1 /*******************************************************************************
3 Intel(R) 82576 Virtual Function Linux driver
4 Copyright(c) 2009 - 2012 Intel Corporation.
6 This program is free software; you can redistribute it and/or modify it
7 under the terms and conditions of the GNU General Public License,
8 version 2, as published by the Free Software Foundation.
10 This program is distributed in the hope it will be useful, but WITHOUT
11 ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 You should have received a copy of the GNU General Public License along with
16 this program; if not, see <http://www.gnu.org/licenses/>.
18 The full GNU General Public License is included in this distribution in
19 the file called "COPYING".
22 e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
23 Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
25 *******************************************************************************/
27 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
29 #include <linux/module.h>
30 #include <linux/types.h>
31 #include <linux/init.h>
32 #include <linux/pci.h>
33 #include <linux/vmalloc.h>
34 #include <linux/pagemap.h>
35 #include <linux/delay.h>
36 #include <linux/netdevice.h>
37 #include <linux/tcp.h>
38 #include <linux/ipv6.h>
39 #include <linux/slab.h>
40 #include <net/checksum.h>
41 #include <net/ip6_checksum.h>
42 #include <linux/mii.h>
43 #include <linux/ethtool.h>
44 #include <linux/if_vlan.h>
45 #include <linux/prefetch.h>
46 #include <linux/sctp.h>
50 #define DRV_VERSION "2.4.0-k"
51 char igbvf_driver_name[] = "igbvf";
52 const char igbvf_driver_version[] = DRV_VERSION;
53 static const char igbvf_driver_string[] =
54 "Intel(R) Gigabit Virtual Function Network Driver";
55 static const char igbvf_copyright[] =
56 "Copyright (c) 2009 - 2012 Intel Corporation.";
58 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
59 static int debug = -1;
60 module_param(debug, int, 0);
61 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
63 static int igbvf_poll(struct napi_struct *napi, int budget);
64 static void igbvf_reset(struct igbvf_adapter *);
65 static void igbvf_set_interrupt_capability(struct igbvf_adapter *);
66 static void igbvf_reset_interrupt_capability(struct igbvf_adapter *);
68 static struct igbvf_info igbvf_vf_info = {
72 .init_ops = e1000_init_function_pointers_vf,
75 static struct igbvf_info igbvf_i350_vf_info = {
76 .mac = e1000_vfadapt_i350,
79 .init_ops = e1000_init_function_pointers_vf,
82 static const struct igbvf_info *igbvf_info_tbl[] = {
83 [board_vf] = &igbvf_vf_info,
84 [board_i350_vf] = &igbvf_i350_vf_info,
88 * igbvf_desc_unused - calculate if we have unused descriptors
89 * @rx_ring: address of receive ring structure
91 static int igbvf_desc_unused(struct igbvf_ring *ring)
93 if (ring->next_to_clean > ring->next_to_use)
94 return ring->next_to_clean - ring->next_to_use - 1;
96 return ring->count + ring->next_to_clean - ring->next_to_use - 1;
100 * igbvf_receive_skb - helper function to handle Rx indications
101 * @adapter: board private structure
102 * @status: descriptor status field as written by hardware
103 * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
104 * @skb: pointer to sk_buff to be indicated to stack
106 static void igbvf_receive_skb(struct igbvf_adapter *adapter,
107 struct net_device *netdev,
109 u32 status, u16 vlan)
113 if (status & E1000_RXD_STAT_VP) {
114 if ((adapter->flags & IGBVF_FLAG_RX_LB_VLAN_BSWAP) &&
115 (status & E1000_RXDEXT_STATERR_LB))
116 vid = be16_to_cpu(vlan) & E1000_RXD_SPC_VLAN_MASK;
118 vid = le16_to_cpu(vlan) & E1000_RXD_SPC_VLAN_MASK;
119 if (test_bit(vid, adapter->active_vlans))
120 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
123 napi_gro_receive(&adapter->rx_ring->napi, skb);
126 static inline void igbvf_rx_checksum_adv(struct igbvf_adapter *adapter,
127 u32 status_err, struct sk_buff *skb)
129 skb_checksum_none_assert(skb);
131 /* Ignore Checksum bit is set or checksum is disabled through ethtool */
132 if ((status_err & E1000_RXD_STAT_IXSM) ||
133 (adapter->flags & IGBVF_FLAG_RX_CSUM_DISABLED))
136 /* TCP/UDP checksum error bit is set */
138 (E1000_RXDEXT_STATERR_TCPE | E1000_RXDEXT_STATERR_IPE)) {
139 /* let the stack verify checksum errors */
140 adapter->hw_csum_err++;
144 /* It must be a TCP or UDP packet with a valid checksum */
145 if (status_err & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS))
146 skb->ip_summed = CHECKSUM_UNNECESSARY;
148 adapter->hw_csum_good++;
152 * igbvf_alloc_rx_buffers - Replace used receive buffers; packet split
153 * @rx_ring: address of ring structure to repopulate
154 * @cleaned_count: number of buffers to repopulate
156 static void igbvf_alloc_rx_buffers(struct igbvf_ring *rx_ring,
159 struct igbvf_adapter *adapter = rx_ring->adapter;
160 struct net_device *netdev = adapter->netdev;
161 struct pci_dev *pdev = adapter->pdev;
162 union e1000_adv_rx_desc *rx_desc;
163 struct igbvf_buffer *buffer_info;
168 i = rx_ring->next_to_use;
169 buffer_info = &rx_ring->buffer_info[i];
171 if (adapter->rx_ps_hdr_size)
172 bufsz = adapter->rx_ps_hdr_size;
174 bufsz = adapter->rx_buffer_len;
176 while (cleaned_count--) {
177 rx_desc = IGBVF_RX_DESC_ADV(*rx_ring, i);
179 if (adapter->rx_ps_hdr_size && !buffer_info->page_dma) {
180 if (!buffer_info->page) {
181 buffer_info->page = alloc_page(GFP_ATOMIC);
182 if (!buffer_info->page) {
183 adapter->alloc_rx_buff_failed++;
186 buffer_info->page_offset = 0;
188 buffer_info->page_offset ^= PAGE_SIZE / 2;
190 buffer_info->page_dma =
191 dma_map_page(&pdev->dev, buffer_info->page,
192 buffer_info->page_offset,
195 if (dma_mapping_error(&pdev->dev,
196 buffer_info->page_dma)) {
197 __free_page(buffer_info->page);
198 buffer_info->page = NULL;
199 dev_err(&pdev->dev, "RX DMA map failed\n");
204 if (!buffer_info->skb) {
205 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
207 adapter->alloc_rx_buff_failed++;
211 buffer_info->skb = skb;
212 buffer_info->dma = dma_map_single(&pdev->dev, skb->data,
215 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
216 dev_kfree_skb(buffer_info->skb);
217 buffer_info->skb = NULL;
218 dev_err(&pdev->dev, "RX DMA map failed\n");
222 /* Refresh the desc even if buffer_addrs didn't change because
223 * each write-back erases this info.
225 if (adapter->rx_ps_hdr_size) {
226 rx_desc->read.pkt_addr =
227 cpu_to_le64(buffer_info->page_dma);
228 rx_desc->read.hdr_addr = cpu_to_le64(buffer_info->dma);
230 rx_desc->read.pkt_addr = cpu_to_le64(buffer_info->dma);
231 rx_desc->read.hdr_addr = 0;
235 if (i == rx_ring->count)
237 buffer_info = &rx_ring->buffer_info[i];
241 if (rx_ring->next_to_use != i) {
242 rx_ring->next_to_use = i;
244 i = (rx_ring->count - 1);
248 /* Force memory writes to complete before letting h/w
249 * know there are new descriptors to fetch. (Only
250 * applicable for weak-ordered memory model archs,
254 writel(i, adapter->hw.hw_addr + rx_ring->tail);
259 * igbvf_clean_rx_irq - Send received data up the network stack; legacy
260 * @adapter: board private structure
262 * the return value indicates whether actual cleaning was done, there
263 * is no guarantee that everything was cleaned
265 static bool igbvf_clean_rx_irq(struct igbvf_adapter *adapter,
266 int *work_done, int work_to_do)
268 struct igbvf_ring *rx_ring = adapter->rx_ring;
269 struct net_device *netdev = adapter->netdev;
270 struct pci_dev *pdev = adapter->pdev;
271 union e1000_adv_rx_desc *rx_desc, *next_rxd;
272 struct igbvf_buffer *buffer_info, *next_buffer;
274 bool cleaned = false;
275 int cleaned_count = 0;
276 unsigned int total_bytes = 0, total_packets = 0;
278 u32 length, hlen, staterr;
280 i = rx_ring->next_to_clean;
281 rx_desc = IGBVF_RX_DESC_ADV(*rx_ring, i);
282 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
284 while (staterr & E1000_RXD_STAT_DD) {
285 if (*work_done >= work_to_do)
288 rmb(); /* read descriptor and rx_buffer_info after status DD */
290 buffer_info = &rx_ring->buffer_info[i];
292 /* HW will not DMA in data larger than the given buffer, even
293 * if it parses the (NFS, of course) header to be larger. In
294 * that case, it fills the header buffer and spills the rest
297 hlen = (le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.hdr_info)
298 & E1000_RXDADV_HDRBUFLEN_MASK) >>
299 E1000_RXDADV_HDRBUFLEN_SHIFT;
300 if (hlen > adapter->rx_ps_hdr_size)
301 hlen = adapter->rx_ps_hdr_size;
303 length = le16_to_cpu(rx_desc->wb.upper.length);
307 skb = buffer_info->skb;
308 prefetch(skb->data - NET_IP_ALIGN);
309 buffer_info->skb = NULL;
310 if (!adapter->rx_ps_hdr_size) {
311 dma_unmap_single(&pdev->dev, buffer_info->dma,
312 adapter->rx_buffer_len,
314 buffer_info->dma = 0;
315 skb_put(skb, length);
319 if (!skb_shinfo(skb)->nr_frags) {
320 dma_unmap_single(&pdev->dev, buffer_info->dma,
321 adapter->rx_ps_hdr_size,
323 buffer_info->dma = 0;
328 dma_unmap_page(&pdev->dev, buffer_info->page_dma,
331 buffer_info->page_dma = 0;
333 skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
335 buffer_info->page_offset,
338 if ((adapter->rx_buffer_len > (PAGE_SIZE / 2)) ||
339 (page_count(buffer_info->page) != 1))
340 buffer_info->page = NULL;
342 get_page(buffer_info->page);
345 skb->data_len += length;
346 skb->truesize += PAGE_SIZE / 2;
350 if (i == rx_ring->count)
352 next_rxd = IGBVF_RX_DESC_ADV(*rx_ring, i);
354 next_buffer = &rx_ring->buffer_info[i];
356 if (!(staterr & E1000_RXD_STAT_EOP)) {
357 buffer_info->skb = next_buffer->skb;
358 buffer_info->dma = next_buffer->dma;
359 next_buffer->skb = skb;
360 next_buffer->dma = 0;
364 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
365 dev_kfree_skb_irq(skb);
369 total_bytes += skb->len;
372 igbvf_rx_checksum_adv(adapter, staterr, skb);
374 skb->protocol = eth_type_trans(skb, netdev);
376 igbvf_receive_skb(adapter, netdev, skb, staterr,
377 rx_desc->wb.upper.vlan);
380 rx_desc->wb.upper.status_error = 0;
382 /* return some buffers to hardware, one at a time is too slow */
383 if (cleaned_count >= IGBVF_RX_BUFFER_WRITE) {
384 igbvf_alloc_rx_buffers(rx_ring, cleaned_count);
388 /* use prefetched values */
390 buffer_info = next_buffer;
392 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
395 rx_ring->next_to_clean = i;
396 cleaned_count = igbvf_desc_unused(rx_ring);
399 igbvf_alloc_rx_buffers(rx_ring, cleaned_count);
401 adapter->total_rx_packets += total_packets;
402 adapter->total_rx_bytes += total_bytes;
403 netdev->stats.rx_bytes += total_bytes;
404 netdev->stats.rx_packets += total_packets;
408 static void igbvf_put_txbuf(struct igbvf_adapter *adapter,
409 struct igbvf_buffer *buffer_info)
411 if (buffer_info->dma) {
412 if (buffer_info->mapped_as_page)
413 dma_unmap_page(&adapter->pdev->dev,
418 dma_unmap_single(&adapter->pdev->dev,
422 buffer_info->dma = 0;
424 if (buffer_info->skb) {
425 dev_kfree_skb_any(buffer_info->skb);
426 buffer_info->skb = NULL;
428 buffer_info->time_stamp = 0;
432 * igbvf_setup_tx_resources - allocate Tx resources (Descriptors)
433 * @adapter: board private structure
435 * Return 0 on success, negative on failure
437 int igbvf_setup_tx_resources(struct igbvf_adapter *adapter,
438 struct igbvf_ring *tx_ring)
440 struct pci_dev *pdev = adapter->pdev;
443 size = sizeof(struct igbvf_buffer) * tx_ring->count;
444 tx_ring->buffer_info = vzalloc(size);
445 if (!tx_ring->buffer_info)
448 /* round up to nearest 4K */
449 tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
450 tx_ring->size = ALIGN(tx_ring->size, 4096);
452 tx_ring->desc = dma_alloc_coherent(&pdev->dev, tx_ring->size,
453 &tx_ring->dma, GFP_KERNEL);
457 tx_ring->adapter = adapter;
458 tx_ring->next_to_use = 0;
459 tx_ring->next_to_clean = 0;
463 vfree(tx_ring->buffer_info);
464 dev_err(&adapter->pdev->dev,
465 "Unable to allocate memory for the transmit descriptor ring\n");
470 * igbvf_setup_rx_resources - allocate Rx resources (Descriptors)
471 * @adapter: board private structure
473 * Returns 0 on success, negative on failure
475 int igbvf_setup_rx_resources(struct igbvf_adapter *adapter,
476 struct igbvf_ring *rx_ring)
478 struct pci_dev *pdev = adapter->pdev;
481 size = sizeof(struct igbvf_buffer) * rx_ring->count;
482 rx_ring->buffer_info = vzalloc(size);
483 if (!rx_ring->buffer_info)
486 desc_len = sizeof(union e1000_adv_rx_desc);
488 /* Round up to nearest 4K */
489 rx_ring->size = rx_ring->count * desc_len;
490 rx_ring->size = ALIGN(rx_ring->size, 4096);
492 rx_ring->desc = dma_alloc_coherent(&pdev->dev, rx_ring->size,
493 &rx_ring->dma, GFP_KERNEL);
497 rx_ring->next_to_clean = 0;
498 rx_ring->next_to_use = 0;
500 rx_ring->adapter = adapter;
505 vfree(rx_ring->buffer_info);
506 rx_ring->buffer_info = NULL;
507 dev_err(&adapter->pdev->dev,
508 "Unable to allocate memory for the receive descriptor ring\n");
513 * igbvf_clean_tx_ring - Free Tx Buffers
514 * @tx_ring: ring to be cleaned
516 static void igbvf_clean_tx_ring(struct igbvf_ring *tx_ring)
518 struct igbvf_adapter *adapter = tx_ring->adapter;
519 struct igbvf_buffer *buffer_info;
523 if (!tx_ring->buffer_info)
526 /* Free all the Tx ring sk_buffs */
527 for (i = 0; i < tx_ring->count; i++) {
528 buffer_info = &tx_ring->buffer_info[i];
529 igbvf_put_txbuf(adapter, buffer_info);
532 size = sizeof(struct igbvf_buffer) * tx_ring->count;
533 memset(tx_ring->buffer_info, 0, size);
535 /* Zero out the descriptor ring */
536 memset(tx_ring->desc, 0, tx_ring->size);
538 tx_ring->next_to_use = 0;
539 tx_ring->next_to_clean = 0;
541 writel(0, adapter->hw.hw_addr + tx_ring->head);
542 writel(0, adapter->hw.hw_addr + tx_ring->tail);
546 * igbvf_free_tx_resources - Free Tx Resources per Queue
547 * @tx_ring: ring to free resources from
549 * Free all transmit software resources
551 void igbvf_free_tx_resources(struct igbvf_ring *tx_ring)
553 struct pci_dev *pdev = tx_ring->adapter->pdev;
555 igbvf_clean_tx_ring(tx_ring);
557 vfree(tx_ring->buffer_info);
558 tx_ring->buffer_info = NULL;
560 dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
563 tx_ring->desc = NULL;
567 * igbvf_clean_rx_ring - Free Rx Buffers per Queue
568 * @adapter: board private structure
570 static void igbvf_clean_rx_ring(struct igbvf_ring *rx_ring)
572 struct igbvf_adapter *adapter = rx_ring->adapter;
573 struct igbvf_buffer *buffer_info;
574 struct pci_dev *pdev = adapter->pdev;
578 if (!rx_ring->buffer_info)
581 /* Free all the Rx ring sk_buffs */
582 for (i = 0; i < rx_ring->count; i++) {
583 buffer_info = &rx_ring->buffer_info[i];
584 if (buffer_info->dma) {
585 if (adapter->rx_ps_hdr_size) {
586 dma_unmap_single(&pdev->dev, buffer_info->dma,
587 adapter->rx_ps_hdr_size,
590 dma_unmap_single(&pdev->dev, buffer_info->dma,
591 adapter->rx_buffer_len,
594 buffer_info->dma = 0;
597 if (buffer_info->skb) {
598 dev_kfree_skb(buffer_info->skb);
599 buffer_info->skb = NULL;
602 if (buffer_info->page) {
603 if (buffer_info->page_dma)
604 dma_unmap_page(&pdev->dev,
605 buffer_info->page_dma,
608 put_page(buffer_info->page);
609 buffer_info->page = NULL;
610 buffer_info->page_dma = 0;
611 buffer_info->page_offset = 0;
615 size = sizeof(struct igbvf_buffer) * rx_ring->count;
616 memset(rx_ring->buffer_info, 0, size);
618 /* Zero out the descriptor ring */
619 memset(rx_ring->desc, 0, rx_ring->size);
621 rx_ring->next_to_clean = 0;
622 rx_ring->next_to_use = 0;
624 writel(0, adapter->hw.hw_addr + rx_ring->head);
625 writel(0, adapter->hw.hw_addr + rx_ring->tail);
629 * igbvf_free_rx_resources - Free Rx Resources
630 * @rx_ring: ring to clean the resources from
632 * Free all receive software resources
635 void igbvf_free_rx_resources(struct igbvf_ring *rx_ring)
637 struct pci_dev *pdev = rx_ring->adapter->pdev;
639 igbvf_clean_rx_ring(rx_ring);
641 vfree(rx_ring->buffer_info);
642 rx_ring->buffer_info = NULL;
644 dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
646 rx_ring->desc = NULL;
650 * igbvf_update_itr - update the dynamic ITR value based on statistics
651 * @adapter: pointer to adapter
652 * @itr_setting: current adapter->itr
653 * @packets: the number of packets during this measurement interval
654 * @bytes: the number of bytes during this measurement interval
656 * Stores a new ITR value based on packets and byte counts during the last
657 * interrupt. The advantage of per interrupt computation is faster updates
658 * and more accurate ITR for the current traffic pattern. Constants in this
659 * function were computed based on theoretical maximum wire speed and thresholds
660 * were set based on testing data as well as attempting to minimize response
661 * time while increasing bulk throughput.
663 static enum latency_range igbvf_update_itr(struct igbvf_adapter *adapter,
664 enum latency_range itr_setting,
665 int packets, int bytes)
667 enum latency_range retval = itr_setting;
670 goto update_itr_done;
672 switch (itr_setting) {
674 /* handle TSO and jumbo frames */
675 if (bytes/packets > 8000)
676 retval = bulk_latency;
677 else if ((packets < 5) && (bytes > 512))
678 retval = low_latency;
680 case low_latency: /* 50 usec aka 20000 ints/s */
682 /* this if handles the TSO accounting */
683 if (bytes/packets > 8000)
684 retval = bulk_latency;
685 else if ((packets < 10) || ((bytes/packets) > 1200))
686 retval = bulk_latency;
687 else if ((packets > 35))
688 retval = lowest_latency;
689 } else if (bytes/packets > 2000) {
690 retval = bulk_latency;
691 } else if (packets <= 2 && bytes < 512) {
692 retval = lowest_latency;
695 case bulk_latency: /* 250 usec aka 4000 ints/s */
698 retval = low_latency;
699 } else if (bytes < 6000) {
700 retval = low_latency;
711 static int igbvf_range_to_itr(enum latency_range current_range)
715 switch (current_range) {
716 /* counts and packets in update_itr are dependent on these numbers */
718 new_itr = IGBVF_70K_ITR;
721 new_itr = IGBVF_20K_ITR;
724 new_itr = IGBVF_4K_ITR;
727 new_itr = IGBVF_START_ITR;
733 static void igbvf_set_itr(struct igbvf_adapter *adapter)
737 adapter->tx_ring->itr_range =
738 igbvf_update_itr(adapter,
739 adapter->tx_ring->itr_val,
740 adapter->total_tx_packets,
741 adapter->total_tx_bytes);
743 /* conservative mode (itr 3) eliminates the lowest_latency setting */
744 if (adapter->requested_itr == 3 &&
745 adapter->tx_ring->itr_range == lowest_latency)
746 adapter->tx_ring->itr_range = low_latency;
748 new_itr = igbvf_range_to_itr(adapter->tx_ring->itr_range);
750 if (new_itr != adapter->tx_ring->itr_val) {
751 u32 current_itr = adapter->tx_ring->itr_val;
752 /* this attempts to bias the interrupt rate towards Bulk
753 * by adding intermediate steps when interrupt rate is
756 new_itr = new_itr > current_itr ?
757 min(current_itr + (new_itr >> 2), new_itr) :
759 adapter->tx_ring->itr_val = new_itr;
761 adapter->tx_ring->set_itr = 1;
764 adapter->rx_ring->itr_range =
765 igbvf_update_itr(adapter, adapter->rx_ring->itr_val,
766 adapter->total_rx_packets,
767 adapter->total_rx_bytes);
768 if (adapter->requested_itr == 3 &&
769 adapter->rx_ring->itr_range == lowest_latency)
770 adapter->rx_ring->itr_range = low_latency;
772 new_itr = igbvf_range_to_itr(adapter->rx_ring->itr_range);
774 if (new_itr != adapter->rx_ring->itr_val) {
775 u32 current_itr = adapter->rx_ring->itr_val;
777 new_itr = new_itr > current_itr ?
778 min(current_itr + (new_itr >> 2), new_itr) :
780 adapter->rx_ring->itr_val = new_itr;
782 adapter->rx_ring->set_itr = 1;
787 * igbvf_clean_tx_irq - Reclaim resources after transmit completes
788 * @adapter: board private structure
790 * returns true if ring is completely cleaned
792 static bool igbvf_clean_tx_irq(struct igbvf_ring *tx_ring)
794 struct igbvf_adapter *adapter = tx_ring->adapter;
795 struct net_device *netdev = adapter->netdev;
796 struct igbvf_buffer *buffer_info;
798 union e1000_adv_tx_desc *tx_desc, *eop_desc;
799 unsigned int total_bytes = 0, total_packets = 0;
800 unsigned int i, count = 0;
801 bool cleaned = false;
803 i = tx_ring->next_to_clean;
804 buffer_info = &tx_ring->buffer_info[i];
805 eop_desc = buffer_info->next_to_watch;
808 /* if next_to_watch is not set then there is no work pending */
812 /* prevent any other reads prior to eop_desc */
815 /* if DD is not set pending work has not been completed */
816 if (!(eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)))
819 /* clear next_to_watch to prevent false hangs */
820 buffer_info->next_to_watch = NULL;
822 for (cleaned = false; !cleaned; count++) {
823 tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
824 cleaned = (tx_desc == eop_desc);
825 skb = buffer_info->skb;
828 unsigned int segs, bytecount;
830 /* gso_segs is currently only valid for tcp */
831 segs = skb_shinfo(skb)->gso_segs ?: 1;
832 /* multiply data chunks by size of headers */
833 bytecount = ((segs - 1) * skb_headlen(skb)) +
835 total_packets += segs;
836 total_bytes += bytecount;
839 igbvf_put_txbuf(adapter, buffer_info);
840 tx_desc->wb.status = 0;
843 if (i == tx_ring->count)
846 buffer_info = &tx_ring->buffer_info[i];
849 eop_desc = buffer_info->next_to_watch;
850 } while (count < tx_ring->count);
852 tx_ring->next_to_clean = i;
854 if (unlikely(count && netif_carrier_ok(netdev) &&
855 igbvf_desc_unused(tx_ring) >= IGBVF_TX_QUEUE_WAKE)) {
856 /* Make sure that anybody stopping the queue after this
857 * sees the new next_to_clean.
860 if (netif_queue_stopped(netdev) &&
861 !(test_bit(__IGBVF_DOWN, &adapter->state))) {
862 netif_wake_queue(netdev);
863 ++adapter->restart_queue;
867 netdev->stats.tx_bytes += total_bytes;
868 netdev->stats.tx_packets += total_packets;
869 return count < tx_ring->count;
872 static irqreturn_t igbvf_msix_other(int irq, void *data)
874 struct net_device *netdev = data;
875 struct igbvf_adapter *adapter = netdev_priv(netdev);
876 struct e1000_hw *hw = &adapter->hw;
878 adapter->int_counter1++;
880 hw->mac.get_link_status = 1;
881 if (!test_bit(__IGBVF_DOWN, &adapter->state))
882 mod_timer(&adapter->watchdog_timer, jiffies + 1);
884 ew32(EIMS, adapter->eims_other);
889 static irqreturn_t igbvf_intr_msix_tx(int irq, void *data)
891 struct net_device *netdev = data;
892 struct igbvf_adapter *adapter = netdev_priv(netdev);
893 struct e1000_hw *hw = &adapter->hw;
894 struct igbvf_ring *tx_ring = adapter->tx_ring;
896 if (tx_ring->set_itr) {
897 writel(tx_ring->itr_val,
898 adapter->hw.hw_addr + tx_ring->itr_register);
899 adapter->tx_ring->set_itr = 0;
902 adapter->total_tx_bytes = 0;
903 adapter->total_tx_packets = 0;
905 /* auto mask will automatically re-enable the interrupt when we write
908 if (!igbvf_clean_tx_irq(tx_ring))
909 /* Ring was not completely cleaned, so fire another interrupt */
910 ew32(EICS, tx_ring->eims_value);
912 ew32(EIMS, tx_ring->eims_value);
917 static irqreturn_t igbvf_intr_msix_rx(int irq, void *data)
919 struct net_device *netdev = data;
920 struct igbvf_adapter *adapter = netdev_priv(netdev);
922 adapter->int_counter0++;
924 /* Write the ITR value calculated at the end of the
925 * previous interrupt.
927 if (adapter->rx_ring->set_itr) {
928 writel(adapter->rx_ring->itr_val,
929 adapter->hw.hw_addr + adapter->rx_ring->itr_register);
930 adapter->rx_ring->set_itr = 0;
933 if (napi_schedule_prep(&adapter->rx_ring->napi)) {
934 adapter->total_rx_bytes = 0;
935 adapter->total_rx_packets = 0;
936 __napi_schedule(&adapter->rx_ring->napi);
942 #define IGBVF_NO_QUEUE -1
944 static void igbvf_assign_vector(struct igbvf_adapter *adapter, int rx_queue,
945 int tx_queue, int msix_vector)
947 struct e1000_hw *hw = &adapter->hw;
950 /* 82576 uses a table-based method for assigning vectors.
951 * Each queue has a single entry in the table to which we write
952 * a vector number along with a "valid" bit. Sadly, the layout
953 * of the table is somewhat counterintuitive.
955 if (rx_queue > IGBVF_NO_QUEUE) {
956 index = (rx_queue >> 1);
957 ivar = array_er32(IVAR0, index);
958 if (rx_queue & 0x1) {
959 /* vector goes into third byte of register */
960 ivar = ivar & 0xFF00FFFF;
961 ivar |= (msix_vector | E1000_IVAR_VALID) << 16;
963 /* vector goes into low byte of register */
964 ivar = ivar & 0xFFFFFF00;
965 ivar |= msix_vector | E1000_IVAR_VALID;
967 adapter->rx_ring[rx_queue].eims_value = BIT(msix_vector);
968 array_ew32(IVAR0, index, ivar);
970 if (tx_queue > IGBVF_NO_QUEUE) {
971 index = (tx_queue >> 1);
972 ivar = array_er32(IVAR0, index);
973 if (tx_queue & 0x1) {
974 /* vector goes into high byte of register */
975 ivar = ivar & 0x00FFFFFF;
976 ivar |= (msix_vector | E1000_IVAR_VALID) << 24;
978 /* vector goes into second byte of register */
979 ivar = ivar & 0xFFFF00FF;
980 ivar |= (msix_vector | E1000_IVAR_VALID) << 8;
982 adapter->tx_ring[tx_queue].eims_value = BIT(msix_vector);
983 array_ew32(IVAR0, index, ivar);
988 * igbvf_configure_msix - Configure MSI-X hardware
989 * @adapter: board private structure
991 * igbvf_configure_msix sets up the hardware to properly
992 * generate MSI-X interrupts.
994 static void igbvf_configure_msix(struct igbvf_adapter *adapter)
997 struct e1000_hw *hw = &adapter->hw;
998 struct igbvf_ring *tx_ring = adapter->tx_ring;
999 struct igbvf_ring *rx_ring = adapter->rx_ring;
1002 adapter->eims_enable_mask = 0;
1004 igbvf_assign_vector(adapter, IGBVF_NO_QUEUE, 0, vector++);
1005 adapter->eims_enable_mask |= tx_ring->eims_value;
1006 writel(tx_ring->itr_val, hw->hw_addr + tx_ring->itr_register);
1007 igbvf_assign_vector(adapter, 0, IGBVF_NO_QUEUE, vector++);
1008 adapter->eims_enable_mask |= rx_ring->eims_value;
1009 writel(rx_ring->itr_val, hw->hw_addr + rx_ring->itr_register);
1011 /* set vector for other causes, i.e. link changes */
1013 tmp = (vector++ | E1000_IVAR_VALID);
1015 ew32(IVAR_MISC, tmp);
1017 adapter->eims_enable_mask = GENMASK(vector - 1, 0);
1018 adapter->eims_other = BIT(vector - 1);
1022 static void igbvf_reset_interrupt_capability(struct igbvf_adapter *adapter)
1024 if (adapter->msix_entries) {
1025 pci_disable_msix(adapter->pdev);
1026 kfree(adapter->msix_entries);
1027 adapter->msix_entries = NULL;
1032 * igbvf_set_interrupt_capability - set MSI or MSI-X if supported
1033 * @adapter: board private structure
1035 * Attempt to configure interrupts using the best available
1036 * capabilities of the hardware and kernel.
1038 static void igbvf_set_interrupt_capability(struct igbvf_adapter *adapter)
1043 /* we allocate 3 vectors, 1 for Tx, 1 for Rx, one for PF messages */
1044 adapter->msix_entries = kcalloc(3, sizeof(struct msix_entry),
1046 if (adapter->msix_entries) {
1047 for (i = 0; i < 3; i++)
1048 adapter->msix_entries[i].entry = i;
1050 err = pci_enable_msix_range(adapter->pdev,
1051 adapter->msix_entries, 3, 3);
1056 dev_err(&adapter->pdev->dev,
1057 "Failed to initialize MSI-X interrupts.\n");
1058 igbvf_reset_interrupt_capability(adapter);
1063 * igbvf_request_msix - Initialize MSI-X interrupts
1064 * @adapter: board private structure
1066 * igbvf_request_msix allocates MSI-X vectors and requests interrupts from the
1069 static int igbvf_request_msix(struct igbvf_adapter *adapter)
1071 struct net_device *netdev = adapter->netdev;
1072 int err = 0, vector = 0;
1074 if (strlen(netdev->name) < (IFNAMSIZ - 5)) {
1075 sprintf(adapter->tx_ring->name, "%s-tx-0", netdev->name);
1076 sprintf(adapter->rx_ring->name, "%s-rx-0", netdev->name);
1078 memcpy(adapter->tx_ring->name, netdev->name, IFNAMSIZ);
1079 memcpy(adapter->rx_ring->name, netdev->name, IFNAMSIZ);
1082 err = request_irq(adapter->msix_entries[vector].vector,
1083 igbvf_intr_msix_tx, 0, adapter->tx_ring->name,
1088 adapter->tx_ring->itr_register = E1000_EITR(vector);
1089 adapter->tx_ring->itr_val = adapter->current_itr;
1092 err = request_irq(adapter->msix_entries[vector].vector,
1093 igbvf_intr_msix_rx, 0, adapter->rx_ring->name,
1098 adapter->rx_ring->itr_register = E1000_EITR(vector);
1099 adapter->rx_ring->itr_val = adapter->current_itr;
1102 err = request_irq(adapter->msix_entries[vector].vector,
1103 igbvf_msix_other, 0, netdev->name, netdev);
1107 igbvf_configure_msix(adapter);
1110 free_irq(adapter->msix_entries[--vector].vector, netdev);
1112 free_irq(adapter->msix_entries[--vector].vector, netdev);
1118 * igbvf_alloc_queues - Allocate memory for all rings
1119 * @adapter: board private structure to initialize
1121 static int igbvf_alloc_queues(struct igbvf_adapter *adapter)
1123 struct net_device *netdev = adapter->netdev;
1125 adapter->tx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL);
1126 if (!adapter->tx_ring)
1129 adapter->rx_ring = kzalloc(sizeof(struct igbvf_ring), GFP_KERNEL);
1130 if (!adapter->rx_ring) {
1131 kfree(adapter->tx_ring);
1135 netif_napi_add(netdev, &adapter->rx_ring->napi, igbvf_poll, 64);
1141 * igbvf_request_irq - initialize interrupts
1142 * @adapter: board private structure
1144 * Attempts to configure interrupts using the best available
1145 * capabilities of the hardware and kernel.
1147 static int igbvf_request_irq(struct igbvf_adapter *adapter)
1151 /* igbvf supports msi-x only */
1152 if (adapter->msix_entries)
1153 err = igbvf_request_msix(adapter);
1158 dev_err(&adapter->pdev->dev,
1159 "Unable to allocate interrupt, Error: %d\n", err);
1164 static void igbvf_free_irq(struct igbvf_adapter *adapter)
1166 struct net_device *netdev = adapter->netdev;
1169 if (adapter->msix_entries) {
1170 for (vector = 0; vector < 3; vector++)
1171 free_irq(adapter->msix_entries[vector].vector, netdev);
1176 * igbvf_irq_disable - Mask off interrupt generation on the NIC
1177 * @adapter: board private structure
1179 static void igbvf_irq_disable(struct igbvf_adapter *adapter)
1181 struct e1000_hw *hw = &adapter->hw;
1185 if (adapter->msix_entries)
1190 * igbvf_irq_enable - Enable default interrupt generation settings
1191 * @adapter: board private structure
1193 static void igbvf_irq_enable(struct igbvf_adapter *adapter)
1195 struct e1000_hw *hw = &adapter->hw;
1197 ew32(EIAC, adapter->eims_enable_mask);
1198 ew32(EIAM, adapter->eims_enable_mask);
1199 ew32(EIMS, adapter->eims_enable_mask);
1203 * igbvf_poll - NAPI Rx polling callback
1204 * @napi: struct associated with this polling callback
1205 * @budget: amount of packets driver is allowed to process this poll
1207 static int igbvf_poll(struct napi_struct *napi, int budget)
1209 struct igbvf_ring *rx_ring = container_of(napi, struct igbvf_ring, napi);
1210 struct igbvf_adapter *adapter = rx_ring->adapter;
1211 struct e1000_hw *hw = &adapter->hw;
1214 igbvf_clean_rx_irq(adapter, &work_done, budget);
1216 /* If not enough Rx work done, exit the polling mode */
1217 if (work_done < budget) {
1218 napi_complete_done(napi, work_done);
1220 if (adapter->requested_itr & 3)
1221 igbvf_set_itr(adapter);
1223 if (!test_bit(__IGBVF_DOWN, &adapter->state))
1224 ew32(EIMS, adapter->rx_ring->eims_value);
1231 * igbvf_set_rlpml - set receive large packet maximum length
1232 * @adapter: board private structure
1234 * Configure the maximum size of packets that will be received
1236 static void igbvf_set_rlpml(struct igbvf_adapter *adapter)
1239 struct e1000_hw *hw = &adapter->hw;
1241 max_frame_size = adapter->max_frame_size + VLAN_TAG_SIZE;
1243 spin_lock_bh(&hw->mbx_lock);
1245 e1000_rlpml_set_vf(hw, max_frame_size);
1247 spin_unlock_bh(&hw->mbx_lock);
1250 static int igbvf_vlan_rx_add_vid(struct net_device *netdev,
1251 __be16 proto, u16 vid)
1253 struct igbvf_adapter *adapter = netdev_priv(netdev);
1254 struct e1000_hw *hw = &adapter->hw;
1256 spin_lock_bh(&hw->mbx_lock);
1258 if (hw->mac.ops.set_vfta(hw, vid, true)) {
1259 dev_err(&adapter->pdev->dev, "Failed to add vlan id %d\n", vid);
1260 spin_unlock_bh(&hw->mbx_lock);
1264 spin_unlock_bh(&hw->mbx_lock);
1266 set_bit(vid, adapter->active_vlans);
1270 static int igbvf_vlan_rx_kill_vid(struct net_device *netdev,
1271 __be16 proto, u16 vid)
1273 struct igbvf_adapter *adapter = netdev_priv(netdev);
1274 struct e1000_hw *hw = &adapter->hw;
1276 spin_lock_bh(&hw->mbx_lock);
1278 if (hw->mac.ops.set_vfta(hw, vid, false)) {
1279 dev_err(&adapter->pdev->dev,
1280 "Failed to remove vlan id %d\n", vid);
1281 spin_unlock_bh(&hw->mbx_lock);
1285 spin_unlock_bh(&hw->mbx_lock);
1287 clear_bit(vid, adapter->active_vlans);
1291 static void igbvf_restore_vlan(struct igbvf_adapter *adapter)
1295 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
1296 igbvf_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
1300 * igbvf_configure_tx - Configure Transmit Unit after Reset
1301 * @adapter: board private structure
1303 * Configure the Tx unit of the MAC after a reset.
1305 static void igbvf_configure_tx(struct igbvf_adapter *adapter)
1307 struct e1000_hw *hw = &adapter->hw;
1308 struct igbvf_ring *tx_ring = adapter->tx_ring;
1310 u32 txdctl, dca_txctrl;
1312 /* disable transmits */
1313 txdctl = er32(TXDCTL(0));
1314 ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1318 /* Setup the HW Tx Head and Tail descriptor pointers */
1319 ew32(TDLEN(0), tx_ring->count * sizeof(union e1000_adv_tx_desc));
1320 tdba = tx_ring->dma;
1321 ew32(TDBAL(0), (tdba & DMA_BIT_MASK(32)));
1322 ew32(TDBAH(0), (tdba >> 32));
1325 tx_ring->head = E1000_TDH(0);
1326 tx_ring->tail = E1000_TDT(0);
1328 /* Turn off Relaxed Ordering on head write-backs. The writebacks
1329 * MUST be delivered in order or it will completely screw up
1332 dca_txctrl = er32(DCA_TXCTRL(0));
1333 dca_txctrl &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN;
1334 ew32(DCA_TXCTRL(0), dca_txctrl);
1336 /* enable transmits */
1337 txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
1338 ew32(TXDCTL(0), txdctl);
1340 /* Setup Transmit Descriptor Settings for eop descriptor */
1341 adapter->txd_cmd = E1000_ADVTXD_DCMD_EOP | E1000_ADVTXD_DCMD_IFCS;
1343 /* enable Report Status bit */
1344 adapter->txd_cmd |= E1000_ADVTXD_DCMD_RS;
1348 * igbvf_setup_srrctl - configure the receive control registers
1349 * @adapter: Board private structure
1351 static void igbvf_setup_srrctl(struct igbvf_adapter *adapter)
1353 struct e1000_hw *hw = &adapter->hw;
1356 srrctl &= ~(E1000_SRRCTL_DESCTYPE_MASK |
1357 E1000_SRRCTL_BSIZEHDR_MASK |
1358 E1000_SRRCTL_BSIZEPKT_MASK);
1360 /* Enable queue drop to avoid head of line blocking */
1361 srrctl |= E1000_SRRCTL_DROP_EN;
1363 /* Setup buffer sizes */
1364 srrctl |= ALIGN(adapter->rx_buffer_len, 1024) >>
1365 E1000_SRRCTL_BSIZEPKT_SHIFT;
1367 if (adapter->rx_buffer_len < 2048) {
1368 adapter->rx_ps_hdr_size = 0;
1369 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
1371 adapter->rx_ps_hdr_size = 128;
1372 srrctl |= adapter->rx_ps_hdr_size <<
1373 E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
1374 srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
1377 ew32(SRRCTL(0), srrctl);
1381 * igbvf_configure_rx - Configure Receive Unit after Reset
1382 * @adapter: board private structure
1384 * Configure the Rx unit of the MAC after a reset.
1386 static void igbvf_configure_rx(struct igbvf_adapter *adapter)
1388 struct e1000_hw *hw = &adapter->hw;
1389 struct igbvf_ring *rx_ring = adapter->rx_ring;
1393 /* disable receives */
1394 rxdctl = er32(RXDCTL(0));
1395 ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1399 /* Setup the HW Rx Head and Tail Descriptor Pointers and
1400 * the Base and Length of the Rx Descriptor Ring
1402 rdba = rx_ring->dma;
1403 ew32(RDBAL(0), (rdba & DMA_BIT_MASK(32)));
1404 ew32(RDBAH(0), (rdba >> 32));
1405 ew32(RDLEN(0), rx_ring->count * sizeof(union e1000_adv_rx_desc));
1406 rx_ring->head = E1000_RDH(0);
1407 rx_ring->tail = E1000_RDT(0);
1411 rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
1412 rxdctl &= 0xFFF00000;
1413 rxdctl |= IGBVF_RX_PTHRESH;
1414 rxdctl |= IGBVF_RX_HTHRESH << 8;
1415 rxdctl |= IGBVF_RX_WTHRESH << 16;
1417 igbvf_set_rlpml(adapter);
1419 /* enable receives */
1420 ew32(RXDCTL(0), rxdctl);
1424 * igbvf_set_multi - Multicast and Promiscuous mode set
1425 * @netdev: network interface device structure
1427 * The set_multi entry point is called whenever the multicast address
1428 * list or the network interface flags are updated. This routine is
1429 * responsible for configuring the hardware for proper multicast,
1430 * promiscuous mode, and all-multi behavior.
1432 static void igbvf_set_multi(struct net_device *netdev)
1434 struct igbvf_adapter *adapter = netdev_priv(netdev);
1435 struct e1000_hw *hw = &adapter->hw;
1436 struct netdev_hw_addr *ha;
1437 u8 *mta_list = NULL;
1440 if (!netdev_mc_empty(netdev)) {
1441 mta_list = kmalloc_array(netdev_mc_count(netdev), ETH_ALEN,
1447 /* prepare a packed array of only addresses. */
1449 netdev_for_each_mc_addr(ha, netdev)
1450 memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
1452 spin_lock_bh(&hw->mbx_lock);
1454 hw->mac.ops.update_mc_addr_list(hw, mta_list, i, 0, 0);
1456 spin_unlock_bh(&hw->mbx_lock);
1461 * igbvf_set_uni - Configure unicast MAC filters
1462 * @netdev: network interface device structure
1464 * This routine is responsible for configuring the hardware for proper
1467 static int igbvf_set_uni(struct net_device *netdev)
1469 struct igbvf_adapter *adapter = netdev_priv(netdev);
1470 struct e1000_hw *hw = &adapter->hw;
1472 if (netdev_uc_count(netdev) > IGBVF_MAX_MAC_FILTERS) {
1473 pr_err("Too many unicast filters - No Space\n");
1477 spin_lock_bh(&hw->mbx_lock);
1479 /* Clear all unicast MAC filters */
1480 hw->mac.ops.set_uc_addr(hw, E1000_VF_MAC_FILTER_CLR, NULL);
1482 spin_unlock_bh(&hw->mbx_lock);
1484 if (!netdev_uc_empty(netdev)) {
1485 struct netdev_hw_addr *ha;
1487 /* Add MAC filters one by one */
1488 netdev_for_each_uc_addr(ha, netdev) {
1489 spin_lock_bh(&hw->mbx_lock);
1491 hw->mac.ops.set_uc_addr(hw, E1000_VF_MAC_FILTER_ADD,
1494 spin_unlock_bh(&hw->mbx_lock);
1502 static void igbvf_set_rx_mode(struct net_device *netdev)
1504 igbvf_set_multi(netdev);
1505 igbvf_set_uni(netdev);
1509 * igbvf_configure - configure the hardware for Rx and Tx
1510 * @adapter: private board structure
1512 static void igbvf_configure(struct igbvf_adapter *adapter)
1514 igbvf_set_rx_mode(adapter->netdev);
1516 igbvf_restore_vlan(adapter);
1518 igbvf_configure_tx(adapter);
1519 igbvf_setup_srrctl(adapter);
1520 igbvf_configure_rx(adapter);
1521 igbvf_alloc_rx_buffers(adapter->rx_ring,
1522 igbvf_desc_unused(adapter->rx_ring));
1525 /* igbvf_reset - bring the hardware into a known good state
1526 * @adapter: private board structure
1528 * This function boots the hardware and enables some settings that
1529 * require a configuration cycle of the hardware - those cannot be
1530 * set/changed during runtime. After reset the device needs to be
1531 * properly configured for Rx, Tx etc.
1533 static void igbvf_reset(struct igbvf_adapter *adapter)
1535 struct e1000_mac_info *mac = &adapter->hw.mac;
1536 struct net_device *netdev = adapter->netdev;
1537 struct e1000_hw *hw = &adapter->hw;
1539 spin_lock_bh(&hw->mbx_lock);
1541 /* Allow time for pending master requests to run */
1542 if (mac->ops.reset_hw(hw))
1543 dev_err(&adapter->pdev->dev, "PF still resetting\n");
1545 mac->ops.init_hw(hw);
1547 spin_unlock_bh(&hw->mbx_lock);
1549 if (is_valid_ether_addr(adapter->hw.mac.addr)) {
1550 memcpy(netdev->dev_addr, adapter->hw.mac.addr,
1552 memcpy(netdev->perm_addr, adapter->hw.mac.addr,
1556 adapter->last_reset = jiffies;
1559 int igbvf_up(struct igbvf_adapter *adapter)
1561 struct e1000_hw *hw = &adapter->hw;
1563 /* hardware has been reset, we need to reload some things */
1564 igbvf_configure(adapter);
1566 clear_bit(__IGBVF_DOWN, &adapter->state);
1568 napi_enable(&adapter->rx_ring->napi);
1569 if (adapter->msix_entries)
1570 igbvf_configure_msix(adapter);
1572 /* Clear any pending interrupts. */
1574 igbvf_irq_enable(adapter);
1576 /* start the watchdog */
1577 hw->mac.get_link_status = 1;
1578 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1583 void igbvf_down(struct igbvf_adapter *adapter)
1585 struct net_device *netdev = adapter->netdev;
1586 struct e1000_hw *hw = &adapter->hw;
1589 /* signal that we're down so the interrupt handler does not
1590 * reschedule our watchdog timer
1592 set_bit(__IGBVF_DOWN, &adapter->state);
1594 /* disable receives in the hardware */
1595 rxdctl = er32(RXDCTL(0));
1596 ew32(RXDCTL(0), rxdctl & ~E1000_RXDCTL_QUEUE_ENABLE);
1598 netif_carrier_off(netdev);
1599 netif_stop_queue(netdev);
1601 /* disable transmits in the hardware */
1602 txdctl = er32(TXDCTL(0));
1603 ew32(TXDCTL(0), txdctl & ~E1000_TXDCTL_QUEUE_ENABLE);
1605 /* flush both disables and wait for them to finish */
1609 napi_disable(&adapter->rx_ring->napi);
1611 igbvf_irq_disable(adapter);
1613 del_timer_sync(&adapter->watchdog_timer);
1615 /* record the stats before reset*/
1616 igbvf_update_stats(adapter);
1618 adapter->link_speed = 0;
1619 adapter->link_duplex = 0;
1621 igbvf_reset(adapter);
1622 igbvf_clean_tx_ring(adapter->tx_ring);
1623 igbvf_clean_rx_ring(adapter->rx_ring);
1626 void igbvf_reinit_locked(struct igbvf_adapter *adapter)
1629 while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
1630 usleep_range(1000, 2000);
1631 igbvf_down(adapter);
1633 clear_bit(__IGBVF_RESETTING, &adapter->state);
1637 * igbvf_sw_init - Initialize general software structures (struct igbvf_adapter)
1638 * @adapter: board private structure to initialize
1640 * igbvf_sw_init initializes the Adapter private data structure.
1641 * Fields are initialized based on PCI device information and
1642 * OS network device settings (MTU size).
1644 static int igbvf_sw_init(struct igbvf_adapter *adapter)
1646 struct net_device *netdev = adapter->netdev;
1649 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN;
1650 adapter->rx_ps_hdr_size = 0;
1651 adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1652 adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
1654 adapter->tx_int_delay = 8;
1655 adapter->tx_abs_int_delay = 32;
1656 adapter->rx_int_delay = 0;
1657 adapter->rx_abs_int_delay = 8;
1658 adapter->requested_itr = 3;
1659 adapter->current_itr = IGBVF_START_ITR;
1661 /* Set various function pointers */
1662 adapter->ei->init_ops(&adapter->hw);
1664 rc = adapter->hw.mac.ops.init_params(&adapter->hw);
1668 rc = adapter->hw.mbx.ops.init_params(&adapter->hw);
1672 igbvf_set_interrupt_capability(adapter);
1674 if (igbvf_alloc_queues(adapter))
1677 spin_lock_init(&adapter->tx_queue_lock);
1679 /* Explicitly disable IRQ since the NIC can be in any state. */
1680 igbvf_irq_disable(adapter);
1682 spin_lock_init(&adapter->stats_lock);
1683 spin_lock_init(&adapter->hw.mbx_lock);
1685 set_bit(__IGBVF_DOWN, &adapter->state);
1689 static void igbvf_initialize_last_counter_stats(struct igbvf_adapter *adapter)
1691 struct e1000_hw *hw = &adapter->hw;
1693 adapter->stats.last_gprc = er32(VFGPRC);
1694 adapter->stats.last_gorc = er32(VFGORC);
1695 adapter->stats.last_gptc = er32(VFGPTC);
1696 adapter->stats.last_gotc = er32(VFGOTC);
1697 adapter->stats.last_mprc = er32(VFMPRC);
1698 adapter->stats.last_gotlbc = er32(VFGOTLBC);
1699 adapter->stats.last_gptlbc = er32(VFGPTLBC);
1700 adapter->stats.last_gorlbc = er32(VFGORLBC);
1701 adapter->stats.last_gprlbc = er32(VFGPRLBC);
1703 adapter->stats.base_gprc = er32(VFGPRC);
1704 adapter->stats.base_gorc = er32(VFGORC);
1705 adapter->stats.base_gptc = er32(VFGPTC);
1706 adapter->stats.base_gotc = er32(VFGOTC);
1707 adapter->stats.base_mprc = er32(VFMPRC);
1708 adapter->stats.base_gotlbc = er32(VFGOTLBC);
1709 adapter->stats.base_gptlbc = er32(VFGPTLBC);
1710 adapter->stats.base_gorlbc = er32(VFGORLBC);
1711 adapter->stats.base_gprlbc = er32(VFGPRLBC);
1715 * igbvf_open - Called when a network interface is made active
1716 * @netdev: network interface device structure
1718 * Returns 0 on success, negative value on failure
1720 * The open entry point is called when a network interface is made
1721 * active by the system (IFF_UP). At this point all resources needed
1722 * for transmit and receive operations are allocated, the interrupt
1723 * handler is registered with the OS, the watchdog timer is started,
1724 * and the stack is notified that the interface is ready.
1726 static int igbvf_open(struct net_device *netdev)
1728 struct igbvf_adapter *adapter = netdev_priv(netdev);
1729 struct e1000_hw *hw = &adapter->hw;
1732 /* disallow open during test */
1733 if (test_bit(__IGBVF_TESTING, &adapter->state))
1736 /* allocate transmit descriptors */
1737 err = igbvf_setup_tx_resources(adapter, adapter->tx_ring);
1741 /* allocate receive descriptors */
1742 err = igbvf_setup_rx_resources(adapter, adapter->rx_ring);
1746 /* before we allocate an interrupt, we must be ready to handle it.
1747 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1748 * as soon as we call pci_request_irq, so we have to setup our
1749 * clean_rx handler before we do so.
1751 igbvf_configure(adapter);
1753 err = igbvf_request_irq(adapter);
1757 /* From here on the code is the same as igbvf_up() */
1758 clear_bit(__IGBVF_DOWN, &adapter->state);
1760 napi_enable(&adapter->rx_ring->napi);
1762 /* clear any pending interrupts */
1765 igbvf_irq_enable(adapter);
1767 /* start the watchdog */
1768 hw->mac.get_link_status = 1;
1769 mod_timer(&adapter->watchdog_timer, jiffies + 1);
1774 igbvf_free_rx_resources(adapter->rx_ring);
1776 igbvf_free_tx_resources(adapter->tx_ring);
1778 igbvf_reset(adapter);
1784 * igbvf_close - Disables a network interface
1785 * @netdev: network interface device structure
1787 * Returns 0, this is not allowed to fail
1789 * The close entry point is called when an interface is de-activated
1790 * by the OS. The hardware is still under the drivers control, but
1791 * needs to be disabled. A global MAC reset is issued to stop the
1792 * hardware, and all transmit and receive resources are freed.
1794 static int igbvf_close(struct net_device *netdev)
1796 struct igbvf_adapter *adapter = netdev_priv(netdev);
1798 WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
1799 igbvf_down(adapter);
1801 igbvf_free_irq(adapter);
1803 igbvf_free_tx_resources(adapter->tx_ring);
1804 igbvf_free_rx_resources(adapter->rx_ring);
1810 * igbvf_set_mac - Change the Ethernet Address of the NIC
1811 * @netdev: network interface device structure
1812 * @p: pointer to an address structure
1814 * Returns 0 on success, negative on failure
1816 static int igbvf_set_mac(struct net_device *netdev, void *p)
1818 struct igbvf_adapter *adapter = netdev_priv(netdev);
1819 struct e1000_hw *hw = &adapter->hw;
1820 struct sockaddr *addr = p;
1822 if (!is_valid_ether_addr(addr->sa_data))
1823 return -EADDRNOTAVAIL;
1825 memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
1827 spin_lock_bh(&hw->mbx_lock);
1829 hw->mac.ops.rar_set(hw, hw->mac.addr, 0);
1831 spin_unlock_bh(&hw->mbx_lock);
1833 if (!ether_addr_equal(addr->sa_data, hw->mac.addr))
1834 return -EADDRNOTAVAIL;
1836 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
1841 #define UPDATE_VF_COUNTER(reg, name) \
1843 u32 current_counter = er32(reg); \
1844 if (current_counter < adapter->stats.last_##name) \
1845 adapter->stats.name += 0x100000000LL; \
1846 adapter->stats.last_##name = current_counter; \
1847 adapter->stats.name &= 0xFFFFFFFF00000000LL; \
1848 adapter->stats.name |= current_counter; \
1852 * igbvf_update_stats - Update the board statistics counters
1853 * @adapter: board private structure
1855 void igbvf_update_stats(struct igbvf_adapter *adapter)
1857 struct e1000_hw *hw = &adapter->hw;
1858 struct pci_dev *pdev = adapter->pdev;
1860 /* Prevent stats update while adapter is being reset, link is down
1861 * or if the pci connection is down.
1863 if (adapter->link_speed == 0)
1866 if (test_bit(__IGBVF_RESETTING, &adapter->state))
1869 if (pci_channel_offline(pdev))
1872 UPDATE_VF_COUNTER(VFGPRC, gprc);
1873 UPDATE_VF_COUNTER(VFGORC, gorc);
1874 UPDATE_VF_COUNTER(VFGPTC, gptc);
1875 UPDATE_VF_COUNTER(VFGOTC, gotc);
1876 UPDATE_VF_COUNTER(VFMPRC, mprc);
1877 UPDATE_VF_COUNTER(VFGOTLBC, gotlbc);
1878 UPDATE_VF_COUNTER(VFGPTLBC, gptlbc);
1879 UPDATE_VF_COUNTER(VFGORLBC, gorlbc);
1880 UPDATE_VF_COUNTER(VFGPRLBC, gprlbc);
1882 /* Fill out the OS statistics structure */
1883 adapter->netdev->stats.multicast = adapter->stats.mprc;
1886 static void igbvf_print_link_info(struct igbvf_adapter *adapter)
1888 dev_info(&adapter->pdev->dev, "Link is Up %d Mbps %s Duplex\n",
1889 adapter->link_speed,
1890 adapter->link_duplex == FULL_DUPLEX ? "Full" : "Half");
1893 static bool igbvf_has_link(struct igbvf_adapter *adapter)
1895 struct e1000_hw *hw = &adapter->hw;
1896 s32 ret_val = E1000_SUCCESS;
1899 /* If interface is down, stay link down */
1900 if (test_bit(__IGBVF_DOWN, &adapter->state))
1903 spin_lock_bh(&hw->mbx_lock);
1905 ret_val = hw->mac.ops.check_for_link(hw);
1907 spin_unlock_bh(&hw->mbx_lock);
1909 link_active = !hw->mac.get_link_status;
1911 /* if check for link returns error we will need to reset */
1912 if (ret_val && time_after(jiffies, adapter->last_reset + (10 * HZ)))
1913 schedule_work(&adapter->reset_task);
1919 * igbvf_watchdog - Timer Call-back
1920 * @data: pointer to adapter cast into an unsigned long
1922 static void igbvf_watchdog(unsigned long data)
1924 struct igbvf_adapter *adapter = (struct igbvf_adapter *)data;
1926 /* Do the rest outside of interrupt context */
1927 schedule_work(&adapter->watchdog_task);
1930 static void igbvf_watchdog_task(struct work_struct *work)
1932 struct igbvf_adapter *adapter = container_of(work,
1933 struct igbvf_adapter,
1935 struct net_device *netdev = adapter->netdev;
1936 struct e1000_mac_info *mac = &adapter->hw.mac;
1937 struct igbvf_ring *tx_ring = adapter->tx_ring;
1938 struct e1000_hw *hw = &adapter->hw;
1942 link = igbvf_has_link(adapter);
1945 if (!netif_carrier_ok(netdev)) {
1946 mac->ops.get_link_up_info(&adapter->hw,
1947 &adapter->link_speed,
1948 &adapter->link_duplex);
1949 igbvf_print_link_info(adapter);
1951 netif_carrier_on(netdev);
1952 netif_wake_queue(netdev);
1955 if (netif_carrier_ok(netdev)) {
1956 adapter->link_speed = 0;
1957 adapter->link_duplex = 0;
1958 dev_info(&adapter->pdev->dev, "Link is Down\n");
1959 netif_carrier_off(netdev);
1960 netif_stop_queue(netdev);
1964 if (netif_carrier_ok(netdev)) {
1965 igbvf_update_stats(adapter);
1967 tx_pending = (igbvf_desc_unused(tx_ring) + 1 <
1970 /* We've lost link, so the controller stops DMA,
1971 * but we've got queued Tx work that's never going
1972 * to get done, so reset controller to flush Tx.
1973 * (Do the reset outside of interrupt context).
1975 adapter->tx_timeout_count++;
1976 schedule_work(&adapter->reset_task);
1980 /* Cause software interrupt to ensure Rx ring is cleaned */
1981 ew32(EICS, adapter->rx_ring->eims_value);
1983 /* Reset the timer */
1984 if (!test_bit(__IGBVF_DOWN, &adapter->state))
1985 mod_timer(&adapter->watchdog_timer,
1986 round_jiffies(jiffies + (2 * HZ)));
1989 #define IGBVF_TX_FLAGS_CSUM 0x00000001
1990 #define IGBVF_TX_FLAGS_VLAN 0x00000002
1991 #define IGBVF_TX_FLAGS_TSO 0x00000004
1992 #define IGBVF_TX_FLAGS_IPV4 0x00000008
1993 #define IGBVF_TX_FLAGS_VLAN_MASK 0xffff0000
1994 #define IGBVF_TX_FLAGS_VLAN_SHIFT 16
1996 static void igbvf_tx_ctxtdesc(struct igbvf_ring *tx_ring, u32 vlan_macip_lens,
1997 u32 type_tucmd, u32 mss_l4len_idx)
1999 struct e1000_adv_tx_context_desc *context_desc;
2000 struct igbvf_buffer *buffer_info;
2001 u16 i = tx_ring->next_to_use;
2003 context_desc = IGBVF_TX_CTXTDESC_ADV(*tx_ring, i);
2004 buffer_info = &tx_ring->buffer_info[i];
2007 tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
2009 /* set bits to identify this as an advanced context descriptor */
2010 type_tucmd |= E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT;
2012 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
2013 context_desc->seqnum_seed = 0;
2014 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd);
2015 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
2017 buffer_info->time_stamp = jiffies;
2018 buffer_info->dma = 0;
2021 static int igbvf_tso(struct igbvf_ring *tx_ring,
2022 struct sk_buff *skb, u32 tx_flags, u8 *hdr_len)
2024 u32 vlan_macip_lens, type_tucmd, mss_l4len_idx;
2034 u32 paylen, l4_offset;
2037 if (skb->ip_summed != CHECKSUM_PARTIAL)
2040 if (!skb_is_gso(skb))
2043 err = skb_cow_head(skb, 0);
2047 ip.hdr = skb_network_header(skb);
2048 l4.hdr = skb_checksum_start(skb);
2050 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
2051 type_tucmd = E1000_ADVTXD_TUCMD_L4T_TCP;
2053 /* initialize outer IP header fields */
2054 if (ip.v4->version == 4) {
2055 unsigned char *csum_start = skb_checksum_start(skb);
2056 unsigned char *trans_start = ip.hdr + (ip.v4->ihl * 4);
2058 /* IP header will have to cancel out any data that
2059 * is not a part of the outer IP header
2061 ip.v4->check = csum_fold(csum_partial(trans_start,
2062 csum_start - trans_start,
2064 type_tucmd |= E1000_ADVTXD_TUCMD_IPV4;
2068 ip.v6->payload_len = 0;
2071 /* determine offset of inner transport header */
2072 l4_offset = l4.hdr - skb->data;
2074 /* compute length of segmentation header */
2075 *hdr_len = (l4.tcp->doff * 4) + l4_offset;
2077 /* remove payload length from inner checksum */
2078 paylen = skb->len - l4_offset;
2079 csum_replace_by_diff(&l4.tcp->check, htonl(paylen));
2082 mss_l4len_idx = (*hdr_len - l4_offset) << E1000_ADVTXD_L4LEN_SHIFT;
2083 mss_l4len_idx |= skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT;
2085 /* VLAN MACLEN IPLEN */
2086 vlan_macip_lens = l4.hdr - ip.hdr;
2087 vlan_macip_lens |= (ip.hdr - skb->data) << E1000_ADVTXD_MACLEN_SHIFT;
2088 vlan_macip_lens |= tx_flags & IGBVF_TX_FLAGS_VLAN_MASK;
2090 igbvf_tx_ctxtdesc(tx_ring, vlan_macip_lens, type_tucmd, mss_l4len_idx);
2095 static inline bool igbvf_ipv6_csum_is_sctp(struct sk_buff *skb)
2097 unsigned int offset = 0;
2099 ipv6_find_hdr(skb, &offset, IPPROTO_SCTP, NULL, NULL);
2101 return offset == skb_checksum_start_offset(skb);
2104 static bool igbvf_tx_csum(struct igbvf_ring *tx_ring, struct sk_buff *skb,
2105 u32 tx_flags, __be16 protocol)
2107 u32 vlan_macip_lens = 0;
2110 if (skb->ip_summed != CHECKSUM_PARTIAL) {
2112 if (!(tx_flags & IGBVF_TX_FLAGS_VLAN))
2117 switch (skb->csum_offset) {
2118 case offsetof(struct tcphdr, check):
2119 type_tucmd = E1000_ADVTXD_TUCMD_L4T_TCP;
2121 case offsetof(struct udphdr, check):
2123 case offsetof(struct sctphdr, checksum):
2124 /* validate that this is actually an SCTP request */
2125 if (((protocol == htons(ETH_P_IP)) &&
2126 (ip_hdr(skb)->protocol == IPPROTO_SCTP)) ||
2127 ((protocol == htons(ETH_P_IPV6)) &&
2128 igbvf_ipv6_csum_is_sctp(skb))) {
2129 type_tucmd = E1000_ADVTXD_TUCMD_L4T_SCTP;
2133 skb_checksum_help(skb);
2137 vlan_macip_lens = skb_checksum_start_offset(skb) -
2138 skb_network_offset(skb);
2140 vlan_macip_lens |= skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT;
2141 vlan_macip_lens |= tx_flags & IGBVF_TX_FLAGS_VLAN_MASK;
2143 igbvf_tx_ctxtdesc(tx_ring, vlan_macip_lens, type_tucmd, 0);
2147 static int igbvf_maybe_stop_tx(struct net_device *netdev, int size)
2149 struct igbvf_adapter *adapter = netdev_priv(netdev);
2151 /* there is enough descriptors then we don't need to worry */
2152 if (igbvf_desc_unused(adapter->tx_ring) >= size)
2155 netif_stop_queue(netdev);
2157 /* Herbert's original patch had:
2158 * smp_mb__after_netif_stop_queue();
2159 * but since that doesn't exist yet, just open code it.
2163 /* We need to check again just in case room has been made available */
2164 if (igbvf_desc_unused(adapter->tx_ring) < size)
2167 netif_wake_queue(netdev);
2169 ++adapter->restart_queue;
2173 #define IGBVF_MAX_TXD_PWR 16
2174 #define IGBVF_MAX_DATA_PER_TXD (1u << IGBVF_MAX_TXD_PWR)
2176 static inline int igbvf_tx_map_adv(struct igbvf_adapter *adapter,
2177 struct igbvf_ring *tx_ring,
2178 struct sk_buff *skb)
2180 struct igbvf_buffer *buffer_info;
2181 struct pci_dev *pdev = adapter->pdev;
2182 unsigned int len = skb_headlen(skb);
2183 unsigned int count = 0, i;
2186 i = tx_ring->next_to_use;
2188 buffer_info = &tx_ring->buffer_info[i];
2189 BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD);
2190 buffer_info->length = len;
2191 /* set time_stamp *before* dma to help avoid a possible race */
2192 buffer_info->time_stamp = jiffies;
2193 buffer_info->mapped_as_page = false;
2194 buffer_info->dma = dma_map_single(&pdev->dev, skb->data, len,
2196 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2199 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) {
2200 const struct skb_frag_struct *frag;
2204 if (i == tx_ring->count)
2207 frag = &skb_shinfo(skb)->frags[f];
2208 len = skb_frag_size(frag);
2210 buffer_info = &tx_ring->buffer_info[i];
2211 BUG_ON(len >= IGBVF_MAX_DATA_PER_TXD);
2212 buffer_info->length = len;
2213 buffer_info->time_stamp = jiffies;
2214 buffer_info->mapped_as_page = true;
2215 buffer_info->dma = skb_frag_dma_map(&pdev->dev, frag, 0, len,
2217 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2221 tx_ring->buffer_info[i].skb = skb;
2226 dev_err(&pdev->dev, "TX DMA map failed\n");
2228 /* clear timestamp and dma mappings for failed buffer_info mapping */
2229 buffer_info->dma = 0;
2230 buffer_info->time_stamp = 0;
2231 buffer_info->length = 0;
2232 buffer_info->mapped_as_page = false;
2236 /* clear timestamp and dma mappings for remaining portion of packet */
2239 i += tx_ring->count;
2241 buffer_info = &tx_ring->buffer_info[i];
2242 igbvf_put_txbuf(adapter, buffer_info);
2248 static inline void igbvf_tx_queue_adv(struct igbvf_adapter *adapter,
2249 struct igbvf_ring *tx_ring,
2250 int tx_flags, int count,
2251 unsigned int first, u32 paylen,
2254 union e1000_adv_tx_desc *tx_desc = NULL;
2255 struct igbvf_buffer *buffer_info;
2256 u32 olinfo_status = 0, cmd_type_len;
2259 cmd_type_len = (E1000_ADVTXD_DTYP_DATA | E1000_ADVTXD_DCMD_IFCS |
2260 E1000_ADVTXD_DCMD_DEXT);
2262 if (tx_flags & IGBVF_TX_FLAGS_VLAN)
2263 cmd_type_len |= E1000_ADVTXD_DCMD_VLE;
2265 if (tx_flags & IGBVF_TX_FLAGS_TSO) {
2266 cmd_type_len |= E1000_ADVTXD_DCMD_TSE;
2268 /* insert tcp checksum */
2269 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2271 /* insert ip checksum */
2272 if (tx_flags & IGBVF_TX_FLAGS_IPV4)
2273 olinfo_status |= E1000_TXD_POPTS_IXSM << 8;
2275 } else if (tx_flags & IGBVF_TX_FLAGS_CSUM) {
2276 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2279 olinfo_status |= ((paylen - hdr_len) << E1000_ADVTXD_PAYLEN_SHIFT);
2281 i = tx_ring->next_to_use;
2283 buffer_info = &tx_ring->buffer_info[i];
2284 tx_desc = IGBVF_TX_DESC_ADV(*tx_ring, i);
2285 tx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
2286 tx_desc->read.cmd_type_len =
2287 cpu_to_le32(cmd_type_len | buffer_info->length);
2288 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
2290 if (i == tx_ring->count)
2294 tx_desc->read.cmd_type_len |= cpu_to_le32(adapter->txd_cmd);
2295 /* Force memory writes to complete before letting h/w
2296 * know there are new descriptors to fetch. (Only
2297 * applicable for weak-ordered memory model archs,
2302 tx_ring->buffer_info[first].next_to_watch = tx_desc;
2303 tx_ring->next_to_use = i;
2304 writel(i, adapter->hw.hw_addr + tx_ring->tail);
2305 /* we need this if more than one processor can write to our tail
2306 * at a time, it synchronizes IO on IA64/Altix systems
2311 static netdev_tx_t igbvf_xmit_frame_ring_adv(struct sk_buff *skb,
2312 struct net_device *netdev,
2313 struct igbvf_ring *tx_ring)
2315 struct igbvf_adapter *adapter = netdev_priv(netdev);
2316 unsigned int first, tx_flags = 0;
2320 __be16 protocol = vlan_get_protocol(skb);
2322 if (test_bit(__IGBVF_DOWN, &adapter->state)) {
2323 dev_kfree_skb_any(skb);
2324 return NETDEV_TX_OK;
2327 if (skb->len <= 0) {
2328 dev_kfree_skb_any(skb);
2329 return NETDEV_TX_OK;
2332 /* need: count + 4 desc gap to keep tail from touching
2333 * + 2 desc gap to keep tail from touching head,
2334 * + 1 desc for skb->data,
2335 * + 1 desc for context descriptor,
2336 * head, otherwise try next time
2338 if (igbvf_maybe_stop_tx(netdev, skb_shinfo(skb)->nr_frags + 4)) {
2339 /* this is a hard error */
2340 return NETDEV_TX_BUSY;
2343 if (skb_vlan_tag_present(skb)) {
2344 tx_flags |= IGBVF_TX_FLAGS_VLAN;
2345 tx_flags |= (skb_vlan_tag_get(skb) <<
2346 IGBVF_TX_FLAGS_VLAN_SHIFT);
2349 if (protocol == htons(ETH_P_IP))
2350 tx_flags |= IGBVF_TX_FLAGS_IPV4;
2352 first = tx_ring->next_to_use;
2354 tso = igbvf_tso(tx_ring, skb, tx_flags, &hdr_len);
2355 if (unlikely(tso < 0)) {
2356 dev_kfree_skb_any(skb);
2357 return NETDEV_TX_OK;
2361 tx_flags |= IGBVF_TX_FLAGS_TSO;
2362 else if (igbvf_tx_csum(tx_ring, skb, tx_flags, protocol) &&
2363 (skb->ip_summed == CHECKSUM_PARTIAL))
2364 tx_flags |= IGBVF_TX_FLAGS_CSUM;
2366 /* count reflects descriptors mapped, if 0 then mapping error
2367 * has occurred and we need to rewind the descriptor queue
2369 count = igbvf_tx_map_adv(adapter, tx_ring, skb);
2372 igbvf_tx_queue_adv(adapter, tx_ring, tx_flags, count,
2373 first, skb->len, hdr_len);
2374 /* Make sure there is space in the ring for the next send. */
2375 igbvf_maybe_stop_tx(netdev, MAX_SKB_FRAGS + 4);
2377 dev_kfree_skb_any(skb);
2378 tx_ring->buffer_info[first].time_stamp = 0;
2379 tx_ring->next_to_use = first;
2382 return NETDEV_TX_OK;
2385 static netdev_tx_t igbvf_xmit_frame(struct sk_buff *skb,
2386 struct net_device *netdev)
2388 struct igbvf_adapter *adapter = netdev_priv(netdev);
2389 struct igbvf_ring *tx_ring;
2391 if (test_bit(__IGBVF_DOWN, &adapter->state)) {
2392 dev_kfree_skb_any(skb);
2393 return NETDEV_TX_OK;
2396 tx_ring = &adapter->tx_ring[0];
2398 return igbvf_xmit_frame_ring_adv(skb, netdev, tx_ring);
2402 * igbvf_tx_timeout - Respond to a Tx Hang
2403 * @netdev: network interface device structure
2405 static void igbvf_tx_timeout(struct net_device *netdev)
2407 struct igbvf_adapter *adapter = netdev_priv(netdev);
2409 /* Do the reset outside of interrupt context */
2410 adapter->tx_timeout_count++;
2411 schedule_work(&adapter->reset_task);
2414 static void igbvf_reset_task(struct work_struct *work)
2416 struct igbvf_adapter *adapter;
2418 adapter = container_of(work, struct igbvf_adapter, reset_task);
2420 igbvf_reinit_locked(adapter);
2424 * igbvf_change_mtu - Change the Maximum Transfer Unit
2425 * @netdev: network interface device structure
2426 * @new_mtu: new value for maximum frame size
2428 * Returns 0 on success, negative on failure
2430 static int igbvf_change_mtu(struct net_device *netdev, int new_mtu)
2432 struct igbvf_adapter *adapter = netdev_priv(netdev);
2433 int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
2435 while (test_and_set_bit(__IGBVF_RESETTING, &adapter->state))
2436 usleep_range(1000, 2000);
2437 /* igbvf_down has a dependency on max_frame_size */
2438 adapter->max_frame_size = max_frame;
2439 if (netif_running(netdev))
2440 igbvf_down(adapter);
2442 /* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
2443 * means we reserve 2 more, this pushes us to allocate from the next
2445 * i.e. RXBUFFER_2048 --> size-4096 slab
2446 * However with the new *_jumbo_rx* routines, jumbo receives will use
2450 if (max_frame <= 1024)
2451 adapter->rx_buffer_len = 1024;
2452 else if (max_frame <= 2048)
2453 adapter->rx_buffer_len = 2048;
2455 #if (PAGE_SIZE / 2) > 16384
2456 adapter->rx_buffer_len = 16384;
2458 adapter->rx_buffer_len = PAGE_SIZE / 2;
2461 /* adjust allocation if LPE protects us, and we aren't using SBP */
2462 if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
2463 (max_frame == ETH_FRAME_LEN + VLAN_HLEN + ETH_FCS_LEN))
2464 adapter->rx_buffer_len = ETH_FRAME_LEN + VLAN_HLEN +
2467 dev_info(&adapter->pdev->dev, "changing MTU from %d to %d\n",
2468 netdev->mtu, new_mtu);
2469 netdev->mtu = new_mtu;
2471 if (netif_running(netdev))
2474 igbvf_reset(adapter);
2476 clear_bit(__IGBVF_RESETTING, &adapter->state);
2481 static int igbvf_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
2489 static int igbvf_suspend(struct pci_dev *pdev, pm_message_t state)
2491 struct net_device *netdev = pci_get_drvdata(pdev);
2492 struct igbvf_adapter *adapter = netdev_priv(netdev);
2497 netif_device_detach(netdev);
2499 if (netif_running(netdev)) {
2500 WARN_ON(test_bit(__IGBVF_RESETTING, &adapter->state));
2501 igbvf_down(adapter);
2502 igbvf_free_irq(adapter);
2506 retval = pci_save_state(pdev);
2511 pci_disable_device(pdev);
2517 static int igbvf_resume(struct pci_dev *pdev)
2519 struct net_device *netdev = pci_get_drvdata(pdev);
2520 struct igbvf_adapter *adapter = netdev_priv(netdev);
2523 pci_restore_state(pdev);
2524 err = pci_enable_device_mem(pdev);
2526 dev_err(&pdev->dev, "Cannot enable PCI device from suspend\n");
2530 pci_set_master(pdev);
2532 if (netif_running(netdev)) {
2533 err = igbvf_request_irq(adapter);
2538 igbvf_reset(adapter);
2540 if (netif_running(netdev))
2543 netif_device_attach(netdev);
2549 static void igbvf_shutdown(struct pci_dev *pdev)
2551 igbvf_suspend(pdev, PMSG_SUSPEND);
2554 #ifdef CONFIG_NET_POLL_CONTROLLER
2555 /* Polling 'interrupt' - used by things like netconsole to send skbs
2556 * without having to re-enable interrupts. It's not called while
2557 * the interrupt routine is executing.
2559 static void igbvf_netpoll(struct net_device *netdev)
2561 struct igbvf_adapter *adapter = netdev_priv(netdev);
2563 disable_irq(adapter->pdev->irq);
2565 igbvf_clean_tx_irq(adapter->tx_ring);
2567 enable_irq(adapter->pdev->irq);
2572 * igbvf_io_error_detected - called when PCI error is detected
2573 * @pdev: Pointer to PCI device
2574 * @state: The current pci connection state
2576 * This function is called after a PCI bus error affecting
2577 * this device has been detected.
2579 static pci_ers_result_t igbvf_io_error_detected(struct pci_dev *pdev,
2580 pci_channel_state_t state)
2582 struct net_device *netdev = pci_get_drvdata(pdev);
2583 struct igbvf_adapter *adapter = netdev_priv(netdev);
2585 netif_device_detach(netdev);
2587 if (state == pci_channel_io_perm_failure)
2588 return PCI_ERS_RESULT_DISCONNECT;
2590 if (netif_running(netdev))
2591 igbvf_down(adapter);
2592 pci_disable_device(pdev);
2594 /* Request a slot slot reset. */
2595 return PCI_ERS_RESULT_NEED_RESET;
2599 * igbvf_io_slot_reset - called after the pci bus has been reset.
2600 * @pdev: Pointer to PCI device
2602 * Restart the card from scratch, as if from a cold-boot. Implementation
2603 * resembles the first-half of the igbvf_resume routine.
2605 static pci_ers_result_t igbvf_io_slot_reset(struct pci_dev *pdev)
2607 struct net_device *netdev = pci_get_drvdata(pdev);
2608 struct igbvf_adapter *adapter = netdev_priv(netdev);
2610 if (pci_enable_device_mem(pdev)) {
2612 "Cannot re-enable PCI device after reset.\n");
2613 return PCI_ERS_RESULT_DISCONNECT;
2615 pci_set_master(pdev);
2617 igbvf_reset(adapter);
2619 return PCI_ERS_RESULT_RECOVERED;
2623 * igbvf_io_resume - called when traffic can start flowing again.
2624 * @pdev: Pointer to PCI device
2626 * This callback is called when the error recovery driver tells us that
2627 * its OK to resume normal operation. Implementation resembles the
2628 * second-half of the igbvf_resume routine.
2630 static void igbvf_io_resume(struct pci_dev *pdev)
2632 struct net_device *netdev = pci_get_drvdata(pdev);
2633 struct igbvf_adapter *adapter = netdev_priv(netdev);
2635 if (netif_running(netdev)) {
2636 if (igbvf_up(adapter)) {
2638 "can't bring device back up after reset\n");
2643 netif_device_attach(netdev);
2646 static void igbvf_print_device_info(struct igbvf_adapter *adapter)
2648 struct e1000_hw *hw = &adapter->hw;
2649 struct net_device *netdev = adapter->netdev;
2650 struct pci_dev *pdev = adapter->pdev;
2652 if (hw->mac.type == e1000_vfadapt_i350)
2653 dev_info(&pdev->dev, "Intel(R) I350 Virtual Function\n");
2655 dev_info(&pdev->dev, "Intel(R) 82576 Virtual Function\n");
2656 dev_info(&pdev->dev, "Address: %pM\n", netdev->dev_addr);
2659 static int igbvf_set_features(struct net_device *netdev,
2660 netdev_features_t features)
2662 struct igbvf_adapter *adapter = netdev_priv(netdev);
2664 if (features & NETIF_F_RXCSUM)
2665 adapter->flags &= ~IGBVF_FLAG_RX_CSUM_DISABLED;
2667 adapter->flags |= IGBVF_FLAG_RX_CSUM_DISABLED;
2672 #define IGBVF_MAX_MAC_HDR_LEN 127
2673 #define IGBVF_MAX_NETWORK_HDR_LEN 511
2675 static netdev_features_t
2676 igbvf_features_check(struct sk_buff *skb, struct net_device *dev,
2677 netdev_features_t features)
2679 unsigned int network_hdr_len, mac_hdr_len;
2681 /* Make certain the headers can be described by a context descriptor */
2682 mac_hdr_len = skb_network_header(skb) - skb->data;
2683 if (unlikely(mac_hdr_len > IGBVF_MAX_MAC_HDR_LEN))
2684 return features & ~(NETIF_F_HW_CSUM |
2686 NETIF_F_HW_VLAN_CTAG_TX |
2690 network_hdr_len = skb_checksum_start(skb) - skb_network_header(skb);
2691 if (unlikely(network_hdr_len > IGBVF_MAX_NETWORK_HDR_LEN))
2692 return features & ~(NETIF_F_HW_CSUM |
2697 /* We can only support IPV4 TSO in tunnels if we can mangle the
2698 * inner IP ID field, so strip TSO if MANGLEID is not supported.
2700 if (skb->encapsulation && !(features & NETIF_F_TSO_MANGLEID))
2701 features &= ~NETIF_F_TSO;
2706 static const struct net_device_ops igbvf_netdev_ops = {
2707 .ndo_open = igbvf_open,
2708 .ndo_stop = igbvf_close,
2709 .ndo_start_xmit = igbvf_xmit_frame,
2710 .ndo_set_rx_mode = igbvf_set_rx_mode,
2711 .ndo_set_mac_address = igbvf_set_mac,
2712 .ndo_change_mtu = igbvf_change_mtu,
2713 .ndo_do_ioctl = igbvf_ioctl,
2714 .ndo_tx_timeout = igbvf_tx_timeout,
2715 .ndo_vlan_rx_add_vid = igbvf_vlan_rx_add_vid,
2716 .ndo_vlan_rx_kill_vid = igbvf_vlan_rx_kill_vid,
2717 #ifdef CONFIG_NET_POLL_CONTROLLER
2718 .ndo_poll_controller = igbvf_netpoll,
2720 .ndo_set_features = igbvf_set_features,
2721 .ndo_features_check = igbvf_features_check,
2725 * igbvf_probe - Device Initialization Routine
2726 * @pdev: PCI device information struct
2727 * @ent: entry in igbvf_pci_tbl
2729 * Returns 0 on success, negative on failure
2731 * igbvf_probe initializes an adapter identified by a pci_dev structure.
2732 * The OS initialization, configuring of the adapter private structure,
2733 * and a hardware reset occur.
2735 static int igbvf_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
2737 struct net_device *netdev;
2738 struct igbvf_adapter *adapter;
2739 struct e1000_hw *hw;
2740 const struct igbvf_info *ei = igbvf_info_tbl[ent->driver_data];
2742 static int cards_found;
2743 int err, pci_using_dac;
2745 err = pci_enable_device_mem(pdev);
2750 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
2754 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
2757 "No usable DMA configuration, aborting\n");
2762 err = pci_request_regions(pdev, igbvf_driver_name);
2766 pci_set_master(pdev);
2769 netdev = alloc_etherdev(sizeof(struct igbvf_adapter));
2771 goto err_alloc_etherdev;
2773 SET_NETDEV_DEV(netdev, &pdev->dev);
2775 pci_set_drvdata(pdev, netdev);
2776 adapter = netdev_priv(netdev);
2778 adapter->netdev = netdev;
2779 adapter->pdev = pdev;
2781 adapter->pba = ei->pba;
2782 adapter->flags = ei->flags;
2783 adapter->hw.back = adapter;
2784 adapter->hw.mac.type = ei->mac;
2785 adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
2787 /* PCI config space info */
2789 hw->vendor_id = pdev->vendor;
2790 hw->device_id = pdev->device;
2791 hw->subsystem_vendor_id = pdev->subsystem_vendor;
2792 hw->subsystem_device_id = pdev->subsystem_device;
2793 hw->revision_id = pdev->revision;
2796 adapter->hw.hw_addr = ioremap(pci_resource_start(pdev, 0),
2797 pci_resource_len(pdev, 0));
2799 if (!adapter->hw.hw_addr)
2802 if (ei->get_variants) {
2803 err = ei->get_variants(adapter);
2805 goto err_get_variants;
2808 /* setup adapter struct */
2809 err = igbvf_sw_init(adapter);
2813 /* construct the net_device struct */
2814 netdev->netdev_ops = &igbvf_netdev_ops;
2816 igbvf_set_ethtool_ops(netdev);
2817 netdev->watchdog_timeo = 5 * HZ;
2818 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
2820 adapter->bd_number = cards_found++;
2822 netdev->hw_features = NETIF_F_SG |
2829 #define IGBVF_GSO_PARTIAL_FEATURES (NETIF_F_GSO_GRE | \
2830 NETIF_F_GSO_GRE_CSUM | \
2831 NETIF_F_GSO_IPXIP4 | \
2832 NETIF_F_GSO_IPXIP6 | \
2833 NETIF_F_GSO_UDP_TUNNEL | \
2834 NETIF_F_GSO_UDP_TUNNEL_CSUM)
2836 netdev->gso_partial_features = IGBVF_GSO_PARTIAL_FEATURES;
2837 netdev->hw_features |= NETIF_F_GSO_PARTIAL |
2838 IGBVF_GSO_PARTIAL_FEATURES;
2840 netdev->features = netdev->hw_features;
2843 netdev->features |= NETIF_F_HIGHDMA;
2845 netdev->vlan_features |= netdev->features | NETIF_F_TSO_MANGLEID;
2846 netdev->mpls_features |= NETIF_F_HW_CSUM;
2847 netdev->hw_enc_features |= netdev->vlan_features;
2849 /* set this bit last since it cannot be part of vlan_features */
2850 netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER |
2851 NETIF_F_HW_VLAN_CTAG_RX |
2852 NETIF_F_HW_VLAN_CTAG_TX;
2854 /* MTU range: 68 - 9216 */
2855 netdev->min_mtu = ETH_MIN_MTU;
2856 netdev->max_mtu = MAX_STD_JUMBO_FRAME_SIZE;
2858 spin_lock_bh(&hw->mbx_lock);
2860 /*reset the controller to put the device in a known good state */
2861 err = hw->mac.ops.reset_hw(hw);
2863 dev_info(&pdev->dev,
2864 "PF still in reset state. Is the PF interface up?\n");
2866 err = hw->mac.ops.read_mac_addr(hw);
2868 dev_info(&pdev->dev, "Error reading MAC address.\n");
2869 else if (is_zero_ether_addr(adapter->hw.mac.addr))
2870 dev_info(&pdev->dev,
2871 "MAC address not assigned by administrator.\n");
2872 memcpy(netdev->dev_addr, adapter->hw.mac.addr,
2876 spin_unlock_bh(&hw->mbx_lock);
2878 if (!is_valid_ether_addr(netdev->dev_addr)) {
2879 dev_info(&pdev->dev, "Assigning random MAC address.\n");
2880 eth_hw_addr_random(netdev);
2881 memcpy(adapter->hw.mac.addr, netdev->dev_addr,
2885 setup_timer(&adapter->watchdog_timer, &igbvf_watchdog,
2886 (unsigned long)adapter);
2888 INIT_WORK(&adapter->reset_task, igbvf_reset_task);
2889 INIT_WORK(&adapter->watchdog_task, igbvf_watchdog_task);
2891 /* ring size defaults */
2892 adapter->rx_ring->count = 1024;
2893 adapter->tx_ring->count = 1024;
2895 /* reset the hardware with the new settings */
2896 igbvf_reset(adapter);
2898 /* set hardware-specific flags */
2899 if (adapter->hw.mac.type == e1000_vfadapt_i350)
2900 adapter->flags |= IGBVF_FLAG_RX_LB_VLAN_BSWAP;
2902 strcpy(netdev->name, "eth%d");
2903 err = register_netdev(netdev);
2907 /* tell the stack to leave us alone until igbvf_open() is called */
2908 netif_carrier_off(netdev);
2909 netif_stop_queue(netdev);
2911 igbvf_print_device_info(adapter);
2913 igbvf_initialize_last_counter_stats(adapter);
2918 netif_napi_del(&adapter->rx_ring->napi);
2919 kfree(adapter->tx_ring);
2920 kfree(adapter->rx_ring);
2922 igbvf_reset_interrupt_capability(adapter);
2924 iounmap(adapter->hw.hw_addr);
2926 free_netdev(netdev);
2928 pci_release_regions(pdev);
2931 pci_disable_device(pdev);
2936 * igbvf_remove - Device Removal Routine
2937 * @pdev: PCI device information struct
2939 * igbvf_remove is called by the PCI subsystem to alert the driver
2940 * that it should release a PCI device. The could be caused by a
2941 * Hot-Plug event, or because the driver is going to be removed from
2944 static void igbvf_remove(struct pci_dev *pdev)
2946 struct net_device *netdev = pci_get_drvdata(pdev);
2947 struct igbvf_adapter *adapter = netdev_priv(netdev);
2948 struct e1000_hw *hw = &adapter->hw;
2950 /* The watchdog timer may be rescheduled, so explicitly
2951 * disable it from being rescheduled.
2953 set_bit(__IGBVF_DOWN, &adapter->state);
2954 del_timer_sync(&adapter->watchdog_timer);
2956 cancel_work_sync(&adapter->reset_task);
2957 cancel_work_sync(&adapter->watchdog_task);
2959 unregister_netdev(netdev);
2961 igbvf_reset_interrupt_capability(adapter);
2963 /* it is important to delete the NAPI struct prior to freeing the
2964 * Rx ring so that you do not end up with null pointer refs
2966 netif_napi_del(&adapter->rx_ring->napi);
2967 kfree(adapter->tx_ring);
2968 kfree(adapter->rx_ring);
2970 iounmap(hw->hw_addr);
2971 if (hw->flash_address)
2972 iounmap(hw->flash_address);
2973 pci_release_regions(pdev);
2975 free_netdev(netdev);
2977 pci_disable_device(pdev);
2980 /* PCI Error Recovery (ERS) */
2981 static const struct pci_error_handlers igbvf_err_handler = {
2982 .error_detected = igbvf_io_error_detected,
2983 .slot_reset = igbvf_io_slot_reset,
2984 .resume = igbvf_io_resume,
2987 static const struct pci_device_id igbvf_pci_tbl[] = {
2988 { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_VF), board_vf },
2989 { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_VF), board_i350_vf },
2990 { } /* terminate list */
2992 MODULE_DEVICE_TABLE(pci, igbvf_pci_tbl);
2994 /* PCI Device API Driver */
2995 static struct pci_driver igbvf_driver = {
2996 .name = igbvf_driver_name,
2997 .id_table = igbvf_pci_tbl,
2998 .probe = igbvf_probe,
2999 .remove = igbvf_remove,
3001 /* Power Management Hooks */
3002 .suspend = igbvf_suspend,
3003 .resume = igbvf_resume,
3005 .shutdown = igbvf_shutdown,
3006 .err_handler = &igbvf_err_handler
3010 * igbvf_init_module - Driver Registration Routine
3012 * igbvf_init_module is the first routine called when the driver is
3013 * loaded. All it does is register with the PCI subsystem.
3015 static int __init igbvf_init_module(void)
3019 pr_info("%s - version %s\n", igbvf_driver_string, igbvf_driver_version);
3020 pr_info("%s\n", igbvf_copyright);
3022 ret = pci_register_driver(&igbvf_driver);
3026 module_init(igbvf_init_module);
3029 * igbvf_exit_module - Driver Exit Cleanup Routine
3031 * igbvf_exit_module is called just before the driver is removed
3034 static void __exit igbvf_exit_module(void)
3036 pci_unregister_driver(&igbvf_driver);
3038 module_exit(igbvf_exit_module);
3040 MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
3041 MODULE_DESCRIPTION("Intel(R) Gigabit Virtual Function Network Driver");
3042 MODULE_LICENSE("GPL");
3043 MODULE_VERSION(DRV_VERSION);