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>
44 #include <linux/prefetch.h>
48 #define DRV_VERSION "2.0.0-k"
49 char igbvf_driver_name[] = "igbvf";
50 const char igbvf_driver_version[] = DRV_VERSION;
51 static const char igbvf_driver_string[] =
52 "Intel(R) Virtual Function Network Driver";
53 static const char igbvf_copyright[] =
54 "Copyright (c) 2009 - 2010 Intel Corporation.";
56 static int igbvf_poll(struct napi_struct *napi, int budget);
57 static void igbvf_reset(struct igbvf_adapter *);
58 static void igbvf_set_interrupt_capability(struct igbvf_adapter *);
59 static void igbvf_reset_interrupt_capability(struct igbvf_adapter *);
61 static struct igbvf_info igbvf_vf_info = {
65 .init_ops = e1000_init_function_pointers_vf,
68 static struct igbvf_info igbvf_i350_vf_info = {
69 .mac = e1000_vfadapt_i350,
72 .init_ops = e1000_init_function_pointers_vf,
75 static const struct igbvf_info *igbvf_info_tbl[] = {
76 [board_vf] = &igbvf_vf_info,
77 [board_i350_vf] = &igbvf_i350_vf_info,
81 * igbvf_desc_unused - calculate if we have unused descriptors
83 static int igbvf_desc_unused(struct igbvf_ring *ring)
85 if (ring->next_to_clean > ring->next_to_use)
86 return ring->next_to_clean - ring->next_to_use - 1;
88 return ring->count + ring->next_to_clean - ring->next_to_use - 1;
92 * igbvf_receive_skb - helper function to handle Rx indications
93 * @adapter: board private structure
94 * @status: descriptor status field as written by hardware
95 * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
96 * @skb: pointer to sk_buff to be indicated to stack
98 static void igbvf_receive_skb(struct igbvf_adapter *adapter,
99 struct net_device *netdev,
101 u32 status, u16 vlan)
103 if (status & E1000_RXD_STAT_VP) {
104 u16 vid = le16_to_cpu(vlan) & E1000_RXD_SPC_VLAN_MASK;
106 __vlan_hwaccel_put_tag(skb, vid);
108 netif_receive_skb(skb);
111 static inline void igbvf_rx_checksum_adv(struct igbvf_adapter *adapter,
112 u32 status_err, struct sk_buff *skb)
114 skb_checksum_none_assert(skb);
116 /* Ignore Checksum bit is set or checksum is disabled through ethtool */
117 if ((status_err & E1000_RXD_STAT_IXSM) ||
118 (adapter->flags & IGBVF_FLAG_RX_CSUM_DISABLED))
121 /* TCP/UDP checksum error bit is set */
123 (E1000_RXDEXT_STATERR_TCPE | E1000_RXDEXT_STATERR_IPE)) {
124 /* let the stack verify checksum errors */
125 adapter->hw_csum_err++;
129 /* It must be a TCP or UDP packet with a valid checksum */
130 if (status_err & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS))
131 skb->ip_summed = CHECKSUM_UNNECESSARY;
133 adapter->hw_csum_good++;
137 * igbvf_alloc_rx_buffers - Replace used receive buffers; packet split
138 * @rx_ring: address of ring structure to repopulate
139 * @cleaned_count: number of buffers to repopulate
141 static void igbvf_alloc_rx_buffers(struct igbvf_ring *rx_ring,
144 struct igbvf_adapter *adapter = rx_ring->adapter;
145 struct net_device *netdev = adapter->netdev;
146 struct pci_dev *pdev = adapter->pdev;
147 union e1000_adv_rx_desc *rx_desc;
148 struct igbvf_buffer *buffer_info;
153 i = rx_ring->next_to_use;
154 buffer_info = &rx_ring->buffer_info[i];
156 if (adapter->rx_ps_hdr_size)
157 bufsz = adapter->rx_ps_hdr_size;
159 bufsz = adapter->rx_buffer_len;
161 while (cleaned_count--) {
162 rx_desc = IGBVF_RX_DESC_ADV(*rx_ring, i);
164 if (adapter->rx_ps_hdr_size && !buffer_info->page_dma) {
165 if (!buffer_info->page) {
166 buffer_info->page = alloc_page(GFP_ATOMIC);
167 if (!buffer_info->page) {
168 adapter->alloc_rx_buff_failed++;
171 buffer_info->page_offset = 0;
173 buffer_info->page_offset ^= PAGE_SIZE / 2;
175 buffer_info->page_dma =
176 dma_map_page(&pdev->dev, buffer_info->page,
177 buffer_info->page_offset,
182 if (!buffer_info->skb) {
183 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
185 adapter->alloc_rx_buff_failed++;
189 buffer_info->skb = skb;
190 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
194 /* Refresh the desc even if buffer_addrs didn't change because
195 * each write-back erases this info. */
196 if (adapter->rx_ps_hdr_size) {
197 rx_desc->read.pkt_addr =
198 cpu_to_le64(buffer_info->page_dma);
199 rx_desc->read.hdr_addr = cpu_to_le64(buffer_info->dma);
201 rx_desc->read.pkt_addr =
202 cpu_to_le64(buffer_info->dma);
203 rx_desc->read.hdr_addr = 0;
207 if (i == rx_ring->count)
209 buffer_info = &rx_ring->buffer_info[i];
213 if (rx_ring->next_to_use != i) {
214 rx_ring->next_to_use = i;
216 i = (rx_ring->count - 1);
220 /* Force memory writes to complete before letting h/w
221 * know there are new descriptors to fetch. (Only
222 * applicable for weak-ordered memory model archs,
225 writel(i, adapter->hw.hw_addr + rx_ring->tail);
230 * igbvf_clean_rx_irq - Send received data up the network stack; legacy
231 * @adapter: board private structure
233 * the return value indicates whether actual cleaning was done, there
234 * is no guarantee that everything was cleaned
236 static bool igbvf_clean_rx_irq(struct igbvf_adapter *adapter,
237 int *work_done, int work_to_do)
239 struct igbvf_ring *rx_ring = adapter->rx_ring;
240 struct net_device *netdev = adapter->netdev;
241 struct pci_dev *pdev = adapter->pdev;
242 union e1000_adv_rx_desc *rx_desc, *next_rxd;
243 struct igbvf_buffer *buffer_info, *next_buffer;
245 bool cleaned = false;
246 int cleaned_count = 0;
247 unsigned int total_bytes = 0, total_packets = 0;
249 u32 length, hlen, staterr;
251 i = rx_ring->next_to_clean;
252 rx_desc = IGBVF_RX_DESC_ADV(*rx_ring, i);
253 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
255 while (staterr & E1000_RXD_STAT_DD) {
256 if (*work_done >= work_to_do)
259 rmb(); /* read descriptor and rx_buffer_info after status DD */
261 buffer_info = &rx_ring->buffer_info[i];
263 /* HW will not DMA in data larger than the given buffer, even
264 * if it parses the (NFS, of course) header to be larger. In
265 * that case, it fills the header buffer and spills the rest
268 hlen = (le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.hdr_info) &
269 E1000_RXDADV_HDRBUFLEN_MASK) >> E1000_RXDADV_HDRBUFLEN_SHIFT;
270 if (hlen > adapter->rx_ps_hdr_size)
271 hlen = adapter->rx_ps_hdr_size;
273 length = le16_to_cpu(rx_desc->wb.upper.length);
277 skb = buffer_info->skb;
278 prefetch(skb->data - NET_IP_ALIGN);
279 buffer_info->skb = NULL;
280 if (!adapter->rx_ps_hdr_size) {
281 dma_unmap_single(&pdev->dev, buffer_info->dma,
282 adapter->rx_buffer_len,
284 buffer_info->dma = 0;
285 skb_put(skb, length);
289 if (!skb_shinfo(skb)->nr_frags) {
290 dma_unmap_single(&pdev->dev, buffer_info->dma,
291 adapter->rx_ps_hdr_size,
297 dma_unmap_page(&pdev->dev, buffer_info->page_dma,
300 buffer_info->page_dma = 0;
302 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
304 buffer_info->page_offset,
307 if ((adapter->rx_buffer_len > (PAGE_SIZE / 2)) ||
308 (page_count(buffer_info->page) != 1))
309 buffer_info->page = NULL;
311 get_page(buffer_info->page);
314 skb->data_len += length;
315 skb->truesize += length;
319 if (i == rx_ring->count)
321 next_rxd = IGBVF_RX_DESC_ADV(*rx_ring, i);
323 next_buffer = &rx_ring->buffer_info[i];
325 if (!(staterr & E1000_RXD_STAT_EOP)) {
326 buffer_info->skb = next_buffer->skb;
327 buffer_info->dma = next_buffer->dma;
328 next_buffer->skb = skb;
329 next_buffer->dma = 0;
333 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
334 dev_kfree_skb_irq(skb);
338 total_bytes += skb->len;
341 igbvf_rx_checksum_adv(adapter, staterr, skb);
343 skb->protocol = eth_type_trans(skb, netdev);
345 igbvf_receive_skb(adapter, netdev, skb, staterr,
346 rx_desc->wb.upper.vlan);
349 rx_desc->wb.upper.status_error = 0;
351 /* return some buffers to hardware, one at a time is too slow */
352 if (cleaned_count >= IGBVF_RX_BUFFER_WRITE) {
353 igbvf_alloc_rx_buffers(rx_ring, cleaned_count);
357 /* use prefetched values */
359 buffer_info = next_buffer;
361 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
364 rx_ring->next_to_clean = i;
365 cleaned_count = igbvf_desc_unused(rx_ring);
368 igbvf_alloc_rx_buffers(rx_ring, cleaned_count);
370 adapter->total_rx_packets += total_packets;
371 adapter->total_rx_bytes += total_bytes;
372 adapter->net_stats.rx_bytes += total_bytes;
373 adapter->net_stats.rx_packets += total_packets;
377 static void igbvf_put_txbuf(struct igbvf_adapter *adapter,
378 struct igbvf_buffer *buffer_info)
380 if (buffer_info->dma) {
381 if (buffer_info->mapped_as_page)
382 dma_unmap_page(&adapter->pdev->dev,
387 dma_unmap_single(&adapter->pdev->dev,
391 buffer_info->dma = 0;
393 if (buffer_info->skb) {
394 dev_kfree_skb_any(buffer_info->skb);
395 buffer_info->skb = NULL;
397 buffer_info->time_stamp = 0;
401 * igbvf_setup_tx_resources - allocate Tx resources (Descriptors)
402 * @adapter: board private structure
404 * Return 0 on success, negative on failure
406 int igbvf_setup_tx_resources(struct igbvf_adapter *adapter,
407 struct igbvf_ring *tx_ring)
409 struct pci_dev *pdev = adapter->pdev;
412 size = sizeof(struct igbvf_buffer) * tx_ring->count;
413 tx_ring->buffer_info = vzalloc(size);
414 if (!tx_ring->buffer_info)
417 /* round up to nearest 4K */
418 tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
419 tx_ring->size = ALIGN(tx_ring->size, 4096);
421 tx_ring->desc = dma_alloc_coherent(&pdev->dev, tx_ring->size,
422 &tx_ring->dma, GFP_KERNEL);
427 tx_ring->adapter = adapter;
428 tx_ring->next_to_use = 0;
429 tx_ring->next_to_clean = 0;
433 vfree(tx_ring->buffer_info);
434 dev_err(&adapter->pdev->dev,
435 "Unable to allocate memory for the transmit descriptor ring\n");
440 * igbvf_setup_rx_resources - allocate Rx resources (Descriptors)
441 * @adapter: board private structure
443 * Returns 0 on success, negative on failure
445 int igbvf_setup_rx_resources(struct igbvf_adapter *adapter,
446 struct igbvf_ring *rx_ring)
448 struct pci_dev *pdev = adapter->pdev;
451 size = sizeof(struct igbvf_buffer) * rx_ring->count;
452 rx_ring->buffer_info = vzalloc(size);
453 if (!rx_ring->buffer_info)
456 desc_len = sizeof(union e1000_adv_rx_desc);
458 /* Round up to nearest 4K */
459 rx_ring->size = rx_ring->count * desc_len;
460 rx_ring->size = ALIGN(rx_ring->size, 4096);
462 rx_ring->desc = dma_alloc_coherent(&pdev->dev, rx_ring->size,
463 &rx_ring->dma, GFP_KERNEL);
468 rx_ring->next_to_clean = 0;
469 rx_ring->next_to_use = 0;
471 rx_ring->adapter = adapter;
476 vfree(rx_ring->buffer_info);
477 rx_ring->buffer_info = NULL;
478 dev_err(&adapter->pdev->dev,
479 "Unable to allocate memory for the receive descriptor ring\n");
484 * igbvf_clean_tx_ring - Free Tx Buffers
485 * @tx_ring: ring to be cleaned
487 static void igbvf_clean_tx_ring(struct igbvf_ring *tx_ring)
489 struct igbvf_adapter *adapter = tx_ring->adapter;
490 struct igbvf_buffer *buffer_info;
494 if (!tx_ring->buffer_info)
497 /* Free all the Tx ring sk_buffs */
498 for (i = 0; i < tx_ring->count; i++) {
499 buffer_info = &tx_ring->buffer_info[i];
500 igbvf_put_txbuf(adapter, buffer_info);
503 size = sizeof(struct igbvf_buffer) * tx_ring->count;
504 memset(tx_ring->buffer_info, 0, size);
506 /* Zero out the descriptor ring */
507 memset(tx_ring->desc, 0, tx_ring->size);
509 tx_ring->next_to_use = 0;
510 tx_ring->next_to_clean = 0;
512 writel(0, adapter->hw.hw_addr + tx_ring->head);
513 writel(0, adapter->hw.hw_addr + tx_ring->tail);
517 * igbvf_free_tx_resources - Free Tx Resources per Queue
518 * @tx_ring: ring to free resources from
520 * Free all transmit software resources
522 void igbvf_free_tx_resources(struct igbvf_ring *tx_ring)
524 struct pci_dev *pdev = tx_ring->adapter->pdev;
526 igbvf_clean_tx_ring(tx_ring);
528 vfree(tx_ring->buffer_info);
529 tx_ring->buffer_info = NULL;
531 dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
534 tx_ring->desc = NULL;
538 * igbvf_clean_rx_ring - Free Rx Buffers per Queue
539 * @adapter: board private structure
541 static void igbvf_clean_rx_ring(struct igbvf_ring *rx_ring)
543 struct igbvf_adapter *adapter = rx_ring->adapter;
544 struct igbvf_buffer *buffer_info;
545 struct pci_dev *pdev = adapter->pdev;
549 if (!rx_ring->buffer_info)
552 /* Free all the Rx ring sk_buffs */
553 for (i = 0; i < rx_ring->count; i++) {
554 buffer_info = &rx_ring->buffer_info[i];
555 if (buffer_info->dma) {
556 if (adapter->rx_ps_hdr_size){
557 dma_unmap_single(&pdev->dev, buffer_info->dma,
558 adapter->rx_ps_hdr_size,
561 dma_unmap_single(&pdev->dev, buffer_info->dma,
562 adapter->rx_buffer_len,
565 buffer_info->dma = 0;
568 if (buffer_info->skb) {
569 dev_kfree_skb(buffer_info->skb);
570 buffer_info->skb = NULL;
573 if (buffer_info->page) {
574 if (buffer_info->page_dma)
575 dma_unmap_page(&pdev->dev,
576 buffer_info->page_dma,
579 put_page(buffer_info->page);
580 buffer_info->page = NULL;
581 buffer_info->page_dma = 0;
582 buffer_info->page_offset = 0;
586 size = sizeof(struct igbvf_buffer) * rx_ring->count;
587 memset(rx_ring->buffer_info, 0, size);
589 /* Zero out the descriptor ring */
590 memset(rx_ring->desc, 0, rx_ring->size);
592 rx_ring->next_to_clean = 0;
593 rx_ring->next_to_use = 0;
595 writel(0, adapter->hw.hw_addr + rx_ring->head);
596 writel(0, adapter->hw.hw_addr + rx_ring->tail);
600 * igbvf_free_rx_resources - Free Rx Resources
601 * @rx_ring: ring to clean the resources from
603 * Free all receive software resources
606 void igbvf_free_rx_resources(struct igbvf_ring *rx_ring)
608 struct pci_dev *pdev = rx_ring->adapter->pdev;
610 igbvf_clean_rx_ring(rx_ring);
612 vfree(rx_ring->buffer_info);
613 rx_ring->buffer_info = NULL;
615 dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
617 rx_ring->desc = NULL;
621 * igbvf_update_itr - update the dynamic ITR value based on statistics
622 * @adapter: pointer to adapter
623 * @itr_setting: current adapter->itr
624 * @packets: the number of packets during this measurement interval
625 * @bytes: the number of bytes during this measurement interval
627 * Stores a new ITR value based on packets and byte
628 * counts during the last interrupt. The advantage of per interrupt
629 * computation is faster updates and more accurate ITR for the current
630 * traffic pattern. Constants in this function were computed
631 * based on theoretical maximum wire speed and thresholds were set based
632 * on testing data as well as attempting to minimize response time
633 * while increasing bulk throughput. This functionality is controlled
634 * by the InterruptThrottleRate module parameter.
636 static unsigned int igbvf_update_itr(struct igbvf_adapter *adapter,
637 u16 itr_setting, int packets,
640 unsigned int retval = itr_setting;
643 goto update_itr_done;
645 switch (itr_setting) {
647 /* handle TSO and jumbo frames */
648 if (bytes/packets > 8000)
649 retval = bulk_latency;
650 else if ((packets < 5) && (bytes > 512))
651 retval = low_latency;
653 case low_latency: /* 50 usec aka 20000 ints/s */
655 /* this if handles the TSO accounting */
656 if (bytes/packets > 8000)
657 retval = bulk_latency;
658 else if ((packets < 10) || ((bytes/packets) > 1200))
659 retval = bulk_latency;
660 else if ((packets > 35))
661 retval = lowest_latency;
662 } else if (bytes/packets > 2000) {
663 retval = bulk_latency;
664 } else if (packets <= 2 && bytes < 512) {
665 retval = lowest_latency;
668 case bulk_latency: /* 250 usec aka 4000 ints/s */
671 retval = low_latency;
672 } else if (bytes < 6000) {
673 retval = low_latency;
682 static void igbvf_set_itr(struct igbvf_adapter *adapter)
684 struct e1000_hw *hw = &adapter->hw;
686 u32 new_itr = adapter->itr;
688 adapter->tx_itr = igbvf_update_itr(adapter, adapter->tx_itr,
689 adapter->total_tx_packets,
690 adapter->total_tx_bytes);
691 /* conservative mode (itr 3) eliminates the lowest_latency setting */
692 if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
693 adapter->tx_itr = low_latency;
695 adapter->rx_itr = igbvf_update_itr(adapter, adapter->rx_itr,
696 adapter->total_rx_packets,
697 adapter->total_rx_bytes);
698 /* conservative mode (itr 3) eliminates the lowest_latency setting */
699 if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
700 adapter->rx_itr = low_latency;
702 current_itr = max(adapter->rx_itr, adapter->tx_itr);
704 switch (current_itr) {
705 /* counts and packets in update_itr are dependent on these numbers */
710 new_itr = 20000; /* aka hwitr = ~200 */
719 if (new_itr != adapter->itr) {
721 * this attempts to bias the interrupt rate towards Bulk
722 * by adding intermediate steps when interrupt rate is
725 new_itr = new_itr > adapter->itr ?
726 min(adapter->itr + (new_itr >> 2), new_itr) :
728 adapter->itr = new_itr;
729 adapter->rx_ring->itr_val = 1952;
731 if (adapter->msix_entries)
732 adapter->rx_ring->set_itr = 1;
739 * igbvf_clean_tx_irq - Reclaim resources after transmit completes
740 * @adapter: board private structure
741 * returns true if ring is completely cleaned
743 static bool igbvf_clean_tx_irq(struct igbvf_ring *tx_ring)
745 struct igbvf_adapter *adapter = tx_ring->adapter;
746 struct net_device *netdev = adapter->netdev;
747 struct igbvf_buffer *buffer_info;
749 union e1000_adv_tx_desc *tx_desc, *eop_desc;
750 unsigned int total_bytes = 0, total_packets = 0;
751 unsigned int i, eop, count = 0;
752 bool cleaned = false;
754 i = tx_ring->next_to_clean;
755 eop = tx_ring->buffer_info[i].next_to_watch;
756 eop_desc = IGBVF_TX_DESC_ADV(*tx_ring, eop);
758 while ((eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)) &&
759 (count < tx_ring->count)) {
760 rmb(); /* read buffer_info after eop_desc status */
761 for (cleaned = false; !cleaned; count++) {
762 tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
763 buffer_info = &tx_ring->buffer_info[i];
764 cleaned = (i == eop);
765 skb = buffer_info->skb;
768 unsigned int segs, bytecount;
770 /* gso_segs is currently only valid for tcp */
771 segs = skb_shinfo(skb)->gso_segs ?: 1;
772 /* multiply data chunks by size of headers */
773 bytecount = ((segs - 1) * skb_headlen(skb)) +
775 total_packets += segs;
776 total_bytes += bytecount;
779 igbvf_put_txbuf(adapter, buffer_info);
780 tx_desc->wb.status = 0;
783 if (i == tx_ring->count)
786 eop = tx_ring->buffer_info[i].next_to_watch;
787 eop_desc = IGBVF_TX_DESC_ADV(*tx_ring, eop);
790 tx_ring->next_to_clean = i;
792 if (unlikely(count &&
793 netif_carrier_ok(netdev) &&
794 igbvf_desc_unused(tx_ring) >= IGBVF_TX_QUEUE_WAKE)) {
795 /* Make sure that anybody stopping the queue after this
796 * sees the new next_to_clean.
799 if (netif_queue_stopped(netdev) &&
800 !(test_bit(__IGBVF_DOWN, &adapter->state))) {
801 netif_wake_queue(netdev);
802 ++adapter->restart_queue;
806 adapter->net_stats.tx_bytes += total_bytes;
807 adapter->net_stats.tx_packets += total_packets;
808 return count < tx_ring->count;
811 static irqreturn_t igbvf_msix_other(int irq, void *data)
813 struct net_device *netdev = data;
814 struct igbvf_adapter *adapter = netdev_priv(netdev);
815 struct e1000_hw *hw = &adapter->hw;
817 adapter->int_counter1++;
819 netif_carrier_off(netdev);
820 hw->mac.get_link_status = 1;
821 if (!test_bit(__IGBVF_DOWN, &adapter->state))
822 mod_timer(&adapter->watchdog_timer, jiffies + 1);
824 ew32(EIMS, adapter->eims_other);
829 static irqreturn_t igbvf_intr_msix_tx(int irq, void *data)
831 struct net_device *netdev = data;
832 struct igbvf_adapter *adapter = netdev_priv(netdev);
833 struct e1000_hw *hw = &adapter->hw;
834 struct igbvf_ring *tx_ring = adapter->tx_ring;
837 adapter->total_tx_bytes = 0;
838 adapter->total_tx_packets = 0;
840 /* auto mask will automatically reenable the interrupt when we write
842 if (!igbvf_clean_tx_irq(tx_ring))
843 /* Ring was not completely cleaned, so fire another interrupt */
844 ew32(EICS, tx_ring->eims_value);
846 ew32(EIMS, tx_ring->eims_value);
851 static irqreturn_t igbvf_intr_msix_rx(int irq, void *data)
853 struct net_device *netdev = data;
854 struct igbvf_adapter *adapter = netdev_priv(netdev);
856 adapter->int_counter0++;
858 /* Write the ITR value calculated at the end of the
859 * previous interrupt.
861 if (adapter->rx_ring->set_itr) {
862 writel(adapter->rx_ring->itr_val,
863 adapter->hw.hw_addr + adapter->rx_ring->itr_register);
864 adapter->rx_ring->set_itr = 0;
867 if (napi_schedule_prep(&adapter->rx_ring->napi)) {
868 adapter->total_rx_bytes = 0;
869 adapter->total_rx_packets = 0;
870 __napi_schedule(&adapter->rx_ring->napi);
876 #define IGBVF_NO_QUEUE -1
878 static void igbvf_assign_vector(struct igbvf_adapter *adapter, int rx_queue,
879 int tx_queue, int msix_vector)
881 struct e1000_hw *hw = &adapter->hw;
884 /* 82576 uses a table-based method for assigning vectors.
885 Each queue has a single entry in the table to which we write
886 a vector number along with a "valid" bit. Sadly, the layout
887 of the table is somewhat counterintuitive. */
888 if (rx_queue > IGBVF_NO_QUEUE) {
889 index = (rx_queue >> 1);
890 ivar = array_er32(IVAR0, index);
891 if (rx_queue & 0x1) {
892 /* vector goes into third byte of register */
893 ivar = ivar & 0xFF00FFFF;
894 ivar |= (msix_vector | E1000_IVAR_VALID) << 16;
896 /* vector goes into low byte of register */
897 ivar = ivar & 0xFFFFFF00;
898 ivar |= msix_vector | E1000_IVAR_VALID;
900 adapter->rx_ring[rx_queue].eims_value = 1 << msix_vector;
901 array_ew32(IVAR0, index, ivar);
903 if (tx_queue > IGBVF_NO_QUEUE) {
904 index = (tx_queue >> 1);
905 ivar = array_er32(IVAR0, index);
906 if (tx_queue & 0x1) {
907 /* vector goes into high byte of register */
908 ivar = ivar & 0x00FFFFFF;
909 ivar |= (msix_vector | E1000_IVAR_VALID) << 24;
911 /* vector goes into second byte of register */
912 ivar = ivar & 0xFFFF00FF;
913 ivar |= (msix_vector | E1000_IVAR_VALID) << 8;
915 adapter->tx_ring[tx_queue].eims_value = 1 << msix_vector;
916 array_ew32(IVAR0, index, ivar);
921 * igbvf_configure_msix - Configure MSI-X hardware
923 * igbvf_configure_msix sets up the hardware to properly
924 * generate MSI-X interrupts.
926 static void igbvf_configure_msix(struct igbvf_adapter *adapter)
929 struct e1000_hw *hw = &adapter->hw;
930 struct igbvf_ring *tx_ring = adapter->tx_ring;
931 struct igbvf_ring *rx_ring = adapter->rx_ring;
934 adapter->eims_enable_mask = 0;
936 igbvf_assign_vector(adapter, IGBVF_NO_QUEUE, 0, vector++);
937 adapter->eims_enable_mask |= tx_ring->eims_value;
938 if (tx_ring->itr_val)
939 writel(tx_ring->itr_val,
940 hw->hw_addr + tx_ring->itr_register);
942 writel(1952, hw->hw_addr + tx_ring->itr_register);
944 igbvf_assign_vector(adapter, 0, IGBVF_NO_QUEUE, vector++);
945 adapter->eims_enable_mask |= rx_ring->eims_value;
946 if (rx_ring->itr_val)
947 writel(rx_ring->itr_val,
948 hw->hw_addr + rx_ring->itr_register);
950 writel(1952, hw->hw_addr + rx_ring->itr_register);
952 /* set vector for other causes, i.e. link changes */
954 tmp = (vector++ | E1000_IVAR_VALID);
956 ew32(IVAR_MISC, tmp);
958 adapter->eims_enable_mask = (1 << (vector)) - 1;
959 adapter->eims_other = 1 << (vector - 1);
963 static void igbvf_reset_interrupt_capability(struct igbvf_adapter *adapter)
965 if (adapter->msix_entries) {
966 pci_disable_msix(adapter->pdev);
967 kfree(adapter->msix_entries);
968 adapter->msix_entries = NULL;
973 * igbvf_set_interrupt_capability - set MSI or MSI-X if supported
975 * Attempt to configure interrupts using the best available
976 * capabilities of the hardware and kernel.
978 static void igbvf_set_interrupt_capability(struct igbvf_adapter *adapter)
983 /* we allocate 3 vectors, 1 for tx, 1 for rx, one for pf messages */
984 adapter->msix_entries = kcalloc(3, sizeof(struct msix_entry),
986 if (adapter->msix_entries) {
987 for (i = 0; i < 3; i++)
988 adapter->msix_entries[i].entry = i;
990 err = pci_enable_msix(adapter->pdev,
991 adapter->msix_entries, 3);
996 dev_err(&adapter->pdev->dev,
997 "Failed to initialize MSI-X interrupts.\n");
998 igbvf_reset_interrupt_capability(adapter);
1003 * igbvf_request_msix - Initialize MSI-X interrupts
1005 * igbvf_request_msix allocates MSI-X vectors and requests interrupts from the
1008 static int igbvf_request_msix(struct igbvf_adapter *adapter)
1010 struct net_device *netdev = adapter->netdev;
1011 int err = 0, vector = 0;
1013 if (strlen(netdev->name) < (IFNAMSIZ - 5)) {
1014 sprintf(adapter->tx_ring->name, "%s-tx-0", netdev->name);
1015 sprintf(adapter->rx_ring->name, "%s-rx-0", netdev->name);
1017 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1018 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1021 err = request_irq(adapter->msix_entries[vector].vector,
1022 igbvf_intr_msix_tx, 0, adapter->tx_ring->name,
1027 adapter->tx_ring->itr_register = E1000_EITR(vector);
1028 adapter->tx_ring->itr_val = 1952;
1031 err = request_irq(adapter->msix_entries[vector].vector,
1032 igbvf_intr_msix_rx, 0, adapter->rx_ring->name,
1037 adapter->rx_ring->itr_register = E1000_EITR(vector);
1038 adapter->rx_ring->itr_val = 1952;
1041 err = request_irq(adapter->msix_entries[vector].vector,
1042 igbvf_msix_other, 0, netdev->name, netdev);
1046 igbvf_configure_msix(adapter);
1053 * igbvf_alloc_queues - Allocate memory for all rings
1054 * @adapter: board private structure to initialize
1056 static int __devinit igbvf_alloc_queues(struct igbvf_adapter *adapter)
1058 struct net_device *netdev = adapter->netdev;
1060 adapter->tx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL);
1061 if (!adapter->tx_ring)
1064 adapter->rx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL);
1065 if (!adapter->rx_ring) {
1066 kfree(adapter->tx_ring);
1070 netif_napi_add(netdev, &adapter->rx_ring->napi, igbvf_poll, 64);
1076 * igbvf_request_irq - initialize interrupts
1078 * Attempts to configure interrupts using the best available
1079 * capabilities of the hardware and kernel.
1081 static int igbvf_request_irq(struct igbvf_adapter *adapter)
1085 /* igbvf supports msi-x only */
1086 if (adapter->msix_entries)
1087 err = igbvf_request_msix(adapter);
1092 dev_err(&adapter->pdev->dev,
1093 "Unable to allocate interrupt, Error: %d\n", err);
1098 static void igbvf_free_irq(struct igbvf_adapter *adapter)
1100 struct net_device *netdev = adapter->netdev;
1103 if (adapter->msix_entries) {
1104 for (vector = 0; vector < 3; vector++)
1105 free_irq(adapter->msix_entries[vector].vector, netdev);
1110 * igbvf_irq_disable - Mask off interrupt generation on the NIC
1112 static void igbvf_irq_disable(struct igbvf_adapter *adapter)
1114 struct e1000_hw *hw = &adapter->hw;
1118 if (adapter->msix_entries)
1123 * igbvf_irq_enable - Enable default interrupt generation settings
1125 static void igbvf_irq_enable(struct igbvf_adapter *adapter)
1127 struct e1000_hw *hw = &adapter->hw;
1129 ew32(EIAC, adapter->eims_enable_mask);
1130 ew32(EIAM, adapter->eims_enable_mask);
1131 ew32(EIMS, adapter->eims_enable_mask);
1135 * igbvf_poll - NAPI Rx polling callback
1136 * @napi: struct associated with this polling callback
1137 * @budget: amount of packets driver is allowed to process this poll
1139 static int igbvf_poll(struct napi_struct *napi, int budget)
1141 struct igbvf_ring *rx_ring = container_of(napi, struct igbvf_ring, napi);
1142 struct igbvf_adapter *adapter = rx_ring->adapter;
1143 struct e1000_hw *hw = &adapter->hw;
1146 igbvf_clean_rx_irq(adapter, &work_done, budget);
1148 /* If not enough Rx work done, exit the polling mode */
1149 if (work_done < budget) {
1150 napi_complete(napi);
1152 if (adapter->itr_setting & 3)
1153 igbvf_set_itr(adapter);
1155 if (!test_bit(__IGBVF_DOWN, &adapter->state))
1156 ew32(EIMS, adapter->rx_ring->eims_value);
1163 * igbvf_set_rlpml - set receive large packet maximum length
1164 * @adapter: board private structure
1166 * Configure the maximum size of packets that will be received
1168 static void igbvf_set_rlpml(struct igbvf_adapter *adapter)
1171 struct e1000_hw *hw = &adapter->hw;
1173 max_frame_size = adapter->max_frame_size + VLAN_TAG_SIZE;
1174 e1000_rlpml_set_vf(hw, max_frame_size);
1177 static void igbvf_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1179 struct igbvf_adapter *adapter = netdev_priv(netdev);
1180 struct e1000_hw *hw = &adapter->hw;
1182 if (hw->mac.ops.set_vfta(hw, vid, true))
1183 dev_err(&adapter->pdev->dev, "Failed to add vlan id %d\n", vid);
1185 set_bit(vid, adapter->active_vlans);
1188 static void igbvf_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
1190 struct igbvf_adapter *adapter = netdev_priv(netdev);
1191 struct e1000_hw *hw = &adapter->hw;
1193 igbvf_irq_disable(adapter);
1195 if (!test_bit(__IGBVF_DOWN, &adapter->state))
1196 igbvf_irq_enable(adapter);
1198 if (hw->mac.ops.set_vfta(hw, vid, false))
1199 dev_err(&adapter->pdev->dev,
1200 "Failed to remove vlan id %d\n", vid);
1202 clear_bit(vid, adapter->active_vlans);
1205 static void igbvf_restore_vlan(struct igbvf_adapter *adapter)
1209 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
1210 igbvf_vlan_rx_add_vid(adapter->netdev, vid);
1214 * igbvf_configure_tx - Configure Transmit Unit after Reset
1215 * @adapter: board private structure
1217 * Configure the Tx unit of the MAC after a reset.
1219 static void igbvf_configure_tx(struct igbvf_adapter *adapter)
1221 struct e1000_hw *hw = &adapter->hw;
1222 struct igbvf_ring *tx_ring = adapter->tx_ring;
1224 u32 txdctl, dca_txctrl;
1226 /* disable transmits */
1227 txdctl = er32(TXDCTL(0));
1228 ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1231 /* Setup the HW Tx Head and Tail descriptor pointers */
1232 ew32(TDLEN(0), tx_ring->count * sizeof(union e1000_adv_tx_desc));
1233 tdba = tx_ring->dma;
1234 ew32(TDBAL(0), (tdba & DMA_BIT_MASK(32)));
1235 ew32(TDBAH(0), (tdba >> 32));
1238 tx_ring->head = E1000_TDH(0);
1239 tx_ring->tail = E1000_TDT(0);
1241 /* Turn off Relaxed Ordering on head write-backs. The writebacks
1242 * MUST be delivered in order or it will completely screw up
1245 dca_txctrl = er32(DCA_TXCTRL(0));
1246 dca_txctrl &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN;
1247 ew32(DCA_TXCTRL(0), dca_txctrl);
1249 /* enable transmits */
1250 txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
1251 ew32(TXDCTL(0), txdctl);
1253 /* Setup Transmit Descriptor Settings for eop descriptor */
1254 adapter->txd_cmd = E1000_ADVTXD_DCMD_EOP | E1000_ADVTXD_DCMD_IFCS;
1256 /* enable Report Status bit */
1257 adapter->txd_cmd |= E1000_ADVTXD_DCMD_RS;
1261 * igbvf_setup_srrctl - configure the receive control registers
1262 * @adapter: Board private structure
1264 static void igbvf_setup_srrctl(struct igbvf_adapter *adapter)
1266 struct e1000_hw *hw = &adapter->hw;
1269 srrctl &= ~(E1000_SRRCTL_DESCTYPE_MASK |
1270 E1000_SRRCTL_BSIZEHDR_MASK |
1271 E1000_SRRCTL_BSIZEPKT_MASK);
1273 /* Enable queue drop to avoid head of line blocking */
1274 srrctl |= E1000_SRRCTL_DROP_EN;
1276 /* Setup buffer sizes */
1277 srrctl |= ALIGN(adapter->rx_buffer_len, 1024) >>
1278 E1000_SRRCTL_BSIZEPKT_SHIFT;
1280 if (adapter->rx_buffer_len < 2048) {
1281 adapter->rx_ps_hdr_size = 0;
1282 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
1284 adapter->rx_ps_hdr_size = 128;
1285 srrctl |= adapter->rx_ps_hdr_size <<
1286 E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
1287 srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
1290 ew32(SRRCTL(0), srrctl);
1294 * igbvf_configure_rx - Configure Receive Unit after Reset
1295 * @adapter: board private structure
1297 * Configure the Rx unit of the MAC after a reset.
1299 static void igbvf_configure_rx(struct igbvf_adapter *adapter)
1301 struct e1000_hw *hw = &adapter->hw;
1302 struct igbvf_ring *rx_ring = adapter->rx_ring;
1306 /* disable receives */
1307 rxdctl = er32(RXDCTL(0));
1308 ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1311 rdlen = rx_ring->count * sizeof(union e1000_adv_rx_desc);
1314 * Setup the HW Rx Head and Tail Descriptor Pointers and
1315 * the Base and Length of the Rx Descriptor Ring
1317 rdba = rx_ring->dma;
1318 ew32(RDBAL(0), (rdba & DMA_BIT_MASK(32)));
1319 ew32(RDBAH(0), (rdba >> 32));
1320 ew32(RDLEN(0), rx_ring->count * sizeof(union e1000_adv_rx_desc));
1321 rx_ring->head = E1000_RDH(0);
1322 rx_ring->tail = E1000_RDT(0);
1326 rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
1327 rxdctl &= 0xFFF00000;
1328 rxdctl |= IGBVF_RX_PTHRESH;
1329 rxdctl |= IGBVF_RX_HTHRESH << 8;
1330 rxdctl |= IGBVF_RX_WTHRESH << 16;
1332 igbvf_set_rlpml(adapter);
1334 /* enable receives */
1335 ew32(RXDCTL(0), rxdctl);
1339 * igbvf_set_multi - Multicast and Promiscuous mode set
1340 * @netdev: network interface device structure
1342 * The set_multi entry point is called whenever the multicast address
1343 * list or the network interface flags are updated. This routine is
1344 * responsible for configuring the hardware for proper multicast,
1345 * promiscuous mode, and all-multi behavior.
1347 static void igbvf_set_multi(struct net_device *netdev)
1349 struct igbvf_adapter *adapter = netdev_priv(netdev);
1350 struct e1000_hw *hw = &adapter->hw;
1351 struct netdev_hw_addr *ha;
1352 u8 *mta_list = NULL;
1355 if (!netdev_mc_empty(netdev)) {
1356 mta_list = kmalloc(netdev_mc_count(netdev) * 6, GFP_ATOMIC);
1358 dev_err(&adapter->pdev->dev,
1359 "failed to allocate multicast filter list\n");
1364 /* prepare a packed array of only addresses. */
1366 netdev_for_each_mc_addr(ha, netdev)
1367 memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
1369 hw->mac.ops.update_mc_addr_list(hw, mta_list, i, 0, 0);
1374 * igbvf_configure - configure the hardware for Rx and Tx
1375 * @adapter: private board structure
1377 static void igbvf_configure(struct igbvf_adapter *adapter)
1379 igbvf_set_multi(adapter->netdev);
1381 igbvf_restore_vlan(adapter);
1383 igbvf_configure_tx(adapter);
1384 igbvf_setup_srrctl(adapter);
1385 igbvf_configure_rx(adapter);
1386 igbvf_alloc_rx_buffers(adapter->rx_ring,
1387 igbvf_desc_unused(adapter->rx_ring));
1390 /* igbvf_reset - bring the hardware into a known good state
1392 * This function boots the hardware and enables some settings that
1393 * require a configuration cycle of the hardware - those cannot be
1394 * set/changed during runtime. After reset the device needs to be
1395 * properly configured for Rx, Tx etc.
1397 static void igbvf_reset(struct igbvf_adapter *adapter)
1399 struct e1000_mac_info *mac = &adapter->hw.mac;
1400 struct net_device *netdev = adapter->netdev;
1401 struct e1000_hw *hw = &adapter->hw;
1403 /* Allow time for pending master requests to run */
1404 if (mac->ops.reset_hw(hw))
1405 dev_err(&adapter->pdev->dev, "PF still resetting\n");
1407 mac->ops.init_hw(hw);
1409 if (is_valid_ether_addr(adapter->hw.mac.addr)) {
1410 memcpy(netdev->dev_addr, adapter->hw.mac.addr,
1412 memcpy(netdev->perm_addr, adapter->hw.mac.addr,
1416 adapter->last_reset = jiffies;
1419 int igbvf_up(struct igbvf_adapter *adapter)
1421 struct e1000_hw *hw = &adapter->hw;
1423 /* hardware has been reset, we need to reload some things */
1424 igbvf_configure(adapter);
1426 clear_bit(__IGBVF_DOWN, &adapter->state);
1428 napi_enable(&adapter->rx_ring->napi);
1429 if (adapter->msix_entries)
1430 igbvf_configure_msix(adapter);
1432 /* Clear any pending interrupts. */
1434 igbvf_irq_enable(adapter);
1436 /* start the watchdog */
1437 hw->mac.get_link_status = 1;
1438 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1444 void igbvf_down(struct igbvf_adapter *adapter)
1446 struct net_device *netdev = adapter->netdev;
1447 struct e1000_hw *hw = &adapter->hw;
1451 * signal that we're down so the interrupt handler does not
1452 * reschedule our watchdog timer
1454 set_bit(__IGBVF_DOWN, &adapter->state);
1456 /* disable receives in the hardware */
1457 rxdctl = er32(RXDCTL(0));
1458 ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1460 netif_stop_queue(netdev);
1462 /* disable transmits in the hardware */
1463 txdctl = er32(TXDCTL(0));
1464 ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1466 /* flush both disables and wait for them to finish */
1470 napi_disable(&adapter->rx_ring->napi);
1472 igbvf_irq_disable(adapter);
1474 del_timer_sync(&adapter->watchdog_timer);
1476 netif_carrier_off(netdev);
1478 /* record the stats before reset*/
1479 igbvf_update_stats(adapter);
1481 adapter->link_speed = 0;
1482 adapter->link_duplex = 0;
1484 igbvf_reset(adapter);
1485 igbvf_clean_tx_ring(adapter->tx_ring);
1486 igbvf_clean_rx_ring(adapter->rx_ring);
1489 void igbvf_reinit_locked(struct igbvf_adapter *adapter)
1492 while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
1494 igbvf_down(adapter);
1496 clear_bit(__IGBVF_RESETTING, &adapter->state);
1500 * igbvf_sw_init - Initialize general software structures (struct igbvf_adapter)
1501 * @adapter: board private structure to initialize
1503 * igbvf_sw_init initializes the Adapter private data structure.
1504 * Fields are initialized based on PCI device information and
1505 * OS network device settings (MTU size).
1507 static int __devinit igbvf_sw_init(struct igbvf_adapter *adapter)
1509 struct net_device *netdev = adapter->netdev;
1512 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
1513 adapter->rx_ps_hdr_size = 0;
1514 adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1515 adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
1517 adapter->tx_int_delay = 8;
1518 adapter->tx_abs_int_delay = 32;
1519 adapter->rx_int_delay = 0;
1520 adapter->rx_abs_int_delay = 8;
1521 adapter->itr_setting = 3;
1522 adapter->itr = 20000;
1524 /* Set various function pointers */
1525 adapter->ei->init_ops(&adapter->hw);
1527 rc = adapter->hw.mac.ops.init_params(&adapter->hw);
1531 rc = adapter->hw.mbx.ops.init_params(&adapter->hw);
1535 igbvf_set_interrupt_capability(adapter);
1537 if (igbvf_alloc_queues(adapter))
1540 spin_lock_init(&adapter->tx_queue_lock);
1542 /* Explicitly disable IRQ since the NIC can be in any state. */
1543 igbvf_irq_disable(adapter);
1545 spin_lock_init(&adapter->stats_lock);
1547 set_bit(__IGBVF_DOWN, &adapter->state);
1551 static void igbvf_initialize_last_counter_stats(struct igbvf_adapter *adapter)
1553 struct e1000_hw *hw = &adapter->hw;
1555 adapter->stats.last_gprc = er32(VFGPRC);
1556 adapter->stats.last_gorc = er32(VFGORC);
1557 adapter->stats.last_gptc = er32(VFGPTC);
1558 adapter->stats.last_gotc = er32(VFGOTC);
1559 adapter->stats.last_mprc = er32(VFMPRC);
1560 adapter->stats.last_gotlbc = er32(VFGOTLBC);
1561 adapter->stats.last_gptlbc = er32(VFGPTLBC);
1562 adapter->stats.last_gorlbc = er32(VFGORLBC);
1563 adapter->stats.last_gprlbc = er32(VFGPRLBC);
1565 adapter->stats.base_gprc = er32(VFGPRC);
1566 adapter->stats.base_gorc = er32(VFGORC);
1567 adapter->stats.base_gptc = er32(VFGPTC);
1568 adapter->stats.base_gotc = er32(VFGOTC);
1569 adapter->stats.base_mprc = er32(VFMPRC);
1570 adapter->stats.base_gotlbc = er32(VFGOTLBC);
1571 adapter->stats.base_gptlbc = er32(VFGPTLBC);
1572 adapter->stats.base_gorlbc = er32(VFGORLBC);
1573 adapter->stats.base_gprlbc = er32(VFGPRLBC);
1577 * igbvf_open - Called when a network interface is made active
1578 * @netdev: network interface device structure
1580 * Returns 0 on success, negative value on failure
1582 * The open entry point is called when a network interface is made
1583 * active by the system (IFF_UP). At this point all resources needed
1584 * for transmit and receive operations are allocated, the interrupt
1585 * handler is registered with the OS, the watchdog timer is started,
1586 * and the stack is notified that the interface is ready.
1588 static int igbvf_open(struct net_device *netdev)
1590 struct igbvf_adapter *adapter = netdev_priv(netdev);
1591 struct e1000_hw *hw = &adapter->hw;
1594 /* disallow open during test */
1595 if (test_bit(__IGBVF_TESTING, &adapter->state))
1598 /* allocate transmit descriptors */
1599 err = igbvf_setup_tx_resources(adapter, adapter->tx_ring);
1603 /* allocate receive descriptors */
1604 err = igbvf_setup_rx_resources(adapter, adapter->rx_ring);
1609 * before we allocate an interrupt, we must be ready to handle it.
1610 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1611 * as soon as we call pci_request_irq, so we have to setup our
1612 * clean_rx handler before we do so.
1614 igbvf_configure(adapter);
1616 err = igbvf_request_irq(adapter);
1620 /* From here on the code is the same as igbvf_up() */
1621 clear_bit(__IGBVF_DOWN, &adapter->state);
1623 napi_enable(&adapter->rx_ring->napi);
1625 /* clear any pending interrupts */
1628 igbvf_irq_enable(adapter);
1630 /* start the watchdog */
1631 hw->mac.get_link_status = 1;
1632 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1637 igbvf_free_rx_resources(adapter->rx_ring);
1639 igbvf_free_tx_resources(adapter->tx_ring);
1641 igbvf_reset(adapter);
1647 * igbvf_close - Disables a network interface
1648 * @netdev: network interface device structure
1650 * Returns 0, this is not allowed to fail
1652 * The close entry point is called when an interface is de-activated
1653 * by the OS. The hardware is still under the drivers control, but
1654 * needs to be disabled. A global MAC reset is issued to stop the
1655 * hardware, and all transmit and receive resources are freed.
1657 static int igbvf_close(struct net_device *netdev)
1659 struct igbvf_adapter *adapter = netdev_priv(netdev);
1661 WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
1662 igbvf_down(adapter);
1664 igbvf_free_irq(adapter);
1666 igbvf_free_tx_resources(adapter->tx_ring);
1667 igbvf_free_rx_resources(adapter->rx_ring);
1672 * igbvf_set_mac - Change the Ethernet Address of the NIC
1673 * @netdev: network interface device structure
1674 * @p: pointer to an address structure
1676 * Returns 0 on success, negative on failure
1678 static int igbvf_set_mac(struct net_device *netdev, void *p)
1680 struct igbvf_adapter *adapter = netdev_priv(netdev);
1681 struct e1000_hw *hw = &adapter->hw;
1682 struct sockaddr *addr = p;
1684 if (!is_valid_ether_addr(addr->sa_data))
1685 return -EADDRNOTAVAIL;
1687 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
1689 hw->mac.ops.rar_set(hw, hw->mac.addr, 0);
1691 if (memcmp(addr->sa_data, hw->mac.addr, 6))
1692 return -EADDRNOTAVAIL;
1694 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
1699 #define UPDATE_VF_COUNTER(reg, name) \
1701 u32 current_counter = er32(reg); \
1702 if (current_counter < adapter->stats.last_##name) \
1703 adapter->stats.name += 0x100000000LL; \
1704 adapter->stats.last_##name = current_counter; \
1705 adapter->stats.name &= 0xFFFFFFFF00000000LL; \
1706 adapter->stats.name |= current_counter; \
1710 * igbvf_update_stats - Update the board statistics counters
1711 * @adapter: board private structure
1713 void igbvf_update_stats(struct igbvf_adapter *adapter)
1715 struct e1000_hw *hw = &adapter->hw;
1716 struct pci_dev *pdev = adapter->pdev;
1719 * Prevent stats update while adapter is being reset, link is down
1720 * or if the pci connection is down.
1722 if (adapter->link_speed == 0)
1725 if (test_bit(__IGBVF_RESETTING, &adapter->state))
1728 if (pci_channel_offline(pdev))
1731 UPDATE_VF_COUNTER(VFGPRC, gprc);
1732 UPDATE_VF_COUNTER(VFGORC, gorc);
1733 UPDATE_VF_COUNTER(VFGPTC, gptc);
1734 UPDATE_VF_COUNTER(VFGOTC, gotc);
1735 UPDATE_VF_COUNTER(VFMPRC, mprc);
1736 UPDATE_VF_COUNTER(VFGOTLBC, gotlbc);
1737 UPDATE_VF_COUNTER(VFGPTLBC, gptlbc);
1738 UPDATE_VF_COUNTER(VFGORLBC, gorlbc);
1739 UPDATE_VF_COUNTER(VFGPRLBC, gprlbc);
1741 /* Fill out the OS statistics structure */
1742 adapter->net_stats.multicast = adapter->stats.mprc;
1745 static void igbvf_print_link_info(struct igbvf_adapter *adapter)
1747 dev_info(&adapter->pdev->dev, "Link is Up %d Mbps %s\n",
1748 adapter->link_speed,
1749 ((adapter->link_duplex == FULL_DUPLEX) ?
1750 "Full Duplex" : "Half Duplex"));
1753 static bool igbvf_has_link(struct igbvf_adapter *adapter)
1755 struct e1000_hw *hw = &adapter->hw;
1756 s32 ret_val = E1000_SUCCESS;
1759 /* If interface is down, stay link down */
1760 if (test_bit(__IGBVF_DOWN, &adapter->state))
1763 ret_val = hw->mac.ops.check_for_link(hw);
1764 link_active = !hw->mac.get_link_status;
1766 /* if check for link returns error we will need to reset */
1767 if (ret_val && time_after(jiffies, adapter->last_reset + (10 * HZ)))
1768 schedule_work(&adapter->reset_task);
1774 * igbvf_watchdog - Timer Call-back
1775 * @data: pointer to adapter cast into an unsigned long
1777 static void igbvf_watchdog(unsigned long data)
1779 struct igbvf_adapter *adapter = (struct igbvf_adapter *) data;
1781 /* Do the rest outside of interrupt context */
1782 schedule_work(&adapter->watchdog_task);
1785 static void igbvf_watchdog_task(struct work_struct *work)
1787 struct igbvf_adapter *adapter = container_of(work,
1788 struct igbvf_adapter,
1790 struct net_device *netdev = adapter->netdev;
1791 struct e1000_mac_info *mac = &adapter->hw.mac;
1792 struct igbvf_ring *tx_ring = adapter->tx_ring;
1793 struct e1000_hw *hw = &adapter->hw;
1797 link = igbvf_has_link(adapter);
1800 if (!netif_carrier_ok(netdev)) {
1801 mac->ops.get_link_up_info(&adapter->hw,
1802 &adapter->link_speed,
1803 &adapter->link_duplex);
1804 igbvf_print_link_info(adapter);
1806 netif_carrier_on(netdev);
1807 netif_wake_queue(netdev);
1810 if (netif_carrier_ok(netdev)) {
1811 adapter->link_speed = 0;
1812 adapter->link_duplex = 0;
1813 dev_info(&adapter->pdev->dev, "Link is Down\n");
1814 netif_carrier_off(netdev);
1815 netif_stop_queue(netdev);
1819 if (netif_carrier_ok(netdev)) {
1820 igbvf_update_stats(adapter);
1822 tx_pending = (igbvf_desc_unused(tx_ring) + 1 <
1826 * We've lost link, so the controller stops DMA,
1827 * but we've got queued Tx work that's never going
1828 * to get done, so reset controller to flush Tx.
1829 * (Do the reset outside of interrupt context).
1831 adapter->tx_timeout_count++;
1832 schedule_work(&adapter->reset_task);
1836 /* Cause software interrupt to ensure Rx ring is cleaned */
1837 ew32(EICS, adapter->rx_ring->eims_value);
1839 /* Reset the timer */
1840 if (!test_bit(__IGBVF_DOWN, &adapter->state))
1841 mod_timer(&adapter->watchdog_timer,
1842 round_jiffies(jiffies + (2 * HZ)));
1845 #define IGBVF_TX_FLAGS_CSUM 0x00000001
1846 #define IGBVF_TX_FLAGS_VLAN 0x00000002
1847 #define IGBVF_TX_FLAGS_TSO 0x00000004
1848 #define IGBVF_TX_FLAGS_IPV4 0x00000008
1849 #define IGBVF_TX_FLAGS_VLAN_MASK 0xffff0000
1850 #define IGBVF_TX_FLAGS_VLAN_SHIFT 16
1852 static int igbvf_tso(struct igbvf_adapter *adapter,
1853 struct igbvf_ring *tx_ring,
1854 struct sk_buff *skb, u32 tx_flags, u8 *hdr_len)
1856 struct e1000_adv_tx_context_desc *context_desc;
1859 struct igbvf_buffer *buffer_info;
1860 u32 info = 0, tu_cmd = 0;
1861 u32 mss_l4len_idx, l4len;
1864 if (skb_header_cloned(skb)) {
1865 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1867 dev_err(&adapter->pdev->dev,
1868 "igbvf_tso returning an error\n");
1873 l4len = tcp_hdrlen(skb);
1876 if (skb->protocol == htons(ETH_P_IP)) {
1877 struct iphdr *iph = ip_hdr(skb);
1880 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
1884 } else if (skb_is_gso_v6(skb)) {
1885 ipv6_hdr(skb)->payload_len = 0;
1886 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
1887 &ipv6_hdr(skb)->daddr,
1891 i = tx_ring->next_to_use;
1893 buffer_info = &tx_ring->buffer_info[i];
1894 context_desc = IGBVF_TX_CTXTDESC_ADV(*tx_ring, i);
1895 /* VLAN MACLEN IPLEN */
1896 if (tx_flags & IGBVF_TX_FLAGS_VLAN)
1897 info |= (tx_flags & IGBVF_TX_FLAGS_VLAN_MASK);
1898 info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
1899 *hdr_len += skb_network_offset(skb);
1900 info |= (skb_transport_header(skb) - skb_network_header(skb));
1901 *hdr_len += (skb_transport_header(skb) - skb_network_header(skb));
1902 context_desc->vlan_macip_lens = cpu_to_le32(info);
1904 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
1905 tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
1907 if (skb->protocol == htons(ETH_P_IP))
1908 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
1909 tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
1911 context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
1914 mss_l4len_idx = (skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT);
1915 mss_l4len_idx |= (l4len << E1000_ADVTXD_L4LEN_SHIFT);
1917 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
1918 context_desc->seqnum_seed = 0;
1920 buffer_info->time_stamp = jiffies;
1921 buffer_info->next_to_watch = i;
1922 buffer_info->dma = 0;
1924 if (i == tx_ring->count)
1927 tx_ring->next_to_use = i;
1932 static inline bool igbvf_tx_csum(struct igbvf_adapter *adapter,
1933 struct igbvf_ring *tx_ring,
1934 struct sk_buff *skb, u32 tx_flags)
1936 struct e1000_adv_tx_context_desc *context_desc;
1938 struct igbvf_buffer *buffer_info;
1939 u32 info = 0, tu_cmd = 0;
1941 if ((skb->ip_summed == CHECKSUM_PARTIAL) ||
1942 (tx_flags & IGBVF_TX_FLAGS_VLAN)) {
1943 i = tx_ring->next_to_use;
1944 buffer_info = &tx_ring->buffer_info[i];
1945 context_desc = IGBVF_TX_CTXTDESC_ADV(*tx_ring, i);
1947 if (tx_flags & IGBVF_TX_FLAGS_VLAN)
1948 info |= (tx_flags & IGBVF_TX_FLAGS_VLAN_MASK);
1950 info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
1951 if (skb->ip_summed == CHECKSUM_PARTIAL)
1952 info |= (skb_transport_header(skb) -
1953 skb_network_header(skb));
1956 context_desc->vlan_macip_lens = cpu_to_le32(info);
1958 tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
1960 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1961 switch (skb->protocol) {
1962 case __constant_htons(ETH_P_IP):
1963 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
1964 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
1965 tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
1967 case __constant_htons(ETH_P_IPV6):
1968 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
1969 tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
1976 context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
1977 context_desc->seqnum_seed = 0;
1978 context_desc->mss_l4len_idx = 0;
1980 buffer_info->time_stamp = jiffies;
1981 buffer_info->next_to_watch = i;
1982 buffer_info->dma = 0;
1984 if (i == tx_ring->count)
1986 tx_ring->next_to_use = i;
1994 static int igbvf_maybe_stop_tx(struct net_device *netdev, int size)
1996 struct igbvf_adapter *adapter = netdev_priv(netdev);
1998 /* there is enough descriptors then we don't need to worry */
1999 if (igbvf_desc_unused(adapter->tx_ring) >= size)
2002 netif_stop_queue(netdev);
2006 /* We need to check again just in case room has been made available */
2007 if (igbvf_desc_unused(adapter->tx_ring) < size)
2010 netif_wake_queue(netdev);
2012 ++adapter->restart_queue;
2016 #define IGBVF_MAX_TXD_PWR 16
2017 #define IGBVF_MAX_DATA_PER_TXD (1 << IGBVF_MAX_TXD_PWR)
2019 static inline int igbvf_tx_map_adv(struct igbvf_adapter *adapter,
2020 struct igbvf_ring *tx_ring,
2021 struct sk_buff *skb,
2024 struct igbvf_buffer *buffer_info;
2025 struct pci_dev *pdev = adapter->pdev;
2026 unsigned int len = skb_headlen(skb);
2027 unsigned int count = 0, i;
2030 i = tx_ring->next_to_use;
2032 buffer_info = &tx_ring->buffer_info[i];
2033 BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD);
2034 buffer_info->length = len;
2035 /* set time_stamp *before* dma to help avoid a possible race */
2036 buffer_info->time_stamp = jiffies;
2037 buffer_info->next_to_watch = i;
2038 buffer_info->mapped_as_page = false;
2039 buffer_info->dma = dma_map_single(&pdev->dev, skb->data, len,
2041 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2045 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) {
2046 struct skb_frag_struct *frag;
2050 if (i == tx_ring->count)
2053 frag = &skb_shinfo(skb)->frags[f];
2056 buffer_info = &tx_ring->buffer_info[i];
2057 BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD);
2058 buffer_info->length = len;
2059 buffer_info->time_stamp = jiffies;
2060 buffer_info->next_to_watch = i;
2061 buffer_info->mapped_as_page = true;
2062 buffer_info->dma = dma_map_page(&pdev->dev,
2067 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2071 tx_ring->buffer_info[i].skb = skb;
2072 tx_ring->buffer_info[first].next_to_watch = i;
2077 dev_err(&pdev->dev, "TX DMA map failed\n");
2079 /* clear timestamp and dma mappings for failed buffer_info mapping */
2080 buffer_info->dma = 0;
2081 buffer_info->time_stamp = 0;
2082 buffer_info->length = 0;
2083 buffer_info->next_to_watch = 0;
2084 buffer_info->mapped_as_page = false;
2088 /* clear timestamp and dma mappings for remaining portion of packet */
2091 i += tx_ring->count;
2093 buffer_info = &tx_ring->buffer_info[i];
2094 igbvf_put_txbuf(adapter, buffer_info);
2100 static inline void igbvf_tx_queue_adv(struct igbvf_adapter *adapter,
2101 struct igbvf_ring *tx_ring,
2102 int tx_flags, int count, u32 paylen,
2105 union e1000_adv_tx_desc *tx_desc = NULL;
2106 struct igbvf_buffer *buffer_info;
2107 u32 olinfo_status = 0, cmd_type_len;
2110 cmd_type_len = (E1000_ADVTXD_DTYP_DATA | E1000_ADVTXD_DCMD_IFCS |
2111 E1000_ADVTXD_DCMD_DEXT);
2113 if (tx_flags & IGBVF_TX_FLAGS_VLAN)
2114 cmd_type_len |= E1000_ADVTXD_DCMD_VLE;
2116 if (tx_flags & IGBVF_TX_FLAGS_TSO) {
2117 cmd_type_len |= E1000_ADVTXD_DCMD_TSE;
2119 /* insert tcp checksum */
2120 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2122 /* insert ip checksum */
2123 if (tx_flags & IGBVF_TX_FLAGS_IPV4)
2124 olinfo_status |= E1000_TXD_POPTS_IXSM << 8;
2126 } else if (tx_flags & IGBVF_TX_FLAGS_CSUM) {
2127 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2130 olinfo_status |= ((paylen - hdr_len) << E1000_ADVTXD_PAYLEN_SHIFT);
2132 i = tx_ring->next_to_use;
2134 buffer_info = &tx_ring->buffer_info[i];
2135 tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
2136 tx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
2137 tx_desc->read.cmd_type_len =
2138 cpu_to_le32(cmd_type_len | buffer_info->length);
2139 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
2141 if (i == tx_ring->count)
2145 tx_desc->read.cmd_type_len |= cpu_to_le32(adapter->txd_cmd);
2146 /* Force memory writes to complete before letting h/w
2147 * know there are new descriptors to fetch. (Only
2148 * applicable for weak-ordered memory model archs,
2149 * such as IA-64). */
2152 tx_ring->next_to_use = i;
2153 writel(i, adapter->hw.hw_addr + tx_ring->tail);
2154 /* we need this if more than one processor can write to our tail
2155 * at a time, it syncronizes IO on IA64/Altix systems */
2159 static netdev_tx_t igbvf_xmit_frame_ring_adv(struct sk_buff *skb,
2160 struct net_device *netdev,
2161 struct igbvf_ring *tx_ring)
2163 struct igbvf_adapter *adapter = netdev_priv(netdev);
2164 unsigned int first, tx_flags = 0;
2169 if (test_bit(__IGBVF_DOWN, &adapter->state)) {
2170 dev_kfree_skb_any(skb);
2171 return NETDEV_TX_OK;
2174 if (skb->len <= 0) {
2175 dev_kfree_skb_any(skb);
2176 return NETDEV_TX_OK;
2180 * need: count + 4 desc gap to keep tail from touching
2181 * + 2 desc gap to keep tail from touching head,
2182 * + 1 desc for skb->data,
2183 * + 1 desc for context descriptor,
2184 * head, otherwise try next time
2186 if (igbvf_maybe_stop_tx(netdev, skb_shinfo(skb)->nr_frags + 4)) {
2187 /* this is a hard error */
2188 return NETDEV_TX_BUSY;
2191 if (vlan_tx_tag_present(skb)) {
2192 tx_flags |= IGBVF_TX_FLAGS_VLAN;
2193 tx_flags |= (vlan_tx_tag_get(skb) << IGBVF_TX_FLAGS_VLAN_SHIFT);
2196 if (skb->protocol == htons(ETH_P_IP))
2197 tx_flags |= IGBVF_TX_FLAGS_IPV4;
2199 first = tx_ring->next_to_use;
2201 tso = skb_is_gso(skb) ?
2202 igbvf_tso(adapter, tx_ring, skb, tx_flags, &hdr_len) : 0;
2203 if (unlikely(tso < 0)) {
2204 dev_kfree_skb_any(skb);
2205 return NETDEV_TX_OK;
2209 tx_flags |= IGBVF_TX_FLAGS_TSO;
2210 else if (igbvf_tx_csum(adapter, tx_ring, skb, tx_flags) &&
2211 (skb->ip_summed == CHECKSUM_PARTIAL))
2212 tx_flags |= IGBVF_TX_FLAGS_CSUM;
2215 * count reflects descriptors mapped, if 0 then mapping error
2216 * has occurred and we need to rewind the descriptor queue
2218 count = igbvf_tx_map_adv(adapter, tx_ring, skb, first);
2221 igbvf_tx_queue_adv(adapter, tx_ring, tx_flags, count,
2223 /* Make sure there is space in the ring for the next send. */
2224 igbvf_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 4);
2226 dev_kfree_skb_any(skb);
2227 tx_ring->buffer_info[first].time_stamp = 0;
2228 tx_ring->next_to_use = first;
2231 return NETDEV_TX_OK;
2234 static netdev_tx_t igbvf_xmit_frame(struct sk_buff *skb,
2235 struct net_device *netdev)
2237 struct igbvf_adapter *adapter = netdev_priv(netdev);
2238 struct igbvf_ring *tx_ring;
2240 if (test_bit(__IGBVF_DOWN, &adapter->state)) {
2241 dev_kfree_skb_any(skb);
2242 return NETDEV_TX_OK;
2245 tx_ring = &adapter->tx_ring[0];
2247 return igbvf_xmit_frame_ring_adv(skb, netdev, tx_ring);
2251 * igbvf_tx_timeout - Respond to a Tx Hang
2252 * @netdev: network interface device structure
2254 static void igbvf_tx_timeout(struct net_device *netdev)
2256 struct igbvf_adapter *adapter = netdev_priv(netdev);
2258 /* Do the reset outside of interrupt context */
2259 adapter->tx_timeout_count++;
2260 schedule_work(&adapter->reset_task);
2263 static void igbvf_reset_task(struct work_struct *work)
2265 struct igbvf_adapter *adapter;
2266 adapter = container_of(work, struct igbvf_adapter, reset_task);
2268 igbvf_reinit_locked(adapter);
2272 * igbvf_get_stats - Get System Network Statistics
2273 * @netdev: network interface device structure
2275 * Returns the address of the device statistics structure.
2276 * The statistics are actually updated from the timer callback.
2278 static struct net_device_stats *igbvf_get_stats(struct net_device *netdev)
2280 struct igbvf_adapter *adapter = netdev_priv(netdev);
2282 /* only return the current stats */
2283 return &adapter->net_stats;
2287 * igbvf_change_mtu - Change the Maximum Transfer Unit
2288 * @netdev: network interface device structure
2289 * @new_mtu: new value for maximum frame size
2291 * Returns 0 on success, negative on failure
2293 static int igbvf_change_mtu(struct net_device *netdev, int new_mtu)
2295 struct igbvf_adapter *adapter = netdev_priv(netdev);
2296 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
2298 if ((new_mtu < 68) || (max_frame > MAX_JUMBO_FRAME_SIZE)) {
2299 dev_err(&adapter->pdev->dev, "Invalid MTU setting\n");
2303 #define MAX_STD_JUMBO_FRAME_SIZE 9234
2304 if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
2305 dev_err(&adapter->pdev->dev, "MTU > 9216 not supported.\n");
2309 while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
2311 /* igbvf_down has a dependency on max_frame_size */
2312 adapter->max_frame_size = max_frame;
2313 if (netif_running(netdev))
2314 igbvf_down(adapter);
2317 * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
2318 * means we reserve 2 more, this pushes us to allocate from the next
2320 * i.e. RXBUFFER_2048 --> size-4096 slab
2321 * However with the new *_jumbo_rx* routines, jumbo receives will use
2325 if (max_frame <= 1024)
2326 adapter->rx_buffer_len = 1024;
2327 else if (max_frame <= 2048)
2328 adapter->rx_buffer_len = 2048;
2330 #if (PAGE_SIZE / 2) > 16384
2331 adapter->rx_buffer_len = 16384;
2333 adapter->rx_buffer_len = PAGE_SIZE / 2;
2337 /* adjust allocation if LPE protects us, and we aren't using SBP */
2338 if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
2339 (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
2340 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN +
2343 dev_info(&adapter->pdev->dev, "changing MTU from %d to %d\n",
2344 netdev->mtu, new_mtu);
2345 netdev->mtu = new_mtu;
2347 if (netif_running(netdev))
2350 igbvf_reset(adapter);
2352 clear_bit(__IGBVF_RESETTING, &adapter->state);
2357 static int igbvf_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
2365 static int igbvf_suspend(struct pci_dev *pdev, pm_message_t state)
2367 struct net_device *netdev = pci_get_drvdata(pdev);
2368 struct igbvf_adapter *adapter = netdev_priv(netdev);
2373 netif_device_detach(netdev);
2375 if (netif_running(netdev)) {
2376 WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
2377 igbvf_down(adapter);
2378 igbvf_free_irq(adapter);
2382 retval = pci_save_state(pdev);
2387 pci_disable_device(pdev);
2393 static int igbvf_resume(struct pci_dev *pdev)
2395 struct net_device *netdev = pci_get_drvdata(pdev);
2396 struct igbvf_adapter *adapter = netdev_priv(netdev);
2399 pci_restore_state(pdev);
2400 err = pci_enable_device_mem(pdev);
2402 dev_err(&pdev->dev, "Cannot enable PCI device from suspend\n");
2406 pci_set_master(pdev);
2408 if (netif_running(netdev)) {
2409 err = igbvf_request_irq(adapter);
2414 igbvf_reset(adapter);
2416 if (netif_running(netdev))
2419 netif_device_attach(netdev);
2425 static void igbvf_shutdown(struct pci_dev *pdev)
2427 igbvf_suspend(pdev, PMSG_SUSPEND);
2430 #ifdef CONFIG_NET_POLL_CONTROLLER
2432 * Polling 'interrupt' - used by things like netconsole to send skbs
2433 * without having to re-enable interrupts. It's not called while
2434 * the interrupt routine is executing.
2436 static void igbvf_netpoll(struct net_device *netdev)
2438 struct igbvf_adapter *adapter = netdev_priv(netdev);
2440 disable_irq(adapter->pdev->irq);
2442 igbvf_clean_tx_irq(adapter->tx_ring);
2444 enable_irq(adapter->pdev->irq);
2449 * igbvf_io_error_detected - called when PCI error is detected
2450 * @pdev: Pointer to PCI device
2451 * @state: The current pci connection state
2453 * This function is called after a PCI bus error affecting
2454 * this device has been detected.
2456 static pci_ers_result_t igbvf_io_error_detected(struct pci_dev *pdev,
2457 pci_channel_state_t state)
2459 struct net_device *netdev = pci_get_drvdata(pdev);
2460 struct igbvf_adapter *adapter = netdev_priv(netdev);
2462 netif_device_detach(netdev);
2464 if (state == pci_channel_io_perm_failure)
2465 return PCI_ERS_RESULT_DISCONNECT;
2467 if (netif_running(netdev))
2468 igbvf_down(adapter);
2469 pci_disable_device(pdev);
2471 /* Request a slot slot reset. */
2472 return PCI_ERS_RESULT_NEED_RESET;
2476 * igbvf_io_slot_reset - called after the pci bus has been reset.
2477 * @pdev: Pointer to PCI device
2479 * Restart the card from scratch, as if from a cold-boot. Implementation
2480 * resembles the first-half of the igbvf_resume routine.
2482 static pci_ers_result_t igbvf_io_slot_reset(struct pci_dev *pdev)
2484 struct net_device *netdev = pci_get_drvdata(pdev);
2485 struct igbvf_adapter *adapter = netdev_priv(netdev);
2487 if (pci_enable_device_mem(pdev)) {
2489 "Cannot re-enable PCI device after reset.\n");
2490 return PCI_ERS_RESULT_DISCONNECT;
2492 pci_set_master(pdev);
2494 igbvf_reset(adapter);
2496 return PCI_ERS_RESULT_RECOVERED;
2500 * igbvf_io_resume - called when traffic can start flowing again.
2501 * @pdev: Pointer to PCI device
2503 * This callback is called when the error recovery driver tells us that
2504 * its OK to resume normal operation. Implementation resembles the
2505 * second-half of the igbvf_resume routine.
2507 static void igbvf_io_resume(struct pci_dev *pdev)
2509 struct net_device *netdev = pci_get_drvdata(pdev);
2510 struct igbvf_adapter *adapter = netdev_priv(netdev);
2512 if (netif_running(netdev)) {
2513 if (igbvf_up(adapter)) {
2515 "can't bring device back up after reset\n");
2520 netif_device_attach(netdev);
2523 static void igbvf_print_device_info(struct igbvf_adapter *adapter)
2525 struct e1000_hw *hw = &adapter->hw;
2526 struct net_device *netdev = adapter->netdev;
2527 struct pci_dev *pdev = adapter->pdev;
2529 dev_info(&pdev->dev, "Intel(R) 82576 Virtual Function\n");
2530 dev_info(&pdev->dev, "Address: %pM\n", netdev->dev_addr);
2531 dev_info(&pdev->dev, "MAC: %d\n", hw->mac.type);
2534 static const struct net_device_ops igbvf_netdev_ops = {
2535 .ndo_open = igbvf_open,
2536 .ndo_stop = igbvf_close,
2537 .ndo_start_xmit = igbvf_xmit_frame,
2538 .ndo_get_stats = igbvf_get_stats,
2539 .ndo_set_multicast_list = igbvf_set_multi,
2540 .ndo_set_mac_address = igbvf_set_mac,
2541 .ndo_change_mtu = igbvf_change_mtu,
2542 .ndo_do_ioctl = igbvf_ioctl,
2543 .ndo_tx_timeout = igbvf_tx_timeout,
2544 .ndo_vlan_rx_add_vid = igbvf_vlan_rx_add_vid,
2545 .ndo_vlan_rx_kill_vid = igbvf_vlan_rx_kill_vid,
2546 #ifdef CONFIG_NET_POLL_CONTROLLER
2547 .ndo_poll_controller = igbvf_netpoll,
2552 * igbvf_probe - Device Initialization Routine
2553 * @pdev: PCI device information struct
2554 * @ent: entry in igbvf_pci_tbl
2556 * Returns 0 on success, negative on failure
2558 * igbvf_probe initializes an adapter identified by a pci_dev structure.
2559 * The OS initialization, configuring of the adapter private structure,
2560 * and a hardware reset occur.
2562 static int __devinit igbvf_probe(struct pci_dev *pdev,
2563 const struct pci_device_id *ent)
2565 struct net_device *netdev;
2566 struct igbvf_adapter *adapter;
2567 struct e1000_hw *hw;
2568 const struct igbvf_info *ei = igbvf_info_tbl[ent->driver_data];
2570 static int cards_found;
2571 int err, pci_using_dac;
2573 err = pci_enable_device_mem(pdev);
2578 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(64));
2580 err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64));
2584 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
2586 err = dma_set_coherent_mask(&pdev->dev,
2589 dev_err(&pdev->dev, "No usable DMA "
2590 "configuration, aborting\n");
2596 err = pci_request_regions(pdev, igbvf_driver_name);
2600 pci_set_master(pdev);
2603 netdev = alloc_etherdev(sizeof(struct igbvf_adapter));
2605 goto err_alloc_etherdev;
2607 SET_NETDEV_DEV(netdev, &pdev->dev);
2609 pci_set_drvdata(pdev, netdev);
2610 adapter = netdev_priv(netdev);
2612 adapter->netdev = netdev;
2613 adapter->pdev = pdev;
2615 adapter->pba = ei->pba;
2616 adapter->flags = ei->flags;
2617 adapter->hw.back = adapter;
2618 adapter->hw.mac.type = ei->mac;
2619 adapter->msg_enable = (1 << NETIF_MSG_DRV | NETIF_MSG_PROBE) - 1;
2621 /* PCI config space info */
2623 hw->vendor_id = pdev->vendor;
2624 hw->device_id = pdev->device;
2625 hw->subsystem_vendor_id = pdev->subsystem_vendor;
2626 hw->subsystem_device_id = pdev->subsystem_device;
2627 hw->revision_id = pdev->revision;
2630 adapter->hw.hw_addr = ioremap(pci_resource_start(pdev, 0),
2631 pci_resource_len(pdev, 0));
2633 if (!adapter->hw.hw_addr)
2636 if (ei->get_variants) {
2637 err = ei->get_variants(adapter);
2642 /* setup adapter struct */
2643 err = igbvf_sw_init(adapter);
2647 /* construct the net_device struct */
2648 netdev->netdev_ops = &igbvf_netdev_ops;
2650 igbvf_set_ethtool_ops(netdev);
2651 netdev->watchdog_timeo = 5 * HZ;
2652 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
2654 adapter->bd_number = cards_found++;
2656 netdev->features = NETIF_F_SG |
2658 NETIF_F_HW_VLAN_TX |
2659 NETIF_F_HW_VLAN_RX |
2660 NETIF_F_HW_VLAN_FILTER;
2662 netdev->features |= NETIF_F_IPV6_CSUM;
2663 netdev->features |= NETIF_F_TSO;
2664 netdev->features |= NETIF_F_TSO6;
2667 netdev->features |= NETIF_F_HIGHDMA;
2669 netdev->vlan_features |= NETIF_F_TSO;
2670 netdev->vlan_features |= NETIF_F_TSO6;
2671 netdev->vlan_features |= NETIF_F_IP_CSUM;
2672 netdev->vlan_features |= NETIF_F_IPV6_CSUM;
2673 netdev->vlan_features |= NETIF_F_SG;
2675 /*reset the controller to put the device in a known good state */
2676 err = hw->mac.ops.reset_hw(hw);
2678 dev_info(&pdev->dev,
2679 "PF still in reset state, assigning new address."
2680 " Is the PF interface up?\n");
2681 dev_hw_addr_random(adapter->netdev, hw->mac.addr);
2683 err = hw->mac.ops.read_mac_addr(hw);
2685 dev_err(&pdev->dev, "Error reading MAC address\n");
2690 memcpy(netdev->dev_addr, adapter->hw.mac.addr, netdev->addr_len);
2691 memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
2693 if (!is_valid_ether_addr(netdev->perm_addr)) {
2694 dev_err(&pdev->dev, "Invalid MAC Address: %pM\n",
2700 setup_timer(&adapter->watchdog_timer, &igbvf_watchdog,
2701 (unsigned long) adapter);
2703 INIT_WORK(&adapter->reset_task, igbvf_reset_task);
2704 INIT_WORK(&adapter->watchdog_task, igbvf_watchdog_task);
2706 /* ring size defaults */
2707 adapter->rx_ring->count = 1024;
2708 adapter->tx_ring->count = 1024;
2710 /* reset the hardware with the new settings */
2711 igbvf_reset(adapter);
2713 strcpy(netdev->name, "eth%d");
2714 err = register_netdev(netdev);
2718 /* tell the stack to leave us alone until igbvf_open() is called */
2719 netif_carrier_off(netdev);
2720 netif_stop_queue(netdev);
2722 igbvf_print_device_info(adapter);
2724 igbvf_initialize_last_counter_stats(adapter);
2729 kfree(adapter->tx_ring);
2730 kfree(adapter->rx_ring);
2732 igbvf_reset_interrupt_capability(adapter);
2733 iounmap(adapter->hw.hw_addr);
2735 free_netdev(netdev);
2737 pci_release_regions(pdev);
2740 pci_disable_device(pdev);
2745 * igbvf_remove - Device Removal Routine
2746 * @pdev: PCI device information struct
2748 * igbvf_remove is called by the PCI subsystem to alert the driver
2749 * that it should release a PCI device. The could be caused by a
2750 * Hot-Plug event, or because the driver is going to be removed from
2753 static void __devexit igbvf_remove(struct pci_dev *pdev)
2755 struct net_device *netdev = pci_get_drvdata(pdev);
2756 struct igbvf_adapter *adapter = netdev_priv(netdev);
2757 struct e1000_hw *hw = &adapter->hw;
2760 * The watchdog timer may be rescheduled, so explicitly
2761 * disable it from being rescheduled.
2763 set_bit(__IGBVF_DOWN, &adapter->state);
2764 del_timer_sync(&adapter->watchdog_timer);
2766 cancel_work_sync(&adapter->reset_task);
2767 cancel_work_sync(&adapter->watchdog_task);
2769 unregister_netdev(netdev);
2771 igbvf_reset_interrupt_capability(adapter);
2774 * it is important to delete the napi struct prior to freeing the
2775 * rx ring so that you do not end up with null pointer refs
2777 netif_napi_del(&adapter->rx_ring->napi);
2778 kfree(adapter->tx_ring);
2779 kfree(adapter->rx_ring);
2781 iounmap(hw->hw_addr);
2782 if (hw->flash_address)
2783 iounmap(hw->flash_address);
2784 pci_release_regions(pdev);
2786 free_netdev(netdev);
2788 pci_disable_device(pdev);
2791 /* PCI Error Recovery (ERS) */
2792 static struct pci_error_handlers igbvf_err_handler = {
2793 .error_detected = igbvf_io_error_detected,
2794 .slot_reset = igbvf_io_slot_reset,
2795 .resume = igbvf_io_resume,
2798 static DEFINE_PCI_DEVICE_TABLE(igbvf_pci_tbl) = {
2799 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_VF), board_vf },
2800 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_VF), board_i350_vf },
2801 { } /* terminate list */
2803 MODULE_DEVICE_TABLE(pci, igbvf_pci_tbl);
2805 /* PCI Device API Driver */
2806 static struct pci_driver igbvf_driver = {
2807 .name = igbvf_driver_name,
2808 .id_table = igbvf_pci_tbl,
2809 .probe = igbvf_probe,
2810 .remove = __devexit_p(igbvf_remove),
2812 /* Power Management Hooks */
2813 .suspend = igbvf_suspend,
2814 .resume = igbvf_resume,
2816 .shutdown = igbvf_shutdown,
2817 .err_handler = &igbvf_err_handler
2821 * igbvf_init_module - Driver Registration Routine
2823 * igbvf_init_module is the first routine called when the driver is
2824 * loaded. All it does is register with the PCI subsystem.
2826 static int __init igbvf_init_module(void)
2829 printk(KERN_INFO "%s - version %s\n",
2830 igbvf_driver_string, igbvf_driver_version);
2831 printk(KERN_INFO "%s\n", igbvf_copyright);
2833 ret = pci_register_driver(&igbvf_driver);
2837 module_init(igbvf_init_module);
2840 * igbvf_exit_module - Driver Exit Cleanup Routine
2842 * igbvf_exit_module is called just before the driver is removed
2845 static void __exit igbvf_exit_module(void)
2847 pci_unregister_driver(&igbvf_driver);
2849 module_exit(igbvf_exit_module);
2852 MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
2853 MODULE_DESCRIPTION("Intel(R) 82576 Virtual Function Network Driver");
2854 MODULE_LICENSE("GPL");
2855 MODULE_VERSION(DRV_VERSION);