Merge branch 'for-next' of git://git.kernel.org/pub/scm/linux/kernel/git/jikos/trivial
[linux-2.6/btrfs-unstable.git] / drivers / net / e1000 / e1000_main.c
blobde69c54301c13fe78652c50e07ccfe9eefad441b
1 /*******************************************************************************
3 Intel PRO/1000 Linux driver
4 Copyright(c) 1999 - 2006 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
13 more details.
15 You should have received a copy of the GNU General Public License along with
16 this program; if not, write to the Free Software Foundation, Inc.,
17 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
19 The full GNU General Public License is included in this distribution in
20 the file called "COPYING".
22 Contact Information:
23 Linux NICS <linux.nics@intel.com>
24 e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
25 Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
27 *******************************************************************************/
29 #include "e1000.h"
30 #include <net/ip6_checksum.h>
31 #include <linux/io.h>
33 /* Intel Media SOC GbE MDIO physical base address */
34 static unsigned long ce4100_gbe_mdio_base_phy;
35 /* Intel Media SOC GbE MDIO virtual base address */
36 void __iomem *ce4100_gbe_mdio_base_virt;
38 char e1000_driver_name[] = "e1000";
39 static char e1000_driver_string[] = "Intel(R) PRO/1000 Network Driver";
40 #define DRV_VERSION "7.3.21-k8-NAPI"
41 const char e1000_driver_version[] = DRV_VERSION;
42 static const char e1000_copyright[] = "Copyright (c) 1999-2006 Intel Corporation.";
44 /* e1000_pci_tbl - PCI Device ID Table
46 * Last entry must be all 0s
48 * Macro expands to...
49 * {PCI_DEVICE(PCI_VENDOR_ID_INTEL, device_id)}
51 static DEFINE_PCI_DEVICE_TABLE(e1000_pci_tbl) = {
52 INTEL_E1000_ETHERNET_DEVICE(0x1000),
53 INTEL_E1000_ETHERNET_DEVICE(0x1001),
54 INTEL_E1000_ETHERNET_DEVICE(0x1004),
55 INTEL_E1000_ETHERNET_DEVICE(0x1008),
56 INTEL_E1000_ETHERNET_DEVICE(0x1009),
57 INTEL_E1000_ETHERNET_DEVICE(0x100C),
58 INTEL_E1000_ETHERNET_DEVICE(0x100D),
59 INTEL_E1000_ETHERNET_DEVICE(0x100E),
60 INTEL_E1000_ETHERNET_DEVICE(0x100F),
61 INTEL_E1000_ETHERNET_DEVICE(0x1010),
62 INTEL_E1000_ETHERNET_DEVICE(0x1011),
63 INTEL_E1000_ETHERNET_DEVICE(0x1012),
64 INTEL_E1000_ETHERNET_DEVICE(0x1013),
65 INTEL_E1000_ETHERNET_DEVICE(0x1014),
66 INTEL_E1000_ETHERNET_DEVICE(0x1015),
67 INTEL_E1000_ETHERNET_DEVICE(0x1016),
68 INTEL_E1000_ETHERNET_DEVICE(0x1017),
69 INTEL_E1000_ETHERNET_DEVICE(0x1018),
70 INTEL_E1000_ETHERNET_DEVICE(0x1019),
71 INTEL_E1000_ETHERNET_DEVICE(0x101A),
72 INTEL_E1000_ETHERNET_DEVICE(0x101D),
73 INTEL_E1000_ETHERNET_DEVICE(0x101E),
74 INTEL_E1000_ETHERNET_DEVICE(0x1026),
75 INTEL_E1000_ETHERNET_DEVICE(0x1027),
76 INTEL_E1000_ETHERNET_DEVICE(0x1028),
77 INTEL_E1000_ETHERNET_DEVICE(0x1075),
78 INTEL_E1000_ETHERNET_DEVICE(0x1076),
79 INTEL_E1000_ETHERNET_DEVICE(0x1077),
80 INTEL_E1000_ETHERNET_DEVICE(0x1078),
81 INTEL_E1000_ETHERNET_DEVICE(0x1079),
82 INTEL_E1000_ETHERNET_DEVICE(0x107A),
83 INTEL_E1000_ETHERNET_DEVICE(0x107B),
84 INTEL_E1000_ETHERNET_DEVICE(0x107C),
85 INTEL_E1000_ETHERNET_DEVICE(0x108A),
86 INTEL_E1000_ETHERNET_DEVICE(0x1099),
87 INTEL_E1000_ETHERNET_DEVICE(0x10B5),
88 INTEL_E1000_ETHERNET_DEVICE(0x2E6E),
89 /* required last entry */
90 {0,}
93 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
95 int e1000_up(struct e1000_adapter *adapter);
96 void e1000_down(struct e1000_adapter *adapter);
97 void e1000_reinit_locked(struct e1000_adapter *adapter);
98 void e1000_reset(struct e1000_adapter *adapter);
99 int e1000_set_spd_dplx(struct e1000_adapter *adapter, u16 spddplx);
100 int e1000_setup_all_tx_resources(struct e1000_adapter *adapter);
101 int e1000_setup_all_rx_resources(struct e1000_adapter *adapter);
102 void e1000_free_all_tx_resources(struct e1000_adapter *adapter);
103 void e1000_free_all_rx_resources(struct e1000_adapter *adapter);
104 static int e1000_setup_tx_resources(struct e1000_adapter *adapter,
105 struct e1000_tx_ring *txdr);
106 static int e1000_setup_rx_resources(struct e1000_adapter *adapter,
107 struct e1000_rx_ring *rxdr);
108 static void e1000_free_tx_resources(struct e1000_adapter *adapter,
109 struct e1000_tx_ring *tx_ring);
110 static void e1000_free_rx_resources(struct e1000_adapter *adapter,
111 struct e1000_rx_ring *rx_ring);
112 void e1000_update_stats(struct e1000_adapter *adapter);
114 static int e1000_init_module(void);
115 static void e1000_exit_module(void);
116 static int e1000_probe(struct pci_dev *pdev, const struct pci_device_id *ent);
117 static void __devexit e1000_remove(struct pci_dev *pdev);
118 static int e1000_alloc_queues(struct e1000_adapter *adapter);
119 static int e1000_sw_init(struct e1000_adapter *adapter);
120 static int e1000_open(struct net_device *netdev);
121 static int e1000_close(struct net_device *netdev);
122 static void e1000_configure_tx(struct e1000_adapter *adapter);
123 static void e1000_configure_rx(struct e1000_adapter *adapter);
124 static void e1000_setup_rctl(struct e1000_adapter *adapter);
125 static void e1000_clean_all_tx_rings(struct e1000_adapter *adapter);
126 static void e1000_clean_all_rx_rings(struct e1000_adapter *adapter);
127 static void e1000_clean_tx_ring(struct e1000_adapter *adapter,
128 struct e1000_tx_ring *tx_ring);
129 static void e1000_clean_rx_ring(struct e1000_adapter *adapter,
130 struct e1000_rx_ring *rx_ring);
131 static void e1000_set_rx_mode(struct net_device *netdev);
132 static void e1000_update_phy_info(unsigned long data);
133 static void e1000_update_phy_info_task(struct work_struct *work);
134 static void e1000_watchdog(unsigned long data);
135 static void e1000_82547_tx_fifo_stall(unsigned long data);
136 static void e1000_82547_tx_fifo_stall_task(struct work_struct *work);
137 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
138 struct net_device *netdev);
139 static struct net_device_stats * e1000_get_stats(struct net_device *netdev);
140 static int e1000_change_mtu(struct net_device *netdev, int new_mtu);
141 static int e1000_set_mac(struct net_device *netdev, void *p);
142 static irqreturn_t e1000_intr(int irq, void *data);
143 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter,
144 struct e1000_tx_ring *tx_ring);
145 static int e1000_clean(struct napi_struct *napi, int budget);
146 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
147 struct e1000_rx_ring *rx_ring,
148 int *work_done, int work_to_do);
149 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
150 struct e1000_rx_ring *rx_ring,
151 int *work_done, int work_to_do);
152 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
153 struct e1000_rx_ring *rx_ring,
154 int cleaned_count);
155 static void e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
156 struct e1000_rx_ring *rx_ring,
157 int cleaned_count);
158 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd);
159 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
160 int cmd);
161 static void e1000_enter_82542_rst(struct e1000_adapter *adapter);
162 static void e1000_leave_82542_rst(struct e1000_adapter *adapter);
163 static void e1000_tx_timeout(struct net_device *dev);
164 static void e1000_reset_task(struct work_struct *work);
165 static void e1000_smartspeed(struct e1000_adapter *adapter);
166 static int e1000_82547_fifo_workaround(struct e1000_adapter *adapter,
167 struct sk_buff *skb);
169 static void e1000_vlan_rx_register(struct net_device *netdev, struct vlan_group *grp);
170 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid);
171 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid);
172 static void e1000_restore_vlan(struct e1000_adapter *adapter);
174 #ifdef CONFIG_PM
175 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state);
176 static int e1000_resume(struct pci_dev *pdev);
177 #endif
178 static void e1000_shutdown(struct pci_dev *pdev);
180 #ifdef CONFIG_NET_POLL_CONTROLLER
181 /* for netdump / net console */
182 static void e1000_netpoll (struct net_device *netdev);
183 #endif
185 #define COPYBREAK_DEFAULT 256
186 static unsigned int copybreak __read_mostly = COPYBREAK_DEFAULT;
187 module_param(copybreak, uint, 0644);
188 MODULE_PARM_DESC(copybreak,
189 "Maximum size of packet that is copied to a new buffer on receive");
191 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
192 pci_channel_state_t state);
193 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev);
194 static void e1000_io_resume(struct pci_dev *pdev);
196 static struct pci_error_handlers e1000_err_handler = {
197 .error_detected = e1000_io_error_detected,
198 .slot_reset = e1000_io_slot_reset,
199 .resume = e1000_io_resume,
202 static struct pci_driver e1000_driver = {
203 .name = e1000_driver_name,
204 .id_table = e1000_pci_tbl,
205 .probe = e1000_probe,
206 .remove = __devexit_p(e1000_remove),
207 #ifdef CONFIG_PM
208 /* Power Managment Hooks */
209 .suspend = e1000_suspend,
210 .resume = e1000_resume,
211 #endif
212 .shutdown = e1000_shutdown,
213 .err_handler = &e1000_err_handler
216 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
217 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
218 MODULE_LICENSE("GPL");
219 MODULE_VERSION(DRV_VERSION);
221 static int debug = NETIF_MSG_DRV | NETIF_MSG_PROBE;
222 module_param(debug, int, 0);
223 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
226 * e1000_get_hw_dev - return device
227 * used by hardware layer to print debugging information
230 struct net_device *e1000_get_hw_dev(struct e1000_hw *hw)
232 struct e1000_adapter *adapter = hw->back;
233 return adapter->netdev;
237 * e1000_init_module - Driver Registration Routine
239 * e1000_init_module is the first routine called when the driver is
240 * loaded. All it does is register with the PCI subsystem.
243 static int __init e1000_init_module(void)
245 int ret;
246 pr_info("%s - version %s\n", e1000_driver_string, e1000_driver_version);
248 pr_info("%s\n", e1000_copyright);
250 ret = pci_register_driver(&e1000_driver);
251 if (copybreak != COPYBREAK_DEFAULT) {
252 if (copybreak == 0)
253 pr_info("copybreak disabled\n");
254 else
255 pr_info("copybreak enabled for "
256 "packets <= %u bytes\n", copybreak);
258 return ret;
261 module_init(e1000_init_module);
264 * e1000_exit_module - Driver Exit Cleanup Routine
266 * e1000_exit_module is called just before the driver is removed
267 * from memory.
270 static void __exit e1000_exit_module(void)
272 pci_unregister_driver(&e1000_driver);
275 module_exit(e1000_exit_module);
277 static int e1000_request_irq(struct e1000_adapter *adapter)
279 struct net_device *netdev = adapter->netdev;
280 irq_handler_t handler = e1000_intr;
281 int irq_flags = IRQF_SHARED;
282 int err;
284 err = request_irq(adapter->pdev->irq, handler, irq_flags, netdev->name,
285 netdev);
286 if (err) {
287 e_err(probe, "Unable to allocate interrupt Error: %d\n", err);
290 return err;
293 static void e1000_free_irq(struct e1000_adapter *adapter)
295 struct net_device *netdev = adapter->netdev;
297 free_irq(adapter->pdev->irq, netdev);
301 * e1000_irq_disable - Mask off interrupt generation on the NIC
302 * @adapter: board private structure
305 static void e1000_irq_disable(struct e1000_adapter *adapter)
307 struct e1000_hw *hw = &adapter->hw;
309 ew32(IMC, ~0);
310 E1000_WRITE_FLUSH();
311 synchronize_irq(adapter->pdev->irq);
315 * e1000_irq_enable - Enable default interrupt generation settings
316 * @adapter: board private structure
319 static void e1000_irq_enable(struct e1000_adapter *adapter)
321 struct e1000_hw *hw = &adapter->hw;
323 ew32(IMS, IMS_ENABLE_MASK);
324 E1000_WRITE_FLUSH();
327 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
329 struct e1000_hw *hw = &adapter->hw;
330 struct net_device *netdev = adapter->netdev;
331 u16 vid = hw->mng_cookie.vlan_id;
332 u16 old_vid = adapter->mng_vlan_id;
333 if (adapter->vlgrp) {
334 if (!vlan_group_get_device(adapter->vlgrp, vid)) {
335 if (hw->mng_cookie.status &
336 E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) {
337 e1000_vlan_rx_add_vid(netdev, vid);
338 adapter->mng_vlan_id = vid;
339 } else
340 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
342 if ((old_vid != (u16)E1000_MNG_VLAN_NONE) &&
343 (vid != old_vid) &&
344 !vlan_group_get_device(adapter->vlgrp, old_vid))
345 e1000_vlan_rx_kill_vid(netdev, old_vid);
346 } else
347 adapter->mng_vlan_id = vid;
351 static void e1000_init_manageability(struct e1000_adapter *adapter)
353 struct e1000_hw *hw = &adapter->hw;
355 if (adapter->en_mng_pt) {
356 u32 manc = er32(MANC);
358 /* disable hardware interception of ARP */
359 manc &= ~(E1000_MANC_ARP_EN);
361 ew32(MANC, manc);
365 static void e1000_release_manageability(struct e1000_adapter *adapter)
367 struct e1000_hw *hw = &adapter->hw;
369 if (adapter->en_mng_pt) {
370 u32 manc = er32(MANC);
372 /* re-enable hardware interception of ARP */
373 manc |= E1000_MANC_ARP_EN;
375 ew32(MANC, manc);
380 * e1000_configure - configure the hardware for RX and TX
381 * @adapter = private board structure
383 static void e1000_configure(struct e1000_adapter *adapter)
385 struct net_device *netdev = adapter->netdev;
386 int i;
388 e1000_set_rx_mode(netdev);
390 e1000_restore_vlan(adapter);
391 e1000_init_manageability(adapter);
393 e1000_configure_tx(adapter);
394 e1000_setup_rctl(adapter);
395 e1000_configure_rx(adapter);
396 /* call E1000_DESC_UNUSED which always leaves
397 * at least 1 descriptor unused to make sure
398 * next_to_use != next_to_clean */
399 for (i = 0; i < adapter->num_rx_queues; i++) {
400 struct e1000_rx_ring *ring = &adapter->rx_ring[i];
401 adapter->alloc_rx_buf(adapter, ring,
402 E1000_DESC_UNUSED(ring));
406 int e1000_up(struct e1000_adapter *adapter)
408 struct e1000_hw *hw = &adapter->hw;
410 /* hardware has been reset, we need to reload some things */
411 e1000_configure(adapter);
413 clear_bit(__E1000_DOWN, &adapter->flags);
415 napi_enable(&adapter->napi);
417 e1000_irq_enable(adapter);
419 netif_wake_queue(adapter->netdev);
421 /* fire a link change interrupt to start the watchdog */
422 ew32(ICS, E1000_ICS_LSC);
423 return 0;
427 * e1000_power_up_phy - restore link in case the phy was powered down
428 * @adapter: address of board private structure
430 * The phy may be powered down to save power and turn off link when the
431 * driver is unloaded and wake on lan is not enabled (among others)
432 * *** this routine MUST be followed by a call to e1000_reset ***
436 void e1000_power_up_phy(struct e1000_adapter *adapter)
438 struct e1000_hw *hw = &adapter->hw;
439 u16 mii_reg = 0;
441 /* Just clear the power down bit to wake the phy back up */
442 if (hw->media_type == e1000_media_type_copper) {
443 /* according to the manual, the phy will retain its
444 * settings across a power-down/up cycle */
445 e1000_read_phy_reg(hw, PHY_CTRL, &mii_reg);
446 mii_reg &= ~MII_CR_POWER_DOWN;
447 e1000_write_phy_reg(hw, PHY_CTRL, mii_reg);
451 static void e1000_power_down_phy(struct e1000_adapter *adapter)
453 struct e1000_hw *hw = &adapter->hw;
455 /* Power down the PHY so no link is implied when interface is down *
456 * The PHY cannot be powered down if any of the following is true *
457 * (a) WoL is enabled
458 * (b) AMT is active
459 * (c) SoL/IDER session is active */
460 if (!adapter->wol && hw->mac_type >= e1000_82540 &&
461 hw->media_type == e1000_media_type_copper) {
462 u16 mii_reg = 0;
464 switch (hw->mac_type) {
465 case e1000_82540:
466 case e1000_82545:
467 case e1000_82545_rev_3:
468 case e1000_82546:
469 case e1000_ce4100:
470 case e1000_82546_rev_3:
471 case e1000_82541:
472 case e1000_82541_rev_2:
473 case e1000_82547:
474 case e1000_82547_rev_2:
475 if (er32(MANC) & E1000_MANC_SMBUS_EN)
476 goto out;
477 break;
478 default:
479 goto out;
481 e1000_read_phy_reg(hw, PHY_CTRL, &mii_reg);
482 mii_reg |= MII_CR_POWER_DOWN;
483 e1000_write_phy_reg(hw, PHY_CTRL, mii_reg);
484 mdelay(1);
486 out:
487 return;
490 void e1000_down(struct e1000_adapter *adapter)
492 struct e1000_hw *hw = &adapter->hw;
493 struct net_device *netdev = adapter->netdev;
494 u32 rctl, tctl;
497 /* disable receives in the hardware */
498 rctl = er32(RCTL);
499 ew32(RCTL, rctl & ~E1000_RCTL_EN);
500 /* flush and sleep below */
502 netif_tx_disable(netdev);
504 /* disable transmits in the hardware */
505 tctl = er32(TCTL);
506 tctl &= ~E1000_TCTL_EN;
507 ew32(TCTL, tctl);
508 /* flush both disables and wait for them to finish */
509 E1000_WRITE_FLUSH();
510 msleep(10);
512 napi_disable(&adapter->napi);
514 e1000_irq_disable(adapter);
517 * Setting DOWN must be after irq_disable to prevent
518 * a screaming interrupt. Setting DOWN also prevents
519 * timers and tasks from rescheduling.
521 set_bit(__E1000_DOWN, &adapter->flags);
523 del_timer_sync(&adapter->tx_fifo_stall_timer);
524 del_timer_sync(&adapter->watchdog_timer);
525 del_timer_sync(&adapter->phy_info_timer);
527 adapter->link_speed = 0;
528 adapter->link_duplex = 0;
529 netif_carrier_off(netdev);
531 e1000_reset(adapter);
532 e1000_clean_all_tx_rings(adapter);
533 e1000_clean_all_rx_rings(adapter);
536 static void e1000_reinit_safe(struct e1000_adapter *adapter)
538 while (test_and_set_bit(__E1000_RESETTING, &adapter->flags))
539 msleep(1);
540 rtnl_lock();
541 e1000_down(adapter);
542 e1000_up(adapter);
543 rtnl_unlock();
544 clear_bit(__E1000_RESETTING, &adapter->flags);
547 void e1000_reinit_locked(struct e1000_adapter *adapter)
549 /* if rtnl_lock is not held the call path is bogus */
550 ASSERT_RTNL();
551 WARN_ON(in_interrupt());
552 while (test_and_set_bit(__E1000_RESETTING, &adapter->flags))
553 msleep(1);
554 e1000_down(adapter);
555 e1000_up(adapter);
556 clear_bit(__E1000_RESETTING, &adapter->flags);
559 void e1000_reset(struct e1000_adapter *adapter)
561 struct e1000_hw *hw = &adapter->hw;
562 u32 pba = 0, tx_space, min_tx_space, min_rx_space;
563 bool legacy_pba_adjust = false;
564 u16 hwm;
566 /* Repartition Pba for greater than 9k mtu
567 * To take effect CTRL.RST is required.
570 switch (hw->mac_type) {
571 case e1000_82542_rev2_0:
572 case e1000_82542_rev2_1:
573 case e1000_82543:
574 case e1000_82544:
575 case e1000_82540:
576 case e1000_82541:
577 case e1000_82541_rev_2:
578 legacy_pba_adjust = true;
579 pba = E1000_PBA_48K;
580 break;
581 case e1000_82545:
582 case e1000_82545_rev_3:
583 case e1000_82546:
584 case e1000_ce4100:
585 case e1000_82546_rev_3:
586 pba = E1000_PBA_48K;
587 break;
588 case e1000_82547:
589 case e1000_82547_rev_2:
590 legacy_pba_adjust = true;
591 pba = E1000_PBA_30K;
592 break;
593 case e1000_undefined:
594 case e1000_num_macs:
595 break;
598 if (legacy_pba_adjust) {
599 if (hw->max_frame_size > E1000_RXBUFFER_8192)
600 pba -= 8; /* allocate more FIFO for Tx */
602 if (hw->mac_type == e1000_82547) {
603 adapter->tx_fifo_head = 0;
604 adapter->tx_head_addr = pba << E1000_TX_HEAD_ADDR_SHIFT;
605 adapter->tx_fifo_size =
606 (E1000_PBA_40K - pba) << E1000_PBA_BYTES_SHIFT;
607 atomic_set(&adapter->tx_fifo_stall, 0);
609 } else if (hw->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
610 /* adjust PBA for jumbo frames */
611 ew32(PBA, pba);
613 /* To maintain wire speed transmits, the Tx FIFO should be
614 * large enough to accommodate two full transmit packets,
615 * rounded up to the next 1KB and expressed in KB. Likewise,
616 * the Rx FIFO should be large enough to accommodate at least
617 * one full receive packet and is similarly rounded up and
618 * expressed in KB. */
619 pba = er32(PBA);
620 /* upper 16 bits has Tx packet buffer allocation size in KB */
621 tx_space = pba >> 16;
622 /* lower 16 bits has Rx packet buffer allocation size in KB */
623 pba &= 0xffff;
625 * the tx fifo also stores 16 bytes of information about the tx
626 * but don't include ethernet FCS because hardware appends it
628 min_tx_space = (hw->max_frame_size +
629 sizeof(struct e1000_tx_desc) -
630 ETH_FCS_LEN) * 2;
631 min_tx_space = ALIGN(min_tx_space, 1024);
632 min_tx_space >>= 10;
633 /* software strips receive CRC, so leave room for it */
634 min_rx_space = hw->max_frame_size;
635 min_rx_space = ALIGN(min_rx_space, 1024);
636 min_rx_space >>= 10;
638 /* If current Tx allocation is less than the min Tx FIFO size,
639 * and the min Tx FIFO size is less than the current Rx FIFO
640 * allocation, take space away from current Rx allocation */
641 if (tx_space < min_tx_space &&
642 ((min_tx_space - tx_space) < pba)) {
643 pba = pba - (min_tx_space - tx_space);
645 /* PCI/PCIx hardware has PBA alignment constraints */
646 switch (hw->mac_type) {
647 case e1000_82545 ... e1000_82546_rev_3:
648 pba &= ~(E1000_PBA_8K - 1);
649 break;
650 default:
651 break;
654 /* if short on rx space, rx wins and must trump tx
655 * adjustment or use Early Receive if available */
656 if (pba < min_rx_space)
657 pba = min_rx_space;
661 ew32(PBA, pba);
664 * flow control settings:
665 * The high water mark must be low enough to fit one full frame
666 * (or the size used for early receive) above it in the Rx FIFO.
667 * Set it to the lower of:
668 * - 90% of the Rx FIFO size, and
669 * - the full Rx FIFO size minus the early receive size (for parts
670 * with ERT support assuming ERT set to E1000_ERT_2048), or
671 * - the full Rx FIFO size minus one full frame
673 hwm = min(((pba << 10) * 9 / 10),
674 ((pba << 10) - hw->max_frame_size));
676 hw->fc_high_water = hwm & 0xFFF8; /* 8-byte granularity */
677 hw->fc_low_water = hw->fc_high_water - 8;
678 hw->fc_pause_time = E1000_FC_PAUSE_TIME;
679 hw->fc_send_xon = 1;
680 hw->fc = hw->original_fc;
682 /* Allow time for pending master requests to run */
683 e1000_reset_hw(hw);
684 if (hw->mac_type >= e1000_82544)
685 ew32(WUC, 0);
687 if (e1000_init_hw(hw))
688 e_dev_err("Hardware Error\n");
689 e1000_update_mng_vlan(adapter);
691 /* if (adapter->hwflags & HWFLAGS_PHY_PWR_BIT) { */
692 if (hw->mac_type >= e1000_82544 &&
693 hw->autoneg == 1 &&
694 hw->autoneg_advertised == ADVERTISE_1000_FULL) {
695 u32 ctrl = er32(CTRL);
696 /* clear phy power management bit if we are in gig only mode,
697 * which if enabled will attempt negotiation to 100Mb, which
698 * can cause a loss of link at power off or driver unload */
699 ctrl &= ~E1000_CTRL_SWDPIN3;
700 ew32(CTRL, ctrl);
703 /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
704 ew32(VET, ETHERNET_IEEE_VLAN_TYPE);
706 e1000_reset_adaptive(hw);
707 e1000_phy_get_info(hw, &adapter->phy_info);
709 e1000_release_manageability(adapter);
713 * Dump the eeprom for users having checksum issues
715 static void e1000_dump_eeprom(struct e1000_adapter *adapter)
717 struct net_device *netdev = adapter->netdev;
718 struct ethtool_eeprom eeprom;
719 const struct ethtool_ops *ops = netdev->ethtool_ops;
720 u8 *data;
721 int i;
722 u16 csum_old, csum_new = 0;
724 eeprom.len = ops->get_eeprom_len(netdev);
725 eeprom.offset = 0;
727 data = kmalloc(eeprom.len, GFP_KERNEL);
728 if (!data) {
729 pr_err("Unable to allocate memory to dump EEPROM data\n");
730 return;
733 ops->get_eeprom(netdev, &eeprom, data);
735 csum_old = (data[EEPROM_CHECKSUM_REG * 2]) +
736 (data[EEPROM_CHECKSUM_REG * 2 + 1] << 8);
737 for (i = 0; i < EEPROM_CHECKSUM_REG * 2; i += 2)
738 csum_new += data[i] + (data[i + 1] << 8);
739 csum_new = EEPROM_SUM - csum_new;
741 pr_err("/*********************/\n");
742 pr_err("Current EEPROM Checksum : 0x%04x\n", csum_old);
743 pr_err("Calculated : 0x%04x\n", csum_new);
745 pr_err("Offset Values\n");
746 pr_err("======== ======\n");
747 print_hex_dump(KERN_ERR, "", DUMP_PREFIX_OFFSET, 16, 1, data, 128, 0);
749 pr_err("Include this output when contacting your support provider.\n");
750 pr_err("This is not a software error! Something bad happened to\n");
751 pr_err("your hardware or EEPROM image. Ignoring this problem could\n");
752 pr_err("result in further problems, possibly loss of data,\n");
753 pr_err("corruption or system hangs!\n");
754 pr_err("The MAC Address will be reset to 00:00:00:00:00:00,\n");
755 pr_err("which is invalid and requires you to set the proper MAC\n");
756 pr_err("address manually before continuing to enable this network\n");
757 pr_err("device. Please inspect the EEPROM dump and report the\n");
758 pr_err("issue to your hardware vendor or Intel Customer Support.\n");
759 pr_err("/*********************/\n");
761 kfree(data);
765 * e1000_is_need_ioport - determine if an adapter needs ioport resources or not
766 * @pdev: PCI device information struct
768 * Return true if an adapter needs ioport resources
770 static int e1000_is_need_ioport(struct pci_dev *pdev)
772 switch (pdev->device) {
773 case E1000_DEV_ID_82540EM:
774 case E1000_DEV_ID_82540EM_LOM:
775 case E1000_DEV_ID_82540EP:
776 case E1000_DEV_ID_82540EP_LOM:
777 case E1000_DEV_ID_82540EP_LP:
778 case E1000_DEV_ID_82541EI:
779 case E1000_DEV_ID_82541EI_MOBILE:
780 case E1000_DEV_ID_82541ER:
781 case E1000_DEV_ID_82541ER_LOM:
782 case E1000_DEV_ID_82541GI:
783 case E1000_DEV_ID_82541GI_LF:
784 case E1000_DEV_ID_82541GI_MOBILE:
785 case E1000_DEV_ID_82544EI_COPPER:
786 case E1000_DEV_ID_82544EI_FIBER:
787 case E1000_DEV_ID_82544GC_COPPER:
788 case E1000_DEV_ID_82544GC_LOM:
789 case E1000_DEV_ID_82545EM_COPPER:
790 case E1000_DEV_ID_82545EM_FIBER:
791 case E1000_DEV_ID_82546EB_COPPER:
792 case E1000_DEV_ID_82546EB_FIBER:
793 case E1000_DEV_ID_82546EB_QUAD_COPPER:
794 return true;
795 default:
796 return false;
800 static const struct net_device_ops e1000_netdev_ops = {
801 .ndo_open = e1000_open,
802 .ndo_stop = e1000_close,
803 .ndo_start_xmit = e1000_xmit_frame,
804 .ndo_get_stats = e1000_get_stats,
805 .ndo_set_rx_mode = e1000_set_rx_mode,
806 .ndo_set_mac_address = e1000_set_mac,
807 .ndo_tx_timeout = e1000_tx_timeout,
808 .ndo_change_mtu = e1000_change_mtu,
809 .ndo_do_ioctl = e1000_ioctl,
810 .ndo_validate_addr = eth_validate_addr,
812 .ndo_vlan_rx_register = e1000_vlan_rx_register,
813 .ndo_vlan_rx_add_vid = e1000_vlan_rx_add_vid,
814 .ndo_vlan_rx_kill_vid = e1000_vlan_rx_kill_vid,
815 #ifdef CONFIG_NET_POLL_CONTROLLER
816 .ndo_poll_controller = e1000_netpoll,
817 #endif
821 * e1000_init_hw_struct - initialize members of hw struct
822 * @adapter: board private struct
823 * @hw: structure used by e1000_hw.c
825 * Factors out initialization of the e1000_hw struct to its own function
826 * that can be called very early at init (just after struct allocation).
827 * Fields are initialized based on PCI device information and
828 * OS network device settings (MTU size).
829 * Returns negative error codes if MAC type setup fails.
831 static int e1000_init_hw_struct(struct e1000_adapter *adapter,
832 struct e1000_hw *hw)
834 struct pci_dev *pdev = adapter->pdev;
836 /* PCI config space info */
837 hw->vendor_id = pdev->vendor;
838 hw->device_id = pdev->device;
839 hw->subsystem_vendor_id = pdev->subsystem_vendor;
840 hw->subsystem_id = pdev->subsystem_device;
841 hw->revision_id = pdev->revision;
843 pci_read_config_word(pdev, PCI_COMMAND, &hw->pci_cmd_word);
845 hw->max_frame_size = adapter->netdev->mtu +
846 ENET_HEADER_SIZE + ETHERNET_FCS_SIZE;
847 hw->min_frame_size = MINIMUM_ETHERNET_FRAME_SIZE;
849 /* identify the MAC */
850 if (e1000_set_mac_type(hw)) {
851 e_err(probe, "Unknown MAC Type\n");
852 return -EIO;
855 switch (hw->mac_type) {
856 default:
857 break;
858 case e1000_82541:
859 case e1000_82547:
860 case e1000_82541_rev_2:
861 case e1000_82547_rev_2:
862 hw->phy_init_script = 1;
863 break;
866 e1000_set_media_type(hw);
867 e1000_get_bus_info(hw);
869 hw->wait_autoneg_complete = false;
870 hw->tbi_compatibility_en = true;
871 hw->adaptive_ifs = true;
873 /* Copper options */
875 if (hw->media_type == e1000_media_type_copper) {
876 hw->mdix = AUTO_ALL_MODES;
877 hw->disable_polarity_correction = false;
878 hw->master_slave = E1000_MASTER_SLAVE;
881 return 0;
885 * e1000_probe - Device Initialization Routine
886 * @pdev: PCI device information struct
887 * @ent: entry in e1000_pci_tbl
889 * Returns 0 on success, negative on failure
891 * e1000_probe initializes an adapter identified by a pci_dev structure.
892 * The OS initialization, configuring of the adapter private structure,
893 * and a hardware reset occur.
895 static int __devinit e1000_probe(struct pci_dev *pdev,
896 const struct pci_device_id *ent)
898 struct net_device *netdev;
899 struct e1000_adapter *adapter;
900 struct e1000_hw *hw;
902 static int cards_found = 0;
903 static int global_quad_port_a = 0; /* global ksp3 port a indication */
904 int i, err, pci_using_dac;
905 u16 eeprom_data = 0;
906 u16 tmp = 0;
907 u16 eeprom_apme_mask = E1000_EEPROM_APME;
908 int bars, need_ioport;
910 /* do not allocate ioport bars when not needed */
911 need_ioport = e1000_is_need_ioport(pdev);
912 if (need_ioport) {
913 bars = pci_select_bars(pdev, IORESOURCE_MEM | IORESOURCE_IO);
914 err = pci_enable_device(pdev);
915 } else {
916 bars = pci_select_bars(pdev, IORESOURCE_MEM);
917 err = pci_enable_device_mem(pdev);
919 if (err)
920 return err;
922 err = pci_request_selected_regions(pdev, bars, e1000_driver_name);
923 if (err)
924 goto err_pci_reg;
926 pci_set_master(pdev);
927 err = pci_save_state(pdev);
928 if (err)
929 goto err_alloc_etherdev;
931 err = -ENOMEM;
932 netdev = alloc_etherdev(sizeof(struct e1000_adapter));
933 if (!netdev)
934 goto err_alloc_etherdev;
936 SET_NETDEV_DEV(netdev, &pdev->dev);
938 pci_set_drvdata(pdev, netdev);
939 adapter = netdev_priv(netdev);
940 adapter->netdev = netdev;
941 adapter->pdev = pdev;
942 adapter->msg_enable = (1 << debug) - 1;
943 adapter->bars = bars;
944 adapter->need_ioport = need_ioport;
946 hw = &adapter->hw;
947 hw->back = adapter;
949 err = -EIO;
950 hw->hw_addr = pci_ioremap_bar(pdev, BAR_0);
951 if (!hw->hw_addr)
952 goto err_ioremap;
954 if (adapter->need_ioport) {
955 for (i = BAR_1; i <= BAR_5; i++) {
956 if (pci_resource_len(pdev, i) == 0)
957 continue;
958 if (pci_resource_flags(pdev, i) & IORESOURCE_IO) {
959 hw->io_base = pci_resource_start(pdev, i);
960 break;
965 /* make ready for any if (hw->...) below */
966 err = e1000_init_hw_struct(adapter, hw);
967 if (err)
968 goto err_sw_init;
971 * there is a workaround being applied below that limits
972 * 64-bit DMA addresses to 64-bit hardware. There are some
973 * 32-bit adapters that Tx hang when given 64-bit DMA addresses
975 pci_using_dac = 0;
976 if ((hw->bus_type == e1000_bus_type_pcix) &&
977 !dma_set_mask(&pdev->dev, DMA_BIT_MASK(64))) {
979 * according to DMA-API-HOWTO, coherent calls will always
980 * succeed if the set call did
982 dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64));
983 pci_using_dac = 1;
984 } else {
985 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
986 if (err) {
987 pr_err("No usable DMA config, aborting\n");
988 goto err_dma;
990 dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(32));
993 netdev->netdev_ops = &e1000_netdev_ops;
994 e1000_set_ethtool_ops(netdev);
995 netdev->watchdog_timeo = 5 * HZ;
996 netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
998 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
1000 adapter->bd_number = cards_found;
1002 /* setup the private structure */
1004 err = e1000_sw_init(adapter);
1005 if (err)
1006 goto err_sw_init;
1008 err = -EIO;
1009 if (hw->mac_type == e1000_ce4100) {
1010 ce4100_gbe_mdio_base_phy = pci_resource_start(pdev, BAR_1);
1011 ce4100_gbe_mdio_base_virt = ioremap(ce4100_gbe_mdio_base_phy,
1012 pci_resource_len(pdev, BAR_1));
1014 if (!ce4100_gbe_mdio_base_virt)
1015 goto err_mdio_ioremap;
1018 if (hw->mac_type >= e1000_82543) {
1019 netdev->features = NETIF_F_SG |
1020 NETIF_F_HW_CSUM |
1021 NETIF_F_HW_VLAN_TX |
1022 NETIF_F_HW_VLAN_RX |
1023 NETIF_F_HW_VLAN_FILTER;
1026 if ((hw->mac_type >= e1000_82544) &&
1027 (hw->mac_type != e1000_82547))
1028 netdev->features |= NETIF_F_TSO;
1030 if (pci_using_dac) {
1031 netdev->features |= NETIF_F_HIGHDMA;
1032 netdev->vlan_features |= NETIF_F_HIGHDMA;
1035 netdev->vlan_features |= NETIF_F_TSO;
1036 netdev->vlan_features |= NETIF_F_HW_CSUM;
1037 netdev->vlan_features |= NETIF_F_SG;
1039 adapter->en_mng_pt = e1000_enable_mng_pass_thru(hw);
1041 /* initialize eeprom parameters */
1042 if (e1000_init_eeprom_params(hw)) {
1043 e_err(probe, "EEPROM initialization failed\n");
1044 goto err_eeprom;
1047 /* before reading the EEPROM, reset the controller to
1048 * put the device in a known good starting state */
1050 e1000_reset_hw(hw);
1052 /* make sure the EEPROM is good */
1053 if (e1000_validate_eeprom_checksum(hw) < 0) {
1054 e_err(probe, "The EEPROM Checksum Is Not Valid\n");
1055 e1000_dump_eeprom(adapter);
1057 * set MAC address to all zeroes to invalidate and temporary
1058 * disable this device for the user. This blocks regular
1059 * traffic while still permitting ethtool ioctls from reaching
1060 * the hardware as well as allowing the user to run the
1061 * interface after manually setting a hw addr using
1062 * `ip set address`
1064 memset(hw->mac_addr, 0, netdev->addr_len);
1065 } else {
1066 /* copy the MAC address out of the EEPROM */
1067 if (e1000_read_mac_addr(hw))
1068 e_err(probe, "EEPROM Read Error\n");
1070 /* don't block initalization here due to bad MAC address */
1071 memcpy(netdev->dev_addr, hw->mac_addr, netdev->addr_len);
1072 memcpy(netdev->perm_addr, hw->mac_addr, netdev->addr_len);
1074 if (!is_valid_ether_addr(netdev->perm_addr))
1075 e_err(probe, "Invalid MAC Address\n");
1077 init_timer(&adapter->tx_fifo_stall_timer);
1078 adapter->tx_fifo_stall_timer.function = e1000_82547_tx_fifo_stall;
1079 adapter->tx_fifo_stall_timer.data = (unsigned long)adapter;
1081 init_timer(&adapter->watchdog_timer);
1082 adapter->watchdog_timer.function = e1000_watchdog;
1083 adapter->watchdog_timer.data = (unsigned long) adapter;
1085 init_timer(&adapter->phy_info_timer);
1086 adapter->phy_info_timer.function = e1000_update_phy_info;
1087 adapter->phy_info_timer.data = (unsigned long)adapter;
1089 INIT_WORK(&adapter->fifo_stall_task, e1000_82547_tx_fifo_stall_task);
1090 INIT_WORK(&adapter->reset_task, e1000_reset_task);
1091 INIT_WORK(&adapter->phy_info_task, e1000_update_phy_info_task);
1093 e1000_check_options(adapter);
1095 /* Initial Wake on LAN setting
1096 * If APM wake is enabled in the EEPROM,
1097 * enable the ACPI Magic Packet filter
1100 switch (hw->mac_type) {
1101 case e1000_82542_rev2_0:
1102 case e1000_82542_rev2_1:
1103 case e1000_82543:
1104 break;
1105 case e1000_82544:
1106 e1000_read_eeprom(hw,
1107 EEPROM_INIT_CONTROL2_REG, 1, &eeprom_data);
1108 eeprom_apme_mask = E1000_EEPROM_82544_APM;
1109 break;
1110 case e1000_82546:
1111 case e1000_82546_rev_3:
1112 if (er32(STATUS) & E1000_STATUS_FUNC_1){
1113 e1000_read_eeprom(hw,
1114 EEPROM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
1115 break;
1117 /* Fall Through */
1118 default:
1119 e1000_read_eeprom(hw,
1120 EEPROM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
1121 break;
1123 if (eeprom_data & eeprom_apme_mask)
1124 adapter->eeprom_wol |= E1000_WUFC_MAG;
1126 /* now that we have the eeprom settings, apply the special cases
1127 * where the eeprom may be wrong or the board simply won't support
1128 * wake on lan on a particular port */
1129 switch (pdev->device) {
1130 case E1000_DEV_ID_82546GB_PCIE:
1131 adapter->eeprom_wol = 0;
1132 break;
1133 case E1000_DEV_ID_82546EB_FIBER:
1134 case E1000_DEV_ID_82546GB_FIBER:
1135 /* Wake events only supported on port A for dual fiber
1136 * regardless of eeprom setting */
1137 if (er32(STATUS) & E1000_STATUS_FUNC_1)
1138 adapter->eeprom_wol = 0;
1139 break;
1140 case E1000_DEV_ID_82546GB_QUAD_COPPER_KSP3:
1141 /* if quad port adapter, disable WoL on all but port A */
1142 if (global_quad_port_a != 0)
1143 adapter->eeprom_wol = 0;
1144 else
1145 adapter->quad_port_a = 1;
1146 /* Reset for multiple quad port adapters */
1147 if (++global_quad_port_a == 4)
1148 global_quad_port_a = 0;
1149 break;
1152 /* initialize the wol settings based on the eeprom settings */
1153 adapter->wol = adapter->eeprom_wol;
1154 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
1156 /* Auto detect PHY address */
1157 if (hw->mac_type == e1000_ce4100) {
1158 for (i = 0; i < 32; i++) {
1159 hw->phy_addr = i;
1160 e1000_read_phy_reg(hw, PHY_ID2, &tmp);
1161 if (tmp == 0 || tmp == 0xFF) {
1162 if (i == 31)
1163 goto err_eeprom;
1164 continue;
1165 } else
1166 break;
1170 /* reset the hardware with the new settings */
1171 e1000_reset(adapter);
1173 strcpy(netdev->name, "eth%d");
1174 err = register_netdev(netdev);
1175 if (err)
1176 goto err_register;
1178 /* print bus type/speed/width info */
1179 e_info(probe, "(PCI%s:%dMHz:%d-bit) %pM\n",
1180 ((hw->bus_type == e1000_bus_type_pcix) ? "-X" : ""),
1181 ((hw->bus_speed == e1000_bus_speed_133) ? 133 :
1182 (hw->bus_speed == e1000_bus_speed_120) ? 120 :
1183 (hw->bus_speed == e1000_bus_speed_100) ? 100 :
1184 (hw->bus_speed == e1000_bus_speed_66) ? 66 : 33),
1185 ((hw->bus_width == e1000_bus_width_64) ? 64 : 32),
1186 netdev->dev_addr);
1188 /* carrier off reporting is important to ethtool even BEFORE open */
1189 netif_carrier_off(netdev);
1191 e_info(probe, "Intel(R) PRO/1000 Network Connection\n");
1193 cards_found++;
1194 return 0;
1196 err_register:
1197 err_eeprom:
1198 e1000_phy_hw_reset(hw);
1200 if (hw->flash_address)
1201 iounmap(hw->flash_address);
1202 kfree(adapter->tx_ring);
1203 kfree(adapter->rx_ring);
1204 err_dma:
1205 err_sw_init:
1206 err_mdio_ioremap:
1207 iounmap(ce4100_gbe_mdio_base_virt);
1208 iounmap(hw->hw_addr);
1209 err_ioremap:
1210 free_netdev(netdev);
1211 err_alloc_etherdev:
1212 pci_release_selected_regions(pdev, bars);
1213 err_pci_reg:
1214 pci_disable_device(pdev);
1215 return err;
1219 * e1000_remove - Device Removal Routine
1220 * @pdev: PCI device information struct
1222 * e1000_remove is called by the PCI subsystem to alert the driver
1223 * that it should release a PCI device. The could be caused by a
1224 * Hot-Plug event, or because the driver is going to be removed from
1225 * memory.
1228 static void __devexit e1000_remove(struct pci_dev *pdev)
1230 struct net_device *netdev = pci_get_drvdata(pdev);
1231 struct e1000_adapter *adapter = netdev_priv(netdev);
1232 struct e1000_hw *hw = &adapter->hw;
1234 set_bit(__E1000_DOWN, &adapter->flags);
1235 del_timer_sync(&adapter->tx_fifo_stall_timer);
1236 del_timer_sync(&adapter->watchdog_timer);
1237 del_timer_sync(&adapter->phy_info_timer);
1239 cancel_work_sync(&adapter->reset_task);
1241 e1000_release_manageability(adapter);
1243 unregister_netdev(netdev);
1245 e1000_phy_hw_reset(hw);
1247 kfree(adapter->tx_ring);
1248 kfree(adapter->rx_ring);
1250 iounmap(hw->hw_addr);
1251 if (hw->flash_address)
1252 iounmap(hw->flash_address);
1253 pci_release_selected_regions(pdev, adapter->bars);
1255 free_netdev(netdev);
1257 pci_disable_device(pdev);
1261 * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
1262 * @adapter: board private structure to initialize
1264 * e1000_sw_init initializes the Adapter private data structure.
1265 * e1000_init_hw_struct MUST be called before this function
1268 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
1270 adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
1272 adapter->num_tx_queues = 1;
1273 adapter->num_rx_queues = 1;
1275 if (e1000_alloc_queues(adapter)) {
1276 e_err(probe, "Unable to allocate memory for queues\n");
1277 return -ENOMEM;
1280 /* Explicitly disable IRQ since the NIC can be in any state. */
1281 e1000_irq_disable(adapter);
1283 spin_lock_init(&adapter->stats_lock);
1285 set_bit(__E1000_DOWN, &adapter->flags);
1287 return 0;
1291 * e1000_alloc_queues - Allocate memory for all rings
1292 * @adapter: board private structure to initialize
1294 * We allocate one ring per queue at run-time since we don't know the
1295 * number of queues at compile-time.
1298 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
1300 adapter->tx_ring = kcalloc(adapter->num_tx_queues,
1301 sizeof(struct e1000_tx_ring), GFP_KERNEL);
1302 if (!adapter->tx_ring)
1303 return -ENOMEM;
1305 adapter->rx_ring = kcalloc(adapter->num_rx_queues,
1306 sizeof(struct e1000_rx_ring), GFP_KERNEL);
1307 if (!adapter->rx_ring) {
1308 kfree(adapter->tx_ring);
1309 return -ENOMEM;
1312 return E1000_SUCCESS;
1316 * e1000_open - Called when a network interface is made active
1317 * @netdev: network interface device structure
1319 * Returns 0 on success, negative value on failure
1321 * The open entry point is called when a network interface is made
1322 * active by the system (IFF_UP). At this point all resources needed
1323 * for transmit and receive operations are allocated, the interrupt
1324 * handler is registered with the OS, the watchdog timer is started,
1325 * and the stack is notified that the interface is ready.
1328 static int e1000_open(struct net_device *netdev)
1330 struct e1000_adapter *adapter = netdev_priv(netdev);
1331 struct e1000_hw *hw = &adapter->hw;
1332 int err;
1334 /* disallow open during test */
1335 if (test_bit(__E1000_TESTING, &adapter->flags))
1336 return -EBUSY;
1338 netif_carrier_off(netdev);
1340 /* allocate transmit descriptors */
1341 err = e1000_setup_all_tx_resources(adapter);
1342 if (err)
1343 goto err_setup_tx;
1345 /* allocate receive descriptors */
1346 err = e1000_setup_all_rx_resources(adapter);
1347 if (err)
1348 goto err_setup_rx;
1350 e1000_power_up_phy(adapter);
1352 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
1353 if ((hw->mng_cookie.status &
1354 E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT)) {
1355 e1000_update_mng_vlan(adapter);
1358 /* before we allocate an interrupt, we must be ready to handle it.
1359 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1360 * as soon as we call pci_request_irq, so we have to setup our
1361 * clean_rx handler before we do so. */
1362 e1000_configure(adapter);
1364 err = e1000_request_irq(adapter);
1365 if (err)
1366 goto err_req_irq;
1368 /* From here on the code is the same as e1000_up() */
1369 clear_bit(__E1000_DOWN, &adapter->flags);
1371 napi_enable(&adapter->napi);
1373 e1000_irq_enable(adapter);
1375 netif_start_queue(netdev);
1377 /* fire a link status change interrupt to start the watchdog */
1378 ew32(ICS, E1000_ICS_LSC);
1380 return E1000_SUCCESS;
1382 err_req_irq:
1383 e1000_power_down_phy(adapter);
1384 e1000_free_all_rx_resources(adapter);
1385 err_setup_rx:
1386 e1000_free_all_tx_resources(adapter);
1387 err_setup_tx:
1388 e1000_reset(adapter);
1390 return err;
1394 * e1000_close - Disables a network interface
1395 * @netdev: network interface device structure
1397 * Returns 0, this is not allowed to fail
1399 * The close entry point is called when an interface is de-activated
1400 * by the OS. The hardware is still under the drivers control, but
1401 * needs to be disabled. A global MAC reset is issued to stop the
1402 * hardware, and all transmit and receive resources are freed.
1405 static int e1000_close(struct net_device *netdev)
1407 struct e1000_adapter *adapter = netdev_priv(netdev);
1408 struct e1000_hw *hw = &adapter->hw;
1410 WARN_ON(test_bit(__E1000_RESETTING, &adapter->flags));
1411 e1000_down(adapter);
1412 e1000_power_down_phy(adapter);
1413 e1000_free_irq(adapter);
1415 e1000_free_all_tx_resources(adapter);
1416 e1000_free_all_rx_resources(adapter);
1418 /* kill manageability vlan ID if supported, but not if a vlan with
1419 * the same ID is registered on the host OS (let 8021q kill it) */
1420 if ((hw->mng_cookie.status &
1421 E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) &&
1422 !(adapter->vlgrp &&
1423 vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id))) {
1424 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
1427 return 0;
1431 * e1000_check_64k_bound - check that memory doesn't cross 64kB boundary
1432 * @adapter: address of board private structure
1433 * @start: address of beginning of memory
1434 * @len: length of memory
1436 static bool e1000_check_64k_bound(struct e1000_adapter *adapter, void *start,
1437 unsigned long len)
1439 struct e1000_hw *hw = &adapter->hw;
1440 unsigned long begin = (unsigned long)start;
1441 unsigned long end = begin + len;
1443 /* First rev 82545 and 82546 need to not allow any memory
1444 * write location to cross 64k boundary due to errata 23 */
1445 if (hw->mac_type == e1000_82545 ||
1446 hw->mac_type == e1000_ce4100 ||
1447 hw->mac_type == e1000_82546) {
1448 return ((begin ^ (end - 1)) >> 16) != 0 ? false : true;
1451 return true;
1455 * e1000_setup_tx_resources - allocate Tx resources (Descriptors)
1456 * @adapter: board private structure
1457 * @txdr: tx descriptor ring (for a specific queue) to setup
1459 * Return 0 on success, negative on failure
1462 static int e1000_setup_tx_resources(struct e1000_adapter *adapter,
1463 struct e1000_tx_ring *txdr)
1465 struct pci_dev *pdev = adapter->pdev;
1466 int size;
1468 size = sizeof(struct e1000_buffer) * txdr->count;
1469 txdr->buffer_info = vzalloc(size);
1470 if (!txdr->buffer_info) {
1471 e_err(probe, "Unable to allocate memory for the Tx descriptor "
1472 "ring\n");
1473 return -ENOMEM;
1476 /* round up to nearest 4K */
1478 txdr->size = txdr->count * sizeof(struct e1000_tx_desc);
1479 txdr->size = ALIGN(txdr->size, 4096);
1481 txdr->desc = dma_alloc_coherent(&pdev->dev, txdr->size, &txdr->dma,
1482 GFP_KERNEL);
1483 if (!txdr->desc) {
1484 setup_tx_desc_die:
1485 vfree(txdr->buffer_info);
1486 e_err(probe, "Unable to allocate memory for the Tx descriptor "
1487 "ring\n");
1488 return -ENOMEM;
1491 /* Fix for errata 23, can't cross 64kB boundary */
1492 if (!e1000_check_64k_bound(adapter, txdr->desc, txdr->size)) {
1493 void *olddesc = txdr->desc;
1494 dma_addr_t olddma = txdr->dma;
1495 e_err(tx_err, "txdr align check failed: %u bytes at %p\n",
1496 txdr->size, txdr->desc);
1497 /* Try again, without freeing the previous */
1498 txdr->desc = dma_alloc_coherent(&pdev->dev, txdr->size,
1499 &txdr->dma, GFP_KERNEL);
1500 /* Failed allocation, critical failure */
1501 if (!txdr->desc) {
1502 dma_free_coherent(&pdev->dev, txdr->size, olddesc,
1503 olddma);
1504 goto setup_tx_desc_die;
1507 if (!e1000_check_64k_bound(adapter, txdr->desc, txdr->size)) {
1508 /* give up */
1509 dma_free_coherent(&pdev->dev, txdr->size, txdr->desc,
1510 txdr->dma);
1511 dma_free_coherent(&pdev->dev, txdr->size, olddesc,
1512 olddma);
1513 e_err(probe, "Unable to allocate aligned memory "
1514 "for the transmit descriptor ring\n");
1515 vfree(txdr->buffer_info);
1516 return -ENOMEM;
1517 } else {
1518 /* Free old allocation, new allocation was successful */
1519 dma_free_coherent(&pdev->dev, txdr->size, olddesc,
1520 olddma);
1523 memset(txdr->desc, 0, txdr->size);
1525 txdr->next_to_use = 0;
1526 txdr->next_to_clean = 0;
1528 return 0;
1532 * e1000_setup_all_tx_resources - wrapper to allocate Tx resources
1533 * (Descriptors) for all queues
1534 * @adapter: board private structure
1536 * Return 0 on success, negative on failure
1539 int e1000_setup_all_tx_resources(struct e1000_adapter *adapter)
1541 int i, err = 0;
1543 for (i = 0; i < adapter->num_tx_queues; i++) {
1544 err = e1000_setup_tx_resources(adapter, &adapter->tx_ring[i]);
1545 if (err) {
1546 e_err(probe, "Allocation for Tx Queue %u failed\n", i);
1547 for (i-- ; i >= 0; i--)
1548 e1000_free_tx_resources(adapter,
1549 &adapter->tx_ring[i]);
1550 break;
1554 return err;
1558 * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
1559 * @adapter: board private structure
1561 * Configure the Tx unit of the MAC after a reset.
1564 static void e1000_configure_tx(struct e1000_adapter *adapter)
1566 u64 tdba;
1567 struct e1000_hw *hw = &adapter->hw;
1568 u32 tdlen, tctl, tipg;
1569 u32 ipgr1, ipgr2;
1571 /* Setup the HW Tx Head and Tail descriptor pointers */
1573 switch (adapter->num_tx_queues) {
1574 case 1:
1575 default:
1576 tdba = adapter->tx_ring[0].dma;
1577 tdlen = adapter->tx_ring[0].count *
1578 sizeof(struct e1000_tx_desc);
1579 ew32(TDLEN, tdlen);
1580 ew32(TDBAH, (tdba >> 32));
1581 ew32(TDBAL, (tdba & 0x00000000ffffffffULL));
1582 ew32(TDT, 0);
1583 ew32(TDH, 0);
1584 adapter->tx_ring[0].tdh = ((hw->mac_type >= e1000_82543) ? E1000_TDH : E1000_82542_TDH);
1585 adapter->tx_ring[0].tdt = ((hw->mac_type >= e1000_82543) ? E1000_TDT : E1000_82542_TDT);
1586 break;
1589 /* Set the default values for the Tx Inter Packet Gap timer */
1590 if ((hw->media_type == e1000_media_type_fiber ||
1591 hw->media_type == e1000_media_type_internal_serdes))
1592 tipg = DEFAULT_82543_TIPG_IPGT_FIBER;
1593 else
1594 tipg = DEFAULT_82543_TIPG_IPGT_COPPER;
1596 switch (hw->mac_type) {
1597 case e1000_82542_rev2_0:
1598 case e1000_82542_rev2_1:
1599 tipg = DEFAULT_82542_TIPG_IPGT;
1600 ipgr1 = DEFAULT_82542_TIPG_IPGR1;
1601 ipgr2 = DEFAULT_82542_TIPG_IPGR2;
1602 break;
1603 default:
1604 ipgr1 = DEFAULT_82543_TIPG_IPGR1;
1605 ipgr2 = DEFAULT_82543_TIPG_IPGR2;
1606 break;
1608 tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
1609 tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
1610 ew32(TIPG, tipg);
1612 /* Set the Tx Interrupt Delay register */
1614 ew32(TIDV, adapter->tx_int_delay);
1615 if (hw->mac_type >= e1000_82540)
1616 ew32(TADV, adapter->tx_abs_int_delay);
1618 /* Program the Transmit Control Register */
1620 tctl = er32(TCTL);
1621 tctl &= ~E1000_TCTL_CT;
1622 tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
1623 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
1625 e1000_config_collision_dist(hw);
1627 /* Setup Transmit Descriptor Settings for eop descriptor */
1628 adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
1630 /* only set IDE if we are delaying interrupts using the timers */
1631 if (adapter->tx_int_delay)
1632 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
1634 if (hw->mac_type < e1000_82543)
1635 adapter->txd_cmd |= E1000_TXD_CMD_RPS;
1636 else
1637 adapter->txd_cmd |= E1000_TXD_CMD_RS;
1639 /* Cache if we're 82544 running in PCI-X because we'll
1640 * need this to apply a workaround later in the send path. */
1641 if (hw->mac_type == e1000_82544 &&
1642 hw->bus_type == e1000_bus_type_pcix)
1643 adapter->pcix_82544 = 1;
1645 ew32(TCTL, tctl);
1650 * e1000_setup_rx_resources - allocate Rx resources (Descriptors)
1651 * @adapter: board private structure
1652 * @rxdr: rx descriptor ring (for a specific queue) to setup
1654 * Returns 0 on success, negative on failure
1657 static int e1000_setup_rx_resources(struct e1000_adapter *adapter,
1658 struct e1000_rx_ring *rxdr)
1660 struct pci_dev *pdev = adapter->pdev;
1661 int size, desc_len;
1663 size = sizeof(struct e1000_buffer) * rxdr->count;
1664 rxdr->buffer_info = vzalloc(size);
1665 if (!rxdr->buffer_info) {
1666 e_err(probe, "Unable to allocate memory for the Rx descriptor "
1667 "ring\n");
1668 return -ENOMEM;
1671 desc_len = sizeof(struct e1000_rx_desc);
1673 /* Round up to nearest 4K */
1675 rxdr->size = rxdr->count * desc_len;
1676 rxdr->size = ALIGN(rxdr->size, 4096);
1678 rxdr->desc = dma_alloc_coherent(&pdev->dev, rxdr->size, &rxdr->dma,
1679 GFP_KERNEL);
1681 if (!rxdr->desc) {
1682 e_err(probe, "Unable to allocate memory for the Rx descriptor "
1683 "ring\n");
1684 setup_rx_desc_die:
1685 vfree(rxdr->buffer_info);
1686 return -ENOMEM;
1689 /* Fix for errata 23, can't cross 64kB boundary */
1690 if (!e1000_check_64k_bound(adapter, rxdr->desc, rxdr->size)) {
1691 void *olddesc = rxdr->desc;
1692 dma_addr_t olddma = rxdr->dma;
1693 e_err(rx_err, "rxdr align check failed: %u bytes at %p\n",
1694 rxdr->size, rxdr->desc);
1695 /* Try again, without freeing the previous */
1696 rxdr->desc = dma_alloc_coherent(&pdev->dev, rxdr->size,
1697 &rxdr->dma, GFP_KERNEL);
1698 /* Failed allocation, critical failure */
1699 if (!rxdr->desc) {
1700 dma_free_coherent(&pdev->dev, rxdr->size, olddesc,
1701 olddma);
1702 e_err(probe, "Unable to allocate memory for the Rx "
1703 "descriptor ring\n");
1704 goto setup_rx_desc_die;
1707 if (!e1000_check_64k_bound(adapter, rxdr->desc, rxdr->size)) {
1708 /* give up */
1709 dma_free_coherent(&pdev->dev, rxdr->size, rxdr->desc,
1710 rxdr->dma);
1711 dma_free_coherent(&pdev->dev, rxdr->size, olddesc,
1712 olddma);
1713 e_err(probe, "Unable to allocate aligned memory for "
1714 "the Rx descriptor ring\n");
1715 goto setup_rx_desc_die;
1716 } else {
1717 /* Free old allocation, new allocation was successful */
1718 dma_free_coherent(&pdev->dev, rxdr->size, olddesc,
1719 olddma);
1722 memset(rxdr->desc, 0, rxdr->size);
1724 rxdr->next_to_clean = 0;
1725 rxdr->next_to_use = 0;
1726 rxdr->rx_skb_top = NULL;
1728 return 0;
1732 * e1000_setup_all_rx_resources - wrapper to allocate Rx resources
1733 * (Descriptors) for all queues
1734 * @adapter: board private structure
1736 * Return 0 on success, negative on failure
1739 int e1000_setup_all_rx_resources(struct e1000_adapter *adapter)
1741 int i, err = 0;
1743 for (i = 0; i < adapter->num_rx_queues; i++) {
1744 err = e1000_setup_rx_resources(adapter, &adapter->rx_ring[i]);
1745 if (err) {
1746 e_err(probe, "Allocation for Rx Queue %u failed\n", i);
1747 for (i-- ; i >= 0; i--)
1748 e1000_free_rx_resources(adapter,
1749 &adapter->rx_ring[i]);
1750 break;
1754 return err;
1758 * e1000_setup_rctl - configure the receive control registers
1759 * @adapter: Board private structure
1761 static void e1000_setup_rctl(struct e1000_adapter *adapter)
1763 struct e1000_hw *hw = &adapter->hw;
1764 u32 rctl;
1766 rctl = er32(RCTL);
1768 rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
1770 rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
1771 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
1772 (hw->mc_filter_type << E1000_RCTL_MO_SHIFT);
1774 if (hw->tbi_compatibility_on == 1)
1775 rctl |= E1000_RCTL_SBP;
1776 else
1777 rctl &= ~E1000_RCTL_SBP;
1779 if (adapter->netdev->mtu <= ETH_DATA_LEN)
1780 rctl &= ~E1000_RCTL_LPE;
1781 else
1782 rctl |= E1000_RCTL_LPE;
1784 /* Setup buffer sizes */
1785 rctl &= ~E1000_RCTL_SZ_4096;
1786 rctl |= E1000_RCTL_BSEX;
1787 switch (adapter->rx_buffer_len) {
1788 case E1000_RXBUFFER_2048:
1789 default:
1790 rctl |= E1000_RCTL_SZ_2048;
1791 rctl &= ~E1000_RCTL_BSEX;
1792 break;
1793 case E1000_RXBUFFER_4096:
1794 rctl |= E1000_RCTL_SZ_4096;
1795 break;
1796 case E1000_RXBUFFER_8192:
1797 rctl |= E1000_RCTL_SZ_8192;
1798 break;
1799 case E1000_RXBUFFER_16384:
1800 rctl |= E1000_RCTL_SZ_16384;
1801 break;
1804 ew32(RCTL, rctl);
1808 * e1000_configure_rx - Configure 8254x Receive Unit after Reset
1809 * @adapter: board private structure
1811 * Configure the Rx unit of the MAC after a reset.
1814 static void e1000_configure_rx(struct e1000_adapter *adapter)
1816 u64 rdba;
1817 struct e1000_hw *hw = &adapter->hw;
1818 u32 rdlen, rctl, rxcsum;
1820 if (adapter->netdev->mtu > ETH_DATA_LEN) {
1821 rdlen = adapter->rx_ring[0].count *
1822 sizeof(struct e1000_rx_desc);
1823 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
1824 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
1825 } else {
1826 rdlen = adapter->rx_ring[0].count *
1827 sizeof(struct e1000_rx_desc);
1828 adapter->clean_rx = e1000_clean_rx_irq;
1829 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
1832 /* disable receives while setting up the descriptors */
1833 rctl = er32(RCTL);
1834 ew32(RCTL, rctl & ~E1000_RCTL_EN);
1836 /* set the Receive Delay Timer Register */
1837 ew32(RDTR, adapter->rx_int_delay);
1839 if (hw->mac_type >= e1000_82540) {
1840 ew32(RADV, adapter->rx_abs_int_delay);
1841 if (adapter->itr_setting != 0)
1842 ew32(ITR, 1000000000 / (adapter->itr * 256));
1845 /* Setup the HW Rx Head and Tail Descriptor Pointers and
1846 * the Base and Length of the Rx Descriptor Ring */
1847 switch (adapter->num_rx_queues) {
1848 case 1:
1849 default:
1850 rdba = adapter->rx_ring[0].dma;
1851 ew32(RDLEN, rdlen);
1852 ew32(RDBAH, (rdba >> 32));
1853 ew32(RDBAL, (rdba & 0x00000000ffffffffULL));
1854 ew32(RDT, 0);
1855 ew32(RDH, 0);
1856 adapter->rx_ring[0].rdh = ((hw->mac_type >= e1000_82543) ? E1000_RDH : E1000_82542_RDH);
1857 adapter->rx_ring[0].rdt = ((hw->mac_type >= e1000_82543) ? E1000_RDT : E1000_82542_RDT);
1858 break;
1861 /* Enable 82543 Receive Checksum Offload for TCP and UDP */
1862 if (hw->mac_type >= e1000_82543) {
1863 rxcsum = er32(RXCSUM);
1864 if (adapter->rx_csum)
1865 rxcsum |= E1000_RXCSUM_TUOFL;
1866 else
1867 /* don't need to clear IPPCSE as it defaults to 0 */
1868 rxcsum &= ~E1000_RXCSUM_TUOFL;
1869 ew32(RXCSUM, rxcsum);
1872 /* Enable Receives */
1873 ew32(RCTL, rctl);
1877 * e1000_free_tx_resources - Free Tx Resources per Queue
1878 * @adapter: board private structure
1879 * @tx_ring: Tx descriptor ring for a specific queue
1881 * Free all transmit software resources
1884 static void e1000_free_tx_resources(struct e1000_adapter *adapter,
1885 struct e1000_tx_ring *tx_ring)
1887 struct pci_dev *pdev = adapter->pdev;
1889 e1000_clean_tx_ring(adapter, tx_ring);
1891 vfree(tx_ring->buffer_info);
1892 tx_ring->buffer_info = NULL;
1894 dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
1895 tx_ring->dma);
1897 tx_ring->desc = NULL;
1901 * e1000_free_all_tx_resources - Free Tx Resources for All Queues
1902 * @adapter: board private structure
1904 * Free all transmit software resources
1907 void e1000_free_all_tx_resources(struct e1000_adapter *adapter)
1909 int i;
1911 for (i = 0; i < adapter->num_tx_queues; i++)
1912 e1000_free_tx_resources(adapter, &adapter->tx_ring[i]);
1915 static void e1000_unmap_and_free_tx_resource(struct e1000_adapter *adapter,
1916 struct e1000_buffer *buffer_info)
1918 if (buffer_info->dma) {
1919 if (buffer_info->mapped_as_page)
1920 dma_unmap_page(&adapter->pdev->dev, buffer_info->dma,
1921 buffer_info->length, DMA_TO_DEVICE);
1922 else
1923 dma_unmap_single(&adapter->pdev->dev, buffer_info->dma,
1924 buffer_info->length,
1925 DMA_TO_DEVICE);
1926 buffer_info->dma = 0;
1928 if (buffer_info->skb) {
1929 dev_kfree_skb_any(buffer_info->skb);
1930 buffer_info->skb = NULL;
1932 buffer_info->time_stamp = 0;
1933 /* buffer_info must be completely set up in the transmit path */
1937 * e1000_clean_tx_ring - Free Tx Buffers
1938 * @adapter: board private structure
1939 * @tx_ring: ring to be cleaned
1942 static void e1000_clean_tx_ring(struct e1000_adapter *adapter,
1943 struct e1000_tx_ring *tx_ring)
1945 struct e1000_hw *hw = &adapter->hw;
1946 struct e1000_buffer *buffer_info;
1947 unsigned long size;
1948 unsigned int i;
1950 /* Free all the Tx ring sk_buffs */
1952 for (i = 0; i < tx_ring->count; i++) {
1953 buffer_info = &tx_ring->buffer_info[i];
1954 e1000_unmap_and_free_tx_resource(adapter, buffer_info);
1957 size = sizeof(struct e1000_buffer) * tx_ring->count;
1958 memset(tx_ring->buffer_info, 0, size);
1960 /* Zero out the descriptor ring */
1962 memset(tx_ring->desc, 0, tx_ring->size);
1964 tx_ring->next_to_use = 0;
1965 tx_ring->next_to_clean = 0;
1966 tx_ring->last_tx_tso = 0;
1968 writel(0, hw->hw_addr + tx_ring->tdh);
1969 writel(0, hw->hw_addr + tx_ring->tdt);
1973 * e1000_clean_all_tx_rings - Free Tx Buffers for all queues
1974 * @adapter: board private structure
1977 static void e1000_clean_all_tx_rings(struct e1000_adapter *adapter)
1979 int i;
1981 for (i = 0; i < adapter->num_tx_queues; i++)
1982 e1000_clean_tx_ring(adapter, &adapter->tx_ring[i]);
1986 * e1000_free_rx_resources - Free Rx Resources
1987 * @adapter: board private structure
1988 * @rx_ring: ring to clean the resources from
1990 * Free all receive software resources
1993 static void e1000_free_rx_resources(struct e1000_adapter *adapter,
1994 struct e1000_rx_ring *rx_ring)
1996 struct pci_dev *pdev = adapter->pdev;
1998 e1000_clean_rx_ring(adapter, rx_ring);
2000 vfree(rx_ring->buffer_info);
2001 rx_ring->buffer_info = NULL;
2003 dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
2004 rx_ring->dma);
2006 rx_ring->desc = NULL;
2010 * e1000_free_all_rx_resources - Free Rx Resources for All Queues
2011 * @adapter: board private structure
2013 * Free all receive software resources
2016 void e1000_free_all_rx_resources(struct e1000_adapter *adapter)
2018 int i;
2020 for (i = 0; i < adapter->num_rx_queues; i++)
2021 e1000_free_rx_resources(adapter, &adapter->rx_ring[i]);
2025 * e1000_clean_rx_ring - Free Rx Buffers per Queue
2026 * @adapter: board private structure
2027 * @rx_ring: ring to free buffers from
2030 static void e1000_clean_rx_ring(struct e1000_adapter *adapter,
2031 struct e1000_rx_ring *rx_ring)
2033 struct e1000_hw *hw = &adapter->hw;
2034 struct e1000_buffer *buffer_info;
2035 struct pci_dev *pdev = adapter->pdev;
2036 unsigned long size;
2037 unsigned int i;
2039 /* Free all the Rx ring sk_buffs */
2040 for (i = 0; i < rx_ring->count; i++) {
2041 buffer_info = &rx_ring->buffer_info[i];
2042 if (buffer_info->dma &&
2043 adapter->clean_rx == e1000_clean_rx_irq) {
2044 dma_unmap_single(&pdev->dev, buffer_info->dma,
2045 buffer_info->length,
2046 DMA_FROM_DEVICE);
2047 } else if (buffer_info->dma &&
2048 adapter->clean_rx == e1000_clean_jumbo_rx_irq) {
2049 dma_unmap_page(&pdev->dev, buffer_info->dma,
2050 buffer_info->length,
2051 DMA_FROM_DEVICE);
2054 buffer_info->dma = 0;
2055 if (buffer_info->page) {
2056 put_page(buffer_info->page);
2057 buffer_info->page = NULL;
2059 if (buffer_info->skb) {
2060 dev_kfree_skb(buffer_info->skb);
2061 buffer_info->skb = NULL;
2065 /* there also may be some cached data from a chained receive */
2066 if (rx_ring->rx_skb_top) {
2067 dev_kfree_skb(rx_ring->rx_skb_top);
2068 rx_ring->rx_skb_top = NULL;
2071 size = sizeof(struct e1000_buffer) * rx_ring->count;
2072 memset(rx_ring->buffer_info, 0, size);
2074 /* Zero out the descriptor ring */
2075 memset(rx_ring->desc, 0, rx_ring->size);
2077 rx_ring->next_to_clean = 0;
2078 rx_ring->next_to_use = 0;
2080 writel(0, hw->hw_addr + rx_ring->rdh);
2081 writel(0, hw->hw_addr + rx_ring->rdt);
2085 * e1000_clean_all_rx_rings - Free Rx Buffers for all queues
2086 * @adapter: board private structure
2089 static void e1000_clean_all_rx_rings(struct e1000_adapter *adapter)
2091 int i;
2093 for (i = 0; i < adapter->num_rx_queues; i++)
2094 e1000_clean_rx_ring(adapter, &adapter->rx_ring[i]);
2097 /* The 82542 2.0 (revision 2) needs to have the receive unit in reset
2098 * and memory write and invalidate disabled for certain operations
2100 static void e1000_enter_82542_rst(struct e1000_adapter *adapter)
2102 struct e1000_hw *hw = &adapter->hw;
2103 struct net_device *netdev = adapter->netdev;
2104 u32 rctl;
2106 e1000_pci_clear_mwi(hw);
2108 rctl = er32(RCTL);
2109 rctl |= E1000_RCTL_RST;
2110 ew32(RCTL, rctl);
2111 E1000_WRITE_FLUSH();
2112 mdelay(5);
2114 if (netif_running(netdev))
2115 e1000_clean_all_rx_rings(adapter);
2118 static void e1000_leave_82542_rst(struct e1000_adapter *adapter)
2120 struct e1000_hw *hw = &adapter->hw;
2121 struct net_device *netdev = adapter->netdev;
2122 u32 rctl;
2124 rctl = er32(RCTL);
2125 rctl &= ~E1000_RCTL_RST;
2126 ew32(RCTL, rctl);
2127 E1000_WRITE_FLUSH();
2128 mdelay(5);
2130 if (hw->pci_cmd_word & PCI_COMMAND_INVALIDATE)
2131 e1000_pci_set_mwi(hw);
2133 if (netif_running(netdev)) {
2134 /* No need to loop, because 82542 supports only 1 queue */
2135 struct e1000_rx_ring *ring = &adapter->rx_ring[0];
2136 e1000_configure_rx(adapter);
2137 adapter->alloc_rx_buf(adapter, ring, E1000_DESC_UNUSED(ring));
2142 * e1000_set_mac - Change the Ethernet Address of the NIC
2143 * @netdev: network interface device structure
2144 * @p: pointer to an address structure
2146 * Returns 0 on success, negative on failure
2149 static int e1000_set_mac(struct net_device *netdev, void *p)
2151 struct e1000_adapter *adapter = netdev_priv(netdev);
2152 struct e1000_hw *hw = &adapter->hw;
2153 struct sockaddr *addr = p;
2155 if (!is_valid_ether_addr(addr->sa_data))
2156 return -EADDRNOTAVAIL;
2158 /* 82542 2.0 needs to be in reset to write receive address registers */
2160 if (hw->mac_type == e1000_82542_rev2_0)
2161 e1000_enter_82542_rst(adapter);
2163 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2164 memcpy(hw->mac_addr, addr->sa_data, netdev->addr_len);
2166 e1000_rar_set(hw, hw->mac_addr, 0);
2168 if (hw->mac_type == e1000_82542_rev2_0)
2169 e1000_leave_82542_rst(adapter);
2171 return 0;
2175 * e1000_set_rx_mode - Secondary Unicast, Multicast and Promiscuous mode set
2176 * @netdev: network interface device structure
2178 * The set_rx_mode entry point is called whenever the unicast or multicast
2179 * address lists or the network interface flags are updated. This routine is
2180 * responsible for configuring the hardware for proper unicast, multicast,
2181 * promiscuous mode, and all-multi behavior.
2184 static void e1000_set_rx_mode(struct net_device *netdev)
2186 struct e1000_adapter *adapter = netdev_priv(netdev);
2187 struct e1000_hw *hw = &adapter->hw;
2188 struct netdev_hw_addr *ha;
2189 bool use_uc = false;
2190 u32 rctl;
2191 u32 hash_value;
2192 int i, rar_entries = E1000_RAR_ENTRIES;
2193 int mta_reg_count = E1000_NUM_MTA_REGISTERS;
2194 u32 *mcarray = kcalloc(mta_reg_count, sizeof(u32), GFP_ATOMIC);
2196 if (!mcarray) {
2197 e_err(probe, "memory allocation failed\n");
2198 return;
2201 /* Check for Promiscuous and All Multicast modes */
2203 rctl = er32(RCTL);
2205 if (netdev->flags & IFF_PROMISC) {
2206 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2207 rctl &= ~E1000_RCTL_VFE;
2208 } else {
2209 if (netdev->flags & IFF_ALLMULTI)
2210 rctl |= E1000_RCTL_MPE;
2211 else
2212 rctl &= ~E1000_RCTL_MPE;
2213 /* Enable VLAN filter if there is a VLAN */
2214 if (adapter->vlgrp)
2215 rctl |= E1000_RCTL_VFE;
2218 if (netdev_uc_count(netdev) > rar_entries - 1) {
2219 rctl |= E1000_RCTL_UPE;
2220 } else if (!(netdev->flags & IFF_PROMISC)) {
2221 rctl &= ~E1000_RCTL_UPE;
2222 use_uc = true;
2225 ew32(RCTL, rctl);
2227 /* 82542 2.0 needs to be in reset to write receive address registers */
2229 if (hw->mac_type == e1000_82542_rev2_0)
2230 e1000_enter_82542_rst(adapter);
2232 /* load the first 14 addresses into the exact filters 1-14. Unicast
2233 * addresses take precedence to avoid disabling unicast filtering
2234 * when possible.
2236 * RAR 0 is used for the station MAC address
2237 * if there are not 14 addresses, go ahead and clear the filters
2239 i = 1;
2240 if (use_uc)
2241 netdev_for_each_uc_addr(ha, netdev) {
2242 if (i == rar_entries)
2243 break;
2244 e1000_rar_set(hw, ha->addr, i++);
2247 netdev_for_each_mc_addr(ha, netdev) {
2248 if (i == rar_entries) {
2249 /* load any remaining addresses into the hash table */
2250 u32 hash_reg, hash_bit, mta;
2251 hash_value = e1000_hash_mc_addr(hw, ha->addr);
2252 hash_reg = (hash_value >> 5) & 0x7F;
2253 hash_bit = hash_value & 0x1F;
2254 mta = (1 << hash_bit);
2255 mcarray[hash_reg] |= mta;
2256 } else {
2257 e1000_rar_set(hw, ha->addr, i++);
2261 for (; i < rar_entries; i++) {
2262 E1000_WRITE_REG_ARRAY(hw, RA, i << 1, 0);
2263 E1000_WRITE_FLUSH();
2264 E1000_WRITE_REG_ARRAY(hw, RA, (i << 1) + 1, 0);
2265 E1000_WRITE_FLUSH();
2268 /* write the hash table completely, write from bottom to avoid
2269 * both stupid write combining chipsets, and flushing each write */
2270 for (i = mta_reg_count - 1; i >= 0 ; i--) {
2272 * If we are on an 82544 has an errata where writing odd
2273 * offsets overwrites the previous even offset, but writing
2274 * backwards over the range solves the issue by always
2275 * writing the odd offset first
2277 E1000_WRITE_REG_ARRAY(hw, MTA, i, mcarray[i]);
2279 E1000_WRITE_FLUSH();
2281 if (hw->mac_type == e1000_82542_rev2_0)
2282 e1000_leave_82542_rst(adapter);
2284 kfree(mcarray);
2287 /* Need to wait a few seconds after link up to get diagnostic information from
2288 * the phy */
2290 static void e1000_update_phy_info(unsigned long data)
2292 struct e1000_adapter *adapter = (struct e1000_adapter *)data;
2293 schedule_work(&adapter->phy_info_task);
2296 static void e1000_update_phy_info_task(struct work_struct *work)
2298 struct e1000_adapter *adapter = container_of(work,
2299 struct e1000_adapter,
2300 phy_info_task);
2301 struct e1000_hw *hw = &adapter->hw;
2303 rtnl_lock();
2304 e1000_phy_get_info(hw, &adapter->phy_info);
2305 rtnl_unlock();
2309 * e1000_82547_tx_fifo_stall - Timer Call-back
2310 * @data: pointer to adapter cast into an unsigned long
2312 static void e1000_82547_tx_fifo_stall(unsigned long data)
2314 struct e1000_adapter *adapter = (struct e1000_adapter *)data;
2315 schedule_work(&adapter->fifo_stall_task);
2319 * e1000_82547_tx_fifo_stall_task - task to complete work
2320 * @work: work struct contained inside adapter struct
2322 static void e1000_82547_tx_fifo_stall_task(struct work_struct *work)
2324 struct e1000_adapter *adapter = container_of(work,
2325 struct e1000_adapter,
2326 fifo_stall_task);
2327 struct e1000_hw *hw = &adapter->hw;
2328 struct net_device *netdev = adapter->netdev;
2329 u32 tctl;
2331 rtnl_lock();
2332 if (atomic_read(&adapter->tx_fifo_stall)) {
2333 if ((er32(TDT) == er32(TDH)) &&
2334 (er32(TDFT) == er32(TDFH)) &&
2335 (er32(TDFTS) == er32(TDFHS))) {
2336 tctl = er32(TCTL);
2337 ew32(TCTL, tctl & ~E1000_TCTL_EN);
2338 ew32(TDFT, adapter->tx_head_addr);
2339 ew32(TDFH, adapter->tx_head_addr);
2340 ew32(TDFTS, adapter->tx_head_addr);
2341 ew32(TDFHS, adapter->tx_head_addr);
2342 ew32(TCTL, tctl);
2343 E1000_WRITE_FLUSH();
2345 adapter->tx_fifo_head = 0;
2346 atomic_set(&adapter->tx_fifo_stall, 0);
2347 netif_wake_queue(netdev);
2348 } else if (!test_bit(__E1000_DOWN, &adapter->flags)) {
2349 mod_timer(&adapter->tx_fifo_stall_timer, jiffies + 1);
2352 rtnl_unlock();
2355 bool e1000_has_link(struct e1000_adapter *adapter)
2357 struct e1000_hw *hw = &adapter->hw;
2358 bool link_active = false;
2360 /* get_link_status is set on LSC (link status) interrupt or
2361 * rx sequence error interrupt. get_link_status will stay
2362 * false until the e1000_check_for_link establishes link
2363 * for copper adapters ONLY
2365 switch (hw->media_type) {
2366 case e1000_media_type_copper:
2367 if (hw->get_link_status) {
2368 e1000_check_for_link(hw);
2369 link_active = !hw->get_link_status;
2370 } else {
2371 link_active = true;
2373 break;
2374 case e1000_media_type_fiber:
2375 e1000_check_for_link(hw);
2376 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
2377 break;
2378 case e1000_media_type_internal_serdes:
2379 e1000_check_for_link(hw);
2380 link_active = hw->serdes_has_link;
2381 break;
2382 default:
2383 break;
2386 return link_active;
2390 * e1000_watchdog - Timer Call-back
2391 * @data: pointer to adapter cast into an unsigned long
2393 static void e1000_watchdog(unsigned long data)
2395 struct e1000_adapter *adapter = (struct e1000_adapter *)data;
2396 struct e1000_hw *hw = &adapter->hw;
2397 struct net_device *netdev = adapter->netdev;
2398 struct e1000_tx_ring *txdr = adapter->tx_ring;
2399 u32 link, tctl;
2401 link = e1000_has_link(adapter);
2402 if ((netif_carrier_ok(netdev)) && link)
2403 goto link_up;
2405 if (link) {
2406 if (!netif_carrier_ok(netdev)) {
2407 u32 ctrl;
2408 bool txb2b = true;
2409 /* update snapshot of PHY registers on LSC */
2410 e1000_get_speed_and_duplex(hw,
2411 &adapter->link_speed,
2412 &adapter->link_duplex);
2414 ctrl = er32(CTRL);
2415 pr_info("%s NIC Link is Up %d Mbps %s, "
2416 "Flow Control: %s\n",
2417 netdev->name,
2418 adapter->link_speed,
2419 adapter->link_duplex == FULL_DUPLEX ?
2420 "Full Duplex" : "Half Duplex",
2421 ((ctrl & E1000_CTRL_TFCE) && (ctrl &
2422 E1000_CTRL_RFCE)) ? "RX/TX" : ((ctrl &
2423 E1000_CTRL_RFCE) ? "RX" : ((ctrl &
2424 E1000_CTRL_TFCE) ? "TX" : "None")));
2426 /* adjust timeout factor according to speed/duplex */
2427 adapter->tx_timeout_factor = 1;
2428 switch (adapter->link_speed) {
2429 case SPEED_10:
2430 txb2b = false;
2431 adapter->tx_timeout_factor = 16;
2432 break;
2433 case SPEED_100:
2434 txb2b = false;
2435 /* maybe add some timeout factor ? */
2436 break;
2439 /* enable transmits in the hardware */
2440 tctl = er32(TCTL);
2441 tctl |= E1000_TCTL_EN;
2442 ew32(TCTL, tctl);
2444 netif_carrier_on(netdev);
2445 if (!test_bit(__E1000_DOWN, &adapter->flags))
2446 mod_timer(&adapter->phy_info_timer,
2447 round_jiffies(jiffies + 2 * HZ));
2448 adapter->smartspeed = 0;
2450 } else {
2451 if (netif_carrier_ok(netdev)) {
2452 adapter->link_speed = 0;
2453 adapter->link_duplex = 0;
2454 pr_info("%s NIC Link is Down\n",
2455 netdev->name);
2456 netif_carrier_off(netdev);
2458 if (!test_bit(__E1000_DOWN, &adapter->flags))
2459 mod_timer(&adapter->phy_info_timer,
2460 round_jiffies(jiffies + 2 * HZ));
2463 e1000_smartspeed(adapter);
2466 link_up:
2467 e1000_update_stats(adapter);
2469 hw->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
2470 adapter->tpt_old = adapter->stats.tpt;
2471 hw->collision_delta = adapter->stats.colc - adapter->colc_old;
2472 adapter->colc_old = adapter->stats.colc;
2474 adapter->gorcl = adapter->stats.gorcl - adapter->gorcl_old;
2475 adapter->gorcl_old = adapter->stats.gorcl;
2476 adapter->gotcl = adapter->stats.gotcl - adapter->gotcl_old;
2477 adapter->gotcl_old = adapter->stats.gotcl;
2479 e1000_update_adaptive(hw);
2481 if (!netif_carrier_ok(netdev)) {
2482 if (E1000_DESC_UNUSED(txdr) + 1 < txdr->count) {
2483 /* We've lost link, so the controller stops DMA,
2484 * but we've got queued Tx work that's never going
2485 * to get done, so reset controller to flush Tx.
2486 * (Do the reset outside of interrupt context). */
2487 adapter->tx_timeout_count++;
2488 schedule_work(&adapter->reset_task);
2489 /* return immediately since reset is imminent */
2490 return;
2494 /* Simple mode for Interrupt Throttle Rate (ITR) */
2495 if (hw->mac_type >= e1000_82540 && adapter->itr_setting == 4) {
2497 * Symmetric Tx/Rx gets a reduced ITR=2000;
2498 * Total asymmetrical Tx or Rx gets ITR=8000;
2499 * everyone else is between 2000-8000.
2501 u32 goc = (adapter->gotcl + adapter->gorcl) / 10000;
2502 u32 dif = (adapter->gotcl > adapter->gorcl ?
2503 adapter->gotcl - adapter->gorcl :
2504 adapter->gorcl - adapter->gotcl) / 10000;
2505 u32 itr = goc > 0 ? (dif * 6000 / goc + 2000) : 8000;
2507 ew32(ITR, 1000000000 / (itr * 256));
2510 /* Cause software interrupt to ensure rx ring is cleaned */
2511 ew32(ICS, E1000_ICS_RXDMT0);
2513 /* Force detection of hung controller every watchdog period */
2514 adapter->detect_tx_hung = true;
2516 /* Reset the timer */
2517 if (!test_bit(__E1000_DOWN, &adapter->flags))
2518 mod_timer(&adapter->watchdog_timer,
2519 round_jiffies(jiffies + 2 * HZ));
2522 enum latency_range {
2523 lowest_latency = 0,
2524 low_latency = 1,
2525 bulk_latency = 2,
2526 latency_invalid = 255
2530 * e1000_update_itr - update the dynamic ITR value based on statistics
2531 * @adapter: pointer to adapter
2532 * @itr_setting: current adapter->itr
2533 * @packets: the number of packets during this measurement interval
2534 * @bytes: the number of bytes during this measurement interval
2536 * Stores a new ITR value based on packets and byte
2537 * counts during the last interrupt. The advantage of per interrupt
2538 * computation is faster updates and more accurate ITR for the current
2539 * traffic pattern. Constants in this function were computed
2540 * based on theoretical maximum wire speed and thresholds were set based
2541 * on testing data as well as attempting to minimize response time
2542 * while increasing bulk throughput.
2543 * this functionality is controlled by the InterruptThrottleRate module
2544 * parameter (see e1000_param.c)
2546 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
2547 u16 itr_setting, int packets, int bytes)
2549 unsigned int retval = itr_setting;
2550 struct e1000_hw *hw = &adapter->hw;
2552 if (unlikely(hw->mac_type < e1000_82540))
2553 goto update_itr_done;
2555 if (packets == 0)
2556 goto update_itr_done;
2558 switch (itr_setting) {
2559 case lowest_latency:
2560 /* jumbo frames get bulk treatment*/
2561 if (bytes/packets > 8000)
2562 retval = bulk_latency;
2563 else if ((packets < 5) && (bytes > 512))
2564 retval = low_latency;
2565 break;
2566 case low_latency: /* 50 usec aka 20000 ints/s */
2567 if (bytes > 10000) {
2568 /* jumbo frames need bulk latency setting */
2569 if (bytes/packets > 8000)
2570 retval = bulk_latency;
2571 else if ((packets < 10) || ((bytes/packets) > 1200))
2572 retval = bulk_latency;
2573 else if ((packets > 35))
2574 retval = lowest_latency;
2575 } else if (bytes/packets > 2000)
2576 retval = bulk_latency;
2577 else if (packets <= 2 && bytes < 512)
2578 retval = lowest_latency;
2579 break;
2580 case bulk_latency: /* 250 usec aka 4000 ints/s */
2581 if (bytes > 25000) {
2582 if (packets > 35)
2583 retval = low_latency;
2584 } else if (bytes < 6000) {
2585 retval = low_latency;
2587 break;
2590 update_itr_done:
2591 return retval;
2594 static void e1000_set_itr(struct e1000_adapter *adapter)
2596 struct e1000_hw *hw = &adapter->hw;
2597 u16 current_itr;
2598 u32 new_itr = adapter->itr;
2600 if (unlikely(hw->mac_type < e1000_82540))
2601 return;
2603 /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2604 if (unlikely(adapter->link_speed != SPEED_1000)) {
2605 current_itr = 0;
2606 new_itr = 4000;
2607 goto set_itr_now;
2610 adapter->tx_itr = e1000_update_itr(adapter,
2611 adapter->tx_itr,
2612 adapter->total_tx_packets,
2613 adapter->total_tx_bytes);
2614 /* conservative mode (itr 3) eliminates the lowest_latency setting */
2615 if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2616 adapter->tx_itr = low_latency;
2618 adapter->rx_itr = e1000_update_itr(adapter,
2619 adapter->rx_itr,
2620 adapter->total_rx_packets,
2621 adapter->total_rx_bytes);
2622 /* conservative mode (itr 3) eliminates the lowest_latency setting */
2623 if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2624 adapter->rx_itr = low_latency;
2626 current_itr = max(adapter->rx_itr, adapter->tx_itr);
2628 switch (current_itr) {
2629 /* counts and packets in update_itr are dependent on these numbers */
2630 case lowest_latency:
2631 new_itr = 70000;
2632 break;
2633 case low_latency:
2634 new_itr = 20000; /* aka hwitr = ~200 */
2635 break;
2636 case bulk_latency:
2637 new_itr = 4000;
2638 break;
2639 default:
2640 break;
2643 set_itr_now:
2644 if (new_itr != adapter->itr) {
2645 /* this attempts to bias the interrupt rate towards Bulk
2646 * by adding intermediate steps when interrupt rate is
2647 * increasing */
2648 new_itr = new_itr > adapter->itr ?
2649 min(adapter->itr + (new_itr >> 2), new_itr) :
2650 new_itr;
2651 adapter->itr = new_itr;
2652 ew32(ITR, 1000000000 / (new_itr * 256));
2656 #define E1000_TX_FLAGS_CSUM 0x00000001
2657 #define E1000_TX_FLAGS_VLAN 0x00000002
2658 #define E1000_TX_FLAGS_TSO 0x00000004
2659 #define E1000_TX_FLAGS_IPV4 0x00000008
2660 #define E1000_TX_FLAGS_VLAN_MASK 0xffff0000
2661 #define E1000_TX_FLAGS_VLAN_SHIFT 16
2663 static int e1000_tso(struct e1000_adapter *adapter,
2664 struct e1000_tx_ring *tx_ring, struct sk_buff *skb)
2666 struct e1000_context_desc *context_desc;
2667 struct e1000_buffer *buffer_info;
2668 unsigned int i;
2669 u32 cmd_length = 0;
2670 u16 ipcse = 0, tucse, mss;
2671 u8 ipcss, ipcso, tucss, tucso, hdr_len;
2672 int err;
2674 if (skb_is_gso(skb)) {
2675 if (skb_header_cloned(skb)) {
2676 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2677 if (err)
2678 return err;
2681 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
2682 mss = skb_shinfo(skb)->gso_size;
2683 if (skb->protocol == htons(ETH_P_IP)) {
2684 struct iphdr *iph = ip_hdr(skb);
2685 iph->tot_len = 0;
2686 iph->check = 0;
2687 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
2688 iph->daddr, 0,
2689 IPPROTO_TCP,
2691 cmd_length = E1000_TXD_CMD_IP;
2692 ipcse = skb_transport_offset(skb) - 1;
2693 } else if (skb->protocol == htons(ETH_P_IPV6)) {
2694 ipv6_hdr(skb)->payload_len = 0;
2695 tcp_hdr(skb)->check =
2696 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
2697 &ipv6_hdr(skb)->daddr,
2698 0, IPPROTO_TCP, 0);
2699 ipcse = 0;
2701 ipcss = skb_network_offset(skb);
2702 ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
2703 tucss = skb_transport_offset(skb);
2704 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
2705 tucse = 0;
2707 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
2708 E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
2710 i = tx_ring->next_to_use;
2711 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
2712 buffer_info = &tx_ring->buffer_info[i];
2714 context_desc->lower_setup.ip_fields.ipcss = ipcss;
2715 context_desc->lower_setup.ip_fields.ipcso = ipcso;
2716 context_desc->lower_setup.ip_fields.ipcse = cpu_to_le16(ipcse);
2717 context_desc->upper_setup.tcp_fields.tucss = tucss;
2718 context_desc->upper_setup.tcp_fields.tucso = tucso;
2719 context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
2720 context_desc->tcp_seg_setup.fields.mss = cpu_to_le16(mss);
2721 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
2722 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
2724 buffer_info->time_stamp = jiffies;
2725 buffer_info->next_to_watch = i;
2727 if (++i == tx_ring->count) i = 0;
2728 tx_ring->next_to_use = i;
2730 return true;
2732 return false;
2735 static bool e1000_tx_csum(struct e1000_adapter *adapter,
2736 struct e1000_tx_ring *tx_ring, struct sk_buff *skb)
2738 struct e1000_context_desc *context_desc;
2739 struct e1000_buffer *buffer_info;
2740 unsigned int i;
2741 u8 css;
2742 u32 cmd_len = E1000_TXD_CMD_DEXT;
2744 if (skb->ip_summed != CHECKSUM_PARTIAL)
2745 return false;
2747 switch (skb->protocol) {
2748 case cpu_to_be16(ETH_P_IP):
2749 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
2750 cmd_len |= E1000_TXD_CMD_TCP;
2751 break;
2752 case cpu_to_be16(ETH_P_IPV6):
2753 /* XXX not handling all IPV6 headers */
2754 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
2755 cmd_len |= E1000_TXD_CMD_TCP;
2756 break;
2757 default:
2758 if (unlikely(net_ratelimit()))
2759 e_warn(drv, "checksum_partial proto=%x!\n",
2760 skb->protocol);
2761 break;
2764 css = skb_checksum_start_offset(skb);
2766 i = tx_ring->next_to_use;
2767 buffer_info = &tx_ring->buffer_info[i];
2768 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
2770 context_desc->lower_setup.ip_config = 0;
2771 context_desc->upper_setup.tcp_fields.tucss = css;
2772 context_desc->upper_setup.tcp_fields.tucso =
2773 css + skb->csum_offset;
2774 context_desc->upper_setup.tcp_fields.tucse = 0;
2775 context_desc->tcp_seg_setup.data = 0;
2776 context_desc->cmd_and_length = cpu_to_le32(cmd_len);
2778 buffer_info->time_stamp = jiffies;
2779 buffer_info->next_to_watch = i;
2781 if (unlikely(++i == tx_ring->count)) i = 0;
2782 tx_ring->next_to_use = i;
2784 return true;
2787 #define E1000_MAX_TXD_PWR 12
2788 #define E1000_MAX_DATA_PER_TXD (1<<E1000_MAX_TXD_PWR)
2790 static int e1000_tx_map(struct e1000_adapter *adapter,
2791 struct e1000_tx_ring *tx_ring,
2792 struct sk_buff *skb, unsigned int first,
2793 unsigned int max_per_txd, unsigned int nr_frags,
2794 unsigned int mss)
2796 struct e1000_hw *hw = &adapter->hw;
2797 struct pci_dev *pdev = adapter->pdev;
2798 struct e1000_buffer *buffer_info;
2799 unsigned int len = skb_headlen(skb);
2800 unsigned int offset = 0, size, count = 0, i;
2801 unsigned int f;
2803 i = tx_ring->next_to_use;
2805 while (len) {
2806 buffer_info = &tx_ring->buffer_info[i];
2807 size = min(len, max_per_txd);
2808 /* Workaround for Controller erratum --
2809 * descriptor for non-tso packet in a linear SKB that follows a
2810 * tso gets written back prematurely before the data is fully
2811 * DMA'd to the controller */
2812 if (!skb->data_len && tx_ring->last_tx_tso &&
2813 !skb_is_gso(skb)) {
2814 tx_ring->last_tx_tso = 0;
2815 size -= 4;
2818 /* Workaround for premature desc write-backs
2819 * in TSO mode. Append 4-byte sentinel desc */
2820 if (unlikely(mss && !nr_frags && size == len && size > 8))
2821 size -= 4;
2822 /* work-around for errata 10 and it applies
2823 * to all controllers in PCI-X mode
2824 * The fix is to make sure that the first descriptor of a
2825 * packet is smaller than 2048 - 16 - 16 (or 2016) bytes
2827 if (unlikely((hw->bus_type == e1000_bus_type_pcix) &&
2828 (size > 2015) && count == 0))
2829 size = 2015;
2831 /* Workaround for potential 82544 hang in PCI-X. Avoid
2832 * terminating buffers within evenly-aligned dwords. */
2833 if (unlikely(adapter->pcix_82544 &&
2834 !((unsigned long)(skb->data + offset + size - 1) & 4) &&
2835 size > 4))
2836 size -= 4;
2838 buffer_info->length = size;
2839 /* set time_stamp *before* dma to help avoid a possible race */
2840 buffer_info->time_stamp = jiffies;
2841 buffer_info->mapped_as_page = false;
2842 buffer_info->dma = dma_map_single(&pdev->dev,
2843 skb->data + offset,
2844 size, DMA_TO_DEVICE);
2845 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2846 goto dma_error;
2847 buffer_info->next_to_watch = i;
2849 len -= size;
2850 offset += size;
2851 count++;
2852 if (len) {
2853 i++;
2854 if (unlikely(i == tx_ring->count))
2855 i = 0;
2859 for (f = 0; f < nr_frags; f++) {
2860 struct skb_frag_struct *frag;
2862 frag = &skb_shinfo(skb)->frags[f];
2863 len = frag->size;
2864 offset = frag->page_offset;
2866 while (len) {
2867 i++;
2868 if (unlikely(i == tx_ring->count))
2869 i = 0;
2871 buffer_info = &tx_ring->buffer_info[i];
2872 size = min(len, max_per_txd);
2873 /* Workaround for premature desc write-backs
2874 * in TSO mode. Append 4-byte sentinel desc */
2875 if (unlikely(mss && f == (nr_frags-1) && size == len && size > 8))
2876 size -= 4;
2877 /* Workaround for potential 82544 hang in PCI-X.
2878 * Avoid terminating buffers within evenly-aligned
2879 * dwords. */
2880 if (unlikely(adapter->pcix_82544 &&
2881 !((unsigned long)(page_to_phys(frag->page) + offset
2882 + size - 1) & 4) &&
2883 size > 4))
2884 size -= 4;
2886 buffer_info->length = size;
2887 buffer_info->time_stamp = jiffies;
2888 buffer_info->mapped_as_page = true;
2889 buffer_info->dma = dma_map_page(&pdev->dev, frag->page,
2890 offset, size,
2891 DMA_TO_DEVICE);
2892 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2893 goto dma_error;
2894 buffer_info->next_to_watch = i;
2896 len -= size;
2897 offset += size;
2898 count++;
2902 tx_ring->buffer_info[i].skb = skb;
2903 tx_ring->buffer_info[first].next_to_watch = i;
2905 return count;
2907 dma_error:
2908 dev_err(&pdev->dev, "TX DMA map failed\n");
2909 buffer_info->dma = 0;
2910 if (count)
2911 count--;
2913 while (count--) {
2914 if (i==0)
2915 i += tx_ring->count;
2916 i--;
2917 buffer_info = &tx_ring->buffer_info[i];
2918 e1000_unmap_and_free_tx_resource(adapter, buffer_info);
2921 return 0;
2924 static void e1000_tx_queue(struct e1000_adapter *adapter,
2925 struct e1000_tx_ring *tx_ring, int tx_flags,
2926 int count)
2928 struct e1000_hw *hw = &adapter->hw;
2929 struct e1000_tx_desc *tx_desc = NULL;
2930 struct e1000_buffer *buffer_info;
2931 u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
2932 unsigned int i;
2934 if (likely(tx_flags & E1000_TX_FLAGS_TSO)) {
2935 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
2936 E1000_TXD_CMD_TSE;
2937 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
2939 if (likely(tx_flags & E1000_TX_FLAGS_IPV4))
2940 txd_upper |= E1000_TXD_POPTS_IXSM << 8;
2943 if (likely(tx_flags & E1000_TX_FLAGS_CSUM)) {
2944 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
2945 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
2948 if (unlikely(tx_flags & E1000_TX_FLAGS_VLAN)) {
2949 txd_lower |= E1000_TXD_CMD_VLE;
2950 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
2953 i = tx_ring->next_to_use;
2955 while (count--) {
2956 buffer_info = &tx_ring->buffer_info[i];
2957 tx_desc = E1000_TX_DESC(*tx_ring, i);
2958 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
2959 tx_desc->lower.data =
2960 cpu_to_le32(txd_lower | buffer_info->length);
2961 tx_desc->upper.data = cpu_to_le32(txd_upper);
2962 if (unlikely(++i == tx_ring->count)) i = 0;
2965 tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
2967 /* Force memory writes to complete before letting h/w
2968 * know there are new descriptors to fetch. (Only
2969 * applicable for weak-ordered memory model archs,
2970 * such as IA-64). */
2971 wmb();
2973 tx_ring->next_to_use = i;
2974 writel(i, hw->hw_addr + tx_ring->tdt);
2975 /* we need this if more than one processor can write to our tail
2976 * at a time, it syncronizes IO on IA64/Altix systems */
2977 mmiowb();
2981 * 82547 workaround to avoid controller hang in half-duplex environment.
2982 * The workaround is to avoid queuing a large packet that would span
2983 * the internal Tx FIFO ring boundary by notifying the stack to resend
2984 * the packet at a later time. This gives the Tx FIFO an opportunity to
2985 * flush all packets. When that occurs, we reset the Tx FIFO pointers
2986 * to the beginning of the Tx FIFO.
2989 #define E1000_FIFO_HDR 0x10
2990 #define E1000_82547_PAD_LEN 0x3E0
2992 static int e1000_82547_fifo_workaround(struct e1000_adapter *adapter,
2993 struct sk_buff *skb)
2995 u32 fifo_space = adapter->tx_fifo_size - adapter->tx_fifo_head;
2996 u32 skb_fifo_len = skb->len + E1000_FIFO_HDR;
2998 skb_fifo_len = ALIGN(skb_fifo_len, E1000_FIFO_HDR);
3000 if (adapter->link_duplex != HALF_DUPLEX)
3001 goto no_fifo_stall_required;
3003 if (atomic_read(&adapter->tx_fifo_stall))
3004 return 1;
3006 if (skb_fifo_len >= (E1000_82547_PAD_LEN + fifo_space)) {
3007 atomic_set(&adapter->tx_fifo_stall, 1);
3008 return 1;
3011 no_fifo_stall_required:
3012 adapter->tx_fifo_head += skb_fifo_len;
3013 if (adapter->tx_fifo_head >= adapter->tx_fifo_size)
3014 adapter->tx_fifo_head -= adapter->tx_fifo_size;
3015 return 0;
3018 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
3020 struct e1000_adapter *adapter = netdev_priv(netdev);
3021 struct e1000_tx_ring *tx_ring = adapter->tx_ring;
3023 netif_stop_queue(netdev);
3024 /* Herbert's original patch had:
3025 * smp_mb__after_netif_stop_queue();
3026 * but since that doesn't exist yet, just open code it. */
3027 smp_mb();
3029 /* We need to check again in a case another CPU has just
3030 * made room available. */
3031 if (likely(E1000_DESC_UNUSED(tx_ring) < size))
3032 return -EBUSY;
3034 /* A reprieve! */
3035 netif_start_queue(netdev);
3036 ++adapter->restart_queue;
3037 return 0;
3040 static int e1000_maybe_stop_tx(struct net_device *netdev,
3041 struct e1000_tx_ring *tx_ring, int size)
3043 if (likely(E1000_DESC_UNUSED(tx_ring) >= size))
3044 return 0;
3045 return __e1000_maybe_stop_tx(netdev, size);
3048 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
3049 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
3050 struct net_device *netdev)
3052 struct e1000_adapter *adapter = netdev_priv(netdev);
3053 struct e1000_hw *hw = &adapter->hw;
3054 struct e1000_tx_ring *tx_ring;
3055 unsigned int first, max_per_txd = E1000_MAX_DATA_PER_TXD;
3056 unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
3057 unsigned int tx_flags = 0;
3058 unsigned int len = skb_headlen(skb);
3059 unsigned int nr_frags;
3060 unsigned int mss;
3061 int count = 0;
3062 int tso;
3063 unsigned int f;
3065 /* This goes back to the question of how to logically map a tx queue
3066 * to a flow. Right now, performance is impacted slightly negatively
3067 * if using multiple tx queues. If the stack breaks away from a
3068 * single qdisc implementation, we can look at this again. */
3069 tx_ring = adapter->tx_ring;
3071 if (unlikely(skb->len <= 0)) {
3072 dev_kfree_skb_any(skb);
3073 return NETDEV_TX_OK;
3076 mss = skb_shinfo(skb)->gso_size;
3077 /* The controller does a simple calculation to
3078 * make sure there is enough room in the FIFO before
3079 * initiating the DMA for each buffer. The calc is:
3080 * 4 = ceil(buffer len/mss). To make sure we don't
3081 * overrun the FIFO, adjust the max buffer len if mss
3082 * drops. */
3083 if (mss) {
3084 u8 hdr_len;
3085 max_per_txd = min(mss << 2, max_per_txd);
3086 max_txd_pwr = fls(max_per_txd) - 1;
3088 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
3089 if (skb->data_len && hdr_len == len) {
3090 switch (hw->mac_type) {
3091 unsigned int pull_size;
3092 case e1000_82544:
3093 /* Make sure we have room to chop off 4 bytes,
3094 * and that the end alignment will work out to
3095 * this hardware's requirements
3096 * NOTE: this is a TSO only workaround
3097 * if end byte alignment not correct move us
3098 * into the next dword */
3099 if ((unsigned long)(skb_tail_pointer(skb) - 1) & 4)
3100 break;
3101 /* fall through */
3102 pull_size = min((unsigned int)4, skb->data_len);
3103 if (!__pskb_pull_tail(skb, pull_size)) {
3104 e_err(drv, "__pskb_pull_tail "
3105 "failed.\n");
3106 dev_kfree_skb_any(skb);
3107 return NETDEV_TX_OK;
3109 len = skb_headlen(skb);
3110 break;
3111 default:
3112 /* do nothing */
3113 break;
3118 /* reserve a descriptor for the offload context */
3119 if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
3120 count++;
3121 count++;
3123 /* Controller Erratum workaround */
3124 if (!skb->data_len && tx_ring->last_tx_tso && !skb_is_gso(skb))
3125 count++;
3127 count += TXD_USE_COUNT(len, max_txd_pwr);
3129 if (adapter->pcix_82544)
3130 count++;
3132 /* work-around for errata 10 and it applies to all controllers
3133 * in PCI-X mode, so add one more descriptor to the count
3135 if (unlikely((hw->bus_type == e1000_bus_type_pcix) &&
3136 (len > 2015)))
3137 count++;
3139 nr_frags = skb_shinfo(skb)->nr_frags;
3140 for (f = 0; f < nr_frags; f++)
3141 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
3142 max_txd_pwr);
3143 if (adapter->pcix_82544)
3144 count += nr_frags;
3146 /* need: count + 2 desc gap to keep tail from touching
3147 * head, otherwise try next time */
3148 if (unlikely(e1000_maybe_stop_tx(netdev, tx_ring, count + 2)))
3149 return NETDEV_TX_BUSY;
3151 if (unlikely(hw->mac_type == e1000_82547)) {
3152 if (unlikely(e1000_82547_fifo_workaround(adapter, skb))) {
3153 netif_stop_queue(netdev);
3154 if (!test_bit(__E1000_DOWN, &adapter->flags))
3155 mod_timer(&adapter->tx_fifo_stall_timer,
3156 jiffies + 1);
3157 return NETDEV_TX_BUSY;
3161 if (unlikely(vlan_tx_tag_present(skb))) {
3162 tx_flags |= E1000_TX_FLAGS_VLAN;
3163 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
3166 first = tx_ring->next_to_use;
3168 tso = e1000_tso(adapter, tx_ring, skb);
3169 if (tso < 0) {
3170 dev_kfree_skb_any(skb);
3171 return NETDEV_TX_OK;
3174 if (likely(tso)) {
3175 if (likely(hw->mac_type != e1000_82544))
3176 tx_ring->last_tx_tso = 1;
3177 tx_flags |= E1000_TX_FLAGS_TSO;
3178 } else if (likely(e1000_tx_csum(adapter, tx_ring, skb)))
3179 tx_flags |= E1000_TX_FLAGS_CSUM;
3181 if (likely(skb->protocol == htons(ETH_P_IP)))
3182 tx_flags |= E1000_TX_FLAGS_IPV4;
3184 count = e1000_tx_map(adapter, tx_ring, skb, first, max_per_txd,
3185 nr_frags, mss);
3187 if (count) {
3188 e1000_tx_queue(adapter, tx_ring, tx_flags, count);
3189 /* Make sure there is space in the ring for the next send. */
3190 e1000_maybe_stop_tx(netdev, tx_ring, MAX_SKB_FRAGS + 2);
3192 } else {
3193 dev_kfree_skb_any(skb);
3194 tx_ring->buffer_info[first].time_stamp = 0;
3195 tx_ring->next_to_use = first;
3198 return NETDEV_TX_OK;
3202 * e1000_tx_timeout - Respond to a Tx Hang
3203 * @netdev: network interface device structure
3206 static void e1000_tx_timeout(struct net_device *netdev)
3208 struct e1000_adapter *adapter = netdev_priv(netdev);
3210 /* Do the reset outside of interrupt context */
3211 adapter->tx_timeout_count++;
3212 schedule_work(&adapter->reset_task);
3215 static void e1000_reset_task(struct work_struct *work)
3217 struct e1000_adapter *adapter =
3218 container_of(work, struct e1000_adapter, reset_task);
3220 e1000_reinit_safe(adapter);
3224 * e1000_get_stats - Get System Network Statistics
3225 * @netdev: network interface device structure
3227 * Returns the address of the device statistics structure.
3228 * The statistics are actually updated from the timer callback.
3231 static struct net_device_stats *e1000_get_stats(struct net_device *netdev)
3233 /* only return the current stats */
3234 return &netdev->stats;
3238 * e1000_change_mtu - Change the Maximum Transfer Unit
3239 * @netdev: network interface device structure
3240 * @new_mtu: new value for maximum frame size
3242 * Returns 0 on success, negative on failure
3245 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
3247 struct e1000_adapter *adapter = netdev_priv(netdev);
3248 struct e1000_hw *hw = &adapter->hw;
3249 int max_frame = new_mtu + ENET_HEADER_SIZE + ETHERNET_FCS_SIZE;
3251 if ((max_frame < MINIMUM_ETHERNET_FRAME_SIZE) ||
3252 (max_frame > MAX_JUMBO_FRAME_SIZE)) {
3253 e_err(probe, "Invalid MTU setting\n");
3254 return -EINVAL;
3257 /* Adapter-specific max frame size limits. */
3258 switch (hw->mac_type) {
3259 case e1000_undefined ... e1000_82542_rev2_1:
3260 if (max_frame > (ETH_FRAME_LEN + ETH_FCS_LEN)) {
3261 e_err(probe, "Jumbo Frames not supported.\n");
3262 return -EINVAL;
3264 break;
3265 default:
3266 /* Capable of supporting up to MAX_JUMBO_FRAME_SIZE limit. */
3267 break;
3270 while (test_and_set_bit(__E1000_RESETTING, &adapter->flags))
3271 msleep(1);
3272 /* e1000_down has a dependency on max_frame_size */
3273 hw->max_frame_size = max_frame;
3274 if (netif_running(netdev))
3275 e1000_down(adapter);
3277 /* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
3278 * means we reserve 2 more, this pushes us to allocate from the next
3279 * larger slab size.
3280 * i.e. RXBUFFER_2048 --> size-4096 slab
3281 * however with the new *_jumbo_rx* routines, jumbo receives will use
3282 * fragmented skbs */
3284 if (max_frame <= E1000_RXBUFFER_2048)
3285 adapter->rx_buffer_len = E1000_RXBUFFER_2048;
3286 else
3287 #if (PAGE_SIZE >= E1000_RXBUFFER_16384)
3288 adapter->rx_buffer_len = E1000_RXBUFFER_16384;
3289 #elif (PAGE_SIZE >= E1000_RXBUFFER_4096)
3290 adapter->rx_buffer_len = PAGE_SIZE;
3291 #endif
3293 /* adjust allocation if LPE protects us, and we aren't using SBP */
3294 if (!hw->tbi_compatibility_on &&
3295 ((max_frame == (ETH_FRAME_LEN + ETH_FCS_LEN)) ||
3296 (max_frame == MAXIMUM_ETHERNET_VLAN_SIZE)))
3297 adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
3299 pr_info("%s changing MTU from %d to %d\n",
3300 netdev->name, netdev->mtu, new_mtu);
3301 netdev->mtu = new_mtu;
3303 if (netif_running(netdev))
3304 e1000_up(adapter);
3305 else
3306 e1000_reset(adapter);
3308 clear_bit(__E1000_RESETTING, &adapter->flags);
3310 return 0;
3314 * e1000_update_stats - Update the board statistics counters
3315 * @adapter: board private structure
3318 void e1000_update_stats(struct e1000_adapter *adapter)
3320 struct net_device *netdev = adapter->netdev;
3321 struct e1000_hw *hw = &adapter->hw;
3322 struct pci_dev *pdev = adapter->pdev;
3323 unsigned long flags;
3324 u16 phy_tmp;
3326 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
3329 * Prevent stats update while adapter is being reset, or if the pci
3330 * connection is down.
3332 if (adapter->link_speed == 0)
3333 return;
3334 if (pci_channel_offline(pdev))
3335 return;
3337 spin_lock_irqsave(&adapter->stats_lock, flags);
3339 /* these counters are modified from e1000_tbi_adjust_stats,
3340 * called from the interrupt context, so they must only
3341 * be written while holding adapter->stats_lock
3344 adapter->stats.crcerrs += er32(CRCERRS);
3345 adapter->stats.gprc += er32(GPRC);
3346 adapter->stats.gorcl += er32(GORCL);
3347 adapter->stats.gorch += er32(GORCH);
3348 adapter->stats.bprc += er32(BPRC);
3349 adapter->stats.mprc += er32(MPRC);
3350 adapter->stats.roc += er32(ROC);
3352 adapter->stats.prc64 += er32(PRC64);
3353 adapter->stats.prc127 += er32(PRC127);
3354 adapter->stats.prc255 += er32(PRC255);
3355 adapter->stats.prc511 += er32(PRC511);
3356 adapter->stats.prc1023 += er32(PRC1023);
3357 adapter->stats.prc1522 += er32(PRC1522);
3359 adapter->stats.symerrs += er32(SYMERRS);
3360 adapter->stats.mpc += er32(MPC);
3361 adapter->stats.scc += er32(SCC);
3362 adapter->stats.ecol += er32(ECOL);
3363 adapter->stats.mcc += er32(MCC);
3364 adapter->stats.latecol += er32(LATECOL);
3365 adapter->stats.dc += er32(DC);
3366 adapter->stats.sec += er32(SEC);
3367 adapter->stats.rlec += er32(RLEC);
3368 adapter->stats.xonrxc += er32(XONRXC);
3369 adapter->stats.xontxc += er32(XONTXC);
3370 adapter->stats.xoffrxc += er32(XOFFRXC);
3371 adapter->stats.xofftxc += er32(XOFFTXC);
3372 adapter->stats.fcruc += er32(FCRUC);
3373 adapter->stats.gptc += er32(GPTC);
3374 adapter->stats.gotcl += er32(GOTCL);
3375 adapter->stats.gotch += er32(GOTCH);
3376 adapter->stats.rnbc += er32(RNBC);
3377 adapter->stats.ruc += er32(RUC);
3378 adapter->stats.rfc += er32(RFC);
3379 adapter->stats.rjc += er32(RJC);
3380 adapter->stats.torl += er32(TORL);
3381 adapter->stats.torh += er32(TORH);
3382 adapter->stats.totl += er32(TOTL);
3383 adapter->stats.toth += er32(TOTH);
3384 adapter->stats.tpr += er32(TPR);
3386 adapter->stats.ptc64 += er32(PTC64);
3387 adapter->stats.ptc127 += er32(PTC127);
3388 adapter->stats.ptc255 += er32(PTC255);
3389 adapter->stats.ptc511 += er32(PTC511);
3390 adapter->stats.ptc1023 += er32(PTC1023);
3391 adapter->stats.ptc1522 += er32(PTC1522);
3393 adapter->stats.mptc += er32(MPTC);
3394 adapter->stats.bptc += er32(BPTC);
3396 /* used for adaptive IFS */
3398 hw->tx_packet_delta = er32(TPT);
3399 adapter->stats.tpt += hw->tx_packet_delta;
3400 hw->collision_delta = er32(COLC);
3401 adapter->stats.colc += hw->collision_delta;
3403 if (hw->mac_type >= e1000_82543) {
3404 adapter->stats.algnerrc += er32(ALGNERRC);
3405 adapter->stats.rxerrc += er32(RXERRC);
3406 adapter->stats.tncrs += er32(TNCRS);
3407 adapter->stats.cexterr += er32(CEXTERR);
3408 adapter->stats.tsctc += er32(TSCTC);
3409 adapter->stats.tsctfc += er32(TSCTFC);
3412 /* Fill out the OS statistics structure */
3413 netdev->stats.multicast = adapter->stats.mprc;
3414 netdev->stats.collisions = adapter->stats.colc;
3416 /* Rx Errors */
3418 /* RLEC on some newer hardware can be incorrect so build
3419 * our own version based on RUC and ROC */
3420 netdev->stats.rx_errors = adapter->stats.rxerrc +
3421 adapter->stats.crcerrs + adapter->stats.algnerrc +
3422 adapter->stats.ruc + adapter->stats.roc +
3423 adapter->stats.cexterr;
3424 adapter->stats.rlerrc = adapter->stats.ruc + adapter->stats.roc;
3425 netdev->stats.rx_length_errors = adapter->stats.rlerrc;
3426 netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
3427 netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
3428 netdev->stats.rx_missed_errors = adapter->stats.mpc;
3430 /* Tx Errors */
3431 adapter->stats.txerrc = adapter->stats.ecol + adapter->stats.latecol;
3432 netdev->stats.tx_errors = adapter->stats.txerrc;
3433 netdev->stats.tx_aborted_errors = adapter->stats.ecol;
3434 netdev->stats.tx_window_errors = adapter->stats.latecol;
3435 netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
3436 if (hw->bad_tx_carr_stats_fd &&
3437 adapter->link_duplex == FULL_DUPLEX) {
3438 netdev->stats.tx_carrier_errors = 0;
3439 adapter->stats.tncrs = 0;
3442 /* Tx Dropped needs to be maintained elsewhere */
3444 /* Phy Stats */
3445 if (hw->media_type == e1000_media_type_copper) {
3446 if ((adapter->link_speed == SPEED_1000) &&
3447 (!e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_tmp))) {
3448 phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
3449 adapter->phy_stats.idle_errors += phy_tmp;
3452 if ((hw->mac_type <= e1000_82546) &&
3453 (hw->phy_type == e1000_phy_m88) &&
3454 !e1000_read_phy_reg(hw, M88E1000_RX_ERR_CNTR, &phy_tmp))
3455 adapter->phy_stats.receive_errors += phy_tmp;
3458 /* Management Stats */
3459 if (hw->has_smbus) {
3460 adapter->stats.mgptc += er32(MGTPTC);
3461 adapter->stats.mgprc += er32(MGTPRC);
3462 adapter->stats.mgpdc += er32(MGTPDC);
3465 spin_unlock_irqrestore(&adapter->stats_lock, flags);
3469 * e1000_intr - Interrupt Handler
3470 * @irq: interrupt number
3471 * @data: pointer to a network interface device structure
3474 static irqreturn_t e1000_intr(int irq, void *data)
3476 struct net_device *netdev = data;
3477 struct e1000_adapter *adapter = netdev_priv(netdev);
3478 struct e1000_hw *hw = &adapter->hw;
3479 u32 icr = er32(ICR);
3481 if (unlikely((!icr) || test_bit(__E1000_DOWN, &adapter->flags)))
3482 return IRQ_NONE; /* Not our interrupt */
3484 if (unlikely(icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC))) {
3485 hw->get_link_status = 1;
3486 /* guard against interrupt when we're going down */
3487 if (!test_bit(__E1000_DOWN, &adapter->flags))
3488 mod_timer(&adapter->watchdog_timer, jiffies + 1);
3491 /* disable interrupts, without the synchronize_irq bit */
3492 ew32(IMC, ~0);
3493 E1000_WRITE_FLUSH();
3495 if (likely(napi_schedule_prep(&adapter->napi))) {
3496 adapter->total_tx_bytes = 0;
3497 adapter->total_tx_packets = 0;
3498 adapter->total_rx_bytes = 0;
3499 adapter->total_rx_packets = 0;
3500 __napi_schedule(&adapter->napi);
3501 } else {
3502 /* this really should not happen! if it does it is basically a
3503 * bug, but not a hard error, so enable ints and continue */
3504 if (!test_bit(__E1000_DOWN, &adapter->flags))
3505 e1000_irq_enable(adapter);
3508 return IRQ_HANDLED;
3512 * e1000_clean - NAPI Rx polling callback
3513 * @adapter: board private structure
3515 static int e1000_clean(struct napi_struct *napi, int budget)
3517 struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
3518 int tx_clean_complete = 0, work_done = 0;
3520 tx_clean_complete = e1000_clean_tx_irq(adapter, &adapter->tx_ring[0]);
3522 adapter->clean_rx(adapter, &adapter->rx_ring[0], &work_done, budget);
3524 if (!tx_clean_complete)
3525 work_done = budget;
3527 /* If budget not fully consumed, exit the polling mode */
3528 if (work_done < budget) {
3529 if (likely(adapter->itr_setting & 3))
3530 e1000_set_itr(adapter);
3531 napi_complete(napi);
3532 if (!test_bit(__E1000_DOWN, &adapter->flags))
3533 e1000_irq_enable(adapter);
3536 return work_done;
3540 * e1000_clean_tx_irq - Reclaim resources after transmit completes
3541 * @adapter: board private structure
3543 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter,
3544 struct e1000_tx_ring *tx_ring)
3546 struct e1000_hw *hw = &adapter->hw;
3547 struct net_device *netdev = adapter->netdev;
3548 struct e1000_tx_desc *tx_desc, *eop_desc;
3549 struct e1000_buffer *buffer_info;
3550 unsigned int i, eop;
3551 unsigned int count = 0;
3552 unsigned int total_tx_bytes=0, total_tx_packets=0;
3554 i = tx_ring->next_to_clean;
3555 eop = tx_ring->buffer_info[i].next_to_watch;
3556 eop_desc = E1000_TX_DESC(*tx_ring, eop);
3558 while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
3559 (count < tx_ring->count)) {
3560 bool cleaned = false;
3561 rmb(); /* read buffer_info after eop_desc */
3562 for ( ; !cleaned; count++) {
3563 tx_desc = E1000_TX_DESC(*tx_ring, i);
3564 buffer_info = &tx_ring->buffer_info[i];
3565 cleaned = (i == eop);
3567 if (cleaned) {
3568 struct sk_buff *skb = buffer_info->skb;
3569 unsigned int segs, bytecount;
3570 segs = skb_shinfo(skb)->gso_segs ?: 1;
3571 /* multiply data chunks by size of headers */
3572 bytecount = ((segs - 1) * skb_headlen(skb)) +
3573 skb->len;
3574 total_tx_packets += segs;
3575 total_tx_bytes += bytecount;
3577 e1000_unmap_and_free_tx_resource(adapter, buffer_info);
3578 tx_desc->upper.data = 0;
3580 if (unlikely(++i == tx_ring->count)) i = 0;
3583 eop = tx_ring->buffer_info[i].next_to_watch;
3584 eop_desc = E1000_TX_DESC(*tx_ring, eop);
3587 tx_ring->next_to_clean = i;
3589 #define TX_WAKE_THRESHOLD 32
3590 if (unlikely(count && netif_carrier_ok(netdev) &&
3591 E1000_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD)) {
3592 /* Make sure that anybody stopping the queue after this
3593 * sees the new next_to_clean.
3595 smp_mb();
3597 if (netif_queue_stopped(netdev) &&
3598 !(test_bit(__E1000_DOWN, &adapter->flags))) {
3599 netif_wake_queue(netdev);
3600 ++adapter->restart_queue;
3604 if (adapter->detect_tx_hung) {
3605 /* Detect a transmit hang in hardware, this serializes the
3606 * check with the clearing of time_stamp and movement of i */
3607 adapter->detect_tx_hung = false;
3608 if (tx_ring->buffer_info[eop].time_stamp &&
3609 time_after(jiffies, tx_ring->buffer_info[eop].time_stamp +
3610 (adapter->tx_timeout_factor * HZ)) &&
3611 !(er32(STATUS) & E1000_STATUS_TXOFF)) {
3613 /* detected Tx unit hang */
3614 e_err(drv, "Detected Tx Unit Hang\n"
3615 " Tx Queue <%lu>\n"
3616 " TDH <%x>\n"
3617 " TDT <%x>\n"
3618 " next_to_use <%x>\n"
3619 " next_to_clean <%x>\n"
3620 "buffer_info[next_to_clean]\n"
3621 " time_stamp <%lx>\n"
3622 " next_to_watch <%x>\n"
3623 " jiffies <%lx>\n"
3624 " next_to_watch.status <%x>\n",
3625 (unsigned long)((tx_ring - adapter->tx_ring) /
3626 sizeof(struct e1000_tx_ring)),
3627 readl(hw->hw_addr + tx_ring->tdh),
3628 readl(hw->hw_addr + tx_ring->tdt),
3629 tx_ring->next_to_use,
3630 tx_ring->next_to_clean,
3631 tx_ring->buffer_info[eop].time_stamp,
3632 eop,
3633 jiffies,
3634 eop_desc->upper.fields.status);
3635 netif_stop_queue(netdev);
3638 adapter->total_tx_bytes += total_tx_bytes;
3639 adapter->total_tx_packets += total_tx_packets;
3640 netdev->stats.tx_bytes += total_tx_bytes;
3641 netdev->stats.tx_packets += total_tx_packets;
3642 return count < tx_ring->count;
3646 * e1000_rx_checksum - Receive Checksum Offload for 82543
3647 * @adapter: board private structure
3648 * @status_err: receive descriptor status and error fields
3649 * @csum: receive descriptor csum field
3650 * @sk_buff: socket buffer with received data
3653 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
3654 u32 csum, struct sk_buff *skb)
3656 struct e1000_hw *hw = &adapter->hw;
3657 u16 status = (u16)status_err;
3658 u8 errors = (u8)(status_err >> 24);
3660 skb_checksum_none_assert(skb);
3662 /* 82543 or newer only */
3663 if (unlikely(hw->mac_type < e1000_82543)) return;
3664 /* Ignore Checksum bit is set */
3665 if (unlikely(status & E1000_RXD_STAT_IXSM)) return;
3666 /* TCP/UDP checksum error bit is set */
3667 if (unlikely(errors & E1000_RXD_ERR_TCPE)) {
3668 /* let the stack verify checksum errors */
3669 adapter->hw_csum_err++;
3670 return;
3672 /* TCP/UDP Checksum has not been calculated */
3673 if (!(status & E1000_RXD_STAT_TCPCS))
3674 return;
3676 /* It must be a TCP or UDP packet with a valid checksum */
3677 if (likely(status & E1000_RXD_STAT_TCPCS)) {
3678 /* TCP checksum is good */
3679 skb->ip_summed = CHECKSUM_UNNECESSARY;
3681 adapter->hw_csum_good++;
3685 * e1000_consume_page - helper function
3687 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
3688 u16 length)
3690 bi->page = NULL;
3691 skb->len += length;
3692 skb->data_len += length;
3693 skb->truesize += length;
3697 * e1000_receive_skb - helper function to handle rx indications
3698 * @adapter: board private structure
3699 * @status: descriptor status field as written by hardware
3700 * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
3701 * @skb: pointer to sk_buff to be indicated to stack
3703 static void e1000_receive_skb(struct e1000_adapter *adapter, u8 status,
3704 __le16 vlan, struct sk_buff *skb)
3706 skb->protocol = eth_type_trans(skb, adapter->netdev);
3708 if ((unlikely(adapter->vlgrp && (status & E1000_RXD_STAT_VP))))
3709 vlan_gro_receive(&adapter->napi, adapter->vlgrp,
3710 le16_to_cpu(vlan) & E1000_RXD_SPC_VLAN_MASK,
3711 skb);
3712 else
3713 napi_gro_receive(&adapter->napi, skb);
3717 * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
3718 * @adapter: board private structure
3719 * @rx_ring: ring to clean
3720 * @work_done: amount of napi work completed this call
3721 * @work_to_do: max amount of work allowed for this call to do
3723 * the return value indicates whether actual cleaning was done, there
3724 * is no guarantee that everything was cleaned
3726 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
3727 struct e1000_rx_ring *rx_ring,
3728 int *work_done, int work_to_do)
3730 struct e1000_hw *hw = &adapter->hw;
3731 struct net_device *netdev = adapter->netdev;
3732 struct pci_dev *pdev = adapter->pdev;
3733 struct e1000_rx_desc *rx_desc, *next_rxd;
3734 struct e1000_buffer *buffer_info, *next_buffer;
3735 unsigned long irq_flags;
3736 u32 length;
3737 unsigned int i;
3738 int cleaned_count = 0;
3739 bool cleaned = false;
3740 unsigned int total_rx_bytes=0, total_rx_packets=0;
3742 i = rx_ring->next_to_clean;
3743 rx_desc = E1000_RX_DESC(*rx_ring, i);
3744 buffer_info = &rx_ring->buffer_info[i];
3746 while (rx_desc->status & E1000_RXD_STAT_DD) {
3747 struct sk_buff *skb;
3748 u8 status;
3750 if (*work_done >= work_to_do)
3751 break;
3752 (*work_done)++;
3753 rmb(); /* read descriptor and rx_buffer_info after status DD */
3755 status = rx_desc->status;
3756 skb = buffer_info->skb;
3757 buffer_info->skb = NULL;
3759 if (++i == rx_ring->count) i = 0;
3760 next_rxd = E1000_RX_DESC(*rx_ring, i);
3761 prefetch(next_rxd);
3763 next_buffer = &rx_ring->buffer_info[i];
3765 cleaned = true;
3766 cleaned_count++;
3767 dma_unmap_page(&pdev->dev, buffer_info->dma,
3768 buffer_info->length, DMA_FROM_DEVICE);
3769 buffer_info->dma = 0;
3771 length = le16_to_cpu(rx_desc->length);
3773 /* errors is only valid for DD + EOP descriptors */
3774 if (unlikely((status & E1000_RXD_STAT_EOP) &&
3775 (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK))) {
3776 u8 last_byte = *(skb->data + length - 1);
3777 if (TBI_ACCEPT(hw, status, rx_desc->errors, length,
3778 last_byte)) {
3779 spin_lock_irqsave(&adapter->stats_lock,
3780 irq_flags);
3781 e1000_tbi_adjust_stats(hw, &adapter->stats,
3782 length, skb->data);
3783 spin_unlock_irqrestore(&adapter->stats_lock,
3784 irq_flags);
3785 length--;
3786 } else {
3787 /* recycle both page and skb */
3788 buffer_info->skb = skb;
3789 /* an error means any chain goes out the window
3790 * too */
3791 if (rx_ring->rx_skb_top)
3792 dev_kfree_skb(rx_ring->rx_skb_top);
3793 rx_ring->rx_skb_top = NULL;
3794 goto next_desc;
3798 #define rxtop rx_ring->rx_skb_top
3799 if (!(status & E1000_RXD_STAT_EOP)) {
3800 /* this descriptor is only the beginning (or middle) */
3801 if (!rxtop) {
3802 /* this is the beginning of a chain */
3803 rxtop = skb;
3804 skb_fill_page_desc(rxtop, 0, buffer_info->page,
3805 0, length);
3806 } else {
3807 /* this is the middle of a chain */
3808 skb_fill_page_desc(rxtop,
3809 skb_shinfo(rxtop)->nr_frags,
3810 buffer_info->page, 0, length);
3811 /* re-use the skb, only consumed the page */
3812 buffer_info->skb = skb;
3814 e1000_consume_page(buffer_info, rxtop, length);
3815 goto next_desc;
3816 } else {
3817 if (rxtop) {
3818 /* end of the chain */
3819 skb_fill_page_desc(rxtop,
3820 skb_shinfo(rxtop)->nr_frags,
3821 buffer_info->page, 0, length);
3822 /* re-use the current skb, we only consumed the
3823 * page */
3824 buffer_info->skb = skb;
3825 skb = rxtop;
3826 rxtop = NULL;
3827 e1000_consume_page(buffer_info, skb, length);
3828 } else {
3829 /* no chain, got EOP, this buf is the packet
3830 * copybreak to save the put_page/alloc_page */
3831 if (length <= copybreak &&
3832 skb_tailroom(skb) >= length) {
3833 u8 *vaddr;
3834 vaddr = kmap_atomic(buffer_info->page,
3835 KM_SKB_DATA_SOFTIRQ);
3836 memcpy(skb_tail_pointer(skb), vaddr, length);
3837 kunmap_atomic(vaddr,
3838 KM_SKB_DATA_SOFTIRQ);
3839 /* re-use the page, so don't erase
3840 * buffer_info->page */
3841 skb_put(skb, length);
3842 } else {
3843 skb_fill_page_desc(skb, 0,
3844 buffer_info->page, 0,
3845 length);
3846 e1000_consume_page(buffer_info, skb,
3847 length);
3852 /* Receive Checksum Offload XXX recompute due to CRC strip? */
3853 e1000_rx_checksum(adapter,
3854 (u32)(status) |
3855 ((u32)(rx_desc->errors) << 24),
3856 le16_to_cpu(rx_desc->csum), skb);
3858 pskb_trim(skb, skb->len - 4);
3860 /* probably a little skewed due to removing CRC */
3861 total_rx_bytes += skb->len;
3862 total_rx_packets++;
3864 /* eth type trans needs skb->data to point to something */
3865 if (!pskb_may_pull(skb, ETH_HLEN)) {
3866 e_err(drv, "pskb_may_pull failed.\n");
3867 dev_kfree_skb(skb);
3868 goto next_desc;
3871 e1000_receive_skb(adapter, status, rx_desc->special, skb);
3873 next_desc:
3874 rx_desc->status = 0;
3876 /* return some buffers to hardware, one at a time is too slow */
3877 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
3878 adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
3879 cleaned_count = 0;
3882 /* use prefetched values */
3883 rx_desc = next_rxd;
3884 buffer_info = next_buffer;
3886 rx_ring->next_to_clean = i;
3888 cleaned_count = E1000_DESC_UNUSED(rx_ring);
3889 if (cleaned_count)
3890 adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
3892 adapter->total_rx_packets += total_rx_packets;
3893 adapter->total_rx_bytes += total_rx_bytes;
3894 netdev->stats.rx_bytes += total_rx_bytes;
3895 netdev->stats.rx_packets += total_rx_packets;
3896 return cleaned;
3900 * this should improve performance for small packets with large amounts
3901 * of reassembly being done in the stack
3903 static void e1000_check_copybreak(struct net_device *netdev,
3904 struct e1000_buffer *buffer_info,
3905 u32 length, struct sk_buff **skb)
3907 struct sk_buff *new_skb;
3909 if (length > copybreak)
3910 return;
3912 new_skb = netdev_alloc_skb_ip_align(netdev, length);
3913 if (!new_skb)
3914 return;
3916 skb_copy_to_linear_data_offset(new_skb, -NET_IP_ALIGN,
3917 (*skb)->data - NET_IP_ALIGN,
3918 length + NET_IP_ALIGN);
3919 /* save the skb in buffer_info as good */
3920 buffer_info->skb = *skb;
3921 *skb = new_skb;
3925 * e1000_clean_rx_irq - Send received data up the network stack; legacy
3926 * @adapter: board private structure
3927 * @rx_ring: ring to clean
3928 * @work_done: amount of napi work completed this call
3929 * @work_to_do: max amount of work allowed for this call to do
3931 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
3932 struct e1000_rx_ring *rx_ring,
3933 int *work_done, int work_to_do)
3935 struct e1000_hw *hw = &adapter->hw;
3936 struct net_device *netdev = adapter->netdev;
3937 struct pci_dev *pdev = adapter->pdev;
3938 struct e1000_rx_desc *rx_desc, *next_rxd;
3939 struct e1000_buffer *buffer_info, *next_buffer;
3940 unsigned long flags;
3941 u32 length;
3942 unsigned int i;
3943 int cleaned_count = 0;
3944 bool cleaned = false;
3945 unsigned int total_rx_bytes=0, total_rx_packets=0;
3947 i = rx_ring->next_to_clean;
3948 rx_desc = E1000_RX_DESC(*rx_ring, i);
3949 buffer_info = &rx_ring->buffer_info[i];
3951 while (rx_desc->status & E1000_RXD_STAT_DD) {
3952 struct sk_buff *skb;
3953 u8 status;
3955 if (*work_done >= work_to_do)
3956 break;
3957 (*work_done)++;
3958 rmb(); /* read descriptor and rx_buffer_info after status DD */
3960 status = rx_desc->status;
3961 skb = buffer_info->skb;
3962 buffer_info->skb = NULL;
3964 prefetch(skb->data - NET_IP_ALIGN);
3966 if (++i == rx_ring->count) i = 0;
3967 next_rxd = E1000_RX_DESC(*rx_ring, i);
3968 prefetch(next_rxd);
3970 next_buffer = &rx_ring->buffer_info[i];
3972 cleaned = true;
3973 cleaned_count++;
3974 dma_unmap_single(&pdev->dev, buffer_info->dma,
3975 buffer_info->length, DMA_FROM_DEVICE);
3976 buffer_info->dma = 0;
3978 length = le16_to_cpu(rx_desc->length);
3979 /* !EOP means multiple descriptors were used to store a single
3980 * packet, if thats the case we need to toss it. In fact, we
3981 * to toss every packet with the EOP bit clear and the next
3982 * frame that _does_ have the EOP bit set, as it is by
3983 * definition only a frame fragment
3985 if (unlikely(!(status & E1000_RXD_STAT_EOP)))
3986 adapter->discarding = true;
3988 if (adapter->discarding) {
3989 /* All receives must fit into a single buffer */
3990 e_dbg("Receive packet consumed multiple buffers\n");
3991 /* recycle */
3992 buffer_info->skb = skb;
3993 if (status & E1000_RXD_STAT_EOP)
3994 adapter->discarding = false;
3995 goto next_desc;
3998 if (unlikely(rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK)) {
3999 u8 last_byte = *(skb->data + length - 1);
4000 if (TBI_ACCEPT(hw, status, rx_desc->errors, length,
4001 last_byte)) {
4002 spin_lock_irqsave(&adapter->stats_lock, flags);
4003 e1000_tbi_adjust_stats(hw, &adapter->stats,
4004 length, skb->data);
4005 spin_unlock_irqrestore(&adapter->stats_lock,
4006 flags);
4007 length--;
4008 } else {
4009 /* recycle */
4010 buffer_info->skb = skb;
4011 goto next_desc;
4015 /* adjust length to remove Ethernet CRC, this must be
4016 * done after the TBI_ACCEPT workaround above */
4017 length -= 4;
4019 /* probably a little skewed due to removing CRC */
4020 total_rx_bytes += length;
4021 total_rx_packets++;
4023 e1000_check_copybreak(netdev, buffer_info, length, &skb);
4025 skb_put(skb, length);
4027 /* Receive Checksum Offload */
4028 e1000_rx_checksum(adapter,
4029 (u32)(status) |
4030 ((u32)(rx_desc->errors) << 24),
4031 le16_to_cpu(rx_desc->csum), skb);
4033 e1000_receive_skb(adapter, status, rx_desc->special, skb);
4035 next_desc:
4036 rx_desc->status = 0;
4038 /* return some buffers to hardware, one at a time is too slow */
4039 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
4040 adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
4041 cleaned_count = 0;
4044 /* use prefetched values */
4045 rx_desc = next_rxd;
4046 buffer_info = next_buffer;
4048 rx_ring->next_to_clean = i;
4050 cleaned_count = E1000_DESC_UNUSED(rx_ring);
4051 if (cleaned_count)
4052 adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
4054 adapter->total_rx_packets += total_rx_packets;
4055 adapter->total_rx_bytes += total_rx_bytes;
4056 netdev->stats.rx_bytes += total_rx_bytes;
4057 netdev->stats.rx_packets += total_rx_packets;
4058 return cleaned;
4062 * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
4063 * @adapter: address of board private structure
4064 * @rx_ring: pointer to receive ring structure
4065 * @cleaned_count: number of buffers to allocate this pass
4068 static void
4069 e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
4070 struct e1000_rx_ring *rx_ring, int cleaned_count)
4072 struct net_device *netdev = adapter->netdev;
4073 struct pci_dev *pdev = adapter->pdev;
4074 struct e1000_rx_desc *rx_desc;
4075 struct e1000_buffer *buffer_info;
4076 struct sk_buff *skb;
4077 unsigned int i;
4078 unsigned int bufsz = 256 - 16 /*for skb_reserve */ ;
4080 i = rx_ring->next_to_use;
4081 buffer_info = &rx_ring->buffer_info[i];
4083 while (cleaned_count--) {
4084 skb = buffer_info->skb;
4085 if (skb) {
4086 skb_trim(skb, 0);
4087 goto check_page;
4090 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
4091 if (unlikely(!skb)) {
4092 /* Better luck next round */
4093 adapter->alloc_rx_buff_failed++;
4094 break;
4097 /* Fix for errata 23, can't cross 64kB boundary */
4098 if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
4099 struct sk_buff *oldskb = skb;
4100 e_err(rx_err, "skb align check failed: %u bytes at "
4101 "%p\n", bufsz, skb->data);
4102 /* Try again, without freeing the previous */
4103 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
4104 /* Failed allocation, critical failure */
4105 if (!skb) {
4106 dev_kfree_skb(oldskb);
4107 adapter->alloc_rx_buff_failed++;
4108 break;
4111 if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
4112 /* give up */
4113 dev_kfree_skb(skb);
4114 dev_kfree_skb(oldskb);
4115 break; /* while (cleaned_count--) */
4118 /* Use new allocation */
4119 dev_kfree_skb(oldskb);
4121 buffer_info->skb = skb;
4122 buffer_info->length = adapter->rx_buffer_len;
4123 check_page:
4124 /* allocate a new page if necessary */
4125 if (!buffer_info->page) {
4126 buffer_info->page = alloc_page(GFP_ATOMIC);
4127 if (unlikely(!buffer_info->page)) {
4128 adapter->alloc_rx_buff_failed++;
4129 break;
4133 if (!buffer_info->dma) {
4134 buffer_info->dma = dma_map_page(&pdev->dev,
4135 buffer_info->page, 0,
4136 buffer_info->length,
4137 DMA_FROM_DEVICE);
4138 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
4139 put_page(buffer_info->page);
4140 dev_kfree_skb(skb);
4141 buffer_info->page = NULL;
4142 buffer_info->skb = NULL;
4143 buffer_info->dma = 0;
4144 adapter->alloc_rx_buff_failed++;
4145 break; /* while !buffer_info->skb */
4149 rx_desc = E1000_RX_DESC(*rx_ring, i);
4150 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4152 if (unlikely(++i == rx_ring->count))
4153 i = 0;
4154 buffer_info = &rx_ring->buffer_info[i];
4157 if (likely(rx_ring->next_to_use != i)) {
4158 rx_ring->next_to_use = i;
4159 if (unlikely(i-- == 0))
4160 i = (rx_ring->count - 1);
4162 /* Force memory writes to complete before letting h/w
4163 * know there are new descriptors to fetch. (Only
4164 * applicable for weak-ordered memory model archs,
4165 * such as IA-64). */
4166 wmb();
4167 writel(i, adapter->hw.hw_addr + rx_ring->rdt);
4172 * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
4173 * @adapter: address of board private structure
4176 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
4177 struct e1000_rx_ring *rx_ring,
4178 int cleaned_count)
4180 struct e1000_hw *hw = &adapter->hw;
4181 struct net_device *netdev = adapter->netdev;
4182 struct pci_dev *pdev = adapter->pdev;
4183 struct e1000_rx_desc *rx_desc;
4184 struct e1000_buffer *buffer_info;
4185 struct sk_buff *skb;
4186 unsigned int i;
4187 unsigned int bufsz = adapter->rx_buffer_len;
4189 i = rx_ring->next_to_use;
4190 buffer_info = &rx_ring->buffer_info[i];
4192 while (cleaned_count--) {
4193 skb = buffer_info->skb;
4194 if (skb) {
4195 skb_trim(skb, 0);
4196 goto map_skb;
4199 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
4200 if (unlikely(!skb)) {
4201 /* Better luck next round */
4202 adapter->alloc_rx_buff_failed++;
4203 break;
4206 /* Fix for errata 23, can't cross 64kB boundary */
4207 if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
4208 struct sk_buff *oldskb = skb;
4209 e_err(rx_err, "skb align check failed: %u bytes at "
4210 "%p\n", bufsz, skb->data);
4211 /* Try again, without freeing the previous */
4212 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
4213 /* Failed allocation, critical failure */
4214 if (!skb) {
4215 dev_kfree_skb(oldskb);
4216 adapter->alloc_rx_buff_failed++;
4217 break;
4220 if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
4221 /* give up */
4222 dev_kfree_skb(skb);
4223 dev_kfree_skb(oldskb);
4224 adapter->alloc_rx_buff_failed++;
4225 break; /* while !buffer_info->skb */
4228 /* Use new allocation */
4229 dev_kfree_skb(oldskb);
4231 buffer_info->skb = skb;
4232 buffer_info->length = adapter->rx_buffer_len;
4233 map_skb:
4234 buffer_info->dma = dma_map_single(&pdev->dev,
4235 skb->data,
4236 buffer_info->length,
4237 DMA_FROM_DEVICE);
4238 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
4239 dev_kfree_skb(skb);
4240 buffer_info->skb = NULL;
4241 buffer_info->dma = 0;
4242 adapter->alloc_rx_buff_failed++;
4243 break; /* while !buffer_info->skb */
4247 * XXX if it was allocated cleanly it will never map to a
4248 * boundary crossing
4251 /* Fix for errata 23, can't cross 64kB boundary */
4252 if (!e1000_check_64k_bound(adapter,
4253 (void *)(unsigned long)buffer_info->dma,
4254 adapter->rx_buffer_len)) {
4255 e_err(rx_err, "dma align check failed: %u bytes at "
4256 "%p\n", adapter->rx_buffer_len,
4257 (void *)(unsigned long)buffer_info->dma);
4258 dev_kfree_skb(skb);
4259 buffer_info->skb = NULL;
4261 dma_unmap_single(&pdev->dev, buffer_info->dma,
4262 adapter->rx_buffer_len,
4263 DMA_FROM_DEVICE);
4264 buffer_info->dma = 0;
4266 adapter->alloc_rx_buff_failed++;
4267 break; /* while !buffer_info->skb */
4269 rx_desc = E1000_RX_DESC(*rx_ring, i);
4270 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4272 if (unlikely(++i == rx_ring->count))
4273 i = 0;
4274 buffer_info = &rx_ring->buffer_info[i];
4277 if (likely(rx_ring->next_to_use != i)) {
4278 rx_ring->next_to_use = i;
4279 if (unlikely(i-- == 0))
4280 i = (rx_ring->count - 1);
4282 /* Force memory writes to complete before letting h/w
4283 * know there are new descriptors to fetch. (Only
4284 * applicable for weak-ordered memory model archs,
4285 * such as IA-64). */
4286 wmb();
4287 writel(i, hw->hw_addr + rx_ring->rdt);
4292 * e1000_smartspeed - Workaround for SmartSpeed on 82541 and 82547 controllers.
4293 * @adapter:
4296 static void e1000_smartspeed(struct e1000_adapter *adapter)
4298 struct e1000_hw *hw = &adapter->hw;
4299 u16 phy_status;
4300 u16 phy_ctrl;
4302 if ((hw->phy_type != e1000_phy_igp) || !hw->autoneg ||
4303 !(hw->autoneg_advertised & ADVERTISE_1000_FULL))
4304 return;
4306 if (adapter->smartspeed == 0) {
4307 /* If Master/Slave config fault is asserted twice,
4308 * we assume back-to-back */
4309 e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_status);
4310 if (!(phy_status & SR_1000T_MS_CONFIG_FAULT)) return;
4311 e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_status);
4312 if (!(phy_status & SR_1000T_MS_CONFIG_FAULT)) return;
4313 e1000_read_phy_reg(hw, PHY_1000T_CTRL, &phy_ctrl);
4314 if (phy_ctrl & CR_1000T_MS_ENABLE) {
4315 phy_ctrl &= ~CR_1000T_MS_ENABLE;
4316 e1000_write_phy_reg(hw, PHY_1000T_CTRL,
4317 phy_ctrl);
4318 adapter->smartspeed++;
4319 if (!e1000_phy_setup_autoneg(hw) &&
4320 !e1000_read_phy_reg(hw, PHY_CTRL,
4321 &phy_ctrl)) {
4322 phy_ctrl |= (MII_CR_AUTO_NEG_EN |
4323 MII_CR_RESTART_AUTO_NEG);
4324 e1000_write_phy_reg(hw, PHY_CTRL,
4325 phy_ctrl);
4328 return;
4329 } else if (adapter->smartspeed == E1000_SMARTSPEED_DOWNSHIFT) {
4330 /* If still no link, perhaps using 2/3 pair cable */
4331 e1000_read_phy_reg(hw, PHY_1000T_CTRL, &phy_ctrl);
4332 phy_ctrl |= CR_1000T_MS_ENABLE;
4333 e1000_write_phy_reg(hw, PHY_1000T_CTRL, phy_ctrl);
4334 if (!e1000_phy_setup_autoneg(hw) &&
4335 !e1000_read_phy_reg(hw, PHY_CTRL, &phy_ctrl)) {
4336 phy_ctrl |= (MII_CR_AUTO_NEG_EN |
4337 MII_CR_RESTART_AUTO_NEG);
4338 e1000_write_phy_reg(hw, PHY_CTRL, phy_ctrl);
4341 /* Restart process after E1000_SMARTSPEED_MAX iterations */
4342 if (adapter->smartspeed++ == E1000_SMARTSPEED_MAX)
4343 adapter->smartspeed = 0;
4347 * e1000_ioctl -
4348 * @netdev:
4349 * @ifreq:
4350 * @cmd:
4353 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4355 switch (cmd) {
4356 case SIOCGMIIPHY:
4357 case SIOCGMIIREG:
4358 case SIOCSMIIREG:
4359 return e1000_mii_ioctl(netdev, ifr, cmd);
4360 default:
4361 return -EOPNOTSUPP;
4366 * e1000_mii_ioctl -
4367 * @netdev:
4368 * @ifreq:
4369 * @cmd:
4372 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
4373 int cmd)
4375 struct e1000_adapter *adapter = netdev_priv(netdev);
4376 struct e1000_hw *hw = &adapter->hw;
4377 struct mii_ioctl_data *data = if_mii(ifr);
4378 int retval;
4379 u16 mii_reg;
4380 u16 spddplx;
4381 unsigned long flags;
4383 if (hw->media_type != e1000_media_type_copper)
4384 return -EOPNOTSUPP;
4386 switch (cmd) {
4387 case SIOCGMIIPHY:
4388 data->phy_id = hw->phy_addr;
4389 break;
4390 case SIOCGMIIREG:
4391 spin_lock_irqsave(&adapter->stats_lock, flags);
4392 if (e1000_read_phy_reg(hw, data->reg_num & 0x1F,
4393 &data->val_out)) {
4394 spin_unlock_irqrestore(&adapter->stats_lock, flags);
4395 return -EIO;
4397 spin_unlock_irqrestore(&adapter->stats_lock, flags);
4398 break;
4399 case SIOCSMIIREG:
4400 if (data->reg_num & ~(0x1F))
4401 return -EFAULT;
4402 mii_reg = data->val_in;
4403 spin_lock_irqsave(&adapter->stats_lock, flags);
4404 if (e1000_write_phy_reg(hw, data->reg_num,
4405 mii_reg)) {
4406 spin_unlock_irqrestore(&adapter->stats_lock, flags);
4407 return -EIO;
4409 spin_unlock_irqrestore(&adapter->stats_lock, flags);
4410 if (hw->media_type == e1000_media_type_copper) {
4411 switch (data->reg_num) {
4412 case PHY_CTRL:
4413 if (mii_reg & MII_CR_POWER_DOWN)
4414 break;
4415 if (mii_reg & MII_CR_AUTO_NEG_EN) {
4416 hw->autoneg = 1;
4417 hw->autoneg_advertised = 0x2F;
4418 } else {
4419 if (mii_reg & 0x40)
4420 spddplx = SPEED_1000;
4421 else if (mii_reg & 0x2000)
4422 spddplx = SPEED_100;
4423 else
4424 spddplx = SPEED_10;
4425 spddplx += (mii_reg & 0x100)
4426 ? DUPLEX_FULL :
4427 DUPLEX_HALF;
4428 retval = e1000_set_spd_dplx(adapter,
4429 spddplx);
4430 if (retval)
4431 return retval;
4433 if (netif_running(adapter->netdev))
4434 e1000_reinit_locked(adapter);
4435 else
4436 e1000_reset(adapter);
4437 break;
4438 case M88E1000_PHY_SPEC_CTRL:
4439 case M88E1000_EXT_PHY_SPEC_CTRL:
4440 if (e1000_phy_reset(hw))
4441 return -EIO;
4442 break;
4444 } else {
4445 switch (data->reg_num) {
4446 case PHY_CTRL:
4447 if (mii_reg & MII_CR_POWER_DOWN)
4448 break;
4449 if (netif_running(adapter->netdev))
4450 e1000_reinit_locked(adapter);
4451 else
4452 e1000_reset(adapter);
4453 break;
4456 break;
4457 default:
4458 return -EOPNOTSUPP;
4460 return E1000_SUCCESS;
4463 void e1000_pci_set_mwi(struct e1000_hw *hw)
4465 struct e1000_adapter *adapter = hw->back;
4466 int ret_val = pci_set_mwi(adapter->pdev);
4468 if (ret_val)
4469 e_err(probe, "Error in setting MWI\n");
4472 void e1000_pci_clear_mwi(struct e1000_hw *hw)
4474 struct e1000_adapter *adapter = hw->back;
4476 pci_clear_mwi(adapter->pdev);
4479 int e1000_pcix_get_mmrbc(struct e1000_hw *hw)
4481 struct e1000_adapter *adapter = hw->back;
4482 return pcix_get_mmrbc(adapter->pdev);
4485 void e1000_pcix_set_mmrbc(struct e1000_hw *hw, int mmrbc)
4487 struct e1000_adapter *adapter = hw->back;
4488 pcix_set_mmrbc(adapter->pdev, mmrbc);
4491 void e1000_io_write(struct e1000_hw *hw, unsigned long port, u32 value)
4493 outl(value, port);
4496 static void e1000_vlan_rx_register(struct net_device *netdev,
4497 struct vlan_group *grp)
4499 struct e1000_adapter *adapter = netdev_priv(netdev);
4500 struct e1000_hw *hw = &adapter->hw;
4501 u32 ctrl, rctl;
4503 if (!test_bit(__E1000_DOWN, &adapter->flags))
4504 e1000_irq_disable(adapter);
4505 adapter->vlgrp = grp;
4507 if (grp) {
4508 /* enable VLAN tag insert/strip */
4509 ctrl = er32(CTRL);
4510 ctrl |= E1000_CTRL_VME;
4511 ew32(CTRL, ctrl);
4513 /* enable VLAN receive filtering */
4514 rctl = er32(RCTL);
4515 rctl &= ~E1000_RCTL_CFIEN;
4516 if (!(netdev->flags & IFF_PROMISC))
4517 rctl |= E1000_RCTL_VFE;
4518 ew32(RCTL, rctl);
4519 e1000_update_mng_vlan(adapter);
4520 } else {
4521 /* disable VLAN tag insert/strip */
4522 ctrl = er32(CTRL);
4523 ctrl &= ~E1000_CTRL_VME;
4524 ew32(CTRL, ctrl);
4526 /* disable VLAN receive filtering */
4527 rctl = er32(RCTL);
4528 rctl &= ~E1000_RCTL_VFE;
4529 ew32(RCTL, rctl);
4531 if (adapter->mng_vlan_id != (u16)E1000_MNG_VLAN_NONE) {
4532 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
4533 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
4537 if (!test_bit(__E1000_DOWN, &adapter->flags))
4538 e1000_irq_enable(adapter);
4541 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
4543 struct e1000_adapter *adapter = netdev_priv(netdev);
4544 struct e1000_hw *hw = &adapter->hw;
4545 u32 vfta, index;
4547 if ((hw->mng_cookie.status &
4548 E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) &&
4549 (vid == adapter->mng_vlan_id))
4550 return;
4551 /* add VID to filter table */
4552 index = (vid >> 5) & 0x7F;
4553 vfta = E1000_READ_REG_ARRAY(hw, VFTA, index);
4554 vfta |= (1 << (vid & 0x1F));
4555 e1000_write_vfta(hw, index, vfta);
4558 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
4560 struct e1000_adapter *adapter = netdev_priv(netdev);
4561 struct e1000_hw *hw = &adapter->hw;
4562 u32 vfta, index;
4564 if (!test_bit(__E1000_DOWN, &adapter->flags))
4565 e1000_irq_disable(adapter);
4566 vlan_group_set_device(adapter->vlgrp, vid, NULL);
4567 if (!test_bit(__E1000_DOWN, &adapter->flags))
4568 e1000_irq_enable(adapter);
4570 /* remove VID from filter table */
4571 index = (vid >> 5) & 0x7F;
4572 vfta = E1000_READ_REG_ARRAY(hw, VFTA, index);
4573 vfta &= ~(1 << (vid & 0x1F));
4574 e1000_write_vfta(hw, index, vfta);
4577 static void e1000_restore_vlan(struct e1000_adapter *adapter)
4579 e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
4581 if (adapter->vlgrp) {
4582 u16 vid;
4583 for (vid = 0; vid < VLAN_N_VID; vid++) {
4584 if (!vlan_group_get_device(adapter->vlgrp, vid))
4585 continue;
4586 e1000_vlan_rx_add_vid(adapter->netdev, vid);
4591 int e1000_set_spd_dplx(struct e1000_adapter *adapter, u16 spddplx)
4593 struct e1000_hw *hw = &adapter->hw;
4595 hw->autoneg = 0;
4597 /* Fiber NICs only allow 1000 gbps Full duplex */
4598 if ((hw->media_type == e1000_media_type_fiber) &&
4599 spddplx != (SPEED_1000 + DUPLEX_FULL)) {
4600 e_err(probe, "Unsupported Speed/Duplex configuration\n");
4601 return -EINVAL;
4604 switch (spddplx) {
4605 case SPEED_10 + DUPLEX_HALF:
4606 hw->forced_speed_duplex = e1000_10_half;
4607 break;
4608 case SPEED_10 + DUPLEX_FULL:
4609 hw->forced_speed_duplex = e1000_10_full;
4610 break;
4611 case SPEED_100 + DUPLEX_HALF:
4612 hw->forced_speed_duplex = e1000_100_half;
4613 break;
4614 case SPEED_100 + DUPLEX_FULL:
4615 hw->forced_speed_duplex = e1000_100_full;
4616 break;
4617 case SPEED_1000 + DUPLEX_FULL:
4618 hw->autoneg = 1;
4619 hw->autoneg_advertised = ADVERTISE_1000_FULL;
4620 break;
4621 case SPEED_1000 + DUPLEX_HALF: /* not supported */
4622 default:
4623 e_err(probe, "Unsupported Speed/Duplex configuration\n");
4624 return -EINVAL;
4626 return 0;
4629 static int __e1000_shutdown(struct pci_dev *pdev, bool *enable_wake)
4631 struct net_device *netdev = pci_get_drvdata(pdev);
4632 struct e1000_adapter *adapter = netdev_priv(netdev);
4633 struct e1000_hw *hw = &adapter->hw;
4634 u32 ctrl, ctrl_ext, rctl, status;
4635 u32 wufc = adapter->wol;
4636 #ifdef CONFIG_PM
4637 int retval = 0;
4638 #endif
4640 netif_device_detach(netdev);
4642 if (netif_running(netdev)) {
4643 WARN_ON(test_bit(__E1000_RESETTING, &adapter->flags));
4644 e1000_down(adapter);
4647 #ifdef CONFIG_PM
4648 retval = pci_save_state(pdev);
4649 if (retval)
4650 return retval;
4651 #endif
4653 status = er32(STATUS);
4654 if (status & E1000_STATUS_LU)
4655 wufc &= ~E1000_WUFC_LNKC;
4657 if (wufc) {
4658 e1000_setup_rctl(adapter);
4659 e1000_set_rx_mode(netdev);
4661 /* turn on all-multi mode if wake on multicast is enabled */
4662 if (wufc & E1000_WUFC_MC) {
4663 rctl = er32(RCTL);
4664 rctl |= E1000_RCTL_MPE;
4665 ew32(RCTL, rctl);
4668 if (hw->mac_type >= e1000_82540) {
4669 ctrl = er32(CTRL);
4670 /* advertise wake from D3Cold */
4671 #define E1000_CTRL_ADVD3WUC 0x00100000
4672 /* phy power management enable */
4673 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
4674 ctrl |= E1000_CTRL_ADVD3WUC |
4675 E1000_CTRL_EN_PHY_PWR_MGMT;
4676 ew32(CTRL, ctrl);
4679 if (hw->media_type == e1000_media_type_fiber ||
4680 hw->media_type == e1000_media_type_internal_serdes) {
4681 /* keep the laser running in D3 */
4682 ctrl_ext = er32(CTRL_EXT);
4683 ctrl_ext |= E1000_CTRL_EXT_SDP7_DATA;
4684 ew32(CTRL_EXT, ctrl_ext);
4687 ew32(WUC, E1000_WUC_PME_EN);
4688 ew32(WUFC, wufc);
4689 } else {
4690 ew32(WUC, 0);
4691 ew32(WUFC, 0);
4694 e1000_release_manageability(adapter);
4696 *enable_wake = !!wufc;
4698 /* make sure adapter isn't asleep if manageability is enabled */
4699 if (adapter->en_mng_pt)
4700 *enable_wake = true;
4702 if (netif_running(netdev))
4703 e1000_free_irq(adapter);
4705 pci_disable_device(pdev);
4707 return 0;
4710 #ifdef CONFIG_PM
4711 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state)
4713 int retval;
4714 bool wake;
4716 retval = __e1000_shutdown(pdev, &wake);
4717 if (retval)
4718 return retval;
4720 if (wake) {
4721 pci_prepare_to_sleep(pdev);
4722 } else {
4723 pci_wake_from_d3(pdev, false);
4724 pci_set_power_state(pdev, PCI_D3hot);
4727 return 0;
4730 static int e1000_resume(struct pci_dev *pdev)
4732 struct net_device *netdev = pci_get_drvdata(pdev);
4733 struct e1000_adapter *adapter = netdev_priv(netdev);
4734 struct e1000_hw *hw = &adapter->hw;
4735 u32 err;
4737 pci_set_power_state(pdev, PCI_D0);
4738 pci_restore_state(pdev);
4739 pci_save_state(pdev);
4741 if (adapter->need_ioport)
4742 err = pci_enable_device(pdev);
4743 else
4744 err = pci_enable_device_mem(pdev);
4745 if (err) {
4746 pr_err("Cannot enable PCI device from suspend\n");
4747 return err;
4749 pci_set_master(pdev);
4751 pci_enable_wake(pdev, PCI_D3hot, 0);
4752 pci_enable_wake(pdev, PCI_D3cold, 0);
4754 if (netif_running(netdev)) {
4755 err = e1000_request_irq(adapter);
4756 if (err)
4757 return err;
4760 e1000_power_up_phy(adapter);
4761 e1000_reset(adapter);
4762 ew32(WUS, ~0);
4764 e1000_init_manageability(adapter);
4766 if (netif_running(netdev))
4767 e1000_up(adapter);
4769 netif_device_attach(netdev);
4771 return 0;
4773 #endif
4775 static void e1000_shutdown(struct pci_dev *pdev)
4777 bool wake;
4779 __e1000_shutdown(pdev, &wake);
4781 if (system_state == SYSTEM_POWER_OFF) {
4782 pci_wake_from_d3(pdev, wake);
4783 pci_set_power_state(pdev, PCI_D3hot);
4787 #ifdef CONFIG_NET_POLL_CONTROLLER
4789 * Polling 'interrupt' - used by things like netconsole to send skbs
4790 * without having to re-enable interrupts. It's not called while
4791 * the interrupt routine is executing.
4793 static void e1000_netpoll(struct net_device *netdev)
4795 struct e1000_adapter *adapter = netdev_priv(netdev);
4797 disable_irq(adapter->pdev->irq);
4798 e1000_intr(adapter->pdev->irq, netdev);
4799 enable_irq(adapter->pdev->irq);
4801 #endif
4804 * e1000_io_error_detected - called when PCI error is detected
4805 * @pdev: Pointer to PCI device
4806 * @state: The current pci connection state
4808 * This function is called after a PCI bus error affecting
4809 * this device has been detected.
4811 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
4812 pci_channel_state_t state)
4814 struct net_device *netdev = pci_get_drvdata(pdev);
4815 struct e1000_adapter *adapter = netdev_priv(netdev);
4817 netif_device_detach(netdev);
4819 if (state == pci_channel_io_perm_failure)
4820 return PCI_ERS_RESULT_DISCONNECT;
4822 if (netif_running(netdev))
4823 e1000_down(adapter);
4824 pci_disable_device(pdev);
4826 /* Request a slot slot reset. */
4827 return PCI_ERS_RESULT_NEED_RESET;
4831 * e1000_io_slot_reset - called after the pci bus has been reset.
4832 * @pdev: Pointer to PCI device
4834 * Restart the card from scratch, as if from a cold-boot. Implementation
4835 * resembles the first-half of the e1000_resume routine.
4837 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
4839 struct net_device *netdev = pci_get_drvdata(pdev);
4840 struct e1000_adapter *adapter = netdev_priv(netdev);
4841 struct e1000_hw *hw = &adapter->hw;
4842 int err;
4844 if (adapter->need_ioport)
4845 err = pci_enable_device(pdev);
4846 else
4847 err = pci_enable_device_mem(pdev);
4848 if (err) {
4849 pr_err("Cannot re-enable PCI device after reset.\n");
4850 return PCI_ERS_RESULT_DISCONNECT;
4852 pci_set_master(pdev);
4854 pci_enable_wake(pdev, PCI_D3hot, 0);
4855 pci_enable_wake(pdev, PCI_D3cold, 0);
4857 e1000_reset(adapter);
4858 ew32(WUS, ~0);
4860 return PCI_ERS_RESULT_RECOVERED;
4864 * e1000_io_resume - called when traffic can start flowing again.
4865 * @pdev: Pointer to PCI device
4867 * This callback is called when the error recovery driver tells us that
4868 * its OK to resume normal operation. Implementation resembles the
4869 * second-half of the e1000_resume routine.
4871 static void e1000_io_resume(struct pci_dev *pdev)
4873 struct net_device *netdev = pci_get_drvdata(pdev);
4874 struct e1000_adapter *adapter = netdev_priv(netdev);
4876 e1000_init_manageability(adapter);
4878 if (netif_running(netdev)) {
4879 if (e1000_up(adapter)) {
4880 pr_info("can't bring device back up after reset\n");
4881 return;
4885 netif_device_attach(netdev);
4888 /* e1000_main.c */