Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dtor/input
[linux-btrfs-devel.git] / drivers / net / ethernet / intel / e1000 / e1000_main.c
blobcf480b55462273d51d4272d61af4e5521d53a379
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 /* Intel Media SOC GbE MDIO physical base address */
37 static unsigned long ce4100_gbe_mdio_base_phy;
38 /* Intel Media SOC GbE MDIO virtual base address */
39 void __iomem *ce4100_gbe_mdio_base_virt;
41 char e1000_driver_name[] = "e1000";
42 static char e1000_driver_string[] = "Intel(R) PRO/1000 Network Driver";
43 #define DRV_VERSION "7.3.21-k8-NAPI"
44 const char e1000_driver_version[] = DRV_VERSION;
45 static const char e1000_copyright[] = "Copyright (c) 1999-2006 Intel Corporation.";
47 /* e1000_pci_tbl - PCI Device ID Table
49 * Last entry must be all 0s
51 * Macro expands to...
52 * {PCI_DEVICE(PCI_VENDOR_ID_INTEL, device_id)}
54 static DEFINE_PCI_DEVICE_TABLE(e1000_pci_tbl) = {
55 INTEL_E1000_ETHERNET_DEVICE(0x1000),
56 INTEL_E1000_ETHERNET_DEVICE(0x1001),
57 INTEL_E1000_ETHERNET_DEVICE(0x1004),
58 INTEL_E1000_ETHERNET_DEVICE(0x1008),
59 INTEL_E1000_ETHERNET_DEVICE(0x1009),
60 INTEL_E1000_ETHERNET_DEVICE(0x100C),
61 INTEL_E1000_ETHERNET_DEVICE(0x100D),
62 INTEL_E1000_ETHERNET_DEVICE(0x100E),
63 INTEL_E1000_ETHERNET_DEVICE(0x100F),
64 INTEL_E1000_ETHERNET_DEVICE(0x1010),
65 INTEL_E1000_ETHERNET_DEVICE(0x1011),
66 INTEL_E1000_ETHERNET_DEVICE(0x1012),
67 INTEL_E1000_ETHERNET_DEVICE(0x1013),
68 INTEL_E1000_ETHERNET_DEVICE(0x1014),
69 INTEL_E1000_ETHERNET_DEVICE(0x1015),
70 INTEL_E1000_ETHERNET_DEVICE(0x1016),
71 INTEL_E1000_ETHERNET_DEVICE(0x1017),
72 INTEL_E1000_ETHERNET_DEVICE(0x1018),
73 INTEL_E1000_ETHERNET_DEVICE(0x1019),
74 INTEL_E1000_ETHERNET_DEVICE(0x101A),
75 INTEL_E1000_ETHERNET_DEVICE(0x101D),
76 INTEL_E1000_ETHERNET_DEVICE(0x101E),
77 INTEL_E1000_ETHERNET_DEVICE(0x1026),
78 INTEL_E1000_ETHERNET_DEVICE(0x1027),
79 INTEL_E1000_ETHERNET_DEVICE(0x1028),
80 INTEL_E1000_ETHERNET_DEVICE(0x1075),
81 INTEL_E1000_ETHERNET_DEVICE(0x1076),
82 INTEL_E1000_ETHERNET_DEVICE(0x1077),
83 INTEL_E1000_ETHERNET_DEVICE(0x1078),
84 INTEL_E1000_ETHERNET_DEVICE(0x1079),
85 INTEL_E1000_ETHERNET_DEVICE(0x107A),
86 INTEL_E1000_ETHERNET_DEVICE(0x107B),
87 INTEL_E1000_ETHERNET_DEVICE(0x107C),
88 INTEL_E1000_ETHERNET_DEVICE(0x108A),
89 INTEL_E1000_ETHERNET_DEVICE(0x1099),
90 INTEL_E1000_ETHERNET_DEVICE(0x10B5),
91 INTEL_E1000_ETHERNET_DEVICE(0x2E6E),
92 /* required last entry */
93 {0,}
96 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
98 int e1000_up(struct e1000_adapter *adapter);
99 void e1000_down(struct e1000_adapter *adapter);
100 void e1000_reinit_locked(struct e1000_adapter *adapter);
101 void e1000_reset(struct e1000_adapter *adapter);
102 int e1000_setup_all_tx_resources(struct e1000_adapter *adapter);
103 int e1000_setup_all_rx_resources(struct e1000_adapter *adapter);
104 void e1000_free_all_tx_resources(struct e1000_adapter *adapter);
105 void e1000_free_all_rx_resources(struct e1000_adapter *adapter);
106 static int e1000_setup_tx_resources(struct e1000_adapter *adapter,
107 struct e1000_tx_ring *txdr);
108 static int e1000_setup_rx_resources(struct e1000_adapter *adapter,
109 struct e1000_rx_ring *rxdr);
110 static void e1000_free_tx_resources(struct e1000_adapter *adapter,
111 struct e1000_tx_ring *tx_ring);
112 static void e1000_free_rx_resources(struct e1000_adapter *adapter,
113 struct e1000_rx_ring *rx_ring);
114 void e1000_update_stats(struct e1000_adapter *adapter);
116 static int e1000_init_module(void);
117 static void e1000_exit_module(void);
118 static int e1000_probe(struct pci_dev *pdev, const struct pci_device_id *ent);
119 static void __devexit e1000_remove(struct pci_dev *pdev);
120 static int e1000_alloc_queues(struct e1000_adapter *adapter);
121 static int e1000_sw_init(struct e1000_adapter *adapter);
122 static int e1000_open(struct net_device *netdev);
123 static int e1000_close(struct net_device *netdev);
124 static void e1000_configure_tx(struct e1000_adapter *adapter);
125 static void e1000_configure_rx(struct e1000_adapter *adapter);
126 static void e1000_setup_rctl(struct e1000_adapter *adapter);
127 static void e1000_clean_all_tx_rings(struct e1000_adapter *adapter);
128 static void e1000_clean_all_rx_rings(struct e1000_adapter *adapter);
129 static void e1000_clean_tx_ring(struct e1000_adapter *adapter,
130 struct e1000_tx_ring *tx_ring);
131 static void e1000_clean_rx_ring(struct e1000_adapter *adapter,
132 struct e1000_rx_ring *rx_ring);
133 static void e1000_set_rx_mode(struct net_device *netdev);
134 static void e1000_update_phy_info_task(struct work_struct *work);
135 static void e1000_watchdog(struct work_struct *work);
136 static void e1000_82547_tx_fifo_stall_task(struct work_struct *work);
137 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
138 struct net_device *netdev);
139 static struct net_device_stats * e1000_get_stats(struct net_device *netdev);
140 static int e1000_change_mtu(struct net_device *netdev, int new_mtu);
141 static int e1000_set_mac(struct net_device *netdev, void *p);
142 static irqreturn_t e1000_intr(int irq, void *data);
143 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter,
144 struct e1000_tx_ring *tx_ring);
145 static int e1000_clean(struct napi_struct *napi, int budget);
146 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
147 struct e1000_rx_ring *rx_ring,
148 int *work_done, int work_to_do);
149 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
150 struct e1000_rx_ring *rx_ring,
151 int *work_done, int work_to_do);
152 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
153 struct e1000_rx_ring *rx_ring,
154 int cleaned_count);
155 static void e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
156 struct e1000_rx_ring *rx_ring,
157 int cleaned_count);
158 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd);
159 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
160 int cmd);
161 static void e1000_enter_82542_rst(struct e1000_adapter *adapter);
162 static void e1000_leave_82542_rst(struct e1000_adapter *adapter);
163 static void e1000_tx_timeout(struct net_device *dev);
164 static void e1000_reset_task(struct work_struct *work);
165 static void e1000_smartspeed(struct e1000_adapter *adapter);
166 static int e1000_82547_fifo_workaround(struct e1000_adapter *adapter,
167 struct sk_buff *skb);
169 static bool e1000_vlan_used(struct e1000_adapter *adapter);
170 static void e1000_vlan_mode(struct net_device *netdev, u32 features);
171 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid);
172 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid);
173 static void e1000_restore_vlan(struct e1000_adapter *adapter);
175 #ifdef CONFIG_PM
176 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state);
177 static int e1000_resume(struct pci_dev *pdev);
178 #endif
179 static void e1000_shutdown(struct pci_dev *pdev);
181 #ifdef CONFIG_NET_POLL_CONTROLLER
182 /* for netdump / net console */
183 static void e1000_netpoll (struct net_device *netdev);
184 #endif
186 #define COPYBREAK_DEFAULT 256
187 static unsigned int copybreak __read_mostly = COPYBREAK_DEFAULT;
188 module_param(copybreak, uint, 0644);
189 MODULE_PARM_DESC(copybreak,
190 "Maximum size of packet that is copied to a new buffer on receive");
192 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
193 pci_channel_state_t state);
194 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev);
195 static void e1000_io_resume(struct pci_dev *pdev);
197 static struct pci_error_handlers e1000_err_handler = {
198 .error_detected = e1000_io_error_detected,
199 .slot_reset = e1000_io_slot_reset,
200 .resume = e1000_io_resume,
203 static struct pci_driver e1000_driver = {
204 .name = e1000_driver_name,
205 .id_table = e1000_pci_tbl,
206 .probe = e1000_probe,
207 .remove = __devexit_p(e1000_remove),
208 #ifdef CONFIG_PM
209 /* Power Management Hooks */
210 .suspend = e1000_suspend,
211 .resume = e1000_resume,
212 #endif
213 .shutdown = e1000_shutdown,
214 .err_handler = &e1000_err_handler
217 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
218 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
219 MODULE_LICENSE("GPL");
220 MODULE_VERSION(DRV_VERSION);
222 static int debug = NETIF_MSG_DRV | NETIF_MSG_PROBE;
223 module_param(debug, int, 0);
224 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
227 * e1000_get_hw_dev - return device
228 * used by hardware layer to print debugging information
231 struct net_device *e1000_get_hw_dev(struct e1000_hw *hw)
233 struct e1000_adapter *adapter = hw->back;
234 return adapter->netdev;
238 * e1000_init_module - Driver Registration Routine
240 * e1000_init_module is the first routine called when the driver is
241 * loaded. All it does is register with the PCI subsystem.
244 static int __init e1000_init_module(void)
246 int ret;
247 pr_info("%s - version %s\n", e1000_driver_string, e1000_driver_version);
249 pr_info("%s\n", e1000_copyright);
251 ret = pci_register_driver(&e1000_driver);
252 if (copybreak != COPYBREAK_DEFAULT) {
253 if (copybreak == 0)
254 pr_info("copybreak disabled\n");
255 else
256 pr_info("copybreak enabled for "
257 "packets <= %u bytes\n", copybreak);
259 return ret;
262 module_init(e1000_init_module);
265 * e1000_exit_module - Driver Exit Cleanup Routine
267 * e1000_exit_module is called just before the driver is removed
268 * from memory.
271 static void __exit e1000_exit_module(void)
273 pci_unregister_driver(&e1000_driver);
276 module_exit(e1000_exit_module);
278 static int e1000_request_irq(struct e1000_adapter *adapter)
280 struct net_device *netdev = adapter->netdev;
281 irq_handler_t handler = e1000_intr;
282 int irq_flags = IRQF_SHARED;
283 int err;
285 err = request_irq(adapter->pdev->irq, handler, irq_flags, netdev->name,
286 netdev);
287 if (err) {
288 e_err(probe, "Unable to allocate interrupt Error: %d\n", err);
291 return err;
294 static void e1000_free_irq(struct e1000_adapter *adapter)
296 struct net_device *netdev = adapter->netdev;
298 free_irq(adapter->pdev->irq, netdev);
302 * e1000_irq_disable - Mask off interrupt generation on the NIC
303 * @adapter: board private structure
306 static void e1000_irq_disable(struct e1000_adapter *adapter)
308 struct e1000_hw *hw = &adapter->hw;
310 ew32(IMC, ~0);
311 E1000_WRITE_FLUSH();
312 synchronize_irq(adapter->pdev->irq);
316 * e1000_irq_enable - Enable default interrupt generation settings
317 * @adapter: board private structure
320 static void e1000_irq_enable(struct e1000_adapter *adapter)
322 struct e1000_hw *hw = &adapter->hw;
324 ew32(IMS, IMS_ENABLE_MASK);
325 E1000_WRITE_FLUSH();
328 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
330 struct e1000_hw *hw = &adapter->hw;
331 struct net_device *netdev = adapter->netdev;
332 u16 vid = hw->mng_cookie.vlan_id;
333 u16 old_vid = adapter->mng_vlan_id;
335 if (!e1000_vlan_used(adapter))
336 return;
338 if (!test_bit(vid, adapter->active_vlans)) {
339 if (hw->mng_cookie.status &
340 E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) {
341 e1000_vlan_rx_add_vid(netdev, vid);
342 adapter->mng_vlan_id = vid;
343 } else {
344 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
346 if ((old_vid != (u16)E1000_MNG_VLAN_NONE) &&
347 (vid != old_vid) &&
348 !test_bit(old_vid, adapter->active_vlans))
349 e1000_vlan_rx_kill_vid(netdev, old_vid);
350 } else {
351 adapter->mng_vlan_id = vid;
355 static void e1000_init_manageability(struct e1000_adapter *adapter)
357 struct e1000_hw *hw = &adapter->hw;
359 if (adapter->en_mng_pt) {
360 u32 manc = er32(MANC);
362 /* disable hardware interception of ARP */
363 manc &= ~(E1000_MANC_ARP_EN);
365 ew32(MANC, manc);
369 static void e1000_release_manageability(struct e1000_adapter *adapter)
371 struct e1000_hw *hw = &adapter->hw;
373 if (adapter->en_mng_pt) {
374 u32 manc = er32(MANC);
376 /* re-enable hardware interception of ARP */
377 manc |= E1000_MANC_ARP_EN;
379 ew32(MANC, manc);
384 * e1000_configure - configure the hardware for RX and TX
385 * @adapter = private board structure
387 static void e1000_configure(struct e1000_adapter *adapter)
389 struct net_device *netdev = adapter->netdev;
390 int i;
392 e1000_set_rx_mode(netdev);
394 e1000_restore_vlan(adapter);
395 e1000_init_manageability(adapter);
397 e1000_configure_tx(adapter);
398 e1000_setup_rctl(adapter);
399 e1000_configure_rx(adapter);
400 /* call E1000_DESC_UNUSED which always leaves
401 * at least 1 descriptor unused to make sure
402 * next_to_use != next_to_clean */
403 for (i = 0; i < adapter->num_rx_queues; i++) {
404 struct e1000_rx_ring *ring = &adapter->rx_ring[i];
405 adapter->alloc_rx_buf(adapter, ring,
406 E1000_DESC_UNUSED(ring));
410 int e1000_up(struct e1000_adapter *adapter)
412 struct e1000_hw *hw = &adapter->hw;
414 /* hardware has been reset, we need to reload some things */
415 e1000_configure(adapter);
417 clear_bit(__E1000_DOWN, &adapter->flags);
419 napi_enable(&adapter->napi);
421 e1000_irq_enable(adapter);
423 netif_wake_queue(adapter->netdev);
425 /* fire a link change interrupt to start the watchdog */
426 ew32(ICS, E1000_ICS_LSC);
427 return 0;
431 * e1000_power_up_phy - restore link in case the phy was powered down
432 * @adapter: address of board private structure
434 * The phy may be powered down to save power and turn off link when the
435 * driver is unloaded and wake on lan is not enabled (among others)
436 * *** this routine MUST be followed by a call to e1000_reset ***
440 void e1000_power_up_phy(struct e1000_adapter *adapter)
442 struct e1000_hw *hw = &adapter->hw;
443 u16 mii_reg = 0;
445 /* Just clear the power down bit to wake the phy back up */
446 if (hw->media_type == e1000_media_type_copper) {
447 /* according to the manual, the phy will retain its
448 * settings across a power-down/up cycle */
449 e1000_read_phy_reg(hw, PHY_CTRL, &mii_reg);
450 mii_reg &= ~MII_CR_POWER_DOWN;
451 e1000_write_phy_reg(hw, PHY_CTRL, mii_reg);
455 static void e1000_power_down_phy(struct e1000_adapter *adapter)
457 struct e1000_hw *hw = &adapter->hw;
459 /* Power down the PHY so no link is implied when interface is down *
460 * The PHY cannot be powered down if any of the following is true *
461 * (a) WoL is enabled
462 * (b) AMT is active
463 * (c) SoL/IDER session is active */
464 if (!adapter->wol && hw->mac_type >= e1000_82540 &&
465 hw->media_type == e1000_media_type_copper) {
466 u16 mii_reg = 0;
468 switch (hw->mac_type) {
469 case e1000_82540:
470 case e1000_82545:
471 case e1000_82545_rev_3:
472 case e1000_82546:
473 case e1000_ce4100:
474 case e1000_82546_rev_3:
475 case e1000_82541:
476 case e1000_82541_rev_2:
477 case e1000_82547:
478 case e1000_82547_rev_2:
479 if (er32(MANC) & E1000_MANC_SMBUS_EN)
480 goto out;
481 break;
482 default:
483 goto out;
485 e1000_read_phy_reg(hw, PHY_CTRL, &mii_reg);
486 mii_reg |= MII_CR_POWER_DOWN;
487 e1000_write_phy_reg(hw, PHY_CTRL, mii_reg);
488 msleep(1);
490 out:
491 return;
494 static void e1000_down_and_stop(struct e1000_adapter *adapter)
496 set_bit(__E1000_DOWN, &adapter->flags);
497 cancel_work_sync(&adapter->reset_task);
498 cancel_delayed_work_sync(&adapter->watchdog_task);
499 cancel_delayed_work_sync(&adapter->phy_info_task);
500 cancel_delayed_work_sync(&adapter->fifo_stall_task);
503 void e1000_down(struct e1000_adapter *adapter)
505 struct e1000_hw *hw = &adapter->hw;
506 struct net_device *netdev = adapter->netdev;
507 u32 rctl, tctl;
510 /* disable receives in the hardware */
511 rctl = er32(RCTL);
512 ew32(RCTL, rctl & ~E1000_RCTL_EN);
513 /* flush and sleep below */
515 netif_tx_disable(netdev);
517 /* disable transmits in the hardware */
518 tctl = er32(TCTL);
519 tctl &= ~E1000_TCTL_EN;
520 ew32(TCTL, tctl);
521 /* flush both disables and wait for them to finish */
522 E1000_WRITE_FLUSH();
523 msleep(10);
525 napi_disable(&adapter->napi);
527 e1000_irq_disable(adapter);
530 * Setting DOWN must be after irq_disable to prevent
531 * a screaming interrupt. Setting DOWN also prevents
532 * tasks from rescheduling.
534 e1000_down_and_stop(adapter);
536 adapter->link_speed = 0;
537 adapter->link_duplex = 0;
538 netif_carrier_off(netdev);
540 e1000_reset(adapter);
541 e1000_clean_all_tx_rings(adapter);
542 e1000_clean_all_rx_rings(adapter);
545 static void e1000_reinit_safe(struct e1000_adapter *adapter)
547 while (test_and_set_bit(__E1000_RESETTING, &adapter->flags))
548 msleep(1);
549 mutex_lock(&adapter->mutex);
550 e1000_down(adapter);
551 e1000_up(adapter);
552 mutex_unlock(&adapter->mutex);
553 clear_bit(__E1000_RESETTING, &adapter->flags);
556 void e1000_reinit_locked(struct e1000_adapter *adapter)
558 /* if rtnl_lock is not held the call path is bogus */
559 ASSERT_RTNL();
560 WARN_ON(in_interrupt());
561 while (test_and_set_bit(__E1000_RESETTING, &adapter->flags))
562 msleep(1);
563 e1000_down(adapter);
564 e1000_up(adapter);
565 clear_bit(__E1000_RESETTING, &adapter->flags);
568 void e1000_reset(struct e1000_adapter *adapter)
570 struct e1000_hw *hw = &adapter->hw;
571 u32 pba = 0, tx_space, min_tx_space, min_rx_space;
572 bool legacy_pba_adjust = false;
573 u16 hwm;
575 /* Repartition Pba for greater than 9k mtu
576 * To take effect CTRL.RST is required.
579 switch (hw->mac_type) {
580 case e1000_82542_rev2_0:
581 case e1000_82542_rev2_1:
582 case e1000_82543:
583 case e1000_82544:
584 case e1000_82540:
585 case e1000_82541:
586 case e1000_82541_rev_2:
587 legacy_pba_adjust = true;
588 pba = E1000_PBA_48K;
589 break;
590 case e1000_82545:
591 case e1000_82545_rev_3:
592 case e1000_82546:
593 case e1000_ce4100:
594 case e1000_82546_rev_3:
595 pba = E1000_PBA_48K;
596 break;
597 case e1000_82547:
598 case e1000_82547_rev_2:
599 legacy_pba_adjust = true;
600 pba = E1000_PBA_30K;
601 break;
602 case e1000_undefined:
603 case e1000_num_macs:
604 break;
607 if (legacy_pba_adjust) {
608 if (hw->max_frame_size > E1000_RXBUFFER_8192)
609 pba -= 8; /* allocate more FIFO for Tx */
611 if (hw->mac_type == e1000_82547) {
612 adapter->tx_fifo_head = 0;
613 adapter->tx_head_addr = pba << E1000_TX_HEAD_ADDR_SHIFT;
614 adapter->tx_fifo_size =
615 (E1000_PBA_40K - pba) << E1000_PBA_BYTES_SHIFT;
616 atomic_set(&adapter->tx_fifo_stall, 0);
618 } else if (hw->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
619 /* adjust PBA for jumbo frames */
620 ew32(PBA, pba);
622 /* To maintain wire speed transmits, the Tx FIFO should be
623 * large enough to accommodate two full transmit packets,
624 * rounded up to the next 1KB and expressed in KB. Likewise,
625 * the Rx FIFO should be large enough to accommodate at least
626 * one full receive packet and is similarly rounded up and
627 * expressed in KB. */
628 pba = er32(PBA);
629 /* upper 16 bits has Tx packet buffer allocation size in KB */
630 tx_space = pba >> 16;
631 /* lower 16 bits has Rx packet buffer allocation size in KB */
632 pba &= 0xffff;
634 * the tx fifo also stores 16 bytes of information about the tx
635 * but don't include ethernet FCS because hardware appends it
637 min_tx_space = (hw->max_frame_size +
638 sizeof(struct e1000_tx_desc) -
639 ETH_FCS_LEN) * 2;
640 min_tx_space = ALIGN(min_tx_space, 1024);
641 min_tx_space >>= 10;
642 /* software strips receive CRC, so leave room for it */
643 min_rx_space = hw->max_frame_size;
644 min_rx_space = ALIGN(min_rx_space, 1024);
645 min_rx_space >>= 10;
647 /* If current Tx allocation is less than the min Tx FIFO size,
648 * and the min Tx FIFO size is less than the current Rx FIFO
649 * allocation, take space away from current Rx allocation */
650 if (tx_space < min_tx_space &&
651 ((min_tx_space - tx_space) < pba)) {
652 pba = pba - (min_tx_space - tx_space);
654 /* PCI/PCIx hardware has PBA alignment constraints */
655 switch (hw->mac_type) {
656 case e1000_82545 ... e1000_82546_rev_3:
657 pba &= ~(E1000_PBA_8K - 1);
658 break;
659 default:
660 break;
663 /* if short on rx space, rx wins and must trump tx
664 * adjustment or use Early Receive if available */
665 if (pba < min_rx_space)
666 pba = min_rx_space;
670 ew32(PBA, pba);
673 * flow control settings:
674 * The high water mark must be low enough to fit one full frame
675 * (or the size used for early receive) above it in the Rx FIFO.
676 * Set it to the lower of:
677 * - 90% of the Rx FIFO size, and
678 * - the full Rx FIFO size minus the early receive size (for parts
679 * with ERT support assuming ERT set to E1000_ERT_2048), or
680 * - the full Rx FIFO size minus one full frame
682 hwm = min(((pba << 10) * 9 / 10),
683 ((pba << 10) - hw->max_frame_size));
685 hw->fc_high_water = hwm & 0xFFF8; /* 8-byte granularity */
686 hw->fc_low_water = hw->fc_high_water - 8;
687 hw->fc_pause_time = E1000_FC_PAUSE_TIME;
688 hw->fc_send_xon = 1;
689 hw->fc = hw->original_fc;
691 /* Allow time for pending master requests to run */
692 e1000_reset_hw(hw);
693 if (hw->mac_type >= e1000_82544)
694 ew32(WUC, 0);
696 if (e1000_init_hw(hw))
697 e_dev_err("Hardware Error\n");
698 e1000_update_mng_vlan(adapter);
700 /* if (adapter->hwflags & HWFLAGS_PHY_PWR_BIT) { */
701 if (hw->mac_type >= e1000_82544 &&
702 hw->autoneg == 1 &&
703 hw->autoneg_advertised == ADVERTISE_1000_FULL) {
704 u32 ctrl = er32(CTRL);
705 /* clear phy power management bit if we are in gig only mode,
706 * which if enabled will attempt negotiation to 100Mb, which
707 * can cause a loss of link at power off or driver unload */
708 ctrl &= ~E1000_CTRL_SWDPIN3;
709 ew32(CTRL, ctrl);
712 /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
713 ew32(VET, ETHERNET_IEEE_VLAN_TYPE);
715 e1000_reset_adaptive(hw);
716 e1000_phy_get_info(hw, &adapter->phy_info);
718 e1000_release_manageability(adapter);
722 * Dump the eeprom for users having checksum issues
724 static void e1000_dump_eeprom(struct e1000_adapter *adapter)
726 struct net_device *netdev = adapter->netdev;
727 struct ethtool_eeprom eeprom;
728 const struct ethtool_ops *ops = netdev->ethtool_ops;
729 u8 *data;
730 int i;
731 u16 csum_old, csum_new = 0;
733 eeprom.len = ops->get_eeprom_len(netdev);
734 eeprom.offset = 0;
736 data = kmalloc(eeprom.len, GFP_KERNEL);
737 if (!data) {
738 pr_err("Unable to allocate memory to dump EEPROM data\n");
739 return;
742 ops->get_eeprom(netdev, &eeprom, data);
744 csum_old = (data[EEPROM_CHECKSUM_REG * 2]) +
745 (data[EEPROM_CHECKSUM_REG * 2 + 1] << 8);
746 for (i = 0; i < EEPROM_CHECKSUM_REG * 2; i += 2)
747 csum_new += data[i] + (data[i + 1] << 8);
748 csum_new = EEPROM_SUM - csum_new;
750 pr_err("/*********************/\n");
751 pr_err("Current EEPROM Checksum : 0x%04x\n", csum_old);
752 pr_err("Calculated : 0x%04x\n", csum_new);
754 pr_err("Offset Values\n");
755 pr_err("======== ======\n");
756 print_hex_dump(KERN_ERR, "", DUMP_PREFIX_OFFSET, 16, 1, data, 128, 0);
758 pr_err("Include this output when contacting your support provider.\n");
759 pr_err("This is not a software error! Something bad happened to\n");
760 pr_err("your hardware or EEPROM image. Ignoring this problem could\n");
761 pr_err("result in further problems, possibly loss of data,\n");
762 pr_err("corruption or system hangs!\n");
763 pr_err("The MAC Address will be reset to 00:00:00:00:00:00,\n");
764 pr_err("which is invalid and requires you to set the proper MAC\n");
765 pr_err("address manually before continuing to enable this network\n");
766 pr_err("device. Please inspect the EEPROM dump and report the\n");
767 pr_err("issue to your hardware vendor or Intel Customer Support.\n");
768 pr_err("/*********************/\n");
770 kfree(data);
774 * e1000_is_need_ioport - determine if an adapter needs ioport resources or not
775 * @pdev: PCI device information struct
777 * Return true if an adapter needs ioport resources
779 static int e1000_is_need_ioport(struct pci_dev *pdev)
781 switch (pdev->device) {
782 case E1000_DEV_ID_82540EM:
783 case E1000_DEV_ID_82540EM_LOM:
784 case E1000_DEV_ID_82540EP:
785 case E1000_DEV_ID_82540EP_LOM:
786 case E1000_DEV_ID_82540EP_LP:
787 case E1000_DEV_ID_82541EI:
788 case E1000_DEV_ID_82541EI_MOBILE:
789 case E1000_DEV_ID_82541ER:
790 case E1000_DEV_ID_82541ER_LOM:
791 case E1000_DEV_ID_82541GI:
792 case E1000_DEV_ID_82541GI_LF:
793 case E1000_DEV_ID_82541GI_MOBILE:
794 case E1000_DEV_ID_82544EI_COPPER:
795 case E1000_DEV_ID_82544EI_FIBER:
796 case E1000_DEV_ID_82544GC_COPPER:
797 case E1000_DEV_ID_82544GC_LOM:
798 case E1000_DEV_ID_82545EM_COPPER:
799 case E1000_DEV_ID_82545EM_FIBER:
800 case E1000_DEV_ID_82546EB_COPPER:
801 case E1000_DEV_ID_82546EB_FIBER:
802 case E1000_DEV_ID_82546EB_QUAD_COPPER:
803 return true;
804 default:
805 return false;
809 static u32 e1000_fix_features(struct net_device *netdev, u32 features)
812 * Since there is no support for separate rx/tx vlan accel
813 * enable/disable make sure tx flag is always in same state as rx.
815 if (features & NETIF_F_HW_VLAN_RX)
816 features |= NETIF_F_HW_VLAN_TX;
817 else
818 features &= ~NETIF_F_HW_VLAN_TX;
820 return features;
823 static int e1000_set_features(struct net_device *netdev, u32 features)
825 struct e1000_adapter *adapter = netdev_priv(netdev);
826 u32 changed = features ^ netdev->features;
828 if (changed & NETIF_F_HW_VLAN_RX)
829 e1000_vlan_mode(netdev, features);
831 if (!(changed & NETIF_F_RXCSUM))
832 return 0;
834 adapter->rx_csum = !!(features & NETIF_F_RXCSUM);
836 if (netif_running(netdev))
837 e1000_reinit_locked(adapter);
838 else
839 e1000_reset(adapter);
841 return 0;
844 static const struct net_device_ops e1000_netdev_ops = {
845 .ndo_open = e1000_open,
846 .ndo_stop = e1000_close,
847 .ndo_start_xmit = e1000_xmit_frame,
848 .ndo_get_stats = e1000_get_stats,
849 .ndo_set_rx_mode = e1000_set_rx_mode,
850 .ndo_set_mac_address = e1000_set_mac,
851 .ndo_tx_timeout = e1000_tx_timeout,
852 .ndo_change_mtu = e1000_change_mtu,
853 .ndo_do_ioctl = e1000_ioctl,
854 .ndo_validate_addr = eth_validate_addr,
855 .ndo_vlan_rx_add_vid = e1000_vlan_rx_add_vid,
856 .ndo_vlan_rx_kill_vid = e1000_vlan_rx_kill_vid,
857 #ifdef CONFIG_NET_POLL_CONTROLLER
858 .ndo_poll_controller = e1000_netpoll,
859 #endif
860 .ndo_fix_features = e1000_fix_features,
861 .ndo_set_features = e1000_set_features,
865 * e1000_init_hw_struct - initialize members of hw struct
866 * @adapter: board private struct
867 * @hw: structure used by e1000_hw.c
869 * Factors out initialization of the e1000_hw struct to its own function
870 * that can be called very early at init (just after struct allocation).
871 * Fields are initialized based on PCI device information and
872 * OS network device settings (MTU size).
873 * Returns negative error codes if MAC type setup fails.
875 static int e1000_init_hw_struct(struct e1000_adapter *adapter,
876 struct e1000_hw *hw)
878 struct pci_dev *pdev = adapter->pdev;
880 /* PCI config space info */
881 hw->vendor_id = pdev->vendor;
882 hw->device_id = pdev->device;
883 hw->subsystem_vendor_id = pdev->subsystem_vendor;
884 hw->subsystem_id = pdev->subsystem_device;
885 hw->revision_id = pdev->revision;
887 pci_read_config_word(pdev, PCI_COMMAND, &hw->pci_cmd_word);
889 hw->max_frame_size = adapter->netdev->mtu +
890 ENET_HEADER_SIZE + ETHERNET_FCS_SIZE;
891 hw->min_frame_size = MINIMUM_ETHERNET_FRAME_SIZE;
893 /* identify the MAC */
894 if (e1000_set_mac_type(hw)) {
895 e_err(probe, "Unknown MAC Type\n");
896 return -EIO;
899 switch (hw->mac_type) {
900 default:
901 break;
902 case e1000_82541:
903 case e1000_82547:
904 case e1000_82541_rev_2:
905 case e1000_82547_rev_2:
906 hw->phy_init_script = 1;
907 break;
910 e1000_set_media_type(hw);
911 e1000_get_bus_info(hw);
913 hw->wait_autoneg_complete = false;
914 hw->tbi_compatibility_en = true;
915 hw->adaptive_ifs = true;
917 /* Copper options */
919 if (hw->media_type == e1000_media_type_copper) {
920 hw->mdix = AUTO_ALL_MODES;
921 hw->disable_polarity_correction = false;
922 hw->master_slave = E1000_MASTER_SLAVE;
925 return 0;
929 * e1000_probe - Device Initialization Routine
930 * @pdev: PCI device information struct
931 * @ent: entry in e1000_pci_tbl
933 * Returns 0 on success, negative on failure
935 * e1000_probe initializes an adapter identified by a pci_dev structure.
936 * The OS initialization, configuring of the adapter private structure,
937 * and a hardware reset occur.
939 static int __devinit e1000_probe(struct pci_dev *pdev,
940 const struct pci_device_id *ent)
942 struct net_device *netdev;
943 struct e1000_adapter *adapter;
944 struct e1000_hw *hw;
946 static int cards_found = 0;
947 static int global_quad_port_a = 0; /* global ksp3 port a indication */
948 int i, err, pci_using_dac;
949 u16 eeprom_data = 0;
950 u16 tmp = 0;
951 u16 eeprom_apme_mask = E1000_EEPROM_APME;
952 int bars, need_ioport;
954 /* do not allocate ioport bars when not needed */
955 need_ioport = e1000_is_need_ioport(pdev);
956 if (need_ioport) {
957 bars = pci_select_bars(pdev, IORESOURCE_MEM | IORESOURCE_IO);
958 err = pci_enable_device(pdev);
959 } else {
960 bars = pci_select_bars(pdev, IORESOURCE_MEM);
961 err = pci_enable_device_mem(pdev);
963 if (err)
964 return err;
966 err = pci_request_selected_regions(pdev, bars, e1000_driver_name);
967 if (err)
968 goto err_pci_reg;
970 pci_set_master(pdev);
971 err = pci_save_state(pdev);
972 if (err)
973 goto err_alloc_etherdev;
975 err = -ENOMEM;
976 netdev = alloc_etherdev(sizeof(struct e1000_adapter));
977 if (!netdev)
978 goto err_alloc_etherdev;
980 SET_NETDEV_DEV(netdev, &pdev->dev);
982 pci_set_drvdata(pdev, netdev);
983 adapter = netdev_priv(netdev);
984 adapter->netdev = netdev;
985 adapter->pdev = pdev;
986 adapter->msg_enable = (1 << debug) - 1;
987 adapter->bars = bars;
988 adapter->need_ioport = need_ioport;
990 hw = &adapter->hw;
991 hw->back = adapter;
993 err = -EIO;
994 hw->hw_addr = pci_ioremap_bar(pdev, BAR_0);
995 if (!hw->hw_addr)
996 goto err_ioremap;
998 if (adapter->need_ioport) {
999 for (i = BAR_1; i <= BAR_5; i++) {
1000 if (pci_resource_len(pdev, i) == 0)
1001 continue;
1002 if (pci_resource_flags(pdev, i) & IORESOURCE_IO) {
1003 hw->io_base = pci_resource_start(pdev, i);
1004 break;
1009 /* make ready for any if (hw->...) below */
1010 err = e1000_init_hw_struct(adapter, hw);
1011 if (err)
1012 goto err_sw_init;
1015 * there is a workaround being applied below that limits
1016 * 64-bit DMA addresses to 64-bit hardware. There are some
1017 * 32-bit adapters that Tx hang when given 64-bit DMA addresses
1019 pci_using_dac = 0;
1020 if ((hw->bus_type == e1000_bus_type_pcix) &&
1021 !dma_set_mask(&pdev->dev, DMA_BIT_MASK(64))) {
1023 * according to DMA-API-HOWTO, coherent calls will always
1024 * succeed if the set call did
1026 dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64));
1027 pci_using_dac = 1;
1028 } else {
1029 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
1030 if (err) {
1031 pr_err("No usable DMA config, aborting\n");
1032 goto err_dma;
1034 dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(32));
1037 netdev->netdev_ops = &e1000_netdev_ops;
1038 e1000_set_ethtool_ops(netdev);
1039 netdev->watchdog_timeo = 5 * HZ;
1040 netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
1042 strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
1044 adapter->bd_number = cards_found;
1046 /* setup the private structure */
1048 err = e1000_sw_init(adapter);
1049 if (err)
1050 goto err_sw_init;
1052 err = -EIO;
1053 if (hw->mac_type == e1000_ce4100) {
1054 ce4100_gbe_mdio_base_phy = pci_resource_start(pdev, BAR_1);
1055 ce4100_gbe_mdio_base_virt = ioremap(ce4100_gbe_mdio_base_phy,
1056 pci_resource_len(pdev, BAR_1));
1058 if (!ce4100_gbe_mdio_base_virt)
1059 goto err_mdio_ioremap;
1062 if (hw->mac_type >= e1000_82543) {
1063 netdev->hw_features = NETIF_F_SG |
1064 NETIF_F_HW_CSUM |
1065 NETIF_F_HW_VLAN_RX;
1066 netdev->features = NETIF_F_HW_VLAN_TX |
1067 NETIF_F_HW_VLAN_FILTER;
1070 if ((hw->mac_type >= e1000_82544) &&
1071 (hw->mac_type != e1000_82547))
1072 netdev->hw_features |= NETIF_F_TSO;
1074 netdev->features |= netdev->hw_features;
1075 netdev->hw_features |= NETIF_F_RXCSUM;
1077 if (pci_using_dac) {
1078 netdev->features |= NETIF_F_HIGHDMA;
1079 netdev->vlan_features |= NETIF_F_HIGHDMA;
1082 netdev->vlan_features |= NETIF_F_TSO;
1083 netdev->vlan_features |= NETIF_F_HW_CSUM;
1084 netdev->vlan_features |= NETIF_F_SG;
1086 netdev->priv_flags |= IFF_UNICAST_FLT;
1088 adapter->en_mng_pt = e1000_enable_mng_pass_thru(hw);
1090 /* initialize eeprom parameters */
1091 if (e1000_init_eeprom_params(hw)) {
1092 e_err(probe, "EEPROM initialization failed\n");
1093 goto err_eeprom;
1096 /* before reading the EEPROM, reset the controller to
1097 * put the device in a known good starting state */
1099 e1000_reset_hw(hw);
1101 /* make sure the EEPROM is good */
1102 if (e1000_validate_eeprom_checksum(hw) < 0) {
1103 e_err(probe, "The EEPROM Checksum Is Not Valid\n");
1104 e1000_dump_eeprom(adapter);
1106 * set MAC address to all zeroes to invalidate and temporary
1107 * disable this device for the user. This blocks regular
1108 * traffic while still permitting ethtool ioctls from reaching
1109 * the hardware as well as allowing the user to run the
1110 * interface after manually setting a hw addr using
1111 * `ip set address`
1113 memset(hw->mac_addr, 0, netdev->addr_len);
1114 } else {
1115 /* copy the MAC address out of the EEPROM */
1116 if (e1000_read_mac_addr(hw))
1117 e_err(probe, "EEPROM Read Error\n");
1119 /* don't block initalization here due to bad MAC address */
1120 memcpy(netdev->dev_addr, hw->mac_addr, netdev->addr_len);
1121 memcpy(netdev->perm_addr, hw->mac_addr, netdev->addr_len);
1123 if (!is_valid_ether_addr(netdev->perm_addr))
1124 e_err(probe, "Invalid MAC Address\n");
1127 INIT_DELAYED_WORK(&adapter->watchdog_task, e1000_watchdog);
1128 INIT_DELAYED_WORK(&adapter->fifo_stall_task,
1129 e1000_82547_tx_fifo_stall_task);
1130 INIT_DELAYED_WORK(&adapter->phy_info_task, e1000_update_phy_info_task);
1131 INIT_WORK(&adapter->reset_task, e1000_reset_task);
1133 e1000_check_options(adapter);
1135 /* Initial Wake on LAN setting
1136 * If APM wake is enabled in the EEPROM,
1137 * enable the ACPI Magic Packet filter
1140 switch (hw->mac_type) {
1141 case e1000_82542_rev2_0:
1142 case e1000_82542_rev2_1:
1143 case e1000_82543:
1144 break;
1145 case e1000_82544:
1146 e1000_read_eeprom(hw,
1147 EEPROM_INIT_CONTROL2_REG, 1, &eeprom_data);
1148 eeprom_apme_mask = E1000_EEPROM_82544_APM;
1149 break;
1150 case e1000_82546:
1151 case e1000_82546_rev_3:
1152 if (er32(STATUS) & E1000_STATUS_FUNC_1){
1153 e1000_read_eeprom(hw,
1154 EEPROM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
1155 break;
1157 /* Fall Through */
1158 default:
1159 e1000_read_eeprom(hw,
1160 EEPROM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
1161 break;
1163 if (eeprom_data & eeprom_apme_mask)
1164 adapter->eeprom_wol |= E1000_WUFC_MAG;
1166 /* now that we have the eeprom settings, apply the special cases
1167 * where the eeprom may be wrong or the board simply won't support
1168 * wake on lan on a particular port */
1169 switch (pdev->device) {
1170 case E1000_DEV_ID_82546GB_PCIE:
1171 adapter->eeprom_wol = 0;
1172 break;
1173 case E1000_DEV_ID_82546EB_FIBER:
1174 case E1000_DEV_ID_82546GB_FIBER:
1175 /* Wake events only supported on port A for dual fiber
1176 * regardless of eeprom setting */
1177 if (er32(STATUS) & E1000_STATUS_FUNC_1)
1178 adapter->eeprom_wol = 0;
1179 break;
1180 case E1000_DEV_ID_82546GB_QUAD_COPPER_KSP3:
1181 /* if quad port adapter, disable WoL on all but port A */
1182 if (global_quad_port_a != 0)
1183 adapter->eeprom_wol = 0;
1184 else
1185 adapter->quad_port_a = 1;
1186 /* Reset for multiple quad port adapters */
1187 if (++global_quad_port_a == 4)
1188 global_quad_port_a = 0;
1189 break;
1192 /* initialize the wol settings based on the eeprom settings */
1193 adapter->wol = adapter->eeprom_wol;
1194 device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
1196 /* Auto detect PHY address */
1197 if (hw->mac_type == e1000_ce4100) {
1198 for (i = 0; i < 32; i++) {
1199 hw->phy_addr = i;
1200 e1000_read_phy_reg(hw, PHY_ID2, &tmp);
1201 if (tmp == 0 || tmp == 0xFF) {
1202 if (i == 31)
1203 goto err_eeprom;
1204 continue;
1205 } else
1206 break;
1210 /* reset the hardware with the new settings */
1211 e1000_reset(adapter);
1213 strcpy(netdev->name, "eth%d");
1214 err = register_netdev(netdev);
1215 if (err)
1216 goto err_register;
1218 e1000_vlan_mode(netdev, netdev->features);
1220 /* print bus type/speed/width info */
1221 e_info(probe, "(PCI%s:%dMHz:%d-bit) %pM\n",
1222 ((hw->bus_type == e1000_bus_type_pcix) ? "-X" : ""),
1223 ((hw->bus_speed == e1000_bus_speed_133) ? 133 :
1224 (hw->bus_speed == e1000_bus_speed_120) ? 120 :
1225 (hw->bus_speed == e1000_bus_speed_100) ? 100 :
1226 (hw->bus_speed == e1000_bus_speed_66) ? 66 : 33),
1227 ((hw->bus_width == e1000_bus_width_64) ? 64 : 32),
1228 netdev->dev_addr);
1230 /* carrier off reporting is important to ethtool even BEFORE open */
1231 netif_carrier_off(netdev);
1233 e_info(probe, "Intel(R) PRO/1000 Network Connection\n");
1235 cards_found++;
1236 return 0;
1238 err_register:
1239 err_eeprom:
1240 e1000_phy_hw_reset(hw);
1242 if (hw->flash_address)
1243 iounmap(hw->flash_address);
1244 kfree(adapter->tx_ring);
1245 kfree(adapter->rx_ring);
1246 err_dma:
1247 err_sw_init:
1248 err_mdio_ioremap:
1249 iounmap(ce4100_gbe_mdio_base_virt);
1250 iounmap(hw->hw_addr);
1251 err_ioremap:
1252 free_netdev(netdev);
1253 err_alloc_etherdev:
1254 pci_release_selected_regions(pdev, bars);
1255 err_pci_reg:
1256 pci_disable_device(pdev);
1257 return err;
1261 * e1000_remove - Device Removal Routine
1262 * @pdev: PCI device information struct
1264 * e1000_remove is called by the PCI subsystem to alert the driver
1265 * that it should release a PCI device. The could be caused by a
1266 * Hot-Plug event, or because the driver is going to be removed from
1267 * memory.
1270 static void __devexit e1000_remove(struct pci_dev *pdev)
1272 struct net_device *netdev = pci_get_drvdata(pdev);
1273 struct e1000_adapter *adapter = netdev_priv(netdev);
1274 struct e1000_hw *hw = &adapter->hw;
1276 e1000_down_and_stop(adapter);
1277 e1000_release_manageability(adapter);
1279 unregister_netdev(netdev);
1281 e1000_phy_hw_reset(hw);
1283 kfree(adapter->tx_ring);
1284 kfree(adapter->rx_ring);
1286 iounmap(hw->hw_addr);
1287 if (hw->flash_address)
1288 iounmap(hw->flash_address);
1289 pci_release_selected_regions(pdev, adapter->bars);
1291 free_netdev(netdev);
1293 pci_disable_device(pdev);
1297 * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
1298 * @adapter: board private structure to initialize
1300 * e1000_sw_init initializes the Adapter private data structure.
1301 * e1000_init_hw_struct MUST be called before this function
1304 static int __devinit e1000_sw_init(struct e1000_adapter *adapter)
1306 adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
1308 adapter->num_tx_queues = 1;
1309 adapter->num_rx_queues = 1;
1311 if (e1000_alloc_queues(adapter)) {
1312 e_err(probe, "Unable to allocate memory for queues\n");
1313 return -ENOMEM;
1316 /* Explicitly disable IRQ since the NIC can be in any state. */
1317 e1000_irq_disable(adapter);
1319 spin_lock_init(&adapter->stats_lock);
1320 mutex_init(&adapter->mutex);
1322 set_bit(__E1000_DOWN, &adapter->flags);
1324 return 0;
1328 * e1000_alloc_queues - Allocate memory for all rings
1329 * @adapter: board private structure to initialize
1331 * We allocate one ring per queue at run-time since we don't know the
1332 * number of queues at compile-time.
1335 static int __devinit e1000_alloc_queues(struct e1000_adapter *adapter)
1337 adapter->tx_ring = kcalloc(adapter->num_tx_queues,
1338 sizeof(struct e1000_tx_ring), GFP_KERNEL);
1339 if (!adapter->tx_ring)
1340 return -ENOMEM;
1342 adapter->rx_ring = kcalloc(adapter->num_rx_queues,
1343 sizeof(struct e1000_rx_ring), GFP_KERNEL);
1344 if (!adapter->rx_ring) {
1345 kfree(adapter->tx_ring);
1346 return -ENOMEM;
1349 return E1000_SUCCESS;
1353 * e1000_open - Called when a network interface is made active
1354 * @netdev: network interface device structure
1356 * Returns 0 on success, negative value on failure
1358 * The open entry point is called when a network interface is made
1359 * active by the system (IFF_UP). At this point all resources needed
1360 * for transmit and receive operations are allocated, the interrupt
1361 * handler is registered with the OS, the watchdog task is started,
1362 * and the stack is notified that the interface is ready.
1365 static int e1000_open(struct net_device *netdev)
1367 struct e1000_adapter *adapter = netdev_priv(netdev);
1368 struct e1000_hw *hw = &adapter->hw;
1369 int err;
1371 /* disallow open during test */
1372 if (test_bit(__E1000_TESTING, &adapter->flags))
1373 return -EBUSY;
1375 netif_carrier_off(netdev);
1377 /* allocate transmit descriptors */
1378 err = e1000_setup_all_tx_resources(adapter);
1379 if (err)
1380 goto err_setup_tx;
1382 /* allocate receive descriptors */
1383 err = e1000_setup_all_rx_resources(adapter);
1384 if (err)
1385 goto err_setup_rx;
1387 e1000_power_up_phy(adapter);
1389 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
1390 if ((hw->mng_cookie.status &
1391 E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT)) {
1392 e1000_update_mng_vlan(adapter);
1395 /* before we allocate an interrupt, we must be ready to handle it.
1396 * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1397 * as soon as we call pci_request_irq, so we have to setup our
1398 * clean_rx handler before we do so. */
1399 e1000_configure(adapter);
1401 err = e1000_request_irq(adapter);
1402 if (err)
1403 goto err_req_irq;
1405 /* From here on the code is the same as e1000_up() */
1406 clear_bit(__E1000_DOWN, &adapter->flags);
1408 napi_enable(&adapter->napi);
1410 e1000_irq_enable(adapter);
1412 netif_start_queue(netdev);
1414 /* fire a link status change interrupt to start the watchdog */
1415 ew32(ICS, E1000_ICS_LSC);
1417 return E1000_SUCCESS;
1419 err_req_irq:
1420 e1000_power_down_phy(adapter);
1421 e1000_free_all_rx_resources(adapter);
1422 err_setup_rx:
1423 e1000_free_all_tx_resources(adapter);
1424 err_setup_tx:
1425 e1000_reset(adapter);
1427 return err;
1431 * e1000_close - Disables a network interface
1432 * @netdev: network interface device structure
1434 * Returns 0, this is not allowed to fail
1436 * The close entry point is called when an interface is de-activated
1437 * by the OS. The hardware is still under the drivers control, but
1438 * needs to be disabled. A global MAC reset is issued to stop the
1439 * hardware, and all transmit and receive resources are freed.
1442 static int e1000_close(struct net_device *netdev)
1444 struct e1000_adapter *adapter = netdev_priv(netdev);
1445 struct e1000_hw *hw = &adapter->hw;
1447 WARN_ON(test_bit(__E1000_RESETTING, &adapter->flags));
1448 e1000_down(adapter);
1449 e1000_power_down_phy(adapter);
1450 e1000_free_irq(adapter);
1452 e1000_free_all_tx_resources(adapter);
1453 e1000_free_all_rx_resources(adapter);
1455 /* kill manageability vlan ID if supported, but not if a vlan with
1456 * the same ID is registered on the host OS (let 8021q kill it) */
1457 if ((hw->mng_cookie.status &
1458 E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) &&
1459 !test_bit(adapter->mng_vlan_id, adapter->active_vlans)) {
1460 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
1463 return 0;
1467 * e1000_check_64k_bound - check that memory doesn't cross 64kB boundary
1468 * @adapter: address of board private structure
1469 * @start: address of beginning of memory
1470 * @len: length of memory
1472 static bool e1000_check_64k_bound(struct e1000_adapter *adapter, void *start,
1473 unsigned long len)
1475 struct e1000_hw *hw = &adapter->hw;
1476 unsigned long begin = (unsigned long)start;
1477 unsigned long end = begin + len;
1479 /* First rev 82545 and 82546 need to not allow any memory
1480 * write location to cross 64k boundary due to errata 23 */
1481 if (hw->mac_type == e1000_82545 ||
1482 hw->mac_type == e1000_ce4100 ||
1483 hw->mac_type == e1000_82546) {
1484 return ((begin ^ (end - 1)) >> 16) != 0 ? false : true;
1487 return true;
1491 * e1000_setup_tx_resources - allocate Tx resources (Descriptors)
1492 * @adapter: board private structure
1493 * @txdr: tx descriptor ring (for a specific queue) to setup
1495 * Return 0 on success, negative on failure
1498 static int e1000_setup_tx_resources(struct e1000_adapter *adapter,
1499 struct e1000_tx_ring *txdr)
1501 struct pci_dev *pdev = adapter->pdev;
1502 int size;
1504 size = sizeof(struct e1000_buffer) * txdr->count;
1505 txdr->buffer_info = vzalloc(size);
1506 if (!txdr->buffer_info) {
1507 e_err(probe, "Unable to allocate memory for the Tx descriptor "
1508 "ring\n");
1509 return -ENOMEM;
1512 /* round up to nearest 4K */
1514 txdr->size = txdr->count * sizeof(struct e1000_tx_desc);
1515 txdr->size = ALIGN(txdr->size, 4096);
1517 txdr->desc = dma_alloc_coherent(&pdev->dev, txdr->size, &txdr->dma,
1518 GFP_KERNEL);
1519 if (!txdr->desc) {
1520 setup_tx_desc_die:
1521 vfree(txdr->buffer_info);
1522 e_err(probe, "Unable to allocate memory for the Tx descriptor "
1523 "ring\n");
1524 return -ENOMEM;
1527 /* Fix for errata 23, can't cross 64kB boundary */
1528 if (!e1000_check_64k_bound(adapter, txdr->desc, txdr->size)) {
1529 void *olddesc = txdr->desc;
1530 dma_addr_t olddma = txdr->dma;
1531 e_err(tx_err, "txdr align check failed: %u bytes at %p\n",
1532 txdr->size, txdr->desc);
1533 /* Try again, without freeing the previous */
1534 txdr->desc = dma_alloc_coherent(&pdev->dev, txdr->size,
1535 &txdr->dma, GFP_KERNEL);
1536 /* Failed allocation, critical failure */
1537 if (!txdr->desc) {
1538 dma_free_coherent(&pdev->dev, txdr->size, olddesc,
1539 olddma);
1540 goto setup_tx_desc_die;
1543 if (!e1000_check_64k_bound(adapter, txdr->desc, txdr->size)) {
1544 /* give up */
1545 dma_free_coherent(&pdev->dev, txdr->size, txdr->desc,
1546 txdr->dma);
1547 dma_free_coherent(&pdev->dev, txdr->size, olddesc,
1548 olddma);
1549 e_err(probe, "Unable to allocate aligned memory "
1550 "for the transmit descriptor ring\n");
1551 vfree(txdr->buffer_info);
1552 return -ENOMEM;
1553 } else {
1554 /* Free old allocation, new allocation was successful */
1555 dma_free_coherent(&pdev->dev, txdr->size, olddesc,
1556 olddma);
1559 memset(txdr->desc, 0, txdr->size);
1561 txdr->next_to_use = 0;
1562 txdr->next_to_clean = 0;
1564 return 0;
1568 * e1000_setup_all_tx_resources - wrapper to allocate Tx resources
1569 * (Descriptors) for all queues
1570 * @adapter: board private structure
1572 * Return 0 on success, negative on failure
1575 int e1000_setup_all_tx_resources(struct e1000_adapter *adapter)
1577 int i, err = 0;
1579 for (i = 0; i < adapter->num_tx_queues; i++) {
1580 err = e1000_setup_tx_resources(adapter, &adapter->tx_ring[i]);
1581 if (err) {
1582 e_err(probe, "Allocation for Tx Queue %u failed\n", i);
1583 for (i-- ; i >= 0; i--)
1584 e1000_free_tx_resources(adapter,
1585 &adapter->tx_ring[i]);
1586 break;
1590 return err;
1594 * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
1595 * @adapter: board private structure
1597 * Configure the Tx unit of the MAC after a reset.
1600 static void e1000_configure_tx(struct e1000_adapter *adapter)
1602 u64 tdba;
1603 struct e1000_hw *hw = &adapter->hw;
1604 u32 tdlen, tctl, tipg;
1605 u32 ipgr1, ipgr2;
1607 /* Setup the HW Tx Head and Tail descriptor pointers */
1609 switch (adapter->num_tx_queues) {
1610 case 1:
1611 default:
1612 tdba = adapter->tx_ring[0].dma;
1613 tdlen = adapter->tx_ring[0].count *
1614 sizeof(struct e1000_tx_desc);
1615 ew32(TDLEN, tdlen);
1616 ew32(TDBAH, (tdba >> 32));
1617 ew32(TDBAL, (tdba & 0x00000000ffffffffULL));
1618 ew32(TDT, 0);
1619 ew32(TDH, 0);
1620 adapter->tx_ring[0].tdh = ((hw->mac_type >= e1000_82543) ? E1000_TDH : E1000_82542_TDH);
1621 adapter->tx_ring[0].tdt = ((hw->mac_type >= e1000_82543) ? E1000_TDT : E1000_82542_TDT);
1622 break;
1625 /* Set the default values for the Tx Inter Packet Gap timer */
1626 if ((hw->media_type == e1000_media_type_fiber ||
1627 hw->media_type == e1000_media_type_internal_serdes))
1628 tipg = DEFAULT_82543_TIPG_IPGT_FIBER;
1629 else
1630 tipg = DEFAULT_82543_TIPG_IPGT_COPPER;
1632 switch (hw->mac_type) {
1633 case e1000_82542_rev2_0:
1634 case e1000_82542_rev2_1:
1635 tipg = DEFAULT_82542_TIPG_IPGT;
1636 ipgr1 = DEFAULT_82542_TIPG_IPGR1;
1637 ipgr2 = DEFAULT_82542_TIPG_IPGR2;
1638 break;
1639 default:
1640 ipgr1 = DEFAULT_82543_TIPG_IPGR1;
1641 ipgr2 = DEFAULT_82543_TIPG_IPGR2;
1642 break;
1644 tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
1645 tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
1646 ew32(TIPG, tipg);
1648 /* Set the Tx Interrupt Delay register */
1650 ew32(TIDV, adapter->tx_int_delay);
1651 if (hw->mac_type >= e1000_82540)
1652 ew32(TADV, adapter->tx_abs_int_delay);
1654 /* Program the Transmit Control Register */
1656 tctl = er32(TCTL);
1657 tctl &= ~E1000_TCTL_CT;
1658 tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
1659 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
1661 e1000_config_collision_dist(hw);
1663 /* Setup Transmit Descriptor Settings for eop descriptor */
1664 adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
1666 /* only set IDE if we are delaying interrupts using the timers */
1667 if (adapter->tx_int_delay)
1668 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
1670 if (hw->mac_type < e1000_82543)
1671 adapter->txd_cmd |= E1000_TXD_CMD_RPS;
1672 else
1673 adapter->txd_cmd |= E1000_TXD_CMD_RS;
1675 /* Cache if we're 82544 running in PCI-X because we'll
1676 * need this to apply a workaround later in the send path. */
1677 if (hw->mac_type == e1000_82544 &&
1678 hw->bus_type == e1000_bus_type_pcix)
1679 adapter->pcix_82544 = 1;
1681 ew32(TCTL, tctl);
1686 * e1000_setup_rx_resources - allocate Rx resources (Descriptors)
1687 * @adapter: board private structure
1688 * @rxdr: rx descriptor ring (for a specific queue) to setup
1690 * Returns 0 on success, negative on failure
1693 static int e1000_setup_rx_resources(struct e1000_adapter *adapter,
1694 struct e1000_rx_ring *rxdr)
1696 struct pci_dev *pdev = adapter->pdev;
1697 int size, desc_len;
1699 size = sizeof(struct e1000_buffer) * rxdr->count;
1700 rxdr->buffer_info = vzalloc(size);
1701 if (!rxdr->buffer_info) {
1702 e_err(probe, "Unable to allocate memory for the Rx descriptor "
1703 "ring\n");
1704 return -ENOMEM;
1707 desc_len = sizeof(struct e1000_rx_desc);
1709 /* Round up to nearest 4K */
1711 rxdr->size = rxdr->count * desc_len;
1712 rxdr->size = ALIGN(rxdr->size, 4096);
1714 rxdr->desc = dma_alloc_coherent(&pdev->dev, rxdr->size, &rxdr->dma,
1715 GFP_KERNEL);
1717 if (!rxdr->desc) {
1718 e_err(probe, "Unable to allocate memory for the Rx descriptor "
1719 "ring\n");
1720 setup_rx_desc_die:
1721 vfree(rxdr->buffer_info);
1722 return -ENOMEM;
1725 /* Fix for errata 23, can't cross 64kB boundary */
1726 if (!e1000_check_64k_bound(adapter, rxdr->desc, rxdr->size)) {
1727 void *olddesc = rxdr->desc;
1728 dma_addr_t olddma = rxdr->dma;
1729 e_err(rx_err, "rxdr align check failed: %u bytes at %p\n",
1730 rxdr->size, rxdr->desc);
1731 /* Try again, without freeing the previous */
1732 rxdr->desc = dma_alloc_coherent(&pdev->dev, rxdr->size,
1733 &rxdr->dma, GFP_KERNEL);
1734 /* Failed allocation, critical failure */
1735 if (!rxdr->desc) {
1736 dma_free_coherent(&pdev->dev, rxdr->size, olddesc,
1737 olddma);
1738 e_err(probe, "Unable to allocate memory for the Rx "
1739 "descriptor ring\n");
1740 goto setup_rx_desc_die;
1743 if (!e1000_check_64k_bound(adapter, rxdr->desc, rxdr->size)) {
1744 /* give up */
1745 dma_free_coherent(&pdev->dev, rxdr->size, rxdr->desc,
1746 rxdr->dma);
1747 dma_free_coherent(&pdev->dev, rxdr->size, olddesc,
1748 olddma);
1749 e_err(probe, "Unable to allocate aligned memory for "
1750 "the Rx descriptor ring\n");
1751 goto setup_rx_desc_die;
1752 } else {
1753 /* Free old allocation, new allocation was successful */
1754 dma_free_coherent(&pdev->dev, rxdr->size, olddesc,
1755 olddma);
1758 memset(rxdr->desc, 0, rxdr->size);
1760 rxdr->next_to_clean = 0;
1761 rxdr->next_to_use = 0;
1762 rxdr->rx_skb_top = NULL;
1764 return 0;
1768 * e1000_setup_all_rx_resources - wrapper to allocate Rx resources
1769 * (Descriptors) for all queues
1770 * @adapter: board private structure
1772 * Return 0 on success, negative on failure
1775 int e1000_setup_all_rx_resources(struct e1000_adapter *adapter)
1777 int i, err = 0;
1779 for (i = 0; i < adapter->num_rx_queues; i++) {
1780 err = e1000_setup_rx_resources(adapter, &adapter->rx_ring[i]);
1781 if (err) {
1782 e_err(probe, "Allocation for Rx Queue %u failed\n", i);
1783 for (i-- ; i >= 0; i--)
1784 e1000_free_rx_resources(adapter,
1785 &adapter->rx_ring[i]);
1786 break;
1790 return err;
1794 * e1000_setup_rctl - configure the receive control registers
1795 * @adapter: Board private structure
1797 static void e1000_setup_rctl(struct e1000_adapter *adapter)
1799 struct e1000_hw *hw = &adapter->hw;
1800 u32 rctl;
1802 rctl = er32(RCTL);
1804 rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
1806 rctl |= E1000_RCTL_BAM | E1000_RCTL_LBM_NO |
1807 E1000_RCTL_RDMTS_HALF |
1808 (hw->mc_filter_type << E1000_RCTL_MO_SHIFT);
1810 if (hw->tbi_compatibility_on == 1)
1811 rctl |= E1000_RCTL_SBP;
1812 else
1813 rctl &= ~E1000_RCTL_SBP;
1815 if (adapter->netdev->mtu <= ETH_DATA_LEN)
1816 rctl &= ~E1000_RCTL_LPE;
1817 else
1818 rctl |= E1000_RCTL_LPE;
1820 /* Setup buffer sizes */
1821 rctl &= ~E1000_RCTL_SZ_4096;
1822 rctl |= E1000_RCTL_BSEX;
1823 switch (adapter->rx_buffer_len) {
1824 case E1000_RXBUFFER_2048:
1825 default:
1826 rctl |= E1000_RCTL_SZ_2048;
1827 rctl &= ~E1000_RCTL_BSEX;
1828 break;
1829 case E1000_RXBUFFER_4096:
1830 rctl |= E1000_RCTL_SZ_4096;
1831 break;
1832 case E1000_RXBUFFER_8192:
1833 rctl |= E1000_RCTL_SZ_8192;
1834 break;
1835 case E1000_RXBUFFER_16384:
1836 rctl |= E1000_RCTL_SZ_16384;
1837 break;
1840 ew32(RCTL, rctl);
1844 * e1000_configure_rx - Configure 8254x Receive Unit after Reset
1845 * @adapter: board private structure
1847 * Configure the Rx unit of the MAC after a reset.
1850 static void e1000_configure_rx(struct e1000_adapter *adapter)
1852 u64 rdba;
1853 struct e1000_hw *hw = &adapter->hw;
1854 u32 rdlen, rctl, rxcsum;
1856 if (adapter->netdev->mtu > ETH_DATA_LEN) {
1857 rdlen = adapter->rx_ring[0].count *
1858 sizeof(struct e1000_rx_desc);
1859 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
1860 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
1861 } else {
1862 rdlen = adapter->rx_ring[0].count *
1863 sizeof(struct e1000_rx_desc);
1864 adapter->clean_rx = e1000_clean_rx_irq;
1865 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
1868 /* disable receives while setting up the descriptors */
1869 rctl = er32(RCTL);
1870 ew32(RCTL, rctl & ~E1000_RCTL_EN);
1872 /* set the Receive Delay Timer Register */
1873 ew32(RDTR, adapter->rx_int_delay);
1875 if (hw->mac_type >= e1000_82540) {
1876 ew32(RADV, adapter->rx_abs_int_delay);
1877 if (adapter->itr_setting != 0)
1878 ew32(ITR, 1000000000 / (adapter->itr * 256));
1881 /* Setup the HW Rx Head and Tail Descriptor Pointers and
1882 * the Base and Length of the Rx Descriptor Ring */
1883 switch (adapter->num_rx_queues) {
1884 case 1:
1885 default:
1886 rdba = adapter->rx_ring[0].dma;
1887 ew32(RDLEN, rdlen);
1888 ew32(RDBAH, (rdba >> 32));
1889 ew32(RDBAL, (rdba & 0x00000000ffffffffULL));
1890 ew32(RDT, 0);
1891 ew32(RDH, 0);
1892 adapter->rx_ring[0].rdh = ((hw->mac_type >= e1000_82543) ? E1000_RDH : E1000_82542_RDH);
1893 adapter->rx_ring[0].rdt = ((hw->mac_type >= e1000_82543) ? E1000_RDT : E1000_82542_RDT);
1894 break;
1897 /* Enable 82543 Receive Checksum Offload for TCP and UDP */
1898 if (hw->mac_type >= e1000_82543) {
1899 rxcsum = er32(RXCSUM);
1900 if (adapter->rx_csum)
1901 rxcsum |= E1000_RXCSUM_TUOFL;
1902 else
1903 /* don't need to clear IPPCSE as it defaults to 0 */
1904 rxcsum &= ~E1000_RXCSUM_TUOFL;
1905 ew32(RXCSUM, rxcsum);
1908 /* Enable Receives */
1909 ew32(RCTL, rctl | E1000_RCTL_EN);
1913 * e1000_free_tx_resources - Free Tx Resources per Queue
1914 * @adapter: board private structure
1915 * @tx_ring: Tx descriptor ring for a specific queue
1917 * Free all transmit software resources
1920 static void e1000_free_tx_resources(struct e1000_adapter *adapter,
1921 struct e1000_tx_ring *tx_ring)
1923 struct pci_dev *pdev = adapter->pdev;
1925 e1000_clean_tx_ring(adapter, tx_ring);
1927 vfree(tx_ring->buffer_info);
1928 tx_ring->buffer_info = NULL;
1930 dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
1931 tx_ring->dma);
1933 tx_ring->desc = NULL;
1937 * e1000_free_all_tx_resources - Free Tx Resources for All Queues
1938 * @adapter: board private structure
1940 * Free all transmit software resources
1943 void e1000_free_all_tx_resources(struct e1000_adapter *adapter)
1945 int i;
1947 for (i = 0; i < adapter->num_tx_queues; i++)
1948 e1000_free_tx_resources(adapter, &adapter->tx_ring[i]);
1951 static void e1000_unmap_and_free_tx_resource(struct e1000_adapter *adapter,
1952 struct e1000_buffer *buffer_info)
1954 if (buffer_info->dma) {
1955 if (buffer_info->mapped_as_page)
1956 dma_unmap_page(&adapter->pdev->dev, buffer_info->dma,
1957 buffer_info->length, DMA_TO_DEVICE);
1958 else
1959 dma_unmap_single(&adapter->pdev->dev, buffer_info->dma,
1960 buffer_info->length,
1961 DMA_TO_DEVICE);
1962 buffer_info->dma = 0;
1964 if (buffer_info->skb) {
1965 dev_kfree_skb_any(buffer_info->skb);
1966 buffer_info->skb = NULL;
1968 buffer_info->time_stamp = 0;
1969 /* buffer_info must be completely set up in the transmit path */
1973 * e1000_clean_tx_ring - Free Tx Buffers
1974 * @adapter: board private structure
1975 * @tx_ring: ring to be cleaned
1978 static void e1000_clean_tx_ring(struct e1000_adapter *adapter,
1979 struct e1000_tx_ring *tx_ring)
1981 struct e1000_hw *hw = &adapter->hw;
1982 struct e1000_buffer *buffer_info;
1983 unsigned long size;
1984 unsigned int i;
1986 /* Free all the Tx ring sk_buffs */
1988 for (i = 0; i < tx_ring->count; i++) {
1989 buffer_info = &tx_ring->buffer_info[i];
1990 e1000_unmap_and_free_tx_resource(adapter, buffer_info);
1993 size = sizeof(struct e1000_buffer) * tx_ring->count;
1994 memset(tx_ring->buffer_info, 0, size);
1996 /* Zero out the descriptor ring */
1998 memset(tx_ring->desc, 0, tx_ring->size);
2000 tx_ring->next_to_use = 0;
2001 tx_ring->next_to_clean = 0;
2002 tx_ring->last_tx_tso = 0;
2004 writel(0, hw->hw_addr + tx_ring->tdh);
2005 writel(0, hw->hw_addr + tx_ring->tdt);
2009 * e1000_clean_all_tx_rings - Free Tx Buffers for all queues
2010 * @adapter: board private structure
2013 static void e1000_clean_all_tx_rings(struct e1000_adapter *adapter)
2015 int i;
2017 for (i = 0; i < adapter->num_tx_queues; i++)
2018 e1000_clean_tx_ring(adapter, &adapter->tx_ring[i]);
2022 * e1000_free_rx_resources - Free Rx Resources
2023 * @adapter: board private structure
2024 * @rx_ring: ring to clean the resources from
2026 * Free all receive software resources
2029 static void e1000_free_rx_resources(struct e1000_adapter *adapter,
2030 struct e1000_rx_ring *rx_ring)
2032 struct pci_dev *pdev = adapter->pdev;
2034 e1000_clean_rx_ring(adapter, rx_ring);
2036 vfree(rx_ring->buffer_info);
2037 rx_ring->buffer_info = NULL;
2039 dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
2040 rx_ring->dma);
2042 rx_ring->desc = NULL;
2046 * e1000_free_all_rx_resources - Free Rx Resources for All Queues
2047 * @adapter: board private structure
2049 * Free all receive software resources
2052 void e1000_free_all_rx_resources(struct e1000_adapter *adapter)
2054 int i;
2056 for (i = 0; i < adapter->num_rx_queues; i++)
2057 e1000_free_rx_resources(adapter, &adapter->rx_ring[i]);
2061 * e1000_clean_rx_ring - Free Rx Buffers per Queue
2062 * @adapter: board private structure
2063 * @rx_ring: ring to free buffers from
2066 static void e1000_clean_rx_ring(struct e1000_adapter *adapter,
2067 struct e1000_rx_ring *rx_ring)
2069 struct e1000_hw *hw = &adapter->hw;
2070 struct e1000_buffer *buffer_info;
2071 struct pci_dev *pdev = adapter->pdev;
2072 unsigned long size;
2073 unsigned int i;
2075 /* Free all the Rx ring sk_buffs */
2076 for (i = 0; i < rx_ring->count; i++) {
2077 buffer_info = &rx_ring->buffer_info[i];
2078 if (buffer_info->dma &&
2079 adapter->clean_rx == e1000_clean_rx_irq) {
2080 dma_unmap_single(&pdev->dev, buffer_info->dma,
2081 buffer_info->length,
2082 DMA_FROM_DEVICE);
2083 } else if (buffer_info->dma &&
2084 adapter->clean_rx == e1000_clean_jumbo_rx_irq) {
2085 dma_unmap_page(&pdev->dev, buffer_info->dma,
2086 buffer_info->length,
2087 DMA_FROM_DEVICE);
2090 buffer_info->dma = 0;
2091 if (buffer_info->page) {
2092 put_page(buffer_info->page);
2093 buffer_info->page = NULL;
2095 if (buffer_info->skb) {
2096 dev_kfree_skb(buffer_info->skb);
2097 buffer_info->skb = NULL;
2101 /* there also may be some cached data from a chained receive */
2102 if (rx_ring->rx_skb_top) {
2103 dev_kfree_skb(rx_ring->rx_skb_top);
2104 rx_ring->rx_skb_top = NULL;
2107 size = sizeof(struct e1000_buffer) * rx_ring->count;
2108 memset(rx_ring->buffer_info, 0, size);
2110 /* Zero out the descriptor ring */
2111 memset(rx_ring->desc, 0, rx_ring->size);
2113 rx_ring->next_to_clean = 0;
2114 rx_ring->next_to_use = 0;
2116 writel(0, hw->hw_addr + rx_ring->rdh);
2117 writel(0, hw->hw_addr + rx_ring->rdt);
2121 * e1000_clean_all_rx_rings - Free Rx Buffers for all queues
2122 * @adapter: board private structure
2125 static void e1000_clean_all_rx_rings(struct e1000_adapter *adapter)
2127 int i;
2129 for (i = 0; i < adapter->num_rx_queues; i++)
2130 e1000_clean_rx_ring(adapter, &adapter->rx_ring[i]);
2133 /* The 82542 2.0 (revision 2) needs to have the receive unit in reset
2134 * and memory write and invalidate disabled for certain operations
2136 static void e1000_enter_82542_rst(struct e1000_adapter *adapter)
2138 struct e1000_hw *hw = &adapter->hw;
2139 struct net_device *netdev = adapter->netdev;
2140 u32 rctl;
2142 e1000_pci_clear_mwi(hw);
2144 rctl = er32(RCTL);
2145 rctl |= E1000_RCTL_RST;
2146 ew32(RCTL, rctl);
2147 E1000_WRITE_FLUSH();
2148 mdelay(5);
2150 if (netif_running(netdev))
2151 e1000_clean_all_rx_rings(adapter);
2154 static void e1000_leave_82542_rst(struct e1000_adapter *adapter)
2156 struct e1000_hw *hw = &adapter->hw;
2157 struct net_device *netdev = adapter->netdev;
2158 u32 rctl;
2160 rctl = er32(RCTL);
2161 rctl &= ~E1000_RCTL_RST;
2162 ew32(RCTL, rctl);
2163 E1000_WRITE_FLUSH();
2164 mdelay(5);
2166 if (hw->pci_cmd_word & PCI_COMMAND_INVALIDATE)
2167 e1000_pci_set_mwi(hw);
2169 if (netif_running(netdev)) {
2170 /* No need to loop, because 82542 supports only 1 queue */
2171 struct e1000_rx_ring *ring = &adapter->rx_ring[0];
2172 e1000_configure_rx(adapter);
2173 adapter->alloc_rx_buf(adapter, ring, E1000_DESC_UNUSED(ring));
2178 * e1000_set_mac - Change the Ethernet Address of the NIC
2179 * @netdev: network interface device structure
2180 * @p: pointer to an address structure
2182 * Returns 0 on success, negative on failure
2185 static int e1000_set_mac(struct net_device *netdev, void *p)
2187 struct e1000_adapter *adapter = netdev_priv(netdev);
2188 struct e1000_hw *hw = &adapter->hw;
2189 struct sockaddr *addr = p;
2191 if (!is_valid_ether_addr(addr->sa_data))
2192 return -EADDRNOTAVAIL;
2194 /* 82542 2.0 needs to be in reset to write receive address registers */
2196 if (hw->mac_type == e1000_82542_rev2_0)
2197 e1000_enter_82542_rst(adapter);
2199 memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2200 memcpy(hw->mac_addr, addr->sa_data, netdev->addr_len);
2202 e1000_rar_set(hw, hw->mac_addr, 0);
2204 if (hw->mac_type == e1000_82542_rev2_0)
2205 e1000_leave_82542_rst(adapter);
2207 return 0;
2211 * e1000_set_rx_mode - Secondary Unicast, Multicast and Promiscuous mode set
2212 * @netdev: network interface device structure
2214 * The set_rx_mode entry point is called whenever the unicast or multicast
2215 * address lists or the network interface flags are updated. This routine is
2216 * responsible for configuring the hardware for proper unicast, multicast,
2217 * promiscuous mode, and all-multi behavior.
2220 static void e1000_set_rx_mode(struct net_device *netdev)
2222 struct e1000_adapter *adapter = netdev_priv(netdev);
2223 struct e1000_hw *hw = &adapter->hw;
2224 struct netdev_hw_addr *ha;
2225 bool use_uc = false;
2226 u32 rctl;
2227 u32 hash_value;
2228 int i, rar_entries = E1000_RAR_ENTRIES;
2229 int mta_reg_count = E1000_NUM_MTA_REGISTERS;
2230 u32 *mcarray = kcalloc(mta_reg_count, sizeof(u32), GFP_ATOMIC);
2232 if (!mcarray) {
2233 e_err(probe, "memory allocation failed\n");
2234 return;
2237 /* Check for Promiscuous and All Multicast modes */
2239 rctl = er32(RCTL);
2241 if (netdev->flags & IFF_PROMISC) {
2242 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2243 rctl &= ~E1000_RCTL_VFE;
2244 } else {
2245 if (netdev->flags & IFF_ALLMULTI)
2246 rctl |= E1000_RCTL_MPE;
2247 else
2248 rctl &= ~E1000_RCTL_MPE;
2249 /* Enable VLAN filter if there is a VLAN */
2250 if (e1000_vlan_used(adapter))
2251 rctl |= E1000_RCTL_VFE;
2254 if (netdev_uc_count(netdev) > rar_entries - 1) {
2255 rctl |= E1000_RCTL_UPE;
2256 } else if (!(netdev->flags & IFF_PROMISC)) {
2257 rctl &= ~E1000_RCTL_UPE;
2258 use_uc = true;
2261 ew32(RCTL, rctl);
2263 /* 82542 2.0 needs to be in reset to write receive address registers */
2265 if (hw->mac_type == e1000_82542_rev2_0)
2266 e1000_enter_82542_rst(adapter);
2268 /* load the first 14 addresses into the exact filters 1-14. Unicast
2269 * addresses take precedence to avoid disabling unicast filtering
2270 * when possible.
2272 * RAR 0 is used for the station MAC address
2273 * if there are not 14 addresses, go ahead and clear the filters
2275 i = 1;
2276 if (use_uc)
2277 netdev_for_each_uc_addr(ha, netdev) {
2278 if (i == rar_entries)
2279 break;
2280 e1000_rar_set(hw, ha->addr, i++);
2283 netdev_for_each_mc_addr(ha, netdev) {
2284 if (i == rar_entries) {
2285 /* load any remaining addresses into the hash table */
2286 u32 hash_reg, hash_bit, mta;
2287 hash_value = e1000_hash_mc_addr(hw, ha->addr);
2288 hash_reg = (hash_value >> 5) & 0x7F;
2289 hash_bit = hash_value & 0x1F;
2290 mta = (1 << hash_bit);
2291 mcarray[hash_reg] |= mta;
2292 } else {
2293 e1000_rar_set(hw, ha->addr, i++);
2297 for (; i < rar_entries; i++) {
2298 E1000_WRITE_REG_ARRAY(hw, RA, i << 1, 0);
2299 E1000_WRITE_FLUSH();
2300 E1000_WRITE_REG_ARRAY(hw, RA, (i << 1) + 1, 0);
2301 E1000_WRITE_FLUSH();
2304 /* write the hash table completely, write from bottom to avoid
2305 * both stupid write combining chipsets, and flushing each write */
2306 for (i = mta_reg_count - 1; i >= 0 ; i--) {
2308 * If we are on an 82544 has an errata where writing odd
2309 * offsets overwrites the previous even offset, but writing
2310 * backwards over the range solves the issue by always
2311 * writing the odd offset first
2313 E1000_WRITE_REG_ARRAY(hw, MTA, i, mcarray[i]);
2315 E1000_WRITE_FLUSH();
2317 if (hw->mac_type == e1000_82542_rev2_0)
2318 e1000_leave_82542_rst(adapter);
2320 kfree(mcarray);
2324 * e1000_update_phy_info_task - get phy info
2325 * @work: work struct contained inside adapter struct
2327 * Need to wait a few seconds after link up to get diagnostic information from
2328 * the phy
2330 static void e1000_update_phy_info_task(struct work_struct *work)
2332 struct e1000_adapter *adapter = container_of(work,
2333 struct e1000_adapter,
2334 phy_info_task.work);
2335 if (test_bit(__E1000_DOWN, &adapter->flags))
2336 return;
2337 mutex_lock(&adapter->mutex);
2338 e1000_phy_get_info(&adapter->hw, &adapter->phy_info);
2339 mutex_unlock(&adapter->mutex);
2343 * e1000_82547_tx_fifo_stall_task - task to complete work
2344 * @work: work struct contained inside adapter struct
2346 static void e1000_82547_tx_fifo_stall_task(struct work_struct *work)
2348 struct e1000_adapter *adapter = container_of(work,
2349 struct e1000_adapter,
2350 fifo_stall_task.work);
2351 struct e1000_hw *hw = &adapter->hw;
2352 struct net_device *netdev = adapter->netdev;
2353 u32 tctl;
2355 if (test_bit(__E1000_DOWN, &adapter->flags))
2356 return;
2357 mutex_lock(&adapter->mutex);
2358 if (atomic_read(&adapter->tx_fifo_stall)) {
2359 if ((er32(TDT) == er32(TDH)) &&
2360 (er32(TDFT) == er32(TDFH)) &&
2361 (er32(TDFTS) == er32(TDFHS))) {
2362 tctl = er32(TCTL);
2363 ew32(TCTL, tctl & ~E1000_TCTL_EN);
2364 ew32(TDFT, adapter->tx_head_addr);
2365 ew32(TDFH, adapter->tx_head_addr);
2366 ew32(TDFTS, adapter->tx_head_addr);
2367 ew32(TDFHS, adapter->tx_head_addr);
2368 ew32(TCTL, tctl);
2369 E1000_WRITE_FLUSH();
2371 adapter->tx_fifo_head = 0;
2372 atomic_set(&adapter->tx_fifo_stall, 0);
2373 netif_wake_queue(netdev);
2374 } else if (!test_bit(__E1000_DOWN, &adapter->flags)) {
2375 schedule_delayed_work(&adapter->fifo_stall_task, 1);
2378 mutex_unlock(&adapter->mutex);
2381 bool e1000_has_link(struct e1000_adapter *adapter)
2383 struct e1000_hw *hw = &adapter->hw;
2384 bool link_active = false;
2386 /* get_link_status is set on LSC (link status) interrupt or rx
2387 * sequence error interrupt (except on intel ce4100).
2388 * get_link_status will stay false until the
2389 * e1000_check_for_link establishes link for copper adapters
2390 * ONLY
2392 switch (hw->media_type) {
2393 case e1000_media_type_copper:
2394 if (hw->mac_type == e1000_ce4100)
2395 hw->get_link_status = 1;
2396 if (hw->get_link_status) {
2397 e1000_check_for_link(hw);
2398 link_active = !hw->get_link_status;
2399 } else {
2400 link_active = true;
2402 break;
2403 case e1000_media_type_fiber:
2404 e1000_check_for_link(hw);
2405 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
2406 break;
2407 case e1000_media_type_internal_serdes:
2408 e1000_check_for_link(hw);
2409 link_active = hw->serdes_has_link;
2410 break;
2411 default:
2412 break;
2415 return link_active;
2419 * e1000_watchdog - work function
2420 * @work: work struct contained inside adapter struct
2422 static void e1000_watchdog(struct work_struct *work)
2424 struct e1000_adapter *adapter = container_of(work,
2425 struct e1000_adapter,
2426 watchdog_task.work);
2427 struct e1000_hw *hw = &adapter->hw;
2428 struct net_device *netdev = adapter->netdev;
2429 struct e1000_tx_ring *txdr = adapter->tx_ring;
2430 u32 link, tctl;
2432 if (test_bit(__E1000_DOWN, &adapter->flags))
2433 return;
2435 mutex_lock(&adapter->mutex);
2436 link = e1000_has_link(adapter);
2437 if ((netif_carrier_ok(netdev)) && link)
2438 goto link_up;
2440 if (link) {
2441 if (!netif_carrier_ok(netdev)) {
2442 u32 ctrl;
2443 bool txb2b = true;
2444 /* update snapshot of PHY registers on LSC */
2445 e1000_get_speed_and_duplex(hw,
2446 &adapter->link_speed,
2447 &adapter->link_duplex);
2449 ctrl = er32(CTRL);
2450 pr_info("%s NIC Link is Up %d Mbps %s, "
2451 "Flow Control: %s\n",
2452 netdev->name,
2453 adapter->link_speed,
2454 adapter->link_duplex == FULL_DUPLEX ?
2455 "Full Duplex" : "Half Duplex",
2456 ((ctrl & E1000_CTRL_TFCE) && (ctrl &
2457 E1000_CTRL_RFCE)) ? "RX/TX" : ((ctrl &
2458 E1000_CTRL_RFCE) ? "RX" : ((ctrl &
2459 E1000_CTRL_TFCE) ? "TX" : "None")));
2461 /* adjust timeout factor according to speed/duplex */
2462 adapter->tx_timeout_factor = 1;
2463 switch (adapter->link_speed) {
2464 case SPEED_10:
2465 txb2b = false;
2466 adapter->tx_timeout_factor = 16;
2467 break;
2468 case SPEED_100:
2469 txb2b = false;
2470 /* maybe add some timeout factor ? */
2471 break;
2474 /* enable transmits in the hardware */
2475 tctl = er32(TCTL);
2476 tctl |= E1000_TCTL_EN;
2477 ew32(TCTL, tctl);
2479 netif_carrier_on(netdev);
2480 if (!test_bit(__E1000_DOWN, &adapter->flags))
2481 schedule_delayed_work(&adapter->phy_info_task,
2482 2 * HZ);
2483 adapter->smartspeed = 0;
2485 } else {
2486 if (netif_carrier_ok(netdev)) {
2487 adapter->link_speed = 0;
2488 adapter->link_duplex = 0;
2489 pr_info("%s NIC Link is Down\n",
2490 netdev->name);
2491 netif_carrier_off(netdev);
2493 if (!test_bit(__E1000_DOWN, &adapter->flags))
2494 schedule_delayed_work(&adapter->phy_info_task,
2495 2 * HZ);
2498 e1000_smartspeed(adapter);
2501 link_up:
2502 e1000_update_stats(adapter);
2504 hw->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
2505 adapter->tpt_old = adapter->stats.tpt;
2506 hw->collision_delta = adapter->stats.colc - adapter->colc_old;
2507 adapter->colc_old = adapter->stats.colc;
2509 adapter->gorcl = adapter->stats.gorcl - adapter->gorcl_old;
2510 adapter->gorcl_old = adapter->stats.gorcl;
2511 adapter->gotcl = adapter->stats.gotcl - adapter->gotcl_old;
2512 adapter->gotcl_old = adapter->stats.gotcl;
2514 e1000_update_adaptive(hw);
2516 if (!netif_carrier_ok(netdev)) {
2517 if (E1000_DESC_UNUSED(txdr) + 1 < txdr->count) {
2518 /* We've lost link, so the controller stops DMA,
2519 * but we've got queued Tx work that's never going
2520 * to get done, so reset controller to flush Tx.
2521 * (Do the reset outside of interrupt context). */
2522 adapter->tx_timeout_count++;
2523 schedule_work(&adapter->reset_task);
2524 /* exit immediately since reset is imminent */
2525 goto unlock;
2529 /* Simple mode for Interrupt Throttle Rate (ITR) */
2530 if (hw->mac_type >= e1000_82540 && adapter->itr_setting == 4) {
2532 * Symmetric Tx/Rx gets a reduced ITR=2000;
2533 * Total asymmetrical Tx or Rx gets ITR=8000;
2534 * everyone else is between 2000-8000.
2536 u32 goc = (adapter->gotcl + adapter->gorcl) / 10000;
2537 u32 dif = (adapter->gotcl > adapter->gorcl ?
2538 adapter->gotcl - adapter->gorcl :
2539 adapter->gorcl - adapter->gotcl) / 10000;
2540 u32 itr = goc > 0 ? (dif * 6000 / goc + 2000) : 8000;
2542 ew32(ITR, 1000000000 / (itr * 256));
2545 /* Cause software interrupt to ensure rx ring is cleaned */
2546 ew32(ICS, E1000_ICS_RXDMT0);
2548 /* Force detection of hung controller every watchdog period */
2549 adapter->detect_tx_hung = true;
2551 /* Reschedule the task */
2552 if (!test_bit(__E1000_DOWN, &adapter->flags))
2553 schedule_delayed_work(&adapter->watchdog_task, 2 * HZ);
2555 unlock:
2556 mutex_unlock(&adapter->mutex);
2559 enum latency_range {
2560 lowest_latency = 0,
2561 low_latency = 1,
2562 bulk_latency = 2,
2563 latency_invalid = 255
2567 * e1000_update_itr - update the dynamic ITR value based on statistics
2568 * @adapter: pointer to adapter
2569 * @itr_setting: current adapter->itr
2570 * @packets: the number of packets during this measurement interval
2571 * @bytes: the number of bytes during this measurement interval
2573 * Stores a new ITR value based on packets and byte
2574 * counts during the last interrupt. The advantage of per interrupt
2575 * computation is faster updates and more accurate ITR for the current
2576 * traffic pattern. Constants in this function were computed
2577 * based on theoretical maximum wire speed and thresholds were set based
2578 * on testing data as well as attempting to minimize response time
2579 * while increasing bulk throughput.
2580 * this functionality is controlled by the InterruptThrottleRate module
2581 * parameter (see e1000_param.c)
2583 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
2584 u16 itr_setting, int packets, int bytes)
2586 unsigned int retval = itr_setting;
2587 struct e1000_hw *hw = &adapter->hw;
2589 if (unlikely(hw->mac_type < e1000_82540))
2590 goto update_itr_done;
2592 if (packets == 0)
2593 goto update_itr_done;
2595 switch (itr_setting) {
2596 case lowest_latency:
2597 /* jumbo frames get bulk treatment*/
2598 if (bytes/packets > 8000)
2599 retval = bulk_latency;
2600 else if ((packets < 5) && (bytes > 512))
2601 retval = low_latency;
2602 break;
2603 case low_latency: /* 50 usec aka 20000 ints/s */
2604 if (bytes > 10000) {
2605 /* jumbo frames need bulk latency setting */
2606 if (bytes/packets > 8000)
2607 retval = bulk_latency;
2608 else if ((packets < 10) || ((bytes/packets) > 1200))
2609 retval = bulk_latency;
2610 else if ((packets > 35))
2611 retval = lowest_latency;
2612 } else if (bytes/packets > 2000)
2613 retval = bulk_latency;
2614 else if (packets <= 2 && bytes < 512)
2615 retval = lowest_latency;
2616 break;
2617 case bulk_latency: /* 250 usec aka 4000 ints/s */
2618 if (bytes > 25000) {
2619 if (packets > 35)
2620 retval = low_latency;
2621 } else if (bytes < 6000) {
2622 retval = low_latency;
2624 break;
2627 update_itr_done:
2628 return retval;
2631 static void e1000_set_itr(struct e1000_adapter *adapter)
2633 struct e1000_hw *hw = &adapter->hw;
2634 u16 current_itr;
2635 u32 new_itr = adapter->itr;
2637 if (unlikely(hw->mac_type < e1000_82540))
2638 return;
2640 /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2641 if (unlikely(adapter->link_speed != SPEED_1000)) {
2642 current_itr = 0;
2643 new_itr = 4000;
2644 goto set_itr_now;
2647 adapter->tx_itr = e1000_update_itr(adapter,
2648 adapter->tx_itr,
2649 adapter->total_tx_packets,
2650 adapter->total_tx_bytes);
2651 /* conservative mode (itr 3) eliminates the lowest_latency setting */
2652 if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2653 adapter->tx_itr = low_latency;
2655 adapter->rx_itr = e1000_update_itr(adapter,
2656 adapter->rx_itr,
2657 adapter->total_rx_packets,
2658 adapter->total_rx_bytes);
2659 /* conservative mode (itr 3) eliminates the lowest_latency setting */
2660 if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2661 adapter->rx_itr = low_latency;
2663 current_itr = max(adapter->rx_itr, adapter->tx_itr);
2665 switch (current_itr) {
2666 /* counts and packets in update_itr are dependent on these numbers */
2667 case lowest_latency:
2668 new_itr = 70000;
2669 break;
2670 case low_latency:
2671 new_itr = 20000; /* aka hwitr = ~200 */
2672 break;
2673 case bulk_latency:
2674 new_itr = 4000;
2675 break;
2676 default:
2677 break;
2680 set_itr_now:
2681 if (new_itr != adapter->itr) {
2682 /* this attempts to bias the interrupt rate towards Bulk
2683 * by adding intermediate steps when interrupt rate is
2684 * increasing */
2685 new_itr = new_itr > adapter->itr ?
2686 min(adapter->itr + (new_itr >> 2), new_itr) :
2687 new_itr;
2688 adapter->itr = new_itr;
2689 ew32(ITR, 1000000000 / (new_itr * 256));
2693 #define E1000_TX_FLAGS_CSUM 0x00000001
2694 #define E1000_TX_FLAGS_VLAN 0x00000002
2695 #define E1000_TX_FLAGS_TSO 0x00000004
2696 #define E1000_TX_FLAGS_IPV4 0x00000008
2697 #define E1000_TX_FLAGS_VLAN_MASK 0xffff0000
2698 #define E1000_TX_FLAGS_VLAN_SHIFT 16
2700 static int e1000_tso(struct e1000_adapter *adapter,
2701 struct e1000_tx_ring *tx_ring, struct sk_buff *skb)
2703 struct e1000_context_desc *context_desc;
2704 struct e1000_buffer *buffer_info;
2705 unsigned int i;
2706 u32 cmd_length = 0;
2707 u16 ipcse = 0, tucse, mss;
2708 u8 ipcss, ipcso, tucss, tucso, hdr_len;
2709 int err;
2711 if (skb_is_gso(skb)) {
2712 if (skb_header_cloned(skb)) {
2713 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2714 if (err)
2715 return err;
2718 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
2719 mss = skb_shinfo(skb)->gso_size;
2720 if (skb->protocol == htons(ETH_P_IP)) {
2721 struct iphdr *iph = ip_hdr(skb);
2722 iph->tot_len = 0;
2723 iph->check = 0;
2724 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
2725 iph->daddr, 0,
2726 IPPROTO_TCP,
2728 cmd_length = E1000_TXD_CMD_IP;
2729 ipcse = skb_transport_offset(skb) - 1;
2730 } else if (skb->protocol == htons(ETH_P_IPV6)) {
2731 ipv6_hdr(skb)->payload_len = 0;
2732 tcp_hdr(skb)->check =
2733 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
2734 &ipv6_hdr(skb)->daddr,
2735 0, IPPROTO_TCP, 0);
2736 ipcse = 0;
2738 ipcss = skb_network_offset(skb);
2739 ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
2740 tucss = skb_transport_offset(skb);
2741 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
2742 tucse = 0;
2744 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
2745 E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
2747 i = tx_ring->next_to_use;
2748 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
2749 buffer_info = &tx_ring->buffer_info[i];
2751 context_desc->lower_setup.ip_fields.ipcss = ipcss;
2752 context_desc->lower_setup.ip_fields.ipcso = ipcso;
2753 context_desc->lower_setup.ip_fields.ipcse = cpu_to_le16(ipcse);
2754 context_desc->upper_setup.tcp_fields.tucss = tucss;
2755 context_desc->upper_setup.tcp_fields.tucso = tucso;
2756 context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
2757 context_desc->tcp_seg_setup.fields.mss = cpu_to_le16(mss);
2758 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
2759 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
2761 buffer_info->time_stamp = jiffies;
2762 buffer_info->next_to_watch = i;
2764 if (++i == tx_ring->count) i = 0;
2765 tx_ring->next_to_use = i;
2767 return true;
2769 return false;
2772 static bool e1000_tx_csum(struct e1000_adapter *adapter,
2773 struct e1000_tx_ring *tx_ring, struct sk_buff *skb)
2775 struct e1000_context_desc *context_desc;
2776 struct e1000_buffer *buffer_info;
2777 unsigned int i;
2778 u8 css;
2779 u32 cmd_len = E1000_TXD_CMD_DEXT;
2781 if (skb->ip_summed != CHECKSUM_PARTIAL)
2782 return false;
2784 switch (skb->protocol) {
2785 case cpu_to_be16(ETH_P_IP):
2786 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
2787 cmd_len |= E1000_TXD_CMD_TCP;
2788 break;
2789 case cpu_to_be16(ETH_P_IPV6):
2790 /* XXX not handling all IPV6 headers */
2791 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
2792 cmd_len |= E1000_TXD_CMD_TCP;
2793 break;
2794 default:
2795 if (unlikely(net_ratelimit()))
2796 e_warn(drv, "checksum_partial proto=%x!\n",
2797 skb->protocol);
2798 break;
2801 css = skb_checksum_start_offset(skb);
2803 i = tx_ring->next_to_use;
2804 buffer_info = &tx_ring->buffer_info[i];
2805 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
2807 context_desc->lower_setup.ip_config = 0;
2808 context_desc->upper_setup.tcp_fields.tucss = css;
2809 context_desc->upper_setup.tcp_fields.tucso =
2810 css + skb->csum_offset;
2811 context_desc->upper_setup.tcp_fields.tucse = 0;
2812 context_desc->tcp_seg_setup.data = 0;
2813 context_desc->cmd_and_length = cpu_to_le32(cmd_len);
2815 buffer_info->time_stamp = jiffies;
2816 buffer_info->next_to_watch = i;
2818 if (unlikely(++i == tx_ring->count)) i = 0;
2819 tx_ring->next_to_use = i;
2821 return true;
2824 #define E1000_MAX_TXD_PWR 12
2825 #define E1000_MAX_DATA_PER_TXD (1<<E1000_MAX_TXD_PWR)
2827 static int e1000_tx_map(struct e1000_adapter *adapter,
2828 struct e1000_tx_ring *tx_ring,
2829 struct sk_buff *skb, unsigned int first,
2830 unsigned int max_per_txd, unsigned int nr_frags,
2831 unsigned int mss)
2833 struct e1000_hw *hw = &adapter->hw;
2834 struct pci_dev *pdev = adapter->pdev;
2835 struct e1000_buffer *buffer_info;
2836 unsigned int len = skb_headlen(skb);
2837 unsigned int offset = 0, size, count = 0, i;
2838 unsigned int f, bytecount, segs;
2840 i = tx_ring->next_to_use;
2842 while (len) {
2843 buffer_info = &tx_ring->buffer_info[i];
2844 size = min(len, max_per_txd);
2845 /* Workaround for Controller erratum --
2846 * descriptor for non-tso packet in a linear SKB that follows a
2847 * tso gets written back prematurely before the data is fully
2848 * DMA'd to the controller */
2849 if (!skb->data_len && tx_ring->last_tx_tso &&
2850 !skb_is_gso(skb)) {
2851 tx_ring->last_tx_tso = 0;
2852 size -= 4;
2855 /* Workaround for premature desc write-backs
2856 * in TSO mode. Append 4-byte sentinel desc */
2857 if (unlikely(mss && !nr_frags && size == len && size > 8))
2858 size -= 4;
2859 /* work-around for errata 10 and it applies
2860 * to all controllers in PCI-X mode
2861 * The fix is to make sure that the first descriptor of a
2862 * packet is smaller than 2048 - 16 - 16 (or 2016) bytes
2864 if (unlikely((hw->bus_type == e1000_bus_type_pcix) &&
2865 (size > 2015) && count == 0))
2866 size = 2015;
2868 /* Workaround for potential 82544 hang in PCI-X. Avoid
2869 * terminating buffers within evenly-aligned dwords. */
2870 if (unlikely(adapter->pcix_82544 &&
2871 !((unsigned long)(skb->data + offset + size - 1) & 4) &&
2872 size > 4))
2873 size -= 4;
2875 buffer_info->length = size;
2876 /* set time_stamp *before* dma to help avoid a possible race */
2877 buffer_info->time_stamp = jiffies;
2878 buffer_info->mapped_as_page = false;
2879 buffer_info->dma = dma_map_single(&pdev->dev,
2880 skb->data + offset,
2881 size, DMA_TO_DEVICE);
2882 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2883 goto dma_error;
2884 buffer_info->next_to_watch = i;
2886 len -= size;
2887 offset += size;
2888 count++;
2889 if (len) {
2890 i++;
2891 if (unlikely(i == tx_ring->count))
2892 i = 0;
2896 for (f = 0; f < nr_frags; f++) {
2897 const struct skb_frag_struct *frag;
2899 frag = &skb_shinfo(skb)->frags[f];
2900 len = skb_frag_size(frag);
2901 offset = 0;
2903 while (len) {
2904 unsigned long bufend;
2905 i++;
2906 if (unlikely(i == tx_ring->count))
2907 i = 0;
2909 buffer_info = &tx_ring->buffer_info[i];
2910 size = min(len, max_per_txd);
2911 /* Workaround for premature desc write-backs
2912 * in TSO mode. Append 4-byte sentinel desc */
2913 if (unlikely(mss && f == (nr_frags-1) && size == len && size > 8))
2914 size -= 4;
2915 /* Workaround for potential 82544 hang in PCI-X.
2916 * Avoid terminating buffers within evenly-aligned
2917 * dwords. */
2918 bufend = (unsigned long)
2919 page_to_phys(skb_frag_page(frag));
2920 bufend += offset + size - 1;
2921 if (unlikely(adapter->pcix_82544 &&
2922 !(bufend & 4) &&
2923 size > 4))
2924 size -= 4;
2926 buffer_info->length = size;
2927 buffer_info->time_stamp = jiffies;
2928 buffer_info->mapped_as_page = true;
2929 buffer_info->dma = skb_frag_dma_map(&pdev->dev, frag,
2930 offset, size, DMA_TO_DEVICE);
2931 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2932 goto dma_error;
2933 buffer_info->next_to_watch = i;
2935 len -= size;
2936 offset += size;
2937 count++;
2941 segs = skb_shinfo(skb)->gso_segs ?: 1;
2942 /* multiply data chunks by size of headers */
2943 bytecount = ((segs - 1) * skb_headlen(skb)) + skb->len;
2945 tx_ring->buffer_info[i].skb = skb;
2946 tx_ring->buffer_info[i].segs = segs;
2947 tx_ring->buffer_info[i].bytecount = bytecount;
2948 tx_ring->buffer_info[first].next_to_watch = i;
2950 return count;
2952 dma_error:
2953 dev_err(&pdev->dev, "TX DMA map failed\n");
2954 buffer_info->dma = 0;
2955 if (count)
2956 count--;
2958 while (count--) {
2959 if (i==0)
2960 i += tx_ring->count;
2961 i--;
2962 buffer_info = &tx_ring->buffer_info[i];
2963 e1000_unmap_and_free_tx_resource(adapter, buffer_info);
2966 return 0;
2969 static void e1000_tx_queue(struct e1000_adapter *adapter,
2970 struct e1000_tx_ring *tx_ring, int tx_flags,
2971 int count)
2973 struct e1000_hw *hw = &adapter->hw;
2974 struct e1000_tx_desc *tx_desc = NULL;
2975 struct e1000_buffer *buffer_info;
2976 u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
2977 unsigned int i;
2979 if (likely(tx_flags & E1000_TX_FLAGS_TSO)) {
2980 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
2981 E1000_TXD_CMD_TSE;
2982 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
2984 if (likely(tx_flags & E1000_TX_FLAGS_IPV4))
2985 txd_upper |= E1000_TXD_POPTS_IXSM << 8;
2988 if (likely(tx_flags & E1000_TX_FLAGS_CSUM)) {
2989 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
2990 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
2993 if (unlikely(tx_flags & E1000_TX_FLAGS_VLAN)) {
2994 txd_lower |= E1000_TXD_CMD_VLE;
2995 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
2998 i = tx_ring->next_to_use;
3000 while (count--) {
3001 buffer_info = &tx_ring->buffer_info[i];
3002 tx_desc = E1000_TX_DESC(*tx_ring, i);
3003 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
3004 tx_desc->lower.data =
3005 cpu_to_le32(txd_lower | buffer_info->length);
3006 tx_desc->upper.data = cpu_to_le32(txd_upper);
3007 if (unlikely(++i == tx_ring->count)) i = 0;
3010 tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
3012 /* Force memory writes to complete before letting h/w
3013 * know there are new descriptors to fetch. (Only
3014 * applicable for weak-ordered memory model archs,
3015 * such as IA-64). */
3016 wmb();
3018 tx_ring->next_to_use = i;
3019 writel(i, hw->hw_addr + tx_ring->tdt);
3020 /* we need this if more than one processor can write to our tail
3021 * at a time, it syncronizes IO on IA64/Altix systems */
3022 mmiowb();
3026 * 82547 workaround to avoid controller hang in half-duplex environment.
3027 * The workaround is to avoid queuing a large packet that would span
3028 * the internal Tx FIFO ring boundary by notifying the stack to resend
3029 * the packet at a later time. This gives the Tx FIFO an opportunity to
3030 * flush all packets. When that occurs, we reset the Tx FIFO pointers
3031 * to the beginning of the Tx FIFO.
3034 #define E1000_FIFO_HDR 0x10
3035 #define E1000_82547_PAD_LEN 0x3E0
3037 static int e1000_82547_fifo_workaround(struct e1000_adapter *adapter,
3038 struct sk_buff *skb)
3040 u32 fifo_space = adapter->tx_fifo_size - adapter->tx_fifo_head;
3041 u32 skb_fifo_len = skb->len + E1000_FIFO_HDR;
3043 skb_fifo_len = ALIGN(skb_fifo_len, E1000_FIFO_HDR);
3045 if (adapter->link_duplex != HALF_DUPLEX)
3046 goto no_fifo_stall_required;
3048 if (atomic_read(&adapter->tx_fifo_stall))
3049 return 1;
3051 if (skb_fifo_len >= (E1000_82547_PAD_LEN + fifo_space)) {
3052 atomic_set(&adapter->tx_fifo_stall, 1);
3053 return 1;
3056 no_fifo_stall_required:
3057 adapter->tx_fifo_head += skb_fifo_len;
3058 if (adapter->tx_fifo_head >= adapter->tx_fifo_size)
3059 adapter->tx_fifo_head -= adapter->tx_fifo_size;
3060 return 0;
3063 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
3065 struct e1000_adapter *adapter = netdev_priv(netdev);
3066 struct e1000_tx_ring *tx_ring = adapter->tx_ring;
3068 netif_stop_queue(netdev);
3069 /* Herbert's original patch had:
3070 * smp_mb__after_netif_stop_queue();
3071 * but since that doesn't exist yet, just open code it. */
3072 smp_mb();
3074 /* We need to check again in a case another CPU has just
3075 * made room available. */
3076 if (likely(E1000_DESC_UNUSED(tx_ring) < size))
3077 return -EBUSY;
3079 /* A reprieve! */
3080 netif_start_queue(netdev);
3081 ++adapter->restart_queue;
3082 return 0;
3085 static int e1000_maybe_stop_tx(struct net_device *netdev,
3086 struct e1000_tx_ring *tx_ring, int size)
3088 if (likely(E1000_DESC_UNUSED(tx_ring) >= size))
3089 return 0;
3090 return __e1000_maybe_stop_tx(netdev, size);
3093 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
3094 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
3095 struct net_device *netdev)
3097 struct e1000_adapter *adapter = netdev_priv(netdev);
3098 struct e1000_hw *hw = &adapter->hw;
3099 struct e1000_tx_ring *tx_ring;
3100 unsigned int first, max_per_txd = E1000_MAX_DATA_PER_TXD;
3101 unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
3102 unsigned int tx_flags = 0;
3103 unsigned int len = skb_headlen(skb);
3104 unsigned int nr_frags;
3105 unsigned int mss;
3106 int count = 0;
3107 int tso;
3108 unsigned int f;
3110 /* This goes back to the question of how to logically map a tx queue
3111 * to a flow. Right now, performance is impacted slightly negatively
3112 * if using multiple tx queues. If the stack breaks away from a
3113 * single qdisc implementation, we can look at this again. */
3114 tx_ring = adapter->tx_ring;
3116 if (unlikely(skb->len <= 0)) {
3117 dev_kfree_skb_any(skb);
3118 return NETDEV_TX_OK;
3121 mss = skb_shinfo(skb)->gso_size;
3122 /* The controller does a simple calculation to
3123 * make sure there is enough room in the FIFO before
3124 * initiating the DMA for each buffer. The calc is:
3125 * 4 = ceil(buffer len/mss). To make sure we don't
3126 * overrun the FIFO, adjust the max buffer len if mss
3127 * drops. */
3128 if (mss) {
3129 u8 hdr_len;
3130 max_per_txd = min(mss << 2, max_per_txd);
3131 max_txd_pwr = fls(max_per_txd) - 1;
3133 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
3134 if (skb->data_len && hdr_len == len) {
3135 switch (hw->mac_type) {
3136 unsigned int pull_size;
3137 case e1000_82544:
3138 /* Make sure we have room to chop off 4 bytes,
3139 * and that the end alignment will work out to
3140 * this hardware's requirements
3141 * NOTE: this is a TSO only workaround
3142 * if end byte alignment not correct move us
3143 * into the next dword */
3144 if ((unsigned long)(skb_tail_pointer(skb) - 1) & 4)
3145 break;
3146 /* fall through */
3147 pull_size = min((unsigned int)4, skb->data_len);
3148 if (!__pskb_pull_tail(skb, pull_size)) {
3149 e_err(drv, "__pskb_pull_tail "
3150 "failed.\n");
3151 dev_kfree_skb_any(skb);
3152 return NETDEV_TX_OK;
3154 len = skb_headlen(skb);
3155 break;
3156 default:
3157 /* do nothing */
3158 break;
3163 /* reserve a descriptor for the offload context */
3164 if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
3165 count++;
3166 count++;
3168 /* Controller Erratum workaround */
3169 if (!skb->data_len && tx_ring->last_tx_tso && !skb_is_gso(skb))
3170 count++;
3172 count += TXD_USE_COUNT(len, max_txd_pwr);
3174 if (adapter->pcix_82544)
3175 count++;
3177 /* work-around for errata 10 and it applies to all controllers
3178 * in PCI-X mode, so add one more descriptor to the count
3180 if (unlikely((hw->bus_type == e1000_bus_type_pcix) &&
3181 (len > 2015)))
3182 count++;
3184 nr_frags = skb_shinfo(skb)->nr_frags;
3185 for (f = 0; f < nr_frags; f++)
3186 count += TXD_USE_COUNT(skb_frag_size(&skb_shinfo(skb)->frags[f]),
3187 max_txd_pwr);
3188 if (adapter->pcix_82544)
3189 count += nr_frags;
3191 /* need: count + 2 desc gap to keep tail from touching
3192 * head, otherwise try next time */
3193 if (unlikely(e1000_maybe_stop_tx(netdev, tx_ring, count + 2)))
3194 return NETDEV_TX_BUSY;
3196 if (unlikely((hw->mac_type == e1000_82547) &&
3197 (e1000_82547_fifo_workaround(adapter, skb)))) {
3198 netif_stop_queue(netdev);
3199 if (!test_bit(__E1000_DOWN, &adapter->flags))
3200 schedule_delayed_work(&adapter->fifo_stall_task, 1);
3201 return NETDEV_TX_BUSY;
3204 if (vlan_tx_tag_present(skb)) {
3205 tx_flags |= E1000_TX_FLAGS_VLAN;
3206 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
3209 first = tx_ring->next_to_use;
3211 tso = e1000_tso(adapter, tx_ring, skb);
3212 if (tso < 0) {
3213 dev_kfree_skb_any(skb);
3214 return NETDEV_TX_OK;
3217 if (likely(tso)) {
3218 if (likely(hw->mac_type != e1000_82544))
3219 tx_ring->last_tx_tso = 1;
3220 tx_flags |= E1000_TX_FLAGS_TSO;
3221 } else if (likely(e1000_tx_csum(adapter, tx_ring, skb)))
3222 tx_flags |= E1000_TX_FLAGS_CSUM;
3224 if (likely(skb->protocol == htons(ETH_P_IP)))
3225 tx_flags |= E1000_TX_FLAGS_IPV4;
3227 count = e1000_tx_map(adapter, tx_ring, skb, first, max_per_txd,
3228 nr_frags, mss);
3230 if (count) {
3231 e1000_tx_queue(adapter, tx_ring, tx_flags, count);
3232 /* Make sure there is space in the ring for the next send. */
3233 e1000_maybe_stop_tx(netdev, tx_ring, MAX_SKB_FRAGS + 2);
3235 } else {
3236 dev_kfree_skb_any(skb);
3237 tx_ring->buffer_info[first].time_stamp = 0;
3238 tx_ring->next_to_use = first;
3241 return NETDEV_TX_OK;
3245 * e1000_tx_timeout - Respond to a Tx Hang
3246 * @netdev: network interface device structure
3249 static void e1000_tx_timeout(struct net_device *netdev)
3251 struct e1000_adapter *adapter = netdev_priv(netdev);
3253 /* Do the reset outside of interrupt context */
3254 adapter->tx_timeout_count++;
3255 schedule_work(&adapter->reset_task);
3258 static void e1000_reset_task(struct work_struct *work)
3260 struct e1000_adapter *adapter =
3261 container_of(work, struct e1000_adapter, reset_task);
3263 if (test_bit(__E1000_DOWN, &adapter->flags))
3264 return;
3265 e1000_reinit_safe(adapter);
3269 * e1000_get_stats - Get System Network Statistics
3270 * @netdev: network interface device structure
3272 * Returns the address of the device statistics structure.
3273 * The statistics are actually updated from the watchdog.
3276 static struct net_device_stats *e1000_get_stats(struct net_device *netdev)
3278 /* only return the current stats */
3279 return &netdev->stats;
3283 * e1000_change_mtu - Change the Maximum Transfer Unit
3284 * @netdev: network interface device structure
3285 * @new_mtu: new value for maximum frame size
3287 * Returns 0 on success, negative on failure
3290 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
3292 struct e1000_adapter *adapter = netdev_priv(netdev);
3293 struct e1000_hw *hw = &adapter->hw;
3294 int max_frame = new_mtu + ENET_HEADER_SIZE + ETHERNET_FCS_SIZE;
3296 if ((max_frame < MINIMUM_ETHERNET_FRAME_SIZE) ||
3297 (max_frame > MAX_JUMBO_FRAME_SIZE)) {
3298 e_err(probe, "Invalid MTU setting\n");
3299 return -EINVAL;
3302 /* Adapter-specific max frame size limits. */
3303 switch (hw->mac_type) {
3304 case e1000_undefined ... e1000_82542_rev2_1:
3305 if (max_frame > (ETH_FRAME_LEN + ETH_FCS_LEN)) {
3306 e_err(probe, "Jumbo Frames not supported.\n");
3307 return -EINVAL;
3309 break;
3310 default:
3311 /* Capable of supporting up to MAX_JUMBO_FRAME_SIZE limit. */
3312 break;
3315 while (test_and_set_bit(__E1000_RESETTING, &adapter->flags))
3316 msleep(1);
3317 /* e1000_down has a dependency on max_frame_size */
3318 hw->max_frame_size = max_frame;
3319 if (netif_running(netdev))
3320 e1000_down(adapter);
3322 /* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
3323 * means we reserve 2 more, this pushes us to allocate from the next
3324 * larger slab size.
3325 * i.e. RXBUFFER_2048 --> size-4096 slab
3326 * however with the new *_jumbo_rx* routines, jumbo receives will use
3327 * fragmented skbs */
3329 if (max_frame <= E1000_RXBUFFER_2048)
3330 adapter->rx_buffer_len = E1000_RXBUFFER_2048;
3331 else
3332 #if (PAGE_SIZE >= E1000_RXBUFFER_16384)
3333 adapter->rx_buffer_len = E1000_RXBUFFER_16384;
3334 #elif (PAGE_SIZE >= E1000_RXBUFFER_4096)
3335 adapter->rx_buffer_len = PAGE_SIZE;
3336 #endif
3338 /* adjust allocation if LPE protects us, and we aren't using SBP */
3339 if (!hw->tbi_compatibility_on &&
3340 ((max_frame == (ETH_FRAME_LEN + ETH_FCS_LEN)) ||
3341 (max_frame == MAXIMUM_ETHERNET_VLAN_SIZE)))
3342 adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
3344 pr_info("%s changing MTU from %d to %d\n",
3345 netdev->name, netdev->mtu, new_mtu);
3346 netdev->mtu = new_mtu;
3348 if (netif_running(netdev))
3349 e1000_up(adapter);
3350 else
3351 e1000_reset(adapter);
3353 clear_bit(__E1000_RESETTING, &adapter->flags);
3355 return 0;
3359 * e1000_update_stats - Update the board statistics counters
3360 * @adapter: board private structure
3363 void e1000_update_stats(struct e1000_adapter *adapter)
3365 struct net_device *netdev = adapter->netdev;
3366 struct e1000_hw *hw = &adapter->hw;
3367 struct pci_dev *pdev = adapter->pdev;
3368 unsigned long flags;
3369 u16 phy_tmp;
3371 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
3374 * Prevent stats update while adapter is being reset, or if the pci
3375 * connection is down.
3377 if (adapter->link_speed == 0)
3378 return;
3379 if (pci_channel_offline(pdev))
3380 return;
3382 spin_lock_irqsave(&adapter->stats_lock, flags);
3384 /* these counters are modified from e1000_tbi_adjust_stats,
3385 * called from the interrupt context, so they must only
3386 * be written while holding adapter->stats_lock
3389 adapter->stats.crcerrs += er32(CRCERRS);
3390 adapter->stats.gprc += er32(GPRC);
3391 adapter->stats.gorcl += er32(GORCL);
3392 adapter->stats.gorch += er32(GORCH);
3393 adapter->stats.bprc += er32(BPRC);
3394 adapter->stats.mprc += er32(MPRC);
3395 adapter->stats.roc += er32(ROC);
3397 adapter->stats.prc64 += er32(PRC64);
3398 adapter->stats.prc127 += er32(PRC127);
3399 adapter->stats.prc255 += er32(PRC255);
3400 adapter->stats.prc511 += er32(PRC511);
3401 adapter->stats.prc1023 += er32(PRC1023);
3402 adapter->stats.prc1522 += er32(PRC1522);
3404 adapter->stats.symerrs += er32(SYMERRS);
3405 adapter->stats.mpc += er32(MPC);
3406 adapter->stats.scc += er32(SCC);
3407 adapter->stats.ecol += er32(ECOL);
3408 adapter->stats.mcc += er32(MCC);
3409 adapter->stats.latecol += er32(LATECOL);
3410 adapter->stats.dc += er32(DC);
3411 adapter->stats.sec += er32(SEC);
3412 adapter->stats.rlec += er32(RLEC);
3413 adapter->stats.xonrxc += er32(XONRXC);
3414 adapter->stats.xontxc += er32(XONTXC);
3415 adapter->stats.xoffrxc += er32(XOFFRXC);
3416 adapter->stats.xofftxc += er32(XOFFTXC);
3417 adapter->stats.fcruc += er32(FCRUC);
3418 adapter->stats.gptc += er32(GPTC);
3419 adapter->stats.gotcl += er32(GOTCL);
3420 adapter->stats.gotch += er32(GOTCH);
3421 adapter->stats.rnbc += er32(RNBC);
3422 adapter->stats.ruc += er32(RUC);
3423 adapter->stats.rfc += er32(RFC);
3424 adapter->stats.rjc += er32(RJC);
3425 adapter->stats.torl += er32(TORL);
3426 adapter->stats.torh += er32(TORH);
3427 adapter->stats.totl += er32(TOTL);
3428 adapter->stats.toth += er32(TOTH);
3429 adapter->stats.tpr += er32(TPR);
3431 adapter->stats.ptc64 += er32(PTC64);
3432 adapter->stats.ptc127 += er32(PTC127);
3433 adapter->stats.ptc255 += er32(PTC255);
3434 adapter->stats.ptc511 += er32(PTC511);
3435 adapter->stats.ptc1023 += er32(PTC1023);
3436 adapter->stats.ptc1522 += er32(PTC1522);
3438 adapter->stats.mptc += er32(MPTC);
3439 adapter->stats.bptc += er32(BPTC);
3441 /* used for adaptive IFS */
3443 hw->tx_packet_delta = er32(TPT);
3444 adapter->stats.tpt += hw->tx_packet_delta;
3445 hw->collision_delta = er32(COLC);
3446 adapter->stats.colc += hw->collision_delta;
3448 if (hw->mac_type >= e1000_82543) {
3449 adapter->stats.algnerrc += er32(ALGNERRC);
3450 adapter->stats.rxerrc += er32(RXERRC);
3451 adapter->stats.tncrs += er32(TNCRS);
3452 adapter->stats.cexterr += er32(CEXTERR);
3453 adapter->stats.tsctc += er32(TSCTC);
3454 adapter->stats.tsctfc += er32(TSCTFC);
3457 /* Fill out the OS statistics structure */
3458 netdev->stats.multicast = adapter->stats.mprc;
3459 netdev->stats.collisions = adapter->stats.colc;
3461 /* Rx Errors */
3463 /* RLEC on some newer hardware can be incorrect so build
3464 * our own version based on RUC and ROC */
3465 netdev->stats.rx_errors = adapter->stats.rxerrc +
3466 adapter->stats.crcerrs + adapter->stats.algnerrc +
3467 adapter->stats.ruc + adapter->stats.roc +
3468 adapter->stats.cexterr;
3469 adapter->stats.rlerrc = adapter->stats.ruc + adapter->stats.roc;
3470 netdev->stats.rx_length_errors = adapter->stats.rlerrc;
3471 netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
3472 netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
3473 netdev->stats.rx_missed_errors = adapter->stats.mpc;
3475 /* Tx Errors */
3476 adapter->stats.txerrc = adapter->stats.ecol + adapter->stats.latecol;
3477 netdev->stats.tx_errors = adapter->stats.txerrc;
3478 netdev->stats.tx_aborted_errors = adapter->stats.ecol;
3479 netdev->stats.tx_window_errors = adapter->stats.latecol;
3480 netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
3481 if (hw->bad_tx_carr_stats_fd &&
3482 adapter->link_duplex == FULL_DUPLEX) {
3483 netdev->stats.tx_carrier_errors = 0;
3484 adapter->stats.tncrs = 0;
3487 /* Tx Dropped needs to be maintained elsewhere */
3489 /* Phy Stats */
3490 if (hw->media_type == e1000_media_type_copper) {
3491 if ((adapter->link_speed == SPEED_1000) &&
3492 (!e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_tmp))) {
3493 phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
3494 adapter->phy_stats.idle_errors += phy_tmp;
3497 if ((hw->mac_type <= e1000_82546) &&
3498 (hw->phy_type == e1000_phy_m88) &&
3499 !e1000_read_phy_reg(hw, M88E1000_RX_ERR_CNTR, &phy_tmp))
3500 adapter->phy_stats.receive_errors += phy_tmp;
3503 /* Management Stats */
3504 if (hw->has_smbus) {
3505 adapter->stats.mgptc += er32(MGTPTC);
3506 adapter->stats.mgprc += er32(MGTPRC);
3507 adapter->stats.mgpdc += er32(MGTPDC);
3510 spin_unlock_irqrestore(&adapter->stats_lock, flags);
3514 * e1000_intr - Interrupt Handler
3515 * @irq: interrupt number
3516 * @data: pointer to a network interface device structure
3519 static irqreturn_t e1000_intr(int irq, void *data)
3521 struct net_device *netdev = data;
3522 struct e1000_adapter *adapter = netdev_priv(netdev);
3523 struct e1000_hw *hw = &adapter->hw;
3524 u32 icr = er32(ICR);
3526 if (unlikely((!icr)))
3527 return IRQ_NONE; /* Not our interrupt */
3530 * we might have caused the interrupt, but the above
3531 * read cleared it, and just in case the driver is
3532 * down there is nothing to do so return handled
3534 if (unlikely(test_bit(__E1000_DOWN, &adapter->flags)))
3535 return IRQ_HANDLED;
3537 if (unlikely(icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC))) {
3538 hw->get_link_status = 1;
3539 /* guard against interrupt when we're going down */
3540 if (!test_bit(__E1000_DOWN, &adapter->flags))
3541 schedule_delayed_work(&adapter->watchdog_task, 1);
3544 /* disable interrupts, without the synchronize_irq bit */
3545 ew32(IMC, ~0);
3546 E1000_WRITE_FLUSH();
3548 if (likely(napi_schedule_prep(&adapter->napi))) {
3549 adapter->total_tx_bytes = 0;
3550 adapter->total_tx_packets = 0;
3551 adapter->total_rx_bytes = 0;
3552 adapter->total_rx_packets = 0;
3553 __napi_schedule(&adapter->napi);
3554 } else {
3555 /* this really should not happen! if it does it is basically a
3556 * bug, but not a hard error, so enable ints and continue */
3557 if (!test_bit(__E1000_DOWN, &adapter->flags))
3558 e1000_irq_enable(adapter);
3561 return IRQ_HANDLED;
3565 * e1000_clean - NAPI Rx polling callback
3566 * @adapter: board private structure
3568 static int e1000_clean(struct napi_struct *napi, int budget)
3570 struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter, napi);
3571 int tx_clean_complete = 0, work_done = 0;
3573 tx_clean_complete = e1000_clean_tx_irq(adapter, &adapter->tx_ring[0]);
3575 adapter->clean_rx(adapter, &adapter->rx_ring[0], &work_done, budget);
3577 if (!tx_clean_complete)
3578 work_done = budget;
3580 /* If budget not fully consumed, exit the polling mode */
3581 if (work_done < budget) {
3582 if (likely(adapter->itr_setting & 3))
3583 e1000_set_itr(adapter);
3584 napi_complete(napi);
3585 if (!test_bit(__E1000_DOWN, &adapter->flags))
3586 e1000_irq_enable(adapter);
3589 return work_done;
3593 * e1000_clean_tx_irq - Reclaim resources after transmit completes
3594 * @adapter: board private structure
3596 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter,
3597 struct e1000_tx_ring *tx_ring)
3599 struct e1000_hw *hw = &adapter->hw;
3600 struct net_device *netdev = adapter->netdev;
3601 struct e1000_tx_desc *tx_desc, *eop_desc;
3602 struct e1000_buffer *buffer_info;
3603 unsigned int i, eop;
3604 unsigned int count = 0;
3605 unsigned int total_tx_bytes=0, total_tx_packets=0;
3607 i = tx_ring->next_to_clean;
3608 eop = tx_ring->buffer_info[i].next_to_watch;
3609 eop_desc = E1000_TX_DESC(*tx_ring, eop);
3611 while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
3612 (count < tx_ring->count)) {
3613 bool cleaned = false;
3614 rmb(); /* read buffer_info after eop_desc */
3615 for ( ; !cleaned; count++) {
3616 tx_desc = E1000_TX_DESC(*tx_ring, i);
3617 buffer_info = &tx_ring->buffer_info[i];
3618 cleaned = (i == eop);
3620 if (cleaned) {
3621 total_tx_packets += buffer_info->segs;
3622 total_tx_bytes += buffer_info->bytecount;
3624 e1000_unmap_and_free_tx_resource(adapter, buffer_info);
3625 tx_desc->upper.data = 0;
3627 if (unlikely(++i == tx_ring->count)) i = 0;
3630 eop = tx_ring->buffer_info[i].next_to_watch;
3631 eop_desc = E1000_TX_DESC(*tx_ring, eop);
3634 tx_ring->next_to_clean = i;
3636 #define TX_WAKE_THRESHOLD 32
3637 if (unlikely(count && netif_carrier_ok(netdev) &&
3638 E1000_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD)) {
3639 /* Make sure that anybody stopping the queue after this
3640 * sees the new next_to_clean.
3642 smp_mb();
3644 if (netif_queue_stopped(netdev) &&
3645 !(test_bit(__E1000_DOWN, &adapter->flags))) {
3646 netif_wake_queue(netdev);
3647 ++adapter->restart_queue;
3651 if (adapter->detect_tx_hung) {
3652 /* Detect a transmit hang in hardware, this serializes the
3653 * check with the clearing of time_stamp and movement of i */
3654 adapter->detect_tx_hung = false;
3655 if (tx_ring->buffer_info[eop].time_stamp &&
3656 time_after(jiffies, tx_ring->buffer_info[eop].time_stamp +
3657 (adapter->tx_timeout_factor * HZ)) &&
3658 !(er32(STATUS) & E1000_STATUS_TXOFF)) {
3660 /* detected Tx unit hang */
3661 e_err(drv, "Detected Tx Unit Hang\n"
3662 " Tx Queue <%lu>\n"
3663 " TDH <%x>\n"
3664 " TDT <%x>\n"
3665 " next_to_use <%x>\n"
3666 " next_to_clean <%x>\n"
3667 "buffer_info[next_to_clean]\n"
3668 " time_stamp <%lx>\n"
3669 " next_to_watch <%x>\n"
3670 " jiffies <%lx>\n"
3671 " next_to_watch.status <%x>\n",
3672 (unsigned long)((tx_ring - adapter->tx_ring) /
3673 sizeof(struct e1000_tx_ring)),
3674 readl(hw->hw_addr + tx_ring->tdh),
3675 readl(hw->hw_addr + tx_ring->tdt),
3676 tx_ring->next_to_use,
3677 tx_ring->next_to_clean,
3678 tx_ring->buffer_info[eop].time_stamp,
3679 eop,
3680 jiffies,
3681 eop_desc->upper.fields.status);
3682 netif_stop_queue(netdev);
3685 adapter->total_tx_bytes += total_tx_bytes;
3686 adapter->total_tx_packets += total_tx_packets;
3687 netdev->stats.tx_bytes += total_tx_bytes;
3688 netdev->stats.tx_packets += total_tx_packets;
3689 return count < tx_ring->count;
3693 * e1000_rx_checksum - Receive Checksum Offload for 82543
3694 * @adapter: board private structure
3695 * @status_err: receive descriptor status and error fields
3696 * @csum: receive descriptor csum field
3697 * @sk_buff: socket buffer with received data
3700 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
3701 u32 csum, struct sk_buff *skb)
3703 struct e1000_hw *hw = &adapter->hw;
3704 u16 status = (u16)status_err;
3705 u8 errors = (u8)(status_err >> 24);
3707 skb_checksum_none_assert(skb);
3709 /* 82543 or newer only */
3710 if (unlikely(hw->mac_type < e1000_82543)) return;
3711 /* Ignore Checksum bit is set */
3712 if (unlikely(status & E1000_RXD_STAT_IXSM)) return;
3713 /* TCP/UDP checksum error bit is set */
3714 if (unlikely(errors & E1000_RXD_ERR_TCPE)) {
3715 /* let the stack verify checksum errors */
3716 adapter->hw_csum_err++;
3717 return;
3719 /* TCP/UDP Checksum has not been calculated */
3720 if (!(status & E1000_RXD_STAT_TCPCS))
3721 return;
3723 /* It must be a TCP or UDP packet with a valid checksum */
3724 if (likely(status & E1000_RXD_STAT_TCPCS)) {
3725 /* TCP checksum is good */
3726 skb->ip_summed = CHECKSUM_UNNECESSARY;
3728 adapter->hw_csum_good++;
3732 * e1000_consume_page - helper function
3734 static void e1000_consume_page(struct e1000_buffer *bi, struct sk_buff *skb,
3735 u16 length)
3737 bi->page = NULL;
3738 skb->len += length;
3739 skb->data_len += length;
3740 skb->truesize += PAGE_SIZE;
3744 * e1000_receive_skb - helper function to handle rx indications
3745 * @adapter: board private structure
3746 * @status: descriptor status field as written by hardware
3747 * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
3748 * @skb: pointer to sk_buff to be indicated to stack
3750 static void e1000_receive_skb(struct e1000_adapter *adapter, u8 status,
3751 __le16 vlan, struct sk_buff *skb)
3753 skb->protocol = eth_type_trans(skb, adapter->netdev);
3755 if (status & E1000_RXD_STAT_VP) {
3756 u16 vid = le16_to_cpu(vlan) & E1000_RXD_SPC_VLAN_MASK;
3758 __vlan_hwaccel_put_tag(skb, vid);
3760 napi_gro_receive(&adapter->napi, skb);
3764 * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
3765 * @adapter: board private structure
3766 * @rx_ring: ring to clean
3767 * @work_done: amount of napi work completed this call
3768 * @work_to_do: max amount of work allowed for this call to do
3770 * the return value indicates whether actual cleaning was done, there
3771 * is no guarantee that everything was cleaned
3773 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
3774 struct e1000_rx_ring *rx_ring,
3775 int *work_done, int work_to_do)
3777 struct e1000_hw *hw = &adapter->hw;
3778 struct net_device *netdev = adapter->netdev;
3779 struct pci_dev *pdev = adapter->pdev;
3780 struct e1000_rx_desc *rx_desc, *next_rxd;
3781 struct e1000_buffer *buffer_info, *next_buffer;
3782 unsigned long irq_flags;
3783 u32 length;
3784 unsigned int i;
3785 int cleaned_count = 0;
3786 bool cleaned = false;
3787 unsigned int total_rx_bytes=0, total_rx_packets=0;
3789 i = rx_ring->next_to_clean;
3790 rx_desc = E1000_RX_DESC(*rx_ring, i);
3791 buffer_info = &rx_ring->buffer_info[i];
3793 while (rx_desc->status & E1000_RXD_STAT_DD) {
3794 struct sk_buff *skb;
3795 u8 status;
3797 if (*work_done >= work_to_do)
3798 break;
3799 (*work_done)++;
3800 rmb(); /* read descriptor and rx_buffer_info after status DD */
3802 status = rx_desc->status;
3803 skb = buffer_info->skb;
3804 buffer_info->skb = NULL;
3806 if (++i == rx_ring->count) i = 0;
3807 next_rxd = E1000_RX_DESC(*rx_ring, i);
3808 prefetch(next_rxd);
3810 next_buffer = &rx_ring->buffer_info[i];
3812 cleaned = true;
3813 cleaned_count++;
3814 dma_unmap_page(&pdev->dev, buffer_info->dma,
3815 buffer_info->length, DMA_FROM_DEVICE);
3816 buffer_info->dma = 0;
3818 length = le16_to_cpu(rx_desc->length);
3820 /* errors is only valid for DD + EOP descriptors */
3821 if (unlikely((status & E1000_RXD_STAT_EOP) &&
3822 (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK))) {
3823 u8 last_byte = *(skb->data + length - 1);
3824 if (TBI_ACCEPT(hw, status, rx_desc->errors, length,
3825 last_byte)) {
3826 spin_lock_irqsave(&adapter->stats_lock,
3827 irq_flags);
3828 e1000_tbi_adjust_stats(hw, &adapter->stats,
3829 length, skb->data);
3830 spin_unlock_irqrestore(&adapter->stats_lock,
3831 irq_flags);
3832 length--;
3833 } else {
3834 /* recycle both page and skb */
3835 buffer_info->skb = skb;
3836 /* an error means any chain goes out the window
3837 * too */
3838 if (rx_ring->rx_skb_top)
3839 dev_kfree_skb(rx_ring->rx_skb_top);
3840 rx_ring->rx_skb_top = NULL;
3841 goto next_desc;
3845 #define rxtop rx_ring->rx_skb_top
3846 if (!(status & E1000_RXD_STAT_EOP)) {
3847 /* this descriptor is only the beginning (or middle) */
3848 if (!rxtop) {
3849 /* this is the beginning of a chain */
3850 rxtop = skb;
3851 skb_fill_page_desc(rxtop, 0, buffer_info->page,
3852 0, length);
3853 } else {
3854 /* this is the middle of a chain */
3855 skb_fill_page_desc(rxtop,
3856 skb_shinfo(rxtop)->nr_frags,
3857 buffer_info->page, 0, length);
3858 /* re-use the skb, only consumed the page */
3859 buffer_info->skb = skb;
3861 e1000_consume_page(buffer_info, rxtop, length);
3862 goto next_desc;
3863 } else {
3864 if (rxtop) {
3865 /* end of the chain */
3866 skb_fill_page_desc(rxtop,
3867 skb_shinfo(rxtop)->nr_frags,
3868 buffer_info->page, 0, length);
3869 /* re-use the current skb, we only consumed the
3870 * page */
3871 buffer_info->skb = skb;
3872 skb = rxtop;
3873 rxtop = NULL;
3874 e1000_consume_page(buffer_info, skb, length);
3875 } else {
3876 /* no chain, got EOP, this buf is the packet
3877 * copybreak to save the put_page/alloc_page */
3878 if (length <= copybreak &&
3879 skb_tailroom(skb) >= length) {
3880 u8 *vaddr;
3881 vaddr = kmap_atomic(buffer_info->page,
3882 KM_SKB_DATA_SOFTIRQ);
3883 memcpy(skb_tail_pointer(skb), vaddr, length);
3884 kunmap_atomic(vaddr,
3885 KM_SKB_DATA_SOFTIRQ);
3886 /* re-use the page, so don't erase
3887 * buffer_info->page */
3888 skb_put(skb, length);
3889 } else {
3890 skb_fill_page_desc(skb, 0,
3891 buffer_info->page, 0,
3892 length);
3893 e1000_consume_page(buffer_info, skb,
3894 length);
3899 /* Receive Checksum Offload XXX recompute due to CRC strip? */
3900 e1000_rx_checksum(adapter,
3901 (u32)(status) |
3902 ((u32)(rx_desc->errors) << 24),
3903 le16_to_cpu(rx_desc->csum), skb);
3905 pskb_trim(skb, skb->len - 4);
3907 /* probably a little skewed due to removing CRC */
3908 total_rx_bytes += skb->len;
3909 total_rx_packets++;
3911 /* eth type trans needs skb->data to point to something */
3912 if (!pskb_may_pull(skb, ETH_HLEN)) {
3913 e_err(drv, "pskb_may_pull failed.\n");
3914 dev_kfree_skb(skb);
3915 goto next_desc;
3918 e1000_receive_skb(adapter, status, rx_desc->special, skb);
3920 next_desc:
3921 rx_desc->status = 0;
3923 /* return some buffers to hardware, one at a time is too slow */
3924 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
3925 adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
3926 cleaned_count = 0;
3929 /* use prefetched values */
3930 rx_desc = next_rxd;
3931 buffer_info = next_buffer;
3933 rx_ring->next_to_clean = i;
3935 cleaned_count = E1000_DESC_UNUSED(rx_ring);
3936 if (cleaned_count)
3937 adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
3939 adapter->total_rx_packets += total_rx_packets;
3940 adapter->total_rx_bytes += total_rx_bytes;
3941 netdev->stats.rx_bytes += total_rx_bytes;
3942 netdev->stats.rx_packets += total_rx_packets;
3943 return cleaned;
3947 * this should improve performance for small packets with large amounts
3948 * of reassembly being done in the stack
3950 static void e1000_check_copybreak(struct net_device *netdev,
3951 struct e1000_buffer *buffer_info,
3952 u32 length, struct sk_buff **skb)
3954 struct sk_buff *new_skb;
3956 if (length > copybreak)
3957 return;
3959 new_skb = netdev_alloc_skb_ip_align(netdev, length);
3960 if (!new_skb)
3961 return;
3963 skb_copy_to_linear_data_offset(new_skb, -NET_IP_ALIGN,
3964 (*skb)->data - NET_IP_ALIGN,
3965 length + NET_IP_ALIGN);
3966 /* save the skb in buffer_info as good */
3967 buffer_info->skb = *skb;
3968 *skb = new_skb;
3972 * e1000_clean_rx_irq - Send received data up the network stack; legacy
3973 * @adapter: board private structure
3974 * @rx_ring: ring to clean
3975 * @work_done: amount of napi work completed this call
3976 * @work_to_do: max amount of work allowed for this call to do
3978 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
3979 struct e1000_rx_ring *rx_ring,
3980 int *work_done, int work_to_do)
3982 struct e1000_hw *hw = &adapter->hw;
3983 struct net_device *netdev = adapter->netdev;
3984 struct pci_dev *pdev = adapter->pdev;
3985 struct e1000_rx_desc *rx_desc, *next_rxd;
3986 struct e1000_buffer *buffer_info, *next_buffer;
3987 unsigned long flags;
3988 u32 length;
3989 unsigned int i;
3990 int cleaned_count = 0;
3991 bool cleaned = false;
3992 unsigned int total_rx_bytes=0, total_rx_packets=0;
3994 i = rx_ring->next_to_clean;
3995 rx_desc = E1000_RX_DESC(*rx_ring, i);
3996 buffer_info = &rx_ring->buffer_info[i];
3998 while (rx_desc->status & E1000_RXD_STAT_DD) {
3999 struct sk_buff *skb;
4000 u8 status;
4002 if (*work_done >= work_to_do)
4003 break;
4004 (*work_done)++;
4005 rmb(); /* read descriptor and rx_buffer_info after status DD */
4007 status = rx_desc->status;
4008 skb = buffer_info->skb;
4009 buffer_info->skb = NULL;
4011 prefetch(skb->data - NET_IP_ALIGN);
4013 if (++i == rx_ring->count) i = 0;
4014 next_rxd = E1000_RX_DESC(*rx_ring, i);
4015 prefetch(next_rxd);
4017 next_buffer = &rx_ring->buffer_info[i];
4019 cleaned = true;
4020 cleaned_count++;
4021 dma_unmap_single(&pdev->dev, buffer_info->dma,
4022 buffer_info->length, DMA_FROM_DEVICE);
4023 buffer_info->dma = 0;
4025 length = le16_to_cpu(rx_desc->length);
4026 /* !EOP means multiple descriptors were used to store a single
4027 * packet, if thats the case we need to toss it. In fact, we
4028 * to toss every packet with the EOP bit clear and the next
4029 * frame that _does_ have the EOP bit set, as it is by
4030 * definition only a frame fragment
4032 if (unlikely(!(status & E1000_RXD_STAT_EOP)))
4033 adapter->discarding = true;
4035 if (adapter->discarding) {
4036 /* All receives must fit into a single buffer */
4037 e_dbg("Receive packet consumed multiple buffers\n");
4038 /* recycle */
4039 buffer_info->skb = skb;
4040 if (status & E1000_RXD_STAT_EOP)
4041 adapter->discarding = false;
4042 goto next_desc;
4045 if (unlikely(rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK)) {
4046 u8 last_byte = *(skb->data + length - 1);
4047 if (TBI_ACCEPT(hw, status, rx_desc->errors, length,
4048 last_byte)) {
4049 spin_lock_irqsave(&adapter->stats_lock, flags);
4050 e1000_tbi_adjust_stats(hw, &adapter->stats,
4051 length, skb->data);
4052 spin_unlock_irqrestore(&adapter->stats_lock,
4053 flags);
4054 length--;
4055 } else {
4056 /* recycle */
4057 buffer_info->skb = skb;
4058 goto next_desc;
4062 /* adjust length to remove Ethernet CRC, this must be
4063 * done after the TBI_ACCEPT workaround above */
4064 length -= 4;
4066 /* probably a little skewed due to removing CRC */
4067 total_rx_bytes += length;
4068 total_rx_packets++;
4070 e1000_check_copybreak(netdev, buffer_info, length, &skb);
4072 skb_put(skb, length);
4074 /* Receive Checksum Offload */
4075 e1000_rx_checksum(adapter,
4076 (u32)(status) |
4077 ((u32)(rx_desc->errors) << 24),
4078 le16_to_cpu(rx_desc->csum), skb);
4080 e1000_receive_skb(adapter, status, rx_desc->special, skb);
4082 next_desc:
4083 rx_desc->status = 0;
4085 /* return some buffers to hardware, one at a time is too slow */
4086 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
4087 adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
4088 cleaned_count = 0;
4091 /* use prefetched values */
4092 rx_desc = next_rxd;
4093 buffer_info = next_buffer;
4095 rx_ring->next_to_clean = i;
4097 cleaned_count = E1000_DESC_UNUSED(rx_ring);
4098 if (cleaned_count)
4099 adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
4101 adapter->total_rx_packets += total_rx_packets;
4102 adapter->total_rx_bytes += total_rx_bytes;
4103 netdev->stats.rx_bytes += total_rx_bytes;
4104 netdev->stats.rx_packets += total_rx_packets;
4105 return cleaned;
4109 * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
4110 * @adapter: address of board private structure
4111 * @rx_ring: pointer to receive ring structure
4112 * @cleaned_count: number of buffers to allocate this pass
4115 static void
4116 e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
4117 struct e1000_rx_ring *rx_ring, int cleaned_count)
4119 struct net_device *netdev = adapter->netdev;
4120 struct pci_dev *pdev = adapter->pdev;
4121 struct e1000_rx_desc *rx_desc;
4122 struct e1000_buffer *buffer_info;
4123 struct sk_buff *skb;
4124 unsigned int i;
4125 unsigned int bufsz = 256 - 16 /*for skb_reserve */ ;
4127 i = rx_ring->next_to_use;
4128 buffer_info = &rx_ring->buffer_info[i];
4130 while (cleaned_count--) {
4131 skb = buffer_info->skb;
4132 if (skb) {
4133 skb_trim(skb, 0);
4134 goto check_page;
4137 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
4138 if (unlikely(!skb)) {
4139 /* Better luck next round */
4140 adapter->alloc_rx_buff_failed++;
4141 break;
4144 /* Fix for errata 23, can't cross 64kB boundary */
4145 if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
4146 struct sk_buff *oldskb = skb;
4147 e_err(rx_err, "skb align check failed: %u bytes at "
4148 "%p\n", bufsz, skb->data);
4149 /* Try again, without freeing the previous */
4150 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
4151 /* Failed allocation, critical failure */
4152 if (!skb) {
4153 dev_kfree_skb(oldskb);
4154 adapter->alloc_rx_buff_failed++;
4155 break;
4158 if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
4159 /* give up */
4160 dev_kfree_skb(skb);
4161 dev_kfree_skb(oldskb);
4162 break; /* while (cleaned_count--) */
4165 /* Use new allocation */
4166 dev_kfree_skb(oldskb);
4168 buffer_info->skb = skb;
4169 buffer_info->length = adapter->rx_buffer_len;
4170 check_page:
4171 /* allocate a new page if necessary */
4172 if (!buffer_info->page) {
4173 buffer_info->page = alloc_page(GFP_ATOMIC);
4174 if (unlikely(!buffer_info->page)) {
4175 adapter->alloc_rx_buff_failed++;
4176 break;
4180 if (!buffer_info->dma) {
4181 buffer_info->dma = dma_map_page(&pdev->dev,
4182 buffer_info->page, 0,
4183 buffer_info->length,
4184 DMA_FROM_DEVICE);
4185 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
4186 put_page(buffer_info->page);
4187 dev_kfree_skb(skb);
4188 buffer_info->page = NULL;
4189 buffer_info->skb = NULL;
4190 buffer_info->dma = 0;
4191 adapter->alloc_rx_buff_failed++;
4192 break; /* while !buffer_info->skb */
4196 rx_desc = E1000_RX_DESC(*rx_ring, i);
4197 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4199 if (unlikely(++i == rx_ring->count))
4200 i = 0;
4201 buffer_info = &rx_ring->buffer_info[i];
4204 if (likely(rx_ring->next_to_use != i)) {
4205 rx_ring->next_to_use = i;
4206 if (unlikely(i-- == 0))
4207 i = (rx_ring->count - 1);
4209 /* Force memory writes to complete before letting h/w
4210 * know there are new descriptors to fetch. (Only
4211 * applicable for weak-ordered memory model archs,
4212 * such as IA-64). */
4213 wmb();
4214 writel(i, adapter->hw.hw_addr + rx_ring->rdt);
4219 * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
4220 * @adapter: address of board private structure
4223 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
4224 struct e1000_rx_ring *rx_ring,
4225 int cleaned_count)
4227 struct e1000_hw *hw = &adapter->hw;
4228 struct net_device *netdev = adapter->netdev;
4229 struct pci_dev *pdev = adapter->pdev;
4230 struct e1000_rx_desc *rx_desc;
4231 struct e1000_buffer *buffer_info;
4232 struct sk_buff *skb;
4233 unsigned int i;
4234 unsigned int bufsz = adapter->rx_buffer_len;
4236 i = rx_ring->next_to_use;
4237 buffer_info = &rx_ring->buffer_info[i];
4239 while (cleaned_count--) {
4240 skb = buffer_info->skb;
4241 if (skb) {
4242 skb_trim(skb, 0);
4243 goto map_skb;
4246 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
4247 if (unlikely(!skb)) {
4248 /* Better luck next round */
4249 adapter->alloc_rx_buff_failed++;
4250 break;
4253 /* Fix for errata 23, can't cross 64kB boundary */
4254 if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
4255 struct sk_buff *oldskb = skb;
4256 e_err(rx_err, "skb align check failed: %u bytes at "
4257 "%p\n", bufsz, skb->data);
4258 /* Try again, without freeing the previous */
4259 skb = netdev_alloc_skb_ip_align(netdev, bufsz);
4260 /* Failed allocation, critical failure */
4261 if (!skb) {
4262 dev_kfree_skb(oldskb);
4263 adapter->alloc_rx_buff_failed++;
4264 break;
4267 if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
4268 /* give up */
4269 dev_kfree_skb(skb);
4270 dev_kfree_skb(oldskb);
4271 adapter->alloc_rx_buff_failed++;
4272 break; /* while !buffer_info->skb */
4275 /* Use new allocation */
4276 dev_kfree_skb(oldskb);
4278 buffer_info->skb = skb;
4279 buffer_info->length = adapter->rx_buffer_len;
4280 map_skb:
4281 buffer_info->dma = dma_map_single(&pdev->dev,
4282 skb->data,
4283 buffer_info->length,
4284 DMA_FROM_DEVICE);
4285 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
4286 dev_kfree_skb(skb);
4287 buffer_info->skb = NULL;
4288 buffer_info->dma = 0;
4289 adapter->alloc_rx_buff_failed++;
4290 break; /* while !buffer_info->skb */
4294 * XXX if it was allocated cleanly it will never map to a
4295 * boundary crossing
4298 /* Fix for errata 23, can't cross 64kB boundary */
4299 if (!e1000_check_64k_bound(adapter,
4300 (void *)(unsigned long)buffer_info->dma,
4301 adapter->rx_buffer_len)) {
4302 e_err(rx_err, "dma align check failed: %u bytes at "
4303 "%p\n", adapter->rx_buffer_len,
4304 (void *)(unsigned long)buffer_info->dma);
4305 dev_kfree_skb(skb);
4306 buffer_info->skb = NULL;
4308 dma_unmap_single(&pdev->dev, buffer_info->dma,
4309 adapter->rx_buffer_len,
4310 DMA_FROM_DEVICE);
4311 buffer_info->dma = 0;
4313 adapter->alloc_rx_buff_failed++;
4314 break; /* while !buffer_info->skb */
4316 rx_desc = E1000_RX_DESC(*rx_ring, i);
4317 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4319 if (unlikely(++i == rx_ring->count))
4320 i = 0;
4321 buffer_info = &rx_ring->buffer_info[i];
4324 if (likely(rx_ring->next_to_use != i)) {
4325 rx_ring->next_to_use = i;
4326 if (unlikely(i-- == 0))
4327 i = (rx_ring->count - 1);
4329 /* Force memory writes to complete before letting h/w
4330 * know there are new descriptors to fetch. (Only
4331 * applicable for weak-ordered memory model archs,
4332 * such as IA-64). */
4333 wmb();
4334 writel(i, hw->hw_addr + rx_ring->rdt);
4339 * e1000_smartspeed - Workaround for SmartSpeed on 82541 and 82547 controllers.
4340 * @adapter:
4343 static void e1000_smartspeed(struct e1000_adapter *adapter)
4345 struct e1000_hw *hw = &adapter->hw;
4346 u16 phy_status;
4347 u16 phy_ctrl;
4349 if ((hw->phy_type != e1000_phy_igp) || !hw->autoneg ||
4350 !(hw->autoneg_advertised & ADVERTISE_1000_FULL))
4351 return;
4353 if (adapter->smartspeed == 0) {
4354 /* If Master/Slave config fault is asserted twice,
4355 * we assume back-to-back */
4356 e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_status);
4357 if (!(phy_status & SR_1000T_MS_CONFIG_FAULT)) return;
4358 e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_status);
4359 if (!(phy_status & SR_1000T_MS_CONFIG_FAULT)) return;
4360 e1000_read_phy_reg(hw, PHY_1000T_CTRL, &phy_ctrl);
4361 if (phy_ctrl & CR_1000T_MS_ENABLE) {
4362 phy_ctrl &= ~CR_1000T_MS_ENABLE;
4363 e1000_write_phy_reg(hw, PHY_1000T_CTRL,
4364 phy_ctrl);
4365 adapter->smartspeed++;
4366 if (!e1000_phy_setup_autoneg(hw) &&
4367 !e1000_read_phy_reg(hw, PHY_CTRL,
4368 &phy_ctrl)) {
4369 phy_ctrl |= (MII_CR_AUTO_NEG_EN |
4370 MII_CR_RESTART_AUTO_NEG);
4371 e1000_write_phy_reg(hw, PHY_CTRL,
4372 phy_ctrl);
4375 return;
4376 } else if (adapter->smartspeed == E1000_SMARTSPEED_DOWNSHIFT) {
4377 /* If still no link, perhaps using 2/3 pair cable */
4378 e1000_read_phy_reg(hw, PHY_1000T_CTRL, &phy_ctrl);
4379 phy_ctrl |= CR_1000T_MS_ENABLE;
4380 e1000_write_phy_reg(hw, PHY_1000T_CTRL, phy_ctrl);
4381 if (!e1000_phy_setup_autoneg(hw) &&
4382 !e1000_read_phy_reg(hw, PHY_CTRL, &phy_ctrl)) {
4383 phy_ctrl |= (MII_CR_AUTO_NEG_EN |
4384 MII_CR_RESTART_AUTO_NEG);
4385 e1000_write_phy_reg(hw, PHY_CTRL, phy_ctrl);
4388 /* Restart process after E1000_SMARTSPEED_MAX iterations */
4389 if (adapter->smartspeed++ == E1000_SMARTSPEED_MAX)
4390 adapter->smartspeed = 0;
4394 * e1000_ioctl -
4395 * @netdev:
4396 * @ifreq:
4397 * @cmd:
4400 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4402 switch (cmd) {
4403 case SIOCGMIIPHY:
4404 case SIOCGMIIREG:
4405 case SIOCSMIIREG:
4406 return e1000_mii_ioctl(netdev, ifr, cmd);
4407 default:
4408 return -EOPNOTSUPP;
4413 * e1000_mii_ioctl -
4414 * @netdev:
4415 * @ifreq:
4416 * @cmd:
4419 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
4420 int cmd)
4422 struct e1000_adapter *adapter = netdev_priv(netdev);
4423 struct e1000_hw *hw = &adapter->hw;
4424 struct mii_ioctl_data *data = if_mii(ifr);
4425 int retval;
4426 u16 mii_reg;
4427 unsigned long flags;
4429 if (hw->media_type != e1000_media_type_copper)
4430 return -EOPNOTSUPP;
4432 switch (cmd) {
4433 case SIOCGMIIPHY:
4434 data->phy_id = hw->phy_addr;
4435 break;
4436 case SIOCGMIIREG:
4437 spin_lock_irqsave(&adapter->stats_lock, flags);
4438 if (e1000_read_phy_reg(hw, data->reg_num & 0x1F,
4439 &data->val_out)) {
4440 spin_unlock_irqrestore(&adapter->stats_lock, flags);
4441 return -EIO;
4443 spin_unlock_irqrestore(&adapter->stats_lock, flags);
4444 break;
4445 case SIOCSMIIREG:
4446 if (data->reg_num & ~(0x1F))
4447 return -EFAULT;
4448 mii_reg = data->val_in;
4449 spin_lock_irqsave(&adapter->stats_lock, flags);
4450 if (e1000_write_phy_reg(hw, data->reg_num,
4451 mii_reg)) {
4452 spin_unlock_irqrestore(&adapter->stats_lock, flags);
4453 return -EIO;
4455 spin_unlock_irqrestore(&adapter->stats_lock, flags);
4456 if (hw->media_type == e1000_media_type_copper) {
4457 switch (data->reg_num) {
4458 case PHY_CTRL:
4459 if (mii_reg & MII_CR_POWER_DOWN)
4460 break;
4461 if (mii_reg & MII_CR_AUTO_NEG_EN) {
4462 hw->autoneg = 1;
4463 hw->autoneg_advertised = 0x2F;
4464 } else {
4465 u32 speed;
4466 if (mii_reg & 0x40)
4467 speed = SPEED_1000;
4468 else if (mii_reg & 0x2000)
4469 speed = SPEED_100;
4470 else
4471 speed = SPEED_10;
4472 retval = e1000_set_spd_dplx(
4473 adapter, speed,
4474 ((mii_reg & 0x100)
4475 ? DUPLEX_FULL :
4476 DUPLEX_HALF));
4477 if (retval)
4478 return retval;
4480 if (netif_running(adapter->netdev))
4481 e1000_reinit_locked(adapter);
4482 else
4483 e1000_reset(adapter);
4484 break;
4485 case M88E1000_PHY_SPEC_CTRL:
4486 case M88E1000_EXT_PHY_SPEC_CTRL:
4487 if (e1000_phy_reset(hw))
4488 return -EIO;
4489 break;
4491 } else {
4492 switch (data->reg_num) {
4493 case PHY_CTRL:
4494 if (mii_reg & MII_CR_POWER_DOWN)
4495 break;
4496 if (netif_running(adapter->netdev))
4497 e1000_reinit_locked(adapter);
4498 else
4499 e1000_reset(adapter);
4500 break;
4503 break;
4504 default:
4505 return -EOPNOTSUPP;
4507 return E1000_SUCCESS;
4510 void e1000_pci_set_mwi(struct e1000_hw *hw)
4512 struct e1000_adapter *adapter = hw->back;
4513 int ret_val = pci_set_mwi(adapter->pdev);
4515 if (ret_val)
4516 e_err(probe, "Error in setting MWI\n");
4519 void e1000_pci_clear_mwi(struct e1000_hw *hw)
4521 struct e1000_adapter *adapter = hw->back;
4523 pci_clear_mwi(adapter->pdev);
4526 int e1000_pcix_get_mmrbc(struct e1000_hw *hw)
4528 struct e1000_adapter *adapter = hw->back;
4529 return pcix_get_mmrbc(adapter->pdev);
4532 void e1000_pcix_set_mmrbc(struct e1000_hw *hw, int mmrbc)
4534 struct e1000_adapter *adapter = hw->back;
4535 pcix_set_mmrbc(adapter->pdev, mmrbc);
4538 void e1000_io_write(struct e1000_hw *hw, unsigned long port, u32 value)
4540 outl(value, port);
4543 static bool e1000_vlan_used(struct e1000_adapter *adapter)
4545 u16 vid;
4547 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
4548 return true;
4549 return false;
4552 static void e1000_vlan_filter_on_off(struct e1000_adapter *adapter,
4553 bool filter_on)
4555 struct e1000_hw *hw = &adapter->hw;
4556 u32 rctl;
4558 if (!test_bit(__E1000_DOWN, &adapter->flags))
4559 e1000_irq_disable(adapter);
4561 if (filter_on) {
4562 /* enable VLAN receive filtering */
4563 rctl = er32(RCTL);
4564 rctl &= ~E1000_RCTL_CFIEN;
4565 if (!(adapter->netdev->flags & IFF_PROMISC))
4566 rctl |= E1000_RCTL_VFE;
4567 ew32(RCTL, rctl);
4568 e1000_update_mng_vlan(adapter);
4569 } else {
4570 /* disable VLAN receive filtering */
4571 rctl = er32(RCTL);
4572 rctl &= ~E1000_RCTL_VFE;
4573 ew32(RCTL, rctl);
4576 if (!test_bit(__E1000_DOWN, &adapter->flags))
4577 e1000_irq_enable(adapter);
4580 static void e1000_vlan_mode(struct net_device *netdev, u32 features)
4582 struct e1000_adapter *adapter = netdev_priv(netdev);
4583 struct e1000_hw *hw = &adapter->hw;
4584 u32 ctrl;
4586 if (!test_bit(__E1000_DOWN, &adapter->flags))
4587 e1000_irq_disable(adapter);
4589 ctrl = er32(CTRL);
4590 if (features & NETIF_F_HW_VLAN_RX) {
4591 /* enable VLAN tag insert/strip */
4592 ctrl |= E1000_CTRL_VME;
4593 } else {
4594 /* disable VLAN tag insert/strip */
4595 ctrl &= ~E1000_CTRL_VME;
4597 ew32(CTRL, ctrl);
4599 if (!test_bit(__E1000_DOWN, &adapter->flags))
4600 e1000_irq_enable(adapter);
4603 static void e1000_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
4605 struct e1000_adapter *adapter = netdev_priv(netdev);
4606 struct e1000_hw *hw = &adapter->hw;
4607 u32 vfta, index;
4609 if ((hw->mng_cookie.status &
4610 E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) &&
4611 (vid == adapter->mng_vlan_id))
4612 return;
4614 if (!e1000_vlan_used(adapter))
4615 e1000_vlan_filter_on_off(adapter, true);
4617 /* add VID to filter table */
4618 index = (vid >> 5) & 0x7F;
4619 vfta = E1000_READ_REG_ARRAY(hw, VFTA, index);
4620 vfta |= (1 << (vid & 0x1F));
4621 e1000_write_vfta(hw, index, vfta);
4623 set_bit(vid, adapter->active_vlans);
4626 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
4628 struct e1000_adapter *adapter = netdev_priv(netdev);
4629 struct e1000_hw *hw = &adapter->hw;
4630 u32 vfta, index;
4632 if (!test_bit(__E1000_DOWN, &adapter->flags))
4633 e1000_irq_disable(adapter);
4634 if (!test_bit(__E1000_DOWN, &adapter->flags))
4635 e1000_irq_enable(adapter);
4637 /* remove VID from filter table */
4638 index = (vid >> 5) & 0x7F;
4639 vfta = E1000_READ_REG_ARRAY(hw, VFTA, index);
4640 vfta &= ~(1 << (vid & 0x1F));
4641 e1000_write_vfta(hw, index, vfta);
4643 clear_bit(vid, adapter->active_vlans);
4645 if (!e1000_vlan_used(adapter))
4646 e1000_vlan_filter_on_off(adapter, false);
4649 static void e1000_restore_vlan(struct e1000_adapter *adapter)
4651 u16 vid;
4653 if (!e1000_vlan_used(adapter))
4654 return;
4656 e1000_vlan_filter_on_off(adapter, true);
4657 for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
4658 e1000_vlan_rx_add_vid(adapter->netdev, vid);
4661 int e1000_set_spd_dplx(struct e1000_adapter *adapter, u32 spd, u8 dplx)
4663 struct e1000_hw *hw = &adapter->hw;
4665 hw->autoneg = 0;
4667 /* Make sure dplx is at most 1 bit and lsb of speed is not set
4668 * for the switch() below to work */
4669 if ((spd & 1) || (dplx & ~1))
4670 goto err_inval;
4672 /* Fiber NICs only allow 1000 gbps Full duplex */
4673 if ((hw->media_type == e1000_media_type_fiber) &&
4674 spd != SPEED_1000 &&
4675 dplx != DUPLEX_FULL)
4676 goto err_inval;
4678 switch (spd + dplx) {
4679 case SPEED_10 + DUPLEX_HALF:
4680 hw->forced_speed_duplex = e1000_10_half;
4681 break;
4682 case SPEED_10 + DUPLEX_FULL:
4683 hw->forced_speed_duplex = e1000_10_full;
4684 break;
4685 case SPEED_100 + DUPLEX_HALF:
4686 hw->forced_speed_duplex = e1000_100_half;
4687 break;
4688 case SPEED_100 + DUPLEX_FULL:
4689 hw->forced_speed_duplex = e1000_100_full;
4690 break;
4691 case SPEED_1000 + DUPLEX_FULL:
4692 hw->autoneg = 1;
4693 hw->autoneg_advertised = ADVERTISE_1000_FULL;
4694 break;
4695 case SPEED_1000 + DUPLEX_HALF: /* not supported */
4696 default:
4697 goto err_inval;
4699 return 0;
4701 err_inval:
4702 e_err(probe, "Unsupported Speed/Duplex configuration\n");
4703 return -EINVAL;
4706 static int __e1000_shutdown(struct pci_dev *pdev, bool *enable_wake)
4708 struct net_device *netdev = pci_get_drvdata(pdev);
4709 struct e1000_adapter *adapter = netdev_priv(netdev);
4710 struct e1000_hw *hw = &adapter->hw;
4711 u32 ctrl, ctrl_ext, rctl, status;
4712 u32 wufc = adapter->wol;
4713 #ifdef CONFIG_PM
4714 int retval = 0;
4715 #endif
4717 netif_device_detach(netdev);
4719 mutex_lock(&adapter->mutex);
4721 if (netif_running(netdev)) {
4722 WARN_ON(test_bit(__E1000_RESETTING, &adapter->flags));
4723 e1000_down(adapter);
4726 #ifdef CONFIG_PM
4727 retval = pci_save_state(pdev);
4728 if (retval) {
4729 mutex_unlock(&adapter->mutex);
4730 return retval;
4732 #endif
4734 status = er32(STATUS);
4735 if (status & E1000_STATUS_LU)
4736 wufc &= ~E1000_WUFC_LNKC;
4738 if (wufc) {
4739 e1000_setup_rctl(adapter);
4740 e1000_set_rx_mode(netdev);
4742 /* turn on all-multi mode if wake on multicast is enabled */
4743 if (wufc & E1000_WUFC_MC) {
4744 rctl = er32(RCTL);
4745 rctl |= E1000_RCTL_MPE;
4746 ew32(RCTL, rctl);
4749 if (hw->mac_type >= e1000_82540) {
4750 ctrl = er32(CTRL);
4751 /* advertise wake from D3Cold */
4752 #define E1000_CTRL_ADVD3WUC 0x00100000
4753 /* phy power management enable */
4754 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
4755 ctrl |= E1000_CTRL_ADVD3WUC |
4756 E1000_CTRL_EN_PHY_PWR_MGMT;
4757 ew32(CTRL, ctrl);
4760 if (hw->media_type == e1000_media_type_fiber ||
4761 hw->media_type == e1000_media_type_internal_serdes) {
4762 /* keep the laser running in D3 */
4763 ctrl_ext = er32(CTRL_EXT);
4764 ctrl_ext |= E1000_CTRL_EXT_SDP7_DATA;
4765 ew32(CTRL_EXT, ctrl_ext);
4768 ew32(WUC, E1000_WUC_PME_EN);
4769 ew32(WUFC, wufc);
4770 } else {
4771 ew32(WUC, 0);
4772 ew32(WUFC, 0);
4775 e1000_release_manageability(adapter);
4777 *enable_wake = !!wufc;
4779 /* make sure adapter isn't asleep if manageability is enabled */
4780 if (adapter->en_mng_pt)
4781 *enable_wake = true;
4783 if (netif_running(netdev))
4784 e1000_free_irq(adapter);
4786 mutex_unlock(&adapter->mutex);
4788 pci_disable_device(pdev);
4790 return 0;
4793 #ifdef CONFIG_PM
4794 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state)
4796 int retval;
4797 bool wake;
4799 retval = __e1000_shutdown(pdev, &wake);
4800 if (retval)
4801 return retval;
4803 if (wake) {
4804 pci_prepare_to_sleep(pdev);
4805 } else {
4806 pci_wake_from_d3(pdev, false);
4807 pci_set_power_state(pdev, PCI_D3hot);
4810 return 0;
4813 static int e1000_resume(struct pci_dev *pdev)
4815 struct net_device *netdev = pci_get_drvdata(pdev);
4816 struct e1000_adapter *adapter = netdev_priv(netdev);
4817 struct e1000_hw *hw = &adapter->hw;
4818 u32 err;
4820 pci_set_power_state(pdev, PCI_D0);
4821 pci_restore_state(pdev);
4822 pci_save_state(pdev);
4824 if (adapter->need_ioport)
4825 err = pci_enable_device(pdev);
4826 else
4827 err = pci_enable_device_mem(pdev);
4828 if (err) {
4829 pr_err("Cannot enable PCI device from suspend\n");
4830 return err;
4832 pci_set_master(pdev);
4834 pci_enable_wake(pdev, PCI_D3hot, 0);
4835 pci_enable_wake(pdev, PCI_D3cold, 0);
4837 if (netif_running(netdev)) {
4838 err = e1000_request_irq(adapter);
4839 if (err)
4840 return err;
4843 e1000_power_up_phy(adapter);
4844 e1000_reset(adapter);
4845 ew32(WUS, ~0);
4847 e1000_init_manageability(adapter);
4849 if (netif_running(netdev))
4850 e1000_up(adapter);
4852 netif_device_attach(netdev);
4854 return 0;
4856 #endif
4858 static void e1000_shutdown(struct pci_dev *pdev)
4860 bool wake;
4862 __e1000_shutdown(pdev, &wake);
4864 if (system_state == SYSTEM_POWER_OFF) {
4865 pci_wake_from_d3(pdev, wake);
4866 pci_set_power_state(pdev, PCI_D3hot);
4870 #ifdef CONFIG_NET_POLL_CONTROLLER
4872 * Polling 'interrupt' - used by things like netconsole to send skbs
4873 * without having to re-enable interrupts. It's not called while
4874 * the interrupt routine is executing.
4876 static void e1000_netpoll(struct net_device *netdev)
4878 struct e1000_adapter *adapter = netdev_priv(netdev);
4880 disable_irq(adapter->pdev->irq);
4881 e1000_intr(adapter->pdev->irq, netdev);
4882 enable_irq(adapter->pdev->irq);
4884 #endif
4887 * e1000_io_error_detected - called when PCI error is detected
4888 * @pdev: Pointer to PCI device
4889 * @state: The current pci connection state
4891 * This function is called after a PCI bus error affecting
4892 * this device has been detected.
4894 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
4895 pci_channel_state_t state)
4897 struct net_device *netdev = pci_get_drvdata(pdev);
4898 struct e1000_adapter *adapter = netdev_priv(netdev);
4900 netif_device_detach(netdev);
4902 if (state == pci_channel_io_perm_failure)
4903 return PCI_ERS_RESULT_DISCONNECT;
4905 if (netif_running(netdev))
4906 e1000_down(adapter);
4907 pci_disable_device(pdev);
4909 /* Request a slot slot reset. */
4910 return PCI_ERS_RESULT_NEED_RESET;
4914 * e1000_io_slot_reset - called after the pci bus has been reset.
4915 * @pdev: Pointer to PCI device
4917 * Restart the card from scratch, as if from a cold-boot. Implementation
4918 * resembles the first-half of the e1000_resume routine.
4920 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
4922 struct net_device *netdev = pci_get_drvdata(pdev);
4923 struct e1000_adapter *adapter = netdev_priv(netdev);
4924 struct e1000_hw *hw = &adapter->hw;
4925 int err;
4927 if (adapter->need_ioport)
4928 err = pci_enable_device(pdev);
4929 else
4930 err = pci_enable_device_mem(pdev);
4931 if (err) {
4932 pr_err("Cannot re-enable PCI device after reset.\n");
4933 return PCI_ERS_RESULT_DISCONNECT;
4935 pci_set_master(pdev);
4937 pci_enable_wake(pdev, PCI_D3hot, 0);
4938 pci_enable_wake(pdev, PCI_D3cold, 0);
4940 e1000_reset(adapter);
4941 ew32(WUS, ~0);
4943 return PCI_ERS_RESULT_RECOVERED;
4947 * e1000_io_resume - called when traffic can start flowing again.
4948 * @pdev: Pointer to PCI device
4950 * This callback is called when the error recovery driver tells us that
4951 * its OK to resume normal operation. Implementation resembles the
4952 * second-half of the e1000_resume routine.
4954 static void e1000_io_resume(struct pci_dev *pdev)
4956 struct net_device *netdev = pci_get_drvdata(pdev);
4957 struct e1000_adapter *adapter = netdev_priv(netdev);
4959 e1000_init_manageability(adapter);
4961 if (netif_running(netdev)) {
4962 if (e1000_up(adapter)) {
4963 pr_info("can't bring device back up after reset\n");
4964 return;
4968 netif_device_attach(netdev);
4971 /* e1000_main.c */