1 /*******************************************************************************
3 Intel(R) 82576 Virtual Function Linux driver
4 Copyright(c) 2009 - 2010 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, write to the Free Software Foundation, Inc.,
17 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
19 The full GNU General Public License is included in this distribution in
20 the file called "COPYING".
23 e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
24 Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
26 *******************************************************************************/
28 #include <linux/module.h>
29 #include <linux/types.h>
30 #include <linux/init.h>
31 #include <linux/pci.h>
32 #include <linux/vmalloc.h>
33 #include <linux/pagemap.h>
34 #include <linux/delay.h>
35 #include <linux/netdevice.h>
36 #include <linux/tcp.h>
37 #include <linux/ipv6.h>
38 #include <linux/slab.h>
39 #include <net/checksum.h>
40 #include <net/ip6_checksum.h>
41 #include <linux/mii.h>
42 #include <linux/ethtool.h>
43 #include <linux/if_vlan.h>
47 #define DRV_VERSION "1.0.8-k0"
48 char igbvf_driver_name[] = "igbvf";
49 const char igbvf_driver_version[] = DRV_VERSION;
50 static const char igbvf_driver_string[] =
51 "Intel(R) Virtual Function Network Driver";
52 static const char igbvf_copyright[] =
53 "Copyright (c) 2009 - 2010 Intel Corporation.";
55 static int igbvf_poll(struct napi_struct *napi, int budget);
56 static void igbvf_reset(struct igbvf_adapter *);
57 static void igbvf_set_interrupt_capability(struct igbvf_adapter *);
58 static void igbvf_reset_interrupt_capability(struct igbvf_adapter *);
60 static struct igbvf_info igbvf_vf_info = {
64 .init_ops = e1000_init_function_pointers_vf,
67 static struct igbvf_info igbvf_i350_vf_info = {
68 .mac = e1000_vfadapt_i350,
71 .init_ops = e1000_init_function_pointers_vf,
74 static const struct igbvf_info *igbvf_info_tbl[] = {
75 [board_vf] = &igbvf_vf_info,
76 [board_i350_vf] = &igbvf_i350_vf_info,
80 * igbvf_desc_unused - calculate if we have unused descriptors
82 static int igbvf_desc_unused(struct igbvf_ring *ring)
84 if (ring->next_to_clean > ring->next_to_use)
85 return ring->next_to_clean - ring->next_to_use - 1;
87 return ring->count + ring->next_to_clean - ring->next_to_use - 1;
91 * igbvf_receive_skb - helper function to handle Rx indications
92 * @adapter: board private structure
93 * @status: descriptor status field as written by hardware
94 * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
95 * @skb: pointer to sk_buff to be indicated to stack
97 static void igbvf_receive_skb(struct igbvf_adapter *adapter,
98 struct net_device *netdev,
100 u32 status, u16 vlan)
102 if (adapter->vlgrp && (status & E1000_RXD_STAT_VP))
103 vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
105 E1000_RXD_SPC_VLAN_MASK);
107 netif_receive_skb(skb);
110 static inline void igbvf_rx_checksum_adv(struct igbvf_adapter *adapter,
111 u32 status_err, struct sk_buff *skb)
113 skb_checksum_none_assert(skb);
115 /* Ignore Checksum bit is set or checksum is disabled through ethtool */
116 if ((status_err & E1000_RXD_STAT_IXSM) ||
117 (adapter->flags & IGBVF_FLAG_RX_CSUM_DISABLED))
120 /* TCP/UDP checksum error bit is set */
122 (E1000_RXDEXT_STATERR_TCPE | E1000_RXDEXT_STATERR_IPE)) {
123 /* let the stack verify checksum errors */
124 adapter->hw_csum_err++;
128 /* It must be a TCP or UDP packet with a valid checksum */
129 if (status_err & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS))
130 skb->ip_summed = CHECKSUM_UNNECESSARY;
132 adapter->hw_csum_good++;
136 * igbvf_alloc_rx_buffers - Replace used receive buffers; packet split
137 * @rx_ring: address of ring structure to repopulate
138 * @cleaned_count: number of buffers to repopulate
140 static void igbvf_alloc_rx_buffers(struct igbvf_ring *rx_ring,
143 struct igbvf_adapter *adapter = rx_ring->adapter;
144 struct net_device *netdev = adapter->netdev;
145 struct pci_dev *pdev = adapter->pdev;
146 union e1000_adv_rx_desc *rx_desc;
147 struct igbvf_buffer *buffer_info;
152 i = rx_ring->next_to_use;
153 buffer_info = &rx_ring->buffer_info[i];
155 if (adapter->rx_ps_hdr_size)
156 bufsz = adapter->rx_ps_hdr_size;
158 bufsz = adapter->rx_buffer_len;
160 while (cleaned_count--) {
161 rx_desc = IGBVF_RX_DESC_ADV(*rx_ring, i);
163 if (adapter->rx_ps_hdr_size && !buffer_info->page_dma) {
164 if (!buffer_info->page) {
165 buffer_info->page = alloc_page(GFP_ATOMIC);
166 if (!buffer_info->page) {
167 adapter->alloc_rx_buff_failed++;
170 buffer_info->page_offset = 0;
172 buffer_info->page_offset ^= PAGE_SIZE / 2;
174 buffer_info->page_dma =
175 dma_map_page(&pdev->dev, buffer_info->page,
176 buffer_info->page_offset,
181 if (!buffer_info->skb) {
182 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
184 adapter->alloc_rx_buff_failed++;
188 buffer_info->skb = skb;
189 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
193 /* Refresh the desc even if buffer_addrs didn't change because
194 * each write-back erases this info. */
195 if (adapter->rx_ps_hdr_size) {
196 rx_desc->read.pkt_addr =
197 cpu_to_le64(buffer_info->page_dma);
198 rx_desc->read.hdr_addr = cpu_to_le64(buffer_info->dma);
200 rx_desc->read.pkt_addr =
201 cpu_to_le64(buffer_info->dma);
202 rx_desc->read.hdr_addr = 0;
206 if (i == rx_ring->count)
208 buffer_info = &rx_ring->buffer_info[i];
212 if (rx_ring->next_to_use != i) {
213 rx_ring->next_to_use = i;
215 i = (rx_ring->count - 1);
219 /* Force memory writes to complete before letting h/w
220 * know there are new descriptors to fetch. (Only
221 * applicable for weak-ordered memory model archs,
224 writel(i, adapter->hw.hw_addr + rx_ring->tail);
229 * igbvf_clean_rx_irq - Send received data up the network stack; legacy
230 * @adapter: board private structure
232 * the return value indicates whether actual cleaning was done, there
233 * is no guarantee that everything was cleaned
235 static bool igbvf_clean_rx_irq(struct igbvf_adapter *adapter,
236 int *work_done, int work_to_do)
238 struct igbvf_ring *rx_ring = adapter->rx_ring;
239 struct net_device *netdev = adapter->netdev;
240 struct pci_dev *pdev = adapter->pdev;
241 union e1000_adv_rx_desc *rx_desc, *next_rxd;
242 struct igbvf_buffer *buffer_info, *next_buffer;
244 bool cleaned = false;
245 int cleaned_count = 0;
246 unsigned int total_bytes = 0, total_packets = 0;
248 u32 length, hlen, staterr;
250 i = rx_ring->next_to_clean;
251 rx_desc = IGBVF_RX_DESC_ADV(*rx_ring, i);
252 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
254 while (staterr & E1000_RXD_STAT_DD) {
255 if (*work_done >= work_to_do)
258 rmb(); /* read descriptor and rx_buffer_info after status DD */
260 buffer_info = &rx_ring->buffer_info[i];
262 /* HW will not DMA in data larger than the given buffer, even
263 * if it parses the (NFS, of course) header to be larger. In
264 * that case, it fills the header buffer and spills the rest
267 hlen = (le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.hdr_info) &
268 E1000_RXDADV_HDRBUFLEN_MASK) >> E1000_RXDADV_HDRBUFLEN_SHIFT;
269 if (hlen > adapter->rx_ps_hdr_size)
270 hlen = adapter->rx_ps_hdr_size;
272 length = le16_to_cpu(rx_desc->wb.upper.length);
276 skb = buffer_info->skb;
277 prefetch(skb->data - NET_IP_ALIGN);
278 buffer_info->skb = NULL;
279 if (!adapter->rx_ps_hdr_size) {
280 dma_unmap_single(&pdev->dev, buffer_info->dma,
281 adapter->rx_buffer_len,
283 buffer_info->dma = 0;
284 skb_put(skb, length);
288 if (!skb_shinfo(skb)->nr_frags) {
289 dma_unmap_single(&pdev->dev, buffer_info->dma,
290 adapter->rx_ps_hdr_size,
296 dma_unmap_page(&pdev->dev, buffer_info->page_dma,
299 buffer_info->page_dma = 0;
301 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
303 buffer_info->page_offset,
306 if ((adapter->rx_buffer_len > (PAGE_SIZE / 2)) ||
307 (page_count(buffer_info->page) != 1))
308 buffer_info->page = NULL;
310 get_page(buffer_info->page);
313 skb->data_len += length;
314 skb->truesize += length;
318 if (i == rx_ring->count)
320 next_rxd = IGBVF_RX_DESC_ADV(*rx_ring, i);
322 next_buffer = &rx_ring->buffer_info[i];
324 if (!(staterr & E1000_RXD_STAT_EOP)) {
325 buffer_info->skb = next_buffer->skb;
326 buffer_info->dma = next_buffer->dma;
327 next_buffer->skb = skb;
328 next_buffer->dma = 0;
332 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
333 dev_kfree_skb_irq(skb);
337 total_bytes += skb->len;
340 igbvf_rx_checksum_adv(adapter, staterr, skb);
342 skb->protocol = eth_type_trans(skb, netdev);
344 igbvf_receive_skb(adapter, netdev, skb, staterr,
345 rx_desc->wb.upper.vlan);
348 rx_desc->wb.upper.status_error = 0;
350 /* return some buffers to hardware, one at a time is too slow */
351 if (cleaned_count >= IGBVF_RX_BUFFER_WRITE) {
352 igbvf_alloc_rx_buffers(rx_ring, cleaned_count);
356 /* use prefetched values */
358 buffer_info = next_buffer;
360 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
363 rx_ring->next_to_clean = i;
364 cleaned_count = igbvf_desc_unused(rx_ring);
367 igbvf_alloc_rx_buffers(rx_ring, cleaned_count);
369 adapter->total_rx_packets += total_packets;
370 adapter->total_rx_bytes += total_bytes;
371 adapter->net_stats.rx_bytes += total_bytes;
372 adapter->net_stats.rx_packets += total_packets;
376 static void igbvf_put_txbuf(struct igbvf_adapter *adapter,
377 struct igbvf_buffer *buffer_info)
379 if (buffer_info->dma) {
380 if (buffer_info->mapped_as_page)
381 dma_unmap_page(&adapter->pdev->dev,
386 dma_unmap_single(&adapter->pdev->dev,
390 buffer_info->dma = 0;
392 if (buffer_info->skb) {
393 dev_kfree_skb_any(buffer_info->skb);
394 buffer_info->skb = NULL;
396 buffer_info->time_stamp = 0;
399 static void igbvf_print_tx_hang(struct igbvf_adapter *adapter)
401 struct igbvf_ring *tx_ring = adapter->tx_ring;
402 unsigned int i = tx_ring->next_to_clean;
403 unsigned int eop = tx_ring->buffer_info[i].next_to_watch;
404 union e1000_adv_tx_desc *eop_desc = IGBVF_TX_DESC_ADV(*tx_ring, eop);
406 /* detected Tx unit hang */
407 dev_err(&adapter->pdev->dev,
408 "Detected Tx Unit Hang:\n"
411 " next_to_use <%x>\n"
412 " next_to_clean <%x>\n"
413 "buffer_info[next_to_clean]:\n"
414 " time_stamp <%lx>\n"
415 " next_to_watch <%x>\n"
417 " next_to_watch.status <%x>\n",
418 readl(adapter->hw.hw_addr + tx_ring->head),
419 readl(adapter->hw.hw_addr + tx_ring->tail),
420 tx_ring->next_to_use,
421 tx_ring->next_to_clean,
422 tx_ring->buffer_info[eop].time_stamp,
425 eop_desc->wb.status);
429 * igbvf_setup_tx_resources - allocate Tx resources (Descriptors)
430 * @adapter: board private structure
432 * Return 0 on success, negative on failure
434 int igbvf_setup_tx_resources(struct igbvf_adapter *adapter,
435 struct igbvf_ring *tx_ring)
437 struct pci_dev *pdev = adapter->pdev;
440 size = sizeof(struct igbvf_buffer) * tx_ring->count;
441 tx_ring->buffer_info = vzalloc(size);
442 if (!tx_ring->buffer_info)
445 /* round up to nearest 4K */
446 tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
447 tx_ring->size = ALIGN(tx_ring->size, 4096);
449 tx_ring->desc = dma_alloc_coherent(&pdev->dev, tx_ring->size,
450 &tx_ring->dma, GFP_KERNEL);
455 tx_ring->adapter = adapter;
456 tx_ring->next_to_use = 0;
457 tx_ring->next_to_clean = 0;
461 vfree(tx_ring->buffer_info);
462 dev_err(&adapter->pdev->dev,
463 "Unable to allocate memory for the transmit descriptor ring\n");
468 * igbvf_setup_rx_resources - allocate Rx resources (Descriptors)
469 * @adapter: board private structure
471 * Returns 0 on success, negative on failure
473 int igbvf_setup_rx_resources(struct igbvf_adapter *adapter,
474 struct igbvf_ring *rx_ring)
476 struct pci_dev *pdev = adapter->pdev;
479 size = sizeof(struct igbvf_buffer) * rx_ring->count;
480 rx_ring->buffer_info = vzalloc(size);
481 if (!rx_ring->buffer_info)
484 desc_len = sizeof(union e1000_adv_rx_desc);
486 /* Round up to nearest 4K */
487 rx_ring->size = rx_ring->count * desc_len;
488 rx_ring->size = ALIGN(rx_ring->size, 4096);
490 rx_ring->desc = dma_alloc_coherent(&pdev->dev, rx_ring->size,
491 &rx_ring->dma, GFP_KERNEL);
496 rx_ring->next_to_clean = 0;
497 rx_ring->next_to_use = 0;
499 rx_ring->adapter = adapter;
504 vfree(rx_ring->buffer_info);
505 rx_ring->buffer_info = NULL;
506 dev_err(&adapter->pdev->dev,
507 "Unable to allocate memory for the receive descriptor ring\n");
512 * igbvf_clean_tx_ring - Free Tx Buffers
513 * @tx_ring: ring to be cleaned
515 static void igbvf_clean_tx_ring(struct igbvf_ring *tx_ring)
517 struct igbvf_adapter *adapter = tx_ring->adapter;
518 struct igbvf_buffer *buffer_info;
522 if (!tx_ring->buffer_info)
525 /* Free all the Tx ring sk_buffs */
526 for (i = 0; i < tx_ring->count; i++) {
527 buffer_info = &tx_ring->buffer_info[i];
528 igbvf_put_txbuf(adapter, buffer_info);
531 size = sizeof(struct igbvf_buffer) * tx_ring->count;
532 memset(tx_ring->buffer_info, 0, size);
534 /* Zero out the descriptor ring */
535 memset(tx_ring->desc, 0, tx_ring->size);
537 tx_ring->next_to_use = 0;
538 tx_ring->next_to_clean = 0;
540 writel(0, adapter->hw.hw_addr + tx_ring->head);
541 writel(0, adapter->hw.hw_addr + tx_ring->tail);
545 * igbvf_free_tx_resources - Free Tx Resources per Queue
546 * @tx_ring: ring to free resources from
548 * Free all transmit software resources
550 void igbvf_free_tx_resources(struct igbvf_ring *tx_ring)
552 struct pci_dev *pdev = tx_ring->adapter->pdev;
554 igbvf_clean_tx_ring(tx_ring);
556 vfree(tx_ring->buffer_info);
557 tx_ring->buffer_info = NULL;
559 dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
562 tx_ring->desc = NULL;
566 * igbvf_clean_rx_ring - Free Rx Buffers per Queue
567 * @adapter: board private structure
569 static void igbvf_clean_rx_ring(struct igbvf_ring *rx_ring)
571 struct igbvf_adapter *adapter = rx_ring->adapter;
572 struct igbvf_buffer *buffer_info;
573 struct pci_dev *pdev = adapter->pdev;
577 if (!rx_ring->buffer_info)
580 /* Free all the Rx ring sk_buffs */
581 for (i = 0; i < rx_ring->count; i++) {
582 buffer_info = &rx_ring->buffer_info[i];
583 if (buffer_info->dma) {
584 if (adapter->rx_ps_hdr_size){
585 dma_unmap_single(&pdev->dev, buffer_info->dma,
586 adapter->rx_ps_hdr_size,
589 dma_unmap_single(&pdev->dev, buffer_info->dma,
590 adapter->rx_buffer_len,
593 buffer_info->dma = 0;
596 if (buffer_info->skb) {
597 dev_kfree_skb(buffer_info->skb);
598 buffer_info->skb = NULL;
601 if (buffer_info->page) {
602 if (buffer_info->page_dma)
603 dma_unmap_page(&pdev->dev,
604 buffer_info->page_dma,
607 put_page(buffer_info->page);
608 buffer_info->page = NULL;
609 buffer_info->page_dma = 0;
610 buffer_info->page_offset = 0;
614 size = sizeof(struct igbvf_buffer) * rx_ring->count;
615 memset(rx_ring->buffer_info, 0, size);
617 /* Zero out the descriptor ring */
618 memset(rx_ring->desc, 0, rx_ring->size);
620 rx_ring->next_to_clean = 0;
621 rx_ring->next_to_use = 0;
623 writel(0, adapter->hw.hw_addr + rx_ring->head);
624 writel(0, adapter->hw.hw_addr + rx_ring->tail);
628 * igbvf_free_rx_resources - Free Rx Resources
629 * @rx_ring: ring to clean the resources from
631 * Free all receive software resources
634 void igbvf_free_rx_resources(struct igbvf_ring *rx_ring)
636 struct pci_dev *pdev = rx_ring->adapter->pdev;
638 igbvf_clean_rx_ring(rx_ring);
640 vfree(rx_ring->buffer_info);
641 rx_ring->buffer_info = NULL;
643 dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
645 rx_ring->desc = NULL;
649 * igbvf_update_itr - update the dynamic ITR value based on statistics
650 * @adapter: pointer to adapter
651 * @itr_setting: current adapter->itr
652 * @packets: the number of packets during this measurement interval
653 * @bytes: the number of bytes during this measurement interval
655 * Stores a new ITR value based on packets and byte
656 * counts during the last interrupt. The advantage of per interrupt
657 * computation is faster updates and more accurate ITR for the current
658 * traffic pattern. Constants in this function were computed
659 * based on theoretical maximum wire speed and thresholds were set based
660 * on testing data as well as attempting to minimize response time
661 * while increasing bulk throughput. This functionality is controlled
662 * by the InterruptThrottleRate module parameter.
664 static unsigned int igbvf_update_itr(struct igbvf_adapter *adapter,
665 u16 itr_setting, int packets,
668 unsigned int retval = itr_setting;
671 goto update_itr_done;
673 switch (itr_setting) {
675 /* handle TSO and jumbo frames */
676 if (bytes/packets > 8000)
677 retval = bulk_latency;
678 else if ((packets < 5) && (bytes > 512))
679 retval = low_latency;
681 case low_latency: /* 50 usec aka 20000 ints/s */
683 /* this if handles the TSO accounting */
684 if (bytes/packets > 8000)
685 retval = bulk_latency;
686 else if ((packets < 10) || ((bytes/packets) > 1200))
687 retval = bulk_latency;
688 else if ((packets > 35))
689 retval = lowest_latency;
690 } else if (bytes/packets > 2000) {
691 retval = bulk_latency;
692 } else if (packets <= 2 && bytes < 512) {
693 retval = lowest_latency;
696 case bulk_latency: /* 250 usec aka 4000 ints/s */
699 retval = low_latency;
700 } else if (bytes < 6000) {
701 retval = low_latency;
710 static void igbvf_set_itr(struct igbvf_adapter *adapter)
712 struct e1000_hw *hw = &adapter->hw;
714 u32 new_itr = adapter->itr;
716 adapter->tx_itr = igbvf_update_itr(adapter, adapter->tx_itr,
717 adapter->total_tx_packets,
718 adapter->total_tx_bytes);
719 /* conservative mode (itr 3) eliminates the lowest_latency setting */
720 if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
721 adapter->tx_itr = low_latency;
723 adapter->rx_itr = igbvf_update_itr(adapter, adapter->rx_itr,
724 adapter->total_rx_packets,
725 adapter->total_rx_bytes);
726 /* conservative mode (itr 3) eliminates the lowest_latency setting */
727 if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
728 adapter->rx_itr = low_latency;
730 current_itr = max(adapter->rx_itr, adapter->tx_itr);
732 switch (current_itr) {
733 /* counts and packets in update_itr are dependent on these numbers */
738 new_itr = 20000; /* aka hwitr = ~200 */
747 if (new_itr != adapter->itr) {
749 * this attempts to bias the interrupt rate towards Bulk
750 * by adding intermediate steps when interrupt rate is
753 new_itr = new_itr > adapter->itr ?
754 min(adapter->itr + (new_itr >> 2), new_itr) :
756 adapter->itr = new_itr;
757 adapter->rx_ring->itr_val = 1952;
759 if (adapter->msix_entries)
760 adapter->rx_ring->set_itr = 1;
767 * igbvf_clean_tx_irq - Reclaim resources after transmit completes
768 * @adapter: board private structure
769 * returns true if ring is completely cleaned
771 static bool igbvf_clean_tx_irq(struct igbvf_ring *tx_ring)
773 struct igbvf_adapter *adapter = tx_ring->adapter;
774 struct e1000_hw *hw = &adapter->hw;
775 struct net_device *netdev = adapter->netdev;
776 struct igbvf_buffer *buffer_info;
778 union e1000_adv_tx_desc *tx_desc, *eop_desc;
779 unsigned int total_bytes = 0, total_packets = 0;
780 unsigned int i, eop, count = 0;
781 bool cleaned = false;
783 i = tx_ring->next_to_clean;
784 eop = tx_ring->buffer_info[i].next_to_watch;
785 eop_desc = IGBVF_TX_DESC_ADV(*tx_ring, eop);
787 while ((eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)) &&
788 (count < tx_ring->count)) {
789 rmb(); /* read buffer_info after eop_desc status */
790 for (cleaned = false; !cleaned; count++) {
791 tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
792 buffer_info = &tx_ring->buffer_info[i];
793 cleaned = (i == eop);
794 skb = buffer_info->skb;
797 unsigned int segs, bytecount;
799 /* gso_segs is currently only valid for tcp */
800 segs = skb_shinfo(skb)->gso_segs ?: 1;
801 /* multiply data chunks by size of headers */
802 bytecount = ((segs - 1) * skb_headlen(skb)) +
804 total_packets += segs;
805 total_bytes += bytecount;
808 igbvf_put_txbuf(adapter, buffer_info);
809 tx_desc->wb.status = 0;
812 if (i == tx_ring->count)
815 eop = tx_ring->buffer_info[i].next_to_watch;
816 eop_desc = IGBVF_TX_DESC_ADV(*tx_ring, eop);
819 tx_ring->next_to_clean = i;
821 if (unlikely(count &&
822 netif_carrier_ok(netdev) &&
823 igbvf_desc_unused(tx_ring) >= IGBVF_TX_QUEUE_WAKE)) {
824 /* Make sure that anybody stopping the queue after this
825 * sees the new next_to_clean.
828 if (netif_queue_stopped(netdev) &&
829 !(test_bit(__IGBVF_DOWN, &adapter->state))) {
830 netif_wake_queue(netdev);
831 ++adapter->restart_queue;
835 if (adapter->detect_tx_hung) {
836 /* Detect a transmit hang in hardware, this serializes the
837 * check with the clearing of time_stamp and movement of i */
838 adapter->detect_tx_hung = false;
839 if (tx_ring->buffer_info[i].time_stamp &&
840 time_after(jiffies, tx_ring->buffer_info[i].time_stamp +
841 (adapter->tx_timeout_factor * HZ)) &&
842 !(er32(STATUS) & E1000_STATUS_TXOFF)) {
844 tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
845 /* detected Tx unit hang */
846 igbvf_print_tx_hang(adapter);
848 netif_stop_queue(netdev);
851 adapter->net_stats.tx_bytes += total_bytes;
852 adapter->net_stats.tx_packets += total_packets;
853 return count < tx_ring->count;
856 static irqreturn_t igbvf_msix_other(int irq, void *data)
858 struct net_device *netdev = data;
859 struct igbvf_adapter *adapter = netdev_priv(netdev);
860 struct e1000_hw *hw = &adapter->hw;
862 adapter->int_counter1++;
864 netif_carrier_off(netdev);
865 hw->mac.get_link_status = 1;
866 if (!test_bit(__IGBVF_DOWN, &adapter->state))
867 mod_timer(&adapter->watchdog_timer, jiffies + 1);
869 ew32(EIMS, adapter->eims_other);
874 static irqreturn_t igbvf_intr_msix_tx(int irq, void *data)
876 struct net_device *netdev = data;
877 struct igbvf_adapter *adapter = netdev_priv(netdev);
878 struct e1000_hw *hw = &adapter->hw;
879 struct igbvf_ring *tx_ring = adapter->tx_ring;
882 adapter->total_tx_bytes = 0;
883 adapter->total_tx_packets = 0;
885 /* auto mask will automatically reenable the interrupt when we write
887 if (!igbvf_clean_tx_irq(tx_ring))
888 /* Ring was not completely cleaned, so fire another interrupt */
889 ew32(EICS, tx_ring->eims_value);
891 ew32(EIMS, tx_ring->eims_value);
896 static irqreturn_t igbvf_intr_msix_rx(int irq, void *data)
898 struct net_device *netdev = data;
899 struct igbvf_adapter *adapter = netdev_priv(netdev);
901 adapter->int_counter0++;
903 /* Write the ITR value calculated at the end of the
904 * previous interrupt.
906 if (adapter->rx_ring->set_itr) {
907 writel(adapter->rx_ring->itr_val,
908 adapter->hw.hw_addr + adapter->rx_ring->itr_register);
909 adapter->rx_ring->set_itr = 0;
912 if (napi_schedule_prep(&adapter->rx_ring->napi)) {
913 adapter->total_rx_bytes = 0;
914 adapter->total_rx_packets = 0;
915 __napi_schedule(&adapter->rx_ring->napi);
921 #define IGBVF_NO_QUEUE -1
923 static void igbvf_assign_vector(struct igbvf_adapter *adapter, int rx_queue,
924 int tx_queue, int msix_vector)
926 struct e1000_hw *hw = &adapter->hw;
929 /* 82576 uses a table-based method for assigning vectors.
930 Each queue has a single entry in the table to which we write
931 a vector number along with a "valid" bit. Sadly, the layout
932 of the table is somewhat counterintuitive. */
933 if (rx_queue > IGBVF_NO_QUEUE) {
934 index = (rx_queue >> 1);
935 ivar = array_er32(IVAR0, index);
936 if (rx_queue & 0x1) {
937 /* vector goes into third byte of register */
938 ivar = ivar & 0xFF00FFFF;
939 ivar |= (msix_vector | E1000_IVAR_VALID) << 16;
941 /* vector goes into low byte of register */
942 ivar = ivar & 0xFFFFFF00;
943 ivar |= msix_vector | E1000_IVAR_VALID;
945 adapter->rx_ring[rx_queue].eims_value = 1 << msix_vector;
946 array_ew32(IVAR0, index, ivar);
948 if (tx_queue > IGBVF_NO_QUEUE) {
949 index = (tx_queue >> 1);
950 ivar = array_er32(IVAR0, index);
951 if (tx_queue & 0x1) {
952 /* vector goes into high byte of register */
953 ivar = ivar & 0x00FFFFFF;
954 ivar |= (msix_vector | E1000_IVAR_VALID) << 24;
956 /* vector goes into second byte of register */
957 ivar = ivar & 0xFFFF00FF;
958 ivar |= (msix_vector | E1000_IVAR_VALID) << 8;
960 adapter->tx_ring[tx_queue].eims_value = 1 << msix_vector;
961 array_ew32(IVAR0, index, ivar);
966 * igbvf_configure_msix - Configure MSI-X hardware
968 * igbvf_configure_msix sets up the hardware to properly
969 * generate MSI-X interrupts.
971 static void igbvf_configure_msix(struct igbvf_adapter *adapter)
974 struct e1000_hw *hw = &adapter->hw;
975 struct igbvf_ring *tx_ring = adapter->tx_ring;
976 struct igbvf_ring *rx_ring = adapter->rx_ring;
979 adapter->eims_enable_mask = 0;
981 igbvf_assign_vector(adapter, IGBVF_NO_QUEUE, 0, vector++);
982 adapter->eims_enable_mask |= tx_ring->eims_value;
983 if (tx_ring->itr_val)
984 writel(tx_ring->itr_val,
985 hw->hw_addr + tx_ring->itr_register);
987 writel(1952, hw->hw_addr + tx_ring->itr_register);
989 igbvf_assign_vector(adapter, 0, IGBVF_NO_QUEUE, vector++);
990 adapter->eims_enable_mask |= rx_ring->eims_value;
991 if (rx_ring->itr_val)
992 writel(rx_ring->itr_val,
993 hw->hw_addr + rx_ring->itr_register);
995 writel(1952, hw->hw_addr + rx_ring->itr_register);
997 /* set vector for other causes, i.e. link changes */
999 tmp = (vector++ | E1000_IVAR_VALID);
1001 ew32(IVAR_MISC, tmp);
1003 adapter->eims_enable_mask = (1 << (vector)) - 1;
1004 adapter->eims_other = 1 << (vector - 1);
1008 static void igbvf_reset_interrupt_capability(struct igbvf_adapter *adapter)
1010 if (adapter->msix_entries) {
1011 pci_disable_msix(adapter->pdev);
1012 kfree(adapter->msix_entries);
1013 adapter->msix_entries = NULL;
1018 * igbvf_set_interrupt_capability - set MSI or MSI-X if supported
1020 * Attempt to configure interrupts using the best available
1021 * capabilities of the hardware and kernel.
1023 static void igbvf_set_interrupt_capability(struct igbvf_adapter *adapter)
1028 /* we allocate 3 vectors, 1 for tx, 1 for rx, one for pf messages */
1029 adapter->msix_entries = kcalloc(3, sizeof(struct msix_entry),
1031 if (adapter->msix_entries) {
1032 for (i = 0; i < 3; i++)
1033 adapter->msix_entries[i].entry = i;
1035 err = pci_enable_msix(adapter->pdev,
1036 adapter->msix_entries, 3);
1041 dev_err(&adapter->pdev->dev,
1042 "Failed to initialize MSI-X interrupts.\n");
1043 igbvf_reset_interrupt_capability(adapter);
1048 * igbvf_request_msix - Initialize MSI-X interrupts
1050 * igbvf_request_msix allocates MSI-X vectors and requests interrupts from the
1053 static int igbvf_request_msix(struct igbvf_adapter *adapter)
1055 struct net_device *netdev = adapter->netdev;
1056 int err = 0, vector = 0;
1058 if (strlen(netdev->name) < (IFNAMSIZ - 5)) {
1059 sprintf(adapter->tx_ring->name, "%s-tx-0", netdev->name);
1060 sprintf(adapter->rx_ring->name, "%s-rx-0", netdev->name);
1062 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1063 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1066 err = request_irq(adapter->msix_entries[vector].vector,
1067 igbvf_intr_msix_tx, 0, adapter->tx_ring->name,
1072 adapter->tx_ring->itr_register = E1000_EITR(vector);
1073 adapter->tx_ring->itr_val = 1952;
1076 err = request_irq(adapter->msix_entries[vector].vector,
1077 igbvf_intr_msix_rx, 0, adapter->rx_ring->name,
1082 adapter->rx_ring->itr_register = E1000_EITR(vector);
1083 adapter->rx_ring->itr_val = 1952;
1086 err = request_irq(adapter->msix_entries[vector].vector,
1087 igbvf_msix_other, 0, netdev->name, netdev);
1091 igbvf_configure_msix(adapter);
1098 * igbvf_alloc_queues - Allocate memory for all rings
1099 * @adapter: board private structure to initialize
1101 static int __devinit igbvf_alloc_queues(struct igbvf_adapter *adapter)
1103 struct net_device *netdev = adapter->netdev;
1105 adapter->tx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL);
1106 if (!adapter->tx_ring)
1109 adapter->rx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL);
1110 if (!adapter->rx_ring) {
1111 kfree(adapter->tx_ring);
1115 netif_napi_add(netdev, &adapter->rx_ring->napi, igbvf_poll, 64);
1121 * igbvf_request_irq - initialize interrupts
1123 * Attempts to configure interrupts using the best available
1124 * capabilities of the hardware and kernel.
1126 static int igbvf_request_irq(struct igbvf_adapter *adapter)
1130 /* igbvf supports msi-x only */
1131 if (adapter->msix_entries)
1132 err = igbvf_request_msix(adapter);
1137 dev_err(&adapter->pdev->dev,
1138 "Unable to allocate interrupt, Error: %d\n", err);
1143 static void igbvf_free_irq(struct igbvf_adapter *adapter)
1145 struct net_device *netdev = adapter->netdev;
1148 if (adapter->msix_entries) {
1149 for (vector = 0; vector < 3; vector++)
1150 free_irq(adapter->msix_entries[vector].vector, netdev);
1155 * igbvf_irq_disable - Mask off interrupt generation on the NIC
1157 static void igbvf_irq_disable(struct igbvf_adapter *adapter)
1159 struct e1000_hw *hw = &adapter->hw;
1163 if (adapter->msix_entries)
1168 * igbvf_irq_enable - Enable default interrupt generation settings
1170 static void igbvf_irq_enable(struct igbvf_adapter *adapter)
1172 struct e1000_hw *hw = &adapter->hw;
1174 ew32(EIAC, adapter->eims_enable_mask);
1175 ew32(EIAM, adapter->eims_enable_mask);
1176 ew32(EIMS, adapter->eims_enable_mask);
1180 * igbvf_poll - NAPI Rx polling callback
1181 * @napi: struct associated with this polling callback
1182 * @budget: amount of packets driver is allowed to process this poll
1184 static int igbvf_poll(struct napi_struct *napi, int budget)
1186 struct igbvf_ring *rx_ring = container_of(napi, struct igbvf_ring, napi);
1187 struct igbvf_adapter *adapter = rx_ring->adapter;
1188 struct e1000_hw *hw = &adapter->hw;
1191 igbvf_clean_rx_irq(adapter, &work_done, budget);
1193 /* If not enough Rx work done, exit the polling mode */
1194 if (work_done < budget) {
1195 napi_complete(napi);
1197 if (adapter->itr_setting & 3)
1198 igbvf_set_itr(adapter);
1200 if (!test_bit(__IGBVF_DOWN, &adapter->state))
1201 ew32(EIMS, adapter->rx_ring->eims_value);
1208 * igbvf_set_rlpml - set receive large packet maximum length
1209 * @adapter: board private structure
1211 * Configure the maximum size of packets that will be received
1213 static void igbvf_set_rlpml(struct igbvf_adapter *adapter)
1215 int max_frame_size = adapter->max_frame_size;
1216 struct e1000_hw *hw = &adapter->hw;
1219 max_frame_size += VLAN_TAG_SIZE;
1221 e1000_rlpml_set_vf(hw, max_frame_size);
1224 static void igbvf_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1226 struct igbvf_adapter *adapter = netdev_priv(netdev);
1227 struct e1000_hw *hw = &adapter->hw;
1229 if (hw->mac.ops.set_vfta(hw, vid, true))
1230 dev_err(&adapter->pdev->dev, "Failed to add vlan id %d\n", vid);
1233 static void igbvf_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
1235 struct igbvf_adapter *adapter = netdev_priv(netdev);
1236 struct e1000_hw *hw = &adapter->hw;
1238 igbvf_irq_disable(adapter);
1239 vlan_group_set_device(adapter->vlgrp, vid, NULL);
1241 if (!test_bit(__IGBVF_DOWN, &adapter->state))
1242 igbvf_irq_enable(adapter);
1244 if (hw->mac.ops.set_vfta(hw, vid, false))
1245 dev_err(&adapter->pdev->dev,
1246 "Failed to remove vlan id %d\n", vid);
1249 static void igbvf_vlan_rx_register(struct net_device *netdev,
1250 struct vlan_group *grp)
1252 struct igbvf_adapter *adapter = netdev_priv(netdev);
1254 adapter->vlgrp = grp;
1257 static void igbvf_restore_vlan(struct igbvf_adapter *adapter)
1261 if (!adapter->vlgrp)
1264 for (vid = 0; vid < VLAN_N_VID; vid++) {
1265 if (!vlan_group_get_device(adapter->vlgrp, vid))
1267 igbvf_vlan_rx_add_vid(adapter->netdev, vid);
1270 igbvf_set_rlpml(adapter);
1274 * igbvf_configure_tx - Configure Transmit Unit after Reset
1275 * @adapter: board private structure
1277 * Configure the Tx unit of the MAC after a reset.
1279 static void igbvf_configure_tx(struct igbvf_adapter *adapter)
1281 struct e1000_hw *hw = &adapter->hw;
1282 struct igbvf_ring *tx_ring = adapter->tx_ring;
1284 u32 txdctl, dca_txctrl;
1286 /* disable transmits */
1287 txdctl = er32(TXDCTL(0));
1288 ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1291 /* Setup the HW Tx Head and Tail descriptor pointers */
1292 ew32(TDLEN(0), tx_ring->count * sizeof(union e1000_adv_tx_desc));
1293 tdba = tx_ring->dma;
1294 ew32(TDBAL(0), (tdba & DMA_BIT_MASK(32)));
1295 ew32(TDBAH(0), (tdba >> 32));
1298 tx_ring->head = E1000_TDH(0);
1299 tx_ring->tail = E1000_TDT(0);
1301 /* Turn off Relaxed Ordering on head write-backs. The writebacks
1302 * MUST be delivered in order or it will completely screw up
1305 dca_txctrl = er32(DCA_TXCTRL(0));
1306 dca_txctrl &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN;
1307 ew32(DCA_TXCTRL(0), dca_txctrl);
1309 /* enable transmits */
1310 txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
1311 ew32(TXDCTL(0), txdctl);
1313 /* Setup Transmit Descriptor Settings for eop descriptor */
1314 adapter->txd_cmd = E1000_ADVTXD_DCMD_EOP | E1000_ADVTXD_DCMD_IFCS;
1316 /* enable Report Status bit */
1317 adapter->txd_cmd |= E1000_ADVTXD_DCMD_RS;
1321 * igbvf_setup_srrctl - configure the receive control registers
1322 * @adapter: Board private structure
1324 static void igbvf_setup_srrctl(struct igbvf_adapter *adapter)
1326 struct e1000_hw *hw = &adapter->hw;
1329 srrctl &= ~(E1000_SRRCTL_DESCTYPE_MASK |
1330 E1000_SRRCTL_BSIZEHDR_MASK |
1331 E1000_SRRCTL_BSIZEPKT_MASK);
1333 /* Enable queue drop to avoid head of line blocking */
1334 srrctl |= E1000_SRRCTL_DROP_EN;
1336 /* Setup buffer sizes */
1337 srrctl |= ALIGN(adapter->rx_buffer_len, 1024) >>
1338 E1000_SRRCTL_BSIZEPKT_SHIFT;
1340 if (adapter->rx_buffer_len < 2048) {
1341 adapter->rx_ps_hdr_size = 0;
1342 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
1344 adapter->rx_ps_hdr_size = 128;
1345 srrctl |= adapter->rx_ps_hdr_size <<
1346 E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
1347 srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
1350 ew32(SRRCTL(0), srrctl);
1354 * igbvf_configure_rx - Configure Receive Unit after Reset
1355 * @adapter: board private structure
1357 * Configure the Rx unit of the MAC after a reset.
1359 static void igbvf_configure_rx(struct igbvf_adapter *adapter)
1361 struct e1000_hw *hw = &adapter->hw;
1362 struct igbvf_ring *rx_ring = adapter->rx_ring;
1366 /* disable receives */
1367 rxdctl = er32(RXDCTL(0));
1368 ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1371 rdlen = rx_ring->count * sizeof(union e1000_adv_rx_desc);
1374 * Setup the HW Rx Head and Tail Descriptor Pointers and
1375 * the Base and Length of the Rx Descriptor Ring
1377 rdba = rx_ring->dma;
1378 ew32(RDBAL(0), (rdba & DMA_BIT_MASK(32)));
1379 ew32(RDBAH(0), (rdba >> 32));
1380 ew32(RDLEN(0), rx_ring->count * sizeof(union e1000_adv_rx_desc));
1381 rx_ring->head = E1000_RDH(0);
1382 rx_ring->tail = E1000_RDT(0);
1386 rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
1387 rxdctl &= 0xFFF00000;
1388 rxdctl |= IGBVF_RX_PTHRESH;
1389 rxdctl |= IGBVF_RX_HTHRESH << 8;
1390 rxdctl |= IGBVF_RX_WTHRESH << 16;
1392 igbvf_set_rlpml(adapter);
1394 /* enable receives */
1395 ew32(RXDCTL(0), rxdctl);
1399 * igbvf_set_multi - Multicast and Promiscuous mode set
1400 * @netdev: network interface device structure
1402 * The set_multi entry point is called whenever the multicast address
1403 * list or the network interface flags are updated. This routine is
1404 * responsible for configuring the hardware for proper multicast,
1405 * promiscuous mode, and all-multi behavior.
1407 static void igbvf_set_multi(struct net_device *netdev)
1409 struct igbvf_adapter *adapter = netdev_priv(netdev);
1410 struct e1000_hw *hw = &adapter->hw;
1411 struct netdev_hw_addr *ha;
1412 u8 *mta_list = NULL;
1415 if (!netdev_mc_empty(netdev)) {
1416 mta_list = kmalloc(netdev_mc_count(netdev) * 6, GFP_ATOMIC);
1418 dev_err(&adapter->pdev->dev,
1419 "failed to allocate multicast filter list\n");
1424 /* prepare a packed array of only addresses. */
1426 netdev_for_each_mc_addr(ha, netdev)
1427 memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
1429 hw->mac.ops.update_mc_addr_list(hw, mta_list, i, 0, 0);
1434 * igbvf_configure - configure the hardware for Rx and Tx
1435 * @adapter: private board structure
1437 static void igbvf_configure(struct igbvf_adapter *adapter)
1439 igbvf_set_multi(adapter->netdev);
1441 igbvf_restore_vlan(adapter);
1443 igbvf_configure_tx(adapter);
1444 igbvf_setup_srrctl(adapter);
1445 igbvf_configure_rx(adapter);
1446 igbvf_alloc_rx_buffers(adapter->rx_ring,
1447 igbvf_desc_unused(adapter->rx_ring));
1450 /* igbvf_reset - bring the hardware into a known good state
1452 * This function boots the hardware and enables some settings that
1453 * require a configuration cycle of the hardware - those cannot be
1454 * set/changed during runtime. After reset the device needs to be
1455 * properly configured for Rx, Tx etc.
1457 static void igbvf_reset(struct igbvf_adapter *adapter)
1459 struct e1000_mac_info *mac = &adapter->hw.mac;
1460 struct net_device *netdev = adapter->netdev;
1461 struct e1000_hw *hw = &adapter->hw;
1463 /* Allow time for pending master requests to run */
1464 if (mac->ops.reset_hw(hw))
1465 dev_err(&adapter->pdev->dev, "PF still resetting\n");
1467 mac->ops.init_hw(hw);
1469 if (is_valid_ether_addr(adapter->hw.mac.addr)) {
1470 memcpy(netdev->dev_addr, adapter->hw.mac.addr,
1472 memcpy(netdev->perm_addr, adapter->hw.mac.addr,
1476 adapter->last_reset = jiffies;
1479 int igbvf_up(struct igbvf_adapter *adapter)
1481 struct e1000_hw *hw = &adapter->hw;
1483 /* hardware has been reset, we need to reload some things */
1484 igbvf_configure(adapter);
1486 clear_bit(__IGBVF_DOWN, &adapter->state);
1488 napi_enable(&adapter->rx_ring->napi);
1489 if (adapter->msix_entries)
1490 igbvf_configure_msix(adapter);
1492 /* Clear any pending interrupts. */
1494 igbvf_irq_enable(adapter);
1496 /* start the watchdog */
1497 hw->mac.get_link_status = 1;
1498 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1504 void igbvf_down(struct igbvf_adapter *adapter)
1506 struct net_device *netdev = adapter->netdev;
1507 struct e1000_hw *hw = &adapter->hw;
1511 * signal that we're down so the interrupt handler does not
1512 * reschedule our watchdog timer
1514 set_bit(__IGBVF_DOWN, &adapter->state);
1516 /* disable receives in the hardware */
1517 rxdctl = er32(RXDCTL(0));
1518 ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1520 netif_stop_queue(netdev);
1522 /* disable transmits in the hardware */
1523 txdctl = er32(TXDCTL(0));
1524 ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1526 /* flush both disables and wait for them to finish */
1530 napi_disable(&adapter->rx_ring->napi);
1532 igbvf_irq_disable(adapter);
1534 del_timer_sync(&adapter->watchdog_timer);
1536 netif_carrier_off(netdev);
1538 /* record the stats before reset*/
1539 igbvf_update_stats(adapter);
1541 adapter->link_speed = 0;
1542 adapter->link_duplex = 0;
1544 igbvf_reset(adapter);
1545 igbvf_clean_tx_ring(adapter->tx_ring);
1546 igbvf_clean_rx_ring(adapter->rx_ring);
1549 void igbvf_reinit_locked(struct igbvf_adapter *adapter)
1552 while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
1554 igbvf_down(adapter);
1556 clear_bit(__IGBVF_RESETTING, &adapter->state);
1560 * igbvf_sw_init - Initialize general software structures (struct igbvf_adapter)
1561 * @adapter: board private structure to initialize
1563 * igbvf_sw_init initializes the Adapter private data structure.
1564 * Fields are initialized based on PCI device information and
1565 * OS network device settings (MTU size).
1567 static int __devinit igbvf_sw_init(struct igbvf_adapter *adapter)
1569 struct net_device *netdev = adapter->netdev;
1572 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
1573 adapter->rx_ps_hdr_size = 0;
1574 adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1575 adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
1577 adapter->tx_int_delay = 8;
1578 adapter->tx_abs_int_delay = 32;
1579 adapter->rx_int_delay = 0;
1580 adapter->rx_abs_int_delay = 8;
1581 adapter->itr_setting = 3;
1582 adapter->itr = 20000;
1584 /* Set various function pointers */
1585 adapter->ei->init_ops(&adapter->hw);
1587 rc = adapter->hw.mac.ops.init_params(&adapter->hw);
1591 rc = adapter->hw.mbx.ops.init_params(&adapter->hw);
1595 igbvf_set_interrupt_capability(adapter);
1597 if (igbvf_alloc_queues(adapter))
1600 spin_lock_init(&adapter->tx_queue_lock);
1602 /* Explicitly disable IRQ since the NIC can be in any state. */
1603 igbvf_irq_disable(adapter);
1605 spin_lock_init(&adapter->stats_lock);
1607 set_bit(__IGBVF_DOWN, &adapter->state);
1611 static void igbvf_initialize_last_counter_stats(struct igbvf_adapter *adapter)
1613 struct e1000_hw *hw = &adapter->hw;
1615 adapter->stats.last_gprc = er32(VFGPRC);
1616 adapter->stats.last_gorc = er32(VFGORC);
1617 adapter->stats.last_gptc = er32(VFGPTC);
1618 adapter->stats.last_gotc = er32(VFGOTC);
1619 adapter->stats.last_mprc = er32(VFMPRC);
1620 adapter->stats.last_gotlbc = er32(VFGOTLBC);
1621 adapter->stats.last_gptlbc = er32(VFGPTLBC);
1622 adapter->stats.last_gorlbc = er32(VFGORLBC);
1623 adapter->stats.last_gprlbc = er32(VFGPRLBC);
1625 adapter->stats.base_gprc = er32(VFGPRC);
1626 adapter->stats.base_gorc = er32(VFGORC);
1627 adapter->stats.base_gptc = er32(VFGPTC);
1628 adapter->stats.base_gotc = er32(VFGOTC);
1629 adapter->stats.base_mprc = er32(VFMPRC);
1630 adapter->stats.base_gotlbc = er32(VFGOTLBC);
1631 adapter->stats.base_gptlbc = er32(VFGPTLBC);
1632 adapter->stats.base_gorlbc = er32(VFGORLBC);
1633 adapter->stats.base_gprlbc = er32(VFGPRLBC);
1637 * igbvf_open - Called when a network interface is made active
1638 * @netdev: network interface device structure
1640 * Returns 0 on success, negative value on failure
1642 * The open entry point is called when a network interface is made
1643 * active by the system (IFF_UP). At this point all resources needed
1644 * for transmit and receive operations are allocated, the interrupt
1645 * handler is registered with the OS, the watchdog timer is started,
1646 * and the stack is notified that the interface is ready.
1648 static int igbvf_open(struct net_device *netdev)
1650 struct igbvf_adapter *adapter = netdev_priv(netdev);
1651 struct e1000_hw *hw = &adapter->hw;
1654 /* disallow open during test */
1655 if (test_bit(__IGBVF_TESTING, &adapter->state))
1658 /* allocate transmit descriptors */
1659 err = igbvf_setup_tx_resources(adapter, adapter->tx_ring);
1663 /* allocate receive descriptors */
1664 err = igbvf_setup_rx_resources(adapter, adapter->rx_ring);
1669 * before we allocate an interrupt, we must be ready to handle it.
1670 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1671 * as soon as we call pci_request_irq, so we have to setup our
1672 * clean_rx handler before we do so.
1674 igbvf_configure(adapter);
1676 err = igbvf_request_irq(adapter);
1680 /* From here on the code is the same as igbvf_up() */
1681 clear_bit(__IGBVF_DOWN, &adapter->state);
1683 napi_enable(&adapter->rx_ring->napi);
1685 /* clear any pending interrupts */
1688 igbvf_irq_enable(adapter);
1690 /* start the watchdog */
1691 hw->mac.get_link_status = 1;
1692 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1697 igbvf_free_rx_resources(adapter->rx_ring);
1699 igbvf_free_tx_resources(adapter->tx_ring);
1701 igbvf_reset(adapter);
1707 * igbvf_close - Disables a network interface
1708 * @netdev: network interface device structure
1710 * Returns 0, this is not allowed to fail
1712 * The close entry point is called when an interface is de-activated
1713 * by the OS. The hardware is still under the drivers control, but
1714 * needs to be disabled. A global MAC reset is issued to stop the
1715 * hardware, and all transmit and receive resources are freed.
1717 static int igbvf_close(struct net_device *netdev)
1719 struct igbvf_adapter *adapter = netdev_priv(netdev);
1721 WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
1722 igbvf_down(adapter);
1724 igbvf_free_irq(adapter);
1726 igbvf_free_tx_resources(adapter->tx_ring);
1727 igbvf_free_rx_resources(adapter->rx_ring);
1732 * igbvf_set_mac - Change the Ethernet Address of the NIC
1733 * @netdev: network interface device structure
1734 * @p: pointer to an address structure
1736 * Returns 0 on success, negative on failure
1738 static int igbvf_set_mac(struct net_device *netdev, void *p)
1740 struct igbvf_adapter *adapter = netdev_priv(netdev);
1741 struct e1000_hw *hw = &adapter->hw;
1742 struct sockaddr *addr = p;
1744 if (!is_valid_ether_addr(addr->sa_data))
1745 return -EADDRNOTAVAIL;
1747 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
1749 hw->mac.ops.rar_set(hw, hw->mac.addr, 0);
1751 if (memcmp(addr->sa_data, hw->mac.addr, 6))
1752 return -EADDRNOTAVAIL;
1754 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
1759 #define UPDATE_VF_COUNTER(reg, name) \
1761 u32 current_counter = er32(reg); \
1762 if (current_counter < adapter->stats.last_##name) \
1763 adapter->stats.name += 0x100000000LL; \
1764 adapter->stats.last_##name = current_counter; \
1765 adapter->stats.name &= 0xFFFFFFFF00000000LL; \
1766 adapter->stats.name |= current_counter; \
1770 * igbvf_update_stats - Update the board statistics counters
1771 * @adapter: board private structure
1773 void igbvf_update_stats(struct igbvf_adapter *adapter)
1775 struct e1000_hw *hw = &adapter->hw;
1776 struct pci_dev *pdev = adapter->pdev;
1779 * Prevent stats update while adapter is being reset, link is down
1780 * or if the pci connection is down.
1782 if (adapter->link_speed == 0)
1785 if (test_bit(__IGBVF_RESETTING, &adapter->state))
1788 if (pci_channel_offline(pdev))
1791 UPDATE_VF_COUNTER(VFGPRC, gprc);
1792 UPDATE_VF_COUNTER(VFGORC, gorc);
1793 UPDATE_VF_COUNTER(VFGPTC, gptc);
1794 UPDATE_VF_COUNTER(VFGOTC, gotc);
1795 UPDATE_VF_COUNTER(VFMPRC, mprc);
1796 UPDATE_VF_COUNTER(VFGOTLBC, gotlbc);
1797 UPDATE_VF_COUNTER(VFGPTLBC, gptlbc);
1798 UPDATE_VF_COUNTER(VFGORLBC, gorlbc);
1799 UPDATE_VF_COUNTER(VFGPRLBC, gprlbc);
1801 /* Fill out the OS statistics structure */
1802 adapter->net_stats.multicast = adapter->stats.mprc;
1805 static void igbvf_print_link_info(struct igbvf_adapter *adapter)
1807 dev_info(&adapter->pdev->dev, "Link is Up %d Mbps %s\n",
1808 adapter->link_speed,
1809 ((adapter->link_duplex == FULL_DUPLEX) ?
1810 "Full Duplex" : "Half Duplex"));
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 /* adjust timeout factor according to speed/duplex */
1867 adapter->tx_timeout_factor = 1;
1868 switch (adapter->link_speed) {
1870 adapter->tx_timeout_factor = 16;
1873 /* maybe add some timeout factor ? */
1877 netif_carrier_on(netdev);
1878 netif_wake_queue(netdev);
1881 if (netif_carrier_ok(netdev)) {
1882 adapter->link_speed = 0;
1883 adapter->link_duplex = 0;
1884 dev_info(&adapter->pdev->dev, "Link is Down\n");
1885 netif_carrier_off(netdev);
1886 netif_stop_queue(netdev);
1890 if (netif_carrier_ok(netdev)) {
1891 igbvf_update_stats(adapter);
1893 tx_pending = (igbvf_desc_unused(tx_ring) + 1 <
1897 * We've lost link, so the controller stops DMA,
1898 * but we've got queued Tx work that's never going
1899 * to get done, so reset controller to flush Tx.
1900 * (Do the reset outside of interrupt context).
1902 adapter->tx_timeout_count++;
1903 schedule_work(&adapter->reset_task);
1907 /* Cause software interrupt to ensure Rx ring is cleaned */
1908 ew32(EICS, adapter->rx_ring->eims_value);
1910 /* Force detection of hung controller every watchdog period */
1911 adapter->detect_tx_hung = 1;
1913 /* Reset the timer */
1914 if (!test_bit(__IGBVF_DOWN, &adapter->state))
1915 mod_timer(&adapter->watchdog_timer,
1916 round_jiffies(jiffies + (2 * HZ)));
1919 #define IGBVF_TX_FLAGS_CSUM 0x00000001
1920 #define IGBVF_TX_FLAGS_VLAN 0x00000002
1921 #define IGBVF_TX_FLAGS_TSO 0x00000004
1922 #define IGBVF_TX_FLAGS_IPV4 0x00000008
1923 #define IGBVF_TX_FLAGS_VLAN_MASK 0xffff0000
1924 #define IGBVF_TX_FLAGS_VLAN_SHIFT 16
1926 static int igbvf_tso(struct igbvf_adapter *adapter,
1927 struct igbvf_ring *tx_ring,
1928 struct sk_buff *skb, u32 tx_flags, u8 *hdr_len)
1930 struct e1000_adv_tx_context_desc *context_desc;
1933 struct igbvf_buffer *buffer_info;
1934 u32 info = 0, tu_cmd = 0;
1935 u32 mss_l4len_idx, l4len;
1938 if (skb_header_cloned(skb)) {
1939 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1941 dev_err(&adapter->pdev->dev,
1942 "igbvf_tso returning an error\n");
1947 l4len = tcp_hdrlen(skb);
1950 if (skb->protocol == htons(ETH_P_IP)) {
1951 struct iphdr *iph = ip_hdr(skb);
1954 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
1958 } else if (skb_is_gso_v6(skb)) {
1959 ipv6_hdr(skb)->payload_len = 0;
1960 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
1961 &ipv6_hdr(skb)->daddr,
1965 i = tx_ring->next_to_use;
1967 buffer_info = &tx_ring->buffer_info[i];
1968 context_desc = IGBVF_TX_CTXTDESC_ADV(*tx_ring, i);
1969 /* VLAN MACLEN IPLEN */
1970 if (tx_flags & IGBVF_TX_FLAGS_VLAN)
1971 info |= (tx_flags & IGBVF_TX_FLAGS_VLAN_MASK);
1972 info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
1973 *hdr_len += skb_network_offset(skb);
1974 info |= (skb_transport_header(skb) - skb_network_header(skb));
1975 *hdr_len += (skb_transport_header(skb) - skb_network_header(skb));
1976 context_desc->vlan_macip_lens = cpu_to_le32(info);
1978 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
1979 tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
1981 if (skb->protocol == htons(ETH_P_IP))
1982 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
1983 tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
1985 context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
1988 mss_l4len_idx = (skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT);
1989 mss_l4len_idx |= (l4len << E1000_ADVTXD_L4LEN_SHIFT);
1991 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
1992 context_desc->seqnum_seed = 0;
1994 buffer_info->time_stamp = jiffies;
1995 buffer_info->next_to_watch = i;
1996 buffer_info->dma = 0;
1998 if (i == tx_ring->count)
2001 tx_ring->next_to_use = i;
2006 static inline bool igbvf_tx_csum(struct igbvf_adapter *adapter,
2007 struct igbvf_ring *tx_ring,
2008 struct sk_buff *skb, u32 tx_flags)
2010 struct e1000_adv_tx_context_desc *context_desc;
2012 struct igbvf_buffer *buffer_info;
2013 u32 info = 0, tu_cmd = 0;
2015 if ((skb->ip_summed == CHECKSUM_PARTIAL) ||
2016 (tx_flags & IGBVF_TX_FLAGS_VLAN)) {
2017 i = tx_ring->next_to_use;
2018 buffer_info = &tx_ring->buffer_info[i];
2019 context_desc = IGBVF_TX_CTXTDESC_ADV(*tx_ring, i);
2021 if (tx_flags & IGBVF_TX_FLAGS_VLAN)
2022 info |= (tx_flags & IGBVF_TX_FLAGS_VLAN_MASK);
2024 info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
2025 if (skb->ip_summed == CHECKSUM_PARTIAL)
2026 info |= (skb_transport_header(skb) -
2027 skb_network_header(skb));
2030 context_desc->vlan_macip_lens = cpu_to_le32(info);
2032 tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
2034 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2035 switch (skb->protocol) {
2036 case __constant_htons(ETH_P_IP):
2037 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
2038 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
2039 tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
2041 case __constant_htons(ETH_P_IPV6):
2042 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
2043 tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
2050 context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
2051 context_desc->seqnum_seed = 0;
2052 context_desc->mss_l4len_idx = 0;
2054 buffer_info->time_stamp = jiffies;
2055 buffer_info->next_to_watch = i;
2056 buffer_info->dma = 0;
2058 if (i == tx_ring->count)
2060 tx_ring->next_to_use = i;
2068 static int igbvf_maybe_stop_tx(struct net_device *netdev, int size)
2070 struct igbvf_adapter *adapter = netdev_priv(netdev);
2072 /* there is enough descriptors then we don't need to worry */
2073 if (igbvf_desc_unused(adapter->tx_ring) >= size)
2076 netif_stop_queue(netdev);
2080 /* We need to check again just in case room has been made available */
2081 if (igbvf_desc_unused(adapter->tx_ring) < size)
2084 netif_wake_queue(netdev);
2086 ++adapter->restart_queue;
2090 #define IGBVF_MAX_TXD_PWR 16
2091 #define IGBVF_MAX_DATA_PER_TXD (1 << IGBVF_MAX_TXD_PWR)
2093 static inline int igbvf_tx_map_adv(struct igbvf_adapter *adapter,
2094 struct igbvf_ring *tx_ring,
2095 struct sk_buff *skb,
2098 struct igbvf_buffer *buffer_info;
2099 struct pci_dev *pdev = adapter->pdev;
2100 unsigned int len = skb_headlen(skb);
2101 unsigned int count = 0, i;
2104 i = tx_ring->next_to_use;
2106 buffer_info = &tx_ring->buffer_info[i];
2107 BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD);
2108 buffer_info->length = len;
2109 /* set time_stamp *before* dma to help avoid a possible race */
2110 buffer_info->time_stamp = jiffies;
2111 buffer_info->next_to_watch = i;
2112 buffer_info->mapped_as_page = false;
2113 buffer_info->dma = dma_map_single(&pdev->dev, skb->data, len,
2115 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2119 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) {
2120 struct skb_frag_struct *frag;
2124 if (i == tx_ring->count)
2127 frag = &skb_shinfo(skb)->frags[f];
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->next_to_watch = i;
2135 buffer_info->mapped_as_page = true;
2136 buffer_info->dma = dma_map_page(&pdev->dev,
2141 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2145 tx_ring->buffer_info[i].skb = skb;
2146 tx_ring->buffer_info[first].next_to_watch = i;
2151 dev_err(&pdev->dev, "TX DMA map failed\n");
2153 /* clear timestamp and dma mappings for failed buffer_info mapping */
2154 buffer_info->dma = 0;
2155 buffer_info->time_stamp = 0;
2156 buffer_info->length = 0;
2157 buffer_info->next_to_watch = 0;
2158 buffer_info->mapped_as_page = false;
2162 /* clear timestamp and dma mappings for remaining portion of packet */
2165 i += tx_ring->count;
2167 buffer_info = &tx_ring->buffer_info[i];
2168 igbvf_put_txbuf(adapter, buffer_info);
2174 static inline void igbvf_tx_queue_adv(struct igbvf_adapter *adapter,
2175 struct igbvf_ring *tx_ring,
2176 int tx_flags, int count, u32 paylen,
2179 union e1000_adv_tx_desc *tx_desc = NULL;
2180 struct igbvf_buffer *buffer_info;
2181 u32 olinfo_status = 0, cmd_type_len;
2184 cmd_type_len = (E1000_ADVTXD_DTYP_DATA | E1000_ADVTXD_DCMD_IFCS |
2185 E1000_ADVTXD_DCMD_DEXT);
2187 if (tx_flags & IGBVF_TX_FLAGS_VLAN)
2188 cmd_type_len |= E1000_ADVTXD_DCMD_VLE;
2190 if (tx_flags & IGBVF_TX_FLAGS_TSO) {
2191 cmd_type_len |= E1000_ADVTXD_DCMD_TSE;
2193 /* insert tcp checksum */
2194 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2196 /* insert ip checksum */
2197 if (tx_flags & IGBVF_TX_FLAGS_IPV4)
2198 olinfo_status |= E1000_TXD_POPTS_IXSM << 8;
2200 } else if (tx_flags & IGBVF_TX_FLAGS_CSUM) {
2201 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2204 olinfo_status |= ((paylen - hdr_len) << E1000_ADVTXD_PAYLEN_SHIFT);
2206 i = tx_ring->next_to_use;
2208 buffer_info = &tx_ring->buffer_info[i];
2209 tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
2210 tx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
2211 tx_desc->read.cmd_type_len =
2212 cpu_to_le32(cmd_type_len | buffer_info->length);
2213 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
2215 if (i == tx_ring->count)
2219 tx_desc->read.cmd_type_len |= cpu_to_le32(adapter->txd_cmd);
2220 /* Force memory writes to complete before letting h/w
2221 * know there are new descriptors to fetch. (Only
2222 * applicable for weak-ordered memory model archs,
2223 * such as IA-64). */
2226 tx_ring->next_to_use = i;
2227 writel(i, adapter->hw.hw_addr + tx_ring->tail);
2228 /* we need this if more than one processor can write to our tail
2229 * at a time, it syncronizes IO on IA64/Altix systems */
2233 static netdev_tx_t igbvf_xmit_frame_ring_adv(struct sk_buff *skb,
2234 struct net_device *netdev,
2235 struct igbvf_ring *tx_ring)
2237 struct igbvf_adapter *adapter = netdev_priv(netdev);
2238 unsigned int first, tx_flags = 0;
2243 if (test_bit(__IGBVF_DOWN, &adapter->state)) {
2244 dev_kfree_skb_any(skb);
2245 return NETDEV_TX_OK;
2248 if (skb->len <= 0) {
2249 dev_kfree_skb_any(skb);
2250 return NETDEV_TX_OK;
2254 * need: count + 4 desc gap to keep tail from touching
2255 * + 2 desc gap to keep tail from touching head,
2256 * + 1 desc for skb->data,
2257 * + 1 desc for context descriptor,
2258 * head, otherwise try next time
2260 if (igbvf_maybe_stop_tx(netdev, skb_shinfo(skb)->nr_frags + 4)) {
2261 /* this is a hard error */
2262 return NETDEV_TX_BUSY;
2265 if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
2266 tx_flags |= IGBVF_TX_FLAGS_VLAN;
2267 tx_flags |= (vlan_tx_tag_get(skb) << IGBVF_TX_FLAGS_VLAN_SHIFT);
2270 if (skb->protocol == htons(ETH_P_IP))
2271 tx_flags |= IGBVF_TX_FLAGS_IPV4;
2273 first = tx_ring->next_to_use;
2275 tso = skb_is_gso(skb) ?
2276 igbvf_tso(adapter, tx_ring, skb, tx_flags, &hdr_len) : 0;
2277 if (unlikely(tso < 0)) {
2278 dev_kfree_skb_any(skb);
2279 return NETDEV_TX_OK;
2283 tx_flags |= IGBVF_TX_FLAGS_TSO;
2284 else if (igbvf_tx_csum(adapter, tx_ring, skb, tx_flags) &&
2285 (skb->ip_summed == CHECKSUM_PARTIAL))
2286 tx_flags |= IGBVF_TX_FLAGS_CSUM;
2289 * count reflects descriptors mapped, if 0 then mapping error
2290 * has occured and we need to rewind the descriptor queue
2292 count = igbvf_tx_map_adv(adapter, tx_ring, skb, first);
2295 igbvf_tx_queue_adv(adapter, tx_ring, tx_flags, count,
2297 /* Make sure there is space in the ring for the next send. */
2298 igbvf_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 4);
2300 dev_kfree_skb_any(skb);
2301 tx_ring->buffer_info[first].time_stamp = 0;
2302 tx_ring->next_to_use = first;
2305 return NETDEV_TX_OK;
2308 static netdev_tx_t igbvf_xmit_frame(struct sk_buff *skb,
2309 struct net_device *netdev)
2311 struct igbvf_adapter *adapter = netdev_priv(netdev);
2312 struct igbvf_ring *tx_ring;
2314 if (test_bit(__IGBVF_DOWN, &adapter->state)) {
2315 dev_kfree_skb_any(skb);
2316 return NETDEV_TX_OK;
2319 tx_ring = &adapter->tx_ring[0];
2321 return igbvf_xmit_frame_ring_adv(skb, netdev, tx_ring);
2325 * igbvf_tx_timeout - Respond to a Tx Hang
2326 * @netdev: network interface device structure
2328 static void igbvf_tx_timeout(struct net_device *netdev)
2330 struct igbvf_adapter *adapter = netdev_priv(netdev);
2332 /* Do the reset outside of interrupt context */
2333 adapter->tx_timeout_count++;
2334 schedule_work(&adapter->reset_task);
2337 static void igbvf_reset_task(struct work_struct *work)
2339 struct igbvf_adapter *adapter;
2340 adapter = container_of(work, struct igbvf_adapter, reset_task);
2342 igbvf_reinit_locked(adapter);
2346 * igbvf_get_stats - Get System Network Statistics
2347 * @netdev: network interface device structure
2349 * Returns the address of the device statistics structure.
2350 * The statistics are actually updated from the timer callback.
2352 static struct net_device_stats *igbvf_get_stats(struct net_device *netdev)
2354 struct igbvf_adapter *adapter = netdev_priv(netdev);
2356 /* only return the current stats */
2357 return &adapter->net_stats;
2361 * igbvf_change_mtu - Change the Maximum Transfer Unit
2362 * @netdev: network interface device structure
2363 * @new_mtu: new value for maximum frame size
2365 * Returns 0 on success, negative on failure
2367 static int igbvf_change_mtu(struct net_device *netdev, int new_mtu)
2369 struct igbvf_adapter *adapter = netdev_priv(netdev);
2370 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
2372 if ((new_mtu < 68) || (max_frame > MAX_JUMBO_FRAME_SIZE)) {
2373 dev_err(&adapter->pdev->dev, "Invalid MTU setting\n");
2377 #define MAX_STD_JUMBO_FRAME_SIZE 9234
2378 if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
2379 dev_err(&adapter->pdev->dev, "MTU > 9216 not supported.\n");
2383 while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
2385 /* igbvf_down has a dependency on max_frame_size */
2386 adapter->max_frame_size = max_frame;
2387 if (netif_running(netdev))
2388 igbvf_down(adapter);
2391 * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
2392 * means we reserve 2 more, this pushes us to allocate from the next
2394 * i.e. RXBUFFER_2048 --> size-4096 slab
2395 * However with the new *_jumbo_rx* routines, jumbo receives will use
2399 if (max_frame <= 1024)
2400 adapter->rx_buffer_len = 1024;
2401 else if (max_frame <= 2048)
2402 adapter->rx_buffer_len = 2048;
2404 #if (PAGE_SIZE / 2) > 16384
2405 adapter->rx_buffer_len = 16384;
2407 adapter->rx_buffer_len = PAGE_SIZE / 2;
2411 /* adjust allocation if LPE protects us, and we aren't using SBP */
2412 if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
2413 (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
2414 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN +
2417 dev_info(&adapter->pdev->dev, "changing MTU from %d to %d\n",
2418 netdev->mtu, new_mtu);
2419 netdev->mtu = new_mtu;
2421 if (netif_running(netdev))
2424 igbvf_reset(adapter);
2426 clear_bit(__IGBVF_RESETTING, &adapter->state);
2431 static int igbvf_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
2439 static int igbvf_suspend(struct pci_dev *pdev, pm_message_t state)
2441 struct net_device *netdev = pci_get_drvdata(pdev);
2442 struct igbvf_adapter *adapter = netdev_priv(netdev);
2447 netif_device_detach(netdev);
2449 if (netif_running(netdev)) {
2450 WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
2451 igbvf_down(adapter);
2452 igbvf_free_irq(adapter);
2456 retval = pci_save_state(pdev);
2461 pci_disable_device(pdev);
2467 static int igbvf_resume(struct pci_dev *pdev)
2469 struct net_device *netdev = pci_get_drvdata(pdev);
2470 struct igbvf_adapter *adapter = netdev_priv(netdev);
2473 pci_restore_state(pdev);
2474 err = pci_enable_device_mem(pdev);
2476 dev_err(&pdev->dev, "Cannot enable PCI device from suspend\n");
2480 pci_set_master(pdev);
2482 if (netif_running(netdev)) {
2483 err = igbvf_request_irq(adapter);
2488 igbvf_reset(adapter);
2490 if (netif_running(netdev))
2493 netif_device_attach(netdev);
2499 static void igbvf_shutdown(struct pci_dev *pdev)
2501 igbvf_suspend(pdev, PMSG_SUSPEND);
2504 #ifdef CONFIG_NET_POLL_CONTROLLER
2506 * Polling 'interrupt' - used by things like netconsole to send skbs
2507 * without having to re-enable interrupts. It's not called while
2508 * the interrupt routine is executing.
2510 static void igbvf_netpoll(struct net_device *netdev)
2512 struct igbvf_adapter *adapter = netdev_priv(netdev);
2514 disable_irq(adapter->pdev->irq);
2516 igbvf_clean_tx_irq(adapter->tx_ring);
2518 enable_irq(adapter->pdev->irq);
2523 * igbvf_io_error_detected - called when PCI error is detected
2524 * @pdev: Pointer to PCI device
2525 * @state: The current pci connection state
2527 * This function is called after a PCI bus error affecting
2528 * this device has been detected.
2530 static pci_ers_result_t igbvf_io_error_detected(struct pci_dev *pdev,
2531 pci_channel_state_t state)
2533 struct net_device *netdev = pci_get_drvdata(pdev);
2534 struct igbvf_adapter *adapter = netdev_priv(netdev);
2536 netif_device_detach(netdev);
2538 if (state == pci_channel_io_perm_failure)
2539 return PCI_ERS_RESULT_DISCONNECT;
2541 if (netif_running(netdev))
2542 igbvf_down(adapter);
2543 pci_disable_device(pdev);
2545 /* Request a slot slot reset. */
2546 return PCI_ERS_RESULT_NEED_RESET;
2550 * igbvf_io_slot_reset - called after the pci bus has been reset.
2551 * @pdev: Pointer to PCI device
2553 * Restart the card from scratch, as if from a cold-boot. Implementation
2554 * resembles the first-half of the igbvf_resume routine.
2556 static pci_ers_result_t igbvf_io_slot_reset(struct pci_dev *pdev)
2558 struct net_device *netdev = pci_get_drvdata(pdev);
2559 struct igbvf_adapter *adapter = netdev_priv(netdev);
2561 if (pci_enable_device_mem(pdev)) {
2563 "Cannot re-enable PCI device after reset.\n");
2564 return PCI_ERS_RESULT_DISCONNECT;
2566 pci_set_master(pdev);
2568 igbvf_reset(adapter);
2570 return PCI_ERS_RESULT_RECOVERED;
2574 * igbvf_io_resume - called when traffic can start flowing again.
2575 * @pdev: Pointer to PCI device
2577 * This callback is called when the error recovery driver tells us that
2578 * its OK to resume normal operation. Implementation resembles the
2579 * second-half of the igbvf_resume routine.
2581 static void igbvf_io_resume(struct pci_dev *pdev)
2583 struct net_device *netdev = pci_get_drvdata(pdev);
2584 struct igbvf_adapter *adapter = netdev_priv(netdev);
2586 if (netif_running(netdev)) {
2587 if (igbvf_up(adapter)) {
2589 "can't bring device back up after reset\n");
2594 netif_device_attach(netdev);
2597 static void igbvf_print_device_info(struct igbvf_adapter *adapter)
2599 struct e1000_hw *hw = &adapter->hw;
2600 struct net_device *netdev = adapter->netdev;
2601 struct pci_dev *pdev = adapter->pdev;
2603 dev_info(&pdev->dev, "Intel(R) 82576 Virtual Function\n");
2604 dev_info(&pdev->dev, "Address: %pM\n", netdev->dev_addr);
2605 dev_info(&pdev->dev, "MAC: %d\n", hw->mac.type);
2608 static const struct net_device_ops igbvf_netdev_ops = {
2609 .ndo_open = igbvf_open,
2610 .ndo_stop = igbvf_close,
2611 .ndo_start_xmit = igbvf_xmit_frame,
2612 .ndo_get_stats = igbvf_get_stats,
2613 .ndo_set_multicast_list = igbvf_set_multi,
2614 .ndo_set_mac_address = igbvf_set_mac,
2615 .ndo_change_mtu = igbvf_change_mtu,
2616 .ndo_do_ioctl = igbvf_ioctl,
2617 .ndo_tx_timeout = igbvf_tx_timeout,
2618 .ndo_vlan_rx_register = igbvf_vlan_rx_register,
2619 .ndo_vlan_rx_add_vid = igbvf_vlan_rx_add_vid,
2620 .ndo_vlan_rx_kill_vid = igbvf_vlan_rx_kill_vid,
2621 #ifdef CONFIG_NET_POLL_CONTROLLER
2622 .ndo_poll_controller = igbvf_netpoll,
2627 * igbvf_probe - Device Initialization Routine
2628 * @pdev: PCI device information struct
2629 * @ent: entry in igbvf_pci_tbl
2631 * Returns 0 on success, negative on failure
2633 * igbvf_probe initializes an adapter identified by a pci_dev structure.
2634 * The OS initialization, configuring of the adapter private structure,
2635 * and a hardware reset occur.
2637 static int __devinit igbvf_probe(struct pci_dev *pdev,
2638 const struct pci_device_id *ent)
2640 struct net_device *netdev;
2641 struct igbvf_adapter *adapter;
2642 struct e1000_hw *hw;
2643 const struct igbvf_info *ei = igbvf_info_tbl[ent->driver_data];
2645 static int cards_found;
2646 int err, pci_using_dac;
2648 err = pci_enable_device_mem(pdev);
2653 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(64));
2655 err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64));
2659 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
2661 err = dma_set_coherent_mask(&pdev->dev,
2664 dev_err(&pdev->dev, "No usable DMA "
2665 "configuration, aborting\n");
2671 err = pci_request_regions(pdev, igbvf_driver_name);
2675 pci_set_master(pdev);
2678 netdev = alloc_etherdev(sizeof(struct igbvf_adapter));
2680 goto err_alloc_etherdev;
2682 SET_NETDEV_DEV(netdev, &pdev->dev);
2684 pci_set_drvdata(pdev, netdev);
2685 adapter = netdev_priv(netdev);
2687 adapter->netdev = netdev;
2688 adapter->pdev = pdev;
2690 adapter->pba = ei->pba;
2691 adapter->flags = ei->flags;
2692 adapter->hw.back = adapter;
2693 adapter->hw.mac.type = ei->mac;
2694 adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
2696 /* PCI config space info */
2698 hw->vendor_id = pdev->vendor;
2699 hw->device_id = pdev->device;
2700 hw->subsystem_vendor_id = pdev->subsystem_vendor;
2701 hw->subsystem_device_id = pdev->subsystem_device;
2703 pci_read_config_byte(pdev, PCI_REVISION_ID, &hw->revision_id);
2706 adapter->hw.hw_addr = ioremap(pci_resource_start(pdev, 0),
2707 pci_resource_len(pdev, 0));
2709 if (!adapter->hw.hw_addr)
2712 if (ei->get_variants) {
2713 err = ei->get_variants(adapter);
2718 /* setup adapter struct */
2719 err = igbvf_sw_init(adapter);
2723 /* construct the net_device struct */
2724 netdev->netdev_ops = &igbvf_netdev_ops;
2726 igbvf_set_ethtool_ops(netdev);
2727 netdev->watchdog_timeo = 5 * HZ;
2728 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
2730 adapter->bd_number = cards_found++;
2732 netdev->features = NETIF_F_SG |
2734 NETIF_F_HW_VLAN_TX |
2735 NETIF_F_HW_VLAN_RX |
2736 NETIF_F_HW_VLAN_FILTER;
2738 netdev->features |= NETIF_F_IPV6_CSUM;
2739 netdev->features |= NETIF_F_TSO;
2740 netdev->features |= NETIF_F_TSO6;
2743 netdev->features |= NETIF_F_HIGHDMA;
2745 netdev->vlan_features |= NETIF_F_TSO;
2746 netdev->vlan_features |= NETIF_F_TSO6;
2747 netdev->vlan_features |= NETIF_F_IP_CSUM;
2748 netdev->vlan_features |= NETIF_F_IPV6_CSUM;
2749 netdev->vlan_features |= NETIF_F_SG;
2751 /*reset the controller to put the device in a known good state */
2752 err = hw->mac.ops.reset_hw(hw);
2754 dev_info(&pdev->dev,
2755 "PF still in reset state, assigning new address."
2756 " Is the PF interface up?\n");
2757 dev_hw_addr_random(adapter->netdev, hw->mac.addr);
2759 err = hw->mac.ops.read_mac_addr(hw);
2761 dev_err(&pdev->dev, "Error reading MAC address\n");
2766 memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
2767 memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
2769 if (!is_valid_ether_addr(netdev->perm_addr)) {
2770 dev_err(&pdev->dev, "Invalid MAC Address: %pM\n",
2776 setup_timer(&adapter->watchdog_timer, &igbvf_watchdog,
2777 (unsigned long) adapter);
2779 INIT_WORK(&adapter->reset_task, igbvf_reset_task);
2780 INIT_WORK(&adapter->watchdog_task, igbvf_watchdog_task);
2782 /* ring size defaults */
2783 adapter->rx_ring->count = 1024;
2784 adapter->tx_ring->count = 1024;
2786 /* reset the hardware with the new settings */
2787 igbvf_reset(adapter);
2789 strcpy(netdev->name, "eth%d");
2790 err = register_netdev(netdev);
2794 /* tell the stack to leave us alone until igbvf_open() is called */
2795 netif_carrier_off(netdev);
2796 netif_stop_queue(netdev);
2798 igbvf_print_device_info(adapter);
2800 igbvf_initialize_last_counter_stats(adapter);
2805 kfree(adapter->tx_ring);
2806 kfree(adapter->rx_ring);
2808 igbvf_reset_interrupt_capability(adapter);
2809 iounmap(adapter->hw.hw_addr);
2811 free_netdev(netdev);
2813 pci_release_regions(pdev);
2816 pci_disable_device(pdev);
2821 * igbvf_remove - Device Removal Routine
2822 * @pdev: PCI device information struct
2824 * igbvf_remove is called by the PCI subsystem to alert the driver
2825 * that it should release a PCI device. The could be caused by a
2826 * Hot-Plug event, or because the driver is going to be removed from
2829 static void __devexit igbvf_remove(struct pci_dev *pdev)
2831 struct net_device *netdev = pci_get_drvdata(pdev);
2832 struct igbvf_adapter *adapter = netdev_priv(netdev);
2833 struct e1000_hw *hw = &adapter->hw;
2836 * The watchdog timer may be rescheduled, so explicitly
2837 * disable it from being rescheduled.
2839 set_bit(__IGBVF_DOWN, &adapter->state);
2840 del_timer_sync(&adapter->watchdog_timer);
2842 cancel_work_sync(&adapter->reset_task);
2843 cancel_work_sync(&adapter->watchdog_task);
2845 unregister_netdev(netdev);
2847 igbvf_reset_interrupt_capability(adapter);
2850 * it is important to delete the napi struct prior to freeing the
2851 * rx ring so that you do not end up with null pointer refs
2853 netif_napi_del(&adapter->rx_ring->napi);
2854 kfree(adapter->tx_ring);
2855 kfree(adapter->rx_ring);
2857 iounmap(hw->hw_addr);
2858 if (hw->flash_address)
2859 iounmap(hw->flash_address);
2860 pci_release_regions(pdev);
2862 free_netdev(netdev);
2864 pci_disable_device(pdev);
2867 /* PCI Error Recovery (ERS) */
2868 static struct pci_error_handlers igbvf_err_handler = {
2869 .error_detected = igbvf_io_error_detected,
2870 .slot_reset = igbvf_io_slot_reset,
2871 .resume = igbvf_io_resume,
2874 static DEFINE_PCI_DEVICE_TABLE(igbvf_pci_tbl) = {
2875 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_VF), board_vf },
2876 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_VF), board_i350_vf },
2877 { } /* terminate list */
2879 MODULE_DEVICE_TABLE(pci, igbvf_pci_tbl);
2881 /* PCI Device API Driver */
2882 static struct pci_driver igbvf_driver = {
2883 .name = igbvf_driver_name,
2884 .id_table = igbvf_pci_tbl,
2885 .probe = igbvf_probe,
2886 .remove = __devexit_p(igbvf_remove),
2888 /* Power Management Hooks */
2889 .suspend = igbvf_suspend,
2890 .resume = igbvf_resume,
2892 .shutdown = igbvf_shutdown,
2893 .err_handler = &igbvf_err_handler
2897 * igbvf_init_module - Driver Registration Routine
2899 * igbvf_init_module is the first routine called when the driver is
2900 * loaded. All it does is register with the PCI subsystem.
2902 static int __init igbvf_init_module(void)
2905 printk(KERN_INFO "%s - version %s\n",
2906 igbvf_driver_string, igbvf_driver_version);
2907 printk(KERN_INFO "%s\n", igbvf_copyright);
2909 ret = pci_register_driver(&igbvf_driver);
2913 module_init(igbvf_init_module);
2916 * igbvf_exit_module - Driver Exit Cleanup Routine
2918 * igbvf_exit_module is called just before the driver is removed
2921 static void __exit igbvf_exit_module(void)
2923 pci_unregister_driver(&igbvf_driver);
2925 module_exit(igbvf_exit_module);
2928 MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
2929 MODULE_DESCRIPTION("Intel(R) 82576 Virtual Function Network Driver");
2930 MODULE_LICENSE("GPL");
2931 MODULE_VERSION(DRV_VERSION);