if_igc.c revision 1.10
1/*	$OpenBSD: if_igc.c,v 1.10 2022/11/11 16:41:44 mbuhl Exp $	*/
2/*-
3 * SPDX-License-Identifier: BSD-2-Clause
4 *
5 * Copyright (c) 2016 Nicole Graziano <nicole@nextbsd.org>
6 * All rights reserved.
7 * Copyright (c) 2021 Rubicon Communications, LLC (Netgate)
8 *
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
11 * are met:
12 * 1. Redistributions of source code must retain the above copyright
13 *    notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 *    notice, this list of conditions and the following disclaimer in the
16 *    documentation and/or other materials provided with the distribution.
17 *
18 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
19 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21 * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
22 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28 * SUCH DAMAGE.
29 */
30
31#include "bpfilter.h"
32#include "vlan.h"
33
34#include <sys/param.h>
35#include <sys/systm.h>
36#include <sys/sockio.h>
37#include <sys/mbuf.h>
38#include <sys/malloc.h>
39#include <sys/kernel.h>
40#include <sys/socket.h>
41#include <sys/device.h>
42#include <sys/endian.h>
43#include <sys/intrmap.h>
44
45#include <net/if.h>
46#include <net/if_media.h>
47#include <net/toeplitz.h>
48
49#include <netinet/in.h>
50#include <netinet/if_ether.h>
51#include <netinet/ip.h>
52#include <netinet/ip6.h>
53
54#if NBPFILTER > 0
55#include <net/bpf.h>
56#endif
57
58#include <machine/bus.h>
59#include <machine/intr.h>
60
61#include <dev/pci/pcivar.h>
62#include <dev/pci/pcireg.h>
63#include <dev/pci/pcidevs.h>
64#include <dev/pci/if_igc.h>
65#include <dev/pci/igc_hw.h>
66
67const struct pci_matchid igc_devices[] = {
68	{ PCI_VENDOR_INTEL, PCI_PRODUCT_INTEL_I220_V },
69	{ PCI_VENDOR_INTEL, PCI_PRODUCT_INTEL_I221_V },
70	{ PCI_VENDOR_INTEL, PCI_PRODUCT_INTEL_I225_BLANK_NVM },
71	{ PCI_VENDOR_INTEL, PCI_PRODUCT_INTEL_I225_I },
72	{ PCI_VENDOR_INTEL, PCI_PRODUCT_INTEL_I225_IT },
73	{ PCI_VENDOR_INTEL, PCI_PRODUCT_INTEL_I225_K },
74	{ PCI_VENDOR_INTEL, PCI_PRODUCT_INTEL_I225_K2 },
75	{ PCI_VENDOR_INTEL, PCI_PRODUCT_INTEL_I225_LM },
76	{ PCI_VENDOR_INTEL, PCI_PRODUCT_INTEL_I225_LMVP },
77	{ PCI_VENDOR_INTEL, PCI_PRODUCT_INTEL_I225_V },
78	{ PCI_VENDOR_INTEL, PCI_PRODUCT_INTEL_I226_BLANK_NVM },
79	{ PCI_VENDOR_INTEL, PCI_PRODUCT_INTEL_I226_IT },
80	{ PCI_VENDOR_INTEL, PCI_PRODUCT_INTEL_I226_LM },
81	{ PCI_VENDOR_INTEL, PCI_PRODUCT_INTEL_I226_K },
82	{ PCI_VENDOR_INTEL, PCI_PRODUCT_INTEL_I226_V }
83};
84
85/*********************************************************************
86 *  Function Prototypes
87 *********************************************************************/
88int	igc_match(struct device *, void *, void *);
89void	igc_attach(struct device *, struct device *, void *);
90int	igc_detach(struct device *, int);
91
92void	igc_identify_hardware(struct igc_softc *);
93int	igc_allocate_pci_resources(struct igc_softc *);
94int	igc_allocate_queues(struct igc_softc *);
95void	igc_free_pci_resources(struct igc_softc *);
96void	igc_reset(struct igc_softc *);
97void	igc_init_dmac(struct igc_softc *, uint32_t);
98int	igc_allocate_msix(struct igc_softc *);
99void	igc_setup_msix(struct igc_softc *);
100int	igc_dma_malloc(struct igc_softc *, bus_size_t, struct igc_dma_alloc *);
101void	igc_dma_free(struct igc_softc *, struct igc_dma_alloc *);
102void	igc_setup_interface(struct igc_softc *);
103
104void	igc_init(void *);
105void	igc_start(struct ifqueue *);
106int	igc_txeof(struct tx_ring *);
107void	igc_stop(struct igc_softc *);
108int	igc_ioctl(struct ifnet *, u_long, caddr_t);
109int	igc_rxrinfo(struct igc_softc *, struct if_rxrinfo *);
110int	igc_rxfill(struct rx_ring *);
111void	igc_rxrefill(void *);
112int	igc_rxeof(struct rx_ring *);
113void	igc_rx_checksum(uint32_t, struct mbuf *, uint32_t);
114void	igc_watchdog(struct ifnet *);
115void	igc_media_status(struct ifnet *, struct ifmediareq *);
116int	igc_media_change(struct ifnet *);
117void	igc_iff(struct igc_softc *);
118void	igc_update_link_status(struct igc_softc *);
119int	igc_get_buf(struct rx_ring *, int);
120int	igc_tx_ctx_setup(struct tx_ring *, struct mbuf *, int, uint32_t *);
121
122void	igc_configure_queues(struct igc_softc *);
123void	igc_set_queues(struct igc_softc *, uint32_t, uint32_t, int);
124void	igc_enable_queue(struct igc_softc *, uint32_t);
125void	igc_enable_intr(struct igc_softc *);
126void	igc_disable_intr(struct igc_softc *);
127int	igc_intr_link(void *);
128int	igc_intr_queue(void *);
129
130int	igc_allocate_transmit_buffers(struct tx_ring *);
131int	igc_setup_transmit_structures(struct igc_softc *);
132int	igc_setup_transmit_ring(struct tx_ring *);
133void	igc_initialize_transmit_unit(struct igc_softc *);
134void	igc_free_transmit_structures(struct igc_softc *);
135void	igc_free_transmit_buffers(struct tx_ring *);
136int	igc_allocate_receive_buffers(struct rx_ring *);
137int	igc_setup_receive_structures(struct igc_softc *);
138int	igc_setup_receive_ring(struct rx_ring *);
139void	igc_initialize_receive_unit(struct igc_softc *);
140void	igc_free_receive_structures(struct igc_softc *);
141void	igc_free_receive_buffers(struct rx_ring *);
142void	igc_initialize_rss_mapping(struct igc_softc *);
143
144void	igc_get_hw_control(struct igc_softc *);
145void	igc_release_hw_control(struct igc_softc *);
146int	igc_is_valid_ether_addr(uint8_t *);
147
148/*********************************************************************
149 *  OpenBSD Device Interface Entry Points
150 *********************************************************************/
151
152struct cfdriver igc_cd = {
153	NULL, "igc", DV_IFNET
154};
155
156const struct cfattach igc_ca = {
157	sizeof(struct igc_softc), igc_match, igc_attach, igc_detach
158};
159
160/*********************************************************************
161 *  Device identification routine
162 *
163 *  igc_match determines if the driver should be loaded on
164 *  adapter based on PCI vendor/device id of the adapter.
165 *
166 *  return 0 on success, positive on failure
167 *********************************************************************/
168int
169igc_match(struct device *parent, void *match, void *aux)
170{
171	return pci_matchbyid((struct pci_attach_args *)aux, igc_devices,
172	    nitems(igc_devices));
173}
174
175/*********************************************************************
176 *  Device initialization routine
177 *
178 *  The attach entry point is called when the driver is being loaded.
179 *  This routine identifies the type of hardware, allocates all resources
180 *  and initializes the hardware.
181 *
182 *  return 0 on success, positive on failure
183 *********************************************************************/
184void
185igc_attach(struct device *parent, struct device *self, void *aux)
186{
187	struct pci_attach_args *pa = (struct pci_attach_args *)aux;
188	struct igc_softc *sc = (struct igc_softc *)self;
189	struct igc_hw *hw = &sc->hw;
190
191	sc->osdep.os_sc = sc;
192	sc->osdep.os_pa = *pa;
193
194	/* Determine hardware and mac info */
195	igc_identify_hardware(sc);
196
197	sc->num_tx_desc = IGC_DEFAULT_TXD;
198	sc->num_rx_desc = IGC_DEFAULT_RXD;
199
200	 /* Setup PCI resources */
201	if (igc_allocate_pci_resources(sc))
202		 goto err_pci;
203
204	/* Allocate TX/RX queues */
205	if (igc_allocate_queues(sc))
206		 goto err_pci;
207
208	/* Do shared code initialization */
209	if (igc_setup_init_funcs(hw, true)) {
210		printf(": Setup of shared code failed\n");
211		goto err_pci;
212	}
213
214	hw->mac.autoneg = DO_AUTO_NEG;
215	hw->phy.autoneg_wait_to_complete = false;
216	hw->phy.autoneg_advertised = AUTONEG_ADV_DEFAULT;
217
218	/* Copper options. */
219	if (hw->phy.media_type == igc_media_type_copper)
220		hw->phy.mdix = AUTO_ALL_MODES;
221
222	/* Set the max frame size. */
223	sc->hw.mac.max_frame_size = 9234;
224
225	/* Allocate multicast array memory. */
226	sc->mta = mallocarray(ETHER_ADDR_LEN, MAX_NUM_MULTICAST_ADDRESSES,
227	    M_DEVBUF, M_NOWAIT);
228	if (sc->mta == NULL) {
229		printf(": Can not allocate multicast setup array\n");
230		goto err_late;
231	}
232
233	/* Check SOL/IDER usage. */
234	if (igc_check_reset_block(hw))
235		printf(": PHY reset is blocked due to SOL/IDER session\n");
236
237	/* Enable Energy Efficient Ethernet. */
238	sc->hw.dev_spec._i225.eee_disable = true;
239
240	igc_reset_hw(hw);
241
242	/* Make sure we have a good EEPROM before we read from it. */
243	if (igc_validate_nvm_checksum(hw) < 0) {
244		/*
245		 * Some PCI-E parts fail the first check due to
246		 * the link being in sleep state, call it again,
247		 * if it fails a second time its a real issue.
248		 */
249		if (igc_validate_nvm_checksum(hw) < 0) {
250			printf(": The EEPROM checksum is not valid\n");
251			goto err_late;
252		}
253	}
254
255	/* Copy the permanent MAC address out of the EEPROM. */
256	if (igc_read_mac_addr(hw) < 0) {
257		printf(": EEPROM read error while reading MAC address\n");
258		goto err_late;
259	}
260
261	if (!igc_is_valid_ether_addr(hw->mac.addr)) {
262		printf(": Invalid MAC address\n");
263		goto err_late;
264	}
265
266	memcpy(sc->sc_ac.ac_enaddr, sc->hw.mac.addr, ETHER_ADDR_LEN);
267
268	if (igc_allocate_msix(sc))
269		goto err_late;
270
271	/* Setup OS specific network interface. */
272	igc_setup_interface(sc);
273
274	igc_reset(sc);
275	hw->mac.get_link_status = true;
276	igc_update_link_status(sc);
277
278	/* The driver can now take control from firmware. */
279	igc_get_hw_control(sc);
280
281	printf(", address %s\n", ether_sprintf(sc->hw.mac.addr));
282	return;
283
284err_late:
285	igc_release_hw_control(sc);
286err_pci:
287	igc_free_pci_resources(sc);
288	free(sc->mta, M_DEVBUF, ETHER_ADDR_LEN * MAX_NUM_MULTICAST_ADDRESSES);
289}
290
291/*********************************************************************
292 *  Device removal routine
293 *
294 *  The detach entry point is called when the driver is being removed.
295 *  This routine stops the adapter and deallocates all the resources
296 *  that were allocated for driver operation.
297 *
298 *  return 0 on success, positive on failure
299 *********************************************************************/
300int
301igc_detach(struct device *self, int flags)
302{
303	struct igc_softc *sc = (struct igc_softc *)self;
304	struct ifnet *ifp = &sc->sc_ac.ac_if;
305
306	igc_stop(sc);
307
308	igc_phy_hw_reset(&sc->hw);
309	igc_release_hw_control(sc);
310
311	ether_ifdetach(ifp);
312	if_detach(ifp);
313
314	igc_free_pci_resources(sc);
315
316	igc_free_transmit_structures(sc);
317	igc_free_receive_structures(sc);
318	free(sc->mta, M_DEVBUF, ETHER_ADDR_LEN * MAX_NUM_MULTICAST_ADDRESSES);
319
320	return 0;
321}
322
323void
324igc_identify_hardware(struct igc_softc *sc)
325{
326	struct igc_osdep *os = &sc->osdep;
327	struct pci_attach_args *pa = &os->os_pa;
328
329	/* Save off the information about this board. */
330	sc->hw.device_id = PCI_PRODUCT(pa->pa_id);
331
332	/* Do shared code init and setup. */
333	if (igc_set_mac_type(&sc->hw)) {
334		printf(": Setup init failure\n");
335		return;
336        }
337}
338
339int
340igc_allocate_pci_resources(struct igc_softc *sc)
341{
342	struct igc_osdep *os = &sc->osdep;
343	struct pci_attach_args *pa = &os->os_pa;
344	pcireg_t memtype;
345
346	memtype = pci_mapreg_type(pa->pa_pc, pa->pa_tag, IGC_PCIREG);
347	if (pci_mapreg_map(pa, IGC_PCIREG, memtype, 0, &os->os_memt,
348	    &os->os_memh, &os->os_membase, &os->os_memsize, 0)) {
349		printf(": unable to map registers\n");
350		return ENXIO;
351	}
352	sc->hw.hw_addr = (uint8_t *)os->os_membase;
353	sc->hw.back = os;
354
355	igc_setup_msix(sc);
356
357	return 0;
358}
359
360int
361igc_allocate_queues(struct igc_softc *sc)
362{
363	struct igc_queue *iq;
364	struct tx_ring *txr;
365	struct rx_ring *rxr;
366	int i, rsize, rxconf, tsize, txconf;
367
368	/* Allocate the top level queue structs. */
369	sc->queues = mallocarray(sc->sc_nqueues, sizeof(struct igc_queue),
370	    M_DEVBUF, M_NOWAIT | M_ZERO);
371	if (sc->queues == NULL) {
372		printf("%s: unable to allocate queue\n", DEVNAME(sc));
373		goto fail;
374	}
375
376	/* Allocate the TX ring. */
377	sc->tx_rings = mallocarray(sc->sc_nqueues, sizeof(struct tx_ring),
378	    M_DEVBUF, M_NOWAIT | M_ZERO);
379	if (sc->tx_rings == NULL) {
380		printf("%s: unable to allocate TX ring\n", DEVNAME(sc));
381		goto fail;
382	}
383
384	/* Allocate the RX ring. */
385	sc->rx_rings = mallocarray(sc->sc_nqueues, sizeof(struct rx_ring),
386	    M_DEVBUF, M_NOWAIT | M_ZERO);
387	if (sc->rx_rings == NULL) {
388		printf("%s: unable to allocate RX ring\n", DEVNAME(sc));
389		goto rx_fail;
390	}
391
392	txconf = rxconf = 0;
393
394	/* Set up the TX queues. */
395	tsize = roundup2(sc->num_tx_desc * sizeof(union igc_adv_tx_desc),
396	    IGC_DBA_ALIGN);
397	for (i = 0; i < sc->sc_nqueues; i++, txconf++) {
398		txr = &sc->tx_rings[i];
399		txr->sc = sc;
400		txr->me = i;
401
402		if (igc_dma_malloc(sc, tsize, &txr->txdma)) {
403			printf("%s: unable to allocate TX descriptor\n",
404			    DEVNAME(sc));
405			goto err_tx_desc;
406		}
407		txr->tx_base = (union igc_adv_tx_desc *)txr->txdma.dma_vaddr;
408		bzero((void *)txr->tx_base, tsize);
409	}
410
411	/* Set up the RX queues. */
412	rsize = roundup2(sc->num_rx_desc * sizeof(union igc_adv_rx_desc),
413	    IGC_DBA_ALIGN);
414	for (i = 0; i < sc->sc_nqueues; i++, rxconf++) {
415		rxr = &sc->rx_rings[i];
416		rxr->sc = sc;
417		rxr->me = i;
418		timeout_set(&rxr->rx_refill, igc_rxrefill, rxr);
419
420		if (igc_dma_malloc(sc, rsize, &rxr->rxdma)) {
421			printf("%s: unable to allocate RX descriptor\n",
422			    DEVNAME(sc));
423			goto err_rx_desc;
424		}
425		rxr->rx_base = (union igc_adv_rx_desc *)rxr->rxdma.dma_vaddr;
426		bzero((void *)rxr->rx_base, rsize);
427	}
428
429	/* Set up the queue holding structs. */
430	for (i = 0; i < sc->sc_nqueues; i++) {
431		iq = &sc->queues[i];
432		iq->sc = sc;
433		iq->txr = &sc->tx_rings[i];
434		iq->rxr = &sc->rx_rings[i];
435		snprintf(iq->name, sizeof(iq->name), "%s:%d", DEVNAME(sc), i);
436	}
437
438	return 0;
439
440err_rx_desc:
441	for (rxr = sc->rx_rings; rxconf > 0; rxr++, rxconf--)
442		igc_dma_free(sc, &rxr->rxdma);
443err_tx_desc:
444	for (txr = sc->tx_rings; txconf > 0; txr++, txconf--)
445		igc_dma_free(sc, &txr->txdma);
446	free(sc->rx_rings, M_DEVBUF, sc->sc_nqueues * sizeof(struct rx_ring));
447	sc->rx_rings = NULL;
448rx_fail:
449	free(sc->tx_rings, M_DEVBUF, sc->sc_nqueues * sizeof(struct tx_ring));
450	sc->tx_rings = NULL;
451fail:
452	return ENOMEM;
453}
454
455void
456igc_free_pci_resources(struct igc_softc *sc)
457{
458	struct igc_osdep *os = &sc->osdep;
459	struct pci_attach_args *pa = &os->os_pa;
460	struct igc_queue *iq = sc->queues;
461	int i;
462
463	/* Release all msix queue resources. */
464	for (i = 0; i < sc->sc_nqueues; i++, iq++) {
465		if (iq->tag)
466			pci_intr_disestablish(pa->pa_pc, iq->tag);
467		iq->tag = NULL;
468	}
469
470	if (sc->tag)
471		pci_intr_disestablish(pa->pa_pc, sc->tag);
472	sc->tag = NULL;
473	if (os->os_membase != 0)
474		bus_space_unmap(os->os_memt, os->os_memh, os->os_memsize);
475	os->os_membase = 0;
476}
477
478/*********************************************************************
479 *
480 *  Initialize the hardware to a configuration as specified by the
481 *  adapter structure.
482 *
483 **********************************************************************/
484void
485igc_reset(struct igc_softc *sc)
486{
487	struct igc_hw *hw = &sc->hw;
488	uint32_t pba;
489	uint16_t rx_buffer_size;
490
491	/* Let the firmware know the OS is in control */
492	igc_get_hw_control(sc);
493
494	/*
495	 * Packet Buffer Allocation (PBA)
496	 * Writing PBA sets the receive portion of the buffer
497	 * the remainder is used for the transmit buffer.
498	 */
499	pba = IGC_PBA_34K;
500
501	/*
502	 * These parameters control the automatic generation (Tx) and
503	 * response (Rx) to Ethernet PAUSE frames.
504	 * - High water mark should allow for at least two frames to be
505	 *   received after sending an XOFF.
506	 * - Low water mark works best when it is very near the high water mark.
507	 *   This allows the receiver to restart by sending XON when it has
508	 *   drained a bit. Here we use an arbitrary value of 1500 which will
509	 *   restart after one full frame is pulled from the buffer. There
510	 *   could be several smaller frames in the buffer and if so they will
511	 *   not trigger the XON until their total number reduces the buffer
512	 *   by 1500.
513	 * - The pause time is fairly large at 1000 x 512ns = 512 usec.
514	 */
515	rx_buffer_size = (pba & 0xffff) << 10;
516	hw->fc.high_water = rx_buffer_size -
517	    roundup2(sc->hw.mac.max_frame_size, 1024);
518	/* 16-byte granularity */
519	hw->fc.low_water = hw->fc.high_water - 16;
520
521	if (sc->fc) /* locally set flow control value? */
522		hw->fc.requested_mode = sc->fc;
523	else
524		hw->fc.requested_mode = igc_fc_full;
525
526	hw->fc.pause_time = IGC_FC_PAUSE_TIME;
527
528	hw->fc.send_xon = true;
529
530	/* Issue a global reset */
531	igc_reset_hw(hw);
532	IGC_WRITE_REG(hw, IGC_WUC, 0);
533
534	/* and a re-init */
535	if (igc_init_hw(hw) < 0) {
536		printf(": Hardware Initialization Failed\n");
537		return;
538	}
539
540	/* Setup DMA Coalescing */
541	igc_init_dmac(sc, pba);
542
543	IGC_WRITE_REG(hw, IGC_VET, ETHERTYPE_VLAN);
544	igc_get_phy_info(hw);
545	igc_check_for_link(hw);
546}
547
548/*********************************************************************
549 *
550 *  Initialize the DMA Coalescing feature
551 *
552 **********************************************************************/
553void
554igc_init_dmac(struct igc_softc *sc, uint32_t pba)
555{
556	struct igc_hw *hw = &sc->hw;
557	uint32_t dmac, reg = ~IGC_DMACR_DMAC_EN;
558	uint16_t hwm, max_frame_size;
559	int status;
560
561	max_frame_size = sc->hw.mac.max_frame_size;
562
563	if (sc->dmac == 0) { /* Disabling it */
564		IGC_WRITE_REG(hw, IGC_DMACR, reg);
565		return;
566	} else
567		printf(": DMA Coalescing enabled\n");
568
569	/* Set starting threshold */
570	IGC_WRITE_REG(hw, IGC_DMCTXTH, 0);
571
572	hwm = 64 * pba - max_frame_size / 16;
573	if (hwm < 64 * (pba - 6))
574		hwm = 64 * (pba - 6);
575	reg = IGC_READ_REG(hw, IGC_FCRTC);
576	reg &= ~IGC_FCRTC_RTH_COAL_MASK;
577	reg |= ((hwm << IGC_FCRTC_RTH_COAL_SHIFT)
578		& IGC_FCRTC_RTH_COAL_MASK);
579	IGC_WRITE_REG(hw, IGC_FCRTC, reg);
580
581	dmac = pba - max_frame_size / 512;
582	if (dmac < pba - 10)
583		dmac = pba - 10;
584	reg = IGC_READ_REG(hw, IGC_DMACR);
585	reg &= ~IGC_DMACR_DMACTHR_MASK;
586	reg |= ((dmac << IGC_DMACR_DMACTHR_SHIFT)
587		& IGC_DMACR_DMACTHR_MASK);
588
589	/* transition to L0x or L1 if available..*/
590	reg |= (IGC_DMACR_DMAC_EN | IGC_DMACR_DMAC_LX_MASK);
591
592	/* Check if status is 2.5Gb backplane connection
593	 * before configuration of watchdog timer, which is
594	 * in msec values in 12.8usec intervals
595	 * watchdog timer= msec values in 32usec intervals
596	 * for non 2.5Gb connection
597	 */
598	status = IGC_READ_REG(hw, IGC_STATUS);
599	if ((status & IGC_STATUS_2P5_SKU) &&
600	    (!(status & IGC_STATUS_2P5_SKU_OVER)))
601		reg |= ((sc->dmac * 5) >> 6);
602	else
603		reg |= (sc->dmac >> 5);
604
605	IGC_WRITE_REG(hw, IGC_DMACR, reg);
606
607	IGC_WRITE_REG(hw, IGC_DMCRTRH, 0);
608
609	/* Set the interval before transition */
610	reg = IGC_READ_REG(hw, IGC_DMCTLX);
611	reg |= IGC_DMCTLX_DCFLUSH_DIS;
612
613	/*
614	** in 2.5Gb connection, TTLX unit is 0.4 usec
615	** which is 0x4*2 = 0xA. But delay is still 4 usec
616	*/
617	status = IGC_READ_REG(hw, IGC_STATUS);
618	if ((status & IGC_STATUS_2P5_SKU) &&
619	    (!(status & IGC_STATUS_2P5_SKU_OVER)))
620		reg |= 0xA;
621	else
622		reg |= 0x4;
623
624	IGC_WRITE_REG(hw, IGC_DMCTLX, reg);
625
626	/* free space in tx packet buffer to wake from DMA coal */
627	IGC_WRITE_REG(hw, IGC_DMCTXTH, (IGC_TXPBSIZE -
628	    (2 * max_frame_size)) >> 6);
629
630	/* make low power state decision controlled by DMA coal */
631	reg = IGC_READ_REG(hw, IGC_PCIEMISC);
632	reg &= ~IGC_PCIEMISC_LX_DECISION;
633	IGC_WRITE_REG(hw, IGC_PCIEMISC, reg);
634}
635
636int
637igc_allocate_msix(struct igc_softc *sc)
638{
639	struct igc_osdep *os = &sc->osdep;
640	struct pci_attach_args *pa = &os->os_pa;
641	struct igc_queue *iq;
642	pci_intr_handle_t ih;
643	int i, error = 0;
644
645	for (i = 0, iq = sc->queues; i < sc->sc_nqueues; i++, iq++) {
646		if (pci_intr_map_msix(pa, i, &ih)) {
647			printf("%s: unable to map msi-x vector %d\n",
648			    DEVNAME(sc), i);
649			error = ENOMEM;
650			goto fail;
651		}
652
653		iq->tag = pci_intr_establish_cpu(pa->pa_pc, ih,
654		    IPL_NET | IPL_MPSAFE, intrmap_cpu(sc->sc_intrmap, i),
655		    igc_intr_queue, iq, iq->name);
656		if (iq->tag == NULL) {
657			printf("%s: unable to establish interrupt %d\n",
658			    DEVNAME(sc), i);
659			error = ENOMEM;
660			goto fail;
661		}
662
663		iq->msix = i;
664		iq->eims = 1 << i;
665	}
666
667	/* Now the link status/control last MSI-X vector. */
668	if (pci_intr_map_msix(pa, i, &ih)) {
669		printf("%s: unable to map link vector\n", DEVNAME(sc));
670		error = ENOMEM;
671		goto fail;
672	}
673
674	sc->tag = pci_intr_establish(pa->pa_pc, ih, IPL_NET | IPL_MPSAFE,
675	    igc_intr_link, sc, sc->sc_dev.dv_xname);
676	if (sc->tag == NULL) {
677		printf("%s: unable to establish link interrupt\n", DEVNAME(sc));
678		error = ENOMEM;
679		goto fail;
680	}
681
682	sc->linkvec = i;
683	printf(", %s, %d queue%s", pci_intr_string(pa->pa_pc, ih),
684	    i, (i > 1) ? "s" : "");
685
686	return 0;
687fail:
688	for (iq = sc->queues; i > 0; i--, iq++) {
689		if (iq->tag == NULL)
690			continue;
691		pci_intr_disestablish(pa->pa_pc, iq->tag);
692		iq->tag = NULL;
693	}
694
695	return error;
696}
697
698void
699igc_setup_msix(struct igc_softc *sc)
700{
701	struct igc_osdep *os = &sc->osdep;
702	struct pci_attach_args *pa = &os->os_pa;
703	int nmsix;
704
705	nmsix = pci_intr_msix_count(pa);
706	if (nmsix <= 1)
707		printf(": not enough msi-x vectors\n");
708
709	/* Give one vector to events. */
710	nmsix--;
711
712	sc->sc_intrmap = intrmap_create(&sc->sc_dev, nmsix, IGC_MAX_VECTORS,
713	    INTRMAP_POWEROF2);
714	sc->sc_nqueues = intrmap_count(sc->sc_intrmap);
715}
716
717int
718igc_dma_malloc(struct igc_softc *sc, bus_size_t size, struct igc_dma_alloc *dma)
719{
720	struct igc_osdep *os = &sc->osdep;
721
722	dma->dma_tag = os->os_pa.pa_dmat;
723
724	if (bus_dmamap_create(dma->dma_tag, size, 1, size, 0, BUS_DMA_NOWAIT,
725	    &dma->dma_map))
726		return 1;
727	if (bus_dmamem_alloc(dma->dma_tag, size, PAGE_SIZE, 0, &dma->dma_seg,
728	    1, &dma->dma_nseg, BUS_DMA_NOWAIT))
729		goto destroy;
730	if (bus_dmamem_map(dma->dma_tag, &dma->dma_seg, dma->dma_nseg, size,
731	    &dma->dma_vaddr, BUS_DMA_NOWAIT | BUS_DMA_COHERENT))
732		goto free;
733	if (bus_dmamap_load(dma->dma_tag, dma->dma_map, dma->dma_vaddr, size,
734	    NULL, BUS_DMA_NOWAIT))
735		goto unmap;
736
737	dma->dma_size = size;
738
739	return 0;
740unmap:
741	bus_dmamem_unmap(dma->dma_tag, dma->dma_vaddr, size);
742free:
743	bus_dmamem_free(dma->dma_tag, &dma->dma_seg, dma->dma_nseg);
744destroy:
745	bus_dmamap_destroy(dma->dma_tag, dma->dma_map);
746	dma->dma_map = NULL;
747	dma->dma_tag = NULL;
748	return 1;
749}
750
751void
752igc_dma_free(struct igc_softc *sc, struct igc_dma_alloc *dma)
753{
754	if (dma->dma_tag == NULL)
755		return;
756
757	if (dma->dma_map != NULL) {
758		bus_dmamap_sync(dma->dma_tag, dma->dma_map, 0,
759		    dma->dma_map->dm_mapsize,
760		    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
761		bus_dmamap_unload(dma->dma_tag, dma->dma_map);
762		bus_dmamem_unmap(dma->dma_tag, dma->dma_vaddr, dma->dma_size);
763		bus_dmamem_free(dma->dma_tag, &dma->dma_seg, dma->dma_nseg);
764		bus_dmamap_destroy(dma->dma_tag, dma->dma_map);
765		dma->dma_map = NULL;
766	}
767}
768
769/*********************************************************************
770 *
771 *  Setup networking device structure and register an interface.
772 *
773 **********************************************************************/
774void
775igc_setup_interface(struct igc_softc *sc)
776{
777	struct ifnet *ifp = &sc->sc_ac.ac_if;
778	int i;
779
780	ifp->if_softc = sc;
781	strlcpy(ifp->if_xname, DEVNAME(sc), IFNAMSIZ);
782	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
783	ifp->if_xflags = IFXF_MPSAFE;
784	ifp->if_ioctl = igc_ioctl;
785	ifp->if_qstart = igc_start;
786	ifp->if_watchdog = igc_watchdog;
787	ifp->if_hardmtu = sc->hw.mac.max_frame_size - ETHER_HDR_LEN -
788	    ETHER_CRC_LEN;
789	ifq_set_maxlen(&ifp->if_snd, sc->num_tx_desc - 1);
790
791	ifp->if_capabilities = IFCAP_VLAN_MTU;
792
793#ifdef notyet
794#if NVLAN > 0
795	ifp->if_capabilities |= IFCAP_VLAN_HWTAGGING;
796#endif
797#endif
798
799	ifp->if_capabilities |= IFCAP_CSUM_IPv4;
800	ifp->if_capabilities |= IFCAP_CSUM_TCPv4 | IFCAP_CSUM_UDPv4;
801	ifp->if_capabilities |= IFCAP_CSUM_TCPv6 | IFCAP_CSUM_UDPv6;
802
803	/* Initialize ifmedia structures. */
804	ifmedia_init(&sc->media, IFM_IMASK, igc_media_change, igc_media_status);
805	ifmedia_add(&sc->media, IFM_ETHER | IFM_10_T, 0, NULL);
806	ifmedia_add(&sc->media, IFM_ETHER | IFM_10_T | IFM_FDX, 0, NULL);
807	ifmedia_add(&sc->media, IFM_ETHER | IFM_100_TX, 0, NULL);
808	ifmedia_add(&sc->media, IFM_ETHER | IFM_100_TX | IFM_FDX, 0, NULL);
809	ifmedia_add(&sc->media, IFM_ETHER | IFM_1000_T | IFM_FDX, 0, NULL);
810	ifmedia_add(&sc->media, IFM_ETHER | IFM_1000_T, 0, NULL);
811	ifmedia_add(&sc->media, IFM_ETHER | IFM_2500_T, 0, NULL);
812
813	ifmedia_add(&sc->media, IFM_ETHER | IFM_AUTO, 0, NULL);
814	ifmedia_set(&sc->media, IFM_ETHER | IFM_AUTO);
815
816	if_attach(ifp);
817	ether_ifattach(ifp);
818
819	if_attach_queues(ifp, sc->sc_nqueues);
820	if_attach_iqueues(ifp, sc->sc_nqueues);
821	for (i = 0; i < sc->sc_nqueues; i++) {
822		struct ifqueue *ifq = ifp->if_ifqs[i];
823		struct ifiqueue *ifiq = ifp->if_iqs[i];
824		struct tx_ring *txr = &sc->tx_rings[i];
825		struct rx_ring *rxr = &sc->rx_rings[i];
826
827		ifq->ifq_softc = txr;
828		txr->ifq = ifq;
829
830		ifiq->ifiq_softc = rxr;
831		rxr->ifiq = ifiq;
832	}
833}
834
835void
836igc_init(void *arg)
837{
838	struct igc_softc *sc = (struct igc_softc *)arg;
839	struct ifnet *ifp = &sc->sc_ac.ac_if;
840	struct rx_ring *rxr;
841	uint32_t ctrl = 0;
842	int i, s;
843
844	s = splnet();
845
846	igc_stop(sc);
847
848	/* Get the latest mac address, user can use a LAA. */
849	bcopy(sc->sc_ac.ac_enaddr, sc->hw.mac.addr, ETHER_ADDR_LEN);
850
851	/* Put the address into the receive address array. */
852	igc_rar_set(&sc->hw, sc->hw.mac.addr, 0);
853
854	/* Initialize the hardware. */
855	igc_reset(sc);
856	igc_update_link_status(sc);
857
858	/* Setup VLAN support, basic and offload if available. */
859	IGC_WRITE_REG(&sc->hw, IGC_VET, ETHERTYPE_VLAN);
860
861	/* Prepare transmit descriptors and buffers. */
862	if (igc_setup_transmit_structures(sc)) {
863		printf("%s: Could not setup transmit structures\n",
864		    DEVNAME(sc));
865		igc_stop(sc);
866		splx(s);
867		return;
868	}
869	igc_initialize_transmit_unit(sc);
870
871	sc->rx_mbuf_sz = MCLBYTES + ETHER_ALIGN;
872	/* Prepare receive descriptors and buffers. */
873	if (igc_setup_receive_structures(sc)) {
874		printf("%s: Could not setup receive structures\n",
875		    DEVNAME(sc));
876		igc_stop(sc);
877		splx(s);
878		return;
879        }
880	igc_initialize_receive_unit(sc);
881
882	if (ifp->if_capabilities & IFCAP_VLAN_HWTAGGING) {
883		ctrl = IGC_READ_REG(&sc->hw, IGC_CTRL);
884		ctrl |= IGC_CTRL_VME;
885		IGC_WRITE_REG(&sc->hw, IGC_CTRL, ctrl);
886	}
887
888	/* Setup multicast table. */
889	igc_iff(sc);
890
891	igc_clear_hw_cntrs_base_generic(&sc->hw);
892
893	igc_configure_queues(sc);
894
895	/* This clears any pending interrupts */
896	IGC_READ_REG(&sc->hw, IGC_ICR);
897	IGC_WRITE_REG(&sc->hw, IGC_ICS, IGC_ICS_LSC);
898
899	/* The driver can now take control from firmware. */
900	igc_get_hw_control(sc);
901
902	/* Set Energy Efficient Ethernet. */
903	igc_set_eee_i225(&sc->hw, true, true, true);
904
905	for (i = 0; i < sc->sc_nqueues; i++) {
906		rxr = &sc->rx_rings[i];
907		igc_rxfill(rxr);
908		if (if_rxr_inuse(&rxr->rx_ring) == 0) {
909			printf("%s: Unable to fill any rx descriptors\n",
910			    DEVNAME(sc));
911			igc_stop(sc);
912			splx(s);
913		}
914		IGC_WRITE_REG(&sc->hw, IGC_RDT(i),
915		    (rxr->last_desc_filled + 1) % sc->num_rx_desc);
916	}
917
918	igc_enable_intr(sc);
919
920	ifp->if_flags |= IFF_RUNNING;
921	for (i = 0; i < sc->sc_nqueues; i++)
922		ifq_clr_oactive(ifp->if_ifqs[i]);
923
924	splx(s);
925}
926
927static inline int
928igc_load_mbuf(bus_dma_tag_t dmat, bus_dmamap_t map, struct mbuf *m)
929{
930	int error;
931
932	error = bus_dmamap_load_mbuf(dmat, map, m,
933	    BUS_DMA_STREAMING | BUS_DMA_NOWAIT);
934	if (error != EFBIG)
935		return (error);
936
937	error = m_defrag(m, M_DONTWAIT);
938	if (error != 0)
939		return (error);
940
941	return (bus_dmamap_load_mbuf(dmat, map, m,
942	    BUS_DMA_STREAMING | BUS_DMA_NOWAIT));
943}
944
945void
946igc_start(struct ifqueue *ifq)
947{
948	struct ifnet *ifp = ifq->ifq_if;
949	struct igc_softc *sc = ifp->if_softc;
950	struct tx_ring *txr = ifq->ifq_softc;
951	union igc_adv_tx_desc *txdesc;
952	struct igc_tx_buf *txbuf;
953	bus_dmamap_t map;
954	struct mbuf *m;
955	unsigned int prod, free, last, i;
956	unsigned int mask;
957	uint32_t cmd_type_len;
958	uint32_t olinfo_status;
959	int post = 0;
960#if NBPFILTER > 0
961	caddr_t if_bpf;
962#endif
963
964	if (!sc->link_active) {
965		ifq_purge(ifq);
966		return;
967	}
968
969	prod = txr->next_avail_desc;
970	free = txr->next_to_clean;
971	if (free <= prod)
972		free += sc->num_tx_desc;
973	free -= prod;
974
975	bus_dmamap_sync(txr->txdma.dma_tag, txr->txdma.dma_map, 0,
976	    txr->txdma.dma_map->dm_mapsize, BUS_DMASYNC_POSTWRITE);
977
978	mask = sc->num_tx_desc - 1;
979
980	for (;;) {
981		if (free <= IGC_MAX_SCATTER + 1) {
982			ifq_set_oactive(ifq);
983			break;
984		}
985
986		m = ifq_dequeue(ifq);
987		if (m == NULL)
988			break;
989
990		txbuf = &txr->tx_buffers[prod];
991		map = txbuf->map;
992
993		if (igc_load_mbuf(txr->txdma.dma_tag, map, m) != 0) {
994			ifq->ifq_errors++;
995			m_freem(m);
996			continue;
997		}
998
999		olinfo_status = m->m_pkthdr.len << IGC_ADVTXD_PAYLEN_SHIFT;
1000
1001		bus_dmamap_sync(txr->txdma.dma_tag, map, 0,
1002		    map->dm_mapsize, BUS_DMASYNC_PREWRITE);
1003
1004		if (igc_tx_ctx_setup(txr, m, prod, &olinfo_status)) {
1005			/* Consume the first descriptor */
1006			prod++;
1007			prod &= mask;
1008			free--;
1009		}
1010
1011		for (i = 0; i < map->dm_nsegs; i++) {
1012			txdesc = &txr->tx_base[prod];
1013
1014			cmd_type_len = IGC_ADVTXD_DCMD_IFCS | IGC_ADVTXD_DTYP_DATA |
1015			    IGC_ADVTXD_DCMD_DEXT | map->dm_segs[i].ds_len;
1016			if (i == map->dm_nsegs - 1)
1017				cmd_type_len |= IGC_ADVTXD_DCMD_EOP |
1018				    IGC_ADVTXD_DCMD_RS;
1019
1020			htolem64(&txdesc->read.buffer_addr, map->dm_segs[i].ds_addr);
1021			htolem32(&txdesc->read.cmd_type_len, cmd_type_len);
1022			htolem32(&txdesc->read.olinfo_status, olinfo_status);
1023
1024			last = prod;
1025
1026			prod++;
1027			prod &= mask;
1028		}
1029
1030		txbuf->m_head = m;
1031		txbuf->eop_index = last;
1032
1033#if NBPFILTER > 0
1034		if_bpf = ifp->if_bpf;
1035		if (if_bpf)
1036			bpf_mtap_ether(if_bpf, m, BPF_DIRECTION_OUT);
1037#endif
1038
1039		free -= i;
1040		post = 1;
1041	}
1042
1043	bus_dmamap_sync(txr->txdma.dma_tag, txr->txdma.dma_map, 0,
1044	    txr->txdma.dma_map->dm_mapsize, BUS_DMASYNC_PREWRITE);
1045
1046	if (post) {
1047		txr->next_avail_desc = prod;
1048		IGC_WRITE_REG(&sc->hw, IGC_TDT(txr->me), prod);
1049	}
1050}
1051
1052int
1053igc_txeof(struct tx_ring *txr)
1054{
1055	struct igc_softc *sc = txr->sc;
1056	struct ifqueue *ifq = txr->ifq;
1057	union igc_adv_tx_desc *txdesc;
1058	struct igc_tx_buf *txbuf;
1059	bus_dmamap_t map;
1060	unsigned int cons, prod, last;
1061	unsigned int mask;
1062	int done = 0;
1063
1064	prod = txr->next_avail_desc;
1065	cons = txr->next_to_clean;
1066
1067	if (cons == prod)
1068		return (0);
1069
1070	bus_dmamap_sync(txr->txdma.dma_tag, txr->txdma.dma_map, 0,
1071	    txr->txdma.dma_map->dm_mapsize, BUS_DMASYNC_POSTREAD);
1072
1073	mask = sc->num_tx_desc - 1;
1074
1075	do {
1076		txbuf = &txr->tx_buffers[cons];
1077		last = txbuf->eop_index;
1078		txdesc = &txr->tx_base[last];
1079
1080		if (!(txdesc->wb.status & htole32(IGC_TXD_STAT_DD)))
1081			break;
1082
1083		map = txbuf->map;
1084
1085		bus_dmamap_sync(txr->txdma.dma_tag, map, 0, map->dm_mapsize,
1086		    BUS_DMASYNC_POSTWRITE);
1087		bus_dmamap_unload(txr->txdma.dma_tag, map);
1088		m_freem(txbuf->m_head);
1089
1090		txbuf->m_head = NULL;
1091		txbuf->eop_index = -1;
1092
1093		cons = last + 1;
1094		cons &= mask;
1095
1096		done = 1;
1097	} while (cons != prod);
1098
1099	bus_dmamap_sync(txr->txdma.dma_tag, txr->txdma.dma_map, 0,
1100	    txr->txdma.dma_map->dm_mapsize, BUS_DMASYNC_PREREAD);
1101
1102	txr->next_to_clean = cons;
1103
1104	if (ifq_is_oactive(ifq))
1105		ifq_restart(ifq);
1106
1107	return (done);
1108}
1109
1110/*********************************************************************
1111 *
1112 *  This routine disables all traffic on the adapter by issuing a
1113 *  global reset on the MAC.
1114 *
1115 **********************************************************************/
1116void
1117igc_stop(struct igc_softc *sc)
1118{
1119	struct ifnet *ifp = &sc->sc_ac.ac_if;
1120	int i;
1121
1122	/* Tell the stack that the interface is no longer active. */
1123        ifp->if_flags &= ~IFF_RUNNING;
1124
1125	igc_disable_intr(sc);
1126
1127	igc_reset_hw(&sc->hw);
1128	IGC_WRITE_REG(&sc->hw, IGC_WUC, 0);
1129
1130	intr_barrier(sc->tag);
1131        for (i = 0; i < sc->sc_nqueues; i++) {
1132                struct ifqueue *ifq = ifp->if_ifqs[i];
1133                ifq_barrier(ifq);
1134                ifq_clr_oactive(ifq);
1135
1136                if (sc->queues[i].tag != NULL)
1137                        intr_barrier(sc->queues[i].tag);
1138                timeout_del(&sc->rx_rings[i].rx_refill);
1139        }
1140
1141        igc_free_transmit_structures(sc);
1142        igc_free_receive_structures(sc);
1143
1144	igc_update_link_status(sc);
1145}
1146
1147/*********************************************************************
1148 *  Ioctl entry point
1149 *
1150 *  igc_ioctl is called when the user wants to configure the
1151 *  interface.
1152 *
1153 *  return 0 on success, positive on failure
1154 **********************************************************************/
1155int
1156igc_ioctl(struct ifnet * ifp, u_long cmd, caddr_t data)
1157{
1158	struct igc_softc *sc = ifp->if_softc;
1159	struct ifreq *ifr = (struct ifreq *)data;
1160	int s, error = 0;
1161
1162	s = splnet();
1163
1164	switch (cmd) {
1165	case SIOCSIFADDR:
1166		ifp->if_flags |= IFF_UP;
1167		if (!(ifp->if_flags & IFF_RUNNING))
1168			igc_init(sc);
1169		break;
1170	case SIOCSIFFLAGS:
1171		if (ifp->if_flags & IFF_UP) {
1172			if (ifp->if_flags & IFF_RUNNING)
1173				error = ENETRESET;
1174			else
1175				igc_init(sc);
1176		} else {
1177			if (ifp->if_flags & IFF_RUNNING)
1178				igc_stop(sc);
1179		}
1180		break;
1181	case SIOCSIFMEDIA:
1182	case SIOCGIFMEDIA:
1183		error = ifmedia_ioctl(ifp, ifr, &sc->media, cmd);
1184		break;
1185	case SIOCGIFRXR:
1186		error = igc_rxrinfo(sc, (struct if_rxrinfo *)ifr->ifr_data);
1187		break;
1188	default:
1189		error = ether_ioctl(ifp, &sc->sc_ac, cmd, data);
1190	}
1191
1192	if (error == ENETRESET) {
1193		if (ifp->if_flags & IFF_RUNNING) {
1194			igc_disable_intr(sc);
1195			igc_iff(sc);
1196			igc_enable_intr(sc);
1197		}
1198		error = 0;
1199	}
1200
1201	splx(s);
1202	return error;
1203}
1204
1205int
1206igc_rxrinfo(struct igc_softc *sc, struct if_rxrinfo *ifri)
1207{
1208	struct if_rxring_info *ifr;
1209	struct rx_ring *rxr;
1210	int error, i, n = 0;
1211
1212	if ((ifr = mallocarray(sc->sc_nqueues, sizeof(*ifr), M_DEVBUF,
1213	    M_WAITOK | M_ZERO)) == NULL)
1214		return ENOMEM;
1215
1216	for (i = 0; i < sc->sc_nqueues; i++) {
1217		rxr = &sc->rx_rings[i];
1218		ifr[n].ifr_size = MCLBYTES;
1219		snprintf(ifr[n].ifr_name, sizeof(ifr[n].ifr_name), "%d", i);
1220		ifr[n].ifr_info = rxr->rx_ring;
1221		n++;
1222	}
1223
1224	error = if_rxr_info_ioctl(ifri, sc->sc_nqueues, ifr);
1225	free(ifr, M_DEVBUF, sc->sc_nqueues * sizeof(*ifr));
1226
1227	return error;
1228}
1229
1230int
1231igc_rxfill(struct rx_ring *rxr)
1232{
1233	struct igc_softc *sc = rxr->sc;
1234	int i, post = 0;
1235	u_int slots;
1236
1237	bus_dmamap_sync(rxr->rxdma.dma_tag, rxr->rxdma.dma_map, 0,
1238	    rxr->rxdma.dma_map->dm_mapsize, BUS_DMASYNC_POSTWRITE);
1239
1240	i = rxr->last_desc_filled;
1241	for (slots = if_rxr_get(&rxr->rx_ring, sc->num_rx_desc); slots > 0;
1242	    slots--) {
1243		if (++i == sc->num_rx_desc)
1244			i = 0;
1245
1246		if (igc_get_buf(rxr, i) != 0)
1247			break;
1248
1249		rxr->last_desc_filled = i;
1250		post = 1;
1251	}
1252
1253	bus_dmamap_sync(rxr->rxdma.dma_tag, rxr->rxdma.dma_map, 0,
1254	    rxr->rxdma.dma_map->dm_mapsize, BUS_DMASYNC_PREWRITE);
1255
1256	if_rxr_put(&rxr->rx_ring, slots);
1257
1258	return post;
1259}
1260
1261void
1262igc_rxrefill(void *xrxr)
1263{
1264	struct rx_ring *rxr = xrxr;
1265	struct igc_softc *sc = rxr->sc;
1266
1267	if (igc_rxfill(rxr)) {
1268		IGC_WRITE_REG(&sc->hw, IGC_RDT(rxr->me),
1269		    (rxr->last_desc_filled + 1) % sc->num_rx_desc);
1270	}
1271	else if (if_rxr_inuse(&rxr->rx_ring) == 0)
1272		timeout_add(&rxr->rx_refill, 1);
1273}
1274
1275/*********************************************************************
1276 *
1277 *  This routine executes in interrupt context. It replenishes
1278 *  the mbufs in the descriptor and sends data which has been
1279 *  dma'ed into host memory to upper layer.
1280 *
1281 *********************************************************************/
1282int
1283igc_rxeof(struct rx_ring *rxr)
1284{
1285	struct igc_softc *sc = rxr->sc;
1286	struct ifnet *ifp = &sc->sc_ac.ac_if;
1287	struct mbuf_list ml = MBUF_LIST_INITIALIZER();
1288	struct mbuf *mp, *m;
1289	struct igc_rx_buf *rxbuf, *nxbuf;
1290	union igc_adv_rx_desc *rxdesc;
1291	uint32_t ptype, staterr = 0;
1292	uint16_t len, vtag;
1293	uint8_t eop = 0;
1294	int i, nextp;
1295
1296	if (!ISSET(ifp->if_flags, IFF_RUNNING))
1297		return 0;
1298
1299	i = rxr->next_to_check;
1300	while (if_rxr_inuse(&rxr->rx_ring) > 0) {
1301		uint32_t hash;
1302		uint16_t hashtype;
1303
1304		bus_dmamap_sync(rxr->rxdma.dma_tag, rxr->rxdma.dma_map,
1305		    i * sizeof(union igc_adv_rx_desc),
1306		    sizeof(union igc_adv_rx_desc), BUS_DMASYNC_POSTREAD);
1307
1308		rxdesc = &rxr->rx_base[i];
1309		staterr = letoh32(rxdesc->wb.upper.status_error);
1310		if (!ISSET(staterr, IGC_RXD_STAT_DD)) {
1311			bus_dmamap_sync(rxr->rxdma.dma_tag, rxr->rxdma.dma_map,
1312			    i * sizeof(union igc_adv_rx_desc),
1313			    sizeof(union igc_adv_rx_desc), BUS_DMASYNC_PREREAD);
1314			break;
1315		}
1316
1317		/* Zero out the receive descriptors status. */
1318		rxdesc->wb.upper.status_error = 0;
1319		rxbuf = &rxr->rx_buffers[i];
1320
1321		/* Pull the mbuf off the ring. */
1322		bus_dmamap_sync(rxr->rxdma.dma_tag, rxbuf->map, 0,
1323		    rxbuf->map->dm_mapsize, BUS_DMASYNC_POSTREAD);
1324		bus_dmamap_unload(rxr->rxdma.dma_tag, rxbuf->map);
1325
1326		mp = rxbuf->buf;
1327		len = letoh16(rxdesc->wb.upper.length);
1328		vtag = letoh16(rxdesc->wb.upper.vlan);
1329		eop = ((staterr & IGC_RXD_STAT_EOP) == IGC_RXD_STAT_EOP);
1330		ptype = letoh32(rxdesc->wb.lower.lo_dword.data) &
1331		    IGC_PKTTYPE_MASK;
1332		hash = letoh32(rxdesc->wb.lower.hi_dword.rss);
1333		hashtype = le16toh(rxdesc->wb.lower.lo_dword.hs_rss.pkt_info) &
1334		    IGC_RXDADV_RSSTYPE_MASK;
1335
1336		if (staterr & IGC_RXDEXT_STATERR_RXE) {
1337			if (rxbuf->fmp) {
1338				m_freem(rxbuf->fmp);
1339				rxbuf->fmp = NULL;
1340			}
1341
1342			m_freem(mp);
1343			rxbuf->buf = NULL;
1344			goto next_desc;
1345		}
1346
1347		if (mp == NULL) {
1348			panic("%s: igc_rxeof: NULL mbuf in slot %d "
1349			    "(nrx %d, filled %d)", DEVNAME(sc), i,
1350			    if_rxr_inuse(&rxr->rx_ring), rxr->last_desc_filled);
1351		}
1352
1353		if (!eop) {
1354			/*
1355			 * Figure out the next descriptor of this frame.
1356			 */
1357			nextp = i + 1;
1358			if (nextp == sc->num_rx_desc)
1359				nextp = 0;
1360			nxbuf = &rxr->rx_buffers[nextp];
1361			/* prefetch(nxbuf); */
1362		}
1363
1364		mp->m_len = len;
1365
1366		m = rxbuf->fmp;
1367		rxbuf->buf = rxbuf->fmp = NULL;
1368
1369		if (m != NULL)
1370			m->m_pkthdr.len += mp->m_len;
1371		else {
1372			m = mp;
1373			m->m_pkthdr.len = mp->m_len;
1374#if NVLAN > 0
1375			if (staterr & IGC_RXD_STAT_VP) {
1376				m->m_pkthdr.ether_vtag = vtag;
1377				m->m_flags |= M_VLANTAG;
1378			}
1379#endif
1380		}
1381
1382		/* Pass the head pointer on */
1383		if (eop == 0) {
1384			nxbuf->fmp = m;
1385			m = NULL;
1386			mp->m_next = nxbuf->buf;
1387		} else {
1388			igc_rx_checksum(staterr, m, ptype);
1389
1390			if (hashtype != IGC_RXDADV_RSSTYPE_NONE) {
1391				m->m_pkthdr.ph_flowid = hash;
1392				SET(m->m_pkthdr.csum_flags, M_FLOWID);
1393			}
1394
1395			ml_enqueue(&ml, m);
1396		}
1397next_desc:
1398		if_rxr_put(&rxr->rx_ring, 1);
1399		bus_dmamap_sync(rxr->rxdma.dma_tag, rxr->rxdma.dma_map,
1400		    i * sizeof(union igc_adv_rx_desc),
1401		    sizeof(union igc_adv_rx_desc), BUS_DMASYNC_PREREAD);
1402
1403		/* Advance our pointers to the next descriptor. */
1404		if (++i == sc->num_rx_desc)
1405			i = 0;
1406	}
1407	rxr->next_to_check = i;
1408
1409	if (ifiq_input(rxr->ifiq, &ml))
1410		if_rxr_livelocked(&rxr->rx_ring);
1411
1412	if (!(staterr & IGC_RXD_STAT_DD))
1413		return 0;
1414
1415	return 1;
1416}
1417
1418/*********************************************************************
1419 *
1420 *  Verify that the hardware indicated that the checksum is valid.
1421 *  Inform the stack about the status of checksum so that stack
1422 *  doesn't spend time verifying the checksum.
1423 *
1424 *********************************************************************/
1425void
1426igc_rx_checksum(uint32_t staterr, struct mbuf *m, uint32_t ptype)
1427{
1428	uint16_t status = (uint16_t)staterr;
1429	uint8_t errors = (uint8_t)(staterr >> 24);
1430
1431	if (status & IGC_RXD_STAT_IPCS) {
1432		if (!(errors & IGC_RXD_ERR_IPE)) {
1433			/* IP Checksum Good */
1434			m->m_pkthdr.csum_flags = M_IPV4_CSUM_IN_OK;
1435		} else
1436			m->m_pkthdr.csum_flags = 0;
1437	}
1438
1439	if (status & (IGC_RXD_STAT_TCPCS | IGC_RXD_STAT_UDPCS)) {
1440		if (!(errors & IGC_RXD_ERR_TCPE))
1441			m->m_pkthdr.csum_flags |=
1442			    M_TCP_CSUM_IN_OK | M_UDP_CSUM_IN_OK;
1443	}
1444}
1445
1446void
1447igc_watchdog(struct ifnet * ifp)
1448{
1449}
1450
1451/*********************************************************************
1452 *
1453 *  Media Ioctl callback
1454 *
1455 *  This routine is called whenever the user queries the status of
1456 *  the interface using ifconfig.
1457 *
1458 **********************************************************************/
1459void
1460igc_media_status(struct ifnet *ifp, struct ifmediareq *ifmr)
1461{
1462	struct igc_softc *sc = ifp->if_softc;
1463
1464	igc_update_link_status(sc);
1465
1466	ifmr->ifm_status = IFM_AVALID;
1467	ifmr->ifm_active = IFM_ETHER;
1468
1469	if (!sc->link_active) {
1470		ifmr->ifm_active |= IFM_NONE;
1471		return;
1472	}
1473
1474	ifmr->ifm_status |= IFM_ACTIVE;
1475
1476	switch (sc->link_speed) {
1477	case 10:
1478		ifmr->ifm_active |= IFM_10_T;
1479		break;
1480	case 100:
1481		ifmr->ifm_active |= IFM_100_TX;
1482                break;
1483	case 1000:
1484		ifmr->ifm_active |= IFM_1000_T;
1485		break;
1486	case 2500:
1487                ifmr->ifm_active |= IFM_2500_T;
1488                break;
1489	}
1490
1491	if (sc->link_duplex == FULL_DUPLEX)
1492		ifmr->ifm_active |= IFM_FDX;
1493	else
1494		ifmr->ifm_active |= IFM_HDX;
1495}
1496
1497/*********************************************************************
1498 *
1499 *  Media Ioctl callback
1500 *
1501 *  This routine is called when the user changes speed/duplex using
1502 *  media/mediopt option with ifconfig.
1503 *
1504 **********************************************************************/
1505int
1506igc_media_change(struct ifnet *ifp)
1507{
1508	struct igc_softc *sc = ifp->if_softc;
1509	struct ifmedia *ifm = &sc->media;
1510
1511	if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER)
1512		return (EINVAL);
1513
1514	sc->hw.mac.autoneg = DO_AUTO_NEG;
1515
1516	switch (IFM_SUBTYPE(ifm->ifm_media)) {
1517	case IFM_AUTO:
1518		sc->hw.phy.autoneg_advertised = AUTONEG_ADV_DEFAULT;
1519		break;
1520        case IFM_2500_T:
1521                sc->hw.phy.autoneg_advertised = ADVERTISE_2500_FULL;
1522                break;
1523	case IFM_1000_T:
1524		sc->hw.phy.autoneg_advertised = ADVERTISE_1000_FULL;
1525		break;
1526	case IFM_100_TX:
1527		if ((ifm->ifm_media & IFM_GMASK) == IFM_HDX)
1528			sc->hw.phy.autoneg_advertised = ADVERTISE_100_HALF;
1529		else
1530			sc->hw.phy.autoneg_advertised = ADVERTISE_100_FULL;
1531		break;
1532	case IFM_10_T:
1533		if ((ifm->ifm_media & IFM_GMASK) == IFM_HDX)
1534			sc->hw.phy.autoneg_advertised = ADVERTISE_10_HALF;
1535		else
1536			sc->hw.phy.autoneg_advertised = ADVERTISE_10_FULL;
1537		break;
1538	default:
1539		return EINVAL;
1540	}
1541
1542	igc_init(sc);
1543
1544	return 0;
1545}
1546
1547void
1548igc_iff(struct igc_softc *sc)
1549{
1550	struct ifnet *ifp = &sc->sc_ac.ac_if;
1551        struct arpcom *ac = &sc->sc_ac;
1552	struct ether_multi *enm;
1553	struct ether_multistep step;
1554	uint32_t reg_rctl = 0;
1555	uint8_t *mta;
1556	int mcnt = 0;
1557
1558	mta = sc->mta;
1559        bzero(mta, sizeof(uint8_t) * ETHER_ADDR_LEN *
1560	    MAX_NUM_MULTICAST_ADDRESSES);
1561
1562	reg_rctl = IGC_READ_REG(&sc->hw, IGC_RCTL);
1563	reg_rctl &= ~(IGC_RCTL_UPE | IGC_RCTL_MPE);
1564	ifp->if_flags &= ~IFF_ALLMULTI;
1565
1566	if (ifp->if_flags & IFF_PROMISC || ac->ac_multirangecnt > 0 ||
1567	    ac->ac_multicnt > MAX_NUM_MULTICAST_ADDRESSES) {
1568		ifp->if_flags |= IFF_ALLMULTI;
1569		reg_rctl |= IGC_RCTL_MPE;
1570		if (ifp->if_flags & IFF_PROMISC)
1571			reg_rctl |= IGC_RCTL_UPE;
1572	} else {
1573		ETHER_FIRST_MULTI(step, ac, enm);
1574		while (enm != NULL) {
1575			bcopy(enm->enm_addrlo,
1576			    &mta[mcnt * ETHER_ADDR_LEN], ETHER_ADDR_LEN);
1577			mcnt++;
1578
1579			ETHER_NEXT_MULTI(step, enm);
1580		}
1581
1582		igc_update_mc_addr_list(&sc->hw, mta, mcnt);
1583	}
1584
1585	IGC_WRITE_REG(&sc->hw, IGC_RCTL, reg_rctl);
1586}
1587
1588void
1589igc_update_link_status(struct igc_softc *sc)
1590{
1591	struct ifnet *ifp = &sc->sc_ac.ac_if;
1592	struct igc_hw *hw = &sc->hw;
1593	int link_state;
1594
1595	if (IGC_READ_REG(&sc->hw, IGC_STATUS) & IGC_STATUS_LU) {
1596		if (sc->link_active == 0) {
1597			igc_get_speed_and_duplex(hw, &sc->link_speed,
1598			    &sc->link_duplex);
1599			sc->link_active = 1;
1600			ifp->if_baudrate = IF_Mbps(sc->link_speed);
1601		}
1602		link_state = (sc->link_duplex == FULL_DUPLEX) ?
1603		    LINK_STATE_FULL_DUPLEX : LINK_STATE_HALF_DUPLEX;
1604	} else {
1605		if (sc->link_active == 1) {
1606			ifp->if_baudrate = sc->link_speed = 0;
1607			sc->link_duplex = 0;
1608			sc->link_active = 0;
1609		}
1610		link_state = LINK_STATE_DOWN;
1611	}
1612	if (ifp->if_link_state != link_state) {
1613		ifp->if_link_state = link_state;
1614		if_link_state_change(ifp);
1615	}
1616}
1617
1618/*********************************************************************
1619 *
1620 *  Get a buffer from system mbuf buffer pool.
1621 *
1622 **********************************************************************/
1623int
1624igc_get_buf(struct rx_ring *rxr, int i)
1625{
1626	struct igc_softc *sc = rxr->sc;
1627	struct igc_rx_buf *rxbuf;
1628	struct mbuf *m;
1629	union igc_adv_rx_desc *rxdesc;
1630	int error;
1631
1632	rxbuf = &rxr->rx_buffers[i];
1633	rxdesc = &rxr->rx_base[i];
1634	if (rxbuf->buf) {
1635		printf("%s: slot %d already has an mbuf\n", DEVNAME(sc), i);
1636		return ENOBUFS;
1637	}
1638
1639	m = MCLGETL(NULL, M_DONTWAIT, sc->rx_mbuf_sz);
1640	if (!m)
1641		return ENOBUFS;
1642
1643	m->m_data += (m->m_ext.ext_size - sc->rx_mbuf_sz);
1644	m->m_len = m->m_pkthdr.len = sc->rx_mbuf_sz;
1645
1646	error = bus_dmamap_load_mbuf(rxr->rxdma.dma_tag, rxbuf->map, m,
1647	    BUS_DMA_NOWAIT);
1648	if (error) {
1649		m_freem(m);
1650		return error;
1651	}
1652
1653	bus_dmamap_sync(rxr->rxdma.dma_tag, rxbuf->map, 0,
1654	    rxbuf->map->dm_mapsize, BUS_DMASYNC_PREREAD);
1655	rxbuf->buf = m;
1656
1657	rxdesc->read.pkt_addr = htole64(rxbuf->map->dm_segs[0].ds_addr);
1658
1659	return 0;
1660}
1661
1662void
1663igc_configure_queues(struct igc_softc *sc)
1664{
1665	struct igc_hw *hw = &sc->hw;
1666	struct igc_queue *iq = sc->queues;
1667	uint32_t ivar, newitr = 0;
1668	int i;
1669
1670	/* First turn on RSS capability */
1671	IGC_WRITE_REG(hw, IGC_GPIE, IGC_GPIE_MSIX_MODE | IGC_GPIE_EIAME |
1672	    IGC_GPIE_PBA | IGC_GPIE_NSICR);
1673
1674	/* Set the starting interrupt rate */
1675	newitr = (4000000 / MAX_INTS_PER_SEC) & 0x7FFC;
1676
1677	newitr |= IGC_EITR_CNT_IGNR;
1678
1679	/* Turn on MSI-X */
1680	for (i = 0; i < sc->sc_nqueues; i++, iq++) {
1681		/* RX entries */
1682		igc_set_queues(sc, i, iq->msix, 0);
1683		/* TX entries */
1684		igc_set_queues(sc, i, iq->msix, 1);
1685		sc->msix_queuesmask |= iq->eims;
1686		IGC_WRITE_REG(hw, IGC_EITR(iq->msix), newitr);
1687	}
1688
1689	/* And for the link interrupt */
1690	ivar = (sc->linkvec | IGC_IVAR_VALID) << 8;
1691	sc->msix_linkmask = 1 << sc->linkvec;
1692	IGC_WRITE_REG(hw, IGC_IVAR_MISC, ivar);
1693}
1694
1695void
1696igc_set_queues(struct igc_softc *sc, uint32_t entry, uint32_t vector, int type)
1697{
1698	struct igc_hw *hw = &sc->hw;
1699	uint32_t ivar, index;
1700
1701	index = entry >> 1;
1702	ivar = IGC_READ_REG_ARRAY(hw, IGC_IVAR0, index);
1703	if (type) {
1704		if (entry & 1) {
1705			ivar &= 0x00FFFFFF;
1706			ivar |= (vector | IGC_IVAR_VALID) << 24;
1707		} else {
1708			ivar &= 0xFFFF00FF;
1709			ivar |= (vector | IGC_IVAR_VALID) << 8;
1710		}
1711	} else {
1712		if (entry & 1) {
1713			ivar &= 0xFF00FFFF;
1714			ivar |= (vector | IGC_IVAR_VALID) << 16;
1715		} else {
1716			ivar &= 0xFFFFFF00;
1717			ivar |= vector | IGC_IVAR_VALID;
1718		}
1719	}
1720	IGC_WRITE_REG_ARRAY(hw, IGC_IVAR0, index, ivar);
1721}
1722
1723void
1724igc_enable_queue(struct igc_softc *sc, uint32_t eims)
1725{
1726	IGC_WRITE_REG(&sc->hw, IGC_EIMS, eims);
1727}
1728
1729void
1730igc_enable_intr(struct igc_softc *sc)
1731{
1732	struct igc_hw *hw = &sc->hw;
1733	uint32_t mask;
1734
1735	mask = (sc->msix_queuesmask | sc->msix_linkmask);
1736	IGC_WRITE_REG(hw, IGC_EIAC, mask);
1737	IGC_WRITE_REG(hw, IGC_EIAM, mask);
1738	IGC_WRITE_REG(hw, IGC_EIMS, mask);
1739	IGC_WRITE_REG(hw, IGC_IMS, IGC_IMS_LSC);
1740	IGC_WRITE_FLUSH(hw);
1741}
1742
1743void
1744igc_disable_intr(struct igc_softc *sc)
1745{
1746	struct igc_hw *hw = &sc->hw;
1747
1748	IGC_WRITE_REG(hw, IGC_EIMC, 0xffffffff);
1749	IGC_WRITE_REG(hw, IGC_EIAC, 0);
1750	IGC_WRITE_REG(hw, IGC_IMC, 0xffffffff);
1751	IGC_WRITE_FLUSH(hw);
1752}
1753
1754int
1755igc_intr_link(void *arg)
1756{
1757	struct igc_softc *sc = (struct igc_softc *)arg;
1758	uint32_t reg_icr = IGC_READ_REG(&sc->hw, IGC_ICR);
1759
1760	if (reg_icr & IGC_ICR_LSC) {
1761		KERNEL_LOCK();
1762		sc->hw.mac.get_link_status = true;
1763		igc_update_link_status(sc);
1764		KERNEL_UNLOCK();
1765	}
1766
1767	IGC_WRITE_REG(&sc->hw, IGC_IMS, IGC_IMS_LSC);
1768	IGC_WRITE_REG(&sc->hw, IGC_EIMS, sc->msix_linkmask);
1769
1770	return 1;
1771}
1772
1773int
1774igc_intr_queue(void *arg)
1775{
1776	struct igc_queue *iq = arg;
1777	struct igc_softc *sc = iq->sc;
1778	struct ifnet *ifp = &sc->sc_ac.ac_if;
1779	struct rx_ring *rxr = iq->rxr;
1780	struct tx_ring *txr = iq->txr;
1781
1782	if (ifp->if_flags & IFF_RUNNING) {
1783		igc_txeof(txr);
1784		igc_rxeof(rxr);
1785		igc_rxrefill(rxr);
1786	}
1787
1788	igc_enable_queue(sc, iq->eims);
1789
1790	return 1;
1791}
1792
1793/*********************************************************************
1794 *
1795 *  Allocate memory for tx_buffer structures. The tx_buffer stores all
1796 *  the information needed to transmit a packet on the wire.
1797 *
1798 **********************************************************************/
1799int
1800igc_allocate_transmit_buffers(struct tx_ring *txr)
1801{
1802	struct igc_softc *sc = txr->sc;
1803	struct igc_tx_buf *txbuf;
1804	int error, i;
1805
1806	txr->tx_buffers = mallocarray(sc->num_tx_desc,
1807	    sizeof(struct igc_tx_buf), M_DEVBUF, M_NOWAIT | M_ZERO);
1808	if (txr->tx_buffers == NULL) {
1809		printf("%s: Unable to allocate tx_buffer memory\n",
1810		    DEVNAME(sc));
1811		error = ENOMEM;
1812		goto fail;
1813	}
1814	txr->txtag = txr->txdma.dma_tag;
1815
1816	/* Create the descriptor buffer dma maps. */
1817	for (i = 0; i < sc->num_tx_desc; i++) {
1818		txbuf = &txr->tx_buffers[i];
1819		error = bus_dmamap_create(txr->txdma.dma_tag, IGC_TSO_SIZE,
1820		    IGC_MAX_SCATTER, PAGE_SIZE, 0, BUS_DMA_NOWAIT, &txbuf->map);
1821		if (error != 0) {
1822			printf("%s: Unable to create TX DMA map\n",
1823			    DEVNAME(sc));
1824			goto fail;
1825		}
1826	}
1827
1828	return 0;
1829fail:
1830	return error;
1831}
1832
1833
1834/*********************************************************************
1835 *
1836 *  Allocate and initialize transmit structures.
1837 *
1838 **********************************************************************/
1839int
1840igc_setup_transmit_structures(struct igc_softc *sc)
1841{
1842	struct tx_ring *txr = sc->tx_rings;
1843	int i;
1844
1845	for (i = 0; i < sc->sc_nqueues; i++, txr++) {
1846		if (igc_setup_transmit_ring(txr))
1847			goto fail;
1848	}
1849
1850	return 0;
1851fail:
1852	igc_free_transmit_structures(sc);
1853	return ENOBUFS;
1854}
1855
1856/*********************************************************************
1857 *
1858 *  Initialize a transmit ring.
1859 *
1860 **********************************************************************/
1861int
1862igc_setup_transmit_ring(struct tx_ring *txr)
1863{
1864	struct igc_softc *sc = txr->sc;
1865
1866	/* Now allocate transmit buffers for the ring. */
1867	if (igc_allocate_transmit_buffers(txr))
1868		return ENOMEM;
1869
1870	/* Clear the old ring contents */
1871	bzero((void *)txr->tx_base,
1872	    (sizeof(union igc_adv_tx_desc)) * sc->num_tx_desc);
1873
1874	/* Reset indices. */
1875	txr->next_avail_desc = 0;
1876	txr->next_to_clean = 0;
1877
1878	bus_dmamap_sync(txr->txdma.dma_tag, txr->txdma.dma_map, 0,
1879	    txr->txdma.dma_map->dm_mapsize,
1880	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
1881
1882	return 0;
1883}
1884
1885/*********************************************************************
1886 *
1887 *  Enable transmit unit.
1888 *
1889 **********************************************************************/
1890void
1891igc_initialize_transmit_unit(struct igc_softc *sc)
1892{
1893	struct ifnet *ifp = &sc->sc_ac.ac_if;
1894	struct tx_ring *txr;
1895	struct igc_hw *hw = &sc->hw;
1896	uint64_t bus_addr;
1897	uint32_t tctl, txdctl = 0;
1898        int i;
1899
1900	/* Setup the Base and Length of the TX descriptor ring. */
1901	for (i = 0; i < sc->sc_nqueues; i++) {
1902		txr = &sc->tx_rings[i];
1903
1904		bus_addr = txr->txdma.dma_map->dm_segs[0].ds_addr;
1905
1906		/* Base and len of TX ring */
1907		IGC_WRITE_REG(hw, IGC_TDLEN(i),
1908		    sc->num_tx_desc * sizeof(union igc_adv_tx_desc));
1909		IGC_WRITE_REG(hw, IGC_TDBAH(i), (uint32_t)(bus_addr >> 32));
1910		IGC_WRITE_REG(hw, IGC_TDBAL(i), (uint32_t)bus_addr);
1911
1912		/* Init the HEAD/TAIL indices */
1913		IGC_WRITE_REG(hw, IGC_TDT(i), 0);
1914		IGC_WRITE_REG(hw, IGC_TDH(i), 0);
1915
1916		txr->watchdog_timer = 0;
1917
1918		txdctl = 0;		/* Clear txdctl */
1919		txdctl |= 0x1f;		/* PTHRESH */
1920		txdctl |= 1 << 8;	/* HTHRESH */
1921		txdctl |= 1 << 16;	/* WTHRESH */
1922		txdctl |= 1 << 22;	/* Reserved bit 22 must always be 1 */
1923		txdctl |= IGC_TXDCTL_GRAN;
1924		txdctl |= 1 << 25;	/* LWTHRESH */
1925
1926		IGC_WRITE_REG(hw, IGC_TXDCTL(i), txdctl);
1927	}
1928	ifp->if_timer = 0;
1929
1930	/* Program the Transmit Control Register */
1931	tctl = IGC_READ_REG(&sc->hw, IGC_TCTL);
1932	tctl &= ~IGC_TCTL_CT;
1933	tctl |= (IGC_TCTL_PSP | IGC_TCTL_RTLC | IGC_TCTL_EN |
1934	    (IGC_COLLISION_THRESHOLD << IGC_CT_SHIFT));
1935
1936	/* This write will effectively turn on the transmit unit. */
1937	IGC_WRITE_REG(&sc->hw, IGC_TCTL, tctl);
1938}
1939
1940/*********************************************************************
1941 *
1942 *  Free all transmit rings.
1943 *
1944 **********************************************************************/
1945void
1946igc_free_transmit_structures(struct igc_softc *sc)
1947{
1948	struct tx_ring *txr = sc->tx_rings;
1949	int i;
1950
1951	for (i = 0; i < sc->sc_nqueues; i++, txr++)
1952		igc_free_transmit_buffers(txr);
1953}
1954
1955/*********************************************************************
1956 *
1957 *  Free transmit ring related data structures.
1958 *
1959 **********************************************************************/
1960void
1961igc_free_transmit_buffers(struct tx_ring *txr)
1962{
1963	struct igc_softc *sc = txr->sc;
1964	struct igc_tx_buf *txbuf;
1965	int i;
1966
1967	if (txr->tx_buffers == NULL)
1968		return;
1969
1970	txbuf = txr->tx_buffers;
1971	for (i = 0; i < sc->num_tx_desc; i++, txbuf++) {
1972		if (txbuf->map != NULL && txbuf->map->dm_nsegs > 0) {
1973			bus_dmamap_sync(txr->txdma.dma_tag, txbuf->map,
1974			    0, txbuf->map->dm_mapsize, BUS_DMASYNC_POSTWRITE);
1975			bus_dmamap_unload(txr->txdma.dma_tag, txbuf->map);
1976		}
1977		if (txbuf->m_head != NULL) {
1978			m_freem(txbuf->m_head);
1979			txbuf->m_head = NULL;
1980		}
1981		if (txbuf->map != NULL) {
1982			bus_dmamap_destroy(txr->txdma.dma_tag, txbuf->map);
1983			txbuf->map = NULL;
1984		}
1985	}
1986
1987	if (txr->tx_buffers != NULL)
1988		free(txr->tx_buffers, M_DEVBUF,
1989		    sc->num_tx_desc * sizeof(struct igc_tx_buf));
1990	txr->tx_buffers = NULL;
1991	txr->txtag = NULL;
1992}
1993
1994
1995/*********************************************************************
1996 *
1997 *  Advanced Context Descriptor setup for VLAN, CSUM or TSO
1998 *
1999 **********************************************************************/
2000
2001int
2002igc_tx_ctx_setup(struct tx_ring *txr, struct mbuf *mp, int prod,
2003    uint32_t *olinfo_status)
2004{
2005	struct igc_adv_tx_context_desc *txdesc;
2006	struct ether_header *eh = mtod(mp, struct ether_header *);
2007	struct mbuf *m;
2008	uint32_t type_tucmd_mlhl = 0;
2009	uint32_t vlan_macip_lens = 0;
2010	uint32_t iphlen;
2011	int hoff;
2012	int off = 0;
2013	uint8_t ipproto;
2014
2015	vlan_macip_lens |= (sizeof(*eh) << IGC_ADVTXD_MACLEN_SHIFT);
2016
2017	/*
2018	 * In advanced descriptors the vlan tag must
2019	 * be placed into the context descriptor. Hence
2020	 * we need to make one even if not doing offloads.
2021	 */
2022#ifdef notyet
2023#if NVLAN > 0
2024	if (ISSET(mp->m_flags, M_VLANTAG)) {
2025		uint32_t vtag = mp->m_pkthdr.ether_vtag;
2026		vlan_macip_lens |= (vtag << IGC_ADVTXD_VLAN_SHIFT);
2027		off = 1;
2028	}
2029#endif
2030#endif
2031
2032	switch (ntohs(eh->ether_type)) {
2033	case ETHERTYPE_IP: {
2034		struct ip *ip;
2035
2036		m = m_getptr(mp, sizeof(*eh), &hoff);
2037		KASSERT(m != NULL && m->m_len - hoff >= sizeof(*ip));
2038		ip = (struct ip *)(mtod(m, caddr_t) + hoff);
2039
2040		iphlen = ip->ip_hl << 2;
2041		ipproto = ip->ip_p;
2042
2043		type_tucmd_mlhl |= IGC_ADVTXD_TUCMD_IPV4;
2044		if (ISSET(mp->m_pkthdr.csum_flags, M_IPV4_CSUM_OUT)) {
2045			*olinfo_status |= IGC_TXD_POPTS_IXSM << 8;
2046			off = 1;
2047		}
2048
2049		break;
2050	}
2051#ifdef INET6
2052	case ETHERTYPE_IPV6: {
2053		struct ip6_hdr *ip6;
2054
2055		m = m_getptr(mp, sizeof(*eh), &hoff);
2056		KASSERT(m != NULL && m->m_len - hoff >= sizeof(*ip6));
2057		ip6 = (struct ip6_hdr *)(mtod(m, caddr_t) + hoff);
2058
2059		iphlen = sizeof(*ip6);
2060		ipproto = ip6->ip6_nxt;
2061
2062		type_tucmd_mlhl |= IGC_ADVTXD_TUCMD_IPV6;
2063		break;
2064	}
2065#endif
2066	default:
2067		return 0;
2068	}
2069
2070	vlan_macip_lens |= iphlen;
2071	type_tucmd_mlhl |= IGC_ADVTXD_DCMD_DEXT | IGC_ADVTXD_DTYP_CTXT;
2072
2073	switch (ipproto) {
2074	case IPPROTO_TCP:
2075		type_tucmd_mlhl |= IGC_ADVTXD_TUCMD_L4T_TCP;
2076		if (ISSET(mp->m_pkthdr.csum_flags, M_TCP_CSUM_OUT)) {
2077			*olinfo_status |= IGC_TXD_POPTS_TXSM << 8;
2078			off = 1;
2079		}
2080		break;
2081	case IPPROTO_UDP:
2082		type_tucmd_mlhl |= IGC_ADVTXD_TUCMD_L4T_UDP;
2083		if (ISSET(mp->m_pkthdr.csum_flags, M_UDP_CSUM_OUT)) {
2084			*olinfo_status |= IGC_TXD_POPTS_TXSM << 8;
2085			off = 1;
2086		}
2087		break;
2088	}
2089
2090	if (off == 0)
2091		return 0;
2092
2093	/* Now ready a context descriptor */
2094	txdesc = (struct igc_adv_tx_context_desc *)&txr->tx_base[prod];
2095
2096	/* Now copy bits into descriptor */
2097	htolem32(&txdesc->vlan_macip_lens, vlan_macip_lens);
2098	htolem32(&txdesc->type_tucmd_mlhl, type_tucmd_mlhl);
2099	htolem32(&txdesc->seqnum_seed, 0);
2100	htolem32(&txdesc->mss_l4len_idx, 0);
2101
2102	return 1;
2103}
2104
2105/*********************************************************************
2106 *
2107 *  Allocate memory for rx_buffer structures. Since we use one
2108 *  rx_buffer per received packet, the maximum number of rx_buffer's
2109 *  that we'll need is equal to the number of receive descriptors
2110 *  that we've allocated.
2111 *
2112 **********************************************************************/
2113int
2114igc_allocate_receive_buffers(struct rx_ring *rxr)
2115{
2116	struct igc_softc *sc = rxr->sc;
2117	struct igc_rx_buf *rxbuf;
2118	int i, error;
2119
2120	rxr->rx_buffers = mallocarray(sc->num_rx_desc,
2121	    sizeof(struct igc_rx_buf), M_DEVBUF, M_NOWAIT | M_ZERO);
2122	if (rxr->rx_buffers == NULL) {
2123		printf("%s: Unable to allocate rx_buffer memory\n",
2124		    DEVNAME(sc));
2125		error = ENOMEM;
2126		goto fail;
2127	}
2128
2129	rxbuf = rxr->rx_buffers;
2130	for (i = 0; i < sc->num_rx_desc; i++, rxbuf++) {
2131		error = bus_dmamap_create(rxr->rxdma.dma_tag,
2132		    MAX_JUMBO_FRAME_SIZE, 1, MAX_JUMBO_FRAME_SIZE, 0,
2133		    BUS_DMA_NOWAIT, &rxbuf->map);
2134		if (error) {
2135			printf("%s: Unable to create RX DMA map\n",
2136			    DEVNAME(sc));
2137			goto fail;
2138		}
2139	}
2140	bus_dmamap_sync(rxr->rxdma.dma_tag, rxr->rxdma.dma_map, 0,
2141	    rxr->rxdma.dma_map->dm_mapsize,
2142	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
2143
2144	return 0;
2145fail:
2146	return error;
2147}
2148
2149/*********************************************************************
2150 *
2151 *  Allocate and initialize receive structures.
2152 *
2153 **********************************************************************/
2154int
2155igc_setup_receive_structures(struct igc_softc *sc)
2156{
2157	struct rx_ring *rxr = sc->rx_rings;
2158	int i;
2159
2160	for (i = 0; i < sc->sc_nqueues; i++, rxr++) {
2161		if (igc_setup_receive_ring(rxr))
2162			goto fail;
2163	}
2164
2165	return 0;
2166fail:
2167	igc_free_receive_structures(sc);
2168	return ENOBUFS;
2169}
2170
2171/*********************************************************************
2172 *
2173 *  Initialize a receive ring and its buffers.
2174 *
2175 **********************************************************************/
2176int
2177igc_setup_receive_ring(struct rx_ring *rxr)
2178{
2179	struct igc_softc *sc = rxr->sc;
2180	struct ifnet *ifp = &sc->sc_ac.ac_if;
2181	int rsize;
2182
2183	rsize = roundup2(sc->num_rx_desc * sizeof(union igc_adv_rx_desc),
2184	    IGC_DBA_ALIGN);
2185
2186	/* Clear the ring contents. */
2187	bzero((void *)rxr->rx_base, rsize);
2188
2189	if (igc_allocate_receive_buffers(rxr))
2190		return ENOMEM;
2191
2192	/* Setup our descriptor indices. */
2193	rxr->next_to_check = 0;
2194	rxr->last_desc_filled = sc->num_rx_desc - 1;
2195
2196	if_rxr_init(&rxr->rx_ring, 2 * ((ifp->if_hardmtu / MCLBYTES) + 1),
2197	    sc->num_rx_desc - 1);
2198
2199	return 0;
2200}
2201
2202/*********************************************************************
2203 *
2204 *  Enable receive unit.
2205 *
2206 **********************************************************************/
2207void
2208igc_initialize_receive_unit(struct igc_softc *sc)
2209{
2210        struct rx_ring *rxr = sc->rx_rings;
2211        struct igc_hw *hw = &sc->hw;
2212	uint32_t rctl, rxcsum, srrctl = 0;
2213	int i;
2214
2215	/*
2216	 * Make sure receives are disabled while setting
2217	 * up the descriptor ring.
2218	 */
2219	rctl = IGC_READ_REG(hw, IGC_RCTL);
2220	IGC_WRITE_REG(hw, IGC_RCTL, rctl & ~IGC_RCTL_EN);
2221
2222	/* Setup the Receive Control Register */
2223	rctl &= ~(3 << IGC_RCTL_MO_SHIFT);
2224	rctl |= IGC_RCTL_EN | IGC_RCTL_BAM | IGC_RCTL_LBM_NO |
2225	    IGC_RCTL_RDMTS_HALF | (hw->mac.mc_filter_type << IGC_RCTL_MO_SHIFT);
2226
2227	/* Do not store bad packets */
2228	rctl &= ~IGC_RCTL_SBP;
2229
2230	/* Enable Long Packet receive */
2231	if (sc->hw.mac.max_frame_size != ETHER_MAX_LEN)
2232		rctl |= IGC_RCTL_LPE;
2233
2234	/* Strip the CRC */
2235	rctl |= IGC_RCTL_SECRC;
2236
2237	/*
2238	 * Set the interrupt throttling rate. Value is calculated
2239	 * as DEFAULT_ITR = 1/(MAX_INTS_PER_SEC * 256ns)
2240	 */
2241	IGC_WRITE_REG(hw, IGC_ITR, DEFAULT_ITR);
2242
2243	rxcsum = IGC_READ_REG(hw, IGC_RXCSUM);
2244	rxcsum &= ~IGC_RXCSUM_PCSD;
2245
2246	if (sc->sc_nqueues > 1)
2247		rxcsum |= IGC_RXCSUM_PCSD;
2248
2249	IGC_WRITE_REG(hw, IGC_RXCSUM, rxcsum);
2250
2251	if (sc->sc_nqueues > 1)
2252		igc_initialize_rss_mapping(sc);
2253
2254#if 0
2255	srrctl |= 4096 >> IGC_SRRCTL_BSIZEPKT_SHIFT;
2256	rctl |= IGC_RCTL_SZ_4096 | IGC_RCTL_BSEX;
2257#endif
2258
2259	srrctl |= 2048 >> IGC_SRRCTL_BSIZEPKT_SHIFT;
2260	rctl |= IGC_RCTL_SZ_2048;
2261
2262	/*
2263	 * If TX flow control is disabled and there's > 1 queue defined,
2264	 * enable DROP.
2265	 *
2266	 * This drops frames rather than hanging the RX MAC for all queues.
2267	 */
2268	if ((sc->sc_nqueues > 1) && (sc->fc == igc_fc_none ||
2269	    sc->fc == igc_fc_rx_pause)) {
2270		srrctl |= IGC_SRRCTL_DROP_EN;
2271	}
2272
2273	/* Setup the Base and Length of the RX descriptor rings. */
2274	for (i = 0; i < sc->sc_nqueues; i++, rxr++) {
2275		IGC_WRITE_REG(hw, IGC_RXDCTL(i), 0);
2276		uint64_t bus_addr = rxr->rxdma.dma_map->dm_segs[0].ds_addr;
2277		uint32_t rxdctl;
2278
2279		srrctl |= IGC_SRRCTL_DESCTYPE_ADV_ONEBUF;
2280
2281		IGC_WRITE_REG(hw, IGC_RDLEN(i),
2282		    sc->num_rx_desc * sizeof(union igc_adv_rx_desc));
2283		IGC_WRITE_REG(hw, IGC_RDBAH(i), (uint32_t)(bus_addr >> 32));
2284		IGC_WRITE_REG(hw, IGC_RDBAL(i), (uint32_t)bus_addr);
2285		IGC_WRITE_REG(hw, IGC_SRRCTL(i), srrctl);
2286
2287		/* Setup the Head and Tail Descriptor Pointers */
2288		IGC_WRITE_REG(hw, IGC_RDH(i), 0);
2289		IGC_WRITE_REG(hw, IGC_RDT(i), 0);
2290
2291		/* Enable this Queue */
2292		rxdctl = IGC_READ_REG(hw, IGC_RXDCTL(i));
2293		rxdctl |= IGC_RXDCTL_QUEUE_ENABLE;
2294		rxdctl &= 0xFFF00000;
2295		rxdctl |= IGC_RX_PTHRESH;
2296		rxdctl |= IGC_RX_HTHRESH << 8;
2297		rxdctl |= IGC_RX_WTHRESH << 16;
2298		IGC_WRITE_REG(hw, IGC_RXDCTL(i), rxdctl);
2299	}
2300
2301	/* Make sure VLAN Filters are off */
2302	rctl &= ~IGC_RCTL_VFE;
2303
2304	/* Write out the settings */
2305	IGC_WRITE_REG(hw, IGC_RCTL, rctl);
2306}
2307
2308/*********************************************************************
2309 *
2310 *  Free all receive rings.
2311 *
2312 **********************************************************************/
2313void
2314igc_free_receive_structures(struct igc_softc *sc)
2315{
2316	struct rx_ring *rxr;
2317	int i;
2318
2319	for (i = 0, rxr = sc->rx_rings; i < sc->sc_nqueues; i++, rxr++)
2320		if_rxr_init(&rxr->rx_ring, 0, 0);
2321
2322	for (i = 0, rxr = sc->rx_rings; i < sc->sc_nqueues; i++, rxr++)
2323		igc_free_receive_buffers(rxr);
2324}
2325
2326/*********************************************************************
2327 *
2328 *  Free receive ring data structures
2329 *
2330 **********************************************************************/
2331void
2332igc_free_receive_buffers(struct rx_ring *rxr)
2333{
2334	struct igc_softc *sc = rxr->sc;
2335	struct igc_rx_buf *rxbuf;
2336	int i;
2337
2338	if (rxr->rx_buffers != NULL) {
2339		for (i = 0; i < sc->num_rx_desc; i++) {
2340			rxbuf = &rxr->rx_buffers[i];
2341			if (rxbuf->buf != NULL) {
2342				bus_dmamap_sync(rxr->rxdma.dma_tag, rxbuf->map,
2343				    0, rxbuf->map->dm_mapsize,
2344				    BUS_DMASYNC_POSTREAD);
2345				bus_dmamap_unload(rxr->rxdma.dma_tag,
2346				    rxbuf->map);
2347				m_freem(rxbuf->buf);
2348				rxbuf->buf = NULL;
2349			}
2350			bus_dmamap_destroy(rxr->rxdma.dma_tag, rxbuf->map);
2351			rxbuf->map = NULL;
2352		}
2353		free(rxr->rx_buffers, M_DEVBUF,
2354		    sc->num_rx_desc * sizeof(struct igc_rx_buf));
2355		rxr->rx_buffers = NULL;
2356	}
2357}
2358
2359/*
2360 * Initialise the RSS mapping for NICs that support multiple transmit/
2361 * receive rings.
2362 */
2363void
2364igc_initialize_rss_mapping(struct igc_softc *sc)
2365{
2366	struct igc_hw *hw = &sc->hw;
2367	uint32_t rss_key[10], mrqc, reta, shift = 0;
2368	int i, queue_id;
2369
2370	/*
2371	 * The redirection table controls which destination
2372	 * queue each bucket redirects traffic to.
2373	 * Each DWORD represents four queues, with the LSB
2374	 * being the first queue in the DWORD.
2375	 *
2376	 * This just allocates buckets to queues using round-robin
2377	 * allocation.
2378	 *
2379	 * NOTE: It Just Happens to line up with the default
2380	 * RSS allocation method.
2381	 */
2382
2383	/* Warning FM follows */
2384	reta = 0;
2385	for (i = 0; i < 128; i++) {
2386		queue_id = (i % sc->sc_nqueues);
2387		/* Adjust if required */
2388		queue_id = queue_id << shift;
2389
2390		/*
2391		 * The low 8 bits are for hash value (n+0);
2392		 * The next 8 bits are for hash value (n+1), etc.
2393		 */
2394		reta = reta >> 8;
2395		reta = reta | ( ((uint32_t) queue_id) << 24);
2396		if ((i & 3) == 3) {
2397			IGC_WRITE_REG(hw, IGC_RETA(i >> 2), reta);
2398			reta = 0;
2399		}
2400	}
2401
2402	/*
2403	 * MRQC: Multiple Receive Queues Command
2404	 * Set queuing to RSS control, number depends on the device.
2405	 */
2406	mrqc = IGC_MRQC_ENABLE_RSS_4Q;
2407
2408	/* Set up random bits */
2409        stoeplitz_to_key(&rss_key, sizeof(rss_key));
2410
2411	/* Now fill our hash function seeds */
2412	for (i = 0; i < 10; i++)
2413		IGC_WRITE_REG_ARRAY(hw, IGC_RSSRK(0), i, rss_key[i]);
2414
2415	/*
2416	 * Configure the RSS fields to hash upon.
2417	 */
2418	mrqc |= (IGC_MRQC_RSS_FIELD_IPV4 | IGC_MRQC_RSS_FIELD_IPV4_TCP);
2419	mrqc |= (IGC_MRQC_RSS_FIELD_IPV6 | IGC_MRQC_RSS_FIELD_IPV6_TCP);
2420	mrqc |= IGC_MRQC_RSS_FIELD_IPV6_TCP_EX;
2421
2422	IGC_WRITE_REG(hw, IGC_MRQC, mrqc);
2423}
2424
2425/*
2426 * igc_get_hw_control sets the {CTRL_EXT|FWSM}:DRV_LOAD bit.
2427 * For ASF and Pass Through versions of f/w this means
2428 * that the driver is loaded. For AMT version type f/w
2429 * this means that the network i/f is open.
2430 */
2431void
2432igc_get_hw_control(struct igc_softc *sc)
2433{
2434	uint32_t ctrl_ext;
2435
2436	ctrl_ext = IGC_READ_REG(&sc->hw, IGC_CTRL_EXT);
2437	IGC_WRITE_REG(&sc->hw, IGC_CTRL_EXT, ctrl_ext | IGC_CTRL_EXT_DRV_LOAD);
2438}
2439
2440/*
2441 * igc_release_hw_control resets {CTRL_EXT|FWSM}:DRV_LOAD bit.
2442 * For ASF and Pass Through versions of f/w this means that
2443 * the driver is no longer loaded. For AMT versions of the
2444 * f/w this means that the network i/f is closed.
2445 */
2446void
2447igc_release_hw_control(struct igc_softc *sc)
2448{
2449	uint32_t ctrl_ext;
2450
2451	ctrl_ext = IGC_READ_REG(&sc->hw, IGC_CTRL_EXT);
2452	IGC_WRITE_REG(&sc->hw, IGC_CTRL_EXT, ctrl_ext & ~IGC_CTRL_EXT_DRV_LOAD);
2453}
2454
2455int
2456igc_is_valid_ether_addr(uint8_t *addr)
2457{
2458	char zero_addr[6] = { 0, 0, 0, 0, 0, 0 };
2459
2460	if ((addr[0] & 1) || (!bcmp(addr, zero_addr, ETHER_ADDR_LEN))) {
2461		return 0;
2462	}
2463
2464	return 1;
2465}
2466