1 /*******************************************************************************
3 Intel(R) 82576 Virtual Function Linux driver
4 Copyright(c) 2009 - 2012 Intel Corporation.
6 This program is free software; you can redistribute it and/or modify it
7 under the terms and conditions of the GNU General Public License,
8 version 2, as published by the Free Software Foundation.
10 This program is distributed in the hope it will be useful, but WITHOUT
11 ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 You should have received a copy of the GNU General Public License along with
16 this program; if not, 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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
30 #include <linux/module.h>
31 #include <linux/types.h>
32 #include <linux/init.h>
33 #include <linux/pci.h>
34 #include <linux/vmalloc.h>
35 #include <linux/pagemap.h>
36 #include <linux/delay.h>
37 #include <linux/netdevice.h>
38 #include <linux/tcp.h>
39 #include <linux/ipv6.h>
40 #include <linux/slab.h>
41 #include <net/checksum.h>
42 #include <net/ip6_checksum.h>
43 #include <linux/mii.h>
44 #include <linux/ethtool.h>
45 #include <linux/if_vlan.h>
46 #include <linux/prefetch.h>
50 #define DRV_VERSION "2.0.2-k"
51 char igbvf_driver_name[] = "igbvf";
52 const char igbvf_driver_version[] = DRV_VERSION;
53 static const char igbvf_driver_string[] =
54 "Intel(R) Gigabit Virtual Function Network Driver";
55 static const char igbvf_copyright[] =
56 "Copyright (c) 2009 - 2012 Intel Corporation.";
58 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
59 static int debug = -1;
60 module_param(debug, int, 0);
61 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
63 static int igbvf_poll(struct napi_struct *napi, int budget);
64 static void igbvf_reset(struct igbvf_adapter *);
65 static void igbvf_set_interrupt_capability(struct igbvf_adapter *);
66 static void igbvf_reset_interrupt_capability(struct igbvf_adapter *);
68 static struct igbvf_info igbvf_vf_info = {
72 .init_ops = e1000_init_function_pointers_vf,
75 static struct igbvf_info igbvf_i350_vf_info = {
76 .mac = e1000_vfadapt_i350,
79 .init_ops = e1000_init_function_pointers_vf,
82 static const struct igbvf_info *igbvf_info_tbl[] = {
83 [board_vf] = &igbvf_vf_info,
84 [board_i350_vf] = &igbvf_i350_vf_info,
88 * igbvf_desc_unused - calculate if we have unused descriptors
90 static int igbvf_desc_unused(struct igbvf_ring *ring)
92 if (ring->next_to_clean > ring->next_to_use)
93 return ring->next_to_clean - ring->next_to_use - 1;
95 return ring->count + ring->next_to_clean - ring->next_to_use - 1;
99 * igbvf_receive_skb - helper function to handle Rx indications
100 * @adapter: board private structure
101 * @status: descriptor status field as written by hardware
102 * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
103 * @skb: pointer to sk_buff to be indicated to stack
105 static void igbvf_receive_skb(struct igbvf_adapter *adapter,
106 struct net_device *netdev,
108 u32 status, u16 vlan)
112 if (status & E1000_RXD_STAT_VP) {
113 if ((adapter->flags & IGBVF_FLAG_RX_LB_VLAN_BSWAP) &&
114 (status & E1000_RXDEXT_STATERR_LB))
115 vid = be16_to_cpu(vlan) & E1000_RXD_SPC_VLAN_MASK;
117 vid = le16_to_cpu(vlan) & E1000_RXD_SPC_VLAN_MASK;
118 if (test_bit(vid, adapter->active_vlans))
119 __vlan_hwaccel_put_tag(skb, vid);
122 napi_gro_receive(&adapter->rx_ring->napi, skb);
125 static inline void igbvf_rx_checksum_adv(struct igbvf_adapter *adapter,
126 u32 status_err, struct sk_buff *skb)
128 skb_checksum_none_assert(skb);
130 /* Ignore Checksum bit is set or checksum is disabled through ethtool */
131 if ((status_err & E1000_RXD_STAT_IXSM) ||
132 (adapter->flags & IGBVF_FLAG_RX_CSUM_DISABLED))
135 /* TCP/UDP checksum error bit is set */
137 (E1000_RXDEXT_STATERR_TCPE | E1000_RXDEXT_STATERR_IPE)) {
138 /* let the stack verify checksum errors */
139 adapter->hw_csum_err++;
143 /* It must be a TCP or UDP packet with a valid checksum */
144 if (status_err & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS))
145 skb->ip_summed = CHECKSUM_UNNECESSARY;
147 adapter->hw_csum_good++;
151 * igbvf_alloc_rx_buffers - Replace used receive buffers; packet split
152 * @rx_ring: address of ring structure to repopulate
153 * @cleaned_count: number of buffers to repopulate
155 static void igbvf_alloc_rx_buffers(struct igbvf_ring *rx_ring,
158 struct igbvf_adapter *adapter = rx_ring->adapter;
159 struct net_device *netdev = adapter->netdev;
160 struct pci_dev *pdev = adapter->pdev;
161 union e1000_adv_rx_desc *rx_desc;
162 struct igbvf_buffer *buffer_info;
167 i = rx_ring->next_to_use;
168 buffer_info = &rx_ring->buffer_info[i];
170 if (adapter->rx_ps_hdr_size)
171 bufsz = adapter->rx_ps_hdr_size;
173 bufsz = adapter->rx_buffer_len;
175 while (cleaned_count--) {
176 rx_desc = IGBVF_RX_DESC_ADV(*rx_ring, i);
178 if (adapter->rx_ps_hdr_size && !buffer_info->page_dma) {
179 if (!buffer_info->page) {
180 buffer_info->page = alloc_page(GFP_ATOMIC);
181 if (!buffer_info->page) {
182 adapter->alloc_rx_buff_failed++;
185 buffer_info->page_offset = 0;
187 buffer_info->page_offset ^= PAGE_SIZE / 2;
189 buffer_info->page_dma =
190 dma_map_page(&pdev->dev, buffer_info->page,
191 buffer_info->page_offset,
194 if (dma_mapping_error(&pdev->dev,
195 buffer_info->page_dma)) {
196 __free_page(buffer_info->page);
197 buffer_info->page = NULL;
198 dev_err(&pdev->dev, "RX DMA map failed\n");
203 if (!buffer_info->skb) {
204 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
206 adapter->alloc_rx_buff_failed++;
210 buffer_info->skb = skb;
211 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
214 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
215 dev_kfree_skb(buffer_info->skb);
216 buffer_info->skb = NULL;
217 dev_err(&pdev->dev, "RX DMA map failed\n");
221 /* Refresh the desc even if buffer_addrs didn't change because
222 * each write-back erases this info. */
223 if (adapter->rx_ps_hdr_size) {
224 rx_desc->read.pkt_addr =
225 cpu_to_le64(buffer_info->page_dma);
226 rx_desc->read.hdr_addr = cpu_to_le64(buffer_info->dma);
228 rx_desc->read.pkt_addr =
229 cpu_to_le64(buffer_info->dma);
230 rx_desc->read.hdr_addr = 0;
234 if (i == rx_ring->count)
236 buffer_info = &rx_ring->buffer_info[i];
240 if (rx_ring->next_to_use != i) {
241 rx_ring->next_to_use = i;
243 i = (rx_ring->count - 1);
247 /* Force memory writes to complete before letting h/w
248 * know there are new descriptors to fetch. (Only
249 * applicable for weak-ordered memory model archs,
252 writel(i, adapter->hw.hw_addr + rx_ring->tail);
257 * igbvf_clean_rx_irq - Send received data up the network stack; legacy
258 * @adapter: board private structure
260 * the return value indicates whether actual cleaning was done, there
261 * is no guarantee that everything was cleaned
263 static bool igbvf_clean_rx_irq(struct igbvf_adapter *adapter,
264 int *work_done, int work_to_do)
266 struct igbvf_ring *rx_ring = adapter->rx_ring;
267 struct net_device *netdev = adapter->netdev;
268 struct pci_dev *pdev = adapter->pdev;
269 union e1000_adv_rx_desc *rx_desc, *next_rxd;
270 struct igbvf_buffer *buffer_info, *next_buffer;
272 bool cleaned = false;
273 int cleaned_count = 0;
274 unsigned int total_bytes = 0, total_packets = 0;
276 u32 length, hlen, staterr;
278 i = rx_ring->next_to_clean;
279 rx_desc = IGBVF_RX_DESC_ADV(*rx_ring, i);
280 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
282 while (staterr & E1000_RXD_STAT_DD) {
283 if (*work_done >= work_to_do)
286 rmb(); /* read descriptor and rx_buffer_info after status DD */
288 buffer_info = &rx_ring->buffer_info[i];
290 /* HW will not DMA in data larger than the given buffer, even
291 * if it parses the (NFS, of course) header to be larger. In
292 * that case, it fills the header buffer and spills the rest
295 hlen = (le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.hdr_info) &
296 E1000_RXDADV_HDRBUFLEN_MASK) >> E1000_RXDADV_HDRBUFLEN_SHIFT;
297 if (hlen > adapter->rx_ps_hdr_size)
298 hlen = adapter->rx_ps_hdr_size;
300 length = le16_to_cpu(rx_desc->wb.upper.length);
304 skb = buffer_info->skb;
305 prefetch(skb->data - NET_IP_ALIGN);
306 buffer_info->skb = NULL;
307 if (!adapter->rx_ps_hdr_size) {
308 dma_unmap_single(&pdev->dev, buffer_info->dma,
309 adapter->rx_buffer_len,
311 buffer_info->dma = 0;
312 skb_put(skb, length);
316 if (!skb_shinfo(skb)->nr_frags) {
317 dma_unmap_single(&pdev->dev, buffer_info->dma,
318 adapter->rx_ps_hdr_size,
324 dma_unmap_page(&pdev->dev, buffer_info->page_dma,
327 buffer_info->page_dma = 0;
329 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
331 buffer_info->page_offset,
334 if ((adapter->rx_buffer_len > (PAGE_SIZE / 2)) ||
335 (page_count(buffer_info->page) != 1))
336 buffer_info->page = NULL;
338 get_page(buffer_info->page);
341 skb->data_len += length;
342 skb->truesize += PAGE_SIZE / 2;
346 if (i == rx_ring->count)
348 next_rxd = IGBVF_RX_DESC_ADV(*rx_ring, i);
350 next_buffer = &rx_ring->buffer_info[i];
352 if (!(staterr & E1000_RXD_STAT_EOP)) {
353 buffer_info->skb = next_buffer->skb;
354 buffer_info->dma = next_buffer->dma;
355 next_buffer->skb = skb;
356 next_buffer->dma = 0;
360 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
361 dev_kfree_skb_irq(skb);
365 total_bytes += skb->len;
368 igbvf_rx_checksum_adv(adapter, staterr, skb);
370 skb->protocol = eth_type_trans(skb, netdev);
372 igbvf_receive_skb(adapter, netdev, skb, staterr,
373 rx_desc->wb.upper.vlan);
376 rx_desc->wb.upper.status_error = 0;
378 /* return some buffers to hardware, one at a time is too slow */
379 if (cleaned_count >= IGBVF_RX_BUFFER_WRITE) {
380 igbvf_alloc_rx_buffers(rx_ring, cleaned_count);
384 /* use prefetched values */
386 buffer_info = next_buffer;
388 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
391 rx_ring->next_to_clean = i;
392 cleaned_count = igbvf_desc_unused(rx_ring);
395 igbvf_alloc_rx_buffers(rx_ring, cleaned_count);
397 adapter->total_rx_packets += total_packets;
398 adapter->total_rx_bytes += total_bytes;
399 adapter->net_stats.rx_bytes += total_bytes;
400 adapter->net_stats.rx_packets += total_packets;
404 static void igbvf_put_txbuf(struct igbvf_adapter *adapter,
405 struct igbvf_buffer *buffer_info)
407 if (buffer_info->dma) {
408 if (buffer_info->mapped_as_page)
409 dma_unmap_page(&adapter->pdev->dev,
414 dma_unmap_single(&adapter->pdev->dev,
418 buffer_info->dma = 0;
420 if (buffer_info->skb) {
421 dev_kfree_skb_any(buffer_info->skb);
422 buffer_info->skb = NULL;
424 buffer_info->time_stamp = 0;
428 * igbvf_setup_tx_resources - allocate Tx resources (Descriptors)
429 * @adapter: board private structure
431 * Return 0 on success, negative on failure
433 int igbvf_setup_tx_resources(struct igbvf_adapter *adapter,
434 struct igbvf_ring *tx_ring)
436 struct pci_dev *pdev = adapter->pdev;
439 size = sizeof(struct igbvf_buffer) * tx_ring->count;
440 tx_ring->buffer_info = vzalloc(size);
441 if (!tx_ring->buffer_info)
444 /* round up to nearest 4K */
445 tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
446 tx_ring->size = ALIGN(tx_ring->size, 4096);
448 tx_ring->desc = dma_alloc_coherent(&pdev->dev, tx_ring->size,
449 &tx_ring->dma, GFP_KERNEL);
453 tx_ring->adapter = adapter;
454 tx_ring->next_to_use = 0;
455 tx_ring->next_to_clean = 0;
459 vfree(tx_ring->buffer_info);
460 dev_err(&adapter->pdev->dev,
461 "Unable to allocate memory for the transmit descriptor ring\n");
466 * igbvf_setup_rx_resources - allocate Rx resources (Descriptors)
467 * @adapter: board private structure
469 * Returns 0 on success, negative on failure
471 int igbvf_setup_rx_resources(struct igbvf_adapter *adapter,
472 struct igbvf_ring *rx_ring)
474 struct pci_dev *pdev = adapter->pdev;
477 size = sizeof(struct igbvf_buffer) * rx_ring->count;
478 rx_ring->buffer_info = vzalloc(size);
479 if (!rx_ring->buffer_info)
482 desc_len = sizeof(union e1000_adv_rx_desc);
484 /* Round up to nearest 4K */
485 rx_ring->size = rx_ring->count * desc_len;
486 rx_ring->size = ALIGN(rx_ring->size, 4096);
488 rx_ring->desc = dma_alloc_coherent(&pdev->dev, rx_ring->size,
489 &rx_ring->dma, GFP_KERNEL);
493 rx_ring->next_to_clean = 0;
494 rx_ring->next_to_use = 0;
496 rx_ring->adapter = adapter;
501 vfree(rx_ring->buffer_info);
502 rx_ring->buffer_info = NULL;
503 dev_err(&adapter->pdev->dev,
504 "Unable to allocate memory for the receive descriptor ring\n");
509 * igbvf_clean_tx_ring - Free Tx Buffers
510 * @tx_ring: ring to be cleaned
512 static void igbvf_clean_tx_ring(struct igbvf_ring *tx_ring)
514 struct igbvf_adapter *adapter = tx_ring->adapter;
515 struct igbvf_buffer *buffer_info;
519 if (!tx_ring->buffer_info)
522 /* Free all the Tx ring sk_buffs */
523 for (i = 0; i < tx_ring->count; i++) {
524 buffer_info = &tx_ring->buffer_info[i];
525 igbvf_put_txbuf(adapter, buffer_info);
528 size = sizeof(struct igbvf_buffer) * tx_ring->count;
529 memset(tx_ring->buffer_info, 0, size);
531 /* Zero out the descriptor ring */
532 memset(tx_ring->desc, 0, tx_ring->size);
534 tx_ring->next_to_use = 0;
535 tx_ring->next_to_clean = 0;
537 writel(0, adapter->hw.hw_addr + tx_ring->head);
538 writel(0, adapter->hw.hw_addr + tx_ring->tail);
542 * igbvf_free_tx_resources - Free Tx Resources per Queue
543 * @tx_ring: ring to free resources from
545 * Free all transmit software resources
547 void igbvf_free_tx_resources(struct igbvf_ring *tx_ring)
549 struct pci_dev *pdev = tx_ring->adapter->pdev;
551 igbvf_clean_tx_ring(tx_ring);
553 vfree(tx_ring->buffer_info);
554 tx_ring->buffer_info = NULL;
556 dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
559 tx_ring->desc = NULL;
563 * igbvf_clean_rx_ring - Free Rx Buffers per Queue
564 * @adapter: board private structure
566 static void igbvf_clean_rx_ring(struct igbvf_ring *rx_ring)
568 struct igbvf_adapter *adapter = rx_ring->adapter;
569 struct igbvf_buffer *buffer_info;
570 struct pci_dev *pdev = adapter->pdev;
574 if (!rx_ring->buffer_info)
577 /* Free all the Rx ring sk_buffs */
578 for (i = 0; i < rx_ring->count; i++) {
579 buffer_info = &rx_ring->buffer_info[i];
580 if (buffer_info->dma) {
581 if (adapter->rx_ps_hdr_size){
582 dma_unmap_single(&pdev->dev, buffer_info->dma,
583 adapter->rx_ps_hdr_size,
586 dma_unmap_single(&pdev->dev, buffer_info->dma,
587 adapter->rx_buffer_len,
590 buffer_info->dma = 0;
593 if (buffer_info->skb) {
594 dev_kfree_skb(buffer_info->skb);
595 buffer_info->skb = NULL;
598 if (buffer_info->page) {
599 if (buffer_info->page_dma)
600 dma_unmap_page(&pdev->dev,
601 buffer_info->page_dma,
604 put_page(buffer_info->page);
605 buffer_info->page = NULL;
606 buffer_info->page_dma = 0;
607 buffer_info->page_offset = 0;
611 size = sizeof(struct igbvf_buffer) * rx_ring->count;
612 memset(rx_ring->buffer_info, 0, size);
614 /* Zero out the descriptor ring */
615 memset(rx_ring->desc, 0, rx_ring->size);
617 rx_ring->next_to_clean = 0;
618 rx_ring->next_to_use = 0;
620 writel(0, adapter->hw.hw_addr + rx_ring->head);
621 writel(0, adapter->hw.hw_addr + rx_ring->tail);
625 * igbvf_free_rx_resources - Free Rx Resources
626 * @rx_ring: ring to clean the resources from
628 * Free all receive software resources
631 void igbvf_free_rx_resources(struct igbvf_ring *rx_ring)
633 struct pci_dev *pdev = rx_ring->adapter->pdev;
635 igbvf_clean_rx_ring(rx_ring);
637 vfree(rx_ring->buffer_info);
638 rx_ring->buffer_info = NULL;
640 dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
642 rx_ring->desc = NULL;
646 * igbvf_update_itr - update the dynamic ITR value based on statistics
647 * @adapter: pointer to adapter
648 * @itr_setting: current adapter->itr
649 * @packets: the number of packets during this measurement interval
650 * @bytes: the number of bytes during this measurement interval
652 * Stores a new ITR value based on packets and byte
653 * counts during the last interrupt. The advantage of per interrupt
654 * computation is faster updates and more accurate ITR for the current
655 * traffic pattern. Constants in this function were computed
656 * based on theoretical maximum wire speed and thresholds were set based
657 * on testing data as well as attempting to minimize response time
658 * while increasing bulk throughput.
660 static enum latency_range igbvf_update_itr(struct igbvf_adapter *adapter,
661 enum latency_range itr_setting,
662 int packets, int bytes)
664 enum latency_range retval = itr_setting;
667 goto update_itr_done;
669 switch (itr_setting) {
671 /* handle TSO and jumbo frames */
672 if (bytes/packets > 8000)
673 retval = bulk_latency;
674 else if ((packets < 5) && (bytes > 512))
675 retval = low_latency;
677 case low_latency: /* 50 usec aka 20000 ints/s */
679 /* this if handles the TSO accounting */
680 if (bytes/packets > 8000)
681 retval = bulk_latency;
682 else if ((packets < 10) || ((bytes/packets) > 1200))
683 retval = bulk_latency;
684 else if ((packets > 35))
685 retval = lowest_latency;
686 } else if (bytes/packets > 2000) {
687 retval = bulk_latency;
688 } else if (packets <= 2 && bytes < 512) {
689 retval = lowest_latency;
692 case bulk_latency: /* 250 usec aka 4000 ints/s */
695 retval = low_latency;
696 } else if (bytes < 6000) {
697 retval = low_latency;
708 static int igbvf_range_to_itr(enum latency_range current_range)
712 switch (current_range) {
713 /* counts and packets in update_itr are dependent on these numbers */
715 new_itr = IGBVF_70K_ITR;
718 new_itr = IGBVF_20K_ITR;
721 new_itr = IGBVF_4K_ITR;
724 new_itr = IGBVF_START_ITR;
730 static void igbvf_set_itr(struct igbvf_adapter *adapter)
734 adapter->tx_ring->itr_range =
735 igbvf_update_itr(adapter,
736 adapter->tx_ring->itr_val,
737 adapter->total_tx_packets,
738 adapter->total_tx_bytes);
740 /* conservative mode (itr 3) eliminates the lowest_latency setting */
741 if (adapter->requested_itr == 3 &&
742 adapter->tx_ring->itr_range == lowest_latency)
743 adapter->tx_ring->itr_range = low_latency;
745 new_itr = igbvf_range_to_itr(adapter->tx_ring->itr_range);
748 if (new_itr != adapter->tx_ring->itr_val) {
749 u32 current_itr = adapter->tx_ring->itr_val;
751 * this attempts to bias the interrupt rate towards Bulk
752 * by adding intermediate steps when interrupt rate is
755 new_itr = new_itr > current_itr ?
756 min(current_itr + (new_itr >> 2), new_itr) :
758 adapter->tx_ring->itr_val = new_itr;
760 adapter->tx_ring->set_itr = 1;
763 adapter->rx_ring->itr_range =
764 igbvf_update_itr(adapter, adapter->rx_ring->itr_val,
765 adapter->total_rx_packets,
766 adapter->total_rx_bytes);
767 if (adapter->requested_itr == 3 &&
768 adapter->rx_ring->itr_range == lowest_latency)
769 adapter->rx_ring->itr_range = low_latency;
771 new_itr = igbvf_range_to_itr(adapter->rx_ring->itr_range);
773 if (new_itr != adapter->rx_ring->itr_val) {
774 u32 current_itr = adapter->rx_ring->itr_val;
775 new_itr = new_itr > current_itr ?
776 min(current_itr + (new_itr >> 2), new_itr) :
778 adapter->rx_ring->itr_val = new_itr;
780 adapter->rx_ring->set_itr = 1;
785 * igbvf_clean_tx_irq - Reclaim resources after transmit completes
786 * @adapter: board private structure
788 * returns true if ring is completely cleaned
790 static bool igbvf_clean_tx_irq(struct igbvf_ring *tx_ring)
792 struct igbvf_adapter *adapter = tx_ring->adapter;
793 struct net_device *netdev = adapter->netdev;
794 struct igbvf_buffer *buffer_info;
796 union e1000_adv_tx_desc *tx_desc, *eop_desc;
797 unsigned int total_bytes = 0, total_packets = 0;
798 unsigned int i, count = 0;
799 bool cleaned = false;
801 i = tx_ring->next_to_clean;
802 buffer_info = &tx_ring->buffer_info[i];
803 eop_desc = buffer_info->next_to_watch;
806 /* if next_to_watch is not set then there is no work pending */
810 /* prevent any other reads prior to eop_desc */
811 read_barrier_depends();
813 /* if DD is not set pending work has not been completed */
814 if (!(eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)))
817 /* clear next_to_watch to prevent false hangs */
818 buffer_info->next_to_watch = NULL;
820 for (cleaned = false; !cleaned; count++) {
821 tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
822 cleaned = (tx_desc == eop_desc);
823 skb = buffer_info->skb;
826 unsigned int segs, bytecount;
828 /* gso_segs is currently only valid for tcp */
829 segs = skb_shinfo(skb)->gso_segs ?: 1;
830 /* multiply data chunks by size of headers */
831 bytecount = ((segs - 1) * skb_headlen(skb)) +
833 total_packets += segs;
834 total_bytes += bytecount;
837 igbvf_put_txbuf(adapter, buffer_info);
838 tx_desc->wb.status = 0;
841 if (i == tx_ring->count)
844 buffer_info = &tx_ring->buffer_info[i];
847 eop_desc = buffer_info->next_to_watch;
848 } while (count < tx_ring->count);
850 tx_ring->next_to_clean = i;
852 if (unlikely(count &&
853 netif_carrier_ok(netdev) &&
854 igbvf_desc_unused(tx_ring) >= IGBVF_TX_QUEUE_WAKE)) {
855 /* Make sure that anybody stopping the queue after this
856 * sees the new next_to_clean.
859 if (netif_queue_stopped(netdev) &&
860 !(test_bit(__IGBVF_DOWN, &adapter->state))) {
861 netif_wake_queue(netdev);
862 ++adapter->restart_queue;
866 adapter->net_stats.tx_bytes += total_bytes;
867 adapter->net_stats.tx_packets += total_packets;
868 return count < tx_ring->count;
871 static irqreturn_t igbvf_msix_other(int irq, void *data)
873 struct net_device *netdev = data;
874 struct igbvf_adapter *adapter = netdev_priv(netdev);
875 struct e1000_hw *hw = &adapter->hw;
877 adapter->int_counter1++;
879 netif_carrier_off(netdev);
880 hw->mac.get_link_status = 1;
881 if (!test_bit(__IGBVF_DOWN, &adapter->state))
882 mod_timer(&adapter->watchdog_timer, jiffies + 1);
884 ew32(EIMS, adapter->eims_other);
889 static irqreturn_t igbvf_intr_msix_tx(int irq, void *data)
891 struct net_device *netdev = data;
892 struct igbvf_adapter *adapter = netdev_priv(netdev);
893 struct e1000_hw *hw = &adapter->hw;
894 struct igbvf_ring *tx_ring = adapter->tx_ring;
896 if (tx_ring->set_itr) {
897 writel(tx_ring->itr_val,
898 adapter->hw.hw_addr + tx_ring->itr_register);
899 adapter->tx_ring->set_itr = 0;
902 adapter->total_tx_bytes = 0;
903 adapter->total_tx_packets = 0;
905 /* auto mask will automatically reenable the interrupt when we write
907 if (!igbvf_clean_tx_irq(tx_ring))
908 /* Ring was not completely cleaned, so fire another interrupt */
909 ew32(EICS, tx_ring->eims_value);
911 ew32(EIMS, tx_ring->eims_value);
916 static irqreturn_t igbvf_intr_msix_rx(int irq, void *data)
918 struct net_device *netdev = data;
919 struct igbvf_adapter *adapter = netdev_priv(netdev);
921 adapter->int_counter0++;
923 /* Write the ITR value calculated at the end of the
924 * previous interrupt.
926 if (adapter->rx_ring->set_itr) {
927 writel(adapter->rx_ring->itr_val,
928 adapter->hw.hw_addr + adapter->rx_ring->itr_register);
929 adapter->rx_ring->set_itr = 0;
932 if (napi_schedule_prep(&adapter->rx_ring->napi)) {
933 adapter->total_rx_bytes = 0;
934 adapter->total_rx_packets = 0;
935 __napi_schedule(&adapter->rx_ring->napi);
941 #define IGBVF_NO_QUEUE -1
943 static void igbvf_assign_vector(struct igbvf_adapter *adapter, int rx_queue,
944 int tx_queue, int msix_vector)
946 struct e1000_hw *hw = &adapter->hw;
949 /* 82576 uses a table-based method for assigning vectors.
950 Each queue has a single entry in the table to which we write
951 a vector number along with a "valid" bit. Sadly, the layout
952 of the table is somewhat counterintuitive. */
953 if (rx_queue > IGBVF_NO_QUEUE) {
954 index = (rx_queue >> 1);
955 ivar = array_er32(IVAR0, index);
956 if (rx_queue & 0x1) {
957 /* vector goes into third byte of register */
958 ivar = ivar & 0xFF00FFFF;
959 ivar |= (msix_vector | E1000_IVAR_VALID) << 16;
961 /* vector goes into low byte of register */
962 ivar = ivar & 0xFFFFFF00;
963 ivar |= msix_vector | E1000_IVAR_VALID;
965 adapter->rx_ring[rx_queue].eims_value = 1 << msix_vector;
966 array_ew32(IVAR0, index, ivar);
968 if (tx_queue > IGBVF_NO_QUEUE) {
969 index = (tx_queue >> 1);
970 ivar = array_er32(IVAR0, index);
971 if (tx_queue & 0x1) {
972 /* vector goes into high byte of register */
973 ivar = ivar & 0x00FFFFFF;
974 ivar |= (msix_vector | E1000_IVAR_VALID) << 24;
976 /* vector goes into second byte of register */
977 ivar = ivar & 0xFFFF00FF;
978 ivar |= (msix_vector | E1000_IVAR_VALID) << 8;
980 adapter->tx_ring[tx_queue].eims_value = 1 << msix_vector;
981 array_ew32(IVAR0, index, ivar);
986 * igbvf_configure_msix - Configure MSI-X hardware
988 * igbvf_configure_msix sets up the hardware to properly
989 * generate MSI-X interrupts.
991 static void igbvf_configure_msix(struct igbvf_adapter *adapter)
994 struct e1000_hw *hw = &adapter->hw;
995 struct igbvf_ring *tx_ring = adapter->tx_ring;
996 struct igbvf_ring *rx_ring = adapter->rx_ring;
999 adapter->eims_enable_mask = 0;
1001 igbvf_assign_vector(adapter, IGBVF_NO_QUEUE, 0, vector++);
1002 adapter->eims_enable_mask |= tx_ring->eims_value;
1003 writel(tx_ring->itr_val, hw->hw_addr + tx_ring->itr_register);
1004 igbvf_assign_vector(adapter, 0, IGBVF_NO_QUEUE, vector++);
1005 adapter->eims_enable_mask |= rx_ring->eims_value;
1006 writel(rx_ring->itr_val, hw->hw_addr + rx_ring->itr_register);
1008 /* set vector for other causes, i.e. link changes */
1010 tmp = (vector++ | E1000_IVAR_VALID);
1012 ew32(IVAR_MISC, tmp);
1014 adapter->eims_enable_mask = (1 << (vector)) - 1;
1015 adapter->eims_other = 1 << (vector - 1);
1019 static void igbvf_reset_interrupt_capability(struct igbvf_adapter *adapter)
1021 if (adapter->msix_entries) {
1022 pci_disable_msix(adapter->pdev);
1023 kfree(adapter->msix_entries);
1024 adapter->msix_entries = NULL;
1029 * igbvf_set_interrupt_capability - set MSI or MSI-X if supported
1031 * Attempt to configure interrupts using the best available
1032 * capabilities of the hardware and kernel.
1034 static void igbvf_set_interrupt_capability(struct igbvf_adapter *adapter)
1039 /* we allocate 3 vectors, 1 for tx, 1 for rx, one for pf messages */
1040 adapter->msix_entries = kcalloc(3, sizeof(struct msix_entry),
1042 if (adapter->msix_entries) {
1043 for (i = 0; i < 3; i++)
1044 adapter->msix_entries[i].entry = i;
1046 err = pci_enable_msix(adapter->pdev,
1047 adapter->msix_entries, 3);
1052 dev_err(&adapter->pdev->dev,
1053 "Failed to initialize MSI-X interrupts.\n");
1054 igbvf_reset_interrupt_capability(adapter);
1059 * igbvf_request_msix - Initialize MSI-X interrupts
1061 * igbvf_request_msix allocates MSI-X vectors and requests interrupts from the
1064 static int igbvf_request_msix(struct igbvf_adapter *adapter)
1066 struct net_device *netdev = adapter->netdev;
1067 int err = 0, vector = 0;
1069 if (strlen(netdev->name) < (IFNAMSIZ - 5)) {
1070 sprintf(adapter->tx_ring->name, "%s-tx-0", netdev->name);
1071 sprintf(adapter->rx_ring->name, "%s-rx-0", netdev->name);
1073 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1074 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1077 err = request_irq(adapter->msix_entries[vector].vector,
1078 igbvf_intr_msix_tx, 0, adapter->tx_ring->name,
1083 adapter->tx_ring->itr_register = E1000_EITR(vector);
1084 adapter->tx_ring->itr_val = adapter->current_itr;
1087 err = request_irq(adapter->msix_entries[vector].vector,
1088 igbvf_intr_msix_rx, 0, adapter->rx_ring->name,
1093 adapter->rx_ring->itr_register = E1000_EITR(vector);
1094 adapter->rx_ring->itr_val = adapter->current_itr;
1097 err = request_irq(adapter->msix_entries[vector].vector,
1098 igbvf_msix_other, 0, netdev->name, netdev);
1102 igbvf_configure_msix(adapter);
1109 * igbvf_alloc_queues - Allocate memory for all rings
1110 * @adapter: board private structure to initialize
1112 static int igbvf_alloc_queues(struct igbvf_adapter *adapter)
1114 struct net_device *netdev = adapter->netdev;
1116 adapter->tx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL);
1117 if (!adapter->tx_ring)
1120 adapter->rx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL);
1121 if (!adapter->rx_ring) {
1122 kfree(adapter->tx_ring);
1126 netif_napi_add(netdev, &adapter->rx_ring->napi, igbvf_poll, 64);
1132 * igbvf_request_irq - initialize interrupts
1134 * Attempts to configure interrupts using the best available
1135 * capabilities of the hardware and kernel.
1137 static int igbvf_request_irq(struct igbvf_adapter *adapter)
1141 /* igbvf supports msi-x only */
1142 if (adapter->msix_entries)
1143 err = igbvf_request_msix(adapter);
1148 dev_err(&adapter->pdev->dev,
1149 "Unable to allocate interrupt, Error: %d\n", err);
1154 static void igbvf_free_irq(struct igbvf_adapter *adapter)
1156 struct net_device *netdev = adapter->netdev;
1159 if (adapter->msix_entries) {
1160 for (vector = 0; vector < 3; vector++)
1161 free_irq(adapter->msix_entries[vector].vector, netdev);
1166 * igbvf_irq_disable - Mask off interrupt generation on the NIC
1168 static void igbvf_irq_disable(struct igbvf_adapter *adapter)
1170 struct e1000_hw *hw = &adapter->hw;
1174 if (adapter->msix_entries)
1179 * igbvf_irq_enable - Enable default interrupt generation settings
1181 static void igbvf_irq_enable(struct igbvf_adapter *adapter)
1183 struct e1000_hw *hw = &adapter->hw;
1185 ew32(EIAC, adapter->eims_enable_mask);
1186 ew32(EIAM, adapter->eims_enable_mask);
1187 ew32(EIMS, adapter->eims_enable_mask);
1191 * igbvf_poll - NAPI Rx polling callback
1192 * @napi: struct associated with this polling callback
1193 * @budget: amount of packets driver is allowed to process this poll
1195 static int igbvf_poll(struct napi_struct *napi, int budget)
1197 struct igbvf_ring *rx_ring = container_of(napi, struct igbvf_ring, napi);
1198 struct igbvf_adapter *adapter = rx_ring->adapter;
1199 struct e1000_hw *hw = &adapter->hw;
1202 igbvf_clean_rx_irq(adapter, &work_done, budget);
1204 /* If not enough Rx work done, exit the polling mode */
1205 if (work_done < budget) {
1206 napi_complete(napi);
1208 if (adapter->requested_itr & 3)
1209 igbvf_set_itr(adapter);
1211 if (!test_bit(__IGBVF_DOWN, &adapter->state))
1212 ew32(EIMS, adapter->rx_ring->eims_value);
1219 * igbvf_set_rlpml - set receive large packet maximum length
1220 * @adapter: board private structure
1222 * Configure the maximum size of packets that will be received
1224 static void igbvf_set_rlpml(struct igbvf_adapter *adapter)
1227 struct e1000_hw *hw = &adapter->hw;
1229 max_frame_size = adapter->max_frame_size + VLAN_TAG_SIZE;
1230 e1000_rlpml_set_vf(hw, max_frame_size);
1233 static int igbvf_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1235 struct igbvf_adapter *adapter = netdev_priv(netdev);
1236 struct e1000_hw *hw = &adapter->hw;
1238 if (hw->mac.ops.set_vfta(hw, vid, true)) {
1239 dev_err(&adapter->pdev->dev, "Failed to add vlan id %d\n", vid);
1242 set_bit(vid, adapter->active_vlans);
1246 static int igbvf_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
1248 struct igbvf_adapter *adapter = netdev_priv(netdev);
1249 struct e1000_hw *hw = &adapter->hw;
1251 if (hw->mac.ops.set_vfta(hw, vid, false)) {
1252 dev_err(&adapter->pdev->dev,
1253 "Failed to remove vlan id %d\n", vid);
1256 clear_bit(vid, adapter->active_vlans);
1260 static void igbvf_restore_vlan(struct igbvf_adapter *adapter)
1264 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
1265 igbvf_vlan_rx_add_vid(adapter->netdev, vid);
1269 * igbvf_configure_tx - Configure Transmit Unit after Reset
1270 * @adapter: board private structure
1272 * Configure the Tx unit of the MAC after a reset.
1274 static void igbvf_configure_tx(struct igbvf_adapter *adapter)
1276 struct e1000_hw *hw = &adapter->hw;
1277 struct igbvf_ring *tx_ring = adapter->tx_ring;
1279 u32 txdctl, dca_txctrl;
1281 /* disable transmits */
1282 txdctl = er32(TXDCTL(0));
1283 ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1287 /* Setup the HW Tx Head and Tail descriptor pointers */
1288 ew32(TDLEN(0), tx_ring->count * sizeof(union e1000_adv_tx_desc));
1289 tdba = tx_ring->dma;
1290 ew32(TDBAL(0), (tdba & DMA_BIT_MASK(32)));
1291 ew32(TDBAH(0), (tdba >> 32));
1294 tx_ring->head = E1000_TDH(0);
1295 tx_ring->tail = E1000_TDT(0);
1297 /* Turn off Relaxed Ordering on head write-backs. The writebacks
1298 * MUST be delivered in order or it will completely screw up
1301 dca_txctrl = er32(DCA_TXCTRL(0));
1302 dca_txctrl &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN;
1303 ew32(DCA_TXCTRL(0), dca_txctrl);
1305 /* enable transmits */
1306 txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
1307 ew32(TXDCTL(0), txdctl);
1309 /* Setup Transmit Descriptor Settings for eop descriptor */
1310 adapter->txd_cmd = E1000_ADVTXD_DCMD_EOP | E1000_ADVTXD_DCMD_IFCS;
1312 /* enable Report Status bit */
1313 adapter->txd_cmd |= E1000_ADVTXD_DCMD_RS;
1317 * igbvf_setup_srrctl - configure the receive control registers
1318 * @adapter: Board private structure
1320 static void igbvf_setup_srrctl(struct igbvf_adapter *adapter)
1322 struct e1000_hw *hw = &adapter->hw;
1325 srrctl &= ~(E1000_SRRCTL_DESCTYPE_MASK |
1326 E1000_SRRCTL_BSIZEHDR_MASK |
1327 E1000_SRRCTL_BSIZEPKT_MASK);
1329 /* Enable queue drop to avoid head of line blocking */
1330 srrctl |= E1000_SRRCTL_DROP_EN;
1332 /* Setup buffer sizes */
1333 srrctl |= ALIGN(adapter->rx_buffer_len, 1024) >>
1334 E1000_SRRCTL_BSIZEPKT_SHIFT;
1336 if (adapter->rx_buffer_len < 2048) {
1337 adapter->rx_ps_hdr_size = 0;
1338 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
1340 adapter->rx_ps_hdr_size = 128;
1341 srrctl |= adapter->rx_ps_hdr_size <<
1342 E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
1343 srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
1346 ew32(SRRCTL(0), srrctl);
1350 * igbvf_configure_rx - Configure Receive Unit after Reset
1351 * @adapter: board private structure
1353 * Configure the Rx unit of the MAC after a reset.
1355 static void igbvf_configure_rx(struct igbvf_adapter *adapter)
1357 struct e1000_hw *hw = &adapter->hw;
1358 struct igbvf_ring *rx_ring = adapter->rx_ring;
1362 /* disable receives */
1363 rxdctl = er32(RXDCTL(0));
1364 ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1368 rdlen = rx_ring->count * sizeof(union e1000_adv_rx_desc);
1371 * Setup the HW Rx Head and Tail Descriptor Pointers and
1372 * the Base and Length of the Rx Descriptor Ring
1374 rdba = rx_ring->dma;
1375 ew32(RDBAL(0), (rdba & DMA_BIT_MASK(32)));
1376 ew32(RDBAH(0), (rdba >> 32));
1377 ew32(RDLEN(0), rx_ring->count * sizeof(union e1000_adv_rx_desc));
1378 rx_ring->head = E1000_RDH(0);
1379 rx_ring->tail = E1000_RDT(0);
1383 rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
1384 rxdctl &= 0xFFF00000;
1385 rxdctl |= IGBVF_RX_PTHRESH;
1386 rxdctl |= IGBVF_RX_HTHRESH << 8;
1387 rxdctl |= IGBVF_RX_WTHRESH << 16;
1389 igbvf_set_rlpml(adapter);
1391 /* enable receives */
1392 ew32(RXDCTL(0), rxdctl);
1396 * igbvf_set_multi - Multicast and Promiscuous mode set
1397 * @netdev: network interface device structure
1399 * The set_multi entry point is called whenever the multicast address
1400 * list or the network interface flags are updated. This routine is
1401 * responsible for configuring the hardware for proper multicast,
1402 * promiscuous mode, and all-multi behavior.
1404 static void igbvf_set_multi(struct net_device *netdev)
1406 struct igbvf_adapter *adapter = netdev_priv(netdev);
1407 struct e1000_hw *hw = &adapter->hw;
1408 struct netdev_hw_addr *ha;
1409 u8 *mta_list = NULL;
1412 if (!netdev_mc_empty(netdev)) {
1413 mta_list = kmalloc_array(netdev_mc_count(netdev), ETH_ALEN,
1419 /* prepare a packed array of only addresses. */
1421 netdev_for_each_mc_addr(ha, netdev)
1422 memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
1424 hw->mac.ops.update_mc_addr_list(hw, mta_list, i, 0, 0);
1429 * igbvf_configure - configure the hardware for Rx and Tx
1430 * @adapter: private board structure
1432 static void igbvf_configure(struct igbvf_adapter *adapter)
1434 igbvf_set_multi(adapter->netdev);
1436 igbvf_restore_vlan(adapter);
1438 igbvf_configure_tx(adapter);
1439 igbvf_setup_srrctl(adapter);
1440 igbvf_configure_rx(adapter);
1441 igbvf_alloc_rx_buffers(adapter->rx_ring,
1442 igbvf_desc_unused(adapter->rx_ring));
1445 /* igbvf_reset - bring the hardware into a known good state
1447 * This function boots the hardware and enables some settings that
1448 * require a configuration cycle of the hardware - those cannot be
1449 * set/changed during runtime. After reset the device needs to be
1450 * properly configured for Rx, Tx etc.
1452 static void igbvf_reset(struct igbvf_adapter *adapter)
1454 struct e1000_mac_info *mac = &adapter->hw.mac;
1455 struct net_device *netdev = adapter->netdev;
1456 struct e1000_hw *hw = &adapter->hw;
1458 /* Allow time for pending master requests to run */
1459 if (mac->ops.reset_hw(hw))
1460 dev_err(&adapter->pdev->dev, "PF still resetting\n");
1462 mac->ops.init_hw(hw);
1464 if (is_valid_ether_addr(adapter->hw.mac.addr)) {
1465 memcpy(netdev->dev_addr, adapter->hw.mac.addr,
1467 memcpy(netdev->perm_addr, adapter->hw.mac.addr,
1471 adapter->last_reset = jiffies;
1474 int igbvf_up(struct igbvf_adapter *adapter)
1476 struct e1000_hw *hw = &adapter->hw;
1478 /* hardware has been reset, we need to reload some things */
1479 igbvf_configure(adapter);
1481 clear_bit(__IGBVF_DOWN, &adapter->state);
1483 napi_enable(&adapter->rx_ring->napi);
1484 if (adapter->msix_entries)
1485 igbvf_configure_msix(adapter);
1487 /* Clear any pending interrupts. */
1489 igbvf_irq_enable(adapter);
1491 /* start the watchdog */
1492 hw->mac.get_link_status = 1;
1493 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1499 void igbvf_down(struct igbvf_adapter *adapter)
1501 struct net_device *netdev = adapter->netdev;
1502 struct e1000_hw *hw = &adapter->hw;
1506 * signal that we're down so the interrupt handler does not
1507 * reschedule our watchdog timer
1509 set_bit(__IGBVF_DOWN, &adapter->state);
1511 /* disable receives in the hardware */
1512 rxdctl = er32(RXDCTL(0));
1513 ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1515 netif_stop_queue(netdev);
1517 /* disable transmits in the hardware */
1518 txdctl = er32(TXDCTL(0));
1519 ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1521 /* flush both disables and wait for them to finish */
1525 napi_disable(&adapter->rx_ring->napi);
1527 igbvf_irq_disable(adapter);
1529 del_timer_sync(&adapter->watchdog_timer);
1531 netif_carrier_off(netdev);
1533 /* record the stats before reset*/
1534 igbvf_update_stats(adapter);
1536 adapter->link_speed = 0;
1537 adapter->link_duplex = 0;
1539 igbvf_reset(adapter);
1540 igbvf_clean_tx_ring(adapter->tx_ring);
1541 igbvf_clean_rx_ring(adapter->rx_ring);
1544 void igbvf_reinit_locked(struct igbvf_adapter *adapter)
1547 while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
1549 igbvf_down(adapter);
1551 clear_bit(__IGBVF_RESETTING, &adapter->state);
1555 * igbvf_sw_init - Initialize general software structures (struct igbvf_adapter)
1556 * @adapter: board private structure to initialize
1558 * igbvf_sw_init initializes the Adapter private data structure.
1559 * Fields are initialized based on PCI device information and
1560 * OS network device settings (MTU size).
1562 static int igbvf_sw_init(struct igbvf_adapter *adapter)
1564 struct net_device *netdev = adapter->netdev;
1567 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
1568 adapter->rx_ps_hdr_size = 0;
1569 adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1570 adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
1572 adapter->tx_int_delay = 8;
1573 adapter->tx_abs_int_delay = 32;
1574 adapter->rx_int_delay = 0;
1575 adapter->rx_abs_int_delay = 8;
1576 adapter->requested_itr = 3;
1577 adapter->current_itr = IGBVF_START_ITR;
1579 /* Set various function pointers */
1580 adapter->ei->init_ops(&adapter->hw);
1582 rc = adapter->hw.mac.ops.init_params(&adapter->hw);
1586 rc = adapter->hw.mbx.ops.init_params(&adapter->hw);
1590 igbvf_set_interrupt_capability(adapter);
1592 if (igbvf_alloc_queues(adapter))
1595 spin_lock_init(&adapter->tx_queue_lock);
1597 /* Explicitly disable IRQ since the NIC can be in any state. */
1598 igbvf_irq_disable(adapter);
1600 spin_lock_init(&adapter->stats_lock);
1602 set_bit(__IGBVF_DOWN, &adapter->state);
1606 static void igbvf_initialize_last_counter_stats(struct igbvf_adapter *adapter)
1608 struct e1000_hw *hw = &adapter->hw;
1610 adapter->stats.last_gprc = er32(VFGPRC);
1611 adapter->stats.last_gorc = er32(VFGORC);
1612 adapter->stats.last_gptc = er32(VFGPTC);
1613 adapter->stats.last_gotc = er32(VFGOTC);
1614 adapter->stats.last_mprc = er32(VFMPRC);
1615 adapter->stats.last_gotlbc = er32(VFGOTLBC);
1616 adapter->stats.last_gptlbc = er32(VFGPTLBC);
1617 adapter->stats.last_gorlbc = er32(VFGORLBC);
1618 adapter->stats.last_gprlbc = er32(VFGPRLBC);
1620 adapter->stats.base_gprc = er32(VFGPRC);
1621 adapter->stats.base_gorc = er32(VFGORC);
1622 adapter->stats.base_gptc = er32(VFGPTC);
1623 adapter->stats.base_gotc = er32(VFGOTC);
1624 adapter->stats.base_mprc = er32(VFMPRC);
1625 adapter->stats.base_gotlbc = er32(VFGOTLBC);
1626 adapter->stats.base_gptlbc = er32(VFGPTLBC);
1627 adapter->stats.base_gorlbc = er32(VFGORLBC);
1628 adapter->stats.base_gprlbc = er32(VFGPRLBC);
1632 * igbvf_open - Called when a network interface is made active
1633 * @netdev: network interface device structure
1635 * Returns 0 on success, negative value on failure
1637 * The open entry point is called when a network interface is made
1638 * active by the system (IFF_UP). At this point all resources needed
1639 * for transmit and receive operations are allocated, the interrupt
1640 * handler is registered with the OS, the watchdog timer is started,
1641 * and the stack is notified that the interface is ready.
1643 static int igbvf_open(struct net_device *netdev)
1645 struct igbvf_adapter *adapter = netdev_priv(netdev);
1646 struct e1000_hw *hw = &adapter->hw;
1649 /* disallow open during test */
1650 if (test_bit(__IGBVF_TESTING, &adapter->state))
1653 /* allocate transmit descriptors */
1654 err = igbvf_setup_tx_resources(adapter, adapter->tx_ring);
1658 /* allocate receive descriptors */
1659 err = igbvf_setup_rx_resources(adapter, adapter->rx_ring);
1664 * before we allocate an interrupt, we must be ready to handle it.
1665 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1666 * as soon as we call pci_request_irq, so we have to setup our
1667 * clean_rx handler before we do so.
1669 igbvf_configure(adapter);
1671 err = igbvf_request_irq(adapter);
1675 /* From here on the code is the same as igbvf_up() */
1676 clear_bit(__IGBVF_DOWN, &adapter->state);
1678 napi_enable(&adapter->rx_ring->napi);
1680 /* clear any pending interrupts */
1683 igbvf_irq_enable(adapter);
1685 /* start the watchdog */
1686 hw->mac.get_link_status = 1;
1687 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1692 igbvf_free_rx_resources(adapter->rx_ring);
1694 igbvf_free_tx_resources(adapter->tx_ring);
1696 igbvf_reset(adapter);
1702 * igbvf_close - Disables a network interface
1703 * @netdev: network interface device structure
1705 * Returns 0, this is not allowed to fail
1707 * The close entry point is called when an interface is de-activated
1708 * by the OS. The hardware is still under the drivers control, but
1709 * needs to be disabled. A global MAC reset is issued to stop the
1710 * hardware, and all transmit and receive resources are freed.
1712 static int igbvf_close(struct net_device *netdev)
1714 struct igbvf_adapter *adapter = netdev_priv(netdev);
1716 WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
1717 igbvf_down(adapter);
1719 igbvf_free_irq(adapter);
1721 igbvf_free_tx_resources(adapter->tx_ring);
1722 igbvf_free_rx_resources(adapter->rx_ring);
1727 * igbvf_set_mac - Change the Ethernet Address of the NIC
1728 * @netdev: network interface device structure
1729 * @p: pointer to an address structure
1731 * Returns 0 on success, negative on failure
1733 static int igbvf_set_mac(struct net_device *netdev, void *p)
1735 struct igbvf_adapter *adapter = netdev_priv(netdev);
1736 struct e1000_hw *hw = &adapter->hw;
1737 struct sockaddr *addr = p;
1739 if (!is_valid_ether_addr(addr->sa_data))
1740 return -EADDRNOTAVAIL;
1742 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
1744 hw->mac.ops.rar_set(hw, hw->mac.addr, 0);
1746 if (memcmp(addr->sa_data, hw->mac.addr, 6))
1747 return -EADDRNOTAVAIL;
1749 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
1754 #define UPDATE_VF_COUNTER(reg, name) \
1756 u32 current_counter = er32(reg); \
1757 if (current_counter < adapter->stats.last_##name) \
1758 adapter->stats.name += 0x100000000LL; \
1759 adapter->stats.last_##name = current_counter; \
1760 adapter->stats.name &= 0xFFFFFFFF00000000LL; \
1761 adapter->stats.name |= current_counter; \
1765 * igbvf_update_stats - Update the board statistics counters
1766 * @adapter: board private structure
1768 void igbvf_update_stats(struct igbvf_adapter *adapter)
1770 struct e1000_hw *hw = &adapter->hw;
1771 struct pci_dev *pdev = adapter->pdev;
1774 * Prevent stats update while adapter is being reset, link is down
1775 * or if the pci connection is down.
1777 if (adapter->link_speed == 0)
1780 if (test_bit(__IGBVF_RESETTING, &adapter->state))
1783 if (pci_channel_offline(pdev))
1786 UPDATE_VF_COUNTER(VFGPRC, gprc);
1787 UPDATE_VF_COUNTER(VFGORC, gorc);
1788 UPDATE_VF_COUNTER(VFGPTC, gptc);
1789 UPDATE_VF_COUNTER(VFGOTC, gotc);
1790 UPDATE_VF_COUNTER(VFMPRC, mprc);
1791 UPDATE_VF_COUNTER(VFGOTLBC, gotlbc);
1792 UPDATE_VF_COUNTER(VFGPTLBC, gptlbc);
1793 UPDATE_VF_COUNTER(VFGORLBC, gorlbc);
1794 UPDATE_VF_COUNTER(VFGPRLBC, gprlbc);
1796 /* Fill out the OS statistics structure */
1797 adapter->net_stats.multicast = adapter->stats.mprc;
1800 static void igbvf_print_link_info(struct igbvf_adapter *adapter)
1802 dev_info(&adapter->pdev->dev, "Link is Up %d Mbps %s Duplex\n",
1803 adapter->link_speed,
1804 adapter->link_duplex == FULL_DUPLEX ? "Full" : "Half");
1807 static bool igbvf_has_link(struct igbvf_adapter *adapter)
1809 struct e1000_hw *hw = &adapter->hw;
1810 s32 ret_val = E1000_SUCCESS;
1813 /* If interface is down, stay link down */
1814 if (test_bit(__IGBVF_DOWN, &adapter->state))
1817 ret_val = hw->mac.ops.check_for_link(hw);
1818 link_active = !hw->mac.get_link_status;
1820 /* if check for link returns error we will need to reset */
1821 if (ret_val && time_after(jiffies, adapter->last_reset + (10 * HZ)))
1822 schedule_work(&adapter->reset_task);
1828 * igbvf_watchdog - Timer Call-back
1829 * @data: pointer to adapter cast into an unsigned long
1831 static void igbvf_watchdog(unsigned long data)
1833 struct igbvf_adapter *adapter = (struct igbvf_adapter *) data;
1835 /* Do the rest outside of interrupt context */
1836 schedule_work(&adapter->watchdog_task);
1839 static void igbvf_watchdog_task(struct work_struct *work)
1841 struct igbvf_adapter *adapter = container_of(work,
1842 struct igbvf_adapter,
1844 struct net_device *netdev = adapter->netdev;
1845 struct e1000_mac_info *mac = &adapter->hw.mac;
1846 struct igbvf_ring *tx_ring = adapter->tx_ring;
1847 struct e1000_hw *hw = &adapter->hw;
1851 link = igbvf_has_link(adapter);
1854 if (!netif_carrier_ok(netdev)) {
1855 mac->ops.get_link_up_info(&adapter->hw,
1856 &adapter->link_speed,
1857 &adapter->link_duplex);
1858 igbvf_print_link_info(adapter);
1860 netif_carrier_on(netdev);
1861 netif_wake_queue(netdev);
1864 if (netif_carrier_ok(netdev)) {
1865 adapter->link_speed = 0;
1866 adapter->link_duplex = 0;
1867 dev_info(&adapter->pdev->dev, "Link is Down\n");
1868 netif_carrier_off(netdev);
1869 netif_stop_queue(netdev);
1873 if (netif_carrier_ok(netdev)) {
1874 igbvf_update_stats(adapter);
1876 tx_pending = (igbvf_desc_unused(tx_ring) + 1 <
1880 * We've lost link, so the controller stops DMA,
1881 * but we've got queued Tx work that's never going
1882 * to get done, so reset controller to flush Tx.
1883 * (Do the reset outside of interrupt context).
1885 adapter->tx_timeout_count++;
1886 schedule_work(&adapter->reset_task);
1890 /* Cause software interrupt to ensure Rx ring is cleaned */
1891 ew32(EICS, adapter->rx_ring->eims_value);
1893 /* Reset the timer */
1894 if (!test_bit(__IGBVF_DOWN, &adapter->state))
1895 mod_timer(&adapter->watchdog_timer,
1896 round_jiffies(jiffies + (2 * HZ)));
1899 #define IGBVF_TX_FLAGS_CSUM 0x00000001
1900 #define IGBVF_TX_FLAGS_VLAN 0x00000002
1901 #define IGBVF_TX_FLAGS_TSO 0x00000004
1902 #define IGBVF_TX_FLAGS_IPV4 0x00000008
1903 #define IGBVF_TX_FLAGS_VLAN_MASK 0xffff0000
1904 #define IGBVF_TX_FLAGS_VLAN_SHIFT 16
1906 static int igbvf_tso(struct igbvf_adapter *adapter,
1907 struct igbvf_ring *tx_ring,
1908 struct sk_buff *skb, u32 tx_flags, u8 *hdr_len)
1910 struct e1000_adv_tx_context_desc *context_desc;
1913 struct igbvf_buffer *buffer_info;
1914 u32 info = 0, tu_cmd = 0;
1915 u32 mss_l4len_idx, l4len;
1918 if (skb_header_cloned(skb)) {
1919 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1921 dev_err(&adapter->pdev->dev,
1922 "igbvf_tso returning an error\n");
1927 l4len = tcp_hdrlen(skb);
1930 if (skb->protocol == htons(ETH_P_IP)) {
1931 struct iphdr *iph = ip_hdr(skb);
1934 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
1938 } else if (skb_is_gso_v6(skb)) {
1939 ipv6_hdr(skb)->payload_len = 0;
1940 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
1941 &ipv6_hdr(skb)->daddr,
1945 i = tx_ring->next_to_use;
1947 buffer_info = &tx_ring->buffer_info[i];
1948 context_desc = IGBVF_TX_CTXTDESC_ADV(*tx_ring, i);
1949 /* VLAN MACLEN IPLEN */
1950 if (tx_flags & IGBVF_TX_FLAGS_VLAN)
1951 info |= (tx_flags & IGBVF_TX_FLAGS_VLAN_MASK);
1952 info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
1953 *hdr_len += skb_network_offset(skb);
1954 info |= (skb_transport_header(skb) - skb_network_header(skb));
1955 *hdr_len += (skb_transport_header(skb) - skb_network_header(skb));
1956 context_desc->vlan_macip_lens = cpu_to_le32(info);
1958 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
1959 tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
1961 if (skb->protocol == htons(ETH_P_IP))
1962 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
1963 tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
1965 context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
1968 mss_l4len_idx = (skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT);
1969 mss_l4len_idx |= (l4len << E1000_ADVTXD_L4LEN_SHIFT);
1971 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
1972 context_desc->seqnum_seed = 0;
1974 buffer_info->time_stamp = jiffies;
1975 buffer_info->dma = 0;
1977 if (i == tx_ring->count)
1980 tx_ring->next_to_use = i;
1985 static inline bool igbvf_tx_csum(struct igbvf_adapter *adapter,
1986 struct igbvf_ring *tx_ring,
1987 struct sk_buff *skb, u32 tx_flags)
1989 struct e1000_adv_tx_context_desc *context_desc;
1991 struct igbvf_buffer *buffer_info;
1992 u32 info = 0, tu_cmd = 0;
1994 if ((skb->ip_summed == CHECKSUM_PARTIAL) ||
1995 (tx_flags & IGBVF_TX_FLAGS_VLAN)) {
1996 i = tx_ring->next_to_use;
1997 buffer_info = &tx_ring->buffer_info[i];
1998 context_desc = IGBVF_TX_CTXTDESC_ADV(*tx_ring, i);
2000 if (tx_flags & IGBVF_TX_FLAGS_VLAN)
2001 info |= (tx_flags & IGBVF_TX_FLAGS_VLAN_MASK);
2003 info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
2004 if (skb->ip_summed == CHECKSUM_PARTIAL)
2005 info |= (skb_transport_header(skb) -
2006 skb_network_header(skb));
2009 context_desc->vlan_macip_lens = cpu_to_le32(info);
2011 tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
2013 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2014 switch (skb->protocol) {
2015 case __constant_htons(ETH_P_IP):
2016 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
2017 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
2018 tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
2020 case __constant_htons(ETH_P_IPV6):
2021 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
2022 tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
2029 context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
2030 context_desc->seqnum_seed = 0;
2031 context_desc->mss_l4len_idx = 0;
2033 buffer_info->time_stamp = jiffies;
2034 buffer_info->dma = 0;
2036 if (i == tx_ring->count)
2038 tx_ring->next_to_use = i;
2046 static int igbvf_maybe_stop_tx(struct net_device *netdev, int size)
2048 struct igbvf_adapter *adapter = netdev_priv(netdev);
2050 /* there is enough descriptors then we don't need to worry */
2051 if (igbvf_desc_unused(adapter->tx_ring) >= size)
2054 netif_stop_queue(netdev);
2058 /* We need to check again just in case room has been made available */
2059 if (igbvf_desc_unused(adapter->tx_ring) < size)
2062 netif_wake_queue(netdev);
2064 ++adapter->restart_queue;
2068 #define IGBVF_MAX_TXD_PWR 16
2069 #define IGBVF_MAX_DATA_PER_TXD (1 << IGBVF_MAX_TXD_PWR)
2071 static inline int igbvf_tx_map_adv(struct igbvf_adapter *adapter,
2072 struct igbvf_ring *tx_ring,
2073 struct sk_buff *skb)
2075 struct igbvf_buffer *buffer_info;
2076 struct pci_dev *pdev = adapter->pdev;
2077 unsigned int len = skb_headlen(skb);
2078 unsigned int count = 0, i;
2081 i = tx_ring->next_to_use;
2083 buffer_info = &tx_ring->buffer_info[i];
2084 BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD);
2085 buffer_info->length = len;
2086 /* set time_stamp *before* dma to help avoid a possible race */
2087 buffer_info->time_stamp = jiffies;
2088 buffer_info->mapped_as_page = false;
2089 buffer_info->dma = dma_map_single(&pdev->dev, skb->data, len,
2091 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2095 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) {
2096 const struct skb_frag_struct *frag;
2100 if (i == tx_ring->count)
2103 frag = &skb_shinfo(skb)->frags[f];
2104 len = skb_frag_size(frag);
2106 buffer_info = &tx_ring->buffer_info[i];
2107 BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD);
2108 buffer_info->length = len;
2109 buffer_info->time_stamp = jiffies;
2110 buffer_info->mapped_as_page = true;
2111 buffer_info->dma = skb_frag_dma_map(&pdev->dev, frag, 0, len,
2113 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2117 tx_ring->buffer_info[i].skb = skb;
2122 dev_err(&pdev->dev, "TX DMA map failed\n");
2124 /* clear timestamp and dma mappings for failed buffer_info mapping */
2125 buffer_info->dma = 0;
2126 buffer_info->time_stamp = 0;
2127 buffer_info->length = 0;
2128 buffer_info->mapped_as_page = false;
2132 /* clear timestamp and dma mappings for remaining portion of packet */
2135 i += tx_ring->count;
2137 buffer_info = &tx_ring->buffer_info[i];
2138 igbvf_put_txbuf(adapter, buffer_info);
2144 static inline void igbvf_tx_queue_adv(struct igbvf_adapter *adapter,
2145 struct igbvf_ring *tx_ring,
2146 int tx_flags, int count,
2147 unsigned int first, u32 paylen,
2150 union e1000_adv_tx_desc *tx_desc = NULL;
2151 struct igbvf_buffer *buffer_info;
2152 u32 olinfo_status = 0, cmd_type_len;
2155 cmd_type_len = (E1000_ADVTXD_DTYP_DATA | E1000_ADVTXD_DCMD_IFCS |
2156 E1000_ADVTXD_DCMD_DEXT);
2158 if (tx_flags & IGBVF_TX_FLAGS_VLAN)
2159 cmd_type_len |= E1000_ADVTXD_DCMD_VLE;
2161 if (tx_flags & IGBVF_TX_FLAGS_TSO) {
2162 cmd_type_len |= E1000_ADVTXD_DCMD_TSE;
2164 /* insert tcp checksum */
2165 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2167 /* insert ip checksum */
2168 if (tx_flags & IGBVF_TX_FLAGS_IPV4)
2169 olinfo_status |= E1000_TXD_POPTS_IXSM << 8;
2171 } else if (tx_flags & IGBVF_TX_FLAGS_CSUM) {
2172 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2175 olinfo_status |= ((paylen - hdr_len) << E1000_ADVTXD_PAYLEN_SHIFT);
2177 i = tx_ring->next_to_use;
2179 buffer_info = &tx_ring->buffer_info[i];
2180 tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
2181 tx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
2182 tx_desc->read.cmd_type_len =
2183 cpu_to_le32(cmd_type_len | buffer_info->length);
2184 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
2186 if (i == tx_ring->count)
2190 tx_desc->read.cmd_type_len |= cpu_to_le32(adapter->txd_cmd);
2191 /* Force memory writes to complete before letting h/w
2192 * know there are new descriptors to fetch. (Only
2193 * applicable for weak-ordered memory model archs,
2194 * such as IA-64). */
2197 tx_ring->buffer_info[first].next_to_watch = tx_desc;
2198 tx_ring->next_to_use = i;
2199 writel(i, adapter->hw.hw_addr + tx_ring->tail);
2200 /* we need this if more than one processor can write to our tail
2201 * at a time, it syncronizes IO on IA64/Altix systems */
2205 static netdev_tx_t igbvf_xmit_frame_ring_adv(struct sk_buff *skb,
2206 struct net_device *netdev,
2207 struct igbvf_ring *tx_ring)
2209 struct igbvf_adapter *adapter = netdev_priv(netdev);
2210 unsigned int first, tx_flags = 0;
2215 if (test_bit(__IGBVF_DOWN, &adapter->state)) {
2216 dev_kfree_skb_any(skb);
2217 return NETDEV_TX_OK;
2220 if (skb->len <= 0) {
2221 dev_kfree_skb_any(skb);
2222 return NETDEV_TX_OK;
2226 * need: count + 4 desc gap to keep tail from touching
2227 * + 2 desc gap to keep tail from touching head,
2228 * + 1 desc for skb->data,
2229 * + 1 desc for context descriptor,
2230 * head, otherwise try next time
2232 if (igbvf_maybe_stop_tx(netdev, skb_shinfo(skb)->nr_frags + 4)) {
2233 /* this is a hard error */
2234 return NETDEV_TX_BUSY;
2237 if (vlan_tx_tag_present(skb)) {
2238 tx_flags |= IGBVF_TX_FLAGS_VLAN;
2239 tx_flags |= (vlan_tx_tag_get(skb) << IGBVF_TX_FLAGS_VLAN_SHIFT);
2242 if (skb->protocol == htons(ETH_P_IP))
2243 tx_flags |= IGBVF_TX_FLAGS_IPV4;
2245 first = tx_ring->next_to_use;
2247 tso = skb_is_gso(skb) ?
2248 igbvf_tso(adapter, tx_ring, skb, tx_flags, &hdr_len) : 0;
2249 if (unlikely(tso < 0)) {
2250 dev_kfree_skb_any(skb);
2251 return NETDEV_TX_OK;
2255 tx_flags |= IGBVF_TX_FLAGS_TSO;
2256 else if (igbvf_tx_csum(adapter, tx_ring, skb, tx_flags) &&
2257 (skb->ip_summed == CHECKSUM_PARTIAL))
2258 tx_flags |= IGBVF_TX_FLAGS_CSUM;
2261 * count reflects descriptors mapped, if 0 then mapping error
2262 * has occurred and we need to rewind the descriptor queue
2264 count = igbvf_tx_map_adv(adapter, tx_ring, skb);
2267 igbvf_tx_queue_adv(adapter, tx_ring, tx_flags, count,
2268 first, skb->len, hdr_len);
2269 /* Make sure there is space in the ring for the next send. */
2270 igbvf_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 4);
2272 dev_kfree_skb_any(skb);
2273 tx_ring->buffer_info[first].time_stamp = 0;
2274 tx_ring->next_to_use = first;
2277 return NETDEV_TX_OK;
2280 static netdev_tx_t igbvf_xmit_frame(struct sk_buff *skb,
2281 struct net_device *netdev)
2283 struct igbvf_adapter *adapter = netdev_priv(netdev);
2284 struct igbvf_ring *tx_ring;
2286 if (test_bit(__IGBVF_DOWN, &adapter->state)) {
2287 dev_kfree_skb_any(skb);
2288 return NETDEV_TX_OK;
2291 tx_ring = &adapter->tx_ring[0];
2293 return igbvf_xmit_frame_ring_adv(skb, netdev, tx_ring);
2297 * igbvf_tx_timeout - Respond to a Tx Hang
2298 * @netdev: network interface device structure
2300 static void igbvf_tx_timeout(struct net_device *netdev)
2302 struct igbvf_adapter *adapter = netdev_priv(netdev);
2304 /* Do the reset outside of interrupt context */
2305 adapter->tx_timeout_count++;
2306 schedule_work(&adapter->reset_task);
2309 static void igbvf_reset_task(struct work_struct *work)
2311 struct igbvf_adapter *adapter;
2312 adapter = container_of(work, struct igbvf_adapter, reset_task);
2314 igbvf_reinit_locked(adapter);
2318 * igbvf_get_stats - Get System Network Statistics
2319 * @netdev: network interface device structure
2321 * Returns the address of the device statistics structure.
2322 * The statistics are actually updated from the timer callback.
2324 static struct net_device_stats *igbvf_get_stats(struct net_device *netdev)
2326 struct igbvf_adapter *adapter = netdev_priv(netdev);
2328 /* only return the current stats */
2329 return &adapter->net_stats;
2333 * igbvf_change_mtu - Change the Maximum Transfer Unit
2334 * @netdev: network interface device structure
2335 * @new_mtu: new value for maximum frame size
2337 * Returns 0 on success, negative on failure
2339 static int igbvf_change_mtu(struct net_device *netdev, int new_mtu)
2341 struct igbvf_adapter *adapter = netdev_priv(netdev);
2342 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
2344 if ((new_mtu < 68) || (max_frame > MAX_JUMBO_FRAME_SIZE)) {
2345 dev_err(&adapter->pdev->dev, "Invalid MTU setting\n");
2349 #define MAX_STD_JUMBO_FRAME_SIZE 9234
2350 if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
2351 dev_err(&adapter->pdev->dev, "MTU > 9216 not supported.\n");
2355 while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
2357 /* igbvf_down has a dependency on max_frame_size */
2358 adapter->max_frame_size = max_frame;
2359 if (netif_running(netdev))
2360 igbvf_down(adapter);
2363 * NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
2364 * means we reserve 2 more, this pushes us to allocate from the next
2366 * i.e. RXBUFFER_2048 --> size-4096 slab
2367 * However with the new *_jumbo_rx* routines, jumbo receives will use
2371 if (max_frame <= 1024)
2372 adapter->rx_buffer_len = 1024;
2373 else if (max_frame <= 2048)
2374 adapter->rx_buffer_len = 2048;
2376 #if (PAGE_SIZE / 2) > 16384
2377 adapter->rx_buffer_len = 16384;
2379 adapter->rx_buffer_len = PAGE_SIZE / 2;
2383 /* adjust allocation if LPE protects us, and we aren't using SBP */
2384 if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
2385 (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
2386 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN +
2389 dev_info(&adapter->pdev->dev, "changing MTU from %d to %d\n",
2390 netdev->mtu, new_mtu);
2391 netdev->mtu = new_mtu;
2393 if (netif_running(netdev))
2396 igbvf_reset(adapter);
2398 clear_bit(__IGBVF_RESETTING, &adapter->state);
2403 static int igbvf_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
2411 static int igbvf_suspend(struct pci_dev *pdev, pm_message_t state)
2413 struct net_device *netdev = pci_get_drvdata(pdev);
2414 struct igbvf_adapter *adapter = netdev_priv(netdev);
2419 netif_device_detach(netdev);
2421 if (netif_running(netdev)) {
2422 WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
2423 igbvf_down(adapter);
2424 igbvf_free_irq(adapter);
2428 retval = pci_save_state(pdev);
2433 pci_disable_device(pdev);
2439 static int igbvf_resume(struct pci_dev *pdev)
2441 struct net_device *netdev = pci_get_drvdata(pdev);
2442 struct igbvf_adapter *adapter = netdev_priv(netdev);
2445 pci_restore_state(pdev);
2446 err = pci_enable_device_mem(pdev);
2448 dev_err(&pdev->dev, "Cannot enable PCI device from suspend\n");
2452 pci_set_master(pdev);
2454 if (netif_running(netdev)) {
2455 err = igbvf_request_irq(adapter);
2460 igbvf_reset(adapter);
2462 if (netif_running(netdev))
2465 netif_device_attach(netdev);
2471 static void igbvf_shutdown(struct pci_dev *pdev)
2473 igbvf_suspend(pdev, PMSG_SUSPEND);
2476 #ifdef CONFIG_NET_POLL_CONTROLLER
2478 * Polling 'interrupt' - used by things like netconsole to send skbs
2479 * without having to re-enable interrupts. It's not called while
2480 * the interrupt routine is executing.
2482 static void igbvf_netpoll(struct net_device *netdev)
2484 struct igbvf_adapter *adapter = netdev_priv(netdev);
2486 disable_irq(adapter->pdev->irq);
2488 igbvf_clean_tx_irq(adapter->tx_ring);
2490 enable_irq(adapter->pdev->irq);
2495 * igbvf_io_error_detected - called when PCI error is detected
2496 * @pdev: Pointer to PCI device
2497 * @state: The current pci connection state
2499 * This function is called after a PCI bus error affecting
2500 * this device has been detected.
2502 static pci_ers_result_t igbvf_io_error_detected(struct pci_dev *pdev,
2503 pci_channel_state_t state)
2505 struct net_device *netdev = pci_get_drvdata(pdev);
2506 struct igbvf_adapter *adapter = netdev_priv(netdev);
2508 netif_device_detach(netdev);
2510 if (state == pci_channel_io_perm_failure)
2511 return PCI_ERS_RESULT_DISCONNECT;
2513 if (netif_running(netdev))
2514 igbvf_down(adapter);
2515 pci_disable_device(pdev);
2517 /* Request a slot slot reset. */
2518 return PCI_ERS_RESULT_NEED_RESET;
2522 * igbvf_io_slot_reset - called after the pci bus has been reset.
2523 * @pdev: Pointer to PCI device
2525 * Restart the card from scratch, as if from a cold-boot. Implementation
2526 * resembles the first-half of the igbvf_resume routine.
2528 static pci_ers_result_t igbvf_io_slot_reset(struct pci_dev *pdev)
2530 struct net_device *netdev = pci_get_drvdata(pdev);
2531 struct igbvf_adapter *adapter = netdev_priv(netdev);
2533 if (pci_enable_device_mem(pdev)) {
2535 "Cannot re-enable PCI device after reset.\n");
2536 return PCI_ERS_RESULT_DISCONNECT;
2538 pci_set_master(pdev);
2540 igbvf_reset(adapter);
2542 return PCI_ERS_RESULT_RECOVERED;
2546 * igbvf_io_resume - called when traffic can start flowing again.
2547 * @pdev: Pointer to PCI device
2549 * This callback is called when the error recovery driver tells us that
2550 * its OK to resume normal operation. Implementation resembles the
2551 * second-half of the igbvf_resume routine.
2553 static void igbvf_io_resume(struct pci_dev *pdev)
2555 struct net_device *netdev = pci_get_drvdata(pdev);
2556 struct igbvf_adapter *adapter = netdev_priv(netdev);
2558 if (netif_running(netdev)) {
2559 if (igbvf_up(adapter)) {
2561 "can't bring device back up after reset\n");
2566 netif_device_attach(netdev);
2569 static void igbvf_print_device_info(struct igbvf_adapter *adapter)
2571 struct e1000_hw *hw = &adapter->hw;
2572 struct net_device *netdev = adapter->netdev;
2573 struct pci_dev *pdev = adapter->pdev;
2575 if (hw->mac.type == e1000_vfadapt_i350)
2576 dev_info(&pdev->dev, "Intel(R) I350 Virtual Function\n");
2578 dev_info(&pdev->dev, "Intel(R) 82576 Virtual Function\n");
2579 dev_info(&pdev->dev, "Address: %pM\n", netdev->dev_addr);
2582 static int igbvf_set_features(struct net_device *netdev,
2583 netdev_features_t features)
2585 struct igbvf_adapter *adapter = netdev_priv(netdev);
2587 if (features & NETIF_F_RXCSUM)
2588 adapter->flags &= ~IGBVF_FLAG_RX_CSUM_DISABLED;
2590 adapter->flags |= IGBVF_FLAG_RX_CSUM_DISABLED;
2595 static const struct net_device_ops igbvf_netdev_ops = {
2596 .ndo_open = igbvf_open,
2597 .ndo_stop = igbvf_close,
2598 .ndo_start_xmit = igbvf_xmit_frame,
2599 .ndo_get_stats = igbvf_get_stats,
2600 .ndo_set_rx_mode = igbvf_set_multi,
2601 .ndo_set_mac_address = igbvf_set_mac,
2602 .ndo_change_mtu = igbvf_change_mtu,
2603 .ndo_do_ioctl = igbvf_ioctl,
2604 .ndo_tx_timeout = igbvf_tx_timeout,
2605 .ndo_vlan_rx_add_vid = igbvf_vlan_rx_add_vid,
2606 .ndo_vlan_rx_kill_vid = igbvf_vlan_rx_kill_vid,
2607 #ifdef CONFIG_NET_POLL_CONTROLLER
2608 .ndo_poll_controller = igbvf_netpoll,
2610 .ndo_set_features = igbvf_set_features,
2614 * igbvf_probe - Device Initialization Routine
2615 * @pdev: PCI device information struct
2616 * @ent: entry in igbvf_pci_tbl
2618 * Returns 0 on success, negative on failure
2620 * igbvf_probe initializes an adapter identified by a pci_dev structure.
2621 * The OS initialization, configuring of the adapter private structure,
2622 * and a hardware reset occur.
2624 static int igbvf_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
2626 struct net_device *netdev;
2627 struct igbvf_adapter *adapter;
2628 struct e1000_hw *hw;
2629 const struct igbvf_info *ei = igbvf_info_tbl[ent->driver_data];
2631 static int cards_found;
2632 int err, pci_using_dac;
2634 err = pci_enable_device_mem(pdev);
2639 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(64));
2641 err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64));
2645 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
2647 err = dma_set_coherent_mask(&pdev->dev,
2650 dev_err(&pdev->dev, "No usable DMA "
2651 "configuration, aborting\n");
2657 err = pci_request_regions(pdev, igbvf_driver_name);
2661 pci_set_master(pdev);
2664 netdev = alloc_etherdev(sizeof(struct igbvf_adapter));
2666 goto err_alloc_etherdev;
2668 SET_NETDEV_DEV(netdev, &pdev->dev);
2670 pci_set_drvdata(pdev, netdev);
2671 adapter = netdev_priv(netdev);
2673 adapter->netdev = netdev;
2674 adapter->pdev = pdev;
2676 adapter->pba = ei->pba;
2677 adapter->flags = ei->flags;
2678 adapter->hw.back = adapter;
2679 adapter->hw.mac.type = ei->mac;
2680 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
2682 /* PCI config space info */
2684 hw->vendor_id = pdev->vendor;
2685 hw->device_id = pdev->device;
2686 hw->subsystem_vendor_id = pdev->subsystem_vendor;
2687 hw->subsystem_device_id = pdev->subsystem_device;
2688 hw->revision_id = pdev->revision;
2691 adapter->hw.hw_addr = ioremap(pci_resource_start(pdev, 0),
2692 pci_resource_len(pdev, 0));
2694 if (!adapter->hw.hw_addr)
2697 if (ei->get_variants) {
2698 err = ei->get_variants(adapter);
2703 /* setup adapter struct */
2704 err = igbvf_sw_init(adapter);
2708 /* construct the net_device struct */
2709 netdev->netdev_ops = &igbvf_netdev_ops;
2711 igbvf_set_ethtool_ops(netdev);
2712 netdev->watchdog_timeo = 5 * HZ;
2713 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
2715 adapter->bd_number = cards_found++;
2717 netdev->hw_features = NETIF_F_SG |
2724 netdev->features = netdev->hw_features |
2725 NETIF_F_HW_VLAN_TX |
2726 NETIF_F_HW_VLAN_RX |
2727 NETIF_F_HW_VLAN_FILTER;
2730 netdev->features |= NETIF_F_HIGHDMA;
2732 netdev->vlan_features |= NETIF_F_TSO;
2733 netdev->vlan_features |= NETIF_F_TSO6;
2734 netdev->vlan_features |= NETIF_F_IP_CSUM;
2735 netdev->vlan_features |= NETIF_F_IPV6_CSUM;
2736 netdev->vlan_features |= NETIF_F_SG;
2738 /*reset the controller to put the device in a known good state */
2739 err = hw->mac.ops.reset_hw(hw);
2741 dev_info(&pdev->dev,
2742 "PF still in reset state. Is the PF interface up?\n");
2744 err = hw->mac.ops.read_mac_addr(hw);
2746 dev_info(&pdev->dev, "Error reading MAC address.\n");
2747 else if (is_zero_ether_addr(adapter->hw.mac.addr))
2748 dev_info(&pdev->dev, "MAC address not assigned by administrator.\n");
2749 memcpy(netdev->dev_addr, adapter->hw.mac.addr,
2753 if (!is_valid_ether_addr(netdev->dev_addr)) {
2754 dev_info(&pdev->dev, "Assigning random MAC address.\n");
2755 eth_hw_addr_random(netdev);
2756 memcpy(adapter->hw.mac.addr, netdev->dev_addr,
2760 setup_timer(&adapter->watchdog_timer, &igbvf_watchdog,
2761 (unsigned long) adapter);
2763 INIT_WORK(&adapter->reset_task, igbvf_reset_task);
2764 INIT_WORK(&adapter->watchdog_task, igbvf_watchdog_task);
2766 /* ring size defaults */
2767 adapter->rx_ring->count = 1024;
2768 adapter->tx_ring->count = 1024;
2770 /* reset the hardware with the new settings */
2771 igbvf_reset(adapter);
2773 /* set hardware-specific flags */
2774 if (adapter->hw.mac.type == e1000_vfadapt_i350)
2775 adapter->flags |= IGBVF_FLAG_RX_LB_VLAN_BSWAP;
2777 strcpy(netdev->name, "eth%d");
2778 err = register_netdev(netdev);
2782 /* tell the stack to leave us alone until igbvf_open() is called */
2783 netif_carrier_off(netdev);
2784 netif_stop_queue(netdev);
2786 igbvf_print_device_info(adapter);
2788 igbvf_initialize_last_counter_stats(adapter);
2793 kfree(adapter->tx_ring);
2794 kfree(adapter->rx_ring);
2796 igbvf_reset_interrupt_capability(adapter);
2797 iounmap(adapter->hw.hw_addr);
2799 free_netdev(netdev);
2801 pci_release_regions(pdev);
2804 pci_disable_device(pdev);
2809 * igbvf_remove - Device Removal Routine
2810 * @pdev: PCI device information struct
2812 * igbvf_remove is called by the PCI subsystem to alert the driver
2813 * that it should release a PCI device. The could be caused by a
2814 * Hot-Plug event, or because the driver is going to be removed from
2817 static void igbvf_remove(struct pci_dev *pdev)
2819 struct net_device *netdev = pci_get_drvdata(pdev);
2820 struct igbvf_adapter *adapter = netdev_priv(netdev);
2821 struct e1000_hw *hw = &adapter->hw;
2824 * The watchdog timer may be rescheduled, so explicitly
2825 * disable it from being rescheduled.
2827 set_bit(__IGBVF_DOWN, &adapter->state);
2828 del_timer_sync(&adapter->watchdog_timer);
2830 cancel_work_sync(&adapter->reset_task);
2831 cancel_work_sync(&adapter->watchdog_task);
2833 unregister_netdev(netdev);
2835 igbvf_reset_interrupt_capability(adapter);
2838 * it is important to delete the napi struct prior to freeing the
2839 * rx ring so that you do not end up with null pointer refs
2841 netif_napi_del(&adapter->rx_ring->napi);
2842 kfree(adapter->tx_ring);
2843 kfree(adapter->rx_ring);
2845 iounmap(hw->hw_addr);
2846 if (hw->flash_address)
2847 iounmap(hw->flash_address);
2848 pci_release_regions(pdev);
2850 free_netdev(netdev);
2852 pci_disable_device(pdev);
2855 /* PCI Error Recovery (ERS) */
2856 static const struct pci_error_handlers igbvf_err_handler = {
2857 .error_detected = igbvf_io_error_detected,
2858 .slot_reset = igbvf_io_slot_reset,
2859 .resume = igbvf_io_resume,
2862 static DEFINE_PCI_DEVICE_TABLE(igbvf_pci_tbl) = {
2863 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_VF), board_vf },
2864 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_VF), board_i350_vf },
2865 { } /* terminate list */
2867 MODULE_DEVICE_TABLE(pci, igbvf_pci_tbl);
2869 /* PCI Device API Driver */
2870 static struct pci_driver igbvf_driver = {
2871 .name = igbvf_driver_name,
2872 .id_table = igbvf_pci_tbl,
2873 .probe = igbvf_probe,
2874 .remove = igbvf_remove,
2876 /* Power Management Hooks */
2877 .suspend = igbvf_suspend,
2878 .resume = igbvf_resume,
2880 .shutdown = igbvf_shutdown,
2881 .err_handler = &igbvf_err_handler
2885 * igbvf_init_module - Driver Registration Routine
2887 * igbvf_init_module is the first routine called when the driver is
2888 * loaded. All it does is register with the PCI subsystem.
2890 static int __init igbvf_init_module(void)
2893 pr_info("%s - version %s\n", igbvf_driver_string, igbvf_driver_version);
2894 pr_info("%s\n", igbvf_copyright);
2896 ret = pci_register_driver(&igbvf_driver);
2900 module_init(igbvf_init_module);
2903 * igbvf_exit_module - Driver Exit Cleanup Routine
2905 * igbvf_exit_module is called just before the driver is removed
2908 static void __exit igbvf_exit_module(void)
2910 pci_unregister_driver(&igbvf_driver);
2912 module_exit(igbvf_exit_module);
2915 MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
2916 MODULE_DESCRIPTION("Intel(R) Gigabit Virtual Function Network Driver");
2917 MODULE_LICENSE("GPL");
2918 MODULE_VERSION(DRV_VERSION);