = m;
2834 
2835 	return (0);
2836 }
2837 
2838 static  __noinline  struct mbuf *
2839 collapse_pkthdr(struct mbuf *m0)
2840 {
2841 	struct mbuf *m, *m_next, *tmp;
2842 
2843 	m = m0;
2844 	m_next = m->m_next;
2845 	while (m_next != NULL && m_next->m_len == 0) {
2846 		m = m_next;
2847 		m->m_next = NULL;
2848 		m_free(m);
2849 		m_next = m_next->m_next;
2850 	}
2851 	m = m0;
2852 	m->m_next = m_next;
2853 	if ((m_next->m_flags & M_EXT) == 0) {
2854 		m = m_defrag(m, M_NOWAIT);
2855 	} else {
2856 		tmp = m_next->m_next;
2857 		memcpy(m_next, m, MPKTHSIZE);
2858 		m = m_next;
2859 		m->m_next = tmp;
2860 	}
2861 	return (m);
2862 }
2863 
2864 /*
2865  * If dodgy hardware rejects the scatter gather chain we've handed it
2866  * we'll need to remove the mbuf chain from ifsg_m[] before we can add the
2867  * m_defrag'd mbufs
2868  */
2869 static __noinline struct mbuf *
2870 iflib_remove_mbuf(iflib_txq_t txq)
2871 {
2872 	int ntxd, i, pidx;
2873 	struct mbuf *m, *mh, **ifsd_m;
2874 
2875 	pidx = txq->ift_pidx;
2876 	ifsd_m = txq->ift_sds.ifsd_m;
2877 	ntxd = txq->ift_size;
2878 	mh = m = ifsd_m[pidx];
2879 	ifsd_m[pidx] = NULL;
2880 #if MEMORY_LOGGING
2881 	txq->ift_dequeued++;
2882 #endif
2883 	i = 1;
2884 
2885 	while (m) {
2886 		ifsd_m[(pidx + i) & (ntxd -1)] = NULL;
2887 #if MEMORY_LOGGING
2888 		txq->ift_dequeued++;
2889 #endif
2890 		m = m->m_next;
2891 		i++;
2892 	}
2893 	return (mh);
2894 }
2895 
2896 static int
2897 iflib_busdma_load_mbuf_sg(iflib_txq_t txq, bus_dma_tag_t tag, bus_dmamap_t map,
2898 			  struct mbuf **m0, bus_dma_segment_t *segs, int *nsegs,
2899 			  int max_segs, int flags)
2900 {
2901 	if_ctx_t ctx;
2902 	if_shared_ctx_t		sctx;
2903 	if_softc_ctx_t		scctx;
2904 	int i, next, pidx, err, ntxd, count;
2905 	struct mbuf *m, *tmp, **ifsd_m;
2906 
2907 	m = *m0;
2908 
2909 	/*
2910 	 * Please don't ever do this
2911 	 */
2912 	if (__predict_false(m->m_len == 0))
2913 		*m0 = m = collapse_pkthdr(m);
2914 
2915 	ctx = txq->ift_ctx;
2916 	sctx = ctx->ifc_sctx;
2917 	scctx = &ctx->ifc_softc_ctx;
2918 	ifsd_m = txq->ift_sds.ifsd_m;
2919 	ntxd = txq->ift_size;
2920 	pidx = txq->ift_pidx;
2921 	if (map != NULL) {
2922 		uint8_t *ifsd_flags = txq->ift_sds.ifsd_flags;
2923 
2924 		err = bus_dmamap_load_mbuf_sg(tag, map,
2925 					      *m0, segs, nsegs, BUS_DMA_NOWAIT);
2926 		if (err)
2927 			return (err);
2928 		ifsd_flags[pidx] |= TX_SW_DESC_MAPPED;
2929 		count = 0;
2930 		m = *m0;
2931 		do {
2932 			if (__predict_false(m->m_len <= 0)) {
2933 				tmp = m;
2934 				m = m->m_next;
2935 				tmp->m_next = NULL;
2936 				m_free(tmp);
2937 				continue;
2938 			}
2939 			m = m->m_next;
2940 			count++;
2941 		} while (m != NULL);
2942 		if (count > *nsegs) {
2943 			ifsd_m[pidx] = *m0;
2944 			ifsd_m[pidx]->m_flags |= M_TOOBIG;
2945 			return (0);
2946 		}
2947 		m = *m0;
2948 		count = 0;
2949 		do {
2950 			next = (pidx + count) & (ntxd-1);
2951 			MPASS(ifsd_m[next] == NULL);
2952 			ifsd_m[next] = m;
2953 			count++;
2954 			tmp = m;
2955 			m = m->m_next;
2956 		} while (m != NULL);
2957 	} else {
2958 		int buflen, sgsize, maxsegsz, max_sgsize;
2959 		vm_offset_t vaddr;
2960 		vm_paddr_t curaddr;
2961 
2962 		count = i = 0;
2963 		m = *m0;
2964 		if (m->m_pkthdr.csum_flags & CSUM_TSO)
2965 			maxsegsz = scctx->isc_tx_tso_segsize_max;
2966 		else
2967 			maxsegsz = sctx->isc_tx_maxsegsize;
2968 
2969 		do {
2970 			if (__predict_false(m->m_len <= 0)) {
2971 				tmp = m;
2972 				m = m->m_next;
2973 				tmp->m_next = NULL;
2974 				m_free(tmp);
2975 				continue;
2976 			}
2977 			buflen = m->m_len;
2978 			vaddr = (vm_offset_t)m->m_data;
2979 			/*
2980 			 * see if we can't be smarter about physically
2981 			 * contiguous mappings
2982 			 */
2983 			next = (pidx + count) & (ntxd-1);
2984 			MPASS(ifsd_m[next] == NULL);
2985 #if MEMORY_LOGGING
2986 			txq->ift_enqueued++;
2987 #endif
2988 			ifsd_m[next] = m;
2989 			while (buflen > 0) {
2990 				if (i >= max_segs)
2991 					goto err;
2992 				max_sgsize = MIN(buflen, maxsegsz);
2993 				curaddr = pmap_kextract(vaddr);
2994 				sgsize = PAGE_SIZE - (curaddr & PAGE_MASK);
2995 				sgsize = MIN(sgsize, max_sgsize);
2996 				segs[i].ds_addr = curaddr;
2997 				segs[i].ds_len = sgsize;
2998 				vaddr += sgsize;
2999 				buflen -= sgsize;
3000 				i++;
3001 			}
3002 			count++;
3003 			tmp = m;
3004 			m = m->m_next;
3005 		} while (m != NULL);
3006 		*nsegs = i;
3007 	}
3008 	return (0);
3009 err:
3010 	*m0 = iflib_remove_mbuf(txq);
3011 	return (EFBIG);
3012 }
3013 
3014 static inline caddr_t
3015 calc_next_txd(iflib_txq_t txq, int cidx, uint8_t qid)
3016 {
3017 	qidx_t size;
3018 	int ntxd;
3019 	caddr_t start, end, cur, next;
3020 
3021 	ntxd = txq->ift_size;
3022 	size = txq->ift_txd_size[qid];
3023 	start = txq->ift_ifdi[qid].idi_vaddr;
3024 
3025 	if (__predict_false(size == 0))
3026 		return (start);
3027 	cur = start + size*cidx;
3028 	end = start + size*ntxd;
3029 	next = CACHE_PTR_NEXT(cur);
3030 	return (next < end ? next : start);
3031 }
3032 
3033 static int
3034 iflib_encap(iflib_txq_t txq, struct mbuf **m_headp)
3035 {
3036 	if_ctx_t		ctx;
3037 	if_shared_ctx_t		sctx;
3038 	if_softc_ctx_t		scctx;
3039 	bus_dma_segment_t	*segs;
3040 	struct mbuf		*m_head;
3041 	void			*next_txd;
3042 	bus_dmamap_t		map;
3043 	struct if_pkt_info	pi;
3044 	int remap = 0;
3045 	int err, nsegs, ndesc, max_segs, pidx, cidx, next, ntxd;
3046 	bus_dma_tag_t desc_tag;
3047 
3048 	segs = txq->ift_segs;
3049 	ctx = txq->ift_ctx;
3050 	sctx = ctx->ifc_sctx;
3051 	scctx = &ctx->ifc_softc_ctx;
3052 	segs = txq->ift_segs;
3053 	ntxd = txq->ift_size;
3054 	m_head = *m_headp;
3055 	map = NULL;
3056 
3057 	/*
3058 	 * If we're doing TSO the next descriptor to clean may be quite far ahead
3059 	 */
3060 	cidx = txq->ift_cidx;
3061 	pidx = txq->ift_pidx;
3062 	if (ctx->ifc_flags & IFC_PREFETCH) {
3063 		next = (cidx + CACHE_PTR_INCREMENT) & (ntxd-1);
3064 		if (!(ctx->ifc_flags & IFLIB_HAS_TXCQ)) {
3065 			next_txd = calc_next_txd(txq, cidx, 0);
3066 			prefetch(next_txd);
3067 		}
3068 
3069 		/* prefetch the next cache line of mbuf pointers and flags */
3070 		prefetch(&txq->ift_sds.ifsd_m[next]);
3071 		if (txq->ift_sds.ifsd_map != NULL) {
3072 			prefetch(&txq->ift_sds.ifsd_map[next]);
3073 			next = (cidx + CACHE_LINE_SIZE) & (ntxd-1);
3074 			prefetch(&txq->ift_sds.ifsd_flags[next]);
3075 		}
3076 	} else if (txq->ift_sds.ifsd_map != NULL)
3077 		map = txq->ift_sds.ifsd_map[pidx];
3078 
3079 	if (m_head->m_pkthdr.csum_flags & CSUM_TSO) {
3080 		desc_tag = txq->ift_tso_desc_tag;
3081 		max_segs = scctx->isc_tx_tso_segments_max;
3082 	} else {
3083 		desc_tag = txq->ift_desc_tag;
3084 		max_segs = scctx->isc_tx_nsegments;
3085 	}
3086 	m_head = *m_headp;
3087 
3088 	pkt_info_zero(&pi);
3089 	pi.ipi_len = m_head->m_pkthdr.len;
3090 	pi.ipi_mflags = (m_head->m_flags & (M_VLANTAG|M_BCAST|M_MCAST));
3091 	pi.ipi_csum_flags = m_head->m_pkthdr.csum_flags;
3092 	pi.ipi_vtag = (m_head->m_flags & M_VLANTAG) ? m_head->m_pkthdr.ether_vtag : 0;
3093 	pi.ipi_pidx = pidx;
3094 	pi.ipi_qsidx = txq->ift_id;
3095 
3096 	/* deliberate bitwise OR to make one condition */
3097 	if (__predict_true((pi.ipi_csum_flags | pi.ipi_vtag))) {
3098 		if (__predict_false((err = iflib_parse_header(txq, &pi, m_headp)) != 0))
3099 			return (err);
3100 		m_head = *m_headp;
3101 	}
3102 
3103 retry:
3104 	err = iflib_busdma_load_mbuf_sg(txq, desc_tag, map, m_headp, segs, &nsegs, max_segs, BUS_DMA_NOWAIT);
3105 defrag:
3106 	if (__predict_false(err)) {
3107 		switch (err) {
3108 		case EFBIG:
3109 			/* try collapse once and defrag once */
3110 			if (remap == 0)
3111 				m_head = m_collapse(*m_headp, M_NOWAIT, max_segs);
3112 			if (remap == 1)
3113 				m_head = m_defrag(*m_headp, M_NOWAIT);
3114 			remap++;
3115 			if (__predict_false(m_head == NULL))
3116 				goto defrag_failed;
3117 			txq->ift_mbuf_defrag++;
3118 			*m_headp = m_head;
3119 			goto retry;
3120 			break;
3121 		case ENOMEM:
3122 			txq->ift_no_tx_dma_setup++;
3123 			break;
3124 		default:
3125 			txq->ift_no_tx_dma_setup++;
3126 			m_freem(*m_headp);
3127 			DBG_COUNTER_INC(tx_frees);
3128 			*m_headp = NULL;
3129 			break;
3130 		}
3131 		txq->ift_map_failed++;
3132 		DBG_COUNTER_INC(encap_load_mbuf_fail);
3133 		return (err);
3134 	}
3135 
3136 	/*
3137 	 * XXX assumes a 1 to 1 relationship between segments and
3138 	 *        descriptors - this does not hold true on all drivers, e.g.
3139 	 *        cxgb
3140 	 */
3141 	if (__predict_false(nsegs + 2 > TXQ_AVAIL(txq))) {
3142 		txq->ift_no_desc_avail++;
3143 		if (map != NULL)
3144 			bus_dmamap_unload(desc_tag, map);
3145 		DBG_COUNTER_INC(encap_txq_avail_fail);
3146 		if ((txq->ift_task.gt_task.ta_flags & TASK_ENQUEUED) == 0)
3147 			GROUPTASK_ENQUEUE(&txq->ift_task);
3148 		return (ENOBUFS);
3149 	}
3150 	/*
3151 	 * On Intel cards we can greatly reduce the number of TX interrupts
3152 	 * we see by only setting report status on every Nth descriptor.
3153 	 * However, this also means that the driver will need to keep track
3154 	 * of the descriptors that RS was set on to check them for the DD bit.
3155 	 */
3156 	txq->ift_rs_pending += nsegs + 1;
3157 	if (txq->ift_rs_pending > TXQ_MAX_RS_DEFERRED(txq) ||
3158 	     iflib_no_tx_batch || (TXQ_AVAIL(txq) - nsegs - 1) <= MAX_TX_DESC(ctx)) {
3159 		pi.ipi_flags |= IPI_TX_INTR;
3160 		txq->ift_rs_pending = 0;
3161 	}
3162 
3163 	pi.ipi_segs = segs;
3164 	pi.ipi_nsegs = nsegs;
3165 
3166 	MPASS(pidx >= 0 && pidx < txq->ift_size);
3167 #ifdef PKT_DEBUG
3168 	print_pkt(&pi);
3169 #endif
3170 	if (map != NULL)
3171 		bus_dmamap_sync(desc_tag, map, BUS_DMASYNC_PREWRITE);
3172 	if ((err = ctx->isc_txd_encap(ctx->ifc_softc, &pi)) == 0) {
3173 		if (map != NULL)
3174 			bus_dmamap_sync(txq->ift_ifdi->idi_tag, txq->ift_ifdi->idi_map,
3175 					BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
3176 		DBG_COUNTER_INC(tx_encap);
3177 		MPASS(pi.ipi_new_pidx < txq->ift_size);
3178 
3179 		ndesc = pi.ipi_new_pidx - pi.ipi_pidx;
3180 		if (pi.ipi_new_pidx < pi.ipi_pidx) {
3181 			ndesc += txq->ift_size;
3182 			txq->ift_gen = 1;
3183 		}
3184 		/*
3185 		 * drivers can need as many as
3186 		 * two sentinels
3187 		 */
3188 		MPASS(ndesc <= pi.ipi_nsegs + 2);
3189 		MPASS(pi.ipi_new_pidx != pidx);
3190 		MPASS(ndesc > 0);
3191 		txq->ift_in_use += ndesc;
3192 
3193 		/*
3194 		 * We update the last software descriptor again here because there may
3195 		 * be a sentinel and/or there may be more mbufs than segments
3196 		 */
3197 		txq->ift_pidx = pi.ipi_new_pidx;
3198 		txq->ift_npending += pi.ipi_ndescs;
3199 	} else if (__predict_false(err == EFBIG && remap < 2)) {
3200 		*m_headp = m_head = iflib_remove_mbuf(txq);
3201 		remap = 1;
3202 		txq->ift_txd_encap_efbig++;
3203 		goto defrag;
3204 	} else
3205 		DBG_COUNTER_INC(encap_txd_encap_fail);
3206 	return (err);
3207 
3208 defrag_failed:
3209 	txq->ift_mbuf_defrag_failed++;
3210 	txq->ift_map_failed++;
3211 	m_freem(*m_headp);
3212 	DBG_COUNTER_INC(tx_frees);
3213 	*m_headp = NULL;
3214 	return (ENOMEM);
3215 }
3216 
3217 static void
3218 iflib_tx_desc_free(iflib_txq_t txq, int n)
3219 {
3220 	int hasmap;
3221 	uint32_t qsize, cidx, mask, gen;
3222 	struct mbuf *m, **ifsd_m;
3223 	uint8_t *ifsd_flags;
3224 	bus_dmamap_t *ifsd_map;
3225 	bool do_prefetch;
3226 
3227 	cidx = txq->ift_cidx;
3228 	gen = txq->ift_gen;
3229 	qsize = txq->ift_size;
3230 	mask = qsize-1;
3231 	hasmap = txq->ift_sds.ifsd_map != NULL;
3232 	ifsd_flags = txq->ift_sds.ifsd_flags;
3233 	ifsd_m = txq->ift_sds.ifsd_m;
3234 	ifsd_map = txq->ift_sds.ifsd_map;
3235 	do_prefetch = (txq->ift_ctx->ifc_flags & IFC_PREFETCH);
3236 
3237 	while (n--) {
3238 		if (do_prefetch) {
3239 			prefetch(ifsd_m[(cidx + 3) & mask]);
3240 			prefetch(ifsd_m[(cidx + 4) & mask]);
3241 		}
3242 		if (ifsd_m[cidx] != NULL) {
3243 			prefetch(&ifsd_m[(cidx + CACHE_PTR_INCREMENT) & mask]);
3244 			prefetch(&ifsd_flags[(cidx + CACHE_PTR_INCREMENT) & mask]);
3245 			if (hasmap && (ifsd_flags[cidx] & TX_SW_DESC_MAPPED)) {
3246 				/*
3247 				 * does it matter if it's not the TSO tag? If so we'll
3248 				 * have to add the type to flags
3249 				 */
3250 				bus_dmamap_unload(txq->ift_desc_tag, ifsd_map[cidx]);
3251 				ifsd_flags[cidx] &= ~TX_SW_DESC_MAPPED;
3252 			}
3253 			if ((m = ifsd_m[cidx]) != NULL) {
3254 				/* XXX we don't support any drivers that batch packets yet */
3255 				MPASS(m->m_nextpkt == NULL);
3256 				/* if the number of clusters exceeds the number of segments
3257 				 * there won't be space on the ring to save a pointer to each
3258 				 * cluster so we simply free the list here
3259 				 */
3260 				if (m->m_flags & M_TOOBIG) {
3261 					m_freem(m);
3262 				} else {
3263 					m_free(m);
3264 				}
3265 				ifsd_m[cidx] = NULL;
3266 #if MEMORY_LOGGING
3267 				txq->ift_dequeued++;
3268 #endif
3269 				DBG_COUNTER_INC(tx_frees);
3270 			}
3271 		}
3272 		if (__predict_false(++cidx == qsize)) {
3273 			cidx = 0;
3274 			gen = 0;
3275 		}
3276 	}
3277 	txq->ift_cidx = cidx;
3278 	txq->ift_gen = gen;
3279 }
3280 
3281 static __inline int
3282 iflib_completed_tx_reclaim(iflib_txq_t txq, int thresh)
3283 {
3284 	int reclaim;
3285 	if_ctx_t ctx = txq->ift_ctx;
3286 
3287 	KASSERT(thresh >= 0, ("invalid threshold to reclaim"));
3288 	MPASS(thresh /*+ MAX_TX_DESC(txq->ift_ctx) */ < txq->ift_size);
3289 
3290 	/*
3291 	 * Need a rate-limiting check so that this isn't called every time
3292 	 */
3293 	iflib_tx_credits_update(ctx, txq);
3294 	reclaim = DESC_RECLAIMABLE(txq);
3295 
3296 	if (reclaim <= thresh /* + MAX_TX_DESC(txq->ift_ctx) */) {
3297 #ifdef INVARIANTS
3298 		if (iflib_verbose_debug) {
3299 			printf("%s processed=%ju cleaned=%ju tx_nsegments=%d reclaim=%d thresh=%d\n", __FUNCTION__,
3300 			       txq->ift_processed, txq->ift_cleaned, txq->ift_ctx->ifc_softc_ctx.isc_tx_nsegments,
3301 			       reclaim, thresh);
3302 
3303 		}
3304 #endif
3305 		return (0);
3306 	}
3307 	iflib_tx_desc_free(txq, reclaim);
3308 	txq->ift_cleaned += reclaim;
3309 	txq->ift_in_use -= reclaim;
3310 
3311 	return (reclaim);
3312 }
3313 
3314 static struct mbuf **
3315 _ring_peek_one(struct ifmp_ring *r, int cidx, int offset, int remaining)
3316 {
3317 	int next, size;
3318 	struct mbuf **items;
3319 
3320 	size = r->size;
3321 	next = (cidx + CACHE_PTR_INCREMENT) & (size-1);
3322 	items = __DEVOLATILE(struct mbuf **, &r->items[0]);
3323 
3324 	prefetch(items[(cidx + offset) & (size-1)]);
3325 	if (remaining > 1) {
3326 		prefetch(&items[next]);
3327 		prefetch(items[(cidx + offset + 1) & (size-1)]);
3328 		prefetch(items[(cidx + offset + 2) & (size-1)]);
3329 		prefetch(items[(cidx + offset + 3) & (size-1)]);
3330 	}
3331 	return (__DEVOLATILE(struct mbuf **, &r->items[(cidx + offset) & (size-1)]));
3332 }
3333 
3334 static void
3335 iflib_txq_check_drain(iflib_txq_t txq, int budget)
3336 {
3337 
3338 	ifmp_ring_check_drainage(txq->ift_br, budget);
3339 }
3340 
3341 static uint32_t
3342 iflib_txq_can_drain(struct ifmp_ring *r)
3343 {
3344 	iflib_txq_t txq = r->cookie;
3345 	if_ctx_t ctx = txq->ift_ctx;
3346 
3347 	return ((TXQ_AVAIL(txq) > MAX_TX_DESC(ctx) + 2) ||
3348 		ctx->isc_txd_credits_update(ctx->ifc_softc, txq->ift_id, false));
3349 }
3350 
3351 static uint32_t
3352 iflib_txq_drain(struct ifmp_ring *r, uint32_t cidx, uint32_t pidx)
3353 {
3354 	iflib_txq_t txq = r->cookie;
3355 	if_ctx_t ctx = txq->ift_ctx;
3356 	struct ifnet *ifp = ctx->ifc_ifp;
3357 	struct mbuf **mp, *m;
3358 	int i, count, consumed, pkt_sent, bytes_sent, mcast_sent, avail;
3359 	int reclaimed, err, in_use_prev, desc_used;
3360 	bool do_prefetch, ring, rang;
3361 
3362 	if (__predict_false(!(if_getdrvflags(ifp) & IFF_DRV_RUNNING) ||
3363 			    !LINK_ACTIVE(ctx))) {
3364 		DBG_COUNTER_INC(txq_drain_notready);
3365 		return (0);
3366 	}
3367 	reclaimed = iflib_completed_tx_reclaim(txq, RECLAIM_THRESH(ctx));
3368 	rang = iflib_txd_db_check(ctx, txq, reclaimed, txq->ift_in_use);
3369 	avail = IDXDIFF(pidx, cidx, r->size);
3370 	if (__predict_false(ctx->ifc_flags & IFC_QFLUSH)) {
3371 		DBG_COUNTER_INC(txq_drain_flushing);
3372 		for (i = 0; i < avail; i++) {
3373 			m_free(r->items[(cidx + i) & (r->size-1)]);
3374 			r->items[(cidx + i) & (r->size-1)] = NULL;
3375 		}
3376 		return (avail);
3377 	}
3378 
3379 	if (__predict_false(if_getdrvflags(ctx->ifc_ifp) & IFF_DRV_OACTIVE)) {
3380 		txq->ift_qstatus = IFLIB_QUEUE_IDLE;
3381 		CALLOUT_LOCK(txq);
3382 		callout_stop(&txq->ift_timer);
3383 		CALLOUT_UNLOCK(txq);
3384 		DBG_COUNTER_INC(txq_drain_oactive);
3385 		return (0);
3386 	}
3387 	if (reclaimed)
3388 		txq->ift_qstatus = IFLIB_QUEUE_IDLE;
3389 	consumed = mcast_sent = bytes_sent = pkt_sent = 0;
3390 	count = MIN(avail, TX_BATCH_SIZE);
3391 #ifdef INVARIANTS
3392 	if (iflib_verbose_debug)
3393 		printf("%s avail=%d ifc_flags=%x txq_avail=%d ", __FUNCTION__,
3394 		       avail, ctx->ifc_flags, TXQ_AVAIL(txq));
3395 #endif
3396 	do_prefetch = (ctx->ifc_flags & IFC_PREFETCH);
3397 	avail = TXQ_AVAIL(txq);
3398 	for (desc_used = i = 0; i < count && avail > MAX_TX_DESC(ctx) + 2; i++) {
3399 		int pidx_prev, rem = do_prefetch ? count - i : 0;
3400 
3401 		mp = _ring_peek_one(r, cidx, i, rem);
3402 		MPASS(mp != NULL && *mp != NULL);
3403 		if (__predict_false(*mp == (struct mbuf *)txq)) {
3404 			consumed++;
3405 			reclaimed++;
3406 			continue;
3407 		}
3408 		in_use_prev = txq->ift_in_use;
3409 		pidx_prev = txq->ift_pidx;
3410 		err = iflib_encap(txq, mp);
3411 		if (__predict_false(err)) {
3412 			DBG_COUNTER_INC(txq_drain_encapfail);
3413 			/* no room - bail out */
3414 			if (err == ENOBUFS)
3415 				break;
3416 			consumed++;
3417 			DBG_COUNTER_INC(txq_drain_encapfail);
3418 			/* we can't send this packet - skip it */
3419 			continue;
3420 		}
3421 		consumed++;
3422 		pkt_sent++;
3423 		m = *mp;
3424 		DBG_COUNTER_INC(tx_sent);
3425 		bytes_sent += m->m_pkthdr.len;
3426 		mcast_sent += !!(m->m_flags & M_MCAST);
3427 		avail = TXQ_AVAIL(txq);
3428 
3429 		txq->ift_db_pending += (txq->ift_in_use - in_use_prev);
3430 		desc_used += (txq->ift_in_use - in_use_prev);
3431 		ETHER_BPF_MTAP(ifp, m);
3432 		if (__predict_false(!(ifp->if_drv_flags & IFF_DRV_RUNNING)))
3433 			break;
3434 		rang = iflib_txd_db_check(ctx, txq, false, in_use_prev);
3435 	}
3436 
3437 	/* deliberate use of bitwise or to avoid gratuitous short-circuit */
3438 	ring = rang ? false  : (iflib_min_tx_latency | err) || (TXQ_AVAIL(txq) < MAX_TX_DESC(ctx));
3439 	iflib_txd_db_check(ctx, txq, ring, txq->ift_in_use);
3440 	if_inc_counter(ifp, IFCOUNTER_OBYTES, bytes_sent);
3441 	if_inc_counter(ifp, IFCOUNTER_OPACKETS, pkt_sent);
3442 	if (mcast_sent)
3443 		if_inc_counter(ifp, IFCOUNTER_OMCASTS, mcast_sent);
3444 #ifdef INVARIANTS
3445 	if (iflib_verbose_debug)
3446 		printf("consumed=%d\n", consumed);
3447 #endif
3448 	return (consumed);
3449 }
3450 
3451 static uint32_t
3452 iflib_txq_drain_always(struct ifmp_ring *r)
3453 {
3454 	return (1);
3455 }
3456 
3457 static uint32_t
3458 iflib_txq_drain_free(struct ifmp_ring *r, uint32_t cidx, uint32_t pidx)
3459 {
3460 	int i, avail;
3461 	struct mbuf **mp;
3462 	iflib_txq_t txq;
3463 
3464 	txq = r->cookie;
3465 
3466 	txq->ift_qstatus = IFLIB_QUEUE_IDLE;
3467 	CALLOUT_LOCK(txq);
3468 	callout_stop(&txq->ift_timer);
3469 	CALLOUT_UNLOCK(txq);
3470 
3471 	avail = IDXDIFF(pidx, cidx, r->size);
3472 	for (i = 0; i < avail; i++) {
3473 		mp = _ring_peek_one(r, cidx, i, avail - i);
3474 		if (__predict_false(*mp == (struct mbuf *)txq))
3475 			continue;
3476 		m_freem(*mp);
3477 	}
3478 	MPASS(ifmp_ring_is_stalled(r) == 0);
3479 	return (avail);
3480 }
3481 
3482 static void
3483 iflib_ifmp_purge(iflib_txq_t txq)
3484 {
3485 	struct ifmp_ring *r;
3486 
3487 	r = txq->ift_br;
3488 	r->drain = iflib_txq_drain_free;
3489 	r->can_drain = iflib_txq_drain_always;
3490 
3491 	ifmp_ring_check_drainage(r, r->size);
3492 
3493 	r->drain = iflib_txq_drain;
3494 	r->can_drain = iflib_txq_can_drain;
3495 }
3496 
3497 static void
3498 _task_fn_tx(void *context)
3499 {
3500 	iflib_txq_t txq = context;
3501 	if_ctx_t ctx = txq->ift_ctx;
3502 	struct ifnet *ifp = ctx->ifc_ifp;
3503 	int rc;
3504 
3505 #ifdef IFLIB_DIAGNOSTICS
3506 	txq->ift_cpu_exec_count[curcpu]++;
3507 #endif
3508 	if (!(if_getdrvflags(ctx->ifc_ifp) & IFF_DRV_RUNNING))
3509 		return;
3510 	if (if_getcapenable(ifp) & IFCAP_NETMAP) {
3511 		if (ctx->isc_txd_credits_update(ctx->ifc_softc, txq->ift_id, false))
3512 			netmap_tx_irq(ifp, txq->ift_id);
3513 		IFDI_TX_QUEUE_INTR_ENABLE(ctx, txq->ift_id);
3514 		return;
3515 	}
3516 	if (txq->ift_db_pending)
3517 		ifmp_ring_enqueue(txq->ift_br, (void **)&txq, 1, TX_BATCH_SIZE);
3518 	ifmp_ring_check_drainage(txq->ift_br, TX_BATCH_SIZE);
3519 	if (ctx->ifc_flags & IFC_LEGACY)
3520 		IFDI_INTR_ENABLE(ctx);
3521 	else {
3522 		rc = IFDI_TX_QUEUE_INTR_ENABLE(ctx, txq->ift_id);
3523 		KASSERT(rc != ENOTSUP, ("MSI-X support requires queue_intr_enable, but not implemented in driver"));
3524 	}
3525 }
3526 
3527 static void
3528 _task_fn_rx(void *context)
3529 {
3530 	iflib_rxq_t rxq = context;
3531 	if_ctx_t ctx = rxq->ifr_ctx;
3532 	bool more;
3533 	int rc;
3534 	uint16_t budget;
3535 
3536 #ifdef IFLIB_DIAGNOSTICS
3537 	rxq->ifr_cpu_exec_count[curcpu]++;
3538 #endif
3539 	DBG_COUNTER_INC(task_fn_rxs);
3540 	if (__predict_false(!(if_getdrvflags(ctx->ifc_ifp) & IFF_DRV_RUNNING)))
3541 		return;
3542 	more = true;
3543 #ifdef DEV_NETMAP
3544 	if (if_getcapenable(ctx->ifc_ifp) & IFCAP_NETMAP) {
3545 		u_int work = 0;
3546 		if (netmap_rx_irq(ctx->ifc_ifp, rxq->ifr_id, &work)) {
3547 			more = false;
3548 		}
3549 	}
3550 #endif
3551 	budget = ctx->ifc_sysctl_rx_budget;
3552 	if (budget == 0)
3553 		budget = 16;	/* XXX */
3554 	if (more == false || (more = iflib_rxeof(rxq, budget)) == false) {
3555 		if (ctx->ifc_flags & IFC_LEGACY)
3556 			IFDI_INTR_ENABLE(ctx);
3557 		else {
3558 			DBG_COUNTER_INC(rx_intr_enables);
3559 			rc = IFDI_RX_QUEUE_INTR_ENABLE(ctx, rxq->ifr_id);
3560 			KASSERT(rc != ENOTSUP, ("MSI-X support requires queue_intr_enable, but not implemented in driver"));
3561 		}
3562 	}
3563 	if (__predict_false(!(if_getdrvflags(ctx->ifc_ifp) & IFF_DRV_RUNNING)))
3564 		return;
3565 	if (more)
3566 		GROUPTASK_ENQUEUE(&rxq->ifr_task);
3567 }
3568 
3569 static void
3570 _task_fn_admin(void *context)
3571 {
3572 	if_ctx_t ctx = context;
3573 	if_softc_ctx_t sctx = &ctx->ifc_softc_ctx;
3574 	iflib_txq_t txq;
3575 	int i;
3576 
3577 	if (!(if_getdrvflags(ctx->ifc_ifp) & IFF_DRV_RUNNING)) {
3578 		if (!(if_getdrvflags(ctx->ifc_ifp) & IFF_DRV_OACTIVE)) {
3579 			return;
3580 		}
3581 	}
3582 
3583 	CTX_LOCK(ctx);
3584 	for (txq = ctx->ifc_txqs, i = 0; i < sctx->isc_ntxqsets; i++, txq++) {
3585 		CALLOUT_LOCK(txq);
3586 		callout_stop(&txq->ift_timer);
3587 		CALLOUT_UNLOCK(txq);
3588 	}
3589 	IFDI_UPDATE_ADMIN_STATUS(ctx);
3590 	for (txq = ctx->ifc_txqs, i = 0; i < sctx->isc_ntxqsets; i++, txq++)
3591 		callout_reset_on(&txq->ift_timer, hz/2, iflib_timer, txq, txq->ift_timer.c_cpu);
3592 	IFDI_LINK_INTR_ENABLE(ctx);
3593 	if (ctx->ifc_flags & IFC_DO_RESET) {
3594 		ctx->ifc_flags &= ~IFC_DO_RESET;
3595 		iflib_if_init_locked(ctx);
3596 	}
3597 	CTX_UNLOCK(ctx);
3598 
3599 	if (LINK_ACTIVE(ctx) == 0)
3600 		return;
3601 	for (txq = ctx->ifc_txqs, i = 0; i < sctx->isc_ntxqsets; i++, txq++)
3602 		iflib_txq_check_drain(txq, IFLIB_RESTART_BUDGET);
3603 }
3604 
3605 
3606 static void
3607 _task_fn_iov(void *context)
3608 {
3609 	if_ctx_t ctx = context;
3610 
3611 	if (!(if_getdrvflags(ctx->ifc_ifp) & IFF_DRV_RUNNING))
3612 		return;
3613 
3614 	CTX_LOCK(ctx);
3615 	IFDI_VFLR_HANDLE(ctx);
3616 	CTX_UNLOCK(ctx);
3617 }
3618 
3619 static int
3620 iflib_sysctl_int_delay(SYSCTL_HANDLER_ARGS)
3621 {
3622 	int err;
3623 	if_int_delay_info_t info;
3624 	if_ctx_t ctx;
3625 
3626 	info = (if_int_delay_info_t)arg1;
3627 	ctx = info->iidi_ctx;
3628 	info->iidi_req = req;
3629 	info->iidi_oidp = oidp;
3630 	CTX_LOCK(ctx);
3631 	err = IFDI_SYSCTL_INT_DELAY(ctx, info);
3632 	CTX_UNLOCK(ctx);
3633 	return (err);
3634 }
3635 
3636 /*********************************************************************
3637  *
3638  *  IFNET FUNCTIONS
3639  *
3640  **********************************************************************/
3641 
3642 static void
3643 iflib_if_init_locked(if_ctx_t ctx)
3644 {
3645 	iflib_stop(ctx);
3646 	iflib_init_locked(ctx);
3647 }
3648 
3649 
3650 static void
3651 iflib_if_init(void *arg)
3652 {
3653 	if_ctx_t ctx = arg;
3654 
3655 	CTX_LOCK(ctx);
3656 	iflib_if_init_locked(ctx);
3657 	CTX_UNLOCK(ctx);
3658 }
3659 
3660 static int
3661 iflib_if_transmit(if_t ifp, struct mbuf *m)
3662 {
3663 	if_ctx_t	ctx = if_getsoftc(ifp);
3664 
3665 	iflib_txq_t txq;
3666 	int err, qidx;
3667 
3668 	if (__predict_false((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 || !LINK_ACTIVE(ctx))) {
3669 		DBG_COUNTER_INC(tx_frees);
3670 		m_freem(m);
3671 		return (ENOBUFS);
3672 	}
3673 
3674 	MPASS(m->m_nextpkt == NULL);
3675 	qidx = 0;
3676 	if ((NTXQSETS(ctx) > 1) && M_HASHTYPE_GET(m))
3677 		qidx = QIDX(ctx, m);
3678 	/*
3679 	 * XXX calculate buf_ring based on flowid (divvy up bits?)
3680 	 */
3681 	txq = &ctx->ifc_txqs[qidx];
3682 
3683 #ifdef DRIVER_BACKPRESSURE
3684 	if (txq->ift_closed) {
3685 		while (m != NULL) {
3686 			next = m->m_nextpkt;
3687 			m->m_nextpkt = NULL;
3688 			m_freem(m);
3689 			m = next;
3690 		}
3691 		return (ENOBUFS);
3692 	}
3693 #endif
3694 #ifdef notyet
3695 	qidx = count = 0;
3696 	mp = marr;
3697 	next = m;
3698 	do {
3699 		count++;
3700 		next = next->m_nextpkt;
3701 	} while (next != NULL);
3702 
3703 	if (count > nitems(marr))
3704 		if ((mp = malloc(count*sizeof(struct mbuf *), M_IFLIB, M_NOWAIT)) == NULL) {
3705 			/* XXX check nextpkt */
3706 			m_freem(m);
3707 			/* XXX simplify for now */
3708 			DBG_COUNTER_INC(tx_frees);
3709 			return (ENOBUFS);
3710 		}
3711 	for (next = m, i = 0; next != NULL; i++) {
3712 		mp[i] = next;
3713 		next = next->m_nextpkt;
3714 		mp[i]->m_nextpkt = NULL;
3715 	}
3716 #endif
3717 	DBG_COUNTER_INC(tx_seen);
3718 	err = ifmp_ring_enqueue(txq->ift_br, (void **)&m, 1, TX_BATCH_SIZE);
3719 
3720 	GROUPTASK_ENQUEUE(&txq->ift_task);
3721 	if (err) {
3722 		/* support forthcoming later */
3723 #ifdef DRIVER_BACKPRESSURE
3724 		txq->ift_closed = TRUE;
3725 #endif
3726 		ifmp_ring_check_drainage(txq->ift_br, TX_BATCH_SIZE);
3727 		m_freem(m);
3728 	}
3729 
3730 	return (err);
3731 }
3732 
3733 static void
3734 iflib_if_qflush(if_t ifp)
3735 {
3736 	if_ctx_t ctx = if_getsoftc(ifp);
3737 	iflib_txq_t txq = ctx->ifc_txqs;
3738 	int i;
3739 
3740 	CTX_LOCK(ctx);
3741 	ctx->ifc_flags |= IFC_QFLUSH;
3742 	CTX_UNLOCK(ctx);
3743 	for (i = 0; i < NTXQSETS(ctx); i++, txq++)
3744 		while (!(ifmp_ring_is_idle(txq->ift_br) || ifmp_ring_is_stalled(txq->ift_br)))
3745 			iflib_txq_check_drain(txq, 0);
3746 	CTX_LOCK(ctx);
3747 	ctx->ifc_flags &= ~IFC_QFLUSH;
3748 	CTX_UNLOCK(ctx);
3749 
3750 	if_qflush(ifp);
3751 }
3752 
3753 
3754 #define IFCAP_FLAGS (IFCAP_TXCSUM_IPV6 | IFCAP_RXCSUM_IPV6 | IFCAP_HWCSUM | IFCAP_LRO | \
3755 		     IFCAP_TSO4 | IFCAP_TSO6 | IFCAP_VLAN_HWTAGGING | IFCAP_HWSTATS | \
3756 		     IFCAP_VLAN_MTU | IFCAP_VLAN_HWFILTER | IFCAP_VLAN_HWTSO)
3757 
3758 static int
3759 iflib_if_ioctl(if_t ifp, u_long command, caddr_t data)
3760 {
3761 	if_ctx_t ctx = if_getsoftc(ifp);
3762 	struct ifreq	*ifr = (struct ifreq *)data;
3763 #if defined(INET) || defined(INET6)
3764 	struct ifaddr	*ifa = (struct ifaddr *)data;
3765 #endif
3766 	bool		avoid_reset = FALSE;
3767 	int		err = 0, reinit = 0, bits;
3768 
3769 	switch (command) {
3770 	case SIOCSIFADDR:
3771 #ifdef INET
3772 		if (ifa->ifa_addr->sa_family == AF_INET)
3773 			avoid_reset = TRUE;
3774 #endif
3775 #ifdef INET6
3776 		if (ifa->ifa_addr->sa_family == AF_INET6)
3777 			avoid_reset = TRUE;
3778 #endif
3779 		/*
3780 		** Calling init results in link renegotiation,
3781 		** so we avoid doing it when possible.
3782 		*/
3783 		if (avoid_reset) {
3784 			if_setflagbits(ifp, IFF_UP,0);
3785 			if (!(if_getdrvflags(ifp)& IFF_DRV_RUNNING))
3786 				reinit = 1;
3787 #ifdef INET
3788 			if (!(if_getflags(ifp) & IFF_NOARP))
3789 				arp_ifinit(ifp, ifa);
3790 #endif
3791 		} else
3792 			err = ether_ioctl(ifp, command, data);
3793 		break;
3794 	case SIOCSIFMTU:
3795 		CTX_LOCK(ctx);
3796 		if (ifr->ifr_mtu == if_getmtu(ifp)) {
3797 			CTX_UNLOCK(ctx);
3798 			break;
3799 		}
3800 		bits = if_getdrvflags(ifp);
3801 		/* stop the driver and free any clusters before proceeding */
3802 		iflib_stop(ctx);
3803 
3804 		if ((err = IFDI_MTU_SET(ctx, ifr->ifr_mtu)) == 0) {
3805 			if (ifr->ifr_mtu > ctx->ifc_max_fl_buf_size)
3806 				ctx->ifc_flags |= IFC_MULTISEG;
3807 			else
3808 				ctx->ifc_flags &= ~IFC_MULTISEG;
3809 			err = if_setmtu(ifp, ifr->ifr_mtu);
3810 		}
3811 		iflib_init_locked(ctx);
3812 		if_setdrvflags(ifp, bits);
3813 		CTX_UNLOCK(ctx);
3814 		break;
3815 	case SIOCSIFFLAGS:
3816 		CTX_LOCK(ctx);
3817 		if (if_getflags(ifp) & IFF_UP) {
3818 			if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) {
3819 				if ((if_getflags(ifp) ^ ctx->ifc_if_flags) &
3820 				    (IFF_PROMISC | IFF_ALLMULTI)) {
3821 					err = IFDI_PROMISC_SET(ctx, if_getflags(ifp));
3822 				}
3823 			} else
3824 				reinit = 1;
3825 		} else if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) {
3826 			iflib_stop(ctx);
3827 		}
3828 		ctx->ifc_if_flags = if_getflags(ifp);
3829 		CTX_UNLOCK(ctx);
3830 		break;
3831 	case SIOCADDMULTI:
3832 	case SIOCDELMULTI:
3833 		if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) {
3834 			CTX_LOCK(ctx);
3835 			IFDI_INTR_DISABLE(ctx);
3836 			IFDI_MULTI_SET(ctx);
3837 			IFDI_INTR_ENABLE(ctx);
3838 			CTX_UNLOCK(ctx);
3839 		}
3840 		break;
3841 	case SIOCSIFMEDIA:
3842 		CTX_LOCK(ctx);
3843 		IFDI_MEDIA_SET(ctx);
3844 		CTX_UNLOCK(ctx);
3845 		/* falls thru */
3846 	case SIOCGIFMEDIA:
3847 		err = ifmedia_ioctl(ifp, ifr, &ctx->ifc_media, command);
3848 		break;
3849 	case SIOCGI2C:
3850 	{
3851 		struct ifi2creq i2c;
3852 
3853 		err = copyin(ifr->ifr_data, &i2c, sizeof(i2c));
3854 		if (err != 0)
3855 			break;
3856 		if (i2c.dev_addr != 0xA0 && i2c.dev_addr != 0xA2) {
3857 			err = EINVAL;
3858 			break;
3859 		}
3860 		if (i2c.len > sizeof(i2c.data)) {
3861 			err = EINVAL;
3862 			break;
3863 		}
3864 
3865 		if ((err = IFDI_I2C_REQ(ctx, &i2c)) == 0)
3866 			err = copyout(&i2c, ifr->ifr_data, sizeof(i2c));
3867 		break;
3868 	}
3869 	case SIOCSIFCAP:
3870 	{
3871 		int mask, setmask;
3872 
3873 		mask = ifr->ifr_reqcap ^ if_getcapenable(ifp);
3874 		setmask = 0;
3875 #ifdef TCP_OFFLOAD
3876 		setmask |= mask & (IFCAP_TOE4|IFCAP_TOE6);
3877 #endif
3878 		setmask |= (mask & IFCAP_FLAGS);
3879 
3880 		if (setmask  & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6))
3881 			setmask |= (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6);
3882 		if ((mask & IFCAP_WOL) &&
3883 		    (if_getcapabilities(ifp) & IFCAP_WOL) != 0)
3884 			setmask |= (mask & (IFCAP_WOL_MCAST|IFCAP_WOL_MAGIC));
3885 		if_vlancap(ifp);
3886 		/*
3887 		 * want to ensure that traffic has stopped before we change any of the flags
3888 		 */
3889 		if (setmask) {
3890 			CTX_LOCK(ctx);
3891 			bits = if_getdrvflags(ifp);
3892 			if (bits & IFF_DRV_RUNNING)
3893 				iflib_stop(ctx);
3894 			if_togglecapenable(ifp, setmask);
3895 			if (bits & IFF_DRV_RUNNING)
3896 				iflib_init_locked(ctx);
3897 			if_setdrvflags(ifp, bits);
3898 			CTX_UNLOCK(ctx);
3899 		}
3900 		break;
3901 	    }
3902 	case SIOCGPRIVATE_0:
3903 	case SIOCSDRVSPEC:
3904 	case SIOCGDRVSPEC:
3905 		CTX_LOCK(ctx);
3906 		err = IFDI_PRIV_IOCTL(ctx, command, data);
3907 		CTX_UNLOCK(ctx);
3908 		break;
3909 	default:
3910 		err = ether_ioctl(ifp, command, data);
3911 		break;
3912 	}
3913 	if (reinit)
3914 		iflib_if_init(ctx);
3915 	return (err);
3916 }
3917 
3918 static uint64_t
3919 iflib_if_get_counter(if_t ifp, ift_counter cnt)
3920 {
3921 	if_ctx_t ctx = if_getsoftc(ifp);
3922 
3923 	return (IFDI_GET_COUNTER(ctx, cnt));
3924 }
3925 
3926 /*********************************************************************
3927  *
3928  *  OTHER FUNCTIONS EXPORTED TO THE STACK
3929  *
3930  **********************************************************************/
3931 
3932 static void
3933 iflib_vlan_register(void *arg, if_t ifp, uint16_t vtag)
3934 {
3935 	if_ctx_t ctx = if_getsoftc(ifp);
3936 
3937 	if ((void *)ctx != arg)
3938 		return;
3939 
3940 	if ((vtag == 0) || (vtag > 4095))
3941 		return;
3942 
3943 	CTX_LOCK(ctx);
3944 	IFDI_VLAN_REGISTER(ctx, vtag);
3945 	/* Re-init to load the changes */
3946 	if (if_getcapenable(ifp) & IFCAP_VLAN_HWFILTER)
3947 		iflib_if_init_locked(ctx);
3948 	CTX_UNLOCK(ctx);
3949 }
3950 
3951 static void
3952 iflib_vlan_unregister(void *arg, if_t ifp, uint16_t vtag)
3953 {
3954 	if_ctx_t ctx = if_getsoftc(ifp);
3955 
3956 	if ((void *)ctx != arg)
3957 		return;
3958 
3959 	if ((vtag == 0) || (vtag > 4095))
3960 		return;
3961 
3962 	CTX_LOCK(ctx);
3963 	IFDI_VLAN_UNREGISTER(ctx, vtag);
3964 	/* Re-init to load the changes */
3965 	if (if_getcapenable(ifp) & IFCAP_VLAN_HWFILTER)
3966 		iflib_if_init_locked(ctx);
3967 	CTX_UNLOCK(ctx);
3968 }
3969 
3970 static void
3971 iflib_led_func(void *arg, int onoff)
3972 {
3973 	if_ctx_t ctx = arg;
3974 
3975 	CTX_LOCK(ctx);
3976 	IFDI_LED_FUNC(ctx, onoff);
3977 	CTX_UNLOCK(ctx);
3978 }
3979 
3980 /*********************************************************************
3981  *
3982  *  BUS FUNCTION DEFINITIONS
3983  *
3984  **********************************************************************/
3985 
3986 int
3987 iflib_device_probe(device_t dev)
3988 {
3989 	pci_vendor_info_t *ent;
3990 
3991 	uint16_t	pci_vendor_id, pci_device_id;
3992 	uint16_t	pci_subvendor_id, pci_subdevice_id;
3993 	uint16_t	pci_rev_id;
3994 	if_shared_ctx_t sctx;
3995 
3996 	if ((sctx = DEVICE_REGISTER(dev)) == NULL || sctx->isc_magic != IFLIB_MAGIC)
3997 		return (ENOTSUP);
3998 
3999 	pci_vendor_id = pci_get_vendor(dev);
4000 	pci_device_id = pci_get_device(dev);
4001 	pci_subvendor_id = pci_get_subvendor(dev);
4002 	pci_subdevice_id = pci_get_subdevice(dev);
4003 	pci_rev_id = pci_get_revid(dev);
4004 	if (sctx->isc_parse_devinfo != NULL)
4005 		sctx->isc_parse_devinfo(&pci_device_id, &pci_subvendor_id, &pci_subdevice_id, &pci_rev_id);
4006 
4007 	ent = sctx->isc_vendor_info;
4008 	while (ent->pvi_vendor_id != 0) {
4009 		if (pci_vendor_id != ent->pvi_vendor_id) {
4010 			ent++;
4011 			continue;
4012 		}
4013 		if ((pci_device_id == ent->pvi_device_id) &&
4014 		    ((pci_subvendor_id == ent->pvi_subvendor_id) ||
4015 		     (ent->pvi_subvendor_id == 0)) &&
4016 		    ((pci_subdevice_id == ent->pvi_subdevice_id) ||
4017 		     (ent->pvi_subdevice_id == 0)) &&
4018 		    ((pci_rev_id == ent->pvi_rev_id) ||
4019 		     (ent->pvi_rev_id == 0))) {
4020 
4021 			device_set_desc_copy(dev, ent->pvi_name);
4022 			/* this needs to be changed to zero if the bus probing code
4023 			 * ever stops re-probing on best match because the sctx
4024 			 * may have its values over written by register calls
4025 			 * in subsequent probes
4026 			 */
4027 			return (BUS_PROBE_DEFAULT);
4028 		}
4029 		ent++;
4030 	}
4031 	return (ENXIO);
4032 }
4033 
4034 int
4035 iflib_device_register(device_t dev, void *sc, if_shared_ctx_t sctx, if_ctx_t *ctxp)
4036 {
4037 	int err, rid, msix, msix_bar;
4038 	if_ctx_t ctx;
4039 	if_t ifp;
4040 	if_softc_ctx_t scctx;
4041 	int i;
4042 	uint16_t main_txq;
4043 	uint16_t main_rxq;
4044 
4045 
4046 	ctx = malloc(sizeof(* ctx), M_IFLIB, M_WAITOK|M_ZERO);
4047 
4048 	if (sc == NULL) {
4049 		sc = malloc(sctx->isc_driver->size, M_IFLIB, M_WAITOK|M_ZERO);
4050 		device_set_softc(dev, ctx);
4051 		ctx->ifc_flags |= IFC_SC_ALLOCATED;
4052 	}
4053 
4054 	ctx->ifc_sctx = sctx;
4055 	ctx->ifc_dev = dev;
4056 	ctx->ifc_softc = sc;
4057 
4058 	if ((err = iflib_register(ctx)) != 0) {
4059 		device_printf(dev, "iflib_register failed %d\n", err);
4060 		return (err);
4061 	}
4062 	iflib_add_device_sysctl_pre(ctx);
4063 
4064 	scctx = &ctx->ifc_softc_ctx;
4065 	ifp = ctx->ifc_ifp;
4066 
4067 	/*
4068 	 * XXX sanity check that ntxd & nrxd are a power of 2
4069 	 */
4070 	if (ctx->ifc_sysctl_ntxqs != 0)
4071 		scctx->isc_ntxqsets = ctx->ifc_sysctl_ntxqs;
4072 	if (ctx->ifc_sysctl_nrxqs != 0)
4073 		scctx->isc_nrxqsets = ctx->ifc_sysctl_nrxqs;
4074 
4075 	for (i = 0; i < sctx->isc_ntxqs; i++) {
4076 		if (ctx->ifc_sysctl_ntxds[i] != 0)
4077 			scctx->isc_ntxd[i] = ctx->ifc_sysctl_ntxds[i];
4078 		else
4079 			scctx->isc_ntxd[i] = sctx->isc_ntxd_default[i];
4080 	}
4081 
4082 	for (i = 0; i < sctx->isc_nrxqs; i++) {
4083 		if (ctx->ifc_sysctl_nrxds[i] != 0)
4084 			scctx->isc_nrxd[i] = ctx->ifc_sysctl_nrxds[i];
4085 		else
4086 			scctx->isc_nrxd[i] = sctx->isc_nrxd_default[i];
4087 	}
4088 
4089 	for (i = 0; i < sctx->isc_nrxqs; i++) {
4090 		if (scctx->isc_nrxd[i] < sctx->isc_nrxd_min[i]) {
4091 			device_printf(dev, "nrxd%d: %d less than nrxd_min %d - resetting to min\n",
4092 				      i, scctx->isc_nrxd[i], sctx->isc_nrxd_min[i]);
4093 			scctx->isc_nrxd[i] = sctx->isc_nrxd_min[i];
4094 		}
4095 		if (scctx->isc_nrxd[i] > sctx->isc_nrxd_max[i]) {
4096 			device_printf(dev, "nrxd%d: %d greater than nrxd_max %d - resetting to max\n",
4097 				      i, scctx->isc_nrxd[i], sctx->isc_nrxd_max[i]);
4098 			scctx->isc_nrxd[i] = sctx->isc_nrxd_max[i];
4099 		}
4100 	}
4101 
4102 	for (i = 0; i < sctx->isc_ntxqs; i++) {
4103 		if (scctx->isc_ntxd[i] < sctx->isc_ntxd_min[i]) {
4104 			device_printf(dev, "ntxd%d: %d less than ntxd_min %d - resetting to min\n",
4105 				      i, scctx->isc_ntxd[i], sctx->isc_ntxd_min[i]);
4106 			scctx->isc_ntxd[i] = sctx->isc_ntxd_min[i];
4107 		}
4108 		if (scctx->isc_ntxd[i] > sctx->isc_ntxd_max[i]) {
4109 			device_printf(dev, "ntxd%d: %d greater than ntxd_max %d - resetting to max\n",
4110 				      i, scctx->isc_ntxd[i], sctx->isc_ntxd_max[i]);
4111 			scctx->isc_ntxd[i] = sctx->isc_ntxd_max[i];
4112 		}
4113 	}
4114 
4115 	if ((err = IFDI_ATTACH_PRE(ctx)) != 0) {
4116 		device_printf(dev, "IFDI_ATTACH_PRE failed %d\n", err);
4117 		return (err);
4118 	}
4119 	_iflib_pre_assert(scctx);
4120 	ctx->ifc_txrx = *scctx->isc_txrx;
4121 
4122 #ifdef INVARIANTS
4123 	MPASS(scctx->isc_capenable);
4124 	if (scctx->isc_capenable & IFCAP_TXCSUM)
4125 		MPASS(scctx->isc_tx_csum_flags);
4126 #endif
4127 
4128 	if_setcapabilities(ifp, scctx->isc_capenable | IFCAP_HWSTATS);
4129 	if_setcapenable(ifp, scctx->isc_capenable | IFCAP_HWSTATS);
4130 
4131 	if (scctx->isc_ntxqsets == 0 || (scctx->isc_ntxqsets_max && scctx->isc_ntxqsets_max < scctx->isc_ntxqsets))
4132 		scctx->isc_ntxqsets = scctx->isc_ntxqsets_max;
4133 	if (scctx->isc_nrxqsets == 0 || (scctx->isc_nrxqsets_max && scctx->isc_nrxqsets_max < scctx->isc_nrxqsets))
4134 		scctx->isc_nrxqsets = scctx->isc_nrxqsets_max;
4135 
4136 #ifdef ACPI_DMAR
4137 	if (dmar_get_dma_tag(device_get_parent(dev), dev) != NULL)
4138 		ctx->ifc_flags |= IFC_DMAR;
4139 #elif !(defined(__i386__) || defined(__amd64__))
4140 	/* set unconditionally for !x86 */
4141 	ctx->ifc_flags |= IFC_DMAR;
4142 #endif
4143 
4144 	msix_bar = scctx->isc_msix_bar;
4145 	main_txq = (sctx->isc_flags & IFLIB_HAS_TXCQ) ? 1 : 0;
4146 	main_rxq = (sctx->isc_flags & IFLIB_HAS_RXCQ) ? 1 : 0;
4147 
4148 	/* XXX change for per-queue sizes */
4149 	device_printf(dev, "using %d tx descriptors and %d rx descriptors\n",
4150 		      scctx->isc_ntxd[main_txq], scctx->isc_nrxd[main_rxq]);
4151 	for (i = 0; i < sctx->isc_nrxqs; i++) {
4152 		if (!powerof2(scctx->isc_nrxd[i])) {
4153 			/* round down instead? */
4154 			device_printf(dev, "# rx descriptors must be a power of 2\n");
4155 			err = EINVAL;
4156 			goto fail;
4157 		}
4158 	}
4159 	for (i = 0; i < sctx->isc_ntxqs; i++) {
4160 		if (!powerof2(scctx->isc_ntxd[i])) {
4161 			device_printf(dev,
4162 			    "# tx descriptors must be a power of 2");
4163 			err = EINVAL;
4164 			goto fail;
4165 		}
4166 	}
4167 
4168 	if (scctx->isc_tx_nsegments > scctx->isc_ntxd[main_txq] /
4169 	    MAX_SINGLE_PACKET_FRACTION)
4170 		scctx->isc_tx_nsegments = max(1, scctx->isc_ntxd[main_txq] /
4171 		    MAX_SINGLE_PACKET_FRACTION);
4172 	if (scctx->isc_tx_tso_segments_max > scctx->isc_ntxd[main_txq] /
4173 	    MAX_SINGLE_PACKET_FRACTION)
4174 		scctx->isc_tx_tso_segments_max = max(1,
4175 		    scctx->isc_ntxd[main_txq] / MAX_SINGLE_PACKET_FRACTION);
4176 
4177 	/*
4178 	 * Protect the stack against modern hardware
4179 	 */
4180 	if (scctx->isc_tx_tso_size_max > FREEBSD_TSO_SIZE_MAX)
4181 		scctx->isc_tx_tso_size_max = FREEBSD_TSO_SIZE_MAX;
4182 
4183 	/* TSO parameters - dig these out of the data sheet - simply correspond to tag setup */
4184 	ifp->if_hw_tsomaxsegcount = scctx->isc_tx_tso_segments_max;
4185 	ifp->if_hw_tsomax = scctx->isc_tx_tso_size_max;
4186 	ifp->if_hw_tsomaxsegsize = scctx->isc_tx_tso_segsize_max;
4187 	if (scctx->isc_rss_table_size == 0)
4188 		scctx->isc_rss_table_size = 64;
4189 	scctx->isc_rss_table_mask = scctx->isc_rss_table_size-1;
4190 
4191 	GROUPTASK_INIT(&ctx->ifc_admin_task, 0, _task_fn_admin, ctx);
4192 	/* XXX format name */
4193 	taskqgroup_attach(qgroup_if_config_tqg, &ctx->ifc_admin_task, ctx, -1, "admin");
4194 	/*
4195 	** Now setup MSI or MSI/X, should
4196 	** return us the number of supported
4197 	** vectors. (Will be 1 for MSI)
4198 	*/
4199 	if (sctx->isc_flags & IFLIB_SKIP_MSIX) {
4200 		msix = scctx->isc_vectors;
4201 	} else if (scctx->isc_msix_bar != 0)
4202 	       /*
4203 		* The simple fact that isc_msix_bar is not 0 does not mean we
4204 		* we have a good value there that is known to work.
4205 		*/
4206 		msix = iflib_msix_init(ctx);
4207 	else {
4208 		scctx->isc_vectors = 1;
4209 		scctx->isc_ntxqsets = 1;
4210 		scctx->isc_nrxqsets = 1;
4211 		scctx->isc_intr = IFLIB_INTR_LEGACY;
4212 		msix = 0;
4213 	}
4214 	/* Get memory for the station queues */
4215 	if ((err = iflib_queues_alloc(ctx))) {
4216 		device_printf(dev, "Unable to allocate queue memory\n");
4217 		goto fail;
4218 	}
4219 
4220 	if ((err = iflib_qset_structures_setup(ctx))) {
4221 		device_printf(dev, "qset structure setup failed %d\n", err);
4222 		goto fail_queues;
4223 	}
4224 
4225 	/*
4226 	 * Group taskqueues aren't properly set up until SMP is started,
4227 	 * so we disable interrupts until we can handle them post
4228 	 * SI_SUB_SMP.
4229 	 *
4230 	 * XXX: disabling interrupts doesn't actually work, at least for
4231 	 * the non-MSI case.  When they occur before SI_SUB_SMP completes,
4232 	 * we do null handling and depend on this not causing too large an
4233 	 * interrupt storm.
4234 	 */
4235 	IFDI_INTR_DISABLE(ctx);
4236 	if (msix > 1 && (err = IFDI_MSIX_INTR_ASSIGN(ctx, msix)) != 0) {
4237 		device_printf(dev, "IFDI_MSIX_INTR_ASSIGN failed %d\n", err);
4238 		goto fail_intr_free;
4239 	}
4240 	if (msix <= 1) {
4241 		rid = 0;
4242 		if (scctx->isc_intr == IFLIB_INTR_MSI) {
4243 			MPASS(msix == 1);
4244 			rid = 1;
4245 		}
4246 		if ((err = iflib_legacy_setup(ctx, ctx->isc_legacy_intr, ctx->ifc_softc, &rid, "irq0")) != 0) {
4247 			device_printf(dev, "iflib_legacy_setup failed %d\n", err);
4248 			goto fail_intr_free;
4249 		}
4250 	}
4251 	ether_ifattach(ctx->ifc_ifp, ctx->ifc_mac);
4252 	if ((err = IFDI_ATTACH_POST(ctx)) != 0) {
4253 		device_printf(dev, "IFDI_ATTACH_POST failed %d\n", err);
4254 		goto fail_detach;
4255 	}
4256 	if ((err = iflib_netmap_attach(ctx))) {
4257 		device_printf(ctx->ifc_dev, "netmap attach failed: %d\n", err);
4258 		goto fail_detach;
4259 	}
4260 	*ctxp = ctx;
4261 
4262 	if_setgetcounterfn(ctx->ifc_ifp, iflib_if_get_counter);
4263 	iflib_add_device_sysctl_post(ctx);
4264 	ctx->ifc_flags |= IFC_INIT_DONE;
4265 	return (0);
4266 fail_detach:
4267 	ether_ifdetach(ctx->ifc_ifp);
4268 fail_intr_free:
4269 	if (scctx->isc_intr == IFLIB_INTR_MSIX || scctx->isc_intr == IFLIB_INTR_MSI)
4270 		pci_release_msi(ctx->ifc_dev);
4271 fail_queues:
4272 	/* XXX free queues */
4273 fail:
4274 	IFDI_DETACH(ctx);
4275 	return (err);
4276 }
4277 
4278 int
4279 iflib_device_attach(device_t dev)
4280 {
4281 	if_ctx_t ctx;
4282 	if_shared_ctx_t sctx;
4283 
4284 	if ((sctx = DEVICE_REGISTER(dev)) == NULL || sctx->isc_magic != IFLIB_MAGIC)
4285 		return (ENOTSUP);
4286 
4287 	pci_enable_busmaster(dev);
4288 
4289 	return (iflib_device_register(dev, NULL, sctx, &ctx));
4290 }
4291 
4292 int
4293 iflib_device_deregister(if_ctx_t ctx)
4294 {
4295 	if_t ifp = ctx->ifc_ifp;
4296 	iflib_txq_t txq;
4297 	iflib_rxq_t rxq;
4298 	device_t dev = ctx->ifc_dev;
4299 	int i, j;
4300 	struct taskqgroup *tqg;
4301 	iflib_fl_t fl;
4302 
4303 	/* Make sure VLANS are not using driver */
4304 	if (if_vlantrunkinuse(ifp)) {
4305 		device_printf(dev,"Vlan in use, detach first\n");
4306 		return (EBUSY);
4307 	}
4308 
4309 	CTX_LOCK(ctx);
4310 	ctx->ifc_in_detach = 1;
4311 	iflib_stop(ctx);
4312 	CTX_UNLOCK(ctx);
4313 
4314 	/* Unregister VLAN events */
4315 	if (ctx->ifc_vlan_attach_event != NULL)
4316 		EVENTHANDLER_DEREGISTER(vlan_config, ctx->ifc_vlan_attach_event);
4317 	if (ctx->ifc_vlan_detach_event != NULL)
4318 		EVENTHANDLER_DEREGISTER(vlan_unconfig, ctx->ifc_vlan_detach_event);
4319 
4320 	iflib_netmap_detach(ifp);
4321 	ether_ifdetach(ifp);
4322 	/* ether_ifdetach calls if_qflush - lock must be destroy afterwards*/
4323 	CTX_LOCK_DESTROY(ctx);
4324 	if (ctx->ifc_led_dev != NULL)
4325 		led_destroy(ctx->ifc_led_dev);
4326 	/* XXX drain any dependent tasks */
4327 	tqg = qgroup_if_io_tqg;
4328 	for (txq = ctx->ifc_txqs, i = 0; i < NTXQSETS(ctx); i++, txq++) {
4329 		callout_drain(&txq->ift_timer);
4330 		if (txq->ift_task.gt_uniq != NULL)
4331 			taskqgroup_detach(tqg, &txq->ift_task);
4332 	}
4333 	for (i = 0, rxq = ctx->ifc_rxqs; i < NRXQSETS(ctx); i++, rxq++) {
4334 		if (rxq->ifr_task.gt_uniq != NULL)
4335 			taskqgroup_detach(tqg, &rxq->ifr_task);
4336 
4337 		for (j = 0, fl = rxq->ifr_fl; j < rxq->ifr_nfl; j++, fl++)
4338 			free(fl->ifl_rx_bitmap, M_IFLIB);
4339 
4340 	}
4341 	tqg = qgroup_if_config_tqg;
4342 	if (ctx->ifc_admin_task.gt_uniq != NULL)
4343 		taskqgroup_detach(tqg, &ctx->ifc_admin_task);
4344 	if (ctx->ifc_vflr_task.gt_uniq != NULL)
4345 		taskqgroup_detach(tqg, &ctx->ifc_vflr_task);
4346 
4347 	IFDI_DETACH(ctx);
4348 	device_set_softc(ctx->ifc_dev, NULL);
4349 	if (ctx->ifc_softc_ctx.isc_intr != IFLIB_INTR_LEGACY) {
4350 		pci_release_msi(dev);
4351 	}
4352 	if (ctx->ifc_softc_ctx.isc_intr != IFLIB_INTR_MSIX) {
4353 		iflib_irq_free(ctx, &ctx->ifc_legacy_irq);
4354 	}
4355 	if (ctx->ifc_msix_mem != NULL) {
4356 		bus_release_resource(ctx->ifc_dev, SYS_RES_MEMORY,
4357 			ctx->ifc_softc_ctx.isc_msix_bar, ctx->ifc_msix_mem);
4358 		ctx->ifc_msix_mem = NULL;
4359 	}
4360 
4361 	bus_generic_detach(dev);
4362 	if_free(ifp);
4363 
4364 	iflib_tx_structures_free(ctx);
4365 	iflib_rx_structures_free(ctx);
4366 	if (ctx->ifc_flags & IFC_SC_ALLOCATED)
4367 		free(ctx->ifc_softc, M_IFLIB);
4368 	free(ctx, M_IFLIB);
4369 	return (0);
4370 }
4371 
4372 
4373 int
4374 iflib_device_detach(device_t dev)
4375 {
4376 	if_ctx_t ctx = device_get_softc(dev);
4377 
4378 	return (iflib_device_deregister(ctx));
4379 }
4380 
4381 int
4382 iflib_device_suspend(device_t dev)
4383 {
4384 	if_ctx_t ctx = device_get_softc(dev);
4385 
4386 	CTX_LOCK(ctx);
4387 	IFDI_SUSPEND(ctx);
4388 	CTX_UNLOCK(ctx);
4389 
4390 	return bus_generic_suspend(dev);
4391 }
4392 int
4393 iflib_device_shutdown(device_t dev)
4394 {
4395 	if_ctx_t ctx = device_get_softc(dev);
4396 
4397 	CTX_LOCK(ctx);
4398 	IFDI_SHUTDOWN(ctx);
4399 	CTX_UNLOCK(ctx);
4400 
4401 	return bus_generic_suspend(dev);
4402 }
4403 
4404 
4405 int
4406 iflib_device_resume(device_t dev)
4407 {
4408 	if_ctx_t ctx = device_get_softc(dev);
4409 	iflib_txq_t txq = ctx->ifc_txqs;
4410 
4411 	CTX_LOCK(ctx);
4412 	IFDI_RESUME(ctx);
4413 	iflib_init_locked(ctx);
4414 	CTX_UNLOCK(ctx);
4415 	for (int i = 0; i < NTXQSETS(ctx); i++, txq++)
4416 		iflib_txq_check_drain(txq, IFLIB_RESTART_BUDGET);
4417 
4418 	return (bus_generic_resume(dev));
4419 }
4420 
4421 int
4422 iflib_device_iov_init(device_t dev, uint16_t num_vfs, const nvlist_t *params)
4423 {
4424 	int error;
4425 	if_ctx_t ctx = device_get_softc(dev);
4426 
4427 	CTX_LOCK(ctx);
4428 	error = IFDI_IOV_INIT(ctx, num_vfs, params);
4429 	CTX_UNLOCK(ctx);
4430 
4431 	return (error);
4432 }
4433 
4434 void
4435 iflib_device_iov_uninit(device_t dev)
4436 {
4437 	if_ctx_t ctx = device_get_softc(dev);
4438 
4439 	CTX_LOCK(ctx);
4440 	IFDI_IOV_UNINIT(ctx);
4441 	CTX_UNLOCK(ctx);
4442 }
4443 
4444 int
4445 iflib_device_iov_add_vf(device_t dev, uint16_t vfnum, const nvlist_t *params)
4446 {
4447 	int error;
4448 	if_ctx_t ctx = device_get_softc(dev);
4449 
4450 	CTX_LOCK(ctx);
4451 	error = IFDI_IOV_VF_ADD(ctx, vfnum, params);
4452 	CTX_UNLOCK(ctx);
4453 
4454 	return (error);
4455 }
4456 
4457 /*********************************************************************
4458  *
4459  *  MODULE FUNCTION DEFINITIONS
4460  *
4461  **********************************************************************/
4462 
4463 /*
4464  * - Start a fast taskqueue thread for each core
4465  * - Start a taskqueue for control operations
4466  */
4467 static int
4468 iflib_module_init(void)
4469 {
4470 	return (0);
4471 }
4472 
4473 static int
4474 iflib_module_event_handler(module_t mod, int what, void *arg)
4475 {
4476 	int err;
4477 
4478 	switch (what) {
4479 	case MOD_LOAD:
4480 		if ((err = iflib_module_init()) != 0)
4481 			return (err);
4482 		break;
4483 	case MOD_UNLOAD:
4484 		return (EBUSY);
4485 	default:
4486 		return (EOPNOTSUPP);
4487 	}
4488 
4489 	return (0);
4490 }
4491 
4492 /*********************************************************************
4493  *
4494  *  PUBLIC FUNCTION DEFINITIONS
4495  *     ordered as in iflib.h
4496  *
4497  **********************************************************************/
4498 
4499 
4500 static void
4501 _iflib_assert(if_shared_ctx_t sctx)
4502 {
4503 	MPASS(sctx->isc_tx_maxsize);
4504 	MPASS(sctx->isc_tx_maxsegsize);
4505 
4506 	MPASS(sctx->isc_rx_maxsize);
4507 	MPASS(sctx->isc_rx_nsegments);
4508 	MPASS(sctx->isc_rx_maxsegsize);
4509 
4510 	MPASS(sctx->isc_nrxd_min[0]);
4511 	MPASS(sctx->isc_nrxd_max[0]);
4512 	MPASS(sctx->isc_nrxd_default[0]);
4513 	MPASS(sctx->isc_ntxd_min[0]);
4514 	MPASS(sctx->isc_ntxd_max[0]);
4515 	MPASS(sctx->isc_ntxd_default[0]);
4516 }
4517 
4518 static void
4519 _iflib_pre_assert(if_softc_ctx_t scctx)
4520 {
4521 
4522 	MPASS(scctx->isc_txrx->ift_txd_encap);
4523 	MPASS(scctx->isc_txrx->ift_txd_flush);
4524 	MPASS(scctx->isc_txrx->ift_txd_credits_update);
4525 	MPASS(scctx->isc_txrx->ift_rxd_available);
4526 	MPASS(scctx->isc_txrx->ift_rxd_pkt_get);
4527 	MPASS(scctx->isc_txrx->ift_rxd_refill);
4528 	MPASS(scctx->isc_txrx->ift_rxd_flush);
4529 }
4530 
4531 static int
4532 iflib_register(if_ctx_t ctx)
4533 {
4534 	if_shared_ctx_t sctx = ctx->ifc_sctx;
4535 	driver_t *driver = sctx->isc_driver;
4536 	device_t dev = ctx->ifc_dev;
4537 	if_t ifp;
4538 
4539 	_iflib_assert(sctx);
4540 
4541 	CTX_LOCK_INIT(ctx, device_get_nameunit(ctx->ifc_dev));
4542 
4543 	ifp = ctx->ifc_ifp = if_gethandle(IFT_ETHER);
4544 	if (ifp == NULL) {
4545 		device_printf(dev, "can not allocate ifnet structure\n");
4546 		return (ENOMEM);
4547 	}
4548 
4549 	/*
4550 	 * Initialize our context's device specific methods
4551 	 */
4552 	kobj_init((kobj_t) ctx, (kobj_class_t) driver);
4553 	kobj_class_compile((kobj_class_t) driver);
4554 	driver->refs++;
4555 
4556 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
4557 	if_setsoftc(ifp, ctx);
4558 	if_setdev(ifp, dev);
4559 	if_setinitfn(ifp, iflib_if_init);
4560 	if_setioctlfn(ifp, iflib_if_ioctl);
4561 	if_settransmitfn(ifp, iflib_if_transmit);
4562 	if_setqflushfn(ifp, iflib_if_qflush);
4563 	if_setflags(ifp, IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST);
4564 
4565 	ctx->ifc_vlan_attach_event =
4566 		EVENTHANDLER_REGISTER(vlan_config, iflib_vlan_register, ctx,
4567 							  EVENTHANDLER_PRI_FIRST);
4568 	ctx->ifc_vlan_detach_event =
4569 		EVENTHANDLER_REGISTER(vlan_unconfig, iflib_vlan_unregister, ctx,
4570 							  EVENTHANDLER_PRI_FIRST);
4571 
4572 	ifmedia_init(&ctx->ifc_media, IFM_IMASK,
4573 					 iflib_media_change, iflib_media_status);
4574 
4575 	return (0);
4576 }
4577 
4578 
4579 static int
4580 iflib_queues_alloc(if_ctx_t ctx)
4581 {
4582 	if_shared_ctx_t sctx = ctx->ifc_sctx;
4583 	if_softc_ctx_t scctx = &ctx->ifc_softc_ctx;
4584 	device_t dev = ctx->ifc_dev;
4585 	int nrxqsets = scctx->isc_nrxqsets;
4586 	int ntxqsets = scctx->isc_ntxqsets;
4587 	iflib_txq_t txq;
4588 	iflib_rxq_t rxq;
4589 	iflib_fl_t fl = NULL;
4590 	int i, j, cpu, err, txconf, rxconf;
4591 	iflib_dma_info_t ifdip;
4592 	uint32_t *rxqsizes = scctx->isc_rxqsizes;
4593 	uint32_t *txqsizes = scctx->isc_txqsizes;
4594 	uint8_t nrxqs = sctx->isc_nrxqs;
4595 	uint8_t ntxqs = sctx->isc_ntxqs;
4596 	int nfree_lists = sctx->isc_nfl ? sctx->isc_nfl : 1;
4597 	caddr_t *vaddrs;
4598 	uint64_t *paddrs;
4599 	struct ifmp_ring **brscp;
4600 
4601 	KASSERT(ntxqs > 0, ("number of queues per qset must be at least 1"));
4602 	KASSERT(nrxqs > 0, ("number of queues per qset must be at least 1"));
4603 
4604 	brscp = NULL;
4605 	txq = NULL;
4606 	rxq = NULL;
4607 
4608 /* Allocate the TX ring struct memory */
4609 	if (!(txq =
4610 	    (iflib_txq_t) malloc(sizeof(struct iflib_txq) *
4611 	    ntxqsets, M_IFLIB, M_NOWAIT | M_ZERO))) {
4612 		device_printf(dev, "Unable to allocate TX ring memory\n");
4613 		err = ENOMEM;
4614 		goto fail;
4615 	}
4616 
4617 	/* Now allocate the RX */
4618 	if (!(rxq =
4619 	    (iflib_rxq_t) malloc(sizeof(struct iflib_rxq) *
4620 	    nrxqsets, M_IFLIB, M_NOWAIT | M_ZERO))) {
4621 		device_printf(dev, "Unable to allocate RX ring memory\n");
4622 		err = ENOMEM;
4623 		goto rx_fail;
4624 	}
4625 
4626 	ctx->ifc_txqs = txq;
4627 	ctx->ifc_rxqs = rxq;
4628 
4629 	/*
4630 	 * XXX handle allocation failure
4631 	 */
4632 	for (txconf = i = 0, cpu = CPU_FIRST(); i < ntxqsets; i++, txconf++, txq++, cpu = CPU_NEXT(cpu)) {
4633 		/* Set up some basics */
4634 
4635 		if ((ifdip = malloc(sizeof(struct iflib_dma_info) * ntxqs, M_IFLIB, M_WAITOK|M_ZERO)) == NULL) {
4636 			device_printf(dev, "failed to allocate iflib_dma_info\n");
4637 			err = ENOMEM;
4638 			goto err_tx_desc;
4639 		}
4640 		txq->ift_ifdi = ifdip;
4641 		for (j = 0; j < ntxqs; j++, ifdip++) {
4642 			if (iflib_dma_alloc(ctx, txqsizes[j], ifdip, BUS_DMA_NOWAIT)) {
4643 				device_printf(dev, "Unable to allocate Descriptor memory\n");
4644 				err = ENOMEM;
4645 				goto err_tx_desc;
4646 			}
4647 			txq->ift_txd_size[j] = scctx->isc_txd_size[j];
4648 			bzero((void *)ifdip->idi_vaddr, txqsizes[j]);
4649 		}
4650 		txq->ift_ctx = ctx;
4651 		txq->ift_id = i;
4652 		if (sctx->isc_flags & IFLIB_HAS_TXCQ) {
4653 			txq->ift_br_offset = 1;
4654 		} else {
4655 			txq->ift_br_offset = 0;
4656 		}
4657 		/* XXX fix this */
4658 		txq->ift_timer.c_cpu = cpu;
4659 
4660 		if (iflib_txsd_alloc(txq)) {
4661 			device_printf(dev, "Critical Failure setting up TX buffers\n");
4662 			err = ENOMEM;
4663 			goto err_tx_desc;
4664 		}
4665 
4666 		/* Initialize the TX lock */
4667 		snprintf(txq->ift_mtx_name, MTX_NAME_LEN, "%s:tx(%d):callout",
4668 		    device_get_nameunit(dev), txq->ift_id);
4669 		mtx_init(&txq->ift_mtx, txq->ift_mtx_name, NULL, MTX_DEF);
4670 		callout_init_mtx(&txq->ift_timer, &txq->ift_mtx, 0);
4671 
4672 		snprintf(txq->ift_db_mtx_name, MTX_NAME_LEN, "%s:tx(%d):db",
4673 			 device_get_nameunit(dev), txq->ift_id);
4674 
4675 		err = ifmp_ring_alloc(&txq->ift_br, 2048, txq, iflib_txq_drain,
4676 				      iflib_txq_can_drain, M_IFLIB, M_WAITOK);
4677 		if (err) {
4678 			/* XXX free any allocated rings */
4679 			device_printf(dev, "Unable to allocate buf_ring\n");
4680 			goto err_tx_desc;
4681 		}
4682 	}
4683 
4684 	for (rxconf = i = 0; i < nrxqsets; i++, rxconf++, rxq++) {
4685 		/* Set up some basics */
4686 
4687 		if ((ifdip = malloc(sizeof(struct iflib_dma_info) * nrxqs, M_IFLIB, M_WAITOK|M_ZERO)) == NULL) {
4688 			device_printf(dev, "failed to allocate iflib_dma_info\n");
4689 			err = ENOMEM;
4690 			goto err_tx_desc;
4691 		}
4692 
4693 		rxq->ifr_ifdi = ifdip;
4694 		/* XXX this needs to be changed if #rx queues != #tx queues */
4695 		rxq->ifr_ntxqirq = 1;
4696 		rxq->ifr_txqid[0] = i;
4697 		for (j = 0; j < nrxqs; j++, ifdip++) {
4698 			if (iflib_dma_alloc(ctx, rxqsizes[j], ifdip, BUS_DMA_NOWAIT)) {
4699 				device_printf(dev, "Unable to allocate Descriptor memory\n");
4700 				err = ENOMEM;
4701 				goto err_tx_desc;
4702 			}
4703 			bzero((void *)ifdip->idi_vaddr, rxqsizes[j]);
4704 		}
4705 		rxq->ifr_ctx = ctx;
4706 		rxq->ifr_id = i;
4707 		if (sctx->isc_flags & IFLIB_HAS_RXCQ) {
4708 			rxq->ifr_fl_offset = 1;
4709 		} else {
4710 			rxq->ifr_fl_offset = 0;
4711 		}
4712 		rxq->ifr_nfl = nfree_lists;
4713 		if (!(fl =
4714 			  (iflib_fl_t) malloc(sizeof(struct iflib_fl) * nfree_lists, M_IFLIB, M_NOWAIT | M_ZERO))) {
4715 			device_printf(dev, "Unable to allocate free list memory\n");
4716 			err = ENOMEM;
4717 			goto err_tx_desc;
4718 		}
4719 		rxq->ifr_fl = fl;
4720 		for (j = 0; j < nfree_lists; j++) {
4721 			fl[j].ifl_rxq = rxq;
4722 			fl[j].ifl_id = j;
4723 			fl[j].ifl_ifdi = &rxq->ifr_ifdi[j + rxq->ifr_fl_offset];
4724 			fl[j].ifl_rxd_size = scctx->isc_rxd_size[j];
4725 		}
4726         /* Allocate receive buffers for the ring*/
4727 		if (iflib_rxsd_alloc(rxq)) {
4728 			device_printf(dev,
4729 			    "Critical Failure setting up receive buffers\n");
4730 			err = ENOMEM;
4731 			goto err_rx_desc;
4732 		}
4733 
4734 		for (j = 0, fl = rxq->ifr_fl; j < rxq->ifr_nfl; j++, fl++)
4735 			fl->ifl_rx_bitmap = bit_alloc(fl->ifl_size, M_IFLIB, M_WAITOK|M_ZERO);
4736 	}
4737 
4738 	/* TXQs */
4739 	vaddrs = malloc(sizeof(caddr_t)*ntxqsets*ntxqs, M_IFLIB, M_WAITOK);
4740 	paddrs = malloc(sizeof(uint64_t)*ntxqsets*ntxqs, M_IFLIB, M_WAITOK);
4741 	for (i = 0; i < ntxqsets; i++) {
4742 		iflib_dma_info_t di = ctx->ifc_txqs[i].ift_ifdi;
4743 
4744 		for (j = 0; j < ntxqs; j++, di++) {
4745 			vaddrs[i*ntxqs + j] = di->idi_vaddr;
4746 			paddrs[i*ntxqs + j] = di->idi_paddr;
4747 		}
4748 	}
4749 	if ((err = IFDI_TX_QUEUES_ALLOC(ctx, vaddrs, paddrs, ntxqs, ntxqsets)) != 0) {
4750 		device_printf(ctx->ifc_dev, "device queue allocation failed\n");
4751 		iflib_tx_structures_free(ctx);
4752 		free(vaddrs, M_IFLIB);
4753 		free(paddrs, M_IFLIB);
4754 		goto err_rx_desc;
4755 	}
4756 	free(vaddrs, M_IFLIB);
4757 	free(paddrs, M_IFLIB);
4758 
4759 	/* RXQs */
4760 	vaddrs = malloc(sizeof(caddr_t)*nrxqsets*nrxqs, M_IFLIB, M_WAITOK);
4761 	paddrs = malloc(sizeof(uint64_t)*nrxqsets*nrxqs, M_IFLIB, M_WAITOK);
4762 	for (i = 0; i < nrxqsets; i++) {
4763 		iflib_dma_info_t di = ctx->ifc_rxqs[i].ifr_ifdi;
4764 
4765 		for (j = 0; j < nrxqs; j++, di++) {
4766 			vaddrs[i*nrxqs + j] = di->idi_vaddr;
4767 			paddrs[i*nrxqs + j] = di->idi_paddr;
4768 		}
4769 	}
4770 	if ((err = IFDI_RX_QUEUES_ALLOC(ctx, vaddrs, paddrs, nrxqs, nrxqsets)) != 0) {
4771 		device_printf(ctx->ifc_dev, "device queue allocation failed\n");
4772 		iflib_tx_structures_free(ctx);
4773 		free(vaddrs, M_IFLIB);
4774 		free(paddrs, M_IFLIB);
4775 		goto err_rx_desc;
4776 	}
4777 	free(vaddrs, M_IFLIB);
4778 	free(paddrs, M_IFLIB);
4779 
4780 	return (0);
4781 
4782 /* XXX handle allocation failure changes */
4783 err_rx_desc:
4784 err_tx_desc:
4785 	if (ctx->ifc_rxqs != NULL)
4786 		free(ctx->ifc_rxqs, M_IFLIB);
4787 	ctx->ifc_rxqs = NULL;
4788 	if (ctx->ifc_txqs != NULL)
4789 		free(ctx->ifc_txqs, M_IFLIB);
4790 	ctx->ifc_txqs = NULL;
4791 rx_fail:
4792 	if (brscp != NULL)
4793 		free(brscp, M_IFLIB);
4794 	if (rxq != NULL)
4795 		free(rxq, M_IFLIB);
4796 	if (txq != NULL)
4797 		free(txq, M_IFLIB);
4798 fail:
4799 	return (err);
4800 }
4801 
4802 static int
4803 iflib_tx_structures_setup(if_ctx_t ctx)
4804 {
4805 	iflib_txq_t txq = ctx->ifc_txqs;
4806 	int i;
4807 
4808 	for (i = 0; i < NTXQSETS(ctx); i++, txq++)
4809 		iflib_txq_setup(txq);
4810 
4811 	return (0);
4812 }
4813 
4814 static void
4815 iflib_tx_structures_free(if_ctx_t ctx)
4816 {
4817 	iflib_txq_t txq = ctx->ifc_txqs;
4818 	int i, j;
4819 
4820 	for (i = 0; i < NTXQSETS(ctx); i++, txq++) {
4821 		iflib_txq_destroy(txq);
4822 		for (j = 0; j < ctx->ifc_nhwtxqs; j++)
4823 			iflib_dma_free(&txq->ift_ifdi[j]);
4824 	}
4825 	free(ctx->ifc_txqs, M_IFLIB);
4826 	ctx->ifc_txqs = NULL;
4827 	IFDI_QUEUES_FREE(ctx);
4828 }
4829 
4830 /*********************************************************************
4831  *
4832  *  Initialize all receive rings.
4833  *
4834  **********************************************************************/
4835 static int
4836 iflib_rx_structures_setup(if_ctx_t ctx)
4837 {
4838 	iflib_rxq_t rxq = ctx->ifc_rxqs;
4839 	int q;
4840 #if defined(INET6) || defined(INET)
4841 	int i, err;
4842 #endif
4843 
4844 	for (q = 0; q < ctx->ifc_softc_ctx.isc_nrxqsets; q++, rxq++) {
4845 #if defined(INET6) || defined(INET)
4846 		tcp_lro_free(&rxq->ifr_lc);
4847 		if ((err = tcp_lro_init_args(&rxq->ifr_lc, ctx->ifc_ifp,
4848 		    TCP_LRO_ENTRIES, min(1024,
4849 		    ctx->ifc_softc_ctx.isc_nrxd[rxq->ifr_fl_offset]))) != 0) {
4850 			device_printf(ctx->ifc_dev, "LRO Initialization failed!\n");
4851 			goto fail;
4852 		}
4853 		rxq->ifr_lro_enabled = TRUE;
4854 #endif
4855 		IFDI_RXQ_SETUP(ctx, rxq->ifr_id);
4856 	}
4857 	return (0);
4858 #if defined(INET6) || defined(INET)
4859 fail:
4860 	/*
4861 	 * Free RX software descriptors allocated so far, we will only handle
4862 	 * the rings that completed, the failing case will have
4863 	 * cleaned up for itself. 'q' failed, so its the terminus.
4864 	 */
4865 	rxq = ctx->ifc_rxqs;
4866 	for (i = 0; i < q; ++i, rxq++) {
4867 		iflib_rx_sds_free(rxq);
4868 		rxq->ifr_cq_gen = rxq->ifr_cq_cidx = rxq->ifr_cq_pidx = 0;
4869 	}
4870 	return (err);
4871 #endif
4872 }
4873 
4874 /*********************************************************************
4875  *
4876  *  Free all receive rings.
4877  *
4878  **********************************************************************/
4879 static void
4880 iflib_rx_structures_free(if_ctx_t ctx)
4881 {
4882 	iflib_rxq_t rxq = ctx->ifc_rxqs;
4883 
4884 	for (int i = 0; i < ctx->ifc_softc_ctx.isc_nrxqsets; i++, rxq++) {
4885 		iflib_rx_sds_free(rxq);
4886 	}
4887 }
4888 
4889 static int
4890 iflib_qset_structures_setup(if_ctx_t ctx)
4891 {
4892 	int err;
4893 
4894 	if ((err = iflib_tx_structures_setup(ctx)) != 0)
4895 		return (err);
4896 
4897 	if ((err = iflib_rx_structures_setup(ctx)) != 0) {
4898 		device_printf(ctx->ifc_dev, "iflib_rx_structures_setup failed: %d\n", err);
4899 		iflib_tx_structures_free(ctx);
4900 		iflib_rx_structures_free(ctx);
4901 	}
4902 	return (err);
4903 }
4904 
4905 int
4906 iflib_irq_alloc(if_ctx_t ctx, if_irq_t irq, int rid,
4907 				driver_filter_t filter, void *filter_arg, driver_intr_t handler, void *arg, char *name)
4908 {
4909 
4910 	return (_iflib_irq_alloc(ctx, irq, rid, filter, handler, arg, name));
4911 }
4912 
4913 static int
4914 find_nth(if_ctx_t ctx, cpuset_t *cpus, int qid)
4915 {
4916 	int i, cpuid, eqid, count;
4917 
4918 	CPU_COPY(&ctx->ifc_cpus, cpus);
4919 	count = CPU_COUNT(&ctx->ifc_cpus);
4920 	eqid = qid % count;
4921 	/* clear up to the qid'th bit */
4922 	for (i = 0; i < eqid; i++) {
4923 		cpuid = CPU_FFS(cpus);
4924 		MPASS(cpuid != 0);
4925 		CPU_CLR(cpuid-1, cpus);
4926 	}
4927 	cpuid = CPU_FFS(cpus);
4928 	MPASS(cpuid != 0);
4929 	return (cpuid-1);
4930 }
4931 
4932 int
4933 iflib_irq_alloc_generic(if_ctx_t ctx, if_irq_t irq, int rid,
4934 						iflib_intr_type_t type, driver_filter_t *filter,
4935 						void *filter_arg, int qid, char *name)
4936 {
4937 	struct grouptask *gtask;
4938 	struct taskqgroup *tqg;
4939 	iflib_filter_info_t info;
4940 	cpuset_t cpus;
4941 	gtask_fn_t *fn;
4942 	int tqrid, err, cpuid;
4943 	driver_filter_t *intr_fast;
4944 	void *q;
4945 
4946 	info = &ctx->ifc_filter_info;
4947 	tqrid = rid;
4948 
4949 	switch (type) {
4950 	/* XXX merge tx/rx for netmap? */
4951 	case IFLIB_INTR_TX:
4952 		q = &ctx->ifc_txqs[qid];
4953 		info = &ctx->ifc_txqs[qid].ift_filter_info;
4954 		gtask = &ctx->ifc_txqs[qid].ift_task;
4955 		tqg = qgroup_if_io_tqg;
4956 		fn = _task_fn_tx;
4957 		intr_fast = iflib_fast_intr;
4958 		GROUPTASK_INIT(gtask, 0, fn, q);
4959 		break;
4960 	case IFLIB_INTR_RX:
4961 		q = &ctx->ifc_rxqs[qid];
4962 		info = &ctx->ifc_rxqs[qid].ifr_filter_info;
4963 		gtask = &ctx->ifc_rxqs[qid].ifr_task;
4964 		tqg = qgroup_if_io_tqg;
4965 		fn = _task_fn_rx;
4966 		intr_fast = iflib_fast_intr;
4967 		GROUPTASK_INIT(gtask, 0, fn, q);
4968 		break;
4969 	case IFLIB_INTR_RXTX:
4970 		q = &ctx->ifc_rxqs[qid];
4971 		info = &ctx->ifc_rxqs[qid].ifr_filter_info;
4972 		gtask = &ctx->ifc_rxqs[qid].ifr_task;
4973 		tqg = qgroup_if_io_tqg;
4974 		fn = _task_fn_rx;
4975 		intr_fast = iflib_fast_intr_rxtx;
4976 		GROUPTASK_INIT(gtask, 0, fn, q);
4977 		break;
4978 	case IFLIB_INTR_ADMIN:
4979 		q = ctx;
4980 		tqrid = -1;
4981 		info = &ctx->ifc_filter_info;
4982 		gtask = &ctx->ifc_admin_task;
4983 		tqg = qgroup_if_config_tqg;
4984 		fn = _task_fn_admin;
4985 		intr_fast = iflib_fast_intr_ctx;
4986 		break;
4987 	default:
4988 		panic("unknown net intr type");
4989 	}
4990 
4991 	info->ifi_filter = filter;
4992 	info->ifi_filter_arg = filter_arg;
4993 	info->ifi_task = gtask;
4994 	info->ifi_ctx = q;
4995 
4996 	err = _iflib_irq_alloc(ctx, irq, rid, intr_fast, NULL, info,  name);
4997 	if (err != 0) {
4998 		device_printf(ctx->ifc_dev, "_iflib_irq_alloc failed %d\n", err);
4999 		return (err);
5000 	}
5001 	if (type == IFLIB_INTR_ADMIN)
5002 		return (0);
5003 
5004 	if (tqrid != -1) {
5005 		cpuid = find_nth(ctx, &cpus, qid);
5006 		taskqgroup_attach_cpu(tqg, gtask, q, cpuid, rman_get_start(irq->ii_res), name);
5007 	} else {
5008 		taskqgroup_attach(tqg, gtask, q, rman_get_start(irq->ii_res), name);
5009 	}
5010 
5011 	return (0);
5012 }
5013 
5014 void
5015 iflib_softirq_alloc_generic(if_ctx_t ctx, if_irq_t irq, iflib_intr_type_t type,  void *arg, int qid, char *name)
5016 {
5017 	struct grouptask *gtask;
5018 	struct taskqgroup *tqg;
5019 	gtask_fn_t *fn;
5020 	void *q;
5021 	int irq_num = -1;
5022 
5023 	switch (type) {
5024 	case IFLIB_INTR_TX:
5025 		q = &ctx->ifc_txqs[qid];
5026 		gtask = &ctx->ifc_txqs[qid].ift_task;
5027 		tqg = qgroup_if_io_tqg;
5028 		fn = _task_fn_tx;
5029 		if (irq != NULL)
5030 			irq_num = rman_get_start(irq->ii_res);
5031 		break;
5032 	case IFLIB_INTR_RX:
5033 		q = &ctx->ifc_rxqs[qid];
5034 		gtask = &ctx->ifc_rxqs[qid].ifr_task;
5035 		tqg = qgroup_if_io_tqg;
5036 		fn = _task_fn_rx;
5037 		if (irq != NULL)
5038 			irq_num = rman_get_start(irq->ii_res);
5039 		break;
5040 	case IFLIB_INTR_IOV:
5041 		q = ctx;
5042 		gtask = &ctx->ifc_vflr_task;
5043 		tqg = qgroup_if_config_tqg;
5044 		fn = _task_fn_iov;
5045 		break;
5046 	default:
5047 		panic("unknown net intr type");
5048 	}
5049 	GROUPTASK_INIT(gtask, 0, fn, q);
5050 	taskqgroup_attach(tqg, gtask, q, irq_num, name);
5051 }
5052 
5053 void
5054 iflib_irq_free(if_ctx_t ctx, if_irq_t irq)
5055 {
5056 	if (irq->ii_tag)
5057 		bus_teardown_intr(ctx->ifc_dev, irq->ii_res, irq->ii_tag);
5058 
5059 	if (irq->ii_res)
5060 		bus_release_resource(ctx->ifc_dev, SYS_RES_IRQ, irq->ii_rid, irq->ii_res);
5061 }
5062 
5063 static int
5064 iflib_legacy_setup(if_ctx_t ctx, driver_filter_t filter, void *filter_arg, int *rid, char *name)
5065 {
5066 	iflib_txq_t txq = ctx->ifc_txqs;
5067 	iflib_rxq_t rxq = ctx->ifc_rxqs;
5068 	if_irq_t irq = &ctx->ifc_legacy_irq;
5069 	iflib_filter_info_t info;
5070 	struct grouptask *gtask;
5071 	struct taskqgroup *tqg;
5072 	gtask_fn_t *fn;
5073 	int tqrid;
5074 	void *q;
5075 	int err;
5076 
5077 	q = &ctx->ifc_rxqs[0];
5078 	info = &rxq[0].ifr_filter_info;
5079 	gtask = &rxq[0].ifr_task;
5080 	tqg = qgroup_if_io_tqg;
5081 	tqrid = irq->ii_rid = *rid;
5082 	fn = _task_fn_rx;
5083 
5084 	ctx->ifc_flags |= IFC_LEGACY;
5085 	info->ifi_filter = filter;
5086 	info->ifi_filter_arg = filter_arg;
5087 	info->ifi_task = gtask;
5088 	info->ifi_ctx = ctx;
5089 
5090 	/* We allocate a single interrupt resource */
5091 	if ((err = _iflib_irq_alloc(ctx, irq, tqrid, iflib_fast_intr_ctx, NULL, info, name)) != 0)
5092 		return (err);
5093 	GROUPTASK_INIT(gtask, 0, fn, q);
5094 	taskqgroup_attach(tqg, gtask, q, tqrid, name);
5095 
5096 	GROUPTASK_INIT(&txq->ift_task, 0, _task_fn_tx, txq);
5097 	taskqgroup_attach(qgroup_if_io_tqg, &txq->ift_task, txq, tqrid, "tx");
5098 	return (0);
5099 }
5100 
5101 void
5102 iflib_led_create(if_ctx_t ctx)
5103 {
5104 
5105 	ctx->ifc_led_dev = led_create(iflib_led_func, ctx,
5106 	    device_get_nameunit(ctx->ifc_dev));
5107 }
5108 
5109 void
5110 iflib_tx_intr_deferred(if_ctx_t ctx, int txqid)
5111 {
5112 
5113 	GROUPTASK_ENQUEUE(&ctx->ifc_txqs[txqid].ift_task);
5114 }
5115 
5116 void
5117 iflib_rx_intr_deferred(if_ctx_t ctx, int rxqid)
5118 {
5119 
5120 	GROUPTASK_ENQUEUE(&ctx->ifc_rxqs[rxqid].ifr_task);
5121 }
5122 
5123 void
5124 iflib_admin_intr_deferred(if_ctx_t ctx)
5125 {
5126 #ifdef INVARIANTS
5127 	struct grouptask *gtask;
5128 
5129 	gtask = &ctx->ifc_admin_task;
5130 	MPASS(gtask != NULL && gtask->gt_taskqueue != NULL);
5131 #endif
5132 
5133 	GROUPTASK_ENQUEUE(&ctx->ifc_admin_task);
5134 }
5135 
5136 void
5137 iflib_iov_intr_deferred(if_ctx_t ctx)
5138 {
5139 
5140 	GROUPTASK_ENQUEUE(&ctx->ifc_vflr_task);
5141 }
5142 
5143 void
5144 iflib_io_tqg_attach(struct grouptask *gt, void *uniq, int cpu, char *name)
5145 {
5146 
5147 	taskqgroup_attach_cpu(qgroup_if_io_tqg, gt, uniq, cpu, -1, name);
5148 }
5149 
5150 void
5151 iflib_config_gtask_init(if_ctx_t ctx, struct grouptask *gtask, gtask_fn_t *fn,
5152 	char *name)
5153 {
5154 
5155 	GROUPTASK_INIT(gtask, 0, fn, ctx);
5156 	taskqgroup_attach(qgroup_if_config_tqg, gtask, gtask, -1, name);
5157 }
5158 
5159 void
5160 iflib_config_gtask_deinit(struct grouptask *gtask)
5161 {
5162 
5163 	taskqgroup_detach(qgroup_if_config_tqg, gtask);
5164 }
5165 
5166 void
5167 iflib_link_state_change(if_ctx_t ctx, int link_state, uint64_t baudrate)
5168 {
5169 	if_t ifp = ctx->ifc_ifp;
5170 	iflib_txq_t txq = ctx->ifc_txqs;
5171 
5172 	if_setbaudrate(ifp, baudrate);
5173 	if (baudrate >= IF_Gbps(10))
5174 		ctx->ifc_flags |= IFC_PREFETCH;
5175 
5176 	/* If link down, disable watchdog */
5177 	if ((ctx->ifc_link_state == LINK_STATE_UP) && (link_state == LINK_STATE_DOWN)) {
5178 		for (int i = 0; i < ctx->ifc_softc_ctx.isc_ntxqsets; i++, txq++)
5179 			txq->ift_qstatus = IFLIB_QUEUE_IDLE;
5180 	}
5181 	ctx->ifc_link_state = link_state;
5182 	if_link_state_change(ifp, link_state);
5183 }
5184 
5185 static int
5186 iflib_tx_credits_update(if_ctx_t ctx, iflib_txq_t txq)
5187 {
5188 	int credits;
5189 #ifdef INVARIANTS
5190 	int credits_pre = txq->ift_cidx_processed;
5191 #endif
5192 
5193 	if (ctx->isc_txd_credits_update == NULL)
5194 		return (0);
5195 
5196 	if ((credits = ctx->isc_txd_credits_update(ctx->ifc_softc, txq->ift_id, true)) == 0)
5197 		return (0);
5198 
5199 	txq->ift_processed += credits;
5200 	txq->ift_cidx_processed += credits;
5201 
5202 	MPASS(credits_pre + credits == txq->ift_cidx_processed);
5203 	if (txq->ift_cidx_processed >= txq->ift_size)
5204 		txq->ift_cidx_processed -= txq->ift_size;
5205 	return (credits);
5206 }
5207 
5208 static int
5209 iflib_rxd_avail(if_ctx_t ctx, iflib_rxq_t rxq, qidx_t cidx, qidx_t budget)
5210 {
5211 
5212 	return (ctx->isc_rxd_available(ctx->ifc_softc, rxq->ifr_id, cidx,
5213 	    budget));
5214 }
5215 
5216 void
5217 iflib_add_int_delay_sysctl(if_ctx_t ctx, const char *name,
5218 	const char *description, if_int_delay_info_t info,
5219 	int offset, int value)
5220 {
5221 	info->iidi_ctx = ctx;
5222 	info->iidi_offset = offset;
5223 	info->iidi_value = value;
5224 	SYSCTL_ADD_PROC(device_get_sysctl_ctx(ctx->ifc_dev),
5225 	    SYSCTL_CHILDREN(device_get_sysctl_tree(ctx->ifc_dev)),
5226 	    OID_AUTO, name, CTLTYPE_INT|CTLFLAG_RW,
5227 	    info, 0, iflib_sysctl_int_delay, "I", description);
5228 }
5229 
5230 struct mtx *
5231 iflib_ctx_lock_get(if_ctx_t ctx)
5232 {
5233 
5234 	return (&ctx->ifc_mtx);
5235 }
5236 
5237 static int
5238 iflib_msix_init(if_ctx_t ctx)
5239 {
5240 	device_t dev = ctx->ifc_dev;
5241 	if_shared_ctx_t sctx = ctx->ifc_sctx;
5242 	if_softc_ctx_t scctx = &ctx->ifc_softc_ctx;
5243 	int vectors, queues, rx_queues, tx_queues, queuemsgs, msgs;
5244 	int iflib_num_tx_queues, iflib_num_rx_queues;
5245 	int err, admincnt, bar;
5246 
5247 	iflib_num_tx_queues = scctx->isc_ntxqsets;
5248 	iflib_num_rx_queues = scctx->isc_nrxqsets;
5249 
5250 	device_printf(dev, "msix_init qsets capped at %d\n", iflib_num_tx_queues);
5251 
5252 	bar = ctx->ifc_softc_ctx.isc_msix_bar;
5253 	admincnt = sctx->isc_admin_intrcnt;
5254 	/* Override by tuneable */
5255 	if (scctx->isc_disable_msix)
5256 		goto msi;
5257 
5258 	/*
5259 	** When used in a virtualized environment
5260 	** PCI BUSMASTER capability may not be set
5261 	** so explicity set it here and rewrite
5262 	** the ENABLE in the MSIX control register
5263 	** at this point to cause the host to
5264 	** successfully initialize us.
5265 	*/
5266 	{
5267 		int msix_ctrl, rid;
5268 
5269  		pci_enable_busmaster(dev);
5270 		rid = 0;
5271 		if (pci_find_cap(dev, PCIY_MSIX, &rid) == 0 && rid != 0) {
5272 			rid += PCIR_MSIX_CTRL;
5273 			msix_ctrl = pci_read_config(dev, rid, 2);
5274 			msix_ctrl |= PCIM_MSIXCTRL_MSIX_ENABLE;
5275 			pci_write_config(dev, rid, msix_ctrl, 2);
5276 		} else {
5277 			device_printf(dev, "PCIY_MSIX capability not found; "
5278 			                   "or rid %d == 0.\n", rid);
5279 			goto msi;
5280 		}
5281 	}
5282 
5283 	/*
5284 	 * bar == -1 => "trust me I know what I'm doing"
5285 	 * Some drivers are for hardware that is so shoddily
5286 	 * documented that no one knows which bars are which
5287 	 * so the developer has to map all bars. This hack
5288 	 * allows shoddy garbage to use msix in this framework.
5289 	 */
5290 	if (bar != -1) {
5291 		ctx->ifc_msix_mem = bus_alloc_resource_any(dev,
5292 	            SYS_RES_MEMORY, &bar, RF_ACTIVE);
5293 		if (ctx->ifc_msix_mem == NULL) {
5294 			/* May not be enabled */
5295 			device_printf(dev, "Unable to map MSIX table \n");
5296 			goto msi;
5297 		}
5298 	}
5299 	/* First try MSI/X */
5300 	if ((msgs = pci_msix_count(dev)) == 0) { /* system has msix disabled */
5301 		device_printf(dev, "System has MSIX disabled \n");
5302 		bus_release_resource(dev, SYS_RES_MEMORY,
5303 		    bar, ctx->ifc_msix_mem);
5304 		ctx->ifc_msix_mem = NULL;
5305 		goto msi;
5306 	}
5307 #if IFLIB_DEBUG
5308 	/* use only 1 qset in debug mode */
5309 	queuemsgs = min(msgs - admincnt, 1);
5310 #else
5311 	queuemsgs = msgs - admincnt;
5312 #endif
5313 	if (bus_get_cpus(dev, INTR_CPUS, sizeof(ctx->ifc_cpus), &ctx->ifc_cpus) == 0) {
5314 #ifdef RSS
5315 		queues = imin(queuemsgs, rss_getnumbuckets());
5316 #else
5317 		queues = queuemsgs;
5318 #endif
5319 		queues = imin(CPU_COUNT(&ctx->ifc_cpus), queues);
5320 		device_printf(dev, "pxm cpus: %d queue msgs: %d admincnt: %d\n",
5321 					  CPU_COUNT(&ctx->ifc_cpus), queuemsgs, admincnt);
5322 	} else {
5323 		device_printf(dev, "Unable to fetch CPU list\n");
5324 		/* Figure out a reasonable auto config value */
5325 		queues = min(queuemsgs, mp_ncpus);
5326 	}
5327 #ifdef  RSS
5328 	/* If we're doing RSS, clamp at the number of RSS buckets */
5329 	if (queues > rss_getnumbuckets())
5330 		queues = rss_getnumbuckets();
5331 #endif
5332 	if (iflib_num_rx_queues > 0 && iflib_num_rx_queues < queuemsgs - admincnt)
5333 		rx_queues = iflib_num_rx_queues;
5334 	else
5335 		rx_queues = queues;
5336 	/*
5337 	 * We want this to be all logical CPUs by default
5338 	 */
5339 	if (iflib_num_tx_queues > 0 && iflib_num_tx_queues < queues)
5340 		tx_queues = iflib_num_tx_queues;
5341 	else
5342 		tx_queues = mp_ncpus;
5343 
5344 	if (ctx->ifc_sysctl_qs_eq_override == 0) {
5345 #ifdef INVARIANTS
5346 		if (tx_queues != rx_queues)
5347 			device_printf(dev, "queue equality override not set, capping rx_queues at %d and tx_queues at %d\n",
5348 				      min(rx_queues, tx_queues), min(rx_queues, tx_queues));
5349 #endif
5350 		tx_queues = min(rx_queues, tx_queues);
5351 		rx_queues = min(rx_queues, tx_queues);
5352 	}
5353 
5354 	device_printf(dev, "using %d rx queues %d tx queues \n", rx_queues, tx_queues);
5355 
5356 	vectors = rx_queues + admincnt;
5357 	if ((err = pci_alloc_msix(dev, &vectors)) == 0) {
5358 		device_printf(dev,
5359 					  "Using MSIX interrupts with %d vectors\n", vectors);
5360 		scctx->isc_vectors = vectors;
5361 		scctx->isc_nrxqsets = rx_queues;
5362 		scctx->isc_ntxqsets = tx_queues;
5363 		scctx->isc_intr = IFLIB_INTR_MSIX;
5364 
5365 		return (vectors);
5366 	} else {
5367 		device_printf(dev, "failed to allocate %d msix vectors, err: %d - using MSI\n", vectors, err);
5368 	}
5369 msi:
5370 	vectors = pci_msi_count(dev);
5371 	scctx->isc_nrxqsets = 1;
5372 	scctx->isc_ntxqsets = 1;
5373 	scctx->isc_vectors = vectors;
5374 	if (vectors == 1 && pci_alloc_msi(dev, &vectors) == 0) {
5375 		device_printf(dev,"Using an MSI interrupt\n");
5376 		scctx->isc_intr = IFLIB_INTR_MSI;
5377 	} else {
5378 		device_printf(dev,"Using a Legacy interrupt\n");
5379 		scctx->isc_intr = IFLIB_INTR_LEGACY;
5380 	}
5381 
5382 	return (vectors);
5383 }
5384 
5385 char * ring_states[] = { "IDLE", "BUSY", "STALLED", "ABDICATED" };
5386 
5387 static int
5388 mp_ring_state_handler(SYSCTL_HANDLER_ARGS)
5389 {
5390 	int rc;
5391 	uint16_t *state = ((uint16_t *)oidp->oid_arg1);
5392 	struct sbuf *sb;
5393 	char *ring_state = "UNKNOWN";
5394 
5395 	/* XXX needed ? */
5396 	rc = sysctl_wire_old_buffer(req, 0);
5397 	MPASS(rc == 0);
5398 	if (rc != 0)
5399 		return (rc);
5400 	sb = sbuf_new_for_sysctl(NULL, NULL, 80, req);
5401 	MPASS(sb != NULL);
5402 	if (sb == NULL)
5403 		return (ENOMEM);
5404 	if (state[3] <= 3)
5405 		ring_state = ring_states[state[3]];
5406 
5407 	sbuf_printf(sb, "pidx_head: %04hd pidx_tail: %04hd cidx: %04hd state: %s",
5408 		    state[0], state[1], state[2], ring_state);
5409 	rc = sbuf_finish(sb);
5410 	sbuf_delete(sb);
5411         return(rc);
5412 }
5413 
5414 enum iflib_ndesc_handler {
5415 	IFLIB_NTXD_HANDLER,
5416 	IFLIB_NRXD_HANDLER,
5417 };
5418 
5419 static int
5420 mp_ndesc_handler(SYSCTL_HANDLER_ARGS)
5421 {
5422 	if_ctx_t ctx = (void *)arg1;
5423 	enum iflib_ndesc_handler type = arg2;
5424 	char buf[256] = {0};
5425 	qidx_t *ndesc;
5426 	char *p, *next;
5427 	int nqs, rc, i;
5428 
5429 	MPASS(type == IFLIB_NTXD_HANDLER || type == IFLIB_NRXD_HANDLER);
5430 
5431 	nqs = 8;
5432 	switch(type) {
5433 	case IFLIB_NTXD_HANDLER:
5434 		ndesc = ctx->ifc_sysctl_ntxds;
5435 		if (ctx->ifc_sctx)
5436 			nqs = ctx->ifc_sctx->isc_ntxqs;
5437 		break;
5438 	case IFLIB_NRXD_HANDLER:
5439 		ndesc = ctx->ifc_sysctl_nrxds;
5440 		if (ctx->ifc_sctx)
5441 			nqs = ctx->ifc_sctx->isc_nrxqs;
5442 		break;
5443 	}
5444 	if (nqs == 0)
5445 		nqs = 8;
5446 
5447 	for (i=0; i<8; i++) {
5448 		if (i >= nqs)
5449 			break;
5450 		if (i)
5451 			strcat(buf, ",");
5452 		sprintf(strchr(buf, 0), "%d", ndesc[i]);
5453 	}
5454 
5455 	rc = sysctl_handle_string(oidp, buf, sizeof(buf), req);
5456 	if (rc || req->newptr == NULL)
5457 		return rc;
5458 
5459 	for (i = 0, next = buf, p = strsep(&next, " ,"); i < 8 && p;
5460 	    i++, p = strsep(&next, " ,")) {
5461 		ndesc[i] = strtoul(p, NULL, 10);
5462 	}
5463 
5464 	return(rc);
5465 }
5466 
5467 #define NAME_BUFLEN 32
5468 static void
5469 iflib_add_device_sysctl_pre(if_ctx_t ctx)
5470 {
5471         device_t dev = iflib_get_dev(ctx);
5472 	struct sysctl_oid_list *child, *oid_list;
5473 	struct sysctl_ctx_list *ctx_list;
5474 	struct sysctl_oid *node;
5475 
5476 	ctx_list = device_get_sysctl_ctx(dev);
5477 	child = SYSCTL_CHILDREN(device_get_sysctl_tree(dev));
5478 	ctx->ifc_sysctl_node = node = SYSCTL_ADD_NODE(ctx_list, child, OID_AUTO, "iflib",
5479 						      CTLFLAG_RD, NULL, "IFLIB fields");
5480 	oid_list = SYSCTL_CHILDREN(node);
5481 
5482 	SYSCTL_ADD_STRING(ctx_list, oid_list, OID_AUTO, "driver_version",
5483 		       CTLFLAG_RD, ctx->ifc_sctx->isc_driver_version, 0,
5484 		       "driver version");
5485 
5486 	SYSCTL_ADD_U16(ctx_list, oid_list, OID_AUTO, "override_ntxqs",
5487 		       CTLFLAG_RWTUN, &ctx->ifc_sysctl_ntxqs, 0,
5488 			"# of txqs to use, 0 => use default #");
5489 	SYSCTL_ADD_U16(ctx_list, oid_list, OID_AUTO, "override_nrxqs",
5490 		       CTLFLAG_RWTUN, &ctx->ifc_sysctl_nrxqs, 0,
5491 			"# of rxqs to use, 0 => use default #");
5492 	SYSCTL_ADD_U16(ctx_list, oid_list, OID_AUTO, "override_qs_enable",
5493 		       CTLFLAG_RWTUN, &ctx->ifc_sysctl_qs_eq_override, 0,
5494                        "permit #txq != #rxq");
5495 	SYSCTL_ADD_INT(ctx_list, oid_list, OID_AUTO, "disable_msix",
5496                       CTLFLAG_RWTUN, &ctx->ifc_softc_ctx.isc_disable_msix, 0,
5497                       "disable MSIX (default 0)");
5498 	SYSCTL_ADD_U16(ctx_list, oid_list, OID_AUTO, "rx_budget",
5499 		       CTLFLAG_RWTUN, &ctx->ifc_sysctl_rx_budget, 0,
5500                        "set the rx budget");
5501 
5502 	/* XXX change for per-queue sizes */
5503 	SYSCTL_ADD_PROC(ctx_list, oid_list, OID_AUTO, "override_ntxds",
5504 		       CTLTYPE_STRING|CTLFLAG_RWTUN, ctx, IFLIB_NTXD_HANDLER,
5505                        mp_ndesc_handler, "A",
5506                        "list of # of tx descriptors to use, 0 = use default #");
5507 	SYSCTL_ADD_PROC(ctx_list, oid_list, OID_AUTO, "override_nrxds",
5508 		       CTLTYPE_STRING|CTLFLAG_RWTUN, ctx, IFLIB_NRXD_HANDLER,
5509                        mp_ndesc_handler, "A",
5510                        "list of # of rx descriptors to use, 0 = use default #");
5511 }
5512 
5513 static void
5514 iflib_add_device_sysctl_post(if_ctx_t ctx)
5515 {
5516 	if_shared_ctx_t sctx = ctx->ifc_sctx;
5517 	if_softc_ctx_t scctx = &ctx->ifc_softc_ctx;
5518         device_t dev = iflib_get_dev(ctx);
5519 	struct sysctl_oid_list *child;
5520 	struct sysctl_ctx_list *ctx_list;
5521 	iflib_fl_t fl;
5522 	iflib_txq_t txq;
5523 	iflib_rxq_t rxq;
5524 	int i, j;
5525 	char namebuf[NAME_BUFLEN];
5526 	char *qfmt;
5527 	struct sysctl_oid *queue_node, *fl_node, *node;
5528 	struct sysctl_oid_list *queue_list, *fl_list;
5529 	ctx_list = device_get_sysctl_ctx(dev);
5530 
5531 	node = ctx->ifc_sysctl_node;
5532 	child = SYSCTL_CHILDREN(node);
5533 
5534 	if (scctx->isc_ntxqsets > 100)
5535 		qfmt = "txq%03d";
5536 	else if (scctx->isc_ntxqsets > 10)
5537 		qfmt = "txq%02d";
5538 	else
5539 		qfmt = "txq%d";
5540 	for (i = 0, txq = ctx->ifc_txqs; i < scctx->isc_ntxqsets; i++, txq++) {
5541 		snprintf(namebuf, NAME_BUFLEN, qfmt, i);
5542 		queue_node = SYSCTL_ADD_NODE(ctx_list, child, OID_AUTO, namebuf,
5543 					     CTLFLAG_RD, NULL, "Queue Name");
5544 		queue_list = SYSCTL_CHILDREN(queue_node);
5545 #if MEMORY_LOGGING
5546 		SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "txq_dequeued",
5547 				CTLFLAG_RD,
5548 				&txq->ift_dequeued, "total mbufs freed");
5549 		SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "txq_enqueued",
5550 				CTLFLAG_RD,
5551 				&txq->ift_enqueued, "total mbufs enqueued");
5552 #endif
5553 		SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "mbuf_defrag",
5554 				   CTLFLAG_RD,
5555 				   &txq->ift_mbuf_defrag, "# of times m_defrag was called");
5556 		SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "m_pullups",
5557 				   CTLFLAG_RD,
5558 				   &txq->ift_pullups, "# of times m_pullup was called");
5559 		SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "mbuf_defrag_failed",
5560 				   CTLFLAG_RD,
5561 				   &txq->ift_mbuf_defrag_failed, "# of times m_defrag failed");
5562 		SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "no_desc_avail",
5563 				   CTLFLAG_RD,
5564 				   &txq->ift_no_desc_avail, "# of times no descriptors were available");
5565 		SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "tx_map_failed",
5566 				   CTLFLAG_RD,
5567 				   &txq->ift_map_failed, "# of times dma map failed");
5568 		SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "txd_encap_efbig",
5569 				   CTLFLAG_RD,
5570 				   &txq->ift_txd_encap_efbig, "# of times txd_encap returned EFBIG");
5571 		SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "no_tx_dma_setup",
5572 				   CTLFLAG_RD,
5573 				   &txq->ift_no_tx_dma_setup, "# of times map failed for other than EFBIG");
5574 		SYSCTL_ADD_U16(ctx_list, queue_list, OID_AUTO, "txq_pidx",
5575 				   CTLFLAG_RD,
5576 				   &txq->ift_pidx, 1, "Producer Index");
5577 		SYSCTL_ADD_U16(ctx_list, queue_list, OID_AUTO, "txq_cidx",
5578 				   CTLFLAG_RD,
5579 				   &txq->ift_cidx, 1, "Consumer Index");
5580 		SYSCTL_ADD_U16(ctx_list, queue_list, OID_AUTO, "txq_cidx_processed",
5581 				   CTLFLAG_RD,
5582 				   &txq->ift_cidx_processed, 1, "Consumer Index seen by credit update");
5583 		SYSCTL_ADD_U16(ctx_list, queue_list, OID_AUTO, "txq_in_use",
5584 				   CTLFLAG_RD,
5585 				   &txq->ift_in_use, 1, "descriptors in use");
5586 		SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "txq_processed",
5587 				   CTLFLAG_RD,
5588 				   &txq->ift_processed, "descriptors procesed for clean");
5589 		SYSCTL_ADD_QUAD(ctx_list, queue_list, OID_AUTO, "txq_cleaned",
5590 				   CTLFLAG_RD,
5591 				   &txq->ift_cleaned, "total cleaned");
5592 		SYSCTL_ADD_PROC(ctx_list, queue_list, OID_AUTO, "ring_state",
5593 				CTLTYPE_STRING | CTLFLAG_RD, __DEVOLATILE(uint64_t *, &txq->ift_br->state),
5594 				0, mp_ring_state_handler, "A", "soft ring state");
5595 		SYSCTL_ADD_COUNTER_U64(ctx_list, queue_list, OID_AUTO, "r_enqueues",
5596 				       CTLFLAG_RD, &txq->ift_br->enqueues,
5597 				       "# of enqueues to the mp_ring for this queue");
5598 		SYSCTL_ADD_COUNTER_U64(ctx_list, queue_list, OID_AUTO, "r_drops",
5599 				       CTLFLAG_RD, &txq->ift_br->drops,
5600 				       "# of drops in the mp_ring for this queue");
5601 		SYSCTL_ADD_COUNTER_U64(ctx_list, queue_list, OID_AUTO, "r_starts",
5602 				       CTLFLAG_RD, &txq->ift_br->starts,
5603 				       "# of normal consumer starts in the mp_ring for this queue");
5604 		SYSCTL_ADD_COUNTER_U64(ctx_list, queue_list, OID_AUTO, "r_stalls",
5605 				       CTLFLAG_RD, &txq->ift_br->stalls,
5606 					       "# of consumer stalls in the mp_ring for this queue");
5607 		SYSCTL_ADD_COUNTER_U64(ctx_list, queue_list, OID_AUTO, "r_restarts",
5608 			       CTLFLAG_RD, &txq->ift_br->restarts,
5609 				       "# of consumer restarts in the mp_ring for this queue");
5610 		SYSCTL_ADD_COUNTER_U64(ctx_list, queue_list, OID_AUTO, "r_abdications",
5611 				       CTLFLAG_RD, &txq->ift_br->abdications,
5612 				       "# of consumer abdications in the mp_ring for this queue");
5613 	}
5614 
5615 	if (scctx->isc_nrxqsets > 100)
5616 		qfmt = "rxq%03d";
5617 	else if (scctx->isc_nrxqsets > 10)
5618 		qfmt = "rxq%02d";
5619 	else
5620 		qfmt = "rxq%d";
5621 	for (i = 0, rxq = ctx->ifc_rxqs; i < scctx->isc_nrxqsets; i++, rxq++) {
5622 		snprintf(namebuf, NAME_BUFLEN, qfmt, i);
5623 		queue_node = SYSCTL_ADD_NODE(ctx_list, child, OID_AUTO, namebuf,
5624 					     CTLFLAG_RD, NULL, "Queue Name");
5625 		queue_list = SYSCTL_CHILDREN(queue_node);
5626 		if (sctx->isc_flags & IFLIB_HAS_RXCQ) {
5627 			SYSCTL_ADD_U16(ctx_list, queue_list, OID_AUTO, "rxq_cq_pidx",
5628 				       CTLFLAG_RD,
5629 				       &rxq->ifr_cq_pidx, 1, "Producer Index");
5630 			SYSCTL_ADD_U16(ctx_list, queue_list, OID_AUTO, "rxq_cq_cidx",
5631 				       CTLFLAG_RD,
5632 				       &rxq->ifr_cq_cidx, 1, "Consumer Index");
5633 		}
5634 
5635 		for (j = 0, fl = rxq->ifr_fl; j < rxq->ifr_nfl; j++, fl++) {
5636 			snprintf(namebuf, NAME_BUFLEN, "rxq_fl%d", j);
5637 			fl_node = SYSCTL_ADD_NODE(ctx_list, queue_list, OID_AUTO, namebuf,
5638 						     CTLFLAG_RD, NULL, "freelist Name");
5639 			fl_list = SYSCTL_CHILDREN(fl_node);
5640 			SYSCTL_ADD_U16(ctx_list, fl_list, OID_AUTO, "pidx",
5641 				       CTLFLAG_RD,
5642 				       &fl->ifl_pidx, 1, "Producer Index");
5643 			SYSCTL_ADD_U16(ctx_list, fl_list, OID_AUTO, "cidx",
5644 				       CTLFLAG_RD,
5645 				       &fl->ifl_cidx, 1, "Consumer Index");
5646 			SYSCTL_ADD_U16(ctx_list, fl_list, OID_AUTO, "credits",
5647 				       CTLFLAG_RD,
5648 				       &fl->ifl_credits, 1, "credits available");
5649 #if MEMORY_LOGGING
5650 			SYSCTL_ADD_QUAD(ctx_list, fl_list, OID_AUTO, "fl_m_enqueued",
5651 					CTLFLAG_RD,
5652 					&fl->ifl_m_enqueued, "mbufs allocated");
5653 			SYSCTL_ADD_QUAD(ctx_list, fl_list, OID_AUTO, "fl_m_dequeued",
5654 					CTLFLAG_RD,
5655 					&fl->ifl_m_dequeued, "mbufs freed");
5656 			SYSCTL_ADD_QUAD(ctx_list, fl_list, OID_AUTO, "fl_cl_enqueued",
5657 					CTLFLAG_RD,
5658 					&fl->ifl_cl_enqueued, "clusters allocated");
5659 			SYSCTL_ADD_QUAD(ctx_list, fl_list, OID_AUTO, "fl_cl_dequeued",
5660 					CTLFLAG_RD,
5661 					&fl->ifl_cl_dequeued, "clusters freed");
5662 #endif
5663 
5664 		}
5665 	}
5666 
5667 }
5668 
5669 #ifndef __NO_STRICT_ALIGNMENT
5670 static struct mbuf *
5671 iflib_fixup_rx(struct mbuf *m)
5672 {
5673 	struct mbuf *n;
5674 
5675 	if (m->m_len <= (MCLBYTES - ETHER_HDR_LEN)) {
5676 		bcopy(m->m_data, m->m_data + ETHER_HDR_LEN, m->m_len);
5677 		m->m_data += ETHER_HDR_LEN;
5678 		n = m;
5679 	} else {
5680 		MGETHDR(n, M_NOWAIT, MT_DATA);
5681 		if (n == NULL) {
5682 			m_freem(m);
5683 			return (NULL);
5684 		}
5685 		bcopy(m->m_data, n->m_data, ETHER_HDR_LEN);
5686 		m->m_data += ETHER_HDR_LEN;
5687 		m->m_len -= ETHER_HDR_LEN;
5688 		n->m_len = ETHER_HDR_LEN;
5689 		M_MOVE_PKTHDR(n, m);
5690 		n->m_next = m;
5691 	}
5692 	return (n);
5693 }
5694 #endif
5695