Merge tag 'v3.3.7' into 3.3/master
[zen-stable.git] / drivers / net / ethernet / intel / e1000 / e1000_main.c
blobb72d97604258359ca7284a27777592f249d9d56d
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>
32 #include <linux/prefetch.h>
33 #include <linux/bitops.h>
34 #include <linux/if_vlan.h>
36 char e1000_driver_name[] = "e1000";
37 static char e1000_driver_string[] = "Intel(R) PRO/1000 Network Driver";
38 #define DRV_VERSION "7.3.21-k8-NAPI"
39 const char e1000_driver_version[] = DRV_VERSION;
40 static const char e1000_copyright[] = "Copyright (c) 1999-2006 Intel Corporation.";
42 /* e1000_pci_tbl - PCI Device ID Table
44 * Last entry must be all 0s
46 * Macro expands to...
47 * {PCI_DEVICE(PCI_VENDOR_ID_INTEL, device_id)}
49 static DEFINE_PCI_DEVICE_TABLE(e1000_pci_tbl) = {
50 INTEL_E1000_ETHERNET_DEVICE(0x1000),
51 INTEL_E1000_ETHERNET_DEVICE(0x1001),
52 INTEL_E1000_ETHERNET_DEVICE(0x1004),
53 INTEL_E1000_ETHERNET_DEVICE(0x1008),
54 INTEL_E1000_ETHERNET_DEVICE(0x1009),
55 INTEL_E1000_ETHERNET_DEVICE(0x100C),
56 INTEL_E1000_ETHERNET_DEVICE(0x100D),
57 INTEL_E1000_ETHERNET_DEVICE(0x100E),
58 INTEL_E1000_ETHERNET_DEVICE(0x100F),
59 INTEL_E1000_ETHERNET_DEVICE(0x1010),
60 INTEL_E1000_ETHERNET_DEVICE(0x1011),
61 INTEL_E1000_ETHERNET_DEVICE(0x1012),
62 INTEL_E1000_ETHERNET_DEVICE(0x1013),
63 INTEL_E1000_ETHERNET_DEVICE(0x1014),
64 INTEL_E1000_ETHERNET_DEVICE(0x1015),
65 INTEL_E1000_ETHERNET_DEVICE(0x1016),
66 INTEL_E1000_ETHERNET_DEVICE(0x1017),
67 INTEL_E1000_ETHERNET_DEVICE(0x1018),
68 INTEL_E1000_ETHERNET_DEVICE(0x1019),
69 INTEL_E1000_ETHERNET_DEVICE(0x101A),
70 INTEL_E1000_ETHERNET_DEVICE(0x101D),
71 INTEL_E1000_ETHERNET_DEVICE(0x101E),
72 INTEL_E1000_ETHERNET_DEVICE(0x1026),
73 INTEL_E1000_ETHERNET_DEVICE(0x1027),
74 INTEL_E1000_ETHERNET_DEVICE(0x1028),
75 INTEL_E1000_ETHERNET_DEVICE(0x1075),
76 INTEL_E1000_ETHERNET_DEVICE(0x1076),
77 INTEL_E1000_ETHERNET_DEVICE(0x1077),
78 INTEL_E1000_ETHERNET_DEVICE(0x1078),
79 INTEL_E1000_ETHERNET_DEVICE(0x1079),
80 INTEL_E1000_ETHERNET_DEVICE(0x107A),
81 INTEL_E1000_ETHERNET_DEVICE(0x107B),
82 INTEL_E1000_ETHERNET_DEVICE(0x107C),
83 INTEL_E1000_ETHERNET_DEVICE(0x108A),
84 INTEL_E1000_ETHERNET_DEVICE(0x1099),
85 INTEL_E1000_ETHERNET_DEVICE(0x10B5),
86 INTEL_E1000_ETHERNET_DEVICE(0x2E6E),
87 /* required last entry */
88 {0,}
91 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
93 int e1000_up(struct e1000_adapter *adapter);
94 void e1000_down(struct e1000_adapter *adapter);
95 void e1000_reinit_locked(struct e1000_adapter *adapter);
96 void e1000_reset(struct e1000_adapter *adapter);
97 int e1000_setup_all_tx_resources(struct e1000_adapter *adapter);
98 int e1000_setup_all_rx_resources(struct e1000_adapter *adapter);
99 void e1000_free_all_tx_resources(struct e1000_adapter *adapter);
100 void e1000_free_all_rx_resources(struct e1000_adapter *adapter);
101 static int e1000_setup_tx_resources(struct e1000_adapter *adapter,
102 struct e1000_tx_ring *txdr);
103 static int e1000_setup_rx_resources(struct e1000_adapter *adapter,
104 struct e1000_rx_ring *rxdr);
105 static void e1000_free_tx_resources(struct e1000_adapter *adapter,
106 struct e1000_tx_ring *tx_ring);
107 static void e1000_free_rx_resources(struct e1000_adapter *adapter,
108 struct e1000_rx_ring *rx_ring);
109 void e1000_update_stats(struct e1000_adapter *adapter);
111 static int e1000_init_module(void);
112 static void e1000_exit_module(void);
113 static int e1000_probe(struct pci_dev *pdev, const struct pci_device_id *ent);
114 static void __devexit e1000_remove(struct pci_dev *pdev);
115 static int e1000_alloc_queues(struct e1000_adapter *adapter);
116 static int e1000_sw_init(struct e1000_adapter *adapter);
117 static int e1000_open(struct net_device *netdev);
118 static int e1000_close(struct net_device *netdev);
119 static void e1000_configure_tx(struct e1000_adapter *adapter);
120 static void e1000_configure_rx(struct e1000_adapter *adapter);
121 static void e1000_setup_rctl(struct e1000_adapter *adapter);
122 static void e1000_clean_all_tx_rings(struct e1000_adapter *adapter);
123 static void e1000_clean_all_rx_rings(struct e1000_adapter *adapter);
124 static void e1000_clean_tx_ring(struct e1000_adapter *adapter,
125 struct e1000_tx_ring *tx_ring);
126 static void e1000_clean_rx_ring(struct e1000_adapter *adapter,
127 struct e1000_rx_ring *rx_ring);
128 static void e1000_set_rx_mode(struct net_device *netdev);
129 static void e1000_update_phy_info_task(struct work_struct *work);
130 static void e1000_watchdog(struct work_struct *work);
131 static void e1000_82547_tx_fifo_stall_task(struct work_struct *work);
132 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
133 struct net_device *netdev);
134 static struct net_device_stats * e1000_get_stats(struct net_device *netdev);
135 static int e1000_change_mtu(struct net_device *netdev, int new_mtu);
136 static int e1000_set_mac(struct net_device *netdev, void *p);
137 static irqreturn_t e1000_intr(int irq, void *data);
138 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter,
139 struct e1000_tx_ring *tx_ring);
140 static int e1000_clean(struct napi_struct *napi, int budget);
141 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
142 struct e1000_rx_ring *rx_ring,
143 int *work_done, int work_to_do);
144 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
145 struct e1000_rx_ring *rx_ring,
146 int *work_done, int work_to_do);
147 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
148 struct e1000_rx_ring *rx_ring,
149 int cleaned_count);
150 static void e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
151 struct e1000_rx_ring *rx_ring,
152 int cleaned_count);
153 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd);
154 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
155 int cmd);
156 static void e1000_enter_82542_rst(struct e1000_adapter *adapter);
157 static void e1000_leave_82542_rst(struct e1000_adapter *adapter);
158 static void e1000_tx_timeout(struct net_device *dev);
159 static void e1000_reset_task(struct work_struct *work);
160 static void e1000_smartspeed(struct e1000_adapter *adapter);
161 static int e1000_82547_fifo_workaround(struct e1000_adapter *adapter,
162 struct sk_buff *skb);
164 static bool e1000_vlan_used(struct e1000_adapter *adapter);
165 static void e1000_vlan_mode(struct net_device *netdev,
166 netdev_features_t features);
167 static void e1000_vlan_filter_on_off(struct e1000_adapter *adapter,
168 bool filter_on);
169 static int e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid);
170 static int e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid);
171 static void e1000_restore_vlan(struct e1000_adapter *adapter);
173 #ifdef CONFIG_PM
174 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state);
175 static int e1000_resume(struct pci_dev *pdev);
176 #endif
177 static void e1000_shutdown(struct pci_dev *pdev);
179 #ifdef CONFIG_NET_POLL_CONTROLLER
180 /* for netdump / net console */
181 static void e1000_netpoll (struct net_device *netdev);
182 #endif
184 #define COPYBREAK_DEFAULT 256
185 static unsigned int copybreak __read_mostly = COPYBREAK_DEFAULT;
186 module_param(copybreak, uint, 0644);
187 MODULE_PARM_DESC(copybreak,
188 "Maximum size of packet that is copied to a new buffer on receive");
190 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
191 pci_channel_state_t state);
192 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev);
193 static void e1000_io_resume(struct pci_dev *pdev);
195 static struct pci_error_handlers e1000_err_handler = {
196 .error_detected = e1000_io_error_detected,
197 .slot_reset = e1000_io_slot_reset,
198 .resume = e1000_io_resume,
201 static struct pci_driver e1000_driver = {
202 .name = e1000_driver_name,
203 .id_table = e1000_pci_tbl,
204 .probe = e1000_probe,
205 .remove = __devexit_p(e1000_remove),
206 #ifdef CONFIG_PM
207 /* Power Management Hooks */
208 .suspend = e1000_suspend,
209 .resume = e1000_resume,
210 #endif
211 .shutdown = e1000_shutdown,
212 .err_handler = &e1000_err_handler
215 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
216 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
217 MODULE_LICENSE("GPL");
218 MODULE_VERSION(DRV_VERSION);
220 static int debug = NETIF_MSG_DRV | NETIF_MSG_PROBE;
221 module_param(debug, int, 0);
222 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
225 * e1000_get_hw_dev - return device
226 * used by hardware layer to print debugging information
229 struct net_device *e1000_get_hw_dev(struct e1000_hw *hw)
231 struct e1000_adapter *adapter = hw->back;
232 return adapter->netdev;
236 * e1000_init_module - Driver Registration Routine
238 * e1000_init_module is the first routine called when the driver is
239 * loaded. All it does is register with the PCI subsystem.
242 static int __init e1000_init_module(void)
244 int ret;
245 pr_info("%s - version %s\n", e1000_driver_string, e1000_driver_version);
247 pr_info("%s\n", e1000_copyright);
249 ret = pci_register_driver(&e1000_driver);
250 if (copybreak != COPYBREAK_DEFAULT) {
251 if (copybreak == 0)
252 pr_info("copybreak disabled\n");
253 else
254 pr_info("copybreak enabled for "
255 "packets <= %u bytes\n", copybreak);
257 return ret;
260 module_init(e1000_init_module);
263 * e1000_exit_module - Driver Exit Cleanup Routine
265 * e1000_exit_module is called just before the driver is removed
266 * from memory.
269 static void __exit e1000_exit_module(void)
271 pci_unregister_driver(&e1000_driver);
274 module_exit(e1000_exit_module);
276 static int e1000_request_irq(struct e1000_adapter *adapter)
278 struct net_device *netdev = adapter->netdev;
279 irq_handler_t handler = e1000_intr;
280 int irq_flags = IRQF_SHARED;
281 int err;
283 err = request_irq(adapter->pdev->irq, handler, irq_flags, netdev->name,
284 netdev);
285 if (err) {
286 e_err(probe, "Unable to allocate interrupt Error: %d\n", err);
289 return err;
292 static void e1000_free_irq(struct e1000_adapter *adapter)
294 struct net_device *netdev = adapter->netdev;
296 free_irq(adapter->pdev->irq, netdev);
300 * e1000_irq_disable - Mask off interrupt generation on the NIC
301 * @adapter: board private structure
304 static void e1000_irq_disable(struct e1000_adapter *adapter)
306 struct e1000_hw *hw = &adapter->hw;
308 ew32(IMC, ~0);
309 E1000_WRITE_FLUSH();
310 synchronize_irq(adapter->pdev->irq);
314 * e1000_irq_enable - Enable default interrupt generation settings
315 * @adapter: board private structure
318 static void e1000_irq_enable(struct e1000_adapter *adapter)
320 struct e1000_hw *hw = &adapter->hw;
322 ew32(IMS, IMS_ENABLE_MASK);
323 E1000_WRITE_FLUSH();
326 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
328 struct e1000_hw *hw = &adapter->hw;
329 struct net_device *netdev = adapter->netdev;
330 u16 vid = hw->mng_cookie.vlan_id;
331 u16 old_vid = adapter->mng_vlan_id;
333 if (!e1000_vlan_used(adapter))
334 return;
336 if (!test_bit(vid, adapter->active_vlans)) {
337 if (hw->mng_cookie.status &
338 E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) {
339 e1000_vlan_rx_add_vid(netdev, vid);
340 adapter->mng_vlan_id = vid;
341 } else {
342 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
344 if ((old_vid != (u16)E1000_MNG_VLAN_NONE) &&
345 (vid != old_vid) &&
346 !test_bit(old_vid, adapter->active_vlans))
347 e1000_vlan_rx_kill_vid(netdev, old_vid);
348 } else {
349 adapter->mng_vlan_id = vid;
353 static void e1000_init_manageability(struct e1000_adapter *adapter)
355 struct e1000_hw *hw = &adapter->hw;
357 if (adapter->en_mng_pt) {
358 u32 manc = er32(MANC);
360 /* disable hardware interception of ARP */
361 manc &= ~(E1000_MANC_ARP_EN);
363 ew32(MANC, manc);
367 static void e1000_release_manageability(struct e1000_adapter *adapter)
369 struct e1000_hw *hw = &adapter->hw;
371 if (adapter->en_mng_pt) {
372 u32 manc = er32(MANC);
374 /* re-enable hardware interception of ARP */
375 manc |= E1000_MANC_ARP_EN;
377 ew32(MANC, manc);
382 * e1000_configure - configure the hardware for RX and TX
383 * @adapter = private board structure
385 static void e1000_configure(struct e1000_adapter *adapter)
387 struct net_device *netdev = adapter->netdev;
388 int i;
390 e1000_set_rx_mode(netdev);
392 e1000_restore_vlan(adapter);
393 e1000_init_manageability(adapter);
395 e1000_configure_tx(adapter);
396 e1000_setup_rctl(adapter);
397 e1000_configure_rx(adapter);
398 /* call E1000_DESC_UNUSED which always leaves
399 * at least 1 descriptor unused to make sure
400 * next_to_use != next_to_clean */
401 for (i = 0; i < adapter->num_rx_queues; i++) {
402 struct e1000_rx_ring *ring = &adapter->rx_ring[i];
403 adapter->alloc_rx_buf(adapter, ring,
404 E1000_DESC_UNUSED(ring));
408 int e1000_up(struct e1000_adapter *adapter)
410 struct e1000_hw *hw = &adapter->hw;
412 /* hardware has been reset, we need to reload some things */
413 e1000_configure(adapter);
415 clear_bit(__E1000_DOWN, &adapter->flags);
417 napi_enable(&adapter->napi);
419 e1000_irq_enable(adapter);
421 netif_wake_queue(adapter->netdev);
423 /* fire a link change interrupt to start the watchdog */
424 ew32(ICS, E1000_ICS_LSC);
425 return 0;
429 * e1000_power_up_phy - restore link in case the phy was powered down
430 * @adapter: address of board private structure
432 * The phy may be powered down to save power and turn off link when the
433 * driver is unloaded and wake on lan is not enabled (among others)
434 * *** this routine MUST be followed by a call to e1000_reset ***
438 void e1000_power_up_phy(struct e1000_adapter *adapter)
440 struct e1000_hw *hw = &adapter->hw;
441 u16 mii_reg = 0;
443 /* Just clear the power down bit to wake the phy back up */
444 if (hw->media_type == e1000_media_type_copper) {
445 /* according to the manual, the phy will retain its
446 * settings across a power-down/up cycle */
447 e1000_read_phy_reg(hw, PHY_CTRL, &mii_reg);
448 mii_reg &= ~MII_CR_POWER_DOWN;
449 e1000_write_phy_reg(hw, PHY_CTRL, mii_reg);
453 static void e1000_power_down_phy(struct e1000_adapter *adapter)
455 struct e1000_hw *hw = &adapter->hw;
457 /* Power down the PHY so no link is implied when interface is down *
458 * The PHY cannot be powered down if any of the following is true *
459 * (a) WoL is enabled
460 * (b) AMT is active
461 * (c) SoL/IDER session is active */
462 if (!adapter->wol && hw->mac_type >= e1000_82540 &&
463 hw->media_type == e1000_media_type_copper) {
464 u16 mii_reg = 0;
466 switch (hw->mac_type) {
467 case e1000_82540:
468 case e1000_82545:
469 case e1000_82545_rev_3:
470 case e1000_82546:
471 case e1000_ce4100:
472 case e1000_82546_rev_3:
473 case e1000_82541:
474 case e1000_82541_rev_2:
475 case e1000_82547:
476 case e1000_82547_rev_2:
477 if (er32(MANC) & E1000_MANC_SMBUS_EN)
478 goto out;
479 break;
480 default:
481 goto out;
483 e1000_read_phy_reg(hw, PHY_CTRL, &mii_reg);
484 mii_reg |= MII_CR_POWER_DOWN;
485 e1000_write_phy_reg(hw, PHY_CTRL, mii_reg);
486 msleep(1);
488 out:
489 return;
492 static void e1000_down_and_stop(struct e1000_adapter *adapter)
494 set_bit(__E1000_DOWN, &adapter->flags);
496 /* Only kill reset task if adapter is not resetting */
497 if (!test_bit(__E1000_RESETTING, &adapter->flags))
498 cancel_work_sync(&adapter->reset_task);
500 cancel_delayed_work_sync(&adapter->watchdog_task);
501 cancel_delayed_work_sync(&adapter->phy_info_task);
502 cancel_delayed_work_sync(&adapter->fifo_stall_task);
505 void e1000_down(struct e1000_adapter *adapter)
507 struct e1000_hw *hw = &adapter->hw;
508 struct net_device *netdev = adapter->netdev;
509 u32 rctl, tctl;
512 /* disable receives in the hardware */
513 rctl = er32(RCTL);
514 ew32(RCTL, rctl & ~E1000_RCTL_EN);
515 /* flush and sleep below */
517 netif_tx_disable(netdev);
519 /* disable transmits in the hardware */
520 tctl = er32(TCTL);
521 tctl &= ~E1000_TCTL_EN;
522 ew32(TCTL, tctl);
523 /* flush both disables and wait for them to finish */
524 E1000_WRITE_FLUSH();
525 msleep(10);
527 napi_disable(&adapter->napi);
529 e1000_irq_disable(adapter);
532 * Setting DOWN must be after irq_disable to prevent
533 * a screaming interrupt. Setting DOWN also prevents
534 * tasks from rescheduling.
536 e1000_down_and_stop(adapter);
538 adapter->link_speed = 0;
539 adapter->link_duplex = 0;
540 netif_carrier_off(netdev);
542 e1000_reset(adapter);
543 e1000_clean_all_tx_rings(adapter);
544 e1000_clean_all_rx_rings(adapter);
547 static void e1000_reinit_safe(struct e1000_adapter *adapter)
549 while (test_and_set_bit(__E1000_RESETTING, &adapter->flags))
550 msleep(1);
551 mutex_lock(&adapter->mutex);
552 e1000_down(adapter);
553 e1000_up(adapter);
554 mutex_unlock(&adapter->mutex);
555 clear_bit(__E1000_RESETTING, &adapter->flags);
558 void e1000_reinit_locked(struct e1000_adapter *adapter)
560 /* if rtnl_lock is not held the call path is bogus */
561 ASSERT_RTNL();
562 WARN_ON(in_interrupt());
563 while (test_and_set_bit(__E1000_RESETTING, &adapter->flags))
564 msleep(1);
565 e1000_down(adapter);
566 e1000_up(adapter);
567 clear_bit(__E1000_RESETTING, &adapter->flags);
570 void e1000_reset(struct e1000_adapter *adapter)
572 struct e1000_hw *hw = &adapter->hw;
573 u32 pba = 0, tx_space, min_tx_space, min_rx_space;
574 bool legacy_pba_adjust = false;
575 u16 hwm;
577 /* Repartition Pba for greater than 9k mtu
578 * To take effect CTRL.RST is required.
581 switch (hw->mac_type) {
582 case e1000_82542_rev2_0:
583 case e1000_82542_rev2_1:
584 case e1000_82543:
585 case e1000_82544:
586 case e1000_82540:
587 case e1000_82541:
588 case e1000_82541_rev_2:
589 legacy_pba_adjust = true;
590 pba = E1000_PBA_48K;
591 break;
592 case e1000_82545:
593 case e1000_82545_rev_3:
594 case e1000_82546:
595 case e1000_ce4100:
596 case e1000_82546_rev_3:
597 pba = E1000_PBA_48K;
598 break;
599 case e1000_82547:
600 case e1000_82547_rev_2:
601 legacy_pba_adjust = true;
602 pba = E1000_PBA_30K;
603 break;
604 case e1000_undefined:
605 case e1000_num_macs:
606 break;
609 if (legacy_pba_adjust) {
610 if (hw->max_frame_size > E1000_RXBUFFER_8192)
611 pba -= 8; /* allocate more FIFO for Tx */
613 if (hw->mac_type == e1000_82547) {
614 adapter->tx_fifo_head = 0;
615 adapter->tx_head_addr = pba << E1000_TX_HEAD_ADDR_SHIFT;
616 adapter->tx_fifo_size =
617 (E1000_PBA_40K - pba) << E1000_PBA_BYTES_SHIFT;
618 atomic_set(&adapter->tx_fifo_stall, 0);
620 } else if (hw->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
621 /* adjust PBA for jumbo frames */
622 ew32(PBA, pba);
624 /* To maintain wire speed transmits, the Tx FIFO should be
625 * large enough to accommodate two full transmit packets,
626 * rounded up to the next 1KB and expressed in KB. Likewise,
627 * the Rx FIFO should be large enough to accommodate at least
628 * one full receive packet and is similarly rounded up and
629 * expressed in KB. */
630 pba = er32(PBA);
631 /* upper 16 bits has Tx packet buffer allocation size in KB */
632 tx_space = pba >> 16;
633 /* lower 16 bits has Rx packet buffer allocation size in KB */
634 pba &= 0xffff;
636 * the tx fifo also stores 16 bytes of information about the tx
637 * but don't include ethernet FCS because hardware appends it
639 min_tx_space = (hw->max_frame_size +
640 sizeof(struct e1000_tx_desc) -
641 ETH_FCS_LEN) * 2;
642 min_tx_space = ALIGN(min_tx_space, 1024);
643 min_tx_space >>= 10;
644 /* software strips receive CRC, so leave room for it */
645 min_rx_space = hw->max_frame_size;
646 min_rx_space = ALIGN(min_rx_space, 1024);
647 min_rx_space >>= 10;
649 /* If current Tx allocation is less than the min Tx FIFO size,
650 * and the min Tx FIFO size is less than the current Rx FIFO
651 * allocation, take space away from current Rx allocation */
652 if (tx_space < min_tx_space &&
653 ((min_tx_space - tx_space) < pba)) {
654 pba = pba - (min_tx_space - tx_space);
656 /* PCI/PCIx hardware has PBA alignment constraints */
657 switch (hw->mac_type) {
658 case e1000_82545 ... e1000_82546_rev_3:
659 pba &= ~(E1000_PBA_8K - 1);
660 break;
661 default:
662 break;
665 /* if short on rx space, rx wins and must trump tx
666 * adjustment or use Early Receive if available */
667 if (pba < min_rx_space)
668 pba = min_rx_space;
672 ew32(PBA, pba);
675 * flow control settings:
676 * The high water mark must be low enough to fit one full frame
677 * (or the size used for early receive) above it in the Rx FIFO.
678 * Set it to the lower of:
679 * - 90% of the Rx FIFO size, and
680 * - the full Rx FIFO size minus the early receive size (for parts
681 * with ERT support assuming ERT set to E1000_ERT_2048), or
682 * - the full Rx FIFO size minus one full frame
684 hwm = min(((pba << 10) * 9 / 10),
685 ((pba << 10) - hw->max_frame_size));
687 hw->fc_high_water = hwm & 0xFFF8; /* 8-byte granularity */
688 hw->fc_low_water = hw->fc_high_water - 8;
689 hw->fc_pause_time = E1000_FC_PAUSE_TIME;
690 hw->fc_send_xon = 1;
691 hw->fc = hw->original_fc;
693 /* Allow time for pending master requests to run */
694 e1000_reset_hw(hw);
695 if (hw->mac_type >= e1000_82544)
696 ew32(WUC, 0);
698 if (e1000_init_hw(hw))
699 e_dev_err("Hardware Error\n");
700 e1000_update_mng_vlan(adapter);
702 /* if (adapter->hwflags & HWFLAGS_PHY_PWR_BIT) { */
703 if (hw->mac_type >= e1000_82544 &&
704 hw->autoneg == 1 &&
705 hw->autoneg_advertised == ADVERTISE_1000_FULL) {
706 u32 ctrl = er32(CTRL);
707 /* clear phy power management bit if we are in gig only mode,
708 * which if enabled will attempt negotiation to 100Mb, which
709 * can cause a loss of link at power off or driver unload */
710 ctrl &= ~E1000_CTRL_SWDPIN3;
711 ew32(CTRL, ctrl);
714 /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
715 ew32(VET, ETHERNET_IEEE_VLAN_TYPE);
717 e1000_reset_adaptive(hw);
718 e1000_phy_get_info(hw, &adapter->phy_info);
720 e1000_release_manageability(adapter);
724 * Dump the eeprom for users having checksum issues
726 static void e1000_dump_eeprom(struct e1000_adapter *adapter)
728 struct net_device *netdev = adapter->netdev;
729 struct ethtool_eeprom eeprom;
730 const struct ethtool_ops *ops = netdev->ethtool_ops;
731 u8 *data;
732 int i;
733 u16 csum_old, csum_new = 0;
735 eeprom.len = ops->get_eeprom_len(netdev);
736 eeprom.offset = 0;
738 data = kmalloc(eeprom.len, GFP_KERNEL);
739 if (!data) {
740 pr_err("Unable to allocate memory to dump EEPROM data\n");
741 return;
744 ops->get_eeprom(netdev, &eeprom, data);
746 csum_old = (data[EEPROM_CHECKSUM_REG * 2]) +
747 (data[EEPROM_CHECKSUM_REG * 2 + 1] << 8);
748 for (i = 0; i < EEPROM_CHECKSUM_REG * 2; i += 2)
749 csum_new += data[i] + (data[i + 1] << 8);
750 csum_new = EEPROM_SUM - csum_new;
752 pr_err("/*********************/\n");
753 pr_err("Current EEPROM Checksum : 0x%04x\n", csum_old);
754 pr_err("Calculated : 0x%04x\n", csum_new);
756 pr_err("Offset Values\n");
757 pr_err("======== ======\n");
758 print_hex_dump(KERN_ERR, "", DUMP_PREFIX_OFFSET, 16, 1, data, 128, 0);
760 pr_err("Include this output when contacting your support provider.\n");
761 pr_err("This is not a software error! Something bad happened to\n");
762 pr_err("your hardware or EEPROM image. Ignoring this problem could\n");
763 pr_err("result in further problems, possibly loss of data,\n");
764 pr_err("corruption or system hangs!\n");
765 pr_err("The MAC Address will be reset to 00:00:00:00:00:00,\n");
766 pr_err("which is invalid and requires you to set the proper MAC\n");
767 pr_err("address manually before continuing to enable this network\n");
768 pr_err("device. Please inspect the EEPROM dump and report the\n");
769 pr_err("issue to your hardware vendor or Intel Customer Support.\n");
770 pr_err("/*********************/\n");
772 kfree(data);
776 * e1000_is_need_ioport - determine if an adapter needs ioport resources or not
777 * @pdev: PCI device information struct
779 * Return true if an adapter needs ioport resources
781 static int e1000_is_need_ioport(struct pci_dev *pdev)
783 switch (pdev->device) {
784 case E1000_DEV_ID_82540EM:
785 case E1000_DEV_ID_82540EM_LOM:
786 case E1000_DEV_ID_82540EP:
787 case E1000_DEV_ID_82540EP_LOM:
788 case E1000_DEV_ID_82540EP_LP:
789 case E1000_DEV_ID_82541EI:
790 case E1000_DEV_ID_82541EI_MOBILE:
791 case E1000_DEV_ID_82541ER:
792 case E1000_DEV_ID_82541ER_LOM:
793 case E1000_DEV_ID_82541GI:
794 case E1000_DEV_ID_82541GI_LF:
795 case E1000_DEV_ID_82541GI_MOBILE:
796 case E1000_DEV_ID_82544EI_COPPER:
797 case E1000_DEV_ID_82544EI_FIBER:
798 case E1000_DEV_ID_82544GC_COPPER:
799 case E1000_DEV_ID_82544GC_LOM:
800 case E1000_DEV_ID_82545EM_COPPER:
801 case E1000_DEV_ID_82545EM_FIBER:
802 case E1000_DEV_ID_82546EB_COPPER:
803 case E1000_DEV_ID_82546EB_FIBER:
804 case E1000_DEV_ID_82546EB_QUAD_COPPER:
805 return true;
806 default:
807 return false;
811 static netdev_features_t e1000_fix_features(struct net_device *netdev,
812 netdev_features_t features)
815 * Since there is no support for separate rx/tx vlan accel
816 * enable/disable make sure tx flag is always in same state as rx.
818 if (features & NETIF_F_HW_VLAN_RX)
819 features |= NETIF_F_HW_VLAN_TX;
820 else
821 features &= ~NETIF_F_HW_VLAN_TX;
823 return features;
826 static int e1000_set_features(struct net_device *netdev,
827 netdev_features_t features)
829 struct e1000_adapter *adapter = netdev_priv(netdev);
830 netdev_features_t changed = features ^ netdev->features;
832 if (changed & NETIF_F_HW_VLAN_RX)
833 e1000_vlan_mode(netdev, features);
835 if (!(changed & NETIF_F_RXCSUM))
836 return 0;
838 adapter->rx_csum = !!(features & NETIF_F_RXCSUM);
840 if (netif_running(netdev))
841 e1000_reinit_locked(adapter);
842 else
843 e1000_reset(adapter);
845 return 0;
848 static const struct net_device_ops e1000_netdev_ops = {
849 .ndo_open = e1000_open,
850 .ndo_stop = e1000_close,
851 .ndo_start_xmit = e1000_xmit_frame,
852 .ndo_get_stats = e1000_get_stats,
853 .ndo_set_rx_mode = e1000_set_rx_mode,
854 .ndo_set_mac_address = e1000_set_mac,
855 .ndo_tx_timeout = e1000_tx_timeout,
856 .ndo_change_mtu = e1000_change_mtu,
857 .ndo_do_ioctl = e1000_ioctl,
858 .ndo_validate_addr = eth_validate_addr,
859 .ndo_vlan_rx_add_vid = e1000_vlan_rx_add_vid,
860 .ndo_vlan_rx_kill_vid = e1000_vlan_rx_kill_vid,
861 #ifdef CONFIG_NET_POLL_CONTROLLER
862 .ndo_poll_controller = e1000_netpoll,
863 #endif
864 .ndo_fix_features = e1000_fix_features,
865 .ndo_set_features = e1000_set_features,
869 * e1000_init_hw_struct - initialize members of hw struct
870 * @adapter: board private struct
871 * @hw: structure used by e1000_hw.c
873 * Factors out initialization of the e1000_hw struct to its own function
874 * that can be called very early at init (just after struct allocation).
875 * Fields are initialized based on PCI device information and
876 * OS network device settings (MTU size).
877 * Returns negative error codes if MAC type setup fails.
879 static int e1000_init_hw_struct(struct e1000_adapter *adapter,
880 struct e1000_hw *hw)
882 struct pci_dev *pdev = adapter->pdev;
884 /* PCI config space info */
885 hw->vendor_id = pdev->vendor;
886 hw->device_id = pdev->device;
887 hw->subsystem_vendor_id = pdev->subsystem_vendor;
888 hw->subsystem_id = pdev->subsystem_device;
889 hw->revision_id = pdev->revision;
891 pci_read_config_word(pdev, PCI_COMMAND, &hw->pci_cmd_word);
893 hw->max_frame_size = adapter->netdev->mtu +
894 ENET_HEADER_SIZE + ETHERNET_FCS_SIZE;
895 hw->min_frame_size = MINIMUM_ETHERNET_FRAME_SIZE;
897 /* identify the MAC */
898 if (e1000_set_mac_type(hw)) {
899 e_err(probe, "Unknown MAC Type\n");
900 return -EIO;
903 switch (hw->mac_type) {
904 default:
905 break;
906 case e1000_82541:
907 case e1000_82547:
908 case e1000_82541_rev_2:
909 case e1000_82547_rev_2:
910 hw->phy_init_script = 1;
911 break;
914 e1000_set_media_type(hw);
915 e1000_get_bus_info(hw);
917 hw->wait_autoneg_complete = false;
918 hw->tbi_compatibility_en = true;
919 hw->adaptive_ifs = true;
921 /* Copper options */
923 if (hw->media_type == e1000_media_type_copper) {
924 hw->mdix = AUTO_ALL_MODES;
925 hw->disable_polarity_correction = false;
926 hw->master_slave = E1000_MASTER_SLAVE;
929 return 0;
933 * e1000_probe - Device Initialization Routine
934 * @pdev: PCI device information struct
935 * @ent: entry in e1000_pci_tbl
937 * Returns 0 on success, negative on failure
939 * e1000_probe initializes an adapter identified by a pci_dev structure.
940 * The OS initialization, configuring of the adapter private structure,
941 * and a hardware reset occur.
943 static int __devinit e1000_probe(struct pci_dev *pdev,
944 const struct pci_device_id *ent)
946 struct net_device *netdev;
947 struct e1000_adapter *adapter;
948 struct e1000_hw *hw;
950 static int cards_found = 0;
951 static int global_quad_port_a = 0; /* global ksp3 port a indication */
952 int i, err, pci_using_dac;
953 u16 eeprom_data = 0;
954 u16 tmp = 0;
955 u16 eeprom_apme_mask = E1000_EEPROM_APME;
956 int bars, need_ioport;
958 /* do not allocate ioport bars when not needed */
959 need_ioport = e1000_is_need_ioport(pdev);
960 if (need_ioport) {
961 bars = pci_select_bars(pdev, IORESOURCE_MEM | IORESOURCE_IO);
962 err = pci_enable_device(pdev);
963 } else {
964 bars = pci_select_bars(pdev, IORESOURCE_MEM);
965 err = pci_enable_device_mem(pdev);
967 if (err)
968 return err;
970 err = pci_request_selected_regions(pdev, bars, e1000_driver_name);
971 if (err)
972 goto err_pci_reg;
974 pci_set_master(pdev);
975 err = pci_save_state(pdev);
976 if (err)
977 goto err_alloc_etherdev;
979 err = -ENOMEM;
980 netdev = alloc_etherdev(sizeof(struct e1000_adapter));
981 if (!netdev)
982 goto err_alloc_etherdev;
984 SET_NETDEV_DEV(netdev, &pdev->dev);
986 pci_set_drvdata(pdev, netdev);
987 adapter = netdev_priv(netdev);
988 adapter->netdev = netdev;
989 adapter->pdev = pdev;
990 adapter->msg_enable = (1 << debug) - 1;
991 adapter->bars = bars;
992 adapter->need_ioport = need_ioport;
994 hw = &adapter->hw;
995 hw->back = adapter;
997 err = -EIO;
998 hw->hw_addr = pci_ioremap_bar(pdev, BAR_0);
999 if (!hw->hw_addr)
1000 goto err_ioremap;
1002 if (adapter->need_ioport) {
1003 for (i = BAR_1; i <= BAR_5; i++) {
1004 if (pci_resource_len(pdev, i) == 0)
1005 continue;
1006 if (pci_resource_flags(pdev, i) & IORESOURCE_IO) {
1007 hw->io_base = pci_resource_start(pdev, i);
1008 break;
1013 /* make ready for any if (hw->...) below */
1014 err = e1000_init_hw_struct(adapter, hw);
1015 if (err)
1016 goto err_sw_init;
1019 * there is a workaround being applied below that limits
1020 * 64-bit DMA addresses to 64-bit hardware. There are some
1021 * 32-bit adapters that Tx hang when given 64-bit DMA addresses
1023 pci_using_dac = 0;
1024 if ((hw->bus_type == e1000_bus_type_pcix) &&
1025 !dma_set_mask(&pdev->dev, DMA_BIT_MASK(64))) {
1027 * according to DMA-API-HOWTO, coherent calls will always
1028 * succeed if the set call did
1030 dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64));
1031 pci_using_dac = 1;
1032 } else {
1033 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
1034 if (err) {
1035 pr_err("No usable DMA config, aborting\n");
1036 goto err_dma;
1038 dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(32));
1041 netdev->netdev_ops = &e1000_netdev_ops;
1042 e1000_set_ethtool_ops(netdev);
1043 netdev->watchdog_timeo = 5 * HZ;
1044 netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
1046 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
1048 adapter->bd_number = cards_found;
1050 /* setup the private structure */
1052 err = e1000_sw_init(adapter);
1053 if (err)
1054 goto err_sw_init;
1056 err = -EIO;
1057 if (hw->mac_type == e1000_ce4100) {
1058 hw->ce4100_gbe_mdio_base_virt =
1059 ioremap(pci_resource_start(pdev, BAR_1),
1060 pci_resource_len(pdev, BAR_1));
1062 if (!hw->ce4100_gbe_mdio_base_virt)
1063 goto err_mdio_ioremap;
1066 if (hw->mac_type >= e1000_82543) {
1067 netdev->hw_features = NETIF_F_SG |
1068 NETIF_F_HW_CSUM |
1069 NETIF_F_HW_VLAN_RX;
1070 netdev->features = NETIF_F_HW_VLAN_TX |
1071 NETIF_F_HW_VLAN_FILTER;
1074 if ((hw->mac_type >= e1000_82544) &&
1075 (hw->mac_type != e1000_82547))
1076 netdev->hw_features |= NETIF_F_TSO;
1078 netdev->features |= netdev->hw_features;
1079 netdev->hw_features |= NETIF_F_RXCSUM;
1081 if (pci_using_dac) {
1082 netdev->features |= NETIF_F_HIGHDMA;
1083 netdev->vlan_features |= NETIF_F_HIGHDMA;
1086 netdev->vlan_features |= NETIF_F_TSO;
1087 netdev->vlan_features |= NETIF_F_HW_CSUM;
1088 netdev->vlan_features |= NETIF_F_SG;
1090 netdev->priv_flags |= IFF_UNICAST_FLT;
1092 adapter->en_mng_pt = e1000_enable_mng_pass_thru(hw);
1094 /* initialize eeprom parameters */
1095 if (e1000_init_eeprom_params(hw)) {
1096 e_err(probe, "EEPROM initialization failed\n");
1097 goto err_eeprom;
1100 /* before reading the EEPROM, reset the controller to
1101 * put the device in a known good starting state */
1103 e1000_reset_hw(hw);
1105 /* make sure the EEPROM is good */
1106 if (e1000_validate_eeprom_checksum(hw) < 0) {
1107 e_err(probe, "The EEPROM Checksum Is Not Valid\n");
1108 e1000_dump_eeprom(adapter);
1110 * set MAC address to all zeroes to invalidate and temporary
1111 * disable this device for the user. This blocks regular
1112 * traffic while still permitting ethtool ioctls from reaching
1113 * the hardware as well as allowing the user to run the
1114 * interface after manually setting a hw addr using
1115 * `ip set address`
1117 memset(hw->mac_addr, 0, netdev->addr_len);
1118 } else {
1119 /* copy the MAC address out of the EEPROM */
1120 if (e1000_read_mac_addr(hw))
1121 e_err(probe, "EEPROM Read Error\n");
1123 /* don't block initalization here due to bad MAC address */
1124 memcpy(netdev->dev_addr, hw->mac_addr, netdev->addr_len);
1125 memcpy(netdev->perm_addr, hw->mac_addr, netdev->addr_len);
1127 if (!is_valid_ether_addr(netdev->perm_addr))
1128 e_err(probe, "Invalid MAC Address\n");
1131 INIT_DELAYED_WORK(&adapter->watchdog_task, e1000_watchdog);
1132 INIT_DELAYED_WORK(&adapter->fifo_stall_task,
1133 e1000_82547_tx_fifo_stall_task);
1134 INIT_DELAYED_WORK(&adapter->phy_info_task, e1000_update_phy_info_task);
1135 INIT_WORK(&adapter->reset_task, e1000_reset_task);
1137 e1000_check_options(adapter);
1139 /* Initial Wake on LAN setting
1140 * If APM wake is enabled in the EEPROM,
1141 * enable the ACPI Magic Packet filter
1144 switch (hw->mac_type) {
1145 case e1000_82542_rev2_0:
1146 case e1000_82542_rev2_1:
1147 case e1000_82543:
1148 break;
1149 case e1000_82544:
1150 e1000_read_eeprom(hw,
1151 EEPROM_INIT_CONTROL2_REG, 1, &eeprom_data);
1152 eeprom_apme_mask = E1000_EEPROM_82544_APM;
1153 break;
1154 case e1000_82546:
1155 case e1000_82546_rev_3:
1156 if (er32(STATUS) & E1000_STATUS_FUNC_1){
1157 e1000_read_eeprom(hw,
1158 EEPROM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
1159 break;
1161 /* Fall Through */
1162 default:
1163 e1000_read_eeprom(hw,
1164 EEPROM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
1165 break;
1167 if (eeprom_data & eeprom_apme_mask)
1168 adapter->eeprom_wol |= E1000_WUFC_MAG;
1170 /* now that we have the eeprom settings, apply the special cases
1171 * where the eeprom may be wrong or the board simply won't support
1172 * wake on lan on a particular port */
1173 switch (pdev->device) {
1174 case E1000_DEV_ID_82546GB_PCIE:
1175 adapter->eeprom_wol = 0;
1176 break;
1177 case E1000_DEV_ID_82546EB_FIBER:
1178 case E1000_DEV_ID_82546GB_FIBER:
1179 /* Wake events only supported on port A for dual fiber
1180 * regardless of eeprom setting */
1181 if (er32(STATUS) & E1000_STATUS_FUNC_1)
1182 adapter->eeprom_wol = 0;
1183 break;
1184 case E1000_DEV_ID_82546GB_QUAD_COPPER_KSP3:
1185 /* if quad port adapter, disable WoL on all but port A */
1186 if (global_quad_port_a != 0)
1187 adapter->eeprom_wol = 0;
1188 else
1189 adapter->quad_port_a = true;
1190 /* Reset for multiple quad port adapters */
1191 if (++global_quad_port_a == 4)
1192 global_quad_port_a = 0;
1193 break;
1196 /* initialize the wol settings based on the eeprom settings */
1197 adapter->wol = adapter->eeprom_wol;
1198 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
1200 /* Auto detect PHY address */
1201 if (hw->mac_type == e1000_ce4100) {
1202 for (i = 0; i < 32; i++) {
1203 hw->phy_addr = i;
1204 e1000_read_phy_reg(hw, PHY_ID2, &tmp);
1205 if (tmp == 0 || tmp == 0xFF) {
1206 if (i == 31)
1207 goto err_eeprom;
1208 continue;
1209 } else
1210 break;
1214 /* reset the hardware with the new settings */
1215 e1000_reset(adapter);
1217 strcpy(netdev->name, "eth%d");
1218 err = register_netdev(netdev);
1219 if (err)
1220 goto err_register;
1222 e1000_vlan_filter_on_off(adapter, false);
1224 /* print bus type/speed/width info */
1225 e_info(probe, "(PCI%s:%dMHz:%d-bit) %pM\n",
1226 ((hw->bus_type == e1000_bus_type_pcix) ? "-X" : ""),
1227 ((hw->bus_speed == e1000_bus_speed_133) ? 133 :
1228 (hw->bus_speed == e1000_bus_speed_120) ? 120 :
1229 (hw->bus_speed == e1000_bus_speed_100) ? 100 :
1230 (hw->bus_speed == e1000_bus_speed_66) ? 66 : 33),
1231 ((hw->bus_width == e1000_bus_width_64) ? 64 : 32),
1232 netdev->dev_addr);
1234 /* carrier off reporting is important to ethtool even BEFORE open */
1235 netif_carrier_off(netdev);
1237 e_info(probe, "Intel(R) PRO/1000 Network Connection\n");
1239 cards_found++;
1240 return 0;
1242 err_register:
1243 err_eeprom:
1244 e1000_phy_hw_reset(hw);
1246 if (hw->flash_address)
1247 iounmap(hw->flash_address);
1248 kfree(adapter->tx_ring);
1249 kfree(adapter->rx_ring);
1250 err_dma:
1251 err_sw_init:
1252 err_mdio_ioremap:
1253 iounmap(hw->ce4100_gbe_mdio_base_virt);
1254 iounmap(hw->hw_addr);
1255 err_ioremap:
1256 free_netdev(netdev);
1257 err_alloc_etherdev:
1258 pci_release_selected_regions(pdev, bars);
1259 err_pci_reg:
1260 pci_disable_device(pdev);
1261 return err;
1265 * e1000_remove - Device Removal Routine
1266 * @pdev: PCI device information struct
1268 * e1000_remove is called by the PCI subsystem to alert the driver
1269 * that it should release a PCI device. The could be caused by a
1270 * Hot-Plug event, or because the driver is going to be removed from
1271 * memory.
1274 static void __devexit e1000_remove(struct pci_dev *pdev)
1276 struct net_device *netdev = pci_get_drvdata(pdev);
1277 struct e1000_adapter *adapter = netdev_priv(netdev);
1278 struct e1000_hw *hw = &adapter->hw;
1280 e1000_down_and_stop(adapter);
1281 e1000_release_manageability(adapter);
1283 unregister_netdev(netdev);
1285 e1000_phy_hw_reset(hw);
1287 kfree(adapter->tx_ring);
1288 kfree(adapter->rx_ring);
1290 if (hw->mac_type == e1000_ce4100)
1291 iounmap(hw->ce4100_gbe_mdio_base_virt);
1292 iounmap(hw->hw_addr);
1293 if (hw->flash_address)
1294 iounmap(hw->flash_address);
1295 pci_release_selected_regions(pdev, adapter->bars);
1297 free_netdev(netdev);
1299 pci_disable_device(pdev);
1303 * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
1304 * @adapter: board private structure to initialize
1306 * e1000_sw_init initializes the Adapter private data structure.
1307 * e1000_init_hw_struct MUST be called before this function
1310 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
1312 adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
1314 adapter->num_tx_queues = 1;
1315 adapter->num_rx_queues = 1;
1317 if (e1000_alloc_queues(adapter)) {
1318 e_err(probe, "Unable to allocate memory for queues\n");
1319 return -ENOMEM;
1322 /* Explicitly disable IRQ since the NIC can be in any state. */
1323 e1000_irq_disable(adapter);
1325 spin_lock_init(&adapter->stats_lock);
1326 mutex_init(&adapter->mutex);
1328 set_bit(__E1000_DOWN, &adapter->flags);
1330 return 0;
1334 * e1000_alloc_queues - Allocate memory for all rings
1335 * @adapter: board private structure to initialize
1337 * We allocate one ring per queue at run-time since we don't know the
1338 * number of queues at compile-time.
1341 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
1343 adapter->tx_ring = kcalloc(adapter->num_tx_queues,
1344 sizeof(struct e1000_tx_ring), GFP_KERNEL);
1345 if (!adapter->tx_ring)
1346 return -ENOMEM;
1348 adapter->rx_ring = kcalloc(adapter->num_rx_queues,
1349 sizeof(struct e1000_rx_ring), GFP_KERNEL);
1350 if (!adapter->rx_ring) {
1351 kfree(adapter->tx_ring);
1352 return -ENOMEM;
1355 return E1000_SUCCESS;
1359 * e1000_open - Called when a network interface is made active
1360 * @netdev: network interface device structure
1362 * Returns 0 on success, negative value on failure
1364 * The open entry point is called when a network interface is made
1365 * active by the system (IFF_UP). At this point all resources needed
1366 * for transmit and receive operations are allocated, the interrupt
1367 * handler is registered with the OS, the watchdog task is started,
1368 * and the stack is notified that the interface is ready.
1371 static int e1000_open(struct net_device *netdev)
1373 struct e1000_adapter *adapter = netdev_priv(netdev);
1374 struct e1000_hw *hw = &adapter->hw;
1375 int err;
1377 /* disallow open during test */
1378 if (test_bit(__E1000_TESTING, &adapter->flags))
1379 return -EBUSY;
1381 netif_carrier_off(netdev);
1383 /* allocate transmit descriptors */
1384 err = e1000_setup_all_tx_resources(adapter);
1385 if (err)
1386 goto err_setup_tx;
1388 /* allocate receive descriptors */
1389 err = e1000_setup_all_rx_resources(adapter);
1390 if (err)
1391 goto err_setup_rx;
1393 e1000_power_up_phy(adapter);
1395 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
1396 if ((hw->mng_cookie.status &
1397 E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT)) {
1398 e1000_update_mng_vlan(adapter);
1401 /* before we allocate an interrupt, we must be ready to handle it.
1402 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1403 * as soon as we call pci_request_irq, so we have to setup our
1404 * clean_rx handler before we do so. */
1405 e1000_configure(adapter);
1407 err = e1000_request_irq(adapter);
1408 if (err)
1409 goto err_req_irq;
1411 /* From here on the code is the same as e1000_up() */
1412 clear_bit(__E1000_DOWN, &adapter->flags);
1414 napi_enable(&adapter->napi);
1416 e1000_irq_enable(adapter);
1418 netif_start_queue(netdev);
1420 /* fire a link status change interrupt to start the watchdog */
1421 ew32(ICS, E1000_ICS_LSC);
1423 return E1000_SUCCESS;
1425 err_req_irq:
1426 e1000_power_down_phy(adapter);
1427 e1000_free_all_rx_resources(adapter);
1428 err_setup_rx:
1429 e1000_free_all_tx_resources(adapter);
1430 err_setup_tx:
1431 e1000_reset(adapter);
1433 return err;
1437 * e1000_close - Disables a network interface
1438 * @netdev: network interface device structure
1440 * Returns 0, this is not allowed to fail
1442 * The close entry point is called when an interface is de-activated
1443 * by the OS. The hardware is still under the drivers control, but
1444 * needs to be disabled. A global MAC reset is issued to stop the
1445 * hardware, and all transmit and receive resources are freed.
1448 static int e1000_close(struct net_device *netdev)
1450 struct e1000_adapter *adapter = netdev_priv(netdev);
1451 struct e1000_hw *hw = &adapter->hw;
1453 WARN_ON(test_bit(__E1000_RESETTING, &adapter->flags));
1454 e1000_down(adapter);
1455 e1000_power_down_phy(adapter);
1456 e1000_free_irq(adapter);
1458 e1000_free_all_tx_resources(adapter);
1459 e1000_free_all_rx_resources(adapter);
1461 /* kill manageability vlan ID if supported, but not if a vlan with
1462 * the same ID is registered on the host OS (let 8021q kill it) */
1463 if ((hw->mng_cookie.status &
1464 E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) &&
1465 !test_bit(adapter->mng_vlan_id, adapter->active_vlans)) {
1466 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
1469 return 0;
1473 * e1000_check_64k_bound - check that memory doesn't cross 64kB boundary
1474 * @adapter: address of board private structure
1475 * @start: address of beginning of memory
1476 * @len: length of memory
1478 static bool e1000_check_64k_bound(struct e1000_adapter *adapter, void *start,
1479 unsigned long len)
1481 struct e1000_hw *hw = &adapter->hw;
1482 unsigned long begin = (unsigned long)start;
1483 unsigned long end = begin + len;
1485 /* First rev 82545 and 82546 need to not allow any memory
1486 * write location to cross 64k boundary due to errata 23 */
1487 if (hw->mac_type == e1000_82545 ||
1488 hw->mac_type == e1000_ce4100 ||
1489 hw->mac_type == e1000_82546) {
1490 return ((begin ^ (end - 1)) >> 16) != 0 ? false : true;
1493 return true;
1497 * e1000_setup_tx_resources - allocate Tx resources (Descriptors)
1498 * @adapter: board private structure
1499 * @txdr: tx descriptor ring (for a specific queue) to setup
1501 * Return 0 on success, negative on failure
1504 static int e1000_setup_tx_resources(struct e1000_adapter *adapter,
1505 struct e1000_tx_ring *txdr)
1507 struct pci_dev *pdev = adapter->pdev;
1508 int size;
1510 size = sizeof(struct e1000_buffer) * txdr->count;
1511 txdr->buffer_info = vzalloc(size);
1512 if (!txdr->buffer_info) {
1513 e_err(probe, "Unable to allocate memory for the Tx descriptor "
1514 "ring\n");
1515 return -ENOMEM;
1518 /* round up to nearest 4K */
1520 txdr->size = txdr->count * sizeof(struct e1000_tx_desc);
1521 txdr->size = ALIGN(txdr->size, 4096);
1523 txdr->desc = dma_alloc_coherent(&pdev->dev, txdr->size, &txdr->dma,
1524 GFP_KERNEL);
1525 if (!txdr->desc) {
1526 setup_tx_desc_die:
1527 vfree(txdr->buffer_info);
1528 e_err(probe, "Unable to allocate memory for the Tx descriptor "
1529 "ring\n");
1530 return -ENOMEM;
1533 /* Fix for errata 23, can't cross 64kB boundary */
1534 if (!e1000_check_64k_bound(adapter, txdr->desc, txdr->size)) {
1535 void *olddesc = txdr->desc;
1536 dma_addr_t olddma = txdr->dma;
1537 e_err(tx_err, "txdr align check failed: %u bytes at %p\n",
1538 txdr->size, txdr->desc);
1539 /* Try again, without freeing the previous */
1540 txdr->desc = dma_alloc_coherent(&pdev->dev, txdr->size,
1541 &txdr->dma, GFP_KERNEL);
1542 /* Failed allocation, critical failure */
1543 if (!txdr->desc) {
1544 dma_free_coherent(&pdev->dev, txdr->size, olddesc,
1545 olddma);
1546 goto setup_tx_desc_die;
1549 if (!e1000_check_64k_bound(adapter, txdr->desc, txdr->size)) {
1550 /* give up */
1551 dma_free_coherent(&pdev->dev, txdr->size, txdr->desc,
1552 txdr->dma);
1553 dma_free_coherent(&pdev->dev, txdr->size, olddesc,
1554 olddma);
1555 e_err(probe, "Unable to allocate aligned memory "
1556 "for the transmit descriptor ring\n");
1557 vfree(txdr->buffer_info);
1558 return -ENOMEM;
1559 } else {
1560 /* Free old allocation, new allocation was successful */
1561 dma_free_coherent(&pdev->dev, txdr->size, olddesc,
1562 olddma);
1565 memset(txdr->desc, 0, txdr->size);
1567 txdr->next_to_use = 0;
1568 txdr->next_to_clean = 0;
1570 return 0;
1574 * e1000_setup_all_tx_resources - wrapper to allocate Tx resources
1575 * (Descriptors) for all queues
1576 * @adapter: board private structure
1578 * Return 0 on success, negative on failure
1581 int e1000_setup_all_tx_resources(struct e1000_adapter *adapter)
1583 int i, err = 0;
1585 for (i = 0; i < adapter->num_tx_queues; i++) {
1586 err = e1000_setup_tx_resources(adapter, &adapter->tx_ring[i]);
1587 if (err) {
1588 e_err(probe, "Allocation for Tx Queue %u failed\n", i);
1589 for (i-- ; i >= 0; i--)
1590 e1000_free_tx_resources(adapter,
1591 &adapter->tx_ring[i]);
1592 break;
1596 return err;
1600 * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
1601 * @adapter: board private structure
1603 * Configure the Tx unit of the MAC after a reset.
1606 static void e1000_configure_tx(struct e1000_adapter *adapter)
1608 u64 tdba;
1609 struct e1000_hw *hw = &adapter->hw;
1610 u32 tdlen, tctl, tipg;
1611 u32 ipgr1, ipgr2;
1613 /* Setup the HW Tx Head and Tail descriptor pointers */
1615 switch (adapter->num_tx_queues) {
1616 case 1:
1617 default:
1618 tdba = adapter->tx_ring[0].dma;
1619 tdlen = adapter->tx_ring[0].count *
1620 sizeof(struct e1000_tx_desc);
1621 ew32(TDLEN, tdlen);
1622 ew32(TDBAH, (tdba >> 32));
1623 ew32(TDBAL, (tdba & 0x00000000ffffffffULL));
1624 ew32(TDT, 0);
1625 ew32(TDH, 0);
1626 adapter->tx_ring[0].tdh = ((hw->mac_type >= e1000_82543) ? E1000_TDH : E1000_82542_TDH);
1627 adapter->tx_ring[0].tdt = ((hw->mac_type >= e1000_82543) ? E1000_TDT : E1000_82542_TDT);
1628 break;
1631 /* Set the default values for the Tx Inter Packet Gap timer */
1632 if ((hw->media_type == e1000_media_type_fiber ||
1633 hw->media_type == e1000_media_type_internal_serdes))
1634 tipg = DEFAULT_82543_TIPG_IPGT_FIBER;
1635 else
1636 tipg = DEFAULT_82543_TIPG_IPGT_COPPER;
1638 switch (hw->mac_type) {
1639 case e1000_82542_rev2_0:
1640 case e1000_82542_rev2_1:
1641 tipg = DEFAULT_82542_TIPG_IPGT;
1642 ipgr1 = DEFAULT_82542_TIPG_IPGR1;
1643 ipgr2 = DEFAULT_82542_TIPG_IPGR2;
1644 break;
1645 default:
1646 ipgr1 = DEFAULT_82543_TIPG_IPGR1;
1647 ipgr2 = DEFAULT_82543_TIPG_IPGR2;
1648 break;
1650 tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
1651 tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
1652 ew32(TIPG, tipg);
1654 /* Set the Tx Interrupt Delay register */
1656 ew32(TIDV, adapter->tx_int_delay);
1657 if (hw->mac_type >= e1000_82540)
1658 ew32(TADV, adapter->tx_abs_int_delay);
1660 /* Program the Transmit Control Register */
1662 tctl = er32(TCTL);
1663 tctl &= ~E1000_TCTL_CT;
1664 tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
1665 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
1667 e1000_config_collision_dist(hw);
1669 /* Setup Transmit Descriptor Settings for eop descriptor */
1670 adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
1672 /* only set IDE if we are delaying interrupts using the timers */
1673 if (adapter->tx_int_delay)
1674 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
1676 if (hw->mac_type < e1000_82543)
1677 adapter->txd_cmd |= E1000_TXD_CMD_RPS;
1678 else
1679 adapter->txd_cmd |= E1000_TXD_CMD_RS;
1681 /* Cache if we're 82544 running in PCI-X because we'll
1682 * need this to apply a workaround later in the send path. */
1683 if (hw->mac_type == e1000_82544 &&
1684 hw->bus_type == e1000_bus_type_pcix)
1685 adapter->pcix_82544 = true;
1687 ew32(TCTL, tctl);
1692 * e1000_setup_rx_resources - allocate Rx resources (Descriptors)
1693 * @adapter: board private structure
1694 * @rxdr: rx descriptor ring (for a specific queue) to setup
1696 * Returns 0 on success, negative on failure
1699 static int e1000_setup_rx_resources(struct e1000_adapter *adapter,
1700 struct e1000_rx_ring *rxdr)
1702 struct pci_dev *pdev = adapter->pdev;
1703 int size, desc_len;
1705 size = sizeof(struct e1000_buffer) * rxdr->count;
1706 rxdr->buffer_info = vzalloc(size);
1707 if (!rxdr->buffer_info) {
1708 e_err(probe, "Unable to allocate memory for the Rx descriptor "
1709 "ring\n");
1710 return -ENOMEM;
1713 desc_len = sizeof(struct e1000_rx_desc);
1715 /* Round up to nearest 4K */
1717 rxdr->size = rxdr->count * desc_len;
1718 rxdr->size = ALIGN(rxdr->size, 4096);
1720 rxdr->desc = dma_alloc_coherent(&pdev->dev, rxdr->size, &rxdr->dma,
1721 GFP_KERNEL);
1723 if (!rxdr->desc) {
1724 e_err(probe, "Unable to allocate memory for the Rx descriptor "
1725 "ring\n");
1726 setup_rx_desc_die:
1727 vfree(rxdr->buffer_info);
1728 return -ENOMEM;
1731 /* Fix for errata 23, can't cross 64kB boundary */
1732 if (!e1000_check_64k_bound(adapter, rxdr->desc, rxdr->size)) {
1733 void *olddesc = rxdr->desc;
1734 dma_addr_t olddma = rxdr->dma;
1735 e_err(rx_err, "rxdr align check failed: %u bytes at %p\n",
1736 rxdr->size, rxdr->desc);
1737 /* Try again, without freeing the previous */
1738 rxdr->desc = dma_alloc_coherent(&pdev->dev, rxdr->size,
1739 &rxdr->dma, GFP_KERNEL);
1740 /* Failed allocation, critical failure */
1741 if (!rxdr->desc) {
1742 dma_free_coherent(&pdev->dev, rxdr->size, olddesc,
1743 olddma);
1744 e_err(probe, "Unable to allocate memory for the Rx "
1745 "descriptor ring\n");
1746 goto setup_rx_desc_die;
1749 if (!e1000_check_64k_bound(adapter, rxdr->desc, rxdr->size)) {
1750 /* give up */
1751 dma_free_coherent(&pdev->dev, rxdr->size, rxdr->desc,
1752 rxdr->dma);
1753 dma_free_coherent(&pdev->dev, rxdr->size, olddesc,
1754 olddma);
1755 e_err(probe, "Unable to allocate aligned memory for "
1756 "the Rx descriptor ring\n");
1757 goto setup_rx_desc_die;
1758 } else {
1759 /* Free old allocation, new allocation was successful */
1760 dma_free_coherent(&pdev->dev, rxdr->size, olddesc,
1761 olddma);
1764 memset(rxdr->desc, 0, rxdr->size);
1766 rxdr->next_to_clean = 0;
1767 rxdr->next_to_use = 0;
1768 rxdr->rx_skb_top = NULL;
1770 return 0;
1774 * e1000_setup_all_rx_resources - wrapper to allocate Rx resources
1775 * (Descriptors) for all queues
1776 * @adapter: board private structure
1778 * Return 0 on success, negative on failure
1781 int e1000_setup_all_rx_resources(struct e1000_adapter *adapter)
1783 int i, err = 0;
1785 for (i = 0; i < adapter->num_rx_queues; i++) {
1786 err = e1000_setup_rx_resources(adapter, &adapter->rx_ring[i]);
1787 if (err) {
1788 e_err(probe, "Allocation for Rx Queue %u failed\n", i);
1789 for (i-- ; i >= 0; i--)
1790 e1000_free_rx_resources(adapter,
1791 &adapter->rx_ring[i]);
1792 break;
1796 return err;
1800 * e1000_setup_rctl - configure the receive control registers
1801 * @adapter: Board private structure
1803 static void e1000_setup_rctl(struct e1000_adapter *adapter)
1805 struct e1000_hw *hw = &adapter->hw;
1806 u32 rctl;
1808 rctl = er32(RCTL);
1810 rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
1812 rctl |= E1000_RCTL_BAM | E1000_RCTL_LBM_NO |
1813 E1000_RCTL_RDMTS_HALF |
1814 (hw->mc_filter_type << E1000_RCTL_MO_SHIFT);
1816 if (hw->tbi_compatibility_on == 1)
1817 rctl |= E1000_RCTL_SBP;
1818 else
1819 rctl &= ~E1000_RCTL_SBP;
1821 if (adapter->netdev->mtu <= ETH_DATA_LEN)
1822 rctl &= ~E1000_RCTL_LPE;
1823 else
1824 rctl |= E1000_RCTL_LPE;
1826 /* Setup buffer sizes */
1827 rctl &= ~E1000_RCTL_SZ_4096;
1828 rctl |= E1000_RCTL_BSEX;
1829 switch (adapter->rx_buffer_len) {
1830 case E1000_RXBUFFER_2048:
1831 default:
1832 rctl |= E1000_RCTL_SZ_2048;
1833 rctl &= ~E1000_RCTL_BSEX;
1834 break;
1835 case E1000_RXBUFFER_4096:
1836 rctl |= E1000_RCTL_SZ_4096;
1837 break;
1838 case E1000_RXBUFFER_8192:
1839 rctl |= E1000_RCTL_SZ_8192;
1840 break;
1841 case E1000_RXBUFFER_16384:
1842 rctl |= E1000_RCTL_SZ_16384;
1843 break;
1846 ew32(RCTL, rctl);
1850 * e1000_configure_rx - Configure 8254x Receive Unit after Reset
1851 * @adapter: board private structure
1853 * Configure the Rx unit of the MAC after a reset.
1856 static void e1000_configure_rx(struct e1000_adapter *adapter)
1858 u64 rdba;
1859 struct e1000_hw *hw = &adapter->hw;
1860 u32 rdlen, rctl, rxcsum;
1862 if (adapter->netdev->mtu > ETH_DATA_LEN) {
1863 rdlen = adapter->rx_ring[0].count *
1864 sizeof(struct e1000_rx_desc);
1865 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
1866 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
1867 } else {
1868 rdlen = adapter->rx_ring[0].count *
1869 sizeof(struct e1000_rx_desc);
1870 adapter->clean_rx = e1000_clean_rx_irq;
1871 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
1874 /* disable receives while setting up the descriptors */
1875 rctl = er32(RCTL);
1876 ew32(RCTL, rctl & ~E1000_RCTL_EN);
1878 /* set the Receive Delay Timer Register */
1879 ew32(RDTR, adapter->rx_int_delay);
1881 if (hw->mac_type >= e1000_82540) {
1882 ew32(RADV, adapter->rx_abs_int_delay);
1883 if (adapter->itr_setting != 0)
1884 ew32(ITR, 1000000000 / (adapter->itr * 256));
1887 /* Setup the HW Rx Head and Tail Descriptor Pointers and
1888 * the Base and Length of the Rx Descriptor Ring */
1889 switch (adapter->num_rx_queues) {
1890 case 1:
1891 default:
1892 rdba = adapter->rx_ring[0].dma;
1893 ew32(RDLEN, rdlen);
1894 ew32(RDBAH, (rdba >> 32));
1895 ew32(RDBAL, (rdba & 0x00000000ffffffffULL));
1896 ew32(RDT, 0);
1897 ew32(RDH, 0);
1898 adapter->rx_ring[0].rdh = ((hw->mac_type >= e1000_82543) ? E1000_RDH : E1000_82542_RDH);
1899 adapter->rx_ring[0].rdt = ((hw->mac_type >= e1000_82543) ? E1000_RDT : E1000_82542_RDT);
1900 break;
1903 /* Enable 82543 Receive Checksum Offload for TCP and UDP */
1904 if (hw->mac_type >= e1000_82543) {
1905 rxcsum = er32(RXCSUM);
1906 if (adapter->rx_csum)
1907 rxcsum |= E1000_RXCSUM_TUOFL;
1908 else
1909 /* don't need to clear IPPCSE as it defaults to 0 */
1910 rxcsum &= ~E1000_RXCSUM_TUOFL;
1911 ew32(RXCSUM, rxcsum);
1914 /* Enable Receives */
1915 ew32(RCTL, rctl | E1000_RCTL_EN);
1919 * e1000_free_tx_resources - Free Tx Resources per Queue
1920 * @adapter: board private structure
1921 * @tx_ring: Tx descriptor ring for a specific queue
1923 * Free all transmit software resources
1926 static void e1000_free_tx_resources(struct e1000_adapter *adapter,
1927 struct e1000_tx_ring *tx_ring)
1929 struct pci_dev *pdev = adapter->pdev;
1931 e1000_clean_tx_ring(adapter, tx_ring);
1933 vfree(tx_ring->buffer_info);
1934 tx_ring->buffer_info = NULL;
1936 dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
1937 tx_ring->dma);
1939 tx_ring->desc = NULL;
1943 * e1000_free_all_tx_resources - Free Tx Resources for All Queues
1944 * @adapter: board private structure
1946 * Free all transmit software resources
1949 void e1000_free_all_tx_resources(struct e1000_adapter *adapter)
1951 int i;
1953 for (i = 0; i < adapter->num_tx_queues; i++)
1954 e1000_free_tx_resources(adapter, &adapter->tx_ring[i]);
1957 static void e1000_unmap_and_free_tx_resource(struct e1000_adapter *adapter,
1958 struct e1000_buffer *buffer_info)
1960 if (buffer_info->dma) {
1961 if (buffer_info->mapped_as_page)
1962 dma_unmap_page(&adapter->pdev->dev, buffer_info->dma,
1963 buffer_info->length, DMA_TO_DEVICE);
1964 else
1965 dma_unmap_single(&adapter->pdev->dev, buffer_info->dma,
1966 buffer_info->length,
1967 DMA_TO_DEVICE);
1968 buffer_info->dma = 0;
1970 if (buffer_info->skb) {
1971 dev_kfree_skb_any(buffer_info->skb);
1972 buffer_info->skb = NULL;
1974 buffer_info->time_stamp = 0;
1975 /* buffer_info must be completely set up in the transmit path */
1979 * e1000_clean_tx_ring - Free Tx Buffers
1980 * @adapter: board private structure
1981 * @tx_ring: ring to be cleaned
1984 static void e1000_clean_tx_ring(struct e1000_adapter *adapter,
1985 struct e1000_tx_ring *tx_ring)
1987 struct e1000_hw *hw = &adapter->hw;
1988 struct e1000_buffer *buffer_info;
1989 unsigned long size;
1990 unsigned int i;
1992 /* Free all the Tx ring sk_buffs */
1994 for (i = 0; i < tx_ring->count; i++) {
1995 buffer_info = &tx_ring->buffer_info[i];
1996 e1000_unmap_and_free_tx_resource(adapter, buffer_info);
1999 size = sizeof(struct e1000_buffer) * tx_ring->count;
2000 memset(tx_ring->buffer_info, 0, size);
2002 /* Zero out the descriptor ring */
2004 memset(tx_ring->desc, 0, tx_ring->size);
2006 tx_ring->next_to_use = 0;
2007 tx_ring->next_to_clean = 0;
2008 tx_ring->last_tx_tso = false;
2010 writel(0, hw->hw_addr + tx_ring->tdh);
2011 writel(0, hw->hw_addr + tx_ring->tdt);
2015 * e1000_clean_all_tx_rings - Free Tx Buffers for all queues
2016 * @adapter: board private structure
2019 static void e1000_clean_all_tx_rings(struct e1000_adapter *adapter)
2021 int i;
2023 for (i = 0; i < adapter->num_tx_queues; i++)
2024 e1000_clean_tx_ring(adapter, &adapter->tx_ring[i]);
2028 * e1000_free_rx_resources - Free Rx Resources
2029 * @adapter: board private structure
2030 * @rx_ring: ring to clean the resources from
2032 * Free all receive software resources
2035 static void e1000_free_rx_resources(struct e1000_adapter *adapter,
2036 struct e1000_rx_ring *rx_ring)
2038 struct pci_dev *pdev = adapter->pdev;
2040 e1000_clean_rx_ring(adapter, rx_ring);
2042 vfree(rx_ring->buffer_info);
2043 rx_ring->buffer_info = NULL;
2045 dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
2046 rx_ring->dma);
2048 rx_ring->desc = NULL;
2052 * e1000_free_all_rx_resources - Free Rx Resources for All Queues
2053 * @adapter: board private structure
2055 * Free all receive software resources
2058 void e1000_free_all_rx_resources(struct e1000_adapter *adapter)
2060 int i;
2062 for (i = 0; i < adapter->num_rx_queues; i++)
2063 e1000_free_rx_resources(adapter, &adapter->rx_ring[i]);
2067 * e1000_clean_rx_ring - Free Rx Buffers per Queue
2068 * @adapter: board private structure
2069 * @rx_ring: ring to free buffers from
2072 static void e1000_clean_rx_ring(struct e1000_adapter *adapter,
2073 struct e1000_rx_ring *rx_ring)
2075 struct e1000_hw *hw = &adapter->hw;
2076 struct e1000_buffer *buffer_info;
2077 struct pci_dev *pdev = adapter->pdev;
2078 unsigned long size;
2079 unsigned int i;
2081 /* Free all the Rx ring sk_buffs */
2082 for (i = 0; i < rx_ring->count; i++) {
2083 buffer_info = &rx_ring->buffer_info[i];
2084 if (buffer_info->dma &&
2085 adapter->clean_rx == e1000_clean_rx_irq) {
2086 dma_unmap_single(&pdev->dev, buffer_info->dma,
2087 buffer_info->length,
2088 DMA_FROM_DEVICE);
2089 } else if (buffer_info->dma &&
2090 adapter->clean_rx == e1000_clean_jumbo_rx_irq) {
2091 dma_unmap_page(&pdev->dev, buffer_info->dma,
2092 buffer_info->length,
2093 DMA_FROM_DEVICE);
2096 buffer_info->dma = 0;
2097 if (buffer_info->page) {
2098 put_page(buffer_info->page);
2099 buffer_info->page = NULL;
2101 if (buffer_info->skb) {
2102 dev_kfree_skb(buffer_info->skb);
2103 buffer_info->skb = NULL;
2107 /* there also may be some cached data from a chained receive */
2108 if (rx_ring->rx_skb_top) {
2109 dev_kfree_skb(rx_ring->rx_skb_top);
2110 rx_ring->rx_skb_top = NULL;
2113 size = sizeof(struct e1000_buffer) * rx_ring->count;
2114 memset(rx_ring->buffer_info, 0, size);
2116 /* Zero out the descriptor ring */
2117 memset(rx_ring->desc, 0, rx_ring->size);
2119 rx_ring->next_to_clean = 0;
2120 rx_ring->next_to_use = 0;
2122 writel(0, hw->hw_addr + rx_ring->rdh);
2123 writel(0, hw->hw_addr + rx_ring->rdt);
2127 * e1000_clean_all_rx_rings - Free Rx Buffers for all queues
2128 * @adapter: board private structure
2131 static void e1000_clean_all_rx_rings(struct e1000_adapter *adapter)
2133 int i;
2135 for (i = 0; i < adapter->num_rx_queues; i++)
2136 e1000_clean_rx_ring(adapter, &adapter->rx_ring[i]);
2139 /* The 82542 2.0 (revision 2) needs to have the receive unit in reset
2140 * and memory write and invalidate disabled for certain operations
2142 static void e1000_enter_82542_rst(struct e1000_adapter *adapter)
2144 struct e1000_hw *hw = &adapter->hw;
2145 struct net_device *netdev = adapter->netdev;
2146 u32 rctl;
2148 e1000_pci_clear_mwi(hw);
2150 rctl = er32(RCTL);
2151 rctl |= E1000_RCTL_RST;
2152 ew32(RCTL, rctl);
2153 E1000_WRITE_FLUSH();
2154 mdelay(5);
2156 if (netif_running(netdev))
2157 e1000_clean_all_rx_rings(adapter);
2160 static void e1000_leave_82542_rst(struct e1000_adapter *adapter)
2162 struct e1000_hw *hw = &adapter->hw;
2163 struct net_device *netdev = adapter->netdev;
2164 u32 rctl;
2166 rctl = er32(RCTL);
2167 rctl &= ~E1000_RCTL_RST;
2168 ew32(RCTL, rctl);
2169 E1000_WRITE_FLUSH();
2170 mdelay(5);
2172 if (hw->pci_cmd_word & PCI_COMMAND_INVALIDATE)
2173 e1000_pci_set_mwi(hw);
2175 if (netif_running(netdev)) {
2176 /* No need to loop, because 82542 supports only 1 queue */
2177 struct e1000_rx_ring *ring = &adapter->rx_ring[0];
2178 e1000_configure_rx(adapter);
2179 adapter->alloc_rx_buf(adapter, ring, E1000_DESC_UNUSED(ring));
2184 * e1000_set_mac - Change the Ethernet Address of the NIC
2185 * @netdev: network interface device structure
2186 * @p: pointer to an address structure
2188 * Returns 0 on success, negative on failure
2191 static int e1000_set_mac(struct net_device *netdev, void *p)
2193 struct e1000_adapter *adapter = netdev_priv(netdev);
2194 struct e1000_hw *hw = &adapter->hw;
2195 struct sockaddr *addr = p;
2197 if (!is_valid_ether_addr(addr->sa_data))
2198 return -EADDRNOTAVAIL;
2200 /* 82542 2.0 needs to be in reset to write receive address registers */
2202 if (hw->mac_type == e1000_82542_rev2_0)
2203 e1000_enter_82542_rst(adapter);
2205 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2206 memcpy(hw->mac_addr, addr->sa_data, netdev->addr_len);
2208 e1000_rar_set(hw, hw->mac_addr, 0);
2210 if (hw->mac_type == e1000_82542_rev2_0)
2211 e1000_leave_82542_rst(adapter);
2213 return 0;
2217 * e1000_set_rx_mode - Secondary Unicast, Multicast and Promiscuous mode set
2218 * @netdev: network interface device structure
2220 * The set_rx_mode entry point is called whenever the unicast or multicast
2221 * address lists or the network interface flags are updated. This routine is
2222 * responsible for configuring the hardware for proper unicast, multicast,
2223 * promiscuous mode, and all-multi behavior.
2226 static void e1000_set_rx_mode(struct net_device *netdev)
2228 struct e1000_adapter *adapter = netdev_priv(netdev);
2229 struct e1000_hw *hw = &adapter->hw;
2230 struct netdev_hw_addr *ha;
2231 bool use_uc = false;
2232 u32 rctl;
2233 u32 hash_value;
2234 int i, rar_entries = E1000_RAR_ENTRIES;
2235 int mta_reg_count = E1000_NUM_MTA_REGISTERS;
2236 u32 *mcarray = kcalloc(mta_reg_count, sizeof(u32), GFP_ATOMIC);
2238 if (!mcarray) {
2239 e_err(probe, "memory allocation failed\n");
2240 return;
2243 /* Check for Promiscuous and All Multicast modes */
2245 rctl = er32(RCTL);
2247 if (netdev->flags & IFF_PROMISC) {
2248 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2249 rctl &= ~E1000_RCTL_VFE;
2250 } else {
2251 if (netdev->flags & IFF_ALLMULTI)
2252 rctl |= E1000_RCTL_MPE;
2253 else
2254 rctl &= ~E1000_RCTL_MPE;
2255 /* Enable VLAN filter if there is a VLAN */
2256 if (e1000_vlan_used(adapter))
2257 rctl |= E1000_RCTL_VFE;
2260 if (netdev_uc_count(netdev) > rar_entries - 1) {
2261 rctl |= E1000_RCTL_UPE;
2262 } else if (!(netdev->flags & IFF_PROMISC)) {
2263 rctl &= ~E1000_RCTL_UPE;
2264 use_uc = true;
2267 ew32(RCTL, rctl);
2269 /* 82542 2.0 needs to be in reset to write receive address registers */
2271 if (hw->mac_type == e1000_82542_rev2_0)
2272 e1000_enter_82542_rst(adapter);
2274 /* load the first 14 addresses into the exact filters 1-14. Unicast
2275 * addresses take precedence to avoid disabling unicast filtering
2276 * when possible.
2278 * RAR 0 is used for the station MAC address
2279 * if there are not 14 addresses, go ahead and clear the filters
2281 i = 1;
2282 if (use_uc)
2283 netdev_for_each_uc_addr(ha, netdev) {
2284 if (i == rar_entries)
2285 break;
2286 e1000_rar_set(hw, ha->addr, i++);
2289 netdev_for_each_mc_addr(ha, netdev) {
2290 if (i == rar_entries) {
2291 /* load any remaining addresses into the hash table */
2292 u32 hash_reg, hash_bit, mta;
2293 hash_value = e1000_hash_mc_addr(hw, ha->addr);
2294 hash_reg = (hash_value >> 5) & 0x7F;
2295 hash_bit = hash_value & 0x1F;
2296 mta = (1 << hash_bit);
2297 mcarray[hash_reg] |= mta;
2298 } else {
2299 e1000_rar_set(hw, ha->addr, i++);
2303 for (; i < rar_entries; i++) {
2304 E1000_WRITE_REG_ARRAY(hw, RA, i << 1, 0);
2305 E1000_WRITE_FLUSH();
2306 E1000_WRITE_REG_ARRAY(hw, RA, (i << 1) + 1, 0);
2307 E1000_WRITE_FLUSH();
2310 /* write the hash table completely, write from bottom to avoid
2311 * both stupid write combining chipsets, and flushing each write */
2312 for (i = mta_reg_count - 1; i >= 0 ; i--) {
2314 * If we are on an 82544 has an errata where writing odd
2315 * offsets overwrites the previous even offset, but writing
2316 * backwards over the range solves the issue by always
2317 * writing the odd offset first
2319 E1000_WRITE_REG_ARRAY(hw, MTA, i, mcarray[i]);
2321 E1000_WRITE_FLUSH();
2323 if (hw->mac_type == e1000_82542_rev2_0)
2324 e1000_leave_82542_rst(adapter);
2326 kfree(mcarray);
2330 * e1000_update_phy_info_task - get phy info
2331 * @work: work struct contained inside adapter struct
2333 * Need to wait a few seconds after link up to get diagnostic information from
2334 * the phy
2336 static void e1000_update_phy_info_task(struct work_struct *work)
2338 struct e1000_adapter *adapter = container_of(work,
2339 struct e1000_adapter,
2340 phy_info_task.work);
2341 if (test_bit(__E1000_DOWN, &adapter->flags))
2342 return;
2343 mutex_lock(&adapter->mutex);
2344 e1000_phy_get_info(&adapter->hw, &adapter->phy_info);
2345 mutex_unlock(&adapter->mutex);
2349 * e1000_82547_tx_fifo_stall_task - task to complete work
2350 * @work: work struct contained inside adapter struct
2352 static void e1000_82547_tx_fifo_stall_task(struct work_struct *work)
2354 struct e1000_adapter *adapter = container_of(work,
2355 struct e1000_adapter,
2356 fifo_stall_task.work);
2357 struct e1000_hw *hw = &adapter->hw;
2358 struct net_device *netdev = adapter->netdev;
2359 u32 tctl;
2361 if (test_bit(__E1000_DOWN, &adapter->flags))
2362 return;
2363 mutex_lock(&adapter->mutex);
2364 if (atomic_read(&adapter->tx_fifo_stall)) {
2365 if ((er32(TDT) == er32(TDH)) &&
2366 (er32(TDFT) == er32(TDFH)) &&
2367 (er32(TDFTS) == er32(TDFHS))) {
2368 tctl = er32(TCTL);
2369 ew32(TCTL, tctl & ~E1000_TCTL_EN);
2370 ew32(TDFT, adapter->tx_head_addr);
2371 ew32(TDFH, adapter->tx_head_addr);
2372 ew32(TDFTS, adapter->tx_head_addr);
2373 ew32(TDFHS, adapter->tx_head_addr);
2374 ew32(TCTL, tctl);
2375 E1000_WRITE_FLUSH();
2377 adapter->tx_fifo_head = 0;
2378 atomic_set(&adapter->tx_fifo_stall, 0);
2379 netif_wake_queue(netdev);
2380 } else if (!test_bit(__E1000_DOWN, &adapter->flags)) {
2381 schedule_delayed_work(&adapter->fifo_stall_task, 1);
2384 mutex_unlock(&adapter->mutex);
2387 bool e1000_has_link(struct e1000_adapter *adapter)
2389 struct e1000_hw *hw = &adapter->hw;
2390 bool link_active = false;
2392 /* get_link_status is set on LSC (link status) interrupt or rx
2393 * sequence error interrupt (except on intel ce4100).
2394 * get_link_status will stay false until the
2395 * e1000_check_for_link establishes link for copper adapters
2396 * ONLY
2398 switch (hw->media_type) {
2399 case e1000_media_type_copper:
2400 if (hw->mac_type == e1000_ce4100)
2401 hw->get_link_status = 1;
2402 if (hw->get_link_status) {
2403 e1000_check_for_link(hw);
2404 link_active = !hw->get_link_status;
2405 } else {
2406 link_active = true;
2408 break;
2409 case e1000_media_type_fiber:
2410 e1000_check_for_link(hw);
2411 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
2412 break;
2413 case e1000_media_type_internal_serdes:
2414 e1000_check_for_link(hw);
2415 link_active = hw->serdes_has_link;
2416 break;
2417 default:
2418 break;
2421 return link_active;
2425 * e1000_watchdog - work function
2426 * @work: work struct contained inside adapter struct
2428 static void e1000_watchdog(struct work_struct *work)
2430 struct e1000_adapter *adapter = container_of(work,
2431 struct e1000_adapter,
2432 watchdog_task.work);
2433 struct e1000_hw *hw = &adapter->hw;
2434 struct net_device *netdev = adapter->netdev;
2435 struct e1000_tx_ring *txdr = adapter->tx_ring;
2436 u32 link, tctl;
2438 if (test_bit(__E1000_DOWN, &adapter->flags))
2439 return;
2441 mutex_lock(&adapter->mutex);
2442 link = e1000_has_link(adapter);
2443 if ((netif_carrier_ok(netdev)) && link)
2444 goto link_up;
2446 if (link) {
2447 if (!netif_carrier_ok(netdev)) {
2448 u32 ctrl;
2449 bool txb2b = true;
2450 /* update snapshot of PHY registers on LSC */
2451 e1000_get_speed_and_duplex(hw,
2452 &adapter->link_speed,
2453 &adapter->link_duplex);
2455 ctrl = er32(CTRL);
2456 pr_info("%s NIC Link is Up %d Mbps %s, "
2457 "Flow Control: %s\n",
2458 netdev->name,
2459 adapter->link_speed,
2460 adapter->link_duplex == FULL_DUPLEX ?
2461 "Full Duplex" : "Half Duplex",
2462 ((ctrl & E1000_CTRL_TFCE) && (ctrl &
2463 E1000_CTRL_RFCE)) ? "RX/TX" : ((ctrl &
2464 E1000_CTRL_RFCE) ? "RX" : ((ctrl &
2465 E1000_CTRL_TFCE) ? "TX" : "None")));
2467 /* adjust timeout factor according to speed/duplex */
2468 adapter->tx_timeout_factor = 1;
2469 switch (adapter->link_speed) {
2470 case SPEED_10:
2471 txb2b = false;
2472 adapter->tx_timeout_factor = 16;
2473 break;
2474 case SPEED_100:
2475 txb2b = false;
2476 /* maybe add some timeout factor ? */
2477 break;
2480 /* enable transmits in the hardware */
2481 tctl = er32(TCTL);
2482 tctl |= E1000_TCTL_EN;
2483 ew32(TCTL, tctl);
2485 netif_carrier_on(netdev);
2486 if (!test_bit(__E1000_DOWN, &adapter->flags))
2487 schedule_delayed_work(&adapter->phy_info_task,
2488 2 * HZ);
2489 adapter->smartspeed = 0;
2491 } else {
2492 if (netif_carrier_ok(netdev)) {
2493 adapter->link_speed = 0;
2494 adapter->link_duplex = 0;
2495 pr_info("%s NIC Link is Down\n",
2496 netdev->name);
2497 netif_carrier_off(netdev);
2499 if (!test_bit(__E1000_DOWN, &adapter->flags))
2500 schedule_delayed_work(&adapter->phy_info_task,
2501 2 * HZ);
2504 e1000_smartspeed(adapter);
2507 link_up:
2508 e1000_update_stats(adapter);
2510 hw->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
2511 adapter->tpt_old = adapter->stats.tpt;
2512 hw->collision_delta = adapter->stats.colc - adapter->colc_old;
2513 adapter->colc_old = adapter->stats.colc;
2515 adapter->gorcl = adapter->stats.gorcl - adapter->gorcl_old;
2516 adapter->gorcl_old = adapter->stats.gorcl;
2517 adapter->gotcl = adapter->stats.gotcl - adapter->gotcl_old;
2518 adapter->gotcl_old = adapter->stats.gotcl;
2520 e1000_update_adaptive(hw);
2522 if (!netif_carrier_ok(netdev)) {
2523 if (E1000_DESC_UNUSED(txdr) + 1 < txdr->count) {
2524 /* We've lost link, so the controller stops DMA,
2525 * but we've got queued Tx work that's never going
2526 * to get done, so reset controller to flush Tx.
2527 * (Do the reset outside of interrupt context). */
2528 adapter->tx_timeout_count++;
2529 schedule_work(&adapter->reset_task);
2530 /* exit immediately since reset is imminent */
2531 goto unlock;
2535 /* Simple mode for Interrupt Throttle Rate (ITR) */
2536 if (hw->mac_type >= e1000_82540 && adapter->itr_setting == 4) {
2538 * Symmetric Tx/Rx gets a reduced ITR=2000;
2539 * Total asymmetrical Tx or Rx gets ITR=8000;
2540 * everyone else is between 2000-8000.
2542 u32 goc = (adapter->gotcl + adapter->gorcl) / 10000;
2543 u32 dif = (adapter->gotcl > adapter->gorcl ?
2544 adapter->gotcl - adapter->gorcl :
2545 adapter->gorcl - adapter->gotcl) / 10000;
2546 u32 itr = goc > 0 ? (dif * 6000 / goc + 2000) : 8000;
2548 ew32(ITR, 1000000000 / (itr * 256));
2551 /* Cause software interrupt to ensure rx ring is cleaned */
2552 ew32(ICS, E1000_ICS_RXDMT0);
2554 /* Force detection of hung controller every watchdog period */
2555 adapter->detect_tx_hung = true;
2557 /* Reschedule the task */
2558 if (!test_bit(__E1000_DOWN, &adapter->flags))
2559 schedule_delayed_work(&adapter->watchdog_task, 2 * HZ);
2561 unlock:
2562 mutex_unlock(&adapter->mutex);
2565 enum latency_range {
2566 lowest_latency = 0,
2567 low_latency = 1,
2568 bulk_latency = 2,
2569 latency_invalid = 255
2573 * e1000_update_itr - update the dynamic ITR value based on statistics
2574 * @adapter: pointer to adapter
2575 * @itr_setting: current adapter->itr
2576 * @packets: the number of packets during this measurement interval
2577 * @bytes: the number of bytes during this measurement interval
2579 * Stores a new ITR value based on packets and byte
2580 * counts during the last interrupt. The advantage of per interrupt
2581 * computation is faster updates and more accurate ITR for the current
2582 * traffic pattern. Constants in this function were computed
2583 * based on theoretical maximum wire speed and thresholds were set based
2584 * on testing data as well as attempting to minimize response time
2585 * while increasing bulk throughput.
2586 * this functionality is controlled by the InterruptThrottleRate module
2587 * parameter (see e1000_param.c)
2589 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
2590 u16 itr_setting, int packets, int bytes)
2592 unsigned int retval = itr_setting;
2593 struct e1000_hw *hw = &adapter->hw;
2595 if (unlikely(hw->mac_type < e1000_82540))
2596 goto update_itr_done;
2598 if (packets == 0)
2599 goto update_itr_done;
2601 switch (itr_setting) {
2602 case lowest_latency:
2603 /* jumbo frames get bulk treatment*/
2604 if (bytes/packets > 8000)
2605 retval = bulk_latency;
2606 else if ((packets < 5) && (bytes > 512))
2607 retval = low_latency;
2608 break;
2609 case low_latency: /* 50 usec aka 20000 ints/s */
2610 if (bytes > 10000) {
2611 /* jumbo frames need bulk latency setting */
2612 if (bytes/packets > 8000)
2613 retval = bulk_latency;
2614 else if ((packets < 10) || ((bytes/packets) > 1200))
2615 retval = bulk_latency;
2616 else if ((packets > 35))
2617 retval = lowest_latency;
2618 } else if (bytes/packets > 2000)
2619 retval = bulk_latency;
2620 else if (packets <= 2 && bytes < 512)
2621 retval = lowest_latency;
2622 break;
2623 case bulk_latency: /* 250 usec aka 4000 ints/s */
2624 if (bytes > 25000) {
2625 if (packets > 35)
2626 retval = low_latency;
2627 } else if (bytes < 6000) {
2628 retval = low_latency;
2630 break;
2633 update_itr_done:
2634 return retval;
2637 static void e1000_set_itr(struct e1000_adapter *adapter)
2639 struct e1000_hw *hw = &adapter->hw;
2640 u16 current_itr;
2641 u32 new_itr = adapter->itr;
2643 if (unlikely(hw->mac_type < e1000_82540))
2644 return;
2646 /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2647 if (unlikely(adapter->link_speed != SPEED_1000)) {
2648 current_itr = 0;
2649 new_itr = 4000;
2650 goto set_itr_now;
2653 adapter->tx_itr = e1000_update_itr(adapter,
2654 adapter->tx_itr,
2655 adapter->total_tx_packets,
2656 adapter->total_tx_bytes);
2657 /* conservative mode (itr 3) eliminates the lowest_latency setting */
2658 if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2659 adapter->tx_itr = low_latency;
2661 adapter->rx_itr = e1000_update_itr(adapter,
2662 adapter->rx_itr,
2663 adapter->total_rx_packets,
2664 adapter->total_rx_bytes);
2665 /* conservative mode (itr 3) eliminates the lowest_latency setting */
2666 if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2667 adapter->rx_itr = low_latency;
2669 current_itr = max(adapter->rx_itr, adapter->tx_itr);
2671 switch (current_itr) {
2672 /* counts and packets in update_itr are dependent on these numbers */
2673 case lowest_latency:
2674 new_itr = 70000;
2675 break;
2676 case low_latency:
2677 new_itr = 20000; /* aka hwitr = ~200 */
2678 break;
2679 case bulk_latency:
2680 new_itr = 4000;
2681 break;
2682 default:
2683 break;
2686 set_itr_now:
2687 if (new_itr != adapter->itr) {
2688 /* this attempts to bias the interrupt rate towards Bulk
2689 * by adding intermediate steps when interrupt rate is
2690 * increasing */
2691 new_itr = new_itr > adapter->itr ?
2692 min(adapter->itr + (new_itr >> 2), new_itr) :
2693 new_itr;
2694 adapter->itr = new_itr;
2695 ew32(ITR, 1000000000 / (new_itr * 256));
2699 #define E1000_TX_FLAGS_CSUM 0x00000001
2700 #define E1000_TX_FLAGS_VLAN 0x00000002
2701 #define E1000_TX_FLAGS_TSO 0x00000004
2702 #define E1000_TX_FLAGS_IPV4 0x00000008
2703 #define E1000_TX_FLAGS_VLAN_MASK 0xffff0000
2704 #define E1000_TX_FLAGS_VLAN_SHIFT 16
2706 static int e1000_tso(struct e1000_adapter *adapter,
2707 struct e1000_tx_ring *tx_ring, struct sk_buff *skb)
2709 struct e1000_context_desc *context_desc;
2710 struct e1000_buffer *buffer_info;
2711 unsigned int i;
2712 u32 cmd_length = 0;
2713 u16 ipcse = 0, tucse, mss;
2714 u8 ipcss, ipcso, tucss, tucso, hdr_len;
2715 int err;
2717 if (skb_is_gso(skb)) {
2718 if (skb_header_cloned(skb)) {
2719 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2720 if (err)
2721 return err;
2724 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
2725 mss = skb_shinfo(skb)->gso_size;
2726 if (skb->protocol == htons(ETH_P_IP)) {
2727 struct iphdr *iph = ip_hdr(skb);
2728 iph->tot_len = 0;
2729 iph->check = 0;
2730 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
2731 iph->daddr, 0,
2732 IPPROTO_TCP,
2734 cmd_length = E1000_TXD_CMD_IP;
2735 ipcse = skb_transport_offset(skb) - 1;
2736 } else if (skb->protocol == htons(ETH_P_IPV6)) {
2737 ipv6_hdr(skb)->payload_len = 0;
2738 tcp_hdr(skb)->check =
2739 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
2740 &ipv6_hdr(skb)->daddr,
2741 0, IPPROTO_TCP, 0);
2742 ipcse = 0;
2744 ipcss = skb_network_offset(skb);
2745 ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
2746 tucss = skb_transport_offset(skb);
2747 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
2748 tucse = 0;
2750 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
2751 E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
2753 i = tx_ring->next_to_use;
2754 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
2755 buffer_info = &tx_ring->buffer_info[i];
2757 context_desc->lower_setup.ip_fields.ipcss = ipcss;
2758 context_desc->lower_setup.ip_fields.ipcso = ipcso;
2759 context_desc->lower_setup.ip_fields.ipcse = cpu_to_le16(ipcse);
2760 context_desc->upper_setup.tcp_fields.tucss = tucss;
2761 context_desc->upper_setup.tcp_fields.tucso = tucso;
2762 context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
2763 context_desc->tcp_seg_setup.fields.mss = cpu_to_le16(mss);
2764 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
2765 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
2767 buffer_info->time_stamp = jiffies;
2768 buffer_info->next_to_watch = i;
2770 if (++i == tx_ring->count) i = 0;
2771 tx_ring->next_to_use = i;
2773 return true;
2775 return false;
2778 static bool e1000_tx_csum(struct e1000_adapter *adapter,
2779 struct e1000_tx_ring *tx_ring, struct sk_buff *skb)
2781 struct e1000_context_desc *context_desc;
2782 struct e1000_buffer *buffer_info;
2783 unsigned int i;
2784 u8 css;
2785 u32 cmd_len = E1000_TXD_CMD_DEXT;
2787 if (skb->ip_summed != CHECKSUM_PARTIAL)
2788 return false;
2790 switch (skb->protocol) {
2791 case cpu_to_be16(ETH_P_IP):
2792 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
2793 cmd_len |= E1000_TXD_CMD_TCP;
2794 break;
2795 case cpu_to_be16(ETH_P_IPV6):
2796 /* XXX not handling all IPV6 headers */
2797 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
2798 cmd_len |= E1000_TXD_CMD_TCP;
2799 break;
2800 default:
2801 if (unlikely(net_ratelimit()))
2802 e_warn(drv, "checksum_partial proto=%x!\n",
2803 skb->protocol);
2804 break;
2807 css = skb_checksum_start_offset(skb);
2809 i = tx_ring->next_to_use;
2810 buffer_info = &tx_ring->buffer_info[i];
2811 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
2813 context_desc->lower_setup.ip_config = 0;
2814 context_desc->upper_setup.tcp_fields.tucss = css;
2815 context_desc->upper_setup.tcp_fields.tucso =
2816 css + skb->csum_offset;
2817 context_desc->upper_setup.tcp_fields.tucse = 0;
2818 context_desc->tcp_seg_setup.data = 0;
2819 context_desc->cmd_and_length = cpu_to_le32(cmd_len);
2821 buffer_info->time_stamp = jiffies;
2822 buffer_info->next_to_watch = i;
2824 if (unlikely(++i == tx_ring->count)) i = 0;
2825 tx_ring->next_to_use = i;
2827 return true;
2830 #define E1000_MAX_TXD_PWR 12
2831 #define E1000_MAX_DATA_PER_TXD (1<<E1000_MAX_TXD_PWR)
2833 static int e1000_tx_map(struct e1000_adapter *adapter,
2834 struct e1000_tx_ring *tx_ring,
2835 struct sk_buff *skb, unsigned int first,
2836 unsigned int max_per_txd, unsigned int nr_frags,
2837 unsigned int mss)
2839 struct e1000_hw *hw = &adapter->hw;
2840 struct pci_dev *pdev = adapter->pdev;
2841 struct e1000_buffer *buffer_info;
2842 unsigned int len = skb_headlen(skb);
2843 unsigned int offset = 0, size, count = 0, i;
2844 unsigned int f, bytecount, segs;
2846 i = tx_ring->next_to_use;
2848 while (len) {
2849 buffer_info = &tx_ring->buffer_info[i];
2850 size = min(len, max_per_txd);
2851 /* Workaround for Controller erratum --
2852 * descriptor for non-tso packet in a linear SKB that follows a
2853 * tso gets written back prematurely before the data is fully
2854 * DMA'd to the controller */
2855 if (!skb->data_len && tx_ring->last_tx_tso &&
2856 !skb_is_gso(skb)) {
2857 tx_ring->last_tx_tso = false;
2858 size -= 4;
2861 /* Workaround for premature desc write-backs
2862 * in TSO mode. Append 4-byte sentinel desc */
2863 if (unlikely(mss && !nr_frags && size == len && size > 8))
2864 size -= 4;
2865 /* work-around for errata 10 and it applies
2866 * to all controllers in PCI-X mode
2867 * The fix is to make sure that the first descriptor of a
2868 * packet is smaller than 2048 - 16 - 16 (or 2016) bytes
2870 if (unlikely((hw->bus_type == e1000_bus_type_pcix) &&
2871 (size > 2015) && count == 0))
2872 size = 2015;
2874 /* Workaround for potential 82544 hang in PCI-X. Avoid
2875 * terminating buffers within evenly-aligned dwords. */
2876 if (unlikely(adapter->pcix_82544 &&
2877 !((unsigned long)(skb->data + offset + size - 1) & 4) &&
2878 size > 4))
2879 size -= 4;
2881 buffer_info->length = size;
2882 /* set time_stamp *before* dma to help avoid a possible race */
2883 buffer_info->time_stamp = jiffies;
2884 buffer_info->mapped_as_page = false;
2885 buffer_info->dma = dma_map_single(&pdev->dev,
2886 skb->data + offset,
2887 size, DMA_TO_DEVICE);
2888 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2889 goto dma_error;
2890 buffer_info->next_to_watch = i;
2892 len -= size;
2893 offset += size;
2894 count++;
2895 if (len) {
2896 i++;
2897 if (unlikely(i == tx_ring->count))
2898 i = 0;
2902 for (f = 0; f < nr_frags; f++) {
2903 const struct skb_frag_struct *frag;
2905 frag = &skb_shinfo(skb)->frags[f];
2906 len = skb_frag_size(frag);
2907 offset = 0;
2909 while (len) {
2910 unsigned long bufend;
2911 i++;
2912 if (unlikely(i == tx_ring->count))
2913 i = 0;
2915 buffer_info = &tx_ring->buffer_info[i];
2916 size = min(len, max_per_txd);
2917 /* Workaround for premature desc write-backs
2918 * in TSO mode. Append 4-byte sentinel desc */
2919 if (unlikely(mss && f == (nr_frags-1) && size == len && size > 8))
2920 size -= 4;
2921 /* Workaround for potential 82544 hang in PCI-X.
2922 * Avoid terminating buffers within evenly-aligned
2923 * dwords. */
2924 bufend = (unsigned long)
2925 page_to_phys(skb_frag_page(frag));
2926 bufend += offset + size - 1;
2927 if (unlikely(adapter->pcix_82544 &&
2928 !(bufend & 4) &&
2929 size > 4))
2930 size -= 4;
2932 buffer_info->length = size;
2933 buffer_info->time_stamp = jiffies;
2934 buffer_info->mapped_as_page = true;
2935 buffer_info->dma = skb_frag_dma_map(&pdev->dev, frag,
2936 offset, size, DMA_TO_DEVICE);
2937 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2938 goto dma_error;
2939 buffer_info->next_to_watch = i;
2941 len -= size;
2942 offset += size;
2943 count++;
2947 segs = skb_shinfo(skb)->gso_segs ?: 1;
2948 /* multiply data chunks by size of headers */
2949 bytecount = ((segs - 1) * skb_headlen(skb)) + skb->len;
2951 tx_ring->buffer_info[i].skb = skb;
2952 tx_ring->buffer_info[i].segs = segs;
2953 tx_ring->buffer_info[i].bytecount = bytecount;
2954 tx_ring->buffer_info[first].next_to_watch = i;
2956 return count;
2958 dma_error:
2959 dev_err(&pdev->dev, "TX DMA map failed\n");
2960 buffer_info->dma = 0;
2961 if (count)
2962 count--;
2964 while (count--) {
2965 if (i==0)
2966 i += tx_ring->count;
2967 i--;
2968 buffer_info = &tx_ring->buffer_info[i];
2969 e1000_unmap_and_free_tx_resource(adapter, buffer_info);
2972 return 0;
2975 static void e1000_tx_queue(struct e1000_adapter *adapter,
2976 struct e1000_tx_ring *tx_ring, int tx_flags,
2977 int count)
2979 struct e1000_hw *hw = &adapter->hw;
2980 struct e1000_tx_desc *tx_desc = NULL;
2981 struct e1000_buffer *buffer_info;
2982 u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
2983 unsigned int i;
2985 if (likely(tx_flags & E1000_TX_FLAGS_TSO)) {
2986 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
2987 E1000_TXD_CMD_TSE;
2988 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
2990 if (likely(tx_flags & E1000_TX_FLAGS_IPV4))
2991 txd_upper |= E1000_TXD_POPTS_IXSM << 8;
2994 if (likely(tx_flags & E1000_TX_FLAGS_CSUM)) {
2995 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
2996 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
2999 if (unlikely(tx_flags & E1000_TX_FLAGS_VLAN)) {
3000 txd_lower |= E1000_TXD_CMD_VLE;
3001 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
3004 i = tx_ring->next_to_use;
3006 while (count--) {
3007 buffer_info = &tx_ring->buffer_info[i];
3008 tx_desc = E1000_TX_DESC(*tx_ring, i);
3009 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
3010 tx_desc->lower.data =
3011 cpu_to_le32(txd_lower | buffer_info->length);
3012 tx_desc->upper.data = cpu_to_le32(txd_upper);
3013 if (unlikely(++i == tx_ring->count)) i = 0;
3016 tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
3018 /* Force memory writes to complete before letting h/w
3019 * know there are new descriptors to fetch. (Only
3020 * applicable for weak-ordered memory model archs,
3021 * such as IA-64). */
3022 wmb();
3024 tx_ring->next_to_use = i;
3025 writel(i, hw->hw_addr + tx_ring->tdt);
3026 /* we need this if more than one processor can write to our tail
3027 * at a time, it syncronizes IO on IA64/Altix systems */
3028 mmiowb();
3032 * 82547 workaround to avoid controller hang in half-duplex environment.
3033 * The workaround is to avoid queuing a large packet that would span
3034 * the internal Tx FIFO ring boundary by notifying the stack to resend
3035 * the packet at a later time. This gives the Tx FIFO an opportunity to
3036 * flush all packets. When that occurs, we reset the Tx FIFO pointers
3037 * to the beginning of the Tx FIFO.
3040 #define E1000_FIFO_HDR 0x10
3041 #define E1000_82547_PAD_LEN 0x3E0
3043 static int e1000_82547_fifo_workaround(struct e1000_adapter *adapter,
3044 struct sk_buff *skb)
3046 u32 fifo_space = adapter->tx_fifo_size - adapter->tx_fifo_head;
3047 u32 skb_fifo_len = skb->len + E1000_FIFO_HDR;
3049 skb_fifo_len = ALIGN(skb_fifo_len, E1000_FIFO_HDR);
3051 if (adapter->link_duplex != HALF_DUPLEX)
3052 goto no_fifo_stall_required;
3054 if (atomic_read(&adapter->tx_fifo_stall))
3055 return 1;
3057 if (skb_fifo_len >= (E1000_82547_PAD_LEN + fifo_space)) {
3058 atomic_set(&adapter->tx_fifo_stall, 1);
3059 return 1;
3062 no_fifo_stall_required:
3063 adapter->tx_fifo_head += skb_fifo_len;
3064 if (adapter->tx_fifo_head >= adapter->tx_fifo_size)
3065 adapter->tx_fifo_head -= adapter->tx_fifo_size;
3066 return 0;
3069 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
3071 struct e1000_adapter *adapter = netdev_priv(netdev);
3072 struct e1000_tx_ring *tx_ring = adapter->tx_ring;
3074 netif_stop_queue(netdev);
3075 /* Herbert's original patch had:
3076 * smp_mb__after_netif_stop_queue();
3077 * but since that doesn't exist yet, just open code it. */
3078 smp_mb();
3080 /* We need to check again in a case another CPU has just
3081 * made room available. */
3082 if (likely(E1000_DESC_UNUSED(tx_ring) < size))
3083 return -EBUSY;
3085 /* A reprieve! */
3086 netif_start_queue(netdev);
3087 ++adapter->restart_queue;
3088 return 0;
3091 static int e1000_maybe_stop_tx(struct net_device *netdev,
3092 struct e1000_tx_ring *tx_ring, int size)
3094 if (likely(E1000_DESC_UNUSED(tx_ring) >= size))
3095 return 0;
3096 return __e1000_maybe_stop_tx(netdev, size);
3099 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
3100 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
3101 struct net_device *netdev)
3103 struct e1000_adapter *adapter = netdev_priv(netdev);
3104 struct e1000_hw *hw = &adapter->hw;
3105 struct e1000_tx_ring *tx_ring;
3106 unsigned int first, max_per_txd = E1000_MAX_DATA_PER_TXD;
3107 unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
3108 unsigned int tx_flags = 0;
3109 unsigned int len = skb_headlen(skb);
3110 unsigned int nr_frags;
3111 unsigned int mss;
3112 int count = 0;
3113 int tso;
3114 unsigned int f;
3116 /* This goes back to the question of how to logically map a tx queue
3117 * to a flow. Right now, performance is impacted slightly negatively
3118 * if using multiple tx queues. If the stack breaks away from a
3119 * single qdisc implementation, we can look at this again. */
3120 tx_ring = adapter->tx_ring;
3122 if (unlikely(skb->len <= 0)) {
3123 dev_kfree_skb_any(skb);
3124 return NETDEV_TX_OK;
3127 mss = skb_shinfo(skb)->gso_size;
3128 /* The controller does a simple calculation to
3129 * make sure there is enough room in the FIFO before
3130 * initiating the DMA for each buffer. The calc is:
3131 * 4 = ceil(buffer len/mss). To make sure we don't
3132 * overrun the FIFO, adjust the max buffer len if mss
3133 * drops. */
3134 if (mss) {
3135 u8 hdr_len;
3136 max_per_txd = min(mss << 2, max_per_txd);
3137 max_txd_pwr = fls(max_per_txd) - 1;
3139 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
3140 if (skb->data_len && hdr_len == len) {
3141 switch (hw->mac_type) {
3142 unsigned int pull_size;
3143 case e1000_82544:
3144 /* Make sure we have room to chop off 4 bytes,
3145 * and that the end alignment will work out to
3146 * this hardware's requirements
3147 * NOTE: this is a TSO only workaround
3148 * if end byte alignment not correct move us
3149 * into the next dword */
3150 if ((unsigned long)(skb_tail_pointer(skb) - 1) & 4)
3151 break;
3152 /* fall through */
3153 pull_size = min((unsigned int)4, skb->data_len);
3154 if (!__pskb_pull_tail(skb, pull_size)) {
3155 e_err(drv, "__pskb_pull_tail "
3156 "failed.\n");
3157 dev_kfree_skb_any(skb);
3158 return NETDEV_TX_OK;
3160 len = skb_headlen(skb);
3161 break;
3162 default:
3163 /* do nothing */
3164 break;
3169 /* reserve a descriptor for the offload context */
3170 if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
3171 count++;
3172 count++;
3174 /* Controller Erratum workaround */
3175 if (!skb->data_len && tx_ring->last_tx_tso && !skb_is_gso(skb))
3176 count++;
3178 count += TXD_USE_COUNT(len, max_txd_pwr);
3180 if (adapter->pcix_82544)
3181 count++;
3183 /* work-around for errata 10 and it applies to all controllers
3184 * in PCI-X mode, so add one more descriptor to the count
3186 if (unlikely((hw->bus_type == e1000_bus_type_pcix) &&
3187 (len > 2015)))
3188 count++;
3190 nr_frags = skb_shinfo(skb)->nr_frags;
3191 for (f = 0; f < nr_frags; f++)
3192 count += TXD_USE_COUNT(skb_frag_size(&skb_shinfo(skb)->frags[f]),
3193 max_txd_pwr);
3194 if (adapter->pcix_82544)
3195 count += nr_frags;
3197 /* need: count + 2 desc gap to keep tail from touching
3198 * head, otherwise try next time */
3199 if (unlikely(e1000_maybe_stop_tx(netdev, tx_ring, count + 2)))
3200 return NETDEV_TX_BUSY;
3202 if (unlikely((hw->mac_type == e1000_82547) &&
3203 (e1000_82547_fifo_workaround(adapter, skb)))) {
3204 netif_stop_queue(netdev);
3205 if (!test_bit(__E1000_DOWN, &adapter->flags))
3206 schedule_delayed_work(&adapter->fifo_stall_task, 1);
3207 return NETDEV_TX_BUSY;
3210 if (vlan_tx_tag_present(skb)) {
3211 tx_flags |= E1000_TX_FLAGS_VLAN;
3212 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
3215 first = tx_ring->next_to_use;
3217 tso = e1000_tso(adapter, tx_ring, skb);
3218 if (tso < 0) {
3219 dev_kfree_skb_any(skb);
3220 return NETDEV_TX_OK;
3223 if (likely(tso)) {
3224 if (likely(hw->mac_type != e1000_82544))
3225 tx_ring->last_tx_tso = true;
3226 tx_flags |= E1000_TX_FLAGS_TSO;
3227 } else if (likely(e1000_tx_csum(adapter, tx_ring, skb)))
3228 tx_flags |= E1000_TX_FLAGS_CSUM;
3230 if (likely(skb->protocol == htons(ETH_P_IP)))
3231 tx_flags |= E1000_TX_FLAGS_IPV4;
3233 count = e1000_tx_map(adapter, tx_ring, skb, first, max_per_txd,
3234 nr_frags, mss);
3236 if (count) {
3237 e1000_tx_queue(adapter, tx_ring, tx_flags, count);
3238 /* Make sure there is space in the ring for the next send. */
3239 e1000_maybe_stop_tx(netdev, tx_ring, MAX_SKB_FRAGS + 2);
3241 } else {
3242 dev_kfree_skb_any(skb);
3243 tx_ring->buffer_info[first].time_stamp = 0;
3244 tx_ring->next_to_use = first;
3247 return NETDEV_TX_OK;
3251 * e1000_tx_timeout - Respond to a Tx Hang
3252 * @netdev: network interface device structure
3255 static void e1000_tx_timeout(struct net_device *netdev)
3257 struct e1000_adapter *adapter = netdev_priv(netdev);
3259 /* Do the reset outside of interrupt context */
3260 adapter->tx_timeout_count++;
3261 schedule_work(&adapter->reset_task);
3264 static void e1000_reset_task(struct work_struct *work)
3266 struct e1000_adapter *adapter =
3267 container_of(work, struct e1000_adapter, reset_task);
3269 if (test_bit(__E1000_DOWN, &adapter->flags))
3270 return;
3271 e1000_reinit_safe(adapter);
3275 * e1000_get_stats - Get System Network Statistics
3276 * @netdev: network interface device structure
3278 * Returns the address of the device statistics structure.
3279 * The statistics are actually updated from the watchdog.
3282 static struct net_device_stats *e1000_get_stats(struct net_device *netdev)
3284 /* only return the current stats */
3285 return &netdev->stats;
3289 * e1000_change_mtu - Change the Maximum Transfer Unit
3290 * @netdev: network interface device structure
3291 * @new_mtu: new value for maximum frame size
3293 * Returns 0 on success, negative on failure
3296 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
3298 struct e1000_adapter *adapter = netdev_priv(netdev);
3299 struct e1000_hw *hw = &adapter->hw;
3300 int max_frame = new_mtu + ENET_HEADER_SIZE + ETHERNET_FCS_SIZE;
3302 if ((max_frame < MINIMUM_ETHERNET_FRAME_SIZE) ||
3303 (max_frame > MAX_JUMBO_FRAME_SIZE)) {
3304 e_err(probe, "Invalid MTU setting\n");
3305 return -EINVAL;
3308 /* Adapter-specific max frame size limits. */
3309 switch (hw->mac_type) {
3310 case e1000_undefined ... e1000_82542_rev2_1:
3311 if (max_frame > (ETH_FRAME_LEN + ETH_FCS_LEN)) {
3312 e_err(probe, "Jumbo Frames not supported.\n");
3313 return -EINVAL;
3315 break;
3316 default:
3317 /* Capable of supporting up to MAX_JUMBO_FRAME_SIZE limit. */
3318 break;
3321 while (test_and_set_bit(__E1000_RESETTING, &adapter->flags))
3322 msleep(1);
3323 /* e1000_down has a dependency on max_frame_size */
3324 hw->max_frame_size = max_frame;
3325 if (netif_running(netdev))
3326 e1000_down(adapter);
3328 /* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
3329 * means we reserve 2 more, this pushes us to allocate from the next
3330 * larger slab size.
3331 * i.e. RXBUFFER_2048 --> size-4096 slab
3332 * however with the new *_jumbo_rx* routines, jumbo receives will use
3333 * fragmented skbs */
3335 if (max_frame <= E1000_RXBUFFER_2048)
3336 adapter->rx_buffer_len = E1000_RXBUFFER_2048;
3337 else
3338 #if (PAGE_SIZE >= E1000_RXBUFFER_16384)
3339 adapter->rx_buffer_len = E1000_RXBUFFER_16384;
3340 #elif (PAGE_SIZE >= E1000_RXBUFFER_4096)
3341 adapter->rx_buffer_len = PAGE_SIZE;
3342 #endif
3344 /* adjust allocation if LPE protects us, and we aren't using SBP */
3345 if (!hw->tbi_compatibility_on &&
3346 ((max_frame == (ETH_FRAME_LEN + ETH_FCS_LEN)) ||
3347 (max_frame == MAXIMUM_ETHERNET_VLAN_SIZE)))
3348 adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
3350 pr_info("%s changing MTU from %d to %d\n",
3351 netdev->name, netdev->mtu, new_mtu);
3352 netdev->mtu = new_mtu;
3354 if (netif_running(netdev))
3355 e1000_up(adapter);
3356 else
3357 e1000_reset(adapter);
3359 clear_bit(__E1000_RESETTING, &adapter->flags);
3361 return 0;
3365 * e1000_update_stats - Update the board statistics counters
3366 * @adapter: board private structure
3369 void e1000_update_stats(struct e1000_adapter *adapter)
3371 struct net_device *netdev = adapter->netdev;
3372 struct e1000_hw *hw = &adapter->hw;
3373 struct pci_dev *pdev = adapter->pdev;
3374 unsigned long flags;
3375 u16 phy_tmp;
3377 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
3380 * Prevent stats update while adapter is being reset, or if the pci
3381 * connection is down.
3383 if (adapter->link_speed == 0)
3384 return;
3385 if (pci_channel_offline(pdev))
3386 return;
3388 spin_lock_irqsave(&adapter->stats_lock, flags);
3390 /* these counters are modified from e1000_tbi_adjust_stats,
3391 * called from the interrupt context, so they must only
3392 * be written while holding adapter->stats_lock
3395 adapter->stats.crcerrs += er32(CRCERRS);
3396 adapter->stats.gprc += er32(GPRC);
3397 adapter->stats.gorcl += er32(GORCL);
3398 adapter->stats.gorch += er32(GORCH);
3399 adapter->stats.bprc += er32(BPRC);
3400 adapter->stats.mprc += er32(MPRC);
3401 adapter->stats.roc += er32(ROC);
3403 adapter->stats.prc64 += er32(PRC64);
3404 adapter->stats.prc127 += er32(PRC127);
3405 adapter->stats.prc255 += er32(PRC255);
3406 adapter->stats.prc511 += er32(PRC511);
3407 adapter->stats.prc1023 += er32(PRC1023);
3408 adapter->stats.prc1522 += er32(PRC1522);
3410 adapter->stats.symerrs += er32(SYMERRS);
3411 adapter->stats.mpc += er32(MPC);
3412 adapter->stats.scc += er32(SCC);
3413 adapter->stats.ecol += er32(ECOL);
3414 adapter->stats.mcc += er32(MCC);
3415 adapter->stats.latecol += er32(LATECOL);
3416 adapter->stats.dc += er32(DC);
3417 adapter->stats.sec += er32(SEC);
3418 adapter->stats.rlec += er32(RLEC);
3419 adapter->stats.xonrxc += er32(XONRXC);
3420 adapter->stats.xontxc += er32(XONTXC);
3421 adapter->stats.xoffrxc += er32(XOFFRXC);
3422 adapter->stats.xofftxc += er32(XOFFTXC);
3423 adapter->stats.fcruc += er32(FCRUC);
3424 adapter->stats.gptc += er32(GPTC);
3425 adapter->stats.gotcl += er32(GOTCL);
3426 adapter->stats.gotch += er32(GOTCH);
3427 adapter->stats.rnbc += er32(RNBC);
3428 adapter->stats.ruc += er32(RUC);
3429 adapter->stats.rfc += er32(RFC);
3430 adapter->stats.rjc += er32(RJC);
3431 adapter->stats.torl += er32(TORL);
3432 adapter->stats.torh += er32(TORH);
3433 adapter->stats.totl += er32(TOTL);
3434 adapter->stats.toth += er32(TOTH);
3435 adapter->stats.tpr += er32(TPR);
3437 adapter->stats.ptc64 += er32(PTC64);
3438 adapter->stats.ptc127 += er32(PTC127);
3439 adapter->stats.ptc255 += er32(PTC255);
3440 adapter->stats.ptc511 += er32(PTC511);
3441 adapter->stats.ptc1023 += er32(PTC1023);
3442 adapter->stats.ptc1522 += er32(PTC1522);
3444 adapter->stats.mptc += er32(MPTC);
3445 adapter->stats.bptc += er32(BPTC);
3447 /* used for adaptive IFS */
3449 hw->tx_packet_delta = er32(TPT);
3450 adapter->stats.tpt += hw->tx_packet_delta;
3451 hw->collision_delta = er32(COLC);
3452 adapter->stats.colc += hw->collision_delta;
3454 if (hw->mac_type >= e1000_82543) {
3455 adapter->stats.algnerrc += er32(ALGNERRC);
3456 adapter->stats.rxerrc += er32(RXERRC);
3457 adapter->stats.tncrs += er32(TNCRS);
3458 adapter->stats.cexterr += er32(CEXTERR);
3459 adapter->stats.tsctc += er32(TSCTC);
3460 adapter->stats.tsctfc += er32(TSCTFC);
3463 /* Fill out the OS statistics structure */
3464 netdev->stats.multicast = adapter->stats.mprc;
3465 netdev->stats.collisions = adapter->stats.colc;
3467 /* Rx Errors */
3469 /* RLEC on some newer hardware can be incorrect so build
3470 * our own version based on RUC and ROC */
3471 netdev->stats.rx_errors = adapter->stats.rxerrc +
3472 adapter->stats.crcerrs + adapter->stats.algnerrc +
3473 adapter->stats.ruc + adapter->stats.roc +
3474 adapter->stats.cexterr;
3475 adapter->stats.rlerrc = adapter->stats.ruc + adapter->stats.roc;
3476 netdev->stats.rx_length_errors = adapter->stats.rlerrc;
3477 netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
3478 netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
3479 netdev->stats.rx_missed_errors = adapter->stats.mpc;
3481 /* Tx Errors */
3482 adapter->stats.txerrc = adapter->stats.ecol + adapter->stats.latecol;
3483 netdev->stats.tx_errors = adapter->stats.txerrc;
3484 netdev->stats.tx_aborted_errors = adapter->stats.ecol;
3485 netdev->stats.tx_window_errors = adapter->stats.latecol;
3486 netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
3487 if (hw->bad_tx_carr_stats_fd &&
3488 adapter->link_duplex == FULL_DUPLEX) {
3489 netdev->stats.tx_carrier_errors = 0;
3490 adapter->stats.tncrs = 0;
3493 /* Tx Dropped needs to be maintained elsewhere */
3495 /* Phy Stats */
3496 if (hw->media_type == e1000_media_type_copper) {
3497 if ((adapter->link_speed == SPEED_1000) &&
3498 (!e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_tmp))) {
3499 phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
3500 adapter->phy_stats.idle_errors += phy_tmp;
3503 if ((hw->mac_type <= e1000_82546) &&
3504 (hw->phy_type == e1000_phy_m88) &&
3505 !e1000_read_phy_reg(hw, M88E1000_RX_ERR_CNTR, &phy_tmp))
3506 adapter->phy_stats.receive_errors += phy_tmp;
3509 /* Management Stats */
3510 if (hw->has_smbus) {
3511 adapter->stats.mgptc += er32(MGTPTC);
3512 adapter->stats.mgprc += er32(MGTPRC);
3513 adapter->stats.mgpdc += er32(MGTPDC);
3516 spin_unlock_irqrestore(&adapter->stats_lock, flags);
3520 * e1000_intr - Interrupt Handler
3521 * @irq: interrupt number
3522 * @data: pointer to a network interface device structure
3525 static irqreturn_t e1000_intr(int irq, void *data)
3527 struct net_device *netdev = data;
3528 struct e1000_adapter *adapter = netdev_priv(netdev);
3529 struct e1000_hw *hw = &adapter->hw;
3530 u32 icr = er32(ICR);
3532 if (unlikely((!icr)))
3533 return IRQ_NONE; /* Not our interrupt */
3536 * we might have caused the interrupt, but the above
3537 * read cleared it, and just in case the driver is
3538 * down there is nothing to do so return handled
3540 if (unlikely(test_bit(__E1000_DOWN, &adapter->flags)))
3541 return IRQ_HANDLED;
3543 if (unlikely(icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC))) {
3544 hw->get_link_status = 1;
3545 /* guard against interrupt when we're going down */
3546 if (!test_bit(__E1000_DOWN, &adapter->flags))
3547 schedule_delayed_work(&adapter->watchdog_task, 1);
3550 /* disable interrupts, without the synchronize_irq bit */
3551 ew32(IMC, ~0);
3552 E1000_WRITE_FLUSH();
3554 if (likely(napi_schedule_prep(&adapter->napi))) {
3555 adapter->total_tx_bytes = 0;
3556 adapter->total_tx_packets = 0;
3557 adapter->total_rx_bytes = 0;
3558 adapter->total_rx_packets = 0;
3559 __napi_schedule(&adapter->napi);
3560 } else {
3561 /* this really should not happen! if it does it is basically a
3562 * bug, but not a hard error, so enable ints and continue */
3563 if (!test_bit(__E1000_DOWN, &adapter->flags))
3564 e1000_irq_enable(adapter);
3567 return IRQ_HANDLED;
3571 * e1000_clean - NAPI Rx polling callback
3572 * @adapter: board private structure
3574 static int e1000_clean(struct napi_struct *napi, int budget)
3576 struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
3577 int tx_clean_complete = 0, work_done = 0;
3579 tx_clean_complete = e1000_clean_tx_irq(adapter, &adapter->tx_ring[0]);
3581 adapter->clean_rx(adapter, &adapter->rx_ring[0], &work_done, budget);
3583 if (!tx_clean_complete)
3584 work_done = budget;
3586 /* If budget not fully consumed, exit the polling mode */
3587 if (work_done < budget) {
3588 if (likely(adapter->itr_setting & 3))
3589 e1000_set_itr(adapter);
3590 napi_complete(napi);
3591 if (!test_bit(__E1000_DOWN, &adapter->flags))
3592 e1000_irq_enable(adapter);
3595 return work_done;
3599 * e1000_clean_tx_irq - Reclaim resources after transmit completes
3600 * @adapter: board private structure
3602 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter,
3603 struct e1000_tx_ring *tx_ring)
3605 struct e1000_hw *hw = &adapter->hw;
3606 struct net_device *netdev = adapter->netdev;
3607 struct e1000_tx_desc *tx_desc, *eop_desc;
3608 struct e1000_buffer *buffer_info;
3609 unsigned int i, eop;
3610 unsigned int count = 0;
3611 unsigned int total_tx_bytes=0, total_tx_packets=0;
3613 i = tx_ring->next_to_clean;
3614 eop = tx_ring->buffer_info[i].next_to_watch;
3615 eop_desc = E1000_TX_DESC(*tx_ring, eop);
3617 while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
3618 (count < tx_ring->count)) {
3619 bool cleaned = false;
3620 rmb(); /* read buffer_info after eop_desc */
3621 for ( ; !cleaned; count++) {
3622 tx_desc = E1000_TX_DESC(*tx_ring, i);
3623 buffer_info = &tx_ring->buffer_info[i];
3624 cleaned = (i == eop);
3626 if (cleaned) {
3627 total_tx_packets += buffer_info->segs;
3628 total_tx_bytes += buffer_info->bytecount;
3630 e1000_unmap_and_free_tx_resource(adapter, buffer_info);
3631 tx_desc->upper.data = 0;
3633 if (unlikely(++i == tx_ring->count)) i = 0;
3636 eop = tx_ring->buffer_info[i].next_to_watch;
3637 eop_desc = E1000_TX_DESC(*tx_ring, eop);
3640 tx_ring->next_to_clean = i;
3642 #define TX_WAKE_THRESHOLD 32
3643 if (unlikely(count && netif_carrier_ok(netdev) &&
3644 E1000_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD)) {
3645 /* Make sure that anybody stopping the queue after this
3646 * sees the new next_to_clean.
3648 smp_mb();
3650 if (netif_queue_stopped(netdev) &&
3651 !(test_bit(__E1000_DOWN, &adapter->flags))) {
3652 netif_wake_queue(netdev);
3653 ++adapter->restart_queue;
3657 if (adapter->detect_tx_hung) {
3658 /* Detect a transmit hang in hardware, this serializes the
3659 * check with the clearing of time_stamp and movement of i */
3660 adapter->detect_tx_hung = false;
3661 if (tx_ring->buffer_info[eop].time_stamp &&
3662 time_after(jiffies, tx_ring->buffer_info[eop].time_stamp +
3663 (adapter->tx_timeout_factor * HZ)) &&
3664 !(er32(STATUS) & E1000_STATUS_TXOFF)) {
3666 /* detected Tx unit hang */
3667 e_err(drv, "Detected Tx Unit Hang\n"
3668 " Tx Queue <%lu>\n"
3669 " TDH <%x>\n"
3670 " TDT <%x>\n"
3671 " next_to_use <%x>\n"
3672 " next_to_clean <%x>\n"
3673 "buffer_info[next_to_clean]\n"
3674 " time_stamp <%lx>\n"
3675 " next_to_watch <%x>\n"
3676 " jiffies <%lx>\n"
3677 " next_to_watch.status <%x>\n",
3678 (unsigned long)((tx_ring - adapter->tx_ring) /
3679 sizeof(struct e1000_tx_ring)),
3680 readl(hw->hw_addr + tx_ring->tdh),
3681 readl(hw->hw_addr + tx_ring->tdt),
3682 tx_ring->next_to_use,
3683 tx_ring->next_to_clean,
3684 tx_ring->buffer_info[eop].time_stamp,
3685 eop,
3686 jiffies,
3687 eop_desc->upper.fields.status);
3688 netif_stop_queue(netdev);
3691 adapter->total_tx_bytes += total_tx_bytes;
3692 adapter->total_tx_packets += total_tx_packets;
3693 netdev->stats.tx_bytes += total_tx_bytes;
3694 netdev->stats.tx_packets += total_tx_packets;
3695 return count < tx_ring->count;
3699 * e1000_rx_checksum - Receive Checksum Offload for 82543
3700 * @adapter: board private structure
3701 * @status_err: receive descriptor status and error fields
3702 * @csum: receive descriptor csum field
3703 * @sk_buff: socket buffer with received data
3706 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
3707 u32 csum, struct sk_buff *skb)
3709 struct e1000_hw *hw = &adapter->hw;
3710 u16 status = (u16)status_err;
3711 u8 errors = (u8)(status_err >> 24);
3713 skb_checksum_none_assert(skb);
3715 /* 82543 or newer only */
3716 if (unlikely(hw->mac_type < e1000_82543)) return;
3717 /* Ignore Checksum bit is set */
3718 if (unlikely(status & E1000_RXD_STAT_IXSM)) return;
3719 /* TCP/UDP checksum error bit is set */
3720 if (unlikely(errors & E1000_RXD_ERR_TCPE)) {
3721 /* let the stack verify checksum errors */
3722 adapter->hw_csum_err++;
3723 return;
3725 /* TCP/UDP Checksum has not been calculated */
3726 if (!(status & E1000_RXD_STAT_TCPCS))
3727 return;
3729 /* It must be a TCP or UDP packet with a valid checksum */
3730 if (likely(status & E1000_RXD_STAT_TCPCS)) {
3731 /* TCP checksum is good */
3732 skb->ip_summed = CHECKSUM_UNNECESSARY;
3734 adapter->hw_csum_good++;
3738 * e1000_consume_page - helper function
3740 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
3741 u16 length)
3743 bi->page = NULL;
3744 skb->len += length;
3745 skb->data_len += length;
3746 skb->truesize += PAGE_SIZE;
3750 * e1000_receive_skb - helper function to handle rx indications
3751 * @adapter: board private structure
3752 * @status: descriptor status field as written by hardware
3753 * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
3754 * @skb: pointer to sk_buff to be indicated to stack
3756 static void e1000_receive_skb(struct e1000_adapter *adapter, u8 status,
3757 __le16 vlan, struct sk_buff *skb)
3759 skb->protocol = eth_type_trans(skb, adapter->netdev);
3761 if (status & E1000_RXD_STAT_VP) {
3762 u16 vid = le16_to_cpu(vlan) & E1000_RXD_SPC_VLAN_MASK;
3764 __vlan_hwaccel_put_tag(skb, vid);
3766 napi_gro_receive(&adapter->napi, skb);
3770 * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
3771 * @adapter: board private structure
3772 * @rx_ring: ring to clean
3773 * @work_done: amount of napi work completed this call
3774 * @work_to_do: max amount of work allowed for this call to do
3776 * the return value indicates whether actual cleaning was done, there
3777 * is no guarantee that everything was cleaned
3779 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
3780 struct e1000_rx_ring *rx_ring,
3781 int *work_done, int work_to_do)
3783 struct e1000_hw *hw = &adapter->hw;
3784 struct net_device *netdev = adapter->netdev;
3785 struct pci_dev *pdev = adapter->pdev;
3786 struct e1000_rx_desc *rx_desc, *next_rxd;
3787 struct e1000_buffer *buffer_info, *next_buffer;
3788 unsigned long irq_flags;
3789 u32 length;
3790 unsigned int i;
3791 int cleaned_count = 0;
3792 bool cleaned = false;
3793 unsigned int total_rx_bytes=0, total_rx_packets=0;
3795 i = rx_ring->next_to_clean;
3796 rx_desc = E1000_RX_DESC(*rx_ring, i);
3797 buffer_info = &rx_ring->buffer_info[i];
3799 while (rx_desc->status & E1000_RXD_STAT_DD) {
3800 struct sk_buff *skb;
3801 u8 status;
3803 if (*work_done >= work_to_do)
3804 break;
3805 (*work_done)++;
3806 rmb(); /* read descriptor and rx_buffer_info after status DD */
3808 status = rx_desc->status;
3809 skb = buffer_info->skb;
3810 buffer_info->skb = NULL;
3812 if (++i == rx_ring->count) i = 0;
3813 next_rxd = E1000_RX_DESC(*rx_ring, i);
3814 prefetch(next_rxd);
3816 next_buffer = &rx_ring->buffer_info[i];
3818 cleaned = true;
3819 cleaned_count++;
3820 dma_unmap_page(&pdev->dev, buffer_info->dma,
3821 buffer_info->length, DMA_FROM_DEVICE);
3822 buffer_info->dma = 0;
3824 length = le16_to_cpu(rx_desc->length);
3826 /* errors is only valid for DD + EOP descriptors */
3827 if (unlikely((status & E1000_RXD_STAT_EOP) &&
3828 (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK))) {
3829 u8 last_byte = *(skb->data + length - 1);
3830 if (TBI_ACCEPT(hw, status, rx_desc->errors, length,
3831 last_byte)) {
3832 spin_lock_irqsave(&adapter->stats_lock,
3833 irq_flags);
3834 e1000_tbi_adjust_stats(hw, &adapter->stats,
3835 length, skb->data);
3836 spin_unlock_irqrestore(&adapter->stats_lock,
3837 irq_flags);
3838 length--;
3839 } else {
3840 /* recycle both page and skb */
3841 buffer_info->skb = skb;
3842 /* an error means any chain goes out the window
3843 * too */
3844 if (rx_ring->rx_skb_top)
3845 dev_kfree_skb(rx_ring->rx_skb_top);
3846 rx_ring->rx_skb_top = NULL;
3847 goto next_desc;
3851 #define rxtop rx_ring->rx_skb_top
3852 if (!(status & E1000_RXD_STAT_EOP)) {
3853 /* this descriptor is only the beginning (or middle) */
3854 if (!rxtop) {
3855 /* this is the beginning of a chain */
3856 rxtop = skb;
3857 skb_fill_page_desc(rxtop, 0, buffer_info->page,
3858 0, length);
3859 } else {
3860 /* this is the middle of a chain */
3861 skb_fill_page_desc(rxtop,
3862 skb_shinfo(rxtop)->nr_frags,
3863 buffer_info->page, 0, length);
3864 /* re-use the skb, only consumed the page */
3865 buffer_info->skb = skb;
3867 e1000_consume_page(buffer_info, rxtop, length);
3868 goto next_desc;
3869 } else {
3870 if (rxtop) {
3871 /* end of the chain */
3872 skb_fill_page_desc(rxtop,
3873 skb_shinfo(rxtop)->nr_frags,
3874 buffer_info->page, 0, length);
3875 /* re-use the current skb, we only consumed the
3876 * page */
3877 buffer_info->skb = skb;
3878 skb = rxtop;
3879 rxtop = NULL;
3880 e1000_consume_page(buffer_info, skb, length);
3881 } else {
3882 /* no chain, got EOP, this buf is the packet
3883 * copybreak to save the put_page/alloc_page */
3884 if (length <= copybreak &&
3885 skb_tailroom(skb) >= length) {
3886 u8 *vaddr;
3887 vaddr = kmap_atomic(buffer_info->page,
3888 KM_SKB_DATA_SOFTIRQ);
3889 memcpy(skb_tail_pointer(skb), vaddr, length);
3890 kunmap_atomic(vaddr,
3891 KM_SKB_DATA_SOFTIRQ);
3892 /* re-use the page, so don't erase
3893 * buffer_info->page */
3894 skb_put(skb, length);
3895 } else {
3896 skb_fill_page_desc(skb, 0,
3897 buffer_info->page, 0,
3898 length);
3899 e1000_consume_page(buffer_info, skb,
3900 length);
3905 /* Receive Checksum Offload XXX recompute due to CRC strip? */
3906 e1000_rx_checksum(adapter,
3907 (u32)(status) |
3908 ((u32)(rx_desc->errors) << 24),
3909 le16_to_cpu(rx_desc->csum), skb);
3911 pskb_trim(skb, skb->len - 4);
3913 /* probably a little skewed due to removing CRC */
3914 total_rx_bytes += skb->len;
3915 total_rx_packets++;
3917 /* eth type trans needs skb->data to point to something */
3918 if (!pskb_may_pull(skb, ETH_HLEN)) {
3919 e_err(drv, "pskb_may_pull failed.\n");
3920 dev_kfree_skb(skb);
3921 goto next_desc;
3924 e1000_receive_skb(adapter, status, rx_desc->special, skb);
3926 next_desc:
3927 rx_desc->status = 0;
3929 /* return some buffers to hardware, one at a time is too slow */
3930 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
3931 adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
3932 cleaned_count = 0;
3935 /* use prefetched values */
3936 rx_desc = next_rxd;
3937 buffer_info = next_buffer;
3939 rx_ring->next_to_clean = i;
3941 cleaned_count = E1000_DESC_UNUSED(rx_ring);
3942 if (cleaned_count)
3943 adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
3945 adapter->total_rx_packets += total_rx_packets;
3946 adapter->total_rx_bytes += total_rx_bytes;
3947 netdev->stats.rx_bytes += total_rx_bytes;
3948 netdev->stats.rx_packets += total_rx_packets;
3949 return cleaned;
3953 * this should improve performance for small packets with large amounts
3954 * of reassembly being done in the stack
3956 static void e1000_check_copybreak(struct net_device *netdev,
3957 struct e1000_buffer *buffer_info,
3958 u32 length, struct sk_buff **skb)
3960 struct sk_buff *new_skb;
3962 if (length > copybreak)
3963 return;
3965 new_skb = netdev_alloc_skb_ip_align(netdev, length);
3966 if (!new_skb)
3967 return;
3969 skb_copy_to_linear_data_offset(new_skb, -NET_IP_ALIGN,
3970 (*skb)->data - NET_IP_ALIGN,
3971 length + NET_IP_ALIGN);
3972 /* save the skb in buffer_info as good */
3973 buffer_info->skb = *skb;
3974 *skb = new_skb;
3978 * e1000_clean_rx_irq - Send received data up the network stack; legacy
3979 * @adapter: board private structure
3980 * @rx_ring: ring to clean
3981 * @work_done: amount of napi work completed this call
3982 * @work_to_do: max amount of work allowed for this call to do
3984 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
3985 struct e1000_rx_ring *rx_ring,
3986 int *work_done, int work_to_do)
3988 struct e1000_hw *hw = &adapter->hw;
3989 struct net_device *netdev = adapter->netdev;
3990 struct pci_dev *pdev = adapter->pdev;
3991 struct e1000_rx_desc *rx_desc, *next_rxd;
3992 struct e1000_buffer *buffer_info, *next_buffer;
3993 unsigned long flags;
3994 u32 length;
3995 unsigned int i;
3996 int cleaned_count = 0;
3997 bool cleaned = false;
3998 unsigned int total_rx_bytes=0, total_rx_packets=0;
4000 i = rx_ring->next_to_clean;
4001 rx_desc = E1000_RX_DESC(*rx_ring, i);
4002 buffer_info = &rx_ring->buffer_info[i];
4004 while (rx_desc->status & E1000_RXD_STAT_DD) {
4005 struct sk_buff *skb;
4006 u8 status;
4008 if (*work_done >= work_to_do)
4009 break;
4010 (*work_done)++;
4011 rmb(); /* read descriptor and rx_buffer_info after status DD */
4013 status = rx_desc->status;
4014 skb = buffer_info->skb;
4015 buffer_info->skb = NULL;
4017 prefetch(skb->data - NET_IP_ALIGN);
4019 if (++i == rx_ring->count) i = 0;
4020 next_rxd = E1000_RX_DESC(*rx_ring, i);
4021 prefetch(next_rxd);
4023 next_buffer = &rx_ring->buffer_info[i];
4025 cleaned = true;
4026 cleaned_count++;
4027 dma_unmap_single(&pdev->dev, buffer_info->dma,
4028 buffer_info->length, DMA_FROM_DEVICE);
4029 buffer_info->dma = 0;
4031 length = le16_to_cpu(rx_desc->length);
4032 /* !EOP means multiple descriptors were used to store a single
4033 * packet, if thats the case we need to toss it. In fact, we
4034 * to toss every packet with the EOP bit clear and the next
4035 * frame that _does_ have the EOP bit set, as it is by
4036 * definition only a frame fragment
4038 if (unlikely(!(status & E1000_RXD_STAT_EOP)))
4039 adapter->discarding = true;
4041 if (adapter->discarding) {
4042 /* All receives must fit into a single buffer */
4043 e_dbg("Receive packet consumed multiple buffers\n");
4044 /* recycle */
4045 buffer_info->skb = skb;
4046 if (status & E1000_RXD_STAT_EOP)
4047 adapter->discarding = false;
4048 goto next_desc;
4051 if (unlikely(rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK)) {
4052 u8 last_byte = *(skb->data + length - 1);
4053 if (TBI_ACCEPT(hw, status, rx_desc->errors, length,
4054 last_byte)) {
4055 spin_lock_irqsave(&adapter->stats_lock, flags);
4056 e1000_tbi_adjust_stats(hw, &adapter->stats,
4057 length, skb->data);
4058 spin_unlock_irqrestore(&adapter->stats_lock,
4059 flags);
4060 length--;
4061 } else {
4062 /* recycle */
4063 buffer_info->skb = skb;
4064 goto next_desc;
4068 /* adjust length to remove Ethernet CRC, this must be
4069 * done after the TBI_ACCEPT workaround above */
4070 length -= 4;
4072 /* probably a little skewed due to removing CRC */
4073 total_rx_bytes += length;
4074 total_rx_packets++;
4076 e1000_check_copybreak(netdev, buffer_info, length, &skb);
4078 skb_put(skb, length);
4080 /* Receive Checksum Offload */
4081 e1000_rx_checksum(adapter,
4082 (u32)(status) |
4083 ((u32)(rx_desc->errors) << 24),
4084 le16_to_cpu(rx_desc->csum), skb);
4086 e1000_receive_skb(adapter, status, rx_desc->special, skb);
4088 next_desc:
4089 rx_desc->status = 0;
4091 /* return some buffers to hardware, one at a time is too slow */
4092 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
4093 adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
4094 cleaned_count = 0;
4097 /* use prefetched values */
4098 rx_desc = next_rxd;
4099 buffer_info = next_buffer;
4101 rx_ring->next_to_clean = i;
4103 cleaned_count = E1000_DESC_UNUSED(rx_ring);
4104 if (cleaned_count)
4105 adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
4107 adapter->total_rx_packets += total_rx_packets;
4108 adapter->total_rx_bytes += total_rx_bytes;
4109 netdev->stats.rx_bytes += total_rx_bytes;
4110 netdev->stats.rx_packets += total_rx_packets;
4111 return cleaned;
4115 * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
4116 * @adapter: address of board private structure
4117 * @rx_ring: pointer to receive ring structure
4118 * @cleaned_count: number of buffers to allocate this pass
4121 static void
4122 e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
4123 struct e1000_rx_ring *rx_ring, int cleaned_count)
4125 struct net_device *netdev = adapter->netdev;
4126 struct pci_dev *pdev = adapter->pdev;
4127 struct e1000_rx_desc *rx_desc;
4128 struct e1000_buffer *buffer_info;
4129 struct sk_buff *skb;
4130 unsigned int i;
4131 unsigned int bufsz = 256 - 16 /*for skb_reserve */ ;
4133 i = rx_ring->next_to_use;
4134 buffer_info = &rx_ring->buffer_info[i];
4136 while (cleaned_count--) {
4137 skb = buffer_info->skb;
4138 if (skb) {
4139 skb_trim(skb, 0);
4140 goto check_page;
4143 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
4144 if (unlikely(!skb)) {
4145 /* Better luck next round */
4146 adapter->alloc_rx_buff_failed++;
4147 break;
4150 /* Fix for errata 23, can't cross 64kB boundary */
4151 if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
4152 struct sk_buff *oldskb = skb;
4153 e_err(rx_err, "skb align check failed: %u bytes at "
4154 "%p\n", bufsz, skb->data);
4155 /* Try again, without freeing the previous */
4156 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
4157 /* Failed allocation, critical failure */
4158 if (!skb) {
4159 dev_kfree_skb(oldskb);
4160 adapter->alloc_rx_buff_failed++;
4161 break;
4164 if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
4165 /* give up */
4166 dev_kfree_skb(skb);
4167 dev_kfree_skb(oldskb);
4168 break; /* while (cleaned_count--) */
4171 /* Use new allocation */
4172 dev_kfree_skb(oldskb);
4174 buffer_info->skb = skb;
4175 buffer_info->length = adapter->rx_buffer_len;
4176 check_page:
4177 /* allocate a new page if necessary */
4178 if (!buffer_info->page) {
4179 buffer_info->page = alloc_page(GFP_ATOMIC);
4180 if (unlikely(!buffer_info->page)) {
4181 adapter->alloc_rx_buff_failed++;
4182 break;
4186 if (!buffer_info->dma) {
4187 buffer_info->dma = dma_map_page(&pdev->dev,
4188 buffer_info->page, 0,
4189 buffer_info->length,
4190 DMA_FROM_DEVICE);
4191 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
4192 put_page(buffer_info->page);
4193 dev_kfree_skb(skb);
4194 buffer_info->page = NULL;
4195 buffer_info->skb = NULL;
4196 buffer_info->dma = 0;
4197 adapter->alloc_rx_buff_failed++;
4198 break; /* while !buffer_info->skb */
4202 rx_desc = E1000_RX_DESC(*rx_ring, i);
4203 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4205 if (unlikely(++i == rx_ring->count))
4206 i = 0;
4207 buffer_info = &rx_ring->buffer_info[i];
4210 if (likely(rx_ring->next_to_use != i)) {
4211 rx_ring->next_to_use = i;
4212 if (unlikely(i-- == 0))
4213 i = (rx_ring->count - 1);
4215 /* Force memory writes to complete before letting h/w
4216 * know there are new descriptors to fetch. (Only
4217 * applicable for weak-ordered memory model archs,
4218 * such as IA-64). */
4219 wmb();
4220 writel(i, adapter->hw.hw_addr + rx_ring->rdt);
4225 * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
4226 * @adapter: address of board private structure
4229 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
4230 struct e1000_rx_ring *rx_ring,
4231 int cleaned_count)
4233 struct e1000_hw *hw = &adapter->hw;
4234 struct net_device *netdev = adapter->netdev;
4235 struct pci_dev *pdev = adapter->pdev;
4236 struct e1000_rx_desc *rx_desc;
4237 struct e1000_buffer *buffer_info;
4238 struct sk_buff *skb;
4239 unsigned int i;
4240 unsigned int bufsz = adapter->rx_buffer_len;
4242 i = rx_ring->next_to_use;
4243 buffer_info = &rx_ring->buffer_info[i];
4245 while (cleaned_count--) {
4246 skb = buffer_info->skb;
4247 if (skb) {
4248 skb_trim(skb, 0);
4249 goto map_skb;
4252 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
4253 if (unlikely(!skb)) {
4254 /* Better luck next round */
4255 adapter->alloc_rx_buff_failed++;
4256 break;
4259 /* Fix for errata 23, can't cross 64kB boundary */
4260 if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
4261 struct sk_buff *oldskb = skb;
4262 e_err(rx_err, "skb align check failed: %u bytes at "
4263 "%p\n", bufsz, skb->data);
4264 /* Try again, without freeing the previous */
4265 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
4266 /* Failed allocation, critical failure */
4267 if (!skb) {
4268 dev_kfree_skb(oldskb);
4269 adapter->alloc_rx_buff_failed++;
4270 break;
4273 if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
4274 /* give up */
4275 dev_kfree_skb(skb);
4276 dev_kfree_skb(oldskb);
4277 adapter->alloc_rx_buff_failed++;
4278 break; /* while !buffer_info->skb */
4281 /* Use new allocation */
4282 dev_kfree_skb(oldskb);
4284 buffer_info->skb = skb;
4285 buffer_info->length = adapter->rx_buffer_len;
4286 map_skb:
4287 buffer_info->dma = dma_map_single(&pdev->dev,
4288 skb->data,
4289 buffer_info->length,
4290 DMA_FROM_DEVICE);
4291 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
4292 dev_kfree_skb(skb);
4293 buffer_info->skb = NULL;
4294 buffer_info->dma = 0;
4295 adapter->alloc_rx_buff_failed++;
4296 break; /* while !buffer_info->skb */
4300 * XXX if it was allocated cleanly it will never map to a
4301 * boundary crossing
4304 /* Fix for errata 23, can't cross 64kB boundary */
4305 if (!e1000_check_64k_bound(adapter,
4306 (void *)(unsigned long)buffer_info->dma,
4307 adapter->rx_buffer_len)) {
4308 e_err(rx_err, "dma align check failed: %u bytes at "
4309 "%p\n", adapter->rx_buffer_len,
4310 (void *)(unsigned long)buffer_info->dma);
4311 dev_kfree_skb(skb);
4312 buffer_info->skb = NULL;
4314 dma_unmap_single(&pdev->dev, buffer_info->dma,
4315 adapter->rx_buffer_len,
4316 DMA_FROM_DEVICE);
4317 buffer_info->dma = 0;
4319 adapter->alloc_rx_buff_failed++;
4320 break; /* while !buffer_info->skb */
4322 rx_desc = E1000_RX_DESC(*rx_ring, i);
4323 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4325 if (unlikely(++i == rx_ring->count))
4326 i = 0;
4327 buffer_info = &rx_ring->buffer_info[i];
4330 if (likely(rx_ring->next_to_use != i)) {
4331 rx_ring->next_to_use = i;
4332 if (unlikely(i-- == 0))
4333 i = (rx_ring->count - 1);
4335 /* Force memory writes to complete before letting h/w
4336 * know there are new descriptors to fetch. (Only
4337 * applicable for weak-ordered memory model archs,
4338 * such as IA-64). */
4339 wmb();
4340 writel(i, hw->hw_addr + rx_ring->rdt);
4345 * e1000_smartspeed - Workaround for SmartSpeed on 82541 and 82547 controllers.
4346 * @adapter:
4349 static void e1000_smartspeed(struct e1000_adapter *adapter)
4351 struct e1000_hw *hw = &adapter->hw;
4352 u16 phy_status;
4353 u16 phy_ctrl;
4355 if ((hw->phy_type != e1000_phy_igp) || !hw->autoneg ||
4356 !(hw->autoneg_advertised & ADVERTISE_1000_FULL))
4357 return;
4359 if (adapter->smartspeed == 0) {
4360 /* If Master/Slave config fault is asserted twice,
4361 * we assume back-to-back */
4362 e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_status);
4363 if (!(phy_status & SR_1000T_MS_CONFIG_FAULT)) return;
4364 e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_status);
4365 if (!(phy_status & SR_1000T_MS_CONFIG_FAULT)) return;
4366 e1000_read_phy_reg(hw, PHY_1000T_CTRL, &phy_ctrl);
4367 if (phy_ctrl & CR_1000T_MS_ENABLE) {
4368 phy_ctrl &= ~CR_1000T_MS_ENABLE;
4369 e1000_write_phy_reg(hw, PHY_1000T_CTRL,
4370 phy_ctrl);
4371 adapter->smartspeed++;
4372 if (!e1000_phy_setup_autoneg(hw) &&
4373 !e1000_read_phy_reg(hw, PHY_CTRL,
4374 &phy_ctrl)) {
4375 phy_ctrl |= (MII_CR_AUTO_NEG_EN |
4376 MII_CR_RESTART_AUTO_NEG);
4377 e1000_write_phy_reg(hw, PHY_CTRL,
4378 phy_ctrl);
4381 return;
4382 } else if (adapter->smartspeed == E1000_SMARTSPEED_DOWNSHIFT) {
4383 /* If still no link, perhaps using 2/3 pair cable */
4384 e1000_read_phy_reg(hw, PHY_1000T_CTRL, &phy_ctrl);
4385 phy_ctrl |= CR_1000T_MS_ENABLE;
4386 e1000_write_phy_reg(hw, PHY_1000T_CTRL, phy_ctrl);
4387 if (!e1000_phy_setup_autoneg(hw) &&
4388 !e1000_read_phy_reg(hw, PHY_CTRL, &phy_ctrl)) {
4389 phy_ctrl |= (MII_CR_AUTO_NEG_EN |
4390 MII_CR_RESTART_AUTO_NEG);
4391 e1000_write_phy_reg(hw, PHY_CTRL, phy_ctrl);
4394 /* Restart process after E1000_SMARTSPEED_MAX iterations */
4395 if (adapter->smartspeed++ == E1000_SMARTSPEED_MAX)
4396 adapter->smartspeed = 0;
4400 * e1000_ioctl -
4401 * @netdev:
4402 * @ifreq:
4403 * @cmd:
4406 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4408 switch (cmd) {
4409 case SIOCGMIIPHY:
4410 case SIOCGMIIREG:
4411 case SIOCSMIIREG:
4412 return e1000_mii_ioctl(netdev, ifr, cmd);
4413 default:
4414 return -EOPNOTSUPP;
4419 * e1000_mii_ioctl -
4420 * @netdev:
4421 * @ifreq:
4422 * @cmd:
4425 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
4426 int cmd)
4428 struct e1000_adapter *adapter = netdev_priv(netdev);
4429 struct e1000_hw *hw = &adapter->hw;
4430 struct mii_ioctl_data *data = if_mii(ifr);
4431 int retval;
4432 u16 mii_reg;
4433 unsigned long flags;
4435 if (hw->media_type != e1000_media_type_copper)
4436 return -EOPNOTSUPP;
4438 switch (cmd) {
4439 case SIOCGMIIPHY:
4440 data->phy_id = hw->phy_addr;
4441 break;
4442 case SIOCGMIIREG:
4443 spin_lock_irqsave(&adapter->stats_lock, flags);
4444 if (e1000_read_phy_reg(hw, data->reg_num & 0x1F,
4445 &data->val_out)) {
4446 spin_unlock_irqrestore(&adapter->stats_lock, flags);
4447 return -EIO;
4449 spin_unlock_irqrestore(&adapter->stats_lock, flags);
4450 break;
4451 case SIOCSMIIREG:
4452 if (data->reg_num & ~(0x1F))
4453 return -EFAULT;
4454 mii_reg = data->val_in;
4455 spin_lock_irqsave(&adapter->stats_lock, flags);
4456 if (e1000_write_phy_reg(hw, data->reg_num,
4457 mii_reg)) {
4458 spin_unlock_irqrestore(&adapter->stats_lock, flags);
4459 return -EIO;
4461 spin_unlock_irqrestore(&adapter->stats_lock, flags);
4462 if (hw->media_type == e1000_media_type_copper) {
4463 switch (data->reg_num) {
4464 case PHY_CTRL:
4465 if (mii_reg & MII_CR_POWER_DOWN)
4466 break;
4467 if (mii_reg & MII_CR_AUTO_NEG_EN) {
4468 hw->autoneg = 1;
4469 hw->autoneg_advertised = 0x2F;
4470 } else {
4471 u32 speed;
4472 if (mii_reg & 0x40)
4473 speed = SPEED_1000;
4474 else if (mii_reg & 0x2000)
4475 speed = SPEED_100;
4476 else
4477 speed = SPEED_10;
4478 retval = e1000_set_spd_dplx(
4479 adapter, speed,
4480 ((mii_reg & 0x100)
4481 ? DUPLEX_FULL :
4482 DUPLEX_HALF));
4483 if (retval)
4484 return retval;
4486 if (netif_running(adapter->netdev))
4487 e1000_reinit_locked(adapter);
4488 else
4489 e1000_reset(adapter);
4490 break;
4491 case M88E1000_PHY_SPEC_CTRL:
4492 case M88E1000_EXT_PHY_SPEC_CTRL:
4493 if (e1000_phy_reset(hw))
4494 return -EIO;
4495 break;
4497 } else {
4498 switch (data->reg_num) {
4499 case PHY_CTRL:
4500 if (mii_reg & MII_CR_POWER_DOWN)
4501 break;
4502 if (netif_running(adapter->netdev))
4503 e1000_reinit_locked(adapter);
4504 else
4505 e1000_reset(adapter);
4506 break;
4509 break;
4510 default:
4511 return -EOPNOTSUPP;
4513 return E1000_SUCCESS;
4516 void e1000_pci_set_mwi(struct e1000_hw *hw)
4518 struct e1000_adapter *adapter = hw->back;
4519 int ret_val = pci_set_mwi(adapter->pdev);
4521 if (ret_val)
4522 e_err(probe, "Error in setting MWI\n");
4525 void e1000_pci_clear_mwi(struct e1000_hw *hw)
4527 struct e1000_adapter *adapter = hw->back;
4529 pci_clear_mwi(adapter->pdev);
4532 int e1000_pcix_get_mmrbc(struct e1000_hw *hw)
4534 struct e1000_adapter *adapter = hw->back;
4535 return pcix_get_mmrbc(adapter->pdev);
4538 void e1000_pcix_set_mmrbc(struct e1000_hw *hw, int mmrbc)
4540 struct e1000_adapter *adapter = hw->back;
4541 pcix_set_mmrbc(adapter->pdev, mmrbc);
4544 void e1000_io_write(struct e1000_hw *hw, unsigned long port, u32 value)
4546 outl(value, port);
4549 static bool e1000_vlan_used(struct e1000_adapter *adapter)
4551 u16 vid;
4553 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
4554 return true;
4555 return false;
4558 static void __e1000_vlan_mode(struct e1000_adapter *adapter,
4559 netdev_features_t features)
4561 struct e1000_hw *hw = &adapter->hw;
4562 u32 ctrl;
4564 ctrl = er32(CTRL);
4565 if (features & NETIF_F_HW_VLAN_RX) {
4566 /* enable VLAN tag insert/strip */
4567 ctrl |= E1000_CTRL_VME;
4568 } else {
4569 /* disable VLAN tag insert/strip */
4570 ctrl &= ~E1000_CTRL_VME;
4572 ew32(CTRL, ctrl);
4574 static void e1000_vlan_filter_on_off(struct e1000_adapter *adapter,
4575 bool filter_on)
4577 struct e1000_hw *hw = &adapter->hw;
4578 u32 rctl;
4580 if (!test_bit(__E1000_DOWN, &adapter->flags))
4581 e1000_irq_disable(adapter);
4583 __e1000_vlan_mode(adapter, adapter->netdev->features);
4584 if (filter_on) {
4585 /* enable VLAN receive filtering */
4586 rctl = er32(RCTL);
4587 rctl &= ~E1000_RCTL_CFIEN;
4588 if (!(adapter->netdev->flags & IFF_PROMISC))
4589 rctl |= E1000_RCTL_VFE;
4590 ew32(RCTL, rctl);
4591 e1000_update_mng_vlan(adapter);
4592 } else {
4593 /* disable VLAN receive filtering */
4594 rctl = er32(RCTL);
4595 rctl &= ~E1000_RCTL_VFE;
4596 ew32(RCTL, rctl);
4599 if (!test_bit(__E1000_DOWN, &adapter->flags))
4600 e1000_irq_enable(adapter);
4603 static void e1000_vlan_mode(struct net_device *netdev,
4604 netdev_features_t features)
4606 struct e1000_adapter *adapter = netdev_priv(netdev);
4608 if (!test_bit(__E1000_DOWN, &adapter->flags))
4609 e1000_irq_disable(adapter);
4611 __e1000_vlan_mode(adapter, features);
4613 if (!test_bit(__E1000_DOWN, &adapter->flags))
4614 e1000_irq_enable(adapter);
4617 static int e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
4619 struct e1000_adapter *adapter = netdev_priv(netdev);
4620 struct e1000_hw *hw = &adapter->hw;
4621 u32 vfta, index;
4623 if ((hw->mng_cookie.status &
4624 E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) &&
4625 (vid == adapter->mng_vlan_id))
4626 return 0;
4628 if (!e1000_vlan_used(adapter))
4629 e1000_vlan_filter_on_off(adapter, true);
4631 /* add VID to filter table */
4632 index = (vid >> 5) & 0x7F;
4633 vfta = E1000_READ_REG_ARRAY(hw, VFTA, index);
4634 vfta |= (1 << (vid & 0x1F));
4635 e1000_write_vfta(hw, index, vfta);
4637 set_bit(vid, adapter->active_vlans);
4639 return 0;
4642 static int e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
4644 struct e1000_adapter *adapter = netdev_priv(netdev);
4645 struct e1000_hw *hw = &adapter->hw;
4646 u32 vfta, index;
4648 if (!test_bit(__E1000_DOWN, &adapter->flags))
4649 e1000_irq_disable(adapter);
4650 if (!test_bit(__E1000_DOWN, &adapter->flags))
4651 e1000_irq_enable(adapter);
4653 /* remove VID from filter table */
4654 index = (vid >> 5) & 0x7F;
4655 vfta = E1000_READ_REG_ARRAY(hw, VFTA, index);
4656 vfta &= ~(1 << (vid & 0x1F));
4657 e1000_write_vfta(hw, index, vfta);
4659 clear_bit(vid, adapter->active_vlans);
4661 if (!e1000_vlan_used(adapter))
4662 e1000_vlan_filter_on_off(adapter, false);
4664 return 0;
4667 static void e1000_restore_vlan(struct e1000_adapter *adapter)
4669 u16 vid;
4671 if (!e1000_vlan_used(adapter))
4672 return;
4674 e1000_vlan_filter_on_off(adapter, true);
4675 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
4676 e1000_vlan_rx_add_vid(adapter->netdev, vid);
4679 int e1000_set_spd_dplx(struct e1000_adapter *adapter, u32 spd, u8 dplx)
4681 struct e1000_hw *hw = &adapter->hw;
4683 hw->autoneg = 0;
4685 /* Make sure dplx is at most 1 bit and lsb of speed is not set
4686 * for the switch() below to work */
4687 if ((spd & 1) || (dplx & ~1))
4688 goto err_inval;
4690 /* Fiber NICs only allow 1000 gbps Full duplex */
4691 if ((hw->media_type == e1000_media_type_fiber) &&
4692 spd != SPEED_1000 &&
4693 dplx != DUPLEX_FULL)
4694 goto err_inval;
4696 switch (spd + dplx) {
4697 case SPEED_10 + DUPLEX_HALF:
4698 hw->forced_speed_duplex = e1000_10_half;
4699 break;
4700 case SPEED_10 + DUPLEX_FULL:
4701 hw->forced_speed_duplex = e1000_10_full;
4702 break;
4703 case SPEED_100 + DUPLEX_HALF:
4704 hw->forced_speed_duplex = e1000_100_half;
4705 break;
4706 case SPEED_100 + DUPLEX_FULL:
4707 hw->forced_speed_duplex = e1000_100_full;
4708 break;
4709 case SPEED_1000 + DUPLEX_FULL:
4710 hw->autoneg = 1;
4711 hw->autoneg_advertised = ADVERTISE_1000_FULL;
4712 break;
4713 case SPEED_1000 + DUPLEX_HALF: /* not supported */
4714 default:
4715 goto err_inval;
4717 return 0;
4719 err_inval:
4720 e_err(probe, "Unsupported Speed/Duplex configuration\n");
4721 return -EINVAL;
4724 static int __e1000_shutdown(struct pci_dev *pdev, bool *enable_wake)
4726 struct net_device *netdev = pci_get_drvdata(pdev);
4727 struct e1000_adapter *adapter = netdev_priv(netdev);
4728 struct e1000_hw *hw = &adapter->hw;
4729 u32 ctrl, ctrl_ext, rctl, status;
4730 u32 wufc = adapter->wol;
4731 #ifdef CONFIG_PM
4732 int retval = 0;
4733 #endif
4735 netif_device_detach(netdev);
4737 if (netif_running(netdev)) {
4738 WARN_ON(test_bit(__E1000_RESETTING, &adapter->flags));
4739 e1000_down(adapter);
4742 #ifdef CONFIG_PM
4743 retval = pci_save_state(pdev);
4744 if (retval)
4745 return retval;
4746 #endif
4748 status = er32(STATUS);
4749 if (status & E1000_STATUS_LU)
4750 wufc &= ~E1000_WUFC_LNKC;
4752 if (wufc) {
4753 e1000_setup_rctl(adapter);
4754 e1000_set_rx_mode(netdev);
4756 rctl = er32(RCTL);
4758 /* turn on all-multi mode if wake on multicast is enabled */
4759 if (wufc & E1000_WUFC_MC)
4760 rctl |= E1000_RCTL_MPE;
4762 /* enable receives in the hardware */
4763 ew32(RCTL, rctl | E1000_RCTL_EN);
4765 if (hw->mac_type >= e1000_82540) {
4766 ctrl = er32(CTRL);
4767 /* advertise wake from D3Cold */
4768 #define E1000_CTRL_ADVD3WUC 0x00100000
4769 /* phy power management enable */
4770 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
4771 ctrl |= E1000_CTRL_ADVD3WUC |
4772 E1000_CTRL_EN_PHY_PWR_MGMT;
4773 ew32(CTRL, ctrl);
4776 if (hw->media_type == e1000_media_type_fiber ||
4777 hw->media_type == e1000_media_type_internal_serdes) {
4778 /* keep the laser running in D3 */
4779 ctrl_ext = er32(CTRL_EXT);
4780 ctrl_ext |= E1000_CTRL_EXT_SDP7_DATA;
4781 ew32(CTRL_EXT, ctrl_ext);
4784 ew32(WUC, E1000_WUC_PME_EN);
4785 ew32(WUFC, wufc);
4786 } else {
4787 ew32(WUC, 0);
4788 ew32(WUFC, 0);
4791 e1000_release_manageability(adapter);
4793 *enable_wake = !!wufc;
4795 /* make sure adapter isn't asleep if manageability is enabled */
4796 if (adapter->en_mng_pt)
4797 *enable_wake = true;
4799 if (netif_running(netdev))
4800 e1000_free_irq(adapter);
4802 pci_disable_device(pdev);
4804 return 0;
4807 #ifdef CONFIG_PM
4808 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state)
4810 int retval;
4811 bool wake;
4813 retval = __e1000_shutdown(pdev, &wake);
4814 if (retval)
4815 return retval;
4817 if (wake) {
4818 pci_prepare_to_sleep(pdev);
4819 } else {
4820 pci_wake_from_d3(pdev, false);
4821 pci_set_power_state(pdev, PCI_D3hot);
4824 return 0;
4827 static int e1000_resume(struct pci_dev *pdev)
4829 struct net_device *netdev = pci_get_drvdata(pdev);
4830 struct e1000_adapter *adapter = netdev_priv(netdev);
4831 struct e1000_hw *hw = &adapter->hw;
4832 u32 err;
4834 pci_set_power_state(pdev, PCI_D0);
4835 pci_restore_state(pdev);
4836 pci_save_state(pdev);
4838 if (adapter->need_ioport)
4839 err = pci_enable_device(pdev);
4840 else
4841 err = pci_enable_device_mem(pdev);
4842 if (err) {
4843 pr_err("Cannot enable PCI device from suspend\n");
4844 return err;
4846 pci_set_master(pdev);
4848 pci_enable_wake(pdev, PCI_D3hot, 0);
4849 pci_enable_wake(pdev, PCI_D3cold, 0);
4851 if (netif_running(netdev)) {
4852 err = e1000_request_irq(adapter);
4853 if (err)
4854 return err;
4857 e1000_power_up_phy(adapter);
4858 e1000_reset(adapter);
4859 ew32(WUS, ~0);
4861 e1000_init_manageability(adapter);
4863 if (netif_running(netdev))
4864 e1000_up(adapter);
4866 netif_device_attach(netdev);
4868 return 0;
4870 #endif
4872 static void e1000_shutdown(struct pci_dev *pdev)
4874 bool wake;
4876 __e1000_shutdown(pdev, &wake);
4878 if (system_state == SYSTEM_POWER_OFF) {
4879 pci_wake_from_d3(pdev, wake);
4880 pci_set_power_state(pdev, PCI_D3hot);
4884 #ifdef CONFIG_NET_POLL_CONTROLLER
4886 * Polling 'interrupt' - used by things like netconsole to send skbs
4887 * without having to re-enable interrupts. It's not called while
4888 * the interrupt routine is executing.
4890 static void e1000_netpoll(struct net_device *netdev)
4892 struct e1000_adapter *adapter = netdev_priv(netdev);
4894 disable_irq(adapter->pdev->irq);
4895 e1000_intr(adapter->pdev->irq, netdev);
4896 enable_irq(adapter->pdev->irq);
4898 #endif
4901 * e1000_io_error_detected - called when PCI error is detected
4902 * @pdev: Pointer to PCI device
4903 * @state: The current pci connection state
4905 * This function is called after a PCI bus error affecting
4906 * this device has been detected.
4908 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
4909 pci_channel_state_t state)
4911 struct net_device *netdev = pci_get_drvdata(pdev);
4912 struct e1000_adapter *adapter = netdev_priv(netdev);
4914 netif_device_detach(netdev);
4916 if (state == pci_channel_io_perm_failure)
4917 return PCI_ERS_RESULT_DISCONNECT;
4919 if (netif_running(netdev))
4920 e1000_down(adapter);
4921 pci_disable_device(pdev);
4923 /* Request a slot slot reset. */
4924 return PCI_ERS_RESULT_NEED_RESET;
4928 * e1000_io_slot_reset - called after the pci bus has been reset.
4929 * @pdev: Pointer to PCI device
4931 * Restart the card from scratch, as if from a cold-boot. Implementation
4932 * resembles the first-half of the e1000_resume routine.
4934 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
4936 struct net_device *netdev = pci_get_drvdata(pdev);
4937 struct e1000_adapter *adapter = netdev_priv(netdev);
4938 struct e1000_hw *hw = &adapter->hw;
4939 int err;
4941 if (adapter->need_ioport)
4942 err = pci_enable_device(pdev);
4943 else
4944 err = pci_enable_device_mem(pdev);
4945 if (err) {
4946 pr_err("Cannot re-enable PCI device after reset.\n");
4947 return PCI_ERS_RESULT_DISCONNECT;
4949 pci_set_master(pdev);
4951 pci_enable_wake(pdev, PCI_D3hot, 0);
4952 pci_enable_wake(pdev, PCI_D3cold, 0);
4954 e1000_reset(adapter);
4955 ew32(WUS, ~0);
4957 return PCI_ERS_RESULT_RECOVERED;
4961 * e1000_io_resume - called when traffic can start flowing again.
4962 * @pdev: Pointer to PCI device
4964 * This callback is called when the error recovery driver tells us that
4965 * its OK to resume normal operation. Implementation resembles the
4966 * second-half of the e1000_resume routine.
4968 static void e1000_io_resume(struct pci_dev *pdev)
4970 struct net_device *netdev = pci_get_drvdata(pdev);
4971 struct e1000_adapter *adapter = netdev_priv(netdev);
4973 e1000_init_manageability(adapter);
4975 if (netif_running(netdev)) {
4976 if (e1000_up(adapter)) {
4977 pr_info("can't bring device back up after reset\n");
4978 return;
4982 netif_device_attach(netdev);
4985 /* e1000_main.c */