1 /*******************************************************************************
3 Intel(R) 82576 Virtual Function Linux driver
4 Copyright(c) 2009 - 2012 Intel Corporation.
6 This program is free software; you can redistribute it and/or modify it
7 under the terms and conditions of the GNU General Public License,
8 version 2, as published by the Free Software Foundation.
10 This program is distributed in the hope it will be useful, but WITHOUT
11 ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 You should have received a copy of the GNU General Public License along with
16 this program; if not, see <http://www.gnu.org/licenses/>.
18 The full GNU General Public License is included in this distribution in
19 the file called "COPYING".
22 e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
23 Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
25 *******************************************************************************/
27 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
29 #include <linux/module.h>
30 #include <linux/types.h>
31 #include <linux/init.h>
32 #include <linux/pci.h>
33 #include <linux/vmalloc.h>
34 #include <linux/pagemap.h>
35 #include <linux/delay.h>
36 #include <linux/netdevice.h>
37 #include <linux/tcp.h>
38 #include <linux/ipv6.h>
39 #include <linux/slab.h>
40 #include <net/checksum.h>
41 #include <net/ip6_checksum.h>
42 #include <linux/mii.h>
43 #include <linux/ethtool.h>
44 #include <linux/if_vlan.h>
45 #include <linux/prefetch.h>
49 #define DRV_VERSION "2.0.2-k"
50 char igbvf_driver_name[] = "igbvf";
51 const char igbvf_driver_version[] = DRV_VERSION;
52 static const char igbvf_driver_string[] =
53 "Intel(R) Gigabit Virtual Function Network Driver";
54 static const char igbvf_copyright[] =
55 "Copyright (c) 2009 - 2012 Intel Corporation.";
57 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
58 static int debug = -1;
59 module_param(debug, int, 0);
60 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
62 static int igbvf_poll(struct napi_struct *napi, int budget);
63 static void igbvf_reset(struct igbvf_adapter *);
64 static void igbvf_set_interrupt_capability(struct igbvf_adapter *);
65 static void igbvf_reset_interrupt_capability(struct igbvf_adapter *);
67 static struct igbvf_info igbvf_vf_info = {
71 .init_ops = e1000_init_function_pointers_vf,
74 static struct igbvf_info igbvf_i350_vf_info = {
75 .mac = e1000_vfadapt_i350,
78 .init_ops = e1000_init_function_pointers_vf,
81 static const struct igbvf_info *igbvf_info_tbl[] = {
82 [board_vf] = &igbvf_vf_info,
83 [board_i350_vf] = &igbvf_i350_vf_info,
87 * igbvf_desc_unused - calculate if we have unused descriptors
88 * @rx_ring: address of receive ring structure
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, htons(ETH_P_8021Q), 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.
224 if (adapter->rx_ps_hdr_size) {
225 rx_desc->read.pkt_addr =
226 cpu_to_le64(buffer_info->page_dma);
227 rx_desc->read.hdr_addr = cpu_to_le64(buffer_info->dma);
229 rx_desc->read.pkt_addr = 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,
253 writel(i, adapter->hw.hw_addr + rx_ring->tail);
258 * igbvf_clean_rx_irq - Send received data up the network stack; legacy
259 * @adapter: board private structure
261 * the return value indicates whether actual cleaning was done, there
262 * is no guarantee that everything was cleaned
264 static bool igbvf_clean_rx_irq(struct igbvf_adapter *adapter,
265 int *work_done, int work_to_do)
267 struct igbvf_ring *rx_ring = adapter->rx_ring;
268 struct net_device *netdev = adapter->netdev;
269 struct pci_dev *pdev = adapter->pdev;
270 union e1000_adv_rx_desc *rx_desc, *next_rxd;
271 struct igbvf_buffer *buffer_info, *next_buffer;
273 bool cleaned = false;
274 int cleaned_count = 0;
275 unsigned int total_bytes = 0, total_packets = 0;
277 u32 length, hlen, staterr;
279 i = rx_ring->next_to_clean;
280 rx_desc = IGBVF_RX_DESC_ADV(*rx_ring, i);
281 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
283 while (staterr & E1000_RXD_STAT_DD) {
284 if (*work_done >= work_to_do)
287 rmb(); /* read descriptor and rx_buffer_info after status DD */
289 buffer_info = &rx_ring->buffer_info[i];
291 /* HW will not DMA in data larger than the given buffer, even
292 * if it parses the (NFS, of course) header to be larger. In
293 * that case, it fills the header buffer and spills the rest
296 hlen = (le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.hdr_info)
297 & E1000_RXDADV_HDRBUFLEN_MASK) >>
298 E1000_RXDADV_HDRBUFLEN_SHIFT;
299 if (hlen > adapter->rx_ps_hdr_size)
300 hlen = adapter->rx_ps_hdr_size;
302 length = le16_to_cpu(rx_desc->wb.upper.length);
306 skb = buffer_info->skb;
307 prefetch(skb->data - NET_IP_ALIGN);
308 buffer_info->skb = NULL;
309 if (!adapter->rx_ps_hdr_size) {
310 dma_unmap_single(&pdev->dev, buffer_info->dma,
311 adapter->rx_buffer_len,
313 buffer_info->dma = 0;
314 skb_put(skb, length);
318 if (!skb_shinfo(skb)->nr_frags) {
319 dma_unmap_single(&pdev->dev, buffer_info->dma,
320 adapter->rx_ps_hdr_size,
326 dma_unmap_page(&pdev->dev, buffer_info->page_dma,
329 buffer_info->page_dma = 0;
331 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
333 buffer_info->page_offset,
336 if ((adapter->rx_buffer_len > (PAGE_SIZE / 2)) ||
337 (page_count(buffer_info->page) != 1))
338 buffer_info->page = NULL;
340 get_page(buffer_info->page);
343 skb->data_len += length;
344 skb->truesize += PAGE_SIZE / 2;
348 if (i == rx_ring->count)
350 next_rxd = IGBVF_RX_DESC_ADV(*rx_ring, i);
352 next_buffer = &rx_ring->buffer_info[i];
354 if (!(staterr & E1000_RXD_STAT_EOP)) {
355 buffer_info->skb = next_buffer->skb;
356 buffer_info->dma = next_buffer->dma;
357 next_buffer->skb = skb;
358 next_buffer->dma = 0;
362 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
363 dev_kfree_skb_irq(skb);
367 total_bytes += skb->len;
370 igbvf_rx_checksum_adv(adapter, staterr, skb);
372 skb->protocol = eth_type_trans(skb, netdev);
374 igbvf_receive_skb(adapter, netdev, skb, staterr,
375 rx_desc->wb.upper.vlan);
378 rx_desc->wb.upper.status_error = 0;
380 /* return some buffers to hardware, one at a time is too slow */
381 if (cleaned_count >= IGBVF_RX_BUFFER_WRITE) {
382 igbvf_alloc_rx_buffers(rx_ring, cleaned_count);
386 /* use prefetched values */
388 buffer_info = next_buffer;
390 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
393 rx_ring->next_to_clean = i;
394 cleaned_count = igbvf_desc_unused(rx_ring);
397 igbvf_alloc_rx_buffers(rx_ring, cleaned_count);
399 adapter->total_rx_packets += total_packets;
400 adapter->total_rx_bytes += total_bytes;
401 adapter->net_stats.rx_bytes += total_bytes;
402 adapter->net_stats.rx_packets += total_packets;
406 static void igbvf_put_txbuf(struct igbvf_adapter *adapter,
407 struct igbvf_buffer *buffer_info)
409 if (buffer_info->dma) {
410 if (buffer_info->mapped_as_page)
411 dma_unmap_page(&adapter->pdev->dev,
416 dma_unmap_single(&adapter->pdev->dev,
420 buffer_info->dma = 0;
422 if (buffer_info->skb) {
423 dev_kfree_skb_any(buffer_info->skb);
424 buffer_info->skb = NULL;
426 buffer_info->time_stamp = 0;
430 * igbvf_setup_tx_resources - allocate Tx resources (Descriptors)
431 * @adapter: board private structure
433 * Return 0 on success, negative on failure
435 int igbvf_setup_tx_resources(struct igbvf_adapter *adapter,
436 struct igbvf_ring *tx_ring)
438 struct pci_dev *pdev = adapter->pdev;
441 size = sizeof(struct igbvf_buffer) * tx_ring->count;
442 tx_ring->buffer_info = vzalloc(size);
443 if (!tx_ring->buffer_info)
446 /* round up to nearest 4K */
447 tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
448 tx_ring->size = ALIGN(tx_ring->size, 4096);
450 tx_ring->desc = dma_alloc_coherent(&pdev->dev, tx_ring->size,
451 &tx_ring->dma, GFP_KERNEL);
455 tx_ring->adapter = adapter;
456 tx_ring->next_to_use = 0;
457 tx_ring->next_to_clean = 0;
461 vfree(tx_ring->buffer_info);
462 dev_err(&adapter->pdev->dev,
463 "Unable to allocate memory for the transmit descriptor ring\n");
468 * igbvf_setup_rx_resources - allocate Rx resources (Descriptors)
469 * @adapter: board private structure
471 * Returns 0 on success, negative on failure
473 int igbvf_setup_rx_resources(struct igbvf_adapter *adapter,
474 struct igbvf_ring *rx_ring)
476 struct pci_dev *pdev = adapter->pdev;
479 size = sizeof(struct igbvf_buffer) * rx_ring->count;
480 rx_ring->buffer_info = vzalloc(size);
481 if (!rx_ring->buffer_info)
484 desc_len = sizeof(union e1000_adv_rx_desc);
486 /* Round up to nearest 4K */
487 rx_ring->size = rx_ring->count * desc_len;
488 rx_ring->size = ALIGN(rx_ring->size, 4096);
490 rx_ring->desc = dma_alloc_coherent(&pdev->dev, rx_ring->size,
491 &rx_ring->dma, GFP_KERNEL);
495 rx_ring->next_to_clean = 0;
496 rx_ring->next_to_use = 0;
498 rx_ring->adapter = adapter;
503 vfree(rx_ring->buffer_info);
504 rx_ring->buffer_info = NULL;
505 dev_err(&adapter->pdev->dev,
506 "Unable to allocate memory for the receive descriptor ring\n");
511 * igbvf_clean_tx_ring - Free Tx Buffers
512 * @tx_ring: ring to be cleaned
514 static void igbvf_clean_tx_ring(struct igbvf_ring *tx_ring)
516 struct igbvf_adapter *adapter = tx_ring->adapter;
517 struct igbvf_buffer *buffer_info;
521 if (!tx_ring->buffer_info)
524 /* Free all the Tx ring sk_buffs */
525 for (i = 0; i < tx_ring->count; i++) {
526 buffer_info = &tx_ring->buffer_info[i];
527 igbvf_put_txbuf(adapter, buffer_info);
530 size = sizeof(struct igbvf_buffer) * tx_ring->count;
531 memset(tx_ring->buffer_info, 0, size);
533 /* Zero out the descriptor ring */
534 memset(tx_ring->desc, 0, tx_ring->size);
536 tx_ring->next_to_use = 0;
537 tx_ring->next_to_clean = 0;
539 writel(0, adapter->hw.hw_addr + tx_ring->head);
540 writel(0, adapter->hw.hw_addr + tx_ring->tail);
544 * igbvf_free_tx_resources - Free Tx Resources per Queue
545 * @tx_ring: ring to free resources from
547 * Free all transmit software resources
549 void igbvf_free_tx_resources(struct igbvf_ring *tx_ring)
551 struct pci_dev *pdev = tx_ring->adapter->pdev;
553 igbvf_clean_tx_ring(tx_ring);
555 vfree(tx_ring->buffer_info);
556 tx_ring->buffer_info = NULL;
558 dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
561 tx_ring->desc = NULL;
565 * igbvf_clean_rx_ring - Free Rx Buffers per Queue
566 * @adapter: board private structure
568 static void igbvf_clean_rx_ring(struct igbvf_ring *rx_ring)
570 struct igbvf_adapter *adapter = rx_ring->adapter;
571 struct igbvf_buffer *buffer_info;
572 struct pci_dev *pdev = adapter->pdev;
576 if (!rx_ring->buffer_info)
579 /* Free all the Rx ring sk_buffs */
580 for (i = 0; i < rx_ring->count; i++) {
581 buffer_info = &rx_ring->buffer_info[i];
582 if (buffer_info->dma) {
583 if (adapter->rx_ps_hdr_size) {
584 dma_unmap_single(&pdev->dev, buffer_info->dma,
585 adapter->rx_ps_hdr_size,
588 dma_unmap_single(&pdev->dev, buffer_info->dma,
589 adapter->rx_buffer_len,
592 buffer_info->dma = 0;
595 if (buffer_info->skb) {
596 dev_kfree_skb(buffer_info->skb);
597 buffer_info->skb = NULL;
600 if (buffer_info->page) {
601 if (buffer_info->page_dma)
602 dma_unmap_page(&pdev->dev,
603 buffer_info->page_dma,
606 put_page(buffer_info->page);
607 buffer_info->page = NULL;
608 buffer_info->page_dma = 0;
609 buffer_info->page_offset = 0;
613 size = sizeof(struct igbvf_buffer) * rx_ring->count;
614 memset(rx_ring->buffer_info, 0, size);
616 /* Zero out the descriptor ring */
617 memset(rx_ring->desc, 0, rx_ring->size);
619 rx_ring->next_to_clean = 0;
620 rx_ring->next_to_use = 0;
622 writel(0, adapter->hw.hw_addr + rx_ring->head);
623 writel(0, adapter->hw.hw_addr + rx_ring->tail);
627 * igbvf_free_rx_resources - Free Rx Resources
628 * @rx_ring: ring to clean the resources from
630 * Free all receive software resources
633 void igbvf_free_rx_resources(struct igbvf_ring *rx_ring)
635 struct pci_dev *pdev = rx_ring->adapter->pdev;
637 igbvf_clean_rx_ring(rx_ring);
639 vfree(rx_ring->buffer_info);
640 rx_ring->buffer_info = NULL;
642 dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
644 rx_ring->desc = NULL;
648 * igbvf_update_itr - update the dynamic ITR value based on statistics
649 * @adapter: pointer to adapter
650 * @itr_setting: current adapter->itr
651 * @packets: the number of packets during this measurement interval
652 * @bytes: the number of bytes during this measurement interval
654 * Stores a new ITR value based on packets and byte counts during the last
655 * interrupt. The advantage of per interrupt computation is faster updates
656 * and more accurate ITR for the current traffic pattern. Constants in this
657 * function were computed based on theoretical maximum wire speed and thresholds
658 * were set based on testing data as well as attempting to minimize response
659 * time while increasing bulk throughput.
661 static enum latency_range igbvf_update_itr(struct igbvf_adapter *adapter,
662 enum latency_range itr_setting,
663 int packets, int bytes)
665 enum latency_range retval = itr_setting;
668 goto update_itr_done;
670 switch (itr_setting) {
672 /* handle TSO and jumbo frames */
673 if (bytes/packets > 8000)
674 retval = bulk_latency;
675 else if ((packets < 5) && (bytes > 512))
676 retval = low_latency;
678 case low_latency: /* 50 usec aka 20000 ints/s */
680 /* this if handles the TSO accounting */
681 if (bytes/packets > 8000)
682 retval = bulk_latency;
683 else if ((packets < 10) || ((bytes/packets) > 1200))
684 retval = bulk_latency;
685 else if ((packets > 35))
686 retval = lowest_latency;
687 } else if (bytes/packets > 2000) {
688 retval = bulk_latency;
689 } else if (packets <= 2 && bytes < 512) {
690 retval = lowest_latency;
693 case bulk_latency: /* 250 usec aka 4000 ints/s */
696 retval = low_latency;
697 } else if (bytes < 6000) {
698 retval = low_latency;
709 static int igbvf_range_to_itr(enum latency_range current_range)
713 switch (current_range) {
714 /* counts and packets in update_itr are dependent on these numbers */
716 new_itr = IGBVF_70K_ITR;
719 new_itr = IGBVF_20K_ITR;
722 new_itr = IGBVF_4K_ITR;
725 new_itr = IGBVF_START_ITR;
731 static void igbvf_set_itr(struct igbvf_adapter *adapter)
735 adapter->tx_ring->itr_range =
736 igbvf_update_itr(adapter,
737 adapter->tx_ring->itr_val,
738 adapter->total_tx_packets,
739 adapter->total_tx_bytes);
741 /* conservative mode (itr 3) eliminates the lowest_latency setting */
742 if (adapter->requested_itr == 3 &&
743 adapter->tx_ring->itr_range == lowest_latency)
744 adapter->tx_ring->itr_range = low_latency;
746 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;
750 /* this attempts to bias the interrupt rate towards Bulk
751 * by adding intermediate steps when interrupt rate is
754 new_itr = new_itr > current_itr ?
755 min(current_itr + (new_itr >> 2), new_itr) :
757 adapter->tx_ring->itr_val = new_itr;
759 adapter->tx_ring->set_itr = 1;
762 adapter->rx_ring->itr_range =
763 igbvf_update_itr(adapter, adapter->rx_ring->itr_val,
764 adapter->total_rx_packets,
765 adapter->total_rx_bytes);
766 if (adapter->requested_itr == 3 &&
767 adapter->rx_ring->itr_range == lowest_latency)
768 adapter->rx_ring->itr_range = low_latency;
770 new_itr = igbvf_range_to_itr(adapter->rx_ring->itr_range);
772 if (new_itr != adapter->rx_ring->itr_val) {
773 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 && netif_carrier_ok(netdev) &&
853 igbvf_desc_unused(tx_ring) >= IGBVF_TX_QUEUE_WAKE)) {
854 /* Make sure that anybody stopping the queue after this
855 * sees the new next_to_clean.
858 if (netif_queue_stopped(netdev) &&
859 !(test_bit(__IGBVF_DOWN, &adapter->state))) {
860 netif_wake_queue(netdev);
861 ++adapter->restart_queue;
865 adapter->net_stats.tx_bytes += total_bytes;
866 adapter->net_stats.tx_packets += total_packets;
867 return count < tx_ring->count;
870 static irqreturn_t igbvf_msix_other(int irq, void *data)
872 struct net_device *netdev = data;
873 struct igbvf_adapter *adapter = netdev_priv(netdev);
874 struct e1000_hw *hw = &adapter->hw;
876 adapter->int_counter1++;
878 netif_carrier_off(netdev);
879 hw->mac.get_link_status = 1;
880 if (!test_bit(__IGBVF_DOWN, &adapter->state))
881 mod_timer(&adapter->watchdog_timer, jiffies + 1);
883 ew32(EIMS, adapter->eims_other);
888 static irqreturn_t igbvf_intr_msix_tx(int irq, void *data)
890 struct net_device *netdev = data;
891 struct igbvf_adapter *adapter = netdev_priv(netdev);
892 struct e1000_hw *hw = &adapter->hw;
893 struct igbvf_ring *tx_ring = adapter->tx_ring;
895 if (tx_ring->set_itr) {
896 writel(tx_ring->itr_val,
897 adapter->hw.hw_addr + tx_ring->itr_register);
898 adapter->tx_ring->set_itr = 0;
901 adapter->total_tx_bytes = 0;
902 adapter->total_tx_packets = 0;
904 /* auto mask will automatically re-enable 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.
954 if (rx_queue > IGBVF_NO_QUEUE) {
955 index = (rx_queue >> 1);
956 ivar = array_er32(IVAR0, index);
957 if (rx_queue & 0x1) {
958 /* vector goes into third byte of register */
959 ivar = ivar & 0xFF00FFFF;
960 ivar |= (msix_vector | E1000_IVAR_VALID) << 16;
962 /* vector goes into low byte of register */
963 ivar = ivar & 0xFFFFFF00;
964 ivar |= msix_vector | E1000_IVAR_VALID;
966 adapter->rx_ring[rx_queue].eims_value = 1 << msix_vector;
967 array_ew32(IVAR0, index, ivar);
969 if (tx_queue > IGBVF_NO_QUEUE) {
970 index = (tx_queue >> 1);
971 ivar = array_er32(IVAR0, index);
972 if (tx_queue & 0x1) {
973 /* vector goes into high byte of register */
974 ivar = ivar & 0x00FFFFFF;
975 ivar |= (msix_vector | E1000_IVAR_VALID) << 24;
977 /* vector goes into second byte of register */
978 ivar = ivar & 0xFFFF00FF;
979 ivar |= (msix_vector | E1000_IVAR_VALID) << 8;
981 adapter->tx_ring[tx_queue].eims_value = 1 << msix_vector;
982 array_ew32(IVAR0, index, ivar);
987 * igbvf_configure_msix - Configure MSI-X hardware
988 * @adapter: board private structure
990 * igbvf_configure_msix sets up the hardware to properly
991 * generate MSI-X interrupts.
993 static void igbvf_configure_msix(struct igbvf_adapter *adapter)
996 struct e1000_hw *hw = &adapter->hw;
997 struct igbvf_ring *tx_ring = adapter->tx_ring;
998 struct igbvf_ring *rx_ring = adapter->rx_ring;
1001 adapter->eims_enable_mask = 0;
1003 igbvf_assign_vector(adapter, IGBVF_NO_QUEUE, 0, vector++);
1004 adapter->eims_enable_mask |= tx_ring->eims_value;
1005 writel(tx_ring->itr_val, hw->hw_addr + tx_ring->itr_register);
1006 igbvf_assign_vector(adapter, 0, IGBVF_NO_QUEUE, vector++);
1007 adapter->eims_enable_mask |= rx_ring->eims_value;
1008 writel(rx_ring->itr_val, hw->hw_addr + rx_ring->itr_register);
1010 /* set vector for other causes, i.e. link changes */
1012 tmp = (vector++ | E1000_IVAR_VALID);
1014 ew32(IVAR_MISC, tmp);
1016 adapter->eims_enable_mask = (1 << (vector)) - 1;
1017 adapter->eims_other = 1 << (vector - 1);
1021 static void igbvf_reset_interrupt_capability(struct igbvf_adapter *adapter)
1023 if (adapter->msix_entries) {
1024 pci_disable_msix(adapter->pdev);
1025 kfree(adapter->msix_entries);
1026 adapter->msix_entries = NULL;
1031 * igbvf_set_interrupt_capability - set MSI or MSI-X if supported
1032 * @adapter: board private structure
1034 * Attempt to configure interrupts using the best available
1035 * capabilities of the hardware and kernel.
1037 static void igbvf_set_interrupt_capability(struct igbvf_adapter *adapter)
1042 /* we allocate 3 vectors, 1 for Tx, 1 for Rx, one for PF messages */
1043 adapter->msix_entries = kcalloc(3, sizeof(struct msix_entry),
1045 if (adapter->msix_entries) {
1046 for (i = 0; i < 3; i++)
1047 adapter->msix_entries[i].entry = i;
1049 err = pci_enable_msix_range(adapter->pdev,
1050 adapter->msix_entries, 3, 3);
1055 dev_err(&adapter->pdev->dev,
1056 "Failed to initialize MSI-X interrupts.\n");
1057 igbvf_reset_interrupt_capability(adapter);
1062 * igbvf_request_msix - Initialize MSI-X interrupts
1063 * @adapter: board private structure
1065 * igbvf_request_msix allocates MSI-X vectors and requests interrupts from the
1068 static int igbvf_request_msix(struct igbvf_adapter *adapter)
1070 struct net_device *netdev = adapter->netdev;
1071 int err = 0, vector = 0;
1073 if (strlen(netdev->name) < (IFNAMSIZ - 5)) {
1074 sprintf(adapter->tx_ring->name, "%s-tx-0", netdev->name);
1075 sprintf(adapter->rx_ring->name, "%s-rx-0", netdev->name);
1077 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1078 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1081 err = request_irq(adapter->msix_entries[vector].vector,
1082 igbvf_intr_msix_tx, 0, adapter->tx_ring->name,
1087 adapter->tx_ring->itr_register = E1000_EITR(vector);
1088 adapter->tx_ring->itr_val = adapter->current_itr;
1091 err = request_irq(adapter->msix_entries[vector].vector,
1092 igbvf_intr_msix_rx, 0, adapter->rx_ring->name,
1097 adapter->rx_ring->itr_register = E1000_EITR(vector);
1098 adapter->rx_ring->itr_val = adapter->current_itr;
1101 err = request_irq(adapter->msix_entries[vector].vector,
1102 igbvf_msix_other, 0, netdev->name, netdev);
1106 igbvf_configure_msix(adapter);
1113 * igbvf_alloc_queues - Allocate memory for all rings
1114 * @adapter: board private structure to initialize
1116 static int igbvf_alloc_queues(struct igbvf_adapter *adapter)
1118 struct net_device *netdev = adapter->netdev;
1120 adapter->tx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL);
1121 if (!adapter->tx_ring)
1124 adapter->rx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL);
1125 if (!adapter->rx_ring) {
1126 kfree(adapter->tx_ring);
1130 netif_napi_add(netdev, &adapter->rx_ring->napi, igbvf_poll, 64);
1136 * igbvf_request_irq - initialize interrupts
1137 * @adapter: board private structure
1139 * Attempts to configure interrupts using the best available
1140 * capabilities of the hardware and kernel.
1142 static int igbvf_request_irq(struct igbvf_adapter *adapter)
1146 /* igbvf supports msi-x only */
1147 if (adapter->msix_entries)
1148 err = igbvf_request_msix(adapter);
1153 dev_err(&adapter->pdev->dev,
1154 "Unable to allocate interrupt, Error: %d\n", err);
1159 static void igbvf_free_irq(struct igbvf_adapter *adapter)
1161 struct net_device *netdev = adapter->netdev;
1164 if (adapter->msix_entries) {
1165 for (vector = 0; vector < 3; vector++)
1166 free_irq(adapter->msix_entries[vector].vector, netdev);
1171 * igbvf_irq_disable - Mask off interrupt generation on the NIC
1172 * @adapter: board private structure
1174 static void igbvf_irq_disable(struct igbvf_adapter *adapter)
1176 struct e1000_hw *hw = &adapter->hw;
1180 if (adapter->msix_entries)
1185 * igbvf_irq_enable - Enable default interrupt generation settings
1186 * @adapter: board private structure
1188 static void igbvf_irq_enable(struct igbvf_adapter *adapter)
1190 struct e1000_hw *hw = &adapter->hw;
1192 ew32(EIAC, adapter->eims_enable_mask);
1193 ew32(EIAM, adapter->eims_enable_mask);
1194 ew32(EIMS, adapter->eims_enable_mask);
1198 * igbvf_poll - NAPI Rx polling callback
1199 * @napi: struct associated with this polling callback
1200 * @budget: amount of packets driver is allowed to process this poll
1202 static int igbvf_poll(struct napi_struct *napi, int budget)
1204 struct igbvf_ring *rx_ring = container_of(napi, struct igbvf_ring, napi);
1205 struct igbvf_adapter *adapter = rx_ring->adapter;
1206 struct e1000_hw *hw = &adapter->hw;
1209 igbvf_clean_rx_irq(adapter, &work_done, budget);
1211 /* If not enough Rx work done, exit the polling mode */
1212 if (work_done < budget) {
1213 napi_complete(napi);
1215 if (adapter->requested_itr & 3)
1216 igbvf_set_itr(adapter);
1218 if (!test_bit(__IGBVF_DOWN, &adapter->state))
1219 ew32(EIMS, adapter->rx_ring->eims_value);
1226 * igbvf_set_rlpml - set receive large packet maximum length
1227 * @adapter: board private structure
1229 * Configure the maximum size of packets that will be received
1231 static void igbvf_set_rlpml(struct igbvf_adapter *adapter)
1234 struct e1000_hw *hw = &adapter->hw;
1236 max_frame_size = adapter->max_frame_size + VLAN_TAG_SIZE;
1237 e1000_rlpml_set_vf(hw, max_frame_size);
1240 static int igbvf_vlan_rx_add_vid(struct net_device *netdev,
1241 __be16 proto, u16 vid)
1243 struct igbvf_adapter *adapter = netdev_priv(netdev);
1244 struct e1000_hw *hw = &adapter->hw;
1246 if (hw->mac.ops.set_vfta(hw, vid, true)) {
1247 dev_err(&adapter->pdev->dev, "Failed to add vlan id %d\n", vid);
1250 set_bit(vid, adapter->active_vlans);
1254 static int igbvf_vlan_rx_kill_vid(struct net_device *netdev,
1255 __be16 proto, u16 vid)
1257 struct igbvf_adapter *adapter = netdev_priv(netdev);
1258 struct e1000_hw *hw = &adapter->hw;
1260 if (hw->mac.ops.set_vfta(hw, vid, false)) {
1261 dev_err(&adapter->pdev->dev,
1262 "Failed to remove vlan id %d\n", vid);
1265 clear_bit(vid, adapter->active_vlans);
1269 static void igbvf_restore_vlan(struct igbvf_adapter *adapter)
1273 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
1274 igbvf_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
1278 * igbvf_configure_tx - Configure Transmit Unit after Reset
1279 * @adapter: board private structure
1281 * Configure the Tx unit of the MAC after a reset.
1283 static void igbvf_configure_tx(struct igbvf_adapter *adapter)
1285 struct e1000_hw *hw = &adapter->hw;
1286 struct igbvf_ring *tx_ring = adapter->tx_ring;
1288 u32 txdctl, dca_txctrl;
1290 /* disable transmits */
1291 txdctl = er32(TXDCTL(0));
1292 ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1296 /* Setup the HW Tx Head and Tail descriptor pointers */
1297 ew32(TDLEN(0), tx_ring->count * sizeof(union e1000_adv_tx_desc));
1298 tdba = tx_ring->dma;
1299 ew32(TDBAL(0), (tdba & DMA_BIT_MASK(32)));
1300 ew32(TDBAH(0), (tdba >> 32));
1303 tx_ring->head = E1000_TDH(0);
1304 tx_ring->tail = E1000_TDT(0);
1306 /* Turn off Relaxed Ordering on head write-backs. The writebacks
1307 * MUST be delivered in order or it will completely screw up
1310 dca_txctrl = er32(DCA_TXCTRL(0));
1311 dca_txctrl &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN;
1312 ew32(DCA_TXCTRL(0), dca_txctrl);
1314 /* enable transmits */
1315 txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
1316 ew32(TXDCTL(0), txdctl);
1318 /* Setup Transmit Descriptor Settings for eop descriptor */
1319 adapter->txd_cmd = E1000_ADVTXD_DCMD_EOP | E1000_ADVTXD_DCMD_IFCS;
1321 /* enable Report Status bit */
1322 adapter->txd_cmd |= E1000_ADVTXD_DCMD_RS;
1326 * igbvf_setup_srrctl - configure the receive control registers
1327 * @adapter: Board private structure
1329 static void igbvf_setup_srrctl(struct igbvf_adapter *adapter)
1331 struct e1000_hw *hw = &adapter->hw;
1334 srrctl &= ~(E1000_SRRCTL_DESCTYPE_MASK |
1335 E1000_SRRCTL_BSIZEHDR_MASK |
1336 E1000_SRRCTL_BSIZEPKT_MASK);
1338 /* Enable queue drop to avoid head of line blocking */
1339 srrctl |= E1000_SRRCTL_DROP_EN;
1341 /* Setup buffer sizes */
1342 srrctl |= ALIGN(adapter->rx_buffer_len, 1024) >>
1343 E1000_SRRCTL_BSIZEPKT_SHIFT;
1345 if (adapter->rx_buffer_len < 2048) {
1346 adapter->rx_ps_hdr_size = 0;
1347 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
1349 adapter->rx_ps_hdr_size = 128;
1350 srrctl |= adapter->rx_ps_hdr_size <<
1351 E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
1352 srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
1355 ew32(SRRCTL(0), srrctl);
1359 * igbvf_configure_rx - Configure Receive Unit after Reset
1360 * @adapter: board private structure
1362 * Configure the Rx unit of the MAC after a reset.
1364 static void igbvf_configure_rx(struct igbvf_adapter *adapter)
1366 struct e1000_hw *hw = &adapter->hw;
1367 struct igbvf_ring *rx_ring = adapter->rx_ring;
1371 /* disable receives */
1372 rxdctl = er32(RXDCTL(0));
1373 ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1377 rdlen = rx_ring->count * sizeof(union e1000_adv_rx_desc);
1379 /* Setup the HW Rx Head and Tail Descriptor Pointers and
1380 * the Base and Length of the Rx Descriptor Ring
1382 rdba = rx_ring->dma;
1383 ew32(RDBAL(0), (rdba & DMA_BIT_MASK(32)));
1384 ew32(RDBAH(0), (rdba >> 32));
1385 ew32(RDLEN(0), rx_ring->count * sizeof(union e1000_adv_rx_desc));
1386 rx_ring->head = E1000_RDH(0);
1387 rx_ring->tail = E1000_RDT(0);
1391 rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
1392 rxdctl &= 0xFFF00000;
1393 rxdctl |= IGBVF_RX_PTHRESH;
1394 rxdctl |= IGBVF_RX_HTHRESH << 8;
1395 rxdctl |= IGBVF_RX_WTHRESH << 16;
1397 igbvf_set_rlpml(adapter);
1399 /* enable receives */
1400 ew32(RXDCTL(0), rxdctl);
1404 * igbvf_set_multi - Multicast and Promiscuous mode set
1405 * @netdev: network interface device structure
1407 * The set_multi entry point is called whenever the multicast address
1408 * list or the network interface flags are updated. This routine is
1409 * responsible for configuring the hardware for proper multicast,
1410 * promiscuous mode, and all-multi behavior.
1412 static void igbvf_set_multi(struct net_device *netdev)
1414 struct igbvf_adapter *adapter = netdev_priv(netdev);
1415 struct e1000_hw *hw = &adapter->hw;
1416 struct netdev_hw_addr *ha;
1417 u8 *mta_list = NULL;
1420 if (!netdev_mc_empty(netdev)) {
1421 mta_list = kmalloc_array(netdev_mc_count(netdev), ETH_ALEN,
1427 /* prepare a packed array of only addresses. */
1429 netdev_for_each_mc_addr(ha, netdev)
1430 memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
1432 hw->mac.ops.update_mc_addr_list(hw, mta_list, i, 0, 0);
1437 * igbvf_configure - configure the hardware for Rx and Tx
1438 * @adapter: private board structure
1440 static void igbvf_configure(struct igbvf_adapter *adapter)
1442 igbvf_set_multi(adapter->netdev);
1444 igbvf_restore_vlan(adapter);
1446 igbvf_configure_tx(adapter);
1447 igbvf_setup_srrctl(adapter);
1448 igbvf_configure_rx(adapter);
1449 igbvf_alloc_rx_buffers(adapter->rx_ring,
1450 igbvf_desc_unused(adapter->rx_ring));
1453 /* igbvf_reset - bring the hardware into a known good state
1454 * @adapter: private board structure
1456 * This function boots the hardware and enables some settings that
1457 * require a configuration cycle of the hardware - those cannot be
1458 * set/changed during runtime. After reset the device needs to be
1459 * properly configured for Rx, Tx etc.
1461 static void igbvf_reset(struct igbvf_adapter *adapter)
1463 struct e1000_mac_info *mac = &adapter->hw.mac;
1464 struct net_device *netdev = adapter->netdev;
1465 struct e1000_hw *hw = &adapter->hw;
1467 /* Allow time for pending master requests to run */
1468 if (mac->ops.reset_hw(hw))
1469 dev_err(&adapter->pdev->dev, "PF still resetting\n");
1471 mac->ops.init_hw(hw);
1473 if (is_valid_ether_addr(adapter->hw.mac.addr)) {
1474 memcpy(netdev->dev_addr, adapter->hw.mac.addr,
1476 memcpy(netdev->perm_addr, adapter->hw.mac.addr,
1480 adapter->last_reset = jiffies;
1483 int igbvf_up(struct igbvf_adapter *adapter)
1485 struct e1000_hw *hw = &adapter->hw;
1487 /* hardware has been reset, we need to reload some things */
1488 igbvf_configure(adapter);
1490 clear_bit(__IGBVF_DOWN, &adapter->state);
1492 napi_enable(&adapter->rx_ring->napi);
1493 if (adapter->msix_entries)
1494 igbvf_configure_msix(adapter);
1496 /* Clear any pending interrupts. */
1498 igbvf_irq_enable(adapter);
1500 /* start the watchdog */
1501 hw->mac.get_link_status = 1;
1502 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1507 void igbvf_down(struct igbvf_adapter *adapter)
1509 struct net_device *netdev = adapter->netdev;
1510 struct e1000_hw *hw = &adapter->hw;
1513 /* signal that we're down so the interrupt handler does not
1514 * reschedule our watchdog timer
1516 set_bit(__IGBVF_DOWN, &adapter->state);
1518 /* disable receives in the hardware */
1519 rxdctl = er32(RXDCTL(0));
1520 ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1522 netif_carrier_off(netdev);
1523 netif_stop_queue(netdev);
1525 /* disable transmits in the hardware */
1526 txdctl = er32(TXDCTL(0));
1527 ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1529 /* flush both disables and wait for them to finish */
1533 napi_disable(&adapter->rx_ring->napi);
1535 igbvf_irq_disable(adapter);
1537 del_timer_sync(&adapter->watchdog_timer);
1539 /* record the stats before reset*/
1540 igbvf_update_stats(adapter);
1542 adapter->link_speed = 0;
1543 adapter->link_duplex = 0;
1545 igbvf_reset(adapter);
1546 igbvf_clean_tx_ring(adapter->tx_ring);
1547 igbvf_clean_rx_ring(adapter->rx_ring);
1550 void igbvf_reinit_locked(struct igbvf_adapter *adapter)
1553 while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
1554 usleep_range(1000, 2000);
1555 igbvf_down(adapter);
1557 clear_bit(__IGBVF_RESETTING, &adapter->state);
1561 * igbvf_sw_init - Initialize general software structures (struct igbvf_adapter)
1562 * @adapter: board private structure to initialize
1564 * igbvf_sw_init initializes the Adapter private data structure.
1565 * Fields are initialized based on PCI device information and
1566 * OS network device settings (MTU size).
1568 static int igbvf_sw_init(struct igbvf_adapter *adapter)
1570 struct net_device *netdev = adapter->netdev;
1573 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
1574 adapter->rx_ps_hdr_size = 0;
1575 adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1576 adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
1578 adapter->tx_int_delay = 8;
1579 adapter->tx_abs_int_delay = 32;
1580 adapter->rx_int_delay = 0;
1581 adapter->rx_abs_int_delay = 8;
1582 adapter->requested_itr = 3;
1583 adapter->current_itr = IGBVF_START_ITR;
1585 /* Set various function pointers */
1586 adapter->ei->init_ops(&adapter->hw);
1588 rc = adapter->hw.mac.ops.init_params(&adapter->hw);
1592 rc = adapter->hw.mbx.ops.init_params(&adapter->hw);
1596 igbvf_set_interrupt_capability(adapter);
1598 if (igbvf_alloc_queues(adapter))
1601 spin_lock_init(&adapter->tx_queue_lock);
1603 /* Explicitly disable IRQ since the NIC can be in any state. */
1604 igbvf_irq_disable(adapter);
1606 spin_lock_init(&adapter->stats_lock);
1608 set_bit(__IGBVF_DOWN, &adapter->state);
1612 static void igbvf_initialize_last_counter_stats(struct igbvf_adapter *adapter)
1614 struct e1000_hw *hw = &adapter->hw;
1616 adapter->stats.last_gprc = er32(VFGPRC);
1617 adapter->stats.last_gorc = er32(VFGORC);
1618 adapter->stats.last_gptc = er32(VFGPTC);
1619 adapter->stats.last_gotc = er32(VFGOTC);
1620 adapter->stats.last_mprc = er32(VFMPRC);
1621 adapter->stats.last_gotlbc = er32(VFGOTLBC);
1622 adapter->stats.last_gptlbc = er32(VFGPTLBC);
1623 adapter->stats.last_gorlbc = er32(VFGORLBC);
1624 adapter->stats.last_gprlbc = er32(VFGPRLBC);
1626 adapter->stats.base_gprc = er32(VFGPRC);
1627 adapter->stats.base_gorc = er32(VFGORC);
1628 adapter->stats.base_gptc = er32(VFGPTC);
1629 adapter->stats.base_gotc = er32(VFGOTC);
1630 adapter->stats.base_mprc = er32(VFMPRC);
1631 adapter->stats.base_gotlbc = er32(VFGOTLBC);
1632 adapter->stats.base_gptlbc = er32(VFGPTLBC);
1633 adapter->stats.base_gorlbc = er32(VFGORLBC);
1634 adapter->stats.base_gprlbc = er32(VFGPRLBC);
1638 * igbvf_open - Called when a network interface is made active
1639 * @netdev: network interface device structure
1641 * Returns 0 on success, negative value on failure
1643 * The open entry point is called when a network interface is made
1644 * active by the system (IFF_UP). At this point all resources needed
1645 * for transmit and receive operations are allocated, the interrupt
1646 * handler is registered with the OS, the watchdog timer is started,
1647 * and the stack is notified that the interface is ready.
1649 static int igbvf_open(struct net_device *netdev)
1651 struct igbvf_adapter *adapter = netdev_priv(netdev);
1652 struct e1000_hw *hw = &adapter->hw;
1655 /* disallow open during test */
1656 if (test_bit(__IGBVF_TESTING, &adapter->state))
1659 /* allocate transmit descriptors */
1660 err = igbvf_setup_tx_resources(adapter, adapter->tx_ring);
1664 /* allocate receive descriptors */
1665 err = igbvf_setup_rx_resources(adapter, adapter->rx_ring);
1669 /* before we allocate an interrupt, we must be ready to handle it.
1670 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1671 * as soon as we call pci_request_irq, so we have to setup our
1672 * clean_rx handler before we do so.
1674 igbvf_configure(adapter);
1676 err = igbvf_request_irq(adapter);
1680 /* From here on the code is the same as igbvf_up() */
1681 clear_bit(__IGBVF_DOWN, &adapter->state);
1683 napi_enable(&adapter->rx_ring->napi);
1685 /* clear any pending interrupts */
1688 igbvf_irq_enable(adapter);
1690 /* start the watchdog */
1691 hw->mac.get_link_status = 1;
1692 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1697 igbvf_free_rx_resources(adapter->rx_ring);
1699 igbvf_free_tx_resources(adapter->tx_ring);
1701 igbvf_reset(adapter);
1707 * igbvf_close - Disables a network interface
1708 * @netdev: network interface device structure
1710 * Returns 0, this is not allowed to fail
1712 * The close entry point is called when an interface is de-activated
1713 * by the OS. The hardware is still under the drivers control, but
1714 * needs to be disabled. A global MAC reset is issued to stop the
1715 * hardware, and all transmit and receive resources are freed.
1717 static int igbvf_close(struct net_device *netdev)
1719 struct igbvf_adapter *adapter = netdev_priv(netdev);
1721 WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
1722 igbvf_down(adapter);
1724 igbvf_free_irq(adapter);
1726 igbvf_free_tx_resources(adapter->tx_ring);
1727 igbvf_free_rx_resources(adapter->rx_ring);
1733 * igbvf_set_mac - Change the Ethernet Address of the NIC
1734 * @netdev: network interface device structure
1735 * @p: pointer to an address structure
1737 * Returns 0 on success, negative on failure
1739 static int igbvf_set_mac(struct net_device *netdev, void *p)
1741 struct igbvf_adapter *adapter = netdev_priv(netdev);
1742 struct e1000_hw *hw = &adapter->hw;
1743 struct sockaddr *addr = p;
1745 if (!is_valid_ether_addr(addr->sa_data))
1746 return -EADDRNOTAVAIL;
1748 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
1750 hw->mac.ops.rar_set(hw, hw->mac.addr, 0);
1752 if (!ether_addr_equal(addr->sa_data, hw->mac.addr))
1753 return -EADDRNOTAVAIL;
1755 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
1760 #define UPDATE_VF_COUNTER(reg, name) \
1762 u32 current_counter = er32(reg); \
1763 if (current_counter < adapter->stats.last_##name) \
1764 adapter->stats.name += 0x100000000LL; \
1765 adapter->stats.last_##name = current_counter; \
1766 adapter->stats.name &= 0xFFFFFFFF00000000LL; \
1767 adapter->stats.name |= current_counter; \
1771 * igbvf_update_stats - Update the board statistics counters
1772 * @adapter: board private structure
1774 void igbvf_update_stats(struct igbvf_adapter *adapter)
1776 struct e1000_hw *hw = &adapter->hw;
1777 struct pci_dev *pdev = adapter->pdev;
1779 /* Prevent stats update while adapter is being reset, link is down
1780 * or if the pci connection is down.
1782 if (adapter->link_speed == 0)
1785 if (test_bit(__IGBVF_RESETTING, &adapter->state))
1788 if (pci_channel_offline(pdev))
1791 UPDATE_VF_COUNTER(VFGPRC, gprc);
1792 UPDATE_VF_COUNTER(VFGORC, gorc);
1793 UPDATE_VF_COUNTER(VFGPTC, gptc);
1794 UPDATE_VF_COUNTER(VFGOTC, gotc);
1795 UPDATE_VF_COUNTER(VFMPRC, mprc);
1796 UPDATE_VF_COUNTER(VFGOTLBC, gotlbc);
1797 UPDATE_VF_COUNTER(VFGPTLBC, gptlbc);
1798 UPDATE_VF_COUNTER(VFGORLBC, gorlbc);
1799 UPDATE_VF_COUNTER(VFGPRLBC, gprlbc);
1801 /* Fill out the OS statistics structure */
1802 adapter->net_stats.multicast = adapter->stats.mprc;
1805 static void igbvf_print_link_info(struct igbvf_adapter *adapter)
1807 dev_info(&adapter->pdev->dev, "Link is Up %d Mbps %s Duplex\n",
1808 adapter->link_speed,
1809 adapter->link_duplex == FULL_DUPLEX ? "Full" : "Half");
1812 static bool igbvf_has_link(struct igbvf_adapter *adapter)
1814 struct e1000_hw *hw = &adapter->hw;
1815 s32 ret_val = E1000_SUCCESS;
1818 /* If interface is down, stay link down */
1819 if (test_bit(__IGBVF_DOWN, &adapter->state))
1822 ret_val = hw->mac.ops.check_for_link(hw);
1823 link_active = !hw->mac.get_link_status;
1825 /* if check for link returns error we will need to reset */
1826 if (ret_val && time_after(jiffies, adapter->last_reset + (10 * HZ)))
1827 schedule_work(&adapter->reset_task);
1833 * igbvf_watchdog - Timer Call-back
1834 * @data: pointer to adapter cast into an unsigned long
1836 static void igbvf_watchdog(unsigned long data)
1838 struct igbvf_adapter *adapter = (struct igbvf_adapter *)data;
1840 /* Do the rest outside of interrupt context */
1841 schedule_work(&adapter->watchdog_task);
1844 static void igbvf_watchdog_task(struct work_struct *work)
1846 struct igbvf_adapter *adapter = container_of(work,
1847 struct igbvf_adapter,
1849 struct net_device *netdev = adapter->netdev;
1850 struct e1000_mac_info *mac = &adapter->hw.mac;
1851 struct igbvf_ring *tx_ring = adapter->tx_ring;
1852 struct e1000_hw *hw = &adapter->hw;
1856 link = igbvf_has_link(adapter);
1859 if (!netif_carrier_ok(netdev)) {
1860 mac->ops.get_link_up_info(&adapter->hw,
1861 &adapter->link_speed,
1862 &adapter->link_duplex);
1863 igbvf_print_link_info(adapter);
1865 netif_carrier_on(netdev);
1866 netif_wake_queue(netdev);
1869 if (netif_carrier_ok(netdev)) {
1870 adapter->link_speed = 0;
1871 adapter->link_duplex = 0;
1872 dev_info(&adapter->pdev->dev, "Link is Down\n");
1873 netif_carrier_off(netdev);
1874 netif_stop_queue(netdev);
1878 if (netif_carrier_ok(netdev)) {
1879 igbvf_update_stats(adapter);
1881 tx_pending = (igbvf_desc_unused(tx_ring) + 1 <
1884 /* We've lost link, so the controller stops DMA,
1885 * but we've got queued Tx work that's never going
1886 * to get done, so reset controller to flush Tx.
1887 * (Do the reset outside of interrupt context).
1889 adapter->tx_timeout_count++;
1890 schedule_work(&adapter->reset_task);
1894 /* Cause software interrupt to ensure Rx ring is cleaned */
1895 ew32(EICS, adapter->rx_ring->eims_value);
1897 /* Reset the timer */
1898 if (!test_bit(__IGBVF_DOWN, &adapter->state))
1899 mod_timer(&adapter->watchdog_timer,
1900 round_jiffies(jiffies + (2 * HZ)));
1903 #define IGBVF_TX_FLAGS_CSUM 0x00000001
1904 #define IGBVF_TX_FLAGS_VLAN 0x00000002
1905 #define IGBVF_TX_FLAGS_TSO 0x00000004
1906 #define IGBVF_TX_FLAGS_IPV4 0x00000008
1907 #define IGBVF_TX_FLAGS_VLAN_MASK 0xffff0000
1908 #define IGBVF_TX_FLAGS_VLAN_SHIFT 16
1910 static int igbvf_tso(struct igbvf_adapter *adapter,
1911 struct igbvf_ring *tx_ring,
1912 struct sk_buff *skb, u32 tx_flags, u8 *hdr_len,
1915 struct e1000_adv_tx_context_desc *context_desc;
1916 struct igbvf_buffer *buffer_info;
1917 u32 info = 0, tu_cmd = 0;
1918 u32 mss_l4len_idx, l4len;
1924 err = skb_cow_head(skb, 0);
1926 dev_err(&adapter->pdev->dev, "igbvf_tso returning an error\n");
1930 l4len = tcp_hdrlen(skb);
1933 if (protocol == htons(ETH_P_IP)) {
1934 struct iphdr *iph = ip_hdr(skb);
1938 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
1942 } else if (skb_is_gso_v6(skb)) {
1943 ipv6_hdr(skb)->payload_len = 0;
1944 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
1945 &ipv6_hdr(skb)->daddr,
1949 i = tx_ring->next_to_use;
1951 buffer_info = &tx_ring->buffer_info[i];
1952 context_desc = IGBVF_TX_CTXTDESC_ADV(*tx_ring, i);
1953 /* VLAN MACLEN IPLEN */
1954 if (tx_flags & IGBVF_TX_FLAGS_VLAN)
1955 info |= (tx_flags & IGBVF_TX_FLAGS_VLAN_MASK);
1956 info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
1957 *hdr_len += skb_network_offset(skb);
1958 info |= (skb_transport_header(skb) - skb_network_header(skb));
1959 *hdr_len += (skb_transport_header(skb) - skb_network_header(skb));
1960 context_desc->vlan_macip_lens = cpu_to_le32(info);
1962 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
1963 tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
1965 if (protocol == htons(ETH_P_IP))
1966 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
1967 tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
1969 context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
1972 mss_l4len_idx = (skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT);
1973 mss_l4len_idx |= (l4len << E1000_ADVTXD_L4LEN_SHIFT);
1975 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
1976 context_desc->seqnum_seed = 0;
1978 buffer_info->time_stamp = jiffies;
1979 buffer_info->dma = 0;
1981 if (i == tx_ring->count)
1984 tx_ring->next_to_use = i;
1989 static inline bool igbvf_tx_csum(struct igbvf_adapter *adapter,
1990 struct igbvf_ring *tx_ring,
1991 struct sk_buff *skb, u32 tx_flags,
1994 struct e1000_adv_tx_context_desc *context_desc;
1996 struct igbvf_buffer *buffer_info;
1997 u32 info = 0, tu_cmd = 0;
1999 if ((skb->ip_summed == CHECKSUM_PARTIAL) ||
2000 (tx_flags & IGBVF_TX_FLAGS_VLAN)) {
2001 i = tx_ring->next_to_use;
2002 buffer_info = &tx_ring->buffer_info[i];
2003 context_desc = IGBVF_TX_CTXTDESC_ADV(*tx_ring, i);
2005 if (tx_flags & IGBVF_TX_FLAGS_VLAN)
2006 info |= (tx_flags & IGBVF_TX_FLAGS_VLAN_MASK);
2008 info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
2009 if (skb->ip_summed == CHECKSUM_PARTIAL)
2010 info |= (skb_transport_header(skb) -
2011 skb_network_header(skb));
2013 context_desc->vlan_macip_lens = cpu_to_le32(info);
2015 tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
2017 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2019 case htons(ETH_P_IP):
2020 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
2021 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
2022 tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
2024 case htons(ETH_P_IPV6):
2025 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
2026 tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
2033 context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
2034 context_desc->seqnum_seed = 0;
2035 context_desc->mss_l4len_idx = 0;
2037 buffer_info->time_stamp = jiffies;
2038 buffer_info->dma = 0;
2040 if (i == tx_ring->count)
2042 tx_ring->next_to_use = i;
2050 static int igbvf_maybe_stop_tx(struct net_device *netdev, int size)
2052 struct igbvf_adapter *adapter = netdev_priv(netdev);
2054 /* there is enough descriptors then we don't need to worry */
2055 if (igbvf_desc_unused(adapter->tx_ring) >= size)
2058 netif_stop_queue(netdev);
2060 /* Herbert's original patch had:
2061 * smp_mb__after_netif_stop_queue();
2062 * but since that doesn't exist yet, just open code it.
2066 /* We need to check again just in case room has been made available */
2067 if (igbvf_desc_unused(adapter->tx_ring) < size)
2070 netif_wake_queue(netdev);
2072 ++adapter->restart_queue;
2076 #define IGBVF_MAX_TXD_PWR 16
2077 #define IGBVF_MAX_DATA_PER_TXD (1 << IGBVF_MAX_TXD_PWR)
2079 static inline int igbvf_tx_map_adv(struct igbvf_adapter *adapter,
2080 struct igbvf_ring *tx_ring,
2081 struct sk_buff *skb)
2083 struct igbvf_buffer *buffer_info;
2084 struct pci_dev *pdev = adapter->pdev;
2085 unsigned int len = skb_headlen(skb);
2086 unsigned int count = 0, i;
2089 i = tx_ring->next_to_use;
2091 buffer_info = &tx_ring->buffer_info[i];
2092 BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD);
2093 buffer_info->length = len;
2094 /* set time_stamp *before* dma to help avoid a possible race */
2095 buffer_info->time_stamp = jiffies;
2096 buffer_info->mapped_as_page = false;
2097 buffer_info->dma = dma_map_single(&pdev->dev, skb->data, len,
2099 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2102 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) {
2103 const struct skb_frag_struct *frag;
2107 if (i == tx_ring->count)
2110 frag = &skb_shinfo(skb)->frags[f];
2111 len = skb_frag_size(frag);
2113 buffer_info = &tx_ring->buffer_info[i];
2114 BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD);
2115 buffer_info->length = len;
2116 buffer_info->time_stamp = jiffies;
2117 buffer_info->mapped_as_page = true;
2118 buffer_info->dma = skb_frag_dma_map(&pdev->dev, frag, 0, len,
2120 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2124 tx_ring->buffer_info[i].skb = skb;
2129 dev_err(&pdev->dev, "TX DMA map failed\n");
2131 /* clear timestamp and dma mappings for failed buffer_info mapping */
2132 buffer_info->dma = 0;
2133 buffer_info->time_stamp = 0;
2134 buffer_info->length = 0;
2135 buffer_info->mapped_as_page = false;
2139 /* clear timestamp and dma mappings for remaining portion of packet */
2142 i += tx_ring->count;
2144 buffer_info = &tx_ring->buffer_info[i];
2145 igbvf_put_txbuf(adapter, buffer_info);
2151 static inline void igbvf_tx_queue_adv(struct igbvf_adapter *adapter,
2152 struct igbvf_ring *tx_ring,
2153 int tx_flags, int count,
2154 unsigned int first, u32 paylen,
2157 union e1000_adv_tx_desc *tx_desc = NULL;
2158 struct igbvf_buffer *buffer_info;
2159 u32 olinfo_status = 0, cmd_type_len;
2162 cmd_type_len = (E1000_ADVTXD_DTYP_DATA | E1000_ADVTXD_DCMD_IFCS |
2163 E1000_ADVTXD_DCMD_DEXT);
2165 if (tx_flags & IGBVF_TX_FLAGS_VLAN)
2166 cmd_type_len |= E1000_ADVTXD_DCMD_VLE;
2168 if (tx_flags & IGBVF_TX_FLAGS_TSO) {
2169 cmd_type_len |= E1000_ADVTXD_DCMD_TSE;
2171 /* insert tcp checksum */
2172 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2174 /* insert ip checksum */
2175 if (tx_flags & IGBVF_TX_FLAGS_IPV4)
2176 olinfo_status |= E1000_TXD_POPTS_IXSM << 8;
2178 } else if (tx_flags & IGBVF_TX_FLAGS_CSUM) {
2179 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2182 olinfo_status |= ((paylen - hdr_len) << E1000_ADVTXD_PAYLEN_SHIFT);
2184 i = tx_ring->next_to_use;
2186 buffer_info = &tx_ring->buffer_info[i];
2187 tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
2188 tx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
2189 tx_desc->read.cmd_type_len =
2190 cpu_to_le32(cmd_type_len | buffer_info->length);
2191 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
2193 if (i == tx_ring->count)
2197 tx_desc->read.cmd_type_len |= cpu_to_le32(adapter->txd_cmd);
2198 /* Force memory writes to complete before letting h/w
2199 * know there are new descriptors to fetch. (Only
2200 * applicable for weak-ordered memory model archs,
2205 tx_ring->buffer_info[first].next_to_watch = tx_desc;
2206 tx_ring->next_to_use = i;
2207 writel(i, adapter->hw.hw_addr + tx_ring->tail);
2208 /* we need this if more than one processor can write to our tail
2209 * at a time, it synchronizes IO on IA64/Altix systems
2214 static netdev_tx_t igbvf_xmit_frame_ring_adv(struct sk_buff *skb,
2215 struct net_device *netdev,
2216 struct igbvf_ring *tx_ring)
2218 struct igbvf_adapter *adapter = netdev_priv(netdev);
2219 unsigned int first, tx_flags = 0;
2223 __be16 protocol = vlan_get_protocol(skb);
2225 if (test_bit(__IGBVF_DOWN, &adapter->state)) {
2226 dev_kfree_skb_any(skb);
2227 return NETDEV_TX_OK;
2230 if (skb->len <= 0) {
2231 dev_kfree_skb_any(skb);
2232 return NETDEV_TX_OK;
2235 /* need: count + 4 desc gap to keep tail from touching
2236 * + 2 desc gap to keep tail from touching head,
2237 * + 1 desc for skb->data,
2238 * + 1 desc for context descriptor,
2239 * head, otherwise try next time
2241 if (igbvf_maybe_stop_tx(netdev, skb_shinfo(skb)->nr_frags + 4)) {
2242 /* this is a hard error */
2243 return NETDEV_TX_BUSY;
2246 if (skb_vlan_tag_present(skb)) {
2247 tx_flags |= IGBVF_TX_FLAGS_VLAN;
2248 tx_flags |= (skb_vlan_tag_get(skb) <<
2249 IGBVF_TX_FLAGS_VLAN_SHIFT);
2252 if (protocol == htons(ETH_P_IP))
2253 tx_flags |= IGBVF_TX_FLAGS_IPV4;
2255 first = tx_ring->next_to_use;
2257 tso = skb_is_gso(skb) ?
2258 igbvf_tso(adapter, tx_ring, skb, tx_flags, &hdr_len, protocol) : 0;
2259 if (unlikely(tso < 0)) {
2260 dev_kfree_skb_any(skb);
2261 return NETDEV_TX_OK;
2265 tx_flags |= IGBVF_TX_FLAGS_TSO;
2266 else if (igbvf_tx_csum(adapter, tx_ring, skb, tx_flags, protocol) &&
2267 (skb->ip_summed == CHECKSUM_PARTIAL))
2268 tx_flags |= IGBVF_TX_FLAGS_CSUM;
2270 /* count reflects descriptors mapped, if 0 then mapping error
2271 * has occurred and we need to rewind the descriptor queue
2273 count = igbvf_tx_map_adv(adapter, tx_ring, skb);
2276 igbvf_tx_queue_adv(adapter, tx_ring, tx_flags, count,
2277 first, skb->len, hdr_len);
2278 /* Make sure there is space in the ring for the next send. */
2279 igbvf_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 4);
2281 dev_kfree_skb_any(skb);
2282 tx_ring->buffer_info[first].time_stamp = 0;
2283 tx_ring->next_to_use = first;
2286 return NETDEV_TX_OK;
2289 static netdev_tx_t igbvf_xmit_frame(struct sk_buff *skb,
2290 struct net_device *netdev)
2292 struct igbvf_adapter *adapter = netdev_priv(netdev);
2293 struct igbvf_ring *tx_ring;
2295 if (test_bit(__IGBVF_DOWN, &adapter->state)) {
2296 dev_kfree_skb_any(skb);
2297 return NETDEV_TX_OK;
2300 tx_ring = &adapter->tx_ring[0];
2302 return igbvf_xmit_frame_ring_adv(skb, netdev, tx_ring);
2306 * igbvf_tx_timeout - Respond to a Tx Hang
2307 * @netdev: network interface device structure
2309 static void igbvf_tx_timeout(struct net_device *netdev)
2311 struct igbvf_adapter *adapter = netdev_priv(netdev);
2313 /* Do the reset outside of interrupt context */
2314 adapter->tx_timeout_count++;
2315 schedule_work(&adapter->reset_task);
2318 static void igbvf_reset_task(struct work_struct *work)
2320 struct igbvf_adapter *adapter;
2322 adapter = container_of(work, struct igbvf_adapter, reset_task);
2324 igbvf_reinit_locked(adapter);
2328 * igbvf_get_stats - Get System Network Statistics
2329 * @netdev: network interface device structure
2331 * Returns the address of the device statistics structure.
2332 * The statistics are actually updated from the timer callback.
2334 static struct net_device_stats *igbvf_get_stats(struct net_device *netdev)
2336 struct igbvf_adapter *adapter = netdev_priv(netdev);
2338 /* only return the current stats */
2339 return &adapter->net_stats;
2343 * igbvf_change_mtu - Change the Maximum Transfer Unit
2344 * @netdev: network interface device structure
2345 * @new_mtu: new value for maximum frame size
2347 * Returns 0 on success, negative on failure
2349 static int igbvf_change_mtu(struct net_device *netdev, int new_mtu)
2351 struct igbvf_adapter *adapter = netdev_priv(netdev);
2352 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
2354 if (new_mtu < 68 || new_mtu > INT_MAX - ETH_HLEN - ETH_FCS_LEN ||
2355 max_frame > MAX_JUMBO_FRAME_SIZE)
2358 #define MAX_STD_JUMBO_FRAME_SIZE 9234
2359 if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
2360 dev_err(&adapter->pdev->dev, "MTU > 9216 not supported.\n");
2364 while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
2365 usleep_range(1000, 2000);
2366 /* igbvf_down has a dependency on max_frame_size */
2367 adapter->max_frame_size = max_frame;
2368 if (netif_running(netdev))
2369 igbvf_down(adapter);
2371 /* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
2372 * means we reserve 2 more, this pushes us to allocate from the next
2374 * i.e. RXBUFFER_2048 --> size-4096 slab
2375 * However with the new *_jumbo_rx* routines, jumbo receives will use
2379 if (max_frame <= 1024)
2380 adapter->rx_buffer_len = 1024;
2381 else if (max_frame <= 2048)
2382 adapter->rx_buffer_len = 2048;
2384 #if (PAGE_SIZE / 2) > 16384
2385 adapter->rx_buffer_len = 16384;
2387 adapter->rx_buffer_len = PAGE_SIZE / 2;
2390 /* adjust allocation if LPE protects us, and we aren't using SBP */
2391 if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
2392 (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
2393 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN +
2396 dev_info(&adapter->pdev->dev, "changing MTU from %d to %d\n",
2397 netdev->mtu, new_mtu);
2398 netdev->mtu = new_mtu;
2400 if (netif_running(netdev))
2403 igbvf_reset(adapter);
2405 clear_bit(__IGBVF_RESETTING, &adapter->state);
2410 static int igbvf_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
2418 static int igbvf_suspend(struct pci_dev *pdev, pm_message_t state)
2420 struct net_device *netdev = pci_get_drvdata(pdev);
2421 struct igbvf_adapter *adapter = netdev_priv(netdev);
2426 netif_device_detach(netdev);
2428 if (netif_running(netdev)) {
2429 WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
2430 igbvf_down(adapter);
2431 igbvf_free_irq(adapter);
2435 retval = pci_save_state(pdev);
2440 pci_disable_device(pdev);
2446 static int igbvf_resume(struct pci_dev *pdev)
2448 struct net_device *netdev = pci_get_drvdata(pdev);
2449 struct igbvf_adapter *adapter = netdev_priv(netdev);
2452 pci_restore_state(pdev);
2453 err = pci_enable_device_mem(pdev);
2455 dev_err(&pdev->dev, "Cannot enable PCI device from suspend\n");
2459 pci_set_master(pdev);
2461 if (netif_running(netdev)) {
2462 err = igbvf_request_irq(adapter);
2467 igbvf_reset(adapter);
2469 if (netif_running(netdev))
2472 netif_device_attach(netdev);
2478 static void igbvf_shutdown(struct pci_dev *pdev)
2480 igbvf_suspend(pdev, PMSG_SUSPEND);
2483 #ifdef CONFIG_NET_POLL_CONTROLLER
2484 /* Polling 'interrupt' - used by things like netconsole to send skbs
2485 * without having to re-enable interrupts. It's not called while
2486 * the interrupt routine is executing.
2488 static void igbvf_netpoll(struct net_device *netdev)
2490 struct igbvf_adapter *adapter = netdev_priv(netdev);
2492 disable_irq(adapter->pdev->irq);
2494 igbvf_clean_tx_irq(adapter->tx_ring);
2496 enable_irq(adapter->pdev->irq);
2501 * igbvf_io_error_detected - called when PCI error is detected
2502 * @pdev: Pointer to PCI device
2503 * @state: The current pci connection state
2505 * This function is called after a PCI bus error affecting
2506 * this device has been detected.
2508 static pci_ers_result_t igbvf_io_error_detected(struct pci_dev *pdev,
2509 pci_channel_state_t state)
2511 struct net_device *netdev = pci_get_drvdata(pdev);
2512 struct igbvf_adapter *adapter = netdev_priv(netdev);
2514 netif_device_detach(netdev);
2516 if (state == pci_channel_io_perm_failure)
2517 return PCI_ERS_RESULT_DISCONNECT;
2519 if (netif_running(netdev))
2520 igbvf_down(adapter);
2521 pci_disable_device(pdev);
2523 /* Request a slot slot reset. */
2524 return PCI_ERS_RESULT_NEED_RESET;
2528 * igbvf_io_slot_reset - called after the pci bus has been reset.
2529 * @pdev: Pointer to PCI device
2531 * Restart the card from scratch, as if from a cold-boot. Implementation
2532 * resembles the first-half of the igbvf_resume routine.
2534 static pci_ers_result_t igbvf_io_slot_reset(struct pci_dev *pdev)
2536 struct net_device *netdev = pci_get_drvdata(pdev);
2537 struct igbvf_adapter *adapter = netdev_priv(netdev);
2539 if (pci_enable_device_mem(pdev)) {
2541 "Cannot re-enable PCI device after reset.\n");
2542 return PCI_ERS_RESULT_DISCONNECT;
2544 pci_set_master(pdev);
2546 igbvf_reset(adapter);
2548 return PCI_ERS_RESULT_RECOVERED;
2552 * igbvf_io_resume - called when traffic can start flowing again.
2553 * @pdev: Pointer to PCI device
2555 * This callback is called when the error recovery driver tells us that
2556 * its OK to resume normal operation. Implementation resembles the
2557 * second-half of the igbvf_resume routine.
2559 static void igbvf_io_resume(struct pci_dev *pdev)
2561 struct net_device *netdev = pci_get_drvdata(pdev);
2562 struct igbvf_adapter *adapter = netdev_priv(netdev);
2564 if (netif_running(netdev)) {
2565 if (igbvf_up(adapter)) {
2567 "can't bring device back up after reset\n");
2572 netif_device_attach(netdev);
2575 static void igbvf_print_device_info(struct igbvf_adapter *adapter)
2577 struct e1000_hw *hw = &adapter->hw;
2578 struct net_device *netdev = adapter->netdev;
2579 struct pci_dev *pdev = adapter->pdev;
2581 if (hw->mac.type == e1000_vfadapt_i350)
2582 dev_info(&pdev->dev, "Intel(R) I350 Virtual Function\n");
2584 dev_info(&pdev->dev, "Intel(R) 82576 Virtual Function\n");
2585 dev_info(&pdev->dev, "Address: %pM\n", netdev->dev_addr);
2588 static int igbvf_set_features(struct net_device *netdev,
2589 netdev_features_t features)
2591 struct igbvf_adapter *adapter = netdev_priv(netdev);
2593 if (features & NETIF_F_RXCSUM)
2594 adapter->flags &= ~IGBVF_FLAG_RX_CSUM_DISABLED;
2596 adapter->flags |= IGBVF_FLAG_RX_CSUM_DISABLED;
2601 static const struct net_device_ops igbvf_netdev_ops = {
2602 .ndo_open = igbvf_open,
2603 .ndo_stop = igbvf_close,
2604 .ndo_start_xmit = igbvf_xmit_frame,
2605 .ndo_get_stats = igbvf_get_stats,
2606 .ndo_set_rx_mode = igbvf_set_multi,
2607 .ndo_set_mac_address = igbvf_set_mac,
2608 .ndo_change_mtu = igbvf_change_mtu,
2609 .ndo_do_ioctl = igbvf_ioctl,
2610 .ndo_tx_timeout = igbvf_tx_timeout,
2611 .ndo_vlan_rx_add_vid = igbvf_vlan_rx_add_vid,
2612 .ndo_vlan_rx_kill_vid = igbvf_vlan_rx_kill_vid,
2613 #ifdef CONFIG_NET_POLL_CONTROLLER
2614 .ndo_poll_controller = igbvf_netpoll,
2616 .ndo_set_features = igbvf_set_features,
2620 * igbvf_probe - Device Initialization Routine
2621 * @pdev: PCI device information struct
2622 * @ent: entry in igbvf_pci_tbl
2624 * Returns 0 on success, negative on failure
2626 * igbvf_probe initializes an adapter identified by a pci_dev structure.
2627 * The OS initialization, configuring of the adapter private structure,
2628 * and a hardware reset occur.
2630 static int igbvf_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
2632 struct net_device *netdev;
2633 struct igbvf_adapter *adapter;
2634 struct e1000_hw *hw;
2635 const struct igbvf_info *ei = igbvf_info_tbl[ent->driver_data];
2637 static int cards_found;
2638 int err, pci_using_dac;
2640 err = pci_enable_device_mem(pdev);
2645 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
2649 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
2652 "No usable DMA 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);
2700 goto err_get_variants;
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_CTAG_TX |
2726 NETIF_F_HW_VLAN_CTAG_RX |
2727 NETIF_F_HW_VLAN_CTAG_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,
2749 "MAC address not assigned by administrator.\n");
2750 memcpy(netdev->dev_addr, adapter->hw.mac.addr,
2754 if (!is_valid_ether_addr(netdev->dev_addr)) {
2755 dev_info(&pdev->dev, "Assigning random MAC address.\n");
2756 eth_hw_addr_random(netdev);
2757 memcpy(adapter->hw.mac.addr, netdev->dev_addr,
2761 setup_timer(&adapter->watchdog_timer, &igbvf_watchdog,
2762 (unsigned long)adapter);
2764 INIT_WORK(&adapter->reset_task, igbvf_reset_task);
2765 INIT_WORK(&adapter->watchdog_task, igbvf_watchdog_task);
2767 /* ring size defaults */
2768 adapter->rx_ring->count = 1024;
2769 adapter->tx_ring->count = 1024;
2771 /* reset the hardware with the new settings */
2772 igbvf_reset(adapter);
2774 /* set hardware-specific flags */
2775 if (adapter->hw.mac.type == e1000_vfadapt_i350)
2776 adapter->flags |= IGBVF_FLAG_RX_LB_VLAN_BSWAP;
2778 strcpy(netdev->name, "eth%d");
2779 err = register_netdev(netdev);
2783 /* tell the stack to leave us alone until igbvf_open() is called */
2784 netif_carrier_off(netdev);
2785 netif_stop_queue(netdev);
2787 igbvf_print_device_info(adapter);
2789 igbvf_initialize_last_counter_stats(adapter);
2794 kfree(adapter->tx_ring);
2795 kfree(adapter->rx_ring);
2797 igbvf_reset_interrupt_capability(adapter);
2799 iounmap(adapter->hw.hw_addr);
2801 free_netdev(netdev);
2803 pci_release_regions(pdev);
2806 pci_disable_device(pdev);
2811 * igbvf_remove - Device Removal Routine
2812 * @pdev: PCI device information struct
2814 * igbvf_remove is called by the PCI subsystem to alert the driver
2815 * that it should release a PCI device. The could be caused by a
2816 * Hot-Plug event, or because the driver is going to be removed from
2819 static void igbvf_remove(struct pci_dev *pdev)
2821 struct net_device *netdev = pci_get_drvdata(pdev);
2822 struct igbvf_adapter *adapter = netdev_priv(netdev);
2823 struct e1000_hw *hw = &adapter->hw;
2825 /* The watchdog timer may be rescheduled, so explicitly
2826 * disable it from being rescheduled.
2828 set_bit(__IGBVF_DOWN, &adapter->state);
2829 del_timer_sync(&adapter->watchdog_timer);
2831 cancel_work_sync(&adapter->reset_task);
2832 cancel_work_sync(&adapter->watchdog_task);
2834 unregister_netdev(netdev);
2836 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 const struct pci_device_id 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)
2894 pr_info("%s - version %s\n", igbvf_driver_string, igbvf_driver_version);
2895 pr_info("%s\n", igbvf_copyright);
2897 ret = pci_register_driver(&igbvf_driver);
2901 module_init(igbvf_init_module);
2904 * igbvf_exit_module - Driver Exit Cleanup Routine
2906 * igbvf_exit_module is called just before the driver is removed
2909 static void __exit igbvf_exit_module(void)
2911 pci_unregister_driver(&igbvf_driver);
2913 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);