1 /*******************************************************************************
3 Intel(R) 82576 Virtual Function Linux driver
4 Copyright(c) 2009 - 2012 Intel Corporation.
6 This program is free software; you can redistribute it and/or modify it
7 under the terms and conditions of the GNU General Public License,
8 version 2, as published by the Free Software Foundation.
10 This program is distributed in the hope it will be useful, but WITHOUT
11 ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 You should have received a copy of the GNU General Public License along with
16 this program; if not, see <http://www.gnu.org/licenses/>.
18 The full GNU General Public License is included in this distribution in
19 the file called "COPYING".
22 e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
23 Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
25 *******************************************************************************/
27 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
29 #include <linux/module.h>
30 #include <linux/types.h>
31 #include <linux/init.h>
32 #include <linux/pci.h>
33 #include <linux/vmalloc.h>
34 #include <linux/pagemap.h>
35 #include <linux/delay.h>
36 #include <linux/netdevice.h>
37 #include <linux/tcp.h>
38 #include <linux/ipv6.h>
39 #include <linux/slab.h>
40 #include <net/checksum.h>
41 #include <net/ip6_checksum.h>
42 #include <linux/mii.h>
43 #include <linux/ethtool.h>
44 #include <linux/if_vlan.h>
45 #include <linux/prefetch.h>
46 #include <linux/sctp.h>
50 #define DRV_VERSION "2.0.2-k"
51 char igbvf_driver_name[] = "igbvf";
52 const char igbvf_driver_version[] = DRV_VERSION;
53 static const char igbvf_driver_string[] =
54 "Intel(R) Gigabit Virtual Function Network Driver";
55 static const char igbvf_copyright[] =
56 "Copyright (c) 2009 - 2012 Intel Corporation.";
58 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
59 static int debug = -1;
60 module_param(debug, int, 0);
61 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
63 static int igbvf_poll(struct napi_struct *napi, int budget);
64 static void igbvf_reset(struct igbvf_adapter *);
65 static void igbvf_set_interrupt_capability(struct igbvf_adapter *);
66 static void igbvf_reset_interrupt_capability(struct igbvf_adapter *);
68 static struct igbvf_info igbvf_vf_info = {
72 .init_ops = e1000_init_function_pointers_vf,
75 static struct igbvf_info igbvf_i350_vf_info = {
76 .mac = e1000_vfadapt_i350,
79 .init_ops = e1000_init_function_pointers_vf,
82 static const struct igbvf_info *igbvf_info_tbl[] = {
83 [board_vf] = &igbvf_vf_info,
84 [board_i350_vf] = &igbvf_i350_vf_info,
88 * igbvf_desc_unused - calculate if we have unused descriptors
89 * @rx_ring: address of receive ring structure
91 static int igbvf_desc_unused(struct igbvf_ring *ring)
93 if (ring->next_to_clean > ring->next_to_use)
94 return ring->next_to_clean - ring->next_to_use - 1;
96 return ring->count + ring->next_to_clean - ring->next_to_use - 1;
100 * igbvf_receive_skb - helper function to handle Rx indications
101 * @adapter: board private structure
102 * @status: descriptor status field as written by hardware
103 * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
104 * @skb: pointer to sk_buff to be indicated to stack
106 static void igbvf_receive_skb(struct igbvf_adapter *adapter,
107 struct net_device *netdev,
109 u32 status, u16 vlan)
113 if (status & E1000_RXD_STAT_VP) {
114 if ((adapter->flags & IGBVF_FLAG_RX_LB_VLAN_BSWAP) &&
115 (status & E1000_RXDEXT_STATERR_LB))
116 vid = be16_to_cpu(vlan) & E1000_RXD_SPC_VLAN_MASK;
118 vid = le16_to_cpu(vlan) & E1000_RXD_SPC_VLAN_MASK;
119 if (test_bit(vid, adapter->active_vlans))
120 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
123 napi_gro_receive(&adapter->rx_ring->napi, skb);
126 static inline void igbvf_rx_checksum_adv(struct igbvf_adapter *adapter,
127 u32 status_err, struct sk_buff *skb)
129 skb_checksum_none_assert(skb);
131 /* Ignore Checksum bit is set or checksum is disabled through ethtool */
132 if ((status_err & E1000_RXD_STAT_IXSM) ||
133 (adapter->flags & IGBVF_FLAG_RX_CSUM_DISABLED))
136 /* TCP/UDP checksum error bit is set */
138 (E1000_RXDEXT_STATERR_TCPE | E1000_RXDEXT_STATERR_IPE)) {
139 /* let the stack verify checksum errors */
140 adapter->hw_csum_err++;
144 /* It must be a TCP or UDP packet with a valid checksum */
145 if (status_err & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS))
146 skb->ip_summed = CHECKSUM_UNNECESSARY;
148 adapter->hw_csum_good++;
152 * igbvf_alloc_rx_buffers - Replace used receive buffers; packet split
153 * @rx_ring: address of ring structure to repopulate
154 * @cleaned_count: number of buffers to repopulate
156 static void igbvf_alloc_rx_buffers(struct igbvf_ring *rx_ring,
159 struct igbvf_adapter *adapter = rx_ring->adapter;
160 struct net_device *netdev = adapter->netdev;
161 struct pci_dev *pdev = adapter->pdev;
162 union e1000_adv_rx_desc *rx_desc;
163 struct igbvf_buffer *buffer_info;
168 i = rx_ring->next_to_use;
169 buffer_info = &rx_ring->buffer_info[i];
171 if (adapter->rx_ps_hdr_size)
172 bufsz = adapter->rx_ps_hdr_size;
174 bufsz = adapter->rx_buffer_len;
176 while (cleaned_count--) {
177 rx_desc = IGBVF_RX_DESC_ADV(*rx_ring, i);
179 if (adapter->rx_ps_hdr_size && !buffer_info->page_dma) {
180 if (!buffer_info->page) {
181 buffer_info->page = alloc_page(GFP_ATOMIC);
182 if (!buffer_info->page) {
183 adapter->alloc_rx_buff_failed++;
186 buffer_info->page_offset = 0;
188 buffer_info->page_offset ^= PAGE_SIZE / 2;
190 buffer_info->page_dma =
191 dma_map_page(&pdev->dev, buffer_info->page,
192 buffer_info->page_offset,
195 if (dma_mapping_error(&pdev->dev,
196 buffer_info->page_dma)) {
197 __free_page(buffer_info->page);
198 buffer_info->page = NULL;
199 dev_err(&pdev->dev, "RX DMA map failed\n");
204 if (!buffer_info->skb) {
205 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
207 adapter->alloc_rx_buff_failed++;
211 buffer_info->skb = skb;
212 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
215 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
216 dev_kfree_skb(buffer_info->skb);
217 buffer_info->skb = NULL;
218 dev_err(&pdev->dev, "RX DMA map failed\n");
222 /* Refresh the desc even if buffer_addrs didn't change because
223 * each write-back erases this info.
225 if (adapter->rx_ps_hdr_size) {
226 rx_desc->read.pkt_addr =
227 cpu_to_le64(buffer_info->page_dma);
228 rx_desc->read.hdr_addr = cpu_to_le64(buffer_info->dma);
230 rx_desc->read.pkt_addr = cpu_to_le64(buffer_info->dma);
231 rx_desc->read.hdr_addr = 0;
235 if (i == rx_ring->count)
237 buffer_info = &rx_ring->buffer_info[i];
241 if (rx_ring->next_to_use != i) {
242 rx_ring->next_to_use = i;
244 i = (rx_ring->count - 1);
248 /* Force memory writes to complete before letting h/w
249 * know there are new descriptors to fetch. (Only
250 * applicable for weak-ordered memory model archs,
254 writel(i, adapter->hw.hw_addr + rx_ring->tail);
259 * igbvf_clean_rx_irq - Send received data up the network stack; legacy
260 * @adapter: board private structure
262 * the return value indicates whether actual cleaning was done, there
263 * is no guarantee that everything was cleaned
265 static bool igbvf_clean_rx_irq(struct igbvf_adapter *adapter,
266 int *work_done, int work_to_do)
268 struct igbvf_ring *rx_ring = adapter->rx_ring;
269 struct net_device *netdev = adapter->netdev;
270 struct pci_dev *pdev = adapter->pdev;
271 union e1000_adv_rx_desc *rx_desc, *next_rxd;
272 struct igbvf_buffer *buffer_info, *next_buffer;
274 bool cleaned = false;
275 int cleaned_count = 0;
276 unsigned int total_bytes = 0, total_packets = 0;
278 u32 length, hlen, staterr;
280 i = rx_ring->next_to_clean;
281 rx_desc = IGBVF_RX_DESC_ADV(*rx_ring, i);
282 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
284 while (staterr & E1000_RXD_STAT_DD) {
285 if (*work_done >= work_to_do)
288 rmb(); /* read descriptor and rx_buffer_info after status DD */
290 buffer_info = &rx_ring->buffer_info[i];
292 /* HW will not DMA in data larger than the given buffer, even
293 * if it parses the (NFS, of course) header to be larger. In
294 * that case, it fills the header buffer and spills the rest
297 hlen = (le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.hdr_info)
298 & E1000_RXDADV_HDRBUFLEN_MASK) >>
299 E1000_RXDADV_HDRBUFLEN_SHIFT;
300 if (hlen > adapter->rx_ps_hdr_size)
301 hlen = adapter->rx_ps_hdr_size;
303 length = le16_to_cpu(rx_desc->wb.upper.length);
307 skb = buffer_info->skb;
308 prefetch(skb->data - NET_IP_ALIGN);
309 buffer_info->skb = NULL;
310 if (!adapter->rx_ps_hdr_size) {
311 dma_unmap_single(&pdev->dev, buffer_info->dma,
312 adapter->rx_buffer_len,
314 buffer_info->dma = 0;
315 skb_put(skb, length);
319 if (!skb_shinfo(skb)->nr_frags) {
320 dma_unmap_single(&pdev->dev, buffer_info->dma,
321 adapter->rx_ps_hdr_size,
323 buffer_info->dma = 0;
328 dma_unmap_page(&pdev->dev, buffer_info->page_dma,
331 buffer_info->page_dma = 0;
333 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
335 buffer_info->page_offset,
338 if ((adapter->rx_buffer_len > (PAGE_SIZE / 2)) ||
339 (page_count(buffer_info->page) != 1))
340 buffer_info->page = NULL;
342 get_page(buffer_info->page);
345 skb->data_len += length;
346 skb->truesize += PAGE_SIZE / 2;
350 if (i == rx_ring->count)
352 next_rxd = IGBVF_RX_DESC_ADV(*rx_ring, i);
354 next_buffer = &rx_ring->buffer_info[i];
356 if (!(staterr & E1000_RXD_STAT_EOP)) {
357 buffer_info->skb = next_buffer->skb;
358 buffer_info->dma = next_buffer->dma;
359 next_buffer->skb = skb;
360 next_buffer->dma = 0;
364 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
365 dev_kfree_skb_irq(skb);
369 total_bytes += skb->len;
372 igbvf_rx_checksum_adv(adapter, staterr, skb);
374 skb->protocol = eth_type_trans(skb, netdev);
376 igbvf_receive_skb(adapter, netdev, skb, staterr,
377 rx_desc->wb.upper.vlan);
380 rx_desc->wb.upper.status_error = 0;
382 /* return some buffers to hardware, one at a time is too slow */
383 if (cleaned_count >= IGBVF_RX_BUFFER_WRITE) {
384 igbvf_alloc_rx_buffers(rx_ring, cleaned_count);
388 /* use prefetched values */
390 buffer_info = next_buffer;
392 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
395 rx_ring->next_to_clean = i;
396 cleaned_count = igbvf_desc_unused(rx_ring);
399 igbvf_alloc_rx_buffers(rx_ring, cleaned_count);
401 adapter->total_rx_packets += total_packets;
402 adapter->total_rx_bytes += total_bytes;
403 adapter->net_stats.rx_bytes += total_bytes;
404 adapter->net_stats.rx_packets += total_packets;
408 static void igbvf_put_txbuf(struct igbvf_adapter *adapter,
409 struct igbvf_buffer *buffer_info)
411 if (buffer_info->dma) {
412 if (buffer_info->mapped_as_page)
413 dma_unmap_page(&adapter->pdev->dev,
418 dma_unmap_single(&adapter->pdev->dev,
422 buffer_info->dma = 0;
424 if (buffer_info->skb) {
425 dev_kfree_skb_any(buffer_info->skb);
426 buffer_info->skb = NULL;
428 buffer_info->time_stamp = 0;
432 * igbvf_setup_tx_resources - allocate Tx resources (Descriptors)
433 * @adapter: board private structure
435 * Return 0 on success, negative on failure
437 int igbvf_setup_tx_resources(struct igbvf_adapter *adapter,
438 struct igbvf_ring *tx_ring)
440 struct pci_dev *pdev = adapter->pdev;
443 size = sizeof(struct igbvf_buffer) * tx_ring->count;
444 tx_ring->buffer_info = vzalloc(size);
445 if (!tx_ring->buffer_info)
448 /* round up to nearest 4K */
449 tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
450 tx_ring->size = ALIGN(tx_ring->size, 4096);
452 tx_ring->desc = dma_alloc_coherent(&pdev->dev, tx_ring->size,
453 &tx_ring->dma, GFP_KERNEL);
457 tx_ring->adapter = adapter;
458 tx_ring->next_to_use = 0;
459 tx_ring->next_to_clean = 0;
463 vfree(tx_ring->buffer_info);
464 dev_err(&adapter->pdev->dev,
465 "Unable to allocate memory for the transmit descriptor ring\n");
470 * igbvf_setup_rx_resources - allocate Rx resources (Descriptors)
471 * @adapter: board private structure
473 * Returns 0 on success, negative on failure
475 int igbvf_setup_rx_resources(struct igbvf_adapter *adapter,
476 struct igbvf_ring *rx_ring)
478 struct pci_dev *pdev = adapter->pdev;
481 size = sizeof(struct igbvf_buffer) * rx_ring->count;
482 rx_ring->buffer_info = vzalloc(size);
483 if (!rx_ring->buffer_info)
486 desc_len = sizeof(union e1000_adv_rx_desc);
488 /* Round up to nearest 4K */
489 rx_ring->size = rx_ring->count * desc_len;
490 rx_ring->size = ALIGN(rx_ring->size, 4096);
492 rx_ring->desc = dma_alloc_coherent(&pdev->dev, rx_ring->size,
493 &rx_ring->dma, GFP_KERNEL);
497 rx_ring->next_to_clean = 0;
498 rx_ring->next_to_use = 0;
500 rx_ring->adapter = adapter;
505 vfree(rx_ring->buffer_info);
506 rx_ring->buffer_info = NULL;
507 dev_err(&adapter->pdev->dev,
508 "Unable to allocate memory for the receive descriptor ring\n");
513 * igbvf_clean_tx_ring - Free Tx Buffers
514 * @tx_ring: ring to be cleaned
516 static void igbvf_clean_tx_ring(struct igbvf_ring *tx_ring)
518 struct igbvf_adapter *adapter = tx_ring->adapter;
519 struct igbvf_buffer *buffer_info;
523 if (!tx_ring->buffer_info)
526 /* Free all the Tx ring sk_buffs */
527 for (i = 0; i < tx_ring->count; i++) {
528 buffer_info = &tx_ring->buffer_info[i];
529 igbvf_put_txbuf(adapter, buffer_info);
532 size = sizeof(struct igbvf_buffer) * tx_ring->count;
533 memset(tx_ring->buffer_info, 0, size);
535 /* Zero out the descriptor ring */
536 memset(tx_ring->desc, 0, tx_ring->size);
538 tx_ring->next_to_use = 0;
539 tx_ring->next_to_clean = 0;
541 writel(0, adapter->hw.hw_addr + tx_ring->head);
542 writel(0, adapter->hw.hw_addr + tx_ring->tail);
546 * igbvf_free_tx_resources - Free Tx Resources per Queue
547 * @tx_ring: ring to free resources from
549 * Free all transmit software resources
551 void igbvf_free_tx_resources(struct igbvf_ring *tx_ring)
553 struct pci_dev *pdev = tx_ring->adapter->pdev;
555 igbvf_clean_tx_ring(tx_ring);
557 vfree(tx_ring->buffer_info);
558 tx_ring->buffer_info = NULL;
560 dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
563 tx_ring->desc = NULL;
567 * igbvf_clean_rx_ring - Free Rx Buffers per Queue
568 * @adapter: board private structure
570 static void igbvf_clean_rx_ring(struct igbvf_ring *rx_ring)
572 struct igbvf_adapter *adapter = rx_ring->adapter;
573 struct igbvf_buffer *buffer_info;
574 struct pci_dev *pdev = adapter->pdev;
578 if (!rx_ring->buffer_info)
581 /* Free all the Rx ring sk_buffs */
582 for (i = 0; i < rx_ring->count; i++) {
583 buffer_info = &rx_ring->buffer_info[i];
584 if (buffer_info->dma) {
585 if (adapter->rx_ps_hdr_size) {
586 dma_unmap_single(&pdev->dev, buffer_info->dma,
587 adapter->rx_ps_hdr_size,
590 dma_unmap_single(&pdev->dev, buffer_info->dma,
591 adapter->rx_buffer_len,
594 buffer_info->dma = 0;
597 if (buffer_info->skb) {
598 dev_kfree_skb(buffer_info->skb);
599 buffer_info->skb = NULL;
602 if (buffer_info->page) {
603 if (buffer_info->page_dma)
604 dma_unmap_page(&pdev->dev,
605 buffer_info->page_dma,
608 put_page(buffer_info->page);
609 buffer_info->page = NULL;
610 buffer_info->page_dma = 0;
611 buffer_info->page_offset = 0;
615 size = sizeof(struct igbvf_buffer) * rx_ring->count;
616 memset(rx_ring->buffer_info, 0, size);
618 /* Zero out the descriptor ring */
619 memset(rx_ring->desc, 0, rx_ring->size);
621 rx_ring->next_to_clean = 0;
622 rx_ring->next_to_use = 0;
624 writel(0, adapter->hw.hw_addr + rx_ring->head);
625 writel(0, adapter->hw.hw_addr + rx_ring->tail);
629 * igbvf_free_rx_resources - Free Rx Resources
630 * @rx_ring: ring to clean the resources from
632 * Free all receive software resources
635 void igbvf_free_rx_resources(struct igbvf_ring *rx_ring)
637 struct pci_dev *pdev = rx_ring->adapter->pdev;
639 igbvf_clean_rx_ring(rx_ring);
641 vfree(rx_ring->buffer_info);
642 rx_ring->buffer_info = NULL;
644 dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
646 rx_ring->desc = NULL;
650 * igbvf_update_itr - update the dynamic ITR value based on statistics
651 * @adapter: pointer to adapter
652 * @itr_setting: current adapter->itr
653 * @packets: the number of packets during this measurement interval
654 * @bytes: the number of bytes during this measurement interval
656 * Stores a new ITR value based on packets and byte counts during the last
657 * interrupt. The advantage of per interrupt computation is faster updates
658 * and more accurate ITR for the current traffic pattern. Constants in this
659 * function were computed based on theoretical maximum wire speed and thresholds
660 * were set based on testing data as well as attempting to minimize response
661 * time while increasing bulk throughput.
663 static enum latency_range igbvf_update_itr(struct igbvf_adapter *adapter,
664 enum latency_range itr_setting,
665 int packets, int bytes)
667 enum latency_range retval = itr_setting;
670 goto update_itr_done;
672 switch (itr_setting) {
674 /* handle TSO and jumbo frames */
675 if (bytes/packets > 8000)
676 retval = bulk_latency;
677 else if ((packets < 5) && (bytes > 512))
678 retval = low_latency;
680 case low_latency: /* 50 usec aka 20000 ints/s */
682 /* this if handles the TSO accounting */
683 if (bytes/packets > 8000)
684 retval = bulk_latency;
685 else if ((packets < 10) || ((bytes/packets) > 1200))
686 retval = bulk_latency;
687 else if ((packets > 35))
688 retval = lowest_latency;
689 } else if (bytes/packets > 2000) {
690 retval = bulk_latency;
691 } else if (packets <= 2 && bytes < 512) {
692 retval = lowest_latency;
695 case bulk_latency: /* 250 usec aka 4000 ints/s */
698 retval = low_latency;
699 } else if (bytes < 6000) {
700 retval = low_latency;
711 static int igbvf_range_to_itr(enum latency_range current_range)
715 switch (current_range) {
716 /* counts and packets in update_itr are dependent on these numbers */
718 new_itr = IGBVF_70K_ITR;
721 new_itr = IGBVF_20K_ITR;
724 new_itr = IGBVF_4K_ITR;
727 new_itr = IGBVF_START_ITR;
733 static void igbvf_set_itr(struct igbvf_adapter *adapter)
737 adapter->tx_ring->itr_range =
738 igbvf_update_itr(adapter,
739 adapter->tx_ring->itr_val,
740 adapter->total_tx_packets,
741 adapter->total_tx_bytes);
743 /* conservative mode (itr 3) eliminates the lowest_latency setting */
744 if (adapter->requested_itr == 3 &&
745 adapter->tx_ring->itr_range == lowest_latency)
746 adapter->tx_ring->itr_range = low_latency;
748 new_itr = igbvf_range_to_itr(adapter->tx_ring->itr_range);
750 if (new_itr != adapter->tx_ring->itr_val) {
751 u32 current_itr = adapter->tx_ring->itr_val;
752 /* this attempts to bias the interrupt rate towards Bulk
753 * by adding intermediate steps when interrupt rate is
756 new_itr = new_itr > current_itr ?
757 min(current_itr + (new_itr >> 2), new_itr) :
759 adapter->tx_ring->itr_val = new_itr;
761 adapter->tx_ring->set_itr = 1;
764 adapter->rx_ring->itr_range =
765 igbvf_update_itr(adapter, adapter->rx_ring->itr_val,
766 adapter->total_rx_packets,
767 adapter->total_rx_bytes);
768 if (adapter->requested_itr == 3 &&
769 adapter->rx_ring->itr_range == lowest_latency)
770 adapter->rx_ring->itr_range = low_latency;
772 new_itr = igbvf_range_to_itr(adapter->rx_ring->itr_range);
774 if (new_itr != adapter->rx_ring->itr_val) {
775 u32 current_itr = adapter->rx_ring->itr_val;
777 new_itr = new_itr > current_itr ?
778 min(current_itr + (new_itr >> 2), new_itr) :
780 adapter->rx_ring->itr_val = new_itr;
782 adapter->rx_ring->set_itr = 1;
787 * igbvf_clean_tx_irq - Reclaim resources after transmit completes
788 * @adapter: board private structure
790 * returns true if ring is completely cleaned
792 static bool igbvf_clean_tx_irq(struct igbvf_ring *tx_ring)
794 struct igbvf_adapter *adapter = tx_ring->adapter;
795 struct net_device *netdev = adapter->netdev;
796 struct igbvf_buffer *buffer_info;
798 union e1000_adv_tx_desc *tx_desc, *eop_desc;
799 unsigned int total_bytes = 0, total_packets = 0;
800 unsigned int i, count = 0;
801 bool cleaned = false;
803 i = tx_ring->next_to_clean;
804 buffer_info = &tx_ring->buffer_info[i];
805 eop_desc = buffer_info->next_to_watch;
808 /* if next_to_watch is not set then there is no work pending */
812 /* prevent any other reads prior to eop_desc */
813 read_barrier_depends();
815 /* if DD is not set pending work has not been completed */
816 if (!(eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)))
819 /* clear next_to_watch to prevent false hangs */
820 buffer_info->next_to_watch = NULL;
822 for (cleaned = false; !cleaned; count++) {
823 tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
824 cleaned = (tx_desc == eop_desc);
825 skb = buffer_info->skb;
828 unsigned int segs, bytecount;
830 /* gso_segs is currently only valid for tcp */
831 segs = skb_shinfo(skb)->gso_segs ?: 1;
832 /* multiply data chunks by size of headers */
833 bytecount = ((segs - 1) * skb_headlen(skb)) +
835 total_packets += segs;
836 total_bytes += bytecount;
839 igbvf_put_txbuf(adapter, buffer_info);
840 tx_desc->wb.status = 0;
843 if (i == tx_ring->count)
846 buffer_info = &tx_ring->buffer_info[i];
849 eop_desc = buffer_info->next_to_watch;
850 } while (count < tx_ring->count);
852 tx_ring->next_to_clean = i;
854 if (unlikely(count && netif_carrier_ok(netdev) &&
855 igbvf_desc_unused(tx_ring) >= IGBVF_TX_QUEUE_WAKE)) {
856 /* Make sure that anybody stopping the queue after this
857 * sees the new next_to_clean.
860 if (netif_queue_stopped(netdev) &&
861 !(test_bit(__IGBVF_DOWN, &adapter->state))) {
862 netif_wake_queue(netdev);
863 ++adapter->restart_queue;
867 adapter->net_stats.tx_bytes += total_bytes;
868 adapter->net_stats.tx_packets += total_packets;
869 return count < tx_ring->count;
872 static irqreturn_t igbvf_msix_other(int irq, void *data)
874 struct net_device *netdev = data;
875 struct igbvf_adapter *adapter = netdev_priv(netdev);
876 struct e1000_hw *hw = &adapter->hw;
878 adapter->int_counter1++;
880 hw->mac.get_link_status = 1;
881 if (!test_bit(__IGBVF_DOWN, &adapter->state))
882 mod_timer(&adapter->watchdog_timer, jiffies + 1);
884 ew32(EIMS, adapter->eims_other);
889 static irqreturn_t igbvf_intr_msix_tx(int irq, void *data)
891 struct net_device *netdev = data;
892 struct igbvf_adapter *adapter = netdev_priv(netdev);
893 struct e1000_hw *hw = &adapter->hw;
894 struct igbvf_ring *tx_ring = adapter->tx_ring;
896 if (tx_ring->set_itr) {
897 writel(tx_ring->itr_val,
898 adapter->hw.hw_addr + tx_ring->itr_register);
899 adapter->tx_ring->set_itr = 0;
902 adapter->total_tx_bytes = 0;
903 adapter->total_tx_packets = 0;
905 /* auto mask will automatically re-enable the interrupt when we write
908 if (!igbvf_clean_tx_irq(tx_ring))
909 /* Ring was not completely cleaned, so fire another interrupt */
910 ew32(EICS, tx_ring->eims_value);
912 ew32(EIMS, tx_ring->eims_value);
917 static irqreturn_t igbvf_intr_msix_rx(int irq, void *data)
919 struct net_device *netdev = data;
920 struct igbvf_adapter *adapter = netdev_priv(netdev);
922 adapter->int_counter0++;
924 /* Write the ITR value calculated at the end of the
925 * previous interrupt.
927 if (adapter->rx_ring->set_itr) {
928 writel(adapter->rx_ring->itr_val,
929 adapter->hw.hw_addr + adapter->rx_ring->itr_register);
930 adapter->rx_ring->set_itr = 0;
933 if (napi_schedule_prep(&adapter->rx_ring->napi)) {
934 adapter->total_rx_bytes = 0;
935 adapter->total_rx_packets = 0;
936 __napi_schedule(&adapter->rx_ring->napi);
942 #define IGBVF_NO_QUEUE -1
944 static void igbvf_assign_vector(struct igbvf_adapter *adapter, int rx_queue,
945 int tx_queue, int msix_vector)
947 struct e1000_hw *hw = &adapter->hw;
950 /* 82576 uses a table-based method for assigning vectors.
951 * Each queue has a single entry in the table to which we write
952 * a vector number along with a "valid" bit. Sadly, the layout
953 * of the table is somewhat counterintuitive.
955 if (rx_queue > IGBVF_NO_QUEUE) {
956 index = (rx_queue >> 1);
957 ivar = array_er32(IVAR0, index);
958 if (rx_queue & 0x1) {
959 /* vector goes into third byte of register */
960 ivar = ivar & 0xFF00FFFF;
961 ivar |= (msix_vector | E1000_IVAR_VALID) << 16;
963 /* vector goes into low byte of register */
964 ivar = ivar & 0xFFFFFF00;
965 ivar |= msix_vector | E1000_IVAR_VALID;
967 adapter->rx_ring[rx_queue].eims_value = 1 << msix_vector;
968 array_ew32(IVAR0, index, ivar);
970 if (tx_queue > IGBVF_NO_QUEUE) {
971 index = (tx_queue >> 1);
972 ivar = array_er32(IVAR0, index);
973 if (tx_queue & 0x1) {
974 /* vector goes into high byte of register */
975 ivar = ivar & 0x00FFFFFF;
976 ivar |= (msix_vector | E1000_IVAR_VALID) << 24;
978 /* vector goes into second byte of register */
979 ivar = ivar & 0xFFFF00FF;
980 ivar |= (msix_vector | E1000_IVAR_VALID) << 8;
982 adapter->tx_ring[tx_queue].eims_value = 1 << msix_vector;
983 array_ew32(IVAR0, index, ivar);
988 * igbvf_configure_msix - Configure MSI-X hardware
989 * @adapter: board private structure
991 * igbvf_configure_msix sets up the hardware to properly
992 * generate MSI-X interrupts.
994 static void igbvf_configure_msix(struct igbvf_adapter *adapter)
997 struct e1000_hw *hw = &adapter->hw;
998 struct igbvf_ring *tx_ring = adapter->tx_ring;
999 struct igbvf_ring *rx_ring = adapter->rx_ring;
1002 adapter->eims_enable_mask = 0;
1004 igbvf_assign_vector(adapter, IGBVF_NO_QUEUE, 0, vector++);
1005 adapter->eims_enable_mask |= tx_ring->eims_value;
1006 writel(tx_ring->itr_val, hw->hw_addr + tx_ring->itr_register);
1007 igbvf_assign_vector(adapter, 0, IGBVF_NO_QUEUE, vector++);
1008 adapter->eims_enable_mask |= rx_ring->eims_value;
1009 writel(rx_ring->itr_val, hw->hw_addr + rx_ring->itr_register);
1011 /* set vector for other causes, i.e. link changes */
1013 tmp = (vector++ | E1000_IVAR_VALID);
1015 ew32(IVAR_MISC, tmp);
1017 adapter->eims_enable_mask = (1 << (vector)) - 1;
1018 adapter->eims_other = 1 << (vector - 1);
1022 static void igbvf_reset_interrupt_capability(struct igbvf_adapter *adapter)
1024 if (adapter->msix_entries) {
1025 pci_disable_msix(adapter->pdev);
1026 kfree(adapter->msix_entries);
1027 adapter->msix_entries = NULL;
1032 * igbvf_set_interrupt_capability - set MSI or MSI-X if supported
1033 * @adapter: board private structure
1035 * Attempt to configure interrupts using the best available
1036 * capabilities of the hardware and kernel.
1038 static void igbvf_set_interrupt_capability(struct igbvf_adapter *adapter)
1043 /* we allocate 3 vectors, 1 for Tx, 1 for Rx, one for PF messages */
1044 adapter->msix_entries = kcalloc(3, sizeof(struct msix_entry),
1046 if (adapter->msix_entries) {
1047 for (i = 0; i < 3; i++)
1048 adapter->msix_entries[i].entry = i;
1050 err = pci_enable_msix_range(adapter->pdev,
1051 adapter->msix_entries, 3, 3);
1056 dev_err(&adapter->pdev->dev,
1057 "Failed to initialize MSI-X interrupts.\n");
1058 igbvf_reset_interrupt_capability(adapter);
1063 * igbvf_request_msix - Initialize MSI-X interrupts
1064 * @adapter: board private structure
1066 * igbvf_request_msix allocates MSI-X vectors and requests interrupts from the
1069 static int igbvf_request_msix(struct igbvf_adapter *adapter)
1071 struct net_device *netdev = adapter->netdev;
1072 int err = 0, vector = 0;
1074 if (strlen(netdev->name) < (IFNAMSIZ - 5)) {
1075 sprintf(adapter->tx_ring->name, "%s-tx-0", netdev->name);
1076 sprintf(adapter->rx_ring->name, "%s-rx-0", netdev->name);
1078 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1079 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1082 err = request_irq(adapter->msix_entries[vector].vector,
1083 igbvf_intr_msix_tx, 0, adapter->tx_ring->name,
1088 adapter->tx_ring->itr_register = E1000_EITR(vector);
1089 adapter->tx_ring->itr_val = adapter->current_itr;
1092 err = request_irq(adapter->msix_entries[vector].vector,
1093 igbvf_intr_msix_rx, 0, adapter->rx_ring->name,
1098 adapter->rx_ring->itr_register = E1000_EITR(vector);
1099 adapter->rx_ring->itr_val = adapter->current_itr;
1102 err = request_irq(adapter->msix_entries[vector].vector,
1103 igbvf_msix_other, 0, netdev->name, netdev);
1107 igbvf_configure_msix(adapter);
1114 * igbvf_alloc_queues - Allocate memory for all rings
1115 * @adapter: board private structure to initialize
1117 static int igbvf_alloc_queues(struct igbvf_adapter *adapter)
1119 struct net_device *netdev = adapter->netdev;
1121 adapter->tx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL);
1122 if (!adapter->tx_ring)
1125 adapter->rx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL);
1126 if (!adapter->rx_ring) {
1127 kfree(adapter->tx_ring);
1131 netif_napi_add(netdev, &adapter->rx_ring->napi, igbvf_poll, 64);
1137 * igbvf_request_irq - initialize interrupts
1138 * @adapter: board private structure
1140 * Attempts to configure interrupts using the best available
1141 * capabilities of the hardware and kernel.
1143 static int igbvf_request_irq(struct igbvf_adapter *adapter)
1147 /* igbvf supports msi-x only */
1148 if (adapter->msix_entries)
1149 err = igbvf_request_msix(adapter);
1154 dev_err(&adapter->pdev->dev,
1155 "Unable to allocate interrupt, Error: %d\n", err);
1160 static void igbvf_free_irq(struct igbvf_adapter *adapter)
1162 struct net_device *netdev = adapter->netdev;
1165 if (adapter->msix_entries) {
1166 for (vector = 0; vector < 3; vector++)
1167 free_irq(adapter->msix_entries[vector].vector, netdev);
1172 * igbvf_irq_disable - Mask off interrupt generation on the NIC
1173 * @adapter: board private structure
1175 static void igbvf_irq_disable(struct igbvf_adapter *adapter)
1177 struct e1000_hw *hw = &adapter->hw;
1181 if (adapter->msix_entries)
1186 * igbvf_irq_enable - Enable default interrupt generation settings
1187 * @adapter: board private structure
1189 static void igbvf_irq_enable(struct igbvf_adapter *adapter)
1191 struct e1000_hw *hw = &adapter->hw;
1193 ew32(EIAC, adapter->eims_enable_mask);
1194 ew32(EIAM, adapter->eims_enable_mask);
1195 ew32(EIMS, adapter->eims_enable_mask);
1199 * igbvf_poll - NAPI Rx polling callback
1200 * @napi: struct associated with this polling callback
1201 * @budget: amount of packets driver is allowed to process this poll
1203 static int igbvf_poll(struct napi_struct *napi, int budget)
1205 struct igbvf_ring *rx_ring = container_of(napi, struct igbvf_ring, napi);
1206 struct igbvf_adapter *adapter = rx_ring->adapter;
1207 struct e1000_hw *hw = &adapter->hw;
1210 igbvf_clean_rx_irq(adapter, &work_done, budget);
1212 /* If not enough Rx work done, exit the polling mode */
1213 if (work_done < budget) {
1214 napi_complete_done(napi, work_done);
1216 if (adapter->requested_itr & 3)
1217 igbvf_set_itr(adapter);
1219 if (!test_bit(__IGBVF_DOWN, &adapter->state))
1220 ew32(EIMS, adapter->rx_ring->eims_value);
1227 * igbvf_set_rlpml - set receive large packet maximum length
1228 * @adapter: board private structure
1230 * Configure the maximum size of packets that will be received
1232 static void igbvf_set_rlpml(struct igbvf_adapter *adapter)
1235 struct e1000_hw *hw = &adapter->hw;
1237 max_frame_size = adapter->max_frame_size + VLAN_TAG_SIZE;
1238 e1000_rlpml_set_vf(hw, max_frame_size);
1241 static int igbvf_vlan_rx_add_vid(struct net_device *netdev,
1242 __be16 proto, u16 vid)
1244 struct igbvf_adapter *adapter = netdev_priv(netdev);
1245 struct e1000_hw *hw = &adapter->hw;
1247 if (hw->mac.ops.set_vfta(hw, vid, true)) {
1248 dev_err(&adapter->pdev->dev, "Failed to add vlan id %d\n", vid);
1251 set_bit(vid, adapter->active_vlans);
1255 static int igbvf_vlan_rx_kill_vid(struct net_device *netdev,
1256 __be16 proto, u16 vid)
1258 struct igbvf_adapter *adapter = netdev_priv(netdev);
1259 struct e1000_hw *hw = &adapter->hw;
1261 if (hw->mac.ops.set_vfta(hw, vid, false)) {
1262 dev_err(&adapter->pdev->dev,
1263 "Failed to remove vlan id %d\n", vid);
1266 clear_bit(vid, adapter->active_vlans);
1270 static void igbvf_restore_vlan(struct igbvf_adapter *adapter)
1274 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
1275 igbvf_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
1279 * igbvf_configure_tx - Configure Transmit Unit after Reset
1280 * @adapter: board private structure
1282 * Configure the Tx unit of the MAC after a reset.
1284 static void igbvf_configure_tx(struct igbvf_adapter *adapter)
1286 struct e1000_hw *hw = &adapter->hw;
1287 struct igbvf_ring *tx_ring = adapter->tx_ring;
1289 u32 txdctl, dca_txctrl;
1291 /* disable transmits */
1292 txdctl = er32(TXDCTL(0));
1293 ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1297 /* Setup the HW Tx Head and Tail descriptor pointers */
1298 ew32(TDLEN(0), tx_ring->count * sizeof(union e1000_adv_tx_desc));
1299 tdba = tx_ring->dma;
1300 ew32(TDBAL(0), (tdba & DMA_BIT_MASK(32)));
1301 ew32(TDBAH(0), (tdba >> 32));
1304 tx_ring->head = E1000_TDH(0);
1305 tx_ring->tail = E1000_TDT(0);
1307 /* Turn off Relaxed Ordering on head write-backs. The writebacks
1308 * MUST be delivered in order or it will completely screw up
1311 dca_txctrl = er32(DCA_TXCTRL(0));
1312 dca_txctrl &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN;
1313 ew32(DCA_TXCTRL(0), dca_txctrl);
1315 /* enable transmits */
1316 txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
1317 ew32(TXDCTL(0), txdctl);
1319 /* Setup Transmit Descriptor Settings for eop descriptor */
1320 adapter->txd_cmd = E1000_ADVTXD_DCMD_EOP | E1000_ADVTXD_DCMD_IFCS;
1322 /* enable Report Status bit */
1323 adapter->txd_cmd |= E1000_ADVTXD_DCMD_RS;
1327 * igbvf_setup_srrctl - configure the receive control registers
1328 * @adapter: Board private structure
1330 static void igbvf_setup_srrctl(struct igbvf_adapter *adapter)
1332 struct e1000_hw *hw = &adapter->hw;
1335 srrctl &= ~(E1000_SRRCTL_DESCTYPE_MASK |
1336 E1000_SRRCTL_BSIZEHDR_MASK |
1337 E1000_SRRCTL_BSIZEPKT_MASK);
1339 /* Enable queue drop to avoid head of line blocking */
1340 srrctl |= E1000_SRRCTL_DROP_EN;
1342 /* Setup buffer sizes */
1343 srrctl |= ALIGN(adapter->rx_buffer_len, 1024) >>
1344 E1000_SRRCTL_BSIZEPKT_SHIFT;
1346 if (adapter->rx_buffer_len < 2048) {
1347 adapter->rx_ps_hdr_size = 0;
1348 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
1350 adapter->rx_ps_hdr_size = 128;
1351 srrctl |= adapter->rx_ps_hdr_size <<
1352 E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
1353 srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
1356 ew32(SRRCTL(0), srrctl);
1360 * igbvf_configure_rx - Configure Receive Unit after Reset
1361 * @adapter: board private structure
1363 * Configure the Rx unit of the MAC after a reset.
1365 static void igbvf_configure_rx(struct igbvf_adapter *adapter)
1367 struct e1000_hw *hw = &adapter->hw;
1368 struct igbvf_ring *rx_ring = adapter->rx_ring;
1372 /* disable receives */
1373 rxdctl = er32(RXDCTL(0));
1374 ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1378 rdlen = rx_ring->count * sizeof(union e1000_adv_rx_desc);
1380 /* Setup the HW Rx Head and Tail Descriptor Pointers and
1381 * the Base and Length of the Rx Descriptor Ring
1383 rdba = rx_ring->dma;
1384 ew32(RDBAL(0), (rdba & DMA_BIT_MASK(32)));
1385 ew32(RDBAH(0), (rdba >> 32));
1386 ew32(RDLEN(0), rx_ring->count * sizeof(union e1000_adv_rx_desc));
1387 rx_ring->head = E1000_RDH(0);
1388 rx_ring->tail = E1000_RDT(0);
1392 rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
1393 rxdctl &= 0xFFF00000;
1394 rxdctl |= IGBVF_RX_PTHRESH;
1395 rxdctl |= IGBVF_RX_HTHRESH << 8;
1396 rxdctl |= IGBVF_RX_WTHRESH << 16;
1398 igbvf_set_rlpml(adapter);
1400 /* enable receives */
1401 ew32(RXDCTL(0), rxdctl);
1405 * igbvf_set_multi - Multicast and Promiscuous mode set
1406 * @netdev: network interface device structure
1408 * The set_multi entry point is called whenever the multicast address
1409 * list or the network interface flags are updated. This routine is
1410 * responsible for configuring the hardware for proper multicast,
1411 * promiscuous mode, and all-multi behavior.
1413 static void igbvf_set_multi(struct net_device *netdev)
1415 struct igbvf_adapter *adapter = netdev_priv(netdev);
1416 struct e1000_hw *hw = &adapter->hw;
1417 struct netdev_hw_addr *ha;
1418 u8 *mta_list = NULL;
1421 if (!netdev_mc_empty(netdev)) {
1422 mta_list = kmalloc_array(netdev_mc_count(netdev), ETH_ALEN,
1428 /* prepare a packed array of only addresses. */
1430 netdev_for_each_mc_addr(ha, netdev)
1431 memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
1433 hw->mac.ops.update_mc_addr_list(hw, mta_list, i, 0, 0);
1438 * igbvf_configure - configure the hardware for Rx and Tx
1439 * @adapter: private board structure
1441 static void igbvf_configure(struct igbvf_adapter *adapter)
1443 igbvf_set_multi(adapter->netdev);
1445 igbvf_restore_vlan(adapter);
1447 igbvf_configure_tx(adapter);
1448 igbvf_setup_srrctl(adapter);
1449 igbvf_configure_rx(adapter);
1450 igbvf_alloc_rx_buffers(adapter->rx_ring,
1451 igbvf_desc_unused(adapter->rx_ring));
1454 /* igbvf_reset - bring the hardware into a known good state
1455 * @adapter: private board structure
1457 * This function boots the hardware and enables some settings that
1458 * require a configuration cycle of the hardware - those cannot be
1459 * set/changed during runtime. After reset the device needs to be
1460 * properly configured for Rx, Tx etc.
1462 static void igbvf_reset(struct igbvf_adapter *adapter)
1464 struct e1000_mac_info *mac = &adapter->hw.mac;
1465 struct net_device *netdev = adapter->netdev;
1466 struct e1000_hw *hw = &adapter->hw;
1468 /* Allow time for pending master requests to run */
1469 if (mac->ops.reset_hw(hw))
1470 dev_err(&adapter->pdev->dev, "PF still resetting\n");
1472 mac->ops.init_hw(hw);
1474 if (is_valid_ether_addr(adapter->hw.mac.addr)) {
1475 memcpy(netdev->dev_addr, adapter->hw.mac.addr,
1477 memcpy(netdev->perm_addr, adapter->hw.mac.addr,
1481 adapter->last_reset = jiffies;
1484 int igbvf_up(struct igbvf_adapter *adapter)
1486 struct e1000_hw *hw = &adapter->hw;
1488 /* hardware has been reset, we need to reload some things */
1489 igbvf_configure(adapter);
1491 clear_bit(__IGBVF_DOWN, &adapter->state);
1493 napi_enable(&adapter->rx_ring->napi);
1494 if (adapter->msix_entries)
1495 igbvf_configure_msix(adapter);
1497 /* Clear any pending interrupts. */
1499 igbvf_irq_enable(adapter);
1501 /* start the watchdog */
1502 hw->mac.get_link_status = 1;
1503 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1508 void igbvf_down(struct igbvf_adapter *adapter)
1510 struct net_device *netdev = adapter->netdev;
1511 struct e1000_hw *hw = &adapter->hw;
1514 /* signal that we're down so the interrupt handler does not
1515 * reschedule our watchdog timer
1517 set_bit(__IGBVF_DOWN, &adapter->state);
1519 /* disable receives in the hardware */
1520 rxdctl = er32(RXDCTL(0));
1521 ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1523 netif_carrier_off(netdev);
1524 netif_stop_queue(netdev);
1526 /* disable transmits in the hardware */
1527 txdctl = er32(TXDCTL(0));
1528 ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1530 /* flush both disables and wait for them to finish */
1534 napi_disable(&adapter->rx_ring->napi);
1536 igbvf_irq_disable(adapter);
1538 del_timer_sync(&adapter->watchdog_timer);
1540 /* record the stats before reset*/
1541 igbvf_update_stats(adapter);
1543 adapter->link_speed = 0;
1544 adapter->link_duplex = 0;
1546 igbvf_reset(adapter);
1547 igbvf_clean_tx_ring(adapter->tx_ring);
1548 igbvf_clean_rx_ring(adapter->rx_ring);
1551 void igbvf_reinit_locked(struct igbvf_adapter *adapter)
1554 while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
1555 usleep_range(1000, 2000);
1556 igbvf_down(adapter);
1558 clear_bit(__IGBVF_RESETTING, &adapter->state);
1562 * igbvf_sw_init - Initialize general software structures (struct igbvf_adapter)
1563 * @adapter: board private structure to initialize
1565 * igbvf_sw_init initializes the Adapter private data structure.
1566 * Fields are initialized based on PCI device information and
1567 * OS network device settings (MTU size).
1569 static int igbvf_sw_init(struct igbvf_adapter *adapter)
1571 struct net_device *netdev = adapter->netdev;
1574 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
1575 adapter->rx_ps_hdr_size = 0;
1576 adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1577 adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
1579 adapter->tx_int_delay = 8;
1580 adapter->tx_abs_int_delay = 32;
1581 adapter->rx_int_delay = 0;
1582 adapter->rx_abs_int_delay = 8;
1583 adapter->requested_itr = 3;
1584 adapter->current_itr = IGBVF_START_ITR;
1586 /* Set various function pointers */
1587 adapter->ei->init_ops(&adapter->hw);
1589 rc = adapter->hw.mac.ops.init_params(&adapter->hw);
1593 rc = adapter->hw.mbx.ops.init_params(&adapter->hw);
1597 igbvf_set_interrupt_capability(adapter);
1599 if (igbvf_alloc_queues(adapter))
1602 spin_lock_init(&adapter->tx_queue_lock);
1604 /* Explicitly disable IRQ since the NIC can be in any state. */
1605 igbvf_irq_disable(adapter);
1607 spin_lock_init(&adapter->stats_lock);
1609 set_bit(__IGBVF_DOWN, &adapter->state);
1613 static void igbvf_initialize_last_counter_stats(struct igbvf_adapter *adapter)
1615 struct e1000_hw *hw = &adapter->hw;
1617 adapter->stats.last_gprc = er32(VFGPRC);
1618 adapter->stats.last_gorc = er32(VFGORC);
1619 adapter->stats.last_gptc = er32(VFGPTC);
1620 adapter->stats.last_gotc = er32(VFGOTC);
1621 adapter->stats.last_mprc = er32(VFMPRC);
1622 adapter->stats.last_gotlbc = er32(VFGOTLBC);
1623 adapter->stats.last_gptlbc = er32(VFGPTLBC);
1624 adapter->stats.last_gorlbc = er32(VFGORLBC);
1625 adapter->stats.last_gprlbc = er32(VFGPRLBC);
1627 adapter->stats.base_gprc = er32(VFGPRC);
1628 adapter->stats.base_gorc = er32(VFGORC);
1629 adapter->stats.base_gptc = er32(VFGPTC);
1630 adapter->stats.base_gotc = er32(VFGOTC);
1631 adapter->stats.base_mprc = er32(VFMPRC);
1632 adapter->stats.base_gotlbc = er32(VFGOTLBC);
1633 adapter->stats.base_gptlbc = er32(VFGPTLBC);
1634 adapter->stats.base_gorlbc = er32(VFGORLBC);
1635 adapter->stats.base_gprlbc = er32(VFGPRLBC);
1639 * igbvf_open - Called when a network interface is made active
1640 * @netdev: network interface device structure
1642 * Returns 0 on success, negative value on failure
1644 * The open entry point is called when a network interface is made
1645 * active by the system (IFF_UP). At this point all resources needed
1646 * for transmit and receive operations are allocated, the interrupt
1647 * handler is registered with the OS, the watchdog timer is started,
1648 * and the stack is notified that the interface is ready.
1650 static int igbvf_open(struct net_device *netdev)
1652 struct igbvf_adapter *adapter = netdev_priv(netdev);
1653 struct e1000_hw *hw = &adapter->hw;
1656 /* disallow open during test */
1657 if (test_bit(__IGBVF_TESTING, &adapter->state))
1660 /* allocate transmit descriptors */
1661 err = igbvf_setup_tx_resources(adapter, adapter->tx_ring);
1665 /* allocate receive descriptors */
1666 err = igbvf_setup_rx_resources(adapter, adapter->rx_ring);
1670 /* before we allocate an interrupt, we must be ready to handle it.
1671 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1672 * as soon as we call pci_request_irq, so we have to setup our
1673 * clean_rx handler before we do so.
1675 igbvf_configure(adapter);
1677 err = igbvf_request_irq(adapter);
1681 /* From here on the code is the same as igbvf_up() */
1682 clear_bit(__IGBVF_DOWN, &adapter->state);
1684 napi_enable(&adapter->rx_ring->napi);
1686 /* clear any pending interrupts */
1689 igbvf_irq_enable(adapter);
1691 /* start the watchdog */
1692 hw->mac.get_link_status = 1;
1693 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1698 igbvf_free_rx_resources(adapter->rx_ring);
1700 igbvf_free_tx_resources(adapter->tx_ring);
1702 igbvf_reset(adapter);
1708 * igbvf_close - Disables a network interface
1709 * @netdev: network interface device structure
1711 * Returns 0, this is not allowed to fail
1713 * The close entry point is called when an interface is de-activated
1714 * by the OS. The hardware is still under the drivers control, but
1715 * needs to be disabled. A global MAC reset is issued to stop the
1716 * hardware, and all transmit and receive resources are freed.
1718 static int igbvf_close(struct net_device *netdev)
1720 struct igbvf_adapter *adapter = netdev_priv(netdev);
1722 WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
1723 igbvf_down(adapter);
1725 igbvf_free_irq(adapter);
1727 igbvf_free_tx_resources(adapter->tx_ring);
1728 igbvf_free_rx_resources(adapter->rx_ring);
1734 * igbvf_set_mac - Change the Ethernet Address of the NIC
1735 * @netdev: network interface device structure
1736 * @p: pointer to an address structure
1738 * Returns 0 on success, negative on failure
1740 static int igbvf_set_mac(struct net_device *netdev, void *p)
1742 struct igbvf_adapter *adapter = netdev_priv(netdev);
1743 struct e1000_hw *hw = &adapter->hw;
1744 struct sockaddr *addr = p;
1746 if (!is_valid_ether_addr(addr->sa_data))
1747 return -EADDRNOTAVAIL;
1749 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
1751 hw->mac.ops.rar_set(hw, hw->mac.addr, 0);
1753 if (!ether_addr_equal(addr->sa_data, hw->mac.addr))
1754 return -EADDRNOTAVAIL;
1756 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
1761 #define UPDATE_VF_COUNTER(reg, name) \
1763 u32 current_counter = er32(reg); \
1764 if (current_counter < adapter->stats.last_##name) \
1765 adapter->stats.name += 0x100000000LL; \
1766 adapter->stats.last_##name = current_counter; \
1767 adapter->stats.name &= 0xFFFFFFFF00000000LL; \
1768 adapter->stats.name |= current_counter; \
1772 * igbvf_update_stats - Update the board statistics counters
1773 * @adapter: board private structure
1775 void igbvf_update_stats(struct igbvf_adapter *adapter)
1777 struct e1000_hw *hw = &adapter->hw;
1778 struct pci_dev *pdev = adapter->pdev;
1780 /* Prevent stats update while adapter is being reset, link is down
1781 * or if the pci connection is down.
1783 if (adapter->link_speed == 0)
1786 if (test_bit(__IGBVF_RESETTING, &adapter->state))
1789 if (pci_channel_offline(pdev))
1792 UPDATE_VF_COUNTER(VFGPRC, gprc);
1793 UPDATE_VF_COUNTER(VFGORC, gorc);
1794 UPDATE_VF_COUNTER(VFGPTC, gptc);
1795 UPDATE_VF_COUNTER(VFGOTC, gotc);
1796 UPDATE_VF_COUNTER(VFMPRC, mprc);
1797 UPDATE_VF_COUNTER(VFGOTLBC, gotlbc);
1798 UPDATE_VF_COUNTER(VFGPTLBC, gptlbc);
1799 UPDATE_VF_COUNTER(VFGORLBC, gorlbc);
1800 UPDATE_VF_COUNTER(VFGPRLBC, gprlbc);
1802 /* Fill out the OS statistics structure */
1803 adapter->net_stats.multicast = adapter->stats.mprc;
1806 static void igbvf_print_link_info(struct igbvf_adapter *adapter)
1808 dev_info(&adapter->pdev->dev, "Link is Up %d Mbps %s Duplex\n",
1809 adapter->link_speed,
1810 adapter->link_duplex == FULL_DUPLEX ? "Full" : "Half");
1813 static bool igbvf_has_link(struct igbvf_adapter *adapter)
1815 struct e1000_hw *hw = &adapter->hw;
1816 s32 ret_val = E1000_SUCCESS;
1819 /* If interface is down, stay link down */
1820 if (test_bit(__IGBVF_DOWN, &adapter->state))
1823 ret_val = hw->mac.ops.check_for_link(hw);
1824 link_active = !hw->mac.get_link_status;
1826 /* if check for link returns error we will need to reset */
1827 if (ret_val && time_after(jiffies, adapter->last_reset + (10 * HZ)))
1828 schedule_work(&adapter->reset_task);
1834 * igbvf_watchdog - Timer Call-back
1835 * @data: pointer to adapter cast into an unsigned long
1837 static void igbvf_watchdog(unsigned long data)
1839 struct igbvf_adapter *adapter = (struct igbvf_adapter *)data;
1841 /* Do the rest outside of interrupt context */
1842 schedule_work(&adapter->watchdog_task);
1845 static void igbvf_watchdog_task(struct work_struct *work)
1847 struct igbvf_adapter *adapter = container_of(work,
1848 struct igbvf_adapter,
1850 struct net_device *netdev = adapter->netdev;
1851 struct e1000_mac_info *mac = &adapter->hw.mac;
1852 struct igbvf_ring *tx_ring = adapter->tx_ring;
1853 struct e1000_hw *hw = &adapter->hw;
1857 link = igbvf_has_link(adapter);
1860 if (!netif_carrier_ok(netdev)) {
1861 mac->ops.get_link_up_info(&adapter->hw,
1862 &adapter->link_speed,
1863 &adapter->link_duplex);
1864 igbvf_print_link_info(adapter);
1866 netif_carrier_on(netdev);
1867 netif_wake_queue(netdev);
1870 if (netif_carrier_ok(netdev)) {
1871 adapter->link_speed = 0;
1872 adapter->link_duplex = 0;
1873 dev_info(&adapter->pdev->dev, "Link is Down\n");
1874 netif_carrier_off(netdev);
1875 netif_stop_queue(netdev);
1879 if (netif_carrier_ok(netdev)) {
1880 igbvf_update_stats(adapter);
1882 tx_pending = (igbvf_desc_unused(tx_ring) + 1 <
1885 /* We've lost link, so the controller stops DMA,
1886 * but we've got queued Tx work that's never going
1887 * to get done, so reset controller to flush Tx.
1888 * (Do the reset outside of interrupt context).
1890 adapter->tx_timeout_count++;
1891 schedule_work(&adapter->reset_task);
1895 /* Cause software interrupt to ensure Rx ring is cleaned */
1896 ew32(EICS, adapter->rx_ring->eims_value);
1898 /* Reset the timer */
1899 if (!test_bit(__IGBVF_DOWN, &adapter->state))
1900 mod_timer(&adapter->watchdog_timer,
1901 round_jiffies(jiffies + (2 * HZ)));
1904 #define IGBVF_TX_FLAGS_CSUM 0x00000001
1905 #define IGBVF_TX_FLAGS_VLAN 0x00000002
1906 #define IGBVF_TX_FLAGS_TSO 0x00000004
1907 #define IGBVF_TX_FLAGS_IPV4 0x00000008
1908 #define IGBVF_TX_FLAGS_VLAN_MASK 0xffff0000
1909 #define IGBVF_TX_FLAGS_VLAN_SHIFT 16
1911 static void igbvf_tx_ctxtdesc(struct igbvf_ring *tx_ring, u32 vlan_macip_lens,
1912 u32 type_tucmd, u32 mss_l4len_idx)
1914 struct e1000_adv_tx_context_desc *context_desc;
1915 struct igbvf_buffer *buffer_info;
1916 u16 i = tx_ring->next_to_use;
1918 context_desc = IGBVF_TX_CTXTDESC_ADV(*tx_ring, i);
1919 buffer_info = &tx_ring->buffer_info[i];
1922 tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
1924 /* set bits to identify this as an advanced context descriptor */
1925 type_tucmd |= E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT;
1927 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
1928 context_desc->seqnum_seed = 0;
1929 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd);
1930 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
1932 buffer_info->time_stamp = jiffies;
1933 buffer_info->dma = 0;
1936 static int igbvf_tso(struct igbvf_adapter *adapter,
1937 struct igbvf_ring *tx_ring,
1938 struct sk_buff *skb, u32 tx_flags, u8 *hdr_len,
1941 struct e1000_adv_tx_context_desc *context_desc;
1942 struct igbvf_buffer *buffer_info;
1943 u32 info = 0, tu_cmd = 0;
1944 u32 mss_l4len_idx, l4len;
1950 err = skb_cow_head(skb, 0);
1952 dev_err(&adapter->pdev->dev, "igbvf_tso returning an error\n");
1956 l4len = tcp_hdrlen(skb);
1959 if (protocol == htons(ETH_P_IP)) {
1960 struct iphdr *iph = ip_hdr(skb);
1964 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
1968 } else if (skb_is_gso_v6(skb)) {
1969 ipv6_hdr(skb)->payload_len = 0;
1970 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
1971 &ipv6_hdr(skb)->daddr,
1975 i = tx_ring->next_to_use;
1977 buffer_info = &tx_ring->buffer_info[i];
1978 context_desc = IGBVF_TX_CTXTDESC_ADV(*tx_ring, i);
1979 /* VLAN MACLEN IPLEN */
1980 if (tx_flags & IGBVF_TX_FLAGS_VLAN)
1981 info |= (tx_flags & IGBVF_TX_FLAGS_VLAN_MASK);
1982 info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
1983 *hdr_len += skb_network_offset(skb);
1984 info |= (skb_transport_header(skb) - skb_network_header(skb));
1985 *hdr_len += (skb_transport_header(skb) - skb_network_header(skb));
1986 context_desc->vlan_macip_lens = cpu_to_le32(info);
1988 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
1989 tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
1991 if (protocol == htons(ETH_P_IP))
1992 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
1993 tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
1995 context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
1998 mss_l4len_idx = (skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT);
1999 mss_l4len_idx |= (l4len << E1000_ADVTXD_L4LEN_SHIFT);
2001 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
2002 context_desc->seqnum_seed = 0;
2004 buffer_info->time_stamp = jiffies;
2005 buffer_info->dma = 0;
2007 if (i == tx_ring->count)
2010 tx_ring->next_to_use = i;
2015 static inline bool igbvf_ipv6_csum_is_sctp(struct sk_buff *skb)
2017 unsigned int offset = 0;
2019 ipv6_find_hdr(skb, &offset, IPPROTO_SCTP, NULL, NULL);
2021 return offset == skb_checksum_start_offset(skb);
2024 static bool igbvf_tx_csum(struct igbvf_ring *tx_ring, struct sk_buff *skb,
2025 u32 tx_flags, __be16 protocol)
2027 u32 vlan_macip_lens = 0;
2030 if (skb->ip_summed != CHECKSUM_PARTIAL) {
2032 if (!(tx_flags & IGBVF_TX_FLAGS_VLAN))
2037 switch (skb->csum_offset) {
2038 case offsetof(struct tcphdr, check):
2039 type_tucmd = E1000_ADVTXD_TUCMD_L4T_TCP;
2041 case offsetof(struct udphdr, check):
2043 case offsetof(struct sctphdr, checksum):
2044 /* validate that this is actually an SCTP request */
2045 if (((protocol == htons(ETH_P_IP)) &&
2046 (ip_hdr(skb)->protocol == IPPROTO_SCTP)) ||
2047 ((protocol == htons(ETH_P_IPV6)) &&
2048 igbvf_ipv6_csum_is_sctp(skb))) {
2049 type_tucmd = E1000_ADVTXD_TUCMD_L4T_SCTP;
2053 skb_checksum_help(skb);
2057 vlan_macip_lens = skb_checksum_start_offset(skb) -
2058 skb_network_offset(skb);
2060 vlan_macip_lens |= skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT;
2061 vlan_macip_lens |= tx_flags & IGBVF_TX_FLAGS_VLAN_MASK;
2063 igbvf_tx_ctxtdesc(tx_ring, vlan_macip_lens, type_tucmd, 0);
2067 static int igbvf_maybe_stop_tx(struct net_device *netdev, int size)
2069 struct igbvf_adapter *adapter = netdev_priv(netdev);
2071 /* there is enough descriptors then we don't need to worry */
2072 if (igbvf_desc_unused(adapter->tx_ring) >= size)
2075 netif_stop_queue(netdev);
2077 /* Herbert's original patch had:
2078 * smp_mb__after_netif_stop_queue();
2079 * but since that doesn't exist yet, just open code it.
2083 /* We need to check again just in case room has been made available */
2084 if (igbvf_desc_unused(adapter->tx_ring) < size)
2087 netif_wake_queue(netdev);
2089 ++adapter->restart_queue;
2093 #define IGBVF_MAX_TXD_PWR 16
2094 #define IGBVF_MAX_DATA_PER_TXD (1 << IGBVF_MAX_TXD_PWR)
2096 static inline int igbvf_tx_map_adv(struct igbvf_adapter *adapter,
2097 struct igbvf_ring *tx_ring,
2098 struct sk_buff *skb)
2100 struct igbvf_buffer *buffer_info;
2101 struct pci_dev *pdev = adapter->pdev;
2102 unsigned int len = skb_headlen(skb);
2103 unsigned int count = 0, i;
2106 i = tx_ring->next_to_use;
2108 buffer_info = &tx_ring->buffer_info[i];
2109 BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD);
2110 buffer_info->length = len;
2111 /* set time_stamp *before* dma to help avoid a possible race */
2112 buffer_info->time_stamp = jiffies;
2113 buffer_info->mapped_as_page = false;
2114 buffer_info->dma = dma_map_single(&pdev->dev, skb->data, len,
2116 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2119 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) {
2120 const struct skb_frag_struct *frag;
2124 if (i == tx_ring->count)
2127 frag = &skb_shinfo(skb)->frags[f];
2128 len = skb_frag_size(frag);
2130 buffer_info = &tx_ring->buffer_info[i];
2131 BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD);
2132 buffer_info->length = len;
2133 buffer_info->time_stamp = jiffies;
2134 buffer_info->mapped_as_page = true;
2135 buffer_info->dma = skb_frag_dma_map(&pdev->dev, frag, 0, len,
2137 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2141 tx_ring->buffer_info[i].skb = skb;
2146 dev_err(&pdev->dev, "TX DMA map failed\n");
2148 /* clear timestamp and dma mappings for failed buffer_info mapping */
2149 buffer_info->dma = 0;
2150 buffer_info->time_stamp = 0;
2151 buffer_info->length = 0;
2152 buffer_info->mapped_as_page = false;
2156 /* clear timestamp and dma mappings for remaining portion of packet */
2159 i += tx_ring->count;
2161 buffer_info = &tx_ring->buffer_info[i];
2162 igbvf_put_txbuf(adapter, buffer_info);
2168 static inline void igbvf_tx_queue_adv(struct igbvf_adapter *adapter,
2169 struct igbvf_ring *tx_ring,
2170 int tx_flags, int count,
2171 unsigned int first, u32 paylen,
2174 union e1000_adv_tx_desc *tx_desc = NULL;
2175 struct igbvf_buffer *buffer_info;
2176 u32 olinfo_status = 0, cmd_type_len;
2179 cmd_type_len = (E1000_ADVTXD_DTYP_DATA | E1000_ADVTXD_DCMD_IFCS |
2180 E1000_ADVTXD_DCMD_DEXT);
2182 if (tx_flags & IGBVF_TX_FLAGS_VLAN)
2183 cmd_type_len |= E1000_ADVTXD_DCMD_VLE;
2185 if (tx_flags & IGBVF_TX_FLAGS_TSO) {
2186 cmd_type_len |= E1000_ADVTXD_DCMD_TSE;
2188 /* insert tcp checksum */
2189 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2191 /* insert ip checksum */
2192 if (tx_flags & IGBVF_TX_FLAGS_IPV4)
2193 olinfo_status |= E1000_TXD_POPTS_IXSM << 8;
2195 } else if (tx_flags & IGBVF_TX_FLAGS_CSUM) {
2196 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2199 olinfo_status |= ((paylen - hdr_len) << E1000_ADVTXD_PAYLEN_SHIFT);
2201 i = tx_ring->next_to_use;
2203 buffer_info = &tx_ring->buffer_info[i];
2204 tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
2205 tx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
2206 tx_desc->read.cmd_type_len =
2207 cpu_to_le32(cmd_type_len | buffer_info->length);
2208 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
2210 if (i == tx_ring->count)
2214 tx_desc->read.cmd_type_len |= cpu_to_le32(adapter->txd_cmd);
2215 /* Force memory writes to complete before letting h/w
2216 * know there are new descriptors to fetch. (Only
2217 * applicable for weak-ordered memory model archs,
2222 tx_ring->buffer_info[first].next_to_watch = tx_desc;
2223 tx_ring->next_to_use = i;
2224 writel(i, adapter->hw.hw_addr + tx_ring->tail);
2225 /* we need this if more than one processor can write to our tail
2226 * at a time, it synchronizes IO on IA64/Altix systems
2231 static netdev_tx_t igbvf_xmit_frame_ring_adv(struct sk_buff *skb,
2232 struct net_device *netdev,
2233 struct igbvf_ring *tx_ring)
2235 struct igbvf_adapter *adapter = netdev_priv(netdev);
2236 unsigned int first, tx_flags = 0;
2240 __be16 protocol = vlan_get_protocol(skb);
2242 if (test_bit(__IGBVF_DOWN, &adapter->state)) {
2243 dev_kfree_skb_any(skb);
2244 return NETDEV_TX_OK;
2247 if (skb->len <= 0) {
2248 dev_kfree_skb_any(skb);
2249 return NETDEV_TX_OK;
2252 /* need: count + 4 desc gap to keep tail from touching
2253 * + 2 desc gap to keep tail from touching head,
2254 * + 1 desc for skb->data,
2255 * + 1 desc for context descriptor,
2256 * head, otherwise try next time
2258 if (igbvf_maybe_stop_tx(netdev, skb_shinfo(skb)->nr_frags + 4)) {
2259 /* this is a hard error */
2260 return NETDEV_TX_BUSY;
2263 if (skb_vlan_tag_present(skb)) {
2264 tx_flags |= IGBVF_TX_FLAGS_VLAN;
2265 tx_flags |= (skb_vlan_tag_get(skb) <<
2266 IGBVF_TX_FLAGS_VLAN_SHIFT);
2269 if (protocol == htons(ETH_P_IP))
2270 tx_flags |= IGBVF_TX_FLAGS_IPV4;
2272 first = tx_ring->next_to_use;
2274 tso = skb_is_gso(skb) ?
2275 igbvf_tso(adapter, tx_ring, skb, tx_flags, &hdr_len, protocol) : 0;
2276 if (unlikely(tso < 0)) {
2277 dev_kfree_skb_any(skb);
2278 return NETDEV_TX_OK;
2282 tx_flags |= IGBVF_TX_FLAGS_TSO;
2283 else if (igbvf_tx_csum(tx_ring, skb, tx_flags, protocol) &&
2284 (skb->ip_summed == CHECKSUM_PARTIAL))
2285 tx_flags |= IGBVF_TX_FLAGS_CSUM;
2287 /* count reflects descriptors mapped, if 0 then mapping error
2288 * has occurred and we need to rewind the descriptor queue
2290 count = igbvf_tx_map_adv(adapter, tx_ring, skb);
2293 igbvf_tx_queue_adv(adapter, tx_ring, tx_flags, count,
2294 first, skb->len, hdr_len);
2295 /* Make sure there is space in the ring for the next send. */
2296 igbvf_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 4);
2298 dev_kfree_skb_any(skb);
2299 tx_ring->buffer_info[first].time_stamp = 0;
2300 tx_ring->next_to_use = first;
2303 return NETDEV_TX_OK;
2306 static netdev_tx_t igbvf_xmit_frame(struct sk_buff *skb,
2307 struct net_device *netdev)
2309 struct igbvf_adapter *adapter = netdev_priv(netdev);
2310 struct igbvf_ring *tx_ring;
2312 if (test_bit(__IGBVF_DOWN, &adapter->state)) {
2313 dev_kfree_skb_any(skb);
2314 return NETDEV_TX_OK;
2317 tx_ring = &adapter->tx_ring[0];
2319 return igbvf_xmit_frame_ring_adv(skb, netdev, tx_ring);
2323 * igbvf_tx_timeout - Respond to a Tx Hang
2324 * @netdev: network interface device structure
2326 static void igbvf_tx_timeout(struct net_device *netdev)
2328 struct igbvf_adapter *adapter = netdev_priv(netdev);
2330 /* Do the reset outside of interrupt context */
2331 adapter->tx_timeout_count++;
2332 schedule_work(&adapter->reset_task);
2335 static void igbvf_reset_task(struct work_struct *work)
2337 struct igbvf_adapter *adapter;
2339 adapter = container_of(work, struct igbvf_adapter, reset_task);
2341 igbvf_reinit_locked(adapter);
2345 * igbvf_get_stats - Get System Network Statistics
2346 * @netdev: network interface device structure
2348 * Returns the address of the device statistics structure.
2349 * The statistics are actually updated from the timer callback.
2351 static struct net_device_stats *igbvf_get_stats(struct net_device *netdev)
2353 struct igbvf_adapter *adapter = netdev_priv(netdev);
2355 /* only return the current stats */
2356 return &adapter->net_stats;
2360 * igbvf_change_mtu - Change the Maximum Transfer Unit
2361 * @netdev: network interface device structure
2362 * @new_mtu: new value for maximum frame size
2364 * Returns 0 on success, negative on failure
2366 static int igbvf_change_mtu(struct net_device *netdev, int new_mtu)
2368 struct igbvf_adapter *adapter = netdev_priv(netdev);
2369 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
2371 if (new_mtu < 68 || new_mtu > INT_MAX - ETH_HLEN - ETH_FCS_LEN ||
2372 max_frame > MAX_JUMBO_FRAME_SIZE)
2375 #define MAX_STD_JUMBO_FRAME_SIZE 9234
2376 if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
2377 dev_err(&adapter->pdev->dev, "MTU > 9216 not supported.\n");
2381 while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
2382 usleep_range(1000, 2000);
2383 /* igbvf_down has a dependency on max_frame_size */
2384 adapter->max_frame_size = max_frame;
2385 if (netif_running(netdev))
2386 igbvf_down(adapter);
2388 /* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
2389 * means we reserve 2 more, this pushes us to allocate from the next
2391 * i.e. RXBUFFER_2048 --> size-4096 slab
2392 * However with the new *_jumbo_rx* routines, jumbo receives will use
2396 if (max_frame <= 1024)
2397 adapter->rx_buffer_len = 1024;
2398 else if (max_frame <= 2048)
2399 adapter->rx_buffer_len = 2048;
2401 #if (PAGE_SIZE / 2) > 16384
2402 adapter->rx_buffer_len = 16384;
2404 adapter->rx_buffer_len = PAGE_SIZE / 2;
2407 /* adjust allocation if LPE protects us, and we aren't using SBP */
2408 if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
2409 (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
2410 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN +
2413 dev_info(&adapter->pdev->dev, "changing MTU from %d to %d\n",
2414 netdev->mtu, new_mtu);
2415 netdev->mtu = new_mtu;
2417 if (netif_running(netdev))
2420 igbvf_reset(adapter);
2422 clear_bit(__IGBVF_RESETTING, &adapter->state);
2427 static int igbvf_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
2435 static int igbvf_suspend(struct pci_dev *pdev, pm_message_t state)
2437 struct net_device *netdev = pci_get_drvdata(pdev);
2438 struct igbvf_adapter *adapter = netdev_priv(netdev);
2443 netif_device_detach(netdev);
2445 if (netif_running(netdev)) {
2446 WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
2447 igbvf_down(adapter);
2448 igbvf_free_irq(adapter);
2452 retval = pci_save_state(pdev);
2457 pci_disable_device(pdev);
2463 static int igbvf_resume(struct pci_dev *pdev)
2465 struct net_device *netdev = pci_get_drvdata(pdev);
2466 struct igbvf_adapter *adapter = netdev_priv(netdev);
2469 pci_restore_state(pdev);
2470 err = pci_enable_device_mem(pdev);
2472 dev_err(&pdev->dev, "Cannot enable PCI device from suspend\n");
2476 pci_set_master(pdev);
2478 if (netif_running(netdev)) {
2479 err = igbvf_request_irq(adapter);
2484 igbvf_reset(adapter);
2486 if (netif_running(netdev))
2489 netif_device_attach(netdev);
2495 static void igbvf_shutdown(struct pci_dev *pdev)
2497 igbvf_suspend(pdev, PMSG_SUSPEND);
2500 #ifdef CONFIG_NET_POLL_CONTROLLER
2501 /* Polling 'interrupt' - used by things like netconsole to send skbs
2502 * without having to re-enable interrupts. It's not called while
2503 * the interrupt routine is executing.
2505 static void igbvf_netpoll(struct net_device *netdev)
2507 struct igbvf_adapter *adapter = netdev_priv(netdev);
2509 disable_irq(adapter->pdev->irq);
2511 igbvf_clean_tx_irq(adapter->tx_ring);
2513 enable_irq(adapter->pdev->irq);
2518 * igbvf_io_error_detected - called when PCI error is detected
2519 * @pdev: Pointer to PCI device
2520 * @state: The current pci connection state
2522 * This function is called after a PCI bus error affecting
2523 * this device has been detected.
2525 static pci_ers_result_t igbvf_io_error_detected(struct pci_dev *pdev,
2526 pci_channel_state_t state)
2528 struct net_device *netdev = pci_get_drvdata(pdev);
2529 struct igbvf_adapter *adapter = netdev_priv(netdev);
2531 netif_device_detach(netdev);
2533 if (state == pci_channel_io_perm_failure)
2534 return PCI_ERS_RESULT_DISCONNECT;
2536 if (netif_running(netdev))
2537 igbvf_down(adapter);
2538 pci_disable_device(pdev);
2540 /* Request a slot slot reset. */
2541 return PCI_ERS_RESULT_NEED_RESET;
2545 * igbvf_io_slot_reset - called after the pci bus has been reset.
2546 * @pdev: Pointer to PCI device
2548 * Restart the card from scratch, as if from a cold-boot. Implementation
2549 * resembles the first-half of the igbvf_resume routine.
2551 static pci_ers_result_t igbvf_io_slot_reset(struct pci_dev *pdev)
2553 struct net_device *netdev = pci_get_drvdata(pdev);
2554 struct igbvf_adapter *adapter = netdev_priv(netdev);
2556 if (pci_enable_device_mem(pdev)) {
2558 "Cannot re-enable PCI device after reset.\n");
2559 return PCI_ERS_RESULT_DISCONNECT;
2561 pci_set_master(pdev);
2563 igbvf_reset(adapter);
2565 return PCI_ERS_RESULT_RECOVERED;
2569 * igbvf_io_resume - called when traffic can start flowing again.
2570 * @pdev: Pointer to PCI device
2572 * This callback is called when the error recovery driver tells us that
2573 * its OK to resume normal operation. Implementation resembles the
2574 * second-half of the igbvf_resume routine.
2576 static void igbvf_io_resume(struct pci_dev *pdev)
2578 struct net_device *netdev = pci_get_drvdata(pdev);
2579 struct igbvf_adapter *adapter = netdev_priv(netdev);
2581 if (netif_running(netdev)) {
2582 if (igbvf_up(adapter)) {
2584 "can't bring device back up after reset\n");
2589 netif_device_attach(netdev);
2592 static void igbvf_print_device_info(struct igbvf_adapter *adapter)
2594 struct e1000_hw *hw = &adapter->hw;
2595 struct net_device *netdev = adapter->netdev;
2596 struct pci_dev *pdev = adapter->pdev;
2598 if (hw->mac.type == e1000_vfadapt_i350)
2599 dev_info(&pdev->dev, "Intel(R) I350 Virtual Function\n");
2601 dev_info(&pdev->dev, "Intel(R) 82576 Virtual Function\n");
2602 dev_info(&pdev->dev, "Address: %pM\n", netdev->dev_addr);
2605 static int igbvf_set_features(struct net_device *netdev,
2606 netdev_features_t features)
2608 struct igbvf_adapter *adapter = netdev_priv(netdev);
2610 if (features & NETIF_F_RXCSUM)
2611 adapter->flags &= ~IGBVF_FLAG_RX_CSUM_DISABLED;
2613 adapter->flags |= IGBVF_FLAG_RX_CSUM_DISABLED;
2618 static const struct net_device_ops igbvf_netdev_ops = {
2619 .ndo_open = igbvf_open,
2620 .ndo_stop = igbvf_close,
2621 .ndo_start_xmit = igbvf_xmit_frame,
2622 .ndo_get_stats = igbvf_get_stats,
2623 .ndo_set_rx_mode = igbvf_set_multi,
2624 .ndo_set_mac_address = igbvf_set_mac,
2625 .ndo_change_mtu = igbvf_change_mtu,
2626 .ndo_do_ioctl = igbvf_ioctl,
2627 .ndo_tx_timeout = igbvf_tx_timeout,
2628 .ndo_vlan_rx_add_vid = igbvf_vlan_rx_add_vid,
2629 .ndo_vlan_rx_kill_vid = igbvf_vlan_rx_kill_vid,
2630 #ifdef CONFIG_NET_POLL_CONTROLLER
2631 .ndo_poll_controller = igbvf_netpoll,
2633 .ndo_set_features = igbvf_set_features,
2634 .ndo_features_check = passthru_features_check,
2638 * igbvf_probe - Device Initialization Routine
2639 * @pdev: PCI device information struct
2640 * @ent: entry in igbvf_pci_tbl
2642 * Returns 0 on success, negative on failure
2644 * igbvf_probe initializes an adapter identified by a pci_dev structure.
2645 * The OS initialization, configuring of the adapter private structure,
2646 * and a hardware reset occur.
2648 static int igbvf_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
2650 struct net_device *netdev;
2651 struct igbvf_adapter *adapter;
2652 struct e1000_hw *hw;
2653 const struct igbvf_info *ei = igbvf_info_tbl[ent->driver_data];
2655 static int cards_found;
2656 int err, pci_using_dac;
2658 err = pci_enable_device_mem(pdev);
2663 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
2667 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
2670 "No usable DMA configuration, aborting\n");
2675 err = pci_request_regions(pdev, igbvf_driver_name);
2679 pci_set_master(pdev);
2682 netdev = alloc_etherdev(sizeof(struct igbvf_adapter));
2684 goto err_alloc_etherdev;
2686 SET_NETDEV_DEV(netdev, &pdev->dev);
2688 pci_set_drvdata(pdev, netdev);
2689 adapter = netdev_priv(netdev);
2691 adapter->netdev = netdev;
2692 adapter->pdev = pdev;
2694 adapter->pba = ei->pba;
2695 adapter->flags = ei->flags;
2696 adapter->hw.back = adapter;
2697 adapter->hw.mac.type = ei->mac;
2698 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
2700 /* PCI config space info */
2702 hw->vendor_id = pdev->vendor;
2703 hw->device_id = pdev->device;
2704 hw->subsystem_vendor_id = pdev->subsystem_vendor;
2705 hw->subsystem_device_id = pdev->subsystem_device;
2706 hw->revision_id = pdev->revision;
2709 adapter->hw.hw_addr = ioremap(pci_resource_start(pdev, 0),
2710 pci_resource_len(pdev, 0));
2712 if (!adapter->hw.hw_addr)
2715 if (ei->get_variants) {
2716 err = ei->get_variants(adapter);
2718 goto err_get_variants;
2721 /* setup adapter struct */
2722 err = igbvf_sw_init(adapter);
2726 /* construct the net_device struct */
2727 netdev->netdev_ops = &igbvf_netdev_ops;
2729 igbvf_set_ethtool_ops(netdev);
2730 netdev->watchdog_timeo = 5 * HZ;
2731 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
2733 adapter->bd_number = cards_found++;
2735 netdev->hw_features = NETIF_F_SG |
2742 netdev->features = netdev->hw_features |
2743 NETIF_F_HW_VLAN_CTAG_TX |
2744 NETIF_F_HW_VLAN_CTAG_RX |
2745 NETIF_F_HW_VLAN_CTAG_FILTER;
2748 netdev->features |= NETIF_F_HIGHDMA;
2750 netdev->vlan_features |= NETIF_F_SG |
2756 netdev->mpls_features |= NETIF_F_HW_CSUM;
2757 netdev->hw_enc_features |= NETIF_F_HW_CSUM;
2759 /*reset the controller to put the device in a known good state */
2760 err = hw->mac.ops.reset_hw(hw);
2762 dev_info(&pdev->dev,
2763 "PF still in reset state. Is the PF interface up?\n");
2765 err = hw->mac.ops.read_mac_addr(hw);
2767 dev_info(&pdev->dev, "Error reading MAC address.\n");
2768 else if (is_zero_ether_addr(adapter->hw.mac.addr))
2769 dev_info(&pdev->dev,
2770 "MAC address not assigned by administrator.\n");
2771 memcpy(netdev->dev_addr, adapter->hw.mac.addr,
2775 if (!is_valid_ether_addr(netdev->dev_addr)) {
2776 dev_info(&pdev->dev, "Assigning random MAC address.\n");
2777 eth_hw_addr_random(netdev);
2778 memcpy(adapter->hw.mac.addr, netdev->dev_addr,
2782 setup_timer(&adapter->watchdog_timer, &igbvf_watchdog,
2783 (unsigned long)adapter);
2785 INIT_WORK(&adapter->reset_task, igbvf_reset_task);
2786 INIT_WORK(&adapter->watchdog_task, igbvf_watchdog_task);
2788 /* ring size defaults */
2789 adapter->rx_ring->count = 1024;
2790 adapter->tx_ring->count = 1024;
2792 /* reset the hardware with the new settings */
2793 igbvf_reset(adapter);
2795 /* set hardware-specific flags */
2796 if (adapter->hw.mac.type == e1000_vfadapt_i350)
2797 adapter->flags |= IGBVF_FLAG_RX_LB_VLAN_BSWAP;
2799 strcpy(netdev->name, "eth%d");
2800 err = register_netdev(netdev);
2804 /* tell the stack to leave us alone until igbvf_open() is called */
2805 netif_carrier_off(netdev);
2806 netif_stop_queue(netdev);
2808 igbvf_print_device_info(adapter);
2810 igbvf_initialize_last_counter_stats(adapter);
2815 kfree(adapter->tx_ring);
2816 kfree(adapter->rx_ring);
2818 igbvf_reset_interrupt_capability(adapter);
2820 iounmap(adapter->hw.hw_addr);
2822 free_netdev(netdev);
2824 pci_release_regions(pdev);
2827 pci_disable_device(pdev);
2832 * igbvf_remove - Device Removal Routine
2833 * @pdev: PCI device information struct
2835 * igbvf_remove is called by the PCI subsystem to alert the driver
2836 * that it should release a PCI device. The could be caused by a
2837 * Hot-Plug event, or because the driver is going to be removed from
2840 static void igbvf_remove(struct pci_dev *pdev)
2842 struct net_device *netdev = pci_get_drvdata(pdev);
2843 struct igbvf_adapter *adapter = netdev_priv(netdev);
2844 struct e1000_hw *hw = &adapter->hw;
2846 /* The watchdog timer may be rescheduled, so explicitly
2847 * disable it from being rescheduled.
2849 set_bit(__IGBVF_DOWN, &adapter->state);
2850 del_timer_sync(&adapter->watchdog_timer);
2852 cancel_work_sync(&adapter->reset_task);
2853 cancel_work_sync(&adapter->watchdog_task);
2855 unregister_netdev(netdev);
2857 igbvf_reset_interrupt_capability(adapter);
2859 /* it is important to delete the NAPI struct prior to freeing the
2860 * Rx ring so that you do not end up with null pointer refs
2862 netif_napi_del(&adapter->rx_ring->napi);
2863 kfree(adapter->tx_ring);
2864 kfree(adapter->rx_ring);
2866 iounmap(hw->hw_addr);
2867 if (hw->flash_address)
2868 iounmap(hw->flash_address);
2869 pci_release_regions(pdev);
2871 free_netdev(netdev);
2873 pci_disable_device(pdev);
2876 /* PCI Error Recovery (ERS) */
2877 static const struct pci_error_handlers igbvf_err_handler = {
2878 .error_detected = igbvf_io_error_detected,
2879 .slot_reset = igbvf_io_slot_reset,
2880 .resume = igbvf_io_resume,
2883 static const struct pci_device_id igbvf_pci_tbl[] = {
2884 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_VF), board_vf },
2885 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_VF), board_i350_vf },
2886 { } /* terminate list */
2888 MODULE_DEVICE_TABLE(pci, igbvf_pci_tbl);
2890 /* PCI Device API Driver */
2891 static struct pci_driver igbvf_driver = {
2892 .name = igbvf_driver_name,
2893 .id_table = igbvf_pci_tbl,
2894 .probe = igbvf_probe,
2895 .remove = igbvf_remove,
2897 /* Power Management Hooks */
2898 .suspend = igbvf_suspend,
2899 .resume = igbvf_resume,
2901 .shutdown = igbvf_shutdown,
2902 .err_handler = &igbvf_err_handler
2906 * igbvf_init_module - Driver Registration Routine
2908 * igbvf_init_module is the first routine called when the driver is
2909 * loaded. All it does is register with the PCI subsystem.
2911 static int __init igbvf_init_module(void)
2915 pr_info("%s - version %s\n", igbvf_driver_string, igbvf_driver_version);
2916 pr_info("%s\n", igbvf_copyright);
2918 ret = pci_register_driver(&igbvf_driver);
2922 module_init(igbvf_init_module);
2925 * igbvf_exit_module - Driver Exit Cleanup Routine
2927 * igbvf_exit_module is called just before the driver is removed
2930 static void __exit igbvf_exit_module(void)
2932 pci_unregister_driver(&igbvf_driver);
2934 module_exit(igbvf_exit_module);
2936 MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
2937 MODULE_DESCRIPTION("Intel(R) Gigabit Virtual Function Network Driver");
2938 MODULE_LICENSE("GPL");
2939 MODULE_VERSION(DRV_VERSION);